1This is gcc.info, produced by makeinfo version 6.5 from gcc.texi. 2 3Copyright (C) 1988-2022 Free Software Foundation, Inc. 4 5 Permission is granted to copy, distribute and/or modify this document 6under the terms of the GNU Free Documentation License, Version 1.3 or 7any later version published by the Free Software Foundation; with the 8Invariant Sections being "Funding Free Software", the Front-Cover Texts 9being (a) (see below), and with the Back-Cover Texts being (b) (see 10below). A copy of the license is included in the section entitled "GNU 11Free Documentation License". 12 13 (a) The FSF's Front-Cover Text is: 14 15 A GNU Manual 16 17 (b) The FSF's Back-Cover Text is: 18 19 You have freedom to copy and modify this GNU Manual, like GNU software. 20Copies published by the Free Software Foundation raise funds for GNU 21development. 22INFO-DIR-SECTION Software development 23START-INFO-DIR-ENTRY 24* gcc: (gcc). The GNU Compiler Collection. 25* g++: (gcc). The GNU C++ compiler. 26* gcov: (gcc) Gcov. 'gcov'--a test coverage program. 27* gcov-tool: (gcc) Gcov-tool. 'gcov-tool'--an offline gcda profile processing program. 28* gcov-dump: (gcc) Gcov-dump. 'gcov-dump'--an offline gcda and gcno profile dump tool. 29* lto-dump: (gcc) lto-dump. 'lto-dump'--Tool for 30dumping LTO object files. 31END-INFO-DIR-ENTRY 32 33 This file documents the use of the GNU compilers. 34 35 Copyright (C) 1988-2022 Free Software Foundation, Inc. 36 37 Permission is granted to copy, distribute and/or modify this document 38under the terms of the GNU Free Documentation License, Version 1.3 or 39any later version published by the Free Software Foundation; with the 40Invariant Sections being "Funding Free Software", the Front-Cover Texts 41being (a) (see below), and with the Back-Cover Texts being (b) (see 42below). A copy of the license is included in the section entitled "GNU 43Free Documentation License". 44 45 (a) The FSF's Front-Cover Text is: 46 47 A GNU Manual 48 49 (b) The FSF's Back-Cover Text is: 50 51 You have freedom to copy and modify this GNU Manual, like GNU software. 52Copies published by the Free Software Foundation raise funds for GNU 53development. 54 55 56File: gcc.info, Node: Top, Next: G++ and GCC 57 58Introduction 59************ 60 61This manual documents how to use the GNU compilers, as well as their 62features and incompatibilities, and how to report bugs. It corresponds 63to the compilers (GCC) version 12.3.0. The internals of the GNU 64compilers, including how to port them to new targets and some 65information about how to write front ends for new languages, are 66documented in a separate manual. *Note Introduction: (gccint)Top. 67 68* Menu: 69 70* G++ and GCC:: You can compile C or C++ programs. 71* Standards:: Language standards supported by GCC. 72* Invoking GCC:: Command options supported by 'gcc'. 73* C Implementation:: How GCC implements the ISO C specification. 74* C++ Implementation:: How GCC implements the ISO C++ specification. 75* C Extensions:: GNU extensions to the C language family. 76* C++ Extensions:: GNU extensions to the C++ language. 77* Objective-C:: GNU Objective-C runtime features. 78* Compatibility:: Binary Compatibility 79* Gcov:: 'gcov'--a test coverage program. 80* Gcov-tool:: 'gcov-tool'--an offline gcda profile processing program. 81* Gcov-dump:: 'gcov-dump'--an offline gcda and gcno profile dump tool. 82* lto-dump:: 'lto-dump'--Tool for dumping LTO 83object files. 84* Trouble:: If you have trouble using GCC. 85* Bugs:: How, why and where to report bugs. 86* Service:: How To Get Help with GCC 87* Contributing:: How to contribute to testing and developing GCC. 88 89* Funding:: How to help assure funding for free software. 90* GNU Project:: The GNU Project and GNU/Linux. 91 92* Copying:: GNU General Public License says 93 how you can copy and share GCC. 94* GNU Free Documentation License:: How you can copy and share this manual. 95* Contributors:: People who have contributed to GCC. 96 97* Option Index:: Index to command line options. 98* Keyword Index:: Index of concepts and symbol names. 99 100 101File: gcc.info, Node: G++ and GCC, Next: Standards, Up: Top 102 1031 Programming Languages Supported by GCC 104**************************************** 105 106GCC stands for "GNU Compiler Collection". GCC is an integrated 107distribution of compilers for several major programming languages. 108These languages currently include C, C++, Objective-C, Objective-C++, 109Fortran, Ada, D, and Go. 110 111 The abbreviation "GCC" has multiple meanings in common use. The 112current official meaning is "GNU Compiler Collection", which refers 113generically to the complete suite of tools. The name historically stood 114for "GNU C Compiler", and this usage is still common when the emphasis 115is on compiling C programs. Finally, the name is also used when 116speaking of the "language-independent" component of GCC: code shared 117among the compilers for all supported languages. 118 119 The language-independent component of GCC includes the majority of the 120optimizers, as well as the "back ends" that generate machine code for 121various processors. 122 123 The part of a compiler that is specific to a particular language is 124called the "front end". In addition to the front ends that are 125integrated components of GCC, there are several other front ends that 126are maintained separately. These support languages such as Mercury, and 127COBOL. To use these, they must be built together with GCC proper. 128 129 Most of the compilers for languages other than C have their own names. 130The C++ compiler is G++, the Ada compiler is GNAT, and so on. When we 131talk about compiling one of those languages, we might refer to that 132compiler by its own name, or as GCC. Either is correct. 133 134 Historically, compilers for many languages, including C++ and Fortran, 135have been implemented as "preprocessors" which emit another high level 136language such as C. None of the compilers included in GCC are 137implemented this way; they all generate machine code directly. This 138sort of preprocessor should not be confused with the "C preprocessor", 139which is an integral feature of the C, C++, Objective-C and 140Objective-C++ languages. 141 142 143File: gcc.info, Node: Standards, Next: Invoking GCC, Prev: G++ and GCC, Up: Top 144 1452 Language Standards Supported by GCC 146************************************* 147 148For each language compiled by GCC for which there is a standard, GCC 149attempts to follow one or more versions of that standard, possibly with 150some exceptions, and possibly with some extensions. 151 1522.1 C Language 153============== 154 155The original ANSI C standard (X3.159-1989) was ratified in 1989 and 156published in 1990. This standard was ratified as an ISO standard 157(ISO/IEC 9899:1990) later in 1990. There were no technical differences 158between these publications, although the sections of the ANSI standard 159were renumbered and became clauses in the ISO standard. The ANSI 160standard, but not the ISO standard, also came with a Rationale document. 161This standard, in both its forms, is commonly known as "C89", or 162occasionally as "C90", from the dates of ratification. To select this 163standard in GCC, use one of the options '-ansi', '-std=c90' or 164'-std=iso9899:1990'; to obtain all the diagnostics required by the 165standard, you should also specify '-pedantic' (or '-pedantic-errors' if 166you want them to be errors rather than warnings). *Note Options 167Controlling C Dialect: C Dialect Options. 168 169 Errors in the 1990 ISO C standard were corrected in two Technical 170Corrigenda published in 1994 and 1996. GCC does not support the 171uncorrected version. 172 173 An amendment to the 1990 standard was published in 1995. This 174amendment added digraphs and '__STDC_VERSION__' to the language, but 175otherwise concerned the library. This amendment is commonly known as 176"AMD1"; the amended standard is sometimes known as "C94" or "C95". To 177select this standard in GCC, use the option '-std=iso9899:199409' (with, 178as for other standard versions, '-pedantic' to receive all required 179diagnostics). 180 181 A new edition of the ISO C standard was published in 1999 as ISO/IEC 1829899:1999, and is commonly known as "C99". (While in development, 183drafts of this standard version were referred to as "C9X".) GCC has 184substantially complete support for this standard version; see 185<https://gcc.gnu.org/c99status.html> for details. To select this 186standard, use '-std=c99' or '-std=iso9899:1999'. 187 188 Errors in the 1999 ISO C standard were corrected in three Technical 189Corrigenda published in 2001, 2004 and 2007. GCC does not support the 190uncorrected version. 191 192 A fourth version of the C standard, known as "C11", was published in 1932011 as ISO/IEC 9899:2011. (While in development, drafts of this 194standard version were referred to as "C1X".) GCC has substantially 195complete support for this standard, enabled with '-std=c11' or 196'-std=iso9899:2011'. A version with corrections integrated was prepared 197in 2017 and published in 2018 as ISO/IEC 9899:2018; it is known as "C17" 198and is supported with '-std=c17' or '-std=iso9899:2017'; the corrections 199are also applied with '-std=c11', and the only difference between the 200options is the value of '__STDC_VERSION__'. 201 202 A further version of the C standard, known as "C2X", is under 203development; experimental and incomplete support for this is enabled 204with '-std=c2x'. 205 206 By default, GCC provides some extensions to the C language that, on 207rare occasions conflict with the C standard. *Note Extensions to the C 208Language Family: C Extensions. Some features that are part of the C99 209standard are accepted as extensions in C90 mode, and some features that 210are part of the C11 standard are accepted as extensions in C90 and C99 211modes. Use of the '-std' options listed above disables these extensions 212where they conflict with the C standard version selected. You may also 213select an extended version of the C language explicitly with 214'-std=gnu90' (for C90 with GNU extensions), '-std=gnu99' (for C99 with 215GNU extensions) or '-std=gnu11' (for C11 with GNU extensions). 216 217 The default, if no C language dialect options are given, is 218'-std=gnu17'. 219 220 The ISO C standard defines (in clause 4) two classes of conforming 221implementation. A "conforming hosted implementation" supports the whole 222standard including all the library facilities; a "conforming 223freestanding implementation" is only required to provide certain library 224facilities: those in '<float.h>', '<limits.h>', '<stdarg.h>', and 225'<stddef.h>'; since AMD1, also those in '<iso646.h>'; since C99, also 226those in '<stdbool.h>' and '<stdint.h>'; and since C11, also those in 227'<stdalign.h>' and '<stdnoreturn.h>'. In addition, complex types, added 228in C99, are not required for freestanding implementations. 229 230 The standard also defines two environments for programs, a 231"freestanding environment", required of all implementations and which 232may not have library facilities beyond those required of freestanding 233implementations, where the handling of program startup and termination 234are implementation-defined; and a "hosted environment", which is not 235required, in which all the library facilities are provided and startup 236is through a function 'int main (void)' or 'int main (int, char *[])'. 237An OS kernel is an example of a program running in a freestanding 238environment; a program using the facilities of an operating system is an 239example of a program running in a hosted environment. 240 241 GCC aims towards being usable as a conforming freestanding 242implementation, or as the compiler for a conforming hosted 243implementation. By default, it acts as the compiler for a hosted 244implementation, defining '__STDC_HOSTED__' as '1' and presuming that 245when the names of ISO C functions are used, they have the semantics 246defined in the standard. To make it act as a conforming freestanding 247implementation for a freestanding environment, use the option 248'-ffreestanding'; it then defines '__STDC_HOSTED__' to '0' and does not 249make assumptions about the meanings of function names from the standard 250library, with exceptions noted below. To build an OS kernel, you may 251well still need to make your own arrangements for linking and startup. 252*Note Options Controlling C Dialect: C Dialect Options. 253 254 GCC does not provide the library facilities required only of hosted 255implementations, nor yet all the facilities required by C99 of 256freestanding implementations on all platforms. To use the facilities of 257a hosted environment, you need to find them elsewhere (for example, in 258the GNU C library). *Note Standard Libraries: Standard Libraries. 259 260 Most of the compiler support routines used by GCC are present in 261'libgcc', but there are a few exceptions. GCC requires the freestanding 262environment provide 'memcpy', 'memmove', 'memset' and 'memcmp'. 263Finally, if '__builtin_trap' is used, and the target does not implement 264the 'trap' pattern, then GCC emits a call to 'abort'. 265 266 For references to Technical Corrigenda, Rationale documents and 267information concerning the history of C that is available online, see 268<https://gcc.gnu.org/readings.html> 269 2702.2 C++ Language 271================ 272 273GCC supports the original ISO C++ standard published in 1998, and the 2742011, 2014, 2017 and mostly 2020 revisions. 275 276 The original ISO C++ standard was published as the ISO standard 277(ISO/IEC 14882:1998) and amended by a Technical Corrigenda published in 2782003 (ISO/IEC 14882:2003). These standards are referred to as C++98 and 279C++03, respectively. GCC implements the majority of C++98 ('export' is 280a notable exception) and most of the changes in C++03. To select this 281standard in GCC, use one of the options '-ansi', '-std=c++98', or 282'-std=c++03'; to obtain all the diagnostics required by the standard, 283you should also specify '-pedantic' (or '-pedantic-errors' if you want 284them to be errors rather than warnings). 285 286 A revised ISO C++ standard was published in 2011 as ISO/IEC 14882:2011, 287and is referred to as C++11; before its publication it was commonly 288referred to as C++0x. C++11 contains several changes to the C++ 289language, all of which have been implemented in GCC. For details see 290<https://gcc.gnu.org/projects/cxx-status.html#cxx11>. To select this 291standard in GCC, use the option '-std=c++11'. 292 293 Another revised ISO C++ standard was published in 2014 as ISO/IEC 29414882:2014, and is referred to as C++14; before its publication it was 295sometimes referred to as C++1y. C++14 contains several further changes 296to the C++ language, all of which have been implemented in GCC. For 297details see <https://gcc.gnu.org/projects/cxx-status.html#cxx14>. To 298select this standard in GCC, use the option '-std=c++14'. 299 300 The C++ language was further revised in 2017 and ISO/IEC 14882:2017 was 301published. This is referred to as C++17, and before publication was 302often referred to as C++1z. GCC supports all the changes in that 303specification. For further details see 304<https://gcc.gnu.org/projects/cxx-status.html#cxx17>. Use the option 305'-std=c++17' to select this variant of C++. 306 307 Another revised ISO C++ standard was published in 2020 as ISO/IEC 30814882:2020, and is referred to as C++20; before its publication it was 309sometimes referred to as C++2a. GCC supports most of the changes in the 310new specification. For further details see 311<https://gcc.gnu.org/projects/cxx-status.html#cxx20>. To select this 312standard in GCC, use the option '-std=c++20'. 313 314 More information about the C++ standards is available on the ISO C++ 315committee's web site at <http://www.open-std.org/jtc1/sc22/wg21/>. 316 317 To obtain all the diagnostics required by any of the standard versions 318described above you should specify '-pedantic' or '-pedantic-errors', 319otherwise GCC will allow some non-ISO C++ features as extensions. *Note 320Warning Options::. 321 322 By default, GCC also provides some additional extensions to the C++ 323language that on rare occasions conflict with the C++ standard. *Note 324Options Controlling C++ Dialect: C++ Dialect Options. Use of the '-std' 325options listed above disables these extensions where they they conflict 326with the C++ standard version selected. You may also select an extended 327version of the C++ language explicitly with '-std=gnu++98' (for C++98 328with GNU extensions), or '-std=gnu++11' (for C++11 with GNU extensions), 329or '-std=gnu++14' (for C++14 with GNU extensions), or '-std=gnu++17' 330(for C++17 with GNU extensions), or '-std=gnu++20' (for C++20 with GNU 331extensions). 332 333 The default, if no C++ language dialect options are given, is 334'-std=gnu++17'. 335 3362.3 Objective-C and Objective-C++ Languages 337=========================================== 338 339GCC supports "traditional" Objective-C (also known as "Objective-C 1.0") 340and contains support for the Objective-C exception and synchronization 341syntax. It has also support for a number of "Objective-C 2.0" language 342extensions, including properties, fast enumeration (only for 343Objective-C), method attributes and the @optional and @required keywords 344in protocols. GCC supports Objective-C++ and features available in 345Objective-C are also available in Objective-C++. 346 347 GCC by default uses the GNU Objective-C runtime library, which is part 348of GCC and is not the same as the Apple/NeXT Objective-C runtime library 349used on Apple systems. There are a number of differences documented in 350this manual. The options '-fgnu-runtime' and '-fnext-runtime' allow you 351to switch between producing output that works with the GNU Objective-C 352runtime library and output that works with the Apple/NeXT Objective-C 353runtime library. 354 355 There is no formal written standard for Objective-C or Objective-C++. 356The authoritative manual on traditional Objective-C (1.0) is 357"Object-Oriented Programming and the Objective-C Language": 358<http://www.gnustep.org/resources/documentation/ObjectivCBook.pdf> is 359the original NeXTstep document. 360 361 The Objective-C exception and synchronization syntax (that is, the 362keywords '@try', '@throw', '@catch', '@finally' and '@synchronized') is 363supported by GCC and is enabled with the option '-fobjc-exceptions'. 364The syntax is briefly documented in this manual and in the Objective-C 3652.0 manuals from Apple. 366 367 The Objective-C 2.0 language extensions and features are automatically 368enabled; they include properties (via the '@property', '@synthesize' and 369'@dynamic keywords'), fast enumeration (not available in Objective-C++), 370attributes for methods (such as 'deprecated', 'noreturn', 'sentinel', 371'format'), the 'unused' attribute for method arguments, the '@package' 372keyword for instance variables and the '@optional' and '@required' 373keywords in protocols. You can disable all these Objective-C 2.0 374language extensions with the option '-fobjc-std=objc1', which causes the 375compiler to recognize the same Objective-C language syntax recognized by 376GCC 4.0, and to produce an error if one of the new features is used. 377 378 GCC has currently no support for non-fragile instance variables. 379 380 The authoritative manual on Objective-C 2.0 is available from Apple: 381 * 382 <https://developer.apple.com/library/archive/documentation/Cocoa/Conceptual/ProgrammingWithObjectiveC/Introduction/Introduction.html> 383 384 For more information concerning the history of Objective-C that is 385available online, see <https://gcc.gnu.org/readings.html> 386 3872.4 Go Language 388=============== 389 390As of the GCC 4.7.1 release, GCC supports the Go 1 language standard, 391described at <https://golang.org/doc/go1>. 392 3932.5 D language 394============== 395 396GCC supports the D 2.0 programming language. The D language itself is 397currently defined by its reference implementation and supporting 398language specification, described at <https://dlang.org/spec/spec.html>. 399 4002.6 References for Other Languages 401================================== 402 403*Note GNAT Reference Manual: (gnat_rm)Top, for information on standard 404conformance and compatibility of the Ada compiler. 405 406 *Note Standards: (gfortran)Standards, for details of standards 407supported by GNU Fortran. 408 409 410File: gcc.info, Node: Invoking GCC, Next: C Implementation, Prev: Standards, Up: Top 411 4123 GCC Command Options 413********************* 414 415When you invoke GCC, it normally does preprocessing, compilation, 416assembly and linking. The "overall options" allow you to stop this 417process at an intermediate stage. For example, the '-c' option says not 418to run the linker. Then the output consists of object files output by 419the assembler. *Note Options Controlling the Kind of Output: Overall 420Options. 421 422 Other options are passed on to one or more stages of processing. Some 423options control the preprocessor and others the compiler itself. Yet 424other options control the assembler and linker; most of these are not 425documented here, since you rarely need to use any of them. 426 427 Most of the command-line options that you can use with GCC are useful 428for C programs; when an option is only useful with another language 429(usually C++), the explanation says so explicitly. If the description 430for a particular option does not mention a source language, you can use 431that option with all supported languages. 432 433 The usual way to run GCC is to run the executable called 'gcc', or 434'MACHINE-gcc' when cross-compiling, or 'MACHINE-gcc-VERSION' to run a 435specific version of GCC. When you compile C++ programs, you should 436invoke GCC as 'g++' instead. *Note Compiling C++ Programs: Invoking 437G++, for information about the differences in behavior between 'gcc' and 438'g++' when compiling C++ programs. 439 440 The 'gcc' program accepts options and file names as operands. Many 441options have multi-letter names; therefore multiple single-letter 442options may _not_ be grouped: '-dv' is very different from '-d -v'. 443 444 You can mix options and other arguments. For the most part, the order 445you use doesn't matter. Order does matter when you use several options 446of the same kind; for example, if you specify '-L' more than once, the 447directories are searched in the order specified. Also, the placement of 448the '-l' option is significant. 449 450 Many options have long names starting with '-f' or with '-W'--for 451example, '-fmove-loop-invariants', '-Wformat' and so on. Most of these 452have both positive and negative forms; the negative form of '-ffoo' is 453'-fno-foo'. This manual documents only one of these two forms, 454whichever one is not the default. 455 456 Some options take one or more arguments typically separated either by a 457space or by the equals sign ('=') from the option name. Unless 458documented otherwise, an argument can be either numeric or a string. 459Numeric arguments must typically be small unsigned decimal or 460hexadecimal integers. Hexadecimal arguments must begin with the '0x' 461prefix. Arguments to options that specify a size threshold of some sort 462may be arbitrarily large decimal or hexadecimal integers followed by a 463byte size suffix designating a multiple of bytes such as 'kB' and 'KiB' 464for kilobyte and kibibyte, respectively, 'MB' and 'MiB' for megabyte and 465mebibyte, 'GB' and 'GiB' for gigabyte and gigibyte, and so on. Such 466arguments are designated by BYTE-SIZE in the following text. Refer to 467the NIST, IEC, and other relevant national and international standards 468for the full listing and explanation of the binary and decimal byte size 469prefixes. 470 471 *Note Option Index::, for an index to GCC's options. 472 473* Menu: 474 475* Option Summary:: Brief list of all options, without explanations. 476* Overall Options:: Controlling the kind of output: 477 an executable, object files, assembler files, 478 or preprocessed source. 479* Invoking G++:: Compiling C++ programs. 480* C Dialect Options:: Controlling the variant of C language compiled. 481* C++ Dialect Options:: Variations on C++. 482* Objective-C and Objective-C++ Dialect Options:: Variations on Objective-C 483 and Objective-C++. 484* Diagnostic Message Formatting Options:: Controlling how diagnostics should 485 be formatted. 486* Warning Options:: How picky should the compiler be? 487* Static Analyzer Options:: More expensive warnings. 488* Debugging Options:: Producing debuggable code. 489* Optimize Options:: How much optimization? 490* Instrumentation Options:: Enabling profiling and extra run-time error checking. 491* Preprocessor Options:: Controlling header files and macro definitions. 492 Also, getting dependency information for Make. 493* Assembler Options:: Passing options to the assembler. 494* Link Options:: Specifying libraries and so on. 495* Directory Options:: Where to find header files and libraries. 496 Where to find the compiler executable files. 497* Code Gen Options:: Specifying conventions for function calls, data layout 498 and register usage. 499* Developer Options:: Printing GCC configuration info, statistics, and 500 debugging dumps. 501* Submodel Options:: Target-specific options, such as compiling for a 502 specific processor variant. 503* Spec Files:: How to pass switches to sub-processes. 504* Environment Variables:: Env vars that affect GCC. 505* Precompiled Headers:: Compiling a header once, and using it many times. 506* C++ Modules:: Experimental C++20 module system. 507 508 509File: gcc.info, Node: Option Summary, Next: Overall Options, Up: Invoking GCC 510 5113.1 Option Summary 512================== 513 514Here is a summary of all the options, grouped by type. Explanations are 515in the following sections. 516 517_Overall Options_ 518 *Note Options Controlling the Kind of Output: Overall Options. 519 -c -S -E -o FILE 520 -dumpbase DUMPBASE -dumpbase-ext AUXDROPSUF 521 -dumpdir DUMPPFX -x LANGUAGE 522 -v -### --help[=CLASS[,...]] --target-help --version 523 -pass-exit-codes -pipe -specs=FILE -wrapper 524 @FILE -ffile-prefix-map=OLD=NEW 525 -fplugin=FILE -fplugin-arg-NAME=ARG 526 -fdump-ada-spec[-slim] -fada-spec-parent=UNIT -fdump-go-spec=FILE 527 528_C Language Options_ 529 *Note Options Controlling C Dialect: C Dialect Options. 530 -ansi -std=STANDARD -aux-info FILENAME 531 -fallow-parameterless-variadic-functions -fno-asm 532 -fno-builtin -fno-builtin-FUNCTION -fcond-mismatch 533 -ffreestanding -fgimple -fgnu-tm -fgnu89-inline -fhosted 534 -flax-vector-conversions -fms-extensions 535 -foffload=ARG -foffload-options=ARG 536 -fopenacc -fopenacc-dim=GEOM 537 -fopenmp -fopenmp-simd 538 -fpermitted-flt-eval-methods=STANDARD 539 -fplan9-extensions -fsigned-bitfields -funsigned-bitfields 540 -fsigned-char -funsigned-char -fsso-struct=ENDIANNESS 541 542_C++ Language Options_ 543 *Note Options Controlling C++ Dialect: C++ Dialect Options. 544 -fabi-version=N -fno-access-control 545 -faligned-new=N -fargs-in-order=N -fchar8_t -fcheck-new 546 -fconstexpr-depth=N -fconstexpr-cache-depth=N 547 -fconstexpr-loop-limit=N -fconstexpr-ops-limit=N 548 -fno-elide-constructors 549 -fno-enforce-eh-specs 550 -fno-gnu-keywords 551 -fno-implicit-templates 552 -fno-implicit-inline-templates 553 -fno-implement-inlines 554 -fmodule-header[=KIND] -fmodule-only -fmodules-ts 555 -fmodule-implicit-inline 556 -fno-module-lazy 557 -fmodule-mapper=SPECIFICATION 558 -fmodule-version-ignore 559 -fms-extensions 560 -fnew-inheriting-ctors 561 -fnew-ttp-matching 562 -fno-nonansi-builtins -fnothrow-opt -fno-operator-names 563 -fno-optional-diags -fpermissive 564 -fno-pretty-templates 565 -fno-rtti -fsized-deallocation 566 -ftemplate-backtrace-limit=N 567 -ftemplate-depth=N 568 -fno-threadsafe-statics -fuse-cxa-atexit 569 -fno-weak -nostdinc++ 570 -fvisibility-inlines-hidden 571 -fvisibility-ms-compat 572 -fext-numeric-literals 573 -flang-info-include-translate[=HEADER] 574 -flang-info-include-translate-not 575 -flang-info-module-cmi[=MODULE] 576 -stdlib=LIBSTDC++,LIBC++ 577 -Wabi-tag -Wcatch-value -Wcatch-value=N 578 -Wno-class-conversion -Wclass-memaccess 579 -Wcomma-subscript -Wconditionally-supported 580 -Wno-conversion-null -Wctad-maybe-unsupported 581 -Wctor-dtor-privacy -Wno-delete-incomplete 582 -Wdelete-non-virtual-dtor -Wno-deprecated-array-compare 583 -Wdeprecated-copy -Wdeprecated-copy-dtor 584 -Wno-deprecated-enum-enum-conversion -Wno-deprecated-enum-float-conversion 585 -Weffc++ -Wno-exceptions -Wextra-semi -Wno-inaccessible-base 586 -Wno-inherited-variadic-ctor -Wno-init-list-lifetime 587 -Winvalid-imported-macros 588 -Wno-invalid-offsetof -Wno-literal-suffix 589 -Wmismatched-new-delete -Wmismatched-tags 590 -Wmultiple-inheritance -Wnamespaces -Wnarrowing 591 -Wnoexcept -Wnoexcept-type -Wnon-virtual-dtor 592 -Wpessimizing-move -Wno-placement-new -Wplacement-new=N 593 -Wrange-loop-construct -Wredundant-move -Wredundant-tags 594 -Wreorder -Wregister 595 -Wstrict-null-sentinel -Wno-subobject-linkage -Wtemplates 596 -Wno-non-template-friend -Wold-style-cast 597 -Woverloaded-virtual -Wno-pmf-conversions -Wsign-promo 598 -Wsized-deallocation -Wsuggest-final-methods 599 -Wsuggest-final-types -Wsuggest-override 600 -Wno-terminate -Wuseless-cast -Wno-vexing-parse 601 -Wvirtual-inheritance 602 -Wno-virtual-move-assign -Wvolatile -Wzero-as-null-pointer-constant 603 604_Objective-C and Objective-C++ Language Options_ 605 *Note Options Controlling Objective-C and Objective-C++ Dialects: 606 Objective-C and Objective-C++ Dialect Options. 607 -fconstant-string-class=CLASS-NAME 608 -fgnu-runtime -fnext-runtime 609 -fno-nil-receivers 610 -fobjc-abi-version=N 611 -fobjc-call-cxx-cdtors 612 -fobjc-direct-dispatch 613 -fobjc-exceptions 614 -fobjc-gc 615 -fobjc-nilcheck 616 -fobjc-std=objc1 617 -fno-local-ivars 618 -fivar-visibility=[public|protected|private|package] 619 -freplace-objc-classes 620 -fzero-link 621 -gen-decls 622 -Wassign-intercept -Wno-property-assign-default 623 -Wno-protocol -Wobjc-root-class -Wselector 624 -Wstrict-selector-match 625 -Wundeclared-selector 626 627_Diagnostic Message Formatting Options_ 628 *Note Options to Control Diagnostic Messages Formatting: Diagnostic 629 Message Formatting Options. 630 -fmessage-length=N 631 -fdiagnostics-plain-output 632 -fdiagnostics-show-location=[once|every-line] 633 -fdiagnostics-color=[auto|never|always] 634 -fdiagnostics-urls=[auto|never|always] 635 -fdiagnostics-format=[text|json] 636 -fno-diagnostics-show-option -fno-diagnostics-show-caret 637 -fno-diagnostics-show-labels -fno-diagnostics-show-line-numbers 638 -fno-diagnostics-show-cwe 639 -fdiagnostics-minimum-margin-width=WIDTH 640 -fdiagnostics-parseable-fixits -fdiagnostics-generate-patch 641 -fdiagnostics-show-template-tree -fno-elide-type 642 -fdiagnostics-path-format=[none|separate-events|inline-events] 643 -fdiagnostics-show-path-depths 644 -fno-show-column 645 -fdiagnostics-column-unit=[display|byte] 646 -fdiagnostics-column-origin=ORIGIN 647 -fdiagnostics-escape-format=[unicode|bytes] 648 649_Warning Options_ 650 *Note Options to Request or Suppress Warnings: Warning Options. 651 -fsyntax-only -fmax-errors=N -Wpedantic 652 -pedantic-errors 653 -w -Wextra -Wall -Wabi=N 654 -Waddress -Wno-address-of-packed-member -Waggregate-return 655 -Walloc-size-larger-than=BYTE-SIZE -Walloc-zero 656 -Walloca -Walloca-larger-than=BYTE-SIZE 657 -Wno-aggressive-loop-optimizations 658 -Warith-conversion 659 -Warray-bounds -Warray-bounds=N -Warray-compare 660 -Wno-attributes -Wattribute-alias=N -Wno-attribute-alias 661 -Wno-attribute-warning 662 -Wbidi-chars=[none|unpaired|any|ucn] 663 -Wbool-compare -Wbool-operation 664 -Wno-builtin-declaration-mismatch 665 -Wno-builtin-macro-redefined -Wc90-c99-compat -Wc99-c11-compat 666 -Wc11-c2x-compat 667 -Wc++-compat -Wc++11-compat -Wc++14-compat -Wc++17-compat 668 -Wc++20-compat 669 -Wno-c++11-extensions -Wno-c++14-extensions -Wno-c++17-extensions 670 -Wno-c++20-extensions -Wno-c++23-extensions 671 -Wcast-align -Wcast-align=strict -Wcast-function-type -Wcast-qual 672 -Wchar-subscripts 673 -Wclobbered -Wcomment 674 -Wconversion -Wno-coverage-mismatch -Wno-cpp 675 -Wdangling-else -Wdangling-pointer -Wdangling-pointer=N 676 -Wdate-time 677 -Wno-deprecated -Wno-deprecated-declarations -Wno-designated-init 678 -Wdisabled-optimization 679 -Wno-discarded-array-qualifiers -Wno-discarded-qualifiers 680 -Wno-div-by-zero -Wdouble-promotion 681 -Wduplicated-branches -Wduplicated-cond 682 -Wempty-body -Wno-endif-labels -Wenum-compare -Wenum-conversion 683 -Werror -Werror=* -Wexpansion-to-defined -Wfatal-errors 684 -Wfloat-conversion -Wfloat-equal -Wformat -Wformat=2 685 -Wno-format-contains-nul -Wno-format-extra-args 686 -Wformat-nonliteral -Wformat-overflow=N 687 -Wformat-security -Wformat-signedness -Wformat-truncation=N 688 -Wformat-y2k -Wframe-address 689 -Wframe-larger-than=BYTE-SIZE -Wno-free-nonheap-object 690 -Wno-if-not-aligned -Wno-ignored-attributes 691 -Wignored-qualifiers -Wno-incompatible-pointer-types 692 -Wimplicit -Wimplicit-fallthrough -Wimplicit-fallthrough=N 693 -Wno-implicit-function-declaration -Wno-implicit-int 694 -Winfinite-recursion 695 -Winit-self -Winline -Wno-int-conversion -Wint-in-bool-context 696 -Wno-int-to-pointer-cast -Wno-invalid-memory-model 697 -Winvalid-pch -Wjump-misses-init -Wlarger-than=BYTE-SIZE 698 -Wlogical-not-parentheses -Wlogical-op -Wlong-long 699 -Wno-lto-type-mismatch -Wmain -Wmaybe-uninitialized 700 -Wmemset-elt-size -Wmemset-transposed-args 701 -Wmisleading-indentation -Wmissing-attributes -Wmissing-braces 702 -Wmissing-field-initializers -Wmissing-format-attribute 703 -Wmissing-include-dirs -Wmissing-noreturn -Wno-missing-profile 704 -Wno-multichar -Wmultistatement-macros -Wnonnull -Wnonnull-compare 705 -Wnormalized=[none|id|nfc|nfkc] 706 -Wnull-dereference -Wno-odr 707 -Wopenacc-parallelism 708 -Wopenmp-simd 709 -Wno-overflow -Woverlength-strings -Wno-override-init-side-effects 710 -Wpacked -Wno-packed-bitfield-compat -Wpacked-not-aligned -Wpadded 711 -Wparentheses -Wno-pedantic-ms-format 712 -Wpointer-arith -Wno-pointer-compare -Wno-pointer-to-int-cast 713 -Wno-pragmas -Wno-prio-ctor-dtor -Wredundant-decls 714 -Wrestrict -Wno-return-local-addr -Wreturn-type 715 -Wno-scalar-storage-order -Wsequence-point 716 -Wshadow -Wshadow=global -Wshadow=local -Wshadow=compatible-local 717 -Wno-shadow-ivar 718 -Wno-shift-count-negative -Wno-shift-count-overflow -Wshift-negative-value 719 -Wno-shift-overflow -Wshift-overflow=N 720 -Wsign-compare -Wsign-conversion 721 -Wno-sizeof-array-argument 722 -Wsizeof-array-div 723 -Wsizeof-pointer-div -Wsizeof-pointer-memaccess 724 -Wstack-protector -Wstack-usage=BYTE-SIZE -Wstrict-aliasing 725 -Wstrict-aliasing=n -Wstrict-overflow -Wstrict-overflow=N 726 -Wstring-compare 727 -Wno-stringop-overflow -Wno-stringop-overread 728 -Wno-stringop-truncation 729 -Wsuggest-attribute=[pure|const|noreturn|format|malloc] 730 -Wswitch -Wno-switch-bool -Wswitch-default -Wswitch-enum 731 -Wno-switch-outside-range -Wno-switch-unreachable -Wsync-nand 732 -Wsystem-headers -Wtautological-compare -Wtrampolines -Wtrigraphs 733 -Wtrivial-auto-var-init -Wtsan -Wtype-limits -Wundef 734 -Wuninitialized -Wunknown-pragmas 735 -Wunsuffixed-float-constants -Wunused 736 -Wunused-but-set-parameter -Wunused-but-set-variable 737 -Wunused-const-variable -Wunused-const-variable=N 738 -Wunused-function -Wunused-label -Wunused-local-typedefs 739 -Wunused-macros 740 -Wunused-parameter -Wno-unused-result 741 -Wunused-value -Wunused-variable 742 -Wno-varargs -Wvariadic-macros 743 -Wvector-operation-performance 744 -Wvla -Wvla-larger-than=BYTE-SIZE -Wno-vla-larger-than 745 -Wvolatile-register-var -Wwrite-strings 746 -Wzero-length-bounds 747 748_Static Analyzer Options_ 749 -fanalyzer 750 -fanalyzer-call-summaries 751 -fanalyzer-checker=NAME 752 -fno-analyzer-feasibility 753 -fanalyzer-fine-grained 754 -fno-analyzer-state-merge 755 -fno-analyzer-state-purge 756 -fanalyzer-transitivity 757 -fanalyzer-verbose-edges 758 -fanalyzer-verbose-state-changes 759 -fanalyzer-verbosity=LEVEL 760 -fdump-analyzer 761 -fdump-analyzer-callgraph 762 -fdump-analyzer-exploded-graph 763 -fdump-analyzer-exploded-nodes 764 -fdump-analyzer-exploded-nodes-2 765 -fdump-analyzer-exploded-nodes-3 766 -fdump-analyzer-exploded-paths 767 -fdump-analyzer-feasibility 768 -fdump-analyzer-json 769 -fdump-analyzer-state-purge 770 -fdump-analyzer-stderr 771 -fdump-analyzer-supergraph 772 -fdump-analyzer-untracked 773 -Wno-analyzer-double-fclose 774 -Wno-analyzer-double-free 775 -Wno-analyzer-exposure-through-output-file 776 -Wno-analyzer-file-leak 777 -Wno-analyzer-free-of-non-heap 778 -Wno-analyzer-malloc-leak 779 -Wno-analyzer-mismatching-deallocation 780 -Wno-analyzer-null-argument 781 -Wno-analyzer-null-dereference 782 -Wno-analyzer-possible-null-argument 783 -Wno-analyzer-possible-null-dereference 784 -Wno-analyzer-shift-count-negative 785 -Wno-analyzer-shift-count-overflow 786 -Wno-analyzer-stale-setjmp-buffer 787 -Wno-analyzer-tainted-allocation-size 788 -Wno-analyzer-tainted-array-index 789 -Wno-analyzer-tainted-divisor 790 -Wno-analyzer-tainted-offset 791 -Wno-analyzer-tainted-size 792 -Wanalyzer-too-complex 793 -Wno-analyzer-unsafe-call-within-signal-handler 794 -Wno-analyzer-use-after-free 795 -Wno-analyzer-use-of-pointer-in-stale-stack-frame 796 -Wno-analyzer-use-of-uninitialized-value 797 -Wno-analyzer-write-to-const 798 -Wno-analyzer-write-to-string-literal 799 800 801_C and Objective-C-only Warning Options_ 802 -Wbad-function-cast -Wmissing-declarations 803 -Wmissing-parameter-type -Wmissing-prototypes -Wnested-externs 804 -Wold-style-declaration -Wold-style-definition 805 -Wstrict-prototypes -Wtraditional -Wtraditional-conversion 806 -Wdeclaration-after-statement -Wpointer-sign 807 808_Debugging Options_ 809 *Note Options for Debugging Your Program: Debugging Options. 810 -g -gLEVEL -gdwarf -gdwarf-VERSION 811 -gbtf -gctf -gctfLEVEL 812 -ggdb -grecord-gcc-switches -gno-record-gcc-switches 813 -gstabs -gstabs+ -gstrict-dwarf -gno-strict-dwarf 814 -gas-loc-support -gno-as-loc-support 815 -gas-locview-support -gno-as-locview-support 816 -gcolumn-info -gno-column-info -gdwarf32 -gdwarf64 817 -gstatement-frontiers -gno-statement-frontiers 818 -gvariable-location-views -gno-variable-location-views 819 -ginternal-reset-location-views -gno-internal-reset-location-views 820 -ginline-points -gno-inline-points 821 -gvms -gxcoff -gxcoff+ -gz[=TYPE] 822 -gsplit-dwarf -gdescribe-dies -gno-describe-dies 823 -fdebug-prefix-map=OLD=NEW -fdebug-types-section 824 -fno-eliminate-unused-debug-types 825 -femit-struct-debug-baseonly -femit-struct-debug-reduced 826 -femit-struct-debug-detailed[=SPEC-LIST] 827 -fno-eliminate-unused-debug-symbols -femit-class-debug-always 828 -fno-merge-debug-strings -fno-dwarf2-cfi-asm 829 -fvar-tracking -fvar-tracking-assignments 830 831_Optimization Options_ 832 *Note Options that Control Optimization: Optimize Options. 833 -faggressive-loop-optimizations 834 -falign-functions[=N[:M:[N2[:M2]]]] 835 -falign-jumps[=N[:M:[N2[:M2]]]] 836 -falign-labels[=N[:M:[N2[:M2]]]] 837 -falign-loops[=N[:M:[N2[:M2]]]] 838 -fno-allocation-dce -fallow-store-data-races 839 -fassociative-math -fauto-profile -fauto-profile[=PATH] 840 -fauto-inc-dec -fbranch-probabilities 841 -fcaller-saves 842 -fcombine-stack-adjustments -fconserve-stack 843 -fcompare-elim -fcprop-registers -fcrossjumping 844 -fcse-follow-jumps -fcse-skip-blocks -fcx-fortran-rules 845 -fcx-limited-range 846 -fdata-sections -fdce -fdelayed-branch 847 -fdelete-null-pointer-checks -fdevirtualize -fdevirtualize-speculatively 848 -fdevirtualize-at-ltrans -fdse 849 -fearly-inlining -fipa-sra -fexpensive-optimizations -ffat-lto-objects 850 -ffast-math -ffinite-math-only -ffloat-store -fexcess-precision=STYLE 851 -ffinite-loops 852 -fforward-propagate -ffp-contract=STYLE -ffunction-sections 853 -fgcse -fgcse-after-reload -fgcse-las -fgcse-lm -fgraphite-identity 854 -fgcse-sm -fhoist-adjacent-loads -fif-conversion 855 -fif-conversion2 -findirect-inlining 856 -finline-functions -finline-functions-called-once -finline-limit=N 857 -finline-small-functions -fipa-modref -fipa-cp -fipa-cp-clone 858 -fipa-bit-cp -fipa-vrp -fipa-pta -fipa-profile -fipa-pure-const 859 -fipa-reference -fipa-reference-addressable 860 -fipa-stack-alignment -fipa-icf -fira-algorithm=ALGORITHM 861 -flive-patching=LEVEL 862 -fira-region=REGION -fira-hoist-pressure 863 -fira-loop-pressure -fno-ira-share-save-slots 864 -fno-ira-share-spill-slots 865 -fisolate-erroneous-paths-dereference -fisolate-erroneous-paths-attribute 866 -fivopts -fkeep-inline-functions -fkeep-static-functions 867 -fkeep-static-consts -flimit-function-alignment -flive-range-shrinkage 868 -floop-block -floop-interchange -floop-strip-mine 869 -floop-unroll-and-jam -floop-nest-optimize 870 -floop-parallelize-all -flra-remat -flto -flto-compression-level 871 -flto-partition=ALG -fmerge-all-constants 872 -fmerge-constants -fmodulo-sched -fmodulo-sched-allow-regmoves 873 -fmove-loop-invariants -fmove-loop-stores -fno-branch-count-reg 874 -fno-defer-pop -fno-fp-int-builtin-inexact -fno-function-cse 875 -fno-guess-branch-probability -fno-inline -fno-math-errno -fno-peephole 876 -fno-peephole2 -fno-printf-return-value -fno-sched-interblock 877 -fno-sched-spec -fno-signed-zeros 878 -fno-toplevel-reorder -fno-trapping-math -fno-zero-initialized-in-bss 879 -fomit-frame-pointer -foptimize-sibling-calls 880 -fpartial-inlining -fpeel-loops -fpredictive-commoning 881 -fprefetch-loop-arrays 882 -fprofile-correction 883 -fprofile-use -fprofile-use=PATH -fprofile-partial-training 884 -fprofile-values -fprofile-reorder-functions 885 -freciprocal-math -free -frename-registers -freorder-blocks 886 -freorder-blocks-algorithm=ALGORITHM 887 -freorder-blocks-and-partition -freorder-functions 888 -frerun-cse-after-loop -freschedule-modulo-scheduled-loops 889 -frounding-math -fsave-optimization-record 890 -fsched2-use-superblocks -fsched-pressure 891 -fsched-spec-load -fsched-spec-load-dangerous 892 -fsched-stalled-insns-dep[=N] -fsched-stalled-insns[=N] 893 -fsched-group-heuristic -fsched-critical-path-heuristic 894 -fsched-spec-insn-heuristic -fsched-rank-heuristic 895 -fsched-last-insn-heuristic -fsched-dep-count-heuristic 896 -fschedule-fusion 897 -fschedule-insns -fschedule-insns2 -fsection-anchors 898 -fselective-scheduling -fselective-scheduling2 899 -fsel-sched-pipelining -fsel-sched-pipelining-outer-loops 900 -fsemantic-interposition -fshrink-wrap -fshrink-wrap-separate 901 -fsignaling-nans 902 -fsingle-precision-constant -fsplit-ivs-in-unroller -fsplit-loops 903 -fsplit-paths 904 -fsplit-wide-types -fsplit-wide-types-early -fssa-backprop -fssa-phiopt 905 -fstdarg-opt -fstore-merging -fstrict-aliasing -fipa-strict-aliasing 906 -fthread-jumps -ftracer -ftree-bit-ccp 907 -ftree-builtin-call-dce -ftree-ccp -ftree-ch 908 -ftree-coalesce-vars -ftree-copy-prop -ftree-dce -ftree-dominator-opts 909 -ftree-dse -ftree-forwprop -ftree-fre -fcode-hoisting 910 -ftree-loop-if-convert -ftree-loop-im 911 -ftree-phiprop -ftree-loop-distribution -ftree-loop-distribute-patterns 912 -ftree-loop-ivcanon -ftree-loop-linear -ftree-loop-optimize 913 -ftree-loop-vectorize 914 -ftree-parallelize-loops=N -ftree-pre -ftree-partial-pre -ftree-pta 915 -ftree-reassoc -ftree-scev-cprop -ftree-sink -ftree-slsr -ftree-sra 916 -ftree-switch-conversion -ftree-tail-merge 917 -ftree-ter -ftree-vectorize -ftree-vrp -ftrivial-auto-var-init 918 -funconstrained-commons -funit-at-a-time -funroll-all-loops 919 -funroll-loops -funsafe-math-optimizations -funswitch-loops 920 -fipa-ra -fvariable-expansion-in-unroller -fvect-cost-model -fvpt 921 -fweb -fwhole-program -fwpa -fuse-linker-plugin -fzero-call-used-regs 922 --param NAME=VALUE 923 -O -O0 -O1 -O2 -O3 -Os -Ofast -Og -Oz 924 925_Program Instrumentation Options_ 926 *Note Program Instrumentation Options: Instrumentation Options. 927 -p -pg -fprofile-arcs --coverage -ftest-coverage 928 -fprofile-abs-path 929 -fprofile-dir=PATH -fprofile-generate -fprofile-generate=PATH 930 -fprofile-info-section -fprofile-info-section=NAME 931 -fprofile-note=PATH -fprofile-prefix-path=PATH 932 -fprofile-update=METHOD -fprofile-filter-files=REGEX 933 -fprofile-exclude-files=REGEX 934 -fprofile-reproducible=[multithreaded|parallel-runs|serial] 935 -fsanitize=STYLE -fsanitize-recover -fsanitize-recover=STYLE 936 -fasan-shadow-offset=NUMBER -fsanitize-sections=S1,S2,... 937 -fsanitize-undefined-trap-on-error -fbounds-check 938 -fcf-protection=[full|branch|return|none|check] 939 -fharden-compares -fharden-conditional-branches 940 -fstack-protector -fstack-protector-all -fstack-protector-strong 941 -fstack-protector-explicit -fstack-check 942 -fstack-limit-register=REG -fstack-limit-symbol=SYM 943 -fno-stack-limit -fsplit-stack 944 -fvtable-verify=[std|preinit|none] 945 -fvtv-counts -fvtv-debug 946 -finstrument-functions 947 -finstrument-functions-exclude-function-list=SYM,SYM,... 948 -finstrument-functions-exclude-file-list=FILE,FILE,... 949 -fprofile-prefix-map=OLD=NEW 950 951_Preprocessor Options_ 952 *Note Options Controlling the Preprocessor: Preprocessor Options. 953 -AQUESTION=ANSWER 954 -A-QUESTION[=ANSWER] 955 -C -CC -DMACRO[=DEFN] 956 -dD -dI -dM -dN -dU 957 -fdebug-cpp -fdirectives-only -fdollars-in-identifiers 958 -fexec-charset=CHARSET -fextended-identifiers 959 -finput-charset=CHARSET -flarge-source-files 960 -fmacro-prefix-map=OLD=NEW -fmax-include-depth=DEPTH 961 -fno-canonical-system-headers -fpch-deps -fpch-preprocess 962 -fpreprocessed -ftabstop=WIDTH -ftrack-macro-expansion 963 -fwide-exec-charset=CHARSET -fworking-directory 964 -H -imacros FILE -include FILE 965 -M -MD -MF -MG -MM -MMD -MP -MQ -MT -Mno-modules 966 -no-integrated-cpp -P -pthread -remap 967 -traditional -traditional-cpp -trigraphs 968 -UMACRO -undef 969 -Wp,OPTION -Xpreprocessor OPTION 970 971_Assembler Options_ 972 *Note Passing Options to the Assembler: Assembler Options. 973 -Wa,OPTION -Xassembler OPTION 974 975_Linker Options_ 976 *Note Options for Linking: Link Options. 977 OBJECT-FILE-NAME -fuse-ld=LINKER -lLIBRARY 978 -nostartfiles -nodefaultlibs -nolibc -nostdlib 979 -e ENTRY --entry=ENTRY 980 -pie -pthread -r -rdynamic 981 -s -static -static-pie -static-libgcc -static-libstdc++ 982 -static-libasan -static-libtsan -static-liblsan -static-libubsan 983 -shared -shared-libgcc -symbolic 984 -T SCRIPT -Wl,OPTION -Xlinker OPTION 985 -u SYMBOL -z KEYWORD 986 987_Directory Options_ 988 *Note Options for Directory Search: Directory Options. 989 -BPREFIX -IDIR -I- 990 -idirafter DIR 991 -imacros FILE -imultilib DIR 992 -iplugindir=DIR -iprefix FILE 993 -iquote DIR -isysroot DIR -isystem DIR 994 -iwithprefix DIR -iwithprefixbefore DIR 995 -LDIR -no-canonical-prefixes --no-sysroot-suffix 996 -nostdinc -nostdinc++ --sysroot=DIR 997 998_Code Generation Options_ 999 *Note Options for Code Generation Conventions: Code Gen Options. 1000 -fcall-saved-REG -fcall-used-REG 1001 -ffixed-REG -fexceptions 1002 -fnon-call-exceptions -fdelete-dead-exceptions -funwind-tables 1003 -fasynchronous-unwind-tables 1004 -fno-gnu-unique 1005 -finhibit-size-directive -fcommon -fno-ident 1006 -fpcc-struct-return -fpic -fPIC -fpie -fPIE -fno-plt 1007 -fno-jump-tables -fno-bit-tests 1008 -frecord-gcc-switches 1009 -freg-struct-return -fshort-enums -fshort-wchar 1010 -fverbose-asm -fpack-struct[=N] 1011 -fleading-underscore -ftls-model=MODEL 1012 -fstack-reuse=REUSE_LEVEL 1013 -ftrampolines -ftrapv -fwrapv 1014 -fvisibility=[default|internal|hidden|protected] 1015 -fstrict-volatile-bitfields -fsync-libcalls 1016 1017_Developer Options_ 1018 *Note GCC Developer Options: Developer Options. 1019 -dLETTERS -dumpspecs -dumpmachine -dumpversion 1020 -dumpfullversion -fcallgraph-info[=su,da] 1021 -fchecking -fchecking=N 1022 -fdbg-cnt-list -fdbg-cnt=COUNTER-VALUE-LIST 1023 -fdisable-ipa-PASS_NAME 1024 -fdisable-rtl-PASS_NAME 1025 -fdisable-rtl-PASS-NAME=RANGE-LIST 1026 -fdisable-tree-PASS_NAME 1027 -fdisable-tree-PASS-NAME=RANGE-LIST 1028 -fdump-debug -fdump-earlydebug 1029 -fdump-noaddr -fdump-unnumbered -fdump-unnumbered-links 1030 -fdump-final-insns[=FILE] 1031 -fdump-ipa-all -fdump-ipa-cgraph -fdump-ipa-inline 1032 -fdump-lang-all 1033 -fdump-lang-SWITCH 1034 -fdump-lang-SWITCH-OPTIONS 1035 -fdump-lang-SWITCH-OPTIONS=FILENAME 1036 -fdump-passes 1037 -fdump-rtl-PASS -fdump-rtl-PASS=FILENAME 1038 -fdump-statistics 1039 -fdump-tree-all 1040 -fdump-tree-SWITCH 1041 -fdump-tree-SWITCH-OPTIONS 1042 -fdump-tree-SWITCH-OPTIONS=FILENAME 1043 -fcompare-debug[=OPTS] -fcompare-debug-second 1044 -fenable-KIND-PASS 1045 -fenable-KIND-PASS=RANGE-LIST 1046 -fira-verbose=N 1047 -flto-report -flto-report-wpa -fmem-report-wpa 1048 -fmem-report -fpre-ipa-mem-report -fpost-ipa-mem-report 1049 -fopt-info -fopt-info-OPTIONS[=FILE] 1050 -fprofile-report 1051 -frandom-seed=STRING -fsched-verbose=N 1052 -fsel-sched-verbose -fsel-sched-dump-cfg -fsel-sched-pipelining-verbose 1053 -fstats -fstack-usage -ftime-report -ftime-report-details 1054 -fvar-tracking-assignments-toggle -gtoggle 1055 -print-file-name=LIBRARY -print-libgcc-file-name 1056 -print-multi-directory -print-multi-lib -print-multi-os-directory 1057 -print-prog-name=PROGRAM -print-search-dirs -Q 1058 -print-sysroot -print-sysroot-headers-suffix 1059 -save-temps -save-temps=cwd -save-temps=obj -time[=FILE] 1060 1061_Machine-Dependent Options_ 1062 *Note Machine-Dependent Options: Submodel Options. 1063 1064 _AArch64 Options_ 1065 -mabi=NAME -mbig-endian -mlittle-endian 1066 -mgeneral-regs-only 1067 -mcmodel=tiny -mcmodel=small -mcmodel=large 1068 -mstrict-align -mno-strict-align 1069 -momit-leaf-frame-pointer 1070 -mtls-dialect=desc -mtls-dialect=traditional 1071 -mtls-size=SIZE 1072 -mfix-cortex-a53-835769 -mfix-cortex-a53-843419 1073 -mlow-precision-recip-sqrt -mlow-precision-sqrt -mlow-precision-div 1074 -mpc-relative-literal-loads 1075 -msign-return-address=SCOPE 1076 -mbranch-protection=NONE|STANDARD|PAC-RET[+LEAF 1077 +B-KEY]|BTI 1078 -mharden-sls=OPTS 1079 -march=NAME -mcpu=NAME -mtune=NAME 1080 -moverride=STRING -mverbose-cost-dump 1081 -mstack-protector-guard=GUARD -mstack-protector-guard-reg=SYSREG 1082 -mstack-protector-guard-offset=OFFSET -mtrack-speculation 1083 -moutline-atomics 1084 1085 _Adapteva Epiphany Options_ 1086 -mhalf-reg-file -mprefer-short-insn-regs 1087 -mbranch-cost=NUM -mcmove -mnops=NUM -msoft-cmpsf 1088 -msplit-lohi -mpost-inc -mpost-modify -mstack-offset=NUM 1089 -mround-nearest -mlong-calls -mshort-calls -msmall16 1090 -mfp-mode=MODE -mvect-double -max-vect-align=NUM 1091 -msplit-vecmove-early -m1reg-REG 1092 1093 _AMD GCN Options_ 1094 -march=GPU -mtune=GPU -mstack-size=BYTES 1095 1096 _ARC Options_ 1097 -mbarrel-shifter -mjli-always 1098 -mcpu=CPU -mA6 -mARC600 -mA7 -mARC700 1099 -mdpfp -mdpfp-compact -mdpfp-fast -mno-dpfp-lrsr 1100 -mea -mno-mpy -mmul32x16 -mmul64 -matomic 1101 -mnorm -mspfp -mspfp-compact -mspfp-fast -msimd -msoft-float -mswap 1102 -mcrc -mdsp-packa -mdvbf -mlock -mmac-d16 -mmac-24 -mrtsc -mswape 1103 -mtelephony -mxy -misize -mannotate-align -marclinux -marclinux_prof 1104 -mlong-calls -mmedium-calls -msdata -mirq-ctrl-saved 1105 -mrgf-banked-regs -mlpc-width=WIDTH -G NUM 1106 -mvolatile-cache -mtp-regno=REGNO 1107 -malign-call -mauto-modify-reg -mbbit-peephole -mno-brcc 1108 -mcase-vector-pcrel -mcompact-casesi -mno-cond-exec -mearly-cbranchsi 1109 -mexpand-adddi -mindexed-loads -mlra -mlra-priority-none 1110 -mlra-priority-compact -mlra-priority-noncompact -mmillicode 1111 -mmixed-code -mq-class -mRcq -mRcw -msize-level=LEVEL 1112 -mtune=CPU -mmultcost=NUM -mcode-density-frame 1113 -munalign-prob-threshold=PROBABILITY -mmpy-option=MULTO 1114 -mdiv-rem -mcode-density -mll64 -mfpu=FPU -mrf16 -mbranch-index 1115 1116 _ARM Options_ 1117 -mapcs-frame -mno-apcs-frame 1118 -mabi=NAME 1119 -mapcs-stack-check -mno-apcs-stack-check 1120 -mapcs-reentrant -mno-apcs-reentrant 1121 -mgeneral-regs-only 1122 -msched-prolog -mno-sched-prolog 1123 -mlittle-endian -mbig-endian 1124 -mbe8 -mbe32 1125 -mfloat-abi=NAME 1126 -mfp16-format=NAME 1127 -mthumb-interwork -mno-thumb-interwork 1128 -mcpu=NAME -march=NAME -mfpu=NAME 1129 -mtune=NAME -mprint-tune-info 1130 -mstructure-size-boundary=N 1131 -mabort-on-noreturn 1132 -mlong-calls -mno-long-calls 1133 -msingle-pic-base -mno-single-pic-base 1134 -mpic-register=REG 1135 -mnop-fun-dllimport 1136 -mpoke-function-name 1137 -mthumb -marm -mflip-thumb 1138 -mtpcs-frame -mtpcs-leaf-frame 1139 -mcaller-super-interworking -mcallee-super-interworking 1140 -mtp=NAME -mtls-dialect=DIALECT 1141 -mword-relocations 1142 -mfix-cortex-m3-ldrd 1143 -mfix-cortex-a57-aes-1742098 1144 -mfix-cortex-a72-aes-1655431 1145 -munaligned-access 1146 -mneon-for-64bits 1147 -mslow-flash-data 1148 -masm-syntax-unified 1149 -mrestrict-it 1150 -mverbose-cost-dump 1151 -mpure-code 1152 -mcmse 1153 -mfix-cmse-cve-2021-35465 1154 -mstack-protector-guard=GUARD -mstack-protector-guard-offset=OFFSET 1155 -mfdpic 1156 1157 _AVR Options_ 1158 -mmcu=MCU -mabsdata -maccumulate-args 1159 -mbranch-cost=COST 1160 -mcall-prologues -mgas-isr-prologues -mint8 1161 -mdouble=BITS -mlong-double=BITS 1162 -mn_flash=SIZE -mno-interrupts 1163 -mmain-is-OS_task -mrelax -mrmw -mstrict-X -mtiny-stack 1164 -mfract-convert-truncate 1165 -mshort-calls -nodevicelib -nodevicespecs 1166 -Waddr-space-convert -Wmisspelled-isr 1167 1168 _Blackfin Options_ 1169 -mcpu=CPU[-SIREVISION] 1170 -msim -momit-leaf-frame-pointer -mno-omit-leaf-frame-pointer 1171 -mspecld-anomaly -mno-specld-anomaly -mcsync-anomaly -mno-csync-anomaly 1172 -mlow-64k -mno-low64k -mstack-check-l1 -mid-shared-library 1173 -mno-id-shared-library -mshared-library-id=N 1174 -mleaf-id-shared-library -mno-leaf-id-shared-library 1175 -msep-data -mno-sep-data -mlong-calls -mno-long-calls 1176 -mfast-fp -minline-plt -mmulticore -mcorea -mcoreb -msdram 1177 -micplb 1178 1179 _C6X Options_ 1180 -mbig-endian -mlittle-endian -march=CPU 1181 -msim -msdata=SDATA-TYPE 1182 1183 _CRIS Options_ 1184 -mcpu=CPU -march=CPU 1185 -mtune=CPU -mmax-stack-frame=N 1186 -metrax4 -metrax100 -mpdebug -mcc-init -mno-side-effects 1187 -mstack-align -mdata-align -mconst-align 1188 -m32-bit -m16-bit -m8-bit -mno-prologue-epilogue 1189 -melf -maout -sim -sim2 1190 -mmul-bug-workaround -mno-mul-bug-workaround 1191 1192 _CR16 Options_ 1193 -mmac 1194 -mcr16cplus -mcr16c 1195 -msim -mint32 -mbit-ops 1196 -mdata-model=MODEL 1197 1198 _C-SKY Options_ 1199 -march=ARCH -mcpu=CPU 1200 -mbig-endian -EB -mlittle-endian -EL 1201 -mhard-float -msoft-float -mfpu=FPU -mdouble-float -mfdivdu 1202 -mfloat-abi=NAME 1203 -melrw -mistack -mmp -mcp -mcache -msecurity -mtrust 1204 -mdsp -medsp -mvdsp 1205 -mdiv -msmart -mhigh-registers -manchor 1206 -mpushpop -mmultiple-stld -mconstpool -mstack-size -mccrt 1207 -mbranch-cost=N -mcse-cc -msched-prolog -msim 1208 1209 _Darwin Options_ 1210 -all_load -allowable_client -arch -arch_errors_fatal 1211 -arch_only -bind_at_load -bundle -bundle_loader 1212 -client_name -compatibility_version -current_version 1213 -dead_strip 1214 -dependency-file -dylib_file -dylinker_install_name 1215 -dynamic -dynamiclib -exported_symbols_list 1216 -filelist -flat_namespace -force_cpusubtype_ALL 1217 -force_flat_namespace -headerpad_max_install_names 1218 -iframework 1219 -image_base -init -install_name -keep_private_externs 1220 -multi_module -multiply_defined -multiply_defined_unused 1221 -noall_load -no_dead_strip_inits_and_terms 1222 -nofixprebinding -nomultidefs -noprebind -noseglinkedit 1223 -pagezero_size -prebind -prebind_all_twolevel_modules 1224 -private_bundle -read_only_relocs -sectalign 1225 -sectobjectsymbols -whyload -seg1addr 1226 -sectcreate -sectobjectsymbols -sectorder 1227 -segaddr -segs_read_only_addr -segs_read_write_addr 1228 -seg_addr_table -seg_addr_table_filename -seglinkedit 1229 -segprot -segs_read_only_addr -segs_read_write_addr 1230 -single_module -static -sub_library -sub_umbrella 1231 -twolevel_namespace -umbrella -undefined 1232 -unexported_symbols_list -weak_reference_mismatches 1233 -whatsloaded -F -gused -gfull -mmacosx-version-min=VERSION 1234 -mkernel -mone-byte-bool 1235 1236 _DEC Alpha Options_ 1237 -mno-fp-regs -msoft-float 1238 -mieee -mieee-with-inexact -mieee-conformant 1239 -mfp-trap-mode=MODE -mfp-rounding-mode=MODE 1240 -mtrap-precision=MODE -mbuild-constants 1241 -mcpu=CPU-TYPE -mtune=CPU-TYPE 1242 -mbwx -mmax -mfix -mcix 1243 -mfloat-vax -mfloat-ieee 1244 -mexplicit-relocs -msmall-data -mlarge-data 1245 -msmall-text -mlarge-text 1246 -mmemory-latency=TIME 1247 1248 _eBPF Options_ 1249 -mbig-endian -mlittle-endian -mkernel=VERSION 1250 -mframe-limit=BYTES -mxbpf -mco-re -mno-co-re 1251 -mjmpext -mjmp32 -malu32 -mcpu=VERSION 1252 1253 _FR30 Options_ 1254 -msmall-model -mno-lsim 1255 1256 _FT32 Options_ 1257 -msim -mlra -mnodiv -mft32b -mcompress -mnopm 1258 1259 _FRV Options_ 1260 -mgpr-32 -mgpr-64 -mfpr-32 -mfpr-64 1261 -mhard-float -msoft-float 1262 -malloc-cc -mfixed-cc -mdword -mno-dword 1263 -mdouble -mno-double 1264 -mmedia -mno-media -mmuladd -mno-muladd 1265 -mfdpic -minline-plt -mgprel-ro -multilib-library-pic 1266 -mlinked-fp -mlong-calls -malign-labels 1267 -mlibrary-pic -macc-4 -macc-8 1268 -mpack -mno-pack -mno-eflags -mcond-move -mno-cond-move 1269 -moptimize-membar -mno-optimize-membar 1270 -mscc -mno-scc -mcond-exec -mno-cond-exec 1271 -mvliw-branch -mno-vliw-branch 1272 -mmulti-cond-exec -mno-multi-cond-exec -mnested-cond-exec 1273 -mno-nested-cond-exec -mtomcat-stats 1274 -mTLS -mtls 1275 -mcpu=CPU 1276 1277 _GNU/Linux Options_ 1278 -mglibc -muclibc -mmusl -mbionic -mandroid 1279 -tno-android-cc -tno-android-ld 1280 1281 _H8/300 Options_ 1282 -mrelax -mh -ms -mn -mexr -mno-exr -mint32 -malign-300 1283 1284 _HPPA Options_ 1285 -march=ARCHITECTURE-TYPE 1286 -mcaller-copies -mdisable-fpregs -mdisable-indexing 1287 -mfast-indirect-calls -mgas -mgnu-ld -mhp-ld 1288 -mfixed-range=REGISTER-RANGE 1289 -mjump-in-delay -mlinker-opt -mlong-calls 1290 -mlong-load-store -mno-disable-fpregs 1291 -mno-disable-indexing -mno-fast-indirect-calls -mno-gas 1292 -mno-jump-in-delay -mno-long-load-store 1293 -mno-portable-runtime -mno-soft-float 1294 -mno-space-regs -msoft-float -mpa-risc-1-0 1295 -mpa-risc-1-1 -mpa-risc-2-0 -mportable-runtime 1296 -mschedule=CPU-TYPE -mspace-regs -msio -mwsio 1297 -munix=UNIX-STD -nolibdld -static -threads 1298 1299 _IA-64 Options_ 1300 -mbig-endian -mlittle-endian -mgnu-as -mgnu-ld -mno-pic 1301 -mvolatile-asm-stop -mregister-names -msdata -mno-sdata 1302 -mconstant-gp -mauto-pic -mfused-madd 1303 -minline-float-divide-min-latency 1304 -minline-float-divide-max-throughput 1305 -mno-inline-float-divide 1306 -minline-int-divide-min-latency 1307 -minline-int-divide-max-throughput 1308 -mno-inline-int-divide 1309 -minline-sqrt-min-latency -minline-sqrt-max-throughput 1310 -mno-inline-sqrt 1311 -mdwarf2-asm -mearly-stop-bits 1312 -mfixed-range=REGISTER-RANGE -mtls-size=TLS-SIZE 1313 -mtune=CPU-TYPE -milp32 -mlp64 1314 -msched-br-data-spec -msched-ar-data-spec -msched-control-spec 1315 -msched-br-in-data-spec -msched-ar-in-data-spec -msched-in-control-spec 1316 -msched-spec-ldc -msched-spec-control-ldc 1317 -msched-prefer-non-data-spec-insns -msched-prefer-non-control-spec-insns 1318 -msched-stop-bits-after-every-cycle -msched-count-spec-in-critical-path 1319 -msel-sched-dont-check-control-spec -msched-fp-mem-deps-zero-cost 1320 -msched-max-memory-insns-hard-limit -msched-max-memory-insns=MAX-INSNS 1321 1322 _LM32 Options_ 1323 -mbarrel-shift-enabled -mdivide-enabled -mmultiply-enabled 1324 -msign-extend-enabled -muser-enabled 1325 1326 _LoongArch Options_ 1327 -march=CPU-TYPE -mtune=CPU-TYPE -mabi=BASE-ABI-TYPE 1328 -mfpu=FPU-TYPE -msoft-float -msingle-float -mdouble-float 1329 -mbranch-cost=N -mcheck-zero-division -mno-check-zero-division 1330 -mcond-move-int -mno-cond-move-int 1331 -mcond-move-float -mno-cond-move-float 1332 -memcpy -mno-memcpy -mstrict-align -mno-strict-align 1333 -mmax-inline-memcpy-size=N 1334 -mcmodel=CODE-MODEL 1335 1336 _M32R/D Options_ 1337 -m32r2 -m32rx -m32r 1338 -mdebug 1339 -malign-loops -mno-align-loops 1340 -missue-rate=NUMBER 1341 -mbranch-cost=NUMBER 1342 -mmodel=CODE-SIZE-MODEL-TYPE 1343 -msdata=SDATA-TYPE 1344 -mno-flush-func -mflush-func=NAME 1345 -mno-flush-trap -mflush-trap=NUMBER 1346 -G NUM 1347 1348 _M32C Options_ 1349 -mcpu=CPU -msim -memregs=NUMBER 1350 1351 _M680x0 Options_ 1352 -march=ARCH -mcpu=CPU -mtune=TUNE 1353 -m68000 -m68020 -m68020-40 -m68020-60 -m68030 -m68040 1354 -m68060 -mcpu32 -m5200 -m5206e -m528x -m5307 -m5407 1355 -mcfv4e -mbitfield -mno-bitfield -mc68000 -mc68020 1356 -mnobitfield -mrtd -mno-rtd -mdiv -mno-div -mshort 1357 -mno-short -mhard-float -m68881 -msoft-float -mpcrel 1358 -malign-int -mstrict-align -msep-data -mno-sep-data 1359 -mshared-library-id=n -mid-shared-library -mno-id-shared-library 1360 -mxgot -mno-xgot -mlong-jump-table-offsets 1361 1362 _MCore Options_ 1363 -mhardlit -mno-hardlit -mdiv -mno-div -mrelax-immediates 1364 -mno-relax-immediates -mwide-bitfields -mno-wide-bitfields 1365 -m4byte-functions -mno-4byte-functions -mcallgraph-data 1366 -mno-callgraph-data -mslow-bytes -mno-slow-bytes -mno-lsim 1367 -mlittle-endian -mbig-endian -m210 -m340 -mstack-increment 1368 1369 _MeP Options_ 1370 -mabsdiff -mall-opts -maverage -mbased=N -mbitops 1371 -mc=N -mclip -mconfig=NAME -mcop -mcop32 -mcop64 -mivc2 1372 -mdc -mdiv -meb -mel -mio-volatile -ml -mleadz -mm -mminmax 1373 -mmult -mno-opts -mrepeat -ms -msatur -msdram -msim -msimnovec -mtf 1374 -mtiny=N 1375 1376 _MicroBlaze Options_ 1377 -msoft-float -mhard-float -msmall-divides -mcpu=CPU 1378 -mmemcpy -mxl-soft-mul -mxl-soft-div -mxl-barrel-shift 1379 -mxl-pattern-compare -mxl-stack-check -mxl-gp-opt -mno-clearbss 1380 -mxl-multiply-high -mxl-float-convert -mxl-float-sqrt 1381 -mbig-endian -mlittle-endian -mxl-reorder -mxl-mode-APP-MODEL 1382 -mpic-data-is-text-relative 1383 1384 _MIPS Options_ 1385 -EL -EB -march=ARCH -mtune=ARCH 1386 -mips1 -mips2 -mips3 -mips4 -mips32 -mips32r2 -mips32r3 -mips32r5 1387 -mips32r6 -mips64 -mips64r2 -mips64r3 -mips64r5 -mips64r6 1388 -mips16 -mno-mips16 -mflip-mips16 1389 -minterlink-compressed -mno-interlink-compressed 1390 -minterlink-mips16 -mno-interlink-mips16 1391 -mabi=ABI -mabicalls -mno-abicalls 1392 -mshared -mno-shared -mplt -mno-plt -mxgot -mno-xgot 1393 -mgp32 -mgp64 -mfp32 -mfpxx -mfp64 -mhard-float -msoft-float 1394 -mno-float -msingle-float -mdouble-float 1395 -modd-spreg -mno-odd-spreg 1396 -mabs=MODE -mnan=ENCODING 1397 -mdsp -mno-dsp -mdspr2 -mno-dspr2 1398 -mmcu -mmno-mcu 1399 -meva -mno-eva 1400 -mvirt -mno-virt 1401 -mxpa -mno-xpa 1402 -mcrc -mno-crc 1403 -mginv -mno-ginv 1404 -mmicromips -mno-micromips 1405 -mmsa -mno-msa 1406 -mloongson-mmi -mno-loongson-mmi 1407 -mloongson-ext -mno-loongson-ext 1408 -mloongson-ext2 -mno-loongson-ext2 1409 -mfpu=FPU-TYPE 1410 -msmartmips -mno-smartmips 1411 -mpaired-single -mno-paired-single -mdmx -mno-mdmx 1412 -mips3d -mno-mips3d -mmt -mno-mt -mllsc -mno-llsc 1413 -mlong64 -mlong32 -msym32 -mno-sym32 1414 -GNUM -mlocal-sdata -mno-local-sdata 1415 -mextern-sdata -mno-extern-sdata -mgpopt -mno-gopt 1416 -membedded-data -mno-embedded-data 1417 -muninit-const-in-rodata -mno-uninit-const-in-rodata 1418 -mcode-readable=SETTING 1419 -msplit-addresses -mno-split-addresses 1420 -mexplicit-relocs -mno-explicit-relocs 1421 -mcheck-zero-division -mno-check-zero-division 1422 -mdivide-traps -mdivide-breaks 1423 -mload-store-pairs -mno-load-store-pairs 1424 -munaligned-access -mno-unaligned-access 1425 -mmemcpy -mno-memcpy -mlong-calls -mno-long-calls 1426 -mmad -mno-mad -mimadd -mno-imadd -mfused-madd -mno-fused-madd -nocpp 1427 -mfix-24k -mno-fix-24k 1428 -mfix-r4000 -mno-fix-r4000 -mfix-r4400 -mno-fix-r4400 1429 -mfix-r5900 -mno-fix-r5900 1430 -mfix-r10000 -mno-fix-r10000 -mfix-rm7000 -mno-fix-rm7000 1431 -mfix-vr4120 -mno-fix-vr4120 1432 -mfix-vr4130 -mno-fix-vr4130 -mfix-sb1 -mno-fix-sb1 1433 -mflush-func=FUNC -mno-flush-func 1434 -mbranch-cost=NUM -mbranch-likely -mno-branch-likely 1435 -mcompact-branches=POLICY 1436 -mfp-exceptions -mno-fp-exceptions 1437 -mvr4130-align -mno-vr4130-align -msynci -mno-synci 1438 -mlxc1-sxc1 -mno-lxc1-sxc1 -mmadd4 -mno-madd4 1439 -mrelax-pic-calls -mno-relax-pic-calls -mmcount-ra-address 1440 -mframe-header-opt -mno-frame-header-opt 1441 1442 _MMIX Options_ 1443 -mlibfuncs -mno-libfuncs -mepsilon -mno-epsilon -mabi=gnu 1444 -mabi=mmixware -mzero-extend -mknuthdiv -mtoplevel-symbols 1445 -melf -mbranch-predict -mno-branch-predict -mbase-addresses 1446 -mno-base-addresses -msingle-exit -mno-single-exit 1447 1448 _MN10300 Options_ 1449 -mmult-bug -mno-mult-bug 1450 -mno-am33 -mam33 -mam33-2 -mam34 1451 -mtune=CPU-TYPE 1452 -mreturn-pointer-on-d0 1453 -mno-crt0 -mrelax -mliw -msetlb 1454 1455 _Moxie Options_ 1456 -meb -mel -mmul.x -mno-crt0 1457 1458 _MSP430 Options_ 1459 -msim -masm-hex -mmcu= -mcpu= -mlarge -msmall -mrelax 1460 -mwarn-mcu 1461 -mcode-region= -mdata-region= 1462 -msilicon-errata= -msilicon-errata-warn= 1463 -mhwmult= -minrt -mtiny-printf -mmax-inline-shift= 1464 1465 _NDS32 Options_ 1466 -mbig-endian -mlittle-endian 1467 -mreduced-regs -mfull-regs 1468 -mcmov -mno-cmov 1469 -mext-perf -mno-ext-perf 1470 -mext-perf2 -mno-ext-perf2 1471 -mext-string -mno-ext-string 1472 -mv3push -mno-v3push 1473 -m16bit -mno-16bit 1474 -misr-vector-size=NUM 1475 -mcache-block-size=NUM 1476 -march=ARCH 1477 -mcmodel=CODE-MODEL 1478 -mctor-dtor -mrelax 1479 1480 _Nios II Options_ 1481 -G NUM -mgpopt=OPTION -mgpopt -mno-gpopt 1482 -mgprel-sec=REGEXP -mr0rel-sec=REGEXP 1483 -mel -meb 1484 -mno-bypass-cache -mbypass-cache 1485 -mno-cache-volatile -mcache-volatile 1486 -mno-fast-sw-div -mfast-sw-div 1487 -mhw-mul -mno-hw-mul -mhw-mulx -mno-hw-mulx -mno-hw-div -mhw-div 1488 -mcustom-INSN=N -mno-custom-INSN 1489 -mcustom-fpu-cfg=NAME 1490 -mhal -msmallc -msys-crt0=NAME -msys-lib=NAME 1491 -march=ARCH -mbmx -mno-bmx -mcdx -mno-cdx 1492 1493 _Nvidia PTX Options_ 1494 -m64 -mmainkernel -moptimize 1495 1496 _OpenRISC Options_ 1497 -mboard=NAME -mnewlib -mhard-mul -mhard-div 1498 -msoft-mul -msoft-div 1499 -msoft-float -mhard-float -mdouble-float -munordered-float 1500 -mcmov -mror -mrori -msext -msfimm -mshftimm 1501 -mcmodel=CODE-MODEL 1502 1503 _PDP-11 Options_ 1504 -mfpu -msoft-float -mac0 -mno-ac0 -m40 -m45 -m10 1505 -mint32 -mno-int16 -mint16 -mno-int32 1506 -msplit -munix-asm -mdec-asm -mgnu-asm -mlra 1507 1508 _picoChip Options_ 1509 -mae=AE_TYPE -mvliw-lookahead=N 1510 -msymbol-as-address -mno-inefficient-warnings 1511 1512 _PowerPC Options_ See RS/6000 and PowerPC Options. 1513 1514 _PRU Options_ 1515 -mmcu=MCU -minrt -mno-relax -mloop 1516 -mabi=VARIANT 1517 1518 _RISC-V Options_ 1519 -mbranch-cost=N-INSTRUCTION 1520 -mplt -mno-plt 1521 -mabi=ABI-STRING 1522 -mfdiv -mno-fdiv 1523 -mdiv -mno-div 1524 -misa-spec=ISA-SPEC-STRING 1525 -march=ISA-STRING 1526 -mtune=PROCESSOR-STRING 1527 -mpreferred-stack-boundary=NUM 1528 -msmall-data-limit=N-BYTES 1529 -msave-restore -mno-save-restore 1530 -mshorten-memrefs -mno-shorten-memrefs 1531 -mstrict-align -mno-strict-align 1532 -mcmodel=medlow -mcmodel=medany 1533 -mexplicit-relocs -mno-explicit-relocs 1534 -mrelax -mno-relax 1535 -mriscv-attribute -mmo-riscv-attribute 1536 -malign-data=TYPE 1537 -mbig-endian -mlittle-endian 1538 -mstack-protector-guard=GUARD -mstack-protector-guard-reg=REG 1539 -mstack-protector-guard-offset=OFFSET 1540 1541 _RL78 Options_ 1542 -msim -mmul=none -mmul=g13 -mmul=g14 -mallregs 1543 -mcpu=g10 -mcpu=g13 -mcpu=g14 -mg10 -mg13 -mg14 1544 -m64bit-doubles -m32bit-doubles -msave-mduc-in-interrupts 1545 1546 _RS/6000 and PowerPC Options_ 1547 -mcpu=CPU-TYPE 1548 -mtune=CPU-TYPE 1549 -mcmodel=CODE-MODEL 1550 -mpowerpc64 1551 -maltivec -mno-altivec 1552 -mpowerpc-gpopt -mno-powerpc-gpopt 1553 -mpowerpc-gfxopt -mno-powerpc-gfxopt 1554 -mmfcrf -mno-mfcrf -mpopcntb -mno-popcntb -mpopcntd -mno-popcntd 1555 -mfprnd -mno-fprnd 1556 -mcmpb -mno-cmpb -mhard-dfp -mno-hard-dfp 1557 -mfull-toc -mminimal-toc -mno-fp-in-toc -mno-sum-in-toc 1558 -m64 -m32 -mxl-compat -mno-xl-compat -mpe 1559 -malign-power -malign-natural 1560 -msoft-float -mhard-float -mmultiple -mno-multiple 1561 -mupdate -mno-update 1562 -mavoid-indexed-addresses -mno-avoid-indexed-addresses 1563 -mfused-madd -mno-fused-madd -mbit-align -mno-bit-align 1564 -mstrict-align -mno-strict-align -mrelocatable 1565 -mno-relocatable -mrelocatable-lib -mno-relocatable-lib 1566 -mtoc -mno-toc -mlittle -mlittle-endian -mbig -mbig-endian 1567 -mdynamic-no-pic -mswdiv -msingle-pic-base 1568 -mprioritize-restricted-insns=PRIORITY 1569 -msched-costly-dep=DEPENDENCE_TYPE 1570 -minsert-sched-nops=SCHEME 1571 -mcall-aixdesc -mcall-eabi -mcall-freebsd 1572 -mcall-linux -mcall-netbsd -mcall-openbsd 1573 -mcall-sysv -mcall-sysv-eabi -mcall-sysv-noeabi 1574 -mtraceback=TRACEBACK_TYPE 1575 -maix-struct-return -msvr4-struct-return 1576 -mabi=ABI-TYPE -msecure-plt -mbss-plt 1577 -mlongcall -mno-longcall -mpltseq -mno-pltseq 1578 -mblock-move-inline-limit=NUM 1579 -mblock-compare-inline-limit=NUM 1580 -mblock-compare-inline-loop-limit=NUM 1581 -mno-block-ops-unaligned-vsx 1582 -mstring-compare-inline-limit=NUM 1583 -misel -mno-isel 1584 -mvrsave -mno-vrsave 1585 -mmulhw -mno-mulhw 1586 -mdlmzb -mno-dlmzb 1587 -mprototype -mno-prototype 1588 -msim -mmvme -mads -myellowknife -memb -msdata 1589 -msdata=OPT -mreadonly-in-sdata -mvxworks -G NUM 1590 -mrecip -mrecip=OPT -mno-recip -mrecip-precision 1591 -mno-recip-precision 1592 -mveclibabi=TYPE -mfriz -mno-friz 1593 -mpointers-to-nested-functions -mno-pointers-to-nested-functions 1594 -msave-toc-indirect -mno-save-toc-indirect 1595 -mpower8-fusion -mno-mpower8-fusion -mpower8-vector -mno-power8-vector 1596 -mcrypto -mno-crypto -mhtm -mno-htm 1597 -mquad-memory -mno-quad-memory 1598 -mquad-memory-atomic -mno-quad-memory-atomic 1599 -mcompat-align-parm -mno-compat-align-parm 1600 -mfloat128 -mno-float128 -mfloat128-hardware -mno-float128-hardware 1601 -mgnu-attribute -mno-gnu-attribute 1602 -mstack-protector-guard=GUARD -mstack-protector-guard-reg=REG 1603 -mstack-protector-guard-offset=OFFSET -mprefixed -mno-prefixed 1604 -mpcrel -mno-pcrel -mmma -mno-mmma -mrop-protect -mno-rop-protect 1605 -mprivileged -mno-privileged 1606 1607 _RX Options_ 1608 -m64bit-doubles -m32bit-doubles -fpu -nofpu 1609 -mcpu= 1610 -mbig-endian-data -mlittle-endian-data 1611 -msmall-data 1612 -msim -mno-sim 1613 -mas100-syntax -mno-as100-syntax 1614 -mrelax 1615 -mmax-constant-size= 1616 -mint-register= 1617 -mpid 1618 -mallow-string-insns -mno-allow-string-insns 1619 -mjsr 1620 -mno-warn-multiple-fast-interrupts 1621 -msave-acc-in-interrupts 1622 1623 _S/390 and zSeries Options_ 1624 -mtune=CPU-TYPE -march=CPU-TYPE 1625 -mhard-float -msoft-float -mhard-dfp -mno-hard-dfp 1626 -mlong-double-64 -mlong-double-128 1627 -mbackchain -mno-backchain -mpacked-stack -mno-packed-stack 1628 -msmall-exec -mno-small-exec -mmvcle -mno-mvcle 1629 -m64 -m31 -mdebug -mno-debug -mesa -mzarch 1630 -mhtm -mvx -mzvector 1631 -mtpf-trace -mno-tpf-trace -mtpf-trace-skip -mno-tpf-trace-skip 1632 -mfused-madd -mno-fused-madd 1633 -mwarn-framesize -mwarn-dynamicstack -mstack-size -mstack-guard 1634 -mhotpatch=HALFWORDS,HALFWORDS 1635 1636 _Score Options_ 1637 -meb -mel 1638 -mnhwloop 1639 -muls 1640 -mmac 1641 -mscore5 -mscore5u -mscore7 -mscore7d 1642 1643 _SH Options_ 1644 -m1 -m2 -m2e 1645 -m2a-nofpu -m2a-single-only -m2a-single -m2a 1646 -m3 -m3e 1647 -m4-nofpu -m4-single-only -m4-single -m4 1648 -m4a-nofpu -m4a-single-only -m4a-single -m4a -m4al 1649 -mb -ml -mdalign -mrelax 1650 -mbigtable -mfmovd -mrenesas -mno-renesas -mnomacsave 1651 -mieee -mno-ieee -mbitops -misize -minline-ic_invalidate -mpadstruct 1652 -mprefergot -musermode -multcost=NUMBER -mdiv=STRATEGY 1653 -mdivsi3_libfunc=NAME -mfixed-range=REGISTER-RANGE 1654 -maccumulate-outgoing-args 1655 -matomic-model=ATOMIC-MODEL 1656 -mbranch-cost=NUM -mzdcbranch -mno-zdcbranch 1657 -mcbranch-force-delay-slot 1658 -mfused-madd -mno-fused-madd -mfsca -mno-fsca -mfsrra -mno-fsrra 1659 -mpretend-cmove -mtas 1660 1661 _Solaris 2 Options_ 1662 -mclear-hwcap -mno-clear-hwcap -mimpure-text -mno-impure-text 1663 -pthreads 1664 1665 _SPARC Options_ 1666 -mcpu=CPU-TYPE 1667 -mtune=CPU-TYPE 1668 -mcmodel=CODE-MODEL 1669 -mmemory-model=MEM-MODEL 1670 -m32 -m64 -mapp-regs -mno-app-regs 1671 -mfaster-structs -mno-faster-structs -mflat -mno-flat 1672 -mfpu -mno-fpu -mhard-float -msoft-float 1673 -mhard-quad-float -msoft-quad-float 1674 -mstack-bias -mno-stack-bias 1675 -mstd-struct-return -mno-std-struct-return 1676 -munaligned-doubles -mno-unaligned-doubles 1677 -muser-mode -mno-user-mode 1678 -mv8plus -mno-v8plus -mvis -mno-vis 1679 -mvis2 -mno-vis2 -mvis3 -mno-vis3 1680 -mvis4 -mno-vis4 -mvis4b -mno-vis4b 1681 -mcbcond -mno-cbcond -mfmaf -mno-fmaf -mfsmuld -mno-fsmuld 1682 -mpopc -mno-popc -msubxc -mno-subxc 1683 -mfix-at697f -mfix-ut699 -mfix-ut700 -mfix-gr712rc 1684 -mlra -mno-lra 1685 1686 _System V Options_ 1687 -Qy -Qn -YP,PATHS -Ym,DIR 1688 1689 _TILE-Gx Options_ 1690 -mcpu=CPU -m32 -m64 -mbig-endian -mlittle-endian 1691 -mcmodel=CODE-MODEL 1692 1693 _TILEPro Options_ 1694 -mcpu=CPU -m32 1695 1696 _V850 Options_ 1697 -mlong-calls -mno-long-calls -mep -mno-ep 1698 -mprolog-function -mno-prolog-function -mspace 1699 -mtda=N -msda=N -mzda=N 1700 -mapp-regs -mno-app-regs 1701 -mdisable-callt -mno-disable-callt 1702 -mv850e2v3 -mv850e2 -mv850e1 -mv850es 1703 -mv850e -mv850 -mv850e3v5 1704 -mloop 1705 -mrelax 1706 -mlong-jumps 1707 -msoft-float 1708 -mhard-float 1709 -mgcc-abi 1710 -mrh850-abi 1711 -mbig-switch 1712 1713 _VAX Options_ 1714 -mg -mgnu -munix -mlra 1715 1716 _Visium Options_ 1717 -mdebug -msim -mfpu -mno-fpu -mhard-float -msoft-float 1718 -mcpu=CPU-TYPE -mtune=CPU-TYPE -msv-mode -muser-mode 1719 1720 _VMS Options_ 1721 -mvms-return-codes -mdebug-main=PREFIX -mmalloc64 1722 -mpointer-size=SIZE 1723 1724 _VxWorks Options_ 1725 -mrtp -non-static -Bstatic -Bdynamic 1726 -Xbind-lazy -Xbind-now 1727 1728 _x86 Options_ 1729 -mtune=CPU-TYPE -march=CPU-TYPE 1730 -mtune-ctrl=FEATURE-LIST -mdump-tune-features -mno-default 1731 -mfpmath=UNIT 1732 -masm=DIALECT -mno-fancy-math-387 1733 -mno-fp-ret-in-387 -m80387 -mhard-float -msoft-float 1734 -mno-wide-multiply -mrtd -malign-double 1735 -mpreferred-stack-boundary=NUM 1736 -mincoming-stack-boundary=NUM 1737 -mcld -mcx16 -msahf -mmovbe -mcrc32 -mmwait 1738 -mrecip -mrecip=OPT 1739 -mvzeroupper -mprefer-avx128 -mprefer-vector-width=OPT 1740 -mmove-max=BITS -mstore-max=BITS 1741 -mmmx -msse -msse2 -msse3 -mssse3 -msse4.1 -msse4.2 -msse4 -mavx 1742 -mavx2 -mavx512f -mavx512pf -mavx512er -mavx512cd -mavx512vl 1743 -mavx512bw -mavx512dq -mavx512ifma -mavx512vbmi -msha -maes 1744 -mpclmul -mfsgsbase -mrdrnd -mf16c -mfma -mpconfig -mwbnoinvd 1745 -mptwrite -mprefetchwt1 -mclflushopt -mclwb -mxsavec -mxsaves 1746 -msse4a -m3dnow -m3dnowa -mpopcnt -mabm -mbmi -mtbm -mfma4 -mxop 1747 -madx -mlzcnt -mbmi2 -mfxsr -mxsave -mxsaveopt -mrtm -mhle -mlwp 1748 -mmwaitx -mclzero -mpku -mthreads -mgfni -mvaes -mwaitpkg 1749 -mshstk -mmanual-endbr -mforce-indirect-call -mavx512vbmi2 -mavx512bf16 -menqcmd 1750 -mvpclmulqdq -mavx512bitalg -mmovdiri -mmovdir64b -mavx512vpopcntdq 1751 -mavx5124fmaps -mavx512vnni -mavx5124vnniw -mprfchw -mrdpid 1752 -mrdseed -msgx -mavx512vp2intersect -mserialize -mtsxldtrk 1753 -mamx-tile -mamx-int8 -mamx-bf16 -muintr -mhreset -mavxvnni 1754 -mavx512fp16 1755 -mcldemote -mms-bitfields -mno-align-stringops -minline-all-stringops 1756 -minline-stringops-dynamically -mstringop-strategy=ALG 1757 -mkl -mwidekl 1758 -mmemcpy-strategy=STRATEGY -mmemset-strategy=STRATEGY 1759 -mpush-args -maccumulate-outgoing-args -m128bit-long-double 1760 -m96bit-long-double -mlong-double-64 -mlong-double-80 -mlong-double-128 1761 -mregparm=NUM -msseregparm 1762 -mveclibabi=TYPE -mvect8-ret-in-mem 1763 -mpc32 -mpc64 -mpc80 -mstackrealign 1764 -momit-leaf-frame-pointer -mno-red-zone -mno-tls-direct-seg-refs 1765 -mcmodel=CODE-MODEL -mabi=NAME -maddress-mode=MODE 1766 -m32 -m64 -mx32 -m16 -miamcu -mlarge-data-threshold=NUM 1767 -msse2avx -mfentry -mrecord-mcount -mnop-mcount -m8bit-idiv 1768 -minstrument-return=TYPE -mfentry-name=NAME -mfentry-section=NAME 1769 -mavx256-split-unaligned-load -mavx256-split-unaligned-store 1770 -malign-data=TYPE -mstack-protector-guard=GUARD 1771 -mstack-protector-guard-reg=REG 1772 -mstack-protector-guard-offset=OFFSET 1773 -mstack-protector-guard-symbol=SYMBOL 1774 -mgeneral-regs-only -mcall-ms2sysv-xlogues -mrelax-cmpxchg-loop 1775 -mindirect-branch=CHOICE -mfunction-return=CHOICE 1776 -mindirect-branch-register -mharden-sls=CHOICE 1777 -mindirect-branch-cs-prefix -mneeded -mno-direct-extern-access 1778 1779 _x86 Windows Options_ 1780 -mconsole -mcygwin -mno-cygwin -mdll 1781 -mnop-fun-dllimport -mthread 1782 -municode -mwin32 -mwindows -fno-set-stack-executable 1783 1784 _Xstormy16 Options_ 1785 -msim 1786 1787 _Xtensa Options_ 1788 -mconst16 -mno-const16 1789 -mfused-madd -mno-fused-madd 1790 -mforce-no-pic 1791 -mserialize-volatile -mno-serialize-volatile 1792 -mtext-section-literals -mno-text-section-literals 1793 -mauto-litpools -mno-auto-litpools 1794 -mtarget-align -mno-target-align 1795 -mlongcalls -mno-longcalls 1796 -mabi=ABI-TYPE 1797 1798 _zSeries Options_ See S/390 and zSeries Options. 1799 1800 1801File: gcc.info, Node: Overall Options, Next: Invoking G++, Prev: Option Summary, Up: Invoking GCC 1802 18033.2 Options Controlling the Kind of Output 1804========================================== 1805 1806Compilation can involve up to four stages: preprocessing, compilation 1807proper, assembly and linking, always in that order. GCC is capable of 1808preprocessing and compiling several files either into several assembler 1809input files, or into one assembler input file; then each assembler input 1810file produces an object file, and linking combines all the object files 1811(those newly compiled, and those specified as input) into an executable 1812file. 1813 1814 For any given input file, the file name suffix determines what kind of 1815compilation is done: 1816 1817'FILE.c' 1818 C source code that must be preprocessed. 1819 1820'FILE.i' 1821 C source code that should not be preprocessed. 1822 1823'FILE.ii' 1824 C++ source code that should not be preprocessed. 1825 1826'FILE.m' 1827 Objective-C source code. Note that you must link with the 1828 'libobjc' library to make an Objective-C program work. 1829 1830'FILE.mi' 1831 Objective-C source code that should not be preprocessed. 1832 1833'FILE.mm' 1834'FILE.M' 1835 Objective-C++ source code. Note that you must link with the 1836 'libobjc' library to make an Objective-C++ program work. Note that 1837 '.M' refers to a literal capital M. 1838 1839'FILE.mii' 1840 Objective-C++ source code that should not be preprocessed. 1841 1842'FILE.h' 1843 C, C++, Objective-C or Objective-C++ header file to be turned into 1844 a precompiled header (default), or C, C++ header file to be turned 1845 into an Ada spec (via the '-fdump-ada-spec' switch). 1846 1847'FILE.cc' 1848'FILE.cp' 1849'FILE.cxx' 1850'FILE.cpp' 1851'FILE.CPP' 1852'FILE.c++' 1853'FILE.C' 1854 C++ source code that must be preprocessed. Note that in '.cxx', 1855 the last two letters must both be literally 'x'. Likewise, '.C' 1856 refers to a literal capital C. 1857 1858'FILE.mm' 1859'FILE.M' 1860 Objective-C++ source code that must be preprocessed. 1861 1862'FILE.mii' 1863 Objective-C++ source code that should not be preprocessed. 1864 1865'FILE.hh' 1866'FILE.H' 1867'FILE.hp' 1868'FILE.hxx' 1869'FILE.hpp' 1870'FILE.HPP' 1871'FILE.h++' 1872'FILE.tcc' 1873 C++ header file to be turned into a precompiled header or Ada spec. 1874 1875'FILE.f' 1876'FILE.for' 1877'FILE.ftn' 1878 Fixed form Fortran source code that should not be preprocessed. 1879 1880'FILE.F' 1881'FILE.FOR' 1882'FILE.fpp' 1883'FILE.FPP' 1884'FILE.FTN' 1885 Fixed form Fortran source code that must be preprocessed (with the 1886 traditional preprocessor). 1887 1888'FILE.f90' 1889'FILE.f95' 1890'FILE.f03' 1891'FILE.f08' 1892 Free form Fortran source code that should not be preprocessed. 1893 1894'FILE.F90' 1895'FILE.F95' 1896'FILE.F03' 1897'FILE.F08' 1898 Free form Fortran source code that must be preprocessed (with the 1899 traditional preprocessor). 1900 1901'FILE.go' 1902 Go source code. 1903 1904'FILE.d' 1905 D source code. 1906 1907'FILE.di' 1908 D interface file. 1909 1910'FILE.dd' 1911 D documentation code (Ddoc). 1912 1913'FILE.ads' 1914 Ada source code file that contains a library unit declaration (a 1915 declaration of a package, subprogram, or generic, or a generic 1916 instantiation), or a library unit renaming declaration (a package, 1917 generic, or subprogram renaming declaration). Such files are also 1918 called "specs". 1919 1920'FILE.adb' 1921 Ada source code file containing a library unit body (a subprogram 1922 or package body). Such files are also called "bodies". 1923 1924'FILE.s' 1925 Assembler code. 1926 1927'FILE.S' 1928'FILE.sx' 1929 Assembler code that must be preprocessed. 1930 1931'OTHER' 1932 An object file to be fed straight into linking. Any file name with 1933 no recognized suffix is treated this way. 1934 1935 You can specify the input language explicitly with the '-x' option: 1936 1937'-x LANGUAGE' 1938 Specify explicitly the LANGUAGE for the following input files 1939 (rather than letting the compiler choose a default based on the 1940 file name suffix). This option applies to all following input 1941 files until the next '-x' option. Possible values for LANGUAGE 1942 are: 1943 c c-header cpp-output 1944 c++ c++-header c++-system-header c++-user-header c++-cpp-output 1945 objective-c objective-c-header objective-c-cpp-output 1946 objective-c++ objective-c++-header objective-c++-cpp-output 1947 assembler assembler-with-cpp 1948 ada 1949 d 1950 f77 f77-cpp-input f95 f95-cpp-input 1951 go 1952 1953'-x none' 1954 Turn off any specification of a language, so that subsequent files 1955 are handled according to their file name suffixes (as they are if 1956 '-x' has not been used at all). 1957 1958 If you only want some of the stages of compilation, you can use '-x' 1959(or filename suffixes) to tell 'gcc' where to start, and one of the 1960options '-c', '-S', or '-E' to say where 'gcc' is to stop. Note that 1961some combinations (for example, '-x cpp-output -E') instruct 'gcc' to do 1962nothing at all. 1963 1964'-c' 1965 Compile or assemble the source files, but do not link. The linking 1966 stage simply is not done. The ultimate output is in the form of an 1967 object file for each source file. 1968 1969 By default, the object file name for a source file is made by 1970 replacing the suffix '.c', '.i', '.s', etc., with '.o'. 1971 1972 Unrecognized input files, not requiring compilation or assembly, 1973 are ignored. 1974 1975'-S' 1976 Stop after the stage of compilation proper; do not assemble. The 1977 output is in the form of an assembler code file for each 1978 non-assembler input file specified. 1979 1980 By default, the assembler file name for a source file is made by 1981 replacing the suffix '.c', '.i', etc., with '.s'. 1982 1983 Input files that don't require compilation are ignored. 1984 1985'-E' 1986 Stop after the preprocessing stage; do not run the compiler proper. 1987 The output is in the form of preprocessed source code, which is 1988 sent to the standard output. 1989 1990 Input files that don't require preprocessing are ignored. 1991 1992'-o FILE' 1993 Place the primary output in file FILE. This applies to whatever 1994 sort of output is being produced, whether it be an executable file, 1995 an object file, an assembler file or preprocessed C code. 1996 1997 If '-o' is not specified, the default is to put an executable file 1998 in 'a.out', the object file for 'SOURCE.SUFFIX' in 'SOURCE.o', its 1999 assembler file in 'SOURCE.s', a precompiled header file in 2000 'SOURCE.SUFFIX.gch', and all preprocessed C source on standard 2001 output. 2002 2003 Though '-o' names only the primary output, it also affects the 2004 naming of auxiliary and dump outputs. See the examples below. 2005 Unless overridden, both auxiliary outputs and dump outputs are 2006 placed in the same directory as the primary output. In auxiliary 2007 outputs, the suffix of the input file is replaced with that of the 2008 auxiliary output file type; in dump outputs, the suffix of the dump 2009 file is appended to the input file suffix. In compilation 2010 commands, the base name of both auxiliary and dump outputs is that 2011 of the primary output; in compile and link commands, the primary 2012 output name, minus the executable suffix, is combined with the 2013 input file name. If both share the same base name, disregarding 2014 the suffix, the result of the combination is that base name, 2015 otherwise, they are concatenated, separated by a dash. 2016 2017 gcc -c foo.c ... 2018 2019 will use 'foo.o' as the primary output, and place aux outputs and 2020 dumps next to it, e.g., aux file 'foo.dwo' for '-gsplit-dwarf', and 2021 dump file 'foo.c.???r.final' for '-fdump-rtl-final'. 2022 2023 If a non-linker output file is explicitly specified, aux and dump 2024 files by default take the same base name: 2025 2026 gcc -c foo.c -o dir/foobar.o ... 2027 2028 will name aux outputs 'dir/foobar.*' and dump outputs 2029 'dir/foobar.c.*'. 2030 2031 A linker output will instead prefix aux and dump outputs: 2032 2033 gcc foo.c bar.c -o dir/foobar ... 2034 2035 will generally name aux outputs 'dir/foobar-foo.*' and 2036 'dir/foobar-bar.*', and dump outputs 'dir/foobar-foo.c.*' and 2037 'dir/foobar-bar.c.*'. 2038 2039 The one exception to the above is when the executable shares the 2040 base name with the single input: 2041 2042 gcc foo.c -o dir/foo ... 2043 2044 in which case aux outputs are named 'dir/foo.*' and dump outputs 2045 named 'dir/foo.c.*'. 2046 2047 The location and the names of auxiliary and dump outputs can be 2048 adjusted by the options '-dumpbase', '-dumpbase-ext', '-dumpdir', 2049 '-save-temps=cwd', and '-save-temps=obj'. 2050 2051'-dumpbase DUMPBASE' 2052 This option sets the base name for auxiliary and dump output files. 2053 It does not affect the name of the primary output file. 2054 Intermediate outputs, when preserved, are not regarded as primary 2055 outputs, but as auxiliary outputs: 2056 2057 gcc -save-temps -S foo.c 2058 2059 saves the (no longer) temporary preprocessed file in 'foo.i', and 2060 then compiles to the (implied) output file 'foo.s', whereas: 2061 2062 gcc -save-temps -dumpbase save-foo -c foo.c 2063 2064 preprocesses to in 'save-foo.i', compiles to 'save-foo.s' (now an 2065 intermediate, thus auxiliary output), and then assembles to the 2066 (implied) output file 'foo.o'. 2067 2068 Absent this option, dump and aux files take their names from the 2069 input file, or from the (non-linker) output file, if one is 2070 explicitly specified: dump output files (e.g. those requested by 2071 '-fdump-*' options) with the input name suffix, and aux output 2072 files (those requested by other non-dump options, e.g. 2073 '-save-temps', '-gsplit-dwarf', '-fcallgraph-info') without it. 2074 2075 Similar suffix differentiation of dump and aux outputs can be 2076 attained for explicitly-given '-dumpbase basename.suf' by also 2077 specifying '-dumpbase-ext .suf'. 2078 2079 If DUMPBASE is explicitly specified with any directory component, 2080 any DUMPPFX specification (e.g. '-dumpdir' or '-save-temps=*') is 2081 ignored, and instead of appending to it, DUMPBASE fully overrides 2082 it: 2083 2084 gcc foo.c -c -o dir/foo.o -dumpbase alt/foo \ 2085 -dumpdir pfx- -save-temps=cwd ... 2086 2087 creates auxiliary and dump outputs named 'alt/foo.*', disregarding 2088 'dir/' in '-o', the './' prefix implied by '-save-temps=cwd', and 2089 'pfx-' in '-dumpdir'. 2090 2091 When '-dumpbase' is specified in a command that compiles multiple 2092 inputs, or that compiles and then links, it may be combined with 2093 DUMPPFX, as specified under '-dumpdir'. Then, each input file is 2094 compiled using the combined DUMPPFX, and default values for 2095 DUMPBASE and AUXDROPSUF are computed for each input file: 2096 2097 gcc foo.c bar.c -c -dumpbase main ... 2098 2099 creates 'foo.o' and 'bar.o' as primary outputs, and avoids 2100 overwriting the auxiliary and dump outputs by using the DUMPBASE as 2101 a prefix, creating auxiliary and dump outputs named 'main-foo.*' 2102 and 'main-bar.*'. 2103 2104 An empty string specified as DUMPBASE avoids the influence of the 2105 output basename in the naming of auxiliary and dump outputs during 2106 compilation, computing default values : 2107 2108 gcc -c foo.c -o dir/foobar.o -dumpbase '' ... 2109 2110 will name aux outputs 'dir/foo.*' and dump outputs 'dir/foo.c.*'. 2111 Note how their basenames are taken from the input name, but the 2112 directory still defaults to that of the output. 2113 2114 The empty-string dumpbase does not prevent the use of the output 2115 basename for outputs during linking: 2116 2117 gcc foo.c bar.c -o dir/foobar -dumpbase '' -flto ... 2118 2119 The compilation of the source files will name auxiliary outputs 2120 'dir/foo.*' and 'dir/bar.*', and dump outputs 'dir/foo.c.*' and 2121 'dir/bar.c.*'. LTO recompilation during linking will use 2122 'dir/foobar.' as the prefix for dumps and auxiliary files. 2123 2124'-dumpbase-ext AUXDROPSUF' 2125 When forming the name of an auxiliary (but not a dump) output file, 2126 drop trailing AUXDROPSUF from DUMPBASE before appending any 2127 suffixes. If not specified, this option defaults to the suffix of 2128 a default DUMPBASE, i.e., the suffix of the input file when 2129 '-dumpbase' is not present in the command line, or DUMPBASE is 2130 combined with DUMPPFX. 2131 2132 gcc foo.c -c -o dir/foo.o -dumpbase x-foo.c -dumpbase-ext .c ... 2133 2134 creates 'dir/foo.o' as the main output, and generates auxiliary 2135 outputs in 'dir/x-foo.*', taking the location of the primary 2136 output, and dropping the '.c' suffix from the DUMPBASE. Dump 2137 outputs retain the suffix: 'dir/x-foo.c.*'. 2138 2139 This option is disregarded if it does not match the suffix of a 2140 specified DUMPBASE, except as an alternative to the executable 2141 suffix when appending the linker output base name to DUMPPFX, as 2142 specified below: 2143 2144 gcc foo.c bar.c -o main.out -dumpbase-ext .out ... 2145 2146 creates 'main.out' as the primary output, and avoids overwriting 2147 the auxiliary and dump outputs by using the executable name minus 2148 AUXDROPSUF as a prefix, creating auxiliary outputs named 2149 'main-foo.*' and 'main-bar.*' and dump outputs named 'main-foo.c.*' 2150 and 'main-bar.c.*'. 2151 2152'-dumpdir DUMPPFX' 2153 When forming the name of an auxiliary or dump output file, use 2154 DUMPPFX as a prefix: 2155 2156 gcc -dumpdir pfx- -c foo.c ... 2157 2158 creates 'foo.o' as the primary output, and auxiliary outputs named 2159 'pfx-foo.*', combining the given DUMPPFX with the default DUMPBASE 2160 derived from the default primary output, derived in turn from the 2161 input name. Dump outputs also take the input name suffix: 2162 'pfx-foo.c.*'. 2163 2164 If DUMPPFX is to be used as a directory name, it must end with a 2165 directory separator: 2166 2167 gcc -dumpdir dir/ -c foo.c -o obj/bar.o ... 2168 2169 creates 'obj/bar.o' as the primary output, and auxiliary outputs 2170 named 'dir/bar.*', combining the given DUMPPFX with the default 2171 DUMPBASE derived from the primary output name. Dump outputs also 2172 take the input name suffix: 'dir/bar.c.*'. 2173 2174 It defaults to the location of the output file, unless the output 2175 file is a special file like '/dev/null'. Options '-save-temps=cwd' 2176 and '-save-temps=obj' override this default, just like an explicit 2177 '-dumpdir' option. In case multiple such options are given, the 2178 last one prevails: 2179 2180 gcc -dumpdir pfx- -c foo.c -save-temps=obj ... 2181 2182 outputs 'foo.o', with auxiliary outputs named 'foo.*' because 2183 '-save-temps=*' overrides the DUMPPFX given by the earlier 2184 '-dumpdir' option. It does not matter that '=obj' is the default 2185 for '-save-temps', nor that the output directory is implicitly the 2186 current directory. Dump outputs are named 'foo.c.*'. 2187 2188 When compiling from multiple input files, if '-dumpbase' is 2189 specified, DUMPBASE, minus a AUXDROPSUF suffix, and a dash are 2190 appended to (or override, if containing any directory components) 2191 an explicit or defaulted DUMPPFX, so that each of the multiple 2192 compilations gets differently-named aux and dump outputs. 2193 2194 gcc foo.c bar.c -c -dumpdir dir/pfx- -dumpbase main ... 2195 2196 outputs auxiliary dumps to 'dir/pfx-main-foo.*' and 2197 'dir/pfx-main-bar.*', appending DUMPBASE- to DUMPPFX. Dump outputs 2198 retain the input file suffix: 'dir/pfx-main-foo.c.*' and 2199 'dir/pfx-main-bar.c.*', respectively. Contrast with the 2200 single-input compilation: 2201 2202 gcc foo.c -c -dumpdir dir/pfx- -dumpbase main ... 2203 2204 that, applying '-dumpbase' to a single source, does not compute and 2205 append a separate DUMPBASE per input file. Its auxiliary and dump 2206 outputs go in 'dir/pfx-main.*'. 2207 2208 When compiling and then linking from multiple input files, a 2209 defaulted or explicitly specified DUMPPFX also undergoes the 2210 DUMPBASE- transformation above (e.g. the compilation of 'foo.c' 2211 and 'bar.c' above, but without '-c'). If neither '-dumpdir' nor 2212 '-dumpbase' are given, the linker output base name, minus 2213 AUXDROPSUF, if specified, or the executable suffix otherwise, plus 2214 a dash is appended to the default DUMPPFX instead. Note, however, 2215 that unlike earlier cases of linking: 2216 2217 gcc foo.c bar.c -dumpdir dir/pfx- -o main ... 2218 2219 does not append the output name 'main' to DUMPPFX, because 2220 '-dumpdir' is explicitly specified. The goal is that the 2221 explicitly-specified DUMPPFX may contain the specified output name 2222 as part of the prefix, if desired; only an explicitly-specified 2223 '-dumpbase' would be combined with it, in order to avoid simply 2224 discarding a meaningful option. 2225 2226 When compiling and then linking from a single input file, the 2227 linker output base name will only be appended to the default 2228 DUMPPFX as above if it does not share the base name with the single 2229 input file name. This has been covered in single-input linking 2230 cases above, but not with an explicit '-dumpdir' that inhibits the 2231 combination, even if overridden by '-save-temps=*': 2232 2233 gcc foo.c -dumpdir alt/pfx- -o dir/main.exe -save-temps=cwd ... 2234 2235 Auxiliary outputs are named 'foo.*', and dump outputs 'foo.c.*', in 2236 the current working directory as ultimately requested by 2237 '-save-temps=cwd'. 2238 2239 Summing it all up for an intuitive though slightly imprecise data 2240 flow: the primary output name is broken into a directory part and a 2241 basename part; DUMPPFX is set to the former, unless overridden by 2242 '-dumpdir' or '-save-temps=*', and DUMPBASE is set to the latter, 2243 unless overriden by '-dumpbase'. If there are multiple inputs or 2244 linking, this DUMPBASE may be combined with DUMPPFX and taken from 2245 each input file. Auxiliary output names for each input are formed 2246 by combining DUMPPFX, DUMPBASE minus suffix, and the auxiliary 2247 output suffix; dump output names are only different in that the 2248 suffix from DUMPBASE is retained. 2249 2250 When it comes to auxiliary and dump outputs created during LTO 2251 recompilation, a combination of DUMPPFX and DUMPBASE, as given or 2252 as derived from the linker output name but not from inputs, even in 2253 cases in which this combination would not otherwise be used as 2254 such, is passed down with a trailing period replacing the 2255 compiler-added dash, if any, as a '-dumpdir' option to 2256 'lto-wrapper'; being involved in linking, this program does not 2257 normally get any '-dumpbase' and '-dumpbase-ext', and it ignores 2258 them. 2259 2260 When running sub-compilers, 'lto-wrapper' appends LTO stage names 2261 to the received DUMPPFX, ensures it contains a directory component 2262 so that it overrides any '-dumpdir', and passes that as '-dumpbase' 2263 to sub-compilers. 2264 2265'-v' 2266 Print (on standard error output) the commands executed to run the 2267 stages of compilation. Also print the version number of the 2268 compiler driver program and of the preprocessor and the compiler 2269 proper. 2270 2271'-###' 2272 Like '-v' except the commands are not executed and arguments are 2273 quoted unless they contain only alphanumeric characters or './-_'. 2274 This is useful for shell scripts to capture the driver-generated 2275 command lines. 2276 2277'--help' 2278 Print (on the standard output) a description of the command-line 2279 options understood by 'gcc'. If the '-v' option is also specified 2280 then '--help' is also passed on to the various processes invoked by 2281 'gcc', so that they can display the command-line options they 2282 accept. If the '-Wextra' option has also been specified (prior to 2283 the '--help' option), then command-line options that have no 2284 documentation associated with them are also displayed. 2285 2286'--target-help' 2287 Print (on the standard output) a description of target-specific 2288 command-line options for each tool. For some targets extra 2289 target-specific information may also be printed. 2290 2291'--help={CLASS|[^]QUALIFIER}[,...]' 2292 Print (on the standard output) a description of the command-line 2293 options understood by the compiler that fit into all specified 2294 classes and qualifiers. These are the supported classes: 2295 2296 'optimizers' 2297 Display all of the optimization options supported by the 2298 compiler. 2299 2300 'warnings' 2301 Display all of the options controlling warning messages 2302 produced by the compiler. 2303 2304 'target' 2305 Display target-specific options. Unlike the '--target-help' 2306 option however, target-specific options of the linker and 2307 assembler are not displayed. This is because those tools do 2308 not currently support the extended '--help=' syntax. 2309 2310 'params' 2311 Display the values recognized by the '--param' option. 2312 2313 LANGUAGE 2314 Display the options supported for LANGUAGE, where LANGUAGE is 2315 the name of one of the languages supported in this version of 2316 GCC. If an option is supported by all languages, one needs to 2317 select 'common' class. 2318 2319 'common' 2320 Display the options that are common to all languages. 2321 2322 These are the supported qualifiers: 2323 2324 'undocumented' 2325 Display only those options that are undocumented. 2326 2327 'joined' 2328 Display options taking an argument that appears after an equal 2329 sign in the same continuous piece of text, such as: 2330 '--help=target'. 2331 2332 'separate' 2333 Display options taking an argument that appears as a separate 2334 word following the original option, such as: '-o output-file'. 2335 2336 Thus for example to display all the undocumented target-specific 2337 switches supported by the compiler, use: 2338 2339 --help=target,undocumented 2340 2341 The sense of a qualifier can be inverted by prefixing it with the 2342 '^' character, so for example to display all binary warning options 2343 (i.e., ones that are either on or off and that do not take an 2344 argument) that have a description, use: 2345 2346 --help=warnings,^joined,^undocumented 2347 2348 The argument to '--help=' should not consist solely of inverted 2349 qualifiers. 2350 2351 Combining several classes is possible, although this usually 2352 restricts the output so much that there is nothing to display. One 2353 case where it does work, however, is when one of the classes is 2354 TARGET. For example, to display all the target-specific 2355 optimization options, use: 2356 2357 --help=target,optimizers 2358 2359 The '--help=' option can be repeated on the command line. Each 2360 successive use displays its requested class of options, skipping 2361 those that have already been displayed. If '--help' is also 2362 specified anywhere on the command line then this takes precedence 2363 over any '--help=' option. 2364 2365 If the '-Q' option appears on the command line before the '--help=' 2366 option, then the descriptive text displayed by '--help=' is 2367 changed. Instead of describing the displayed options, an 2368 indication is given as to whether the option is enabled, disabled 2369 or set to a specific value (assuming that the compiler knows this 2370 at the point where the '--help=' option is used). 2371 2372 Here is a truncated example from the ARM port of 'gcc': 2373 2374 % gcc -Q -mabi=2 --help=target -c 2375 The following options are target specific: 2376 -mabi= 2 2377 -mabort-on-noreturn [disabled] 2378 -mapcs [disabled] 2379 2380 The output is sensitive to the effects of previous command-line 2381 options, so for example it is possible to find out which 2382 optimizations are enabled at '-O2' by using: 2383 2384 -Q -O2 --help=optimizers 2385 2386 Alternatively you can discover which binary optimizations are 2387 enabled by '-O3' by using: 2388 2389 gcc -c -Q -O3 --help=optimizers > /tmp/O3-opts 2390 gcc -c -Q -O2 --help=optimizers > /tmp/O2-opts 2391 diff /tmp/O2-opts /tmp/O3-opts | grep enabled 2392 2393'--version' 2394 Display the version number and copyrights of the invoked GCC. 2395 2396'-pass-exit-codes' 2397 Normally the 'gcc' program exits with the code of 1 if any phase of 2398 the compiler returns a non-success return code. If you specify 2399 '-pass-exit-codes', the 'gcc' program instead returns with the 2400 numerically highest error produced by any phase returning an error 2401 indication. The C, C++, and Fortran front ends return 4 if an 2402 internal compiler error is encountered. 2403 2404'-pipe' 2405 Use pipes rather than temporary files for communication between the 2406 various stages of compilation. This fails to work on some systems 2407 where the assembler is unable to read from a pipe; but the GNU 2408 assembler has no trouble. 2409 2410'-specs=FILE' 2411 Process FILE after the compiler reads in the standard 'specs' file, 2412 in order to override the defaults which the 'gcc' driver program 2413 uses when determining what switches to pass to 'cc1', 'cc1plus', 2414 'as', 'ld', etc. More than one '-specs=FILE' can be specified on 2415 the command line, and they are processed in order, from left to 2416 right. *Note Spec Files::, for information about the format of the 2417 FILE. 2418 2419'-wrapper' 2420 Invoke all subcommands under a wrapper program. The name of the 2421 wrapper program and its parameters are passed as a comma separated 2422 list. 2423 2424 gcc -c t.c -wrapper gdb,--args 2425 2426 This invokes all subprograms of 'gcc' under 'gdb --args', thus the 2427 invocation of 'cc1' is 'gdb --args cc1 ...'. 2428 2429'-ffile-prefix-map=OLD=NEW' 2430 When compiling files residing in directory 'OLD', record any 2431 references to them in the result of the compilation as if the files 2432 resided in directory 'NEW' instead. Specifying this option is 2433 equivalent to specifying all the individual '-f*-prefix-map' 2434 options. This can be used to make reproducible builds that are 2435 location independent. See also '-fmacro-prefix-map', 2436 '-fdebug-prefix-map' and '-fprofile-prefix-map'. 2437 2438'-fplugin=NAME.so' 2439 Load the plugin code in file NAME.so, assumed to be a shared object 2440 to be dlopen'd by the compiler. The base name of the shared object 2441 file is used to identify the plugin for the purposes of argument 2442 parsing (See '-fplugin-arg-NAME-KEY=VALUE' below). Each plugin 2443 should define the callback functions specified in the Plugins API. 2444 2445'-fplugin-arg-NAME-KEY=VALUE' 2446 Define an argument called KEY with a value of VALUE for the plugin 2447 called NAME. 2448 2449'-fdump-ada-spec[-slim]' 2450 For C and C++ source and include files, generate corresponding Ada 2451 specs. *Note (gnat_ugn)Generating Ada Bindings for C and C++ 2452 headers::, which provides detailed documentation on this feature. 2453 2454'-fada-spec-parent=UNIT' 2455 In conjunction with '-fdump-ada-spec[-slim]' above, generate Ada 2456 specs as child units of parent UNIT. 2457 2458'-fdump-go-spec=FILE' 2459 For input files in any language, generate corresponding Go 2460 declarations in FILE. This generates Go 'const', 'type', 'var', 2461 and 'func' declarations which may be a useful way to start writing 2462 a Go interface to code written in some other language. 2463 2464'@FILE' 2465 Read command-line options from FILE. The options read are inserted 2466 in place of the original @FILE option. If FILE does not exist, or 2467 cannot be read, then the option will be treated literally, and not 2468 removed. 2469 2470 Options in FILE are separated by whitespace. A whitespace 2471 character may be included in an option by surrounding the entire 2472 option in either single or double quotes. Any character (including 2473 a backslash) may be included by prefixing the character to be 2474 included with a backslash. The FILE may itself contain additional 2475 @FILE options; any such options will be processed recursively. 2476 2477 2478File: gcc.info, Node: Invoking G++, Next: C Dialect Options, Prev: Overall Options, Up: Invoking GCC 2479 24803.3 Compiling C++ Programs 2481========================== 2482 2483C++ source files conventionally use one of the suffixes '.C', '.cc', 2484'.cpp', '.CPP', '.c++', '.cp', or '.cxx'; C++ header files often use 2485'.hh', '.hpp', '.H', or (for shared template code) '.tcc'; and 2486preprocessed C++ files use the suffix '.ii'. GCC recognizes files with 2487these names and compiles them as C++ programs even if you call the 2488compiler the same way as for compiling C programs (usually with the name 2489'gcc'). 2490 2491 However, the use of 'gcc' does not add the C++ library. 'g++' is a 2492program that calls GCC and automatically specifies linking against the 2493C++ library. It treats '.c', '.h' and '.i' files as C++ source files 2494instead of C source files unless '-x' is used. This program is also 2495useful when precompiling a C header file with a '.h' extension for use 2496in C++ compilations. On many systems, 'g++' is also installed with the 2497name 'c++'. 2498 2499 When you compile C++ programs, you may specify many of the same 2500command-line options that you use for compiling programs in any 2501language; or command-line options meaningful for C and related 2502languages; or options that are meaningful only for C++ programs. *Note 2503Options Controlling C Dialect: C Dialect Options, for explanations of 2504options for languages related to C. *Note Options Controlling C++ 2505Dialect: C++ Dialect Options, for explanations of options that are 2506meaningful only for C++ programs. 2507 2508 2509File: gcc.info, Node: C Dialect Options, Next: C++ Dialect Options, Prev: Invoking G++, Up: Invoking GCC 2510 25113.4 Options Controlling C Dialect 2512================================= 2513 2514The following options control the dialect of C (or languages derived 2515from C, such as C++, Objective-C and Objective-C++) that the compiler 2516accepts: 2517 2518'-ansi' 2519 In C mode, this is equivalent to '-std=c90'. In C++ mode, it is 2520 equivalent to '-std=c++98'. 2521 2522 This turns off certain features of GCC that are incompatible with 2523 ISO C90 (when compiling C code), or of standard C++ (when compiling 2524 C++ code), such as the 'asm' and 'typeof' keywords, and predefined 2525 macros such as 'unix' and 'vax' that identify the type of system 2526 you are using. It also enables the undesirable and rarely used ISO 2527 trigraph feature. For the C compiler, it disables recognition of 2528 C++ style '//' comments as well as the 'inline' keyword. 2529 2530 The alternate keywords '__asm__', '__extension__', '__inline__' and 2531 '__typeof__' continue to work despite '-ansi'. You would not want 2532 to use them in an ISO C program, of course, but it is useful to put 2533 them in header files that might be included in compilations done 2534 with '-ansi'. Alternate predefined macros such as '__unix__' and 2535 '__vax__' are also available, with or without '-ansi'. 2536 2537 The '-ansi' option does not cause non-ISO programs to be rejected 2538 gratuitously. For that, '-Wpedantic' is required in addition to 2539 '-ansi'. *Note Warning Options::. 2540 2541 The macro '__STRICT_ANSI__' is predefined when the '-ansi' option 2542 is used. Some header files may notice this macro and refrain from 2543 declaring certain functions or defining certain macros that the ISO 2544 standard doesn't call for; this is to avoid interfering with any 2545 programs that might use these names for other things. 2546 2547 Functions that are normally built in but do not have semantics 2548 defined by ISO C (such as 'alloca' and 'ffs') are not built-in 2549 functions when '-ansi' is used. *Note Other built-in functions 2550 provided by GCC: Other Builtins, for details of the functions 2551 affected. 2552 2553'-std=' 2554 Determine the language standard. *Note Language Standards 2555 Supported by GCC: Standards, for details of these standard 2556 versions. This option is currently only supported when compiling C 2557 or C++. 2558 2559 The compiler can accept several base standards, such as 'c90' or 2560 'c++98', and GNU dialects of those standards, such as 'gnu90' or 2561 'gnu++98'. When a base standard is specified, the compiler accepts 2562 all programs following that standard plus those using GNU 2563 extensions that do not contradict it. For example, '-std=c90' 2564 turns off certain features of GCC that are incompatible with ISO 2565 C90, such as the 'asm' and 'typeof' keywords, but not other GNU 2566 extensions that do not have a meaning in ISO C90, such as omitting 2567 the middle term of a '?:' expression. On the other hand, when a 2568 GNU dialect of a standard is specified, all features supported by 2569 the compiler are enabled, even when those features change the 2570 meaning of the base standard. As a result, some strict-conforming 2571 programs may be rejected. The particular standard is used by 2572 '-Wpedantic' to identify which features are GNU extensions given 2573 that version of the standard. For example '-std=gnu90 -Wpedantic' 2574 warns about C++ style '//' comments, while '-std=gnu99 -Wpedantic' 2575 does not. 2576 2577 A value for this option must be provided; possible values are 2578 2579 'c90' 2580 'c89' 2581 'iso9899:1990' 2582 Support all ISO C90 programs (certain GNU extensions that 2583 conflict with ISO C90 are disabled). Same as '-ansi' for C 2584 code. 2585 2586 'iso9899:199409' 2587 ISO C90 as modified in amendment 1. 2588 2589 'c99' 2590 'c9x' 2591 'iso9899:1999' 2592 'iso9899:199x' 2593 ISO C99. This standard is substantially completely supported, 2594 modulo bugs and floating-point issues (mainly but not entirely 2595 relating to optional C99 features from Annexes F and G). See 2596 <https://gcc.gnu.org/c99status.html> for more information. 2597 The names 'c9x' and 'iso9899:199x' are deprecated. 2598 2599 'c11' 2600 'c1x' 2601 'iso9899:2011' 2602 ISO C11, the 2011 revision of the ISO C standard. This 2603 standard is substantially completely supported, modulo bugs, 2604 floating-point issues (mainly but not entirely relating to 2605 optional C11 features from Annexes F and G) and the optional 2606 Annexes K (Bounds-checking interfaces) and L (Analyzability). 2607 The name 'c1x' is deprecated. 2608 2609 'c17' 2610 'c18' 2611 'iso9899:2017' 2612 'iso9899:2018' 2613 ISO C17, the 2017 revision of the ISO C standard (published in 2614 2018). This standard is same as C11 except for corrections of 2615 defects (all of which are also applied with '-std=c11') and a 2616 new value of '__STDC_VERSION__', and so is supported to the 2617 same extent as C11. 2618 2619 'c2x' 2620 The next version of the ISO C standard, still under 2621 development. The support for this version is experimental and 2622 incomplete. 2623 2624 'gnu90' 2625 'gnu89' 2626 GNU dialect of ISO C90 (including some C99 features). 2627 2628 'gnu99' 2629 'gnu9x' 2630 GNU dialect of ISO C99. The name 'gnu9x' is deprecated. 2631 2632 'gnu11' 2633 'gnu1x' 2634 GNU dialect of ISO C11. The name 'gnu1x' is deprecated. 2635 2636 'gnu17' 2637 'gnu18' 2638 GNU dialect of ISO C17. This is the default for C code. 2639 2640 'gnu2x' 2641 The next version of the ISO C standard, still under 2642 development, plus GNU extensions. The support for this 2643 version is experimental and incomplete. 2644 2645 'c++98' 2646 'c++03' 2647 The 1998 ISO C++ standard plus the 2003 technical corrigendum 2648 and some additional defect reports. Same as '-ansi' for C++ 2649 code. 2650 2651 'gnu++98' 2652 'gnu++03' 2653 GNU dialect of '-std=c++98'. 2654 2655 'c++11' 2656 'c++0x' 2657 The 2011 ISO C++ standard plus amendments. The name 'c++0x' 2658 is deprecated. 2659 2660 'gnu++11' 2661 'gnu++0x' 2662 GNU dialect of '-std=c++11'. The name 'gnu++0x' is 2663 deprecated. 2664 2665 'c++14' 2666 'c++1y' 2667 The 2014 ISO C++ standard plus amendments. The name 'c++1y' 2668 is deprecated. 2669 2670 'gnu++14' 2671 'gnu++1y' 2672 GNU dialect of '-std=c++14'. The name 'gnu++1y' is 2673 deprecated. 2674 2675 'c++17' 2676 'c++1z' 2677 The 2017 ISO C++ standard plus amendments. The name 'c++1z' 2678 is deprecated. 2679 2680 'gnu++17' 2681 'gnu++1z' 2682 GNU dialect of '-std=c++17'. This is the default for C++ 2683 code. The name 'gnu++1z' is deprecated. 2684 2685 'c++20' 2686 'c++2a' 2687 The 2020 ISO C++ standard plus amendments. Support is 2688 experimental, and could change in incompatible ways in future 2689 releases. The name 'c++2a' is deprecated. 2690 2691 'gnu++20' 2692 'gnu++2a' 2693 GNU dialect of '-std=c++20'. Support is experimental, and 2694 could change in incompatible ways in future releases. The 2695 name 'gnu++2a' is deprecated. 2696 2697 'c++2b' 2698 'c++23' 2699 The next revision of the ISO C++ standard, planned for 2023. 2700 Support is highly experimental, and will almost certainly 2701 change in incompatible ways in future releases. 2702 2703 'gnu++2b' 2704 'gnu++23' 2705 GNU dialect of '-std=c++2b'. Support is highly experimental, 2706 and will almost certainly change in incompatible ways in 2707 future releases. 2708 2709'-aux-info FILENAME' 2710 Output to the given filename prototyped declarations for all 2711 functions declared and/or defined in a translation unit, including 2712 those in header files. This option is silently ignored in any 2713 language other than C. 2714 2715 Besides declarations, the file indicates, in comments, the origin 2716 of each declaration (source file and line), whether the declaration 2717 was implicit, prototyped or unprototyped ('I', 'N' for new or 'O' 2718 for old, respectively, in the first character after the line number 2719 and the colon), and whether it came from a declaration or a 2720 definition ('C' or 'F', respectively, in the following character). 2721 In the case of function definitions, a K&R-style list of arguments 2722 followed by their declarations is also provided, inside comments, 2723 after the declaration. 2724 2725'-fallow-parameterless-variadic-functions' 2726 Accept variadic functions without named parameters. 2727 2728 Although it is possible to define such a function, this is not very 2729 useful as it is not possible to read the arguments. This is only 2730 supported for C as this construct is allowed by C++. 2731 2732'-fno-asm' 2733 Do not recognize 'asm', 'inline' or 'typeof' as a keyword, so that 2734 code can use these words as identifiers. You can use the keywords 2735 '__asm__', '__inline__' and '__typeof__' instead. In C, '-ansi' 2736 implies '-fno-asm'. 2737 2738 In C++, 'inline' is a standard keyword and is not affected by this 2739 switch. You may want to use the '-fno-gnu-keywords' flag instead, 2740 which disables 'typeof' but not 'asm' and 'inline'. In C99 mode 2741 ('-std=c99' or '-std=gnu99'), this switch only affects the 'asm' 2742 and 'typeof' keywords, since 'inline' is a standard keyword in ISO 2743 C99. 2744 2745'-fno-builtin' 2746'-fno-builtin-FUNCTION' 2747 Don't recognize built-in functions that do not begin with 2748 '__builtin_' as prefix. *Note Other built-in functions provided by 2749 GCC: Other Builtins, for details of the functions affected, 2750 including those which are not built-in functions when '-ansi' or 2751 '-std' options for strict ISO C conformance are used because they 2752 do not have an ISO standard meaning. 2753 2754 GCC normally generates special code to handle certain built-in 2755 functions more efficiently; for instance, calls to 'alloca' may 2756 become single instructions which adjust the stack directly, and 2757 calls to 'memcpy' may become inline copy loops. The resulting code 2758 is often both smaller and faster, but since the function calls no 2759 longer appear as such, you cannot set a breakpoint on those calls, 2760 nor can you change the behavior of the functions by linking with a 2761 different library. In addition, when a function is recognized as a 2762 built-in function, GCC may use information about that function to 2763 warn about problems with calls to that function, or to generate 2764 more efficient code, even if the resulting code still contains 2765 calls to that function. For example, warnings are given with 2766 '-Wformat' for bad calls to 'printf' when 'printf' is built in and 2767 'strlen' is known not to modify global memory. 2768 2769 With the '-fno-builtin-FUNCTION' option only the built-in function 2770 FUNCTION is disabled. FUNCTION must not begin with '__builtin_'. 2771 If a function is named that is not built-in in this version of GCC, 2772 this option is ignored. There is no corresponding 2773 '-fbuiltin-FUNCTION' option; if you wish to enable built-in 2774 functions selectively when using '-fno-builtin' or 2775 '-ffreestanding', you may define macros such as: 2776 2777 #define abs(n) __builtin_abs ((n)) 2778 #define strcpy(d, s) __builtin_strcpy ((d), (s)) 2779 2780'-fcond-mismatch' 2781 Allow conditional expressions with mismatched types in the second 2782 and third arguments. The value of such an expression is void. 2783 This option is not supported for C++. 2784 2785'-ffreestanding' 2786 2787 Assert that compilation targets a freestanding environment. This 2788 implies '-fno-builtin'. A freestanding environment is one in which 2789 the standard library may not exist, and program startup may not 2790 necessarily be at 'main'. The most obvious example is an OS 2791 kernel. This is equivalent to '-fno-hosted'. 2792 2793 *Note Language Standards Supported by GCC: Standards, for details 2794 of freestanding and hosted environments. 2795 2796'-fgimple' 2797 2798 Enable parsing of function definitions marked with '__GIMPLE'. 2799 This is an experimental feature that allows unit testing of GIMPLE 2800 passes. 2801 2802'-fgnu-tm' 2803 When the option '-fgnu-tm' is specified, the compiler generates 2804 code for the Linux variant of Intel's current Transactional Memory 2805 ABI specification document (Revision 1.1, May 6 2009). This is an 2806 experimental feature whose interface may change in future versions 2807 of GCC, as the official specification changes. Please note that 2808 not all architectures are supported for this feature. 2809 2810 For more information on GCC's support for transactional memory, 2811 *Note The GNU Transactional Memory Library: (libitm)Enabling 2812 libitm. 2813 2814 Note that the transactional memory feature is not supported with 2815 non-call exceptions ('-fnon-call-exceptions'). 2816 2817'-fgnu89-inline' 2818 The option '-fgnu89-inline' tells GCC to use the traditional GNU 2819 semantics for 'inline' functions when in C99 mode. *Note An Inline 2820 Function is As Fast As a Macro: Inline. Using this option is 2821 roughly equivalent to adding the 'gnu_inline' function attribute to 2822 all inline functions (*note Function Attributes::). 2823 2824 The option '-fno-gnu89-inline' explicitly tells GCC to use the C99 2825 semantics for 'inline' when in C99 or gnu99 mode (i.e., it 2826 specifies the default behavior). This option is not supported in 2827 '-std=c90' or '-std=gnu90' mode. 2828 2829 The preprocessor macros '__GNUC_GNU_INLINE__' and 2830 '__GNUC_STDC_INLINE__' may be used to check which semantics are in 2831 effect for 'inline' functions. *Note (cpp)Common Predefined 2832 Macros::. 2833 2834'-fhosted' 2835 2836 Assert that compilation targets a hosted environment. This implies 2837 '-fbuiltin'. A hosted environment is one in which the entire 2838 standard library is available, and in which 'main' has a return 2839 type of 'int'. Examples are nearly everything except a kernel. 2840 This is equivalent to '-fno-freestanding'. 2841 2842'-flax-vector-conversions' 2843 Allow implicit conversions between vectors with differing numbers 2844 of elements and/or incompatible element types. This option should 2845 not be used for new code. 2846 2847'-fms-extensions' 2848 Accept some non-standard constructs used in Microsoft header files. 2849 2850 In C++ code, this allows member names in structures to be similar 2851 to previous types declarations. 2852 2853 typedef int UOW; 2854 struct ABC { 2855 UOW UOW; 2856 }; 2857 2858 Some cases of unnamed fields in structures and unions are only 2859 accepted with this option. *Note Unnamed struct/union fields 2860 within structs/unions: Unnamed Fields, for details. 2861 2862 Note that this option is off for all targets except for x86 targets 2863 using ms-abi. 2864 2865'-foffload=disable' 2866'-foffload=default' 2867'-foffload=TARGET-LIST' 2868 Specify for which OpenMP and OpenACC offload targets code should be 2869 generated. The default behavior, equivalent to 2870 '-foffload=default', is to generate code for all supported offload 2871 targets. The '-foffload=disable' form generates code only for the 2872 host fallback, while '-foffload=TARGET-LIST' generates code only 2873 for the specified comma-separated list of offload targets. 2874 2875 Offload targets are specified in GCC's internal target-triplet 2876 format. You can run the compiler with '-v' to show the list of 2877 configured offload targets under 'OFFLOAD_TARGET_NAMES'. 2878 2879'-foffload-options=OPTIONS' 2880'-foffload-options=TARGET-TRIPLET-LIST=OPTIONS' 2881 2882 With '-foffload-options=OPTIONS', GCC passes the specified OPTIONS 2883 to the compilers for all enabled offloading targets. You can 2884 specify options that apply only to a specific target or targets by 2885 using the '-foffload-options=TARGET-LIST=OPTIONS' form. The 2886 TARGET-LIST is a comma-separated list in the same format as for the 2887 '-foffload=' option. 2888 2889 Typical command lines are 2890 2891 -foffload-options=-lgfortran -foffload-options=-lm 2892 -foffload-options="-lgfortran -lm" -foffload-options=nvptx-none=-latomic 2893 -foffload-options=amdgcn-amdhsa=-march=gfx906 -foffload-options=-lm 2894 2895'-fopenacc' 2896 Enable handling of OpenACC directives '#pragma acc' in C/C++ and 2897 '!$acc' in Fortran. When '-fopenacc' is specified, the compiler 2898 generates accelerated code according to the OpenACC Application 2899 Programming Interface v2.6 <https://www.openacc.org>. This option 2900 implies '-pthread', and thus is only supported on targets that have 2901 support for '-pthread'. 2902 2903'-fopenacc-dim=GEOM' 2904 Specify default compute dimensions for parallel offload regions 2905 that do not explicitly specify. The GEOM value is a triple of 2906 ':'-separated sizes, in order 'gang', 'worker' and, 'vector'. A 2907 size can be omitted, to use a target-specific default value. 2908 2909'-fopenmp' 2910 Enable handling of OpenMP directives '#pragma omp' in C/C++ and 2911 '!$omp' in Fortran. When '-fopenmp' is specified, the compiler 2912 generates parallel code according to the OpenMP Application Program 2913 Interface v4.5 <https://www.openmp.org>. This option implies 2914 '-pthread', and thus is only supported on targets that have support 2915 for '-pthread'. '-fopenmp' implies '-fopenmp-simd'. 2916 2917'-fopenmp-simd' 2918 Enable handling of OpenMP's SIMD directives with '#pragma omp' in 2919 C/C++ and '!$omp' in Fortran. Other OpenMP directives are ignored. 2920 2921'-fpermitted-flt-eval-methods=STYLE' 2922 ISO/IEC TS 18661-3 defines new permissible values for 2923 'FLT_EVAL_METHOD' that indicate that operations and constants with 2924 a semantic type that is an interchange or extended format should be 2925 evaluated to the precision and range of that type. These new 2926 values are a superset of those permitted under C99/C11, which does 2927 not specify the meaning of other positive values of 2928 'FLT_EVAL_METHOD'. As such, code conforming to C11 may not have 2929 been written expecting the possibility of the new values. 2930 2931 '-fpermitted-flt-eval-methods' specifies whether the compiler 2932 should allow only the values of 'FLT_EVAL_METHOD' specified in 2933 C99/C11, or the extended set of values specified in ISO/IEC TS 2934 18661-3. 2935 2936 STYLE is either 'c11' or 'ts-18661-3' as appropriate. 2937 2938 The default when in a standards compliant mode ('-std=c11' or 2939 similar) is '-fpermitted-flt-eval-methods=c11'. The default when 2940 in a GNU dialect ('-std=gnu11' or similar) is 2941 '-fpermitted-flt-eval-methods=ts-18661-3'. 2942 2943'-fplan9-extensions' 2944 Accept some non-standard constructs used in Plan 9 code. 2945 2946 This enables '-fms-extensions', permits passing pointers to 2947 structures with anonymous fields to functions that expect pointers 2948 to elements of the type of the field, and permits referring to 2949 anonymous fields declared using a typedef. *Note Unnamed 2950 struct/union fields within structs/unions: Unnamed Fields, for 2951 details. This is only supported for C, not C++. 2952 2953'-fsigned-bitfields' 2954'-funsigned-bitfields' 2955'-fno-signed-bitfields' 2956'-fno-unsigned-bitfields' 2957 These options control whether a bit-field is signed or unsigned, 2958 when the declaration does not use either 'signed' or 'unsigned'. 2959 By default, such a bit-field is signed, because this is consistent: 2960 the basic integer types such as 'int' are signed types. 2961 2962'-fsigned-char' 2963 Let the type 'char' be signed, like 'signed char'. 2964 2965 Note that this is equivalent to '-fno-unsigned-char', which is the 2966 negative form of '-funsigned-char'. Likewise, the option 2967 '-fno-signed-char' is equivalent to '-funsigned-char'. 2968 2969'-funsigned-char' 2970 Let the type 'char' be unsigned, like 'unsigned char'. 2971 2972 Each kind of machine has a default for what 'char' should be. It 2973 is either like 'unsigned char' by default or like 'signed char' by 2974 default. 2975 2976 Ideally, a portable program should always use 'signed char' or 2977 'unsigned char' when it depends on the signedness of an object. 2978 But many programs have been written to use plain 'char' and expect 2979 it to be signed, or expect it to be unsigned, depending on the 2980 machines they were written for. This option, and its inverse, let 2981 you make such a program work with the opposite default. 2982 2983 The type 'char' is always a distinct type from each of 'signed 2984 char' or 'unsigned char', even though its behavior is always just 2985 like one of those two. 2986 2987'-fsso-struct=ENDIANNESS' 2988 Set the default scalar storage order of structures and unions to 2989 the specified endianness. The accepted values are 'big-endian', 2990 'little-endian' and 'native' for the native endianness of the 2991 target (the default). This option is not supported for C++. 2992 2993 *Warning:* the '-fsso-struct' switch causes GCC to generate code 2994 that is not binary compatible with code generated without it if the 2995 specified endianness is not the native endianness of the target. 2996 2997 2998File: gcc.info, Node: C++ Dialect Options, Next: Objective-C and Objective-C++ Dialect Options, Prev: C Dialect Options, Up: Invoking GCC 2999 30003.5 Options Controlling C++ Dialect 3001=================================== 3002 3003This section describes the command-line options that are only meaningful 3004for C++ programs. You can also use most of the GNU compiler options 3005regardless of what language your program is in. For example, you might 3006compile a file 'firstClass.C' like this: 3007 3008 g++ -g -fstrict-enums -O -c firstClass.C 3009 3010In this example, only '-fstrict-enums' is an option meant only for C++ 3011programs; you can use the other options with any language supported by 3012GCC. 3013 3014 Some options for compiling C programs, such as '-std', are also 3015relevant for C++ programs. *Note Options Controlling C Dialect: C 3016Dialect Options. 3017 3018 Here is a list of options that are _only_ for compiling C++ programs: 3019 3020'-fabi-version=N' 3021 Use version N of the C++ ABI. The default is version 0. 3022 3023 Version 0 refers to the version conforming most closely to the C++ 3024 ABI specification. Therefore, the ABI obtained using version 0 3025 will change in different versions of G++ as ABI bugs are fixed. 3026 3027 Version 1 is the version of the C++ ABI that first appeared in G++ 3028 3.2. 3029 3030 Version 2 is the version of the C++ ABI that first appeared in G++ 3031 3.4, and was the default through G++ 4.9. 3032 3033 Version 3 corrects an error in mangling a constant address as a 3034 template argument. 3035 3036 Version 4, which first appeared in G++ 4.5, implements a standard 3037 mangling for vector types. 3038 3039 Version 5, which first appeared in G++ 4.6, corrects the mangling 3040 of attribute const/volatile on function pointer types, decltype of 3041 a plain decl, and use of a function parameter in the declaration of 3042 another parameter. 3043 3044 Version 6, which first appeared in G++ 4.7, corrects the promotion 3045 behavior of C++11 scoped enums and the mangling of template 3046 argument packs, const/static_cast, prefix ++ and -, and a class 3047 scope function used as a template argument. 3048 3049 Version 7, which first appeared in G++ 4.8, that treats nullptr_t 3050 as a builtin type and corrects the mangling of lambdas in default 3051 argument scope. 3052 3053 Version 8, which first appeared in G++ 4.9, corrects the 3054 substitution behavior of function types with 3055 function-cv-qualifiers. 3056 3057 Version 9, which first appeared in G++ 5.2, corrects the alignment 3058 of 'nullptr_t'. 3059 3060 Version 10, which first appeared in G++ 6.1, adds mangling of 3061 attributes that affect type identity, such as ia32 calling 3062 convention attributes (e.g. 'stdcall'). 3063 3064 Version 11, which first appeared in G++ 7, corrects the mangling of 3065 sizeof... expressions and operator names. For multiple entities 3066 with the same name within a function, that are declared in 3067 different scopes, the mangling now changes starting with the 3068 twelfth occurrence. It also implies '-fnew-inheriting-ctors'. 3069 3070 Version 12, which first appeared in G++ 8, corrects the calling 3071 conventions for empty classes on the x86_64 target and for classes 3072 with only deleted copy/move constructors. It accidentally changes 3073 the calling convention for classes with a deleted copy constructor 3074 and a trivial move constructor. 3075 3076 Version 13, which first appeared in G++ 8.2, fixes the accidental 3077 change in version 12. 3078 3079 Version 14, which first appeared in G++ 10, corrects the mangling 3080 of the nullptr expression. 3081 3082 Version 15, which first appeared in G++ 11, changes the mangling of 3083 '__alignof__' to be distinct from that of 'alignof', and dependent 3084 operator names. 3085 3086 See also '-Wabi'. 3087 3088'-fabi-compat-version=N' 3089 On targets that support strong aliases, G++ works around mangling 3090 changes by creating an alias with the correct mangled name when 3091 defining a symbol with an incorrect mangled name. This switch 3092 specifies which ABI version to use for the alias. 3093 3094 With '-fabi-version=0' (the default), this defaults to 11 (GCC 7 3095 compatibility). If another ABI version is explicitly selected, 3096 this defaults to 0. For compatibility with GCC versions 3.2 3097 through 4.9, use '-fabi-compat-version=2'. 3098 3099 If this option is not provided but '-Wabi=N' is, that version is 3100 used for compatibility aliases. If this option is provided along 3101 with '-Wabi' (without the version), the version from this option is 3102 used for the warning. 3103 3104'-fno-access-control' 3105 Turn off all access checking. This switch is mainly useful for 3106 working around bugs in the access control code. 3107 3108'-faligned-new' 3109 Enable support for C++17 'new' of types that require more alignment 3110 than 'void* ::operator new(std::size_t)' provides. A numeric 3111 argument such as '-faligned-new=32' can be used to specify how much 3112 alignment (in bytes) is provided by that function, but few users 3113 will need to override the default of 'alignof(std::max_align_t)'. 3114 3115 This flag is enabled by default for '-std=c++17'. 3116 3117'-fchar8_t' 3118'-fno-char8_t' 3119 Enable support for 'char8_t' as adopted for C++20. This includes 3120 the addition of a new 'char8_t' fundamental type, changes to the 3121 types of UTF-8 string and character literals, new signatures for 3122 user-defined literals, associated standard library updates, and new 3123 '__cpp_char8_t' and '__cpp_lib_char8_t' feature test macros. 3124 3125 This option enables functions to be overloaded for ordinary and 3126 UTF-8 strings: 3127 3128 int f(const char *); // #1 3129 int f(const char8_t *); // #2 3130 int v1 = f("text"); // Calls #1 3131 int v2 = f(u8"text"); // Calls #2 3132 3133 and introduces new signatures for user-defined literals: 3134 3135 int operator""_udl1(char8_t); 3136 int v3 = u8'x'_udl1; 3137 int operator""_udl2(const char8_t*, std::size_t); 3138 int v4 = u8"text"_udl2; 3139 template<typename T, T...> int operator""_udl3(); 3140 int v5 = u8"text"_udl3; 3141 3142 The change to the types of UTF-8 string and character literals 3143 introduces incompatibilities with ISO C++11 and later standards. 3144 For example, the following code is well-formed under ISO C++11, but 3145 is ill-formed when '-fchar8_t' is specified. 3146 3147 char ca[] = u8"xx"; // error: char-array initialized from wide 3148 // string 3149 const char *cp = u8"xx";// error: invalid conversion from 3150 // `const char8_t*' to `const char*' 3151 int f(const char*); 3152 auto v = f(u8"xx"); // error: invalid conversion from 3153 // `const char8_t*' to `const char*' 3154 std::string s{u8"xx"}; // error: no matching function for call to 3155 // `std::basic_string<char>::basic_string()' 3156 using namespace std::literals; 3157 s = u8"xx"s; // error: conversion from 3158 // `basic_string<char8_t>' to non-scalar 3159 // type `basic_string<char>' requested 3160 3161'-fcheck-new' 3162 Check that the pointer returned by 'operator new' is non-null 3163 before attempting to modify the storage allocated. This check is 3164 normally unnecessary because the C++ standard specifies that 3165 'operator new' only returns '0' if it is declared 'throw()', in 3166 which case the compiler always checks the return value even without 3167 this option. In all other cases, when 'operator new' has a 3168 non-empty exception specification, memory exhaustion is signalled 3169 by throwing 'std::bad_alloc'. See also 'new (nothrow)'. 3170 3171'-fconcepts' 3172'-fconcepts-ts' 3173 Below '-std=c++20', '-fconcepts' enables support for the C++ 3174 Extensions for Concepts Technical Specification, ISO 19217 (2015). 3175 3176 With '-std=c++20' and above, Concepts are part of the language 3177 standard, so '-fconcepts' defaults to on. But the standard 3178 specification of Concepts differs significantly from the TS, so 3179 some constructs that were allowed in the TS but didn't make it into 3180 the standard can still be enabled by '-fconcepts-ts'. 3181 3182'-fconstexpr-depth=N' 3183 Set the maximum nested evaluation depth for C++11 constexpr 3184 functions to N. A limit is needed to detect endless recursion 3185 during constant expression evaluation. The minimum specified by 3186 the standard is 512. 3187 3188'-fconstexpr-cache-depth=N' 3189 Set the maximum level of nested evaluation depth for C++11 3190 constexpr functions that will be cached to N. This is a heuristic 3191 that trades off compilation speed (when the cache avoids repeated 3192 calculations) against memory consumption (when the cache grows very 3193 large from highly recursive evaluations). The default is 8. Very 3194 few users are likely to want to adjust it, but if your code does 3195 heavy constexpr calculations you might want to experiment to find 3196 which value works best for you. 3197 3198'-fconstexpr-fp-except' 3199 Annex F of the C standard specifies that IEC559 floating point 3200 exceptions encountered at compile time should not stop compilation. 3201 C++ compilers have historically not followed this guidance, instead 3202 treating floating point division by zero as non-constant even 3203 though it has a well defined value. This flag tells the compiler 3204 to give Annex F priority over other rules saying that a particular 3205 operation is undefined. 3206 3207 constexpr float inf = 1./0.; // OK with -fconstexpr-fp-except 3208 3209'-fconstexpr-loop-limit=N' 3210 Set the maximum number of iterations for a loop in C++14 constexpr 3211 functions to N. A limit is needed to detect infinite loops during 3212 constant expression evaluation. The default is 262144 (1<<18). 3213 3214'-fconstexpr-ops-limit=N' 3215 Set the maximum number of operations during a single constexpr 3216 evaluation. Even when number of iterations of a single loop is 3217 limited with the above limit, if there are several nested loops and 3218 each of them has many iterations but still smaller than the above 3219 limit, or if in a body of some loop or even outside of a loop too 3220 many expressions need to be evaluated, the resulting constexpr 3221 evaluation might take too long. The default is 33554432 (1<<25). 3222 3223'-fcoroutines' 3224 Enable support for the C++ coroutines extension (experimental). 3225 3226'-fno-elide-constructors' 3227 The C++ standard allows an implementation to omit creating a 3228 temporary that is only used to initialize another object of the 3229 same type. Specifying this option disables that optimization, and 3230 forces G++ to call the copy constructor in all cases. This option 3231 also causes G++ to call trivial member functions which otherwise 3232 would be expanded inline. 3233 3234 In C++17, the compiler is required to omit these temporaries, but 3235 this option still affects trivial member functions. 3236 3237'-fno-enforce-eh-specs' 3238 Don't generate code to check for violation of exception 3239 specifications at run time. This option violates the C++ standard, 3240 but may be useful for reducing code size in production builds, much 3241 like defining 'NDEBUG'. This does not give user code permission to 3242 throw exceptions in violation of the exception specifications; the 3243 compiler still optimizes based on the specifications, so throwing 3244 an unexpected exception results in undefined behavior at run time. 3245 3246'-fextern-tls-init' 3247'-fno-extern-tls-init' 3248 The C++11 and OpenMP standards allow 'thread_local' and 3249 'threadprivate' variables to have dynamic (runtime) initialization. 3250 To support this, any use of such a variable goes through a wrapper 3251 function that performs any necessary initialization. When the use 3252 and definition of the variable are in the same translation unit, 3253 this overhead can be optimized away, but when the use is in a 3254 different translation unit there is significant overhead even if 3255 the variable doesn't actually need dynamic initialization. If the 3256 programmer can be sure that no use of the variable in a 3257 non-defining TU needs to trigger dynamic initialization (either 3258 because the variable is statically initialized, or a use of the 3259 variable in the defining TU will be executed before any uses in 3260 another TU), they can avoid this overhead with the 3261 '-fno-extern-tls-init' option. 3262 3263 On targets that support symbol aliases, the default is 3264 '-fextern-tls-init'. On targets that do not support symbol 3265 aliases, the default is '-fno-extern-tls-init'. 3266 3267'-ffold-simple-inlines' 3268'-fno-fold-simple-inlines' 3269 Permit the C++ frontend to fold calls to 'std::move', 3270 'std::forward', 'std::addressof' and 'std::as_const'. In contrast 3271 to inlining, this means no debug information will be generated for 3272 such calls. Since these functions are rarely interesting to debug, 3273 this flag is enabled by default unless '-fno-inline' is active. 3274 3275'-fno-gnu-keywords' 3276 Do not recognize 'typeof' as a keyword, so that code can use this 3277 word as an identifier. You can use the keyword '__typeof__' 3278 instead. This option is implied by the strict ISO C++ dialects: 3279 '-ansi', '-std=c++98', '-std=c++11', etc. 3280 3281'-fimplicit-constexpr' 3282 Make inline functions implicitly constexpr, if they satisfy the 3283 requirements for a constexpr function. This option can be used in 3284 C++14 mode or later. This can result in initialization changing 3285 from dynamic to static and other optimizations. 3286 3287'-fno-implicit-templates' 3288 Never emit code for non-inline templates that are instantiated 3289 implicitly (i.e. by use); only emit code for explicit 3290 instantiations. If you use this option, you must take care to 3291 structure your code to include all the necessary explicit 3292 instantiations to avoid getting undefined symbols at link time. 3293 *Note Template Instantiation::, for more information. 3294 3295'-fno-implicit-inline-templates' 3296 Don't emit code for implicit instantiations of inline templates, 3297 either. The default is to handle inlines differently so that 3298 compiles with and without optimization need the same set of 3299 explicit instantiations. 3300 3301'-fno-implement-inlines' 3302 To save space, do not emit out-of-line copies of inline functions 3303 controlled by '#pragma implementation'. This causes linker errors 3304 if these functions are not inlined everywhere they are called. 3305 3306'-fmodules-ts' 3307'-fno-modules-ts' 3308 Enable support for C++20 modules (*note C++ Modules::). The 3309 '-fno-modules-ts' is usually not needed, as that is the default. 3310 Even though this is a C++20 feature, it is not currently implicitly 3311 enabled by selecting that standard version. 3312 3313'-fmodule-header' 3314'-fmodule-header=user' 3315'-fmodule-header=system' 3316 Compile a header file to create an importable header unit. 3317 3318'-fmodule-implicit-inline' 3319 Member functions defined in their class definitions are not 3320 implicitly inline for modular code. This is different to 3321 traditional C++ behavior, for good reasons. However, it may result 3322 in a difficulty during code porting. This option makes such 3323 function definitions implicitly inline. It does however generate 3324 an ABI incompatibility, so you must use it everywhere or nowhere. 3325 (Such definitions outside of a named module remain implicitly 3326 inline, regardless.) 3327 3328'-fno-module-lazy' 3329 Disable lazy module importing and module mapper creation. 3330 3331'-fmodule-mapper=[HOSTNAME]:PORT[?IDENT]' 3332'-fmodule-mapper=|PROGRAM[?IDENT] ARGS...' 3333'-fmodule-mapper==SOCKET[?IDENT]' 3334'-fmodule-mapper=<>[INOUT][?IDENT]' 3335'-fmodule-mapper=<IN>OUT[?IDENT]' 3336'-fmodule-mapper=FILE[?IDENT]' 3337 An oracle to query for module name to filename mappings. If 3338 unspecified the 'CXX_MODULE_MAPPER' environment variable is used, 3339 and if that is unset, an in-process default is provided. 3340 3341'-fmodule-only' 3342 Only emit the Compiled Module Interface, inhibiting any object 3343 file. 3344 3345'-fms-extensions' 3346 Disable Wpedantic warnings about constructs used in MFC, such as 3347 implicit int and getting a pointer to member function via 3348 non-standard syntax. 3349 3350'-fnew-inheriting-ctors' 3351 Enable the P0136 adjustment to the semantics of C++11 constructor 3352 inheritance. This is part of C++17 but also considered to be a 3353 Defect Report against C++11 and C++14. This flag is enabled by 3354 default unless '-fabi-version=10' or lower is specified. 3355 3356'-fnew-ttp-matching' 3357 Enable the P0522 resolution to Core issue 150, template template 3358 parameters and default arguments: this allows a template with 3359 default template arguments as an argument for a template template 3360 parameter with fewer template parameters. This flag is enabled by 3361 default for '-std=c++17'. 3362 3363'-fno-nonansi-builtins' 3364 Disable built-in declarations of functions that are not mandated by 3365 ANSI/ISO C. These include 'ffs', 'alloca', '_exit', 'index', 3366 'bzero', 'conjf', and other related functions. 3367 3368'-fnothrow-opt' 3369 Treat a 'throw()' exception specification as if it were a 3370 'noexcept' specification to reduce or eliminate the text size 3371 overhead relative to a function with no exception specification. 3372 If the function has local variables of types with non-trivial 3373 destructors, the exception specification actually makes the 3374 function smaller because the EH cleanups for those variables can be 3375 optimized away. The semantic effect is that an exception thrown 3376 out of a function with such an exception specification results in a 3377 call to 'terminate' rather than 'unexpected'. 3378 3379'-fno-operator-names' 3380 Do not treat the operator name keywords 'and', 'bitand', 'bitor', 3381 'compl', 'not', 'or' and 'xor' as synonyms as keywords. 3382 3383'-fno-optional-diags' 3384 Disable diagnostics that the standard says a compiler does not need 3385 to issue. Currently, the only such diagnostic issued by G++ is the 3386 one for a name having multiple meanings within a class. 3387 3388'-fpermissive' 3389 Downgrade some diagnostics about nonconformant code from errors to 3390 warnings. Thus, using '-fpermissive' allows some nonconforming 3391 code to compile. 3392 3393'-fno-pretty-templates' 3394 When an error message refers to a specialization of a function 3395 template, the compiler normally prints the signature of the 3396 template followed by the template arguments and any typedefs or 3397 typenames in the signature (e.g. 'void f(T) [with T = int]' rather 3398 than 'void f(int)') so that it's clear which template is involved. 3399 When an error message refers to a specialization of a class 3400 template, the compiler omits any template arguments that match the 3401 default template arguments for that template. If either of these 3402 behaviors make it harder to understand the error message rather 3403 than easier, you can use '-fno-pretty-templates' to disable them. 3404 3405'-fno-rtti' 3406 Disable generation of information about every class with virtual 3407 functions for use by the C++ run-time type identification features 3408 ('dynamic_cast' and 'typeid'). If you don't use those parts of the 3409 language, you can save some space by using this flag. Note that 3410 exception handling uses the same information, but G++ generates it 3411 as needed. The 'dynamic_cast' operator can still be used for casts 3412 that do not require run-time type information, i.e. casts to 'void 3413 *' or to unambiguous base classes. 3414 3415 Mixing code compiled with '-frtti' with that compiled with 3416 '-fno-rtti' may not work. For example, programs may fail to link 3417 if a class compiled with '-fno-rtti' is used as a base for a class 3418 compiled with '-frtti'. 3419 3420'-fsized-deallocation' 3421 Enable the built-in global declarations 3422 void operator delete (void *, std::size_t) noexcept; 3423 void operator delete[] (void *, std::size_t) noexcept; 3424 as introduced in C++14. This is useful for user-defined 3425 replacement deallocation functions that, for example, use the size 3426 of the object to make deallocation faster. Enabled by default 3427 under '-std=c++14' and above. The flag '-Wsized-deallocation' 3428 warns about places that might want to add a definition. 3429 3430'-fstrict-enums' 3431 Allow the compiler to optimize using the assumption that a value of 3432 enumerated type can only be one of the values of the enumeration 3433 (as defined in the C++ standard; basically, a value that can be 3434 represented in the minimum number of bits needed to represent all 3435 the enumerators). This assumption may not be valid if the program 3436 uses a cast to convert an arbitrary integer value to the enumerated 3437 type. 3438 3439'-fstrong-eval-order' 3440 Evaluate member access, array subscripting, and shift expressions 3441 in left-to-right order, and evaluate assignment in right-to-left 3442 order, as adopted for C++17. Enabled by default with '-std=c++17'. 3443 '-fstrong-eval-order=some' enables just the ordering of member 3444 access and shift expressions, and is the default without 3445 '-std=c++17'. 3446 3447'-ftemplate-backtrace-limit=N' 3448 Set the maximum number of template instantiation notes for a single 3449 warning or error to N. The default value is 10. 3450 3451'-ftemplate-depth=N' 3452 Set the maximum instantiation depth for template classes to N. A 3453 limit on the template instantiation depth is needed to detect 3454 endless recursions during template class instantiation. ANSI/ISO 3455 C++ conforming programs must not rely on a maximum depth greater 3456 than 17 (changed to 1024 in C++11). The default value is 900, as 3457 the compiler can run out of stack space before hitting 1024 in some 3458 situations. 3459 3460'-fno-threadsafe-statics' 3461 Do not emit the extra code to use the routines specified in the C++ 3462 ABI for thread-safe initialization of local statics. You can use 3463 this option to reduce code size slightly in code that doesn't need 3464 to be thread-safe. 3465 3466'-fuse-cxa-atexit' 3467 Register destructors for objects with static storage duration with 3468 the '__cxa_atexit' function rather than the 'atexit' function. 3469 This option is required for fully standards-compliant handling of 3470 static destructors, but only works if your C library supports 3471 '__cxa_atexit'. 3472 3473'-fno-use-cxa-get-exception-ptr' 3474 Don't use the '__cxa_get_exception_ptr' runtime routine. This 3475 causes 'std::uncaught_exception' to be incorrect, but is necessary 3476 if the runtime routine is not available. 3477 3478'-fvisibility-inlines-hidden' 3479 This switch declares that the user does not attempt to compare 3480 pointers to inline functions or methods where the addresses of the 3481 two functions are taken in different shared objects. 3482 3483 The effect of this is that GCC may, effectively, mark inline 3484 methods with '__attribute__ ((visibility ("hidden")))' so that they 3485 do not appear in the export table of a DSO and do not require a PLT 3486 indirection when used within the DSO. Enabling this option can 3487 have a dramatic effect on load and link times of a DSO as it 3488 massively reduces the size of the dynamic export table when the 3489 library makes heavy use of templates. 3490 3491 The behavior of this switch is not quite the same as marking the 3492 methods as hidden directly, because it does not affect static 3493 variables local to the function or cause the compiler to deduce 3494 that the function is defined in only one shared object. 3495 3496 You may mark a method as having a visibility explicitly to negate 3497 the effect of the switch for that method. For example, if you do 3498 want to compare pointers to a particular inline method, you might 3499 mark it as having default visibility. Marking the enclosing class 3500 with explicit visibility has no effect. 3501 3502 Explicitly instantiated inline methods are unaffected by this 3503 option as their linkage might otherwise cross a shared library 3504 boundary. *Note Template Instantiation::. 3505 3506'-fvisibility-ms-compat' 3507 This flag attempts to use visibility settings to make GCC's C++ 3508 linkage model compatible with that of Microsoft Visual Studio. 3509 3510 The flag makes these changes to GCC's linkage model: 3511 3512 1. It sets the default visibility to 'hidden', like 3513 '-fvisibility=hidden'. 3514 3515 2. Types, but not their members, are not hidden by default. 3516 3517 3. The One Definition Rule is relaxed for types without explicit 3518 visibility specifications that are defined in more than one 3519 shared object: those declarations are permitted if they are 3520 permitted when this option is not used. 3521 3522 In new code it is better to use '-fvisibility=hidden' and export 3523 those classes that are intended to be externally visible. 3524 Unfortunately it is possible for code to rely, perhaps 3525 accidentally, on the Visual Studio behavior. 3526 3527 Among the consequences of these changes are that static data 3528 members of the same type with the same name but defined in 3529 different shared objects are different, so changing one does not 3530 change the other; and that pointers to function members defined in 3531 different shared objects may not compare equal. When this flag is 3532 given, it is a violation of the ODR to define types with the same 3533 name differently. 3534 3535'-fno-weak' 3536 Do not use weak symbol support, even if it is provided by the 3537 linker. By default, G++ uses weak symbols if they are available. 3538 This option exists only for testing, and should not be used by 3539 end-users; it results in inferior code and has no benefits. This 3540 option may be removed in a future release of G++. 3541 3542'-fext-numeric-literals (C++ and Objective-C++ only)' 3543 Accept imaginary, fixed-point, or machine-defined literal number 3544 suffixes as GNU extensions. When this option is turned off these 3545 suffixes are treated as C++11 user-defined literal numeric 3546 suffixes. This is on by default for all pre-C++11 dialects and all 3547 GNU dialects: '-std=c++98', '-std=gnu++98', '-std=gnu++11', 3548 '-std=gnu++14'. This option is off by default for ISO C++11 3549 onwards ('-std=c++11', ...). 3550 3551'-nostdinc++' 3552 Do not search for header files in the standard directories specific 3553 to C++, but do still search the other standard directories. (This 3554 option is used when building the C++ library.) 3555 3556'-flang-info-include-translate' 3557'-flang-info-include-translate-not' 3558'-flang-info-include-translate=HEADER' 3559 Inform of include translation events. The first will note accepted 3560 include translations, the second will note declined include 3561 translations. The HEADER form will inform of include translations 3562 relating to that specific header. If HEADER is of the form 3563 '"user"' or '<system>' it will be resolved to a specific user or 3564 system header using the include path. 3565 3566'-flang-info-module-cmi' 3567'-flang-info-module-cmi=MODULE' 3568 Inform of Compiled Module Interface pathnames. The first will note 3569 all read CMI pathnames. The MODULE form will not reading a 3570 specific module's CMI. MODULE may be a named module or a 3571 header-unit (the latter indicated by either being a pathname 3572 containing directory separators or enclosed in '<>' or '""'). 3573 3574'-stdlib=LIBSTDC++,LIBC++' 3575 When G++ is configured to support this option, it allows 3576 specification of alternate C++ runtime libraries. Two options are 3577 available: LIBSTDC++ (the default, native C++ runtime for G++) and 3578 LIBC++ which is the C++ runtime installed on some operating systems 3579 (e.g. Darwin versions from Darwin11 onwards). The option switches 3580 G++ to use the headers from the specified library and to emit 3581 '-lstdc++' or '-lc++' respectively, when a C++ runtime is required 3582 for linking. 3583 3584 In addition, these warning options have meanings only for C++ programs: 3585 3586'-Wabi-tag (C++ and Objective-C++ only)' 3587 Warn when a type with an ABI tag is used in a context that does not 3588 have that ABI tag. See *note C++ Attributes:: for more information 3589 about ABI tags. 3590 3591'-Wcomma-subscript (C++ and Objective-C++ only)' 3592 Warn about uses of a comma expression within a subscripting 3593 expression. This usage was deprecated in C++20 and is going to be 3594 removed in C++23. However, a comma expression wrapped in '( )' is 3595 not deprecated. Example: 3596 3597 void f(int *a, int b, int c) { 3598 a[b,c]; // deprecated in C++20, invalid in C++23 3599 a[(b,c)]; // OK 3600 } 3601 3602 In C++23 it is valid to have comma separated expressions in a 3603 subscript when an overloaded subscript operator is found and 3604 supports the right number and types of arguments. G++ will accept 3605 the formerly valid syntax for code that is not valid in C++23 but 3606 used to be valid but deprecated in C++20 with a pedantic warning 3607 that can be disabled with '-Wno-comma-subscript'. 3608 3609 Enabled by default with '-std=c++20' unless '-Wno-deprecated', and 3610 with '-std=c++23' regardless of '-Wno-deprecated'. 3611 3612'-Wctad-maybe-unsupported (C++ and Objective-C++ only)' 3613 Warn when performing class template argument deduction (CTAD) on a 3614 type with no explicitly written deduction guides. This warning 3615 will point out cases where CTAD succeeded only because the compiler 3616 synthesized the implicit deduction guides, which might not be what 3617 the programmer intended. Certain style guides allow CTAD only on 3618 types that specifically "opt-in"; i.e., on types that are designed 3619 to support CTAD. This warning can be suppressed with the following 3620 pattern: 3621 3622 struct allow_ctad_t; // any name works 3623 template <typename T> struct S { 3624 S(T) { } 3625 }; 3626 S(allow_ctad_t) -> S<void>; // guide with incomplete parameter type will never be considered 3627 3628'-Wctor-dtor-privacy (C++ and Objective-C++ only)' 3629 Warn when a class seems unusable because all the constructors or 3630 destructors in that class are private, and it has neither friends 3631 nor public static member functions. Also warn if there are no 3632 non-private methods, and there's at least one private member 3633 function that isn't a constructor or destructor. 3634 3635'-Wdelete-non-virtual-dtor (C++ and Objective-C++ only)' 3636 Warn when 'delete' is used to destroy an instance of a class that 3637 has virtual functions and non-virtual destructor. It is unsafe to 3638 delete an instance of a derived class through a pointer to a base 3639 class if the base class does not have a virtual destructor. This 3640 warning is enabled by '-Wall'. 3641 3642'-Wdeprecated-copy (C++ and Objective-C++ only)' 3643 Warn that the implicit declaration of a copy constructor or copy 3644 assignment operator is deprecated if the class has a user-provided 3645 copy constructor or copy assignment operator, in C++11 and up. 3646 This warning is enabled by '-Wextra'. With 3647 '-Wdeprecated-copy-dtor', also deprecate if the class has a 3648 user-provided destructor. 3649 3650'-Wno-deprecated-enum-enum-conversion (C++ and Objective-C++ only)' 3651 Disable the warning about the case when the usual arithmetic 3652 conversions are applied on operands where one is of enumeration 3653 type and the other is of a different enumeration type. This 3654 conversion was deprecated in C++20. For example: 3655 3656 enum E1 { e }; 3657 enum E2 { f }; 3658 int k = f - e; 3659 3660 '-Wdeprecated-enum-enum-conversion' is enabled by default with 3661 '-std=c++20'. In pre-C++20 dialects, this warning can be enabled 3662 by '-Wenum-conversion'. 3663 3664'-Wno-deprecated-enum-float-conversion (C++ and Objective-C++ only)' 3665 Disable the warning about the case when the usual arithmetic 3666 conversions are applied on operands where one is of enumeration 3667 type and the other is of a floating-point type. This conversion 3668 was deprecated in C++20. For example: 3669 3670 enum E1 { e }; 3671 enum E2 { f }; 3672 bool b = e <= 3.7; 3673 3674 '-Wdeprecated-enum-float-conversion' is enabled by default with 3675 '-std=c++20'. In pre-C++20 dialects, this warning can be enabled 3676 by '-Wenum-conversion'. 3677 3678'-Wno-init-list-lifetime (C++ and Objective-C++ only)' 3679 Do not warn about uses of 'std::initializer_list' that are likely 3680 to result in dangling pointers. Since the underlying array for an 3681 'initializer_list' is handled like a normal C++ temporary object, 3682 it is easy to inadvertently keep a pointer to the array past the 3683 end of the array's lifetime. For example: 3684 3685 * If a function returns a temporary 'initializer_list', or a 3686 local 'initializer_list' variable, the array's lifetime ends 3687 at the end of the return statement, so the value returned has 3688 a dangling pointer. 3689 3690 * If a new-expression creates an 'initializer_list', the array 3691 only lives until the end of the enclosing full-expression, so 3692 the 'initializer_list' in the heap has a dangling pointer. 3693 3694 * When an 'initializer_list' variable is assigned from a 3695 brace-enclosed initializer list, the temporary array created 3696 for the right side of the assignment only lives until the end 3697 of the full-expression, so at the next statement the 3698 'initializer_list' variable has a dangling pointer. 3699 3700 // li's initial underlying array lives as long as li 3701 std::initializer_list<int> li = { 1,2,3 }; 3702 // assignment changes li to point to a temporary array 3703 li = { 4, 5 }; 3704 // now the temporary is gone and li has a dangling pointer 3705 int i = li.begin()[0] // undefined behavior 3706 3707 * When a list constructor stores the 'begin' pointer from the 3708 'initializer_list' argument, this doesn't extend the lifetime 3709 of the array, so if a class variable is constructed from a 3710 temporary 'initializer_list', the pointer is left dangling by 3711 the end of the variable declaration statement. 3712 3713'-Winvalid-imported-macros' 3714 Verify all imported macro definitions are valid at the end of 3715 compilation. This is not enabled by default, as it requires 3716 additional processing to determine. It may be useful when 3717 preparing sets of header-units to ensure consistent macros. 3718 3719'-Wno-literal-suffix (C++ and Objective-C++ only)' 3720 Do not warn when a string or character literal is followed by a 3721 ud-suffix which does not begin with an underscore. As a conforming 3722 extension, GCC treats such suffixes as separate preprocessing 3723 tokens in order to maintain backwards compatibility with code that 3724 uses formatting macros from '<inttypes.h>'. For example: 3725 3726 #define __STDC_FORMAT_MACROS 3727 #include <inttypes.h> 3728 #include <stdio.h> 3729 3730 int main() { 3731 int64_t i64 = 123; 3732 printf("My int64: %" PRId64"\n", i64); 3733 } 3734 3735 In this case, 'PRId64' is treated as a separate preprocessing 3736 token. 3737 3738 This option also controls warnings when a user-defined literal 3739 operator is declared with a literal suffix identifier that doesn't 3740 begin with an underscore. Literal suffix identifiers that don't 3741 begin with an underscore are reserved for future standardization. 3742 3743 These warnings are enabled by default. 3744 3745'-Wno-narrowing (C++ and Objective-C++ only)' 3746 For C++11 and later standards, narrowing conversions are diagnosed 3747 by default, as required by the standard. A narrowing conversion 3748 from a constant produces an error, and a narrowing conversion from 3749 a non-constant produces a warning, but '-Wno-narrowing' suppresses 3750 the diagnostic. Note that this does not affect the meaning of 3751 well-formed code; narrowing conversions are still considered 3752 ill-formed in SFINAE contexts. 3753 3754 With '-Wnarrowing' in C++98, warn when a narrowing conversion 3755 prohibited by C++11 occurs within '{ }', e.g. 3756 3757 int i = { 2.2 }; // error: narrowing from double to int 3758 3759 This flag is included in '-Wall' and '-Wc++11-compat'. 3760 3761'-Wnoexcept (C++ and Objective-C++ only)' 3762 Warn when a noexcept-expression evaluates to false because of a 3763 call to a function that does not have a non-throwing exception 3764 specification (i.e. 'throw()' or 'noexcept') but is known by the 3765 compiler to never throw an exception. 3766 3767'-Wnoexcept-type (C++ and Objective-C++ only)' 3768 Warn if the C++17 feature making 'noexcept' part of a function type 3769 changes the mangled name of a symbol relative to C++14. Enabled by 3770 '-Wabi' and '-Wc++17-compat'. 3771 3772 As an example: 3773 3774 template <class T> void f(T t) { t(); }; 3775 void g() noexcept; 3776 void h() { f(g); } 3777 3778 In C++14, 'f' calls 'f<void(*)()>', but in C++17 it calls 3779 'f<void(*)()noexcept>'. 3780 3781'-Wclass-memaccess (C++ and Objective-C++ only)' 3782 Warn when the destination of a call to a raw memory function such 3783 as 'memset' or 'memcpy' is an object of class type, and when 3784 writing into such an object might bypass the class non-trivial or 3785 deleted constructor or copy assignment, violate const-correctness 3786 or encapsulation, or corrupt virtual table pointers. Modifying the 3787 representation of such objects may violate invariants maintained by 3788 member functions of the class. For example, the call to 'memset' 3789 below is undefined because it modifies a non-trivial class object 3790 and is, therefore, diagnosed. The safe way to either initialize or 3791 clear the storage of objects of such types is by using the 3792 appropriate constructor or assignment operator, if one is 3793 available. 3794 std::string str = "abc"; 3795 memset (&str, 0, sizeof str); 3796 The '-Wclass-memaccess' option is enabled by '-Wall'. Explicitly 3797 casting the pointer to the class object to 'void *' or to a type 3798 that can be safely accessed by the raw memory function suppresses 3799 the warning. 3800 3801'-Wnon-virtual-dtor (C++ and Objective-C++ only)' 3802 Warn when a class has virtual functions and an accessible 3803 non-virtual destructor itself or in an accessible polymorphic base 3804 class, in which case it is possible but unsafe to delete an 3805 instance of a derived class through a pointer to the class itself 3806 or base class. This warning is automatically enabled if '-Weffc++' 3807 is specified. 3808 3809'-Wregister (C++ and Objective-C++ only)' 3810 Warn on uses of the 'register' storage class specifier, except when 3811 it is part of the GNU *note Explicit Register Variables:: 3812 extension. The use of the 'register' keyword as storage class 3813 specifier has been deprecated in C++11 and removed in C++17. 3814 Enabled by default with '-std=c++17'. 3815 3816'-Wreorder (C++ and Objective-C++ only)' 3817 Warn when the order of member initializers given in the code does 3818 not match the order in which they must be executed. For instance: 3819 3820 struct A { 3821 int i; 3822 int j; 3823 A(): j (0), i (1) { } 3824 }; 3825 3826 The compiler rearranges the member initializers for 'i' and 'j' to 3827 match the declaration order of the members, emitting a warning to 3828 that effect. This warning is enabled by '-Wall'. 3829 3830'-Wno-pessimizing-move (C++ and Objective-C++ only)' 3831 This warning warns when a call to 'std::move' prevents copy 3832 elision. A typical scenario when copy elision can occur is when 3833 returning in a function with a class return type, when the 3834 expression being returned is the name of a non-volatile automatic 3835 object, and is not a function parameter, and has the same type as 3836 the function return type. 3837 3838 struct T { 3839 ... 3840 }; 3841 T fn() 3842 { 3843 T t; 3844 ... 3845 return std::move (t); 3846 } 3847 3848 But in this example, the 'std::move' call prevents copy elision. 3849 3850 This warning is enabled by '-Wall'. 3851 3852'-Wno-redundant-move (C++ and Objective-C++ only)' 3853 This warning warns about redundant calls to 'std::move'; that is, 3854 when a move operation would have been performed even without the 3855 'std::move' call. This happens because the compiler is forced to 3856 treat the object as if it were an rvalue in certain situations such 3857 as returning a local variable, where copy elision isn't applicable. 3858 Consider: 3859 3860 struct T { 3861 ... 3862 }; 3863 T fn(T t) 3864 { 3865 ... 3866 return std::move (t); 3867 } 3868 3869 Here, the 'std::move' call is redundant. Because G++ implements 3870 Core Issue 1579, another example is: 3871 3872 struct T { // convertible to U 3873 ... 3874 }; 3875 struct U { 3876 ... 3877 }; 3878 U fn() 3879 { 3880 T t; 3881 ... 3882 return std::move (t); 3883 } 3884 In this example, copy elision isn't applicable because the type of 3885 the expression being returned and the function return type differ, 3886 yet G++ treats the return value as if it were designated by an 3887 rvalue. 3888 3889 This warning is enabled by '-Wextra'. 3890 3891'-Wrange-loop-construct (C++ and Objective-C++ only)' 3892 This warning warns when a C++ range-based for-loop is creating an 3893 unnecessary copy. This can happen when the range declaration is 3894 not a reference, but probably should be. For example: 3895 3896 struct S { char arr[128]; }; 3897 void fn () { 3898 S arr[5]; 3899 for (const auto x : arr) { ... } 3900 } 3901 3902 It does not warn when the type being copied is a trivially-copyable 3903 type whose size is less than 64 bytes. 3904 3905 This warning also warns when a loop variable in a range-based 3906 for-loop is initialized with a value of a different type resulting 3907 in a copy. For example: 3908 3909 void fn() { 3910 int arr[10]; 3911 for (const double &x : arr) { ... } 3912 } 3913 3914 In the example above, in every iteration of the loop a temporary 3915 value of type 'double' is created and destroyed, to which the 3916 reference 'const double &' is bound. 3917 3918 This warning is enabled by '-Wall'. 3919 3920'-Wredundant-tags (C++ and Objective-C++ only)' 3921 Warn about redundant class-key and enum-key in references to class 3922 types and enumerated types in contexts where the key can be 3923 eliminated without causing an ambiguity. For example: 3924 3925 struct foo; 3926 struct foo *p; // warn that keyword struct can be eliminated 3927 3928 On the other hand, in this example there is no warning: 3929 3930 struct foo; 3931 void foo (); // "hides" struct foo 3932 void bar (struct foo&); // no warning, keyword struct is necessary 3933 3934'-Wno-subobject-linkage (C++ and Objective-C++ only)' 3935 Do not warn if a class type has a base or a field whose type uses 3936 the anonymous namespace or depends on a type with no linkage. If a 3937 type A depends on a type B with no or internal linkage, defining it 3938 in multiple translation units would be an ODR violation because the 3939 meaning of B is different in each translation unit. If A only 3940 appears in a single translation unit, the best way to silence the 3941 warning is to give it internal linkage by putting it in an 3942 anonymous namespace as well. The compiler doesn't give this 3943 warning for types defined in the main .C file, as those are 3944 unlikely to have multiple definitions. '-Wsubobject-linkage' is 3945 enabled by default. 3946 3947'-Weffc++ (C++ and Objective-C++ only)' 3948 Warn about violations of the following style guidelines from Scott 3949 Meyers' 'Effective C++' series of books: 3950 3951 * Define a copy constructor and an assignment operator for 3952 classes with dynamically-allocated memory. 3953 3954 * Prefer initialization to assignment in constructors. 3955 3956 * Have 'operator=' return a reference to '*this'. 3957 3958 * Don't try to return a reference when you must return an 3959 object. 3960 3961 * Distinguish between prefix and postfix forms of increment and 3962 decrement operators. 3963 3964 * Never overload '&&', '||', or ','. 3965 3966 This option also enables '-Wnon-virtual-dtor', which is also one of 3967 the effective C++ recommendations. However, the check is extended 3968 to warn about the lack of virtual destructor in accessible 3969 non-polymorphic bases classes too. 3970 3971 When selecting this option, be aware that the standard library 3972 headers do not obey all of these guidelines; use 'grep -v' to 3973 filter out those warnings. 3974 3975'-Wno-exceptions (C++ and Objective-C++ only)' 3976 Disable the warning about the case when an exception handler is 3977 shadowed by another handler, which can point out a wrong ordering 3978 of exception handlers. 3979 3980'-Wstrict-null-sentinel (C++ and Objective-C++ only)' 3981 Warn about the use of an uncasted 'NULL' as sentinel. When 3982 compiling only with GCC this is a valid sentinel, as 'NULL' is 3983 defined to '__null'. Although it is a null pointer constant rather 3984 than a null pointer, it is guaranteed to be of the same size as a 3985 pointer. But this use is not portable across different compilers. 3986 3987'-Wno-non-template-friend (C++ and Objective-C++ only)' 3988 Disable warnings when non-template friend functions are declared 3989 within a template. In very old versions of GCC that predate 3990 implementation of the ISO standard, declarations such as 'friend 3991 int foo(int)', where the name of the friend is an unqualified-id, 3992 could be interpreted as a particular specialization of a template 3993 function; the warning exists to diagnose compatibility problems, 3994 and is enabled by default. 3995 3996'-Wold-style-cast (C++ and Objective-C++ only)' 3997 Warn if an old-style (C-style) cast to a non-void type is used 3998 within a C++ program. The new-style casts ('dynamic_cast', 3999 'static_cast', 'reinterpret_cast', and 'const_cast') are less 4000 vulnerable to unintended effects and much easier to search for. 4001 4002'-Woverloaded-virtual (C++ and Objective-C++ only)' 4003 Warn when a function declaration hides virtual functions from a 4004 base class. For example, in: 4005 4006 struct A { 4007 virtual void f(); 4008 }; 4009 4010 struct B: public A { 4011 void f(int); 4012 }; 4013 4014 the 'A' class version of 'f' is hidden in 'B', and code like: 4015 4016 B* b; 4017 b->f(); 4018 4019 fails to compile. 4020 4021'-Wno-pmf-conversions (C++ and Objective-C++ only)' 4022 Disable the diagnostic for converting a bound pointer to member 4023 function to a plain pointer. 4024 4025'-Wsign-promo (C++ and Objective-C++ only)' 4026 Warn when overload resolution chooses a promotion from unsigned or 4027 enumerated type to a signed type, over a conversion to an unsigned 4028 type of the same size. Previous versions of G++ tried to preserve 4029 unsignedness, but the standard mandates the current behavior. 4030 4031'-Wtemplates (C++ and Objective-C++ only)' 4032 Warn when a primary template declaration is encountered. Some 4033 coding rules disallow templates, and this may be used to enforce 4034 that rule. The warning is inactive inside a system header file, 4035 such as the STL, so one can still use the STL. One may also 4036 instantiate or specialize templates. 4037 4038'-Wmismatched-new-delete (C++ and Objective-C++ only)' 4039 Warn for mismatches between calls to 'operator new' or 'operator 4040 delete' and the corresponding call to the allocation or 4041 deallocation function. This includes invocations of C++ 'operator 4042 delete' with pointers returned from either mismatched forms of 4043 'operator new', or from other functions that allocate objects for 4044 which the 'operator delete' isn't a suitable deallocator, as well 4045 as calls to other deallocation functions with pointers returned 4046 from 'operator new' for which the deallocation function isn't 4047 suitable. 4048 4049 For example, the 'delete' expression in the function below is 4050 diagnosed because it doesn't match the array form of the 'new' 4051 expression the pointer argument was returned from. Similarly, the 4052 call to 'free' is also diagnosed. 4053 4054 void f () 4055 { 4056 int *a = new int[n]; 4057 delete a; // warning: mismatch in array forms of expressions 4058 4059 char *p = new char[n]; 4060 free (p); // warning: mismatch between new and free 4061 } 4062 4063 The related option '-Wmismatched-dealloc' diagnoses mismatches 4064 involving allocation and deallocation functions other than 4065 'operator new' and 'operator delete'. 4066 4067 '-Wmismatched-new-delete' is included in '-Wall'. 4068 4069'-Wmismatched-tags (C++ and Objective-C++ only)' 4070 Warn for declarations of structs, classes, and class templates and 4071 their specializations with a class-key that does not match either 4072 the definition or the first declaration if no definition is 4073 provided. 4074 4075 For example, the declaration of 'struct Object' in the argument 4076 list of 'draw' triggers the warning. To avoid it, either remove 4077 the redundant class-key 'struct' or replace it with 'class' to 4078 match its definition. 4079 class Object { 4080 public: 4081 virtual ~Object () = 0; 4082 }; 4083 void draw (struct Object*); 4084 4085 It is not wrong to declare a class with the class-key 'struct' as 4086 the example above shows. The '-Wmismatched-tags' option is 4087 intended to help achieve a consistent style of class declarations. 4088 In code that is intended to be portable to Windows-based compilers 4089 the warning helps prevent unresolved references due to the 4090 difference in the mangling of symbols declared with different 4091 class-keys. The option can be used either on its own or in 4092 conjunction with '-Wredundant-tags'. 4093 4094'-Wmultiple-inheritance (C++ and Objective-C++ only)' 4095 Warn when a class is defined with multiple direct base classes. 4096 Some coding rules disallow multiple inheritance, and this may be 4097 used to enforce that rule. The warning is inactive inside a system 4098 header file, such as the STL, so one can still use the STL. One may 4099 also define classes that indirectly use multiple inheritance. 4100 4101'-Wvirtual-inheritance' 4102 Warn when a class is defined with a virtual direct base class. 4103 Some coding rules disallow multiple inheritance, and this may be 4104 used to enforce that rule. The warning is inactive inside a system 4105 header file, such as the STL, so one can still use the STL. One may 4106 also define classes that indirectly use virtual inheritance. 4107 4108'-Wno-virtual-move-assign' 4109 Suppress warnings about inheriting from a virtual base with a 4110 non-trivial C++11 move assignment operator. This is dangerous 4111 because if the virtual base is reachable along more than one path, 4112 it is moved multiple times, which can mean both objects end up in 4113 the moved-from state. If the move assignment operator is written 4114 to avoid moving from a moved-from object, this warning can be 4115 disabled. 4116 4117'-Wnamespaces' 4118 Warn when a namespace definition is opened. Some coding rules 4119 disallow namespaces, and this may be used to enforce that rule. 4120 The warning is inactive inside a system header file, such as the 4121 STL, so one can still use the STL. One may also use using 4122 directives and qualified names. 4123 4124'-Wno-terminate (C++ and Objective-C++ only)' 4125 Disable the warning about a throw-expression that will immediately 4126 result in a call to 'terminate'. 4127 4128'-Wno-vexing-parse (C++ and Objective-C++ only)' 4129 Warn about the most vexing parse syntactic ambiguity. This warns 4130 about the cases when a declaration looks like a variable 4131 definition, but the C++ language requires it to be interpreted as a 4132 function declaration. For instance: 4133 4134 void f(double a) { 4135 int i(); // extern int i (void); 4136 int n(int(a)); // extern int n (int); 4137 } 4138 4139 Another example: 4140 4141 struct S { S(int); }; 4142 void f(double a) { 4143 S x(int(a)); // extern struct S x (int); 4144 S y(int()); // extern struct S y (int (*) (void)); 4145 S z(); // extern struct S z (void); 4146 } 4147 4148 The warning will suggest options how to deal with such an 4149 ambiguity; e.g., it can suggest removing the parentheses or using 4150 braces instead. 4151 4152 This warning is enabled by default. 4153 4154'-Wno-class-conversion (C++ and Objective-C++ only)' 4155 Do not warn when a conversion function converts an object to the 4156 same type, to a base class of that type, or to void; such a 4157 conversion function will never be called. 4158 4159'-Wvolatile (C++ and Objective-C++ only)' 4160 Warn about deprecated uses of the 'volatile' qualifier. This 4161 includes postfix and prefix '++' and '--' expressions of 4162 'volatile'-qualified types, using simple assignments where the left 4163 operand is a 'volatile'-qualified non-class type for their value, 4164 compound assignments where the left operand is a 4165 'volatile'-qualified non-class type, 'volatile'-qualified function 4166 return type, 'volatile'-qualified parameter type, and structured 4167 bindings of a 'volatile'-qualified type. This usage was deprecated 4168 in C++20. 4169 4170 Enabled by default with '-std=c++20'. 4171 4172'-Wzero-as-null-pointer-constant (C++ and Objective-C++ only)' 4173 Warn when a literal '0' is used as null pointer constant. This can 4174 be useful to facilitate the conversion to 'nullptr' in C++11. 4175 4176'-Waligned-new' 4177 Warn about a new-expression of a type that requires greater 4178 alignment than the 'alignof(std::max_align_t)' but uses an 4179 allocation function without an explicit alignment parameter. This 4180 option is enabled by '-Wall'. 4181 4182 Normally this only warns about global allocation functions, but 4183 '-Waligned-new=all' also warns about class member allocation 4184 functions. 4185 4186'-Wno-placement-new' 4187'-Wplacement-new=N' 4188 Warn about placement new expressions with undefined behavior, such 4189 as constructing an object in a buffer that is smaller than the type 4190 of the object. For example, the placement new expression below is 4191 diagnosed because it attempts to construct an array of 64 integers 4192 in a buffer only 64 bytes large. 4193 char buf [64]; 4194 new (buf) int[64]; 4195 This warning is enabled by default. 4196 4197 '-Wplacement-new=1' 4198 This is the default warning level of '-Wplacement-new'. At 4199 this level the warning is not issued for some strictly 4200 undefined constructs that GCC allows as extensions for 4201 compatibility with legacy code. For example, the following 4202 'new' expression is not diagnosed at this level even though it 4203 has undefined behavior according to the C++ standard because 4204 it writes past the end of the one-element array. 4205 struct S { int n, a[1]; }; 4206 S *s = (S *)malloc (sizeof *s + 31 * sizeof s->a[0]); 4207 new (s->a)int [32](); 4208 4209 '-Wplacement-new=2' 4210 At this level, in addition to diagnosing all the same 4211 constructs as at level 1, a diagnostic is also issued for 4212 placement new expressions that construct an object in the last 4213 member of structure whose type is an array of a single element 4214 and whose size is less than the size of the object being 4215 constructed. While the previous example would be diagnosed, 4216 the following construct makes use of the flexible member array 4217 extension to avoid the warning at level 2. 4218 struct S { int n, a[]; }; 4219 S *s = (S *)malloc (sizeof *s + 32 * sizeof s->a[0]); 4220 new (s->a)int [32](); 4221 4222'-Wcatch-value' 4223'-Wcatch-value=N (C++ and Objective-C++ only)' 4224 Warn about catch handlers that do not catch via reference. With 4225 '-Wcatch-value=1' (or '-Wcatch-value' for short) warn about 4226 polymorphic class types that are caught by value. With 4227 '-Wcatch-value=2' warn about all class types that are caught by 4228 value. With '-Wcatch-value=3' warn about all types that are not 4229 caught by reference. '-Wcatch-value' is enabled by '-Wall'. 4230 4231'-Wconditionally-supported (C++ and Objective-C++ only)' 4232 Warn for conditionally-supported (C++11 [intro.defs]) constructs. 4233 4234'-Wno-delete-incomplete (C++ and Objective-C++ only)' 4235 Do not warn when deleting a pointer to incomplete type, which may 4236 cause undefined behavior at runtime. This warning is enabled by 4237 default. 4238 4239'-Wextra-semi (C++, Objective-C++ only)' 4240 Warn about redundant semicolons after in-class function 4241 definitions. 4242 4243'-Wno-inaccessible-base (C++, Objective-C++ only)' 4244 This option controls warnings when a base class is inaccessible in 4245 a class derived from it due to ambiguity. The warning is enabled 4246 by default. Note that the warning for ambiguous virtual bases is 4247 enabled by the '-Wextra' option. 4248 struct A { int a; }; 4249 4250 struct B : A { }; 4251 4252 struct C : B, A { }; 4253 4254'-Wno-inherited-variadic-ctor' 4255 Suppress warnings about use of C++11 inheriting constructors when 4256 the base class inherited from has a C variadic constructor; the 4257 warning is on by default because the ellipsis is not inherited. 4258 4259'-Wno-invalid-offsetof (C++ and Objective-C++ only)' 4260 Suppress warnings from applying the 'offsetof' macro to a non-POD 4261 type. According to the 2014 ISO C++ standard, applying 'offsetof' 4262 to a non-standard-layout type is undefined. In existing C++ 4263 implementations, however, 'offsetof' typically gives meaningful 4264 results. This flag is for users who are aware that they are 4265 writing nonportable code and who have deliberately chosen to ignore 4266 the warning about it. 4267 4268 The restrictions on 'offsetof' may be relaxed in a future version 4269 of the C++ standard. 4270 4271'-Wsized-deallocation (C++ and Objective-C++ only)' 4272 Warn about a definition of an unsized deallocation function 4273 void operator delete (void *) noexcept; 4274 void operator delete[] (void *) noexcept; 4275 without a definition of the corresponding sized deallocation 4276 function 4277 void operator delete (void *, std::size_t) noexcept; 4278 void operator delete[] (void *, std::size_t) noexcept; 4279 or vice versa. Enabled by '-Wextra' along with 4280 '-fsized-deallocation'. 4281 4282'-Wsuggest-final-types' 4283 Warn about types with virtual methods where code quality would be 4284 improved if the type were declared with the C++11 'final' 4285 specifier, or, if possible, declared in an anonymous namespace. 4286 This allows GCC to more aggressively devirtualize the polymorphic 4287 calls. This warning is more effective with link-time optimization, 4288 where the information about the class hierarchy graph is more 4289 complete. 4290 4291'-Wsuggest-final-methods' 4292 Warn about virtual methods where code quality would be improved if 4293 the method were declared with the C++11 'final' specifier, or, if 4294 possible, its type were declared in an anonymous namespace or with 4295 the 'final' specifier. This warning is more effective with 4296 link-time optimization, where the information about the class 4297 hierarchy graph is more complete. It is recommended to first 4298 consider suggestions of '-Wsuggest-final-types' and then rebuild 4299 with new annotations. 4300 4301'-Wsuggest-override' 4302 Warn about overriding virtual functions that are not marked with 4303 the 'override' keyword. 4304 4305'-Wuse-after-free' 4306'-Wuse-after-free=N' 4307 Warn about uses of pointers to dynamically allocated objects that 4308 have been rendered indeterminate by a call to a deallocation 4309 function. The warning is enabled at all optimization levels but 4310 may yield different results with optimization than without. 4311 4312 '-Wuse-after-free=1' 4313 At level 1 the warning attempts to diagnose only unconditional 4314 uses of pointers made indeterminate by a deallocation call or 4315 a successful call to 'realloc', regardless of whether or not 4316 the call resulted in an actual reallocatio of memory. This 4317 includes double-'free' calls as well as uses in arithmetic and 4318 relational expressions. Although undefined, uses of 4319 indeterminate pointers in equality (or inequality) expressions 4320 are not diagnosed at this level. 4321 '-Wuse-after-free=2' 4322 At level 2, in addition to unconditional uses, the warning 4323 also diagnoses conditional uses of pointers made indeterminate 4324 by a deallocation call. As at level 2, uses in equality (or 4325 inequality) expressions are not diagnosed. For example, the 4326 second call to 'free' in the following function is diagnosed 4327 at this level: 4328 struct A { int refcount; void *data; }; 4329 4330 void release (struct A *p) 4331 { 4332 int refcount = --p->refcount; 4333 free (p); 4334 if (refcount == 0) 4335 free (p->data); // warning: p may be used after free 4336 } 4337 '-Wuse-after-free=3' 4338 At level 3, the warning also diagnoses uses of indeterminate 4339 pointers in equality expressions. All uses of indeterminate 4340 pointers are undefined but equality tests sometimes appear 4341 after calls to 'realloc' as an attempt to determine whether 4342 the call resulted in relocating the object to a different 4343 address. They are diagnosed at a separate level to aid legacy 4344 code gradually transition to safe alternatives. For example, 4345 the equality test in the function below is diagnosed at this 4346 level: 4347 void adjust_pointers (int**, int); 4348 4349 void grow (int **p, int n) 4350 { 4351 int **q = (int**)realloc (p, n *= 2); 4352 if (q == p) 4353 return; 4354 adjust_pointers ((int**)q, n); 4355 } 4356 To avoid the warning at this level, store offsets into 4357 allocated memory instead of pointers. This approach obviates 4358 needing to adjust the stored pointers after reallocation. 4359 4360 '-Wuse-after-free=2' is included in '-Wall'. 4361 4362'-Wuseless-cast (C++ and Objective-C++ only)' 4363 Warn when an expression is casted to its own type. 4364 4365'-Wno-conversion-null (C++ and Objective-C++ only)' 4366 Do not warn for conversions between 'NULL' and non-pointer types. 4367 '-Wconversion-null' is enabled by default. 4368 4369 4370File: gcc.info, Node: Objective-C and Objective-C++ Dialect Options, Next: Diagnostic Message Formatting Options, Prev: C++ Dialect Options, Up: Invoking GCC 4371 43723.6 Options Controlling Objective-C and Objective-C++ Dialects 4373============================================================== 4374 4375(NOTE: This manual does not describe the Objective-C and Objective-C++ 4376languages themselves. *Note Language Standards Supported by GCC: 4377Standards, for references.) 4378 4379 This section describes the command-line options that are only 4380meaningful for Objective-C and Objective-C++ programs. You can also use 4381most of the language-independent GNU compiler options. For example, you 4382might compile a file 'some_class.m' like this: 4383 4384 gcc -g -fgnu-runtime -O -c some_class.m 4385 4386In this example, '-fgnu-runtime' is an option meant only for Objective-C 4387and Objective-C++ programs; you can use the other options with any 4388language supported by GCC. 4389 4390 Note that since Objective-C is an extension of the C language, 4391Objective-C compilations may also use options specific to the C 4392front-end (e.g., '-Wtraditional'). Similarly, Objective-C++ 4393compilations may use C++-specific options (e.g., '-Wabi'). 4394 4395 Here is a list of options that are _only_ for compiling Objective-C and 4396Objective-C++ programs: 4397 4398'-fconstant-string-class=CLASS-NAME' 4399 Use CLASS-NAME as the name of the class to instantiate for each 4400 literal string specified with the syntax '@"..."'. The default 4401 class name is 'NXConstantString' if the GNU runtime is being used, 4402 and 'NSConstantString' if the NeXT runtime is being used (see 4403 below). The '-fconstant-cfstrings' option, if also present, 4404 overrides the '-fconstant-string-class' setting and cause '@"..."' 4405 literals to be laid out as constant CoreFoundation strings. 4406 4407'-fgnu-runtime' 4408 Generate object code compatible with the standard GNU Objective-C 4409 runtime. This is the default for most types of systems. 4410 4411'-fnext-runtime' 4412 Generate output compatible with the NeXT runtime. This is the 4413 default for NeXT-based systems, including Darwin and Mac OS X. The 4414 macro '__NEXT_RUNTIME__' is predefined if (and only if) this option 4415 is used. 4416 4417'-fno-nil-receivers' 4418 Assume that all Objective-C message dispatches ('[receiver 4419 message:arg]') in this translation unit ensure that the receiver is 4420 not 'nil'. This allows for more efficient entry points in the 4421 runtime to be used. This option is only available in conjunction 4422 with the NeXT runtime and ABI version 0 or 1. 4423 4424'-fobjc-abi-version=N' 4425 Use version N of the Objective-C ABI for the selected runtime. 4426 This option is currently supported only for the NeXT runtime. In 4427 that case, Version 0 is the traditional (32-bit) ABI without 4428 support for properties and other Objective-C 2.0 additions. 4429 Version 1 is the traditional (32-bit) ABI with support for 4430 properties and other Objective-C 2.0 additions. Version 2 is the 4431 modern (64-bit) ABI. If nothing is specified, the default is 4432 Version 0 on 32-bit target machines, and Version 2 on 64-bit target 4433 machines. 4434 4435'-fobjc-call-cxx-cdtors' 4436 For each Objective-C class, check if any of its instance variables 4437 is a C++ object with a non-trivial default constructor. If so, 4438 synthesize a special '- (id) .cxx_construct' instance method which 4439 runs non-trivial default constructors on any such instance 4440 variables, in order, and then return 'self'. Similarly, check if 4441 any instance variable is a C++ object with a non-trivial 4442 destructor, and if so, synthesize a special '- (void) 4443 .cxx_destruct' method which runs all such default destructors, in 4444 reverse order. 4445 4446 The '- (id) .cxx_construct' and '- (void) .cxx_destruct' methods 4447 thusly generated only operate on instance variables declared in the 4448 current Objective-C class, and not those inherited from 4449 superclasses. It is the responsibility of the Objective-C runtime 4450 to invoke all such methods in an object's inheritance hierarchy. 4451 The '- (id) .cxx_construct' methods are invoked by the runtime 4452 immediately after a new object instance is allocated; the '- (void) 4453 .cxx_destruct' methods are invoked immediately before the runtime 4454 deallocates an object instance. 4455 4456 As of this writing, only the NeXT runtime on Mac OS X 10.4 and 4457 later has support for invoking the '- (id) .cxx_construct' and '- 4458 (void) .cxx_destruct' methods. 4459 4460'-fobjc-direct-dispatch' 4461 Allow fast jumps to the message dispatcher. On Darwin this is 4462 accomplished via the comm page. 4463 4464'-fobjc-exceptions' 4465 Enable syntactic support for structured exception handling in 4466 Objective-C, similar to what is offered by C++. This option is 4467 required to use the Objective-C keywords '@try', '@throw', 4468 '@catch', '@finally' and '@synchronized'. This option is available 4469 with both the GNU runtime and the NeXT runtime (but not available 4470 in conjunction with the NeXT runtime on Mac OS X 10.2 and earlier). 4471 4472'-fobjc-gc' 4473 Enable garbage collection (GC) in Objective-C and Objective-C++ 4474 programs. This option is only available with the NeXT runtime; the 4475 GNU runtime has a different garbage collection implementation that 4476 does not require special compiler flags. 4477 4478'-fobjc-nilcheck' 4479 For the NeXT runtime with version 2 of the ABI, check for a nil 4480 receiver in method invocations before doing the actual method call. 4481 This is the default and can be disabled using '-fno-objc-nilcheck'. 4482 Class methods and super calls are never checked for nil in this way 4483 no matter what this flag is set to. Currently this flag does 4484 nothing when the GNU runtime, or an older version of the NeXT 4485 runtime ABI, is used. 4486 4487'-fobjc-std=objc1' 4488 Conform to the language syntax of Objective-C 1.0, the language 4489 recognized by GCC 4.0. This only affects the Objective-C additions 4490 to the C/C++ language; it does not affect conformance to C/C++ 4491 standards, which is controlled by the separate C/C++ dialect option 4492 flags. When this option is used with the Objective-C or 4493 Objective-C++ compiler, any Objective-C syntax that is not 4494 recognized by GCC 4.0 is rejected. This is useful if you need to 4495 make sure that your Objective-C code can be compiled with older 4496 versions of GCC. 4497 4498'-freplace-objc-classes' 4499 Emit a special marker instructing 'ld(1)' not to statically link in 4500 the resulting object file, and allow 'dyld(1)' to load it in at run 4501 time instead. This is used in conjunction with the 4502 Fix-and-Continue debugging mode, where the object file in question 4503 may be recompiled and dynamically reloaded in the course of program 4504 execution, without the need to restart the program itself. 4505 Currently, Fix-and-Continue functionality is only available in 4506 conjunction with the NeXT runtime on Mac OS X 10.3 and later. 4507 4508'-fzero-link' 4509 When compiling for the NeXT runtime, the compiler ordinarily 4510 replaces calls to 'objc_getClass("...")' (when the name of the 4511 class is known at compile time) with static class references that 4512 get initialized at load time, which improves run-time performance. 4513 Specifying the '-fzero-link' flag suppresses this behavior and 4514 causes calls to 'objc_getClass("...")' to be retained. This is 4515 useful in Zero-Link debugging mode, since it allows for individual 4516 class implementations to be modified during program execution. The 4517 GNU runtime currently always retains calls to 4518 'objc_get_class("...")' regardless of command-line options. 4519 4520'-fno-local-ivars' 4521 By default instance variables in Objective-C can be accessed as if 4522 they were local variables from within the methods of the class 4523 they're declared in. This can lead to shadowing between instance 4524 variables and other variables declared either locally inside a 4525 class method or globally with the same name. Specifying the 4526 '-fno-local-ivars' flag disables this behavior thus avoiding 4527 variable shadowing issues. 4528 4529'-fivar-visibility=[public|protected|private|package]' 4530 Set the default instance variable visibility to the specified 4531 option so that instance variables declared outside the scope of any 4532 access modifier directives default to the specified visibility. 4533 4534'-gen-decls' 4535 Dump interface declarations for all classes seen in the source file 4536 to a file named 'SOURCENAME.decl'. 4537 4538'-Wassign-intercept (Objective-C and Objective-C++ only)' 4539 Warn whenever an Objective-C assignment is being intercepted by the 4540 garbage collector. 4541 4542'-Wno-property-assign-default (Objective-C and Objective-C++ only)' 4543 Do not warn if a property for an Objective-C object has no assign 4544 semantics specified. 4545 4546'-Wno-protocol (Objective-C and Objective-C++ only)' 4547 If a class is declared to implement a protocol, a warning is issued 4548 for every method in the protocol that is not implemented by the 4549 class. The default behavior is to issue a warning for every method 4550 not explicitly implemented in the class, even if a method 4551 implementation is inherited from the superclass. If you use the 4552 '-Wno-protocol' option, then methods inherited from the superclass 4553 are considered to be implemented, and no warning is issued for 4554 them. 4555 4556'-Wobjc-root-class (Objective-C and Objective-C++ only)' 4557 Warn if a class interface lacks a superclass. Most classes will 4558 inherit from 'NSObject' (or 'Object') for example. When declaring 4559 classes intended to be root classes, the warning can be suppressed 4560 by marking their interfaces with 4561 '__attribute__((objc_root_class))'. 4562 4563'-Wselector (Objective-C and Objective-C++ only)' 4564 Warn if multiple methods of different types for the same selector 4565 are found during compilation. The check is performed on the list 4566 of methods in the final stage of compilation. Additionally, a 4567 check is performed for each selector appearing in a 4568 '@selector(...)' expression, and a corresponding method for that 4569 selector has been found during compilation. Because these checks 4570 scan the method table only at the end of compilation, these 4571 warnings are not produced if the final stage of compilation is not 4572 reached, for example because an error is found during compilation, 4573 or because the '-fsyntax-only' option is being used. 4574 4575'-Wstrict-selector-match (Objective-C and Objective-C++ only)' 4576 Warn if multiple methods with differing argument and/or return 4577 types are found for a given selector when attempting to send a 4578 message using this selector to a receiver of type 'id' or 'Class'. 4579 When this flag is off (which is the default behavior), the compiler 4580 omits such warnings if any differences found are confined to types 4581 that share the same size and alignment. 4582 4583'-Wundeclared-selector (Objective-C and Objective-C++ only)' 4584 Warn if a '@selector(...)' expression referring to an undeclared 4585 selector is found. A selector is considered undeclared if no 4586 method with that name has been declared before the '@selector(...)' 4587 expression, either explicitly in an '@interface' or '@protocol' 4588 declaration, or implicitly in an '@implementation' section. This 4589 option always performs its checks as soon as a '@selector(...)' 4590 expression is found, while '-Wselector' only performs its checks in 4591 the final stage of compilation. This also enforces the coding 4592 style convention that methods and selectors must be declared before 4593 being used. 4594 4595'-print-objc-runtime-info' 4596 Generate C header describing the largest structure that is passed 4597 by value, if any. 4598 4599 4600File: gcc.info, Node: Diagnostic Message Formatting Options, Next: Warning Options, Prev: Objective-C and Objective-C++ Dialect Options, Up: Invoking GCC 4601 46023.7 Options to Control Diagnostic Messages Formatting 4603===================================================== 4604 4605Traditionally, diagnostic messages have been formatted irrespective of 4606the output device's aspect (e.g. its width, ...). You can use the 4607options described below to control the formatting algorithm for 4608diagnostic messages, e.g. how many characters per line, how often source 4609location information should be reported. Note that some language front 4610ends may not honor these options. 4611 4612'-fmessage-length=N' 4613 Try to format error messages so that they fit on lines of about N 4614 characters. If N is zero, then no line-wrapping is done; each 4615 error message appears on a single line. This is the default for 4616 all front ends. 4617 4618 Note - this option also affects the display of the '#error' and 4619 '#warning' pre-processor directives, and the 'deprecated' 4620 function/type/variable attribute. It does not however affect the 4621 'pragma GCC warning' and 'pragma GCC error' pragmas. 4622 4623'-fdiagnostics-plain-output' 4624 This option requests that diagnostic output look as plain as 4625 possible, which may be useful when running 'dejagnu' or other 4626 utilities that need to parse diagnostics output and prefer that it 4627 remain more stable over time. '-fdiagnostics-plain-output' is 4628 currently equivalent to the following options: 4629 -fno-diagnostics-show-caret 4630 -fno-diagnostics-show-line-numbers 4631 -fdiagnostics-color=never 4632 -fdiagnostics-urls=never 4633 -fdiagnostics-path-format=separate-events 4634 In the future, if GCC changes the default appearance of its 4635 diagnostics, the corresponding option to disable the new behavior 4636 will be added to this list. 4637 4638'-fdiagnostics-show-location=once' 4639 Only meaningful in line-wrapping mode. Instructs the diagnostic 4640 messages reporter to emit source location information _once_; that 4641 is, in case the message is too long to fit on a single physical 4642 line and has to be wrapped, the source location won't be emitted 4643 (as prefix) again, over and over, in subsequent continuation lines. 4644 This is the default behavior. 4645 4646'-fdiagnostics-show-location=every-line' 4647 Only meaningful in line-wrapping mode. Instructs the diagnostic 4648 messages reporter to emit the same source location information (as 4649 prefix) for physical lines that result from the process of breaking 4650 a message which is too long to fit on a single line. 4651 4652'-fdiagnostics-color[=WHEN]' 4653'-fno-diagnostics-color' 4654 Use color in diagnostics. WHEN is 'never', 'always', or 'auto'. 4655 The default depends on how the compiler has been configured, it can 4656 be any of the above WHEN options or also 'never' if 'GCC_COLORS' 4657 environment variable isn't present in the environment, and 'auto' 4658 otherwise. 'auto' makes GCC use color only when the standard error 4659 is a terminal, and when not executing in an emacs shell. The forms 4660 '-fdiagnostics-color' and '-fno-diagnostics-color' are aliases for 4661 '-fdiagnostics-color=always' and '-fdiagnostics-color=never', 4662 respectively. 4663 4664 The colors are defined by the environment variable 'GCC_COLORS'. 4665 Its value is a colon-separated list of capabilities and Select 4666 Graphic Rendition (SGR) substrings. SGR commands are interpreted 4667 by the terminal or terminal emulator. (See the section in the 4668 documentation of your text terminal for permitted values and their 4669 meanings as character attributes.) These substring values are 4670 integers in decimal representation and can be concatenated with 4671 semicolons. Common values to concatenate include '1' for bold, '4' 4672 for underline, '5' for blink, '7' for inverse, '39' for default 4673 foreground color, '30' to '37' for foreground colors, '90' to '97' 4674 for 16-color mode foreground colors, '38;5;0' to '38;5;255' for 4675 88-color and 256-color modes foreground colors, '49' for default 4676 background color, '40' to '47' for background colors, '100' to 4677 '107' for 16-color mode background colors, and '48;5;0' to 4678 '48;5;255' for 88-color and 256-color modes background colors. 4679 4680 The default 'GCC_COLORS' is 4681 error=01;31:warning=01;35:note=01;36:range1=32:range2=34:locus=01:\ 4682 quote=01:path=01;36:fixit-insert=32:fixit-delete=31:\ 4683 diff-filename=01:diff-hunk=32:diff-delete=31:diff-insert=32:\ 4684 type-diff=01;32 4685 where '01;31' is bold red, '01;35' is bold magenta, '01;36' is bold 4686 cyan, '32' is green, '34' is blue, '01' is bold, and '31' is red. 4687 Setting 'GCC_COLORS' to the empty string disables colors. 4688 Supported capabilities are as follows. 4689 4690 'error=' 4691 SGR substring for error: markers. 4692 4693 'warning=' 4694 SGR substring for warning: markers. 4695 4696 'note=' 4697 SGR substring for note: markers. 4698 4699 'path=' 4700 SGR substring for colorizing paths of control-flow events as 4701 printed via '-fdiagnostics-path-format=', such as the 4702 identifiers of individual events and lines indicating 4703 interprocedural calls and returns. 4704 4705 'range1=' 4706 SGR substring for first additional range. 4707 4708 'range2=' 4709 SGR substring for second additional range. 4710 4711 'locus=' 4712 SGR substring for location information, 'file:line' or 4713 'file:line:column' etc. 4714 4715 'quote=' 4716 SGR substring for information printed within quotes. 4717 4718 'fixit-insert=' 4719 SGR substring for fix-it hints suggesting text to be inserted 4720 or replaced. 4721 4722 'fixit-delete=' 4723 SGR substring for fix-it hints suggesting text to be deleted. 4724 4725 'diff-filename=' 4726 SGR substring for filename headers within generated patches. 4727 4728 'diff-hunk=' 4729 SGR substring for the starts of hunks within generated 4730 patches. 4731 4732 'diff-delete=' 4733 SGR substring for deleted lines within generated patches. 4734 4735 'diff-insert=' 4736 SGR substring for inserted lines within generated patches. 4737 4738 'type-diff=' 4739 SGR substring for highlighting mismatching types within 4740 template arguments in the C++ frontend. 4741 4742'-fdiagnostics-urls[=WHEN]' 4743 Use escape sequences to embed URLs in diagnostics. For example, 4744 when '-fdiagnostics-show-option' emits text showing the 4745 command-line option controlling a diagnostic, embed a URL for 4746 documentation of that option. 4747 4748 WHEN is 'never', 'always', or 'auto'. 'auto' makes GCC use URL 4749 escape sequences only when the standard error is a terminal, and 4750 when not executing in an emacs shell or any graphical terminal 4751 which is known to be incompatible with this feature, see below. 4752 4753 The default depends on how the compiler has been configured. It 4754 can be any of the above WHEN options. 4755 4756 GCC can also be configured (via the 4757 '--with-diagnostics-urls=auto-if-env' configure-time option) so 4758 that the default is affected by environment variables. Under such 4759 a configuration, GCC defaults to using 'auto' if either 'GCC_URLS' 4760 or 'TERM_URLS' environment variables are present and non-empty in 4761 the environment of the compiler, or 'never' if neither are. 4762 4763 However, even with '-fdiagnostics-urls=always' the behavior is 4764 dependent on those environment variables: If 'GCC_URLS' is set to 4765 empty or 'no', do not embed URLs in diagnostics. If set to 'st', 4766 URLs use ST escape sequences. If set to 'bel', the default, URLs 4767 use BEL escape sequences. Any other non-empty value enables the 4768 feature. If 'GCC_URLS' is not set, use 'TERM_URLS' as a fallback. 4769 Note: ST is an ANSI escape sequence, string terminator 'ESC \', BEL 4770 is an ASCII character, CTRL-G that usually sounds like a beep. 4771 4772 At this time GCC tries to detect also a few terminals that are 4773 known to not implement the URL feature, and have bugs or at least 4774 had bugs in some versions that are still in use, where the URL 4775 escapes are likely to misbehave, i.e. print garbage on the screen. 4776 That list is currently xfce4-terminal, certain known to be buggy 4777 gnome-terminal versions, the linux console, and mingw. This check 4778 can be skipped with the '-fdiagnostics-urls=always'. 4779 4780'-fno-diagnostics-show-option' 4781 By default, each diagnostic emitted includes text indicating the 4782 command-line option that directly controls the diagnostic (if such 4783 an option is known to the diagnostic machinery). Specifying the 4784 '-fno-diagnostics-show-option' flag suppresses that behavior. 4785 4786'-fno-diagnostics-show-caret' 4787 By default, each diagnostic emitted includes the original source 4788 line and a caret '^' indicating the column. This option suppresses 4789 this information. The source line is truncated to N characters, if 4790 the '-fmessage-length=n' option is given. When the output is done 4791 to the terminal, the width is limited to the width given by the 4792 'COLUMNS' environment variable or, if not set, to the terminal 4793 width. 4794 4795'-fno-diagnostics-show-labels' 4796 By default, when printing source code (via 4797 '-fdiagnostics-show-caret'), diagnostics can label ranges of source 4798 code with pertinent information, such as the types of expressions: 4799 4800 printf ("foo %s bar", long_i + long_j); 4801 ~^ ~~~~~~~~~~~~~~~ 4802 | | 4803 char * long int 4804 4805 This option suppresses the printing of these labels (in the example 4806 above, the vertical bars and the "char *" and "long int" text). 4807 4808'-fno-diagnostics-show-cwe' 4809 Diagnostic messages can optionally have an associated CWE 4810 (https://cwe.mitre.org/index.html) identifier. GCC itself only 4811 provides such metadata for some of the '-fanalyzer' diagnostics. 4812 GCC plugins may also provide diagnostics with such metadata. By 4813 default, if this information is present, it will be printed with 4814 the diagnostic. This option suppresses the printing of this 4815 metadata. 4816 4817'-fno-diagnostics-show-line-numbers' 4818 By default, when printing source code (via 4819 '-fdiagnostics-show-caret'), a left margin is printed, showing line 4820 numbers. This option suppresses this left margin. 4821 4822'-fdiagnostics-minimum-margin-width=WIDTH' 4823 This option controls the minimum width of the left margin printed 4824 by '-fdiagnostics-show-line-numbers'. It defaults to 6. 4825 4826'-fdiagnostics-parseable-fixits' 4827 Emit fix-it hints in a machine-parseable format, suitable for 4828 consumption by IDEs. For each fix-it, a line will be printed after 4829 the relevant diagnostic, starting with the string "fix-it:". For 4830 example: 4831 4832 fix-it:"test.c":{45:3-45:21}:"gtk_widget_show_all" 4833 4834 The location is expressed as a half-open range, expressed as a 4835 count of bytes, starting at byte 1 for the initial column. In the 4836 above example, bytes 3 through 20 of line 45 of "test.c" are to be 4837 replaced with the given string: 4838 4839 00000000011111111112222222222 4840 12345678901234567890123456789 4841 gtk_widget_showall (dlg); 4842 ^^^^^^^^^^^^^^^^^^ 4843 gtk_widget_show_all 4844 4845 The filename and replacement string escape backslash as "\\", tab 4846 as "\t", newline as "\n", double quotes as "\"", non-printable 4847 characters as octal (e.g. vertical tab as "\013"). 4848 4849 An empty replacement string indicates that the given range is to be 4850 removed. An empty range (e.g. "45:3-45:3") indicates that the 4851 string is to be inserted at the given position. 4852 4853'-fdiagnostics-generate-patch' 4854 Print fix-it hints to stderr in unified diff format, after any 4855 diagnostics are printed. For example: 4856 4857 --- test.c 4858 +++ test.c 4859 @ -42,5 +42,5 @ 4860 4861 void show_cb(GtkDialog *dlg) 4862 { 4863 - gtk_widget_showall(dlg); 4864 + gtk_widget_show_all(dlg); 4865 } 4866 4867 4868 The diff may or may not be colorized, following the same rules as 4869 for diagnostics (see '-fdiagnostics-color'). 4870 4871'-fdiagnostics-show-template-tree' 4872 4873 In the C++ frontend, when printing diagnostics showing mismatching 4874 template types, such as: 4875 4876 could not convert 'std::map<int, std::vector<double> >()' 4877 from 'map<[...],vector<double>>' to 'map<[...],vector<float>> 4878 4879 the '-fdiagnostics-show-template-tree' flag enables printing a 4880 tree-like structure showing the common and differing parts of the 4881 types, such as: 4882 4883 map< 4884 [...], 4885 vector< 4886 [double != float]>> 4887 4888 The parts that differ are highlighted with color ("double" and 4889 "float" in this case). 4890 4891'-fno-elide-type' 4892 By default when the C++ frontend prints diagnostics showing 4893 mismatching template types, common parts of the types are printed 4894 as "[...]" to simplify the error message. For example: 4895 4896 could not convert 'std::map<int, std::vector<double> >()' 4897 from 'map<[...],vector<double>>' to 'map<[...],vector<float>> 4898 4899 Specifying the '-fno-elide-type' flag suppresses that behavior. 4900 This flag also affects the output of the 4901 '-fdiagnostics-show-template-tree' flag. 4902 4903'-fdiagnostics-path-format=KIND' 4904 Specify how to print paths of control-flow events for diagnostics 4905 that have such a path associated with them. 4906 4907 KIND is 'none', 'separate-events', or 'inline-events', the default. 4908 4909 'none' means to not print diagnostic paths. 4910 4911 'separate-events' means to print a separate "note" diagnostic for 4912 each event within the diagnostic. For example: 4913 4914 test.c:29:5: error: passing NULL as argument 1 to 'PyList_Append' which requires a non-NULL parameter 4915 test.c:25:10: note: (1) when 'PyList_New' fails, returning NULL 4916 test.c:27:3: note: (2) when 'i < count' 4917 test.c:29:5: note: (3) when calling 'PyList_Append', passing NULL from (1) as argument 1 4918 4919 'inline-events' means to print the events "inline" within the 4920 source code. This view attempts to consolidate the events into 4921 runs of sufficiently-close events, printing them as labelled ranges 4922 within the source. 4923 4924 For example, the same events as above might be printed as: 4925 4926 'test': events 1-3 4927 | 4928 | 25 | list = PyList_New(0); 4929 | | ^~~~~~~~~~~~~ 4930 | | | 4931 | | (1) when 'PyList_New' fails, returning NULL 4932 | 26 | 4933 | 27 | for (i = 0; i < count; i++) { 4934 | | ~~~ 4935 | | | 4936 | | (2) when 'i < count' 4937 | 28 | item = PyLong_FromLong(random()); 4938 | 29 | PyList_Append(list, item); 4939 | | ~~~~~~~~~~~~~~~~~~~~~~~~~ 4940 | | | 4941 | | (3) when calling 'PyList_Append', passing NULL from (1) as argument 1 4942 | 4943 4944 Interprocedural control flow is shown by grouping the events by 4945 stack frame, and using indentation to show how stack frames are 4946 nested, pushed, and popped. 4947 4948 For example: 4949 4950 'test': events 1-2 4951 | 4952 | 133 | { 4953 | | ^ 4954 | | | 4955 | | (1) entering 'test' 4956 | 134 | boxed_int *obj = make_boxed_int (i); 4957 | | ~~~~~~~~~~~~~~~~~~ 4958 | | | 4959 | | (2) calling 'make_boxed_int' 4960 | 4961 +--> 'make_boxed_int': events 3-4 4962 | 4963 | 120 | { 4964 | | ^ 4965 | | | 4966 | | (3) entering 'make_boxed_int' 4967 | 121 | boxed_int *result = (boxed_int *)wrapped_malloc (sizeof (boxed_int)); 4968 | | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 4969 | | | 4970 | | (4) calling 'wrapped_malloc' 4971 | 4972 +--> 'wrapped_malloc': events 5-6 4973 | 4974 | 7 | { 4975 | | ^ 4976 | | | 4977 | | (5) entering 'wrapped_malloc' 4978 | 8 | return malloc (size); 4979 | | ~~~~~~~~~~~~~ 4980 | | | 4981 | | (6) calling 'malloc' 4982 | 4983 <-------------+ 4984 | 4985 'test': event 7 4986 | 4987 | 138 | free_boxed_int (obj); 4988 | | ^~~~~~~~~~~~~~~~~~~~ 4989 | | | 4990 | | (7) calling 'free_boxed_int' 4991 | 4992 (etc) 4993 4994'-fdiagnostics-show-path-depths' 4995 This option provides additional information when printing 4996 control-flow paths associated with a diagnostic. 4997 4998 If this is option is provided then the stack depth will be printed 4999 for each run of events within 5000 '-fdiagnostics-path-format=separate-events'. 5001 5002 This is intended for use by GCC developers and plugin developers 5003 when debugging diagnostics that report interprocedural control 5004 flow. 5005 5006'-fno-show-column' 5007 Do not print column numbers in diagnostics. This may be necessary 5008 if diagnostics are being scanned by a program that does not 5009 understand the column numbers, such as 'dejagnu'. 5010 5011'-fdiagnostics-column-unit=UNIT' 5012 Select the units for the column number. This affects traditional 5013 diagnostics (in the absence of '-fno-show-column'), as well as JSON 5014 format diagnostics if requested. 5015 5016 The default UNIT, 'display', considers the number of display 5017 columns occupied by each character. This may be larger than the 5018 number of bytes required to encode the character, in the case of 5019 tab characters, or it may be smaller, in the case of multibyte 5020 characters. For example, the character "GREEK SMALL LETTER PI 5021 (U+03C0)" occupies one display column, and its UTF-8 encoding 5022 requires two bytes; the character "SLIGHTLY SMILING FACE (U+1F642)" 5023 occupies two display columns, and its UTF-8 encoding requires four 5024 bytes. 5025 5026 Setting UNIT to 'byte' changes the column number to the raw byte 5027 count in all cases, as was traditionally output by GCC prior to 5028 version 11.1.0. 5029 5030'-fdiagnostics-column-origin=ORIGIN' 5031 Select the origin for column numbers, i.e. the column number 5032 assigned to the first column. The default value of 1 corresponds 5033 to traditional GCC behavior and to the GNU style guide. Some 5034 utilities may perform better with an origin of 0; any non-negative 5035 value may be specified. 5036 5037'-fdiagnostics-escape-format=FORMAT' 5038 When GCC prints pertinent source lines for a diagnostic it normally 5039 attempts to print the source bytes directly. However, some 5040 diagnostics relate to encoding issues in the source file, such as 5041 malformed UTF-8, or issues with Unicode normalization. These 5042 diagnostics are flagged so that GCC will escape bytes that are not 5043 printable ASCII when printing their pertinent source lines. 5044 5045 This option controls how such bytes should be escaped. 5046 5047 The default FORMAT, 'unicode' displays Unicode characters that are 5048 not printable ASCII in the form '<U+XXXX>', and bytes that do not 5049 correspond to a Unicode character validly-encoded in UTF-8-encoded 5050 will be displayed as hexadecimal in the form '<XX>'. 5051 5052 For example, a source line containing the string 'before' followed 5053 by the Unicode character U+03C0 ("GREEK SMALL LETTER PI", with 5054 UTF-8 encoding 0xCF 0x80) followed by the byte 0xBF (a stray UTF-8 5055 trailing byte), followed by the string 'after' will be printed for 5056 such a diagnostic as: 5057 5058 before<U+03C0><BF>after 5059 5060 Setting FORMAT to 'bytes' will display all non-printable-ASCII 5061 bytes in the form '<XX>', thus showing the underlying encoding of 5062 non-ASCII Unicode characters. For the example above, the following 5063 will be printed: 5064 5065 before<CF><80><BF>after 5066 5067'-fdiagnostics-format=FORMAT' 5068 Select a different format for printing diagnostics. FORMAT is 5069 'text' or 'json'. The default is 'text'. 5070 5071 The 'json' format consists of a top-level JSON array containing 5072 JSON objects representing the diagnostics. 5073 5074 The JSON is emitted as one line, without formatting; the examples 5075 below have been formatted for clarity. 5076 5077 Diagnostics can have child diagnostics. For example, this error 5078 and note: 5079 5080 misleading-indentation.c:15:3: warning: this 'if' clause does not 5081 guard... [-Wmisleading-indentation] 5082 15 | if (flag) 5083 | ^~ 5084 misleading-indentation.c:17:5: note: ...this statement, but the latter 5085 is misleadingly indented as if it were guarded by the 'if' 5086 17 | y = 2; 5087 | ^ 5088 5089 might be printed in JSON form (after formatting) like this: 5090 5091 [ 5092 { 5093 "kind": "warning", 5094 "locations": [ 5095 { 5096 "caret": { 5097 "display-column": 3, 5098 "byte-column": 3, 5099 "column": 3, 5100 "file": "misleading-indentation.c", 5101 "line": 15 5102 }, 5103 "finish": { 5104 "display-column": 4, 5105 "byte-column": 4, 5106 "column": 4, 5107 "file": "misleading-indentation.c", 5108 "line": 15 5109 } 5110 } 5111 ], 5112 "message": "this \u2018if\u2019 clause does not guard...", 5113 "option": "-Wmisleading-indentation", 5114 "option_url": "https://gcc.gnu.org/onlinedocs/gcc/Warning-Options.html#index-Wmisleading-indentation", 5115 "children": [ 5116 { 5117 "kind": "note", 5118 "locations": [ 5119 { 5120 "caret": { 5121 "display-column": 5, 5122 "byte-column": 5, 5123 "column": 5, 5124 "file": "misleading-indentation.c", 5125 "line": 17 5126 } 5127 } 5128 ], 5129 "escape-source": false, 5130 "message": "...this statement, but the latter is ..." 5131 } 5132 ] 5133 "escape-source": false, 5134 "column-origin": 1, 5135 } 5136 ] 5137 5138 where the 'note' is a child of the 'warning'. 5139 5140 A diagnostic has a 'kind'. If this is 'warning', then there is an 5141 'option' key describing the command-line option controlling the 5142 warning. 5143 5144 A diagnostic can contain zero or more locations. Each location has 5145 an optional 'label' string and up to three positions within it: a 5146 'caret' position and optional 'start' and 'finish' positions. A 5147 position is described by a 'file' name, a 'line' number, and three 5148 numbers indicating a column position: 5149 5150 * 'display-column' counts display columns, accounting for tabs 5151 and multibyte characters. 5152 5153 * 'byte-column' counts raw bytes. 5154 5155 * 'column' is equal to one of the previous two, as dictated by 5156 the '-fdiagnostics-column-unit' option. 5157 5158 All three columns are relative to the origin specified by 5159 '-fdiagnostics-column-origin', which is typically equal to 1 but 5160 may be set, for instance, to 0 for compatibility with other 5161 utilities that number columns from 0. The column origin is 5162 recorded in the JSON output in the 'column-origin' tag. In the 5163 remaining examples below, the extra column number outputs have been 5164 omitted for brevity. 5165 5166 For example, this error: 5167 5168 bad-binary-ops.c:64:23: error: invalid operands to binary + (have 'S' {aka 5169 'struct s'} and 'T' {aka 'struct t'}) 5170 64 | return callee_4a () + callee_4b (); 5171 | ~~~~~~~~~~~~ ^ ~~~~~~~~~~~~ 5172 | | | 5173 | | T {aka struct t} 5174 | S {aka struct s} 5175 5176 has three locations. Its primary location is at the "+" token at 5177 column 23. It has two secondary locations, describing the left and 5178 right-hand sides of the expression, which have labels. It might be 5179 printed in JSON form as: 5180 5181 { 5182 "children": [], 5183 "kind": "error", 5184 "locations": [ 5185 { 5186 "caret": { 5187 "column": 23, "file": "bad-binary-ops.c", "line": 64 5188 } 5189 }, 5190 { 5191 "caret": { 5192 "column": 10, "file": "bad-binary-ops.c", "line": 64 5193 }, 5194 "finish": { 5195 "column": 21, "file": "bad-binary-ops.c", "line": 64 5196 }, 5197 "label": "S {aka struct s}" 5198 }, 5199 { 5200 "caret": { 5201 "column": 25, "file": "bad-binary-ops.c", "line": 64 5202 }, 5203 "finish": { 5204 "column": 36, "file": "bad-binary-ops.c", "line": 64 5205 }, 5206 "label": "T {aka struct t}" 5207 } 5208 ], 5209 "escape-source": false, 5210 "message": "invalid operands to binary + ..." 5211 } 5212 5213 If a diagnostic contains fix-it hints, it has a 'fixits' array, 5214 consisting of half-open intervals, similar to the output of 5215 '-fdiagnostics-parseable-fixits'. For example, this diagnostic 5216 with a replacement fix-it hint: 5217 5218 demo.c:8:15: error: 'struct s' has no member named 'colour'; did you 5219 mean 'color'? 5220 8 | return ptr->colour; 5221 | ^~~~~~ 5222 | color 5223 5224 might be printed in JSON form as: 5225 5226 { 5227 "children": [], 5228 "fixits": [ 5229 { 5230 "next": { 5231 "column": 21, 5232 "file": "demo.c", 5233 "line": 8 5234 }, 5235 "start": { 5236 "column": 15, 5237 "file": "demo.c", 5238 "line": 8 5239 }, 5240 "string": "color" 5241 } 5242 ], 5243 "kind": "error", 5244 "locations": [ 5245 { 5246 "caret": { 5247 "column": 15, 5248 "file": "demo.c", 5249 "line": 8 5250 }, 5251 "finish": { 5252 "column": 20, 5253 "file": "demo.c", 5254 "line": 8 5255 } 5256 } 5257 ], 5258 "escape-source": false, 5259 "message": "\u2018struct s\u2019 has no member named ..." 5260 } 5261 5262 where the fix-it hint suggests replacing the text from 'start' up 5263 to but not including 'next' with 'string''s value. Deletions are 5264 expressed via an empty value for 'string', insertions by having 5265 'start' equal 'next'. 5266 5267 If the diagnostic has a path of control-flow events associated with 5268 it, it has a 'path' array of objects representing the events. Each 5269 event object has a 'description' string, a 'location' object, along 5270 with a 'function' string and a 'depth' number for representing 5271 interprocedural paths. The 'function' represents the current 5272 function at that event, and the 'depth' represents the stack depth 5273 relative to some baseline: the higher, the more frames are within 5274 the stack. 5275 5276 For example, the intraprocedural example shown for 5277 '-fdiagnostics-path-format=' might have this JSON for its path: 5278 5279 "path": [ 5280 { 5281 "depth": 0, 5282 "description": "when 'PyList_New' fails, returning NULL", 5283 "function": "test", 5284 "location": { 5285 "column": 10, 5286 "file": "test.c", 5287 "line": 25 5288 } 5289 }, 5290 { 5291 "depth": 0, 5292 "description": "when 'i < count'", 5293 "function": "test", 5294 "location": { 5295 "column": 3, 5296 "file": "test.c", 5297 "line": 27 5298 } 5299 }, 5300 { 5301 "depth": 0, 5302 "description": "when calling 'PyList_Append', passing NULL from (1) as argument 1", 5303 "function": "test", 5304 "location": { 5305 "column": 5, 5306 "file": "test.c", 5307 "line": 29 5308 } 5309 } 5310 ] 5311 5312 Diagnostics have a boolean attribute 'escape-source', hinting 5313 whether non-ASCII bytes should be escaped when printing the 5314 pertinent lines of source code ('true' for diagnostics involving 5315 source encoding issues). 5316 5317 5318File: gcc.info, Node: Warning Options, Next: Static Analyzer Options, Prev: Diagnostic Message Formatting Options, Up: Invoking GCC 5319 53203.8 Options to Request or Suppress Warnings 5321=========================================== 5322 5323Warnings are diagnostic messages that report constructions that are not 5324inherently erroneous but that are risky or suggest there may have been 5325an error. 5326 5327 The following language-independent options do not enable specific 5328warnings but control the kinds of diagnostics produced by GCC. 5329 5330'-fsyntax-only' 5331 Check the code for syntax errors, but don't do anything beyond 5332 that. 5333 5334'-fmax-errors=N' 5335 Limits the maximum number of error messages to N, at which point 5336 GCC bails out rather than attempting to continue processing the 5337 source code. If N is 0 (the default), there is no limit on the 5338 number of error messages produced. If '-Wfatal-errors' is also 5339 specified, then '-Wfatal-errors' takes precedence over this option. 5340 5341'-w' 5342 Inhibit all warning messages. 5343 5344'-Werror' 5345 Make all warnings into errors. 5346 5347'-Werror=' 5348 Make the specified warning into an error. The specifier for a 5349 warning is appended; for example '-Werror=switch' turns the 5350 warnings controlled by '-Wswitch' into errors. This switch takes a 5351 negative form, to be used to negate '-Werror' for specific 5352 warnings; for example '-Wno-error=switch' makes '-Wswitch' warnings 5353 not be errors, even when '-Werror' is in effect. 5354 5355 The warning message for each controllable warning includes the 5356 option that controls the warning. That option can then be used 5357 with '-Werror=' and '-Wno-error=' as described above. (Printing of 5358 the option in the warning message can be disabled using the 5359 '-fno-diagnostics-show-option' flag.) 5360 5361 Note that specifying '-Werror='FOO automatically implies '-W'FOO. 5362 However, '-Wno-error='FOO does not imply anything. 5363 5364'-Wfatal-errors' 5365 This option causes the compiler to abort compilation on the first 5366 error occurred rather than trying to keep going and printing 5367 further error messages. 5368 5369 You can request many specific warnings with options beginning with 5370'-W', for example '-Wimplicit' to request warnings on implicit 5371declarations. Each of these specific warning options also has a 5372negative form beginning '-Wno-' to turn off warnings; for example, 5373'-Wno-implicit'. This manual lists only one of the two forms, whichever 5374is not the default. For further language-specific options also refer to 5375*note C++ Dialect Options:: and *note Objective-C and Objective-C++ 5376Dialect Options::. Additional warnings can be produced by enabling the 5377static analyzer; *Note Static Analyzer Options::. 5378 5379 Some options, such as '-Wall' and '-Wextra', turn on other options, 5380such as '-Wunused', which may turn on further options, such as 5381'-Wunused-value'. The combined effect of positive and negative forms is 5382that more specific options have priority over less specific ones, 5383independently of their position in the command-line. For options of the 5384same specificity, the last one takes effect. Options enabled or 5385disabled via pragmas (*note Diagnostic Pragmas::) take effect as if they 5386appeared at the end of the command-line. 5387 5388 When an unrecognized warning option is requested (e.g., 5389'-Wunknown-warning'), GCC emits a diagnostic stating that the option is 5390not recognized. However, if the '-Wno-' form is used, the behavior is 5391slightly different: no diagnostic is produced for '-Wno-unknown-warning' 5392unless other diagnostics are being produced. This allows the use of new 5393'-Wno-' options with old compilers, but if something goes wrong, the 5394compiler warns that an unrecognized option is present. 5395 5396 The effectiveness of some warnings depends on optimizations also being 5397enabled. For example '-Wsuggest-final-types' is more effective with 5398link-time optimization and some instances of other warnings may not be 5399issued at all unless optimization is enabled. While optimization in 5400general improves the efficacy of control and data flow sensitive 5401warnings, in some cases it may also cause false positives. 5402 5403'-Wpedantic' 5404'-pedantic' 5405 Issue all the warnings demanded by strict ISO C and ISO C++; reject 5406 all programs that use forbidden extensions, and some other programs 5407 that do not follow ISO C and ISO C++. For ISO C, follows the 5408 version of the ISO C standard specified by any '-std' option used. 5409 5410 Valid ISO C and ISO C++ programs should compile properly with or 5411 without this option (though a rare few require '-ansi' or a '-std' 5412 option specifying the required version of ISO C). However, without 5413 this option, certain GNU extensions and traditional C and C++ 5414 features are supported as well. With this option, they are 5415 rejected. 5416 5417 '-Wpedantic' does not cause warning messages for use of the 5418 alternate keywords whose names begin and end with '__'. This 5419 alternate format can also be used to disable warnings for non-ISO 5420 '__intN' types, i.e. '__intN__'. Pedantic warnings are also 5421 disabled in the expression that follows '__extension__'. However, 5422 only system header files should use these escape routes; 5423 application programs should avoid them. *Note Alternate 5424 Keywords::. 5425 5426 Some users try to use '-Wpedantic' to check programs for strict ISO 5427 C conformance. They soon find that it does not do quite what they 5428 want: it finds some non-ISO practices, but not all--only those for 5429 which ISO C _requires_ a diagnostic, and some others for which 5430 diagnostics have been added. 5431 5432 A feature to report any failure to conform to ISO C might be useful 5433 in some instances, but would require considerable additional work 5434 and would be quite different from '-Wpedantic'. We don't have 5435 plans to support such a feature in the near future. 5436 5437 Where the standard specified with '-std' represents a GNU extended 5438 dialect of C, such as 'gnu90' or 'gnu99', there is a corresponding 5439 "base standard", the version of ISO C on which the GNU extended 5440 dialect is based. Warnings from '-Wpedantic' are given where they 5441 are required by the base standard. (It does not make sense for 5442 such warnings to be given only for features not in the specified 5443 GNU C dialect, since by definition the GNU dialects of C include 5444 all features the compiler supports with the given option, and there 5445 would be nothing to warn about.) 5446 5447'-pedantic-errors' 5448 Give an error whenever the "base standard" (see '-Wpedantic') 5449 requires a diagnostic, in some cases where there is undefined 5450 behavior at compile-time and in some other cases that do not 5451 prevent compilation of programs that are valid according to the 5452 standard. This is not equivalent to '-Werror=pedantic', since 5453 there are errors enabled by this option and not enabled by the 5454 latter and vice versa. 5455 5456'-Wall' 5457 This enables all the warnings about constructions that some users 5458 consider questionable, and that are easy to avoid (or modify to 5459 prevent the warning), even in conjunction with macros. This also 5460 enables some language-specific warnings described in *note C++ 5461 Dialect Options:: and *note Objective-C and Objective-C++ Dialect 5462 Options::. 5463 5464 '-Wall' turns on the following warning flags: 5465 5466 -Waddress 5467 -Warray-bounds=1 (only with -O2) 5468 -Warray-compare 5469 -Warray-parameter=2 (C and Objective-C only) 5470 -Wbool-compare 5471 -Wbool-operation 5472 -Wc++11-compat -Wc++14-compat 5473 -Wcatch-value (C++ and Objective-C++ only) 5474 -Wchar-subscripts 5475 -Wcomment 5476 -Wdangling-pointer=2 5477 -Wduplicate-decl-specifier (C and Objective-C only) 5478 -Wenum-compare (in C/ObjC; this is on by default in C++) 5479 -Wformat 5480 -Wformat-overflow 5481 -Wformat-truncation 5482 -Wint-in-bool-context 5483 -Wimplicit (C and Objective-C only) 5484 -Wimplicit-int (C and Objective-C only) 5485 -Wimplicit-function-declaration (C and Objective-C only) 5486 -Winit-self (only for C++) 5487 -Wlogical-not-parentheses 5488 -Wmain (only for C/ObjC and unless -ffreestanding) 5489 -Wmaybe-uninitialized 5490 -Wmemset-elt-size 5491 -Wmemset-transposed-args 5492 -Wmisleading-indentation (only for C/C++) 5493 -Wmismatched-dealloc 5494 -Wmismatched-new-delete (only for C/C++) 5495 -Wmissing-attributes 5496 -Wmissing-braces (only for C/ObjC) 5497 -Wmultistatement-macros 5498 -Wnarrowing (only for C++) 5499 -Wnonnull 5500 -Wnonnull-compare 5501 -Wopenmp-simd 5502 -Wparentheses 5503 -Wpessimizing-move (only for C++) 5504 -Wpointer-sign 5505 -Wrange-loop-construct (only for C++) 5506 -Wreorder 5507 -Wrestrict 5508 -Wreturn-type 5509 -Wsequence-point 5510 -Wsign-compare (only in C++) 5511 -Wsizeof-array-div 5512 -Wsizeof-pointer-div 5513 -Wsizeof-pointer-memaccess 5514 -Wstrict-aliasing 5515 -Wstrict-overflow=1 5516 -Wswitch 5517 -Wtautological-compare 5518 -Wtrigraphs 5519 -Wuninitialized 5520 -Wunknown-pragmas 5521 -Wunused-function 5522 -Wunused-label 5523 -Wunused-value 5524 -Wunused-variable 5525 -Wuse-after-free=3 5526 -Wvla-parameter (C and Objective-C only) 5527 -Wvolatile-register-var 5528 -Wzero-length-bounds 5529 5530 Note that some warning flags are not implied by '-Wall'. Some of 5531 them warn about constructions that users generally do not consider 5532 questionable, but which occasionally you might wish to check for; 5533 others warn about constructions that are necessary or hard to avoid 5534 in some cases, and there is no simple way to modify the code to 5535 suppress the warning. Some of them are enabled by '-Wextra' but 5536 many of them must be enabled individually. 5537 5538'-Wextra' 5539 This enables some extra warning flags that are not enabled by 5540 '-Wall'. (This option used to be called '-W'. The older name is 5541 still supported, but the newer name is more descriptive.) 5542 5543 -Wclobbered 5544 -Wcast-function-type 5545 -Wdeprecated-copy (C++ only) 5546 -Wempty-body 5547 -Wenum-conversion (C only) 5548 -Wignored-qualifiers 5549 -Wimplicit-fallthrough=3 5550 -Wmissing-field-initializers 5551 -Wmissing-parameter-type (C only) 5552 -Wold-style-declaration (C only) 5553 -Woverride-init 5554 -Wsign-compare (C only) 5555 -Wstring-compare 5556 -Wredundant-move (only for C++) 5557 -Wtype-limits 5558 -Wuninitialized 5559 -Wshift-negative-value (in C++11 to C++17 and in C99 and newer) 5560 -Wunused-parameter (only with -Wunused or -Wall) 5561 -Wunused-but-set-parameter (only with -Wunused or -Wall) 5562 5563 The option '-Wextra' also prints warning messages for the following 5564 cases: 5565 5566 * A pointer is compared against integer zero with '<', '<=', 5567 '>', or '>='. 5568 5569 * (C++ only) An enumerator and a non-enumerator both appear in a 5570 conditional expression. 5571 5572 * (C++ only) Ambiguous virtual bases. 5573 5574 * (C++ only) Subscripting an array that has been declared 5575 'register'. 5576 5577 * (C++ only) Taking the address of a variable that has been 5578 declared 'register'. 5579 5580 * (C++ only) A base class is not initialized in the copy 5581 constructor of a derived class. 5582 5583'-Wabi (C, Objective-C, C++ and Objective-C++ only)' 5584 5585 Warn about code affected by ABI changes. This includes code that 5586 may not be compatible with the vendor-neutral C++ ABI as well as 5587 the psABI for the particular target. 5588 5589 Since G++ now defaults to updating the ABI with each major release, 5590 normally '-Wabi' warns only about C++ ABI compatibility problems if 5591 there is a check added later in a release series for an ABI issue 5592 discovered since the initial release. '-Wabi' warns about more 5593 things if an older ABI version is selected (with 5594 '-fabi-version=N'). 5595 5596 '-Wabi' can also be used with an explicit version number to warn 5597 about C++ ABI compatibility with a particular '-fabi-version' 5598 level, e.g. '-Wabi=2' to warn about changes relative to 5599 '-fabi-version=2'. 5600 5601 If an explicit version number is provided and 5602 '-fabi-compat-version' is not specified, the version number from 5603 this option is used for compatibility aliases. If no explicit 5604 version number is provided with this option, but 5605 '-fabi-compat-version' is specified, that version number is used 5606 for C++ ABI warnings. 5607 5608 Although an effort has been made to warn about all such cases, 5609 there are probably some cases that are not warned about, even 5610 though G++ is generating incompatible code. There may also be 5611 cases where warnings are emitted even though the code that is 5612 generated is compatible. 5613 5614 You should rewrite your code to avoid these warnings if you are 5615 concerned about the fact that code generated by G++ may not be 5616 binary compatible with code generated by other compilers. 5617 5618 Known incompatibilities in '-fabi-version=2' (which was the default 5619 from GCC 3.4 to 4.9) include: 5620 5621 * A template with a non-type template parameter of reference 5622 type was mangled incorrectly: 5623 extern int N; 5624 template <int &> struct S {}; 5625 void n (S<N>) {2} 5626 5627 This was fixed in '-fabi-version=3'. 5628 5629 * SIMD vector types declared using '__attribute ((vector_size))' 5630 were mangled in a non-standard way that does not allow for 5631 overloading of functions taking vectors of different sizes. 5632 5633 The mangling was changed in '-fabi-version=4'. 5634 5635 * '__attribute ((const))' and 'noreturn' were mangled as type 5636 qualifiers, and 'decltype' of a plain declaration was folded 5637 away. 5638 5639 These mangling issues were fixed in '-fabi-version=5'. 5640 5641 * Scoped enumerators passed as arguments to a variadic function 5642 are promoted like unscoped enumerators, causing 'va_arg' to 5643 complain. On most targets this does not actually affect the 5644 parameter passing ABI, as there is no way to pass an argument 5645 smaller than 'int'. 5646 5647 Also, the ABI changed the mangling of template argument packs, 5648 'const_cast', 'static_cast', prefix increment/decrement, and a 5649 class scope function used as a template argument. 5650 5651 These issues were corrected in '-fabi-version=6'. 5652 5653 * Lambdas in default argument scope were mangled incorrectly, 5654 and the ABI changed the mangling of 'nullptr_t'. 5655 5656 These issues were corrected in '-fabi-version=7'. 5657 5658 * When mangling a function type with function-cv-qualifiers, the 5659 un-qualified function type was incorrectly treated as a 5660 substitution candidate. 5661 5662 This was fixed in '-fabi-version=8', the default for GCC 5.1. 5663 5664 * 'decltype(nullptr)' incorrectly had an alignment of 1, leading 5665 to unaligned accesses. Note that this did not affect the ABI 5666 of a function with a 'nullptr_t' parameter, as parameters have 5667 a minimum alignment. 5668 5669 This was fixed in '-fabi-version=9', the default for GCC 5.2. 5670 5671 * Target-specific attributes that affect the identity of a type, 5672 such as ia32 calling conventions on a function type (stdcall, 5673 regparm, etc.), did not affect the mangled name, leading to 5674 name collisions when function pointers were used as template 5675 arguments. 5676 5677 This was fixed in '-fabi-version=10', the default for GCC 6.1. 5678 5679 This option also enables warnings about psABI-related changes. The 5680 known psABI changes at this point include: 5681 5682 * For SysV/x86-64, unions with 'long double' members are passed 5683 in memory as specified in psABI. Prior to GCC 4.4, this was 5684 not the case. For example: 5685 5686 union U { 5687 long double ld; 5688 int i; 5689 }; 5690 5691 'union U' is now always passed in memory. 5692 5693'-Wchar-subscripts' 5694 Warn if an array subscript has type 'char'. This is a common cause 5695 of error, as programmers often forget that this type is signed on 5696 some machines. This warning is enabled by '-Wall'. 5697 5698'-Wno-coverage-mismatch' 5699 Warn if feedback profiles do not match when using the 5700 '-fprofile-use' option. If a source file is changed between 5701 compiling with '-fprofile-generate' and with '-fprofile-use', the 5702 files with the profile feedback can fail to match the source file 5703 and GCC cannot use the profile feedback information. By default, 5704 this warning is enabled and is treated as an error. 5705 '-Wno-coverage-mismatch' can be used to disable the warning or 5706 '-Wno-error=coverage-mismatch' can be used to disable the error. 5707 Disabling the error for this warning can result in poorly optimized 5708 code and is useful only in the case of very minor changes such as 5709 bug fixes to an existing code-base. Completely disabling the 5710 warning is not recommended. 5711 5712'-Wno-coverage-invalid-line-number' 5713 Warn in case a function ends earlier than it begins due to an 5714 invalid linenum macros. The warning is emitted only with 5715 '--coverage' enabled. 5716 5717 By default, this warning is enabled and is treated as an error. 5718 '-Wno-coverage-invalid-line-number' can be used to disable the 5719 warning or '-Wno-error=coverage-invalid-line-number' can be used to 5720 disable the error. 5721 5722'-Wno-cpp (C, Objective-C, C++, Objective-C++ and Fortran only)' 5723 Suppress warning messages emitted by '#warning' directives. 5724 5725'-Wdouble-promotion (C, C++, Objective-C and Objective-C++ only)' 5726 Give a warning when a value of type 'float' is implicitly promoted 5727 to 'double'. CPUs with a 32-bit "single-precision" floating-point 5728 unit implement 'float' in hardware, but emulate 'double' in 5729 software. On such a machine, doing computations using 'double' 5730 values is much more expensive because of the overhead required for 5731 software emulation. 5732 5733 It is easy to accidentally do computations with 'double' because 5734 floating-point literals are implicitly of type 'double'. For 5735 example, in: 5736 float area(float radius) 5737 { 5738 return 3.14159 * radius * radius; 5739 } 5740 the compiler performs the entire computation with 'double' because 5741 the floating-point literal is a 'double'. 5742 5743'-Wduplicate-decl-specifier (C and Objective-C only)' 5744 Warn if a declaration has duplicate 'const', 'volatile', 'restrict' 5745 or '_Atomic' specifier. This warning is enabled by '-Wall'. 5746 5747'-Wformat' 5748'-Wformat=N' 5749 Check calls to 'printf' and 'scanf', etc., to make sure that the 5750 arguments supplied have types appropriate to the format string 5751 specified, and that the conversions specified in the format string 5752 make sense. This includes standard functions, and others specified 5753 by format attributes (*note Function Attributes::), in the 5754 'printf', 'scanf', 'strftime' and 'strfmon' (an X/Open extension, 5755 not in the C standard) families (or other target-specific 5756 families). Which functions are checked without format attributes 5757 having been specified depends on the standard version selected, and 5758 such checks of functions without the attribute specified are 5759 disabled by '-ffreestanding' or '-fno-builtin'. 5760 5761 The formats are checked against the format features supported by 5762 GNU libc version 2.2. These include all ISO C90 and C99 features, 5763 as well as features from the Single Unix Specification and some BSD 5764 and GNU extensions. Other library implementations may not support 5765 all these features; GCC does not support warning about features 5766 that go beyond a particular library's limitations. However, if 5767 '-Wpedantic' is used with '-Wformat', warnings are given about 5768 format features not in the selected standard version (but not for 5769 'strfmon' formats, since those are not in any version of the C 5770 standard). *Note Options Controlling C Dialect: C Dialect Options. 5771 5772 '-Wformat=1' 5773 '-Wformat' 5774 Option '-Wformat' is equivalent to '-Wformat=1', and 5775 '-Wno-format' is equivalent to '-Wformat=0'. Since '-Wformat' 5776 also checks for null format arguments for several functions, 5777 '-Wformat' also implies '-Wnonnull'. Some aspects of this 5778 level of format checking can be disabled by the options: 5779 '-Wno-format-contains-nul', '-Wno-format-extra-args', and 5780 '-Wno-format-zero-length'. '-Wformat' is enabled by '-Wall'. 5781 5782 '-Wformat=2' 5783 Enable '-Wformat' plus additional format checks. Currently 5784 equivalent to '-Wformat -Wformat-nonliteral -Wformat-security 5785 -Wformat-y2k'. 5786 5787'-Wno-format-contains-nul' 5788 If '-Wformat' is specified, do not warn about format strings that 5789 contain NUL bytes. 5790 5791'-Wno-format-extra-args' 5792 If '-Wformat' is specified, do not warn about excess arguments to a 5793 'printf' or 'scanf' format function. The C standard specifies that 5794 such arguments are ignored. 5795 5796 Where the unused arguments lie between used arguments that are 5797 specified with '$' operand number specifications, normally warnings 5798 are still given, since the implementation could not know what type 5799 to pass to 'va_arg' to skip the unused arguments. However, in the 5800 case of 'scanf' formats, this option suppresses the warning if the 5801 unused arguments are all pointers, since the Single Unix 5802 Specification says that such unused arguments are allowed. 5803 5804'-Wformat-overflow' 5805'-Wformat-overflow=LEVEL' 5806 Warn about calls to formatted input/output functions such as 5807 'sprintf' and 'vsprintf' that might overflow the destination 5808 buffer. When the exact number of bytes written by a format 5809 directive cannot be determined at compile-time it is estimated 5810 based on heuristics that depend on the LEVEL argument and on 5811 optimization. While enabling optimization will in most cases 5812 improve the accuracy of the warning, it may also result in false 5813 positives. 5814 5815 '-Wformat-overflow' 5816 '-Wformat-overflow=1' 5817 Level 1 of '-Wformat-overflow' enabled by '-Wformat' employs a 5818 conservative approach that warns only about calls that most 5819 likely overflow the buffer. At this level, numeric arguments 5820 to format directives with unknown values are assumed to have 5821 the value of one, and strings of unknown length to be empty. 5822 Numeric arguments that are known to be bounded to a subrange 5823 of their type, or string arguments whose output is bounded 5824 either by their directive's precision or by a finite set of 5825 string literals, are assumed to take on the value within the 5826 range that results in the most bytes on output. For example, 5827 the call to 'sprintf' below is diagnosed because even with 5828 both A and B equal to zero, the terminating NUL character 5829 (''\0'') appended by the function to the destination buffer 5830 will be written past its end. Increasing the size of the 5831 buffer by a single byte is sufficient to avoid the warning, 5832 though it may not be sufficient to avoid the overflow. 5833 5834 void f (int a, int b) 5835 { 5836 char buf [13]; 5837 sprintf (buf, "a = %i, b = %i\n", a, b); 5838 } 5839 5840 '-Wformat-overflow=2' 5841 Level 2 warns also about calls that might overflow the 5842 destination buffer given an argument of sufficient length or 5843 magnitude. At level 2, unknown numeric arguments are assumed 5844 to have the minimum representable value for signed types with 5845 a precision greater than 1, and the maximum representable 5846 value otherwise. Unknown string arguments whose length cannot 5847 be assumed to be bounded either by the directive's precision, 5848 or by a finite set of string literals they may evaluate to, or 5849 the character array they may point to, are assumed to be 1 5850 character long. 5851 5852 At level 2, the call in the example above is again diagnosed, 5853 but this time because with A equal to a 32-bit 'INT_MIN' the 5854 first '%i' directive will write some of its digits beyond the 5855 end of the destination buffer. To make the call safe 5856 regardless of the values of the two variables, the size of the 5857 destination buffer must be increased to at least 34 bytes. 5858 GCC includes the minimum size of the buffer in an 5859 informational note following the warning. 5860 5861 An alternative to increasing the size of the destination 5862 buffer is to constrain the range of formatted values. The 5863 maximum length of string arguments can be bounded by 5864 specifying the precision in the format directive. When 5865 numeric arguments of format directives can be assumed to be 5866 bounded by less than the precision of their type, choosing an 5867 appropriate length modifier to the format specifier will 5868 reduce the required buffer size. For example, if A and B in 5869 the example above can be assumed to be within the precision of 5870 the 'short int' type then using either the '%hi' format 5871 directive or casting the argument to 'short' reduces the 5872 maximum required size of the buffer to 24 bytes. 5873 5874 void f (int a, int b) 5875 { 5876 char buf [23]; 5877 sprintf (buf, "a = %hi, b = %i\n", a, (short)b); 5878 } 5879 5880'-Wno-format-zero-length' 5881 If '-Wformat' is specified, do not warn about zero-length formats. 5882 The C standard specifies that zero-length formats are allowed. 5883 5884'-Wformat-nonliteral' 5885 If '-Wformat' is specified, also warn if the format string is not a 5886 string literal and so cannot be checked, unless the format function 5887 takes its format arguments as a 'va_list'. 5888 5889'-Wformat-security' 5890 If '-Wformat' is specified, also warn about uses of format 5891 functions that represent possible security problems. At present, 5892 this warns about calls to 'printf' and 'scanf' functions where the 5893 format string is not a string literal and there are no format 5894 arguments, as in 'printf (foo);'. This may be a security hole if 5895 the format string came from untrusted input and contains '%n'. 5896 (This is currently a subset of what '-Wformat-nonliteral' warns 5897 about, but in future warnings may be added to '-Wformat-security' 5898 that are not included in '-Wformat-nonliteral'.) 5899 5900'-Wformat-signedness' 5901 If '-Wformat' is specified, also warn if the format string requires 5902 an unsigned argument and the argument is signed and vice versa. 5903 5904'-Wformat-truncation' 5905'-Wformat-truncation=LEVEL' 5906 Warn about calls to formatted input/output functions such as 5907 'snprintf' and 'vsnprintf' that might result in output truncation. 5908 When the exact number of bytes written by a format directive cannot 5909 be determined at compile-time it is estimated based on heuristics 5910 that depend on the LEVEL argument and on optimization. While 5911 enabling optimization will in most cases improve the accuracy of 5912 the warning, it may also result in false positives. Except as 5913 noted otherwise, the option uses the same logic 5914 '-Wformat-overflow'. 5915 5916 '-Wformat-truncation' 5917 '-Wformat-truncation=1' 5918 Level 1 of '-Wformat-truncation' enabled by '-Wformat' employs 5919 a conservative approach that warns only about calls to bounded 5920 functions whose return value is unused and that will most 5921 likely result in output truncation. 5922 5923 '-Wformat-truncation=2' 5924 Level 2 warns also about calls to bounded functions whose 5925 return value is used and that might result in truncation given 5926 an argument of sufficient length or magnitude. 5927 5928'-Wformat-y2k' 5929 If '-Wformat' is specified, also warn about 'strftime' formats that 5930 may yield only a two-digit year. 5931 5932'-Wnonnull' 5933 Warn about passing a null pointer for arguments marked as requiring 5934 a non-null value by the 'nonnull' function attribute. 5935 5936 '-Wnonnull' is included in '-Wall' and '-Wformat'. It can be 5937 disabled with the '-Wno-nonnull' option. 5938 5939'-Wnonnull-compare' 5940 Warn when comparing an argument marked with the 'nonnull' function 5941 attribute against null inside the function. 5942 5943 '-Wnonnull-compare' is included in '-Wall'. It can be disabled 5944 with the '-Wno-nonnull-compare' option. 5945 5946'-Wnull-dereference' 5947 Warn if the compiler detects paths that trigger erroneous or 5948 undefined behavior due to dereferencing a null pointer. This 5949 option is only active when '-fdelete-null-pointer-checks' is 5950 active, which is enabled by optimizations in most targets. The 5951 precision of the warnings depends on the optimization options used. 5952 5953'-Winfinite-recursion' 5954 Warn about infinitely recursive calls. The warning is effective at 5955 all optimization levels but requires optimization in order to 5956 detect infinite recursion in calls between two or more functions. 5957 '-Winfinite-recursion' is included in '-Wall'. 5958 5959'-Winit-self (C, C++, Objective-C and Objective-C++ only)' 5960 Warn about uninitialized variables that are initialized with 5961 themselves. Note this option can only be used with the 5962 '-Wuninitialized' option. 5963 5964 For example, GCC warns about 'i' being uninitialized in the 5965 following snippet only when '-Winit-self' has been specified: 5966 int f() 5967 { 5968 int i = i; 5969 return i; 5970 } 5971 5972 This warning is enabled by '-Wall' in C++. 5973 5974'-Wno-implicit-int (C and Objective-C only)' 5975 This option controls warnings when a declaration does not specify a 5976 type. This warning is enabled by default in C99 and later dialects 5977 of C, and also by '-Wall'. 5978 5979'-Wno-implicit-function-declaration (C and Objective-C only)' 5980 This option controls warnings when a function is used before being 5981 declared. This warning is enabled by default in C99 and later 5982 dialects of C, and also by '-Wall'. The warning is made into an 5983 error by '-pedantic-errors'. 5984 5985'-Wimplicit (C and Objective-C only)' 5986 Same as '-Wimplicit-int' and '-Wimplicit-function-declaration'. 5987 This warning is enabled by '-Wall'. 5988 5989'-Wimplicit-fallthrough' 5990 '-Wimplicit-fallthrough' is the same as '-Wimplicit-fallthrough=3' 5991 and '-Wno-implicit-fallthrough' is the same as 5992 '-Wimplicit-fallthrough=0'. 5993 5994'-Wimplicit-fallthrough=N' 5995 Warn when a switch case falls through. For example: 5996 5997 switch (cond) 5998 { 5999 case 1: 6000 a = 1; 6001 break; 6002 case 2: 6003 a = 2; 6004 case 3: 6005 a = 3; 6006 break; 6007 } 6008 6009 This warning does not warn when the last statement of a case cannot 6010 fall through, e.g. when there is a return statement or a call to 6011 function declared with the noreturn attribute. 6012 '-Wimplicit-fallthrough=' also takes into account control flow 6013 statements, such as ifs, and only warns when appropriate. E.g. 6014 6015 switch (cond) 6016 { 6017 case 1: 6018 if (i > 3) { 6019 bar (5); 6020 break; 6021 } else if (i < 1) { 6022 bar (0); 6023 } else 6024 return; 6025 default: 6026 ... 6027 } 6028 6029 Since there are occasions where a switch case fall through is 6030 desirable, GCC provides an attribute, '__attribute__ 6031 ((fallthrough))', that is to be used along with a null statement to 6032 suppress this warning that would normally occur: 6033 6034 switch (cond) 6035 { 6036 case 1: 6037 bar (0); 6038 __attribute__ ((fallthrough)); 6039 default: 6040 ... 6041 } 6042 6043 C++17 provides a standard way to suppress the 6044 '-Wimplicit-fallthrough' warning using '[[fallthrough]];' instead 6045 of the GNU attribute. In C++11 or C++14 users can use 6046 '[[gnu::fallthrough]];', which is a GNU extension. Instead of 6047 these attributes, it is also possible to add a fallthrough comment 6048 to silence the warning. The whole body of the C or C++ style 6049 comment should match the given regular expressions listed below. 6050 The option argument N specifies what kind of comments are accepted: 6051 6052 * '-Wimplicit-fallthrough=0' disables the warning altogether. 6053 6054 * '-Wimplicit-fallthrough=1' matches '.*' regular expression, 6055 any comment is used as fallthrough comment. 6056 6057 * '-Wimplicit-fallthrough=2' case insensitively matches 6058 '.*falls?[ \t-]*thr(ough|u).*' regular expression. 6059 6060 * '-Wimplicit-fallthrough=3' case sensitively matches one of the 6061 following regular expressions: 6062 6063 * '-fallthrough' 6064 6065 * '@fallthrough@' 6066 6067 * 'lint -fallthrough[ \t]*' 6068 6069 * '[ \t.!]*(ELSE,? |INTENTIONAL(LY)? )? 6070 FALL(S | |-)?THR(OUGH|U)[ \t.!]*(-[^\n\r]*)?' 6071 6072 * '[ \t.!]*(Else,? |Intentional(ly)? )? 6073 Fall((s | |-)[Tt]|t)hr(ough|u)[ \t.!]*(-[^\n\r]*)?' 6074 6075 * '[ \t.!]*([Ee]lse,? |[Ii]ntentional(ly)? )? 6076 fall(s | |-)?thr(ough|u)[ \t.!]*(-[^\n\r]*)?' 6077 6078 * '-Wimplicit-fallthrough=4' case sensitively matches one of the 6079 following regular expressions: 6080 6081 * '-fallthrough' 6082 6083 * '@fallthrough@' 6084 6085 * 'lint -fallthrough[ \t]*' 6086 6087 * '[ \t]*FALLTHR(OUGH|U)[ \t]*' 6088 6089 * '-Wimplicit-fallthrough=5' doesn't recognize any comments as 6090 fallthrough comments, only attributes disable the warning. 6091 6092 The comment needs to be followed after optional whitespace and 6093 other comments by 'case' or 'default' keywords or by a user label 6094 that precedes some 'case' or 'default' label. 6095 6096 switch (cond) 6097 { 6098 case 1: 6099 bar (0); 6100 /* FALLTHRU */ 6101 default: 6102 ... 6103 } 6104 6105 The '-Wimplicit-fallthrough=3' warning is enabled by '-Wextra'. 6106 6107'-Wno-if-not-aligned (C, C++, Objective-C and Objective-C++ only)' 6108 Control if warnings triggered by the 'warn_if_not_aligned' 6109 attribute should be issued. These warnings are enabled by default. 6110 6111'-Wignored-qualifiers (C and C++ only)' 6112 Warn if the return type of a function has a type qualifier such as 6113 'const'. For ISO C such a type qualifier has no effect, since the 6114 value returned by a function is not an lvalue. For C++, the 6115 warning is only emitted for scalar types or 'void'. ISO C 6116 prohibits qualified 'void' return types on function definitions, so 6117 such return types always receive a warning even without this 6118 option. 6119 6120 This warning is also enabled by '-Wextra'. 6121 6122'-Wno-ignored-attributes (C and C++ only)' 6123 This option controls warnings when an attribute is ignored. This 6124 is different from the '-Wattributes' option in that it warns 6125 whenever the compiler decides to drop an attribute, not that the 6126 attribute is either unknown, used in a wrong place, etc. This 6127 warning is enabled by default. 6128 6129'-Wmain' 6130 Warn if the type of 'main' is suspicious. 'main' should be a 6131 function with external linkage, returning int, taking either zero 6132 arguments, two, or three arguments of appropriate types. This 6133 warning is enabled by default in C++ and is enabled by either 6134 '-Wall' or '-Wpedantic'. 6135 6136'-Wmisleading-indentation (C and C++ only)' 6137 Warn when the indentation of the code does not reflect the block 6138 structure. Specifically, a warning is issued for 'if', 'else', 6139 'while', and 'for' clauses with a guarded statement that does not 6140 use braces, followed by an unguarded statement with the same 6141 indentation. 6142 6143 In the following example, the call to "bar" is misleadingly 6144 indented as if it were guarded by the "if" conditional. 6145 6146 if (some_condition ()) 6147 foo (); 6148 bar (); /* Gotcha: this is not guarded by the "if". */ 6149 6150 In the case of mixed tabs and spaces, the warning uses the 6151 '-ftabstop=' option to determine if the statements line up 6152 (defaulting to 8). 6153 6154 The warning is not issued for code involving multiline preprocessor 6155 logic such as the following example. 6156 6157 if (flagA) 6158 foo (0); 6159 #if SOME_CONDITION_THAT_DOES_NOT_HOLD 6160 if (flagB) 6161 #endif 6162 foo (1); 6163 6164 The warning is not issued after a '#line' directive, since this 6165 typically indicates autogenerated code, and no assumptions can be 6166 made about the layout of the file that the directive references. 6167 6168 This warning is enabled by '-Wall' in C and C++. 6169 6170'-Wmissing-attributes' 6171 Warn when a declaration of a function is missing one or more 6172 attributes that a related function is declared with and whose 6173 absence may adversely affect the correctness or efficiency of 6174 generated code. For example, the warning is issued for 6175 declarations of aliases that use attributes to specify less 6176 restrictive requirements than those of their targets. This 6177 typically represents a potential optimization opportunity. By 6178 contrast, the '-Wattribute-alias=2' option controls warnings issued 6179 when the alias is more restrictive than the target, which could 6180 lead to incorrect code generation. Attributes considered include 6181 'alloc_align', 'alloc_size', 'cold', 'const', 'hot', 'leaf', 6182 'malloc', 'nonnull', 'noreturn', 'nothrow', 'pure', 6183 'returns_nonnull', and 'returns_twice'. 6184 6185 In C++, the warning is issued when an explicit specialization of a 6186 primary template declared with attribute 'alloc_align', 6187 'alloc_size', 'assume_aligned', 'format', 'format_arg', 'malloc', 6188 or 'nonnull' is declared without it. Attributes 'deprecated', 6189 'error', and 'warning' suppress the warning. (*note Function 6190 Attributes::). 6191 6192 You can use the 'copy' attribute to apply the same set of 6193 attributes to a declaration as that on another declaration without 6194 explicitly enumerating the attributes. This attribute can be 6195 applied to declarations of functions (*note Common Function 6196 Attributes::), variables (*note Common Variable Attributes::), or 6197 types (*note Common Type Attributes::). 6198 6199 '-Wmissing-attributes' is enabled by '-Wall'. 6200 6201 For example, since the declaration of the primary function template 6202 below makes use of both attribute 'malloc' and 'alloc_size' the 6203 declaration of the explicit specialization of the template is 6204 diagnosed because it is missing one of the attributes. 6205 6206 template <class T> 6207 T* __attribute__ ((malloc, alloc_size (1))) 6208 allocate (size_t); 6209 6210 template <> 6211 void* __attribute__ ((malloc)) // missing alloc_size 6212 allocate<void> (size_t); 6213 6214'-Wmissing-braces' 6215 Warn if an aggregate or union initializer is not fully bracketed. 6216 In the following example, the initializer for 'a' is not fully 6217 bracketed, but that for 'b' is fully bracketed. 6218 6219 int a[2][2] = { 0, 1, 2, 3 }; 6220 int b[2][2] = { { 0, 1 }, { 2, 3 } }; 6221 6222 This warning is enabled by '-Wall'. 6223 6224'-Wmissing-include-dirs (C, C++, Objective-C, Objective-C++ and Fortran only)' 6225 Warn if a user-supplied include directory does not exist. This 6226 opions is disabled by default for C, C++, Objective-C and 6227 Objective-C++. For Fortran, it is partially enabled by default by 6228 warning for -I and -J, only. 6229 6230'-Wno-missing-profile' 6231 This option controls warnings if feedback profiles are missing when 6232 using the '-fprofile-use' option. This option diagnoses those 6233 cases where a new function or a new file is added between compiling 6234 with '-fprofile-generate' and with '-fprofile-use', without 6235 regenerating the profiles. In these cases, the profile feedback 6236 data files do not contain any profile feedback information for the 6237 newly added function or file respectively. Also, in the case when 6238 profile count data (.gcda) files are removed, GCC cannot use any 6239 profile feedback information. In all these cases, warnings are 6240 issued to inform you that a profile generation step is due. 6241 Ignoring the warning can result in poorly optimized code. 6242 '-Wno-missing-profile' can be used to disable the warning, but this 6243 is not recommended and should be done only when non-existent 6244 profile data is justified. 6245 6246'-Wmismatched-dealloc' 6247 6248 Warn for calls to deallocation functions with pointer arguments 6249 returned from from allocations functions for which the former isn't 6250 a suitable deallocator. A pair of functions can be associated as 6251 matching allocators and deallocators by use of attribute 'malloc'. 6252 Unless disabled by the '-fno-builtin' option the standard functions 6253 'calloc', 'malloc', 'realloc', and 'free', as well as the 6254 corresponding forms of C++ 'operator new' and 'operator delete' are 6255 implicitly associated as matching allocators and deallocators. In 6256 the following example 'mydealloc' is the deallocator for pointers 6257 returned from 'myalloc'. 6258 6259 void mydealloc (void*); 6260 6261 __attribute__ ((malloc (mydealloc, 1))) void* 6262 myalloc (size_t); 6263 6264 void f (void) 6265 { 6266 void *p = myalloc (32); 6267 // ...use p... 6268 free (p); // warning: not a matching deallocator for myalloc 6269 mydealloc (p); // ok 6270 } 6271 6272 In C++, the related option '-Wmismatched-new-delete' diagnoses 6273 mismatches involving either 'operator new' or 'operator delete'. 6274 6275 Option '-Wmismatched-dealloc' is included in '-Wall'. 6276 6277'-Wmultistatement-macros' 6278 Warn about unsafe multiple statement macros that appear to be 6279 guarded by a clause such as 'if', 'else', 'for', 'switch', or 6280 'while', in which only the first statement is actually guarded 6281 after the macro is expanded. 6282 6283 For example: 6284 6285 #define DOIT x++; y++ 6286 if (c) 6287 DOIT; 6288 6289 will increment 'y' unconditionally, not just when 'c' holds. The 6290 can usually be fixed by wrapping the macro in a do-while loop: 6291 #define DOIT do { x++; y++; } while (0) 6292 if (c) 6293 DOIT; 6294 6295 This warning is enabled by '-Wall' in C and C++. 6296 6297'-Wparentheses' 6298 Warn if parentheses are omitted in certain contexts, such as when 6299 there is an assignment in a context where a truth value is 6300 expected, or when operators are nested whose precedence people 6301 often get confused about. 6302 6303 Also warn if a comparison like 'x<=y<=z' appears; this is 6304 equivalent to '(x<=y ? 1 : 0) <= z', which is a different 6305 interpretation from that of ordinary mathematical notation. 6306 6307 Also warn for dangerous uses of the GNU extension to '?:' with 6308 omitted middle operand. When the condition in the '?': operator is 6309 a boolean expression, the omitted value is always 1. Often 6310 programmers expect it to be a value computed inside the conditional 6311 expression instead. 6312 6313 For C++ this also warns for some cases of unnecessary parentheses 6314 in declarations, which can indicate an attempt at a function call 6315 instead of a declaration: 6316 { 6317 // Declares a local variable called mymutex. 6318 std::unique_lock<std::mutex> (mymutex); 6319 // User meant std::unique_lock<std::mutex> lock (mymutex); 6320 } 6321 6322 This warning is enabled by '-Wall'. 6323 6324'-Wsequence-point' 6325 Warn about code that may have undefined semantics because of 6326 violations of sequence point rules in the C and C++ standards. 6327 6328 The C and C++ standards define the order in which expressions in a 6329 C/C++ program are evaluated in terms of "sequence points", which 6330 represent a partial ordering between the execution of parts of the 6331 program: those executed before the sequence point, and those 6332 executed after it. These occur after the evaluation of a full 6333 expression (one which is not part of a larger expression), after 6334 the evaluation of the first operand of a '&&', '||', '? :' or ',' 6335 (comma) operator, before a function is called (but after the 6336 evaluation of its arguments and the expression denoting the called 6337 function), and in certain other places. Other than as expressed by 6338 the sequence point rules, the order of evaluation of subexpressions 6339 of an expression is not specified. All these rules describe only a 6340 partial order rather than a total order, since, for example, if two 6341 functions are called within one expression with no sequence point 6342 between them, the order in which the functions are called is not 6343 specified. However, the standards committee have ruled that 6344 function calls do not overlap. 6345 6346 It is not specified when between sequence points modifications to 6347 the values of objects take effect. Programs whose behavior depends 6348 on this have undefined behavior; the C and C++ standards specify 6349 that "Between the previous and next sequence point an object shall 6350 have its stored value modified at most once by the evaluation of an 6351 expression. Furthermore, the prior value shall be read only to 6352 determine the value to be stored.". If a program breaks these 6353 rules, the results on any particular implementation are entirely 6354 unpredictable. 6355 6356 Examples of code with undefined behavior are 'a = a++;', 'a[n] = 6357 b[n++]' and 'a[i++] = i;'. Some more complicated cases are not 6358 diagnosed by this option, and it may give an occasional false 6359 positive result, but in general it has been found fairly effective 6360 at detecting this sort of problem in programs. 6361 6362 The C++17 standard will define the order of evaluation of operands 6363 in more cases: in particular it requires that the right-hand side 6364 of an assignment be evaluated before the left-hand side, so the 6365 above examples are no longer undefined. But this option will still 6366 warn about them, to help people avoid writing code that is 6367 undefined in C and earlier revisions of C++. 6368 6369 The standard is worded confusingly, therefore there is some debate 6370 over the precise meaning of the sequence point rules in subtle 6371 cases. Links to discussions of the problem, including proposed 6372 formal definitions, may be found on the GCC readings page, at 6373 <https://gcc.gnu.org/readings.html>. 6374 6375 This warning is enabled by '-Wall' for C and C++. 6376 6377'-Wno-return-local-addr' 6378 Do not warn about returning a pointer (or in C++, a reference) to a 6379 variable that goes out of scope after the function returns. 6380 6381'-Wreturn-type' 6382 Warn whenever a function is defined with a return type that 6383 defaults to 'int'. Also warn about any 'return' statement with no 6384 return value in a function whose return type is not 'void' (falling 6385 off the end of the function body is considered returning without a 6386 value). 6387 6388 For C only, warn about a 'return' statement with an expression in a 6389 function whose return type is 'void', unless the expression type is 6390 also 'void'. As a GNU extension, the latter case is accepted 6391 without a warning unless '-Wpedantic' is used. Attempting to use 6392 the return value of a non-'void' function other than 'main' that 6393 flows off the end by reaching the closing curly brace that 6394 terminates the function is undefined. 6395 6396 Unlike in C, in C++, flowing off the end of a non-'void' function 6397 other than 'main' results in undefined behavior even when the value 6398 of the function is not used. 6399 6400 This warning is enabled by default in C++ and by '-Wall' otherwise. 6401 6402'-Wno-shift-count-negative' 6403 Controls warnings if a shift count is negative. This warning is 6404 enabled by default. 6405 6406'-Wno-shift-count-overflow' 6407 Controls warnings if a shift count is greater than or equal to the 6408 bit width of the type. This warning is enabled by default. 6409 6410'-Wshift-negative-value' 6411 Warn if left shifting a negative value. This warning is enabled by 6412 '-Wextra' in C99 (and newer) and C++11 to C++17 modes. 6413 6414'-Wno-shift-overflow' 6415'-Wshift-overflow=N' 6416 These options control warnings about left shift overflows. 6417 6418 '-Wshift-overflow=1' 6419 This is the warning level of '-Wshift-overflow' and is enabled 6420 by default in C99 and C++11 modes (and newer). This warning 6421 level does not warn about left-shifting 1 into the sign bit. 6422 (However, in C, such an overflow is still rejected in contexts 6423 where an integer constant expression is required.) No warning 6424 is emitted in C++20 mode (and newer), as signed left shifts 6425 always wrap. 6426 6427 '-Wshift-overflow=2' 6428 This warning level also warns about left-shifting 1 into the 6429 sign bit, unless C++14 mode (or newer) is active. 6430 6431'-Wswitch' 6432 Warn whenever a 'switch' statement has an index of enumerated type 6433 and lacks a 'case' for one or more of the named codes of that 6434 enumeration. (The presence of a 'default' label prevents this 6435 warning.) 'case' labels outside the enumeration range also provoke 6436 warnings when this option is used (even if there is a 'default' 6437 label). This warning is enabled by '-Wall'. 6438 6439'-Wswitch-default' 6440 Warn whenever a 'switch' statement does not have a 'default' case. 6441 6442'-Wswitch-enum' 6443 Warn whenever a 'switch' statement has an index of enumerated type 6444 and lacks a 'case' for one or more of the named codes of that 6445 enumeration. 'case' labels outside the enumeration range also 6446 provoke warnings when this option is used. The only difference 6447 between '-Wswitch' and this option is that this option gives a 6448 warning about an omitted enumeration code even if there is a 6449 'default' label. 6450 6451'-Wno-switch-bool' 6452 Do not warn when a 'switch' statement has an index of boolean type 6453 and the case values are outside the range of a boolean type. It is 6454 possible to suppress this warning by casting the controlling 6455 expression to a type other than 'bool'. For example: 6456 switch ((int) (a == 4)) 6457 { 6458 ... 6459 } 6460 This warning is enabled by default for C and C++ programs. 6461 6462'-Wno-switch-outside-range' 6463 This option controls warnings when a 'switch' case has a value that 6464 is outside of its respective type range. This warning is enabled 6465 by default for C and C++ programs. 6466 6467'-Wno-switch-unreachable' 6468 Do not warn when a 'switch' statement contains statements between 6469 the controlling expression and the first case label, which will 6470 never be executed. For example: 6471 switch (cond) 6472 { 6473 i = 15; 6474 ... 6475 case 5: 6476 ... 6477 } 6478 '-Wswitch-unreachable' does not warn if the statement between the 6479 controlling expression and the first case label is just a 6480 declaration: 6481 switch (cond) 6482 { 6483 int i; 6484 ... 6485 case 5: 6486 i = 5; 6487 ... 6488 } 6489 This warning is enabled by default for C and C++ programs. 6490 6491'-Wsync-nand (C and C++ only)' 6492 Warn when '__sync_fetch_and_nand' and '__sync_nand_and_fetch' 6493 built-in functions are used. These functions changed semantics in 6494 GCC 4.4. 6495 6496'-Wtrivial-auto-var-init' 6497 Warn when '-ftrivial-auto-var-init' cannot initialize the automatic 6498 variable. A common situation is an automatic variable that is 6499 declared between the controlling expression and the first case 6500 label of a 'switch' statement. 6501 6502'-Wunused-but-set-parameter' 6503 Warn whenever a function parameter is assigned to, but otherwise 6504 unused (aside from its declaration). 6505 6506 To suppress this warning use the 'unused' attribute (*note Variable 6507 Attributes::). 6508 6509 This warning is also enabled by '-Wunused' together with '-Wextra'. 6510 6511'-Wunused-but-set-variable' 6512 Warn whenever a local variable is assigned to, but otherwise unused 6513 (aside from its declaration). This warning is enabled by '-Wall'. 6514 6515 To suppress this warning use the 'unused' attribute (*note Variable 6516 Attributes::). 6517 6518 This warning is also enabled by '-Wunused', which is enabled by 6519 '-Wall'. 6520 6521'-Wunused-function' 6522 Warn whenever a static function is declared but not defined or a 6523 non-inline static function is unused. This warning is enabled by 6524 '-Wall'. 6525 6526'-Wunused-label' 6527 Warn whenever a label is declared but not used. This warning is 6528 enabled by '-Wall'. 6529 6530 To suppress this warning use the 'unused' attribute (*note Variable 6531 Attributes::). 6532 6533'-Wunused-local-typedefs (C, Objective-C, C++ and Objective-C++ only)' 6534 Warn when a typedef locally defined in a function is not used. 6535 This warning is enabled by '-Wall'. 6536 6537'-Wunused-parameter' 6538 Warn whenever a function parameter is unused aside from its 6539 declaration. 6540 6541 To suppress this warning use the 'unused' attribute (*note Variable 6542 Attributes::). 6543 6544'-Wno-unused-result' 6545 Do not warn if a caller of a function marked with attribute 6546 'warn_unused_result' (*note Function Attributes::) does not use its 6547 return value. The default is '-Wunused-result'. 6548 6549'-Wunused-variable' 6550 Warn whenever a local or static variable is unused aside from its 6551 declaration. This option implies '-Wunused-const-variable=1' for 6552 C, but not for C++. This warning is enabled by '-Wall'. 6553 6554 To suppress this warning use the 'unused' attribute (*note Variable 6555 Attributes::). 6556 6557'-Wunused-const-variable' 6558'-Wunused-const-variable=N' 6559 Warn whenever a constant static variable is unused aside from its 6560 declaration. '-Wunused-const-variable=1' is enabled by 6561 '-Wunused-variable' for C, but not for C++. In C this declares 6562 variable storage, but in C++ this is not an error since const 6563 variables take the place of '#define's. 6564 6565 To suppress this warning use the 'unused' attribute (*note Variable 6566 Attributes::). 6567 6568 '-Wunused-const-variable=1' 6569 This is the warning level that is enabled by 6570 '-Wunused-variable' for C. It warns only about unused static 6571 const variables defined in the main compilation unit, but not 6572 about static const variables declared in any header included. 6573 6574 '-Wunused-const-variable=2' 6575 This warning level also warns for unused constant static 6576 variables in headers (excluding system headers). This is the 6577 warning level of '-Wunused-const-variable' and must be 6578 explicitly requested since in C++ this isn't an error and in C 6579 it might be harder to clean up all headers included. 6580 6581'-Wunused-value' 6582 Warn whenever a statement computes a result that is explicitly not 6583 used. To suppress this warning cast the unused expression to 6584 'void'. This includes an expression-statement or the left-hand 6585 side of a comma expression that contains no side effects. For 6586 example, an expression such as 'x[i,j]' causes a warning, while 6587 'x[(void)i,j]' does not. 6588 6589 This warning is enabled by '-Wall'. 6590 6591'-Wunused' 6592 All the above '-Wunused' options combined. 6593 6594 In order to get a warning about an unused function parameter, you 6595 must either specify '-Wextra -Wunused' (note that '-Wall' implies 6596 '-Wunused'), or separately specify '-Wunused-parameter'. 6597 6598'-Wuninitialized' 6599 Warn if an object with automatic or allocated storage duration is 6600 used without having been initialized. In C++, also warn if a 6601 non-static reference or non-static 'const' member appears in a 6602 class without constructors. 6603 6604 In addition, passing a pointer (or in C++, a reference) to an 6605 uninitialized object to a 'const'-qualified argument of a built-in 6606 function known to read the object is also diagnosed by this 6607 warning. ('-Wmaybe-uninitialized' is issued for ordinary 6608 functions.) 6609 6610 If you want to warn about code that uses the uninitialized value of 6611 the variable in its own initializer, use the '-Winit-self' option. 6612 6613 These warnings occur for individual uninitialized elements of 6614 structure, union or array variables as well as for variables that 6615 are uninitialized as a whole. They do not occur for variables or 6616 elements declared 'volatile'. Because these warnings depend on 6617 optimization, the exact variables or elements for which there are 6618 warnings depend on the precise optimization options and version of 6619 GCC used. 6620 6621 Note that there may be no warning about a variable that is used 6622 only to compute a value that itself is never used, because such 6623 computations may be deleted by data flow analysis before the 6624 warnings are printed. 6625 6626 In C++, this warning also warns about using uninitialized objects 6627 in member-initializer-lists. For example, GCC warns about 'b' 6628 being uninitialized in the following snippet: 6629 6630 struct A { 6631 int a; 6632 int b; 6633 A() : a(b) { } 6634 }; 6635 6636'-Wno-invalid-memory-model' 6637 This option controls warnings for invocations of *note __atomic 6638 Builtins::, *note __sync Builtins::, and the C11 atomic generic 6639 functions with a memory consistency argument that is either invalid 6640 for the operation or outside the range of values of the 6641 'memory_order' enumeration. For example, since the 6642 '__atomic_store' and '__atomic_store_n' built-ins are only defined 6643 for the relaxed, release, and sequentially consistent memory orders 6644 the following code is diagnosed: 6645 6646 void store (int *i) 6647 { 6648 __atomic_store_n (i, 0, memory_order_consume); 6649 } 6650 6651 '-Winvalid-memory-model' is enabled by default. 6652 6653'-Wmaybe-uninitialized' 6654 For an object with automatic or allocated storage duration, if 6655 there exists a path from the function entry to a use of the object 6656 that is initialized, but there exist some other paths for which the 6657 object is not initialized, the compiler emits a warning if it 6658 cannot prove the uninitialized paths are not executed at run time. 6659 6660 In addition, passing a pointer (or in C++, a reference) to an 6661 uninitialized object to a 'const'-qualified function argument is 6662 also diagnosed by this warning. ('-Wuninitialized' is issued for 6663 built-in functions known to read the object.) Annotating the 6664 function with attribute 'access (none)' indicates that the argument 6665 isn't used to access the object and avoids the warning (*note 6666 Common Function Attributes::). 6667 6668 These warnings are only possible in optimizing compilation, because 6669 otherwise GCC does not keep track of the state of variables. 6670 6671 These warnings are made optional because GCC may not be able to 6672 determine when the code is correct in spite of appearing to have an 6673 error. Here is one example of how this can happen: 6674 6675 { 6676 int x; 6677 switch (y) 6678 { 6679 case 1: x = 1; 6680 break; 6681 case 2: x = 4; 6682 break; 6683 case 3: x = 5; 6684 } 6685 foo (x); 6686 } 6687 6688 If the value of 'y' is always 1, 2 or 3, then 'x' is always 6689 initialized, but GCC doesn't know this. To suppress the warning, 6690 you need to provide a default case with assert(0) or similar code. 6691 6692 This option also warns when a non-volatile automatic variable might 6693 be changed by a call to 'longjmp'. The compiler sees only the 6694 calls to 'setjmp'. It cannot know where 'longjmp' will be called; 6695 in fact, a signal handler could call it at any point in the code. 6696 As a result, you may get a warning even when there is in fact no 6697 problem because 'longjmp' cannot in fact be called at the place 6698 that would cause a problem. 6699 6700 Some spurious warnings can be avoided if you declare all the 6701 functions you use that never return as 'noreturn'. *Note Function 6702 Attributes::. 6703 6704 This warning is enabled by '-Wall' or '-Wextra'. 6705 6706'-Wunknown-pragmas' 6707 Warn when a '#pragma' directive is encountered that is not 6708 understood by GCC. If this command-line option is used, warnings 6709 are even issued for unknown pragmas in system header files. This 6710 is not the case if the warnings are only enabled by the '-Wall' 6711 command-line option. 6712 6713'-Wno-pragmas' 6714 Do not warn about misuses of pragmas, such as incorrect parameters, 6715 invalid syntax, or conflicts between pragmas. See also 6716 '-Wunknown-pragmas'. 6717 6718'-Wno-prio-ctor-dtor' 6719 Do not warn if a priority from 0 to 100 is used for constructor or 6720 destructor. The use of constructor and destructor attributes allow 6721 you to assign a priority to the constructor/destructor to control 6722 its order of execution before 'main' is called or after it returns. 6723 The priority values must be greater than 100 as the compiler 6724 reserves priority values between 0-100 for the implementation. 6725 6726'-Wstrict-aliasing' 6727 This option is only active when '-fstrict-aliasing' is active. It 6728 warns about code that might break the strict aliasing rules that 6729 the compiler is using for optimization. The warning does not catch 6730 all cases, but does attempt to catch the more common pitfalls. It 6731 is included in '-Wall'. It is equivalent to '-Wstrict-aliasing=3' 6732 6733'-Wstrict-aliasing=n' 6734 This option is only active when '-fstrict-aliasing' is active. It 6735 warns about code that might break the strict aliasing rules that 6736 the compiler is using for optimization. Higher levels correspond 6737 to higher accuracy (fewer false positives). Higher levels also 6738 correspond to more effort, similar to the way '-O' works. 6739 '-Wstrict-aliasing' is equivalent to '-Wstrict-aliasing=3'. 6740 6741 Level 1: Most aggressive, quick, least accurate. Possibly useful 6742 when higher levels do not warn but '-fstrict-aliasing' still breaks 6743 the code, as it has very few false negatives. However, it has many 6744 false positives. Warns for all pointer conversions between 6745 possibly incompatible types, even if never dereferenced. Runs in 6746 the front end only. 6747 6748 Level 2: Aggressive, quick, not too precise. May still have many 6749 false positives (not as many as level 1 though), and few false 6750 negatives (but possibly more than level 1). Unlike level 1, it 6751 only warns when an address is taken. Warns about incomplete types. 6752 Runs in the front end only. 6753 6754 Level 3 (default for '-Wstrict-aliasing'): Should have very few 6755 false positives and few false negatives. Slightly slower than 6756 levels 1 or 2 when optimization is enabled. Takes care of the 6757 common pun+dereference pattern in the front end: 6758 '*(int*)&some_float'. If optimization is enabled, it also runs in 6759 the back end, where it deals with multiple statement cases using 6760 flow-sensitive points-to information. Only warns when the 6761 converted pointer is dereferenced. Does not warn about incomplete 6762 types. 6763 6764'-Wstrict-overflow' 6765'-Wstrict-overflow=N' 6766 This option is only active when signed overflow is undefined. It 6767 warns about cases where the compiler optimizes based on the 6768 assumption that signed overflow does not occur. Note that it does 6769 not warn about all cases where the code might overflow: it only 6770 warns about cases where the compiler implements some optimization. 6771 Thus this warning depends on the optimization level. 6772 6773 An optimization that assumes that signed overflow does not occur is 6774 perfectly safe if the values of the variables involved are such 6775 that overflow never does, in fact, occur. Therefore this warning 6776 can easily give a false positive: a warning about code that is not 6777 actually a problem. To help focus on important issues, several 6778 warning levels are defined. No warnings are issued for the use of 6779 undefined signed overflow when estimating how many iterations a 6780 loop requires, in particular when determining whether a loop will 6781 be executed at all. 6782 6783 '-Wstrict-overflow=1' 6784 Warn about cases that are both questionable and easy to avoid. 6785 For example the compiler simplifies 'x + 1 > x' to '1'. This 6786 level of '-Wstrict-overflow' is enabled by '-Wall'; higher 6787 levels are not, and must be explicitly requested. 6788 6789 '-Wstrict-overflow=2' 6790 Also warn about other cases where a comparison is simplified 6791 to a constant. For example: 'abs (x) >= 0'. This can only be 6792 simplified when signed integer overflow is undefined, because 6793 'abs (INT_MIN)' overflows to 'INT_MIN', which is less than 6794 zero. '-Wstrict-overflow' (with no level) is the same as 6795 '-Wstrict-overflow=2'. 6796 6797 '-Wstrict-overflow=3' 6798 Also warn about other cases where a comparison is simplified. 6799 For example: 'x + 1 > 1' is simplified to 'x > 0'. 6800 6801 '-Wstrict-overflow=4' 6802 Also warn about other simplifications not covered by the above 6803 cases. For example: '(x * 10) / 5' is simplified to 'x * 2'. 6804 6805 '-Wstrict-overflow=5' 6806 Also warn about cases where the compiler reduces the magnitude 6807 of a constant involved in a comparison. For example: 'x + 2 > 6808 y' is simplified to 'x + 1 >= y'. This is reported only at 6809 the highest warning level because this simplification applies 6810 to many comparisons, so this warning level gives a very large 6811 number of false positives. 6812 6813'-Wstring-compare' 6814 Warn for calls to 'strcmp' and 'strncmp' whose result is determined 6815 to be either zero or non-zero in tests for such equality owing to 6816 the length of one argument being greater than the size of the array 6817 the other argument is stored in (or the bound in the case of 6818 'strncmp'). Such calls could be mistakes. For example, the call 6819 to 'strcmp' below is diagnosed because its result is necessarily 6820 non-zero irrespective of the contents of the array 'a'. 6821 6822 extern char a[4]; 6823 void f (char *d) 6824 { 6825 strcpy (d, "string"); 6826 ... 6827 if (0 == strcmp (a, d)) // cannot be true 6828 puts ("a and d are the same"); 6829 } 6830 6831 '-Wstring-compare' is enabled by '-Wextra'. 6832 6833'-Wno-stringop-overflow' 6834'-Wstringop-overflow' 6835'-Wstringop-overflow=TYPE' 6836 Warn for calls to string manipulation functions such as 'memcpy' 6837 and 'strcpy' that are determined to overflow the destination 6838 buffer. The optional argument is one greater than the type of 6839 Object Size Checking to perform to determine the size of the 6840 destination. *Note Object Size Checking::. The argument is 6841 meaningful only for functions that operate on character arrays but 6842 not for raw memory functions like 'memcpy' which always make use of 6843 Object Size type-0. The option also warns for calls that specify a 6844 size in excess of the largest possible object or at most 'SIZE_MAX 6845 / 2' bytes. The option produces the best results with optimization 6846 enabled but can detect a small subset of simple buffer overflows 6847 even without optimization in calls to the GCC built-in functions 6848 like '__builtin_memcpy' that correspond to the standard functions. 6849 In any case, the option warns about just a subset of buffer 6850 overflows detected by the corresponding overflow checking 6851 built-ins. For example, the option issues a warning for the 6852 'strcpy' call below because it copies at least 5 characters (the 6853 string '"blue"' including the terminating NUL) into the buffer of 6854 size 4. 6855 6856 enum Color { blue, purple, yellow }; 6857 const char* f (enum Color clr) 6858 { 6859 static char buf [4]; 6860 const char *str; 6861 switch (clr) 6862 { 6863 case blue: str = "blue"; break; 6864 case purple: str = "purple"; break; 6865 case yellow: str = "yellow"; break; 6866 } 6867 6868 return strcpy (buf, str); // warning here 6869 } 6870 6871 Option '-Wstringop-overflow=2' is enabled by default. 6872 6873 '-Wstringop-overflow' 6874 '-Wstringop-overflow=1' 6875 The '-Wstringop-overflow=1' option uses type-zero Object Size 6876 Checking to determine the sizes of destination objects. At 6877 this setting the option does not warn for writes past the end 6878 of subobjects of larger objects accessed by pointers unless 6879 the size of the largest surrounding object is known. When the 6880 destination may be one of several objects it is assumed to be 6881 the largest one of them. On Linux systems, when optimization 6882 is enabled at this setting the option warns for the same code 6883 as when the '_FORTIFY_SOURCE' macro is defined to a non-zero 6884 value. 6885 6886 '-Wstringop-overflow=2' 6887 The '-Wstringop-overflow=2' option uses type-one Object Size 6888 Checking to determine the sizes of destination objects. At 6889 this setting the option warns about overflows when writing to 6890 members of the largest complete objects whose exact size is 6891 known. However, it does not warn for excessive writes to the 6892 same members of unknown objects referenced by pointers since 6893 they may point to arrays containing unknown numbers of 6894 elements. This is the default setting of the option. 6895 6896 '-Wstringop-overflow=3' 6897 The '-Wstringop-overflow=3' option uses type-two Object Size 6898 Checking to determine the sizes of destination objects. At 6899 this setting the option warns about overflowing the smallest 6900 object or data member. This is the most restrictive setting 6901 of the option that may result in warnings for safe code. 6902 6903 '-Wstringop-overflow=4' 6904 The '-Wstringop-overflow=4' option uses type-three Object Size 6905 Checking to determine the sizes of destination objects. At 6906 this setting the option warns about overflowing any data 6907 members, and when the destination is one of several objects it 6908 uses the size of the largest of them to decide whether to 6909 issue a warning. Similarly to '-Wstringop-overflow=3' this 6910 setting of the option may result in warnings for benign code. 6911 6912'-Wno-stringop-overread' 6913 Warn for calls to string manipulation functions such as 'memchr', 6914 or 'strcpy' that are determined to read past the end of the source 6915 sequence. 6916 6917 Option '-Wstringop-overread' is enabled by default. 6918 6919'-Wno-stringop-truncation' 6920 Do not warn for calls to bounded string manipulation functions such 6921 as 'strncat', 'strncpy', and 'stpncpy' that may either truncate the 6922 copied string or leave the destination unchanged. 6923 6924 In the following example, the call to 'strncat' specifies a bound 6925 that is less than the length of the source string. As a result, 6926 the copy of the source will be truncated and so the call is 6927 diagnosed. To avoid the warning use 'bufsize - strlen (buf) - 1)' 6928 as the bound. 6929 6930 void append (char *buf, size_t bufsize) 6931 { 6932 strncat (buf, ".txt", 3); 6933 } 6934 6935 As another example, the following call to 'strncpy' results in 6936 copying to 'd' just the characters preceding the terminating NUL, 6937 without appending the NUL to the end. Assuming the result of 6938 'strncpy' is necessarily a NUL-terminated string is a common 6939 mistake, and so the call is diagnosed. To avoid the warning when 6940 the result is not expected to be NUL-terminated, call 'memcpy' 6941 instead. 6942 6943 void copy (char *d, const char *s) 6944 { 6945 strncpy (d, s, strlen (s)); 6946 } 6947 6948 In the following example, the call to 'strncpy' specifies the size 6949 of the destination buffer as the bound. If the length of the 6950 source string is equal to or greater than this size the result of 6951 the copy will not be NUL-terminated. Therefore, the call is also 6952 diagnosed. To avoid the warning, specify 'sizeof buf - 1' as the 6953 bound and set the last element of the buffer to 'NUL'. 6954 6955 void copy (const char *s) 6956 { 6957 char buf[80]; 6958 strncpy (buf, s, sizeof buf); 6959 ... 6960 } 6961 6962 In situations where a character array is intended to store a 6963 sequence of bytes with no terminating 'NUL' such an array may be 6964 annotated with attribute 'nonstring' to avoid this warning. Such 6965 arrays, however, are not suitable arguments to functions that 6966 expect 'NUL'-terminated strings. To help detect accidental misuses 6967 of such arrays GCC issues warnings unless it can prove that the use 6968 is safe. *Note Common Variable Attributes::. 6969 6970'-Wsuggest-attribute=[pure|const|noreturn|format|cold|malloc]' 6971 Warn for cases where adding an attribute may be beneficial. The 6972 attributes currently supported are listed below. 6973 6974 '-Wsuggest-attribute=pure' 6975 '-Wsuggest-attribute=const' 6976 '-Wsuggest-attribute=noreturn' 6977 '-Wmissing-noreturn' 6978 '-Wsuggest-attribute=malloc' 6979 6980 Warn about functions that might be candidates for attributes 6981 'pure', 'const' or 'noreturn' or 'malloc'. The compiler only 6982 warns for functions visible in other compilation units or (in 6983 the case of 'pure' and 'const') if it cannot prove that the 6984 function returns normally. A function returns normally if it 6985 doesn't contain an infinite loop or return abnormally by 6986 throwing, calling 'abort' or trapping. This analysis requires 6987 option '-fipa-pure-const', which is enabled by default at '-O' 6988 and higher. Higher optimization levels improve the accuracy 6989 of the analysis. 6990 6991 '-Wsuggest-attribute=format' 6992 '-Wmissing-format-attribute' 6993 6994 Warn about function pointers that might be candidates for 6995 'format' attributes. Note these are only possible candidates, 6996 not absolute ones. GCC guesses that function pointers with 6997 'format' attributes that are used in assignment, 6998 initialization, parameter passing or return statements should 6999 have a corresponding 'format' attribute in the resulting type. 7000 I.e. the left-hand side of the assignment or initialization, 7001 the type of the parameter variable, or the return type of the 7002 containing function respectively should also have a 'format' 7003 attribute to avoid the warning. 7004 7005 GCC also warns about function definitions that might be 7006 candidates for 'format' attributes. Again, these are only 7007 possible candidates. GCC guesses that 'format' attributes 7008 might be appropriate for any function that calls a function 7009 like 'vprintf' or 'vscanf', but this might not always be the 7010 case, and some functions for which 'format' attributes are 7011 appropriate may not be detected. 7012 7013 '-Wsuggest-attribute=cold' 7014 7015 Warn about functions that might be candidates for 'cold' 7016 attribute. This is based on static detection and generally 7017 only warns about functions which always leads to a call to 7018 another 'cold' function such as wrappers of C++ 'throw' or 7019 fatal error reporting functions leading to 'abort'. 7020 7021'-Walloc-zero' 7022 Warn about calls to allocation functions decorated with attribute 7023 'alloc_size' that specify zero bytes, including those to the 7024 built-in forms of the functions 'aligned_alloc', 'alloca', 7025 'calloc', 'malloc', and 'realloc'. Because the behavior of these 7026 functions when called with a zero size differs among 7027 implementations (and in the case of 'realloc' has been deprecated) 7028 relying on it may result in subtle portability bugs and should be 7029 avoided. 7030 7031'-Walloc-size-larger-than=BYTE-SIZE' 7032 Warn about calls to functions decorated with attribute 'alloc_size' 7033 that attempt to allocate objects larger than the specified number 7034 of bytes, or where the result of the size computation in an integer 7035 type with infinite precision would exceed the value of 7036 'PTRDIFF_MAX' on the target. 7037 '-Walloc-size-larger-than=''PTRDIFF_MAX' is enabled by default. 7038 Warnings controlled by the option can be disabled either by 7039 specifying BYTE-SIZE of 'SIZE_MAX' or more or by 7040 '-Wno-alloc-size-larger-than'. *Note Function Attributes::. 7041 7042'-Wno-alloc-size-larger-than' 7043 Disable '-Walloc-size-larger-than=' warnings. The option is 7044 equivalent to '-Walloc-size-larger-than=''SIZE_MAX' or larger. 7045 7046'-Walloca' 7047 This option warns on all uses of 'alloca' in the source. 7048 7049'-Walloca-larger-than=BYTE-SIZE' 7050 This option warns on calls to 'alloca' with an integer argument 7051 whose value is either zero, or that is not bounded by a controlling 7052 predicate that limits its value to at most BYTE-SIZE. It also 7053 warns for calls to 'alloca' where the bound value is unknown. 7054 Arguments of non-integer types are considered unbounded even if 7055 they appear to be constrained to the expected range. 7056 7057 For example, a bounded case of 'alloca' could be: 7058 7059 void func (size_t n) 7060 { 7061 void *p; 7062 if (n <= 1000) 7063 p = alloca (n); 7064 else 7065 p = malloc (n); 7066 f (p); 7067 } 7068 7069 In the above example, passing '-Walloca-larger-than=1000' would not 7070 issue a warning because the call to 'alloca' is known to be at most 7071 1000 bytes. However, if '-Walloca-larger-than=500' were passed, 7072 the compiler would emit a warning. 7073 7074 Unbounded uses, on the other hand, are uses of 'alloca' with no 7075 controlling predicate constraining its integer argument. For 7076 example: 7077 7078 void func () 7079 { 7080 void *p = alloca (n); 7081 f (p); 7082 } 7083 7084 If '-Walloca-larger-than=500' were passed, the above would trigger 7085 a warning, but this time because of the lack of bounds checking. 7086 7087 Note, that even seemingly correct code involving signed integers 7088 could cause a warning: 7089 7090 void func (signed int n) 7091 { 7092 if (n < 500) 7093 { 7094 p = alloca (n); 7095 f (p); 7096 } 7097 } 7098 7099 In the above example, N could be negative, causing a larger than 7100 expected argument to be implicitly cast into the 'alloca' call. 7101 7102 This option also warns when 'alloca' is used in a loop. 7103 7104 '-Walloca-larger-than=''PTRDIFF_MAX' is enabled by default but is 7105 usually only effective when '-ftree-vrp' is active (default for 7106 '-O2' and above). 7107 7108 See also '-Wvla-larger-than=''byte-size'. 7109 7110'-Wno-alloca-larger-than' 7111 Disable '-Walloca-larger-than=' warnings. The option is equivalent 7112 to '-Walloca-larger-than=''SIZE_MAX' or larger. 7113 7114'-Warith-conversion' 7115 Do warn about implicit conversions from arithmetic operations even 7116 when conversion of the operands to the same type cannot change 7117 their values. This affects warnings from '-Wconversion', 7118 '-Wfloat-conversion', and '-Wsign-conversion'. 7119 7120 void f (char c, int i) 7121 { 7122 c = c + i; // warns with -Wconversion 7123 c = c + 1; // only warns with -Warith-conversion 7124 } 7125 7126'-Warray-bounds' 7127'-Warray-bounds=N' 7128 Warn about out of bounds subscripts or offsets into arrays. This 7129 warning is enabled by '-Wall'. It is more effective when 7130 '-ftree-vrp' is active (the default for '-O2' and above) but a 7131 subset of instances are issued even without optimization. 7132 7133 '-Warray-bounds=1' 7134 This is the default warning level of '-Warray-bounds' and is 7135 enabled by '-Wall'; higher levels are not, and must be 7136 explicitly requested. 7137 7138 '-Warray-bounds=2' 7139 This warning level also warns about out of bounds accesses to 7140 trailing struct members of one-element array types (*note Zero 7141 Length::) and about the intermediate results of pointer 7142 arithmetic that may yield out of bounds values. This warning 7143 level may give a larger number of false positives and is 7144 deactivated by default. 7145 7146'-Warray-compare' 7147 Warn about equality and relational comparisons between two operands 7148 of array type. This comparison was deprecated in C++20. For 7149 example: 7150 7151 int arr1[5]; 7152 int arr2[5]; 7153 bool same = arr1 == arr2; 7154 7155 '-Warray-compare' is enabled by '-Wall'. 7156 7157'-Warray-parameter' 7158'-Warray-parameter=N' 7159 Warn about redeclarations of functions involving arguments of array 7160 or pointer types of inconsistent kinds or forms, and enable the 7161 detection of out-of-bounds accesses to such parameters by warnings 7162 such as '-Warray-bounds'. 7163 7164 If the first function declaration uses the array form the bound 7165 specified in the array is assumed to be the minimum number of 7166 elements expected to be provided in calls to the function and the 7167 maximum number of elements accessed by it. Failing to provide 7168 arguments of sufficient size or accessing more than the maximum 7169 number of elements may be diagnosed by warnings such as 7170 '-Warray-bounds'. At level 1 the warning diagnoses inconsistencies 7171 involving array parameters declared using the 'T[static N]' form. 7172 7173 For example, the warning triggers for the following redeclarations 7174 because the first one allows an array of any size to be passed to 7175 'f' while the second one with the keyword 'static' specifies that 7176 the array argument must have at least four elements. 7177 7178 void f (int[static 4]); 7179 void f (int[]); // warning (inconsistent array form) 7180 7181 void g (void) 7182 { 7183 int *p = (int *)malloc (4); 7184 f (p); // warning (array too small) 7185 ... 7186 } 7187 7188 At level 2 the warning also triggers for redeclarations involving 7189 any other inconsistency in array or pointer argument forms denoting 7190 array sizes. Pointers and arrays of unspecified bound are 7191 considered equivalent and do not trigger a warning. 7192 7193 void g (int*); 7194 void g (int[]); // no warning 7195 void g (int[8]); // warning (inconsistent array bound) 7196 7197 '-Warray-parameter=2' is included in '-Wall'. The 7198 '-Wvla-parameter' option triggers warnings for similar 7199 inconsistencies involving Variable Length Array arguments. 7200 7201'-Wattribute-alias=N' 7202'-Wno-attribute-alias' 7203 Warn about declarations using the 'alias' and similar attributes 7204 whose target is incompatible with the type of the alias. *Note 7205 Declaring Attributes of Functions: Function Attributes. 7206 7207 '-Wattribute-alias=1' 7208 The default warning level of the '-Wattribute-alias' option 7209 diagnoses incompatibilities between the type of the alias 7210 declaration and that of its target. Such incompatibilities 7211 are typically indicative of bugs. 7212 7213 '-Wattribute-alias=2' 7214 7215 At this level '-Wattribute-alias' also diagnoses cases where 7216 the attributes of the alias declaration are more restrictive 7217 than the attributes applied to its target. These mismatches 7218 can potentially result in incorrect code generation. In other 7219 cases they may be benign and could be resolved simply by 7220 adding the missing attribute to the target. For comparison, 7221 see the '-Wmissing-attributes' option, which controls 7222 diagnostics when the alias declaration is less restrictive 7223 than the target, rather than more restrictive. 7224 7225 Attributes considered include 'alloc_align', 'alloc_size', 7226 'cold', 'const', 'hot', 'leaf', 'malloc', 'nonnull', 7227 'noreturn', 'nothrow', 'pure', 'returns_nonnull', and 7228 'returns_twice'. 7229 7230 '-Wattribute-alias' is equivalent to '-Wattribute-alias=1'. This 7231 is the default. You can disable these warnings with either 7232 '-Wno-attribute-alias' or '-Wattribute-alias=0'. 7233 7234'-Wbidi-chars=[none|unpaired|any|ucn]' 7235 Warn about possibly misleading UTF-8 bidirectional control 7236 characters in comments, string literals, character constants, and 7237 identifiers. Such characters can change left-to-right writing 7238 direction into right-to-left (and vice versa), which can cause 7239 confusion between the logical order and visual order. This may be 7240 dangerous; for instance, it may seem that a piece of code is not 7241 commented out, whereas it in fact is. 7242 7243 There are three levels of warning supported by GCC. The default is 7244 '-Wbidi-chars=unpaired', which warns about improperly terminated 7245 bidi contexts. '-Wbidi-chars=none' turns the warning off. 7246 '-Wbidi-chars=any' warns about any use of bidirectional control 7247 characters. 7248 7249 By default, this warning does not warn about UCNs. It is, however, 7250 possible to turn on such checking by using 7251 '-Wbidi-chars=unpaired,ucn' or '-Wbidi-chars=any,ucn'. Using 7252 '-Wbidi-chars=ucn' is valid, and is equivalent to 7253 '-Wbidi-chars=unpaired,ucn', if no previous '-Wbidi-chars=any' was 7254 specified. 7255 7256'-Wbool-compare' 7257 Warn about boolean expression compared with an integer value 7258 different from 'true'/'false'. For instance, the following 7259 comparison is always false: 7260 int n = 5; 7261 ... 7262 if ((n > 1) == 2) { ... } 7263 This warning is enabled by '-Wall'. 7264 7265'-Wbool-operation' 7266 Warn about suspicious operations on expressions of a boolean type. 7267 For instance, bitwise negation of a boolean is very likely a bug in 7268 the program. For C, this warning also warns about incrementing or 7269 decrementing a boolean, which rarely makes sense. (In C++, 7270 decrementing a boolean is always invalid. Incrementing a boolean 7271 is invalid in C++17, and deprecated otherwise.) 7272 7273 This warning is enabled by '-Wall'. 7274 7275'-Wduplicated-branches' 7276 Warn when an if-else has identical branches. This warning detects 7277 cases like 7278 if (p != NULL) 7279 return 0; 7280 else 7281 return 0; 7282 It doesn't warn when both branches contain just a null statement. 7283 This warning also warn for conditional operators: 7284 int i = x ? *p : *p; 7285 7286'-Wduplicated-cond' 7287 Warn about duplicated conditions in an if-else-if chain. For 7288 instance, warn for the following code: 7289 if (p->q != NULL) { ... } 7290 else if (p->q != NULL) { ... } 7291 7292'-Wframe-address' 7293 Warn when the '__builtin_frame_address' or 7294 '__builtin_return_address' is called with an argument greater than 7295 0. Such calls may return indeterminate values or crash the 7296 program. The warning is included in '-Wall'. 7297 7298'-Wno-discarded-qualifiers (C and Objective-C only)' 7299 Do not warn if type qualifiers on pointers are being discarded. 7300 Typically, the compiler warns if a 'const char *' variable is 7301 passed to a function that takes a 'char *' parameter. This option 7302 can be used to suppress such a warning. 7303 7304'-Wno-discarded-array-qualifiers (C and Objective-C only)' 7305 Do not warn if type qualifiers on arrays which are pointer targets 7306 are being discarded. Typically, the compiler warns if a 'const int 7307 (*)[]' variable is passed to a function that takes a 'int (*)[]' 7308 parameter. This option can be used to suppress such a warning. 7309 7310'-Wno-incompatible-pointer-types (C and Objective-C only)' 7311 Do not warn when there is a conversion between pointers that have 7312 incompatible types. This warning is for cases not covered by 7313 '-Wno-pointer-sign', which warns for pointer argument passing or 7314 assignment with different signedness. 7315 7316'-Wno-int-conversion (C and Objective-C only)' 7317 Do not warn about incompatible integer to pointer and pointer to 7318 integer conversions. This warning is about implicit conversions; 7319 for explicit conversions the warnings '-Wno-int-to-pointer-cast' 7320 and '-Wno-pointer-to-int-cast' may be used. 7321 7322'-Wzero-length-bounds' 7323 Warn about accesses to elements of zero-length array members that 7324 might overlap other members of the same object. Declaring interior 7325 zero-length arrays is discouraged because accesses to them are 7326 undefined. See *Note Zero Length::. 7327 7328 For example, the first two stores in function 'bad' are diagnosed 7329 because the array elements overlap the subsequent members 'b' and 7330 'c'. The third store is diagnosed by '-Warray-bounds' because it 7331 is beyond the bounds of the enclosing object. 7332 7333 struct X { int a[0]; int b, c; }; 7334 struct X x; 7335 7336 void bad (void) 7337 { 7338 x.a[0] = 0; // -Wzero-length-bounds 7339 x.a[1] = 1; // -Wzero-length-bounds 7340 x.a[2] = 2; // -Warray-bounds 7341 } 7342 7343 Option '-Wzero-length-bounds' is enabled by '-Warray-bounds'. 7344 7345'-Wno-div-by-zero' 7346 Do not warn about compile-time integer division by zero. 7347 Floating-point division by zero is not warned about, as it can be a 7348 legitimate way of obtaining infinities and NaNs. 7349 7350'-Wsystem-headers' 7351 Print warning messages for constructs found in system header files. 7352 Warnings from system headers are normally suppressed, on the 7353 assumption that they usually do not indicate real problems and 7354 would only make the compiler output harder to read. Using this 7355 command-line option tells GCC to emit warnings from system headers 7356 as if they occurred in user code. However, note that using '-Wall' 7357 in conjunction with this option does _not_ warn about unknown 7358 pragmas in system headers--for that, '-Wunknown-pragmas' must also 7359 be used. 7360 7361'-Wtautological-compare' 7362 Warn if a self-comparison always evaluates to true or false. This 7363 warning detects various mistakes such as: 7364 int i = 1; 7365 ... 7366 if (i > i) { ... } 7367 7368 This warning also warns about bitwise comparisons that always 7369 evaluate to true or false, for instance: 7370 if ((a & 16) == 10) { ... } 7371 will always be false. 7372 7373 This warning is enabled by '-Wall'. 7374 7375'-Wtrampolines' 7376 Warn about trampolines generated for pointers to nested functions. 7377 A trampoline is a small piece of data or code that is created at 7378 run time on the stack when the address of a nested function is 7379 taken, and is used to call the nested function indirectly. For 7380 some targets, it is made up of data only and thus requires no 7381 special treatment. But, for most targets, it is made up of code 7382 and thus requires the stack to be made executable in order for the 7383 program to work properly. 7384 7385'-Wfloat-equal' 7386 Warn if floating-point values are used in equality comparisons. 7387 7388 The idea behind this is that sometimes it is convenient (for the 7389 programmer) to consider floating-point values as approximations to 7390 infinitely precise real numbers. If you are doing this, then you 7391 need to compute (by analyzing the code, or in some other way) the 7392 maximum or likely maximum error that the computation introduces, 7393 and allow for it when performing comparisons (and when producing 7394 output, but that's a different problem). In particular, instead of 7395 testing for equality, you should check to see whether the two 7396 values have ranges that overlap; and this is done with the 7397 relational operators, so equality comparisons are probably 7398 mistaken. 7399 7400'-Wtraditional (C and Objective-C only)' 7401 Warn about certain constructs that behave differently in 7402 traditional and ISO C. Also warn about ISO C constructs that have 7403 no traditional C equivalent, and/or problematic constructs that 7404 should be avoided. 7405 7406 * Macro parameters that appear within string literals in the 7407 macro body. In traditional C macro replacement takes place 7408 within string literals, but in ISO C it does not. 7409 7410 * In traditional C, some preprocessor directives did not exist. 7411 Traditional preprocessors only considered a line to be a 7412 directive if the '#' appeared in column 1 on the line. 7413 Therefore '-Wtraditional' warns about directives that 7414 traditional C understands but ignores because the '#' does not 7415 appear as the first character on the line. It also suggests 7416 you hide directives like '#pragma' not understood by 7417 traditional C by indenting them. Some traditional 7418 implementations do not recognize '#elif', so this option 7419 suggests avoiding it altogether. 7420 7421 * A function-like macro that appears without arguments. 7422 7423 * The unary plus operator. 7424 7425 * The 'U' integer constant suffix, or the 'F' or 'L' 7426 floating-point constant suffixes. (Traditional C does support 7427 the 'L' suffix on integer constants.) Note, these suffixes 7428 appear in macros defined in the system headers of most modern 7429 systems, e.g. the '_MIN'/'_MAX' macros in '<limits.h>'. Use 7430 of these macros in user code might normally lead to spurious 7431 warnings, however GCC's integrated preprocessor has enough 7432 context to avoid warning in these cases. 7433 7434 * A function declared external in one block and then used after 7435 the end of the block. 7436 7437 * A 'switch' statement has an operand of type 'long'. 7438 7439 * A non-'static' function declaration follows a 'static' one. 7440 This construct is not accepted by some traditional C 7441 compilers. 7442 7443 * The ISO type of an integer constant has a different width or 7444 signedness from its traditional type. This warning is only 7445 issued if the base of the constant is ten. I.e. hexadecimal 7446 or octal values, which typically represent bit patterns, are 7447 not warned about. 7448 7449 * Usage of ISO string concatenation is detected. 7450 7451 * Initialization of automatic aggregates. 7452 7453 * Identifier conflicts with labels. Traditional C lacks a 7454 separate namespace for labels. 7455 7456 * Initialization of unions. If the initializer is zero, the 7457 warning is omitted. This is done under the assumption that 7458 the zero initializer in user code appears conditioned on e.g. 7459 '__STDC__' to avoid missing initializer warnings and relies on 7460 default initialization to zero in the traditional C case. 7461 7462 * Conversions by prototypes between fixed/floating-point values 7463 and vice versa. The absence of these prototypes when 7464 compiling with traditional C causes serious problems. This is 7465 a subset of the possible conversion warnings; for the full set 7466 use '-Wtraditional-conversion'. 7467 7468 * Use of ISO C style function definitions. This warning 7469 intentionally is _not_ issued for prototype declarations or 7470 variadic functions because these ISO C features appear in your 7471 code when using libiberty's traditional C compatibility 7472 macros, 'PARAMS' and 'VPARAMS'. This warning is also bypassed 7473 for nested functions because that feature is already a GCC 7474 extension and thus not relevant to traditional C 7475 compatibility. 7476 7477'-Wtraditional-conversion (C and Objective-C only)' 7478 Warn if a prototype causes a type conversion that is different from 7479 what would happen to the same argument in the absence of a 7480 prototype. This includes conversions of fixed point to floating 7481 and vice versa, and conversions changing the width or signedness of 7482 a fixed-point argument except when the same as the default 7483 promotion. 7484 7485'-Wdeclaration-after-statement (C and Objective-C only)' 7486 Warn when a declaration is found after a statement in a block. 7487 This construct, known from C++, was introduced with ISO C99 and is 7488 by default allowed in GCC. It is not supported by ISO C90. *Note 7489 Mixed Labels and Declarations::. 7490 7491'-Wshadow' 7492 Warn whenever a local variable or type declaration shadows another 7493 variable, parameter, type, class member (in C++), or instance 7494 variable (in Objective-C) or whenever a built-in function is 7495 shadowed. Note that in C++, the compiler warns if a local variable 7496 shadows an explicit typedef, but not if it shadows a 7497 struct/class/enum. If this warning is enabled, it includes also 7498 all instances of local shadowing. This means that 7499 '-Wno-shadow=local' and '-Wno-shadow=compatible-local' are ignored 7500 when '-Wshadow' is used. Same as '-Wshadow=global'. 7501 7502'-Wno-shadow-ivar (Objective-C only)' 7503 Do not warn whenever a local variable shadows an instance variable 7504 in an Objective-C method. 7505 7506'-Wshadow=global' 7507 Warn for any shadowing. Same as '-Wshadow'. 7508 7509'-Wshadow=local' 7510 Warn when a local variable shadows another local variable or 7511 parameter. 7512 7513'-Wshadow=compatible-local' 7514 Warn when a local variable shadows another local variable or 7515 parameter whose type is compatible with that of the shadowing 7516 variable. In C++, type compatibility here means the type of the 7517 shadowing variable can be converted to that of the shadowed 7518 variable. The creation of this flag (in addition to 7519 '-Wshadow=local') is based on the idea that when a local variable 7520 shadows another one of incompatible type, it is most likely 7521 intentional, not a bug or typo, as shown in the following example: 7522 7523 for (SomeIterator i = SomeObj.begin(); i != SomeObj.end(); ++i) 7524 { 7525 for (int i = 0; i < N; ++i) 7526 { 7527 ... 7528 } 7529 ... 7530 } 7531 7532 Since the two variable 'i' in the example above have incompatible 7533 types, enabling only '-Wshadow=compatible-local' does not emit a 7534 warning. Because their types are incompatible, if a programmer 7535 accidentally uses one in place of the other, type checking is 7536 expected to catch that and emit an error or warning. Use of this 7537 flag instead of '-Wshadow=local' can possibly reduce the number of 7538 warnings triggered by intentional shadowing. Note that this also 7539 means that shadowing 'const char *i' by 'char *i' does not emit a 7540 warning. 7541 7542 This warning is also enabled by '-Wshadow=local'. 7543 7544'-Wlarger-than=BYTE-SIZE' 7545 Warn whenever an object is defined whose size exceeds BYTE-SIZE. 7546 '-Wlarger-than=''PTRDIFF_MAX' is enabled by default. Warnings 7547 controlled by the option can be disabled either by specifying 7548 BYTE-SIZE of 'SIZE_MAX' or more or by '-Wno-larger-than'. 7549 7550 Also warn for calls to bounded functions such as 'memchr' or 7551 'strnlen' that specify a bound greater than the largest possible 7552 object, which is 'PTRDIFF_MAX' bytes by default. These warnings 7553 can only be disabled by '-Wno-larger-than'. 7554 7555'-Wno-larger-than' 7556 Disable '-Wlarger-than=' warnings. The option is equivalent to 7557 '-Wlarger-than=''SIZE_MAX' or larger. 7558 7559'-Wframe-larger-than=BYTE-SIZE' 7560 Warn if the size of a function frame exceeds BYTE-SIZE. The 7561 computation done to determine the stack frame size is approximate 7562 and not conservative. The actual requirements may be somewhat 7563 greater than BYTE-SIZE even if you do not get a warning. In 7564 addition, any space allocated via 'alloca', variable-length arrays, 7565 or related constructs is not included by the compiler when 7566 determining whether or not to issue a warning. 7567 '-Wframe-larger-than=''PTRDIFF_MAX' is enabled by default. 7568 Warnings controlled by the option can be disabled either by 7569 specifying BYTE-SIZE of 'SIZE_MAX' or more or by 7570 '-Wno-frame-larger-than'. 7571 7572'-Wno-frame-larger-than' 7573 Disable '-Wframe-larger-than=' warnings. The option is equivalent 7574 to '-Wframe-larger-than=''SIZE_MAX' or larger. 7575 7576'-Wfree-nonheap-object' 7577 Warn when attempting to deallocate an object that was either not 7578 allocated on the heap, or by using a pointer that was not returned 7579 from a prior call to the corresponding allocation function. For 7580 example, because the call to 'stpcpy' returns a pointer to the 7581 terminating nul character and not to the beginning of the object, 7582 the call to 'free' below is diagnosed. 7583 7584 void f (char *p) 7585 { 7586 p = stpcpy (p, "abc"); 7587 // ... 7588 free (p); // warning 7589 } 7590 7591 '-Wfree-nonheap-object' is included in '-Wall'. 7592 7593'-Wstack-usage=BYTE-SIZE' 7594 Warn if the stack usage of a function might exceed BYTE-SIZE. The 7595 computation done to determine the stack usage is conservative. Any 7596 space allocated via 'alloca', variable-length arrays, or related 7597 constructs is included by the compiler when determining whether or 7598 not to issue a warning. 7599 7600 The message is in keeping with the output of '-fstack-usage'. 7601 7602 * If the stack usage is fully static but exceeds the specified 7603 amount, it's: 7604 7605 warning: stack usage is 1120 bytes 7606 * If the stack usage is (partly) dynamic but bounded, it's: 7607 7608 warning: stack usage might be 1648 bytes 7609 * If the stack usage is (partly) dynamic and not bounded, it's: 7610 7611 warning: stack usage might be unbounded 7612 7613 '-Wstack-usage=''PTRDIFF_MAX' is enabled by default. Warnings 7614 controlled by the option can be disabled either by specifying 7615 BYTE-SIZE of 'SIZE_MAX' or more or by '-Wno-stack-usage'. 7616 7617'-Wno-stack-usage' 7618 Disable '-Wstack-usage=' warnings. The option is equivalent to 7619 '-Wstack-usage=''SIZE_MAX' or larger. 7620 7621'-Wunsafe-loop-optimizations' 7622 Warn if the loop cannot be optimized because the compiler cannot 7623 assume anything on the bounds of the loop indices. With 7624 '-funsafe-loop-optimizations' warn if the compiler makes such 7625 assumptions. 7626 7627'-Wno-pedantic-ms-format (MinGW targets only)' 7628 When used in combination with '-Wformat' and '-pedantic' without 7629 GNU extensions, this option disables the warnings about non-ISO 7630 'printf' / 'scanf' format width specifiers 'I32', 'I64', and 'I' 7631 used on Windows targets, which depend on the MS runtime. 7632 7633'-Wpointer-arith' 7634 Warn about anything that depends on the "size of" a function type 7635 or of 'void'. GNU C assigns these types a size of 1, for 7636 convenience in calculations with 'void *' pointers and pointers to 7637 functions. In C++, warn also when an arithmetic operation involves 7638 'NULL'. This warning is also enabled by '-Wpedantic'. 7639 7640'-Wno-pointer-compare' 7641 Do not warn if a pointer is compared with a zero character 7642 constant. This usually means that the pointer was meant to be 7643 dereferenced. For example: 7644 7645 const char *p = foo (); 7646 if (p == '\0') 7647 return 42; 7648 7649 Note that the code above is invalid in C++11. 7650 7651 This warning is enabled by default. 7652 7653'-Wtsan' 7654 Warn about unsupported features in ThreadSanitizer. 7655 7656 ThreadSanitizer does not support 'std::atomic_thread_fence' and can 7657 report false positives. 7658 7659 This warning is enabled by default. 7660 7661'-Wtype-limits' 7662 Warn if a comparison is always true or always false due to the 7663 limited range of the data type, but do not warn for constant 7664 expressions. For example, warn if an unsigned variable is compared 7665 against zero with '<' or '>='. This warning is also enabled by 7666 '-Wextra'. 7667 7668'-Wabsolute-value (C and Objective-C only)' 7669 Warn for calls to standard functions that compute the absolute 7670 value of an argument when a more appropriate standard function is 7671 available. For example, calling 'abs(3.14)' triggers the warning 7672 because the appropriate function to call to compute the absolute 7673 value of a double argument is 'fabs'. The option also triggers 7674 warnings when the argument in a call to such a function has an 7675 unsigned type. This warning can be suppressed with an explicit 7676 type cast and it is also enabled by '-Wextra'. 7677 7678'-Wcomment' 7679'-Wcomments' 7680 Warn whenever a comment-start sequence '/*' appears in a '/*' 7681 comment, or whenever a backslash-newline appears in a '//' comment. 7682 This warning is enabled by '-Wall'. 7683 7684'-Wtrigraphs' 7685 Warn if any trigraphs are encountered that might change the meaning 7686 of the program. Trigraphs within comments are not warned about, 7687 except those that would form escaped newlines. 7688 7689 This option is implied by '-Wall'. If '-Wall' is not given, this 7690 option is still enabled unless trigraphs are enabled. To get 7691 trigraph conversion without warnings, but get the other '-Wall' 7692 warnings, use '-trigraphs -Wall -Wno-trigraphs'. 7693 7694'-Wundef' 7695 Warn if an undefined identifier is evaluated in an '#if' directive. 7696 Such identifiers are replaced with zero. 7697 7698'-Wexpansion-to-defined' 7699 Warn whenever 'defined' is encountered in the expansion of a macro 7700 (including the case where the macro is expanded by an '#if' 7701 directive). Such usage is not portable. This warning is also 7702 enabled by '-Wpedantic' and '-Wextra'. 7703 7704'-Wunused-macros' 7705 Warn about macros defined in the main file that are unused. A 7706 macro is "used" if it is expanded or tested for existence at least 7707 once. The preprocessor also warns if the macro has not been used 7708 at the time it is redefined or undefined. 7709 7710 Built-in macros, macros defined on the command line, and macros 7711 defined in include files are not warned about. 7712 7713 _Note:_ If a macro is actually used, but only used in skipped 7714 conditional blocks, then the preprocessor reports it as unused. To 7715 avoid the warning in such a case, you might improve the scope of 7716 the macro's definition by, for example, moving it into the first 7717 skipped block. Alternatively, you could provide a dummy use with 7718 something like: 7719 7720 #if defined the_macro_causing_the_warning 7721 #endif 7722 7723'-Wno-endif-labels' 7724 Do not warn whenever an '#else' or an '#endif' are followed by 7725 text. This sometimes happens in older programs with code of the 7726 form 7727 7728 #if FOO 7729 ... 7730 #else FOO 7731 ... 7732 #endif FOO 7733 7734 The second and third 'FOO' should be in comments. This warning is 7735 on by default. 7736 7737'-Wbad-function-cast (C and Objective-C only)' 7738 Warn when a function call is cast to a non-matching type. For 7739 example, warn if a call to a function returning an integer type is 7740 cast to a pointer type. 7741 7742'-Wc90-c99-compat (C and Objective-C only)' 7743 Warn about features not present in ISO C90, but present in ISO C99. 7744 For instance, warn about use of variable length arrays, 'long long' 7745 type, 'bool' type, compound literals, designated initializers, and 7746 so on. This option is independent of the standards mode. Warnings 7747 are disabled in the expression that follows '__extension__'. 7748 7749'-Wc99-c11-compat (C and Objective-C only)' 7750 Warn about features not present in ISO C99, but present in ISO C11. 7751 For instance, warn about use of anonymous structures and unions, 7752 '_Atomic' type qualifier, '_Thread_local' storage-class specifier, 7753 '_Alignas' specifier, 'Alignof' operator, '_Generic' keyword, and 7754 so on. This option is independent of the standards mode. Warnings 7755 are disabled in the expression that follows '__extension__'. 7756 7757'-Wc11-c2x-compat (C and Objective-C only)' 7758 Warn about features not present in ISO C11, but present in ISO C2X. 7759 For instance, warn about omitting the string in '_Static_assert', 7760 use of '[[]]' syntax for attributes, use of decimal floating-point 7761 types, and so on. This option is independent of the standards 7762 mode. Warnings are disabled in the expression that follows 7763 '__extension__'. 7764 7765'-Wc++-compat (C and Objective-C only)' 7766 Warn about ISO C constructs that are outside of the common subset 7767 of ISO C and ISO C++, e.g. request for implicit conversion from 7768 'void *' to a pointer to non-'void' type. 7769 7770'-Wc++11-compat (C++ and Objective-C++ only)' 7771 Warn about C++ constructs whose meaning differs between ISO C++ 7772 1998 and ISO C++ 2011, e.g., identifiers in ISO C++ 1998 that are 7773 keywords in ISO C++ 2011. This warning turns on '-Wnarrowing' and 7774 is enabled by '-Wall'. 7775 7776'-Wc++14-compat (C++ and Objective-C++ only)' 7777 Warn about C++ constructs whose meaning differs between ISO C++ 7778 2011 and ISO C++ 2014. This warning is enabled by '-Wall'. 7779 7780'-Wc++17-compat (C++ and Objective-C++ only)' 7781 Warn about C++ constructs whose meaning differs between ISO C++ 7782 2014 and ISO C++ 2017. This warning is enabled by '-Wall'. 7783 7784'-Wc++20-compat (C++ and Objective-C++ only)' 7785 Warn about C++ constructs whose meaning differs between ISO C++ 7786 2017 and ISO C++ 2020. This warning is enabled by '-Wall'. 7787 7788'-Wno-c++11-extensions (C++ and Objective-C++ only)' 7789 Do not warn about C++11 constructs in code being compiled using an 7790 older C++ standard. Even without this option, some C++11 7791 constructs will only be diagnosed if '-Wpedantic' is used. 7792 7793'-Wno-c++14-extensions (C++ and Objective-C++ only)' 7794 Do not warn about C++14 constructs in code being compiled using an 7795 older C++ standard. Even without this option, some C++14 7796 constructs will only be diagnosed if '-Wpedantic' is used. 7797 7798'-Wno-c++17-extensions (C++ and Objective-C++ only)' 7799 Do not warn about C++17 constructs in code being compiled using an 7800 older C++ standard. Even without this option, some C++17 7801 constructs will only be diagnosed if '-Wpedantic' is used. 7802 7803'-Wno-c++20-extensions (C++ and Objective-C++ only)' 7804 Do not warn about C++20 constructs in code being compiled using an 7805 older C++ standard. Even without this option, some C++20 7806 constructs will only be diagnosed if '-Wpedantic' is used. 7807 7808'-Wno-c++23-extensions (C++ and Objective-C++ only)' 7809 Do not warn about C++23 constructs in code being compiled using an 7810 older C++ standard. Even without this option, some C++23 7811 constructs will only be diagnosed if '-Wpedantic' is used. 7812 7813'-Wcast-qual' 7814 Warn whenever a pointer is cast so as to remove a type qualifier 7815 from the target type. For example, warn if a 'const char *' is 7816 cast to an ordinary 'char *'. 7817 7818 Also warn when making a cast that introduces a type qualifier in an 7819 unsafe way. For example, casting 'char **' to 'const char **' is 7820 unsafe, as in this example: 7821 7822 /* p is char ** value. */ 7823 const char **q = (const char **) p; 7824 /* Assignment of readonly string to const char * is OK. */ 7825 *q = "string"; 7826 /* Now char** pointer points to read-only memory. */ 7827 **p = 'b'; 7828 7829'-Wcast-align' 7830 Warn whenever a pointer is cast such that the required alignment of 7831 the target is increased. For example, warn if a 'char *' is cast 7832 to an 'int *' on machines where integers can only be accessed at 7833 two- or four-byte boundaries. 7834 7835'-Wcast-align=strict' 7836 Warn whenever a pointer is cast such that the required alignment of 7837 the target is increased. For example, warn if a 'char *' is cast 7838 to an 'int *' regardless of the target machine. 7839 7840'-Wcast-function-type' 7841 Warn when a function pointer is cast to an incompatible function 7842 pointer. In a cast involving function types with a variable 7843 argument list only the types of initial arguments that are provided 7844 are considered. Any parameter of pointer-type matches any other 7845 pointer-type. Any benign differences in integral types are 7846 ignored, like 'int' vs. 'long' on ILP32 targets. Likewise type 7847 qualifiers are ignored. The function type 'void (*) (void)' is 7848 special and matches everything, which can be used to suppress this 7849 warning. In a cast involving pointer to member types this warning 7850 warns whenever the type cast is changing the pointer to member 7851 type. This warning is enabled by '-Wextra'. 7852 7853'-Wwrite-strings' 7854 When compiling C, give string constants the type 'const 7855 char[LENGTH]' so that copying the address of one into a non-'const' 7856 'char *' pointer produces a warning. These warnings help you find 7857 at compile time code that can try to write into a string constant, 7858 but only if you have been very careful about using 'const' in 7859 declarations and prototypes. Otherwise, it is just a nuisance. 7860 This is why we did not make '-Wall' request these warnings. 7861 7862 When compiling C++, warn about the deprecated conversion from 7863 string literals to 'char *'. This warning is enabled by default 7864 for C++ programs. 7865 7866'-Wclobbered' 7867 Warn for variables that might be changed by 'longjmp' or 'vfork'. 7868 This warning is also enabled by '-Wextra'. 7869 7870'-Wconversion' 7871 Warn for implicit conversions that may alter a value. This 7872 includes conversions between real and integer, like 'abs (x)' when 7873 'x' is 'double'; conversions between signed and unsigned, like 7874 'unsigned ui = -1'; and conversions to smaller types, like 'sqrtf 7875 (M_PI)'. Do not warn for explicit casts like 'abs ((int) x)' and 7876 'ui = (unsigned) -1', or if the value is not changed by the 7877 conversion like in 'abs (2.0)'. Warnings about conversions between 7878 signed and unsigned integers can be disabled by using 7879 '-Wno-sign-conversion'. 7880 7881 For C++, also warn for confusing overload resolution for 7882 user-defined conversions; and conversions that never use a type 7883 conversion operator: conversions to 'void', the same type, a base 7884 class or a reference to them. Warnings about conversions between 7885 signed and unsigned integers are disabled by default in C++ unless 7886 '-Wsign-conversion' is explicitly enabled. 7887 7888 Warnings about conversion from arithmetic on a small type back to 7889 that type are only given with '-Warith-conversion'. 7890 7891'-Wdangling-else' 7892 Warn about constructions where there may be confusion to which 'if' 7893 statement an 'else' branch belongs. Here is an example of such a 7894 case: 7895 7896 { 7897 if (a) 7898 if (b) 7899 foo (); 7900 else 7901 bar (); 7902 } 7903 7904 In C/C++, every 'else' branch belongs to the innermost possible 7905 'if' statement, which in this example is 'if (b)'. This is often 7906 not what the programmer expected, as illustrated in the above 7907 example by indentation the programmer chose. When there is the 7908 potential for this confusion, GCC issues a warning when this flag 7909 is specified. To eliminate the warning, add explicit braces around 7910 the innermost 'if' statement so there is no way the 'else' can 7911 belong to the enclosing 'if'. The resulting code looks like this: 7912 7913 { 7914 if (a) 7915 { 7916 if (b) 7917 foo (); 7918 else 7919 bar (); 7920 } 7921 } 7922 7923 This warning is enabled by '-Wparentheses'. 7924 7925'-Wdangling-pointer' 7926'-Wdangling-pointer=N' 7927 Warn about uses of pointers (or C++ references) to objects with 7928 automatic storage duration after their lifetime has ended. This 7929 includes local variables declared in nested blocks, compound 7930 literals and other unnamed temporary objects. In addition, warn 7931 about storing the address of such objects in escaped pointers. The 7932 warning is enabled at all optimization levels but may yield 7933 different results with optimization than without. 7934 7935 '-Wdangling-pointer=1' 7936 At level 1 the warning diagnoses only unconditional uses of 7937 dangling pointers. For example 7938 int f (int c1, int c2, x) 7939 { 7940 char *p = strchr ((char[]){ c1, c2 }, c3); 7941 return p ? *p : 'x'; // warning: dangling pointer to a compound literal 7942 } 7943 In the following function the store of the address of the 7944 local variable 'x' in the escaped pointer '*p' also triggers 7945 the warning. 7946 void g (int **p) 7947 { 7948 int x = 7; 7949 *p = &x; // warning: storing the address of a local variable in *p 7950 } 7951 7952 '-Wdangling-pointer=2' 7953 At level 2, in addition to unconditional uses the warning also 7954 diagnoses conditional uses of dangling pointers. 7955 7956 For example, because the array A in the following function is 7957 out of scope when the pointer S that was set to point is used, 7958 the warning triggers at this level. 7959 7960 void f (char *s) 7961 { 7962 if (!s) 7963 { 7964 char a[12] = "tmpname"; 7965 s = a; 7966 } 7967 strcat (s, ".tmp"); // warning: dangling pointer to a may be used 7968 ... 7969 } 7970 7971 '-Wdangling-pointer=2' is included in '-Wall'. 7972 7973'-Wdate-time' 7974 Warn when macros '__TIME__', '__DATE__' or '__TIMESTAMP__' are 7975 encountered as they might prevent bit-wise-identical reproducible 7976 compilations. 7977 7978'-Wempty-body' 7979 Warn if an empty body occurs in an 'if', 'else' or 'do while' 7980 statement. This warning is also enabled by '-Wextra'. 7981 7982'-Wno-endif-labels' 7983 Do not warn about stray tokens after '#else' and '#endif'. 7984 7985'-Wenum-compare' 7986 Warn about a comparison between values of different enumerated 7987 types. In C++ enumerated type mismatches in conditional 7988 expressions are also diagnosed and the warning is enabled by 7989 default. In C this warning is enabled by '-Wall'. 7990 7991'-Wenum-conversion' 7992 Warn when a value of enumerated type is implicitly converted to a 7993 different enumerated type. This warning is enabled by '-Wextra' in 7994 C. 7995 7996'-Wjump-misses-init (C, Objective-C only)' 7997 Warn if a 'goto' statement or a 'switch' statement jumps forward 7998 across the initialization of a variable, or jumps backward to a 7999 label after the variable has been initialized. This only warns 8000 about variables that are initialized when they are declared. This 8001 warning is only supported for C and Objective-C; in C++ this sort 8002 of branch is an error in any case. 8003 8004 '-Wjump-misses-init' is included in '-Wc++-compat'. It can be 8005 disabled with the '-Wno-jump-misses-init' option. 8006 8007'-Wsign-compare' 8008 Warn when a comparison between signed and unsigned values could 8009 produce an incorrect result when the signed value is converted to 8010 unsigned. In C++, this warning is also enabled by '-Wall'. In C, 8011 it is also enabled by '-Wextra'. 8012 8013'-Wsign-conversion' 8014 Warn for implicit conversions that may change the sign of an 8015 integer value, like assigning a signed integer expression to an 8016 unsigned integer variable. An explicit cast silences the warning. 8017 In C, this option is enabled also by '-Wconversion'. 8018 8019'-Wfloat-conversion' 8020 Warn for implicit conversions that reduce the precision of a real 8021 value. This includes conversions from real to integer, and from 8022 higher precision real to lower precision real values. This option 8023 is also enabled by '-Wconversion'. 8024 8025'-Wno-scalar-storage-order' 8026 Do not warn on suspicious constructs involving reverse scalar 8027 storage order. 8028 8029'-Wsizeof-array-div' 8030 Warn about divisions of two sizeof operators when the first one is 8031 applied to an array and the divisor does not equal the size of the 8032 array element. In such a case, the computation will not yield the 8033 number of elements in the array, which is likely what the user 8034 intended. This warning warns e.g. about 8035 int fn () 8036 { 8037 int arr[10]; 8038 return sizeof (arr) / sizeof (short); 8039 } 8040 8041 This warning is enabled by '-Wall'. 8042 8043'-Wsizeof-pointer-div' 8044 Warn for suspicious divisions of two sizeof expressions that divide 8045 the pointer size by the element size, which is the usual way to 8046 compute the array size but won't work out correctly with pointers. 8047 This warning warns e.g. about 'sizeof (ptr) / sizeof (ptr[0])' if 8048 'ptr' is not an array, but a pointer. This warning is enabled by 8049 '-Wall'. 8050 8051'-Wsizeof-pointer-memaccess' 8052 Warn for suspicious length parameters to certain string and memory 8053 built-in functions if the argument uses 'sizeof'. This warning 8054 triggers for example for 'memset (ptr, 0, sizeof (ptr));' if 'ptr' 8055 is not an array, but a pointer, and suggests a possible fix, or 8056 about 'memcpy (&foo, ptr, sizeof (&foo));'. 8057 '-Wsizeof-pointer-memaccess' also warns about calls to bounded 8058 string copy functions like 'strncat' or 'strncpy' that specify as 8059 the bound a 'sizeof' expression of the source array. For example, 8060 in the following function the call to 'strncat' specifies the size 8061 of the source string as the bound. That is almost certainly a 8062 mistake and so the call is diagnosed. 8063 void make_file (const char *name) 8064 { 8065 char path[PATH_MAX]; 8066 strncpy (path, name, sizeof path - 1); 8067 strncat (path, ".text", sizeof ".text"); 8068 ... 8069 } 8070 8071 The '-Wsizeof-pointer-memaccess' option is enabled by '-Wall'. 8072 8073'-Wno-sizeof-array-argument' 8074 Do not warn when the 'sizeof' operator is applied to a parameter 8075 that is declared as an array in a function definition. This 8076 warning is enabled by default for C and C++ programs. 8077 8078'-Wmemset-elt-size' 8079 Warn for suspicious calls to the 'memset' built-in function, if the 8080 first argument references an array, and the third argument is a 8081 number equal to the number of elements, but not equal to the size 8082 of the array in memory. This indicates that the user has omitted a 8083 multiplication by the element size. This warning is enabled by 8084 '-Wall'. 8085 8086'-Wmemset-transposed-args' 8087 Warn for suspicious calls to the 'memset' built-in function where 8088 the second argument is not zero and the third argument is zero. 8089 For example, the call 'memset (buf, sizeof buf, 0)' is diagnosed 8090 because 'memset (buf, 0, sizeof buf)' was meant instead. The 8091 diagnostic is only emitted if the third argument is a literal zero. 8092 Otherwise, if it is an expression that is folded to zero, or a cast 8093 of zero to some type, it is far less likely that the arguments have 8094 been mistakenly transposed and no warning is emitted. This warning 8095 is enabled by '-Wall'. 8096 8097'-Waddress' 8098 Warn about suspicious uses of address expressions. These include 8099 comparing the address of a function or a declared object to the 8100 null pointer constant such as in 8101 void f (void); 8102 void g (void) 8103 { 8104 if (!func) // warning: expression evaluates to false 8105 abort (); 8106 } 8107 comparisons of a pointer to a string literal, such as in 8108 void f (const char *x) 8109 { 8110 if (x == "abc") // warning: expression evaluates to false 8111 puts ("equal"); 8112 } 8113 and tests of the results of pointer addition or subtraction for 8114 equality to null, such as in 8115 void f (const int *p, int i) 8116 { 8117 return p + i == NULL; 8118 } 8119 Such uses typically indicate a programmer error: the address of 8120 most functions and objects necessarily evaluates to true (the 8121 exception are weak symbols), so their use in a conditional might 8122 indicate missing parentheses in a function call or a missing 8123 dereference in an array expression. The subset of the warning for 8124 object pointers can be suppressed by casting the pointer operand to 8125 an integer type such as 'inptr_t' or 'uinptr_t'. Comparisons 8126 against string literals result in unspecified behavior and are not 8127 portable, and suggest the intent was to call 'strcmp'. The warning 8128 is suppressed if the suspicious expression is the result of macro 8129 expansion. '-Waddress' warning is enabled by '-Wall'. 8130 8131'-Wno-address-of-packed-member' 8132 Do not warn when the address of packed member of struct or union is 8133 taken, which usually results in an unaligned pointer value. This 8134 is enabled by default. 8135 8136'-Wlogical-op' 8137 Warn about suspicious uses of logical operators in expressions. 8138 This includes using logical operators in contexts where a bit-wise 8139 operator is likely to be expected. Also warns when the operands of 8140 a logical operator are the same: 8141 extern int a; 8142 if (a < 0 && a < 0) { ... } 8143 8144'-Wlogical-not-parentheses' 8145 Warn about logical not used on the left hand side operand of a 8146 comparison. This option does not warn if the right operand is 8147 considered to be a boolean expression. Its purpose is to detect 8148 suspicious code like the following: 8149 int a; 8150 ... 8151 if (!a > 1) { ... } 8152 8153 It is possible to suppress the warning by wrapping the LHS into 8154 parentheses: 8155 if ((!a) > 1) { ... } 8156 8157 This warning is enabled by '-Wall'. 8158 8159'-Waggregate-return' 8160 Warn if any functions that return structures or unions are defined 8161 or called. (In languages where you can return an array, this also 8162 elicits a warning.) 8163 8164'-Wno-aggressive-loop-optimizations' 8165 Warn if in a loop with constant number of iterations the compiler 8166 detects undefined behavior in some statement during one or more of 8167 the iterations. 8168 8169'-Wno-attributes' 8170 Do not warn if an unexpected '__attribute__' is used, such as 8171 unrecognized attributes, function attributes applied to variables, 8172 etc. This does not stop errors for incorrect use of supported 8173 attributes. 8174 8175 Additionally, using '-Wno-attributes=', it is possible to suppress 8176 warnings about unknown scoped attributes (in C++11 and C2X). For 8177 example, '-Wno-attributes=vendor::attr' disables warning about the 8178 following declaration: 8179 8180 [[vendor::attr]] void f(); 8181 8182 It is also possible to disable warning about all attributes in a 8183 namespace using '-Wno-attributes=vendor::' which prevents warning 8184 about both of these declarations: 8185 8186 [[vendor::safe]] void f(); 8187 [[vendor::unsafe]] void f2(); 8188 8189 Note that '-Wno-attributes=' does not imply '-Wno-attributes'. 8190 8191'-Wno-builtin-declaration-mismatch' 8192 Warn if a built-in function is declared with an incompatible 8193 signature or as a non-function, or when a built-in function 8194 declared with a type that does not include a prototype is called 8195 with arguments whose promoted types do not match those expected by 8196 the function. When '-Wextra' is specified, also warn when a 8197 built-in function that takes arguments is declared without a 8198 prototype. The '-Wbuiltin-declaration-mismatch' warning is enabled 8199 by default. To avoid the warning include the appropriate header to 8200 bring the prototypes of built-in functions into scope. 8201 8202 For example, the call to 'memset' below is diagnosed by the warning 8203 because the function expects a value of type 'size_t' as its 8204 argument but the type of '32' is 'int'. With '-Wextra', the 8205 declaration of the function is diagnosed as well. 8206 extern void* memset (); 8207 void f (void *d) 8208 { 8209 memset (d, '\0', 32); 8210 } 8211 8212'-Wno-builtin-macro-redefined' 8213 Do not warn if certain built-in macros are redefined. This 8214 suppresses warnings for redefinition of '__TIMESTAMP__', 8215 '__TIME__', '__DATE__', '__FILE__', and '__BASE_FILE__'. 8216 8217'-Wstrict-prototypes (C and Objective-C only)' 8218 Warn if a function is declared or defined without specifying the 8219 argument types. (An old-style function definition is permitted 8220 without a warning if preceded by a declaration that specifies the 8221 argument types.) 8222 8223'-Wold-style-declaration (C and Objective-C only)' 8224 Warn for obsolescent usages, according to the C Standard, in a 8225 declaration. For example, warn if storage-class specifiers like 8226 'static' are not the first things in a declaration. This warning 8227 is also enabled by '-Wextra'. 8228 8229'-Wold-style-definition (C and Objective-C only)' 8230 Warn if an old-style function definition is used. A warning is 8231 given even if there is a previous prototype. A definition using 8232 '()' is not considered an old-style definition in C2X mode, because 8233 it is equivalent to '(void)' in that case, but is considered an 8234 old-style definition for older standards. 8235 8236'-Wmissing-parameter-type (C and Objective-C only)' 8237 A function parameter is declared without a type specifier in 8238 K&R-style functions: 8239 8240 void foo(bar) { } 8241 8242 This warning is also enabled by '-Wextra'. 8243 8244'-Wmissing-prototypes (C and Objective-C only)' 8245 Warn if a global function is defined without a previous prototype 8246 declaration. This warning is issued even if the definition itself 8247 provides a prototype. Use this option to detect global functions 8248 that do not have a matching prototype declaration in a header file. 8249 This option is not valid for C++ because all function declarations 8250 provide prototypes and a non-matching declaration declares an 8251 overload rather than conflict with an earlier declaration. Use 8252 '-Wmissing-declarations' to detect missing declarations in C++. 8253 8254'-Wmissing-declarations' 8255 Warn if a global function is defined without a previous 8256 declaration. Do so even if the definition itself provides a 8257 prototype. Use this option to detect global functions that are not 8258 declared in header files. In C, no warnings are issued for 8259 functions with previous non-prototype declarations; use 8260 '-Wmissing-prototypes' to detect missing prototypes. In C++, no 8261 warnings are issued for function templates, or for inline 8262 functions, or for functions in anonymous namespaces. 8263 8264'-Wmissing-field-initializers' 8265 Warn if a structure's initializer has some fields missing. For 8266 example, the following code causes such a warning, because 'x.h' is 8267 implicitly zero: 8268 8269 struct s { int f, g, h; }; 8270 struct s x = { 3, 4 }; 8271 8272 This option does not warn about designated initializers, so the 8273 following modification does not trigger a warning: 8274 8275 struct s { int f, g, h; }; 8276 struct s x = { .f = 3, .g = 4 }; 8277 8278 In C this option does not warn about the universal zero initializer 8279 '{ 0 }': 8280 8281 struct s { int f, g, h; }; 8282 struct s x = { 0 }; 8283 8284 Likewise, in C++ this option does not warn about the empty { } 8285 initializer, for example: 8286 8287 struct s { int f, g, h; }; 8288 s x = { }; 8289 8290 This warning is included in '-Wextra'. To get other '-Wextra' 8291 warnings without this one, use '-Wextra 8292 -Wno-missing-field-initializers'. 8293 8294'-Wno-missing-requires' 8295 8296 By default, the compiler warns about a concept-id appearing as a 8297 C++20 simple-requirement: 8298 8299 bool satisfied = requires { C<T> }; 8300 8301 Here 'satisfied' will be true if 'C<T>' is a valid expression, 8302 which it is for all T. Presumably the user meant to write 8303 8304 bool satisfied = requires { requires C<T> }; 8305 8306 so 'satisfied' is only true if concept 'C' is satisfied for type 8307 'T'. 8308 8309 This warning can be disabled with '-Wno-missing-requires'. 8310 8311'-Wno-missing-template-keyword' 8312 8313 The member access tokens ., -> and :: must be followed by the 8314 'template' keyword if the parent object is dependent and the member 8315 being named is a template. 8316 8317 template <class X> 8318 void DoStuff (X x) 8319 { 8320 x.template DoSomeOtherStuff<X>(); // Good. 8321 x.DoMoreStuff<X>(); // Warning, x is dependent. 8322 } 8323 8324 In rare cases it is possible to get false positives. To silence 8325 this, wrap the expression in parentheses. For example, the 8326 following is treated as a template, even where m and N are 8327 integers: 8328 8329 void NotATemplate (my_class t) 8330 { 8331 int N = 5; 8332 8333 bool test = t.m < N > (0); // Treated as a template. 8334 test = (t.m < N) > (0); // Same meaning, but not treated as a template. 8335 } 8336 8337 This warning can be disabled with '-Wno-missing-template-keyword'. 8338 8339'-Wno-multichar' 8340 Do not warn if a multicharacter constant (''FOOF'') is used. 8341 Usually they indicate a typo in the user's code, as they have 8342 implementation-defined values, and should not be used in portable 8343 code. 8344 8345'-Wnormalized=[none|id|nfc|nfkc]' 8346 In ISO C and ISO C++, two identifiers are different if they are 8347 different sequences of characters. However, sometimes when 8348 characters outside the basic ASCII character set are used, you can 8349 have two different character sequences that look the same. To 8350 avoid confusion, the ISO 10646 standard sets out some 8351 "normalization rules" which when applied ensure that two sequences 8352 that look the same are turned into the same sequence. GCC can warn 8353 you if you are using identifiers that have not been normalized; 8354 this option controls that warning. 8355 8356 There are four levels of warning supported by GCC. The default is 8357 '-Wnormalized=nfc', which warns about any identifier that is not in 8358 the ISO 10646 "C" normalized form, "NFC". NFC is the recommended 8359 form for most uses. It is equivalent to '-Wnormalized'. 8360 8361 Unfortunately, there are some characters allowed in identifiers by 8362 ISO C and ISO C++ that, when turned into NFC, are not allowed in 8363 identifiers. That is, there's no way to use these symbols in 8364 portable ISO C or C++ and have all your identifiers in NFC. 8365 '-Wnormalized=id' suppresses the warning for these characters. It 8366 is hoped that future versions of the standards involved will 8367 correct this, which is why this option is not the default. 8368 8369 You can switch the warning off for all characters by writing 8370 '-Wnormalized=none' or '-Wno-normalized'. You should only do this 8371 if you are using some other normalization scheme (like "D"), 8372 because otherwise you can easily create bugs that are literally 8373 impossible to see. 8374 8375 Some characters in ISO 10646 have distinct meanings but look 8376 identical in some fonts or display methodologies, especially once 8377 formatting has been applied. For instance '\u207F', "SUPERSCRIPT 8378 LATIN SMALL LETTER N", displays just like a regular 'n' that has 8379 been placed in a superscript. ISO 10646 defines the "NFKC" 8380 normalization scheme to convert all these into a standard form as 8381 well, and GCC warns if your code is not in NFKC if you use 8382 '-Wnormalized=nfkc'. This warning is comparable to warning about 8383 every identifier that contains the letter O because it might be 8384 confused with the digit 0, and so is not the default, but may be 8385 useful as a local coding convention if the programming environment 8386 cannot be fixed to display these characters distinctly. 8387 8388'-Wno-attribute-warning' 8389 Do not warn about usage of functions (*note Function Attributes::) 8390 declared with 'warning' attribute. By default, this warning is 8391 enabled. '-Wno-attribute-warning' can be used to disable the 8392 warning or '-Wno-error=attribute-warning' can be used to disable 8393 the error when compiled with '-Werror' flag. 8394 8395'-Wno-deprecated' 8396 Do not warn about usage of deprecated features. *Note Deprecated 8397 Features::. 8398 8399'-Wno-deprecated-declarations' 8400 Do not warn about uses of functions (*note Function Attributes::), 8401 variables (*note Variable Attributes::), and types (*note Type 8402 Attributes::) marked as deprecated by using the 'deprecated' 8403 attribute. 8404 8405'-Wno-overflow' 8406 Do not warn about compile-time overflow in constant expressions. 8407 8408'-Wno-odr' 8409 Warn about One Definition Rule violations during link-time 8410 optimization. Enabled by default. 8411 8412'-Wopenacc-parallelism' 8413 Warn about potentially suboptimal choices related to OpenACC 8414 parallelism. 8415 8416'-Wopenmp-simd' 8417 Warn if the vectorizer cost model overrides the OpenMP simd 8418 directive set by user. The '-fsimd-cost-model=unlimited' option 8419 can be used to relax the cost model. 8420 8421'-Woverride-init (C and Objective-C only)' 8422 Warn if an initialized field without side effects is overridden 8423 when using designated initializers (*note Designated Initializers: 8424 Designated Inits.). 8425 8426 This warning is included in '-Wextra'. To get other '-Wextra' 8427 warnings without this one, use '-Wextra -Wno-override-init'. 8428 8429'-Wno-override-init-side-effects (C and Objective-C only)' 8430 Do not warn if an initialized field with side effects is overridden 8431 when using designated initializers (*note Designated Initializers: 8432 Designated Inits.). This warning is enabled by default. 8433 8434'-Wpacked' 8435 Warn if a structure is given the packed attribute, but the packed 8436 attribute has no effect on the layout or size of the structure. 8437 Such structures may be mis-aligned for little benefit. For 8438 instance, in this code, the variable 'f.x' in 'struct bar' is 8439 misaligned even though 'struct bar' does not itself have the packed 8440 attribute: 8441 8442 struct foo { 8443 int x; 8444 char a, b, c, d; 8445 } __attribute__((packed)); 8446 struct bar { 8447 char z; 8448 struct foo f; 8449 }; 8450 8451'-Wnopacked-bitfield-compat' 8452 The 4.1, 4.2 and 4.3 series of GCC ignore the 'packed' attribute on 8453 bit-fields of type 'char'. This was fixed in GCC 4.4 but the 8454 change can lead to differences in the structure layout. GCC 8455 informs you when the offset of such a field has changed in GCC 4.4. 8456 For example there is no longer a 4-bit padding between field 'a' 8457 and 'b' in this structure: 8458 8459 struct foo 8460 { 8461 char a:4; 8462 char b:8; 8463 } __attribute__ ((packed)); 8464 8465 This warning is enabled by default. Use 8466 '-Wno-packed-bitfield-compat' to disable this warning. 8467 8468'-Wpacked-not-aligned (C, C++, Objective-C and Objective-C++ only)' 8469 Warn if a structure field with explicitly specified alignment in a 8470 packed struct or union is misaligned. For example, a warning will 8471 be issued on 'struct S', like, 'warning: alignment 1 of 'struct S' 8472 is less than 8', in this code: 8473 8474 struct __attribute__ ((aligned (8))) S8 { char a[8]; }; 8475 struct __attribute__ ((packed)) S { 8476 struct S8 s8; 8477 }; 8478 8479 This warning is enabled by '-Wall'. 8480 8481'-Wpadded' 8482 Warn if padding is included in a structure, either to align an 8483 element of the structure or to align the whole structure. 8484 Sometimes when this happens it is possible to rearrange the fields 8485 of the structure to reduce the padding and so make the structure 8486 smaller. 8487 8488'-Wredundant-decls' 8489 Warn if anything is declared more than once in the same scope, even 8490 in cases where multiple declaration is valid and changes nothing. 8491 8492'-Wrestrict' 8493 Warn when an object referenced by a 'restrict'-qualified parameter 8494 (or, in C++, a '__restrict'-qualified parameter) is aliased by 8495 another argument, or when copies between such objects overlap. For 8496 example, the call to the 'strcpy' function below attempts to 8497 truncate the string by replacing its initial characters with the 8498 last four. However, because the call writes the terminating NUL 8499 into 'a[4]', the copies overlap and the call is diagnosed. 8500 8501 void foo (void) 8502 { 8503 char a[] = "abcd1234"; 8504 strcpy (a, a + 4); 8505 ... 8506 } 8507 The '-Wrestrict' option detects some instances of simple overlap 8508 even without optimization but works best at '-O2' and above. It is 8509 included in '-Wall'. 8510 8511'-Wnested-externs (C and Objective-C only)' 8512 Warn if an 'extern' declaration is encountered within a function. 8513 8514'-Winline' 8515 Warn if a function that is declared as inline cannot be inlined. 8516 Even with this option, the compiler does not warn about failures to 8517 inline functions declared in system headers. 8518 8519 The compiler uses a variety of heuristics to determine whether or 8520 not to inline a function. For example, the compiler takes into 8521 account the size of the function being inlined and the amount of 8522 inlining that has already been done in the current function. 8523 Therefore, seemingly insignificant changes in the source program 8524 can cause the warnings produced by '-Winline' to appear or 8525 disappear. 8526 8527'-Winterference-size' 8528 Warn about use of C++17 8529 'std::hardware_destructive_interference_size' without specifying 8530 its value with '--param destructive-interference-size'. Also warn 8531 about questionable values for that option. 8532 8533 This variable is intended to be used for controlling class layout, 8534 to avoid false sharing in concurrent code: 8535 8536 struct independent_fields { 8537 alignas(std::hardware_destructive_interference_size) std::atomic<int> one; 8538 alignas(std::hardware_destructive_interference_size) std::atomic<int> two; 8539 }; 8540 8541 Here 'one' and 'two' are intended to be far enough apart that 8542 stores to one won't require accesses to the other to reload the 8543 cache line. 8544 8545 By default, '--param destructive-interference-size' and '--param 8546 constructive-interference-size' are set based on the current 8547 '-mtune' option, typically to the L1 cache line size for the 8548 particular target CPU, sometimes to a range if tuning for a generic 8549 target. So all translation units that depend on ABI compatibility 8550 for the use of these variables must be compiled with the same 8551 '-mtune' (or '-mcpu'). 8552 8553 If ABI stability is important, such as if the use is in a header 8554 for a library, you should probably not use the hardware 8555 interference size variables at all. Alternatively, you can force a 8556 particular value with '--param'. 8557 8558 If you are confident that your use of the variable does not affect 8559 ABI outside a single build of your project, you can turn off the 8560 warning with '-Wno-interference-size'. 8561 8562'-Wint-in-bool-context' 8563 Warn for suspicious use of integer values where boolean values are 8564 expected, such as conditional expressions (?:) using non-boolean 8565 integer constants in boolean context, like 'if (a <= b ? 2 : 3)'. 8566 Or left shifting of signed integers in boolean context, like 'for 8567 (a = 0; 1 << a; a++);'. Likewise for all kinds of multiplications 8568 regardless of the data type. This warning is enabled by '-Wall'. 8569 8570'-Wno-int-to-pointer-cast' 8571 Suppress warnings from casts to pointer type of an integer of a 8572 different size. In C++, casting to a pointer type of smaller size 8573 is an error. 'Wint-to-pointer-cast' is enabled by default. 8574 8575'-Wno-pointer-to-int-cast (C and Objective-C only)' 8576 Suppress warnings from casts from a pointer to an integer type of a 8577 different size. 8578 8579'-Winvalid-pch' 8580 Warn if a precompiled header (*note Precompiled Headers::) is found 8581 in the search path but cannot be used. 8582 8583'-Wlong-long' 8584 Warn if 'long long' type is used. This is enabled by either 8585 '-Wpedantic' or '-Wtraditional' in ISO C90 and C++98 modes. To 8586 inhibit the warning messages, use '-Wno-long-long'. 8587 8588'-Wvariadic-macros' 8589 Warn if variadic macros are used in ISO C90 mode, or if the GNU 8590 alternate syntax is used in ISO C99 mode. This is enabled by 8591 either '-Wpedantic' or '-Wtraditional'. To inhibit the warning 8592 messages, use '-Wno-variadic-macros'. 8593 8594'-Wno-varargs' 8595 Do not warn upon questionable usage of the macros used to handle 8596 variable arguments like 'va_start'. These warnings are enabled by 8597 default. 8598 8599'-Wvector-operation-performance' 8600 Warn if vector operation is not implemented via SIMD capabilities 8601 of the architecture. Mainly useful for the performance tuning. 8602 Vector operation can be implemented 'piecewise', which means that 8603 the scalar operation is performed on every vector element; 'in 8604 parallel', which means that the vector operation is implemented 8605 using scalars of wider type, which normally is more performance 8606 efficient; and 'as a single scalar', which means that vector fits 8607 into a scalar type. 8608 8609'-Wvla' 8610 Warn if a variable-length array is used in the code. '-Wno-vla' 8611 prevents the '-Wpedantic' warning of the variable-length array. 8612 8613'-Wvla-larger-than=BYTE-SIZE' 8614 If this option is used, the compiler warns for declarations of 8615 variable-length arrays whose size is either unbounded, or bounded 8616 by an argument that allows the array size to exceed BYTE-SIZE 8617 bytes. This is similar to how '-Walloca-larger-than='BYTE-SIZE 8618 works, but with variable-length arrays. 8619 8620 Note that GCC may optimize small variable-length arrays of a known 8621 value into plain arrays, so this warning may not get triggered for 8622 such arrays. 8623 8624 '-Wvla-larger-than=''PTRDIFF_MAX' is enabled by default but is 8625 typically only effective when '-ftree-vrp' is active (default for 8626 '-O2' and above). 8627 8628 See also '-Walloca-larger-than=BYTE-SIZE'. 8629 8630'-Wno-vla-larger-than' 8631 Disable '-Wvla-larger-than=' warnings. The option is equivalent to 8632 '-Wvla-larger-than=''SIZE_MAX' or larger. 8633 8634'-Wvla-parameter' 8635 Warn about redeclarations of functions involving arguments of 8636 Variable Length Array types of inconsistent kinds or forms, and 8637 enable the detection of out-of-bounds accesses to such parameters 8638 by warnings such as '-Warray-bounds'. 8639 8640 If the first function declaration uses the VLA form the bound 8641 specified in the array is assumed to be the minimum number of 8642 elements expected to be provided in calls to the function and the 8643 maximum number of elements accessed by it. Failing to provide 8644 arguments of sufficient size or accessing more than the maximum 8645 number of elements may be diagnosed. 8646 8647 For example, the warning triggers for the following redeclarations 8648 because the first one allows an array of any size to be passed to 8649 'f' while the second one specifies that the array argument must 8650 have at least 'n' elements. In addition, calling 'f' with the 8651 associated VLA bound parameter in excess of the actual VLA bound 8652 triggers a warning as well. 8653 8654 void f (int n, int[n]); 8655 void f (int, int[]); // warning: argument 2 previously declared as a VLA 8656 8657 void g (int n) 8658 { 8659 if (n > 4) 8660 return; 8661 int a[n]; 8662 f (sizeof a, a); // warning: access to a by f may be out of bounds 8663 ... 8664 } 8665 8666 8667 '-Wvla-parameter' is included in '-Wall'. The '-Warray-parameter' 8668 option triggers warnings for similar problems involving ordinary 8669 array arguments. 8670 8671'-Wvolatile-register-var' 8672 Warn if a register variable is declared volatile. The volatile 8673 modifier does not inhibit all optimizations that may eliminate 8674 reads and/or writes to register variables. This warning is enabled 8675 by '-Wall'. 8676 8677'-Wdisabled-optimization' 8678 Warn if a requested optimization pass is disabled. This warning 8679 does not generally indicate that there is anything wrong with your 8680 code; it merely indicates that GCC's optimizers are unable to 8681 handle the code effectively. Often, the problem is that your code 8682 is too big or too complex; GCC refuses to optimize programs when 8683 the optimization itself is likely to take inordinate amounts of 8684 time. 8685 8686'-Wpointer-sign (C and Objective-C only)' 8687 Warn for pointer argument passing or assignment with different 8688 signedness. This option is only supported for C and Objective-C. 8689 It is implied by '-Wall' and by '-Wpedantic', which can be disabled 8690 with '-Wno-pointer-sign'. 8691 8692'-Wstack-protector' 8693 This option is only active when '-fstack-protector' is active. It 8694 warns about functions that are not protected against stack 8695 smashing. 8696 8697'-Woverlength-strings' 8698 Warn about string constants that are longer than the "minimum 8699 maximum" length specified in the C standard. Modern compilers 8700 generally allow string constants that are much longer than the 8701 standard's minimum limit, but very portable programs should avoid 8702 using longer strings. 8703 8704 The limit applies _after_ string constant concatenation, and does 8705 not count the trailing NUL. In C90, the limit was 509 characters; 8706 in C99, it was raised to 4095. C++98 does not specify a normative 8707 minimum maximum, so we do not diagnose overlength strings in C++. 8708 8709 This option is implied by '-Wpedantic', and can be disabled with 8710 '-Wno-overlength-strings'. 8711 8712'-Wunsuffixed-float-constants (C and Objective-C only)' 8713 8714 Issue a warning for any floating constant that does not have a 8715 suffix. When used together with '-Wsystem-headers' it warns about 8716 such constants in system header files. This can be useful when 8717 preparing code to use with the 'FLOAT_CONST_DECIMAL64' pragma from 8718 the decimal floating-point extension to C99. 8719 8720'-Wno-lto-type-mismatch' 8721 8722 During the link-time optimization, do not warn about type 8723 mismatches in global declarations from different compilation units. 8724 Requires '-flto' to be enabled. Enabled by default. 8725 8726'-Wno-designated-init (C and Objective-C only)' 8727 Suppress warnings when a positional initializer is used to 8728 initialize a structure that has been marked with the 8729 'designated_init' attribute. 8730 8731 8732File: gcc.info, Node: Static Analyzer Options, Next: Debugging Options, Prev: Warning Options, Up: Invoking GCC 8733 87343.9 Options That Control Static Analysis 8735======================================== 8736 8737'-fanalyzer' 8738 This option enables an static analysis of program flow which looks 8739 for "interesting" interprocedural paths through the code, and 8740 issues warnings for problems found on them. 8741 8742 This analysis is much more expensive than other GCC warnings. 8743 8744 In technical terms, it performs coverage-guided symbolic execution 8745 of the code being compiled. It is neither sound nor complete: it 8746 can have false positives and false negatives. It is a bug-finding 8747 tool, rather than a tool for proving program correctness. 8748 8749 The analyzer is only suitable for use on C code in this release. 8750 8751 Enabling this option effectively enables the following warnings: 8752 8753 8754 -Wanalyzer-double-fclose 8755 -Wanalyzer-double-free 8756 -Wanalyzer-exposure-through-output-file 8757 -Wanalyzer-file-leak 8758 -Wanalyzer-free-of-non-heap 8759 -Wanalyzer-malloc-leak 8760 -Wanalyzer-mismatching-deallocation 8761 -Wanalyzer-null-argument 8762 -Wanalyzer-null-dereference 8763 -Wanalyzer-possible-null-argument 8764 -Wanalyzer-possible-null-dereference 8765 -Wanalyzer-shift-count-negative 8766 -Wanalyzer-shift-count-overflow 8767 -Wanalyzer-stale-setjmp-buffer 8768 -Wanalyzer-unsafe-call-within-signal-handler 8769 -Wanalyzer-use-after-free 8770 -Wanalyzer-use-of-pointer-in-stale-stack-frame 8771 -Wanalyzer-use-of-uninitialized-value 8772 -Wanalyzer-write-to-const 8773 -Wanalyzer-write-to-string-literal 8774 8775 8776 This option is only available if GCC was configured with analyzer 8777 support enabled. 8778 8779'-Wanalyzer-too-complex' 8780 If '-fanalyzer' is enabled, the analyzer uses various heuristics to 8781 attempt to explore the control flow and data flow in the program, 8782 but these can be defeated by sufficiently complicated code. 8783 8784 By default, the analysis silently stops if the code is too 8785 complicated for the analyzer to fully explore and it reaches an 8786 internal limit. The '-Wanalyzer-too-complex' option warns if this 8787 occurs. 8788 8789'-Wno-analyzer-double-fclose' 8790 This warning requires '-fanalyzer', which enables it; use 8791 '-Wno-analyzer-double-fclose' to disable it. 8792 8793 This diagnostic warns for paths through the code in which a 'FILE 8794 *' can have 'fclose' called on it more than once. 8795 8796'-Wno-analyzer-double-free' 8797 This warning requires '-fanalyzer', which enables it; use 8798 '-Wno-analyzer-double-free' to disable it. 8799 8800 This diagnostic warns for paths through the code in which a pointer 8801 can have a deallocator called on it more than once, either 'free', 8802 or a deallocator referenced by attribute 'malloc'. 8803 8804'-Wno-analyzer-exposure-through-output-file' 8805 This warning requires '-fanalyzer', which enables it; use 8806 '-Wno-analyzer-exposure-through-output-file' to disable it. 8807 8808 This diagnostic warns for paths through the code in which a 8809 security-sensitive value is written to an output file (such as 8810 writing a password to a log file). 8811 8812'-Wno-analyzer-file-leak' 8813 This warning requires '-fanalyzer', which enables it; use 8814 '-Wno-analyzer-file-leak' to disable it. 8815 8816 This diagnostic warns for paths through the code in which a 8817 '<stdio.h>' 'FILE *' stream object is leaked. 8818 8819'-Wno-analyzer-free-of-non-heap' 8820 This warning requires '-fanalyzer', which enables it; use 8821 '-Wno-analyzer-free-of-non-heap' to disable it. 8822 8823 This diagnostic warns for paths through the code in which 'free' is 8824 called on a non-heap pointer (e.g. an on-stack buffer, or a 8825 global). 8826 8827'-Wno-analyzer-malloc-leak' 8828 This warning requires '-fanalyzer', which enables it; use 8829 '-Wno-analyzer-malloc-leak' to disable it. 8830 8831 This diagnostic warns for paths through the code in which a pointer 8832 allocated via an allocator is leaked: either 'malloc', or a 8833 function marked with attribute 'malloc'. 8834 8835'-Wno-analyzer-mismatching-deallocation' 8836 This warning requires '-fanalyzer', which enables it; use 8837 '-Wno-analyzer-mismatching-deallocation' to disable it. 8838 8839 This diagnostic warns for paths through the code in which the wrong 8840 deallocation function is called on a pointer value, based on which 8841 function was used to allocate the pointer value. The diagnostic 8842 will warn about mismatches between 'free', scalar 'delete' and 8843 vector 'delete[]', and those marked as allocator/deallocator pairs 8844 using attribute 'malloc'. 8845 8846'-Wno-analyzer-possible-null-argument' 8847 This warning requires '-fanalyzer', which enables it; use 8848 '-Wno-analyzer-possible-null-argument' to disable it. 8849 8850 This diagnostic warns for paths through the code in which a 8851 possibly-NULL value is passed to a function argument marked with 8852 '__attribute__((nonnull))' as requiring a non-NULL value. 8853 8854'-Wno-analyzer-possible-null-dereference' 8855 This warning requires '-fanalyzer', which enables it; use 8856 '-Wno-analyzer-possible-null-dereference' to disable it. 8857 8858 This diagnostic warns for paths through the code in which a 8859 possibly-NULL value is dereferenced. 8860 8861'-Wno-analyzer-null-argument' 8862 This warning requires '-fanalyzer', which enables it; use 8863 '-Wno-analyzer-null-argument' to disable it. 8864 8865 This diagnostic warns for paths through the code in which a value 8866 known to be NULL is passed to a function argument marked with 8867 '__attribute__((nonnull))' as requiring a non-NULL value. 8868 8869'-Wno-analyzer-null-dereference' 8870 This warning requires '-fanalyzer', which enables it; use 8871 '-Wno-analyzer-null-dereference' to disable it. 8872 8873 This diagnostic warns for paths through the code in which a value 8874 known to be NULL is dereferenced. 8875 8876'-Wno-analyzer-shift-count-negative' 8877 This warning requires '-fanalyzer', which enables it; use 8878 '-Wno-analyzer-shift-count-negative' to disable it. 8879 8880 This diagnostic warns for paths through the code in which a shift 8881 is attempted with a negative count. It is analogous to the 8882 '-Wshift-count-negative' diagnostic implemented in the C/C++ front 8883 ends, but is implemented based on analyzing interprocedural paths, 8884 rather than merely parsing the syntax tree. However, the analyzer 8885 does not prioritize detection of such paths, so false negatives are 8886 more likely relative to other warnings. 8887 8888'-Wno-analyzer-shift-count-overflow' 8889 This warning requires '-fanalyzer', which enables it; use 8890 '-Wno-analyzer-shift-count-overflow' to disable it. 8891 8892 This diagnostic warns for paths through the code in which a shift 8893 is attempted with a count greater than or equal to the precision of 8894 the operand's type. It is analogous to the 8895 '-Wshift-count-overflow' diagnostic implemented in the C/C++ front 8896 ends, but is implemented based on analyzing interprocedural paths, 8897 rather than merely parsing the syntax tree. However, the analyzer 8898 does not prioritize detection of such paths, so false negatives are 8899 more likely relative to other warnings. 8900 8901'-Wno-analyzer-stale-setjmp-buffer' 8902 This warning requires '-fanalyzer', which enables it; use 8903 '-Wno-analyzer-stale-setjmp-buffer' to disable it. 8904 8905 This diagnostic warns for paths through the code in which 'longjmp' 8906 is called to rewind to a 'jmp_buf' relating to a 'setjmp' call in a 8907 function that has returned. 8908 8909 When 'setjmp' is called on a 'jmp_buf' to record a rewind location, 8910 it records the stack frame. The stack frame becomes invalid when 8911 the function containing the 'setjmp' call returns. Attempting to 8912 rewind to it via 'longjmp' would reference a stack frame that no 8913 longer exists, and likely lead to a crash (or worse). 8914 8915'-Wno-analyzer-tainted-allocation-size' 8916 This warning requires both '-fanalyzer' and 8917 '-fanalyzer-checker=taint' to enable it; use 8918 '-Wno-analyzer-tainted-allocation-size' to disable it. 8919 8920 This diagnostic warns for paths through the code in which a value 8921 that could be under an attacker's control is used as the size of an 8922 allocation without being sanitized, so that an attacker could 8923 inject an excessively large allocation and potentially cause a 8924 denial of service attack. 8925 8926 See CWE-789: Memory Allocation with Excessive Size Value 8927 (https://cwe.mitre.org/data/definitions/789.html). 8928 8929'-Wno-analyzer-tainted-array-index' 8930 This warning requires both '-fanalyzer' and 8931 '-fanalyzer-checker=taint' to enable it; use 8932 '-Wno-analyzer-tainted-array-index' to disable it. 8933 8934 This diagnostic warns for paths through the code in which a value 8935 that could be under an attacker's control is used as the index of 8936 an array access without being sanitized, so that an attacker could 8937 inject an out-of-bounds access. 8938 8939 See CWE-129: Improper Validation of Array Index 8940 (https://cwe.mitre.org/data/definitions/129.html). 8941 8942'-Wno-analyzer-tainted-divisor' 8943 This warning requires both '-fanalyzer' and 8944 '-fanalyzer-checker=taint' to enable it; use 8945 '-Wno-analyzer-tainted-divisor' to disable it. 8946 8947 This diagnostic warns for paths through the code in which a value 8948 that could be under an attacker's control is used as the divisor in 8949 a division or modulus operation without being sanitized, so that an 8950 attacker could inject a division-by-zero. 8951 8952'-Wno-analyzer-tainted-offset' 8953 This warning requires both '-fanalyzer' and 8954 '-fanalyzer-checker=taint' to enable it; use 8955 '-Wno-analyzer-tainted-offset' to disable it. 8956 8957 This diagnostic warns for paths through the code in which a value 8958 that could be under an attacker's control is used as a pointer 8959 offset without being sanitized, so that an attacker could inject an 8960 out-of-bounds access. 8961 8962 See CWE-823: Use of Out-of-range Pointer Offset 8963 (https://cwe.mitre.org/data/definitions/823.html). 8964 8965'-Wno-analyzer-tainted-size' 8966 This warning requires both '-fanalyzer' and 8967 '-fanalyzer-checker=taint' to enable it; use 8968 '-Wno-analyzer-tainted-size' to disable it. 8969 8970 This diagnostic warns for paths through the code in which a value 8971 that could be under an attacker's control is used as the size of an 8972 operation such as 'memset' without being sanitized, so that an 8973 attacker could inject an out-of-bounds access. 8974 8975'-Wno-analyzer-unsafe-call-within-signal-handler' 8976 This warning requires '-fanalyzer', which enables it; use 8977 '-Wno-analyzer-unsafe-call-within-signal-handler' to disable it. 8978 8979 This diagnostic warns for paths through the code in which a 8980 function known to be async-signal-unsafe (such as 'fprintf') is 8981 called from a signal handler. 8982 8983'-Wno-analyzer-use-after-free' 8984 This warning requires '-fanalyzer', which enables it; use 8985 '-Wno-analyzer-use-after-free' to disable it. 8986 8987 This diagnostic warns for paths through the code in which a pointer 8988 is used after a deallocator is called on it: either 'free', or a 8989 deallocator referenced by attribute 'malloc'. 8990 8991'-Wno-analyzer-use-of-pointer-in-stale-stack-frame' 8992 This warning requires '-fanalyzer', which enables it; use 8993 '-Wno-analyzer-use-of-pointer-in-stale-stack-frame' to disable it. 8994 8995 This diagnostic warns for paths through the code in which a pointer 8996 is dereferenced that points to a variable in a stale stack frame. 8997 8998'-Wno-analyzer-write-to-const' 8999 This warning requires '-fanalyzer', which enables it; use 9000 '-Wno-analyzer-write-to-const' to disable it. 9001 9002 This diagnostic warns for paths through the code in which the 9003 analyzer detects an attempt to write through a pointer to a 'const' 9004 object. However, the analyzer does not prioritize detection of 9005 such paths, so false negatives are more likely relative to other 9006 warnings. 9007 9008'-Wno-analyzer-write-to-string-literal' 9009 This warning requires '-fanalyzer', which enables it; use 9010 '-Wno-analyzer-write-to-string-literal' to disable it. 9011 9012 This diagnostic warns for paths through the code in which the 9013 analyzer detects an attempt to write through a pointer to a string 9014 literal. However, the analyzer does not prioritize detection of 9015 such paths, so false negatives are more likely relative to other 9016 warnings. 9017 9018'-Wno-analyzer-use-of-uninitialized-value' 9019 This warning requires '-fanalyzer', which enables it; use 9020 '-Wno-analyzer-use-of-uninitialized-value' to disable it. 9021 9022 This diagnostic warns for paths through the code in which an 9023 uninitialized value is used. 9024 9025 Pertinent parameters for controlling the exploration are: '--param 9026analyzer-bb-explosion-factor=VALUE', '--param 9027analyzer-max-enodes-per-program-point=VALUE', '--param 9028analyzer-max-recursion-depth=VALUE', and '--param 9029analyzer-min-snodes-for-call-summary=VALUE'. 9030 9031 The following options control the analyzer. 9032 9033'-fanalyzer-call-summaries' 9034 Simplify interprocedural analysis by computing the effect of 9035 certain calls, rather than exploring all paths through the function 9036 from callsite to each possible return. 9037 9038 If enabled, call summaries are only used for functions with more 9039 than one call site, and that are sufficiently complicated (as per 9040 '--param analyzer-min-snodes-for-call-summary=VALUE'). 9041 9042'-fanalyzer-checker=NAME' 9043 Restrict the analyzer to run just the named checker, and enable it. 9044 9045 Some checkers are disabled by default (even with '-fanalyzer'), 9046 such as the 'taint' checker that implements 9047 '-Wanalyzer-tainted-array-index', and this option is required to 9048 enable them. 9049 9050 _Note:_ currently, '-fanalyzer-checker=taint' disables the 9051 following warnings from '-fanalyzer': 9052 9053 9054 -Wanalyzer-double-fclose 9055 -Wanalyzer-double-free 9056 -Wanalyzer-exposure-through-output-file 9057 -Wanalyzer-file-leak 9058 -Wanalyzer-free-of-non-heap 9059 -Wanalyzer-malloc-leak 9060 -Wanalyzer-mismatching-deallocation 9061 -Wanalyzer-null-argument 9062 -Wanalyzer-null-dereference 9063 -Wanalyzer-possible-null-argument 9064 -Wanalyzer-possible-null-dereference 9065 -Wanalyzer-unsafe-call-within-signal-handler 9066 -Wanalyzer-use-after-free 9067 9068 9069'-fno-analyzer-feasibility' 9070 This option is intended for analyzer developers. 9071 9072 By default the analyzer verifies that there is a feasible control 9073 flow path for each diagnostic it emits: that the conditions that 9074 hold are not mutually exclusive. Diagnostics for which no feasible 9075 path can be found are rejected. This filtering can be suppressed 9076 with '-fno-analyzer-feasibility', for debugging issues in this 9077 code. 9078 9079'-fanalyzer-fine-grained' 9080 This option is intended for analyzer developers. 9081 9082 Internally the analyzer builds an "exploded graph" that combines 9083 control flow graphs with data flow information. 9084 9085 By default, an edge in this graph can contain the effects of a run 9086 of multiple statements within a basic block. With 9087 '-fanalyzer-fine-grained', each statement gets its own edge. 9088 9089'-fanalyzer-show-duplicate-count' 9090 This option is intended for analyzer developers: if multiple 9091 diagnostics have been detected as being duplicates of each other, 9092 it emits a note when reporting the best diagnostic, giving the 9093 number of additional diagnostics that were suppressed by the 9094 deduplication logic. 9095 9096'-fno-analyzer-state-merge' 9097 This option is intended for analyzer developers. 9098 9099 By default the analyzer attempts to simplify analysis by merging 9100 sufficiently similar states at each program point as it builds its 9101 "exploded graph". With '-fno-analyzer-state-merge' this merging 9102 can be suppressed, for debugging state-handling issues. 9103 9104'-fno-analyzer-state-purge' 9105 This option is intended for analyzer developers. 9106 9107 By default the analyzer attempts to simplify analysis by purging 9108 aspects of state at a program point that appear to no longer be 9109 relevant e.g. the values of locals that aren't accessed later in 9110 the function and which aren't relevant to leak analysis. 9111 9112 With '-fno-analyzer-state-purge' this purging of state can be 9113 suppressed, for debugging state-handling issues. 9114 9115'-fanalyzer-transitivity' 9116 This option enables transitivity of constraints within the 9117 analyzer. 9118 9119'-fanalyzer-verbose-edges' 9120 This option is intended for analyzer developers. It enables more 9121 verbose, lower-level detail in the descriptions of control flow 9122 within diagnostic paths. 9123 9124'-fanalyzer-verbose-state-changes' 9125 This option is intended for analyzer developers. It enables more 9126 verbose, lower-level detail in the descriptions of events relating 9127 to state machines within diagnostic paths. 9128 9129'-fanalyzer-verbosity=LEVEL' 9130 This option controls the complexity of the control flow paths that 9131 are emitted for analyzer diagnostics. 9132 9133 The LEVEL can be one of: 9134 9135 '0' 9136 At this level, interprocedural call and return events are 9137 displayed, along with the most pertinent state-change events 9138 relating to a diagnostic. For example, for a double-'free' 9139 diagnostic, both calls to 'free' will be shown. 9140 9141 '1' 9142 As per the previous level, but also show events for the entry 9143 to each function. 9144 9145 '2' 9146 As per the previous level, but also show events relating to 9147 control flow that are significant to triggering the issue 9148 (e.g. "true path taken" at a conditional). 9149 9150 This level is the default. 9151 9152 '3' 9153 As per the previous level, but show all control flow events, 9154 not just significant ones. 9155 9156 '4' 9157 This level is intended for analyzer developers; it adds 9158 various other events intended for debugging the analyzer. 9159 9160'-fdump-analyzer' 9161 Dump internal details about what the analyzer is doing to 9162 'FILE.analyzer.txt'. This option is overridden by 9163 '-fdump-analyzer-stderr'. 9164 9165'-fdump-analyzer-stderr' 9166 Dump internal details about what the analyzer is doing to stderr. 9167 This option overrides '-fdump-analyzer'. 9168 9169'-fdump-analyzer-callgraph' 9170 Dump a representation of the call graph suitable for viewing with 9171 GraphViz to 'FILE.callgraph.dot'. 9172 9173'-fdump-analyzer-exploded-graph' 9174 Dump a representation of the "exploded graph" suitable for viewing 9175 with GraphViz to 'FILE.eg.dot'. Nodes are color-coded based on 9176 state-machine states to emphasize state changes. 9177 9178'-fdump-analyzer-exploded-nodes' 9179 Emit diagnostics showing where nodes in the "exploded graph" are in 9180 relation to the program source. 9181 9182'-fdump-analyzer-exploded-nodes-2' 9183 Dump a textual representation of the "exploded graph" to 9184 'FILE.eg.txt'. 9185 9186'-fdump-analyzer-exploded-nodes-3' 9187 Dump a textual representation of the "exploded graph" to one dump 9188 file per node, to 'FILE.eg-ID.txt'. This is typically a large 9189 number of dump files. 9190 9191'-fdump-analyzer-exploded-paths' 9192 Dump a textual representation of the "exploded path" for each 9193 diagnostic to 'FILE.IDX.KIND.epath.txt'. 9194 9195'-fdump-analyzer-feasibility' 9196 Dump internal details about the analyzer's search for feasible 9197 paths. The details are written in a form suitable for viewing with 9198 GraphViz to filenames of the form 'FILE.*.fg.dot', 'FILE.*.tg.dot', 9199 and 'FILE.*.fpath.txt'. 9200 9201'-fdump-analyzer-json' 9202 Dump a compressed JSON representation of analyzer internals to 9203 'FILE.analyzer.json.gz'. The precise format is subject to change. 9204 9205'-fdump-analyzer-state-purge' 9206 As per '-fdump-analyzer-supergraph', dump a representation of the 9207 "supergraph" suitable for viewing with GraphViz, but annotate the 9208 graph with information on what state will be purged at each node. 9209 The graph is written to 'FILE.state-purge.dot'. 9210 9211'-fdump-analyzer-supergraph' 9212 Dump representations of the "supergraph" suitable for viewing with 9213 GraphViz to 'FILE.supergraph.dot' and to 'FILE.supergraph-eg.dot'. 9214 These show all of the control flow graphs in the program, with 9215 interprocedural edges for calls and returns. The second dump 9216 contains annotations showing nodes in the "exploded graph" and 9217 diagnostics associated with them. 9218 9219'-fdump-analyzer-untracked' 9220 Emit custom warnings with internal details intended for analyzer 9221 developers. 9222 9223 9224File: gcc.info, Node: Debugging Options, Next: Optimize Options, Prev: Static Analyzer Options, Up: Invoking GCC 9225 92263.10 Options for Debugging Your Program 9227======================================= 9228 9229To tell GCC to emit extra information for use by a debugger, in almost 9230all cases you need only to add '-g' to your other options. Some debug 9231formats can co-exist (like DWARF with CTF) when each of them is enabled 9232explicitly by adding the respective command line option to your other 9233options. 9234 9235 GCC allows you to use '-g' with '-O'. The shortcuts taken by optimized 9236code may occasionally be surprising: some variables you declared may not 9237exist at all; flow of control may briefly move where you did not expect 9238it; some statements may not be executed because they compute constant 9239results or their values are already at hand; some statements may execute 9240in different places because they have been moved out of loops. 9241Nevertheless it is possible to debug optimized output. This makes it 9242reasonable to use the optimizer for programs that might have bugs. 9243 9244 If you are not using some other optimization option, consider using 9245'-Og' (*note Optimize Options::) with '-g'. With no '-O' option at all, 9246some compiler passes that collect information useful for debugging do 9247not run at all, so that '-Og' may result in a better debugging 9248experience. 9249 9250'-g' 9251 Produce debugging information in the operating system's native 9252 format (stabs, COFF, XCOFF, or DWARF). GDB can work with this 9253 debugging information. 9254 9255 On most systems that use stabs format, '-g' enables use of extra 9256 debugging information that only GDB can use; this extra information 9257 makes debugging work better in GDB but probably makes other 9258 debuggers crash or refuse to read the program. If you want to 9259 control for certain whether to generate the extra information, use 9260 '-gstabs+', '-gstabs', '-gxcoff+', '-gxcoff', or '-gvms' (see 9261 below). 9262 9263'-ggdb' 9264 Produce debugging information for use by GDB. This means to use 9265 the most expressive format available (DWARF, stabs, or the native 9266 format if neither of those are supported), including GDB extensions 9267 if at all possible. 9268 9269'-gdwarf' 9270'-gdwarf-VERSION' 9271 Produce debugging information in DWARF format (if that is 9272 supported). The value of VERSION may be either 2, 3, 4 or 5; the 9273 default version for most targets is 5 (with the exception of 9274 VxWorks, TPF and Darwin/Mac OS X, which default to version 2, and 9275 AIX, which defaults to version 4). 9276 9277 Note that with DWARF Version 2, some ports require and always use 9278 some non-conflicting DWARF 3 extensions in the unwind tables. 9279 9280 Version 4 may require GDB 7.0 and '-fvar-tracking-assignments' for 9281 maximum benefit. Version 5 requires GDB 8.0 or higher. 9282 9283 GCC no longer supports DWARF Version 1, which is substantially 9284 different than Version 2 and later. For historical reasons, some 9285 other DWARF-related options such as '-fno-dwarf2-cfi-asm') retain a 9286 reference to DWARF Version 2 in their names, but apply to all 9287 currently-supported versions of DWARF. 9288 9289'-gbtf' 9290 Request BTF debug information. BTF is the default debugging format 9291 for the eBPF target. On other targets, like x86, BTF debug 9292 information can be generated along with DWARF debug information 9293 when both of the debug formats are enabled explicitly via their 9294 respective command line options. 9295 9296'-gctf' 9297'-gctfLEVEL' 9298 Request CTF debug information and use level to specify how much CTF 9299 debug information should be produced. If '-gctf' is specified 9300 without a value for level, the default level of CTF debug 9301 information is 2. 9302 9303 CTF debug information can be generated along with DWARF debug 9304 information when both of the debug formats are enabled explicitly 9305 via their respective command line options. 9306 9307 Level 0 produces no CTF debug information at all. Thus, '-gctf0' 9308 negates '-gctf'. 9309 9310 Level 1 produces CTF information for tracebacks only. This 9311 includes callsite information, but does not include type 9312 information. 9313 9314 Level 2 produces type information for entities (functions, data 9315 objects etc.) at file-scope or global-scope only. 9316 9317'-gstabs' 9318 Produce debugging information in stabs format (if that is 9319 supported), without GDB extensions. This is the format used by DBX 9320 on most BSD systems. On MIPS, Alpha and System V Release 4 systems 9321 this option produces stabs debugging output that is not understood 9322 by DBX. On System V Release 4 systems this option requires the GNU 9323 assembler. 9324 9325'-gstabs+' 9326 Produce debugging information in stabs format (if that is 9327 supported), using GNU extensions understood only by the GNU 9328 debugger (GDB). The use of these extensions is likely to make 9329 other debuggers crash or refuse to read the program. 9330 9331'-gxcoff' 9332 Produce debugging information in XCOFF format (if that is 9333 supported). This is the format used by the DBX debugger on IBM 9334 RS/6000 systems. 9335 9336'-gxcoff+' 9337 Produce debugging information in XCOFF format (if that is 9338 supported), using GNU extensions understood only by the GNU 9339 debugger (GDB). The use of these extensions is likely to make 9340 other debuggers crash or refuse to read the program, and may cause 9341 assemblers other than the GNU assembler (GAS) to fail with an 9342 error. 9343 9344'-gvms' 9345 Produce debugging information in Alpha/VMS debug format (if that is 9346 supported). This is the format used by DEBUG on Alpha/VMS systems. 9347 9348'-gLEVEL' 9349'-ggdbLEVEL' 9350'-gstabsLEVEL' 9351'-gxcoffLEVEL' 9352'-gvmsLEVEL' 9353 Request debugging information and also use LEVEL to specify how 9354 much information. The default level is 2. 9355 9356 Level 0 produces no debug information at all. Thus, '-g0' negates 9357 '-g'. 9358 9359 Level 1 produces minimal information, enough for making backtraces 9360 in parts of the program that you don't plan to debug. This 9361 includes descriptions of functions and external variables, and line 9362 number tables, but no information about local variables. 9363 9364 Level 3 includes extra information, such as all the macro 9365 definitions present in the program. Some debuggers support macro 9366 expansion when you use '-g3'. 9367 9368 If you use multiple '-g' options, with or without level numbers, 9369 the last such option is the one that is effective. 9370 9371 '-gdwarf' does not accept a concatenated debug level, to avoid 9372 confusion with '-gdwarf-LEVEL'. Instead use an additional 9373 '-gLEVEL' option to change the debug level for DWARF. 9374 9375'-fno-eliminate-unused-debug-symbols' 9376 By default, no debug information is produced for symbols that are 9377 not actually used. Use this option if you want debug information 9378 for all symbols. 9379 9380'-femit-class-debug-always' 9381 Instead of emitting debugging information for a C++ class in only 9382 one object file, emit it in all object files using the class. This 9383 option should be used only with debuggers that are unable to handle 9384 the way GCC normally emits debugging information for classes 9385 because using this option increases the size of debugging 9386 information by as much as a factor of two. 9387 9388'-fno-merge-debug-strings' 9389 Direct the linker to not merge together strings in the debugging 9390 information that are identical in different object files. Merging 9391 is not supported by all assemblers or linkers. Merging decreases 9392 the size of the debug information in the output file at the cost of 9393 increasing link processing time. Merging is enabled by default. 9394 9395'-fdebug-prefix-map=OLD=NEW' 9396 When compiling files residing in directory 'OLD', record debugging 9397 information describing them as if the files resided in directory 9398 'NEW' instead. This can be used to replace a build-time path with 9399 an install-time path in the debug info. It can also be used to 9400 change an absolute path to a relative path by using '.' for NEW. 9401 This can give more reproducible builds, which are location 9402 independent, but may require an extra command to tell GDB where to 9403 find the source files. See also '-ffile-prefix-map'. 9404 9405'-fvar-tracking' 9406 Run variable tracking pass. It computes where variables are stored 9407 at each position in code. Better debugging information is then 9408 generated (if the debugging information format supports this 9409 information). 9410 9411 It is enabled by default when compiling with optimization ('-Os', 9412 '-O', '-O2', ...), debugging information ('-g') and the debug info 9413 format supports it. 9414 9415'-fvar-tracking-assignments' 9416 Annotate assignments to user variables early in the compilation and 9417 attempt to carry the annotations over throughout the compilation 9418 all the way to the end, in an attempt to improve debug information 9419 while optimizing. Use of '-gdwarf-4' is recommended along with it. 9420 9421 It can be enabled even if var-tracking is disabled, in which case 9422 annotations are created and maintained, but discarded at the end. 9423 By default, this flag is enabled together with '-fvar-tracking', 9424 except when selective scheduling is enabled. 9425 9426'-gsplit-dwarf' 9427 If DWARF debugging information is enabled, separate as much 9428 debugging information as possible into a separate output file with 9429 the extension '.dwo'. This option allows the build system to avoid 9430 linking files with debug information. To be useful, this option 9431 requires a debugger capable of reading '.dwo' files. 9432 9433'-gdwarf32' 9434'-gdwarf64' 9435 If DWARF debugging information is enabled, the '-gdwarf32' selects 9436 the 32-bit DWARF format and the '-gdwarf64' selects the 64-bit 9437 DWARF format. The default is target specific, on most targets it 9438 is '-gdwarf32' though. The 32-bit DWARF format is smaller, but 9439 can't support more than 2GiB of debug information in any of the 9440 DWARF debug information sections. The 64-bit DWARF format allows 9441 larger debug information and might not be well supported by all 9442 consumers yet. 9443 9444'-gdescribe-dies' 9445 Add description attributes to some DWARF DIEs that have no name 9446 attribute, such as artificial variables, external references and 9447 call site parameter DIEs. 9448 9449'-gpubnames' 9450 Generate DWARF '.debug_pubnames' and '.debug_pubtypes' sections. 9451 9452'-ggnu-pubnames' 9453 Generate '.debug_pubnames' and '.debug_pubtypes' sections in a 9454 format suitable for conversion into a GDB index. This option is 9455 only useful with a linker that can produce GDB index version 7. 9456 9457'-fdebug-types-section' 9458 When using DWARF Version 4 or higher, type DIEs can be put into 9459 their own '.debug_types' section instead of making them part of the 9460 '.debug_info' section. It is more efficient to put them in a 9461 separate comdat section since the linker can then remove 9462 duplicates. But not all DWARF consumers support '.debug_types' 9463 sections yet and on some objects '.debug_types' produces larger 9464 instead of smaller debugging information. 9465 9466'-grecord-gcc-switches' 9467'-gno-record-gcc-switches' 9468 This switch causes the command-line options used to invoke the 9469 compiler that may affect code generation to be appended to the 9470 DW_AT_producer attribute in DWARF debugging information. The 9471 options are concatenated with spaces separating them from each 9472 other and from the compiler version. It is enabled by default. 9473 See also '-frecord-gcc-switches' for another way of storing 9474 compiler options into the object file. 9475 9476'-gstrict-dwarf' 9477 Disallow using extensions of later DWARF standard version than 9478 selected with '-gdwarf-VERSION'. On most targets using 9479 non-conflicting DWARF extensions from later standard versions is 9480 allowed. 9481 9482'-gno-strict-dwarf' 9483 Allow using extensions of later DWARF standard version than 9484 selected with '-gdwarf-VERSION'. 9485 9486'-gas-loc-support' 9487 Inform the compiler that the assembler supports '.loc' directives. 9488 It may then use them for the assembler to generate DWARF2+ line 9489 number tables. 9490 9491 This is generally desirable, because assembler-generated 9492 line-number tables are a lot more compact than those the compiler 9493 can generate itself. 9494 9495 This option will be enabled by default if, at GCC configure time, 9496 the assembler was found to support such directives. 9497 9498'-gno-as-loc-support' 9499 Force GCC to generate DWARF2+ line number tables internally, if 9500 DWARF2+ line number tables are to be generated. 9501 9502'-gas-locview-support' 9503 Inform the compiler that the assembler supports 'view' assignment 9504 and reset assertion checking in '.loc' directives. 9505 9506 This option will be enabled by default if, at GCC configure time, 9507 the assembler was found to support them. 9508 9509'-gno-as-locview-support' 9510 Force GCC to assign view numbers internally, if 9511 '-gvariable-location-views' are explicitly requested. 9512 9513'-gcolumn-info' 9514'-gno-column-info' 9515 Emit location column information into DWARF debugging information, 9516 rather than just file and line. This option is enabled by default. 9517 9518'-gstatement-frontiers' 9519'-gno-statement-frontiers' 9520 This option causes GCC to create markers in the internal 9521 representation at the beginning of statements, and to keep them 9522 roughly in place throughout compilation, using them to guide the 9523 output of 'is_stmt' markers in the line number table. This is 9524 enabled by default when compiling with optimization ('-Os', '-O1', 9525 '-O2', ...), and outputting DWARF 2 debug information at the normal 9526 level. 9527 9528'-gvariable-location-views' 9529'-gvariable-location-views=incompat5' 9530'-gno-variable-location-views' 9531 Augment variable location lists with progressive view numbers 9532 implied from the line number table. This enables debug information 9533 consumers to inspect state at certain points of the program, even 9534 if no instructions associated with the corresponding source 9535 locations are present at that point. If the assembler lacks 9536 support for view numbers in line number tables, this will cause the 9537 compiler to emit the line number table, which generally makes them 9538 somewhat less compact. The augmented line number tables and 9539 location lists are fully backward-compatible, so they can be 9540 consumed by debug information consumers that are not aware of these 9541 augmentations, but they won't derive any benefit from them either. 9542 9543 This is enabled by default when outputting DWARF 2 debug 9544 information at the normal level, as long as there is assembler 9545 support, '-fvar-tracking-assignments' is enabled and 9546 '-gstrict-dwarf' is not. When assembler support is not available, 9547 this may still be enabled, but it will force GCC to output internal 9548 line number tables, and if '-ginternal-reset-location-views' is not 9549 enabled, that will most certainly lead to silently mismatching 9550 location views. 9551 9552 There is a proposed representation for view numbers that is not 9553 backward compatible with the location list format introduced in 9554 DWARF 5, that can be enabled with 9555 '-gvariable-location-views=incompat5'. This option may be removed 9556 in the future, is only provided as a reference implementation of 9557 the proposed representation. Debug information consumers are not 9558 expected to support this extended format, and they would be 9559 rendered unable to decode location lists using it. 9560 9561'-ginternal-reset-location-views' 9562'-gno-internal-reset-location-views' 9563 Attempt to determine location views that can be omitted from 9564 location view lists. This requires the compiler to have very 9565 accurate insn length estimates, which isn't always the case, and it 9566 may cause incorrect view lists to be generated silently when using 9567 an assembler that does not support location view lists. The GNU 9568 assembler will flag any such error as a 'view number mismatch'. 9569 This is only enabled on ports that define a reliable estimation 9570 function. 9571 9572'-ginline-points' 9573'-gno-inline-points' 9574 Generate extended debug information for inlined functions. 9575 Location view tracking markers are inserted at inlined entry 9576 points, so that address and view numbers can be computed and output 9577 in debug information. This can be enabled independently of 9578 location views, in which case the view numbers won't be output, but 9579 it can only be enabled along with statement frontiers, and it is 9580 only enabled by default if location views are enabled. 9581 9582'-gz[=TYPE]' 9583 Produce compressed debug sections in DWARF format, if that is 9584 supported. If TYPE is not given, the default type depends on the 9585 capabilities of the assembler and linker used. TYPE may be one of 9586 'none' (don't compress debug sections), 'zlib' (use zlib 9587 compression in ELF gABI format), or 'zlib-gnu' (use zlib 9588 compression in traditional GNU format). If the linker doesn't 9589 support writing compressed debug sections, the option is rejected. 9590 Otherwise, if the assembler does not support them, '-gz' is 9591 silently ignored when producing object files. 9592 9593'-femit-struct-debug-baseonly' 9594 Emit debug information for struct-like types only when the base 9595 name of the compilation source file matches the base name of file 9596 in which the struct is defined. 9597 9598 This option substantially reduces the size of debugging 9599 information, but at significant potential loss in type information 9600 to the debugger. See '-femit-struct-debug-reduced' for a less 9601 aggressive option. See '-femit-struct-debug-detailed' for more 9602 detailed control. 9603 9604 This option works only with DWARF debug output. 9605 9606'-femit-struct-debug-reduced' 9607 Emit debug information for struct-like types only when the base 9608 name of the compilation source file matches the base name of file 9609 in which the type is defined, unless the struct is a template or 9610 defined in a system header. 9611 9612 This option significantly reduces the size of debugging 9613 information, with some potential loss in type information to the 9614 debugger. See '-femit-struct-debug-baseonly' for a more aggressive 9615 option. See '-femit-struct-debug-detailed' for more detailed 9616 control. 9617 9618 This option works only with DWARF debug output. 9619 9620'-femit-struct-debug-detailed[=SPEC-LIST]' 9621 Specify the struct-like types for which the compiler generates 9622 debug information. The intent is to reduce duplicate struct debug 9623 information between different object files within the same program. 9624 9625 This option is a detailed version of '-femit-struct-debug-reduced' 9626 and '-femit-struct-debug-baseonly', which serves for most needs. 9627 9628 A specification has the syntax 9629 ['dir:'|'ind:']['ord:'|'gen:']('any'|'sys'|'base'|'none') 9630 9631 The optional first word limits the specification to structs that 9632 are used directly ('dir:') or used indirectly ('ind:'). A struct 9633 type is used directly when it is the type of a variable, member. 9634 Indirect uses arise through pointers to structs. That is, when use 9635 of an incomplete struct is valid, the use is indirect. An example 9636 is 'struct one direct; struct two * indirect;'. 9637 9638 The optional second word limits the specification to ordinary 9639 structs ('ord:') or generic structs ('gen:'). Generic structs are 9640 a bit complicated to explain. For C++, these are non-explicit 9641 specializations of template classes, or non-template classes within 9642 the above. Other programming languages have generics, but 9643 '-femit-struct-debug-detailed' does not yet implement them. 9644 9645 The third word specifies the source files for those structs for 9646 which the compiler should emit debug information. The values 9647 'none' and 'any' have the normal meaning. The value 'base' means 9648 that the base of name of the file in which the type declaration 9649 appears must match the base of the name of the main compilation 9650 file. In practice, this means that when compiling 'foo.c', debug 9651 information is generated for types declared in that file and 9652 'foo.h', but not other header files. The value 'sys' means those 9653 types satisfying 'base' or declared in system or compiler headers. 9654 9655 You may need to experiment to determine the best settings for your 9656 application. 9657 9658 The default is '-femit-struct-debug-detailed=all'. 9659 9660 This option works only with DWARF debug output. 9661 9662'-fno-dwarf2-cfi-asm' 9663 Emit DWARF unwind info as compiler generated '.eh_frame' section 9664 instead of using GAS '.cfi_*' directives. 9665 9666'-fno-eliminate-unused-debug-types' 9667 Normally, when producing DWARF output, GCC avoids producing debug 9668 symbol output for types that are nowhere used in the source file 9669 being compiled. Sometimes it is useful to have GCC emit debugging 9670 information for all types declared in a compilation unit, 9671 regardless of whether or not they are actually used in that 9672 compilation unit, for example if, in the debugger, you want to cast 9673 a value to a type that is not actually used in your program (but is 9674 declared). More often, however, this results in a significant 9675 amount of wasted space. 9676 9677 9678File: gcc.info, Node: Optimize Options, Next: Instrumentation Options, Prev: Debugging Options, Up: Invoking GCC 9679 96803.11 Options That Control Optimization 9681====================================== 9682 9683These options control various sorts of optimizations. 9684 9685 Without any optimization option, the compiler's goal is to reduce the 9686cost of compilation and to make debugging produce the expected results. 9687Statements are independent: if you stop the program with a breakpoint 9688between statements, you can then assign a new value to any variable or 9689change the program counter to any other statement in the function and 9690get exactly the results you expect from the source code. 9691 9692 Turning on optimization flags makes the compiler attempt to improve the 9693performance and/or code size at the expense of compilation time and 9694possibly the ability to debug the program. 9695 9696 The compiler performs optimization based on the knowledge it has of the 9697program. Compiling multiple files at once to a single output file mode 9698allows the compiler to use information gained from all of the files when 9699compiling each of them. 9700 9701 Not all optimizations are controlled directly by a flag. Only 9702optimizations that have a flag are listed in this section. 9703 9704 Most optimizations are completely disabled at '-O0' or if an '-O' level 9705is not set on the command line, even if individual optimization flags 9706are specified. Similarly, '-Og' suppresses many optimization passes. 9707 9708 Depending on the target and how GCC was configured, a slightly 9709different set of optimizations may be enabled at each '-O' level than 9710those listed here. You can invoke GCC with '-Q --help=optimizers' to 9711find out the exact set of optimizations that are enabled at each level. 9712*Note Overall Options::, for examples. 9713 9714'-O' 9715'-O1' 9716 Optimize. Optimizing compilation takes somewhat more time, and a 9717 lot more memory for a large function. 9718 9719 With '-O', the compiler tries to reduce code size and execution 9720 time, without performing any optimizations that take a great deal 9721 of compilation time. 9722 9723 '-O' turns on the following optimization flags: 9724 9725 -fauto-inc-dec 9726 -fbranch-count-reg 9727 -fcombine-stack-adjustments 9728 -fcompare-elim 9729 -fcprop-registers 9730 -fdce 9731 -fdefer-pop 9732 -fdelayed-branch 9733 -fdse 9734 -fforward-propagate 9735 -fguess-branch-probability 9736 -fif-conversion 9737 -fif-conversion2 9738 -finline-functions-called-once 9739 -fipa-modref 9740 -fipa-profile 9741 -fipa-pure-const 9742 -fipa-reference 9743 -fipa-reference-addressable 9744 -fmerge-constants 9745 -fmove-loop-invariants 9746 -fmove-loop-stores 9747 -fomit-frame-pointer 9748 -freorder-blocks 9749 -fshrink-wrap 9750 -fshrink-wrap-separate 9751 -fsplit-wide-types 9752 -fssa-backprop 9753 -fssa-phiopt 9754 -ftree-bit-ccp 9755 -ftree-ccp 9756 -ftree-ch 9757 -ftree-coalesce-vars 9758 -ftree-copy-prop 9759 -ftree-dce 9760 -ftree-dominator-opts 9761 -ftree-dse 9762 -ftree-forwprop 9763 -ftree-fre 9764 -ftree-phiprop 9765 -ftree-pta 9766 -ftree-scev-cprop 9767 -ftree-sink 9768 -ftree-slsr 9769 -ftree-sra 9770 -ftree-ter 9771 -funit-at-a-time 9772 9773'-O2' 9774 Optimize even more. GCC performs nearly all supported 9775 optimizations that do not involve a space-speed tradeoff. As 9776 compared to '-O', this option increases both compilation time and 9777 the performance of the generated code. 9778 9779 '-O2' turns on all optimization flags specified by '-O1'. It also 9780 turns on the following optimization flags: 9781 9782 -falign-functions -falign-jumps 9783 -falign-labels -falign-loops 9784 -fcaller-saves 9785 -fcode-hoisting 9786 -fcrossjumping 9787 -fcse-follow-jumps -fcse-skip-blocks 9788 -fdelete-null-pointer-checks 9789 -fdevirtualize -fdevirtualize-speculatively 9790 -fexpensive-optimizations 9791 -ffinite-loops 9792 -fgcse -fgcse-lm 9793 -fhoist-adjacent-loads 9794 -finline-functions 9795 -finline-small-functions 9796 -findirect-inlining 9797 -fipa-bit-cp -fipa-cp -fipa-icf 9798 -fipa-ra -fipa-sra -fipa-vrp 9799 -fisolate-erroneous-paths-dereference 9800 -flra-remat 9801 -foptimize-sibling-calls 9802 -foptimize-strlen 9803 -fpartial-inlining 9804 -fpeephole2 9805 -freorder-blocks-algorithm=stc 9806 -freorder-blocks-and-partition -freorder-functions 9807 -frerun-cse-after-loop 9808 -fschedule-insns -fschedule-insns2 9809 -fsched-interblock -fsched-spec 9810 -fstore-merging 9811 -fstrict-aliasing 9812 -fthread-jumps 9813 -ftree-builtin-call-dce 9814 -ftree-loop-vectorize 9815 -ftree-pre 9816 -ftree-slp-vectorize 9817 -ftree-switch-conversion -ftree-tail-merge 9818 -ftree-vrp 9819 -fvect-cost-model=very-cheap 9820 9821 Please note the warning under '-fgcse' about invoking '-O2' on 9822 programs that use computed gotos. 9823 9824'-O3' 9825 Optimize yet more. '-O3' turns on all optimizations specified by 9826 '-O2' and also turns on the following optimization flags: 9827 9828 -fgcse-after-reload 9829 -fipa-cp-clone 9830 -floop-interchange 9831 -floop-unroll-and-jam 9832 -fpeel-loops 9833 -fpredictive-commoning 9834 -fsplit-loops 9835 -fsplit-paths 9836 -ftree-loop-distribution 9837 -ftree-partial-pre 9838 -funswitch-loops 9839 -fvect-cost-model=dynamic 9840 -fversion-loops-for-strides 9841 9842'-O0' 9843 Reduce compilation time and make debugging produce the expected 9844 results. This is the default. 9845 9846'-Os' 9847 Optimize for size. '-Os' enables all '-O2' optimizations except 9848 those that often increase code size: 9849 9850 -falign-functions -falign-jumps 9851 -falign-labels -falign-loops 9852 -fprefetch-loop-arrays -freorder-blocks-algorithm=stc 9853 9854 It also enables '-finline-functions', causes the compiler to tune 9855 for code size rather than execution speed, and performs further 9856 optimizations designed to reduce code size. 9857 9858'-Ofast' 9859 Disregard strict standards compliance. '-Ofast' enables all '-O3' 9860 optimizations. It also enables optimizations that are not valid 9861 for all standard-compliant programs. It turns on '-ffast-math', 9862 '-fallow-store-data-races' and the Fortran-specific 9863 '-fstack-arrays', unless '-fmax-stack-var-size' is specified, and 9864 '-fno-protect-parens'. It turns off '-fsemantic-interposition'. 9865 9866'-Og' 9867 Optimize debugging experience. '-Og' should be the optimization 9868 level of choice for the standard edit-compile-debug cycle, offering 9869 a reasonable level of optimization while maintaining fast 9870 compilation and a good debugging experience. It is a better choice 9871 than '-O0' for producing debuggable code because some compiler 9872 passes that collect debug information are disabled at '-O0'. 9873 9874 Like '-O0', '-Og' completely disables a number of optimization 9875 passes so that individual options controlling them have no effect. 9876 Otherwise '-Og' enables all '-O1' optimization flags except for 9877 those that may interfere with debugging: 9878 9879 -fbranch-count-reg -fdelayed-branch 9880 -fdse -fif-conversion -fif-conversion2 9881 -finline-functions-called-once 9882 -fmove-loop-invariants -fmove-loop-stores -fssa-phiopt 9883 -ftree-bit-ccp -ftree-dse -ftree-pta -ftree-sra 9884 9885'-Oz' 9886 Optimize aggressively for size rather than speed. This may 9887 increase the number of instructions executed if those instructions 9888 require fewer bytes to encode. '-Oz' behaves similarly to '-Os' 9889 including enabling most '-O2' optimizations. 9890 9891 If you use multiple '-O' options, with or without level numbers, the 9892last such option is the one that is effective. 9893 9894 Options of the form '-fFLAG' specify machine-independent flags. Most 9895flags have both positive and negative forms; the negative form of 9896'-ffoo' is '-fno-foo'. In the table below, only one of the forms is 9897listed--the one you typically use. You can figure out the other form by 9898either removing 'no-' or adding it. 9899 9900 The following options control specific optimizations. They are either 9901activated by '-O' options or are related to ones that are. You can use 9902the following flags in the rare cases when "fine-tuning" of 9903optimizations to be performed is desired. 9904 9905'-fno-defer-pop' 9906 For machines that must pop arguments after a function call, always 9907 pop the arguments as soon as each function returns. At levels 9908 '-O1' and higher, '-fdefer-pop' is the default; this allows the 9909 compiler to let arguments accumulate on the stack for several 9910 function calls and pop them all at once. 9911 9912'-fforward-propagate' 9913 Perform a forward propagation pass on RTL. The pass tries to 9914 combine two instructions and checks if the result can be 9915 simplified. If loop unrolling is active, two passes are performed 9916 and the second is scheduled after loop unrolling. 9917 9918 This option is enabled by default at optimization levels '-O1', 9919 '-O2', '-O3', '-Os'. 9920 9921'-ffp-contract=STYLE' 9922 '-ffp-contract=off' disables floating-point expression contraction. 9923 '-ffp-contract=fast' enables floating-point expression contraction 9924 such as forming of fused multiply-add operations if the target has 9925 native support for them. '-ffp-contract=on' enables floating-point 9926 expression contraction if allowed by the language standard. This 9927 is currently not implemented and treated equal to 9928 '-ffp-contract=off'. 9929 9930 The default is '-ffp-contract=fast'. 9931 9932'-fomit-frame-pointer' 9933 Omit the frame pointer in functions that don't need one. This 9934 avoids the instructions to save, set up and restore the frame 9935 pointer; on many targets it also makes an extra register available. 9936 9937 On some targets this flag has no effect because the standard 9938 calling sequence always uses a frame pointer, so it cannot be 9939 omitted. 9940 9941 Note that '-fno-omit-frame-pointer' doesn't guarantee the frame 9942 pointer is used in all functions. Several targets always omit the 9943 frame pointer in leaf functions. 9944 9945 Enabled by default at '-O1' and higher. 9946 9947'-foptimize-sibling-calls' 9948 Optimize sibling and tail recursive calls. 9949 9950 Enabled at levels '-O2', '-O3', '-Os'. 9951 9952'-foptimize-strlen' 9953 Optimize various standard C string functions (e.g. 'strlen', 9954 'strchr' or 'strcpy') and their '_FORTIFY_SOURCE' counterparts into 9955 faster alternatives. 9956 9957 Enabled at levels '-O2', '-O3'. 9958 9959'-fno-inline' 9960 Do not expand any functions inline apart from those marked with the 9961 'always_inline' attribute. This is the default when not 9962 optimizing. 9963 9964 Single functions can be exempted from inlining by marking them with 9965 the 'noinline' attribute. 9966 9967'-finline-small-functions' 9968 Integrate functions into their callers when their body is smaller 9969 than expected function call code (so overall size of program gets 9970 smaller). The compiler heuristically decides which functions are 9971 simple enough to be worth integrating in this way. This inlining 9972 applies to all functions, even those not declared inline. 9973 9974 Enabled at levels '-O2', '-O3', '-Os'. 9975 9976'-findirect-inlining' 9977 Inline also indirect calls that are discovered to be known at 9978 compile time thanks to previous inlining. This option has any 9979 effect only when inlining itself is turned on by the 9980 '-finline-functions' or '-finline-small-functions' options. 9981 9982 Enabled at levels '-O2', '-O3', '-Os'. 9983 9984'-finline-functions' 9985 Consider all functions for inlining, even if they are not declared 9986 inline. The compiler heuristically decides which functions are 9987 worth integrating in this way. 9988 9989 If all calls to a given function are integrated, and the function 9990 is declared 'static', then the function is normally not output as 9991 assembler code in its own right. 9992 9993 Enabled at levels '-O2', '-O3', '-Os'. Also enabled by 9994 '-fprofile-use' and '-fauto-profile'. 9995 9996'-finline-functions-called-once' 9997 Consider all 'static' functions called once for inlining into their 9998 caller even if they are not marked 'inline'. If a call to a given 9999 function is integrated, then the function is not output as 10000 assembler code in its own right. 10001 10002 Enabled at levels '-O1', '-O2', '-O3' and '-Os', but not '-Og'. 10003 10004'-fearly-inlining' 10005 Inline functions marked by 'always_inline' and functions whose body 10006 seems smaller than the function call overhead early before doing 10007 '-fprofile-generate' instrumentation and real inlining pass. Doing 10008 so makes profiling significantly cheaper and usually inlining 10009 faster on programs having large chains of nested wrapper functions. 10010 10011 Enabled by default. 10012 10013'-fipa-sra' 10014 Perform interprocedural scalar replacement of aggregates, removal 10015 of unused parameters and replacement of parameters passed by 10016 reference by parameters passed by value. 10017 10018 Enabled at levels '-O2', '-O3' and '-Os'. 10019 10020'-finline-limit=N' 10021 By default, GCC limits the size of functions that can be inlined. 10022 This flag allows coarse control of this limit. N is the size of 10023 functions that can be inlined in number of pseudo instructions. 10024 10025 Inlining is actually controlled by a number of parameters, which 10026 may be specified individually by using '--param NAME=VALUE'. The 10027 '-finline-limit=N' option sets some of these parameters as follows: 10028 10029 'max-inline-insns-single' 10030 is set to N/2. 10031 'max-inline-insns-auto' 10032 is set to N/2. 10033 10034 See below for a documentation of the individual parameters 10035 controlling inlining and for the defaults of these parameters. 10036 10037 _Note:_ there may be no value to '-finline-limit' that results in 10038 default behavior. 10039 10040 _Note:_ pseudo instruction represents, in this particular context, 10041 an abstract measurement of function's size. In no way does it 10042 represent a count of assembly instructions and as such its exact 10043 meaning might change from one release to an another. 10044 10045'-fno-keep-inline-dllexport' 10046 This is a more fine-grained version of '-fkeep-inline-functions', 10047 which applies only to functions that are declared using the 10048 'dllexport' attribute or declspec. *Note Declaring Attributes of 10049 Functions: Function Attributes. 10050 10051'-fkeep-inline-functions' 10052 In C, emit 'static' functions that are declared 'inline' into the 10053 object file, even if the function has been inlined into all of its 10054 callers. This switch does not affect functions using the 'extern 10055 inline' extension in GNU C90. In C++, emit any and all inline 10056 functions into the object file. 10057 10058'-fkeep-static-functions' 10059 Emit 'static' functions into the object file, even if the function 10060 is never used. 10061 10062'-fkeep-static-consts' 10063 Emit variables declared 'static const' when optimization isn't 10064 turned on, even if the variables aren't referenced. 10065 10066 GCC enables this option by default. If you want to force the 10067 compiler to check if a variable is referenced, regardless of 10068 whether or not optimization is turned on, use the 10069 '-fno-keep-static-consts' option. 10070 10071'-fmerge-constants' 10072 Attempt to merge identical constants (string constants and 10073 floating-point constants) across compilation units. 10074 10075 This option is the default for optimized compilation if the 10076 assembler and linker support it. Use '-fno-merge-constants' to 10077 inhibit this behavior. 10078 10079 Enabled at levels '-O1', '-O2', '-O3', '-Os'. 10080 10081'-fmerge-all-constants' 10082 Attempt to merge identical constants and identical variables. 10083 10084 This option implies '-fmerge-constants'. In addition to 10085 '-fmerge-constants' this considers e.g. even constant initialized 10086 arrays or initialized constant variables with integral or 10087 floating-point types. Languages like C or C++ require each 10088 variable, including multiple instances of the same variable in 10089 recursive calls, to have distinct locations, so using this option 10090 results in non-conforming behavior. 10091 10092'-fmodulo-sched' 10093 Perform swing modulo scheduling immediately before the first 10094 scheduling pass. This pass looks at innermost loops and reorders 10095 their instructions by overlapping different iterations. 10096 10097'-fmodulo-sched-allow-regmoves' 10098 Perform more aggressive SMS-based modulo scheduling with register 10099 moves allowed. By setting this flag certain anti-dependences edges 10100 are deleted, which triggers the generation of reg-moves based on 10101 the life-range analysis. This option is effective only with 10102 '-fmodulo-sched' enabled. 10103 10104'-fno-branch-count-reg' 10105 Disable the optimization pass that scans for opportunities to use 10106 "decrement and branch" instructions on a count register instead of 10107 instruction sequences that decrement a register, compare it against 10108 zero, and then branch based upon the result. This option is only 10109 meaningful on architectures that support such instructions, which 10110 include x86, PowerPC, IA-64 and S/390. Note that the 10111 '-fno-branch-count-reg' option doesn't remove the decrement and 10112 branch instructions from the generated instruction stream 10113 introduced by other optimization passes. 10114 10115 The default is '-fbranch-count-reg' at '-O1' and higher, except for 10116 '-Og'. 10117 10118'-fno-function-cse' 10119 Do not put function addresses in registers; make each instruction 10120 that calls a constant function contain the function's address 10121 explicitly. 10122 10123 This option results in less efficient code, but some strange hacks 10124 that alter the assembler output may be confused by the 10125 optimizations performed when this option is not used. 10126 10127 The default is '-ffunction-cse' 10128 10129'-fno-zero-initialized-in-bss' 10130 If the target supports a BSS section, GCC by default puts variables 10131 that are initialized to zero into BSS. This can save space in the 10132 resulting code. 10133 10134 This option turns off this behavior because some programs 10135 explicitly rely on variables going to the data section--e.g., so 10136 that the resulting executable can find the beginning of that 10137 section and/or make assumptions based on that. 10138 10139 The default is '-fzero-initialized-in-bss'. 10140 10141'-fthread-jumps' 10142 Perform optimizations that check to see if a jump branches to a 10143 location where another comparison subsumed by the first is found. 10144 If so, the first branch is redirected to either the destination of 10145 the second branch or a point immediately following it, depending on 10146 whether the condition is known to be true or false. 10147 10148 Enabled at levels '-O1', '-O2', '-O3', '-Os'. 10149 10150'-fsplit-wide-types' 10151 When using a type that occupies multiple registers, such as 'long 10152 long' on a 32-bit system, split the registers apart and allocate 10153 them independently. This normally generates better code for those 10154 types, but may make debugging more difficult. 10155 10156 Enabled at levels '-O1', '-O2', '-O3', '-Os'. 10157 10158'-fsplit-wide-types-early' 10159 Fully split wide types early, instead of very late. This option 10160 has no effect unless '-fsplit-wide-types' is turned on. 10161 10162 This is the default on some targets. 10163 10164'-fcse-follow-jumps' 10165 In common subexpression elimination (CSE), scan through jump 10166 instructions when the target of the jump is not reached by any 10167 other path. For example, when CSE encounters an 'if' statement 10168 with an 'else' clause, CSE follows the jump when the condition 10169 tested is false. 10170 10171 Enabled at levels '-O2', '-O3', '-Os'. 10172 10173'-fcse-skip-blocks' 10174 This is similar to '-fcse-follow-jumps', but causes CSE to follow 10175 jumps that conditionally skip over blocks. When CSE encounters a 10176 simple 'if' statement with no else clause, '-fcse-skip-blocks' 10177 causes CSE to follow the jump around the body of the 'if'. 10178 10179 Enabled at levels '-O2', '-O3', '-Os'. 10180 10181'-frerun-cse-after-loop' 10182 Re-run common subexpression elimination after loop optimizations 10183 are performed. 10184 10185 Enabled at levels '-O2', '-O3', '-Os'. 10186 10187'-fgcse' 10188 Perform a global common subexpression elimination pass. This pass 10189 also performs global constant and copy propagation. 10190 10191 _Note:_ When compiling a program using computed gotos, a GCC 10192 extension, you may get better run-time performance if you disable 10193 the global common subexpression elimination pass by adding 10194 '-fno-gcse' to the command line. 10195 10196 Enabled at levels '-O2', '-O3', '-Os'. 10197 10198'-fgcse-lm' 10199 When '-fgcse-lm' is enabled, global common subexpression 10200 elimination attempts to move loads that are only killed by stores 10201 into themselves. This allows a loop containing a load/store 10202 sequence to be changed to a load outside the loop, and a copy/store 10203 within the loop. 10204 10205 Enabled by default when '-fgcse' is enabled. 10206 10207'-fgcse-sm' 10208 When '-fgcse-sm' is enabled, a store motion pass is run after 10209 global common subexpression elimination. This pass attempts to 10210 move stores out of loops. When used in conjunction with 10211 '-fgcse-lm', loops containing a load/store sequence can be changed 10212 to a load before the loop and a store after the loop. 10213 10214 Not enabled at any optimization level. 10215 10216'-fgcse-las' 10217 When '-fgcse-las' is enabled, the global common subexpression 10218 elimination pass eliminates redundant loads that come after stores 10219 to the same memory location (both partial and full redundancies). 10220 10221 Not enabled at any optimization level. 10222 10223'-fgcse-after-reload' 10224 When '-fgcse-after-reload' is enabled, a redundant load elimination 10225 pass is performed after reload. The purpose of this pass is to 10226 clean up redundant spilling. 10227 10228 Enabled by '-O3', '-fprofile-use' and '-fauto-profile'. 10229 10230'-faggressive-loop-optimizations' 10231 This option tells the loop optimizer to use language constraints to 10232 derive bounds for the number of iterations of a loop. This assumes 10233 that loop code does not invoke undefined behavior by for example 10234 causing signed integer overflows or out-of-bound array accesses. 10235 The bounds for the number of iterations of a loop are used to guide 10236 loop unrolling and peeling and loop exit test optimizations. This 10237 option is enabled by default. 10238 10239'-funconstrained-commons' 10240 This option tells the compiler that variables declared in common 10241 blocks (e.g. Fortran) may later be overridden with longer trailing 10242 arrays. This prevents certain optimizations that depend on knowing 10243 the array bounds. 10244 10245'-fcrossjumping' 10246 Perform cross-jumping transformation. This transformation unifies 10247 equivalent code and saves code size. The resulting code may or may 10248 not perform better than without cross-jumping. 10249 10250 Enabled at levels '-O2', '-O3', '-Os'. 10251 10252'-fauto-inc-dec' 10253 Combine increments or decrements of addresses with memory accesses. 10254 This pass is always skipped on architectures that do not have 10255 instructions to support this. Enabled by default at '-O1' and 10256 higher on architectures that support this. 10257 10258'-fdce' 10259 Perform dead code elimination (DCE) on RTL. Enabled by default at 10260 '-O1' and higher. 10261 10262'-fdse' 10263 Perform dead store elimination (DSE) on RTL. Enabled by default at 10264 '-O1' and higher. 10265 10266'-fif-conversion' 10267 Attempt to transform conditional jumps into branch-less 10268 equivalents. This includes use of conditional moves, min, max, set 10269 flags and abs instructions, and some tricks doable by standard 10270 arithmetics. The use of conditional execution on chips where it is 10271 available is controlled by '-fif-conversion2'. 10272 10273 Enabled at levels '-O1', '-O2', '-O3', '-Os', but not with '-Og'. 10274 10275'-fif-conversion2' 10276 Use conditional execution (where available) to transform 10277 conditional jumps into branch-less equivalents. 10278 10279 Enabled at levels '-O1', '-O2', '-O3', '-Os', but not with '-Og'. 10280 10281'-fdeclone-ctor-dtor' 10282 The C++ ABI requires multiple entry points for constructors and 10283 destructors: one for a base subobject, one for a complete object, 10284 and one for a virtual destructor that calls operator delete 10285 afterwards. For a hierarchy with virtual bases, the base and 10286 complete variants are clones, which means two copies of the 10287 function. With this option, the base and complete variants are 10288 changed to be thunks that call a common implementation. 10289 10290 Enabled by '-Os'. 10291 10292'-fdelete-null-pointer-checks' 10293 Assume that programs cannot safely dereference null pointers, and 10294 that no code or data element resides at address zero. This option 10295 enables simple constant folding optimizations at all optimization 10296 levels. In addition, other optimization passes in GCC use this 10297 flag to control global dataflow analyses that eliminate useless 10298 checks for null pointers; these assume that a memory access to 10299 address zero always results in a trap, so that if a pointer is 10300 checked after it has already been dereferenced, it cannot be null. 10301 10302 Note however that in some environments this assumption is not true. 10303 Use '-fno-delete-null-pointer-checks' to disable this optimization 10304 for programs that depend on that behavior. 10305 10306 This option is enabled by default on most targets. On Nios II ELF, 10307 it defaults to off. On AVR, CR16, and MSP430, this option is 10308 completely disabled. 10309 10310 Passes that use the dataflow information are enabled independently 10311 at different optimization levels. 10312 10313'-fdevirtualize' 10314 Attempt to convert calls to virtual functions to direct calls. 10315 This is done both within a procedure and interprocedurally as part 10316 of indirect inlining ('-findirect-inlining') and interprocedural 10317 constant propagation ('-fipa-cp'). Enabled at levels '-O2', '-O3', 10318 '-Os'. 10319 10320'-fdevirtualize-speculatively' 10321 Attempt to convert calls to virtual functions to speculative direct 10322 calls. Based on the analysis of the type inheritance graph, 10323 determine for a given call the set of likely targets. If the set 10324 is small, preferably of size 1, change the call into a conditional 10325 deciding between direct and indirect calls. The speculative calls 10326 enable more optimizations, such as inlining. When they seem 10327 useless after further optimization, they are converted back into 10328 original form. 10329 10330'-fdevirtualize-at-ltrans' 10331 Stream extra information needed for aggressive devirtualization 10332 when running the link-time optimizer in local transformation mode. 10333 This option enables more devirtualization but significantly 10334 increases the size of streamed data. For this reason it is 10335 disabled by default. 10336 10337'-fexpensive-optimizations' 10338 Perform a number of minor optimizations that are relatively 10339 expensive. 10340 10341 Enabled at levels '-O2', '-O3', '-Os'. 10342 10343'-free' 10344 Attempt to remove redundant extension instructions. This is 10345 especially helpful for the x86-64 architecture, which implicitly 10346 zero-extends in 64-bit registers after writing to their lower 10347 32-bit half. 10348 10349 Enabled for Alpha, AArch64 and x86 at levels '-O2', '-O3', '-Os'. 10350 10351'-fno-lifetime-dse' 10352 In C++ the value of an object is only affected by changes within 10353 its lifetime: when the constructor begins, the object has an 10354 indeterminate value, and any changes during the lifetime of the 10355 object are dead when the object is destroyed. Normally dead store 10356 elimination will take advantage of this; if your code relies on the 10357 value of the object storage persisting beyond the lifetime of the 10358 object, you can use this flag to disable this optimization. To 10359 preserve stores before the constructor starts (e.g. because your 10360 operator new clears the object storage) but still treat the object 10361 as dead after the destructor, you can use '-flifetime-dse=1'. The 10362 default behavior can be explicitly selected with 10363 '-flifetime-dse=2'. '-flifetime-dse=0' is equivalent to 10364 '-fno-lifetime-dse'. 10365 10366'-flive-range-shrinkage' 10367 Attempt to decrease register pressure through register live range 10368 shrinkage. This is helpful for fast processors with small or 10369 moderate size register sets. 10370 10371'-fira-algorithm=ALGORITHM' 10372 Use the specified coloring algorithm for the integrated register 10373 allocator. The ALGORITHM argument can be 'priority', which 10374 specifies Chow's priority coloring, or 'CB', which specifies 10375 Chaitin-Briggs coloring. Chaitin-Briggs coloring is not 10376 implemented for all architectures, but for those targets that do 10377 support it, it is the default because it generates better code. 10378 10379'-fira-region=REGION' 10380 Use specified regions for the integrated register allocator. The 10381 REGION argument should be one of the following: 10382 10383 'all' 10384 Use all loops as register allocation regions. This can give 10385 the best results for machines with a small and/or irregular 10386 register set. 10387 10388 'mixed' 10389 Use all loops except for loops with small register pressure as 10390 the regions. This value usually gives the best results in 10391 most cases and for most architectures, and is enabled by 10392 default when compiling with optimization for speed ('-O', 10393 '-O2', ...). 10394 10395 'one' 10396 Use all functions as a single region. This typically results 10397 in the smallest code size, and is enabled by default for '-Os' 10398 or '-O0'. 10399 10400'-fira-hoist-pressure' 10401 Use IRA to evaluate register pressure in the code hoisting pass for 10402 decisions to hoist expressions. This option usually results in 10403 smaller code, but it can slow the compiler down. 10404 10405 This option is enabled at level '-Os' for all targets. 10406 10407'-fira-loop-pressure' 10408 Use IRA to evaluate register pressure in loops for decisions to 10409 move loop invariants. This option usually results in generation of 10410 faster and smaller code on machines with large register files (>= 10411 32 registers), but it can slow the compiler down. 10412 10413 This option is enabled at level '-O3' for some targets. 10414 10415'-fno-ira-share-save-slots' 10416 Disable sharing of stack slots used for saving call-used hard 10417 registers living through a call. Each hard register gets a 10418 separate stack slot, and as a result function stack frames are 10419 larger. 10420 10421'-fno-ira-share-spill-slots' 10422 Disable sharing of stack slots allocated for pseudo-registers. 10423 Each pseudo-register that does not get a hard register gets a 10424 separate stack slot, and as a result function stack frames are 10425 larger. 10426 10427'-flra-remat' 10428 Enable CFG-sensitive rematerialization in LRA. Instead of loading 10429 values of spilled pseudos, LRA tries to rematerialize (recalculate) 10430 values if it is profitable. 10431 10432 Enabled at levels '-O2', '-O3', '-Os'. 10433 10434'-fdelayed-branch' 10435 If supported for the target machine, attempt to reorder 10436 instructions to exploit instruction slots available after delayed 10437 branch instructions. 10438 10439 Enabled at levels '-O1', '-O2', '-O3', '-Os', but not at '-Og'. 10440 10441'-fschedule-insns' 10442 If supported for the target machine, attempt to reorder 10443 instructions to eliminate execution stalls due to required data 10444 being unavailable. This helps machines that have slow floating 10445 point or memory load instructions by allowing other instructions to 10446 be issued until the result of the load or floating-point 10447 instruction is required. 10448 10449 Enabled at levels '-O2', '-O3'. 10450 10451'-fschedule-insns2' 10452 Similar to '-fschedule-insns', but requests an additional pass of 10453 instruction scheduling after register allocation has been done. 10454 This is especially useful on machines with a relatively small 10455 number of registers and where memory load instructions take more 10456 than one cycle. 10457 10458 Enabled at levels '-O2', '-O3', '-Os'. 10459 10460'-fno-sched-interblock' 10461 Disable instruction scheduling across basic blocks, which is 10462 normally enabled when scheduling before register allocation, i.e. 10463 with '-fschedule-insns' or at '-O2' or higher. 10464 10465'-fno-sched-spec' 10466 Disable speculative motion of non-load instructions, which is 10467 normally enabled when scheduling before register allocation, i.e. 10468 with '-fschedule-insns' or at '-O2' or higher. 10469 10470'-fsched-pressure' 10471 Enable register pressure sensitive insn scheduling before register 10472 allocation. This only makes sense when scheduling before register 10473 allocation is enabled, i.e. with '-fschedule-insns' or at '-O2' or 10474 higher. Usage of this option can improve the generated code and 10475 decrease its size by preventing register pressure increase above 10476 the number of available hard registers and subsequent spills in 10477 register allocation. 10478 10479'-fsched-spec-load' 10480 Allow speculative motion of some load instructions. This only 10481 makes sense when scheduling before register allocation, i.e. with 10482 '-fschedule-insns' or at '-O2' or higher. 10483 10484'-fsched-spec-load-dangerous' 10485 Allow speculative motion of more load instructions. This only 10486 makes sense when scheduling before register allocation, i.e. with 10487 '-fschedule-insns' or at '-O2' or higher. 10488 10489'-fsched-stalled-insns' 10490'-fsched-stalled-insns=N' 10491 Define how many insns (if any) can be moved prematurely from the 10492 queue of stalled insns into the ready list during the second 10493 scheduling pass. '-fno-sched-stalled-insns' means that no insns 10494 are moved prematurely, '-fsched-stalled-insns=0' means there is no 10495 limit on how many queued insns can be moved prematurely. 10496 '-fsched-stalled-insns' without a value is equivalent to 10497 '-fsched-stalled-insns=1'. 10498 10499'-fsched-stalled-insns-dep' 10500'-fsched-stalled-insns-dep=N' 10501 Define how many insn groups (cycles) are examined for a dependency 10502 on a stalled insn that is a candidate for premature removal from 10503 the queue of stalled insns. This has an effect only during the 10504 second scheduling pass, and only if '-fsched-stalled-insns' is 10505 used. '-fno-sched-stalled-insns-dep' is equivalent to 10506 '-fsched-stalled-insns-dep=0'. '-fsched-stalled-insns-dep' without 10507 a value is equivalent to '-fsched-stalled-insns-dep=1'. 10508 10509'-fsched2-use-superblocks' 10510 When scheduling after register allocation, use superblock 10511 scheduling. This allows motion across basic block boundaries, 10512 resulting in faster schedules. This option is experimental, as not 10513 all machine descriptions used by GCC model the CPU closely enough 10514 to avoid unreliable results from the algorithm. 10515 10516 This only makes sense when scheduling after register allocation, 10517 i.e. with '-fschedule-insns2' or at '-O2' or higher. 10518 10519'-fsched-group-heuristic' 10520 Enable the group heuristic in the scheduler. This heuristic favors 10521 the instruction that belongs to a schedule group. This is enabled 10522 by default when scheduling is enabled, i.e. with '-fschedule-insns' 10523 or '-fschedule-insns2' or at '-O2' or higher. 10524 10525'-fsched-critical-path-heuristic' 10526 Enable the critical-path heuristic in the scheduler. This 10527 heuristic favors instructions on the critical path. This is 10528 enabled by default when scheduling is enabled, i.e. with 10529 '-fschedule-insns' or '-fschedule-insns2' or at '-O2' or higher. 10530 10531'-fsched-spec-insn-heuristic' 10532 Enable the speculative instruction heuristic in the scheduler. 10533 This heuristic favors speculative instructions with greater 10534 dependency weakness. This is enabled by default when scheduling is 10535 enabled, i.e. with '-fschedule-insns' or '-fschedule-insns2' or at 10536 '-O2' or higher. 10537 10538'-fsched-rank-heuristic' 10539 Enable the rank heuristic in the scheduler. This heuristic favors 10540 the instruction belonging to a basic block with greater size or 10541 frequency. This is enabled by default when scheduling is enabled, 10542 i.e. with '-fschedule-insns' or '-fschedule-insns2' or at '-O2' or 10543 higher. 10544 10545'-fsched-last-insn-heuristic' 10546 Enable the last-instruction heuristic in the scheduler. This 10547 heuristic favors the instruction that is less dependent on the last 10548 instruction scheduled. This is enabled by default when scheduling 10549 is enabled, i.e. with '-fschedule-insns' or '-fschedule-insns2' or 10550 at '-O2' or higher. 10551 10552'-fsched-dep-count-heuristic' 10553 Enable the dependent-count heuristic in the scheduler. This 10554 heuristic favors the instruction that has more instructions 10555 depending on it. This is enabled by default when scheduling is 10556 enabled, i.e. with '-fschedule-insns' or '-fschedule-insns2' or at 10557 '-O2' or higher. 10558 10559'-freschedule-modulo-scheduled-loops' 10560 Modulo scheduling is performed before traditional scheduling. If a 10561 loop is modulo scheduled, later scheduling passes may change its 10562 schedule. Use this option to control that behavior. 10563 10564'-fselective-scheduling' 10565 Schedule instructions using selective scheduling algorithm. 10566 Selective scheduling runs instead of the first scheduler pass. 10567 10568'-fselective-scheduling2' 10569 Schedule instructions using selective scheduling algorithm. 10570 Selective scheduling runs instead of the second scheduler pass. 10571 10572'-fsel-sched-pipelining' 10573 Enable software pipelining of innermost loops during selective 10574 scheduling. This option has no effect unless one of 10575 '-fselective-scheduling' or '-fselective-scheduling2' is turned on. 10576 10577'-fsel-sched-pipelining-outer-loops' 10578 When pipelining loops during selective scheduling, also pipeline 10579 outer loops. This option has no effect unless 10580 '-fsel-sched-pipelining' is turned on. 10581 10582'-fsemantic-interposition' 10583 Some object formats, like ELF, allow interposing of symbols by the 10584 dynamic linker. This means that for symbols exported from the DSO, 10585 the compiler cannot perform interprocedural propagation, inlining 10586 and other optimizations in anticipation that the function or 10587 variable in question may change. While this feature is useful, for 10588 example, to rewrite memory allocation functions by a debugging 10589 implementation, it is expensive in the terms of code quality. With 10590 '-fno-semantic-interposition' the compiler assumes that if 10591 interposition happens for functions the overwriting function will 10592 have precisely the same semantics (and side effects). Similarly if 10593 interposition happens for variables, the constructor of the 10594 variable will be the same. The flag has no effect for functions 10595 explicitly declared inline (where it is never allowed for 10596 interposition to change semantics) and for symbols explicitly 10597 declared weak. 10598 10599'-fshrink-wrap' 10600 Emit function prologues only before parts of the function that need 10601 it, rather than at the top of the function. This flag is enabled 10602 by default at '-O' and higher. 10603 10604'-fshrink-wrap-separate' 10605 Shrink-wrap separate parts of the prologue and epilogue separately, 10606 so that those parts are only executed when needed. This option is 10607 on by default, but has no effect unless '-fshrink-wrap' is also 10608 turned on and the target supports this. 10609 10610'-fcaller-saves' 10611 Enable allocation of values to registers that are clobbered by 10612 function calls, by emitting extra instructions to save and restore 10613 the registers around such calls. Such allocation is done only when 10614 it seems to result in better code. 10615 10616 This option is always enabled by default on certain machines, 10617 usually those which have no call-preserved registers to use 10618 instead. 10619 10620 Enabled at levels '-O2', '-O3', '-Os'. 10621 10622'-fcombine-stack-adjustments' 10623 Tracks stack adjustments (pushes and pops) and stack memory 10624 references and then tries to find ways to combine them. 10625 10626 Enabled by default at '-O1' and higher. 10627 10628'-fipa-ra' 10629 Use caller save registers for allocation if those registers are not 10630 used by any called function. In that case it is not necessary to 10631 save and restore them around calls. This is only possible if 10632 called functions are part of same compilation unit as current 10633 function and they are compiled before it. 10634 10635 Enabled at levels '-O2', '-O3', '-Os', however the option is 10636 disabled if generated code will be instrumented for profiling 10637 ('-p', or '-pg') or if callee's register usage cannot be known 10638 exactly (this happens on targets that do not expose prologues and 10639 epilogues in RTL). 10640 10641'-fconserve-stack' 10642 Attempt to minimize stack usage. The compiler attempts to use less 10643 stack space, even if that makes the program slower. This option 10644 implies setting the 'large-stack-frame' parameter to 100 and the 10645 'large-stack-frame-growth' parameter to 400. 10646 10647'-ftree-reassoc' 10648 Perform reassociation on trees. This flag is enabled by default at 10649 '-O1' and higher. 10650 10651'-fcode-hoisting' 10652 Perform code hoisting. Code hoisting tries to move the evaluation 10653 of expressions executed on all paths to the function exit as early 10654 as possible. This is especially useful as a code size 10655 optimization, but it often helps for code speed as well. This flag 10656 is enabled by default at '-O2' and higher. 10657 10658'-ftree-pre' 10659 Perform partial redundancy elimination (PRE) on trees. This flag 10660 is enabled by default at '-O2' and '-O3'. 10661 10662'-ftree-partial-pre' 10663 Make partial redundancy elimination (PRE) more aggressive. This 10664 flag is enabled by default at '-O3'. 10665 10666'-ftree-forwprop' 10667 Perform forward propagation on trees. This flag is enabled by 10668 default at '-O1' and higher. 10669 10670'-ftree-fre' 10671 Perform full redundancy elimination (FRE) on trees. The difference 10672 between FRE and PRE is that FRE only considers expressions that are 10673 computed on all paths leading to the redundant computation. This 10674 analysis is faster than PRE, though it exposes fewer redundancies. 10675 This flag is enabled by default at '-O1' and higher. 10676 10677'-ftree-phiprop' 10678 Perform hoisting of loads from conditional pointers on trees. This 10679 pass is enabled by default at '-O1' and higher. 10680 10681'-fhoist-adjacent-loads' 10682 Speculatively hoist loads from both branches of an if-then-else if 10683 the loads are from adjacent locations in the same structure and the 10684 target architecture has a conditional move instruction. This flag 10685 is enabled by default at '-O2' and higher. 10686 10687'-ftree-copy-prop' 10688 Perform copy propagation on trees. This pass eliminates 10689 unnecessary copy operations. This flag is enabled by default at 10690 '-O1' and higher. 10691 10692'-fipa-pure-const' 10693 Discover which functions are pure or constant. Enabled by default 10694 at '-O1' and higher. 10695 10696'-fipa-reference' 10697 Discover which static variables do not escape the compilation unit. 10698 Enabled by default at '-O1' and higher. 10699 10700'-fipa-reference-addressable' 10701 Discover read-only, write-only and non-addressable static 10702 variables. Enabled by default at '-O1' and higher. 10703 10704'-fipa-stack-alignment' 10705 Reduce stack alignment on call sites if possible. Enabled by 10706 default. 10707 10708'-fipa-pta' 10709 Perform interprocedural pointer analysis and interprocedural 10710 modification and reference analysis. This option can cause 10711 excessive memory and compile-time usage on large compilation units. 10712 It is not enabled by default at any optimization level. 10713 10714'-fipa-profile' 10715 Perform interprocedural profile propagation. The functions called 10716 only from cold functions are marked as cold. Also functions 10717 executed once (such as 'cold', 'noreturn', static constructors or 10718 destructors) are identified. Cold functions and loop less parts of 10719 functions executed once are then optimized for size. Enabled by 10720 default at '-O1' and higher. 10721 10722'-fipa-modref' 10723 Perform interprocedural mod/ref analysis. This optimization 10724 analyzes the side effects of functions (memory locations that are 10725 modified or referenced) and enables better optimization across the 10726 function call boundary. This flag is enabled by default at '-O1' 10727 and higher. 10728 10729'-fipa-cp' 10730 Perform interprocedural constant propagation. This optimization 10731 analyzes the program to determine when values passed to functions 10732 are constants and then optimizes accordingly. This optimization 10733 can substantially increase performance if the application has 10734 constants passed to functions. This flag is enabled by default at 10735 '-O2', '-Os' and '-O3'. It is also enabled by '-fprofile-use' and 10736 '-fauto-profile'. 10737 10738'-fipa-cp-clone' 10739 Perform function cloning to make interprocedural constant 10740 propagation stronger. When enabled, interprocedural constant 10741 propagation performs function cloning when externally visible 10742 function can be called with constant arguments. Because this 10743 optimization can create multiple copies of functions, it may 10744 significantly increase code size (see '--param 10745 ipa-cp-unit-growth=VALUE'). This flag is enabled by default at 10746 '-O3'. It is also enabled by '-fprofile-use' and '-fauto-profile'. 10747 10748'-fipa-bit-cp' 10749 When enabled, perform interprocedural bitwise constant propagation. 10750 This flag is enabled by default at '-O2' and by '-fprofile-use' and 10751 '-fauto-profile'. It requires that '-fipa-cp' is enabled. 10752 10753'-fipa-vrp' 10754 When enabled, perform interprocedural propagation of value ranges. 10755 This flag is enabled by default at '-O2'. It requires that 10756 '-fipa-cp' is enabled. 10757 10758'-fipa-icf' 10759 Perform Identical Code Folding for functions and read-only 10760 variables. The optimization reduces code size and may disturb 10761 unwind stacks by replacing a function by equivalent one with a 10762 different name. The optimization works more effectively with 10763 link-time optimization enabled. 10764 10765 Although the behavior is similar to the Gold Linker's ICF 10766 optimization, GCC ICF works on different levels and thus the 10767 optimizations are not same - there are equivalences that are found 10768 only by GCC and equivalences found only by Gold. 10769 10770 This flag is enabled by default at '-O2' and '-Os'. 10771 10772'-flive-patching=LEVEL' 10773 Control GCC's optimizations to produce output suitable for 10774 live-patching. 10775 10776 If the compiler's optimization uses a function's body or 10777 information extracted from its body to optimize/change another 10778 function, the latter is called an impacted function of the former. 10779 If a function is patched, its impacted functions should be patched 10780 too. 10781 10782 The impacted functions are determined by the compiler's 10783 interprocedural optimizations. For example, a caller is impacted 10784 when inlining a function into its caller, cloning a function and 10785 changing its caller to call this new clone, or extracting a 10786 function's pureness/constness information to optimize its direct or 10787 indirect callers, etc. 10788 10789 Usually, the more IPA optimizations enabled, the larger the number 10790 of impacted functions for each function. In order to control the 10791 number of impacted functions and more easily compute the list of 10792 impacted function, IPA optimizations can be partially enabled at 10793 two different levels. 10794 10795 The LEVEL argument should be one of the following: 10796 10797 'inline-clone' 10798 10799 Only enable inlining and cloning optimizations, which includes 10800 inlining, cloning, interprocedural scalar replacement of 10801 aggregates and partial inlining. As a result, when patching a 10802 function, all its callers and its clones' callers are 10803 impacted, therefore need to be patched as well. 10804 10805 '-flive-patching=inline-clone' disables the following 10806 optimization flags: 10807 -fwhole-program -fipa-pta -fipa-reference -fipa-ra 10808 -fipa-icf -fipa-icf-functions -fipa-icf-variables 10809 -fipa-bit-cp -fipa-vrp -fipa-pure-const -fipa-reference-addressable 10810 -fipa-stack-alignment -fipa-modref 10811 10812 'inline-only-static' 10813 10814 Only enable inlining of static functions. As a result, when 10815 patching a static function, all its callers are impacted and 10816 so need to be patched as well. 10817 10818 In addition to all the flags that 10819 '-flive-patching=inline-clone' disables, 10820 '-flive-patching=inline-only-static' disables the following 10821 additional optimization flags: 10822 -fipa-cp-clone -fipa-sra -fpartial-inlining -fipa-cp 10823 10824 When '-flive-patching' is specified without any value, the default 10825 value is INLINE-CLONE. 10826 10827 This flag is disabled by default. 10828 10829 Note that '-flive-patching' is not supported with link-time 10830 optimization ('-flto'). 10831 10832'-fisolate-erroneous-paths-dereference' 10833 Detect paths that trigger erroneous or undefined behavior due to 10834 dereferencing a null pointer. Isolate those paths from the main 10835 control flow and turn the statement with erroneous or undefined 10836 behavior into a trap. This flag is enabled by default at '-O2' and 10837 higher and depends on '-fdelete-null-pointer-checks' also being 10838 enabled. 10839 10840'-fisolate-erroneous-paths-attribute' 10841 Detect paths that trigger erroneous or undefined behavior due to a 10842 null value being used in a way forbidden by a 'returns_nonnull' or 10843 'nonnull' attribute. Isolate those paths from the main control 10844 flow and turn the statement with erroneous or undefined behavior 10845 into a trap. This is not currently enabled, but may be enabled by 10846 '-O2' in the future. 10847 10848'-ftree-sink' 10849 Perform forward store motion on trees. This flag is enabled by 10850 default at '-O1' and higher. 10851 10852'-ftree-bit-ccp' 10853 Perform sparse conditional bit constant propagation on trees and 10854 propagate pointer alignment information. This pass only operates 10855 on local scalar variables and is enabled by default at '-O1' and 10856 higher, except for '-Og'. It requires that '-ftree-ccp' is 10857 enabled. 10858 10859'-ftree-ccp' 10860 Perform sparse conditional constant propagation (CCP) on trees. 10861 This pass only operates on local scalar variables and is enabled by 10862 default at '-O1' and higher. 10863 10864'-fssa-backprop' 10865 Propagate information about uses of a value up the definition chain 10866 in order to simplify the definitions. For example, this pass 10867 strips sign operations if the sign of a value never matters. The 10868 flag is enabled by default at '-O1' and higher. 10869 10870'-fssa-phiopt' 10871 Perform pattern matching on SSA PHI nodes to optimize conditional 10872 code. This pass is enabled by default at '-O1' and higher, except 10873 for '-Og'. 10874 10875'-ftree-switch-conversion' 10876 Perform conversion of simple initializations in a switch to 10877 initializations from a scalar array. This flag is enabled by 10878 default at '-O2' and higher. 10879 10880'-ftree-tail-merge' 10881 Look for identical code sequences. When found, replace one with a 10882 jump to the other. This optimization is known as tail merging or 10883 cross jumping. This flag is enabled by default at '-O2' and 10884 higher. The compilation time in this pass can be limited using 10885 'max-tail-merge-comparisons' parameter and 10886 'max-tail-merge-iterations' parameter. 10887 10888'-ftree-dce' 10889 Perform dead code elimination (DCE) on trees. This flag is enabled 10890 by default at '-O1' and higher. 10891 10892'-ftree-builtin-call-dce' 10893 Perform conditional dead code elimination (DCE) for calls to 10894 built-in functions that may set 'errno' but are otherwise free of 10895 side effects. This flag is enabled by default at '-O2' and higher 10896 if '-Os' is not also specified. 10897 10898'-ffinite-loops' 10899 Assume that a loop with an exit will eventually take the exit and 10900 not loop indefinitely. This allows the compiler to remove loops 10901 that otherwise have no side-effects, not considering eventual 10902 endless looping as such. 10903 10904 This option is enabled by default at '-O2' for C++ with -std=c++11 10905 or higher. 10906 10907'-ftree-dominator-opts' 10908 Perform a variety of simple scalar cleanups (constant/copy 10909 propagation, redundancy elimination, range propagation and 10910 expression simplification) based on a dominator tree traversal. 10911 This also performs jump threading (to reduce jumps to jumps). This 10912 flag is enabled by default at '-O1' and higher. 10913 10914'-ftree-dse' 10915 Perform dead store elimination (DSE) on trees. A dead store is a 10916 store into a memory location that is later overwritten by another 10917 store without any intervening loads. In this case the earlier 10918 store can be deleted. This flag is enabled by default at '-O1' and 10919 higher. 10920 10921'-ftree-ch' 10922 Perform loop header copying on trees. This is beneficial since it 10923 increases effectiveness of code motion optimizations. It also 10924 saves one jump. This flag is enabled by default at '-O1' and 10925 higher. It is not enabled for '-Os', since it usually increases 10926 code size. 10927 10928'-ftree-loop-optimize' 10929 Perform loop optimizations on trees. This flag is enabled by 10930 default at '-O1' and higher. 10931 10932'-ftree-loop-linear' 10933'-floop-strip-mine' 10934'-floop-block' 10935 Perform loop nest optimizations. Same as '-floop-nest-optimize'. 10936 To use this code transformation, GCC has to be configured with 10937 '--with-isl' to enable the Graphite loop transformation 10938 infrastructure. 10939 10940'-fgraphite-identity' 10941 Enable the identity transformation for graphite. For every SCoP we 10942 generate the polyhedral representation and transform it back to 10943 gimple. Using '-fgraphite-identity' we can check the costs or 10944 benefits of the GIMPLE -> GRAPHITE -> GIMPLE transformation. Some 10945 minimal optimizations are also performed by the code generator isl, 10946 like index splitting and dead code elimination in loops. 10947 10948'-floop-nest-optimize' 10949 Enable the isl based loop nest optimizer. This is a generic loop 10950 nest optimizer based on the Pluto optimization algorithms. It 10951 calculates a loop structure optimized for data-locality and 10952 parallelism. This option is experimental. 10953 10954'-floop-parallelize-all' 10955 Use the Graphite data dependence analysis to identify loops that 10956 can be parallelized. Parallelize all the loops that can be 10957 analyzed to not contain loop carried dependences without checking 10958 that it is profitable to parallelize the loops. 10959 10960'-ftree-coalesce-vars' 10961 While transforming the program out of the SSA representation, 10962 attempt to reduce copying by coalescing versions of different 10963 user-defined variables, instead of just compiler temporaries. This 10964 may severely limit the ability to debug an optimized program 10965 compiled with '-fno-var-tracking-assignments'. In the negated 10966 form, this flag prevents SSA coalescing of user variables. This 10967 option is enabled by default if optimization is enabled, and it 10968 does very little otherwise. 10969 10970'-ftree-loop-if-convert' 10971 Attempt to transform conditional jumps in the innermost loops to 10972 branch-less equivalents. The intent is to remove control-flow from 10973 the innermost loops in order to improve the ability of the 10974 vectorization pass to handle these loops. This is enabled by 10975 default if vectorization is enabled. 10976 10977'-ftree-loop-distribution' 10978 Perform loop distribution. This flag can improve cache performance 10979 on big loop bodies and allow further loop optimizations, like 10980 parallelization or vectorization, to take place. For example, the 10981 loop 10982 DO I = 1, N 10983 A(I) = B(I) + C 10984 D(I) = E(I) * F 10985 ENDDO 10986 is transformed to 10987 DO I = 1, N 10988 A(I) = B(I) + C 10989 ENDDO 10990 DO I = 1, N 10991 D(I) = E(I) * F 10992 ENDDO 10993 This flag is enabled by default at '-O3'. It is also enabled by 10994 '-fprofile-use' and '-fauto-profile'. 10995 10996'-ftree-loop-distribute-patterns' 10997 Perform loop distribution of patterns that can be code generated 10998 with calls to a library. This flag is enabled by default at '-O2' 10999 and higher, and by '-fprofile-use' and '-fauto-profile'. 11000 11001 This pass distributes the initialization loops and generates a call 11002 to memset zero. For example, the loop 11003 DO I = 1, N 11004 A(I) = 0 11005 B(I) = A(I) + I 11006 ENDDO 11007 is transformed to 11008 DO I = 1, N 11009 A(I) = 0 11010 ENDDO 11011 DO I = 1, N 11012 B(I) = A(I) + I 11013 ENDDO 11014 and the initialization loop is transformed into a call to memset 11015 zero. This flag is enabled by default at '-O3'. It is also 11016 enabled by '-fprofile-use' and '-fauto-profile'. 11017 11018'-floop-interchange' 11019 Perform loop interchange outside of graphite. This flag can 11020 improve cache performance on loop nest and allow further loop 11021 optimizations, like vectorization, to take place. For example, the 11022 loop 11023 for (int i = 0; i < N; i++) 11024 for (int j = 0; j < N; j++) 11025 for (int k = 0; k < N; k++) 11026 c[i][j] = c[i][j] + a[i][k]*b[k][j]; 11027 is transformed to 11028 for (int i = 0; i < N; i++) 11029 for (int k = 0; k < N; k++) 11030 for (int j = 0; j < N; j++) 11031 c[i][j] = c[i][j] + a[i][k]*b[k][j]; 11032 This flag is enabled by default at '-O3'. It is also enabled by 11033 '-fprofile-use' and '-fauto-profile'. 11034 11035'-floop-unroll-and-jam' 11036 Apply unroll and jam transformations on feasible loops. In a loop 11037 nest this unrolls the outer loop by some factor and fuses the 11038 resulting multiple inner loops. This flag is enabled by default at 11039 '-O3'. It is also enabled by '-fprofile-use' and '-fauto-profile'. 11040 11041'-ftree-loop-im' 11042 Perform loop invariant motion on trees. This pass moves only 11043 invariants that are hard to handle at RTL level (function calls, 11044 operations that expand to nontrivial sequences of insns). With 11045 '-funswitch-loops' it also moves operands of conditions that are 11046 invariant out of the loop, so that we can use just trivial 11047 invariantness analysis in loop unswitching. The pass also includes 11048 store motion. 11049 11050'-ftree-loop-ivcanon' 11051 Create a canonical counter for number of iterations in loops for 11052 which determining number of iterations requires complicated 11053 analysis. Later optimizations then may determine the number 11054 easily. Useful especially in connection with unrolling. 11055 11056'-ftree-scev-cprop' 11057 Perform final value replacement. If a variable is modified in a 11058 loop in such a way that its value when exiting the loop can be 11059 determined using only its initial value and the number of loop 11060 iterations, replace uses of the final value by such a computation, 11061 provided it is sufficiently cheap. This reduces data dependencies 11062 and may allow further simplifications. Enabled by default at '-O1' 11063 and higher. 11064 11065'-fivopts' 11066 Perform induction variable optimizations (strength reduction, 11067 induction variable merging and induction variable elimination) on 11068 trees. 11069 11070'-ftree-parallelize-loops=n' 11071 Parallelize loops, i.e., split their iteration space to run in n 11072 threads. This is only possible for loops whose iterations are 11073 independent and can be arbitrarily reordered. The optimization is 11074 only profitable on multiprocessor machines, for loops that are 11075 CPU-intensive, rather than constrained e.g. by memory bandwidth. 11076 This option implies '-pthread', and thus is only supported on 11077 targets that have support for '-pthread'. 11078 11079'-ftree-pta' 11080 Perform function-local points-to analysis on trees. This flag is 11081 enabled by default at '-O1' and higher, except for '-Og'. 11082 11083'-ftree-sra' 11084 Perform scalar replacement of aggregates. This pass replaces 11085 structure references with scalars to prevent committing structures 11086 to memory too early. This flag is enabled by default at '-O1' and 11087 higher, except for '-Og'. 11088 11089'-fstore-merging' 11090 Perform merging of narrow stores to consecutive memory addresses. 11091 This pass merges contiguous stores of immediate values narrower 11092 than a word into fewer wider stores to reduce the number of 11093 instructions. This is enabled by default at '-O2' and higher as 11094 well as '-Os'. 11095 11096'-ftree-ter' 11097 Perform temporary expression replacement during the SSA->normal 11098 phase. Single use/single def temporaries are replaced at their use 11099 location with their defining expression. This results in 11100 non-GIMPLE code, but gives the expanders much more complex trees to 11101 work on resulting in better RTL generation. This is enabled by 11102 default at '-O1' and higher. 11103 11104'-ftree-slsr' 11105 Perform straight-line strength reduction on trees. This recognizes 11106 related expressions involving multiplications and replaces them by 11107 less expensive calculations when possible. This is enabled by 11108 default at '-O1' and higher. 11109 11110'-ftree-vectorize' 11111 Perform vectorization on trees. This flag enables 11112 '-ftree-loop-vectorize' and '-ftree-slp-vectorize' if not 11113 explicitly specified. 11114 11115'-ftree-loop-vectorize' 11116 Perform loop vectorization on trees. This flag is enabled by 11117 default at '-O2' and by '-ftree-vectorize', '-fprofile-use', and 11118 '-fauto-profile'. 11119 11120'-ftree-slp-vectorize' 11121 Perform basic block vectorization on trees. This flag is enabled 11122 by default at '-O2' and by '-ftree-vectorize', '-fprofile-use', and 11123 '-fauto-profile'. 11124 11125'-ftrivial-auto-var-init=CHOICE' 11126 Initialize automatic variables with either a pattern or with zeroes 11127 to increase the security and predictability of a program by 11128 preventing uninitialized memory disclosure and use. GCC still 11129 considers an automatic variable that doesn't have an explicit 11130 initializer as uninitialized, '-Wuninitialized' and 11131 '-Wanalyzer-use-of-uninitialized-value' will still report warning 11132 messages on such automatic variables. With this option, GCC will 11133 also initialize any padding of automatic variables that have 11134 structure or union types to zeroes. However, the current 11135 implementation cannot initialize automatic variables that are 11136 declared between the controlling expression and the first case of a 11137 'switch' statement. Using '-Wtrivial-auto-var-init' to report all 11138 such cases. 11139 11140 The three values of CHOICE are: 11141 11142 * 'uninitialized' doesn't initialize any automatic variables. 11143 This is C and C++'s default. 11144 11145 * 'pattern' Initialize automatic variables with values which 11146 will likely transform logic bugs into crashes down the line, 11147 are easily recognized in a crash dump and without being values 11148 that programmers can rely on for useful program semantics. 11149 The current value is byte-repeatable pattern with byte "0xFE". 11150 The values used for pattern initialization might be changed in 11151 the future. 11152 11153 * 'zero' Initialize automatic variables with zeroes. 11154 11155 The default is 'uninitialized'. 11156 11157 You can control this behavior for a specific variable by using the 11158 variable attribute 'uninitialized' (*note Variable Attributes::). 11159 11160'-fvect-cost-model=MODEL' 11161 Alter the cost model used for vectorization. The MODEL argument 11162 should be one of 'unlimited', 'dynamic', 'cheap' or 'very-cheap'. 11163 With the 'unlimited' model the vectorized code-path is assumed to 11164 be profitable while with the 'dynamic' model a runtime check guards 11165 the vectorized code-path to enable it only for iteration counts 11166 that will likely execute faster than when executing the original 11167 scalar loop. The 'cheap' model disables vectorization of loops 11168 where doing so would be cost prohibitive for example due to 11169 required runtime checks for data dependence or alignment but 11170 otherwise is equal to the 'dynamic' model. The 'very-cheap' model 11171 only allows vectorization if the vector code would entirely replace 11172 the scalar code that is being vectorized. For example, if each 11173 iteration of a vectorized loop would only be able to handle exactly 11174 four iterations of the scalar loop, the 'very-cheap' model would 11175 only allow vectorization if the scalar iteration count is known to 11176 be a multiple of four. 11177 11178 The default cost model depends on other optimization flags and is 11179 either 'dynamic' or 'cheap'. 11180 11181'-fsimd-cost-model=MODEL' 11182 Alter the cost model used for vectorization of loops marked with 11183 the OpenMP simd directive. The MODEL argument should be one of 11184 'unlimited', 'dynamic', 'cheap'. All values of MODEL have the same 11185 meaning as described in '-fvect-cost-model' and by default a cost 11186 model defined with '-fvect-cost-model' is used. 11187 11188'-ftree-vrp' 11189 Perform Value Range Propagation on trees. This is similar to the 11190 constant propagation pass, but instead of values, ranges of values 11191 are propagated. This allows the optimizers to remove unnecessary 11192 range checks like array bound checks and null pointer checks. This 11193 is enabled by default at '-O2' and higher. Null pointer check 11194 elimination is only done if '-fdelete-null-pointer-checks' is 11195 enabled. 11196 11197'-fsplit-paths' 11198 Split paths leading to loop backedges. This can improve dead code 11199 elimination and common subexpression elimination. This is enabled 11200 by default at '-O3' and above. 11201 11202'-fsplit-ivs-in-unroller' 11203 Enables expression of values of induction variables in later 11204 iterations of the unrolled loop using the value in the first 11205 iteration. This breaks long dependency chains, thus improving 11206 efficiency of the scheduling passes. 11207 11208 A combination of '-fweb' and CSE is often sufficient to obtain the 11209 same effect. However, that is not reliable in cases where the loop 11210 body is more complicated than a single basic block. It also does 11211 not work at all on some architectures due to restrictions in the 11212 CSE pass. 11213 11214 This optimization is enabled by default. 11215 11216'-fvariable-expansion-in-unroller' 11217 With this option, the compiler creates multiple copies of some 11218 local variables when unrolling a loop, which can result in superior 11219 code. 11220 11221 This optimization is enabled by default for PowerPC targets, but 11222 disabled by default otherwise. 11223 11224'-fpartial-inlining' 11225 Inline parts of functions. This option has any effect only when 11226 inlining itself is turned on by the '-finline-functions' or 11227 '-finline-small-functions' options. 11228 11229 Enabled at levels '-O2', '-O3', '-Os'. 11230 11231'-fpredictive-commoning' 11232 Perform predictive commoning optimization, i.e., reusing 11233 computations (especially memory loads and stores) performed in 11234 previous iterations of loops. 11235 11236 This option is enabled at level '-O3'. It is also enabled by 11237 '-fprofile-use' and '-fauto-profile'. 11238 11239'-fprefetch-loop-arrays' 11240 If supported by the target machine, generate instructions to 11241 prefetch memory to improve the performance of loops that access 11242 large arrays. 11243 11244 This option may generate better or worse code; results are highly 11245 dependent on the structure of loops within the source code. 11246 11247 Disabled at level '-Os'. 11248 11249'-fno-printf-return-value' 11250 Do not substitute constants for known return value of formatted 11251 output functions such as 'sprintf', 'snprintf', 'vsprintf', and 11252 'vsnprintf' (but not 'printf' of 'fprintf'). This transformation 11253 allows GCC to optimize or even eliminate branches based on the 11254 known return value of these functions called with arguments that 11255 are either constant, or whose values are known to be in a range 11256 that makes determining the exact return value possible. For 11257 example, when '-fprintf-return-value' is in effect, both the branch 11258 and the body of the 'if' statement (but not the call to 'snprint') 11259 can be optimized away when 'i' is a 32-bit or smaller integer 11260 because the return value is guaranteed to be at most 8. 11261 11262 char buf[9]; 11263 if (snprintf (buf, "%08x", i) >= sizeof buf) 11264 ... 11265 11266 The '-fprintf-return-value' option relies on other optimizations 11267 and yields best results with '-O2' and above. It works in tandem 11268 with the '-Wformat-overflow' and '-Wformat-truncation' options. 11269 The '-fprintf-return-value' option is enabled by default. 11270 11271'-fno-peephole' 11272'-fno-peephole2' 11273 Disable any machine-specific peephole optimizations. The 11274 difference between '-fno-peephole' and '-fno-peephole2' is in how 11275 they are implemented in the compiler; some targets use one, some 11276 use the other, a few use both. 11277 11278 '-fpeephole' is enabled by default. '-fpeephole2' enabled at 11279 levels '-O2', '-O3', '-Os'. 11280 11281'-fno-guess-branch-probability' 11282 Do not guess branch probabilities using heuristics. 11283 11284 GCC uses heuristics to guess branch probabilities if they are not 11285 provided by profiling feedback ('-fprofile-arcs'). These 11286 heuristics are based on the control flow graph. If some branch 11287 probabilities are specified by '__builtin_expect', then the 11288 heuristics are used to guess branch probabilities for the rest of 11289 the control flow graph, taking the '__builtin_expect' info into 11290 account. The interactions between the heuristics and 11291 '__builtin_expect' can be complex, and in some cases, it may be 11292 useful to disable the heuristics so that the effects of 11293 '__builtin_expect' are easier to understand. 11294 11295 It is also possible to specify expected probability of the 11296 expression with '__builtin_expect_with_probability' built-in 11297 function. 11298 11299 The default is '-fguess-branch-probability' at levels '-O', '-O2', 11300 '-O3', '-Os'. 11301 11302'-freorder-blocks' 11303 Reorder basic blocks in the compiled function in order to reduce 11304 number of taken branches and improve code locality. 11305 11306 Enabled at levels '-O1', '-O2', '-O3', '-Os'. 11307 11308'-freorder-blocks-algorithm=ALGORITHM' 11309 Use the specified algorithm for basic block reordering. The 11310 ALGORITHM argument can be 'simple', which does not increase code 11311 size (except sometimes due to secondary effects like alignment), or 11312 'stc', the "software trace cache" algorithm, which tries to put all 11313 often executed code together, minimizing the number of branches 11314 executed by making extra copies of code. 11315 11316 The default is 'simple' at levels '-O1', '-Os', and 'stc' at levels 11317 '-O2', '-O3'. 11318 11319'-freorder-blocks-and-partition' 11320 In addition to reordering basic blocks in the compiled function, in 11321 order to reduce number of taken branches, partitions hot and cold 11322 basic blocks into separate sections of the assembly and '.o' files, 11323 to improve paging and cache locality performance. 11324 11325 This optimization is automatically turned off in the presence of 11326 exception handling or unwind tables (on targets using 11327 setjump/longjump or target specific scheme), for linkonce sections, 11328 for functions with a user-defined section attribute and on any 11329 architecture that does not support named sections. When 11330 '-fsplit-stack' is used this option is not enabled by default (to 11331 avoid linker errors), but may be enabled explicitly (if using a 11332 working linker). 11333 11334 Enabled for x86 at levels '-O2', '-O3', '-Os'. 11335 11336'-freorder-functions' 11337 Reorder functions in the object file in order to improve code 11338 locality. This is implemented by using special subsections 11339 '.text.hot' for most frequently executed functions and 11340 '.text.unlikely' for unlikely executed functions. Reordering is 11341 done by the linker so object file format must support named 11342 sections and linker must place them in a reasonable way. 11343 11344 This option isn't effective unless you either provide profile 11345 feedback (see '-fprofile-arcs' for details) or manually annotate 11346 functions with 'hot' or 'cold' attributes (*note Common Function 11347 Attributes::). 11348 11349 Enabled at levels '-O2', '-O3', '-Os'. 11350 11351'-fstrict-aliasing' 11352 Allow the compiler to assume the strictest aliasing rules 11353 applicable to the language being compiled. For C (and C++), this 11354 activates optimizations based on the type of expressions. In 11355 particular, an object of one type is assumed never to reside at the 11356 same address as an object of a different type, unless the types are 11357 almost the same. For example, an 'unsigned int' can alias an 11358 'int', but not a 'void*' or a 'double'. A character type may alias 11359 any other type. 11360 11361 Pay special attention to code like this: 11362 union a_union { 11363 int i; 11364 double d; 11365 }; 11366 11367 int f() { 11368 union a_union t; 11369 t.d = 3.0; 11370 return t.i; 11371 } 11372 The practice of reading from a different union member than the one 11373 most recently written to (called "type-punning") is common. Even 11374 with '-fstrict-aliasing', type-punning is allowed, provided the 11375 memory is accessed through the union type. So, the code above 11376 works as expected. *Note Structures unions enumerations and 11377 bit-fields implementation::. However, this code might not: 11378 int f() { 11379 union a_union t; 11380 int* ip; 11381 t.d = 3.0; 11382 ip = &t.i; 11383 return *ip; 11384 } 11385 11386 Similarly, access by taking the address, casting the resulting 11387 pointer and dereferencing the result has undefined behavior, even 11388 if the cast uses a union type, e.g.: 11389 int f() { 11390 double d = 3.0; 11391 return ((union a_union *) &d)->i; 11392 } 11393 11394 The '-fstrict-aliasing' option is enabled at levels '-O2', '-O3', 11395 '-Os'. 11396 11397'-fipa-strict-aliasing' 11398 Controls whether rules of '-fstrict-aliasing' are applied across 11399 function boundaries. Note that if multiple functions gets inlined 11400 into a single function the memory accesses are no longer considered 11401 to be crossing a function boundary. 11402 11403 The '-fipa-strict-aliasing' option is enabled by default and is 11404 effective only in combination with '-fstrict-aliasing'. 11405 11406'-falign-functions' 11407'-falign-functions=N' 11408'-falign-functions=N:M' 11409'-falign-functions=N:M:N2' 11410'-falign-functions=N:M:N2:M2' 11411 Align the start of functions to the next power-of-two greater than 11412 or equal to N, skipping up to M-1 bytes. This ensures that at 11413 least the first M bytes of the function can be fetched by the CPU 11414 without crossing an N-byte alignment boundary. 11415 11416 If M is not specified, it defaults to N. 11417 11418 Examples: '-falign-functions=32' aligns functions to the next 11419 32-byte boundary, '-falign-functions=24' aligns to the next 32-byte 11420 boundary only if this can be done by skipping 23 bytes or less, 11421 '-falign-functions=32:7' aligns to the next 32-byte boundary only 11422 if this can be done by skipping 6 bytes or less. 11423 11424 The second pair of N2:M2 values allows you to specify a secondary 11425 alignment: '-falign-functions=64:7:32:3' aligns to the next 64-byte 11426 boundary if this can be done by skipping 6 bytes or less, otherwise 11427 aligns to the next 32-byte boundary if this can be done by skipping 11428 2 bytes or less. If M2 is not specified, it defaults to N2. 11429 11430 Some assemblers only support this flag when N is a power of two; in 11431 that case, it is rounded up. 11432 11433 '-fno-align-functions' and '-falign-functions=1' are equivalent and 11434 mean that functions are not aligned. 11435 11436 If N is not specified or is zero, use a machine-dependent default. 11437 The maximum allowed N option value is 65536. 11438 11439 Enabled at levels '-O2', '-O3'. 11440 11441'-flimit-function-alignment' 11442 If this option is enabled, the compiler tries to avoid 11443 unnecessarily overaligning functions. It attempts to instruct the 11444 assembler to align by the amount specified by '-falign-functions', 11445 but not to skip more bytes than the size of the function. 11446 11447'-falign-labels' 11448'-falign-labels=N' 11449'-falign-labels=N:M' 11450'-falign-labels=N:M:N2' 11451'-falign-labels=N:M:N2:M2' 11452 Align all branch targets to a power-of-two boundary. 11453 11454 Parameters of this option are analogous to the '-falign-functions' 11455 option. '-fno-align-labels' and '-falign-labels=1' are equivalent 11456 and mean that labels are not aligned. 11457 11458 If '-falign-loops' or '-falign-jumps' are applicable and are 11459 greater than this value, then their values are used instead. 11460 11461 If N is not specified or is zero, use a machine-dependent default 11462 which is very likely to be '1', meaning no alignment. The maximum 11463 allowed N option value is 65536. 11464 11465 Enabled at levels '-O2', '-O3'. 11466 11467'-falign-loops' 11468'-falign-loops=N' 11469'-falign-loops=N:M' 11470'-falign-loops=N:M:N2' 11471'-falign-loops=N:M:N2:M2' 11472 Align loops to a power-of-two boundary. If the loops are executed 11473 many times, this makes up for any execution of the dummy padding 11474 instructions. 11475 11476 If '-falign-labels' is greater than this value, then its value is 11477 used instead. 11478 11479 Parameters of this option are analogous to the '-falign-functions' 11480 option. '-fno-align-loops' and '-falign-loops=1' are equivalent 11481 and mean that loops are not aligned. The maximum allowed N option 11482 value is 65536. 11483 11484 If N is not specified or is zero, use a machine-dependent default. 11485 11486 Enabled at levels '-O2', '-O3'. 11487 11488'-falign-jumps' 11489'-falign-jumps=N' 11490'-falign-jumps=N:M' 11491'-falign-jumps=N:M:N2' 11492'-falign-jumps=N:M:N2:M2' 11493 Align branch targets to a power-of-two boundary, for branch targets 11494 where the targets can only be reached by jumping. In this case, no 11495 dummy operations need be executed. 11496 11497 If '-falign-labels' is greater than this value, then its value is 11498 used instead. 11499 11500 Parameters of this option are analogous to the '-falign-functions' 11501 option. '-fno-align-jumps' and '-falign-jumps=1' are equivalent 11502 and mean that loops are not aligned. 11503 11504 If N is not specified or is zero, use a machine-dependent default. 11505 The maximum allowed N option value is 65536. 11506 11507 Enabled at levels '-O2', '-O3'. 11508 11509'-fno-allocation-dce' 11510 Do not remove unused C++ allocations in dead code elimination. 11511 11512'-fallow-store-data-races' 11513 Allow the compiler to perform optimizations that may introduce new 11514 data races on stores, without proving that the variable cannot be 11515 concurrently accessed by other threads. Does not affect 11516 optimization of local data. It is safe to use this option if it is 11517 known that global data will not be accessed by multiple threads. 11518 11519 Examples of optimizations enabled by '-fallow-store-data-races' 11520 include hoisting or if-conversions that may cause a value that was 11521 already in memory to be re-written with that same value. Such 11522 re-writing is safe in a single threaded context but may be unsafe 11523 in a multi-threaded context. Note that on some processors, 11524 if-conversions may be required in order to enable vectorization. 11525 11526 Enabled at level '-Ofast'. 11527 11528'-funit-at-a-time' 11529 This option is left for compatibility reasons. '-funit-at-a-time' 11530 has no effect, while '-fno-unit-at-a-time' implies 11531 '-fno-toplevel-reorder' and '-fno-section-anchors'. 11532 11533 Enabled by default. 11534 11535'-fno-toplevel-reorder' 11536 Do not reorder top-level functions, variables, and 'asm' 11537 statements. Output them in the same order that they appear in the 11538 input file. When this option is used, unreferenced static 11539 variables are not removed. This option is intended to support 11540 existing code that relies on a particular ordering. For new code, 11541 it is better to use attributes when possible. 11542 11543 '-ftoplevel-reorder' is the default at '-O1' and higher, and also 11544 at '-O0' if '-fsection-anchors' is explicitly requested. 11545 Additionally '-fno-toplevel-reorder' implies 11546 '-fno-section-anchors'. 11547 11548'-fweb' 11549 Constructs webs as commonly used for register allocation purposes 11550 and assign each web individual pseudo register. This allows the 11551 register allocation pass to operate on pseudos directly, but also 11552 strengthens several other optimization passes, such as CSE, loop 11553 optimizer and trivial dead code remover. It can, however, make 11554 debugging impossible, since variables no longer stay in a "home 11555 register". 11556 11557 Enabled by default with '-funroll-loops'. 11558 11559'-fwhole-program' 11560 Assume that the current compilation unit represents the whole 11561 program being compiled. All public functions and variables with 11562 the exception of 'main' and those merged by attribute 11563 'externally_visible' become static functions and in effect are 11564 optimized more aggressively by interprocedural optimizers. 11565 11566 This option should not be used in combination with '-flto'. 11567 Instead relying on a linker plugin should provide safer and more 11568 precise information. 11569 11570'-flto[=N]' 11571 This option runs the standard link-time optimizer. When invoked 11572 with source code, it generates GIMPLE (one of GCC's internal 11573 representations) and writes it to special ELF sections in the 11574 object file. When the object files are linked together, all the 11575 function bodies are read from these ELF sections and instantiated 11576 as if they had been part of the same translation unit. 11577 11578 To use the link-time optimizer, '-flto' and optimization options 11579 should be specified at compile time and during the final link. It 11580 is recommended that you compile all the files participating in the 11581 same link with the same options and also specify those options at 11582 link time. For example: 11583 11584 gcc -c -O2 -flto foo.c 11585 gcc -c -O2 -flto bar.c 11586 gcc -o myprog -flto -O2 foo.o bar.o 11587 11588 The first two invocations to GCC save a bytecode representation of 11589 GIMPLE into special ELF sections inside 'foo.o' and 'bar.o'. The 11590 final invocation reads the GIMPLE bytecode from 'foo.o' and 11591 'bar.o', merges the two files into a single internal image, and 11592 compiles the result as usual. Since both 'foo.o' and 'bar.o' are 11593 merged into a single image, this causes all the interprocedural 11594 analyses and optimizations in GCC to work across the two files as 11595 if they were a single one. This means, for example, that the 11596 inliner is able to inline functions in 'bar.o' into functions in 11597 'foo.o' and vice-versa. 11598 11599 Another (simpler) way to enable link-time optimization is: 11600 11601 gcc -o myprog -flto -O2 foo.c bar.c 11602 11603 The above generates bytecode for 'foo.c' and 'bar.c', merges them 11604 together into a single GIMPLE representation and optimizes them as 11605 usual to produce 'myprog'. 11606 11607 The important thing to keep in mind is that to enable link-time 11608 optimizations you need to use the GCC driver to perform the link 11609 step. GCC automatically performs link-time optimization if any of 11610 the objects involved were compiled with the '-flto' command-line 11611 option. You can always override the automatic decision to do 11612 link-time optimization by passing '-fno-lto' to the link command. 11613 11614 To make whole program optimization effective, it is necessary to 11615 make certain whole program assumptions. The compiler needs to know 11616 what functions and variables can be accessed by libraries and 11617 runtime outside of the link-time optimized unit. When supported by 11618 the linker, the linker plugin (see '-fuse-linker-plugin') passes 11619 information to the compiler about used and externally visible 11620 symbols. When the linker plugin is not available, 11621 '-fwhole-program' should be used to allow the compiler to make 11622 these assumptions, which leads to more aggressive optimization 11623 decisions. 11624 11625 When a file is compiled with '-flto' without '-fuse-linker-plugin', 11626 the generated object file is larger than a regular object file 11627 because it contains GIMPLE bytecodes and the usual final code (see 11628 '-ffat-lto-objects'). This means that object files with LTO 11629 information can be linked as normal object files; if '-fno-lto' is 11630 passed to the linker, no interprocedural optimizations are applied. 11631 Note that when '-fno-fat-lto-objects' is enabled the compile stage 11632 is faster but you cannot perform a regular, non-LTO link on them. 11633 11634 When producing the final binary, GCC only applies link-time 11635 optimizations to those files that contain bytecode. Therefore, you 11636 can mix and match object files and libraries with GIMPLE bytecodes 11637 and final object code. GCC automatically selects which files to 11638 optimize in LTO mode and which files to link without further 11639 processing. 11640 11641 Generally, options specified at link time override those specified 11642 at compile time, although in some cases GCC attempts to infer 11643 link-time options from the settings used to compile the input 11644 files. 11645 11646 If you do not specify an optimization level option '-O' at link 11647 time, then GCC uses the highest optimization level used when 11648 compiling the object files. Note that it is generally ineffective 11649 to specify an optimization level option only at link time and not 11650 at compile time, for two reasons. First, compiling without 11651 optimization suppresses compiler passes that gather information 11652 needed for effective optimization at link time. Second, some early 11653 optimization passes can be performed only at compile time and not 11654 at link time. 11655 11656 There are some code generation flags preserved by GCC when 11657 generating bytecodes, as they need to be used during the final 11658 link. Currently, the following options and their settings are 11659 taken from the first object file that explicitly specifies them: 11660 '-fcommon', '-fexceptions', '-fnon-call-exceptions', '-fgnu-tm' and 11661 all the '-m' target flags. 11662 11663 The following options '-fPIC', '-fpic', '-fpie' and '-fPIE' are 11664 combined based on the following scheme: 11665 11666 -fPIC + -fpic = -fpic 11667 -fPIC + -fno-pic = -fno-pic 11668 -fpic/-fPIC + (no option) = (no option) 11669 -fPIC + -fPIE = -fPIE 11670 -fpic + -fPIE = -fpie 11671 -fPIC/-fpic + -fpie = -fpie 11672 11673 Certain ABI-changing flags are required to match in all compilation 11674 units, and trying to override this at link time with a conflicting 11675 value is ignored. This includes options such as 11676 '-freg-struct-return' and '-fpcc-struct-return'. 11677 11678 Other options such as '-ffp-contract', '-fno-strict-overflow', 11679 '-fwrapv', '-fno-trapv' or '-fno-strict-aliasing' are passed 11680 through to the link stage and merged conservatively for conflicting 11681 translation units. Specifically '-fno-strict-overflow', '-fwrapv' 11682 and '-fno-trapv' take precedence; and for example 11683 '-ffp-contract=off' takes precedence over '-ffp-contract=fast'. 11684 You can override them at link time. 11685 11686 Diagnostic options such as '-Wstringop-overflow' are passed through 11687 to the link stage and their setting matches that of the 11688 compile-step at function granularity. Note that this matters only 11689 for diagnostics emitted during optimization. Note that code 11690 transforms such as inlining can lead to warnings being enabled or 11691 disabled for regions if code not consistent with the setting at 11692 compile time. 11693 11694 When you need to pass options to the assembler via '-Wa' or 11695 '-Xassembler' make sure to either compile such translation units 11696 with '-fno-lto' or consistently use the same assembler options on 11697 all translation units. You can alternatively also specify 11698 assembler options at LTO link time. 11699 11700 To enable debug info generation you need to supply '-g' at compile 11701 time. If any of the input files at link time were built with debug 11702 info generation enabled the link will enable debug info generation 11703 as well. Any elaborate debug info settings like the dwarf level 11704 '-gdwarf-5' need to be explicitly repeated at the linker command 11705 line and mixing different settings in different translation units 11706 is discouraged. 11707 11708 If LTO encounters objects with C linkage declared with incompatible 11709 types in separate translation units to be linked together 11710 (undefined behavior according to ISO C99 6.2.7), a non-fatal 11711 diagnostic may be issued. The behavior is still undefined at run 11712 time. Similar diagnostics may be raised for other languages. 11713 11714 Another feature of LTO is that it is possible to apply 11715 interprocedural optimizations on files written in different 11716 languages: 11717 11718 gcc -c -flto foo.c 11719 g++ -c -flto bar.cc 11720 gfortran -c -flto baz.f90 11721 g++ -o myprog -flto -O3 foo.o bar.o baz.o -lgfortran 11722 11723 Notice that the final link is done with 'g++' to get the C++ 11724 runtime libraries and '-lgfortran' is added to get the Fortran 11725 runtime libraries. In general, when mixing languages in LTO mode, 11726 you should use the same link command options as when mixing 11727 languages in a regular (non-LTO) compilation. 11728 11729 If object files containing GIMPLE bytecode are stored in a library 11730 archive, say 'libfoo.a', it is possible to extract and use them in 11731 an LTO link if you are using a linker with plugin support. To 11732 create static libraries suitable for LTO, use 'gcc-ar' and 11733 'gcc-ranlib' instead of 'ar' and 'ranlib'; to show the symbols of 11734 object files with GIMPLE bytecode, use 'gcc-nm'. Those commands 11735 require that 'ar', 'ranlib' and 'nm' have been compiled with plugin 11736 support. At link time, use the flag '-fuse-linker-plugin' to 11737 ensure that the library participates in the LTO optimization 11738 process: 11739 11740 gcc -o myprog -O2 -flto -fuse-linker-plugin a.o b.o -lfoo 11741 11742 With the linker plugin enabled, the linker extracts the needed 11743 GIMPLE files from 'libfoo.a' and passes them on to the running GCC 11744 to make them part of the aggregated GIMPLE image to be optimized. 11745 11746 If you are not using a linker with plugin support and/or do not 11747 enable the linker plugin, then the objects inside 'libfoo.a' are 11748 extracted and linked as usual, but they do not participate in the 11749 LTO optimization process. In order to make a static library 11750 suitable for both LTO optimization and usual linkage, compile its 11751 object files with '-flto' '-ffat-lto-objects'. 11752 11753 Link-time optimizations do not require the presence of the whole 11754 program to operate. If the program does not require any symbols to 11755 be exported, it is possible to combine '-flto' and 11756 '-fwhole-program' to allow the interprocedural optimizers to use 11757 more aggressive assumptions which may lead to improved optimization 11758 opportunities. Use of '-fwhole-program' is not needed when linker 11759 plugin is active (see '-fuse-linker-plugin'). 11760 11761 The current implementation of LTO makes no attempt to generate 11762 bytecode that is portable between different types of hosts. The 11763 bytecode files are versioned and there is a strict version check, 11764 so bytecode files generated in one version of GCC do not work with 11765 an older or newer version of GCC. 11766 11767 Link-time optimization does not work well with generation of 11768 debugging information on systems other than those using a 11769 combination of ELF and DWARF. 11770 11771 If you specify the optional N, the optimization and code generation 11772 done at link time is executed in parallel using N parallel jobs by 11773 utilizing an installed 'make' program. The environment variable 11774 'MAKE' may be used to override the program used. 11775 11776 You can also specify '-flto=jobserver' to use GNU make's job server 11777 mode to determine the number of parallel jobs. This is useful when 11778 the Makefile calling GCC is already executing in parallel. You 11779 must prepend a '+' to the command recipe in the parent Makefile for 11780 this to work. This option likely only works if 'MAKE' is GNU make. 11781 Even without the option value, GCC tries to automatically detect a 11782 running GNU make's job server. 11783 11784 Use '-flto=auto' to use GNU make's job server, if available, or 11785 otherwise fall back to autodetection of the number of CPU threads 11786 present in your system. 11787 11788'-flto-partition=ALG' 11789 Specify the partitioning algorithm used by the link-time optimizer. 11790 The value is either '1to1' to specify a partitioning mirroring the 11791 original source files or 'balanced' to specify partitioning into 11792 equally sized chunks (whenever possible) or 'max' to create new 11793 partition for every symbol where possible. Specifying 'none' as an 11794 algorithm disables partitioning and streaming completely. The 11795 default value is 'balanced'. While '1to1' can be used as an 11796 workaround for various code ordering issues, the 'max' partitioning 11797 is intended for internal testing only. The value 'one' specifies 11798 that exactly one partition should be used while the value 'none' 11799 bypasses partitioning and executes the link-time optimization step 11800 directly from the WPA phase. 11801 11802'-flto-compression-level=N' 11803 This option specifies the level of compression used for 11804 intermediate language written to LTO object files, and is only 11805 meaningful in conjunction with LTO mode ('-flto'). GCC currently 11806 supports two LTO compression algorithms. For zstd, valid values 11807 are 0 (no compression) to 19 (maximum compression), while zlib 11808 supports values from 0 to 9. Values outside this range are clamped 11809 to either minimum or maximum of the supported values. If the 11810 option is not given, a default balanced compression setting is 11811 used. 11812 11813'-fuse-linker-plugin' 11814 Enables the use of a linker plugin during link-time optimization. 11815 This option relies on plugin support in the linker, which is 11816 available in gold or in GNU ld 2.21 or newer. 11817 11818 This option enables the extraction of object files with GIMPLE 11819 bytecode out of library archives. This improves the quality of 11820 optimization by exposing more code to the link-time optimizer. 11821 This information specifies what symbols can be accessed externally 11822 (by non-LTO object or during dynamic linking). Resulting code 11823 quality improvements on binaries (and shared libraries that use 11824 hidden visibility) are similar to '-fwhole-program'. See '-flto' 11825 for a description of the effect of this flag and how to use it. 11826 11827 This option is enabled by default when LTO support in GCC is 11828 enabled and GCC was configured for use with a linker supporting 11829 plugins (GNU ld 2.21 or newer or gold). 11830 11831'-ffat-lto-objects' 11832 Fat LTO objects are object files that contain both the intermediate 11833 language and the object code. This makes them usable for both LTO 11834 linking and normal linking. This option is effective only when 11835 compiling with '-flto' and is ignored at link time. 11836 11837 '-fno-fat-lto-objects' improves compilation time over plain LTO, 11838 but requires the complete toolchain to be aware of LTO. It requires 11839 a linker with linker plugin support for basic functionality. 11840 Additionally, 'nm', 'ar' and 'ranlib' need to support linker 11841 plugins to allow a full-featured build environment (capable of 11842 building static libraries etc). GCC provides the 'gcc-ar', 11843 'gcc-nm', 'gcc-ranlib' wrappers to pass the right options to these 11844 tools. With non fat LTO makefiles need to be modified to use them. 11845 11846 Note that modern binutils provide plugin auto-load mechanism. 11847 Installing the linker plugin into '$libdir/bfd-plugins' has the 11848 same effect as usage of the command wrappers ('gcc-ar', 'gcc-nm' 11849 and 'gcc-ranlib'). 11850 11851 The default is '-fno-fat-lto-objects' on targets with linker plugin 11852 support. 11853 11854'-fcompare-elim' 11855 After register allocation and post-register allocation instruction 11856 splitting, identify arithmetic instructions that compute processor 11857 flags similar to a comparison operation based on that arithmetic. 11858 If possible, eliminate the explicit comparison operation. 11859 11860 This pass only applies to certain targets that cannot explicitly 11861 represent the comparison operation before register allocation is 11862 complete. 11863 11864 Enabled at levels '-O1', '-O2', '-O3', '-Os'. 11865 11866'-fcprop-registers' 11867 After register allocation and post-register allocation instruction 11868 splitting, perform a copy-propagation pass to try to reduce 11869 scheduling dependencies and occasionally eliminate the copy. 11870 11871 Enabled at levels '-O1', '-O2', '-O3', '-Os'. 11872 11873'-fprofile-correction' 11874 Profiles collected using an instrumented binary for multi-threaded 11875 programs may be inconsistent due to missed counter updates. When 11876 this option is specified, GCC uses heuristics to correct or smooth 11877 out such inconsistencies. By default, GCC emits an error message 11878 when an inconsistent profile is detected. 11879 11880 This option is enabled by '-fauto-profile'. 11881 11882'-fprofile-partial-training' 11883 With '-fprofile-use' all portions of programs not executed during 11884 train run are optimized agressively for size rather than speed. In 11885 some cases it is not practical to train all possible hot paths in 11886 the program. (For example, program may contain functions specific 11887 for a given hardware and trianing may not cover all hardware 11888 configurations program is run on.) With 11889 '-fprofile-partial-training' profile feedback will be ignored for 11890 all functions not executed during the train run leading them to be 11891 optimized as if they were compiled without profile feedback. This 11892 leads to better performance when train run is not representative 11893 but also leads to significantly bigger code. 11894 11895'-fprofile-use' 11896'-fprofile-use=PATH' 11897 Enable profile feedback-directed optimizations, and the following 11898 optimizations, many of which are generally profitable only with 11899 profile feedback available: 11900 11901 -fbranch-probabilities -fprofile-values 11902 -funroll-loops -fpeel-loops -ftracer -fvpt 11903 -finline-functions -fipa-cp -fipa-cp-clone -fipa-bit-cp 11904 -fpredictive-commoning -fsplit-loops -funswitch-loops 11905 -fgcse-after-reload -ftree-loop-vectorize -ftree-slp-vectorize 11906 -fvect-cost-model=dynamic -ftree-loop-distribute-patterns 11907 -fprofile-reorder-functions 11908 11909 Before you can use this option, you must first generate profiling 11910 information. *Note Instrumentation Options::, for information 11911 about the '-fprofile-generate' option. 11912 11913 By default, GCC emits an error message if the feedback profiles do 11914 not match the source code. This error can be turned into a warning 11915 by using '-Wno-error=coverage-mismatch'. Note this may result in 11916 poorly optimized code. Additionally, by default, GCC also emits a 11917 warning message if the feedback profiles do not exist (see 11918 '-Wmissing-profile'). 11919 11920 If PATH is specified, GCC looks at the PATH to find the profile 11921 feedback data files. See '-fprofile-dir'. 11922 11923'-fauto-profile' 11924'-fauto-profile=PATH' 11925 Enable sampling-based feedback-directed optimizations, and the 11926 following optimizations, many of which are generally profitable 11927 only with profile feedback available: 11928 11929 -fbranch-probabilities -fprofile-values 11930 -funroll-loops -fpeel-loops -ftracer -fvpt 11931 -finline-functions -fipa-cp -fipa-cp-clone -fipa-bit-cp 11932 -fpredictive-commoning -fsplit-loops -funswitch-loops 11933 -fgcse-after-reload -ftree-loop-vectorize -ftree-slp-vectorize 11934 -fvect-cost-model=dynamic -ftree-loop-distribute-patterns 11935 -fprofile-correction 11936 11937 PATH is the name of a file containing AutoFDO profile information. 11938 If omitted, it defaults to 'fbdata.afdo' in the current directory. 11939 11940 Producing an AutoFDO profile data file requires running your 11941 program with the 'perf' utility on a supported GNU/Linux target 11942 system. For more information, see <https://perf.wiki.kernel.org/>. 11943 11944 E.g. 11945 perf record -e br_inst_retired:near_taken -b -o perf.data \ 11946 -- your_program 11947 11948 Then use the 'create_gcov' tool to convert the raw profile data to 11949 a format that can be used by GCC. You must also supply the 11950 unstripped binary for your program to this tool. See 11951 <https://github.com/google/autofdo>. 11952 11953 E.g. 11954 create_gcov --binary=your_program.unstripped --profile=perf.data \ 11955 --gcov=profile.afdo 11956 11957 The following options control compiler behavior regarding 11958floating-point arithmetic. These options trade off between speed and 11959correctness. All must be specifically enabled. 11960 11961'-ffloat-store' 11962 Do not store floating-point variables in registers, and inhibit 11963 other options that might change whether a floating-point value is 11964 taken from a register or memory. 11965 11966 This option prevents undesirable excess precision on machines such 11967 as the 68000 where the floating registers (of the 68881) keep more 11968 precision than a 'double' is supposed to have. Similarly for the 11969 x86 architecture. For most programs, the excess precision does 11970 only good, but a few programs rely on the precise definition of 11971 IEEE floating point. Use '-ffloat-store' for such programs, after 11972 modifying them to store all pertinent intermediate computations 11973 into variables. 11974 11975'-fexcess-precision=STYLE' 11976 This option allows further control over excess precision on 11977 machines where floating-point operations occur in a format with 11978 more precision or range than the IEEE standard and interchange 11979 floating-point types. By default, '-fexcess-precision=fast' is in 11980 effect; this means that operations may be carried out in a wider 11981 precision than the types specified in the source if that would 11982 result in faster code, and it is unpredictable when rounding to the 11983 types specified in the source code takes place. When compiling C, 11984 if '-fexcess-precision=standard' is specified then excess precision 11985 follows the rules specified in ISO C99; in particular, both casts 11986 and assignments cause values to be rounded to their semantic types 11987 (whereas '-ffloat-store' only affects assignments). This option is 11988 enabled by default for C if a strict conformance option such as 11989 '-std=c99' is used. '-ffast-math' enables 11990 '-fexcess-precision=fast' by default regardless of whether a strict 11991 conformance option is used. 11992 11993 '-fexcess-precision=standard' is not implemented for languages 11994 other than C. On the x86, it has no effect if '-mfpmath=sse' or 11995 '-mfpmath=sse+387' is specified; in the former case, IEEE semantics 11996 apply without excess precision, and in the latter, rounding is 11997 unpredictable. 11998 11999'-ffast-math' 12000 Sets the options '-fno-math-errno', '-funsafe-math-optimizations', 12001 '-ffinite-math-only', '-fno-rounding-math', '-fno-signaling-nans', 12002 '-fcx-limited-range' and '-fexcess-precision=fast'. 12003 12004 This option causes the preprocessor macro '__FAST_MATH__' to be 12005 defined. 12006 12007 This option is not turned on by any '-O' option besides '-Ofast' 12008 since it can result in incorrect output for programs that depend on 12009 an exact implementation of IEEE or ISO rules/specifications for 12010 math functions. It may, however, yield faster code for programs 12011 that do not require the guarantees of these specifications. 12012 12013'-fno-math-errno' 12014 Do not set 'errno' after calling math functions that are executed 12015 with a single instruction, e.g., 'sqrt'. A program that relies on 12016 IEEE exceptions for math error handling may want to use this flag 12017 for speed while maintaining IEEE arithmetic compatibility. 12018 12019 This option is not turned on by any '-O' option since it can result 12020 in incorrect output for programs that depend on an exact 12021 implementation of IEEE or ISO rules/specifications for math 12022 functions. It may, however, yield faster code for programs that do 12023 not require the guarantees of these specifications. 12024 12025 The default is '-fmath-errno'. 12026 12027 On Darwin systems, the math library never sets 'errno'. There is 12028 therefore no reason for the compiler to consider the possibility 12029 that it might, and '-fno-math-errno' is the default. 12030 12031'-funsafe-math-optimizations' 12032 12033 Allow optimizations for floating-point arithmetic that (a) assume 12034 that arguments and results are valid and (b) may violate IEEE or 12035 ANSI standards. When used at link time, it may include libraries 12036 or startup files that change the default FPU control word or other 12037 similar optimizations. 12038 12039 This option is not turned on by any '-O' option since it can result 12040 in incorrect output for programs that depend on an exact 12041 implementation of IEEE or ISO rules/specifications for math 12042 functions. It may, however, yield faster code for programs that do 12043 not require the guarantees of these specifications. Enables 12044 '-fno-signed-zeros', '-fno-trapping-math', '-fassociative-math' and 12045 '-freciprocal-math'. 12046 12047 The default is '-fno-unsafe-math-optimizations'. 12048 12049'-fassociative-math' 12050 12051 Allow re-association of operands in series of floating-point 12052 operations. This violates the ISO C and C++ language standard by 12053 possibly changing computation result. NOTE: re-ordering may change 12054 the sign of zero as well as ignore NaNs and inhibit or create 12055 underflow or overflow (and thus cannot be used on code that relies 12056 on rounding behavior like '(x + 2**52) - 2**52'. May also reorder 12057 floating-point comparisons and thus may not be used when ordered 12058 comparisons are required. This option requires that both 12059 '-fno-signed-zeros' and '-fno-trapping-math' be in effect. 12060 Moreover, it doesn't make much sense with '-frounding-math'. For 12061 Fortran the option is automatically enabled when both 12062 '-fno-signed-zeros' and '-fno-trapping-math' are in effect. 12063 12064 The default is '-fno-associative-math'. 12065 12066'-freciprocal-math' 12067 12068 Allow the reciprocal of a value to be used instead of dividing by 12069 the value if this enables optimizations. For example 'x / y' can 12070 be replaced with 'x * (1/y)', which is useful if '(1/y)' is subject 12071 to common subexpression elimination. Note that this loses 12072 precision and increases the number of flops operating on the value. 12073 12074 The default is '-fno-reciprocal-math'. 12075 12076'-ffinite-math-only' 12077 Allow optimizations for floating-point arithmetic that assume that 12078 arguments and results are not NaNs or +-Infs. 12079 12080 This option is not turned on by any '-O' option since it can result 12081 in incorrect output for programs that depend on an exact 12082 implementation of IEEE or ISO rules/specifications for math 12083 functions. It may, however, yield faster code for programs that do 12084 not require the guarantees of these specifications. 12085 12086 The default is '-fno-finite-math-only'. 12087 12088'-fno-signed-zeros' 12089 Allow optimizations for floating-point arithmetic that ignore the 12090 signedness of zero. IEEE arithmetic specifies the behavior of 12091 distinct +0.0 and -0.0 values, which then prohibits simplification 12092 of expressions such as x+0.0 or 0.0*x (even with 12093 '-ffinite-math-only'). This option implies that the sign of a zero 12094 result isn't significant. 12095 12096 The default is '-fsigned-zeros'. 12097 12098'-fno-trapping-math' 12099 Compile code assuming that floating-point operations cannot 12100 generate user-visible traps. These traps include division by zero, 12101 overflow, underflow, inexact result and invalid operation. This 12102 option requires that '-fno-signaling-nans' be in effect. Setting 12103 this option may allow faster code if one relies on "non-stop" IEEE 12104 arithmetic, for example. 12105 12106 This option should never be turned on by any '-O' option since it 12107 can result in incorrect output for programs that depend on an exact 12108 implementation of IEEE or ISO rules/specifications for math 12109 functions. 12110 12111 The default is '-ftrapping-math'. 12112 12113'-frounding-math' 12114 Disable transformations and optimizations that assume default 12115 floating-point rounding behavior. This is round-to-zero for all 12116 floating point to integer conversions, and round-to-nearest for all 12117 other arithmetic truncations. This option should be specified for 12118 programs that change the FP rounding mode dynamically, or that may 12119 be executed with a non-default rounding mode. This option disables 12120 constant folding of floating-point expressions at compile time 12121 (which may be affected by rounding mode) and arithmetic 12122 transformations that are unsafe in the presence of sign-dependent 12123 rounding modes. 12124 12125 The default is '-fno-rounding-math'. 12126 12127 This option is experimental and does not currently guarantee to 12128 disable all GCC optimizations that are affected by rounding mode. 12129 Future versions of GCC may provide finer control of this setting 12130 using C99's 'FENV_ACCESS' pragma. This command-line option will be 12131 used to specify the default state for 'FENV_ACCESS'. 12132 12133'-fsignaling-nans' 12134 Compile code assuming that IEEE signaling NaNs may generate 12135 user-visible traps during floating-point operations. Setting this 12136 option disables optimizations that may change the number of 12137 exceptions visible with signaling NaNs. This option implies 12138 '-ftrapping-math'. 12139 12140 This option causes the preprocessor macro '__SUPPORT_SNAN__' to be 12141 defined. 12142 12143 The default is '-fno-signaling-nans'. 12144 12145 This option is experimental and does not currently guarantee to 12146 disable all GCC optimizations that affect signaling NaN behavior. 12147 12148'-fno-fp-int-builtin-inexact' 12149 Do not allow the built-in functions 'ceil', 'floor', 'round' and 12150 'trunc', and their 'float' and 'long double' variants, to generate 12151 code that raises the "inexact" floating-point exception for 12152 noninteger arguments. ISO C99 and C11 allow these functions to 12153 raise the "inexact" exception, but ISO/IEC TS 18661-1:2014, the C 12154 bindings to IEEE 754-2008, as integrated into ISO C2X, does not 12155 allow these functions to do so. 12156 12157 The default is '-ffp-int-builtin-inexact', allowing the exception 12158 to be raised, unless C2X or a later C standard is selected. This 12159 option does nothing unless '-ftrapping-math' is in effect. 12160 12161 Even if '-fno-fp-int-builtin-inexact' is used, if the functions 12162 generate a call to a library function then the "inexact" exception 12163 may be raised if the library implementation does not follow TS 12164 18661. 12165 12166'-fsingle-precision-constant' 12167 Treat floating-point constants as single precision instead of 12168 implicitly converting them to double-precision constants. 12169 12170'-fcx-limited-range' 12171 When enabled, this option states that a range reduction step is not 12172 needed when performing complex division. Also, there is no 12173 checking whether the result of a complex multiplication or division 12174 is 'NaN + I*NaN', with an attempt to rescue the situation in that 12175 case. The default is '-fno-cx-limited-range', but is enabled by 12176 '-ffast-math'. 12177 12178 This option controls the default setting of the ISO C99 12179 'CX_LIMITED_RANGE' pragma. Nevertheless, the option applies to all 12180 languages. 12181 12182'-fcx-fortran-rules' 12183 Complex multiplication and division follow Fortran rules. Range 12184 reduction is done as part of complex division, but there is no 12185 checking whether the result of a complex multiplication or division 12186 is 'NaN + I*NaN', with an attempt to rescue the situation in that 12187 case. 12188 12189 The default is '-fno-cx-fortran-rules'. 12190 12191 The following options control optimizations that may improve 12192performance, but are not enabled by any '-O' options. This section 12193includes experimental options that may produce broken code. 12194 12195'-fbranch-probabilities' 12196 After running a program compiled with '-fprofile-arcs' (*note 12197 Instrumentation Options::), you can compile it a second time using 12198 '-fbranch-probabilities', to improve optimizations based on the 12199 number of times each branch was taken. When a program compiled 12200 with '-fprofile-arcs' exits, it saves arc execution counts to a 12201 file called 'SOURCENAME.gcda' for each source file. The 12202 information in this data file is very dependent on the structure of 12203 the generated code, so you must use the same source code and the 12204 same optimization options for both compilations. See details about 12205 the file naming in '-fprofile-arcs'. 12206 12207 With '-fbranch-probabilities', GCC puts a 'REG_BR_PROB' note on 12208 each 'JUMP_INSN' and 'CALL_INSN'. These can be used to improve 12209 optimization. Currently, they are only used in one place: in 12210 'reorg.cc', instead of guessing which path a branch is most likely 12211 to take, the 'REG_BR_PROB' values are used to exactly determine 12212 which path is taken more often. 12213 12214 Enabled by '-fprofile-use' and '-fauto-profile'. 12215 12216'-fprofile-values' 12217 If combined with '-fprofile-arcs', it adds code so that some data 12218 about values of expressions in the program is gathered. 12219 12220 With '-fbranch-probabilities', it reads back the data gathered from 12221 profiling values of expressions for usage in optimizations. 12222 12223 Enabled by '-fprofile-generate', '-fprofile-use', and 12224 '-fauto-profile'. 12225 12226'-fprofile-reorder-functions' 12227 Function reordering based on profile instrumentation collects first 12228 time of execution of a function and orders these functions in 12229 ascending order. 12230 12231 Enabled with '-fprofile-use'. 12232 12233'-fvpt' 12234 If combined with '-fprofile-arcs', this option instructs the 12235 compiler to add code to gather information about values of 12236 expressions. 12237 12238 With '-fbranch-probabilities', it reads back the data gathered and 12239 actually performs the optimizations based on them. Currently the 12240 optimizations include specialization of division operations using 12241 the knowledge about the value of the denominator. 12242 12243 Enabled with '-fprofile-use' and '-fauto-profile'. 12244 12245'-frename-registers' 12246 Attempt to avoid false dependencies in scheduled code by making use 12247 of registers left over after register allocation. This 12248 optimization most benefits processors with lots of registers. 12249 Depending on the debug information format adopted by the target, 12250 however, it can make debugging impossible, since variables no 12251 longer stay in a "home register". 12252 12253 Enabled by default with '-funroll-loops'. 12254 12255'-fschedule-fusion' 12256 Performs a target dependent pass over the instruction stream to 12257 schedule instructions of same type together because target machine 12258 can execute them more efficiently if they are adjacent to each 12259 other in the instruction flow. 12260 12261 Enabled at levels '-O2', '-O3', '-Os'. 12262 12263'-ftracer' 12264 Perform tail duplication to enlarge superblock size. This 12265 transformation simplifies the control flow of the function allowing 12266 other optimizations to do a better job. 12267 12268 Enabled by '-fprofile-use' and '-fauto-profile'. 12269 12270'-funroll-loops' 12271 Unroll loops whose number of iterations can be determined at 12272 compile time or upon entry to the loop. '-funroll-loops' implies 12273 '-frerun-cse-after-loop', '-fweb' and '-frename-registers'. It 12274 also turns on complete loop peeling (i.e. complete removal of loops 12275 with a small constant number of iterations). This option makes 12276 code larger, and may or may not make it run faster. 12277 12278 Enabled by '-fprofile-use' and '-fauto-profile'. 12279 12280'-funroll-all-loops' 12281 Unroll all loops, even if their number of iterations is uncertain 12282 when the loop is entered. This usually makes programs run more 12283 slowly. '-funroll-all-loops' implies the same options as 12284 '-funroll-loops'. 12285 12286'-fpeel-loops' 12287 Peels loops for which there is enough information that they do not 12288 roll much (from profile feedback or static analysis). It also 12289 turns on complete loop peeling (i.e. complete removal of loops with 12290 small constant number of iterations). 12291 12292 Enabled by '-O3', '-fprofile-use', and '-fauto-profile'. 12293 12294'-fmove-loop-invariants' 12295 Enables the loop invariant motion pass in the RTL loop optimizer. 12296 Enabled at level '-O1' and higher, except for '-Og'. 12297 12298'-fmove-loop-stores' 12299 Enables the loop store motion pass in the GIMPLE loop optimizer. 12300 This moves invariant stores to after the end of the loop in 12301 exchange for carrying the stored value in a register across the 12302 iteration. Note for this option to have an effect '-ftree-loop-im' 12303 has to be enabled as well. Enabled at level '-O1' and higher, 12304 except for '-Og'. 12305 12306'-fsplit-loops' 12307 Split a loop into two if it contains a condition that's always true 12308 for one side of the iteration space and false for the other. 12309 12310 Enabled by '-fprofile-use' and '-fauto-profile'. 12311 12312'-funswitch-loops' 12313 Move branches with loop invariant conditions out of the loop, with 12314 duplicates of the loop on both branches (modified according to 12315 result of the condition). 12316 12317 Enabled by '-fprofile-use' and '-fauto-profile'. 12318 12319'-fversion-loops-for-strides' 12320 If a loop iterates over an array with a variable stride, create 12321 another version of the loop that assumes the stride is always one. 12322 For example: 12323 12324 for (int i = 0; i < n; ++i) 12325 x[i * stride] = ...; 12326 12327 becomes: 12328 12329 if (stride == 1) 12330 for (int i = 0; i < n; ++i) 12331 x[i] = ...; 12332 else 12333 for (int i = 0; i < n; ++i) 12334 x[i * stride] = ...; 12335 12336 This is particularly useful for assumed-shape arrays in Fortran 12337 where (for example) it allows better vectorization assuming 12338 contiguous accesses. This flag is enabled by default at '-O3'. It 12339 is also enabled by '-fprofile-use' and '-fauto-profile'. 12340 12341'-ffunction-sections' 12342'-fdata-sections' 12343 Place each function or data item into its own section in the output 12344 file if the target supports arbitrary sections. The name of the 12345 function or the name of the data item determines the section's name 12346 in the output file. 12347 12348 Use these options on systems where the linker can perform 12349 optimizations to improve locality of reference in the instruction 12350 space. Most systems using the ELF object format have linkers with 12351 such optimizations. On AIX, the linker rearranges sections 12352 (CSECTs) based on the call graph. The performance impact varies. 12353 12354 Together with a linker garbage collection (linker '--gc-sections' 12355 option) these options may lead to smaller statically-linked 12356 executables (after stripping). 12357 12358 On ELF/DWARF systems these options do not degenerate the quality of 12359 the debug information. There could be issues with other object 12360 files/debug info formats. 12361 12362 Only use these options when there are significant benefits from 12363 doing so. When you specify these options, the assembler and linker 12364 create larger object and executable files and are also slower. 12365 These options affect code generation. They prevent optimizations 12366 by the compiler and assembler using relative locations inside a 12367 translation unit since the locations are unknown until link time. 12368 An example of such an optimization is relaxing calls to short call 12369 instructions. 12370 12371'-fstdarg-opt' 12372 Optimize the prologue of variadic argument functions with respect 12373 to usage of those arguments. 12374 12375'-fsection-anchors' 12376 Try to reduce the number of symbolic address calculations by using 12377 shared "anchor" symbols to address nearby objects. This 12378 transformation can help to reduce the number of GOT entries and GOT 12379 accesses on some targets. 12380 12381 For example, the implementation of the following function 'foo': 12382 12383 static int a, b, c; 12384 int foo (void) { return a + b + c; } 12385 12386 usually calculates the addresses of all three variables, but if you 12387 compile it with '-fsection-anchors', it accesses the variables from 12388 a common anchor point instead. The effect is similar to the 12389 following pseudocode (which isn't valid C): 12390 12391 int foo (void) 12392 { 12393 register int *xr = &x; 12394 return xr[&a - &x] + xr[&b - &x] + xr[&c - &x]; 12395 } 12396 12397 Not all targets support this option. 12398 12399'-fzero-call-used-regs=CHOICE' 12400 Zero call-used registers at function return to increase program 12401 security by either mitigating Return-Oriented Programming (ROP) 12402 attacks or preventing information leakage through registers. 12403 12404 The possible values of CHOICE are the same as for the 12405 'zero_call_used_regs' attribute (*note Function Attributes::). The 12406 default is 'skip'. 12407 12408 You can control this behavior for a specific function by using the 12409 function attribute 'zero_call_used_regs' (*note Function 12410 Attributes::). 12411 12412'--param NAME=VALUE' 12413 In some places, GCC uses various constants to control the amount of 12414 optimization that is done. For example, GCC does not inline 12415 functions that contain more than a certain number of instructions. 12416 You can control some of these constants on the command line using 12417 the '--param' option. 12418 12419 The names of specific parameters, and the meaning of the values, 12420 are tied to the internals of the compiler, and are subject to 12421 change without notice in future releases. 12422 12423 In order to get minimal, maximal and default value of a parameter, 12424 one can use '--help=param -Q' options. 12425 12426 In each case, the VALUE is an integer. The following choices of 12427 NAME are recognized for all targets: 12428 12429 'predictable-branch-outcome' 12430 When branch is predicted to be taken with probability lower 12431 than this threshold (in percent), then it is considered well 12432 predictable. 12433 12434 'max-rtl-if-conversion-insns' 12435 RTL if-conversion tries to remove conditional branches around 12436 a block and replace them with conditionally executed 12437 instructions. This parameter gives the maximum number of 12438 instructions in a block which should be considered for 12439 if-conversion. The compiler will also use other heuristics to 12440 decide whether if-conversion is likely to be profitable. 12441 12442 'max-rtl-if-conversion-predictable-cost' 12443 RTL if-conversion will try to remove conditional branches 12444 around a block and replace them with conditionally executed 12445 instructions. These parameters give the maximum permissible 12446 cost for the sequence that would be generated by if-conversion 12447 depending on whether the branch is statically determined to be 12448 predictable or not. The units for this parameter are the same 12449 as those for the GCC internal seq_cost metric. The compiler 12450 will try to provide a reasonable default for this parameter 12451 using the BRANCH_COST target macro. 12452 12453 'max-crossjump-edges' 12454 The maximum number of incoming edges to consider for 12455 cross-jumping. The algorithm used by '-fcrossjumping' is 12456 O(N^2) in the number of edges incoming to each block. 12457 Increasing values mean more aggressive optimization, making 12458 the compilation time increase with probably small improvement 12459 in executable size. 12460 12461 'min-crossjump-insns' 12462 The minimum number of instructions that must be matched at the 12463 end of two blocks before cross-jumping is performed on them. 12464 This value is ignored in the case where all instructions in 12465 the block being cross-jumped from are matched. 12466 12467 'max-grow-copy-bb-insns' 12468 The maximum code size expansion factor when copying basic 12469 blocks instead of jumping. The expansion is relative to a 12470 jump instruction. 12471 12472 'max-goto-duplication-insns' 12473 The maximum number of instructions to duplicate to a block 12474 that jumps to a computed goto. To avoid O(N^2) behavior in a 12475 number of passes, GCC factors computed gotos early in the 12476 compilation process, and unfactors them as late as possible. 12477 Only computed jumps at the end of a basic blocks with no more 12478 than max-goto-duplication-insns are unfactored. 12479 12480 'max-delay-slot-insn-search' 12481 The maximum number of instructions to consider when looking 12482 for an instruction to fill a delay slot. If more than this 12483 arbitrary number of instructions are searched, the time 12484 savings from filling the delay slot are minimal, so stop 12485 searching. Increasing values mean more aggressive 12486 optimization, making the compilation time increase with 12487 probably small improvement in execution time. 12488 12489 'max-delay-slot-live-search' 12490 When trying to fill delay slots, the maximum number of 12491 instructions to consider when searching for a block with valid 12492 live register information. Increasing this arbitrarily chosen 12493 value means more aggressive optimization, increasing the 12494 compilation time. This parameter should be removed when the 12495 delay slot code is rewritten to maintain the control-flow 12496 graph. 12497 12498 'max-gcse-memory' 12499 The approximate maximum amount of memory in 'kB' that can be 12500 allocated in order to perform the global common subexpression 12501 elimination optimization. If more memory than specified is 12502 required, the optimization is not done. 12503 12504 'max-gcse-insertion-ratio' 12505 If the ratio of expression insertions to deletions is larger 12506 than this value for any expression, then RTL PRE inserts or 12507 removes the expression and thus leaves partially redundant 12508 computations in the instruction stream. 12509 12510 'max-pending-list-length' 12511 The maximum number of pending dependencies scheduling allows 12512 before flushing the current state and starting over. Large 12513 functions with few branches or calls can create excessively 12514 large lists which needlessly consume memory and resources. 12515 12516 'max-modulo-backtrack-attempts' 12517 The maximum number of backtrack attempts the scheduler should 12518 make when modulo scheduling a loop. Larger values can 12519 exponentially increase compilation time. 12520 12521 'max-inline-functions-called-once-loop-depth' 12522 Maximal loop depth of a call considered by inline heuristics 12523 that tries to inline all functions called once. 12524 12525 'max-inline-functions-called-once-insns' 12526 Maximal estimated size of functions produced while inlining 12527 functions called once. 12528 12529 'max-inline-insns-single' 12530 Several parameters control the tree inliner used in GCC. This 12531 number sets the maximum number of instructions (counted in 12532 GCC's internal representation) in a single function that the 12533 tree inliner considers for inlining. This only affects 12534 functions declared inline and methods implemented in a class 12535 declaration (C++). 12536 12537 'max-inline-insns-auto' 12538 When you use '-finline-functions' (included in '-O3'), a lot 12539 of functions that would otherwise not be considered for 12540 inlining by the compiler are investigated. To those 12541 functions, a different (more restrictive) limit compared to 12542 functions declared inline can be applied ('--param 12543 max-inline-insns-auto'). 12544 12545 'max-inline-insns-small' 12546 This is bound applied to calls which are considered relevant 12547 with '-finline-small-functions'. 12548 12549 'max-inline-insns-size' 12550 This is bound applied to calls which are optimized for size. 12551 Small growth may be desirable to anticipate optimization 12552 oppurtunities exposed by inlining. 12553 12554 'uninlined-function-insns' 12555 Number of instructions accounted by inliner for function 12556 overhead such as function prologue and epilogue. 12557 12558 'uninlined-function-time' 12559 Extra time accounted by inliner for function overhead such as 12560 time needed to execute function prologue and epilogue. 12561 12562 'inline-heuristics-hint-percent' 12563 The scale (in percents) applied to 'inline-insns-single', 12564 'inline-insns-single-O2', 'inline-insns-auto' when inline 12565 heuristics hints that inlining is very profitable (will enable 12566 later optimizations). 12567 12568 'uninlined-thunk-insns' 12569 'uninlined-thunk-time' 12570 Same as '--param uninlined-function-insns' and '--param 12571 uninlined-function-time' but applied to function thunks. 12572 12573 'inline-min-speedup' 12574 When estimated performance improvement of caller + callee 12575 runtime exceeds this threshold (in percent), the function can 12576 be inlined regardless of the limit on '--param 12577 max-inline-insns-single' and '--param max-inline-insns-auto'. 12578 12579 'large-function-insns' 12580 The limit specifying really large functions. For functions 12581 larger than this limit after inlining, inlining is constrained 12582 by '--param large-function-growth'. This parameter is useful 12583 primarily to avoid extreme compilation time caused by 12584 non-linear algorithms used by the back end. 12585 12586 'large-function-growth' 12587 Specifies maximal growth of large function caused by inlining 12588 in percents. For example, parameter value 100 limits large 12589 function growth to 2.0 times the original size. 12590 12591 'large-unit-insns' 12592 The limit specifying large translation unit. Growth caused by 12593 inlining of units larger than this limit is limited by 12594 '--param inline-unit-growth'. For small units this might be 12595 too tight. For example, consider a unit consisting of 12596 function A that is inline and B that just calls A three times. 12597 If B is small relative to A, the growth of unit is 300\% and 12598 yet such inlining is very sane. For very large units 12599 consisting of small inlineable functions, however, the overall 12600 unit growth limit is needed to avoid exponential explosion of 12601 code size. Thus for smaller units, the size is increased to 12602 '--param large-unit-insns' before applying '--param 12603 inline-unit-growth'. 12604 12605 'lazy-modules' 12606 Maximum number of concurrently open C++ module files when lazy 12607 loading. 12608 12609 'inline-unit-growth' 12610 Specifies maximal overall growth of the compilation unit 12611 caused by inlining. For example, parameter value 20 limits 12612 unit growth to 1.2 times the original size. Cold functions 12613 (either marked cold via an attribute or by profile feedback) 12614 are not accounted into the unit size. 12615 12616 'ipa-cp-unit-growth' 12617 Specifies maximal overall growth of the compilation unit 12618 caused by interprocedural constant propagation. For example, 12619 parameter value 10 limits unit growth to 1.1 times the 12620 original size. 12621 12622 'ipa-cp-large-unit-insns' 12623 The size of translation unit that IPA-CP pass considers large. 12624 12625 'large-stack-frame' 12626 The limit specifying large stack frames. While inlining the 12627 algorithm is trying to not grow past this limit too much. 12628 12629 'large-stack-frame-growth' 12630 Specifies maximal growth of large stack frames caused by 12631 inlining in percents. For example, parameter value 1000 12632 limits large stack frame growth to 11 times the original size. 12633 12634 'max-inline-insns-recursive' 12635 'max-inline-insns-recursive-auto' 12636 Specifies the maximum number of instructions an out-of-line 12637 copy of a self-recursive inline function can grow into by 12638 performing recursive inlining. 12639 12640 '--param max-inline-insns-recursive' applies to functions 12641 declared inline. For functions not declared inline, recursive 12642 inlining happens only when '-finline-functions' (included in 12643 '-O3') is enabled; '--param max-inline-insns-recursive-auto' 12644 applies instead. 12645 12646 'max-inline-recursive-depth' 12647 'max-inline-recursive-depth-auto' 12648 Specifies the maximum recursion depth used for recursive 12649 inlining. 12650 12651 '--param max-inline-recursive-depth' applies to functions 12652 declared inline. For functions not declared inline, recursive 12653 inlining happens only when '-finline-functions' (included in 12654 '-O3') is enabled; '--param max-inline-recursive-depth-auto' 12655 applies instead. 12656 12657 'min-inline-recursive-probability' 12658 Recursive inlining is profitable only for function having deep 12659 recursion in average and can hurt for function having little 12660 recursion depth by increasing the prologue size or complexity 12661 of function body to other optimizers. 12662 12663 When profile feedback is available (see '-fprofile-generate') 12664 the actual recursion depth can be guessed from the probability 12665 that function recurses via a given call expression. This 12666 parameter limits inlining only to call expressions whose 12667 probability exceeds the given threshold (in percents). 12668 12669 'early-inlining-insns' 12670 Specify growth that the early inliner can make. In effect it 12671 increases the amount of inlining for code having a large 12672 abstraction penalty. 12673 12674 'max-early-inliner-iterations' 12675 Limit of iterations of the early inliner. This basically 12676 bounds the number of nested indirect calls the early inliner 12677 can resolve. Deeper chains are still handled by late 12678 inlining. 12679 12680 'comdat-sharing-probability' 12681 Probability (in percent) that C++ inline function with comdat 12682 visibility are shared across multiple compilation units. 12683 12684 'modref-max-bases' 12685 'modref-max-refs' 12686 'modref-max-accesses' 12687 Specifies the maximal number of base pointers, references and 12688 accesses stored for a single function by mod/ref analysis. 12689 12690 'modref-max-tests' 12691 Specifies the maxmal number of tests alias oracle can perform 12692 to disambiguate memory locations using the mod/ref 12693 information. This parameter ought to be bigger than '--param 12694 modref-max-bases' and '--param modref-max-refs'. 12695 12696 'modref-max-depth' 12697 Specifies the maximum depth of DFS walk used by modref escape 12698 analysis. Setting to 0 disables the analysis completely. 12699 12700 'modref-max-escape-points' 12701 Specifies the maximum number of escape points tracked by 12702 modref per SSA-name. 12703 12704 'modref-max-adjustments' 12705 Specifies the maximum number the access range is enlarged 12706 during modref dataflow analysis. 12707 12708 'profile-func-internal-id' 12709 A parameter to control whether to use function internal id in 12710 profile database lookup. If the value is 0, the compiler uses 12711 an id that is based on function assembler name and filename, 12712 which makes old profile data more tolerant to source changes 12713 such as function reordering etc. 12714 12715 'min-vect-loop-bound' 12716 The minimum number of iterations under which loops are not 12717 vectorized when '-ftree-vectorize' is used. The number of 12718 iterations after vectorization needs to be greater than the 12719 value specified by this option to allow vectorization. 12720 12721 'gcse-cost-distance-ratio' 12722 Scaling factor in calculation of maximum distance an 12723 expression can be moved by GCSE optimizations. This is 12724 currently supported only in the code hoisting pass. The 12725 bigger the ratio, the more aggressive code hoisting is with 12726 simple expressions, i.e., the expressions that have cost less 12727 than 'gcse-unrestricted-cost'. Specifying 0 disables hoisting 12728 of simple expressions. 12729 12730 'gcse-unrestricted-cost' 12731 Cost, roughly measured as the cost of a single typical machine 12732 instruction, at which GCSE optimizations do not constrain the 12733 distance an expression can travel. This is currently 12734 supported only in the code hoisting pass. The lesser the 12735 cost, the more aggressive code hoisting is. Specifying 0 12736 allows all expressions to travel unrestricted distances. 12737 12738 'max-hoist-depth' 12739 The depth of search in the dominator tree for expressions to 12740 hoist. This is used to avoid quadratic behavior in hoisting 12741 algorithm. The value of 0 does not limit on the search, but 12742 may slow down compilation of huge functions. 12743 12744 'max-tail-merge-comparisons' 12745 The maximum amount of similar bbs to compare a bb with. This 12746 is used to avoid quadratic behavior in tree tail merging. 12747 12748 'max-tail-merge-iterations' 12749 The maximum amount of iterations of the pass over the 12750 function. This is used to limit compilation time in tree tail 12751 merging. 12752 12753 'store-merging-allow-unaligned' 12754 Allow the store merging pass to introduce unaligned stores if 12755 it is legal to do so. 12756 12757 'max-stores-to-merge' 12758 The maximum number of stores to attempt to merge into wider 12759 stores in the store merging pass. 12760 12761 'max-store-chains-to-track' 12762 The maximum number of store chains to track at the same time 12763 in the attempt to merge them into wider stores in the store 12764 merging pass. 12765 12766 'max-stores-to-track' 12767 The maximum number of stores to track at the same time in the 12768 attemt to to merge them into wider stores in the store merging 12769 pass. 12770 12771 'max-unrolled-insns' 12772 The maximum number of instructions that a loop may have to be 12773 unrolled. If a loop is unrolled, this parameter also 12774 determines how many times the loop code is unrolled. 12775 12776 'max-average-unrolled-insns' 12777 The maximum number of instructions biased by probabilities of 12778 their execution that a loop may have to be unrolled. If a 12779 loop is unrolled, this parameter also determines how many 12780 times the loop code is unrolled. 12781 12782 'max-unroll-times' 12783 The maximum number of unrollings of a single loop. 12784 12785 'max-peeled-insns' 12786 The maximum number of instructions that a loop may have to be 12787 peeled. If a loop is peeled, this parameter also determines 12788 how many times the loop code is peeled. 12789 12790 'max-peel-times' 12791 The maximum number of peelings of a single loop. 12792 12793 'max-peel-branches' 12794 The maximum number of branches on the hot path through the 12795 peeled sequence. 12796 12797 'max-completely-peeled-insns' 12798 The maximum number of insns of a completely peeled loop. 12799 12800 'max-completely-peel-times' 12801 The maximum number of iterations of a loop to be suitable for 12802 complete peeling. 12803 12804 'max-completely-peel-loop-nest-depth' 12805 The maximum depth of a loop nest suitable for complete 12806 peeling. 12807 12808 'max-unswitch-insns' 12809 The maximum number of insns of an unswitched loop. 12810 12811 'max-unswitch-level' 12812 The maximum number of branches unswitched in a single loop. 12813 12814 'lim-expensive' 12815 The minimum cost of an expensive expression in the loop 12816 invariant motion. 12817 12818 'min-loop-cond-split-prob' 12819 When FDO profile information is available, 12820 'min-loop-cond-split-prob' specifies minimum threshold for 12821 probability of semi-invariant condition statement to trigger 12822 loop split. 12823 12824 'iv-consider-all-candidates-bound' 12825 Bound on number of candidates for induction variables, below 12826 which all candidates are considered for each use in induction 12827 variable optimizations. If there are more candidates than 12828 this, only the most relevant ones are considered to avoid 12829 quadratic time complexity. 12830 12831 'iv-max-considered-uses' 12832 The induction variable optimizations give up on loops that 12833 contain more induction variable uses. 12834 12835 'iv-always-prune-cand-set-bound' 12836 If the number of candidates in the set is smaller than this 12837 value, always try to remove unnecessary ivs from the set when 12838 adding a new one. 12839 12840 'avg-loop-niter' 12841 Average number of iterations of a loop. 12842 12843 'dse-max-object-size' 12844 Maximum size (in bytes) of objects tracked bytewise by dead 12845 store elimination. Larger values may result in larger 12846 compilation times. 12847 12848 'dse-max-alias-queries-per-store' 12849 Maximum number of queries into the alias oracle per store. 12850 Larger values result in larger compilation times and may 12851 result in more removed dead stores. 12852 12853 'scev-max-expr-size' 12854 Bound on size of expressions used in the scalar evolutions 12855 analyzer. Large expressions slow the analyzer. 12856 12857 'scev-max-expr-complexity' 12858 Bound on the complexity of the expressions in the scalar 12859 evolutions analyzer. Complex expressions slow the analyzer. 12860 12861 'max-tree-if-conversion-phi-args' 12862 Maximum number of arguments in a PHI supported by TREE if 12863 conversion unless the loop is marked with simd pragma. 12864 12865 'vect-max-version-for-alignment-checks' 12866 The maximum number of run-time checks that can be performed 12867 when doing loop versioning for alignment in the vectorizer. 12868 12869 'vect-max-version-for-alias-checks' 12870 The maximum number of run-time checks that can be performed 12871 when doing loop versioning for alias in the vectorizer. 12872 12873 'vect-max-peeling-for-alignment' 12874 The maximum number of loop peels to enhance access alignment 12875 for vectorizer. Value -1 means no limit. 12876 12877 'max-iterations-to-track' 12878 The maximum number of iterations of a loop the brute-force 12879 algorithm for analysis of the number of iterations of the loop 12880 tries to evaluate. 12881 12882 'hot-bb-count-fraction' 12883 The denominator n of fraction 1/n of the maximal execution 12884 count of a basic block in the entire program that a basic 12885 block needs to at least have in order to be considered hot. 12886 The default is 10000, which means that a basic block is 12887 considered hot if its execution count is greater than 1/10000 12888 of the maximal execution count. 0 means that it is never 12889 considered hot. Used in non-LTO mode. 12890 12891 'hot-bb-count-ws-permille' 12892 The number of most executed permilles, ranging from 0 to 1000, 12893 of the profiled execution of the entire program to which the 12894 execution count of a basic block must be part of in order to 12895 be considered hot. The default is 990, which means that a 12896 basic block is considered hot if its execution count 12897 contributes to the upper 990 permilles, or 99.0%, of the 12898 profiled execution of the entire program. 0 means that it is 12899 never considered hot. Used in LTO mode. 12900 12901 'hot-bb-frequency-fraction' 12902 The denominator n of fraction 1/n of the execution frequency 12903 of the entry block of a function that a basic block of this 12904 function needs to at least have in order to be considered hot. 12905 The default is 1000, which means that a basic block is 12906 considered hot in a function if it is executed more frequently 12907 than 1/1000 of the frequency of the entry block of the 12908 function. 0 means that it is never considered hot. 12909 12910 'unlikely-bb-count-fraction' 12911 The denominator n of fraction 1/n of the number of profiled 12912 runs of the entire program below which the execution count of 12913 a basic block must be in order for the basic block to be 12914 considered unlikely executed. The default is 20, which means 12915 that a basic block is considered unlikely executed if it is 12916 executed in fewer than 1/20, or 5%, of the runs of the 12917 program. 0 means that it is always considered unlikely 12918 executed. 12919 12920 'max-predicted-iterations' 12921 The maximum number of loop iterations we predict statically. 12922 This is useful in cases where a function contains a single 12923 loop with known bound and another loop with unknown bound. 12924 The known number of iterations is predicted correctly, while 12925 the unknown number of iterations average to roughly 10. This 12926 means that the loop without bounds appears artificially cold 12927 relative to the other one. 12928 12929 'builtin-expect-probability' 12930 Control the probability of the expression having the specified 12931 value. This parameter takes a percentage (i.e. 0 ... 100) as 12932 input. 12933 12934 'builtin-string-cmp-inline-length' 12935 The maximum length of a constant string for a builtin string 12936 cmp call eligible for inlining. 12937 12938 'align-threshold' 12939 12940 Select fraction of the maximal frequency of executions of a 12941 basic block in a function to align the basic block. 12942 12943 'align-loop-iterations' 12944 12945 A loop expected to iterate at least the selected number of 12946 iterations is aligned. 12947 12948 'tracer-dynamic-coverage' 12949 'tracer-dynamic-coverage-feedback' 12950 12951 This value is used to limit superblock formation once the 12952 given percentage of executed instructions is covered. This 12953 limits unnecessary code size expansion. 12954 12955 The 'tracer-dynamic-coverage-feedback' parameter is used only 12956 when profile feedback is available. The real profiles (as 12957 opposed to statically estimated ones) are much less balanced 12958 allowing the threshold to be larger value. 12959 12960 'tracer-max-code-growth' 12961 Stop tail duplication once code growth has reached given 12962 percentage. This is a rather artificial limit, as most of the 12963 duplicates are eliminated later in cross jumping, so it may be 12964 set to much higher values than is the desired code growth. 12965 12966 'tracer-min-branch-ratio' 12967 12968 Stop reverse growth when the reverse probability of best edge 12969 is less than this threshold (in percent). 12970 12971 'tracer-min-branch-probability' 12972 'tracer-min-branch-probability-feedback' 12973 12974 Stop forward growth if the best edge has probability lower 12975 than this threshold. 12976 12977 Similarly to 'tracer-dynamic-coverage' two parameters are 12978 provided. 'tracer-min-branch-probability-feedback' is used 12979 for compilation with profile feedback and 12980 'tracer-min-branch-probability' compilation without. The 12981 value for compilation with profile feedback needs to be more 12982 conservative (higher) in order to make tracer effective. 12983 12984 'stack-clash-protection-guard-size' 12985 Specify the size of the operating system provided stack guard 12986 as 2 raised to NUM bytes. Higher values may reduce the number 12987 of explicit probes, but a value larger than the operating 12988 system provided guard will leave code vulnerable to stack 12989 clash style attacks. 12990 12991 'stack-clash-protection-probe-interval' 12992 Stack clash protection involves probing stack space as it is 12993 allocated. This param controls the maximum distance between 12994 probes into the stack as 2 raised to NUM bytes. Higher values 12995 may reduce the number of explicit probes, but a value larger 12996 than the operating system provided guard will leave code 12997 vulnerable to stack clash style attacks. 12998 12999 'max-cse-path-length' 13000 13001 The maximum number of basic blocks on path that CSE considers. 13002 13003 'max-cse-insns' 13004 The maximum number of instructions CSE processes before 13005 flushing. 13006 13007 'ggc-min-expand' 13008 13009 GCC uses a garbage collector to manage its own memory 13010 allocation. This parameter specifies the minimum percentage 13011 by which the garbage collector's heap should be allowed to 13012 expand between collections. Tuning this may improve 13013 compilation speed; it has no effect on code generation. 13014 13015 The default is 30% + 70% * (RAM/1GB) with an upper bound of 13016 100% when RAM >= 1GB. If 'getrlimit' is available, the notion 13017 of "RAM" is the smallest of actual RAM and 'RLIMIT_DATA' or 13018 'RLIMIT_AS'. If GCC is not able to calculate RAM on a 13019 particular platform, the lower bound of 30% is used. Setting 13020 this parameter and 'ggc-min-heapsize' to zero causes a full 13021 collection to occur at every opportunity. This is extremely 13022 slow, but can be useful for debugging. 13023 13024 'ggc-min-heapsize' 13025 13026 Minimum size of the garbage collector's heap before it begins 13027 bothering to collect garbage. The first collection occurs 13028 after the heap expands by 'ggc-min-expand'% beyond 13029 'ggc-min-heapsize'. Again, tuning this may improve 13030 compilation speed, and has no effect on code generation. 13031 13032 The default is the smaller of RAM/8, RLIMIT_RSS, or a limit 13033 that tries to ensure that RLIMIT_DATA or RLIMIT_AS are not 13034 exceeded, but with a lower bound of 4096 (four megabytes) and 13035 an upper bound of 131072 (128 megabytes). If GCC is not able 13036 to calculate RAM on a particular platform, the lower bound is 13037 used. Setting this parameter very large effectively disables 13038 garbage collection. Setting this parameter and 13039 'ggc-min-expand' to zero causes a full collection to occur at 13040 every opportunity. 13041 13042 'max-reload-search-insns' 13043 The maximum number of instruction reload should look backward 13044 for equivalent register. Increasing values mean more 13045 aggressive optimization, making the compilation time increase 13046 with probably slightly better performance. 13047 13048 'max-cselib-memory-locations' 13049 The maximum number of memory locations cselib should take into 13050 account. Increasing values mean more aggressive optimization, 13051 making the compilation time increase with probably slightly 13052 better performance. 13053 13054 'max-sched-ready-insns' 13055 The maximum number of instructions ready to be issued the 13056 scheduler should consider at any given time during the first 13057 scheduling pass. Increasing values mean more thorough 13058 searches, making the compilation time increase with probably 13059 little benefit. 13060 13061 'max-sched-region-blocks' 13062 The maximum number of blocks in a region to be considered for 13063 interblock scheduling. 13064 13065 'max-pipeline-region-blocks' 13066 The maximum number of blocks in a region to be considered for 13067 pipelining in the selective scheduler. 13068 13069 'max-sched-region-insns' 13070 The maximum number of insns in a region to be considered for 13071 interblock scheduling. 13072 13073 'max-pipeline-region-insns' 13074 The maximum number of insns in a region to be considered for 13075 pipelining in the selective scheduler. 13076 13077 'min-spec-prob' 13078 The minimum probability (in percents) of reaching a source 13079 block for interblock speculative scheduling. 13080 13081 'max-sched-extend-regions-iters' 13082 The maximum number of iterations through CFG to extend 13083 regions. A value of 0 disables region extensions. 13084 13085 'max-sched-insn-conflict-delay' 13086 The maximum conflict delay for an insn to be considered for 13087 speculative motion. 13088 13089 'sched-spec-prob-cutoff' 13090 The minimal probability of speculation success (in percents), 13091 so that speculative insns are scheduled. 13092 13093 'sched-state-edge-prob-cutoff' 13094 The minimum probability an edge must have for the scheduler to 13095 save its state across it. 13096 13097 'sched-mem-true-dep-cost' 13098 Minimal distance (in CPU cycles) between store and load 13099 targeting same memory locations. 13100 13101 'selsched-max-lookahead' 13102 The maximum size of the lookahead window of selective 13103 scheduling. It is a depth of search for available 13104 instructions. 13105 13106 'selsched-max-sched-times' 13107 The maximum number of times that an instruction is scheduled 13108 during selective scheduling. This is the limit on the number 13109 of iterations through which the instruction may be pipelined. 13110 13111 'selsched-insns-to-rename' 13112 The maximum number of best instructions in the ready list that 13113 are considered for renaming in the selective scheduler. 13114 13115 'sms-min-sc' 13116 The minimum value of stage count that swing modulo scheduler 13117 generates. 13118 13119 'max-last-value-rtl' 13120 The maximum size measured as number of RTLs that can be 13121 recorded in an expression in combiner for a pseudo register as 13122 last known value of that register. 13123 13124 'max-combine-insns' 13125 The maximum number of instructions the RTL combiner tries to 13126 combine. 13127 13128 'integer-share-limit' 13129 Small integer constants can use a shared data structure, 13130 reducing the compiler's memory usage and increasing its speed. 13131 This sets the maximum value of a shared integer constant. 13132 13133 'ssp-buffer-size' 13134 The minimum size of buffers (i.e. arrays) that receive stack 13135 smashing protection when '-fstack-protector' is used. 13136 13137 'min-size-for-stack-sharing' 13138 The minimum size of variables taking part in stack slot 13139 sharing when not optimizing. 13140 13141 'max-jump-thread-duplication-stmts' 13142 Maximum number of statements allowed in a block that needs to 13143 be duplicated when threading jumps. 13144 13145 'max-fields-for-field-sensitive' 13146 Maximum number of fields in a structure treated in a field 13147 sensitive manner during pointer analysis. 13148 13149 'prefetch-latency' 13150 Estimate on average number of instructions that are executed 13151 before prefetch finishes. The distance prefetched ahead is 13152 proportional to this constant. Increasing this number may 13153 also lead to less streams being prefetched (see 13154 'simultaneous-prefetches'). 13155 13156 'simultaneous-prefetches' 13157 Maximum number of prefetches that can run at the same time. 13158 13159 'l1-cache-line-size' 13160 The size of cache line in L1 data cache, in bytes. 13161 13162 'l1-cache-size' 13163 The size of L1 data cache, in kilobytes. 13164 13165 'l2-cache-size' 13166 The size of L2 data cache, in kilobytes. 13167 13168 'prefetch-dynamic-strides' 13169 Whether the loop array prefetch pass should issue software 13170 prefetch hints for strides that are non-constant. In some 13171 cases this may be beneficial, though the fact the stride is 13172 non-constant may make it hard to predict when there is clear 13173 benefit to issuing these hints. 13174 13175 Set to 1 if the prefetch hints should be issued for 13176 non-constant strides. Set to 0 if prefetch hints should be 13177 issued only for strides that are known to be constant and 13178 below 'prefetch-minimum-stride'. 13179 13180 'prefetch-minimum-stride' 13181 Minimum constant stride, in bytes, to start using prefetch 13182 hints for. If the stride is less than this threshold, 13183 prefetch hints will not be issued. 13184 13185 This setting is useful for processors that have hardware 13186 prefetchers, in which case there may be conflicts between the 13187 hardware prefetchers and the software prefetchers. If the 13188 hardware prefetchers have a maximum stride they can handle, it 13189 should be used here to improve the use of software 13190 prefetchers. 13191 13192 A value of -1 means we don't have a threshold and therefore 13193 prefetch hints can be issued for any constant stride. 13194 13195 This setting is only useful for strides that are known and 13196 constant. 13197 13198 'destructive-interference-size' 13199 'constructive-interference-size' 13200 The values for the C++17 variables 13201 'std::hardware_destructive_interference_size' and 13202 'std::hardware_constructive_interference_size'. The 13203 destructive interference size is the minimum recommended 13204 offset between two independent concurrently-accessed objects; 13205 the constructive interference size is the maximum recommended 13206 size of contiguous memory accessed together. Typically both 13207 will be the size of an L1 cache line for the target, in bytes. 13208 For a generic target covering a range of L1 cache line sizes, 13209 typically the constructive interference size will be the small 13210 end of the range and the destructive size will be the large 13211 end. 13212 13213 The destructive interference size is intended to be used for 13214 layout, and thus has ABI impact. The default value is not 13215 expected to be stable, and on some targets varies with 13216 '-mtune', so use of this variable in a context where ABI 13217 stability is important, such as the public interface of a 13218 library, is strongly discouraged; if it is used in that 13219 context, users can stabilize the value using this option. 13220 13221 The constructive interference size is less sensitive, as it is 13222 typically only used in a 'static_assert' to make sure that a 13223 type fits within a cache line. 13224 13225 See also '-Winterference-size'. 13226 13227 'loop-interchange-max-num-stmts' 13228 The maximum number of stmts in a loop to be interchanged. 13229 13230 'loop-interchange-stride-ratio' 13231 The minimum ratio between stride of two loops for interchange 13232 to be profitable. 13233 13234 'min-insn-to-prefetch-ratio' 13235 The minimum ratio between the number of instructions and the 13236 number of prefetches to enable prefetching in a loop. 13237 13238 'prefetch-min-insn-to-mem-ratio' 13239 The minimum ratio between the number of instructions and the 13240 number of memory references to enable prefetching in a loop. 13241 13242 'use-canonical-types' 13243 Whether the compiler should use the "canonical" type system. 13244 Should always be 1, which uses a more efficient internal 13245 mechanism for comparing types in C++ and Objective-C++. 13246 However, if bugs in the canonical type system are causing 13247 compilation failures, set this value to 0 to disable canonical 13248 types. 13249 13250 'switch-conversion-max-branch-ratio' 13251 Switch initialization conversion refuses to create arrays that 13252 are bigger than 'switch-conversion-max-branch-ratio' times the 13253 number of branches in the switch. 13254 13255 'max-partial-antic-length' 13256 Maximum length of the partial antic set computed during the 13257 tree partial redundancy elimination optimization 13258 ('-ftree-pre') when optimizing at '-O3' and above. For some 13259 sorts of source code the enhanced partial redundancy 13260 elimination optimization can run away, consuming all of the 13261 memory available on the host machine. This parameter sets a 13262 limit on the length of the sets that are computed, which 13263 prevents the runaway behavior. Setting a value of 0 for this 13264 parameter allows an unlimited set length. 13265 13266 'rpo-vn-max-loop-depth' 13267 Maximum loop depth that is value-numbered optimistically. 13268 When the limit hits the innermost RPO-VN-MAX-LOOP-DEPTH loops 13269 and the outermost loop in the loop nest are value-numbered 13270 optimistically and the remaining ones not. 13271 13272 'sccvn-max-alias-queries-per-access' 13273 Maximum number of alias-oracle queries we perform when looking 13274 for redundancies for loads and stores. If this limit is hit 13275 the search is aborted and the load or store is not considered 13276 redundant. The number of queries is algorithmically limited 13277 to the number of stores on all paths from the load to the 13278 function entry. 13279 13280 'ira-max-loops-num' 13281 IRA uses regional register allocation by default. If a 13282 function contains more loops than the number given by this 13283 parameter, only at most the given number of the most 13284 frequently-executed loops form regions for regional register 13285 allocation. 13286 13287 'ira-max-conflict-table-size' 13288 Although IRA uses a sophisticated algorithm to compress the 13289 conflict table, the table can still require excessive amounts 13290 of memory for huge functions. If the conflict table for a 13291 function could be more than the size in MB given by this 13292 parameter, the register allocator instead uses a faster, 13293 simpler, and lower-quality algorithm that does not require 13294 building a pseudo-register conflict table. 13295 13296 'ira-loop-reserved-regs' 13297 IRA can be used to evaluate more accurate register pressure in 13298 loops for decisions to move loop invariants (see '-O3'). The 13299 number of available registers reserved for some other purposes 13300 is given by this parameter. Default of the parameter is the 13301 best found from numerous experiments. 13302 13303 'ira-consider-dup-in-all-alts' 13304 Make IRA to consider matching constraint (duplicated operand 13305 number) heavily in all available alternatives for preferred 13306 register class. If it is set as zero, it means IRA only 13307 respects the matching constraint when it's in the only 13308 available alternative with an appropriate register class. 13309 Otherwise, it means IRA will check all available alternatives 13310 for preferred register class even if it has found some choice 13311 with an appropriate register class and respect the found 13312 qualified matching constraint. 13313 13314 'lra-inheritance-ebb-probability-cutoff' 13315 LRA tries to reuse values reloaded in registers in subsequent 13316 insns. This optimization is called inheritance. EBB is used 13317 as a region to do this optimization. The parameter defines a 13318 minimal fall-through edge probability in percentage used to 13319 add BB to inheritance EBB in LRA. The default value was chosen 13320 from numerous runs of SPEC2000 on x86-64. 13321 13322 'loop-invariant-max-bbs-in-loop' 13323 Loop invariant motion can be very expensive, both in 13324 compilation time and in amount of needed compile-time memory, 13325 with very large loops. Loops with more basic blocks than this 13326 parameter won't have loop invariant motion optimization 13327 performed on them. 13328 13329 'loop-max-datarefs-for-datadeps' 13330 Building data dependencies is expensive for very large loops. 13331 This parameter limits the number of data references in loops 13332 that are considered for data dependence analysis. These large 13333 loops are no handled by the optimizations using loop data 13334 dependencies. 13335 13336 'max-vartrack-size' 13337 Sets a maximum number of hash table slots to use during 13338 variable tracking dataflow analysis of any function. If this 13339 limit is exceeded with variable tracking at assignments 13340 enabled, analysis for that function is retried without it, 13341 after removing all debug insns from the function. If the 13342 limit is exceeded even without debug insns, var tracking 13343 analysis is completely disabled for the function. Setting the 13344 parameter to zero makes it unlimited. 13345 13346 'max-vartrack-expr-depth' 13347 Sets a maximum number of recursion levels when attempting to 13348 map variable names or debug temporaries to value expressions. 13349 This trades compilation time for more complete debug 13350 information. If this is set too low, value expressions that 13351 are available and could be represented in debug information 13352 may end up not being used; setting this higher may enable the 13353 compiler to find more complex debug expressions, but compile 13354 time and memory use may grow. 13355 13356 'max-debug-marker-count' 13357 Sets a threshold on the number of debug markers (e.g. begin 13358 stmt markers) to avoid complexity explosion at inlining or 13359 expanding to RTL. If a function has more such gimple stmts 13360 than the set limit, such stmts will be dropped from the 13361 inlined copy of a function, and from its RTL expansion. 13362 13363 'min-nondebug-insn-uid' 13364 Use uids starting at this parameter for nondebug insns. The 13365 range below the parameter is reserved exclusively for debug 13366 insns created by '-fvar-tracking-assignments', but debug insns 13367 may get (non-overlapping) uids above it if the reserved range 13368 is exhausted. 13369 13370 'ipa-sra-ptr-growth-factor' 13371 IPA-SRA replaces a pointer to an aggregate with one or more 13372 new parameters only when their cumulative size is less or 13373 equal to 'ipa-sra-ptr-growth-factor' times the size of the 13374 original pointer parameter. 13375 13376 'ipa-sra-max-replacements' 13377 Maximum pieces of an aggregate that IPA-SRA tracks. As a 13378 consequence, it is also the maximum number of replacements of 13379 a formal parameter. 13380 13381 'sra-max-scalarization-size-Ospeed' 13382 'sra-max-scalarization-size-Osize' 13383 The two Scalar Reduction of Aggregates passes (SRA and 13384 IPA-SRA) aim to replace scalar parts of aggregates with uses 13385 of independent scalar variables. These parameters control the 13386 maximum size, in storage units, of aggregate which is 13387 considered for replacement when compiling for speed 13388 ('sra-max-scalarization-size-Ospeed') or size 13389 ('sra-max-scalarization-size-Osize') respectively. 13390 13391 'sra-max-propagations' 13392 The maximum number of artificial accesses that Scalar 13393 Replacement of Aggregates (SRA) will track, per one local 13394 variable, in order to facilitate copy propagation. 13395 13396 'tm-max-aggregate-size' 13397 When making copies of thread-local variables in a transaction, 13398 this parameter specifies the size in bytes after which 13399 variables are saved with the logging functions as opposed to 13400 save/restore code sequence pairs. This option only applies 13401 when using '-fgnu-tm'. 13402 13403 'graphite-max-nb-scop-params' 13404 To avoid exponential effects in the Graphite loop transforms, 13405 the number of parameters in a Static Control Part (SCoP) is 13406 bounded. A value of zero can be used to lift the bound. A 13407 variable whose value is unknown at compilation time and 13408 defined outside a SCoP is a parameter of the SCoP. 13409 13410 'loop-block-tile-size' 13411 Loop blocking or strip mining transforms, enabled with 13412 '-floop-block' or '-floop-strip-mine', strip mine each loop in 13413 the loop nest by a given number of iterations. The strip 13414 length can be changed using the 'loop-block-tile-size' 13415 parameter. 13416 13417 'ipa-jump-function-lookups' 13418 Specifies number of statements visited during jump function 13419 offset discovery. 13420 13421 'ipa-cp-value-list-size' 13422 IPA-CP attempts to track all possible values and types passed 13423 to a function's parameter in order to propagate them and 13424 perform devirtualization. 'ipa-cp-value-list-size' is the 13425 maximum number of values and types it stores per one formal 13426 parameter of a function. 13427 13428 'ipa-cp-eval-threshold' 13429 IPA-CP calculates its own score of cloning profitability 13430 heuristics and performs those cloning opportunities with 13431 scores that exceed 'ipa-cp-eval-threshold'. 13432 13433 'ipa-cp-max-recursive-depth' 13434 Maximum depth of recursive cloning for self-recursive 13435 function. 13436 13437 'ipa-cp-min-recursive-probability' 13438 Recursive cloning only when the probability of call being 13439 executed exceeds the parameter. 13440 13441 'ipa-cp-profile-count-base' 13442 When using '-fprofile-use' option, IPA-CP will consider the 13443 measured execution count of a call graph edge at this 13444 percentage position in their histogram as the basis for its 13445 heuristics calculation. 13446 13447 'ipa-cp-recursive-freq-factor' 13448 The number of times interprocedural copy propagation expects 13449 recursive functions to call themselves. 13450 13451 'ipa-cp-recursion-penalty' 13452 Percentage penalty the recursive functions will receive when 13453 they are evaluated for cloning. 13454 13455 'ipa-cp-single-call-penalty' 13456 Percentage penalty functions containing a single call to 13457 another function will receive when they are evaluated for 13458 cloning. 13459 13460 'ipa-max-agg-items' 13461 IPA-CP is also capable to propagate a number of scalar values 13462 passed in an aggregate. 'ipa-max-agg-items' controls the 13463 maximum number of such values per one parameter. 13464 13465 'ipa-cp-loop-hint-bonus' 13466 When IPA-CP determines that a cloning candidate would make the 13467 number of iterations of a loop known, it adds a bonus of 13468 'ipa-cp-loop-hint-bonus' to the profitability score of the 13469 candidate. 13470 13471 'ipa-max-loop-predicates' 13472 The maximum number of different predicates IPA will use to 13473 describe when loops in a function have known properties. 13474 13475 'ipa-max-aa-steps' 13476 During its analysis of function bodies, IPA-CP employs alias 13477 analysis in order to track values pointed to by function 13478 parameters. In order not spend too much time analyzing huge 13479 functions, it gives up and consider all memory clobbered after 13480 examining 'ipa-max-aa-steps' statements modifying memory. 13481 13482 'ipa-max-switch-predicate-bounds' 13483 Maximal number of boundary endpoints of case ranges of switch 13484 statement. For switch exceeding this limit, IPA-CP will not 13485 construct cloning cost predicate, which is used to estimate 13486 cloning benefit, for default case of the switch statement. 13487 13488 'ipa-max-param-expr-ops' 13489 IPA-CP will analyze conditional statement that references some 13490 function parameter to estimate benefit for cloning upon 13491 certain constant value. But if number of operations in a 13492 parameter expression exceeds 'ipa-max-param-expr-ops', the 13493 expression is treated as complicated one, and is not handled 13494 by IPA analysis. 13495 13496 'lto-partitions' 13497 Specify desired number of partitions produced during WHOPR 13498 compilation. The number of partitions should exceed the 13499 number of CPUs used for compilation. 13500 13501 'lto-min-partition' 13502 Size of minimal partition for WHOPR (in estimated 13503 instructions). This prevents expenses of splitting very small 13504 programs into too many partitions. 13505 13506 'lto-max-partition' 13507 Size of max partition for WHOPR (in estimated instructions). 13508 to provide an upper bound for individual size of partition. 13509 Meant to be used only with balanced partitioning. 13510 13511 'lto-max-streaming-parallelism' 13512 Maximal number of parallel processes used for LTO streaming. 13513 13514 'cxx-max-namespaces-for-diagnostic-help' 13515 The maximum number of namespaces to consult for suggestions 13516 when C++ name lookup fails for an identifier. 13517 13518 'sink-frequency-threshold' 13519 The maximum relative execution frequency (in percents) of the 13520 target block relative to a statement's original block to allow 13521 statement sinking of a statement. Larger numbers result in 13522 more aggressive statement sinking. A small positive 13523 adjustment is applied for statements with memory operands as 13524 those are even more profitable so sink. 13525 13526 'max-stores-to-sink' 13527 The maximum number of conditional store pairs that can be 13528 sunk. Set to 0 if either vectorization ('-ftree-vectorize') 13529 or if-conversion ('-ftree-loop-if-convert') is disabled. 13530 13531 'case-values-threshold' 13532 The smallest number of different values for which it is best 13533 to use a jump-table instead of a tree of conditional branches. 13534 If the value is 0, use the default for the machine. 13535 13536 'jump-table-max-growth-ratio-for-size' 13537 The maximum code size growth ratio when expanding into a jump 13538 table (in percent). The parameter is used when optimizing for 13539 size. 13540 13541 'jump-table-max-growth-ratio-for-speed' 13542 The maximum code size growth ratio when expanding into a jump 13543 table (in percent). The parameter is used when optimizing for 13544 speed. 13545 13546 'tree-reassoc-width' 13547 Set the maximum number of instructions executed in parallel in 13548 reassociated tree. This parameter overrides target dependent 13549 heuristics used by default if has non zero value. 13550 13551 'sched-pressure-algorithm' 13552 Choose between the two available implementations of 13553 '-fsched-pressure'. Algorithm 1 is the original 13554 implementation and is the more likely to prevent instructions 13555 from being reordered. Algorithm 2 was designed to be a 13556 compromise between the relatively conservative approach taken 13557 by algorithm 1 and the rather aggressive approach taken by the 13558 default scheduler. It relies more heavily on having a regular 13559 register file and accurate register pressure classes. See 13560 'haifa-sched.cc' in the GCC sources for more details. 13561 13562 The default choice depends on the target. 13563 13564 'max-slsr-cand-scan' 13565 Set the maximum number of existing candidates that are 13566 considered when seeking a basis for a new straight-line 13567 strength reduction candidate. 13568 13569 'asan-globals' 13570 Enable buffer overflow detection for global objects. This 13571 kind of protection is enabled by default if you are using 13572 '-fsanitize=address' option. To disable global objects 13573 protection use '--param asan-globals=0'. 13574 13575 'asan-stack' 13576 Enable buffer overflow detection for stack objects. This kind 13577 of protection is enabled by default when using 13578 '-fsanitize=address'. To disable stack protection use 13579 '--param asan-stack=0' option. 13580 13581 'asan-instrument-reads' 13582 Enable buffer overflow detection for memory reads. This kind 13583 of protection is enabled by default when using 13584 '-fsanitize=address'. To disable memory reads protection use 13585 '--param asan-instrument-reads=0'. 13586 13587 'asan-instrument-writes' 13588 Enable buffer overflow detection for memory writes. This kind 13589 of protection is enabled by default when using 13590 '-fsanitize=address'. To disable memory writes protection use 13591 '--param asan-instrument-writes=0' option. 13592 13593 'asan-memintrin' 13594 Enable detection for built-in functions. This kind of 13595 protection is enabled by default when using 13596 '-fsanitize=address'. To disable built-in functions 13597 protection use '--param asan-memintrin=0'. 13598 13599 'asan-use-after-return' 13600 Enable detection of use-after-return. This kind of protection 13601 is enabled by default when using the '-fsanitize=address' 13602 option. To disable it use '--param asan-use-after-return=0'. 13603 13604 Note: By default the check is disabled at run time. To enable 13605 it, add 'detect_stack_use_after_return=1' to the environment 13606 variable 'ASAN_OPTIONS'. 13607 13608 'asan-instrumentation-with-call-threshold' 13609 If number of memory accesses in function being instrumented is 13610 greater or equal to this number, use callbacks instead of 13611 inline checks. E.g. to disable inline code use '--param 13612 asan-instrumentation-with-call-threshold=0'. 13613 13614 'hwasan-instrument-stack' 13615 Enable hwasan instrumentation of statically sized 13616 stack-allocated variables. This kind of instrumentation is 13617 enabled by default when using '-fsanitize=hwaddress' and 13618 disabled by default when using '-fsanitize=kernel-hwaddress'. 13619 To disable stack instrumentation use '--param 13620 hwasan-instrument-stack=0', and to enable it use '--param 13621 hwasan-instrument-stack=1'. 13622 13623 'hwasan-random-frame-tag' 13624 When using stack instrumentation, decide tags for stack 13625 variables using a deterministic sequence beginning at a random 13626 tag for each frame. With this parameter unset tags are chosen 13627 using the same sequence but beginning from 1. This is enabled 13628 by default for '-fsanitize=hwaddress' and unavailable for 13629 '-fsanitize=kernel-hwaddress'. To disable it use '--param 13630 hwasan-random-frame-tag=0'. 13631 13632 'hwasan-instrument-allocas' 13633 Enable hwasan instrumentation of dynamically sized 13634 stack-allocated variables. This kind of instrumentation is 13635 enabled by default when using '-fsanitize=hwaddress' and 13636 disabled by default when using '-fsanitize=kernel-hwaddress'. 13637 To disable instrumentation of such variables use '--param 13638 hwasan-instrument-allocas=0', and to enable it use '--param 13639 hwasan-instrument-allocas=1'. 13640 13641 'hwasan-instrument-reads' 13642 Enable hwasan checks on memory reads. Instrumentation of 13643 reads is enabled by default for both '-fsanitize=hwaddress' 13644 and '-fsanitize=kernel-hwaddress'. To disable checking memory 13645 reads use '--param hwasan-instrument-reads=0'. 13646 13647 'hwasan-instrument-writes' 13648 Enable hwasan checks on memory writes. Instrumentation of 13649 writes is enabled by default for both '-fsanitize=hwaddress' 13650 and '-fsanitize=kernel-hwaddress'. To disable checking memory 13651 writes use '--param hwasan-instrument-writes=0'. 13652 13653 'hwasan-instrument-mem-intrinsics' 13654 Enable hwasan instrumentation of builtin functions. 13655 Instrumentation of these builtin functions is enabled by 13656 default for both '-fsanitize=hwaddress' and 13657 '-fsanitize=kernel-hwaddress'. To disable instrumentation of 13658 builtin functions use '--param 13659 hwasan-instrument-mem-intrinsics=0'. 13660 13661 'use-after-scope-direct-emission-threshold' 13662 If the size of a local variable in bytes is smaller or equal 13663 to this number, directly poison (or unpoison) shadow memory 13664 instead of using run-time callbacks. 13665 13666 'tsan-distinguish-volatile' 13667 Emit special instrumentation for accesses to volatiles. 13668 13669 'tsan-instrument-func-entry-exit' 13670 Emit instrumentation calls to __tsan_func_entry() and 13671 __tsan_func_exit(). 13672 13673 'max-fsm-thread-path-insns' 13674 Maximum number of instructions to copy when duplicating blocks 13675 on a finite state automaton jump thread path. 13676 13677 'max-fsm-thread-length' 13678 Maximum number of basic blocks on a jump thread path. 13679 13680 'threader-debug' 13681 threader-debug=[none|all] Enables verbose dumping of the 13682 threader solver. 13683 13684 'parloops-chunk-size' 13685 Chunk size of omp schedule for loops parallelized by parloops. 13686 13687 'parloops-schedule' 13688 Schedule type of omp schedule for loops parallelized by 13689 parloops (static, dynamic, guided, auto, runtime). 13690 13691 'parloops-min-per-thread' 13692 The minimum number of iterations per thread of an innermost 13693 parallelized loop for which the parallelized variant is 13694 preferred over the single threaded one. Note that for a 13695 parallelized loop nest the minimum number of iterations of the 13696 outermost loop per thread is two. 13697 13698 'max-ssa-name-query-depth' 13699 Maximum depth of recursion when querying properties of SSA 13700 names in things like fold routines. One level of recursion 13701 corresponds to following a use-def chain. 13702 13703 'max-speculative-devirt-maydefs' 13704 The maximum number of may-defs we analyze when looking for a 13705 must-def specifying the dynamic type of an object that invokes 13706 a virtual call we may be able to devirtualize speculatively. 13707 13708 'max-vrp-switch-assertions' 13709 The maximum number of assertions to add along the default edge 13710 of a switch statement during VRP. 13711 13712 'evrp-sparse-threshold' 13713 Maximum number of basic blocks before EVRP uses a sparse 13714 cache. 13715 13716 'evrp-mode' 13717 Specifies the mode Early VRP should operate in. 13718 13719 'vrp1-mode' 13720 Specifies the mode VRP pass 1 should operate in. 13721 13722 'vrp2-mode' 13723 Specifies the mode VRP pass 2 should operate in. 13724 13725 'ranger-debug' 13726 Specifies the type of debug output to be issued for ranges. 13727 13728 'evrp-switch-limit' 13729 Specifies the maximum number of switch cases before EVRP 13730 ignores a switch. 13731 13732 'unroll-jam-min-percent' 13733 The minimum percentage of memory references that must be 13734 optimized away for the unroll-and-jam transformation to be 13735 considered profitable. 13736 13737 'unroll-jam-max-unroll' 13738 The maximum number of times the outer loop should be unrolled 13739 by the unroll-and-jam transformation. 13740 13741 'max-rtl-if-conversion-unpredictable-cost' 13742 Maximum permissible cost for the sequence that would be 13743 generated by the RTL if-conversion pass for a branch that is 13744 considered unpredictable. 13745 13746 'max-variable-expansions-in-unroller' 13747 If '-fvariable-expansion-in-unroller' is used, the maximum 13748 number of times that an individual variable will be expanded 13749 during loop unrolling. 13750 13751 'partial-inlining-entry-probability' 13752 Maximum probability of the entry BB of split region (in 13753 percent relative to entry BB of the function) to make partial 13754 inlining happen. 13755 13756 'max-tracked-strlens' 13757 Maximum number of strings for which strlen optimization pass 13758 will track string lengths. 13759 13760 'gcse-after-reload-partial-fraction' 13761 The threshold ratio for performing partial redundancy 13762 elimination after reload. 13763 13764 'gcse-after-reload-critical-fraction' 13765 The threshold ratio of critical edges execution count that 13766 permit performing redundancy elimination after reload. 13767 13768 'max-loop-header-insns' 13769 The maximum number of insns in loop header duplicated by the 13770 copy loop headers pass. 13771 13772 'vect-epilogues-nomask' 13773 Enable loop epilogue vectorization using smaller vector size. 13774 13775 'vect-partial-vector-usage' 13776 Controls when the loop vectorizer considers using partial 13777 vector loads and stores as an alternative to falling back to 13778 scalar code. 0 stops the vectorizer from ever using partial 13779 vector loads and stores. 1 allows partial vector loads and 13780 stores if vectorization removes the need for the code to 13781 iterate. 2 allows partial vector loads and stores in all 13782 loops. The parameter only has an effect on targets that 13783 support partial vector loads and stores. 13784 13785 'vect-inner-loop-cost-factor' 13786 The maximum factor which the loop vectorizer applies to the 13787 cost of statements in an inner loop relative to the loop being 13788 vectorized. The factor applied is the maximum of the 13789 estimated number of iterations of the inner loop and this 13790 parameter. The default value of this parameter is 50. 13791 13792 'vect-induction-float' 13793 Enable loop vectorization of floating point inductions. 13794 13795 'avoid-fma-max-bits' 13796 Maximum number of bits for which we avoid creating FMAs. 13797 13798 'sms-loop-average-count-threshold' 13799 A threshold on the average loop count considered by the swing 13800 modulo scheduler. 13801 13802 'sms-dfa-history' 13803 The number of cycles the swing modulo scheduler considers when 13804 checking conflicts using DFA. 13805 13806 'graphite-allow-codegen-errors' 13807 Whether codegen errors should be ICEs when '-fchecking'. 13808 13809 'sms-max-ii-factor' 13810 A factor for tuning the upper bound that swing modulo 13811 scheduler uses for scheduling a loop. 13812 13813 'lra-max-considered-reload-pseudos' 13814 The max number of reload pseudos which are considered during 13815 spilling a non-reload pseudo. 13816 13817 'max-pow-sqrt-depth' 13818 Maximum depth of sqrt chains to use when synthesizing 13819 exponentiation by a real constant. 13820 13821 'max-dse-active-local-stores' 13822 Maximum number of active local stores in RTL dead store 13823 elimination. 13824 13825 'asan-instrument-allocas' 13826 Enable asan allocas/VLAs protection. 13827 13828 'max-iterations-computation-cost' 13829 Bound on the cost of an expression to compute the number of 13830 iterations. 13831 13832 'max-isl-operations' 13833 Maximum number of isl operations, 0 means unlimited. 13834 13835 'graphite-max-arrays-per-scop' 13836 Maximum number of arrays per scop. 13837 13838 'max-vartrack-reverse-op-size' 13839 Max. size of loc list for which reverse ops should be added. 13840 13841 'fsm-scale-path-stmts' 13842 Scale factor to apply to the number of statements in a 13843 threading path when comparing to the number of (scaled) 13844 blocks. 13845 13846 'uninit-control-dep-attempts' 13847 Maximum number of nested calls to search for control 13848 dependencies during uninitialized variable analysis. 13849 13850 'fsm-scale-path-blocks' 13851 Scale factor to apply to the number of blocks in a threading 13852 path when comparing to the number of (scaled) statements. 13853 13854 'sched-autopref-queue-depth' 13855 Hardware autoprefetcher scheduler model control flag. Number 13856 of lookahead cycles the model looks into; at ' ' only enable 13857 instruction sorting heuristic. 13858 13859 'loop-versioning-max-inner-insns' 13860 The maximum number of instructions that an inner loop can have 13861 before the loop versioning pass considers it too big to copy. 13862 13863 'loop-versioning-max-outer-insns' 13864 The maximum number of instructions that an outer loop can have 13865 before the loop versioning pass considers it too big to copy, 13866 discounting any instructions in inner loops that directly 13867 benefit from versioning. 13868 13869 'ssa-name-def-chain-limit' 13870 The maximum number of SSA_NAME assignments to follow in 13871 determining a property of a variable such as its value. This 13872 limits the number of iterations or recursive calls GCC 13873 performs when optimizing certain statements or when 13874 determining their validity prior to issuing diagnostics. 13875 13876 'store-merging-max-size' 13877 Maximum size of a single store merging region in bytes. 13878 13879 'hash-table-verification-limit' 13880 The number of elements for which hash table verification is 13881 done for each searched element. 13882 13883 'max-find-base-term-values' 13884 Maximum number of VALUEs handled during a single 13885 find_base_term call. 13886 13887 'analyzer-max-enodes-per-program-point' 13888 The maximum number of exploded nodes per program point within 13889 the analyzer, before terminating analysis of that point. 13890 13891 'analyzer-max-constraints' 13892 The maximum number of constraints per state. 13893 13894 'analyzer-min-snodes-for-call-summary' 13895 The minimum number of supernodes within a function for the 13896 analyzer to consider summarizing its effects at call sites. 13897 13898 'analyzer-max-enodes-for-full-dump' 13899 The maximum depth of exploded nodes that should appear in a 13900 dot dump before switching to a less verbose format. 13901 13902 'analyzer-max-recursion-depth' 13903 The maximum number of times a callsite can appear in a call 13904 stack within the analyzer, before terminating analysis of a 13905 call that would recurse deeper. 13906 13907 'analyzer-max-svalue-depth' 13908 The maximum depth of a symbolic value, before approximating 13909 the value as unknown. 13910 13911 'analyzer-max-infeasible-edges' 13912 The maximum number of infeasible edges to reject before 13913 declaring a diagnostic as infeasible. 13914 13915 'gimple-fe-computed-hot-bb-threshold' 13916 The number of executions of a basic block which is considered 13917 hot. The parameter is used only in GIMPLE FE. 13918 13919 'analyzer-bb-explosion-factor' 13920 The maximum number of 'after supernode' exploded nodes within 13921 the analyzer per supernode, before terminating analysis. 13922 13923 'ranger-logical-depth' 13924 Maximum depth of logical expression evaluation ranger will 13925 look through when evaluating outgoing edge ranges. 13926 13927 'relation-block-limit' 13928 Maximum number of relations the oracle will register in a 13929 basic block. 13930 13931 'min-pagesize' 13932 Minimum page size for warning purposes. 13933 13934 'openacc-kernels' 13935 Specify mode of OpenACC 'kernels' constructs handling. With 13936 '--param=openacc-kernels=decompose', OpenACC 'kernels' 13937 constructs are decomposed into parts, a sequence of compute 13938 constructs, each then handled individually. This is work in 13939 progress. With '--param=openacc-kernels=parloops', OpenACC 13940 'kernels' constructs are handled by the 'parloops' pass, en 13941 bloc. This is the current default. 13942 13943 'openacc-privatization' 13944 Specify mode of OpenACC privatization diagnostics for 13945 '-fopt-info-omp-note' and applicable '-fdump-tree-*-details'. 13946 With '--param=openacc-privatization=quiet', don't diagnose. 13947 This is the current default. With 13948 '--param=openacc-privatization=noisy', do diagnose. 13949 13950 The following choices of NAME are available on AArch64 targets: 13951 13952 'aarch64-sve-compare-costs' 13953 When vectorizing for SVE, consider using "unpacked" vectors 13954 for smaller elements and use the cost model to pick the 13955 cheapest approach. Also use the cost model to choose between 13956 SVE and Advanced SIMD vectorization. 13957 13958 Using unpacked vectors includes storing smaller elements in 13959 larger containers and accessing elements with extending loads 13960 and truncating stores. 13961 13962 'aarch64-float-recp-precision' 13963 The number of Newton iterations for calculating the reciprocal 13964 for float type. The precision of division is proportional to 13965 this param when division approximation is enabled. The 13966 default value is 1. 13967 13968 'aarch64-double-recp-precision' 13969 The number of Newton iterations for calculating the reciprocal 13970 for double type. The precision of division is propotional to 13971 this param when division approximation is enabled. The 13972 default value is 2. 13973 13974 'aarch64-autovec-preference' 13975 Force an ISA selection strategy for auto-vectorization. 13976 Accepts values from 0 to 4, inclusive. 13977 '0' 13978 Use the default heuristics. 13979 '1' 13980 Use only Advanced SIMD for auto-vectorization. 13981 '2' 13982 Use only SVE for auto-vectorization. 13983 '3' 13984 Use both Advanced SIMD and SVE. Prefer Advanced SIMD when 13985 the costs are deemed equal. 13986 '4' 13987 Use both Advanced SIMD and SVE. Prefer SVE when the costs 13988 are deemed equal. 13989 The default value is 0. 13990 13991 'aarch64-loop-vect-issue-rate-niters' 13992 The tuning for some AArch64 CPUs tries to take both latencies 13993 and issue rates into account when deciding whether a loop 13994 should be vectorized using SVE, vectorized using Advanced 13995 SIMD, or not vectorized at all. If this parameter is set to 13996 N, GCC will not use this heuristic for loops that are known to 13997 execute in fewer than N Advanced SIMD iterations. 13998 13999 'aarch64-vect-unroll-limit' 14000 The vectorizer will use available tuning information to 14001 determine whether it would be beneficial to unroll the main 14002 vectorized loop and by how much. This parameter set's the 14003 upper bound of how much the vectorizer will unroll the main 14004 loop. The default value is four. 14005 14006 The following choices of NAME are available on i386 and x86_64 14007 targets: 14008 14009 'x86-stlf-window-ninsns' 14010 Instructions number above which STFL stall penalty can be 14011 compensated. 14012 14013 14014File: gcc.info, Node: Instrumentation Options, Next: Preprocessor Options, Prev: Optimize Options, Up: Invoking GCC 14015 140163.12 Program Instrumentation Options 14017==================================== 14018 14019GCC supports a number of command-line options that control adding 14020run-time instrumentation to the code it normally generates. For 14021example, one purpose of instrumentation is collect profiling statistics 14022for use in finding program hot spots, code coverage analysis, or 14023profile-guided optimizations. Another class of program instrumentation 14024is adding run-time checking to detect programming errors like invalid 14025pointer dereferences or out-of-bounds array accesses, as well as 14026deliberately hostile attacks such as stack smashing or C++ vtable 14027hijacking. There is also a general hook which can be used to implement 14028other forms of tracing or function-level instrumentation for debug or 14029program analysis purposes. 14030 14031'-p' 14032'-pg' 14033 Generate extra code to write profile information suitable for the 14034 analysis program 'prof' (for '-p') or 'gprof' (for '-pg'). You 14035 must use this option when compiling the source files you want data 14036 about, and you must also use it when linking. 14037 14038 You can use the function attribute 'no_instrument_function' to 14039 suppress profiling of individual functions when compiling with 14040 these options. *Note Common Function Attributes::. 14041 14042'-fprofile-arcs' 14043 Add code so that program flow "arcs" are instrumented. During 14044 execution the program records how many times each branch and call 14045 is executed and how many times it is taken or returns. On targets 14046 that support constructors with priority support, profiling properly 14047 handles constructors, destructors and C++ constructors (and 14048 destructors) of classes which are used as a type of a global 14049 variable. 14050 14051 When the compiled program exits it saves this data to a file called 14052 'AUXNAME.gcda' for each source file. The data may be used for 14053 profile-directed optimizations ('-fbranch-probabilities'), or for 14054 test coverage analysis ('-ftest-coverage'). Each object file's 14055 AUXNAME is generated from the name of the output file, if 14056 explicitly specified and it is not the final executable, otherwise 14057 it is the basename of the source file. In both cases any suffix is 14058 removed (e.g. 'foo.gcda' for input file 'dir/foo.c', or 14059 'dir/foo.gcda' for output file specified as '-o dir/foo.o'). 14060 14061 Note that if a command line directly links source files, the 14062 corresponding .GCDA files will be prefixed with the unsuffixed name 14063 of the output file. E.g. 'gcc a.c b.c -o binary' would generate 14064 'binary-a.gcda' and 'binary-b.gcda' files. 14065 14066 *Note Cross-profiling::. 14067 14068'--coverage' 14069 14070 This option is used to compile and link code instrumented for 14071 coverage analysis. The option is a synonym for '-fprofile-arcs' 14072 '-ftest-coverage' (when compiling) and '-lgcov' (when linking). 14073 See the documentation for those options for more details. 14074 14075 * Compile the source files with '-fprofile-arcs' plus 14076 optimization and code generation options. For test coverage 14077 analysis, use the additional '-ftest-coverage' option. You do 14078 not need to profile every source file in a program. 14079 14080 * Compile the source files additionally with 14081 '-fprofile-abs-path' to create absolute path names in the 14082 '.gcno' files. This allows 'gcov' to find the correct sources 14083 in projects where compilations occur with different working 14084 directories. 14085 14086 * Link your object files with '-lgcov' or '-fprofile-arcs' (the 14087 latter implies the former). 14088 14089 * Run the program on a representative workload to generate the 14090 arc profile information. This may be repeated any number of 14091 times. You can run concurrent instances of your program, and 14092 provided that the file system supports locking, the data files 14093 will be correctly updated. Unless a strict ISO C dialect 14094 option is in effect, 'fork' calls are detected and correctly 14095 handled without double counting. 14096 14097 Moreover, an object file can be recompiled multiple times and 14098 the corresponding '.gcda' file merges as long as the source 14099 file and the compiler options are unchanged. 14100 14101 * For profile-directed optimizations, compile the source files 14102 again with the same optimization and code generation options 14103 plus '-fbranch-probabilities' (*note Options that Control 14104 Optimization: Optimize Options.). 14105 14106 * For test coverage analysis, use 'gcov' to produce human 14107 readable information from the '.gcno' and '.gcda' files. 14108 Refer to the 'gcov' documentation for further information. 14109 14110 With '-fprofile-arcs', for each function of your program GCC 14111 creates a program flow graph, then finds a spanning tree for the 14112 graph. Only arcs that are not on the spanning tree have to be 14113 instrumented: the compiler adds code to count the number of times 14114 that these arcs are executed. When an arc is the only exit or only 14115 entrance to a block, the instrumentation code can be added to the 14116 block; otherwise, a new basic block must be created to hold the 14117 instrumentation code. 14118 14119'-ftest-coverage' 14120 Produce a notes file that the 'gcov' code-coverage utility (*note 14121 'gcov'--a Test Coverage Program: Gcov.) can use to show program 14122 coverage. Each source file's note file is called 'AUXNAME.gcno'. 14123 Refer to the '-fprofile-arcs' option above for a description of 14124 AUXNAME and instructions on how to generate test coverage data. 14125 Coverage data matches the source files more closely if you do not 14126 optimize. 14127 14128'-fprofile-abs-path' 14129 Automatically convert relative source file names to absolute path 14130 names in the '.gcno' files. This allows 'gcov' to find the correct 14131 sources in projects where compilations occur with different working 14132 directories. 14133 14134'-fprofile-dir=PATH' 14135 14136 Set the directory to search for the profile data files in to PATH. 14137 This option affects only the profile data generated by 14138 '-fprofile-generate', '-ftest-coverage', '-fprofile-arcs' and used 14139 by '-fprofile-use' and '-fbranch-probabilities' and its related 14140 options. Both absolute and relative paths can be used. By 14141 default, GCC uses the current directory as PATH, thus the profile 14142 data file appears in the same directory as the object file. In 14143 order to prevent the file name clashing, if the object file name is 14144 not an absolute path, we mangle the absolute path of the 14145 'SOURCENAME.gcda' file and use it as the file name of a '.gcda' 14146 file. See details about the file naming in '-fprofile-arcs'. See 14147 similar option '-fprofile-note'. 14148 14149 When an executable is run in a massive parallel environment, it is 14150 recommended to save profile to different folders. That can be done 14151 with variables in PATH that are exported during run-time: 14152 14153 '%p' 14154 process ID. 14155 14156 '%q{VAR}' 14157 value of environment variable VAR 14158 14159'-fprofile-generate' 14160'-fprofile-generate=PATH' 14161 14162 Enable options usually used for instrumenting application to 14163 produce profile useful for later recompilation with profile 14164 feedback based optimization. You must use '-fprofile-generate' 14165 both when compiling and when linking your program. 14166 14167 The following options are enabled: '-fprofile-arcs', 14168 '-fprofile-values', '-finline-functions', and '-fipa-bit-cp'. 14169 14170 If PATH is specified, GCC looks at the PATH to find the profile 14171 feedback data files. See '-fprofile-dir'. 14172 14173 To optimize the program based on the collected profile information, 14174 use '-fprofile-use'. *Note Optimize Options::, for more 14175 information. 14176 14177'-fprofile-info-section' 14178'-fprofile-info-section=NAME' 14179 14180 Register the profile information in the specified section instead 14181 of using a constructor/destructor. The section name is NAME if it 14182 is specified, otherwise the section name defaults to '.gcov_info'. 14183 A pointer to the profile information generated by '-fprofile-arcs' 14184 is placed in the specified section for each translation unit. This 14185 option disables the profile information registration through a 14186 constructor and it disables the profile information processing 14187 through a destructor. This option is not intended to be used in 14188 hosted environments such as GNU/Linux. It targets free-standing 14189 environments (for example embedded systems) with limited resources 14190 which do not support constructors/destructors or the C library file 14191 I/O. 14192 14193 The linker could collect the input sections in a continuous memory 14194 block and define start and end symbols. A GNU linker script 14195 example which defines a linker output section follows: 14196 14197 .gcov_info : 14198 { 14199 PROVIDE (__gcov_info_start = .); 14200 KEEP (*(.gcov_info)) 14201 PROVIDE (__gcov_info_end = .); 14202 } 14203 14204 The program could dump the profiling information registered in this 14205 linker set for example like this: 14206 14207 #include <gcov.h> 14208 #include <stdio.h> 14209 #include <stdlib.h> 14210 14211 extern const struct gcov_info *__gcov_info_start[]; 14212 extern const struct gcov_info *__gcov_info_end[]; 14213 14214 static void 14215 filename (const char *f, void *arg) 14216 { 14217 puts (f); 14218 } 14219 14220 static void 14221 dump (const void *d, unsigned n, void *arg) 14222 { 14223 const unsigned char *c = d; 14224 14225 for (unsigned i = 0; i < n; ++i) 14226 printf ("%02x", c[i]); 14227 } 14228 14229 static void * 14230 allocate (unsigned length, void *arg) 14231 { 14232 return malloc (length); 14233 } 14234 14235 static void 14236 dump_gcov_info (void) 14237 { 14238 const struct gcov_info **info = __gcov_info_start; 14239 const struct gcov_info **end = __gcov_info_end; 14240 14241 /* Obfuscate variable to prevent compiler optimizations. */ 14242 __asm__ ("" : "+r" (info)); 14243 14244 while (info != end) 14245 { 14246 void *arg = NULL; 14247 __gcov_info_to_gcda (*info, filename, dump, allocate, arg); 14248 putchar ('\n'); 14249 ++info; 14250 } 14251 } 14252 14253 int 14254 main() 14255 { 14256 dump_gcov_info(); 14257 return 0; 14258 } 14259 14260'-fprofile-note=PATH' 14261 14262 If PATH is specified, GCC saves '.gcno' file into PATH location. 14263 If you combine the option with multiple source files, the '.gcno' 14264 file will be overwritten. 14265 14266'-fprofile-prefix-path=PATH' 14267 14268 This option can be used in combination with 14269 'profile-generate='PROFILE_DIR and 'profile-use='PROFILE_DIR to 14270 inform GCC where is the base directory of built source tree. By 14271 default PROFILE_DIR will contain files with mangled absolute paths 14272 of all object files in the built project. This is not desirable 14273 when directory used to build the instrumented binary differs from 14274 the directory used to build the binary optimized with profile 14275 feedback because the profile data will not be found during the 14276 optimized build. In such setups '-fprofile-prefix-path='PATH with 14277 PATH pointing to the base directory of the build can be used to 14278 strip the irrelevant part of the path and keep all file names 14279 relative to the main build directory. 14280 14281'-fprofile-prefix-map=OLD=NEW' 14282 When compiling files residing in directory 'OLD', record profiling 14283 information (with '--coverage') describing them as if the files 14284 resided in directory 'NEW' instead. See also '-ffile-prefix-map'. 14285 14286'-fprofile-update=METHOD' 14287 14288 Alter the update method for an application instrumented for profile 14289 feedback based optimization. The METHOD argument should be one of 14290 'single', 'atomic' or 'prefer-atomic'. The first one is useful for 14291 single-threaded applications, while the second one prevents profile 14292 corruption by emitting thread-safe code. 14293 14294 *Warning:* When an application does not properly join all threads 14295 (or creates an detached thread), a profile file can be still 14296 corrupted. 14297 14298 Using 'prefer-atomic' would be transformed either to 'atomic', when 14299 supported by a target, or to 'single' otherwise. The GCC driver 14300 automatically selects 'prefer-atomic' when '-pthread' is present in 14301 the command line. 14302 14303'-fprofile-filter-files=REGEX' 14304 14305 Instrument only functions from files whose name matches any of the 14306 regular expressions (separated by semi-colons). 14307 14308 For example, '-fprofile-filter-files=main\.c;module.*\.c' will 14309 instrument only 'main.c' and all C files starting with 'module'. 14310 14311'-fprofile-exclude-files=REGEX' 14312 14313 Instrument only functions from files whose name does not match any 14314 of the regular expressions (separated by semi-colons). 14315 14316 For example, '-fprofile-exclude-files=/usr/.*' will prevent 14317 instrumentation of all files that are located in the '/usr/' 14318 folder. 14319 14320'-fprofile-reproducible=[multithreaded|parallel-runs|serial]' 14321 Control level of reproducibility of profile gathered by 14322 '-fprofile-generate'. This makes it possible to rebuild program 14323 with same outcome which is useful, for example, for distribution 14324 packages. 14325 14326 With '-fprofile-reproducible=serial' the profile gathered by 14327 '-fprofile-generate' is reproducible provided the trained program 14328 behaves the same at each invocation of the train run, it is not 14329 multi-threaded and profile data streaming is always done in the 14330 same order. Note that profile streaming happens at the end of 14331 program run but also before 'fork' function is invoked. 14332 14333 Note that it is quite common that execution counts of some part of 14334 programs depends, for example, on length of temporary file names or 14335 memory space randomization (that may affect hash-table collision 14336 rate). Such non-reproducible part of programs may be annotated by 14337 'no_instrument_function' function attribute. 'gcov-dump' with '-l' 14338 can be used to dump gathered data and verify that they are indeed 14339 reproducible. 14340 14341 With '-fprofile-reproducible=parallel-runs' collected profile stays 14342 reproducible regardless the order of streaming of the data into 14343 gcda files. This setting makes it possible to run multiple 14344 instances of instrumented program in parallel (such as with 'make 14345 -j'). This reduces quality of gathered data, in particular of 14346 indirect call profiling. 14347 14348'-fsanitize=address' 14349 Enable AddressSanitizer, a fast memory error detector. Memory 14350 access instructions are instrumented to detect out-of-bounds and 14351 use-after-free bugs. The option enables 14352 '-fsanitize-address-use-after-scope'. See 14353 <https://github.com/google/sanitizers/wiki/AddressSanitizer> for 14354 more details. The run-time behavior can be influenced using the 14355 'ASAN_OPTIONS' environment variable. When set to 'help=1', the 14356 available options are shown at startup of the instrumented program. 14357 See 14358 <https://github.com/google/sanitizers/wiki/AddressSanitizerFlags#run-time-flags> 14359 for a list of supported options. The option cannot be combined 14360 with '-fsanitize=thread' or '-fsanitize=hwaddress'. Note that the 14361 only target '-fsanitize=hwaddress' is currently supported on is 14362 AArch64. 14363 14364'-fsanitize=kernel-address' 14365 Enable AddressSanitizer for Linux kernel. See 14366 <https://github.com/google/kasan> for more details. 14367 14368'-fsanitize=hwaddress' 14369 Enable Hardware-assisted AddressSanitizer, which uses a hardware 14370 ability to ignore the top byte of a pointer to allow the detection 14371 of memory errors with a low memory overhead. Memory access 14372 instructions are instrumented to detect out-of-bounds and 14373 use-after-free bugs. The option enables 14374 '-fsanitize-address-use-after-scope'. See 14375 <https://clang.llvm.org/docs/HardwareAssistedAddressSanitizerDesign.html> 14376 for more details. The run-time behavior can be influenced using 14377 the 'HWASAN_OPTIONS' environment variable. When set to 'help=1', 14378 the available options are shown at startup of the instrumented 14379 program. The option cannot be combined with '-fsanitize=thread' or 14380 '-fsanitize=address', and is currently only available on AArch64. 14381 14382'-fsanitize=kernel-hwaddress' 14383 Enable Hardware-assisted AddressSanitizer for compilation of the 14384 Linux kernel. Similar to '-fsanitize=kernel-address' but using an 14385 alternate instrumentation method, and similar to 14386 '-fsanitize=hwaddress' but with instrumentation differences 14387 necessary for compiling the Linux kernel. These differences are to 14388 avoid hwasan library initialization calls and to account for the 14389 stack pointer having a different value in its top byte. 14390 14391 _Note:_ This option has different defaults to the 14392 '-fsanitize=hwaddress'. Instrumenting the stack and alloca calls 14393 are not on by default but are still possible by specifying the 14394 command-line options '--param hwasan-instrument-stack=1' and 14395 '--param hwasan-instrument-allocas=1' respectively. Using a random 14396 frame tag is not implemented for kernel instrumentation. 14397 14398'-fsanitize=pointer-compare' 14399 Instrument comparison operation (<, <=, >, >=) with pointer 14400 operands. The option must be combined with either 14401 '-fsanitize=kernel-address' or '-fsanitize=address' The option 14402 cannot be combined with '-fsanitize=thread'. Note: By default the 14403 check is disabled at run time. To enable it, add 14404 'detect_invalid_pointer_pairs=2' to the environment variable 14405 'ASAN_OPTIONS'. Using 'detect_invalid_pointer_pairs=1' detects 14406 invalid operation only when both pointers are non-null. 14407 14408'-fsanitize=pointer-subtract' 14409 Instrument subtraction with pointer operands. The option must be 14410 combined with either '-fsanitize=kernel-address' or 14411 '-fsanitize=address' The option cannot be combined with 14412 '-fsanitize=thread'. Note: By default the check is disabled at run 14413 time. To enable it, add 'detect_invalid_pointer_pairs=2' to the 14414 environment variable 'ASAN_OPTIONS'. Using 14415 'detect_invalid_pointer_pairs=1' detects invalid operation only 14416 when both pointers are non-null. 14417 14418'-fsanitize=shadow-call-stack' 14419 Enable ShadowCallStack, a security enhancement mechanism used to 14420 protect programs against return address overwrites (e.g. stack 14421 buffer overflows.) It works by saving a function's return address 14422 to a separately allocated shadow call stack in the function 14423 prologue and restoring the return address from the shadow call 14424 stack in the function epilogue. Instrumentation only occurs in 14425 functions that need to save the return address to the stack. 14426 14427 Currently it only supports the aarch64 platform. It is 14428 specifically designed for linux kernels that enable the 14429 CONFIG_SHADOW_CALL_STACK option. For the user space programs, 14430 runtime support is not currently provided in libc and libgcc. 14431 Users who want to use this feature in user space need to provide 14432 their own support for the runtime. It should be noted that this 14433 may cause the ABI rules to be broken. 14434 14435 On aarch64, the instrumentation makes use of the platform register 14436 'x18'. This generally means that any code that may run on the same 14437 thread as code compiled with ShadowCallStack must be compiled with 14438 the flag '-ffixed-x18', otherwise functions compiled without 14439 '-ffixed-x18' might clobber 'x18' and so corrupt the shadow stack 14440 pointer. 14441 14442 Also, because there is no userspace runtime support, code compiled 14443 with ShadowCallStack cannot use exception handling. Use 14444 '-fno-exceptions' to turn off exceptions. 14445 14446 See <https://clang.llvm.org/docs/ShadowCallStack.html> for more 14447 details. 14448 14449'-fsanitize=thread' 14450 Enable ThreadSanitizer, a fast data race detector. Memory access 14451 instructions are instrumented to detect data race bugs. See 14452 <https://github.com/google/sanitizers/wiki#threadsanitizer> for 14453 more details. The run-time behavior can be influenced using the 14454 'TSAN_OPTIONS' environment variable; see 14455 <https://github.com/google/sanitizers/wiki/ThreadSanitizerFlags> 14456 for a list of supported options. The option cannot be combined 14457 with '-fsanitize=address', '-fsanitize=leak'. 14458 14459 Note that sanitized atomic builtins cannot throw exceptions when 14460 operating on invalid memory addresses with non-call exceptions 14461 ('-fnon-call-exceptions'). 14462 14463'-fsanitize=leak' 14464 Enable LeakSanitizer, a memory leak detector. This option only 14465 matters for linking of executables and the executable is linked 14466 against a library that overrides 'malloc' and other allocator 14467 functions. See 14468 <https://github.com/google/sanitizers/wiki/AddressSanitizerLeakSanitizer> 14469 for more details. The run-time behavior can be influenced using 14470 the 'LSAN_OPTIONS' environment variable. The option cannot be 14471 combined with '-fsanitize=thread'. 14472 14473'-fsanitize=undefined' 14474 Enable UndefinedBehaviorSanitizer, a fast undefined behavior 14475 detector. Various computations are instrumented to detect 14476 undefined behavior at runtime. See 14477 <https://clang.llvm.org/docs/UndefinedBehaviorSanitizer.html> for 14478 more details. The run-time behavior can be influenced using the 14479 'UBSAN_OPTIONS' environment variable. Current suboptions are: 14480 14481 '-fsanitize=shift' 14482 This option enables checking that the result of a shift 14483 operation is not undefined. Note that what exactly is 14484 considered undefined differs slightly between C and C++, as 14485 well as between ISO C90 and C99, etc. This option has two 14486 suboptions, '-fsanitize=shift-base' and 14487 '-fsanitize=shift-exponent'. 14488 14489 '-fsanitize=shift-exponent' 14490 This option enables checking that the second argument of a 14491 shift operation is not negative and is smaller than the 14492 precision of the promoted first argument. 14493 14494 '-fsanitize=shift-base' 14495 If the second argument of a shift operation is within range, 14496 check that the result of a shift operation is not undefined. 14497 Note that what exactly is considered undefined differs 14498 slightly between C and C++, as well as between ISO C90 and 14499 C99, etc. 14500 14501 '-fsanitize=integer-divide-by-zero' 14502 Detect integer division by zero. 14503 14504 '-fsanitize=unreachable' 14505 With this option, the compiler turns the 14506 '__builtin_unreachable' call into a diagnostics message call 14507 instead. When reaching the '__builtin_unreachable' call, the 14508 behavior is undefined. 14509 14510 '-fsanitize=vla-bound' 14511 This option instructs the compiler to check that the size of a 14512 variable length array is positive. 14513 14514 '-fsanitize=null' 14515 This option enables pointer checking. Particularly, the 14516 application built with this option turned on will issue an 14517 error message when it tries to dereference a NULL pointer, or 14518 if a reference (possibly an rvalue reference) is bound to a 14519 NULL pointer, or if a method is invoked on an object pointed 14520 by a NULL pointer. 14521 14522 '-fsanitize=return' 14523 This option enables return statement checking. Programs built 14524 with this option turned on will issue an error message when 14525 the end of a non-void function is reached without actually 14526 returning a value. This option works in C++ only. 14527 14528 '-fsanitize=signed-integer-overflow' 14529 This option enables signed integer overflow checking. We 14530 check that the result of '+', '*', and both unary and binary 14531 '-' does not overflow in the signed arithmetics. This also 14532 detects 'INT_MIN / -1' signed division. Note, integer 14533 promotion rules must be taken into account. That is, the 14534 following is not an overflow: 14535 signed char a = SCHAR_MAX; 14536 a++; 14537 14538 '-fsanitize=bounds' 14539 This option enables instrumentation of array bounds. Various 14540 out of bounds accesses are detected. Flexible array members, 14541 flexible array member-like arrays, and initializers of 14542 variables with static storage are not instrumented. 14543 14544 '-fsanitize=bounds-strict' 14545 This option enables strict instrumentation of array bounds. 14546 Most out of bounds accesses are detected, including flexible 14547 array members and flexible array member-like arrays. 14548 Initializers of variables with static storage are not 14549 instrumented. 14550 14551 '-fsanitize=alignment' 14552 14553 This option enables checking of alignment of pointers when 14554 they are dereferenced, or when a reference is bound to 14555 insufficiently aligned target, or when a method or constructor 14556 is invoked on insufficiently aligned object. 14557 14558 '-fsanitize=object-size' 14559 This option enables instrumentation of memory references using 14560 the '__builtin_object_size' function. Various out of bounds 14561 pointer accesses are detected. 14562 14563 '-fsanitize=float-divide-by-zero' 14564 Detect floating-point division by zero. Unlike other similar 14565 options, '-fsanitize=float-divide-by-zero' is not enabled by 14566 '-fsanitize=undefined', since floating-point division by zero 14567 can be a legitimate way of obtaining infinities and NaNs. 14568 14569 '-fsanitize=float-cast-overflow' 14570 This option enables floating-point type to integer conversion 14571 checking. We check that the result of the conversion does not 14572 overflow. Unlike other similar options, 14573 '-fsanitize=float-cast-overflow' is not enabled by 14574 '-fsanitize=undefined'. This option does not work well with 14575 'FE_INVALID' exceptions enabled. 14576 14577 '-fsanitize=nonnull-attribute' 14578 14579 This option enables instrumentation of calls, checking whether 14580 null values are not passed to arguments marked as requiring a 14581 non-null value by the 'nonnull' function attribute. 14582 14583 '-fsanitize=returns-nonnull-attribute' 14584 14585 This option enables instrumentation of return statements in 14586 functions marked with 'returns_nonnull' function attribute, to 14587 detect returning of null values from such functions. 14588 14589 '-fsanitize=bool' 14590 14591 This option enables instrumentation of loads from bool. If a 14592 value other than 0/1 is loaded, a run-time error is issued. 14593 14594 '-fsanitize=enum' 14595 14596 This option enables instrumentation of loads from an enum 14597 type. If a value outside the range of values for the enum 14598 type is loaded, a run-time error is issued. 14599 14600 '-fsanitize=vptr' 14601 14602 This option enables instrumentation of C++ member function 14603 calls, member accesses and some conversions between pointers 14604 to base and derived classes, to verify the referenced object 14605 has the correct dynamic type. 14606 14607 '-fsanitize=pointer-overflow' 14608 14609 This option enables instrumentation of pointer arithmetics. 14610 If the pointer arithmetics overflows, a run-time error is 14611 issued. 14612 14613 '-fsanitize=builtin' 14614 14615 This option enables instrumentation of arguments to selected 14616 builtin functions. If an invalid value is passed to such 14617 arguments, a run-time error is issued. E.g. passing 0 as the 14618 argument to '__builtin_ctz' or '__builtin_clz' invokes 14619 undefined behavior and is diagnosed by this option. 14620 14621 While '-ftrapv' causes traps for signed overflows to be emitted, 14622 '-fsanitize=undefined' gives a diagnostic message. This currently 14623 works only for the C family of languages. 14624 14625'-fno-sanitize=all' 14626 14627 This option disables all previously enabled sanitizers. 14628 '-fsanitize=all' is not allowed, as some sanitizers cannot be used 14629 together. 14630 14631'-fasan-shadow-offset=NUMBER' 14632 This option forces GCC to use custom shadow offset in 14633 AddressSanitizer checks. It is useful for experimenting with 14634 different shadow memory layouts in Kernel AddressSanitizer. 14635 14636'-fsanitize-sections=S1,S2,...' 14637 Sanitize global variables in selected user-defined sections. SI 14638 may contain wildcards. 14639 14640'-fsanitize-recover[=OPTS]' 14641 '-fsanitize-recover=' controls error recovery mode for sanitizers 14642 mentioned in comma-separated list of OPTS. Enabling this option 14643 for a sanitizer component causes it to attempt to continue running 14644 the program as if no error happened. This means multiple runtime 14645 errors can be reported in a single program run, and the exit code 14646 of the program may indicate success even when errors have been 14647 reported. The '-fno-sanitize-recover=' option can be used to alter 14648 this behavior: only the first detected error is reported and 14649 program then exits with a non-zero exit code. 14650 14651 Currently this feature only works for '-fsanitize=undefined' (and 14652 its suboptions except for '-fsanitize=unreachable' and 14653 '-fsanitize=return'), '-fsanitize=float-cast-overflow', 14654 '-fsanitize=float-divide-by-zero', '-fsanitize=bounds-strict', 14655 '-fsanitize=kernel-address' and '-fsanitize=address'. For these 14656 sanitizers error recovery is turned on by default, except 14657 '-fsanitize=address', for which this feature is experimental. 14658 '-fsanitize-recover=all' and '-fno-sanitize-recover=all' is also 14659 accepted, the former enables recovery for all sanitizers that 14660 support it, the latter disables recovery for all sanitizers that 14661 support it. 14662 14663 Even if a recovery mode is turned on the compiler side, it needs to 14664 be also enabled on the runtime library side, otherwise the failures 14665 are still fatal. The runtime library defaults to 'halt_on_error=0' 14666 for ThreadSanitizer and UndefinedBehaviorSanitizer, while default 14667 value for AddressSanitizer is 'halt_on_error=1'. This can be 14668 overridden through setting the 'halt_on_error' flag in the 14669 corresponding environment variable. 14670 14671 Syntax without an explicit OPTS parameter is deprecated. It is 14672 equivalent to specifying an OPTS list of: 14673 14674 undefined,float-cast-overflow,float-divide-by-zero,bounds-strict 14675 14676'-fsanitize-address-use-after-scope' 14677 Enable sanitization of local variables to detect use-after-scope 14678 bugs. The option sets '-fstack-reuse' to 'none'. 14679 14680'-fsanitize-undefined-trap-on-error' 14681 The '-fsanitize-undefined-trap-on-error' option instructs the 14682 compiler to report undefined behavior using '__builtin_trap' rather 14683 than a 'libubsan' library routine. The advantage of this is that 14684 the 'libubsan' library is not needed and is not linked in, so this 14685 is usable even in freestanding environments. 14686 14687'-fsanitize-coverage=trace-pc' 14688 Enable coverage-guided fuzzing code instrumentation. Inserts a 14689 call to '__sanitizer_cov_trace_pc' into every basic block. 14690 14691'-fsanitize-coverage=trace-cmp' 14692 Enable dataflow guided fuzzing code instrumentation. Inserts a 14693 call to '__sanitizer_cov_trace_cmp1', '__sanitizer_cov_trace_cmp2', 14694 '__sanitizer_cov_trace_cmp4' or '__sanitizer_cov_trace_cmp8' for 14695 integral comparison with both operands variable or 14696 '__sanitizer_cov_trace_const_cmp1', 14697 '__sanitizer_cov_trace_const_cmp2', 14698 '__sanitizer_cov_trace_const_cmp4' or 14699 '__sanitizer_cov_trace_const_cmp8' for integral comparison with one 14700 operand constant, '__sanitizer_cov_trace_cmpf' or 14701 '__sanitizer_cov_trace_cmpd' for float or double comparisons and 14702 '__sanitizer_cov_trace_switch' for switch statements. 14703 14704'-fcf-protection=[full|branch|return|none|check]' 14705 Enable code instrumentation of control-flow transfers to increase 14706 program security by checking that target addresses of control-flow 14707 transfer instructions (such as indirect function call, function 14708 return, indirect jump) are valid. This prevents diverting the flow 14709 of control to an unexpected target. This is intended to protect 14710 against such threats as Return-oriented Programming (ROP), and 14711 similarly call/jmp-oriented programming (COP/JOP). 14712 14713 The value 'branch' tells the compiler to implement checking of 14714 validity of control-flow transfer at the point of indirect branch 14715 instructions, i.e. call/jmp instructions. The value 'return' 14716 implements checking of validity at the point of returning from a 14717 function. The value 'full' is an alias for specifying both 14718 'branch' and 'return'. The value 'none' turns off instrumentation. 14719 14720 The value 'check' is used for the final link with link-time 14721 optimization (LTO). An error is issued if LTO object files are 14722 compiled with different '-fcf-protection' values. The value 14723 'check' is ignored at the compile time. 14724 14725 The macro '__CET__' is defined when '-fcf-protection' is used. The 14726 first bit of '__CET__' is set to 1 for the value 'branch' and the 14727 second bit of '__CET__' is set to 1 for the 'return'. 14728 14729 You can also use the 'nocf_check' attribute to identify which 14730 functions and calls should be skipped from instrumentation (*note 14731 Function Attributes::). 14732 14733 Currently the x86 GNU/Linux target provides an implementation based 14734 on Intel Control-flow Enforcement Technology (CET) which works for 14735 i686 processor or newer. 14736 14737'-fharden-compares' 14738 For every logical test that survives gimple optimizations and is 14739 _not_ the condition in a conditional branch (for example, 14740 conditions tested for conditional moves, or to store in boolean 14741 variables), emit extra code to compute and verify the reversed 14742 condition, and to call '__builtin_trap' if the results do not 14743 match. Use with '-fharden-conditional-branches' to cover all 14744 conditionals. 14745 14746'-fharden-conditional-branches' 14747 For every non-vectorized conditional branch that survives gimple 14748 optimizations, emit extra code to compute and verify the reversed 14749 condition, and to call '__builtin_trap' if the result is 14750 unexpected. Use with '-fharden-compares' to cover all 14751 conditionals. 14752 14753'-fstack-protector' 14754 Emit extra code to check for buffer overflows, such as stack 14755 smashing attacks. This is done by adding a guard variable to 14756 functions with vulnerable objects. This includes functions that 14757 call 'alloca', and functions with buffers larger than or equal to 8 14758 bytes. The guards are initialized when a function is entered and 14759 then checked when the function exits. If a guard check fails, an 14760 error message is printed and the program exits. Only variables 14761 that are actually allocated on the stack are considered, optimized 14762 away variables or variables allocated in registers don't count. 14763 14764'-fstack-protector-all' 14765 Like '-fstack-protector' except that all functions are protected. 14766 14767'-fstack-protector-strong' 14768 Like '-fstack-protector' but includes additional functions to be 14769 protected -- those that have local array definitions, or have 14770 references to local frame addresses. Only variables that are 14771 actually allocated on the stack are considered, optimized away 14772 variables or variables allocated in registers don't count. 14773 14774'-fstack-protector-explicit' 14775 Like '-fstack-protector' but only protects those functions which 14776 have the 'stack_protect' attribute. 14777 14778'-fstack-check' 14779 Generate code to verify that you do not go beyond the boundary of 14780 the stack. You should specify this flag if you are running in an 14781 environment with multiple threads, but you only rarely need to 14782 specify it in a single-threaded environment since stack overflow is 14783 automatically detected on nearly all systems if there is only one 14784 stack. 14785 14786 Note that this switch does not actually cause checking to be done; 14787 the operating system or the language runtime must do that. The 14788 switch causes generation of code to ensure that they see the stack 14789 being extended. 14790 14791 You can additionally specify a string parameter: 'no' means no 14792 checking, 'generic' means force the use of old-style checking, 14793 'specific' means use the best checking method and is equivalent to 14794 bare '-fstack-check'. 14795 14796 Old-style checking is a generic mechanism that requires no specific 14797 target support in the compiler but comes with the following 14798 drawbacks: 14799 14800 1. Modified allocation strategy for large objects: they are 14801 always allocated dynamically if their size exceeds a fixed 14802 threshold. Note this may change the semantics of some code. 14803 14804 2. Fixed limit on the size of the static frame of functions: when 14805 it is topped by a particular function, stack checking is not 14806 reliable and a warning is issued by the compiler. 14807 14808 3. Inefficiency: because of both the modified allocation strategy 14809 and the generic implementation, code performance is hampered. 14810 14811 Note that old-style stack checking is also the fallback method for 14812 'specific' if no target support has been added in the compiler. 14813 14814 '-fstack-check=' is designed for Ada's needs to detect infinite 14815 recursion and stack overflows. 'specific' is an excellent choice 14816 when compiling Ada code. It is not generally sufficient to protect 14817 against stack-clash attacks. To protect against those you want 14818 '-fstack-clash-protection'. 14819 14820'-fstack-clash-protection' 14821 Generate code to prevent stack clash style attacks. When this 14822 option is enabled, the compiler will only allocate one page of 14823 stack space at a time and each page is accessed immediately after 14824 allocation. Thus, it prevents allocations from jumping over any 14825 stack guard page provided by the operating system. 14826 14827 Most targets do not fully support stack clash protection. However, 14828 on those targets '-fstack-clash-protection' will protect dynamic 14829 stack allocations. '-fstack-clash-protection' may also provide 14830 limited protection for static stack allocations if the target 14831 supports '-fstack-check=specific'. 14832 14833'-fstack-limit-register=REG' 14834'-fstack-limit-symbol=SYM' 14835'-fno-stack-limit' 14836 Generate code to ensure that the stack does not grow beyond a 14837 certain value, either the value of a register or the address of a 14838 symbol. If a larger stack is required, a signal is raised at run 14839 time. For most targets, the signal is raised before the stack 14840 overruns the boundary, so it is possible to catch the signal 14841 without taking special precautions. 14842 14843 For instance, if the stack starts at absolute address '0x80000000' 14844 and grows downwards, you can use the flags 14845 '-fstack-limit-symbol=__stack_limit' and 14846 '-Wl,--defsym,__stack_limit=0x7ffe0000' to enforce a stack limit of 14847 128KB. Note that this may only work with the GNU linker. 14848 14849 You can locally override stack limit checking by using the 14850 'no_stack_limit' function attribute (*note Function Attributes::). 14851 14852'-fsplit-stack' 14853 Generate code to automatically split the stack before it overflows. 14854 The resulting program has a discontiguous stack which can only 14855 overflow if the program is unable to allocate any more memory. 14856 This is most useful when running threaded programs, as it is no 14857 longer necessary to calculate a good stack size to use for each 14858 thread. This is currently only implemented for the x86 targets 14859 running GNU/Linux. 14860 14861 When code compiled with '-fsplit-stack' calls code compiled without 14862 '-fsplit-stack', there may not be much stack space available for 14863 the latter code to run. If compiling all code, including library 14864 code, with '-fsplit-stack' is not an option, then the linker can 14865 fix up these calls so that the code compiled without 14866 '-fsplit-stack' always has a large stack. Support for this is 14867 implemented in the gold linker in GNU binutils release 2.21 and 14868 later. 14869 14870'-fvtable-verify=[std|preinit|none]' 14871 This option is only available when compiling C++ code. It turns on 14872 (or off, if using '-fvtable-verify=none') the security feature that 14873 verifies at run time, for every virtual call, that the vtable 14874 pointer through which the call is made is valid for the type of the 14875 object, and has not been corrupted or overwritten. If an invalid 14876 vtable pointer is detected at run time, an error is reported and 14877 execution of the program is immediately halted. 14878 14879 This option causes run-time data structures to be built at program 14880 startup, which are used for verifying the vtable pointers. The 14881 options 'std' and 'preinit' control the timing of when these data 14882 structures are built. In both cases the data structures are built 14883 before execution reaches 'main'. Using '-fvtable-verify=std' 14884 causes the data structures to be built after shared libraries have 14885 been loaded and initialized. '-fvtable-verify=preinit' causes them 14886 to be built before shared libraries have been loaded and 14887 initialized. 14888 14889 If this option appears multiple times in the command line with 14890 different values specified, 'none' takes highest priority over both 14891 'std' and 'preinit'; 'preinit' takes priority over 'std'. 14892 14893'-fvtv-debug' 14894 When used in conjunction with '-fvtable-verify=std' or 14895 '-fvtable-verify=preinit', causes debug versions of the runtime 14896 functions for the vtable verification feature to be called. This 14897 flag also causes the compiler to log information about which vtable 14898 pointers it finds for each class. This information is written to a 14899 file named 'vtv_set_ptr_data.log' in the directory named by the 14900 environment variable 'VTV_LOGS_DIR' if that is defined or the 14901 current working directory otherwise. 14902 14903 Note: This feature _appends_ data to the log file. If you want a 14904 fresh log file, be sure to delete any existing one. 14905 14906'-fvtv-counts' 14907 This is a debugging flag. When used in conjunction with 14908 '-fvtable-verify=std' or '-fvtable-verify=preinit', this causes the 14909 compiler to keep track of the total number of virtual calls it 14910 encounters and the number of verifications it inserts. It also 14911 counts the number of calls to certain run-time library functions 14912 that it inserts and logs this information for each compilation 14913 unit. The compiler writes this information to a file named 14914 'vtv_count_data.log' in the directory named by the environment 14915 variable 'VTV_LOGS_DIR' if that is defined or the current working 14916 directory otherwise. It also counts the size of the vtable pointer 14917 sets for each class, and writes this information to 14918 'vtv_class_set_sizes.log' in the same directory. 14919 14920 Note: This feature _appends_ data to the log files. To get fresh 14921 log files, be sure to delete any existing ones. 14922 14923'-finstrument-functions' 14924 Generate instrumentation calls for entry and exit to functions. 14925 Just after function entry and just before function exit, the 14926 following profiling functions are called with the address of the 14927 current function and its call site. (On some platforms, 14928 '__builtin_return_address' does not work beyond the current 14929 function, so the call site information may not be available to the 14930 profiling functions otherwise.) 14931 14932 void __cyg_profile_func_enter (void *this_fn, 14933 void *call_site); 14934 void __cyg_profile_func_exit (void *this_fn, 14935 void *call_site); 14936 14937 The first argument is the address of the start of the current 14938 function, which may be looked up exactly in the symbol table. 14939 14940 This instrumentation is also done for functions expanded inline in 14941 other functions. The profiling calls indicate where, conceptually, 14942 the inline function is entered and exited. This means that 14943 addressable versions of such functions must be available. If all 14944 your uses of a function are expanded inline, this may mean an 14945 additional expansion of code size. If you use 'extern inline' in 14946 your C code, an addressable version of such functions must be 14947 provided. (This is normally the case anyway, but if you get lucky 14948 and the optimizer always expands the functions inline, you might 14949 have gotten away without providing static copies.) 14950 14951 A function may be given the attribute 'no_instrument_function', in 14952 which case this instrumentation is not done. This can be used, for 14953 example, for the profiling functions listed above, high-priority 14954 interrupt routines, and any functions from which the profiling 14955 functions cannot safely be called (perhaps signal handlers, if the 14956 profiling routines generate output or allocate memory). *Note 14957 Common Function Attributes::. 14958 14959'-finstrument-functions-exclude-file-list=FILE,FILE,...' 14960 14961 Set the list of functions that are excluded from instrumentation 14962 (see the description of '-finstrument-functions'). If the file 14963 that contains a function definition matches with one of FILE, then 14964 that function is not instrumented. The match is done on 14965 substrings: if the FILE parameter is a substring of the file name, 14966 it is considered to be a match. 14967 14968 For example: 14969 14970 -finstrument-functions-exclude-file-list=/bits/stl,include/sys 14971 14972 excludes any inline function defined in files whose pathnames 14973 contain '/bits/stl' or 'include/sys'. 14974 14975 If, for some reason, you want to include letter ',' in one of SYM, 14976 write '\,'. For example, 14977 '-finstrument-functions-exclude-file-list='\,\,tmp'' (note the 14978 single quote surrounding the option). 14979 14980'-finstrument-functions-exclude-function-list=SYM,SYM,...' 14981 14982 This is similar to '-finstrument-functions-exclude-file-list', but 14983 this option sets the list of function names to be excluded from 14984 instrumentation. The function name to be matched is its 14985 user-visible name, such as 'vector<int> blah(const vector<int> &)', 14986 not the internal mangled name (e.g., '_Z4blahRSt6vectorIiSaIiEE'). 14987 The match is done on substrings: if the SYM parameter is a 14988 substring of the function name, it is considered to be a match. 14989 For C99 and C++ extended identifiers, the function name must be 14990 given in UTF-8, not using universal character names. 14991 14992'-fpatchable-function-entry=N[,M]' 14993 Generate N NOPs right at the beginning of each function, with the 14994 function entry point before the Mth NOP. If M is omitted, it 14995 defaults to '0' so the function entry points to the address just at 14996 the first NOP. The NOP instructions reserve extra space which can 14997 be used to patch in any desired instrumentation at run time, 14998 provided that the code segment is writable. The amount of space is 14999 controllable indirectly via the number of NOPs; the NOP instruction 15000 used corresponds to the instruction emitted by the internal GCC 15001 back-end interface 'gen_nop'. This behavior is target-specific and 15002 may also depend on the architecture variant and/or other 15003 compilation options. 15004 15005 For run-time identification, the starting addresses of these areas, 15006 which correspond to their respective function entries minus M, are 15007 additionally collected in the '__patchable_function_entries' 15008 section of the resulting binary. 15009 15010 Note that the value of '__attribute__ ((patchable_function_entry 15011 (N,M)))' takes precedence over command-line option 15012 '-fpatchable-function-entry=N,M'. This can be used to increase the 15013 area size or to remove it completely on a single function. If 15014 'N=0', no pad location is recorded. 15015 15016 The NOP instructions are inserted at--and maybe before, depending 15017 on M--the function entry address, even before the prologue. 15018 15019 The maximum value of N and M is 65535. 15020 15021 15022File: gcc.info, Node: Preprocessor Options, Next: Assembler Options, Prev: Instrumentation Options, Up: Invoking GCC 15023 150243.13 Options Controlling the Preprocessor 15025========================================= 15026 15027These options control the C preprocessor, which is run on each C source 15028file before actual compilation. 15029 15030 If you use the '-E' option, nothing is done except preprocessing. Some 15031of these options make sense only together with '-E' because they cause 15032the preprocessor output to be unsuitable for actual compilation. 15033 15034 In addition to the options listed here, there are a number of options 15035to control search paths for include files documented in *note Directory 15036Options::. Options to control preprocessor diagnostics are listed in 15037*note Warning Options::. 15038 15039'-D NAME' 15040 Predefine NAME as a macro, with definition '1'. 15041 15042'-D NAME=DEFINITION' 15043 The contents of DEFINITION are tokenized and processed as if they 15044 appeared during translation phase three in a '#define' directive. 15045 In particular, the definition is truncated by embedded newline 15046 characters. 15047 15048 If you are invoking the preprocessor from a shell or shell-like 15049 program you may need to use the shell's quoting syntax to protect 15050 characters such as spaces that have a meaning in the shell syntax. 15051 15052 If you wish to define a function-like macro on the command line, 15053 write its argument list with surrounding parentheses before the 15054 equals sign (if any). Parentheses are meaningful to most shells, 15055 so you should quote the option. With 'sh' and 'csh', 15056 '-D'NAME(ARGS...)=DEFINITION'' works. 15057 15058 '-D' and '-U' options are processed in the order they are given on 15059 the command line. All '-imacros FILE' and '-include FILE' options 15060 are processed after all '-D' and '-U' options. 15061 15062'-U NAME' 15063 Cancel any previous definition of NAME, either built in or provided 15064 with a '-D' option. 15065 15066'-include FILE' 15067 Process FILE as if '#include "file"' appeared as the first line of 15068 the primary source file. However, the first directory searched for 15069 FILE is the preprocessor's working directory _instead of_ the 15070 directory containing the main source file. If not found there, it 15071 is searched for in the remainder of the '#include "..."' search 15072 chain as normal. 15073 15074 If multiple '-include' options are given, the files are included in 15075 the order they appear on the command line. 15076 15077'-imacros FILE' 15078 Exactly like '-include', except that any output produced by 15079 scanning FILE is thrown away. Macros it defines remain defined. 15080 This allows you to acquire all the macros from a header without 15081 also processing its declarations. 15082 15083 All files specified by '-imacros' are processed before all files 15084 specified by '-include'. 15085 15086'-undef' 15087 Do not predefine any system-specific or GCC-specific macros. The 15088 standard predefined macros remain defined. 15089 15090'-pthread' 15091 Define additional macros required for using the POSIX threads 15092 library. You should use this option consistently for both 15093 compilation and linking. This option is supported on GNU/Linux 15094 targets, most other Unix derivatives, and also on x86 Cygwin and 15095 MinGW targets. 15096 15097'-M' 15098 Instead of outputting the result of preprocessing, output a rule 15099 suitable for 'make' describing the dependencies of the main source 15100 file. The preprocessor outputs one 'make' rule containing the 15101 object file name for that source file, a colon, and the names of 15102 all the included files, including those coming from '-include' or 15103 '-imacros' command-line options. 15104 15105 Unless specified explicitly (with '-MT' or '-MQ'), the object file 15106 name consists of the name of the source file with any suffix 15107 replaced with object file suffix and with any leading directory 15108 parts removed. If there are many included files then the rule is 15109 split into several lines using '\'-newline. The rule has no 15110 commands. 15111 15112 This option does not suppress the preprocessor's debug output, such 15113 as '-dM'. To avoid mixing such debug output with the dependency 15114 rules you should explicitly specify the dependency output file with 15115 '-MF', or use an environment variable like 'DEPENDENCIES_OUTPUT' 15116 (*note Environment Variables::). Debug output is still sent to the 15117 regular output stream as normal. 15118 15119 Passing '-M' to the driver implies '-E', and suppresses warnings 15120 with an implicit '-w'. 15121 15122'-MM' 15123 Like '-M' but do not mention header files that are found in system 15124 header directories, nor header files that are included, directly or 15125 indirectly, from such a header. 15126 15127 This implies that the choice of angle brackets or double quotes in 15128 an '#include' directive does not in itself determine whether that 15129 header appears in '-MM' dependency output. 15130 15131'-MF FILE' 15132 When used with '-M' or '-MM', specifies a file to write the 15133 dependencies to. If no '-MF' switch is given the preprocessor 15134 sends the rules to the same place it would send preprocessed 15135 output. 15136 15137 When used with the driver options '-MD' or '-MMD', '-MF' overrides 15138 the default dependency output file. 15139 15140 If FILE is '-', then the dependencies are written to 'stdout'. 15141 15142'-MG' 15143 In conjunction with an option such as '-M' requesting dependency 15144 generation, '-MG' assumes missing header files are generated files 15145 and adds them to the dependency list without raising an error. The 15146 dependency filename is taken directly from the '#include' directive 15147 without prepending any path. '-MG' also suppresses preprocessed 15148 output, as a missing header file renders this useless. 15149 15150 This feature is used in automatic updating of makefiles. 15151 15152'-Mno-modules' 15153 Disable dependency generation for compiled module interfaces. 15154 15155'-MP' 15156 This option instructs CPP to add a phony target for each dependency 15157 other than the main file, causing each to depend on nothing. These 15158 dummy rules work around errors 'make' gives if you remove header 15159 files without updating the 'Makefile' to match. 15160 15161 This is typical output: 15162 15163 test.o: test.c test.h 15164 15165 test.h: 15166 15167'-MT TARGET' 15168 15169 Change the target of the rule emitted by dependency generation. By 15170 default CPP takes the name of the main input file, deletes any 15171 directory components and any file suffix such as '.c', and appends 15172 the platform's usual object suffix. The result is the target. 15173 15174 An '-MT' option sets the target to be exactly the string you 15175 specify. If you want multiple targets, you can specify them as a 15176 single argument to '-MT', or use multiple '-MT' options. 15177 15178 For example, '-MT '$(objpfx)foo.o'' might give 15179 15180 $(objpfx)foo.o: foo.c 15181 15182'-MQ TARGET' 15183 15184 Same as '-MT', but it quotes any characters which are special to 15185 Make. '-MQ '$(objpfx)foo.o'' gives 15186 15187 $$(objpfx)foo.o: foo.c 15188 15189 The default target is automatically quoted, as if it were given 15190 with '-MQ'. 15191 15192'-MD' 15193 '-MD' is equivalent to '-M -MF FILE', except that '-E' is not 15194 implied. The driver determines FILE based on whether an '-o' 15195 option is given. If it is, the driver uses its argument but with a 15196 suffix of '.d', otherwise it takes the name of the input file, 15197 removes any directory components and suffix, and applies a '.d' 15198 suffix. 15199 15200 If '-MD' is used in conjunction with '-E', any '-o' switch is 15201 understood to specify the dependency output file (*note -MF: 15202 dashMF.), but if used without '-E', each '-o' is understood to 15203 specify a target object file. 15204 15205 Since '-E' is not implied, '-MD' can be used to generate a 15206 dependency output file as a side effect of the compilation process. 15207 15208'-MMD' 15209 Like '-MD' except mention only user header files, not system header 15210 files. 15211 15212'-fpreprocessed' 15213 Indicate to the preprocessor that the input file has already been 15214 preprocessed. This suppresses things like macro expansion, 15215 trigraph conversion, escaped newline splicing, and processing of 15216 most directives. The preprocessor still recognizes and removes 15217 comments, so that you can pass a file preprocessed with '-C' to the 15218 compiler without problems. In this mode the integrated 15219 preprocessor is little more than a tokenizer for the front ends. 15220 15221 '-fpreprocessed' is implicit if the input file has one of the 15222 extensions '.i', '.ii' or '.mi'. These are the extensions that GCC 15223 uses for preprocessed files created by '-save-temps'. 15224 15225'-fdirectives-only' 15226 When preprocessing, handle directives, but do not expand macros. 15227 15228 The option's behavior depends on the '-E' and '-fpreprocessed' 15229 options. 15230 15231 With '-E', preprocessing is limited to the handling of directives 15232 such as '#define', '#ifdef', and '#error'. Other preprocessor 15233 operations, such as macro expansion and trigraph conversion are not 15234 performed. In addition, the '-dD' option is implicitly enabled. 15235 15236 With '-fpreprocessed', predefinition of command line and most 15237 builtin macros is disabled. Macros such as '__LINE__', which are 15238 contextually dependent, are handled normally. This enables 15239 compilation of files previously preprocessed with '-E 15240 -fdirectives-only'. 15241 15242 With both '-E' and '-fpreprocessed', the rules for '-fpreprocessed' 15243 take precedence. This enables full preprocessing of files 15244 previously preprocessed with '-E -fdirectives-only'. 15245 15246'-fdollars-in-identifiers' 15247 Accept '$' in identifiers. 15248 15249'-fextended-identifiers' 15250 Accept universal character names and extended characters in 15251 identifiers. This option is enabled by default for C99 (and later 15252 C standard versions) and C++. 15253 15254'-fno-canonical-system-headers' 15255 When preprocessing, do not shorten system header paths with 15256 canonicalization. 15257 15258'-fmax-include-depth=DEPTH' 15259 Set the maximum depth of the nested #include. The default is 200. 15260 15261'-ftabstop=WIDTH' 15262 Set the distance between tab stops. This helps the preprocessor 15263 report correct column numbers in warnings or errors, even if tabs 15264 appear on the line. If the value is less than 1 or greater than 15265 100, the option is ignored. The default is 8. 15266 15267'-ftrack-macro-expansion[=LEVEL]' 15268 Track locations of tokens across macro expansions. This allows the 15269 compiler to emit diagnostic about the current macro expansion stack 15270 when a compilation error occurs in a macro expansion. Using this 15271 option makes the preprocessor and the compiler consume more memory. 15272 The LEVEL parameter can be used to choose the level of precision of 15273 token location tracking thus decreasing the memory consumption if 15274 necessary. Value '0' of LEVEL de-activates this option. Value '1' 15275 tracks tokens locations in a degraded mode for the sake of minimal 15276 memory overhead. In this mode all tokens resulting from the 15277 expansion of an argument of a function-like macro have the same 15278 location. Value '2' tracks tokens locations completely. This 15279 value is the most memory hungry. When this option is given no 15280 argument, the default parameter value is '2'. 15281 15282 Note that '-ftrack-macro-expansion=2' is activated by default. 15283 15284'-fmacro-prefix-map=OLD=NEW' 15285 When preprocessing files residing in directory 'OLD', expand the 15286 '__FILE__' and '__BASE_FILE__' macros as if the files resided in 15287 directory 'NEW' instead. This can be used to change an absolute 15288 path to a relative path by using '.' for NEW which can result in 15289 more reproducible builds that are location independent. This 15290 option also affects '__builtin_FILE()' during compilation. See 15291 also '-ffile-prefix-map'. 15292 15293'-fexec-charset=CHARSET' 15294 Set the execution character set, used for string and character 15295 constants. The default is UTF-8. CHARSET can be any encoding 15296 supported by the system's 'iconv' library routine. 15297 15298'-fwide-exec-charset=CHARSET' 15299 Set the wide execution character set, used for wide string and 15300 character constants. The default is one of UTF-32BE, UTF-32LE, 15301 UTF-16BE, or UTF-16LE, whichever corresponds to the width of 15302 'wchar_t' and the big-endian or little-endian byte order being used 15303 for code generation. As with '-fexec-charset', CHARSET can be any 15304 encoding supported by the system's 'iconv' library routine; 15305 however, you will have problems with encodings that do not fit 15306 exactly in 'wchar_t'. 15307 15308'-finput-charset=CHARSET' 15309 Set the input character set, used for translation from the 15310 character set of the input file to the source character set used by 15311 GCC. If the locale does not specify, or GCC cannot get this 15312 information from the locale, the default is UTF-8. This can be 15313 overridden by either the locale or this command-line option. 15314 Currently the command-line option takes precedence if there's a 15315 conflict. CHARSET can be any encoding supported by the system's 15316 'iconv' library routine. 15317 15318'-fpch-deps' 15319 When using precompiled headers (*note Precompiled Headers::), this 15320 flag causes the dependency-output flags to also list the files from 15321 the precompiled header's dependencies. If not specified, only the 15322 precompiled header are listed and not the files that were used to 15323 create it, because those files are not consulted when a precompiled 15324 header is used. 15325 15326'-fpch-preprocess' 15327 This option allows use of a precompiled header (*note Precompiled 15328 Headers::) together with '-E'. It inserts a special '#pragma', 15329 '#pragma GCC pch_preprocess "FILENAME"' in the output to mark the 15330 place where the precompiled header was found, and its FILENAME. 15331 When '-fpreprocessed' is in use, GCC recognizes this '#pragma' and 15332 loads the PCH. 15333 15334 This option is off by default, because the resulting preprocessed 15335 output is only really suitable as input to GCC. It is switched on 15336 by '-save-temps'. 15337 15338 You should not write this '#pragma' in your own code, but it is 15339 safe to edit the filename if the PCH file is available in a 15340 different location. The filename may be absolute or it may be 15341 relative to GCC's current directory. 15342 15343'-fworking-directory' 15344 Enable generation of linemarkers in the preprocessor output that 15345 let the compiler know the current working directory at the time of 15346 preprocessing. When this option is enabled, the preprocessor 15347 emits, after the initial linemarker, a second linemarker with the 15348 current working directory followed by two slashes. GCC uses this 15349 directory, when it's present in the preprocessed input, as the 15350 directory emitted as the current working directory in some 15351 debugging information formats. This option is implicitly enabled 15352 if debugging information is enabled, but this can be inhibited with 15353 the negated form '-fno-working-directory'. If the '-P' flag is 15354 present in the command line, this option has no effect, since no 15355 '#line' directives are emitted whatsoever. 15356 15357'-A PREDICATE=ANSWER' 15358 Make an assertion with the predicate PREDICATE and answer ANSWER. 15359 This form is preferred to the older form '-A PREDICATE(ANSWER)', 15360 which is still supported, because it does not use shell special 15361 characters. 15362 15363'-A -PREDICATE=ANSWER' 15364 Cancel an assertion with the predicate PREDICATE and answer ANSWER. 15365 15366'-C' 15367 Do not discard comments. All comments are passed through to the 15368 output file, except for comments in processed directives, which are 15369 deleted along with the directive. 15370 15371 You should be prepared for side effects when using '-C'; it causes 15372 the preprocessor to treat comments as tokens in their own right. 15373 For example, comments appearing at the start of what would be a 15374 directive line have the effect of turning that line into an 15375 ordinary source line, since the first token on the line is no 15376 longer a '#'. 15377 15378'-CC' 15379 Do not discard comments, including during macro expansion. This is 15380 like '-C', except that comments contained within macros are also 15381 passed through to the output file where the macro is expanded. 15382 15383 In addition to the side effects of the '-C' option, the '-CC' 15384 option causes all C++-style comments inside a macro to be converted 15385 to C-style comments. This is to prevent later use of that macro 15386 from inadvertently commenting out the remainder of the source line. 15387 15388 The '-CC' option is generally used to support lint comments. 15389 15390'-P' 15391 Inhibit generation of linemarkers in the output from the 15392 preprocessor. This might be useful when running the preprocessor 15393 on something that is not C code, and will be sent to a program 15394 which might be confused by the linemarkers. 15395 15396'-traditional' 15397'-traditional-cpp' 15398 15399 Try to imitate the behavior of pre-standard C preprocessors, as 15400 opposed to ISO C preprocessors. See the GNU CPP manual for 15401 details. 15402 15403 Note that GCC does not otherwise attempt to emulate a pre-standard 15404 C compiler, and these options are only supported with the '-E' 15405 switch, or when invoking CPP explicitly. 15406 15407'-trigraphs' 15408 Support ISO C trigraphs. These are three-character sequences, all 15409 starting with '??', that are defined by ISO C to stand for single 15410 characters. For example, '??/' stands for '\', so ''??/n'' is a 15411 character constant for a newline. 15412 15413 The nine trigraphs and their replacements are 15414 15415 Trigraph: ??( ??) ??< ??> ??= ??/ ??' ??! ??- 15416 Replacement: [ ] { } # \ ^ | ~ 15417 15418 By default, GCC ignores trigraphs, but in standard-conforming modes 15419 it converts them. See the '-std' and '-ansi' options. 15420 15421'-remap' 15422 Enable special code to work around file systems which only permit 15423 very short file names, such as MS-DOS. 15424 15425'-H' 15426 Print the name of each header file used, in addition to other 15427 normal activities. Each name is indented to show how deep in the 15428 '#include' stack it is. Precompiled header files are also printed, 15429 even if they are found to be invalid; an invalid precompiled header 15430 file is printed with '...x' and a valid one with '...!' . 15431 15432'-dLETTERS' 15433 Says to make debugging dumps during compilation as specified by 15434 LETTERS. The flags documented here are those relevant to the 15435 preprocessor. Other LETTERS are interpreted by the compiler 15436 proper, or reserved for future versions of GCC, and so are silently 15437 ignored. If you specify LETTERS whose behavior conflicts, the 15438 result is undefined. *Note Developer Options::, for more 15439 information. 15440 15441 '-dM' 15442 Instead of the normal output, generate a list of '#define' 15443 directives for all the macros defined during the execution of 15444 the preprocessor, including predefined macros. This gives you 15445 a way of finding out what is predefined in your version of the 15446 preprocessor. Assuming you have no file 'foo.h', the command 15447 15448 touch foo.h; cpp -dM foo.h 15449 15450 shows all the predefined macros. 15451 15452 If you use '-dM' without the '-E' option, '-dM' is interpreted 15453 as a synonym for '-fdump-rtl-mach'. *Note (gcc)Developer 15454 Options::. 15455 15456 '-dD' 15457 Like '-dM' except in two respects: it does _not_ include the 15458 predefined macros, and it outputs _both_ the '#define' 15459 directives and the result of preprocessing. Both kinds of 15460 output go to the standard output file. 15461 15462 '-dN' 15463 Like '-dD', but emit only the macro names, not their 15464 expansions. 15465 15466 '-dI' 15467 Output '#include' directives in addition to the result of 15468 preprocessing. 15469 15470 '-dU' 15471 Like '-dD' except that only macros that are expanded, or whose 15472 definedness is tested in preprocessor directives, are output; 15473 the output is delayed until the use or test of the macro; and 15474 '#undef' directives are also output for macros tested but 15475 undefined at the time. 15476 15477'-fdebug-cpp' 15478 This option is only useful for debugging GCC. When used from CPP or 15479 with '-E', it dumps debugging information about location maps. 15480 Every token in the output is preceded by the dump of the map its 15481 location belongs to. 15482 15483 When used from GCC without '-E', this option has no effect. 15484 15485'-Wp,OPTION' 15486 You can use '-Wp,OPTION' to bypass the compiler driver and pass 15487 OPTION directly through to the preprocessor. If OPTION contains 15488 commas, it is split into multiple options at the commas. However, 15489 many options are modified, translated or interpreted by the 15490 compiler driver before being passed to the preprocessor, and '-Wp' 15491 forcibly bypasses this phase. The preprocessor's direct interface 15492 is undocumented and subject to change, so whenever possible you 15493 should avoid using '-Wp' and let the driver handle the options 15494 instead. 15495 15496'-Xpreprocessor OPTION' 15497 Pass OPTION as an option to the preprocessor. You can use this to 15498 supply system-specific preprocessor options that GCC does not 15499 recognize. 15500 15501 If you want to pass an option that takes an argument, you must use 15502 '-Xpreprocessor' twice, once for the option and once for the 15503 argument. 15504 15505'-no-integrated-cpp' 15506 Perform preprocessing as a separate pass before compilation. By 15507 default, GCC performs preprocessing as an integrated part of input 15508 tokenization and parsing. If this option is provided, the 15509 appropriate language front end ('cc1', 'cc1plus', or 'cc1obj' for 15510 C, C++, and Objective-C, respectively) is instead invoked twice, 15511 once for preprocessing only and once for actual compilation of the 15512 preprocessed input. This option may be useful in conjunction with 15513 the '-B' or '-wrapper' options to specify an alternate preprocessor 15514 or perform additional processing of the program source between 15515 normal preprocessing and compilation. 15516 15517'-flarge-source-files' 15518 Adjust GCC to expect large source files, at the expense of slower 15519 compilation and higher memory usage. 15520 15521 Specifically, GCC normally tracks both column numbers and line 15522 numbers within source files and it normally prints both of these 15523 numbers in diagnostics. However, once it has processed a certain 15524 number of source lines, it stops tracking column numbers and only 15525 tracks line numbers. This means that diagnostics for later lines 15526 do not include column numbers. It also means that options like 15527 '-Wmisleading-indentation' cease to work at that point, although 15528 the compiler prints a note if this happens. Passing 15529 '-flarge-source-files' significantly increases the number of source 15530 lines that GCC can process before it stops tracking columns. 15531 15532 15533File: gcc.info, Node: Assembler Options, Next: Link Options, Prev: Preprocessor Options, Up: Invoking GCC 15534 155353.14 Passing Options to the Assembler 15536===================================== 15537 15538You can pass options to the assembler. 15539 15540'-Wa,OPTION' 15541 Pass OPTION as an option to the assembler. If OPTION contains 15542 commas, it is split into multiple options at the commas. 15543 15544'-Xassembler OPTION' 15545 Pass OPTION as an option to the assembler. You can use this to 15546 supply system-specific assembler options that GCC does not 15547 recognize. 15548 15549 If you want to pass an option that takes an argument, you must use 15550 '-Xassembler' twice, once for the option and once for the argument. 15551 15552 15553File: gcc.info, Node: Link Options, Next: Directory Options, Prev: Assembler Options, Up: Invoking GCC 15554 155553.15 Options for Linking 15556======================== 15557 15558These options come into play when the compiler links object files into 15559an executable output file. They are meaningless if the compiler is not 15560doing a link step. 15561 15562'OBJECT-FILE-NAME' 15563 A file name that does not end in a special recognized suffix is 15564 considered to name an object file or library. (Object files are 15565 distinguished from libraries by the linker according to the file 15566 contents.) If linking is done, these object files are used as 15567 input to the linker. 15568 15569'-c' 15570'-S' 15571'-E' 15572 If any of these options is used, then the linker is not run, and 15573 object file names should not be used as arguments. *Note Overall 15574 Options::. 15575 15576'-flinker-output=TYPE' 15577 This option controls code generation of the link-time optimizer. 15578 By default the linker output is automatically determined by the 15579 linker plugin. For debugging the compiler and if incremental 15580 linking with a non-LTO object file is desired, it may be useful to 15581 control the type manually. 15582 15583 If TYPE is 'exec', code generation produces a static binary. In 15584 this case '-fpic' and '-fpie' are both disabled. 15585 15586 If TYPE is 'dyn', code generation produces a shared library. In 15587 this case '-fpic' or '-fPIC' is preserved, but not enabled 15588 automatically. This allows to build shared libraries without 15589 position-independent code on architectures where this is possible, 15590 i.e. on x86. 15591 15592 If TYPE is 'pie', code generation produces an '-fpie' executable. 15593 This results in similar optimizations as 'exec' except that '-fpie' 15594 is not disabled if specified at compilation time. 15595 15596 If TYPE is 'rel', the compiler assumes that incremental linking is 15597 done. The sections containing intermediate code for link-time 15598 optimization are merged, pre-optimized, and output to the resulting 15599 object file. In addition, if '-ffat-lto-objects' is specified, 15600 binary code is produced for future non-LTO linking. The object 15601 file produced by incremental linking is smaller than a static 15602 library produced from the same object files. At link time the 15603 result of incremental linking also loads faster than a static 15604 library assuming that the majority of objects in the library are 15605 used. 15606 15607 Finally 'nolto-rel' configures the compiler for incremental linking 15608 where code generation is forced, a final binary is produced, and 15609 the intermediate code for later link-time optimization is stripped. 15610 When multiple object files are linked together the resulting code 15611 is better optimized than with link-time optimizations disabled (for 15612 example, cross-module inlining happens), but most of benefits of 15613 whole program optimizations are lost. 15614 15615 During the incremental link (by '-r') the linker plugin defaults to 15616 'rel'. With current interfaces to GNU Binutils it is however not 15617 possible to incrementally link LTO objects and non-LTO objects into 15618 a single mixed object file. If any of object files in incremental 15619 link cannot be used for link-time optimization, the linker plugin 15620 issues a warning and uses 'nolto-rel'. To maintain whole program 15621 optimization, it is recommended to link such objects into static 15622 library instead. Alternatively it is possible to use H.J. Lu's 15623 binutils with support for mixed objects. 15624 15625'-fuse-ld=bfd' 15626 Use the 'bfd' linker instead of the default linker. 15627 15628'-fuse-ld=gold' 15629 Use the 'gold' linker instead of the default linker. 15630 15631'-fuse-ld=lld' 15632 Use the LLVM 'lld' linker instead of the default linker. 15633 15634'-fuse-ld=mold' 15635 Use the Modern Linker ('mold') instead of the default linker. 15636 15637'-lLIBRARY' 15638'-l LIBRARY' 15639 Search the library named LIBRARY when linking. (The second 15640 alternative with the library as a separate argument is only for 15641 POSIX compliance and is not recommended.) 15642 15643 The '-l' option is passed directly to the linker by GCC. Refer to 15644 your linker documentation for exact details. The general 15645 description below applies to the GNU linker. 15646 15647 The linker searches a standard list of directories for the library. 15648 The directories searched include several standard system 15649 directories plus any that you specify with '-L'. 15650 15651 Static libraries are archives of object files, and have file names 15652 like 'libLIBRARY.a'. Some targets also support shared libraries, 15653 which typically have names like 'libLIBRARY.so'. If both static 15654 and shared libraries are found, the linker gives preference to 15655 linking with the shared library unless the '-static' option is 15656 used. 15657 15658 It makes a difference where in the command you write this option; 15659 the linker searches and processes libraries and object files in the 15660 order they are specified. Thus, 'foo.o -lz bar.o' searches library 15661 'z' after file 'foo.o' but before 'bar.o'. If 'bar.o' refers to 15662 functions in 'z', those functions may not be loaded. 15663 15664'-lobjc' 15665 You need this special case of the '-l' option in order to link an 15666 Objective-C or Objective-C++ program. 15667 15668'-nostartfiles' 15669 Do not use the standard system startup files when linking. The 15670 standard system libraries are used normally, unless '-nostdlib', 15671 '-nolibc', or '-nodefaultlibs' is used. 15672 15673'-nodefaultlibs' 15674 Do not use the standard system libraries when linking. Only the 15675 libraries you specify are passed to the linker, and options 15676 specifying linkage of the system libraries, such as 15677 '-static-libgcc' or '-shared-libgcc', are ignored. The standard 15678 startup files are used normally, unless '-nostartfiles' is used. 15679 15680 The compiler may generate calls to 'memcmp', 'memset', 'memcpy' and 15681 'memmove'. These entries are usually resolved by entries in libc. 15682 These entry points should be supplied through some other mechanism 15683 when this option is specified. 15684 15685'-nolibc' 15686 Do not use the C library or system libraries tightly coupled with 15687 it when linking. Still link with the startup files, 'libgcc' or 15688 toolchain provided language support libraries such as 'libgnat', 15689 'libgfortran' or 'libstdc++' unless options preventing their 15690 inclusion are used as well. This typically removes '-lc' from the 15691 link command line, as well as system libraries that normally go 15692 with it and become meaningless when absence of a C library is 15693 assumed, for example '-lpthread' or '-lm' in some configurations. 15694 This is intended for bare-board targets when there is indeed no C 15695 library available. 15696 15697'-nostdlib' 15698 Do not use the standard system startup files or libraries when 15699 linking. No startup files and only the libraries you specify are 15700 passed to the linker, and options specifying linkage of the system 15701 libraries, such as '-static-libgcc' or '-shared-libgcc', are 15702 ignored. 15703 15704 The compiler may generate calls to 'memcmp', 'memset', 'memcpy' and 15705 'memmove'. These entries are usually resolved by entries in libc. 15706 These entry points should be supplied through some other mechanism 15707 when this option is specified. 15708 15709 One of the standard libraries bypassed by '-nostdlib' and 15710 '-nodefaultlibs' is 'libgcc.a', a library of internal subroutines 15711 which GCC uses to overcome shortcomings of particular machines, or 15712 special needs for some languages. (*Note Interfacing to GCC 15713 Output: (gccint)Interface, for more discussion of 'libgcc.a'.) In 15714 most cases, you need 'libgcc.a' even when you want to avoid other 15715 standard libraries. In other words, when you specify '-nostdlib' 15716 or '-nodefaultlibs' you should usually specify '-lgcc' as well. 15717 This ensures that you have no unresolved references to internal GCC 15718 library subroutines. (An example of such an internal subroutine is 15719 '__main', used to ensure C++ constructors are called; *note 15720 'collect2': (gccint)Collect2.) 15721 15722'-e ENTRY' 15723'--entry=ENTRY' 15724 15725 Specify that the program entry point is ENTRY. The argument is 15726 interpreted by the linker; the GNU linker accepts either a symbol 15727 name or an address. 15728 15729'-pie' 15730 Produce a dynamically linked position independent executable on 15731 targets that support it. For predictable results, you must also 15732 specify the same set of options used for compilation ('-fpie', 15733 '-fPIE', or model suboptions) when you specify this linker option. 15734 15735'-no-pie' 15736 Don't produce a dynamically linked position independent executable. 15737 15738'-static-pie' 15739 Produce a static position independent executable on targets that 15740 support it. A static position independent executable is similar to 15741 a static executable, but can be loaded at any address without a 15742 dynamic linker. For predictable results, you must also specify the 15743 same set of options used for compilation ('-fpie', '-fPIE', or 15744 model suboptions) when you specify this linker option. 15745 15746'-pthread' 15747 Link with the POSIX threads library. This option is supported on 15748 GNU/Linux targets, most other Unix derivatives, and also on x86 15749 Cygwin and MinGW targets. On some targets this option also sets 15750 flags for the preprocessor, so it should be used consistently for 15751 both compilation and linking. 15752 15753'-r' 15754 Produce a relocatable object as output. This is also known as 15755 partial linking. 15756 15757'-rdynamic' 15758 Pass the flag '-export-dynamic' to the ELF linker, on targets that 15759 support it. This instructs the linker to add all symbols, not only 15760 used ones, to the dynamic symbol table. This option is needed for 15761 some uses of 'dlopen' or to allow obtaining backtraces from within 15762 a program. 15763 15764'-s' 15765 Remove all symbol table and relocation information from the 15766 executable. 15767 15768'-static' 15769 On systems that support dynamic linking, this overrides '-pie' and 15770 prevents linking with the shared libraries. On other systems, this 15771 option has no effect. 15772 15773'-shared' 15774 Produce a shared object which can then be linked with other objects 15775 to form an executable. Not all systems support this option. For 15776 predictable results, you must also specify the same set of options 15777 used for compilation ('-fpic', '-fPIC', or model suboptions) when 15778 you specify this linker option.(1) 15779 15780'-shared-libgcc' 15781'-static-libgcc' 15782 On systems that provide 'libgcc' as a shared library, these options 15783 force the use of either the shared or static version, respectively. 15784 If no shared version of 'libgcc' was built when the compiler was 15785 configured, these options have no effect. 15786 15787 There are several situations in which an application should use the 15788 shared 'libgcc' instead of the static version. The most common of 15789 these is when the application wishes to throw and catch exceptions 15790 across different shared libraries. In that case, each of the 15791 libraries as well as the application itself should use the shared 15792 'libgcc'. 15793 15794 Therefore, the G++ driver automatically adds '-shared-libgcc' 15795 whenever you build a shared library or a main executable, because 15796 C++ programs typically use exceptions, so this is the right thing 15797 to do. 15798 15799 If, instead, you use the GCC driver to create shared libraries, you 15800 may find that they are not always linked with the shared 'libgcc'. 15801 If GCC finds, at its configuration time, that you have a non-GNU 15802 linker or a GNU linker that does not support option 15803 '--eh-frame-hdr', it links the shared version of 'libgcc' into 15804 shared libraries by default. Otherwise, it takes advantage of the 15805 linker and optimizes away the linking with the shared version of 15806 'libgcc', linking with the static version of libgcc by default. 15807 This allows exceptions to propagate through such shared libraries, 15808 without incurring relocation costs at library load time. 15809 15810 However, if a library or main executable is supposed to throw or 15811 catch exceptions, you must link it using the G++ driver, or using 15812 the option '-shared-libgcc', such that it is linked with the shared 15813 'libgcc'. 15814 15815'-static-libasan' 15816 When the '-fsanitize=address' option is used to link a program, the 15817 GCC driver automatically links against 'libasan'. If 'libasan' is 15818 available as a shared library, and the '-static' option is not 15819 used, then this links against the shared version of 'libasan'. The 15820 '-static-libasan' option directs the GCC driver to link 'libasan' 15821 statically, without necessarily linking other libraries statically. 15822 15823'-static-libtsan' 15824 When the '-fsanitize=thread' option is used to link a program, the 15825 GCC driver automatically links against 'libtsan'. If 'libtsan' is 15826 available as a shared library, and the '-static' option is not 15827 used, then this links against the shared version of 'libtsan'. The 15828 '-static-libtsan' option directs the GCC driver to link 'libtsan' 15829 statically, without necessarily linking other libraries statically. 15830 15831'-static-liblsan' 15832 When the '-fsanitize=leak' option is used to link a program, the 15833 GCC driver automatically links against 'liblsan'. If 'liblsan' is 15834 available as a shared library, and the '-static' option is not 15835 used, then this links against the shared version of 'liblsan'. The 15836 '-static-liblsan' option directs the GCC driver to link 'liblsan' 15837 statically, without necessarily linking other libraries statically. 15838 15839'-static-libubsan' 15840 When the '-fsanitize=undefined' option is used to link a program, 15841 the GCC driver automatically links against 'libubsan'. If 15842 'libubsan' is available as a shared library, and the '-static' 15843 option is not used, then this links against the shared version of 15844 'libubsan'. The '-static-libubsan' option directs the GCC driver 15845 to link 'libubsan' statically, without necessarily linking other 15846 libraries statically. 15847 15848'-static-libstdc++' 15849 When the 'g++' program is used to link a C++ program, it normally 15850 automatically links against 'libstdc++'. If 'libstdc++' is 15851 available as a shared library, and the '-static' option is not 15852 used, then this links against the shared version of 'libstdc++'. 15853 That is normally fine. However, it is sometimes useful to freeze 15854 the version of 'libstdc++' used by the program without going all 15855 the way to a fully static link. The '-static-libstdc++' option 15856 directs the 'g++' driver to link 'libstdc++' statically, without 15857 necessarily linking other libraries statically. 15858 15859'-symbolic' 15860 Bind references to global symbols when building a shared object. 15861 Warn about any unresolved references (unless overridden by the link 15862 editor option '-Xlinker -z -Xlinker defs'). Only a few systems 15863 support this option. 15864 15865'-T SCRIPT' 15866 Use SCRIPT as the linker script. This option is supported by most 15867 systems using the GNU linker. On some targets, such as bare-board 15868 targets without an operating system, the '-T' option may be 15869 required when linking to avoid references to undefined symbols. 15870 15871'-Xlinker OPTION' 15872 Pass OPTION as an option to the linker. You can use this to supply 15873 system-specific linker options that GCC does not recognize. 15874 15875 If you want to pass an option that takes a separate argument, you 15876 must use '-Xlinker' twice, once for the option and once for the 15877 argument. For example, to pass '-assert definitions', you must 15878 write '-Xlinker -assert -Xlinker definitions'. It does not work to 15879 write '-Xlinker "-assert definitions"', because this passes the 15880 entire string as a single argument, which is not what the linker 15881 expects. 15882 15883 When using the GNU linker, it is usually more convenient to pass 15884 arguments to linker options using the 'OPTION=VALUE' syntax than as 15885 separate arguments. For example, you can specify '-Xlinker 15886 -Map=output.map' rather than '-Xlinker -Map -Xlinker output.map'. 15887 Other linkers may not support this syntax for command-line options. 15888 15889'-Wl,OPTION' 15890 Pass OPTION as an option to the linker. If OPTION contains commas, 15891 it is split into multiple options at the commas. You can use this 15892 syntax to pass an argument to the option. For example, 15893 '-Wl,-Map,output.map' passes '-Map output.map' to the linker. When 15894 using the GNU linker, you can also get the same effect with 15895 '-Wl,-Map=output.map'. 15896 15897'-u SYMBOL' 15898 Pretend the symbol SYMBOL is undefined, to force linking of library 15899 modules to define it. You can use '-u' multiple times with 15900 different symbols to force loading of additional library modules. 15901 15902'-z KEYWORD' 15903 '-z' is passed directly on to the linker along with the keyword 15904 KEYWORD. See the section in the documentation of your linker for 15905 permitted values and their meanings. 15906 15907 ---------- Footnotes ---------- 15908 15909 (1) On some systems, 'gcc -shared' needs to build supplementary stub 15910code for constructors to work. On multi-libbed systems, 'gcc -shared' 15911must select the correct support libraries to link against. Failing to 15912supply the correct flags may lead to subtle defects. Supplying them in 15913cases where they are not necessary is innocuous. 15914 15915 15916File: gcc.info, Node: Directory Options, Next: Code Gen Options, Prev: Link Options, Up: Invoking GCC 15917 159183.16 Options for Directory Search 15919================================= 15920 15921These options specify directories to search for header files, for 15922libraries and for parts of the compiler: 15923 15924'-I DIR' 15925'-iquote DIR' 15926'-isystem DIR' 15927'-idirafter DIR' 15928 Add the directory DIR to the list of directories to be searched for 15929 header files during preprocessing. If DIR begins with '=' or 15930 '$SYSROOT', then the '=' or '$SYSROOT' is replaced by the sysroot 15931 prefix; see '--sysroot' and '-isysroot'. 15932 15933 Directories specified with '-iquote' apply only to the quote form 15934 of the directive, '#include "FILE"'. Directories specified with 15935 '-I', '-isystem', or '-idirafter' apply to lookup for both the 15936 '#include "FILE"' and '#include <FILE>' directives. 15937 15938 You can specify any number or combination of these options on the 15939 command line to search for header files in several directories. 15940 The lookup order is as follows: 15941 15942 1. For the quote form of the include directive, the directory of 15943 the current file is searched first. 15944 15945 2. For the quote form of the include directive, the directories 15946 specified by '-iquote' options are searched in left-to-right 15947 order, as they appear on the command line. 15948 15949 3. Directories specified with '-I' options are scanned in 15950 left-to-right order. 15951 15952 4. Directories specified with '-isystem' options are scanned in 15953 left-to-right order. 15954 15955 5. Standard system directories are scanned. 15956 15957 6. Directories specified with '-idirafter' options are scanned in 15958 left-to-right order. 15959 15960 You can use '-I' to override a system header file, substituting 15961 your own version, since these directories are searched before the 15962 standard system header file directories. However, you should not 15963 use this option to add directories that contain vendor-supplied 15964 system header files; use '-isystem' for that. 15965 15966 The '-isystem' and '-idirafter' options also mark the directory as 15967 a system directory, so that it gets the same special treatment that 15968 is applied to the standard system directories. 15969 15970 If a standard system include directory, or a directory specified 15971 with '-isystem', is also specified with '-I', the '-I' option is 15972 ignored. The directory is still searched but as a system directory 15973 at its normal position in the system include chain. This is to 15974 ensure that GCC's procedure to fix buggy system headers and the 15975 ordering for the '#include_next' directive are not inadvertently 15976 changed. If you really need to change the search order for system 15977 directories, use the '-nostdinc' and/or '-isystem' options. 15978 15979'-I-' 15980 Split the include path. This option has been deprecated. Please 15981 use '-iquote' instead for '-I' directories before the '-I-' and 15982 remove the '-I-' option. 15983 15984 Any directories specified with '-I' options before '-I-' are 15985 searched only for headers requested with '#include "FILE"'; they 15986 are not searched for '#include <FILE>'. If additional directories 15987 are specified with '-I' options after the '-I-', those directories 15988 are searched for all '#include' directives. 15989 15990 In addition, '-I-' inhibits the use of the directory of the current 15991 file directory as the first search directory for '#include "FILE"'. 15992 There is no way to override this effect of '-I-'. 15993 15994'-iprefix PREFIX' 15995 Specify PREFIX as the prefix for subsequent '-iwithprefix' options. 15996 If the prefix represents a directory, you should include the final 15997 '/'. 15998 15999'-iwithprefix DIR' 16000'-iwithprefixbefore DIR' 16001 Append DIR to the prefix specified previously with '-iprefix', and 16002 add the resulting directory to the include search path. 16003 '-iwithprefixbefore' puts it in the same place '-I' would; 16004 '-iwithprefix' puts it where '-idirafter' would. 16005 16006'-isysroot DIR' 16007 This option is like the '--sysroot' option, but applies only to 16008 header files (except for Darwin targets, where it applies to both 16009 header files and libraries). See the '--sysroot' option for more 16010 information. 16011 16012'-imultilib DIR' 16013 Use DIR as a subdirectory of the directory containing 16014 target-specific C++ headers. 16015 16016'-nostdinc' 16017 Do not search the standard system directories for header files. 16018 Only the directories explicitly specified with '-I', '-iquote', 16019 '-isystem', and/or '-idirafter' options (and the directory of the 16020 current file, if appropriate) are searched. 16021 16022'-nostdinc++' 16023 Do not search for header files in the C++-specific standard 16024 directories, but do still search the other standard directories. 16025 (This option is used when building the C++ library.) 16026 16027'-iplugindir=DIR' 16028 Set the directory to search for plugins that are passed by 16029 '-fplugin=NAME' instead of '-fplugin=PATH/NAME.so'. This option is 16030 not meant to be used by the user, but only passed by the driver. 16031 16032'-LDIR' 16033 Add directory DIR to the list of directories to be searched for 16034 '-l'. 16035 16036'-BPREFIX' 16037 This option specifies where to find the executables, libraries, 16038 include files, and data files of the compiler itself. 16039 16040 The compiler driver program runs one or more of the subprograms 16041 'cpp', 'cc1', 'as' and 'ld'. It tries PREFIX as a prefix for each 16042 program it tries to run, both with and without 'MACHINE/VERSION/' 16043 for the corresponding target machine and compiler version. 16044 16045 For each subprogram to be run, the compiler driver first tries the 16046 '-B' prefix, if any. If that name is not found, or if '-B' is not 16047 specified, the driver tries two standard prefixes, '/usr/lib/gcc/' 16048 and '/usr/local/lib/gcc/'. If neither of those results in a file 16049 name that is found, the unmodified program name is searched for 16050 using the directories specified in your 'PATH' environment 16051 variable. 16052 16053 The compiler checks to see if the path provided by '-B' refers to a 16054 directory, and if necessary it adds a directory separator character 16055 at the end of the path. 16056 16057 '-B' prefixes that effectively specify directory names also apply 16058 to libraries in the linker, because the compiler translates these 16059 options into '-L' options for the linker. They also apply to 16060 include files in the preprocessor, because the compiler translates 16061 these options into '-isystem' options for the preprocessor. In 16062 this case, the compiler appends 'include' to the prefix. 16063 16064 The runtime support file 'libgcc.a' can also be searched for using 16065 the '-B' prefix, if needed. If it is not found there, the two 16066 standard prefixes above are tried, and that is all. The file is 16067 left out of the link if it is not found by those means. 16068 16069 Another way to specify a prefix much like the '-B' prefix is to use 16070 the environment variable 'GCC_EXEC_PREFIX'. *Note Environment 16071 Variables::. 16072 16073 As a special kludge, if the path provided by '-B' is 16074 '[dir/]stageN/', where N is a number in the range 0 to 9, then it 16075 is replaced by '[dir/]include'. This is to help with 16076 boot-strapping the compiler. 16077 16078'-no-canonical-prefixes' 16079 Do not expand any symbolic links, resolve references to '/../' or 16080 '/./', or make the path absolute when generating a relative prefix. 16081 16082'--sysroot=DIR' 16083 Use DIR as the logical root directory for headers and libraries. 16084 For example, if the compiler normally searches for headers in 16085 '/usr/include' and libraries in '/usr/lib', it instead searches 16086 'DIR/usr/include' and 'DIR/usr/lib'. 16087 16088 If you use both this option and the '-isysroot' option, then the 16089 '--sysroot' option applies to libraries, but the '-isysroot' option 16090 applies to header files. 16091 16092 The GNU linker (beginning with version 2.16) has the necessary 16093 support for this option. If your linker does not support this 16094 option, the header file aspect of '--sysroot' still works, but the 16095 library aspect does not. 16096 16097'--no-sysroot-suffix' 16098 For some targets, a suffix is added to the root directory specified 16099 with '--sysroot', depending on the other options used, so that 16100 headers may for example be found in 'DIR/SUFFIX/usr/include' 16101 instead of 'DIR/usr/include'. This option disables the addition of 16102 such a suffix. 16103 16104 16105File: gcc.info, Node: Code Gen Options, Next: Developer Options, Prev: Directory Options, Up: Invoking GCC 16106 161073.17 Options for Code Generation Conventions 16108============================================ 16109 16110These machine-independent options control the interface conventions used 16111in code generation. 16112 16113 Most of them have both positive and negative forms; the negative form 16114of '-ffoo' is '-fno-foo'. In the table below, only one of the forms is 16115listed--the one that is not the default. You can figure out the other 16116form by either removing 'no-' or adding it. 16117 16118'-fstack-reuse=REUSE-LEVEL' 16119 This option controls stack space reuse for user declared local/auto 16120 variables and compiler generated temporaries. REUSE_LEVEL can be 16121 'all', 'named_vars', or 'none'. 'all' enables stack reuse for all 16122 local variables and temporaries, 'named_vars' enables the reuse 16123 only for user defined local variables with names, and 'none' 16124 disables stack reuse completely. The default value is 'all'. The 16125 option is needed when the program extends the lifetime of a scoped 16126 local variable or a compiler generated temporary beyond the end 16127 point defined by the language. When a lifetime of a variable ends, 16128 and if the variable lives in memory, the optimizing compiler has 16129 the freedom to reuse its stack space with other temporaries or 16130 scoped local variables whose live range does not overlap with it. 16131 Legacy code extending local lifetime is likely to break with the 16132 stack reuse optimization. 16133 16134 For example, 16135 16136 int *p; 16137 { 16138 int local1; 16139 16140 p = &local1; 16141 local1 = 10; 16142 .... 16143 } 16144 { 16145 int local2; 16146 local2 = 20; 16147 ... 16148 } 16149 16150 if (*p == 10) // out of scope use of local1 16151 { 16152 16153 } 16154 16155 Another example: 16156 16157 struct A 16158 { 16159 A(int k) : i(k), j(k) { } 16160 int i; 16161 int j; 16162 }; 16163 16164 A *ap; 16165 16166 void foo(const A& ar) 16167 { 16168 ap = &ar; 16169 } 16170 16171 void bar() 16172 { 16173 foo(A(10)); // temp object's lifetime ends when foo returns 16174 16175 { 16176 A a(20); 16177 .... 16178 } 16179 ap->i+= 10; // ap references out of scope temp whose space 16180 // is reused with a. What is the value of ap->i? 16181 } 16182 16183 16184 The lifetime of a compiler generated temporary is well defined by 16185 the C++ standard. When a lifetime of a temporary ends, and if the 16186 temporary lives in memory, the optimizing compiler has the freedom 16187 to reuse its stack space with other temporaries or scoped local 16188 variables whose live range does not overlap with it. However some 16189 of the legacy code relies on the behavior of older compilers in 16190 which temporaries' stack space is not reused, the aggressive stack 16191 reuse can lead to runtime errors. This option is used to control 16192 the temporary stack reuse optimization. 16193 16194'-ftrapv' 16195 This option generates traps for signed overflow on addition, 16196 subtraction, multiplication operations. The options '-ftrapv' and 16197 '-fwrapv' override each other, so using '-ftrapv' '-fwrapv' on the 16198 command-line results in '-fwrapv' being effective. Note that only 16199 active options override, so using '-ftrapv' '-fwrapv' '-fno-wrapv' 16200 on the command-line results in '-ftrapv' being effective. 16201 16202'-fwrapv' 16203 This option instructs the compiler to assume that signed arithmetic 16204 overflow of addition, subtraction and multiplication wraps around 16205 using twos-complement representation. This flag enables some 16206 optimizations and disables others. The options '-ftrapv' and 16207 '-fwrapv' override each other, so using '-ftrapv' '-fwrapv' on the 16208 command-line results in '-fwrapv' being effective. Note that only 16209 active options override, so using '-ftrapv' '-fwrapv' '-fno-wrapv' 16210 on the command-line results in '-ftrapv' being effective. 16211 16212'-fwrapv-pointer' 16213 This option instructs the compiler to assume that pointer 16214 arithmetic overflow on addition and subtraction wraps around using 16215 twos-complement representation. This flag disables some 16216 optimizations which assume pointer overflow is invalid. 16217 16218'-fstrict-overflow' 16219 This option implies '-fno-wrapv' '-fno-wrapv-pointer' and when 16220 negated implies '-fwrapv' '-fwrapv-pointer'. 16221 16222'-fexceptions' 16223 Enable exception handling. Generates extra code needed to 16224 propagate exceptions. For some targets, this implies GCC generates 16225 frame unwind information for all functions, which can produce 16226 significant data size overhead, although it does not affect 16227 execution. If you do not specify this option, GCC enables it by 16228 default for languages like C++ that normally require exception 16229 handling, and disables it for languages like C that do not normally 16230 require it. However, you may need to enable this option when 16231 compiling C code that needs to interoperate properly with exception 16232 handlers written in C++. You may also wish to disable this option 16233 if you are compiling older C++ programs that don't use exception 16234 handling. 16235 16236'-fnon-call-exceptions' 16237 Generate code that allows trapping instructions to throw 16238 exceptions. Note that this requires platform-specific runtime 16239 support that does not exist everywhere. Moreover, it only allows 16240 _trapping_ instructions to throw exceptions, i.e. memory references 16241 or floating-point instructions. It does not allow exceptions to be 16242 thrown from arbitrary signal handlers such as 'SIGALRM'. This 16243 enables '-fexceptions'. 16244 16245'-fdelete-dead-exceptions' 16246 Consider that instructions that may throw exceptions but don't 16247 otherwise contribute to the execution of the program can be 16248 optimized away. This does not affect calls to functions except 16249 those with the 'pure' or 'const' attributes. This option is 16250 enabled by default for the Ada and C++ compilers, as permitted by 16251 the language specifications. Optimization passes that cause dead 16252 exceptions to be removed are enabled independently at different 16253 optimization levels. 16254 16255'-funwind-tables' 16256 Similar to '-fexceptions', except that it just generates any needed 16257 static data, but does not affect the generated code in any other 16258 way. You normally do not need to enable this option; instead, a 16259 language processor that needs this handling enables it on your 16260 behalf. 16261 16262'-fasynchronous-unwind-tables' 16263 Generate unwind table in DWARF format, if supported by target 16264 machine. The table is exact at each instruction boundary, so it 16265 can be used for stack unwinding from asynchronous events (such as 16266 debugger or garbage collector). 16267 16268'-fno-gnu-unique' 16269 On systems with recent GNU assembler and C library, the C++ 16270 compiler uses the 'STB_GNU_UNIQUE' binding to make sure that 16271 definitions of template static data members and static local 16272 variables in inline functions are unique even in the presence of 16273 'RTLD_LOCAL'; this is necessary to avoid problems with a library 16274 used by two different 'RTLD_LOCAL' plugins depending on a 16275 definition in one of them and therefore disagreeing with the other 16276 one about the binding of the symbol. But this causes 'dlclose' to 16277 be ignored for affected DSOs; if your program relies on 16278 reinitialization of a DSO via 'dlclose' and 'dlopen', you can use 16279 '-fno-gnu-unique'. 16280 16281'-fpcc-struct-return' 16282 Return "short" 'struct' and 'union' values in memory like longer 16283 ones, rather than in registers. This convention is less efficient, 16284 but it has the advantage of allowing intercallability between 16285 GCC-compiled files and files compiled with other compilers, 16286 particularly the Portable C Compiler (pcc). 16287 16288 The precise convention for returning structures in memory depends 16289 on the target configuration macros. 16290 16291 Short structures and unions are those whose size and alignment 16292 match that of some integer type. 16293 16294 *Warning:* code compiled with the '-fpcc-struct-return' switch is 16295 not binary compatible with code compiled with the 16296 '-freg-struct-return' switch. Use it to conform to a non-default 16297 application binary interface. 16298 16299'-freg-struct-return' 16300 Return 'struct' and 'union' values in registers when possible. 16301 This is more efficient for small structures than 16302 '-fpcc-struct-return'. 16303 16304 If you specify neither '-fpcc-struct-return' nor 16305 '-freg-struct-return', GCC defaults to whichever convention is 16306 standard for the target. If there is no standard convention, GCC 16307 defaults to '-fpcc-struct-return', except on targets where GCC is 16308 the principal compiler. In those cases, we can choose the 16309 standard, and we chose the more efficient register return 16310 alternative. 16311 16312 *Warning:* code compiled with the '-freg-struct-return' switch is 16313 not binary compatible with code compiled with the 16314 '-fpcc-struct-return' switch. Use it to conform to a non-default 16315 application binary interface. 16316 16317'-fshort-enums' 16318 Allocate to an 'enum' type only as many bytes as it needs for the 16319 declared range of possible values. Specifically, the 'enum' type 16320 is equivalent to the smallest integer type that has enough room. 16321 16322 *Warning:* the '-fshort-enums' switch causes GCC to generate code 16323 that is not binary compatible with code generated without that 16324 switch. Use it to conform to a non-default application binary 16325 interface. 16326 16327'-fshort-wchar' 16328 Override the underlying type for 'wchar_t' to be 'short unsigned 16329 int' instead of the default for the target. This option is useful 16330 for building programs to run under WINE. 16331 16332 *Warning:* the '-fshort-wchar' switch causes GCC to generate code 16333 that is not binary compatible with code generated without that 16334 switch. Use it to conform to a non-default application binary 16335 interface. 16336 16337'-fcommon' 16338 In C code, this option controls the placement of global variables 16339 defined without an initializer, known as "tentative definitions" in 16340 the C standard. Tentative definitions are distinct from 16341 declarations of a variable with the 'extern' keyword, which do not 16342 allocate storage. 16343 16344 The default is '-fno-common', which specifies that the compiler 16345 places uninitialized global variables in the BSS section of the 16346 object file. This inhibits the merging of tentative definitions by 16347 the linker so you get a multiple-definition error if the same 16348 variable is accidentally defined in more than one compilation unit. 16349 16350 The '-fcommon' places uninitialized global variables in a common 16351 block. This allows the linker to resolve all tentative definitions 16352 of the same variable in different compilation units to the same 16353 object, or to a non-tentative definition. This behavior is 16354 inconsistent with C++, and on many targets implies a speed and code 16355 size penalty on global variable references. It is mainly useful to 16356 enable legacy code to link without errors. 16357 16358'-fno-ident' 16359 Ignore the '#ident' directive. 16360 16361'-finhibit-size-directive' 16362 Don't output a '.size' assembler directive, or anything else that 16363 would cause trouble if the function is split in the middle, and the 16364 two halves are placed at locations far apart in memory. This 16365 option is used when compiling 'crtstuff.c'; you should not need to 16366 use it for anything else. 16367 16368'-fverbose-asm' 16369 Put extra commentary information in the generated assembly code to 16370 make it more readable. This option is generally only of use to 16371 those who actually need to read the generated assembly code 16372 (perhaps while debugging the compiler itself). 16373 16374 '-fno-verbose-asm', the default, causes the extra information to be 16375 omitted and is useful when comparing two assembler files. 16376 16377 The added comments include: 16378 16379 * information on the compiler version and command-line options, 16380 16381 * the source code lines associated with the assembly 16382 instructions, in the form FILENAME:LINENUMBER:CONTENT OF LINE, 16383 16384 * hints on which high-level expressions correspond to the 16385 various assembly instruction operands. 16386 16387 For example, given this C source file: 16388 16389 int test (int n) 16390 { 16391 int i; 16392 int total = 0; 16393 16394 for (i = 0; i < n; i++) 16395 total += i * i; 16396 16397 return total; 16398 } 16399 16400 compiling to (x86_64) assembly via '-S' and emitting the result 16401 direct to stdout via '-o' '-' 16402 16403 gcc -S test.c -fverbose-asm -Os -o - 16404 16405 gives output similar to this: 16406 16407 .file "test.c" 16408 # GNU C11 (GCC) version 7.0.0 20160809 (experimental) (x86_64-pc-linux-gnu) 16409 [...snip...] 16410 # options passed: 16411 [...snip...] 16412 16413 .text 16414 .globl test 16415 .type test, @function 16416 test: 16417 .LFB0: 16418 .cfi_startproc 16419 # test.c:4: int total = 0; 16420 xorl %eax, %eax # <retval> 16421 # test.c:6: for (i = 0; i < n; i++) 16422 xorl %edx, %edx # i 16423 .L2: 16424 # test.c:6: for (i = 0; i < n; i++) 16425 cmpl %edi, %edx # n, i 16426 jge .L5 #, 16427 # test.c:7: total += i * i; 16428 movl %edx, %ecx # i, tmp92 16429 imull %edx, %ecx # i, tmp92 16430 # test.c:6: for (i = 0; i < n; i++) 16431 incl %edx # i 16432 # test.c:7: total += i * i; 16433 addl %ecx, %eax # tmp92, <retval> 16434 jmp .L2 # 16435 .L5: 16436 # test.c:10: } 16437 ret 16438 .cfi_endproc 16439 .LFE0: 16440 .size test, .-test 16441 .ident "GCC: (GNU) 7.0.0 20160809 (experimental)" 16442 .section .note.GNU-stack,"",@progbits 16443 16444 The comments are intended for humans rather than machines and hence 16445 the precise format of the comments is subject to change. 16446 16447'-frecord-gcc-switches' 16448 This switch causes the command line used to invoke the compiler to 16449 be recorded into the object file that is being created. This 16450 switch is only implemented on some targets and the exact format of 16451 the recording is target and binary file format dependent, but it 16452 usually takes the form of a section containing ASCII text. This 16453 switch is related to the '-fverbose-asm' switch, but that switch 16454 only records information in the assembler output file as comments, 16455 so it never reaches the object file. See also 16456 '-grecord-gcc-switches' for another way of storing compiler options 16457 into the object file. 16458 16459'-fpic' 16460 Generate position-independent code (PIC) suitable for use in a 16461 shared library, if supported for the target machine. Such code 16462 accesses all constant addresses through a global offset table 16463 (GOT). The dynamic loader resolves the GOT entries when the 16464 program starts (the dynamic loader is not part of GCC; it is part 16465 of the operating system). If the GOT size for the linked 16466 executable exceeds a machine-specific maximum size, you get an 16467 error message from the linker indicating that '-fpic' does not 16468 work; in that case, recompile with '-fPIC' instead. (These 16469 maximums are 8k on the SPARC, 28k on AArch64 and 32k on the m68k 16470 and RS/6000. The x86 has no such limit.) 16471 16472 Position-independent code requires special support, and therefore 16473 works only on certain machines. For the x86, GCC supports PIC for 16474 System V but not for the Sun 386i. Code generated for the IBM 16475 RS/6000 is always position-independent. 16476 16477 When this flag is set, the macros '__pic__' and '__PIC__' are 16478 defined to 1. 16479 16480'-fPIC' 16481 If supported for the target machine, emit position-independent 16482 code, suitable for dynamic linking and avoiding any limit on the 16483 size of the global offset table. This option makes a difference on 16484 AArch64, m68k, PowerPC and SPARC. 16485 16486 Position-independent code requires special support, and therefore 16487 works only on certain machines. 16488 16489 When this flag is set, the macros '__pic__' and '__PIC__' are 16490 defined to 2. 16491 16492'-fpie' 16493'-fPIE' 16494 These options are similar to '-fpic' and '-fPIC', but the generated 16495 position-independent code can be only linked into executables. 16496 Usually these options are used to compile code that will be linked 16497 using the '-pie' GCC option. 16498 16499 '-fpie' and '-fPIE' both define the macros '__pie__' and '__PIE__'. 16500 The macros have the value 1 for '-fpie' and 2 for '-fPIE'. 16501 16502'-fno-plt' 16503 Do not use the PLT for external function calls in 16504 position-independent code. Instead, load the callee address at 16505 call sites from the GOT and branch to it. This leads to more 16506 efficient code by eliminating PLT stubs and exposing GOT loads to 16507 optimizations. On architectures such as 32-bit x86 where PLT stubs 16508 expect the GOT pointer in a specific register, this gives more 16509 register allocation freedom to the compiler. Lazy binding requires 16510 use of the PLT; with '-fno-plt' all external symbols are resolved 16511 at load time. 16512 16513 Alternatively, the function attribute 'noplt' can be used to avoid 16514 calls through the PLT for specific external functions. 16515 16516 In position-dependent code, a few targets also convert calls to 16517 functions that are marked to not use the PLT to use the GOT 16518 instead. 16519 16520'-fno-jump-tables' 16521 Do not use jump tables for switch statements even where it would be 16522 more efficient than other code generation strategies. This option 16523 is of use in conjunction with '-fpic' or '-fPIC' for building code 16524 that forms part of a dynamic linker and cannot reference the 16525 address of a jump table. On some targets, jump tables do not 16526 require a GOT and this option is not needed. 16527 16528'-fno-bit-tests' 16529 Do not use bit tests for switch statements even where it would be 16530 more efficient than other code generation strategies. 16531 16532'-ffixed-REG' 16533 Treat the register named REG as a fixed register; generated code 16534 should never refer to it (except perhaps as a stack pointer, frame 16535 pointer or in some other fixed role). 16536 16537 REG must be the name of a register. The register names accepted 16538 are machine-specific and are defined in the 'REGISTER_NAMES' macro 16539 in the machine description macro file. 16540 16541 This flag does not have a negative form, because it specifies a 16542 three-way choice. 16543 16544'-fcall-used-REG' 16545 Treat the register named REG as an allocable register that is 16546 clobbered by function calls. It may be allocated for temporaries 16547 or variables that do not live across a call. Functions compiled 16548 this way do not save and restore the register REG. 16549 16550 It is an error to use this flag with the frame pointer or stack 16551 pointer. Use of this flag for other registers that have fixed 16552 pervasive roles in the machine's execution model produces 16553 disastrous results. 16554 16555 This flag does not have a negative form, because it specifies a 16556 three-way choice. 16557 16558'-fcall-saved-REG' 16559 Treat the register named REG as an allocable register saved by 16560 functions. It may be allocated even for temporaries or variables 16561 that live across a call. Functions compiled this way save and 16562 restore the register REG if they use it. 16563 16564 It is an error to use this flag with the frame pointer or stack 16565 pointer. Use of this flag for other registers that have fixed 16566 pervasive roles in the machine's execution model produces 16567 disastrous results. 16568 16569 A different sort of disaster results from the use of this flag for 16570 a register in which function values may be returned. 16571 16572 This flag does not have a negative form, because it specifies a 16573 three-way choice. 16574 16575'-fpack-struct[=N]' 16576 Without a value specified, pack all structure members together 16577 without holes. When a value is specified (which must be a small 16578 power of two), pack structure members according to this value, 16579 representing the maximum alignment (that is, objects with default 16580 alignment requirements larger than this are output potentially 16581 unaligned at the next fitting location. 16582 16583 *Warning:* the '-fpack-struct' switch causes GCC to generate code 16584 that is not binary compatible with code generated without that 16585 switch. Additionally, it makes the code suboptimal. Use it to 16586 conform to a non-default application binary interface. 16587 16588'-fleading-underscore' 16589 This option and its counterpart, '-fno-leading-underscore', 16590 forcibly change the way C symbols are represented in the object 16591 file. One use is to help link with legacy assembly code. 16592 16593 *Warning:* the '-fleading-underscore' switch causes GCC to generate 16594 code that is not binary compatible with code generated without that 16595 switch. Use it to conform to a non-default application binary 16596 interface. Not all targets provide complete support for this 16597 switch. 16598 16599'-ftls-model=MODEL' 16600 Alter the thread-local storage model to be used (*note 16601 Thread-Local::). The MODEL argument should be one of 16602 'global-dynamic', 'local-dynamic', 'initial-exec' or 'local-exec'. 16603 Note that the choice is subject to optimization: the compiler may 16604 use a more efficient model for symbols not visible outside of the 16605 translation unit, or if '-fpic' is not given on the command line. 16606 16607 The default without '-fpic' is 'initial-exec'; with '-fpic' the 16608 default is 'global-dynamic'. 16609 16610'-ftrampolines' 16611 For targets that normally need trampolines for nested functions, 16612 always generate them instead of using descriptors. Otherwise, for 16613 targets that do not need them, like for example HP-PA or IA-64, do 16614 nothing. 16615 16616 A trampoline is a small piece of code that is created at run time 16617 on the stack when the address of a nested function is taken, and is 16618 used to call the nested function indirectly. Therefore, it 16619 requires the stack to be made executable in order for the program 16620 to work properly. 16621 16622 '-fno-trampolines' is enabled by default on a language by language 16623 basis to let the compiler avoid generating them, if it computes 16624 that this is safe, and replace them with descriptors. Descriptors 16625 are made up of data only, but the generated code must be prepared 16626 to deal with them. As of this writing, '-fno-trampolines' is 16627 enabled by default only for Ada. 16628 16629 Moreover, code compiled with '-ftrampolines' and code compiled with 16630 '-fno-trampolines' are not binary compatible if nested functions 16631 are present. This option must therefore be used on a program-wide 16632 basis and be manipulated with extreme care. 16633 16634 For languages other than Ada, the '-ftrampolines' and 16635 '-fno-trampolines' options currently have no effect, and 16636 trampolines are always generated on platforms that need them for 16637 nested functions. 16638 16639'-fvisibility=[default|internal|hidden|protected]' 16640 Set the default ELF image symbol visibility to the specified 16641 option--all symbols are marked with this unless overridden within 16642 the code. Using this feature can very substantially improve 16643 linking and load times of shared object libraries, produce more 16644 optimized code, provide near-perfect API export and prevent symbol 16645 clashes. It is *strongly* recommended that you use this in any 16646 shared objects you distribute. 16647 16648 Despite the nomenclature, 'default' always means public; i.e., 16649 available to be linked against from outside the shared object. 16650 'protected' and 'internal' are pretty useless in real-world usage 16651 so the only other commonly used option is 'hidden'. The default if 16652 '-fvisibility' isn't specified is 'default', i.e., make every 16653 symbol public. 16654 16655 A good explanation of the benefits offered by ensuring ELF symbols 16656 have the correct visibility is given by "How To Write Shared 16657 Libraries" by Ulrich Drepper (which can be found at 16658 <https://www.akkadia.org/drepper/>)--however a superior solution 16659 made possible by this option to marking things hidden when the 16660 default is public is to make the default hidden and mark things 16661 public. This is the norm with DLLs on Windows and with 16662 '-fvisibility=hidden' and '__attribute__ ((visibility("default")))' 16663 instead of '__declspec(dllexport)' you get almost identical 16664 semantics with identical syntax. This is a great boon to those 16665 working with cross-platform projects. 16666 16667 For those adding visibility support to existing code, you may find 16668 '#pragma GCC visibility' of use. This works by you enclosing the 16669 declarations you wish to set visibility for with (for example) 16670 '#pragma GCC visibility push(hidden)' and '#pragma GCC visibility 16671 pop'. Bear in mind that symbol visibility should be viewed *as 16672 part of the API interface contract* and thus all new code should 16673 always specify visibility when it is not the default; i.e., 16674 declarations only for use within the local DSO should *always* be 16675 marked explicitly as hidden as so to avoid PLT indirection 16676 overheads--making this abundantly clear also aids readability and 16677 self-documentation of the code. Note that due to ISO C++ 16678 specification requirements, 'operator new' and 'operator delete' 16679 must always be of default visibility. 16680 16681 Be aware that headers from outside your project, in particular 16682 system headers and headers from any other library you use, may not 16683 be expecting to be compiled with visibility other than the default. 16684 You may need to explicitly say '#pragma GCC visibility 16685 push(default)' before including any such headers. 16686 16687 'extern' declarations are not affected by '-fvisibility', so a lot 16688 of code can be recompiled with '-fvisibility=hidden' with no 16689 modifications. However, this means that calls to 'extern' 16690 functions with no explicit visibility use the PLT, so it is more 16691 effective to use '__attribute ((visibility))' and/or '#pragma GCC 16692 visibility' to tell the compiler which 'extern' declarations should 16693 be treated as hidden. 16694 16695 Note that '-fvisibility' does affect C++ vague linkage entities. 16696 This means that, for instance, an exception class that is be thrown 16697 between DSOs must be explicitly marked with default visibility so 16698 that the 'type_info' nodes are unified between the DSOs. 16699 16700 An overview of these techniques, their benefits and how to use them 16701 is at <https://gcc.gnu.org/wiki/Visibility>. 16702 16703'-fstrict-volatile-bitfields' 16704 This option should be used if accesses to volatile bit-fields (or 16705 other structure fields, although the compiler usually honors those 16706 types anyway) should use a single access of the width of the 16707 field's type, aligned to a natural alignment if possible. For 16708 example, targets with memory-mapped peripheral registers might 16709 require all such accesses to be 16 bits wide; with this flag you 16710 can declare all peripheral bit-fields as 'unsigned short' (assuming 16711 short is 16 bits on these targets) to force GCC to use 16-bit 16712 accesses instead of, perhaps, a more efficient 32-bit access. 16713 16714 If this option is disabled, the compiler uses the most efficient 16715 instruction. In the previous example, that might be a 32-bit load 16716 instruction, even though that accesses bytes that do not contain 16717 any portion of the bit-field, or memory-mapped registers unrelated 16718 to the one being updated. 16719 16720 In some cases, such as when the 'packed' attribute is applied to a 16721 structure field, it may not be possible to access the field with a 16722 single read or write that is correctly aligned for the target 16723 machine. In this case GCC falls back to generating multiple 16724 accesses rather than code that will fault or truncate the result at 16725 run time. 16726 16727 Note: Due to restrictions of the C/C++11 memory model, write 16728 accesses are not allowed to touch non bit-field members. It is 16729 therefore recommended to define all bits of the field's type as 16730 bit-field members. 16731 16732 The default value of this option is determined by the application 16733 binary interface for the target processor. 16734 16735'-fsync-libcalls' 16736 This option controls whether any out-of-line instance of the 16737 '__sync' family of functions may be used to implement the C++11 16738 '__atomic' family of functions. 16739 16740 The default value of this option is enabled, thus the only useful 16741 form of the option is '-fno-sync-libcalls'. This option is used in 16742 the implementation of the 'libatomic' runtime library. 16743 16744 16745File: gcc.info, Node: Developer Options, Next: Submodel Options, Prev: Code Gen Options, Up: Invoking GCC 16746 167473.18 GCC Developer Options 16748========================== 16749 16750This section describes command-line options that are primarily of 16751interest to GCC developers, including options to support compiler 16752testing and investigation of compiler bugs and compile-time performance 16753problems. This includes options that produce debug dumps at various 16754points in the compilation; that print statistics such as memory use and 16755execution time; and that print information about GCC's configuration, 16756such as where it searches for libraries. You should rarely need to use 16757any of these options for ordinary compilation and linking tasks. 16758 16759 Many developer options that cause GCC to dump output to a file take an 16760optional '=FILENAME' suffix. You can specify 'stdout' or '-' to dump to 16761standard output, and 'stderr' for standard error. 16762 16763 If '=FILENAME' is omitted, a default dump file name is constructed by 16764concatenating the base dump file name, a pass number, phase letter, and 16765pass name. The base dump file name is the name of output file produced 16766by the compiler if explicitly specified and not an executable; otherwise 16767it is the source file name. The pass number is determined by the order 16768passes are registered with the compiler's pass manager. This is 16769generally the same as the order of execution, but passes registered by 16770plugins, target-specific passes, or passes that are otherwise registered 16771late are numbered higher than the pass named 'final', even if they are 16772executed earlier. The phase letter is one of 'i' (inter-procedural 16773analysis), 'l' (language-specific), 'r' (RTL), or 't' (tree). The files 16774are created in the directory of the output file. 16775 16776'-fcallgraph-info' 16777'-fcallgraph-info=MARKERS' 16778 Makes the compiler output callgraph information for the program, on 16779 a per-object-file basis. The information is generated in the 16780 common VCG format. It can be decorated with additional, per-node 16781 and/or per-edge information, if a list of comma-separated markers 16782 is additionally specified. When the 'su' marker is specified, the 16783 callgraph is decorated with stack usage information; it is 16784 equivalent to '-fstack-usage'. When the 'da' marker is specified, 16785 the callgraph is decorated with information about dynamically 16786 allocated objects. 16787 16788 When compiling with '-flto', no callgraph information is output 16789 along with the object file. At LTO link time, '-fcallgraph-info' 16790 may generate multiple callgraph information files next to 16791 intermediate LTO output files. 16792 16793'-dLETTERS' 16794'-fdump-rtl-PASS' 16795'-fdump-rtl-PASS=FILENAME' 16796 Says to make debugging dumps during compilation at times specified 16797 by LETTERS. This is used for debugging the RTL-based passes of the 16798 compiler. 16799 16800 Some '-dLETTERS' switches have different meaning when '-E' is used 16801 for preprocessing. *Note Preprocessor Options::, for information 16802 about preprocessor-specific dump options. 16803 16804 Debug dumps can be enabled with a '-fdump-rtl' switch or some '-d' 16805 option LETTERS. Here are the possible letters for use in PASS and 16806 LETTERS, and their meanings: 16807 16808 '-fdump-rtl-alignments' 16809 Dump after branch alignments have been computed. 16810 16811 '-fdump-rtl-asmcons' 16812 Dump after fixing rtl statements that have unsatisfied in/out 16813 constraints. 16814 16815 '-fdump-rtl-auto_inc_dec' 16816 Dump after auto-inc-dec discovery. This pass is only run on 16817 architectures that have auto inc or auto dec instructions. 16818 16819 '-fdump-rtl-barriers' 16820 Dump after cleaning up the barrier instructions. 16821 16822 '-fdump-rtl-bbpart' 16823 Dump after partitioning hot and cold basic blocks. 16824 16825 '-fdump-rtl-bbro' 16826 Dump after block reordering. 16827 16828 '-fdump-rtl-btl1' 16829 '-fdump-rtl-btl2' 16830 '-fdump-rtl-btl1' and '-fdump-rtl-btl2' enable dumping after 16831 the two branch target load optimization passes. 16832 16833 '-fdump-rtl-bypass' 16834 Dump after jump bypassing and control flow optimizations. 16835 16836 '-fdump-rtl-combine' 16837 Dump after the RTL instruction combination pass. 16838 16839 '-fdump-rtl-compgotos' 16840 Dump after duplicating the computed gotos. 16841 16842 '-fdump-rtl-ce1' 16843 '-fdump-rtl-ce2' 16844 '-fdump-rtl-ce3' 16845 '-fdump-rtl-ce1', '-fdump-rtl-ce2', and '-fdump-rtl-ce3' 16846 enable dumping after the three if conversion passes. 16847 16848 '-fdump-rtl-cprop_hardreg' 16849 Dump after hard register copy propagation. 16850 16851 '-fdump-rtl-csa' 16852 Dump after combining stack adjustments. 16853 16854 '-fdump-rtl-cse1' 16855 '-fdump-rtl-cse2' 16856 '-fdump-rtl-cse1' and '-fdump-rtl-cse2' enable dumping after 16857 the two common subexpression elimination passes. 16858 16859 '-fdump-rtl-dce' 16860 Dump after the standalone dead code elimination passes. 16861 16862 '-fdump-rtl-dbr' 16863 Dump after delayed branch scheduling. 16864 16865 '-fdump-rtl-dce1' 16866 '-fdump-rtl-dce2' 16867 '-fdump-rtl-dce1' and '-fdump-rtl-dce2' enable dumping after 16868 the two dead store elimination passes. 16869 16870 '-fdump-rtl-eh' 16871 Dump after finalization of EH handling code. 16872 16873 '-fdump-rtl-eh_ranges' 16874 Dump after conversion of EH handling range regions. 16875 16876 '-fdump-rtl-expand' 16877 Dump after RTL generation. 16878 16879 '-fdump-rtl-fwprop1' 16880 '-fdump-rtl-fwprop2' 16881 '-fdump-rtl-fwprop1' and '-fdump-rtl-fwprop2' enable dumping 16882 after the two forward propagation passes. 16883 16884 '-fdump-rtl-gcse1' 16885 '-fdump-rtl-gcse2' 16886 '-fdump-rtl-gcse1' and '-fdump-rtl-gcse2' enable dumping after 16887 global common subexpression elimination. 16888 16889 '-fdump-rtl-init-regs' 16890 Dump after the initialization of the registers. 16891 16892 '-fdump-rtl-initvals' 16893 Dump after the computation of the initial value sets. 16894 16895 '-fdump-rtl-into_cfglayout' 16896 Dump after converting to cfglayout mode. 16897 16898 '-fdump-rtl-ira' 16899 Dump after iterated register allocation. 16900 16901 '-fdump-rtl-jump' 16902 Dump after the second jump optimization. 16903 16904 '-fdump-rtl-loop2' 16905 '-fdump-rtl-loop2' enables dumping after the rtl loop 16906 optimization passes. 16907 16908 '-fdump-rtl-mach' 16909 Dump after performing the machine dependent reorganization 16910 pass, if that pass exists. 16911 16912 '-fdump-rtl-mode_sw' 16913 Dump after removing redundant mode switches. 16914 16915 '-fdump-rtl-rnreg' 16916 Dump after register renumbering. 16917 16918 '-fdump-rtl-outof_cfglayout' 16919 Dump after converting from cfglayout mode. 16920 16921 '-fdump-rtl-peephole2' 16922 Dump after the peephole pass. 16923 16924 '-fdump-rtl-postreload' 16925 Dump after post-reload optimizations. 16926 16927 '-fdump-rtl-pro_and_epilogue' 16928 Dump after generating the function prologues and epilogues. 16929 16930 '-fdump-rtl-sched1' 16931 '-fdump-rtl-sched2' 16932 '-fdump-rtl-sched1' and '-fdump-rtl-sched2' enable dumping 16933 after the basic block scheduling passes. 16934 16935 '-fdump-rtl-ree' 16936 Dump after sign/zero extension elimination. 16937 16938 '-fdump-rtl-seqabstr' 16939 Dump after common sequence discovery. 16940 16941 '-fdump-rtl-shorten' 16942 Dump after shortening branches. 16943 16944 '-fdump-rtl-sibling' 16945 Dump after sibling call optimizations. 16946 16947 '-fdump-rtl-split1' 16948 '-fdump-rtl-split2' 16949 '-fdump-rtl-split3' 16950 '-fdump-rtl-split4' 16951 '-fdump-rtl-split5' 16952 These options enable dumping after five rounds of instruction 16953 splitting. 16954 16955 '-fdump-rtl-sms' 16956 Dump after modulo scheduling. This pass is only run on some 16957 architectures. 16958 16959 '-fdump-rtl-stack' 16960 Dump after conversion from GCC's "flat register file" 16961 registers to the x87's stack-like registers. This pass is 16962 only run on x86 variants. 16963 16964 '-fdump-rtl-subreg1' 16965 '-fdump-rtl-subreg2' 16966 '-fdump-rtl-subreg1' and '-fdump-rtl-subreg2' enable dumping 16967 after the two subreg expansion passes. 16968 16969 '-fdump-rtl-unshare' 16970 Dump after all rtl has been unshared. 16971 16972 '-fdump-rtl-vartrack' 16973 Dump after variable tracking. 16974 16975 '-fdump-rtl-vregs' 16976 Dump after converting virtual registers to hard registers. 16977 16978 '-fdump-rtl-web' 16979 Dump after live range splitting. 16980 16981 '-fdump-rtl-regclass' 16982 '-fdump-rtl-subregs_of_mode_init' 16983 '-fdump-rtl-subregs_of_mode_finish' 16984 '-fdump-rtl-dfinit' 16985 '-fdump-rtl-dfinish' 16986 These dumps are defined but always produce empty files. 16987 16988 '-da' 16989 '-fdump-rtl-all' 16990 Produce all the dumps listed above. 16991 16992 '-dA' 16993 Annotate the assembler output with miscellaneous debugging 16994 information. 16995 16996 '-dD' 16997 Dump all macro definitions, at the end of preprocessing, in 16998 addition to normal output. 16999 17000 '-dH' 17001 Produce a core dump whenever an error occurs. 17002 17003 '-dp' 17004 Annotate the assembler output with a comment indicating which 17005 pattern and alternative is used. The length and cost of each 17006 instruction are also printed. 17007 17008 '-dP' 17009 Dump the RTL in the assembler output as a comment before each 17010 instruction. Also turns on '-dp' annotation. 17011 17012 '-dx' 17013 Just generate RTL for a function instead of compiling it. 17014 Usually used with '-fdump-rtl-expand'. 17015 17016'-fdump-debug' 17017 Dump debugging information generated during the debug generation 17018 phase. 17019 17020'-fdump-earlydebug' 17021 Dump debugging information generated during the early debug 17022 generation phase. 17023 17024'-fdump-noaddr' 17025 When doing debugging dumps, suppress address output. This makes it 17026 more feasible to use diff on debugging dumps for compiler 17027 invocations with different compiler binaries and/or different text 17028 / bss / data / heap / stack / dso start locations. 17029 17030'-freport-bug' 17031 Collect and dump debug information into a temporary file if an 17032 internal compiler error (ICE) occurs. 17033 17034'-fdump-unnumbered' 17035 When doing debugging dumps, suppress instruction numbers and 17036 address output. This makes it more feasible to use diff on 17037 debugging dumps for compiler invocations with different options, in 17038 particular with and without '-g'. 17039 17040'-fdump-unnumbered-links' 17041 When doing debugging dumps (see '-d' option above), suppress 17042 instruction numbers for the links to the previous and next 17043 instructions in a sequence. 17044 17045'-fdump-ipa-SWITCH' 17046'-fdump-ipa-SWITCH-OPTIONS' 17047 Control the dumping at various stages of inter-procedural analysis 17048 language tree to a file. The file name is generated by appending a 17049 switch specific suffix to the source file name, and the file is 17050 created in the same directory as the output file. The following 17051 dumps are possible: 17052 17053 'all' 17054 Enables all inter-procedural analysis dumps. 17055 17056 'cgraph' 17057 Dumps information about call-graph optimization, unused 17058 function removal, and inlining decisions. 17059 17060 'inline' 17061 Dump after function inlining. 17062 17063 Additionally, the options '-optimized', '-missed', '-note', and 17064 '-all' can be provided, with the same meaning as for '-fopt-info', 17065 defaulting to '-optimized'. 17066 17067 For example, '-fdump-ipa-inline-optimized-missed' will emit 17068 information on callsites that were inlined, along with callsites 17069 that were not inlined. 17070 17071 By default, the dump will contain messages about successful 17072 optimizations (equivalent to '-optimized') together with low-level 17073 details about the analysis. 17074 17075'-fdump-lang' 17076 Dump language-specific information. The file name is made by 17077 appending '.lang' to the source file name. 17078 17079'-fdump-lang-all' 17080'-fdump-lang-SWITCH' 17081'-fdump-lang-SWITCH-OPTIONS' 17082'-fdump-lang-SWITCH-OPTIONS=FILENAME' 17083 Control the dumping of language-specific information. The OPTIONS 17084 and FILENAME portions behave as described in the '-fdump-tree' 17085 option. The following SWITCH values are accepted: 17086 17087 'all' 17088 17089 Enable all language-specific dumps. 17090 17091 'class' 17092 Dump class hierarchy information. Virtual table information 17093 is emitted unless ''slim'' is specified. This option is 17094 applicable to C++ only. 17095 17096 'module' 17097 Dump module information. Options 'lineno' (locations), 17098 'graph' (reachability), 'blocks' (clusters), 'uid' 17099 (serialization), 'alias' (mergeable), 'asmname' (Elrond), 'eh' 17100 (mapper) & 'vops' (macros) may provide additional information. 17101 This option is applicable to C++ only. 17102 17103 'raw' 17104 Dump the raw internal tree data. This option is applicable to 17105 C++ only. 17106 17107'-fdump-passes' 17108 Print on 'stderr' the list of optimization passes that are turned 17109 on and off by the current command-line options. 17110 17111'-fdump-statistics-OPTION' 17112 Enable and control dumping of pass statistics in a separate file. 17113 The file name is generated by appending a suffix ending in 17114 '.statistics' to the source file name, and the file is created in 17115 the same directory as the output file. If the '-OPTION' form is 17116 used, '-stats' causes counters to be summed over the whole 17117 compilation unit while '-details' dumps every event as the passes 17118 generate them. The default with no option is to sum counters for 17119 each function compiled. 17120 17121'-fdump-tree-all' 17122'-fdump-tree-SWITCH' 17123'-fdump-tree-SWITCH-OPTIONS' 17124'-fdump-tree-SWITCH-OPTIONS=FILENAME' 17125 Control the dumping at various stages of processing the 17126 intermediate language tree to a file. If the '-OPTIONS' form is 17127 used, OPTIONS is a list of '-' separated options which control the 17128 details of the dump. Not all options are applicable to all dumps; 17129 those that are not meaningful are ignored. The following options 17130 are available 17131 17132 'address' 17133 Print the address of each node. Usually this is not 17134 meaningful as it changes according to the environment and 17135 source file. Its primary use is for tying up a dump file with 17136 a debug environment. 17137 'asmname' 17138 If 'DECL_ASSEMBLER_NAME' has been set for a given decl, use 17139 that in the dump instead of 'DECL_NAME'. Its primary use is 17140 ease of use working backward from mangled names in the 17141 assembly file. 17142 'slim' 17143 When dumping front-end intermediate representations, inhibit 17144 dumping of members of a scope or body of a function merely 17145 because that scope has been reached. Only dump such items 17146 when they are directly reachable by some other path. 17147 17148 When dumping pretty-printed trees, this option inhibits 17149 dumping the bodies of control structures. 17150 17151 When dumping RTL, print the RTL in slim (condensed) form 17152 instead of the default LISP-like representation. 17153 'raw' 17154 Print a raw representation of the tree. By default, trees are 17155 pretty-printed into a C-like representation. 17156 'details' 17157 Enable more detailed dumps (not honored by every dump option). 17158 Also include information from the optimization passes. 17159 'stats' 17160 Enable dumping various statistics about the pass (not honored 17161 by every dump option). 17162 'blocks' 17163 Enable showing basic block boundaries (disabled in raw dumps). 17164 'graph' 17165 For each of the other indicated dump files 17166 ('-fdump-rtl-PASS'), dump a representation of the control flow 17167 graph suitable for viewing with GraphViz to 17168 'FILE.PASSID.PASS.dot'. Each function in the file is 17169 pretty-printed as a subgraph, so that GraphViz can render them 17170 all in a single plot. 17171 17172 This option currently only works for RTL dumps, and the RTL is 17173 always dumped in slim form. 17174 'vops' 17175 Enable showing virtual operands for every statement. 17176 'lineno' 17177 Enable showing line numbers for statements. 17178 'uid' 17179 Enable showing the unique ID ('DECL_UID') for each variable. 17180 'verbose' 17181 Enable showing the tree dump for each statement. 17182 'eh' 17183 Enable showing the EH region number holding each statement. 17184 'scev' 17185 Enable showing scalar evolution analysis details. 17186 'optimized' 17187 Enable showing optimization information (only available in 17188 certain passes). 17189 'missed' 17190 Enable showing missed optimization information (only available 17191 in certain passes). 17192 'note' 17193 Enable other detailed optimization information (only available 17194 in certain passes). 17195 'all' 17196 Turn on all options, except 'raw', 'slim', 'verbose' and 17197 'lineno'. 17198 'optall' 17199 Turn on all optimization options, i.e., 'optimized', 'missed', 17200 and 'note'. 17201 17202 To determine what tree dumps are available or find the dump for a 17203 pass of interest follow the steps below. 17204 17205 1. Invoke GCC with '-fdump-passes' and in the 'stderr' output 17206 look for a code that corresponds to the pass you are 17207 interested in. For example, the codes 'tree-evrp', 17208 'tree-vrp1', and 'tree-vrp2' correspond to the three Value 17209 Range Propagation passes. The number at the end distinguishes 17210 distinct invocations of the same pass. 17211 2. To enable the creation of the dump file, append the pass code 17212 to the '-fdump-' option prefix and invoke GCC with it. For 17213 example, to enable the dump from the Early Value Range 17214 Propagation pass, invoke GCC with the '-fdump-tree-evrp' 17215 option. Optionally, you may specify the name of the dump 17216 file. If you don't specify one, GCC creates as described 17217 below. 17218 3. Find the pass dump in a file whose name is composed of three 17219 components separated by a period: the name of the source file 17220 GCC was invoked to compile, a numeric suffix indicating the 17221 pass number followed by the letter 't' for tree passes (and 17222 the letter 'r' for RTL passes), and finally the pass code. 17223 For example, the Early VRP pass dump might be in a file named 17224 'myfile.c.038t.evrp' in the current working directory. Note 17225 that the numeric codes are not stable and may change from one 17226 version of GCC to another. 17227 17228'-fopt-info' 17229'-fopt-info-OPTIONS' 17230'-fopt-info-OPTIONS=FILENAME' 17231 Controls optimization dumps from various optimization passes. If 17232 the '-OPTIONS' form is used, OPTIONS is a list of '-' separated 17233 option keywords to select the dump details and optimizations. 17234 17235 The OPTIONS can be divided into three groups: 17236 1. options describing what kinds of messages should be emitted, 17237 2. options describing the verbosity of the dump, and 17238 3. options describing which optimizations should be included. 17239 The options from each group can be freely mixed as they are 17240 non-overlapping. However, in case of any conflicts, the later 17241 options override the earlier options on the command line. 17242 17243 The following options control which kinds of messages should be 17244 emitted: 17245 17246 'optimized' 17247 Print information when an optimization is successfully 17248 applied. It is up to a pass to decide which information is 17249 relevant. For example, the vectorizer passes print the source 17250 location of loops which are successfully vectorized. 17251 'missed' 17252 Print information about missed optimizations. Individual 17253 passes control which information to include in the output. 17254 'note' 17255 Print verbose information about optimizations, such as certain 17256 transformations, more detailed messages about decisions etc. 17257 'all' 17258 Print detailed optimization information. This includes 17259 'optimized', 'missed', and 'note'. 17260 17261 The following option controls the dump verbosity: 17262 17263 'internals' 17264 By default, only "high-level" messages are emitted. This 17265 option enables additional, more detailed, messages, which are 17266 likely to only be of interest to GCC developers. 17267 17268 One or more of the following option keywords can be used to 17269 describe a group of optimizations: 17270 17271 'ipa' 17272 Enable dumps from all interprocedural optimizations. 17273 'loop' 17274 Enable dumps from all loop optimizations. 17275 'inline' 17276 Enable dumps from all inlining optimizations. 17277 'omp' 17278 Enable dumps from all OMP (Offloading and Multi Processing) 17279 optimizations. 17280 'vec' 17281 Enable dumps from all vectorization optimizations. 17282 'optall' 17283 Enable dumps from all optimizations. This is a superset of 17284 the optimization groups listed above. 17285 17286 If OPTIONS is omitted, it defaults to 'optimized-optall', which 17287 means to dump messages about successful optimizations from all the 17288 passes, omitting messages that are treated as "internals". 17289 17290 If the FILENAME is provided, then the dumps from all the applicable 17291 optimizations are concatenated into the FILENAME. Otherwise the 17292 dump is output onto 'stderr'. Though multiple '-fopt-info' options 17293 are accepted, only one of them can include a FILENAME. If other 17294 filenames are provided then all but the first such option are 17295 ignored. 17296 17297 Note that the output FILENAME is overwritten in case of multiple 17298 translation units. If a combined output from multiple translation 17299 units is desired, 'stderr' should be used instead. 17300 17301 In the following example, the optimization info is output to 17302 'stderr': 17303 17304 gcc -O3 -fopt-info 17305 17306 This example: 17307 gcc -O3 -fopt-info-missed=missed.all 17308 17309 outputs missed optimization report from all the passes into 17310 'missed.all', and this one: 17311 17312 gcc -O2 -ftree-vectorize -fopt-info-vec-missed 17313 17314 prints information about missed optimization opportunities from 17315 vectorization passes on 'stderr'. Note that 17316 '-fopt-info-vec-missed' is equivalent to '-fopt-info-missed-vec'. 17317 The order of the optimization group names and message types listed 17318 after '-fopt-info' does not matter. 17319 17320 As another example, 17321 gcc -O3 -fopt-info-inline-optimized-missed=inline.txt 17322 17323 outputs information about missed optimizations as well as optimized 17324 locations from all the inlining passes into 'inline.txt'. 17325 17326 Finally, consider: 17327 17328 gcc -fopt-info-vec-missed=vec.miss -fopt-info-loop-optimized=loop.opt 17329 17330 Here the two output filenames 'vec.miss' and 'loop.opt' are in 17331 conflict since only one output file is allowed. In this case, only 17332 the first option takes effect and the subsequent options are 17333 ignored. Thus only 'vec.miss' is produced which contains dumps 17334 from the vectorizer about missed opportunities. 17335 17336'-fsave-optimization-record' 17337 Write a SRCFILE.opt-record.json.gz file detailing what 17338 optimizations were performed, for those optimizations that support 17339 '-fopt-info'. 17340 17341 This option is experimental and the format of the data within the 17342 compressed JSON file is subject to change. 17343 17344 It is roughly equivalent to a machine-readable version of 17345 '-fopt-info-all', as a collection of messages with source file, 17346 line number and column number, with the following additional data 17347 for each message: 17348 17349 * the execution count of the code being optimized, along with 17350 metadata about whether this was from actual profile data, or 17351 just an estimate, allowing consumers to prioritize messages by 17352 code hotness, 17353 17354 * the function name of the code being optimized, where 17355 applicable, 17356 17357 * the "inlining chain" for the code being optimized, so that 17358 when a function is inlined into several different places 17359 (which might themselves be inlined), the reader can 17360 distinguish between the copies, 17361 17362 * objects identifying those parts of the message that refer to 17363 expressions, statements or symbol-table nodes, which of these 17364 categories they are, and, when available, their source code 17365 location, 17366 17367 * the GCC pass that emitted the message, and 17368 17369 * the location in GCC's own code from which the message was 17370 emitted 17371 17372 Additionally, some messages are logically nested within other 17373 messages, reflecting implementation details of the optimization 17374 passes. 17375 17376'-fsched-verbose=N' 17377 On targets that use instruction scheduling, this option controls 17378 the amount of debugging output the scheduler prints to the dump 17379 files. 17380 17381 For N greater than zero, '-fsched-verbose' outputs the same 17382 information as '-fdump-rtl-sched1' and '-fdump-rtl-sched2'. For N 17383 greater than one, it also output basic block probabilities, 17384 detailed ready list information and unit/insn info. For N greater 17385 than two, it includes RTL at abort point, control-flow and regions 17386 info. And for N over four, '-fsched-verbose' also includes 17387 dependence info. 17388 17389'-fenable-KIND-PASS' 17390'-fdisable-KIND-PASS=RANGE-LIST' 17391 17392 This is a set of options that are used to explicitly disable/enable 17393 optimization passes. These options are intended for use for 17394 debugging GCC. Compiler users should use regular options for 17395 enabling/disabling passes instead. 17396 17397 '-fdisable-ipa-PASS' 17398 Disable IPA pass PASS. PASS is the pass name. If the same 17399 pass is statically invoked in the compiler multiple times, the 17400 pass name should be appended with a sequential number starting 17401 from 1. 17402 17403 '-fdisable-rtl-PASS' 17404 '-fdisable-rtl-PASS=RANGE-LIST' 17405 Disable RTL pass PASS. PASS is the pass name. If the same 17406 pass is statically invoked in the compiler multiple times, the 17407 pass name should be appended with a sequential number starting 17408 from 1. RANGE-LIST is a comma-separated list of function 17409 ranges or assembler names. Each range is a number pair 17410 separated by a colon. The range is inclusive in both ends. 17411 If the range is trivial, the number pair can be simplified as 17412 a single number. If the function's call graph node's UID 17413 falls within one of the specified ranges, the PASS is disabled 17414 for that function. The UID is shown in the function header of 17415 a dump file, and the pass names can be dumped by using option 17416 '-fdump-passes'. 17417 17418 '-fdisable-tree-PASS' 17419 '-fdisable-tree-PASS=RANGE-LIST' 17420 Disable tree pass PASS. See '-fdisable-rtl' for the 17421 description of option arguments. 17422 17423 '-fenable-ipa-PASS' 17424 Enable IPA pass PASS. PASS is the pass name. If the same 17425 pass is statically invoked in the compiler multiple times, the 17426 pass name should be appended with a sequential number starting 17427 from 1. 17428 17429 '-fenable-rtl-PASS' 17430 '-fenable-rtl-PASS=RANGE-LIST' 17431 Enable RTL pass PASS. See '-fdisable-rtl' for option argument 17432 description and examples. 17433 17434 '-fenable-tree-PASS' 17435 '-fenable-tree-PASS=RANGE-LIST' 17436 Enable tree pass PASS. See '-fdisable-rtl' for the 17437 description of option arguments. 17438 17439 Here are some examples showing uses of these options. 17440 17441 17442 # disable ccp1 for all functions 17443 -fdisable-tree-ccp1 17444 # disable complete unroll for function whose cgraph node uid is 1 17445 -fenable-tree-cunroll=1 17446 # disable gcse2 for functions at the following ranges [1,1], 17447 # [300,400], and [400,1000] 17448 # disable gcse2 for functions foo and foo2 17449 -fdisable-rtl-gcse2=foo,foo2 17450 # disable early inlining 17451 -fdisable-tree-einline 17452 # disable ipa inlining 17453 -fdisable-ipa-inline 17454 # enable tree full unroll 17455 -fenable-tree-unroll 17456 17457 17458'-fchecking' 17459'-fchecking=N' 17460 Enable internal consistency checking. The default depends on the 17461 compiler configuration. '-fchecking=2' enables further internal 17462 consistency checking that might affect code generation. 17463 17464'-frandom-seed=STRING' 17465 This option provides a seed that GCC uses in place of random 17466 numbers in generating certain symbol names that have to be 17467 different in every compiled file. It is also used to place unique 17468 stamps in coverage data files and the object files that produce 17469 them. You can use the '-frandom-seed' option to produce 17470 reproducibly identical object files. 17471 17472 The STRING can either be a number (decimal, octal or hex) or an 17473 arbitrary string (in which case it's converted to a number by 17474 computing CRC32). 17475 17476 The STRING should be different for every file you compile. 17477 17478'-save-temps' 17479 Store the usual "temporary" intermediate files permanently; name 17480 them as auxiliary output files, as specified described under 17481 '-dumpbase' and '-dumpdir'. 17482 17483 When used in combination with the '-x' command-line option, 17484 '-save-temps' is sensible enough to avoid overwriting an input 17485 source file with the same extension as an intermediate file. The 17486 corresponding intermediate file may be obtained by renaming the 17487 source file before using '-save-temps'. 17488 17489'-save-temps=cwd' 17490 Equivalent to '-save-temps -dumpdir ./'. 17491 17492'-save-temps=obj' 17493 Equivalent to '-save-temps -dumpdir outdir/', where 'outdir/' is 17494 the directory of the output file specified after the '-o' option, 17495 including any directory separators. If the '-o' option is not 17496 used, the '-save-temps=obj' switch behaves like '-save-temps=cwd'. 17497 17498'-time[=FILE]' 17499 Report the CPU time taken by each subprocess in the compilation 17500 sequence. For C source files, this is the compiler proper and 17501 assembler (plus the linker if linking is done). 17502 17503 Without the specification of an output file, the output looks like 17504 this: 17505 17506 # cc1 0.12 0.01 17507 # as 0.00 0.01 17508 17509 The first number on each line is the "user time", that is time 17510 spent executing the program itself. The second number is "system 17511 time", time spent executing operating system routines on behalf of 17512 the program. Both numbers are in seconds. 17513 17514 With the specification of an output file, the output is appended to 17515 the named file, and it looks like this: 17516 17517 0.12 0.01 cc1 OPTIONS 17518 0.00 0.01 as OPTIONS 17519 17520 The "user time" and the "system time" are moved before the program 17521 name, and the options passed to the program are displayed, so that 17522 one can later tell what file was being compiled, and with which 17523 options. 17524 17525'-fdump-final-insns[=FILE]' 17526 Dump the final internal representation (RTL) to FILE. If the 17527 optional argument is omitted (or if FILE is '.'), the name of the 17528 dump file is determined by appending '.gkd' to the dump base name, 17529 see '-dumpbase'. 17530 17531'-fcompare-debug[=OPTS]' 17532 If no error occurs during compilation, run the compiler a second 17533 time, adding OPTS and '-fcompare-debug-second' to the arguments 17534 passed to the second compilation. Dump the final internal 17535 representation in both compilations, and print an error if they 17536 differ. 17537 17538 If the equal sign is omitted, the default '-gtoggle' is used. 17539 17540 The environment variable 'GCC_COMPARE_DEBUG', if defined, non-empty 17541 and nonzero, implicitly enables '-fcompare-debug'. If 17542 'GCC_COMPARE_DEBUG' is defined to a string starting with a dash, 17543 then it is used for OPTS, otherwise the default '-gtoggle' is used. 17544 17545 '-fcompare-debug=', with the equal sign but without OPTS, is 17546 equivalent to '-fno-compare-debug', which disables the dumping of 17547 the final representation and the second compilation, preventing 17548 even 'GCC_COMPARE_DEBUG' from taking effect. 17549 17550 To verify full coverage during '-fcompare-debug' testing, set 17551 'GCC_COMPARE_DEBUG' to say '-fcompare-debug-not-overridden', which 17552 GCC rejects as an invalid option in any actual compilation (rather 17553 than preprocessing, assembly or linking). To get just a warning, 17554 setting 'GCC_COMPARE_DEBUG' to '-w%n-fcompare-debug not overridden' 17555 will do. 17556 17557'-fcompare-debug-second' 17558 This option is implicitly passed to the compiler for the second 17559 compilation requested by '-fcompare-debug', along with options to 17560 silence warnings, and omitting other options that would cause the 17561 compiler to produce output to files or to standard output as a side 17562 effect. Dump files and preserved temporary files are renamed so as 17563 to contain the '.gk' additional extension during the second 17564 compilation, to avoid overwriting those generated by the first. 17565 17566 When this option is passed to the compiler driver, it causes the 17567 _first_ compilation to be skipped, which makes it useful for little 17568 other than debugging the compiler proper. 17569 17570'-gtoggle' 17571 Turn off generation of debug info, if leaving out this option 17572 generates it, or turn it on at level 2 otherwise. The position of 17573 this argument in the command line does not matter; it takes effect 17574 after all other options are processed, and it does so only once, no 17575 matter how many times it is given. This is mainly intended to be 17576 used with '-fcompare-debug'. 17577 17578'-fvar-tracking-assignments-toggle' 17579 Toggle '-fvar-tracking-assignments', in the same way that 17580 '-gtoggle' toggles '-g'. 17581 17582'-Q' 17583 Makes the compiler print out each function name as it is compiled, 17584 and print some statistics about each pass when it finishes. 17585 17586'-ftime-report' 17587 Makes the compiler print some statistics about the time consumed by 17588 each pass when it finishes. 17589 17590'-ftime-report-details' 17591 Record the time consumed by infrastructure parts separately for 17592 each pass. 17593 17594'-fira-verbose=N' 17595 Control the verbosity of the dump file for the integrated register 17596 allocator. The default value is 5. If the value N is greater or 17597 equal to 10, the dump output is sent to stderr using the same 17598 format as N minus 10. 17599 17600'-flto-report' 17601 Prints a report with internal details on the workings of the 17602 link-time optimizer. The contents of this report vary from version 17603 to version. It is meant to be useful to GCC developers when 17604 processing object files in LTO mode (via '-flto'). 17605 17606 Disabled by default. 17607 17608'-flto-report-wpa' 17609 Like '-flto-report', but only print for the WPA phase of link-time 17610 optimization. 17611 17612'-fmem-report' 17613 Makes the compiler print some statistics about permanent memory 17614 allocation when it finishes. 17615 17616'-fmem-report-wpa' 17617 Makes the compiler print some statistics about permanent memory 17618 allocation for the WPA phase only. 17619 17620'-fpre-ipa-mem-report' 17621'-fpost-ipa-mem-report' 17622 Makes the compiler print some statistics about permanent memory 17623 allocation before or after interprocedural optimization. 17624 17625'-fprofile-report' 17626 Makes the compiler print some statistics about consistency of the 17627 (estimated) profile and effect of individual passes. 17628 17629'-fstack-usage' 17630 Makes the compiler output stack usage information for the program, 17631 on a per-function basis. The filename for the dump is made by 17632 appending '.su' to the AUXNAME. AUXNAME is generated from the name 17633 of the output file, if explicitly specified and it is not an 17634 executable, otherwise it is the basename of the source file. An 17635 entry is made up of three fields: 17636 17637 * The name of the function. 17638 * A number of bytes. 17639 * One or more qualifiers: 'static', 'dynamic', 'bounded'. 17640 17641 The qualifier 'static' means that the function manipulates the 17642 stack statically: a fixed number of bytes are allocated for the 17643 frame on function entry and released on function exit; no stack 17644 adjustments are otherwise made in the function. The second field 17645 is this fixed number of bytes. 17646 17647 The qualifier 'dynamic' means that the function manipulates the 17648 stack dynamically: in addition to the static allocation described 17649 above, stack adjustments are made in the body of the function, for 17650 example to push/pop arguments around function calls. If the 17651 qualifier 'bounded' is also present, the amount of these 17652 adjustments is bounded at compile time and the second field is an 17653 upper bound of the total amount of stack used by the function. If 17654 it is not present, the amount of these adjustments is not bounded 17655 at compile time and the second field only represents the bounded 17656 part. 17657 17658'-fstats' 17659 Emit statistics about front-end processing at the end of the 17660 compilation. This option is supported only by the C++ front end, 17661 and the information is generally only useful to the G++ development 17662 team. 17663 17664'-fdbg-cnt-list' 17665 Print the name and the counter upper bound for all debug counters. 17666 17667'-fdbg-cnt=COUNTER-VALUE-LIST' 17668 Set the internal debug counter lower and upper bound. 17669 COUNTER-VALUE-LIST is a comma-separated list of 17670 NAME:LOWER_BOUND1-UPPER_BOUND1 [:LOWER_BOUND2-UPPER_BOUND2...] 17671 tuples which sets the name of the counter and list of closed 17672 intervals. The LOWER_BOUND is optional and is zero initialized if 17673 not set. For example, with '-fdbg-cnt=dce:2-4:10-11,tail_call:10', 17674 'dbg_cnt(dce)' returns true only for second, third, fourth, tenth 17675 and eleventh invocation. For 'dbg_cnt(tail_call)' true is returned 17676 for first 10 invocations. 17677 17678'-print-file-name=LIBRARY' 17679 Print the full absolute name of the library file LIBRARY that would 17680 be used when linking--and don't do anything else. With this 17681 option, GCC does not compile or link anything; it just prints the 17682 file name. 17683 17684'-print-multi-directory' 17685 Print the directory name corresponding to the multilib selected by 17686 any other switches present in the command line. This directory is 17687 supposed to exist in 'GCC_EXEC_PREFIX'. 17688 17689'-print-multi-lib' 17690 Print the mapping from multilib directory names to compiler 17691 switches that enable them. The directory name is separated from 17692 the switches by ';', and each switch starts with an '@' instead of 17693 the '-', without spaces between multiple switches. This is 17694 supposed to ease shell processing. 17695 17696'-print-multi-os-directory' 17697 Print the path to OS libraries for the selected multilib, relative 17698 to some 'lib' subdirectory. If OS libraries are present in the 17699 'lib' subdirectory and no multilibs are used, this is usually just 17700 '.', if OS libraries are present in 'libSUFFIX' sibling directories 17701 this prints e.g. '../lib64', '../lib' or '../lib32', or if OS 17702 libraries are present in 'lib/SUBDIR' subdirectories it prints e.g. 17703 'amd64', 'sparcv9' or 'ev6'. 17704 17705'-print-multiarch' 17706 Print the path to OS libraries for the selected multiarch, relative 17707 to some 'lib' subdirectory. 17708 17709'-print-prog-name=PROGRAM' 17710 Like '-print-file-name', but searches for a program such as 'cpp'. 17711 17712'-print-libgcc-file-name' 17713 Same as '-print-file-name=libgcc.a'. 17714 17715 This is useful when you use '-nostdlib' or '-nodefaultlibs' but you 17716 do want to link with 'libgcc.a'. You can do: 17717 17718 gcc -nostdlib FILES... `gcc -print-libgcc-file-name` 17719 17720'-print-search-dirs' 17721 Print the name of the configured installation directory and a list 17722 of program and library directories 'gcc' searches--and don't do 17723 anything else. 17724 17725 This is useful when 'gcc' prints the error message 'installation 17726 problem, cannot exec cpp0: No such file or directory'. To resolve 17727 this you either need to put 'cpp0' and the other compiler 17728 components where 'gcc' expects to find them, or you can set the 17729 environment variable 'GCC_EXEC_PREFIX' to the directory where you 17730 installed them. Don't forget the trailing '/'. *Note Environment 17731 Variables::. 17732 17733'-print-sysroot' 17734 Print the target sysroot directory that is used during compilation. 17735 This is the target sysroot specified either at configure time or 17736 using the '--sysroot' option, possibly with an extra suffix that 17737 depends on compilation options. If no target sysroot is specified, 17738 the option prints nothing. 17739 17740'-print-sysroot-headers-suffix' 17741 Print the suffix added to the target sysroot when searching for 17742 headers, or give an error if the compiler is not configured with 17743 such a suffix--and don't do anything else. 17744 17745'-dumpmachine' 17746 Print the compiler's target machine (for example, 17747 'i686-pc-linux-gnu')--and don't do anything else. 17748 17749'-dumpversion' 17750 Print the compiler version (for example, '3.0', '6.3.0' or 17751 '7')--and don't do anything else. This is the compiler version 17752 used in filesystem paths and specs. Depending on how the compiler 17753 has been configured it can be just a single number (major version), 17754 two numbers separated by a dot (major and minor version) or three 17755 numbers separated by dots (major, minor and patchlevel version). 17756 17757'-dumpfullversion' 17758 Print the full compiler version--and don't do anything else. The 17759 output is always three numbers separated by dots, major, minor and 17760 patchlevel version. 17761 17762'-dumpspecs' 17763 Print the compiler's built-in specs--and don't do anything else. 17764 (This is used when GCC itself is being built.) *Note Spec Files::. 17765 17766 17767File: gcc.info, Node: Submodel Options, Next: Spec Files, Prev: Developer Options, Up: Invoking GCC 17768 177693.19 Machine-Dependent Options 17770============================== 17771 17772Each target machine supported by GCC can have its own options--for 17773example, to allow you to compile for a particular processor variant or 17774ABI, or to control optimizations specific to that machine. By 17775convention, the names of machine-specific options start with '-m'. 17776 17777 Some configurations of the compiler also support additional 17778target-specific options, usually for compatibility with other compilers 17779on the same platform. 17780 17781* Menu: 17782 17783* AArch64 Options:: 17784* Adapteva Epiphany Options:: 17785* AMD GCN Options:: 17786* ARC Options:: 17787* ARM Options:: 17788* AVR Options:: 17789* Blackfin Options:: 17790* C6X Options:: 17791* CRIS Options:: 17792* CR16 Options:: 17793* C-SKY Options:: 17794* Darwin Options:: 17795* DEC Alpha Options:: 17796* eBPF Options:: 17797* FR30 Options:: 17798* FT32 Options:: 17799* FRV Options:: 17800* GNU/Linux Options:: 17801* H8/300 Options:: 17802* HPPA Options:: 17803* IA-64 Options:: 17804* LM32 Options:: 17805* LoongArch Options:: 17806* M32C Options:: 17807* M32R/D Options:: 17808* M680x0 Options:: 17809* MCore Options:: 17810* MeP Options:: 17811* MicroBlaze Options:: 17812* MIPS Options:: 17813* MMIX Options:: 17814* MN10300 Options:: 17815* Moxie Options:: 17816* MSP430 Options:: 17817* NDS32 Options:: 17818* Nios II Options:: 17819* Nvidia PTX Options:: 17820* OpenRISC Options:: 17821* PDP-11 Options:: 17822* picoChip Options:: 17823* PowerPC Options:: 17824* PRU Options:: 17825* RISC-V Options:: 17826* RL78 Options:: 17827* RS/6000 and PowerPC Options:: 17828* RX Options:: 17829* S/390 and zSeries Options:: 17830* Score Options:: 17831* SH Options:: 17832* Solaris 2 Options:: 17833* SPARC Options:: 17834* System V Options:: 17835* TILE-Gx Options:: 17836* TILEPro Options:: 17837* V850 Options:: 17838* VAX Options:: 17839* Visium Options:: 17840* VMS Options:: 17841* VxWorks Options:: 17842* x86 Options:: 17843* x86 Windows Options:: 17844* Xstormy16 Options:: 17845* Xtensa Options:: 17846* zSeries Options:: 17847 17848 17849File: gcc.info, Node: AArch64 Options, Next: Adapteva Epiphany Options, Up: Submodel Options 17850 178513.19.1 AArch64 Options 17852---------------------- 17853 17854These options are defined for AArch64 implementations: 17855 17856'-mabi=NAME' 17857 Generate code for the specified data model. Permissible values are 17858 'ilp32' for SysV-like data model where int, long int and pointers 17859 are 32 bits, and 'lp64' for SysV-like data model where int is 32 17860 bits, but long int and pointers are 64 bits. 17861 17862 The default depends on the specific target configuration. Note 17863 that the LP64 and ILP32 ABIs are not link-compatible; you must 17864 compile your entire program with the same ABI, and link with a 17865 compatible set of libraries. 17866 17867'-mbig-endian' 17868 Generate big-endian code. This is the default when GCC is 17869 configured for an 'aarch64_be-*-*' target. 17870 17871'-mgeneral-regs-only' 17872 Generate code which uses only the general-purpose registers. This 17873 will prevent the compiler from using floating-point and Advanced 17874 SIMD registers but will not impose any restrictions on the 17875 assembler. 17876 17877'-mlittle-endian' 17878 Generate little-endian code. This is the default when GCC is 17879 configured for an 'aarch64-*-*' but not an 'aarch64_be-*-*' target. 17880 17881'-mcmodel=tiny' 17882 Generate code for the tiny code model. The program and its 17883 statically defined symbols must be within 1MB of each other. 17884 Programs can be statically or dynamically linked. 17885 17886'-mcmodel=small' 17887 Generate code for the small code model. The program and its 17888 statically defined symbols must be within 4GB of each other. 17889 Programs can be statically or dynamically linked. This is the 17890 default code model. 17891 17892'-mcmodel=large' 17893 Generate code for the large code model. This makes no assumptions 17894 about addresses and sizes of sections. Programs can be statically 17895 linked only. The '-mcmodel=large' option is incompatible with 17896 '-mabi=ilp32', '-fpic' and '-fPIC'. 17897 17898'-mstrict-align' 17899'-mno-strict-align' 17900 Avoid or allow generating memory accesses that may not be aligned 17901 on a natural object boundary as described in the architecture 17902 specification. 17903 17904'-momit-leaf-frame-pointer' 17905'-mno-omit-leaf-frame-pointer' 17906 Omit or keep the frame pointer in leaf functions. The former 17907 behavior is the default. 17908 17909'-mstack-protector-guard=GUARD' 17910'-mstack-protector-guard-reg=REG' 17911'-mstack-protector-guard-offset=OFFSET' 17912 Generate stack protection code using canary at GUARD. Supported 17913 locations are 'global' for a global canary or 'sysreg' for a canary 17914 in an appropriate system register. 17915 17916 With the latter choice the options 17917 '-mstack-protector-guard-reg=REG' and 17918 '-mstack-protector-guard-offset=OFFSET' furthermore specify which 17919 system register to use as base register for reading the canary, and 17920 from what offset from that base register. There is no default 17921 register or offset as this is entirely for use within the Linux 17922 kernel. 17923 17924'-mtls-dialect=desc' 17925 Use TLS descriptors as the thread-local storage mechanism for 17926 dynamic accesses of TLS variables. This is the default. 17927 17928'-mtls-dialect=traditional' 17929 Use traditional TLS as the thread-local storage mechanism for 17930 dynamic accesses of TLS variables. 17931 17932'-mtls-size=SIZE' 17933 Specify bit size of immediate TLS offsets. Valid values are 12, 17934 24, 32, 48. This option requires binutils 2.26 or newer. 17935 17936'-mfix-cortex-a53-835769' 17937'-mno-fix-cortex-a53-835769' 17938 Enable or disable the workaround for the ARM Cortex-A53 erratum 17939 number 835769. This involves inserting a NOP instruction between 17940 memory instructions and 64-bit integer multiply-accumulate 17941 instructions. 17942 17943'-mfix-cortex-a53-843419' 17944'-mno-fix-cortex-a53-843419' 17945 Enable or disable the workaround for the ARM Cortex-A53 erratum 17946 number 843419. This erratum workaround is made at link time and 17947 this will only pass the corresponding flag to the linker. 17948 17949'-mlow-precision-recip-sqrt' 17950'-mno-low-precision-recip-sqrt' 17951 Enable or disable the reciprocal square root approximation. This 17952 option only has an effect if '-ffast-math' or 17953 '-funsafe-math-optimizations' is used as well. Enabling this 17954 reduces precision of reciprocal square root results to about 16 17955 bits for single precision and to 32 bits for double precision. 17956 17957'-mlow-precision-sqrt' 17958'-mno-low-precision-sqrt' 17959 Enable or disable the square root approximation. This option only 17960 has an effect if '-ffast-math' or '-funsafe-math-optimizations' is 17961 used as well. Enabling this reduces precision of square root 17962 results to about 16 bits for single precision and to 32 bits for 17963 double precision. If enabled, it implies 17964 '-mlow-precision-recip-sqrt'. 17965 17966'-mlow-precision-div' 17967'-mno-low-precision-div' 17968 Enable or disable the division approximation. This option only has 17969 an effect if '-ffast-math' or '-funsafe-math-optimizations' is used 17970 as well. Enabling this reduces precision of division results to 17971 about 16 bits for single precision and to 32 bits for double 17972 precision. 17973 17974'-mtrack-speculation' 17975'-mno-track-speculation' 17976 Enable or disable generation of additional code to track 17977 speculative execution through conditional branches. The tracking 17978 state can then be used by the compiler when expanding calls to 17979 '__builtin_speculation_safe_copy' to permit a more efficient code 17980 sequence to be generated. 17981 17982'-moutline-atomics' 17983'-mno-outline-atomics' 17984 Enable or disable calls to out-of-line helpers to implement atomic 17985 operations. These helpers will, at runtime, determine if the LSE 17986 instructions from ARMv8.1-A can be used; if not, they will use the 17987 load/store-exclusive instructions that are present in the base 17988 ARMv8.0 ISA. 17989 17990 This option is only applicable when compiling for the base ARMv8.0 17991 instruction set. If using a later revision, e.g. 17992 '-march=armv8.1-a' or '-march=armv8-a+lse', the ARMv8.1-Atomics 17993 instructions will be used directly. The same applies when using 17994 '-mcpu=' when the selected cpu supports the 'lse' feature. This 17995 option is on by default. 17996 17997'-march=NAME' 17998 Specify the name of the target architecture and, optionally, one or 17999 more feature modifiers. This option has the form 18000 '-march=ARCH{+[no]FEATURE}*'. 18001 18002 The table below summarizes the permissible values for ARCH and the 18003 features that they enable by default: 18004 18005 ARCH value Architecture Includes by default 18006 -------------------------------------------------------------------------- 18007 'armv8-a' Armv8-A '+fp', '+simd' 18008 'armv8.1-a' Armv8.1-A 'armv8-a', '+crc', '+lse', '+rdma' 18009 'armv8.2-a' Armv8.2-A 'armv8.1-a' 18010 'armv8.3-a' Armv8.3-A 'armv8.2-a', '+pauth' 18011 'armv8.4-a' Armv8.4-A 'armv8.3-a', '+flagm', '+fp16fml', 18012 '+dotprod' 18013 'armv8.5-a' Armv8.5-A 'armv8.4-a', '+sb', '+ssbs', '+predres' 18014 'armv8.6-a' Armv8.6-A 'armv8.5-a', '+bf16', '+i8mm' 18015 'armv8.7-a' Armv8.7-A 'armv8.6-a', '+ls64' 18016 'armv8.8-a' Armv8.8-a 'armv8.7-a', '+mops' 18017 'armv9-a' Armv9-A 'armv8.5-a', '+sve', '+sve2' 18018 'armv8-r' Armv8-R 'armv8-r' 18019 18020 The value 'native' is available on native AArch64 GNU/Linux and 18021 causes the compiler to pick the architecture of the host system. 18022 This option has no effect if the compiler is unable to recognize 18023 the architecture of the host system, 18024 18025 The permissible values for FEATURE are listed in the sub-section on 18026 *note '-march' and '-mcpu' Feature Modifiers: 18027 aarch64-feature-modifiers. Where conflicting feature modifiers are 18028 specified, the right-most feature is used. 18029 18030 GCC uses NAME to determine what kind of instructions it can emit 18031 when generating assembly code. If '-march' is specified without 18032 either of '-mtune' or '-mcpu' also being specified, the code is 18033 tuned to perform well across a range of target processors 18034 implementing the target architecture. 18035 18036'-mtune=NAME' 18037 Specify the name of the target processor for which GCC should tune 18038 the performance of the code. Permissible values for this option 18039 are: 'generic', 'cortex-a35', 'cortex-a53', 'cortex-a55', 18040 'cortex-a57', 'cortex-a72', 'cortex-a73', 'cortex-a75', 18041 'cortex-a76', 'cortex-a76ae', 'cortex-a77', 'cortex-a65', 18042 'cortex-a65ae', 'cortex-a34', 'cortex-a78', 'cortex-a78ae', 18043 'cortex-a78c', 'ares', 'exynos-m1', 'emag', 'falkor', 18044 'neoverse-512tvb', 'neoverse-e1', 'neoverse-n1', 'neoverse-n2', 18045 'neoverse-v1', 'neoverse-v2', 'qdf24xx', 'saphira', 'phecda', 18046 'xgene1', 'vulcan', 'octeontx', 'octeontx81', 'octeontx83', 18047 'octeontx2', 'octeontx2t98', 'octeontx2t96' 'octeontx2t93', 18048 'octeontx2f95', 'octeontx2f95n', 'octeontx2f95mm', 'a64fx', 18049 'thunderx', 'thunderxt88', 'thunderxt88p1', 'thunderxt81', 18050 'tsv110', 'thunderxt83', 'thunderx2t99', 'thunderx3t110', 'zeus', 18051 'cortex-a57.cortex-a53', 'cortex-a72.cortex-a53', 18052 'cortex-a73.cortex-a35', 'cortex-a73.cortex-a53', 18053 'cortex-a75.cortex-a55', 'cortex-a76.cortex-a55', 'cortex-r82', 18054 'cortex-x1', 'cortex-x2', 'cortex-a510', 'cortex-a710', 'ampere1', 18055 'ampere1a', 'native'. 18056 18057 The values 'cortex-a57.cortex-a53', 'cortex-a72.cortex-a53', 18058 'cortex-a73.cortex-a35', 'cortex-a73.cortex-a53', 18059 'cortex-a75.cortex-a55', 'cortex-a76.cortex-a55' specify that GCC 18060 should tune for a big.LITTLE system. 18061 18062 The value 'neoverse-512tvb' specifies that GCC should tune for 18063 Neoverse cores that (a) implement SVE and (b) have a total vector 18064 bandwidth of 512 bits per cycle. In other words, the option tells 18065 GCC to tune for Neoverse cores that can execute 4 128-bit Advanced 18066 SIMD arithmetic instructions a cycle and that can execute an 18067 equivalent number of SVE arithmetic instructions per cycle (2 for 18068 256-bit SVE, 4 for 128-bit SVE). This is more general than tuning 18069 for a specific core like Neoverse V1 but is more specific than the 18070 default tuning described below. 18071 18072 Additionally on native AArch64 GNU/Linux systems the value 'native' 18073 tunes performance to the host system. This option has no effect if 18074 the compiler is unable to recognize the processor of the host 18075 system. 18076 18077 Where none of '-mtune=', '-mcpu=' or '-march=' are specified, the 18078 code is tuned to perform well across a range of target processors. 18079 18080 This option cannot be suffixed by feature modifiers. 18081 18082'-mcpu=NAME' 18083 Specify the name of the target processor, optionally suffixed by 18084 one or more feature modifiers. This option has the form 18085 '-mcpu=CPU{+[no]FEATURE}*', where the permissible values for CPU 18086 are the same as those available for '-mtune'. The permissible 18087 values for FEATURE are documented in the sub-section on *note 18088 '-march' and '-mcpu' Feature Modifiers: aarch64-feature-modifiers. 18089 Where conflicting feature modifiers are specified, the right-most 18090 feature is used. 18091 18092 GCC uses NAME to determine what kind of instructions it can emit 18093 when generating assembly code (as if by '-march') and to determine 18094 the target processor for which to tune for performance (as if by 18095 '-mtune'). Where this option is used in conjunction with '-march' 18096 or '-mtune', those options take precedence over the appropriate 18097 part of this option. 18098 18099 '-mcpu=neoverse-512tvb' is special in that it does not refer to a 18100 specific core, but instead refers to all Neoverse cores that (a) 18101 implement SVE and (b) have a total vector bandwidth of 512 bits a 18102 cycle. Unless overridden by '-march', '-mcpu=neoverse-512tvb' 18103 generates code that can run on a Neoverse V1 core, since Neoverse 18104 V1 is the first Neoverse core with these properties. Unless 18105 overridden by '-mtune', '-mcpu=neoverse-512tvb' tunes code in the 18106 same way as for '-mtune=neoverse-512tvb'. 18107 18108'-moverride=STRING' 18109 Override tuning decisions made by the back-end in response to a 18110 '-mtune=' switch. The syntax, semantics, and accepted values for 18111 STRING in this option are not guaranteed to be consistent across 18112 releases. 18113 18114 This option is only intended to be useful when developing GCC. 18115 18116'-mverbose-cost-dump' 18117 Enable verbose cost model dumping in the debug dump files. This 18118 option is provided for use in debugging the compiler. 18119 18120'-mpc-relative-literal-loads' 18121'-mno-pc-relative-literal-loads' 18122 Enable or disable PC-relative literal loads. With this option 18123 literal pools are accessed using a single instruction and emitted 18124 after each function. This limits the maximum size of functions to 18125 1MB. This is enabled by default for '-mcmodel=tiny'. 18126 18127'-msign-return-address=SCOPE' 18128 Select the function scope on which return address signing will be 18129 applied. Permissible values are 'none', which disables return 18130 address signing, 'non-leaf', which enables pointer signing for 18131 functions which are not leaf functions, and 'all', which enables 18132 pointer signing for all functions. The default value is 'none'. 18133 This option has been deprecated by -mbranch-protection. 18134 18135'-mbranch-protection=NONE|STANDARD|PAC-RET[+LEAF+B-KEY]|BTI' 18136 Select the branch protection features to use. 'none' is the 18137 default and turns off all types of branch protection. 'standard' 18138 turns on all types of branch protection features. If a feature has 18139 additional tuning options, then 'standard' sets it to its standard 18140 level. 'pac-ret[+LEAF]' turns on return address signing to its 18141 standard level: signing functions that save the return address to 18142 memory (non-leaf functions will practically always do this) using 18143 the a-key. The optional argument 'leaf' can be used to extend the 18144 signing to include leaf functions. The optional argument 'b-key' 18145 can be used to sign the functions with the B-key instead of the 18146 A-key. 'bti' turns on branch target identification mechanism. 18147 18148'-mharden-sls=OPTS' 18149 Enable compiler hardening against straight line speculation (SLS). 18150 OPTS is a comma-separated list of the following options: 18151 'retbr' 18152 'blr' 18153 In addition, '-mharden-sls=all' enables all SLS hardening while 18154 '-mharden-sls=none' disables all SLS hardening. 18155 18156'-msve-vector-bits=BITS' 18157 Specify the number of bits in an SVE vector register. This option 18158 only has an effect when SVE is enabled. 18159 18160 GCC supports two forms of SVE code generation: "vector-length 18161 agnostic" output that works with any size of vector register and 18162 "vector-length specific" output that allows GCC to make assumptions 18163 about the vector length when it is useful for optimization reasons. 18164 The possible values of 'bits' are: 'scalable', '128', '256', '512', 18165 '1024' and '2048'. Specifying 'scalable' selects vector-length 18166 agnostic output. At present '-msve-vector-bits=128' also generates 18167 vector-length agnostic output for big-endian targets. All other 18168 values generate vector-length specific code. The behavior of these 18169 values may change in future releases and no value except 'scalable' 18170 should be relied on for producing code that is portable across 18171 different hardware SVE vector lengths. 18172 18173 The default is '-msve-vector-bits=scalable', which produces 18174 vector-length agnostic code. 18175 181763.19.1.1 '-march' and '-mcpu' Feature Modifiers 18177............................................... 18178 18179Feature modifiers used with '-march' and '-mcpu' can be any of the 18180following and their inverses 'noFEATURE': 18181 18182'crc' 18183 Enable CRC extension. This is on by default for 18184 '-march=armv8.1-a'. 18185'crypto' 18186 Enable Crypto extension. This also enables Advanced SIMD and 18187 floating-point instructions. 18188'fp' 18189 Enable floating-point instructions. This is on by default for all 18190 possible values for options '-march' and '-mcpu'. 18191'simd' 18192 Enable Advanced SIMD instructions. This also enables 18193 floating-point instructions. This is on by default for all 18194 possible values for options '-march' and '-mcpu'. 18195'sve' 18196 Enable Scalable Vector Extension instructions. This also enables 18197 Advanced SIMD and floating-point instructions. 18198'lse' 18199 Enable Large System Extension instructions. This is on by default 18200 for '-march=armv8.1-a'. 18201'rdma' 18202 Enable Round Double Multiply Accumulate instructions. This is on 18203 by default for '-march=armv8.1-a'. 18204'fp16' 18205 Enable FP16 extension. This also enables floating-point 18206 instructions. 18207'fp16fml' 18208 Enable FP16 fmla extension. This also enables FP16 extensions and 18209 floating-point instructions. This option is enabled by default for 18210 '-march=armv8.4-a'. Use of this option with architectures prior to 18211 Armv8.2-A is not supported. 18212 18213'rcpc' 18214 Enable the RcPc extension. This does not change code generation 18215 from GCC, but is passed on to the assembler, enabling inline asm 18216 statements to use instructions from the RcPc extension. 18217'dotprod' 18218 Enable the Dot Product extension. This also enables Advanced SIMD 18219 instructions. 18220'aes' 18221 Enable the Armv8-a aes and pmull crypto extension. This also 18222 enables Advanced SIMD instructions. 18223'sha2' 18224 Enable the Armv8-a sha2 crypto extension. This also enables 18225 Advanced SIMD instructions. 18226'sha3' 18227 Enable the sha512 and sha3 crypto extension. This also enables 18228 Advanced SIMD instructions. Use of this option with architectures 18229 prior to Armv8.2-A is not supported. 18230'sm4' 18231 Enable the sm3 and sm4 crypto extension. This also enables 18232 Advanced SIMD instructions. Use of this option with architectures 18233 prior to Armv8.2-A is not supported. 18234'profile' 18235 Enable the Statistical Profiling extension. This option is only to 18236 enable the extension at the assembler level and does not affect 18237 code generation. 18238'rng' 18239 Enable the Armv8.5-a Random Number instructions. This option is 18240 only to enable the extension at the assembler level and does not 18241 affect code generation. 18242'memtag' 18243 Enable the Armv8.5-a Memory Tagging Extensions. Use of this option 18244 with architectures prior to Armv8.5-A is not supported. 18245'sb' 18246 Enable the Armv8-a Speculation Barrier instruction. This option is 18247 only to enable the extension at the assembler level and does not 18248 affect code generation. This option is enabled by default for 18249 '-march=armv8.5-a'. 18250'ssbs' 18251 Enable the Armv8-a Speculative Store Bypass Safe instruction. This 18252 option is only to enable the extension at the assembler level and 18253 does not affect code generation. This option is enabled by default 18254 for '-march=armv8.5-a'. 18255'predres' 18256 Enable the Armv8-a Execution and Data Prediction Restriction 18257 instructions. This option is only to enable the extension at the 18258 assembler level and does not affect code generation. This option 18259 is enabled by default for '-march=armv8.5-a'. 18260'sve2' 18261 Enable the Armv8-a Scalable Vector Extension 2. This also enables 18262 SVE instructions. 18263'sve2-bitperm' 18264 Enable SVE2 bitperm instructions. This also enables SVE2 18265 instructions. 18266'sve2-sm4' 18267 Enable SVE2 sm4 instructions. This also enables SVE2 instructions. 18268'sve2-aes' 18269 Enable SVE2 aes instructions. This also enables SVE2 instructions. 18270'sve2-sha3' 18271 Enable SVE2 sha3 instructions. This also enables SVE2 18272 instructions. 18273'tme' 18274 Enable the Transactional Memory Extension. 18275'i8mm' 18276 Enable 8-bit Integer Matrix Multiply instructions. This also 18277 enables Advanced SIMD and floating-point instructions. This option 18278 is enabled by default for '-march=armv8.6-a'. Use of this option 18279 with architectures prior to Armv8.2-A is not supported. 18280'f32mm' 18281 Enable 32-bit Floating point Matrix Multiply instructions. This 18282 also enables SVE instructions. Use of this option with 18283 architectures prior to Armv8.2-A is not supported. 18284'f64mm' 18285 Enable 64-bit Floating point Matrix Multiply instructions. This 18286 also enables SVE instructions. Use of this option with 18287 architectures prior to Armv8.2-A is not supported. 18288'bf16' 18289 Enable brain half-precision floating-point instructions. This also 18290 enables Advanced SIMD and floating-point instructions. This option 18291 is enabled by default for '-march=armv8.6-a'. Use of this option 18292 with architectures prior to Armv8.2-A is not supported. 18293'ls64' 18294 Enable the 64-byte atomic load and store instructions for 18295 accelerators. This option is enabled by default for 18296 '-march=armv8.7-a'. 18297'mops' 18298 Enable the instructions to accelerate memory operations like 18299 'memcpy', 'memmove', 'memset'. This option is enabled by default 18300 for '-march=armv8.8-a' 18301'flagm' 18302 Enable the Flag Manipulation instructions Extension. 18303'pauth' 18304 Enable the Pointer Authentication Extension. 18305 18306 Feature 'crypto' implies 'aes', 'sha2', and 'simd', which implies 'fp'. 18307Conversely, 'nofp' implies 'nosimd', which implies 'nocrypto', 'noaes' 18308and 'nosha2'. 18309 18310 18311File: gcc.info, Node: Adapteva Epiphany Options, Next: AMD GCN Options, Prev: AArch64 Options, Up: Submodel Options 18312 183133.19.2 Adapteva Epiphany Options 18314-------------------------------- 18315 18316These '-m' options are defined for Adapteva Epiphany: 18317 18318'-mhalf-reg-file' 18319 Don't allocate any register in the range 'r32'...'r63'. That 18320 allows code to run on hardware variants that lack these registers. 18321 18322'-mprefer-short-insn-regs' 18323 Preferentially allocate registers that allow short instruction 18324 generation. This can result in increased instruction count, so 18325 this may either reduce or increase overall code size. 18326 18327'-mbranch-cost=NUM' 18328 Set the cost of branches to roughly NUM "simple" instructions. 18329 This cost is only a heuristic and is not guaranteed to produce 18330 consistent results across releases. 18331 18332'-mcmove' 18333 Enable the generation of conditional moves. 18334 18335'-mnops=NUM' 18336 Emit NUM NOPs before every other generated instruction. 18337 18338'-mno-soft-cmpsf' 18339 For single-precision floating-point comparisons, emit an 'fsub' 18340 instruction and test the flags. This is faster than a software 18341 comparison, but can get incorrect results in the presence of NaNs, 18342 or when two different small numbers are compared such that their 18343 difference is calculated as zero. The default is '-msoft-cmpsf', 18344 which uses slower, but IEEE-compliant, software comparisons. 18345 18346'-mstack-offset=NUM' 18347 Set the offset between the top of the stack and the stack pointer. 18348 E.g., a value of 8 means that the eight bytes in the range 18349 'sp+0...sp+7' can be used by leaf functions without stack 18350 allocation. Values other than '8' or '16' are untested and 18351 unlikely to work. Note also that this option changes the ABI; 18352 compiling a program with a different stack offset than the 18353 libraries have been compiled with generally does not work. This 18354 option can be useful if you want to evaluate if a different stack 18355 offset would give you better code, but to actually use a different 18356 stack offset to build working programs, it is recommended to 18357 configure the toolchain with the appropriate 18358 '--with-stack-offset=NUM' option. 18359 18360'-mno-round-nearest' 18361 Make the scheduler assume that the rounding mode has been set to 18362 truncating. The default is '-mround-nearest'. 18363 18364'-mlong-calls' 18365 If not otherwise specified by an attribute, assume all calls might 18366 be beyond the offset range of the 'b' / 'bl' instructions, and 18367 therefore load the function address into a register before 18368 performing a (otherwise direct) call. This is the default. 18369 18370'-mshort-calls' 18371 If not otherwise specified by an attribute, assume all direct calls 18372 are in the range of the 'b' / 'bl' instructions, so use these 18373 instructions for direct calls. The default is '-mlong-calls'. 18374 18375'-msmall16' 18376 Assume addresses can be loaded as 16-bit unsigned values. This 18377 does not apply to function addresses for which '-mlong-calls' 18378 semantics are in effect. 18379 18380'-mfp-mode=MODE' 18381 Set the prevailing mode of the floating-point unit. This 18382 determines the floating-point mode that is provided and expected at 18383 function call and return time. Making this mode match the mode you 18384 predominantly need at function start can make your programs smaller 18385 and faster by avoiding unnecessary mode switches. 18386 18387 MODE can be set to one the following values: 18388 18389 'caller' 18390 Any mode at function entry is valid, and retained or restored 18391 when the function returns, and when it calls other functions. 18392 This mode is useful for compiling libraries or other 18393 compilation units you might want to incorporate into different 18394 programs with different prevailing FPU modes, and the 18395 convenience of being able to use a single object file 18396 outweighs the size and speed overhead for any extra mode 18397 switching that might be needed, compared with what would be 18398 needed with a more specific choice of prevailing FPU mode. 18399 18400 'truncate' 18401 This is the mode used for floating-point calculations with 18402 truncating (i.e. round towards zero) rounding mode. That 18403 includes conversion from floating point to integer. 18404 18405 'round-nearest' 18406 This is the mode used for floating-point calculations with 18407 round-to-nearest-or-even rounding mode. 18408 18409 'int' 18410 This is the mode used to perform integer calculations in the 18411 FPU, e.g. integer multiply, or integer 18412 multiply-and-accumulate. 18413 18414 The default is '-mfp-mode=caller' 18415 18416'-mno-split-lohi' 18417'-mno-postinc' 18418'-mno-postmodify' 18419 Code generation tweaks that disable, respectively, splitting of 18420 32-bit loads, generation of post-increment addresses, and 18421 generation of post-modify addresses. The defaults are 18422 'msplit-lohi', '-mpost-inc', and '-mpost-modify'. 18423 18424'-mnovect-double' 18425 Change the preferred SIMD mode to SImode. The default is 18426 '-mvect-double', which uses DImode as preferred SIMD mode. 18427 18428'-max-vect-align=NUM' 18429 The maximum alignment for SIMD vector mode types. NUM may be 4 or 18430 8. The default is 8. Note that this is an ABI change, even though 18431 many library function interfaces are unaffected if they don't use 18432 SIMD vector modes in places that affect size and/or alignment of 18433 relevant types. 18434 18435'-msplit-vecmove-early' 18436 Split vector moves into single word moves before reload. In theory 18437 this can give better register allocation, but so far the reverse 18438 seems to be generally the case. 18439 18440'-m1reg-REG' 18441 Specify a register to hold the constant -1, which makes loading 18442 small negative constants and certain bitmasks faster. Allowable 18443 values for REG are 'r43' and 'r63', which specify use of that 18444 register as a fixed register, and 'none', which means that no 18445 register is used for this purpose. The default is '-m1reg-none'. 18446 18447 18448File: gcc.info, Node: AMD GCN Options, Next: ARC Options, Prev: Adapteva Epiphany Options, Up: Submodel Options 18449 184503.19.3 AMD GCN Options 18451---------------------- 18452 18453These options are defined specifically for the AMD GCN port. 18454 18455'-march=GPU' 18456'-mtune=GPU' 18457 Set architecture type or tuning for GPU. Supported values for GPU 18458 are 18459 18460 'fiji' 18461 Compile for GCN3 Fiji devices (gfx803). 18462 18463 'gfx900' 18464 Compile for GCN5 Vega 10 devices (gfx900). 18465 18466 'gfx906' 18467 Compile for GCN5 Vega 20 devices (gfx906). 18468 18469'-msram-ecc=on' 18470'-msram-ecc=off' 18471'-msram-ecc=any' 18472 Compile binaries suitable for devices with the SRAM-ECC feature 18473 enabled, disabled, or either mode. This feature can be enabled 18474 per-process on some devices. The compiled code must match the 18475 device mode. The default is 'any', for devices that support it. 18476 18477'-mstack-size=BYTES' 18478 Specify how many BYTES of stack space will be requested for each 18479 GPU thread (wave-front). Beware that there may be many threads and 18480 limited memory available. The size of the stack allocation may 18481 also have an impact on run-time performance. The default is 32KB 18482 when using OpenACC or OpenMP, and 1MB otherwise. 18483 18484'-mxnack' 18485 Compile binaries suitable for devices with the XNACK feature 18486 enabled. Some devices always require XNACK and some allow the user 18487 to configure XNACK. The compiled code must match the device mode. 18488 The default is '-mno-xnack'. At present this option is a 18489 placeholder for support that is not yet implemented. 18490 18491 18492File: gcc.info, Node: ARC Options, Next: ARM Options, Prev: AMD GCN Options, Up: Submodel Options 18493 184943.19.4 ARC Options 18495------------------ 18496 18497The following options control the architecture variant for which code is 18498being compiled: 18499 18500'-mbarrel-shifter' 18501 Generate instructions supported by barrel shifter. This is the 18502 default unless '-mcpu=ARC601' or '-mcpu=ARCEM' is in effect. 18503 18504'-mjli-always' 18505 Force to call a function using jli_s instruction. This option is 18506 valid only for ARCv2 architecture. 18507 18508'-mcpu=CPU' 18509 Set architecture type, register usage, and instruction scheduling 18510 parameters for CPU. There are also shortcut alias options 18511 available for backward compatibility and convenience. Supported 18512 values for CPU are 18513 18514 'arc600' 18515 Compile for ARC600. Aliases: '-mA6', '-mARC600'. 18516 18517 'arc601' 18518 Compile for ARC601. Alias: '-mARC601'. 18519 18520 'arc700' 18521 Compile for ARC700. Aliases: '-mA7', '-mARC700'. This is the 18522 default when configured with '--with-cpu=arc700'. 18523 18524 'arcem' 18525 Compile for ARC EM. 18526 18527 'archs' 18528 Compile for ARC HS. 18529 18530 'em' 18531 Compile for ARC EM CPU with no hardware extensions. 18532 18533 'em4' 18534 Compile for ARC EM4 CPU. 18535 18536 'em4_dmips' 18537 Compile for ARC EM4 DMIPS CPU. 18538 18539 'em4_fpus' 18540 Compile for ARC EM4 DMIPS CPU with the single-precision 18541 floating-point extension. 18542 18543 'em4_fpuda' 18544 Compile for ARC EM4 DMIPS CPU with single-precision 18545 floating-point and double assist instructions. 18546 18547 'hs' 18548 Compile for ARC HS CPU with no hardware extensions except the 18549 atomic instructions. 18550 18551 'hs34' 18552 Compile for ARC HS34 CPU. 18553 18554 'hs38' 18555 Compile for ARC HS38 CPU. 18556 18557 'hs38_linux' 18558 Compile for ARC HS38 CPU with all hardware extensions on. 18559 18560 'arc600_norm' 18561 Compile for ARC 600 CPU with 'norm' instructions enabled. 18562 18563 'arc600_mul32x16' 18564 Compile for ARC 600 CPU with 'norm' and 32x16-bit multiply 18565 instructions enabled. 18566 18567 'arc600_mul64' 18568 Compile for ARC 600 CPU with 'norm' and 'mul64'-family 18569 instructions enabled. 18570 18571 'arc601_norm' 18572 Compile for ARC 601 CPU with 'norm' instructions enabled. 18573 18574 'arc601_mul32x16' 18575 Compile for ARC 601 CPU with 'norm' and 32x16-bit multiply 18576 instructions enabled. 18577 18578 'arc601_mul64' 18579 Compile for ARC 601 CPU with 'norm' and 'mul64'-family 18580 instructions enabled. 18581 18582 'nps400' 18583 Compile for ARC 700 on NPS400 chip. 18584 18585 'em_mini' 18586 Compile for ARC EM minimalist configuration featuring reduced 18587 register set. 18588 18589'-mdpfp' 18590'-mdpfp-compact' 18591 Generate double-precision FPX instructions, tuned for the compact 18592 implementation. 18593 18594'-mdpfp-fast' 18595 Generate double-precision FPX instructions, tuned for the fast 18596 implementation. 18597 18598'-mno-dpfp-lrsr' 18599 Disable 'lr' and 'sr' instructions from using FPX extension aux 18600 registers. 18601 18602'-mea' 18603 Generate extended arithmetic instructions. Currently only 'divaw', 18604 'adds', 'subs', and 'sat16' are supported. Only valid for 18605 '-mcpu=ARC700'. 18606 18607'-mno-mpy' 18608 Do not generate 'mpy'-family instructions for ARC700. This option 18609 is deprecated. 18610 18611'-mmul32x16' 18612 Generate 32x16-bit multiply and multiply-accumulate instructions. 18613 18614'-mmul64' 18615 Generate 'mul64' and 'mulu64' instructions. Only valid for 18616 '-mcpu=ARC600'. 18617 18618'-mnorm' 18619 Generate 'norm' instructions. This is the default if 18620 '-mcpu=ARC700' is in effect. 18621 18622'-mspfp' 18623'-mspfp-compact' 18624 Generate single-precision FPX instructions, tuned for the compact 18625 implementation. 18626 18627'-mspfp-fast' 18628 Generate single-precision FPX instructions, tuned for the fast 18629 implementation. 18630 18631'-msimd' 18632 Enable generation of ARC SIMD instructions via target-specific 18633 builtins. Only valid for '-mcpu=ARC700'. 18634 18635'-msoft-float' 18636 This option ignored; it is provided for compatibility purposes 18637 only. Software floating-point code is emitted by default, and this 18638 default can overridden by FPX options; '-mspfp', '-mspfp-compact', 18639 or '-mspfp-fast' for single precision, and '-mdpfp', 18640 '-mdpfp-compact', or '-mdpfp-fast' for double precision. 18641 18642'-mswap' 18643 Generate 'swap' instructions. 18644 18645'-matomic' 18646 This enables use of the locked load/store conditional extension to 18647 implement atomic memory built-in functions. Not available for ARC 18648 6xx or ARC EM cores. 18649 18650'-mdiv-rem' 18651 Enable 'div' and 'rem' instructions for ARCv2 cores. 18652 18653'-mcode-density' 18654 Enable code density instructions for ARC EM. This option is on by 18655 default for ARC HS. 18656 18657'-mll64' 18658 Enable double load/store operations for ARC HS cores. 18659 18660'-mtp-regno=REGNO' 18661 Specify thread pointer register number. 18662 18663'-mmpy-option=MULTO' 18664 Compile ARCv2 code with a multiplier design option. You can 18665 specify the option using either a string or numeric value for 18666 MULTO. 'wlh1' is the default value. The recognized values are: 18667 18668 '0' 18669 'none' 18670 No multiplier available. 18671 18672 '1' 18673 'w' 18674 16x16 multiplier, fully pipelined. The following instructions 18675 are enabled: 'mpyw' and 'mpyuw'. 18676 18677 '2' 18678 'wlh1' 18679 32x32 multiplier, fully pipelined (1 stage). The following 18680 instructions are additionally enabled: 'mpy', 'mpyu', 'mpym', 18681 'mpymu', and 'mpy_s'. 18682 18683 '3' 18684 'wlh2' 18685 32x32 multiplier, fully pipelined (2 stages). The following 18686 instructions are additionally enabled: 'mpy', 'mpyu', 'mpym', 18687 'mpymu', and 'mpy_s'. 18688 18689 '4' 18690 'wlh3' 18691 Two 16x16 multipliers, blocking, sequential. The following 18692 instructions are additionally enabled: 'mpy', 'mpyu', 'mpym', 18693 'mpymu', and 'mpy_s'. 18694 18695 '5' 18696 'wlh4' 18697 One 16x16 multiplier, blocking, sequential. The following 18698 instructions are additionally enabled: 'mpy', 'mpyu', 'mpym', 18699 'mpymu', and 'mpy_s'. 18700 18701 '6' 18702 'wlh5' 18703 One 32x4 multiplier, blocking, sequential. The following 18704 instructions are additionally enabled: 'mpy', 'mpyu', 'mpym', 18705 'mpymu', and 'mpy_s'. 18706 18707 '7' 18708 'plus_dmpy' 18709 ARC HS SIMD support. 18710 18711 '8' 18712 'plus_macd' 18713 ARC HS SIMD support. 18714 18715 '9' 18716 'plus_qmacw' 18717 ARC HS SIMD support. 18718 18719 This option is only available for ARCv2 cores. 18720 18721'-mfpu=FPU' 18722 Enables support for specific floating-point hardware extensions for 18723 ARCv2 cores. Supported values for FPU are: 18724 18725 'fpus' 18726 Enables support for single-precision floating-point hardware 18727 extensions. 18728 18729 'fpud' 18730 Enables support for double-precision floating-point hardware 18731 extensions. The single-precision floating-point extension is 18732 also enabled. Not available for ARC EM. 18733 18734 'fpuda' 18735 Enables support for double-precision floating-point hardware 18736 extensions using double-precision assist instructions. The 18737 single-precision floating-point extension is also enabled. 18738 This option is only available for ARC EM. 18739 18740 'fpuda_div' 18741 Enables support for double-precision floating-point hardware 18742 extensions using double-precision assist instructions. The 18743 single-precision floating-point, square-root, and divide 18744 extensions are also enabled. This option is only available 18745 for ARC EM. 18746 18747 'fpuda_fma' 18748 Enables support for double-precision floating-point hardware 18749 extensions using double-precision assist instructions. The 18750 single-precision floating-point and fused multiply and add 18751 hardware extensions are also enabled. This option is only 18752 available for ARC EM. 18753 18754 'fpuda_all' 18755 Enables support for double-precision floating-point hardware 18756 extensions using double-precision assist instructions. All 18757 single-precision floating-point hardware extensions are also 18758 enabled. This option is only available for ARC EM. 18759 18760 'fpus_div' 18761 Enables support for single-precision floating-point, 18762 square-root and divide hardware extensions. 18763 18764 'fpud_div' 18765 Enables support for double-precision floating-point, 18766 square-root and divide hardware extensions. This option 18767 includes option 'fpus_div'. Not available for ARC EM. 18768 18769 'fpus_fma' 18770 Enables support for single-precision floating-point and fused 18771 multiply and add hardware extensions. 18772 18773 'fpud_fma' 18774 Enables support for double-precision floating-point and fused 18775 multiply and add hardware extensions. This option includes 18776 option 'fpus_fma'. Not available for ARC EM. 18777 18778 'fpus_all' 18779 Enables support for all single-precision floating-point 18780 hardware extensions. 18781 18782 'fpud_all' 18783 Enables support for all single- and double-precision 18784 floating-point hardware extensions. Not available for ARC EM. 18785 18786'-mirq-ctrl-saved=REGISTER-RANGE, BLINK, LP_COUNT' 18787 Specifies general-purposes registers that the processor 18788 automatically saves/restores on interrupt entry and exit. 18789 REGISTER-RANGE is specified as two registers separated by a dash. 18790 The register range always starts with 'r0', the upper limit is 'fp' 18791 register. BLINK and LP_COUNT are optional. This option is only 18792 valid for ARC EM and ARC HS cores. 18793 18794'-mrgf-banked-regs=NUMBER' 18795 Specifies the number of registers replicated in second register 18796 bank on entry to fast interrupt. Fast interrupts are interrupts 18797 with the highest priority level P0. These interrupts save only PC 18798 and STATUS32 registers to avoid memory transactions during 18799 interrupt entry and exit sequences. Use this option when you are 18800 using fast interrupts in an ARC V2 family processor. Permitted 18801 values are 4, 8, 16, and 32. 18802 18803'-mlpc-width=WIDTH' 18804 Specify the width of the 'lp_count' register. Valid values for 18805 WIDTH are 8, 16, 20, 24, 28 and 32 bits. The default width is 18806 fixed to 32 bits. If the width is less than 32, the compiler does 18807 not attempt to transform loops in your program to use the 18808 zero-delay loop mechanism unless it is known that the 'lp_count' 18809 register can hold the required loop-counter value. Depending on 18810 the width specified, the compiler and run-time library might 18811 continue to use the loop mechanism for various needs. This option 18812 defines macro '__ARC_LPC_WIDTH__' with the value of WIDTH. 18813 18814'-mrf16' 18815 This option instructs the compiler to generate code for a 16-entry 18816 register file. This option defines the '__ARC_RF16__' preprocessor 18817 macro. 18818 18819'-mbranch-index' 18820 Enable use of 'bi' or 'bih' instructions to implement jump tables. 18821 18822 The following options are passed through to the assembler, and also 18823define preprocessor macro symbols. 18824 18825'-mdsp-packa' 18826 Passed down to the assembler to enable the DSP Pack A extensions. 18827 Also sets the preprocessor symbol '__Xdsp_packa'. This option is 18828 deprecated. 18829 18830'-mdvbf' 18831 Passed down to the assembler to enable the dual Viterbi butterfly 18832 extension. Also sets the preprocessor symbol '__Xdvbf'. This 18833 option is deprecated. 18834 18835'-mlock' 18836 Passed down to the assembler to enable the locked load/store 18837 conditional extension. Also sets the preprocessor symbol 18838 '__Xlock'. 18839 18840'-mmac-d16' 18841 Passed down to the assembler. Also sets the preprocessor symbol 18842 '__Xxmac_d16'. This option is deprecated. 18843 18844'-mmac-24' 18845 Passed down to the assembler. Also sets the preprocessor symbol 18846 '__Xxmac_24'. This option is deprecated. 18847 18848'-mrtsc' 18849 Passed down to the assembler to enable the 64-bit time-stamp 18850 counter extension instruction. Also sets the preprocessor symbol 18851 '__Xrtsc'. This option is deprecated. 18852 18853'-mswape' 18854 Passed down to the assembler to enable the swap byte ordering 18855 extension instruction. Also sets the preprocessor symbol 18856 '__Xswape'. 18857 18858'-mtelephony' 18859 Passed down to the assembler to enable dual- and single-operand 18860 instructions for telephony. Also sets the preprocessor symbol 18861 '__Xtelephony'. This option is deprecated. 18862 18863'-mxy' 18864 Passed down to the assembler to enable the XY memory extension. 18865 Also sets the preprocessor symbol '__Xxy'. 18866 18867 The following options control how the assembly code is annotated: 18868 18869'-misize' 18870 Annotate assembler instructions with estimated addresses. 18871 18872'-mannotate-align' 18873 Explain what alignment considerations lead to the decision to make 18874 an instruction short or long. 18875 18876 The following options are passed through to the linker: 18877 18878'-marclinux' 18879 Passed through to the linker, to specify use of the 'arclinux' 18880 emulation. This option is enabled by default in tool chains built 18881 for 'arc-linux-uclibc' and 'arceb-linux-uclibc' targets when 18882 profiling is not requested. 18883 18884'-marclinux_prof' 18885 Passed through to the linker, to specify use of the 'arclinux_prof' 18886 emulation. This option is enabled by default in tool chains built 18887 for 'arc-linux-uclibc' and 'arceb-linux-uclibc' targets when 18888 profiling is requested. 18889 18890 The following options control the semantics of generated code: 18891 18892'-mlong-calls' 18893 Generate calls as register indirect calls, thus providing access to 18894 the full 32-bit address range. 18895 18896'-mmedium-calls' 18897 Don't use less than 25-bit addressing range for calls, which is the 18898 offset available for an unconditional branch-and-link instruction. 18899 Conditional execution of function calls is suppressed, to allow use 18900 of the 25-bit range, rather than the 21-bit range with conditional 18901 branch-and-link. This is the default for tool chains built for 18902 'arc-linux-uclibc' and 'arceb-linux-uclibc' targets. 18903 18904'-G NUM' 18905 Put definitions of externally-visible data in a small data section 18906 if that data is no bigger than NUM bytes. The default value of NUM 18907 is 4 for any ARC configuration, or 8 when we have double load/store 18908 operations. 18909 18910'-mno-sdata' 18911 Do not generate sdata references. This is the default for tool 18912 chains built for 'arc-linux-uclibc' and 'arceb-linux-uclibc' 18913 targets. 18914 18915'-mvolatile-cache' 18916 Use ordinarily cached memory accesses for volatile references. 18917 This is the default. 18918 18919'-mno-volatile-cache' 18920 Enable cache bypass for volatile references. 18921 18922 The following options fine tune code generation: 18923'-malign-call' 18924 Does nothing. Preserved for backward compatibility. 18925 18926'-mauto-modify-reg' 18927 Enable the use of pre/post modify with register displacement. 18928 18929'-mbbit-peephole' 18930 Enable bbit peephole2. 18931 18932'-mno-brcc' 18933 This option disables a target-specific pass in 'arc_reorg' to 18934 generate compare-and-branch ('brCC') instructions. It has no 18935 effect on generation of these instructions driven by the combiner 18936 pass. 18937 18938'-mcase-vector-pcrel' 18939 Use PC-relative switch case tables to enable case table shortening. 18940 This is the default for '-Os'. 18941 18942'-mcompact-casesi' 18943 Enable compact 'casesi' pattern. This is the default for '-Os', 18944 and only available for ARCv1 cores. This option is deprecated. 18945 18946'-mno-cond-exec' 18947 Disable the ARCompact-specific pass to generate conditional 18948 execution instructions. 18949 18950 Due to delay slot scheduling and interactions between operand 18951 numbers, literal sizes, instruction lengths, and the support for 18952 conditional execution, the target-independent pass to generate 18953 conditional execution is often lacking, so the ARC port has kept a 18954 special pass around that tries to find more conditional execution 18955 generation opportunities after register allocation, branch 18956 shortening, and delay slot scheduling have been done. This pass 18957 generally, but not always, improves performance and code size, at 18958 the cost of extra compilation time, which is why there is an option 18959 to switch it off. If you have a problem with call instructions 18960 exceeding their allowable offset range because they are 18961 conditionalized, you should consider using '-mmedium-calls' 18962 instead. 18963 18964'-mearly-cbranchsi' 18965 Enable pre-reload use of the 'cbranchsi' pattern. 18966 18967'-mexpand-adddi' 18968 Expand 'adddi3' and 'subdi3' at RTL generation time into 'add.f', 18969 'adc' etc. This option is deprecated. 18970 18971'-mindexed-loads' 18972 Enable the use of indexed loads. This can be problematic because 18973 some optimizers then assume that indexed stores exist, which is not 18974 the case. 18975 18976'-mlra' 18977 Enable Local Register Allocation. This is still experimental for 18978 ARC, so by default the compiler uses standard reload (i.e. 18979 '-mno-lra'). 18980 18981'-mlra-priority-none' 18982 Don't indicate any priority for target registers. 18983 18984'-mlra-priority-compact' 18985 Indicate target register priority for r0..r3 / r12..r15. 18986 18987'-mlra-priority-noncompact' 18988 Reduce target register priority for r0..r3 / r12..r15. 18989 18990'-mmillicode' 18991 When optimizing for size (using '-Os'), prologues and epilogues 18992 that have to save or restore a large number of registers are often 18993 shortened by using call to a special function in libgcc; this is 18994 referred to as a _millicode_ call. As these calls can pose 18995 performance issues, and/or cause linking issues when linking in a 18996 nonstandard way, this option is provided to turn on or off 18997 millicode call generation. 18998 18999'-mcode-density-frame' 19000 This option enable the compiler to emit 'enter' and 'leave' 19001 instructions. These instructions are only valid for CPUs with 19002 code-density feature. 19003 19004'-mmixed-code' 19005 Does nothing. Preserved for backward compatibility. 19006 19007'-mq-class' 19008 Ths option is deprecated. Enable 'q' instruction alternatives. 19009 This is the default for '-Os'. 19010 19011'-mRcq' 19012 Enable 'Rcq' constraint handling. Most short code generation 19013 depends on this. This is the default. 19014 19015'-mRcw' 19016 Enable 'Rcw' constraint handling. Most ccfsm condexec mostly 19017 depends on this. This is the default. 19018 19019'-msize-level=LEVEL' 19020 Fine-tune size optimization with regards to instruction lengths and 19021 alignment. The recognized values for LEVEL are: 19022 '0' 19023 No size optimization. This level is deprecated and treated 19024 like '1'. 19025 19026 '1' 19027 Short instructions are used opportunistically. 19028 19029 '2' 19030 In addition, alignment of loops and of code after barriers are 19031 dropped. 19032 19033 '3' 19034 In addition, optional data alignment is dropped, and the 19035 option 'Os' is enabled. 19036 19037 This defaults to '3' when '-Os' is in effect. Otherwise, the 19038 behavior when this is not set is equivalent to level '1'. 19039 19040'-mtune=CPU' 19041 Set instruction scheduling parameters for CPU, overriding any 19042 implied by '-mcpu='. 19043 19044 Supported values for CPU are 19045 19046 'ARC600' 19047 Tune for ARC600 CPU. 19048 19049 'ARC601' 19050 Tune for ARC601 CPU. 19051 19052 'ARC700' 19053 Tune for ARC700 CPU with standard multiplier block. 19054 19055 'ARC700-xmac' 19056 Tune for ARC700 CPU with XMAC block. 19057 19058 'ARC725D' 19059 Tune for ARC725D CPU. 19060 19061 'ARC750D' 19062 Tune for ARC750D CPU. 19063 19064'-mmultcost=NUM' 19065 Cost to assume for a multiply instruction, with '4' being equal to 19066 a normal instruction. 19067 19068'-munalign-prob-threshold=PROBABILITY' 19069 Does nothing. Preserved for backward compatibility. 19070 19071 The following options are maintained for backward compatibility, but 19072are now deprecated and will be removed in a future release: 19073 19074'-margonaut' 19075 Obsolete FPX. 19076 19077'-mbig-endian' 19078'-EB' 19079 Compile code for big-endian targets. Use of these options is now 19080 deprecated. Big-endian code is supported by configuring GCC to 19081 build 'arceb-elf32' and 'arceb-linux-uclibc' targets, for which big 19082 endian is the default. 19083 19084'-mlittle-endian' 19085'-EL' 19086 Compile code for little-endian targets. Use of these options is 19087 now deprecated. Little-endian code is supported by configuring GCC 19088 to build 'arc-elf32' and 'arc-linux-uclibc' targets, for which 19089 little endian is the default. 19090 19091'-mbarrel_shifter' 19092 Replaced by '-mbarrel-shifter'. 19093 19094'-mdpfp_compact' 19095 Replaced by '-mdpfp-compact'. 19096 19097'-mdpfp_fast' 19098 Replaced by '-mdpfp-fast'. 19099 19100'-mdsp_packa' 19101 Replaced by '-mdsp-packa'. 19102 19103'-mEA' 19104 Replaced by '-mea'. 19105 19106'-mmac_24' 19107 Replaced by '-mmac-24'. 19108 19109'-mmac_d16' 19110 Replaced by '-mmac-d16'. 19111 19112'-mspfp_compact' 19113 Replaced by '-mspfp-compact'. 19114 19115'-mspfp_fast' 19116 Replaced by '-mspfp-fast'. 19117 19118'-mtune=CPU' 19119 Values 'arc600', 'arc601', 'arc700' and 'arc700-xmac' for CPU are 19120 replaced by 'ARC600', 'ARC601', 'ARC700' and 'ARC700-xmac' 19121 respectively. 19122 19123'-multcost=NUM' 19124 Replaced by '-mmultcost'. 19125 19126 19127File: gcc.info, Node: ARM Options, Next: AVR Options, Prev: ARC Options, Up: Submodel Options 19128 191293.19.5 ARM Options 19130------------------ 19131 19132These '-m' options are defined for the ARM port: 19133 19134'-mabi=NAME' 19135 Generate code for the specified ABI. Permissible values are: 19136 'apcs-gnu', 'atpcs', 'aapcs', 'aapcs-linux' and 'iwmmxt'. 19137 19138'-mapcs-frame' 19139 Generate a stack frame that is compliant with the ARM Procedure 19140 Call Standard for all functions, even if this is not strictly 19141 necessary for correct execution of the code. Specifying 19142 '-fomit-frame-pointer' with this option causes the stack frames not 19143 to be generated for leaf functions. The default is 19144 '-mno-apcs-frame'. This option is deprecated. 19145 19146'-mapcs' 19147 This is a synonym for '-mapcs-frame' and is deprecated. 19148 19149'-mthumb-interwork' 19150 Generate code that supports calling between the ARM and Thumb 19151 instruction sets. Without this option, on pre-v5 architectures, 19152 the two instruction sets cannot be reliably used inside one 19153 program. The default is '-mno-thumb-interwork', since slightly 19154 larger code is generated when '-mthumb-interwork' is specified. In 19155 AAPCS configurations this option is meaningless. 19156 19157'-mno-sched-prolog' 19158 Prevent the reordering of instructions in the function prologue, or 19159 the merging of those instruction with the instructions in the 19160 function's body. This means that all functions start with a 19161 recognizable set of instructions (or in fact one of a choice from a 19162 small set of different function prologues), and this information 19163 can be used to locate the start of functions inside an executable 19164 piece of code. The default is '-msched-prolog'. 19165 19166'-mfloat-abi=NAME' 19167 Specifies which floating-point ABI to use. Permissible values are: 19168 'soft', 'softfp' and 'hard'. 19169 19170 Specifying 'soft' causes GCC to generate output containing library 19171 calls for floating-point operations. 'softfp' allows the 19172 generation of code using hardware floating-point instructions, but 19173 still uses the soft-float calling conventions. 'hard' allows 19174 generation of floating-point instructions and uses FPU-specific 19175 calling conventions. 19176 19177 The default depends on the specific target configuration. Note 19178 that the hard-float and soft-float ABIs are not link-compatible; 19179 you must compile your entire program with the same ABI, and link 19180 with a compatible set of libraries. 19181 19182'-mgeneral-regs-only' 19183 Generate code which uses only the general-purpose registers. This 19184 will prevent the compiler from using floating-point and Advanced 19185 SIMD registers but will not impose any restrictions on the 19186 assembler. 19187 19188'-mlittle-endian' 19189 Generate code for a processor running in little-endian mode. This 19190 is the default for all standard configurations. 19191 19192'-mbig-endian' 19193 Generate code for a processor running in big-endian mode; the 19194 default is to compile code for a little-endian processor. 19195 19196'-mbe8' 19197'-mbe32' 19198 When linking a big-endian image select between BE8 and BE32 19199 formats. The option has no effect for little-endian images and is 19200 ignored. The default is dependent on the selected target 19201 architecture. For ARMv6 and later architectures the default is 19202 BE8, for older architectures the default is BE32. BE32 format has 19203 been deprecated by ARM. 19204 19205'-march=NAME[+extension...]' 19206 This specifies the name of the target ARM architecture. GCC uses 19207 this name to determine what kind of instructions it can emit when 19208 generating assembly code. This option can be used in conjunction 19209 with or instead of the '-mcpu=' option. 19210 19211 Permissible names are: 'armv4t', 'armv5t', 'armv5te', 'armv6', 19212 'armv6j', 'armv6k', 'armv6kz', 'armv6t2', 'armv6z', 'armv6zk', 19213 'armv7', 'armv7-a', 'armv7ve', 'armv8-a', 'armv8.1-a', 'armv8.2-a', 19214 'armv8.3-a', 'armv8.4-a', 'armv8.5-a', 'armv8.6-a', 'armv9-a', 19215 'armv7-r', 'armv8-r', 'armv6-m', 'armv6s-m', 'armv7-m', 'armv7e-m', 19216 'armv8-m.base', 'armv8-m.main', 'armv8.1-m.main', 'armv9-a', 19217 'iwmmxt' and 'iwmmxt2'. 19218 19219 Additionally, the following architectures, which lack support for 19220 the Thumb execution state, are recognized but support is 19221 deprecated: 'armv4'. 19222 19223 Many of the architectures support extensions. These can be added 19224 by appending '+EXTENSION' to the architecture name. Extension 19225 options are processed in order and capabilities accumulate. An 19226 extension will also enable any necessary base extensions upon which 19227 it depends. For example, the '+crypto' extension will always 19228 enable the '+simd' extension. The exception to the additive 19229 construction is for extensions that are prefixed with '+no...': 19230 these extensions disable the specified option and any other 19231 extensions that may depend on the presence of that extension. 19232 19233 For example, '-march=armv7-a+simd+nofp+vfpv4' is equivalent to 19234 writing '-march=armv7-a+vfpv4' since the '+simd' option is entirely 19235 disabled by the '+nofp' option that follows it. 19236 19237 Most extension names are generically named, but have an effect that 19238 is dependent upon the architecture to which it is applied. For 19239 example, the '+simd' option can be applied to both 'armv7-a' and 19240 'armv8-a' architectures, but will enable the original ARMv7-A 19241 Advanced SIMD (Neon) extensions for 'armv7-a' and the ARMv8-A 19242 variant for 'armv8-a'. 19243 19244 The table below lists the supported extensions for each 19245 architecture. Architectures not mentioned do not support any 19246 extensions. 19247 19248 'armv5te' 19249 'armv6' 19250 'armv6j' 19251 'armv6k' 19252 'armv6kz' 19253 'armv6t2' 19254 'armv6z' 19255 'armv6zk' 19256 '+fp' 19257 The VFPv2 floating-point instructions. The extension 19258 '+vfpv2' can be used as an alias for this extension. 19259 19260 '+nofp' 19261 Disable the floating-point instructions. 19262 19263 'armv7' 19264 The common subset of the ARMv7-A, ARMv7-R and ARMv7-M 19265 architectures. 19266 '+fp' 19267 The VFPv3 floating-point instructions, with 16 19268 double-precision registers. The extension '+vfpv3-d16' 19269 can be used as an alias for this extension. Note that 19270 floating-point is not supported by the base ARMv7-M 19271 architecture, but is compatible with both the ARMv7-A and 19272 ARMv7-R architectures. 19273 19274 '+nofp' 19275 Disable the floating-point instructions. 19276 19277 'armv7-a' 19278 '+mp' 19279 The multiprocessing extension. 19280 19281 '+sec' 19282 The security extension. 19283 19284 '+fp' 19285 The VFPv3 floating-point instructions, with 16 19286 double-precision registers. The extension '+vfpv3-d16' 19287 can be used as an alias for this extension. 19288 19289 '+simd' 19290 The Advanced SIMD (Neon) v1 and the VFPv3 floating-point 19291 instructions. The extensions '+neon' and '+neon-vfpv3' 19292 can be used as aliases for this extension. 19293 19294 '+vfpv3' 19295 The VFPv3 floating-point instructions, with 32 19296 double-precision registers. 19297 19298 '+vfpv3-d16-fp16' 19299 The VFPv3 floating-point instructions, with 16 19300 double-precision registers and the half-precision 19301 floating-point conversion operations. 19302 19303 '+vfpv3-fp16' 19304 The VFPv3 floating-point instructions, with 32 19305 double-precision registers and the half-precision 19306 floating-point conversion operations. 19307 19308 '+vfpv4-d16' 19309 The VFPv4 floating-point instructions, with 16 19310 double-precision registers. 19311 19312 '+vfpv4' 19313 The VFPv4 floating-point instructions, with 32 19314 double-precision registers. 19315 19316 '+neon-fp16' 19317 The Advanced SIMD (Neon) v1 and the VFPv3 floating-point 19318 instructions, with the half-precision floating-point 19319 conversion operations. 19320 19321 '+neon-vfpv4' 19322 The Advanced SIMD (Neon) v2 and the VFPv4 floating-point 19323 instructions. 19324 19325 '+nosimd' 19326 Disable the Advanced SIMD instructions (does not disable 19327 floating point). 19328 19329 '+nofp' 19330 Disable the floating-point and Advanced SIMD 19331 instructions. 19332 19333 'armv7ve' 19334 The extended version of the ARMv7-A architecture with support 19335 for virtualization. 19336 '+fp' 19337 The VFPv4 floating-point instructions, with 16 19338 double-precision registers. The extension '+vfpv4-d16' 19339 can be used as an alias for this extension. 19340 19341 '+simd' 19342 The Advanced SIMD (Neon) v2 and the VFPv4 floating-point 19343 instructions. The extension '+neon-vfpv4' can be used as 19344 an alias for this extension. 19345 19346 '+vfpv3-d16' 19347 The VFPv3 floating-point instructions, with 16 19348 double-precision registers. 19349 19350 '+vfpv3' 19351 The VFPv3 floating-point instructions, with 32 19352 double-precision registers. 19353 19354 '+vfpv3-d16-fp16' 19355 The VFPv3 floating-point instructions, with 16 19356 double-precision registers and the half-precision 19357 floating-point conversion operations. 19358 19359 '+vfpv3-fp16' 19360 The VFPv3 floating-point instructions, with 32 19361 double-precision registers and the half-precision 19362 floating-point conversion operations. 19363 19364 '+vfpv4-d16' 19365 The VFPv4 floating-point instructions, with 16 19366 double-precision registers. 19367 19368 '+vfpv4' 19369 The VFPv4 floating-point instructions, with 32 19370 double-precision registers. 19371 19372 '+neon' 19373 The Advanced SIMD (Neon) v1 and the VFPv3 floating-point 19374 instructions. The extension '+neon-vfpv3' can be used as 19375 an alias for this extension. 19376 19377 '+neon-fp16' 19378 The Advanced SIMD (Neon) v1 and the VFPv3 floating-point 19379 instructions, with the half-precision floating-point 19380 conversion operations. 19381 19382 '+nosimd' 19383 Disable the Advanced SIMD instructions (does not disable 19384 floating point). 19385 19386 '+nofp' 19387 Disable the floating-point and Advanced SIMD 19388 instructions. 19389 19390 'armv8-a' 19391 '+crc' 19392 The Cyclic Redundancy Check (CRC) instructions. 19393 '+simd' 19394 The ARMv8-A Advanced SIMD and floating-point 19395 instructions. 19396 '+crypto' 19397 The cryptographic instructions. 19398 '+nocrypto' 19399 Disable the cryptographic instructions. 19400 '+nofp' 19401 Disable the floating-point, Advanced SIMD and 19402 cryptographic instructions. 19403 '+sb' 19404 Speculation Barrier Instruction. 19405 '+predres' 19406 Execution and Data Prediction Restriction Instructions. 19407 19408 'armv8.1-a' 19409 '+simd' 19410 The ARMv8.1-A Advanced SIMD and floating-point 19411 instructions. 19412 19413 '+crypto' 19414 The cryptographic instructions. This also enables the 19415 Advanced SIMD and floating-point instructions. 19416 19417 '+nocrypto' 19418 Disable the cryptographic instructions. 19419 19420 '+nofp' 19421 Disable the floating-point, Advanced SIMD and 19422 cryptographic instructions. 19423 19424 '+sb' 19425 Speculation Barrier Instruction. 19426 19427 '+predres' 19428 Execution and Data Prediction Restriction Instructions. 19429 19430 'armv8.2-a' 19431 'armv8.3-a' 19432 '+fp16' 19433 The half-precision floating-point data processing 19434 instructions. This also enables the Advanced SIMD and 19435 floating-point instructions. 19436 19437 '+fp16fml' 19438 The half-precision floating-point fmla extension. This 19439 also enables the half-precision floating-point extension 19440 and Advanced SIMD and floating-point instructions. 19441 19442 '+simd' 19443 The ARMv8.1-A Advanced SIMD and floating-point 19444 instructions. 19445 19446 '+crypto' 19447 The cryptographic instructions. This also enables the 19448 Advanced SIMD and floating-point instructions. 19449 19450 '+dotprod' 19451 Enable the Dot Product extension. This also enables 19452 Advanced SIMD instructions. 19453 19454 '+nocrypto' 19455 Disable the cryptographic extension. 19456 19457 '+nofp' 19458 Disable the floating-point, Advanced SIMD and 19459 cryptographic instructions. 19460 19461 '+sb' 19462 Speculation Barrier Instruction. 19463 19464 '+predres' 19465 Execution and Data Prediction Restriction Instructions. 19466 19467 '+i8mm' 19468 8-bit Integer Matrix Multiply instructions. This also 19469 enables Advanced SIMD and floating-point instructions. 19470 19471 '+bf16' 19472 Brain half-precision floating-point instructions. This 19473 also enables Advanced SIMD and floating-point 19474 instructions. 19475 19476 'armv8.4-a' 19477 '+fp16' 19478 The half-precision floating-point data processing 19479 instructions. This also enables the Advanced SIMD and 19480 floating-point instructions as well as the Dot Product 19481 extension and the half-precision floating-point fmla 19482 extension. 19483 19484 '+simd' 19485 The ARMv8.3-A Advanced SIMD and floating-point 19486 instructions as well as the Dot Product extension. 19487 19488 '+crypto' 19489 The cryptographic instructions. This also enables the 19490 Advanced SIMD and floating-point instructions as well as 19491 the Dot Product extension. 19492 19493 '+nocrypto' 19494 Disable the cryptographic extension. 19495 19496 '+nofp' 19497 Disable the floating-point, Advanced SIMD and 19498 cryptographic instructions. 19499 19500 '+sb' 19501 Speculation Barrier Instruction. 19502 19503 '+predres' 19504 Execution and Data Prediction Restriction Instructions. 19505 19506 '+i8mm' 19507 8-bit Integer Matrix Multiply instructions. This also 19508 enables Advanced SIMD and floating-point instructions. 19509 19510 '+bf16' 19511 Brain half-precision floating-point instructions. This 19512 also enables Advanced SIMD and floating-point 19513 instructions. 19514 19515 'armv8.5-a' 19516 '+fp16' 19517 The half-precision floating-point data processing 19518 instructions. This also enables the Advanced SIMD and 19519 floating-point instructions as well as the Dot Product 19520 extension and the half-precision floating-point fmla 19521 extension. 19522 19523 '+simd' 19524 The ARMv8.3-A Advanced SIMD and floating-point 19525 instructions as well as the Dot Product extension. 19526 19527 '+crypto' 19528 The cryptographic instructions. This also enables the 19529 Advanced SIMD and floating-point instructions as well as 19530 the Dot Product extension. 19531 19532 '+nocrypto' 19533 Disable the cryptographic extension. 19534 19535 '+nofp' 19536 Disable the floating-point, Advanced SIMD and 19537 cryptographic instructions. 19538 19539 '+i8mm' 19540 8-bit Integer Matrix Multiply instructions. This also 19541 enables Advanced SIMD and floating-point instructions. 19542 19543 '+bf16' 19544 Brain half-precision floating-point instructions. This 19545 also enables Advanced SIMD and floating-point 19546 instructions. 19547 19548 'armv8.6-a' 19549 '+fp16' 19550 The half-precision floating-point data processing 19551 instructions. This also enables the Advanced SIMD and 19552 floating-point instructions as well as the Dot Product 19553 extension and the half-precision floating-point fmla 19554 extension. 19555 19556 '+simd' 19557 The ARMv8.3-A Advanced SIMD and floating-point 19558 instructions as well as the Dot Product extension. 19559 19560 '+crypto' 19561 The cryptographic instructions. This also enables the 19562 Advanced SIMD and floating-point instructions as well as 19563 the Dot Product extension. 19564 19565 '+nocrypto' 19566 Disable the cryptographic extension. 19567 19568 '+nofp' 19569 Disable the floating-point, Advanced SIMD and 19570 cryptographic instructions. 19571 19572 '+i8mm' 19573 8-bit Integer Matrix Multiply instructions. This also 19574 enables Advanced SIMD and floating-point instructions. 19575 19576 '+bf16' 19577 Brain half-precision floating-point instructions. This 19578 also enables Advanced SIMD and floating-point 19579 instructions. 19580 19581 'armv7-r' 19582 '+fp.sp' 19583 The single-precision VFPv3 floating-point instructions. 19584 The extension '+vfpv3xd' can be used as an alias for this 19585 extension. 19586 19587 '+fp' 19588 The VFPv3 floating-point instructions with 16 19589 double-precision registers. The extension +vfpv3-d16 can 19590 be used as an alias for this extension. 19591 19592 '+vfpv3xd-d16-fp16' 19593 The single-precision VFPv3 floating-point instructions 19594 with 16 double-precision registers and the half-precision 19595 floating-point conversion operations. 19596 19597 '+vfpv3-d16-fp16' 19598 The VFPv3 floating-point instructions with 16 19599 double-precision registers and the half-precision 19600 floating-point conversion operations. 19601 19602 '+nofp' 19603 Disable the floating-point extension. 19604 19605 '+idiv' 19606 The ARM-state integer division instructions. 19607 19608 '+noidiv' 19609 Disable the ARM-state integer division extension. 19610 19611 'armv7e-m' 19612 '+fp' 19613 The single-precision VFPv4 floating-point instructions. 19614 19615 '+fpv5' 19616 The single-precision FPv5 floating-point instructions. 19617 19618 '+fp.dp' 19619 The single- and double-precision FPv5 floating-point 19620 instructions. 19621 19622 '+nofp' 19623 Disable the floating-point extensions. 19624 19625 'armv8.1-m.main' 19626 19627 '+dsp' 19628 The DSP instructions. 19629 19630 '+mve' 19631 The M-Profile Vector Extension (MVE) integer 19632 instructions. 19633 19634 '+mve.fp' 19635 The M-Profile Vector Extension (MVE) integer and single 19636 precision floating-point instructions. 19637 19638 '+fp' 19639 The single-precision floating-point instructions. 19640 19641 '+fp.dp' 19642 The single- and double-precision floating-point 19643 instructions. 19644 19645 '+nofp' 19646 Disable the floating-point extension. 19647 19648 '+cdecp0, +cdecp1, ... , +cdecp7' 19649 Enable the Custom Datapath Extension (CDE) on selected 19650 coprocessors according to the numbers given in the 19651 options in the range 0 to 7. 19652 19653 'armv8-m.main' 19654 '+dsp' 19655 The DSP instructions. 19656 19657 '+nodsp' 19658 Disable the DSP extension. 19659 19660 '+fp' 19661 The single-precision floating-point instructions. 19662 19663 '+fp.dp' 19664 The single- and double-precision floating-point 19665 instructions. 19666 19667 '+nofp' 19668 Disable the floating-point extension. 19669 19670 '+cdecp0, +cdecp1, ... , +cdecp7' 19671 Enable the Custom Datapath Extension (CDE) on selected 19672 coprocessors according to the numbers given in the 19673 options in the range 0 to 7. 19674 19675 'armv8-r' 19676 '+crc' 19677 The Cyclic Redundancy Check (CRC) instructions. 19678 '+fp.sp' 19679 The single-precision FPv5 floating-point instructions. 19680 '+simd' 19681 The ARMv8-A Advanced SIMD and floating-point 19682 instructions. 19683 '+crypto' 19684 The cryptographic instructions. 19685 '+nocrypto' 19686 Disable the cryptographic instructions. 19687 '+nofp' 19688 Disable the floating-point, Advanced SIMD and 19689 cryptographic instructions. 19690 19691 '-march=native' causes the compiler to auto-detect the architecture 19692 of the build computer. At present, this feature is only supported 19693 on GNU/Linux, and not all architectures are recognized. If the 19694 auto-detect is unsuccessful the option has no effect. 19695 19696'-mtune=NAME' 19697 This option specifies the name of the target ARM processor for 19698 which GCC should tune the performance of the code. For some ARM 19699 implementations better performance can be obtained by using this 19700 option. Permissible names are: 'arm7tdmi', 'arm7tdmi-s', 19701 'arm710t', 'arm720t', 'arm740t', 'strongarm', 'strongarm110', 19702 'strongarm1100', 'strongarm1110', 'arm8', 'arm810', 'arm9', 19703 'arm9e', 'arm920', 'arm920t', 'arm922t', 'arm946e-s', 'arm966e-s', 19704 'arm968e-s', 'arm926ej-s', 'arm940t', 'arm9tdmi', 'arm10tdmi', 19705 'arm1020t', 'arm1026ej-s', 'arm10e', 'arm1020e', 'arm1022e', 19706 'arm1136j-s', 'arm1136jf-s', 'mpcore', 'mpcorenovfp', 19707 'arm1156t2-s', 'arm1156t2f-s', 'arm1176jz-s', 'arm1176jzf-s', 19708 'generic-armv7-a', 'cortex-a5', 'cortex-a7', 'cortex-a8', 19709 'cortex-a9', 'cortex-a12', 'cortex-a15', 'cortex-a17', 19710 'cortex-a32', 'cortex-a35', 'cortex-a53', 'cortex-a55', 19711 'cortex-a57', 'cortex-a72', 'cortex-a73', 'cortex-a75', 19712 'cortex-a76', 'cortex-a76ae', 'cortex-a77', 'cortex-a78', 19713 'cortex-a78ae', 'cortex-a78c', 'cortex-a710', 'ares', 'cortex-r4', 19714 'cortex-r4f', 'cortex-r5', 'cortex-r7', 'cortex-r8', 'cortex-r52', 19715 'cortex-r52plus', 'cortex-m0', 'cortex-m0plus', 'cortex-m1', 19716 'cortex-m3', 'cortex-m4', 'cortex-m7', 'cortex-m23', 'cortex-m33', 19717 'cortex-m35p', 'cortex-m55', 'cortex-x1', 19718 'cortex-m1.small-multiply', 'cortex-m0.small-multiply', 19719 'cortex-m0plus.small-multiply', 'exynos-m1', 'marvell-pj4', 19720 'neoverse-n1', 'neoverse-n2', 'neoverse-v1', 'xscale', 'iwmmxt', 19721 'iwmmxt2', 'ep9312', 'fa526', 'fa626', 'fa606te', 'fa626te', 19722 'fmp626', 'fa726te', 'xgene1'. 19723 19724 Additionally, this option can specify that GCC should tune the 19725 performance of the code for a big.LITTLE system. Permissible names 19726 are: 'cortex-a15.cortex-a7', 'cortex-a17.cortex-a7', 19727 'cortex-a57.cortex-a53', 'cortex-a72.cortex-a53', 19728 'cortex-a72.cortex-a35', 'cortex-a73.cortex-a53', 19729 'cortex-a75.cortex-a55', 'cortex-a76.cortex-a55'. 19730 19731 '-mtune=generic-ARCH' specifies that GCC should tune the 19732 performance for a blend of processors within architecture ARCH. 19733 The aim is to generate code that run well on the current most 19734 popular processors, balancing between optimizations that benefit 19735 some CPUs in the range, and avoiding performance pitfalls of other 19736 CPUs. The effects of this option may change in future GCC versions 19737 as CPU models come and go. 19738 19739 '-mtune' permits the same extension options as '-mcpu', but the 19740 extension options do not affect the tuning of the generated code. 19741 19742 '-mtune=native' causes the compiler to auto-detect the CPU of the 19743 build computer. At present, this feature is only supported on 19744 GNU/Linux, and not all architectures are recognized. If the 19745 auto-detect is unsuccessful the option has no effect. 19746 19747'-mcpu=NAME[+extension...]' 19748 This specifies the name of the target ARM processor. GCC uses this 19749 name to derive the name of the target ARM architecture (as if 19750 specified by '-march') and the ARM processor type for which to tune 19751 for performance (as if specified by '-mtune'). Where this option 19752 is used in conjunction with '-march' or '-mtune', those options 19753 take precedence over the appropriate part of this option. 19754 19755 Many of the supported CPUs implement optional architectural 19756 extensions. Where this is so the architectural extensions are 19757 normally enabled by default. If implementations that lack the 19758 extension exist, then the extension syntax can be used to disable 19759 those extensions that have been omitted. For floating-point and 19760 Advanced SIMD (Neon) instructions, the settings of the options 19761 '-mfloat-abi' and '-mfpu' must also be considered: floating-point 19762 and Advanced SIMD instructions will only be used if '-mfloat-abi' 19763 is not set to 'soft'; and any setting of '-mfpu' other than 'auto' 19764 will override the available floating-point and SIMD extension 19765 instructions. 19766 19767 For example, 'cortex-a9' can be found in three major 19768 configurations: integer only, with just a floating-point unit or 19769 with floating-point and Advanced SIMD. The default is to enable all 19770 the instructions, but the extensions '+nosimd' and '+nofp' can be 19771 used to disable just the SIMD or both the SIMD and floating-point 19772 instructions respectively. 19773 19774 Permissible names for this option are the same as those for 19775 '-mtune'. 19776 19777 The following extension options are common to the listed CPUs: 19778 19779 '+nodsp' 19780 Disable the DSP instructions on 'cortex-m33', 'cortex-m35p'. 19781 19782 '+nofp' 19783 Disables the floating-point instructions on 'arm9e', 19784 'arm946e-s', 'arm966e-s', 'arm968e-s', 'arm10e', 'arm1020e', 19785 'arm1022e', 'arm926ej-s', 'arm1026ej-s', 'cortex-r5', 19786 'cortex-r7', 'cortex-r8', 'cortex-m4', 'cortex-m7', 19787 'cortex-m33' and 'cortex-m35p'. Disables the floating-point 19788 and SIMD instructions on 'generic-armv7-a', 'cortex-a5', 19789 'cortex-a7', 'cortex-a8', 'cortex-a9', 'cortex-a12', 19790 'cortex-a15', 'cortex-a17', 'cortex-a15.cortex-a7', 19791 'cortex-a17.cortex-a7', 'cortex-a32', 'cortex-a35', 19792 'cortex-a53' and 'cortex-a55'. 19793 19794 '+nofp.dp' 19795 Disables the double-precision component of the floating-point 19796 instructions on 'cortex-r5', 'cortex-r7', 'cortex-r8', 19797 'cortex-r52', 'cortex-r52plus' and 'cortex-m7'. 19798 19799 '+nosimd' 19800 Disables the SIMD (but not floating-point) instructions on 19801 'generic-armv7-a', 'cortex-a5', 'cortex-a7' and 'cortex-a9'. 19802 19803 '+crypto' 19804 Enables the cryptographic instructions on 'cortex-a32', 19805 'cortex-a35', 'cortex-a53', 'cortex-a55', 'cortex-a57', 19806 'cortex-a72', 'cortex-a73', 'cortex-a75', 'exynos-m1', 19807 'xgene1', 'cortex-a57.cortex-a53', 'cortex-a72.cortex-a53', 19808 'cortex-a73.cortex-a35', 'cortex-a73.cortex-a53' and 19809 'cortex-a75.cortex-a55'. 19810 19811 Additionally the 'generic-armv7-a' pseudo target defaults to VFPv3 19812 with 16 double-precision registers. It supports the following 19813 extension options: 'mp', 'sec', 'vfpv3-d16', 'vfpv3', 19814 'vfpv3-d16-fp16', 'vfpv3-fp16', 'vfpv4-d16', 'vfpv4', 'neon', 19815 'neon-vfpv3', 'neon-fp16', 'neon-vfpv4'. The meanings are the same 19816 as for the extensions to '-march=armv7-a'. 19817 19818 '-mcpu=generic-ARCH' is also permissible, and is equivalent to 19819 '-march=ARCH -mtune=generic-ARCH'. See '-mtune' for more 19820 information. 19821 19822 '-mcpu=native' causes the compiler to auto-detect the CPU of the 19823 build computer. At present, this feature is only supported on 19824 GNU/Linux, and not all architectures are recognized. If the 19825 auto-detect is unsuccessful the option has no effect. 19826 19827'-mfpu=NAME' 19828 This specifies what floating-point hardware (or hardware emulation) 19829 is available on the target. Permissible names are: 'auto', 19830 'vfpv2', 'vfpv3', 'vfpv3-fp16', 'vfpv3-d16', 'vfpv3-d16-fp16', 19831 'vfpv3xd', 'vfpv3xd-fp16', 'neon-vfpv3', 'neon-fp16', 'vfpv4', 19832 'vfpv4-d16', 'fpv4-sp-d16', 'neon-vfpv4', 'fpv5-d16', 19833 'fpv5-sp-d16', 'fp-armv8', 'neon-fp-armv8' and 19834 'crypto-neon-fp-armv8'. Note that 'neon' is an alias for 19835 'neon-vfpv3' and 'vfp' is an alias for 'vfpv2'. 19836 19837 The setting 'auto' is the default and is special. It causes the 19838 compiler to select the floating-point and Advanced SIMD 19839 instructions based on the settings of '-mcpu' and '-march'. 19840 19841 If the selected floating-point hardware includes the NEON extension 19842 (e.g. '-mfpu=neon'), note that floating-point operations are not 19843 generated by GCC's auto-vectorization pass unless 19844 '-funsafe-math-optimizations' is also specified. This is because 19845 NEON hardware does not fully implement the IEEE 754 standard for 19846 floating-point arithmetic (in particular denormal values are 19847 treated as zero), so the use of NEON instructions may lead to a 19848 loss of precision. 19849 19850 You can also set the fpu name at function level by using the 19851 'target("fpu=")' function attributes (*note ARM Function 19852 Attributes::) or pragmas (*note Function Specific Option 19853 Pragmas::). 19854 19855'-mfp16-format=NAME' 19856 Specify the format of the '__fp16' half-precision floating-point 19857 type. Permissible names are 'none', 'ieee', and 'alternative'; the 19858 default is 'none', in which case the '__fp16' type is not defined. 19859 *Note Half-Precision::, for more information. 19860 19861'-mstructure-size-boundary=N' 19862 The sizes of all structures and unions are rounded up to a multiple 19863 of the number of bits set by this option. Permissible values are 19864 8, 32 and 64. The default value varies for different toolchains. 19865 For the COFF targeted toolchain the default value is 8. A value of 19866 64 is only allowed if the underlying ABI supports it. 19867 19868 Specifying a larger number can produce faster, more efficient code, 19869 but can also increase the size of the program. Different values 19870 are potentially incompatible. Code compiled with one value cannot 19871 necessarily expect to work with code or libraries compiled with 19872 another value, if they exchange information using structures or 19873 unions. 19874 19875 This option is deprecated. 19876 19877'-mabort-on-noreturn' 19878 Generate a call to the function 'abort' at the end of a 'noreturn' 19879 function. It is executed if the function tries to return. 19880 19881'-mlong-calls' 19882'-mno-long-calls' 19883 Tells the compiler to perform function calls by first loading the 19884 address of the function into a register and then performing a 19885 subroutine call on this register. This switch is needed if the 19886 target function lies outside of the 64-megabyte addressing range of 19887 the offset-based version of subroutine call instruction. 19888 19889 Even if this switch is enabled, not all function calls are turned 19890 into long calls. The heuristic is that static functions, functions 19891 that have the 'short_call' attribute, functions that are inside the 19892 scope of a '#pragma no_long_calls' directive, and functions whose 19893 definitions have already been compiled within the current 19894 compilation unit are not turned into long calls. The exceptions to 19895 this rule are that weak function definitions, functions with the 19896 'long_call' attribute or the 'section' attribute, and functions 19897 that are within the scope of a '#pragma long_calls' directive are 19898 always turned into long calls. 19899 19900 This feature is not enabled by default. Specifying 19901 '-mno-long-calls' restores the default behavior, as does placing 19902 the function calls within the scope of a '#pragma long_calls_off' 19903 directive. Note these switches have no effect on how the compiler 19904 generates code to handle function calls via function pointers. 19905 19906'-msingle-pic-base' 19907 Treat the register used for PIC addressing as read-only, rather 19908 than loading it in the prologue for each function. The runtime 19909 system is responsible for initializing this register with an 19910 appropriate value before execution begins. 19911 19912'-mpic-register=REG' 19913 Specify the register to be used for PIC addressing. For standard 19914 PIC base case, the default is any suitable register determined by 19915 compiler. For single PIC base case, the default is 'R9' if target 19916 is EABI based or stack-checking is enabled, otherwise the default 19917 is 'R10'. 19918 19919'-mpic-data-is-text-relative' 19920 Assume that the displacement between the text and data segments is 19921 fixed at static link time. This permits using PC-relative 19922 addressing operations to access data known to be in the data 19923 segment. For non-VxWorks RTP targets, this option is enabled by 19924 default. When disabled on such targets, it will enable 19925 '-msingle-pic-base' by default. 19926 19927'-mpoke-function-name' 19928 Write the name of each function into the text section, directly 19929 preceding the function prologue. The generated code is similar to 19930 this: 19931 19932 t0 19933 .ascii "arm_poke_function_name", 0 19934 .align 19935 t1 19936 .word 0xff000000 + (t1 - t0) 19937 arm_poke_function_name 19938 mov ip, sp 19939 stmfd sp!, {fp, ip, lr, pc} 19940 sub fp, ip, #4 19941 19942 When performing a stack backtrace, code can inspect the value of 19943 'pc' stored at 'fp + 0'. If the trace function then looks at 19944 location 'pc - 12' and the top 8 bits are set, then we know that 19945 there is a function name embedded immediately preceding this 19946 location and has length '((pc[-3]) & 0xff000000)'. 19947 19948'-mthumb' 19949'-marm' 19950 19951 Select between generating code that executes in ARM and Thumb 19952 states. The default for most configurations is to generate code 19953 that executes in ARM state, but the default can be changed by 19954 configuring GCC with the '--with-mode='STATE configure option. 19955 19956 You can also override the ARM and Thumb mode for each function by 19957 using the 'target("thumb")' and 'target("arm")' function attributes 19958 (*note ARM Function Attributes::) or pragmas (*note Function 19959 Specific Option Pragmas::). 19960 19961'-mflip-thumb' 19962 Switch ARM/Thumb modes on alternating functions. This option is 19963 provided for regression testing of mixed Thumb/ARM code generation, 19964 and is not intended for ordinary use in compiling code. 19965 19966'-mtpcs-frame' 19967 Generate a stack frame that is compliant with the Thumb Procedure 19968 Call Standard for all non-leaf functions. (A leaf function is one 19969 that does not call any other functions.) The default is 19970 '-mno-tpcs-frame'. 19971 19972'-mtpcs-leaf-frame' 19973 Generate a stack frame that is compliant with the Thumb Procedure 19974 Call Standard for all leaf functions. (A leaf function is one that 19975 does not call any other functions.) The default is 19976 '-mno-apcs-leaf-frame'. 19977 19978'-mcallee-super-interworking' 19979 Gives all externally visible functions in the file being compiled 19980 an ARM instruction set header which switches to Thumb mode before 19981 executing the rest of the function. This allows these functions to 19982 be called from non-interworking code. This option is not valid in 19983 AAPCS configurations because interworking is enabled by default. 19984 19985'-mcaller-super-interworking' 19986 Allows calls via function pointers (including virtual functions) to 19987 execute correctly regardless of whether the target code has been 19988 compiled for interworking or not. There is a small overhead in the 19989 cost of executing a function pointer if this option is enabled. 19990 This option is not valid in AAPCS configurations because 19991 interworking is enabled by default. 19992 19993'-mtp=NAME' 19994 Specify the access model for the thread local storage pointer. The 19995 valid models are 'soft', which generates calls to 19996 '__aeabi_read_tp', 'cp15', which fetches the thread pointer from 19997 'cp15' directly (supported in the arm6k architecture), and 'auto', 19998 which uses the best available method for the selected processor. 19999 The default setting is 'auto'. 20000 20001'-mtls-dialect=DIALECT' 20002 Specify the dialect to use for accessing thread local storage. Two 20003 DIALECTs are supported--'gnu' and 'gnu2'. The 'gnu' dialect 20004 selects the original GNU scheme for supporting local and global 20005 dynamic TLS models. The 'gnu2' dialect selects the GNU descriptor 20006 scheme, which provides better performance for shared libraries. 20007 The GNU descriptor scheme is compatible with the original scheme, 20008 but does require new assembler, linker and library support. 20009 Initial and local exec TLS models are unaffected by this option and 20010 always use the original scheme. 20011 20012'-mword-relocations' 20013 Only generate absolute relocations on word-sized values (i.e. 20014 R_ARM_ABS32). This is enabled by default on targets (uClinux, 20015 SymbianOS) where the runtime loader imposes this restriction, and 20016 when '-fpic' or '-fPIC' is specified. This option conflicts with 20017 '-mslow-flash-data'. 20018 20019'-mfix-cortex-m3-ldrd' 20020 Some Cortex-M3 cores can cause data corruption when 'ldrd' 20021 instructions with overlapping destination and base registers are 20022 used. This option avoids generating these instructions. This 20023 option is enabled by default when '-mcpu=cortex-m3' is specified. 20024 20025'-mfix-cortex-a57-aes-1742098' 20026'-mno-fix-cortex-a57-aes-1742098' 20027'-mfix-cortex-a72-aes-1655431' 20028'-mno-fix-cortex-a72-aes-1655431' 20029 Enable (disable) mitigation for an erratum on Cortex-A57 and 20030 Cortex-A72 that affects the AES cryptographic instructions. This 20031 option is enabled by default when either '-mcpu=cortex-a57' or 20032 '-mcpu=cortex-a72' is specified. 20033 20034'-munaligned-access' 20035'-mno-unaligned-access' 20036 Enables (or disables) reading and writing of 16- and 32- bit values 20037 from addresses that are not 16- or 32- bit aligned. By default 20038 unaligned access is disabled for all pre-ARMv6, all ARMv6-M and for 20039 ARMv8-M Baseline architectures, and enabled for all other 20040 architectures. If unaligned access is not enabled then words in 20041 packed data structures are accessed a byte at a time. 20042 20043 The ARM attribute 'Tag_CPU_unaligned_access' is set in the 20044 generated object file to either true or false, depending upon the 20045 setting of this option. If unaligned access is enabled then the 20046 preprocessor symbol '__ARM_FEATURE_UNALIGNED' is also defined. 20047 20048'-mneon-for-64bits' 20049 This option is deprecated and has no effect. 20050 20051'-mslow-flash-data' 20052 Assume loading data from flash is slower than fetching instruction. 20053 Therefore literal load is minimized for better performance. This 20054 option is only supported when compiling for ARMv7 M-profile and off 20055 by default. It conflicts with '-mword-relocations'. 20056 20057'-masm-syntax-unified' 20058 Assume inline assembler is using unified asm syntax. The default 20059 is currently off which implies divided syntax. This option has no 20060 impact on Thumb2. However, this may change in future releases of 20061 GCC. Divided syntax should be considered deprecated. 20062 20063'-mrestrict-it' 20064 Restricts generation of IT blocks to conform to the rules of 20065 ARMv8-A. IT blocks can only contain a single 16-bit instruction 20066 from a select set of instructions. This option is on by default 20067 for ARMv8-A Thumb mode. 20068 20069'-mprint-tune-info' 20070 Print CPU tuning information as comment in assembler file. This is 20071 an option used only for regression testing of the compiler and not 20072 intended for ordinary use in compiling code. This option is 20073 disabled by default. 20074 20075'-mverbose-cost-dump' 20076 Enable verbose cost model dumping in the debug dump files. This 20077 option is provided for use in debugging the compiler. 20078 20079'-mpure-code' 20080 Do not allow constant data to be placed in code sections. 20081 Additionally, when compiling for ELF object format give all text 20082 sections the ELF processor-specific section attribute 20083 'SHF_ARM_PURECODE'. This option is only available when generating 20084 non-pic code for M-profile targets. 20085 20086'-mcmse' 20087 Generate secure code as per the "ARMv8-M Security Extensions: 20088 Requirements on Development Tools Engineering Specification", which 20089 can be found on 20090 <https://developer.arm.com/documentation/ecm0359818/latest/>. 20091 20092'-mfix-cmse-cve-2021-35465' 20093 Mitigate against a potential security issue with the 'VLLDM' 20094 instruction in some M-profile devices when using CMSE 20095 (CVE-2021-365465). This option is enabled by default when the 20096 option '-mcpu=' is used with 'cortex-m33', 'cortex-m35p' or 20097 'cortex-m55'. The option '-mno-fix-cmse-cve-2021-35465' can be 20098 used to disable the mitigation. 20099 20100'-mstack-protector-guard=GUARD' 20101'-mstack-protector-guard-offset=OFFSET' 20102 Generate stack protection code using canary at GUARD. Supported 20103 locations are 'global' for a global canary or 'tls' for a canary 20104 accessible via the TLS register. The option 20105 '-mstack-protector-guard-offset=' is for use with 20106 '-fstack-protector-guard=tls' and not for use in user-land code. 20107 20108'-mfdpic' 20109'-mno-fdpic' 20110 Select the FDPIC ABI, which uses 64-bit function descriptors to 20111 represent pointers to functions. When the compiler is configured 20112 for 'arm-*-uclinuxfdpiceabi' targets, this option is on by default 20113 and implies '-fPIE' if none of the PIC/PIE-related options is 20114 provided. On other targets, it only enables the FDPIC-specific 20115 code generation features, and the user should explicitly provide 20116 the PIC/PIE-related options as needed. 20117 20118 Note that static linking is not supported because it would still 20119 involve the dynamic linker when the program self-relocates. If 20120 such behavior is acceptable, use -static and -Wl,-dynamic-linker 20121 options. 20122 20123 The opposite '-mno-fdpic' option is useful (and required) to build 20124 the Linux kernel using the same ('arm-*-uclinuxfdpiceabi') 20125 toolchain as the one used to build the userland programs. 20126 20127 20128File: gcc.info, Node: AVR Options, Next: Blackfin Options, Prev: ARM Options, Up: Submodel Options 20129 201303.19.6 AVR Options 20131------------------ 20132 20133These options are defined for AVR implementations: 20134 20135'-mmcu=MCU' 20136 Specify Atmel AVR instruction set architectures (ISA) or MCU type. 20137 20138 The default for this option is 'avr2'. 20139 20140 GCC supports the following AVR devices and ISAs: 20141 20142 'avr2' 20143 "Classic" devices with up to 8 KiB of program memory. 20144 MCU = 'attiny22', 'attiny26', 'at90s2313', 'at90s2323', 20145 'at90s2333', 'at90s2343', 'at90s4414', 'at90s4433', 20146 'at90s4434', 'at90c8534', 'at90s8515', 'at90s8535'. 20147 20148 'avr25' 20149 "Classic" devices with up to 8 KiB of program memory and with 20150 the 'MOVW' instruction. 20151 MCU = 'attiny13', 'attiny13a', 'attiny24', 'attiny24a', 20152 'attiny25', 'attiny261', 'attiny261a', 'attiny2313', 20153 'attiny2313a', 'attiny43u', 'attiny44', 'attiny44a', 20154 'attiny45', 'attiny48', 'attiny441', 'attiny461', 20155 'attiny461a', 'attiny4313', 'attiny84', 'attiny84a', 20156 'attiny85', 'attiny87', 'attiny88', 'attiny828', 'attiny841', 20157 'attiny861', 'attiny861a', 'ata5272', 'ata6616c', 'at86rf401'. 20158 20159 'avr3' 20160 "Classic" devices with 16 KiB up to 64 KiB of program memory. 20161 MCU = 'at76c711', 'at43usb355'. 20162 20163 'avr31' 20164 "Classic" devices with 128 KiB of program memory. 20165 MCU = 'atmega103', 'at43usb320'. 20166 20167 'avr35' 20168 "Classic" devices with 16 KiB up to 64 KiB of program memory 20169 and with the 'MOVW' instruction. 20170 MCU = 'attiny167', 'attiny1634', 'atmega8u2', 'atmega16u2', 20171 'atmega32u2', 'ata5505', 'ata6617c', 'ata664251', 'at90usb82', 20172 'at90usb162'. 20173 20174 'avr4' 20175 "Enhanced" devices with up to 8 KiB of program memory. 20176 MCU = 'atmega48', 'atmega48a', 'atmega48p', 'atmega48pa', 20177 'atmega48pb', 'atmega8', 'atmega8a', 'atmega8hva', 'atmega88', 20178 'atmega88a', 'atmega88p', 'atmega88pa', 'atmega88pb', 20179 'atmega8515', 'atmega8535', 'ata6285', 'ata6286', 'ata6289', 20180 'ata6612c', 'at90pwm1', 'at90pwm2', 'at90pwm2b', 'at90pwm3', 20181 'at90pwm3b', 'at90pwm81'. 20182 20183 'avr5' 20184 "Enhanced" devices with 16 KiB up to 64 KiB of program memory. 20185 20186 MCU = 'atmega16', 'atmega16a', 'atmega16hva', 'atmega16hva2', 20187 'atmega16hvb', 'atmega16hvbrevb', 'atmega16m1', 'atmega16u4', 20188 'atmega161', 'atmega162', 'atmega163', 'atmega164a', 20189 'atmega164p', 'atmega164pa', 'atmega165', 'atmega165a', 20190 'atmega165p', 'atmega165pa', 'atmega168', 'atmega168a', 20191 'atmega168p', 'atmega168pa', 'atmega168pb', 'atmega169', 20192 'atmega169a', 'atmega169p', 'atmega169pa', 'atmega32', 20193 'atmega32a', 'atmega32c1', 'atmega32hvb', 'atmega32hvbrevb', 20194 'atmega32m1', 'atmega32u4', 'atmega32u6', 'atmega323', 20195 'atmega324a', 'atmega324p', 'atmega324pa', 'atmega324pb', 20196 'atmega325', 'atmega325a', 'atmega325p', 'atmega325pa', 20197 'atmega328', 'atmega328p', 'atmega328pb', 'atmega329', 20198 'atmega329a', 'atmega329p', 'atmega329pa', 'atmega3250', 20199 'atmega3250a', 'atmega3250p', 'atmega3250pa', 'atmega3290', 20200 'atmega3290a', 'atmega3290p', 'atmega3290pa', 'atmega406', 20201 'atmega64', 'atmega64a', 'atmega64c1', 'atmega64hve', 20202 'atmega64hve2', 'atmega64m1', 'atmega64rfr2', 'atmega640', 20203 'atmega644', 'atmega644a', 'atmega644p', 'atmega644pa', 20204 'atmega644rfr2', 'atmega645', 'atmega645a', 'atmega645p', 20205 'atmega649', 'atmega649a', 'atmega649p', 'atmega6450', 20206 'atmega6450a', 'atmega6450p', 'atmega6490', 'atmega6490a', 20207 'atmega6490p', 'ata5795', 'ata5790', 'ata5790n', 'ata5791', 20208 'ata6613c', 'ata6614q', 'ata5782', 'ata5831', 'ata8210', 20209 'ata8510', 'ata5702m322', 'at90pwm161', 'at90pwm216', 20210 'at90pwm316', 'at90can32', 'at90can64', 'at90scr100', 20211 'at90usb646', 'at90usb647', 'at94k', 'm3000'. 20212 20213 'avr51' 20214 "Enhanced" devices with 128 KiB of program memory. 20215 MCU = 'atmega128', 'atmega128a', 'atmega128rfa1', 20216 'atmega128rfr2', 'atmega1280', 'atmega1281', 'atmega1284', 20217 'atmega1284p', 'atmega1284rfr2', 'at90can128', 'at90usb1286', 20218 'at90usb1287'. 20219 20220 'avr6' 20221 "Enhanced" devices with 3-byte PC, i.e. with more than 128 KiB 20222 of program memory. 20223 MCU = 'atmega256rfr2', 'atmega2560', 'atmega2561', 20224 'atmega2564rfr2'. 20225 20226 'avrxmega2' 20227 "XMEGA" devices with more than 8 KiB and up to 64 KiB of 20228 program memory. 20229 MCU = 'atxmega8e5', 'atxmega16a4', 'atxmega16a4u', 20230 'atxmega16c4', 'atxmega16d4', 'atxmega16e5', 'atxmega32a4', 20231 'atxmega32a4u', 'atxmega32c3', 'atxmega32c4', 'atxmega32d3', 20232 'atxmega32d4', 'atxmega32e5'. 20233 20234 'avrxmega3' 20235 "XMEGA" devices with up to 64 KiB of combined program memory 20236 and RAM, and with program memory visible in the RAM address 20237 space. 20238 MCU = 'attiny202', 'attiny204', 'attiny212', 'attiny214', 20239 'attiny402', 'attiny404', 'attiny406', 'attiny412', 20240 'attiny414', 'attiny416', 'attiny417', 'attiny804', 20241 'attiny806', 'attiny807', 'attiny814', 'attiny816', 20242 'attiny817', 'attiny1604', 'attiny1606', 'attiny1607', 20243 'attiny1614', 'attiny1616', 'attiny1617', 'attiny3214', 20244 'attiny3216', 'attiny3217', 'atmega808', 'atmega809', 20245 'atmega1608', 'atmega1609', 'atmega3208', 'atmega3209', 20246 'atmega4808', 'atmega4809'. 20247 20248 'avrxmega4' 20249 "XMEGA" devices with more than 64 KiB and up to 128 KiB of 20250 program memory. 20251 MCU = 'atxmega64a3', 'atxmega64a3u', 'atxmega64a4u', 20252 'atxmega64b1', 'atxmega64b3', 'atxmega64c3', 'atxmega64d3', 20253 'atxmega64d4'. 20254 20255 'avrxmega5' 20256 "XMEGA" devices with more than 64 KiB and up to 128 KiB of 20257 program memory and more than 64 KiB of RAM. 20258 MCU = 'atxmega64a1', 'atxmega64a1u'. 20259 20260 'avrxmega6' 20261 "XMEGA" devices with more than 128 KiB of program memory. 20262 MCU = 'atxmega128a3', 'atxmega128a3u', 'atxmega128b1', 20263 'atxmega128b3', 'atxmega128c3', 'atxmega128d3', 20264 'atxmega128d4', 'atxmega192a3', 'atxmega192a3u', 20265 'atxmega192c3', 'atxmega192d3', 'atxmega256a3', 20266 'atxmega256a3b', 'atxmega256a3bu', 'atxmega256a3u', 20267 'atxmega256c3', 'atxmega256d3', 'atxmega384c3', 20268 'atxmega384d3'. 20269 20270 'avrxmega7' 20271 "XMEGA" devices with more than 128 KiB of program memory and 20272 more than 64 KiB of RAM. 20273 MCU = 'atxmega128a1', 'atxmega128a1u', 'atxmega128a4u'. 20274 20275 'avrtiny' 20276 "TINY" Tiny core devices with 512 B up to 4 KiB of program 20277 memory. 20278 MCU = 'attiny4', 'attiny5', 'attiny9', 'attiny10', 'attiny20', 20279 'attiny40'. 20280 20281 'avr1' 20282 This ISA is implemented by the minimal AVR core and supported 20283 for assembler only. 20284 MCU = 'attiny11', 'attiny12', 'attiny15', 'attiny28', 20285 'at90s1200'. 20286 20287'-mabsdata' 20288 20289 Assume that all data in static storage can be accessed by LDS / STS 20290 instructions. This option has only an effect on reduced Tiny 20291 devices like ATtiny40. See also the 'absdata' *note variable 20292 attribute: AVR Variable Attributes. 20293 20294'-maccumulate-args' 20295 Accumulate outgoing function arguments and acquire/release the 20296 needed stack space for outgoing function arguments once in function 20297 prologue/epilogue. Without this option, outgoing arguments are 20298 pushed before calling a function and popped afterwards. 20299 20300 Popping the arguments after the function call can be expensive on 20301 AVR so that accumulating the stack space might lead to smaller 20302 executables because arguments need not be removed from the stack 20303 after such a function call. 20304 20305 This option can lead to reduced code size for functions that 20306 perform several calls to functions that get their arguments on the 20307 stack like calls to printf-like functions. 20308 20309'-mbranch-cost=COST' 20310 Set the branch costs for conditional branch instructions to COST. 20311 Reasonable values for COST are small, non-negative integers. The 20312 default branch cost is 0. 20313 20314'-mcall-prologues' 20315 Functions prologues/epilogues are expanded as calls to appropriate 20316 subroutines. Code size is smaller. 20317 20318'-mdouble=BITS' 20319'-mlong-double=BITS' 20320 Set the size (in bits) of the 'double' or 'long double' type, 20321 respectively. Possible values for BITS are 32 and 64. Whether or 20322 not a specific value for BITS is allowed depends on the 20323 '--with-double=' and '--with-long-double=' 20324 configure options (https://gcc.gnu.org/install/configure.html#avr), 20325 and the same applies for the default values of the options. 20326 20327'-mgas-isr-prologues' 20328 Interrupt service routines (ISRs) may use the '__gcc_isr' pseudo 20329 instruction supported by GNU Binutils. If this option is on, the 20330 feature can still be disabled for individual ISRs by means of the 20331 *note 'no_gccisr': AVR Function Attributes. function attribute. 20332 This feature is activated per default if optimization is on (but 20333 not with '-Og', *note Optimize Options::), and if GNU Binutils 20334 support PR21683 (https://sourceware.org/PR21683). 20335 20336'-mint8' 20337 Assume 'int' to be 8-bit integer. This affects the sizes of all 20338 types: a 'char' is 1 byte, an 'int' is 1 byte, a 'long' is 2 bytes, 20339 and 'long long' is 4 bytes. Please note that this option does not 20340 conform to the C standards, but it results in smaller code size. 20341 20342'-mmain-is-OS_task' 20343 Do not save registers in 'main'. The effect is the same like 20344 attaching attribute *note 'OS_task': AVR Function Attributes. to 20345 'main'. It is activated per default if optimization is on. 20346 20347'-mn-flash=NUM' 20348 Assume that the flash memory has a size of NUM times 64 KiB. 20349 20350'-mno-interrupts' 20351 Generated code is not compatible with hardware interrupts. Code 20352 size is smaller. 20353 20354'-mrelax' 20355 Try to replace 'CALL' resp. 'JMP' instruction by the shorter 20356 'RCALL' resp. 'RJMP' instruction if applicable. Setting '-mrelax' 20357 just adds the '--mlink-relax' option to the assembler's command 20358 line and the '--relax' option to the linker's command line. 20359 20360 Jump relaxing is performed by the linker because jump offsets are 20361 not known before code is located. Therefore, the assembler code 20362 generated by the compiler is the same, but the instructions in the 20363 executable may differ from instructions in the assembler code. 20364 20365 Relaxing must be turned on if linker stubs are needed, see the 20366 section on 'EIND' and linker stubs below. 20367 20368'-mrmw' 20369 Assume that the device supports the Read-Modify-Write instructions 20370 'XCH', 'LAC', 'LAS' and 'LAT'. 20371 20372'-mshort-calls' 20373 20374 Assume that 'RJMP' and 'RCALL' can target the whole program memory. 20375 20376 This option is used internally for multilib selection. It is not 20377 an optimization option, and you don't need to set it by hand. 20378 20379'-msp8' 20380 Treat the stack pointer register as an 8-bit register, i.e. assume 20381 the high byte of the stack pointer is zero. In general, you don't 20382 need to set this option by hand. 20383 20384 This option is used internally by the compiler to select and build 20385 multilibs for architectures 'avr2' and 'avr25'. These 20386 architectures mix devices with and without 'SPH'. For any setting 20387 other than '-mmcu=avr2' or '-mmcu=avr25' the compiler driver adds 20388 or removes this option from the compiler proper's command line, 20389 because the compiler then knows if the device or architecture has 20390 an 8-bit stack pointer and thus no 'SPH' register or not. 20391 20392'-mstrict-X' 20393 Use address register 'X' in a way proposed by the hardware. This 20394 means that 'X' is only used in indirect, post-increment or 20395 pre-decrement addressing. 20396 20397 Without this option, the 'X' register may be used in the same way 20398 as 'Y' or 'Z' which then is emulated by additional instructions. 20399 For example, loading a value with 'X+const' addressing with a small 20400 non-negative 'const < 64' to a register RN is performed as 20401 20402 adiw r26, const ; X += const 20403 ld RN, X ; RN = *X 20404 sbiw r26, const ; X -= const 20405 20406'-mtiny-stack' 20407 Only change the lower 8 bits of the stack pointer. 20408 20409'-mfract-convert-truncate' 20410 Allow to use truncation instead of rounding towards zero for 20411 fractional fixed-point types. 20412 20413'-nodevicelib' 20414 Don't link against AVR-LibC's device specific library 'lib<mcu>.a'. 20415 20416'-nodevicespecs' 20417 Don't add '-specs=device-specs/specs-MCU' to the compiler driver's 20418 command line. The user takes responsibility for supplying the 20419 sub-processes like compiler proper, assembler and linker with 20420 appropriate command line options. This means that the user has to 20421 supply her private device specs file by means of 20422 '-specs=PATH-TO-SPECS-FILE'. There is no more need for option 20423 '-mmcu=MCU'. 20424 20425 This option can also serve as a replacement for the older way of 20426 specifying custom device-specs files that needed '-B SOME-PATH' to 20427 point to a directory which contains a folder named 'device-specs' 20428 which contains a specs file named 'specs-MCU', where MCU was 20429 specified by '-mmcu=MCU'. 20430 20431'-Waddr-space-convert' 20432 Warn about conversions between address spaces in the case where the 20433 resulting address space is not contained in the incoming address 20434 space. 20435 20436'-Wmisspelled-isr' 20437 Warn if the ISR is misspelled, i.e. without __vector prefix. 20438 Enabled by default. 20439 204403.19.6.1 'EIND' and Devices with More Than 128 Ki Bytes of Flash 20441................................................................ 20442 20443Pointers in the implementation are 16 bits wide. The address of a 20444function or label is represented as word address so that indirect jumps 20445and calls can target any code address in the range of 64 Ki words. 20446 20447 In order to facilitate indirect jump on devices with more than 128 Ki 20448bytes of program memory space, there is a special function register 20449called 'EIND' that serves as most significant part of the target address 20450when 'EICALL' or 'EIJMP' instructions are used. 20451 20452 Indirect jumps and calls on these devices are handled as follows by the 20453compiler and are subject to some limitations: 20454 20455 * The compiler never sets 'EIND'. 20456 20457 * The compiler uses 'EIND' implicitly in 'EICALL'/'EIJMP' 20458 instructions or might read 'EIND' directly in order to emulate an 20459 indirect call/jump by means of a 'RET' instruction. 20460 20461 * The compiler assumes that 'EIND' never changes during the startup 20462 code or during the application. In particular, 'EIND' is not 20463 saved/restored in function or interrupt service routine 20464 prologue/epilogue. 20465 20466 * For indirect calls to functions and computed goto, the linker 20467 generates _stubs_. Stubs are jump pads sometimes also called 20468 _trampolines_. Thus, the indirect call/jump jumps to such a stub. 20469 The stub contains a direct jump to the desired address. 20470 20471 * Linker relaxation must be turned on so that the linker generates 20472 the stubs correctly in all situations. See the compiler option 20473 '-mrelax' and the linker option '--relax'. There are corner cases 20474 where the linker is supposed to generate stubs but aborts without 20475 relaxation and without a helpful error message. 20476 20477 * The default linker script is arranged for code with 'EIND = 0'. If 20478 code is supposed to work for a setup with 'EIND != 0', a custom 20479 linker script has to be used in order to place the sections whose 20480 name start with '.trampolines' into the segment where 'EIND' points 20481 to. 20482 20483 * The startup code from libgcc never sets 'EIND'. Notice that 20484 startup code is a blend of code from libgcc and AVR-LibC. For the 20485 impact of AVR-LibC on 'EIND', see the 20486 AVR-LibC user manual (http://nongnu.org/avr-libc/user-manual/). 20487 20488 * It is legitimate for user-specific startup code to set up 'EIND' 20489 early, for example by means of initialization code located in 20490 section '.init3'. Such code runs prior to general startup code 20491 that initializes RAM and calls constructors, but after the bit of 20492 startup code from AVR-LibC that sets 'EIND' to the segment where 20493 the vector table is located. 20494 #include <avr/io.h> 20495 20496 static void 20497 __attribute__((section(".init3"),naked,used,no_instrument_function)) 20498 init3_set_eind (void) 20499 { 20500 __asm volatile ("ldi r24,pm_hh8(__trampolines_start)\n\t" 20501 "out %i0,r24" :: "n" (&EIND) : "r24","memory"); 20502 } 20503 20504 The '__trampolines_start' symbol is defined in the linker script. 20505 20506 * Stubs are generated automatically by the linker if the following 20507 two conditions are met: 20508 20509 - The address of a label is taken by means of the 'gs' modifier 20510 (short for _generate stubs_) like so: 20511 LDI r24, lo8(gs(FUNC)) 20512 LDI r25, hi8(gs(FUNC)) 20513 - The final location of that label is in a code segment 20514 _outside_ the segment where the stubs are located. 20515 20516 * The compiler emits such 'gs' modifiers for code labels in the 20517 following situations: 20518 - Taking address of a function or code label. 20519 - Computed goto. 20520 - If prologue-save function is used, see '-mcall-prologues' 20521 command-line option. 20522 - Switch/case dispatch tables. If you do not want such dispatch 20523 tables you can specify the '-fno-jump-tables' command-line 20524 option. 20525 - C and C++ constructors/destructors called during 20526 startup/shutdown. 20527 - If the tools hit a 'gs()' modifier explained above. 20528 20529 * Jumping to non-symbolic addresses like so is _not_ supported: 20530 20531 int main (void) 20532 { 20533 /* Call function at word address 0x2 */ 20534 return ((int(*)(void)) 0x2)(); 20535 } 20536 20537 Instead, a stub has to be set up, i.e. the function has to be 20538 called through a symbol ('func_4' in the example): 20539 20540 int main (void) 20541 { 20542 extern int func_4 (void); 20543 20544 /* Call function at byte address 0x4 */ 20545 return func_4(); 20546 } 20547 20548 and the application be linked with '-Wl,--defsym,func_4=0x4'. 20549 Alternatively, 'func_4' can be defined in the linker script. 20550 205513.19.6.2 Handling of the 'RAMPD', 'RAMPX', 'RAMPY' and 'RAMPZ' Special Function Registers 20552......................................................................................... 20553 20554Some AVR devices support memories larger than the 64 KiB range that can 20555be accessed with 16-bit pointers. To access memory locations outside 20556this 64 KiB range, the content of a 'RAMP' register is used as high part 20557of the address: The 'X', 'Y', 'Z' address register is concatenated with 20558the 'RAMPX', 'RAMPY', 'RAMPZ' special function register, respectively, 20559to get a wide address. Similarly, 'RAMPD' is used together with direct 20560addressing. 20561 20562 * The startup code initializes the 'RAMP' special function registers 20563 with zero. 20564 20565 * If a *note named address space: AVR Named Address Spaces. other 20566 than generic or '__flash' is used, then 'RAMPZ' is set as needed 20567 before the operation. 20568 20569 * If the device supports RAM larger than 64 KiB and the compiler 20570 needs to change 'RAMPZ' to accomplish an operation, 'RAMPZ' is 20571 reset to zero after the operation. 20572 20573 * If the device comes with a specific 'RAMP' register, the ISR 20574 prologue/epilogue saves/restores that SFR and initializes it with 20575 zero in case the ISR code might (implicitly) use it. 20576 20577 * RAM larger than 64 KiB is not supported by GCC for AVR targets. If 20578 you use inline assembler to read from locations outside the 16-bit 20579 address range and change one of the 'RAMP' registers, you must 20580 reset it to zero after the access. 20581 205823.19.6.3 AVR Built-in Macros 20583............................ 20584 20585GCC defines several built-in macros so that the user code can test for 20586the presence or absence of features. Almost any of the following 20587built-in macros are deduced from device capabilities and thus triggered 20588by the '-mmcu=' command-line option. 20589 20590 For even more AVR-specific built-in macros see *note AVR Named Address 20591Spaces:: and *note AVR Built-in Functions::. 20592 20593'__AVR_ARCH__' 20594 Build-in macro that resolves to a decimal number that identifies 20595 the architecture and depends on the '-mmcu=MCU' option. Possible 20596 values are: 20597 20598 '2', '25', '3', '31', '35', '4', '5', '51', '6' 20599 20600 for MCU='avr2', 'avr25', 'avr3', 'avr31', 'avr35', 'avr4', 'avr5', 20601 'avr51', 'avr6', 20602 20603 respectively and 20604 20605 '100', '102', '103', '104', '105', '106', '107' 20606 20607 for MCU='avrtiny', 'avrxmega2', 'avrxmega3', 'avrxmega4', 20608 'avrxmega5', 'avrxmega6', 'avrxmega7', respectively. If MCU 20609 specifies a device, this built-in macro is set accordingly. For 20610 example, with '-mmcu=atmega8' the macro is defined to '4'. 20611 20612'__AVR_DEVICE__' 20613 Setting '-mmcu=DEVICE' defines this built-in macro which reflects 20614 the device's name. For example, '-mmcu=atmega8' defines the 20615 built-in macro '__AVR_ATmega8__', '-mmcu=attiny261a' defines 20616 '__AVR_ATtiny261A__', etc. 20617 20618 The built-in macros' names follow the scheme '__AVR_DEVICE__' where 20619 DEVICE is the device name as from the AVR user manual. The 20620 difference between DEVICE in the built-in macro and DEVICE in 20621 '-mmcu=DEVICE' is that the latter is always lowercase. 20622 20623 If DEVICE is not a device but only a core architecture like 20624 'avr51', this macro is not defined. 20625 20626'__AVR_DEVICE_NAME__' 20627 Setting '-mmcu=DEVICE' defines this built-in macro to the device's 20628 name. For example, with '-mmcu=atmega8' the macro is defined to 20629 'atmega8'. 20630 20631 If DEVICE is not a device but only a core architecture like 20632 'avr51', this macro is not defined. 20633 20634'__AVR_XMEGA__' 20635 The device / architecture belongs to the XMEGA family of devices. 20636 20637'__AVR_HAVE_ELPM__' 20638 The device has the 'ELPM' instruction. 20639 20640'__AVR_HAVE_ELPMX__' 20641 The device has the 'ELPM RN,Z' and 'ELPM RN,Z+' instructions. 20642 20643'__AVR_HAVE_MOVW__' 20644 The device has the 'MOVW' instruction to perform 16-bit 20645 register-register moves. 20646 20647'__AVR_HAVE_LPMX__' 20648 The device has the 'LPM RN,Z' and 'LPM RN,Z+' instructions. 20649 20650'__AVR_HAVE_MUL__' 20651 The device has a hardware multiplier. 20652 20653'__AVR_HAVE_JMP_CALL__' 20654 The device has the 'JMP' and 'CALL' instructions. This is the case 20655 for devices with more than 8 KiB of program memory. 20656 20657'__AVR_HAVE_EIJMP_EICALL__' 20658'__AVR_3_BYTE_PC__' 20659 The device has the 'EIJMP' and 'EICALL' instructions. This is the 20660 case for devices with more than 128 KiB of program memory. This 20661 also means that the program counter (PC) is 3 bytes wide. 20662 20663'__AVR_2_BYTE_PC__' 20664 The program counter (PC) is 2 bytes wide. This is the case for 20665 devices with up to 128 KiB of program memory. 20666 20667'__AVR_HAVE_8BIT_SP__' 20668'__AVR_HAVE_16BIT_SP__' 20669 The stack pointer (SP) register is treated as 8-bit respectively 20670 16-bit register by the compiler. The definition of these macros is 20671 affected by '-mtiny-stack'. 20672 20673'__AVR_HAVE_SPH__' 20674'__AVR_SP8__' 20675 The device has the SPH (high part of stack pointer) special 20676 function register or has an 8-bit stack pointer, respectively. The 20677 definition of these macros is affected by '-mmcu=' and in the cases 20678 of '-mmcu=avr2' and '-mmcu=avr25' also by '-msp8'. 20679 20680'__AVR_HAVE_RAMPD__' 20681'__AVR_HAVE_RAMPX__' 20682'__AVR_HAVE_RAMPY__' 20683'__AVR_HAVE_RAMPZ__' 20684 The device has the 'RAMPD', 'RAMPX', 'RAMPY', 'RAMPZ' special 20685 function register, respectively. 20686 20687'__NO_INTERRUPTS__' 20688 This macro reflects the '-mno-interrupts' command-line option. 20689 20690'__AVR_ERRATA_SKIP__' 20691'__AVR_ERRATA_SKIP_JMP_CALL__' 20692 Some AVR devices (AT90S8515, ATmega103) must not skip 32-bit 20693 instructions because of a hardware erratum. Skip instructions are 20694 'SBRS', 'SBRC', 'SBIS', 'SBIC' and 'CPSE'. The second macro is 20695 only defined if '__AVR_HAVE_JMP_CALL__' is also set. 20696 20697'__AVR_ISA_RMW__' 20698 The device has Read-Modify-Write instructions (XCH, LAC, LAS and 20699 LAT). 20700 20701'__AVR_SFR_OFFSET__=OFFSET' 20702 Instructions that can address I/O special function registers 20703 directly like 'IN', 'OUT', 'SBI', etc. may use a different address 20704 as if addressed by an instruction to access RAM like 'LD' or 'STS'. 20705 This offset depends on the device architecture and has to be 20706 subtracted from the RAM address in order to get the respective 20707 I/O address. 20708 20709'__AVR_SHORT_CALLS__' 20710 The '-mshort-calls' command line option is set. 20711 20712'__AVR_PM_BASE_ADDRESS__=ADDR' 20713 Some devices support reading from flash memory by means of 'LD*' 20714 instructions. The flash memory is seen in the data address space 20715 at an offset of '__AVR_PM_BASE_ADDRESS__'. If this macro is not 20716 defined, this feature is not available. If defined, the address 20717 space is linear and there is no need to put '.rodata' into RAM. 20718 This is handled by the default linker description file, and is 20719 currently available for 'avrtiny' and 'avrxmega3'. Even more 20720 convenient, there is no need to use address spaces like '__flash' 20721 or features like attribute 'progmem' and 'pgm_read_*'. 20722 20723'__WITH_AVRLIBC__' 20724 The compiler is configured to be used together with AVR-Libc. See 20725 the '--with-avrlibc' configure option. 20726 20727'__HAVE_DOUBLE_MULTILIB__' 20728 Defined if '-mdouble=' acts as a multilib option. 20729 20730'__HAVE_DOUBLE32__' 20731'__HAVE_DOUBLE64__' 20732 Defined if the compiler supports 32-bit double resp. 64-bit 20733 double. The actual layout is specified by option '-mdouble='. 20734 20735'__DEFAULT_DOUBLE__' 20736 The size in bits of 'double' if '-mdouble=' is not set. To test 20737 the layout of 'double' in a program, use the built-in macro 20738 '__SIZEOF_DOUBLE__'. 20739 20740'__HAVE_LONG_DOUBLE32__' 20741'__HAVE_LONG_DOUBLE64__' 20742'__HAVE_LONG_DOUBLE_MULTILIB__' 20743'__DEFAULT_LONG_DOUBLE__' 20744 Same as above, but for 'long double' instead of 'double'. 20745 20746'__WITH_DOUBLE_COMPARISON__' 20747 Reflects the '--with-double-comparison={tristate|bool|libf7}' 20748 configure option (https://gcc.gnu.org/install/configure.html#avr) 20749 and is defined to '2' or '3'. 20750 20751'__WITH_LIBF7_LIBGCC__' 20752'__WITH_LIBF7_MATH__' 20753'__WITH_LIBF7_MATH_SYMBOLS__' 20754 Reflects the '--with-libf7={libgcc|math|math-symbols}' 20755 configure option (https://gcc.gnu.org/install/configure.html#avr). 20756 20757 20758File: gcc.info, Node: Blackfin Options, Next: C6X Options, Prev: AVR Options, Up: Submodel Options 20759 207603.19.7 Blackfin Options 20761----------------------- 20762 20763'-mcpu=CPU[-SIREVISION]' 20764 Specifies the name of the target Blackfin processor. Currently, 20765 CPU can be one of 'bf512', 'bf514', 'bf516', 'bf518', 'bf522', 20766 'bf523', 'bf524', 'bf525', 'bf526', 'bf527', 'bf531', 'bf532', 20767 'bf533', 'bf534', 'bf536', 'bf537', 'bf538', 'bf539', 'bf542', 20768 'bf544', 'bf547', 'bf548', 'bf549', 'bf542m', 'bf544m', 'bf547m', 20769 'bf548m', 'bf549m', 'bf561', 'bf592'. 20770 20771 The optional SIREVISION specifies the silicon revision of the 20772 target Blackfin processor. Any workarounds available for the 20773 targeted silicon revision are enabled. If SIREVISION is 'none', no 20774 workarounds are enabled. If SIREVISION is 'any', all workarounds 20775 for the targeted processor are enabled. The '__SILICON_REVISION__' 20776 macro is defined to two hexadecimal digits representing the major 20777 and minor numbers in the silicon revision. If SIREVISION is 20778 'none', the '__SILICON_REVISION__' is not defined. If SIREVISION 20779 is 'any', the '__SILICON_REVISION__' is defined to be '0xffff'. If 20780 this optional SIREVISION is not used, GCC assumes the latest known 20781 silicon revision of the targeted Blackfin processor. 20782 20783 GCC defines a preprocessor macro for the specified CPU. For the 20784 'bfin-elf' toolchain, this option causes the hardware BSP provided 20785 by libgloss to be linked in if '-msim' is not given. 20786 20787 Without this option, 'bf532' is used as the processor by default. 20788 20789 Note that support for 'bf561' is incomplete. For 'bf561', only the 20790 preprocessor macro is defined. 20791 20792'-msim' 20793 Specifies that the program will be run on the simulator. This 20794 causes the simulator BSP provided by libgloss to be linked in. 20795 This option has effect only for 'bfin-elf' toolchain. Certain 20796 other options, such as '-mid-shared-library' and '-mfdpic', imply 20797 '-msim'. 20798 20799'-momit-leaf-frame-pointer' 20800 Don't keep the frame pointer in a register for leaf functions. 20801 This avoids the instructions to save, set up and restore frame 20802 pointers and makes an extra register available in leaf functions. 20803 20804'-mspecld-anomaly' 20805 When enabled, the compiler ensures that the generated code does not 20806 contain speculative loads after jump instructions. If this option 20807 is used, '__WORKAROUND_SPECULATIVE_LOADS' is defined. 20808 20809'-mno-specld-anomaly' 20810 Don't generate extra code to prevent speculative loads from 20811 occurring. 20812 20813'-mcsync-anomaly' 20814 When enabled, the compiler ensures that the generated code does not 20815 contain CSYNC or SSYNC instructions too soon after conditional 20816 branches. If this option is used, '__WORKAROUND_SPECULATIVE_SYNCS' 20817 is defined. 20818 20819'-mno-csync-anomaly' 20820 Don't generate extra code to prevent CSYNC or SSYNC instructions 20821 from occurring too soon after a conditional branch. 20822 20823'-mlow64k' 20824 When enabled, the compiler is free to take advantage of the 20825 knowledge that the entire program fits into the low 64k of memory. 20826 20827'-mno-low64k' 20828 Assume that the program is arbitrarily large. This is the default. 20829 20830'-mstack-check-l1' 20831 Do stack checking using information placed into L1 scratchpad 20832 memory by the uClinux kernel. 20833 20834'-mid-shared-library' 20835 Generate code that supports shared libraries via the library ID 20836 method. This allows for execute in place and shared libraries in 20837 an environment without virtual memory management. This option 20838 implies '-fPIC'. With a 'bfin-elf' target, this option implies 20839 '-msim'. 20840 20841'-mno-id-shared-library' 20842 Generate code that doesn't assume ID-based shared libraries are 20843 being used. This is the default. 20844 20845'-mleaf-id-shared-library' 20846 Generate code that supports shared libraries via the library ID 20847 method, but assumes that this library or executable won't link 20848 against any other ID shared libraries. That allows the compiler to 20849 use faster code for jumps and calls. 20850 20851'-mno-leaf-id-shared-library' 20852 Do not assume that the code being compiled won't link against any 20853 ID shared libraries. Slower code is generated for jump and call 20854 insns. 20855 20856'-mshared-library-id=n' 20857 Specifies the identification number of the ID-based shared library 20858 being compiled. Specifying a value of 0 generates more compact 20859 code; specifying other values forces the allocation of that number 20860 to the current library but is no more space- or time-efficient than 20861 omitting this option. 20862 20863'-msep-data' 20864 Generate code that allows the data segment to be located in a 20865 different area of memory from the text segment. This allows for 20866 execute in place in an environment without virtual memory 20867 management by eliminating relocations against the text section. 20868 20869'-mno-sep-data' 20870 Generate code that assumes that the data segment follows the text 20871 segment. This is the default. 20872 20873'-mlong-calls' 20874'-mno-long-calls' 20875 Tells the compiler to perform function calls by first loading the 20876 address of the function into a register and then performing a 20877 subroutine call on this register. This switch is needed if the 20878 target function lies outside of the 24-bit addressing range of the 20879 offset-based version of subroutine call instruction. 20880 20881 This feature is not enabled by default. Specifying 20882 '-mno-long-calls' restores the default behavior. Note these 20883 switches have no effect on how the compiler generates code to 20884 handle function calls via function pointers. 20885 20886'-mfast-fp' 20887 Link with the fast floating-point library. This library relaxes 20888 some of the IEEE floating-point standard's rules for checking 20889 inputs against Not-a-Number (NAN), in the interest of performance. 20890 20891'-minline-plt' 20892 Enable inlining of PLT entries in function calls to functions that 20893 are not known to bind locally. It has no effect without '-mfdpic'. 20894 20895'-mmulticore' 20896 Build a standalone application for multicore Blackfin processors. 20897 This option causes proper start files and link scripts supporting 20898 multicore to be used, and defines the macro '__BFIN_MULTICORE'. It 20899 can only be used with '-mcpu=bf561[-SIREVISION]'. 20900 20901 This option can be used with '-mcorea' or '-mcoreb', which selects 20902 the one-application-per-core programming model. Without '-mcorea' 20903 or '-mcoreb', the single-application/dual-core programming model is 20904 used. In this model, the main function of Core B should be named 20905 as 'coreb_main'. 20906 20907 If this option is not used, the single-core application programming 20908 model is used. 20909 20910'-mcorea' 20911 Build a standalone application for Core A of BF561 when using the 20912 one-application-per-core programming model. Proper start files and 20913 link scripts are used to support Core A, and the macro 20914 '__BFIN_COREA' is defined. This option can only be used in 20915 conjunction with '-mmulticore'. 20916 20917'-mcoreb' 20918 Build a standalone application for Core B of BF561 when using the 20919 one-application-per-core programming model. Proper start files and 20920 link scripts are used to support Core B, and the macro 20921 '__BFIN_COREB' is defined. When this option is used, 'coreb_main' 20922 should be used instead of 'main'. This option can only be used in 20923 conjunction with '-mmulticore'. 20924 20925'-msdram' 20926 Build a standalone application for SDRAM. Proper start files and 20927 link scripts are used to put the application into SDRAM, and the 20928 macro '__BFIN_SDRAM' is defined. The loader should initialize 20929 SDRAM before loading the application. 20930 20931'-micplb' 20932 Assume that ICPLBs are enabled at run time. This has an effect on 20933 certain anomaly workarounds. For Linux targets, the default is to 20934 assume ICPLBs are enabled; for standalone applications the default 20935 is off. 20936 20937 20938File: gcc.info, Node: C6X Options, Next: CRIS Options, Prev: Blackfin Options, Up: Submodel Options 20939 209403.19.8 C6X Options 20941------------------ 20942 20943'-march=NAME' 20944 This specifies the name of the target architecture. GCC uses this 20945 name to determine what kind of instructions it can emit when 20946 generating assembly code. Permissible names are: 'c62x', 'c64x', 20947 'c64x+', 'c67x', 'c67x+', 'c674x'. 20948 20949'-mbig-endian' 20950 Generate code for a big-endian target. 20951 20952'-mlittle-endian' 20953 Generate code for a little-endian target. This is the default. 20954 20955'-msim' 20956 Choose startup files and linker script suitable for the simulator. 20957 20958'-msdata=default' 20959 Put small global and static data in the '.neardata' section, which 20960 is pointed to by register 'B14'. Put small uninitialized global 20961 and static data in the '.bss' section, which is adjacent to the 20962 '.neardata' section. Put small read-only data into the '.rodata' 20963 section. The corresponding sections used for large pieces of data 20964 are '.fardata', '.far' and '.const'. 20965 20966'-msdata=all' 20967 Put all data, not just small objects, into the sections reserved 20968 for small data, and use addressing relative to the 'B14' register 20969 to access them. 20970 20971'-msdata=none' 20972 Make no use of the sections reserved for small data, and use 20973 absolute addresses to access all data. Put all initialized global 20974 and static data in the '.fardata' section, and all uninitialized 20975 data in the '.far' section. Put all constant data into the 20976 '.const' section. 20977 20978 20979File: gcc.info, Node: CRIS Options, Next: CR16 Options, Prev: C6X Options, Up: Submodel Options 20980 209813.19.9 CRIS Options 20982------------------- 20983 20984These options are defined specifically for the CRIS ports. 20985 20986'-march=ARCHITECTURE-TYPE' 20987'-mcpu=ARCHITECTURE-TYPE' 20988 Generate code for the specified architecture. The choices for 20989 ARCHITECTURE-TYPE are 'v3', 'v8' and 'v10' for respectively 20990 ETRAX 4, ETRAX 100, and ETRAX 100 LX. Default is 'v0'. 20991 20992'-mtune=ARCHITECTURE-TYPE' 20993 Tune to ARCHITECTURE-TYPE everything applicable about the generated 20994 code, except for the ABI and the set of available instructions. 20995 The choices for ARCHITECTURE-TYPE are the same as for 20996 '-march=ARCHITECTURE-TYPE'. 20997 20998'-mmax-stack-frame=N' 20999 Warn when the stack frame of a function exceeds N bytes. 21000 21001'-metrax4' 21002'-metrax100' 21003 The options '-metrax4' and '-metrax100' are synonyms for 21004 '-march=v3' and '-march=v8' respectively. 21005 21006'-mmul-bug-workaround' 21007'-mno-mul-bug-workaround' 21008 Work around a bug in the 'muls' and 'mulu' instructions for CPU 21009 models where it applies. This option is disabled by default. 21010 21011'-mpdebug' 21012 Enable CRIS-specific verbose debug-related information in the 21013 assembly code. This option also has the effect of turning off the 21014 '#NO_APP' formatted-code indicator to the assembler at the 21015 beginning of the assembly file. 21016 21017'-mcc-init' 21018 Do not use condition-code results from previous instruction; always 21019 emit compare and test instructions before use of condition codes. 21020 21021'-mno-side-effects' 21022 Do not emit instructions with side effects in addressing modes 21023 other than post-increment. 21024 21025'-mstack-align' 21026'-mno-stack-align' 21027'-mdata-align' 21028'-mno-data-align' 21029'-mconst-align' 21030'-mno-const-align' 21031 These options ('no-' options) arrange (eliminate arrangements) for 21032 the stack frame, individual data and constants to be aligned for 21033 the maximum single data access size for the chosen CPU model. The 21034 default is to arrange for 32-bit alignment. ABI details such as 21035 structure layout are not affected by these options. 21036 21037'-m32-bit' 21038'-m16-bit' 21039'-m8-bit' 21040 Similar to the stack- data- and const-align options above, these 21041 options arrange for stack frame, writable data and constants to all 21042 be 32-bit, 16-bit or 8-bit aligned. The default is 32-bit 21043 alignment. 21044 21045'-mno-prologue-epilogue' 21046'-mprologue-epilogue' 21047 With '-mno-prologue-epilogue', the normal function prologue and 21048 epilogue which set up the stack frame are omitted and no return 21049 instructions or return sequences are generated in the code. Use 21050 this option only together with visual inspection of the compiled 21051 code: no warnings or errors are generated when call-saved registers 21052 must be saved, or storage for local variables needs to be 21053 allocated. 21054 21055'-melf' 21056 Legacy no-op option. 21057 21058'-sim' 21059 This option arranges to link with input-output functions from a 21060 simulator library. Code, initialized data and zero-initialized 21061 data are allocated consecutively. 21062 21063'-sim2' 21064 Like '-sim', but pass linker options to locate initialized data at 21065 0x40000000 and zero-initialized data at 0x80000000. 21066 21067 21068File: gcc.info, Node: CR16 Options, Next: C-SKY Options, Prev: CRIS Options, Up: Submodel Options 21069 210703.19.10 CR16 Options 21071-------------------- 21072 21073These options are defined specifically for the CR16 ports. 21074 21075'-mmac' 21076 Enable the use of multiply-accumulate instructions. Disabled by 21077 default. 21078 21079'-mcr16cplus' 21080'-mcr16c' 21081 Generate code for CR16C or CR16C+ architecture. CR16C+ 21082 architecture is default. 21083 21084'-msim' 21085 Links the library libsim.a which is in compatible with simulator. 21086 Applicable to ELF compiler only. 21087 21088'-mint32' 21089 Choose integer type as 32-bit wide. 21090 21091'-mbit-ops' 21092 Generates 'sbit'/'cbit' instructions for bit manipulations. 21093 21094'-mdata-model=MODEL' 21095 Choose a data model. The choices for MODEL are 'near', 'far' or 21096 'medium'. 'medium' is default. However, 'far' is not valid with 21097 '-mcr16c', as the CR16C architecture does not support the far data 21098 model. 21099 21100 21101File: gcc.info, Node: C-SKY Options, Next: Darwin Options, Prev: CR16 Options, Up: Submodel Options 21102 211033.19.11 C-SKY Options 21104--------------------- 21105 21106GCC supports these options when compiling for C-SKY V2 processors. 21107 21108'-march=ARCH' 21109 Specify the C-SKY target architecture. Valid values for ARCH are: 21110 'ck801', 'ck802', 'ck803', 'ck807', and 'ck810'. The default is 21111 'ck810'. 21112 21113'-mcpu=CPU' 21114 Specify the C-SKY target processor. Valid values for CPU are: 21115 'ck801', 'ck801t', 'ck802', 'ck802t', 'ck802j', 'ck803', 'ck803h', 21116 'ck803t', 'ck803ht', 'ck803f', 'ck803fh', 'ck803e', 'ck803eh', 21117 'ck803et', 'ck803eht', 'ck803ef', 'ck803efh', 'ck803ft', 21118 'ck803eft', 'ck803efht', 'ck803r1', 'ck803hr1', 'ck803tr1', 21119 'ck803htr1', 'ck803fr1', 'ck803fhr1', 'ck803er1', 'ck803ehr1', 21120 'ck803etr1', 'ck803ehtr1', 'ck803efr1', 'ck803efhr1', 'ck803ftr1', 21121 'ck803eftr1', 'ck803efhtr1', 'ck803s', 'ck803st', 'ck803se', 21122 'ck803sf', 'ck803sef', 'ck803seft', 'ck807e', 'ck807ef', 'ck807', 21123 'ck807f', 'ck810e', 'ck810et', 'ck810ef', 'ck810eft', 'ck810', 21124 'ck810v', 'ck810f', 'ck810t', 'ck810fv', 'ck810tv', 'ck810ft', and 21125 'ck810ftv'. 21126 21127'-mbig-endian' 21128'-EB' 21129'-mlittle-endian' 21130'-EL' 21131 21132 Select big- or little-endian code. The default is little-endian. 21133 21134'-mfloat-abi=NAME' 21135 Specifies which floating-point ABI to use. Permissible values are: 21136 'soft', 'softfp' and 'hard'. 21137 21138 Specifying 'soft' causes GCC to generate output containing library 21139 calls for floating-point operations. 'softfp' allows the 21140 generation of code using hardware floating-point instructions, but 21141 still uses the soft-float calling conventions. 'hard' allows 21142 generation of floating-point instructions and uses FPU-specific 21143 calling conventions. 21144 21145 The default depends on the specific target configuration. Note 21146 that the hard-float and soft-float ABIs are not link-compatible; 21147 you must compile your entire program with the same ABI, and link 21148 with a compatible set of libraries. 21149 21150'-mhard-float' 21151'-msoft-float' 21152 21153 Select hardware or software floating-point implementations. The 21154 default is soft float. 21155 21156'-mdouble-float' 21157'-mno-double-float' 21158 When '-mhard-float' is in effect, enable generation of 21159 double-precision float instructions. This is the default except 21160 when compiling for CK803. 21161 21162'-mfdivdu' 21163'-mno-fdivdu' 21164 When '-mhard-float' is in effect, enable generation of 'frecipd', 21165 'fsqrtd', and 'fdivd' instructions. This is the default except 21166 when compiling for CK803. 21167 21168'-mfpu=FPU' 21169 Select the floating-point processor. This option can only be used 21170 with '-mhard-float'. Values for FPU are 'fpv2_sf' (equivalent to 21171 '-mno-double-float -mno-fdivdu'), 'fpv2' ('-mdouble-float 21172 -mno-divdu'), and 'fpv2_divd' ('-mdouble-float -mdivdu'). 21173 21174'-melrw' 21175'-mno-elrw' 21176 Enable the extended 'lrw' instruction. This option defaults to on 21177 for CK801 and off otherwise. 21178 21179'-mistack' 21180'-mno-istack' 21181 Enable interrupt stack instructions; the default is off. 21182 21183 The '-mistack' option is required to handle the 'interrupt' and 21184 'isr' function attributes (*note C-SKY Function Attributes::). 21185 21186'-mmp' 21187 Enable multiprocessor instructions; the default is off. 21188 21189'-mcp' 21190 Enable coprocessor instructions; the default is off. 21191 21192'-mcache' 21193 Enable coprocessor instructions; the default is off. 21194 21195'-msecurity' 21196 Enable C-SKY security instructions; the default is off. 21197 21198'-mtrust' 21199 Enable C-SKY trust instructions; the default is off. 21200 21201'-mdsp' 21202'-medsp' 21203'-mvdsp' 21204 Enable C-SKY DSP, Enhanced DSP, or Vector DSP instructions, 21205 respectively. All of these options default to off. 21206 21207'-mdiv' 21208'-mno-div' 21209 Generate divide instructions. Default is off. 21210 21211'-msmart' 21212'-mno-smart' 21213 Generate code for Smart Mode, using only registers numbered 0-7 to 21214 allow use of 16-bit instructions. This option is ignored for CK801 21215 where this is the required behavior, and it defaults to on for 21216 CK802. For other targets, the default is off. 21217 21218'-mhigh-registers' 21219'-mno-high-registers' 21220 Generate code using the high registers numbered 16-31. This option 21221 is not supported on CK801, CK802, or CK803, and is enabled by 21222 default for other processors. 21223 21224'-manchor' 21225'-mno-anchor' 21226 Generate code using global anchor symbol addresses. 21227 21228'-mpushpop' 21229'-mno-pushpop' 21230 Generate code using 'push' and 'pop' instructions. This option 21231 defaults to on. 21232 21233'-mmultiple-stld' 21234'-mstm' 21235'-mno-multiple-stld' 21236'-mno-stm' 21237 Generate code using 'stm' and 'ldm' instructions. This option 21238 isn't supported on CK801 but is enabled by default on other 21239 processors. 21240 21241'-mconstpool' 21242'-mno-constpool' 21243 Create constant pools in the compiler instead of deferring it to 21244 the assembler. This option is the default and required for correct 21245 code generation on CK801 and CK802, and is optional on other 21246 processors. 21247 21248'-mstack-size' 21249'-mno-stack-size' 21250 Emit '.stack_size' directives for each function in the assembly 21251 output. This option defaults to off. 21252 21253'-mccrt' 21254'-mno-ccrt' 21255 Generate code for the C-SKY compiler runtime instead of libgcc. 21256 This option defaults to off. 21257 21258'-mbranch-cost=N' 21259 Set the branch costs to roughly 'n' instructions. The default is 21260 1. 21261 21262'-msched-prolog' 21263'-mno-sched-prolog' 21264 Permit scheduling of function prologue and epilogue sequences. 21265 Using this option can result in code that is not compliant with the 21266 C-SKY V2 ABI prologue requirements and that cannot be debugged or 21267 backtraced. It is disabled by default. 21268 21269'-msim' 21270 Links the library libsemi.a which is in compatible with simulator. 21271 Applicable to ELF compiler only. 21272 21273 21274File: gcc.info, Node: Darwin Options, Next: DEC Alpha Options, Prev: C-SKY Options, Up: Submodel Options 21275 212763.19.12 Darwin Options 21277---------------------- 21278 21279These options are defined for all architectures running the Darwin 21280operating system. 21281 21282 FSF GCC on Darwin does not create "fat" object files; it creates an 21283object file for the single architecture that GCC was built to target. 21284Apple's GCC on Darwin does create "fat" files if multiple '-arch' 21285options are used; it does so by running the compiler or linker multiple 21286times and joining the results together with 'lipo'. 21287 21288 The subtype of the file created (like 'ppc7400' or 'ppc970' or 'i686') 21289is determined by the flags that specify the ISA that GCC is targeting, 21290like '-mcpu' or '-march'. The '-force_cpusubtype_ALL' option can be 21291used to override this. 21292 21293 The Darwin tools vary in their behavior when presented with an ISA 21294mismatch. The assembler, 'as', only permits instructions to be used 21295that are valid for the subtype of the file it is generating, so you 21296cannot put 64-bit instructions in a 'ppc750' object file. The linker 21297for shared libraries, '/usr/bin/libtool', fails and prints an error if 21298asked to create a shared library with a less restrictive subtype than 21299its input files (for instance, trying to put a 'ppc970' object file in a 21300'ppc7400' library). The linker for executables, 'ld', quietly gives the 21301executable the most restrictive subtype of any of its input files. 21302 21303'-FDIR' 21304 Add the framework directory DIR to the head of the list of 21305 directories to be searched for header files. These directories are 21306 interleaved with those specified by '-I' options and are scanned in 21307 a left-to-right order. 21308 21309 A framework directory is a directory with frameworks in it. A 21310 framework is a directory with a 'Headers' and/or 'PrivateHeaders' 21311 directory contained directly in it that ends in '.framework'. The 21312 name of a framework is the name of this directory excluding the 21313 '.framework'. Headers associated with the framework are found in 21314 one of those two directories, with 'Headers' being searched first. 21315 A subframework is a framework directory that is in a framework's 21316 'Frameworks' directory. Includes of subframework headers can only 21317 appear in a header of a framework that contains the subframework, 21318 or in a sibling subframework header. Two subframeworks are 21319 siblings if they occur in the same framework. A subframework 21320 should not have the same name as a framework; a warning is issued 21321 if this is violated. Currently a subframework cannot have 21322 subframeworks; in the future, the mechanism may be extended to 21323 support this. The standard frameworks can be found in 21324 '/System/Library/Frameworks' and '/Library/Frameworks'. An example 21325 include looks like '#include <Framework/header.h>', where 21326 'Framework' denotes the name of the framework and 'header.h' is 21327 found in the 'PrivateHeaders' or 'Headers' directory. 21328 21329'-iframeworkDIR' 21330 Like '-F' except the directory is a treated as a system directory. 21331 The main difference between this '-iframework' and '-F' is that 21332 with '-iframework' the compiler does not warn about constructs 21333 contained within header files found via DIR. This option is valid 21334 only for the C family of languages. 21335 21336'-gused' 21337 Emit debugging information for symbols that are used. For stabs 21338 debugging format, this enables '-feliminate-unused-debug-symbols'. 21339 This is by default ON. 21340 21341'-gfull' 21342 Emit debugging information for all symbols and types. 21343 21344'-mmacosx-version-min=VERSION' 21345 The earliest version of MacOS X that this executable will run on is 21346 VERSION. Typical values of VERSION include '10.1', '10.2', and 21347 '10.3.9'. 21348 21349 If the compiler was built to use the system's headers by default, 21350 then the default for this option is the system version on which the 21351 compiler is running, otherwise the default is to make choices that 21352 are compatible with as many systems and code bases as possible. 21353 21354'-mkernel' 21355 Enable kernel development mode. The '-mkernel' option sets 21356 '-static', '-fno-common', '-fno-use-cxa-atexit', '-fno-exceptions', 21357 '-fno-non-call-exceptions', '-fapple-kext', '-fno-weak' and 21358 '-fno-rtti' where applicable. This mode also sets '-mno-altivec', 21359 '-msoft-float', '-fno-builtin' and '-mlong-branch' for PowerPC 21360 targets. 21361 21362'-mone-byte-bool' 21363 Override the defaults for 'bool' so that 'sizeof(bool)==1'. By 21364 default 'sizeof(bool)' is '4' when compiling for Darwin/PowerPC and 21365 '1' when compiling for Darwin/x86, so this option has no effect on 21366 x86. 21367 21368 *Warning:* The '-mone-byte-bool' switch causes GCC to generate code 21369 that is not binary compatible with code generated without that 21370 switch. Using this switch may require recompiling all other 21371 modules in a program, including system libraries. Use this switch 21372 to conform to a non-default data model. 21373 21374'-mfix-and-continue' 21375'-ffix-and-continue' 21376'-findirect-data' 21377 Generate code suitable for fast turnaround development, such as to 21378 allow GDB to dynamically load '.o' files into already-running 21379 programs. '-findirect-data' and '-ffix-and-continue' are provided 21380 for backwards compatibility. 21381 21382'-all_load' 21383 Loads all members of static archive libraries. See man ld(1) for 21384 more information. 21385 21386'-arch_errors_fatal' 21387 Cause the errors having to do with files that have the wrong 21388 architecture to be fatal. 21389 21390'-bind_at_load' 21391 Causes the output file to be marked such that the dynamic linker 21392 will bind all undefined references when the file is loaded or 21393 launched. 21394 21395'-bundle' 21396 Produce a Mach-o bundle format file. See man ld(1) for more 21397 information. 21398 21399'-bundle_loader EXECUTABLE' 21400 This option specifies the EXECUTABLE that will load the build 21401 output file being linked. See man ld(1) for more information. 21402 21403'-dynamiclib' 21404 When passed this option, GCC produces a dynamic library instead of 21405 an executable when linking, using the Darwin 'libtool' command. 21406 21407'-force_cpusubtype_ALL' 21408 This causes GCC's output file to have the 'ALL' subtype, instead of 21409 one controlled by the '-mcpu' or '-march' option. 21410 21411'-allowable_client CLIENT_NAME' 21412'-client_name' 21413'-compatibility_version' 21414'-current_version' 21415'-dead_strip' 21416'-dependency-file' 21417'-dylib_file' 21418'-dylinker_install_name' 21419'-dynamic' 21420'-exported_symbols_list' 21421'-filelist' 21422'-flat_namespace' 21423'-force_flat_namespace' 21424'-headerpad_max_install_names' 21425'-image_base' 21426'-init' 21427'-install_name' 21428'-keep_private_externs' 21429'-multi_module' 21430'-multiply_defined' 21431'-multiply_defined_unused' 21432'-noall_load' 21433'-no_dead_strip_inits_and_terms' 21434'-nofixprebinding' 21435'-nomultidefs' 21436'-noprebind' 21437'-noseglinkedit' 21438'-pagezero_size' 21439'-prebind' 21440'-prebind_all_twolevel_modules' 21441'-private_bundle' 21442'-read_only_relocs' 21443'-sectalign' 21444'-sectobjectsymbols' 21445'-whyload' 21446'-seg1addr' 21447'-sectcreate' 21448'-sectobjectsymbols' 21449'-sectorder' 21450'-segaddr' 21451'-segs_read_only_addr' 21452'-segs_read_write_addr' 21453'-seg_addr_table' 21454'-seg_addr_table_filename' 21455'-seglinkedit' 21456'-segprot' 21457'-segs_read_only_addr' 21458'-segs_read_write_addr' 21459'-single_module' 21460'-static' 21461'-sub_library' 21462'-sub_umbrella' 21463'-twolevel_namespace' 21464'-umbrella' 21465'-undefined' 21466'-unexported_symbols_list' 21467'-weak_reference_mismatches' 21468'-whatsloaded' 21469 These options are passed to the Darwin linker. The Darwin linker 21470 man page describes them in detail. 21471 21472 21473File: gcc.info, Node: DEC Alpha Options, Next: eBPF Options, Prev: Darwin Options, Up: Submodel Options 21474 214753.19.13 DEC Alpha Options 21476------------------------- 21477 21478These '-m' options are defined for the DEC Alpha implementations: 21479 21480'-mno-soft-float' 21481'-msoft-float' 21482 Use (do not use) the hardware floating-point instructions for 21483 floating-point operations. When '-msoft-float' is specified, 21484 functions in 'libgcc.a' are used to perform floating-point 21485 operations. Unless they are replaced by routines that emulate the 21486 floating-point operations, or compiled in such a way as to call 21487 such emulations routines, these routines issue floating-point 21488 operations. If you are compiling for an Alpha without 21489 floating-point operations, you must ensure that the library is 21490 built so as not to call them. 21491 21492 Note that Alpha implementations without floating-point operations 21493 are required to have floating-point registers. 21494 21495'-mfp-reg' 21496'-mno-fp-regs' 21497 Generate code that uses (does not use) the floating-point register 21498 set. '-mno-fp-regs' implies '-msoft-float'. If the floating-point 21499 register set is not used, floating-point operands are passed in 21500 integer registers as if they were integers and floating-point 21501 results are passed in '$0' instead of '$f0'. This is a 21502 non-standard calling sequence, so any function with a 21503 floating-point argument or return value called by code compiled 21504 with '-mno-fp-regs' must also be compiled with that option. 21505 21506 A typical use of this option is building a kernel that does not 21507 use, and hence need not save and restore, any floating-point 21508 registers. 21509 21510'-mieee' 21511 The Alpha architecture implements floating-point hardware optimized 21512 for maximum performance. It is mostly compliant with the IEEE 21513 floating-point standard. However, for full compliance, software 21514 assistance is required. This option generates code fully 21515 IEEE-compliant code _except_ that the INEXACT-FLAG is not 21516 maintained (see below). If this option is turned on, the 21517 preprocessor macro '_IEEE_FP' is defined during compilation. The 21518 resulting code is less efficient but is able to correctly support 21519 denormalized numbers and exceptional IEEE values such as 21520 not-a-number and plus/minus infinity. Other Alpha compilers call 21521 this option '-ieee_with_no_inexact'. 21522 21523'-mieee-with-inexact' 21524 This is like '-mieee' except the generated code also maintains the 21525 IEEE INEXACT-FLAG. Turning on this option causes the generated 21526 code to implement fully-compliant IEEE math. In addition to 21527 '_IEEE_FP', '_IEEE_FP_EXACT' is defined as a preprocessor macro. 21528 On some Alpha implementations the resulting code may execute 21529 significantly slower than the code generated by default. Since 21530 there is very little code that depends on the INEXACT-FLAG, you 21531 should normally not specify this option. Other Alpha compilers 21532 call this option '-ieee_with_inexact'. 21533 21534'-mfp-trap-mode=TRAP-MODE' 21535 This option controls what floating-point related traps are enabled. 21536 Other Alpha compilers call this option '-fptm TRAP-MODE'. The trap 21537 mode can be set to one of four values: 21538 21539 'n' 21540 This is the default (normal) setting. The only traps that are 21541 enabled are the ones that cannot be disabled in software 21542 (e.g., division by zero trap). 21543 21544 'u' 21545 In addition to the traps enabled by 'n', underflow traps are 21546 enabled as well. 21547 21548 'su' 21549 Like 'u', but the instructions are marked to be safe for 21550 software completion (see Alpha architecture manual for 21551 details). 21552 21553 'sui' 21554 Like 'su', but inexact traps are enabled as well. 21555 21556'-mfp-rounding-mode=ROUNDING-MODE' 21557 Selects the IEEE rounding mode. Other Alpha compilers call this 21558 option '-fprm ROUNDING-MODE'. The ROUNDING-MODE can be one of: 21559 21560 'n' 21561 Normal IEEE rounding mode. Floating-point numbers are rounded 21562 towards the nearest machine number or towards the even machine 21563 number in case of a tie. 21564 21565 'm' 21566 Round towards minus infinity. 21567 21568 'c' 21569 Chopped rounding mode. Floating-point numbers are rounded 21570 towards zero. 21571 21572 'd' 21573 Dynamic rounding mode. A field in the floating-point control 21574 register (FPCR, see Alpha architecture reference manual) 21575 controls the rounding mode in effect. The C library 21576 initializes this register for rounding towards plus infinity. 21577 Thus, unless your program modifies the FPCR, 'd' corresponds 21578 to round towards plus infinity. 21579 21580'-mtrap-precision=TRAP-PRECISION' 21581 In the Alpha architecture, floating-point traps are imprecise. 21582 This means without software assistance it is impossible to recover 21583 from a floating trap and program execution normally needs to be 21584 terminated. GCC can generate code that can assist operating system 21585 trap handlers in determining the exact location that caused a 21586 floating-point trap. Depending on the requirements of an 21587 application, different levels of precisions can be selected: 21588 21589 'p' 21590 Program precision. This option is the default and means a 21591 trap handler can only identify which program caused a 21592 floating-point exception. 21593 21594 'f' 21595 Function precision. The trap handler can determine the 21596 function that caused a floating-point exception. 21597 21598 'i' 21599 Instruction precision. The trap handler can determine the 21600 exact instruction that caused a floating-point exception. 21601 21602 Other Alpha compilers provide the equivalent options called 21603 '-scope_safe' and '-resumption_safe'. 21604 21605'-mieee-conformant' 21606 This option marks the generated code as IEEE conformant. You must 21607 not use this option unless you also specify '-mtrap-precision=i' 21608 and either '-mfp-trap-mode=su' or '-mfp-trap-mode=sui'. Its only 21609 effect is to emit the line '.eflag 48' in the function prologue of 21610 the generated assembly file. 21611 21612'-mbuild-constants' 21613 Normally GCC examines a 32- or 64-bit integer constant to see if it 21614 can construct it from smaller constants in two or three 21615 instructions. If it cannot, it outputs the constant as a literal 21616 and generates code to load it from the data segment at run time. 21617 21618 Use this option to require GCC to construct _all_ integer constants 21619 using code, even if it takes more instructions (the maximum is 21620 six). 21621 21622 You typically use this option to build a shared library dynamic 21623 loader. Itself a shared library, it must relocate itself in memory 21624 before it can find the variables and constants in its own data 21625 segment. 21626 21627'-mbwx' 21628'-mno-bwx' 21629'-mcix' 21630'-mno-cix' 21631'-mfix' 21632'-mno-fix' 21633'-mmax' 21634'-mno-max' 21635 Indicate whether GCC should generate code to use the optional BWX, 21636 CIX, FIX and MAX instruction sets. The default is to use the 21637 instruction sets supported by the CPU type specified via '-mcpu=' 21638 option or that of the CPU on which GCC was built if none is 21639 specified. 21640 21641'-mfloat-vax' 21642'-mfloat-ieee' 21643 Generate code that uses (does not use) VAX F and G floating-point 21644 arithmetic instead of IEEE single and double precision. 21645 21646'-mexplicit-relocs' 21647'-mno-explicit-relocs' 21648 Older Alpha assemblers provided no way to generate symbol 21649 relocations except via assembler macros. Use of these macros does 21650 not allow optimal instruction scheduling. GNU binutils as of 21651 version 2.12 supports a new syntax that allows the compiler to 21652 explicitly mark which relocations should apply to which 21653 instructions. This option is mostly useful for debugging, as GCC 21654 detects the capabilities of the assembler when it is built and sets 21655 the default accordingly. 21656 21657'-msmall-data' 21658'-mlarge-data' 21659 When '-mexplicit-relocs' is in effect, static data is accessed via 21660 "gp-relative" relocations. When '-msmall-data' is used, objects 8 21661 bytes long or smaller are placed in a "small data area" (the 21662 '.sdata' and '.sbss' sections) and are accessed via 16-bit 21663 relocations off of the '$gp' register. This limits the size of the 21664 small data area to 64KB, but allows the variables to be directly 21665 accessed via a single instruction. 21666 21667 The default is '-mlarge-data'. With this option the data area is 21668 limited to just below 2GB. Programs that require more than 2GB of 21669 data must use 'malloc' or 'mmap' to allocate the data in the heap 21670 instead of in the program's data segment. 21671 21672 When generating code for shared libraries, '-fpic' implies 21673 '-msmall-data' and '-fPIC' implies '-mlarge-data'. 21674 21675'-msmall-text' 21676'-mlarge-text' 21677 When '-msmall-text' is used, the compiler assumes that the code of 21678 the entire program (or shared library) fits in 4MB, and is thus 21679 reachable with a branch instruction. When '-msmall-data' is used, 21680 the compiler can assume that all local symbols share the same '$gp' 21681 value, and thus reduce the number of instructions required for a 21682 function call from 4 to 1. 21683 21684 The default is '-mlarge-text'. 21685 21686'-mcpu=CPU_TYPE' 21687 Set the instruction set and instruction scheduling parameters for 21688 machine type CPU_TYPE. You can specify either the 'EV' style name 21689 or the corresponding chip number. GCC supports scheduling 21690 parameters for the EV4, EV5 and EV6 family of processors and 21691 chooses the default values for the instruction set from the 21692 processor you specify. If you do not specify a processor type, GCC 21693 defaults to the processor on which the compiler was built. 21694 21695 Supported values for CPU_TYPE are 21696 21697 'ev4' 21698 'ev45' 21699 '21064' 21700 Schedules as an EV4 and has no instruction set extensions. 21701 21702 'ev5' 21703 '21164' 21704 Schedules as an EV5 and has no instruction set extensions. 21705 21706 'ev56' 21707 '21164a' 21708 Schedules as an EV5 and supports the BWX extension. 21709 21710 'pca56' 21711 '21164pc' 21712 '21164PC' 21713 Schedules as an EV5 and supports the BWX and MAX extensions. 21714 21715 'ev6' 21716 '21264' 21717 Schedules as an EV6 and supports the BWX, FIX, and MAX 21718 extensions. 21719 21720 'ev67' 21721 '21264a' 21722 Schedules as an EV6 and supports the BWX, CIX, FIX, and MAX 21723 extensions. 21724 21725 Native toolchains also support the value 'native', which selects 21726 the best architecture option for the host processor. 21727 '-mcpu=native' has no effect if GCC does not recognize the 21728 processor. 21729 21730'-mtune=CPU_TYPE' 21731 Set only the instruction scheduling parameters for machine type 21732 CPU_TYPE. The instruction set is not changed. 21733 21734 Native toolchains also support the value 'native', which selects 21735 the best architecture option for the host processor. 21736 '-mtune=native' has no effect if GCC does not recognize the 21737 processor. 21738 21739'-mmemory-latency=TIME' 21740 Sets the latency the scheduler should assume for typical memory 21741 references as seen by the application. This number is highly 21742 dependent on the memory access patterns used by the application and 21743 the size of the external cache on the machine. 21744 21745 Valid options for TIME are 21746 21747 'NUMBER' 21748 A decimal number representing clock cycles. 21749 21750 'L1' 21751 'L2' 21752 'L3' 21753 'main' 21754 The compiler contains estimates of the number of clock cycles 21755 for "typical" EV4 & EV5 hardware for the Level 1, 2 & 3 caches 21756 (also called Dcache, Scache, and Bcache), as well as to main 21757 memory. Note that L3 is only valid for EV5. 21758 21759 21760File: gcc.info, Node: eBPF Options, Next: FR30 Options, Prev: DEC Alpha Options, Up: Submodel Options 21761 217623.19.14 eBPF Options 21763-------------------- 21764 21765'-mframe-limit=BYTES' 21766 This specifies the hard limit for frame sizes, in bytes. 21767 Currently, the value that can be specified should be less than or 21768 equal to '32767'. Defaults to whatever limit is imposed by the 21769 version of the Linux kernel targeted. 21770 21771'-mkernel=VERSION' 21772 This specifies the minimum version of the kernel that will run the 21773 compiled program. GCC uses this version to determine which 21774 instructions to use, what kernel helpers to allow, etc. Currently, 21775 VERSION can be one of '4.0', '4.1', '4.2', '4.3', '4.4', '4.5', 21776 '4.6', '4.7', '4.8', '4.9', '4.10', '4.11', '4.12', '4.13', '4.14', 21777 '4.15', '4.16', '4.17', '4.18', '4.19', '4.20', '5.0', '5.1', 21778 '5.2', 'latest' and 'native'. 21779 21780'-mbig-endian' 21781 Generate code for a big-endian target. 21782 21783'-mlittle-endian' 21784 Generate code for a little-endian target. This is the default. 21785 21786'-mjmpext' 21787 Enable generation of extra conditional-branch instructions. 21788 Enabled for CPU v2 and above. 21789 21790'-mjmp32' 21791 Enable 32-bit jump instructions. Enabled for CPU v3 and above. 21792 21793'-malu32' 21794 Enable 32-bit ALU instructions. Enabled for CPU v3 and above. 21795 21796'-mcpu=VERSION' 21797 This specifies which version of the eBPF ISA to target. Newer 21798 versions may not be supported by all kernels. The default is 'v3'. 21799 21800 Supported values for VERSION are: 21801 21802 'v1' 21803 The first stable eBPF ISA with no special features or 21804 extensions. 21805 21806 'v2' 21807 Supports the jump extensions, as in '-mjmpext'. 21808 21809 'v3' 21810 All features of v2, plus: 21811 - 32-bit jump operations, as in '-mjmp32' 21812 - 32-bit ALU operations, as in '-malu32' 21813 21814'-mco-re' 21815 Enable BPF Compile Once - Run Everywhere (CO-RE) support. Requires 21816 and is implied by '-gbtf'. 21817 21818'-mno-co-re' 21819 Disable BPF Compile Once - Run Everywhere (CO-RE) support. BPF 21820 CO-RE support is enabled by default when generating BTF debug 21821 information for the BPF target. 21822 21823'-mxbpf' 21824 Generate code for an expanded version of BPF, which relaxes some of 21825 the restrictions imposed by the BPF architecture: 21826 - Save and restore callee-saved registers at function entry and 21827 exit, respectively. 21828 21829 21830File: gcc.info, Node: FR30 Options, Next: FT32 Options, Prev: eBPF Options, Up: Submodel Options 21831 218323.19.15 FR30 Options 21833-------------------- 21834 21835These options are defined specifically for the FR30 port. 21836 21837'-msmall-model' 21838 Use the small address space model. This can produce smaller code, 21839 but it does assume that all symbolic values and addresses fit into 21840 a 20-bit range. 21841 21842'-mno-lsim' 21843 Assume that runtime support has been provided and so there is no 21844 need to include the simulator library ('libsim.a') on the linker 21845 command line. 21846 21847 21848File: gcc.info, Node: FT32 Options, Next: FRV Options, Prev: FR30 Options, Up: Submodel Options 21849 218503.19.16 FT32 Options 21851-------------------- 21852 21853These options are defined specifically for the FT32 port. 21854 21855'-msim' 21856 Specifies that the program will be run on the simulator. This 21857 causes an alternate runtime startup and library to be linked. You 21858 must not use this option when generating programs that will run on 21859 real hardware; you must provide your own runtime library for 21860 whatever I/O functions are needed. 21861 21862'-mlra' 21863 Enable Local Register Allocation. This is still experimental for 21864 FT32, so by default the compiler uses standard reload. 21865 21866'-mnodiv' 21867 Do not use div and mod instructions. 21868 21869'-mft32b' 21870 Enable use of the extended instructions of the FT32B processor. 21871 21872'-mcompress' 21873 Compress all code using the Ft32B code compression scheme. 21874 21875'-mnopm' 21876 Do not generate code that reads program memory. 21877 21878 21879File: gcc.info, Node: FRV Options, Next: GNU/Linux Options, Prev: FT32 Options, Up: Submodel Options 21880 218813.19.17 FRV Options 21882------------------- 21883 21884'-mgpr-32' 21885 21886 Only use the first 32 general-purpose registers. 21887 21888'-mgpr-64' 21889 21890 Use all 64 general-purpose registers. 21891 21892'-mfpr-32' 21893 21894 Use only the first 32 floating-point registers. 21895 21896'-mfpr-64' 21897 21898 Use all 64 floating-point registers. 21899 21900'-mhard-float' 21901 21902 Use hardware instructions for floating-point operations. 21903 21904'-msoft-float' 21905 21906 Use library routines for floating-point operations. 21907 21908'-malloc-cc' 21909 21910 Dynamically allocate condition code registers. 21911 21912'-mfixed-cc' 21913 21914 Do not try to dynamically allocate condition code registers, only 21915 use 'icc0' and 'fcc0'. 21916 21917'-mdword' 21918 21919 Change ABI to use double word insns. 21920 21921'-mno-dword' 21922 21923 Do not use double word instructions. 21924 21925'-mdouble' 21926 21927 Use floating-point double instructions. 21928 21929'-mno-double' 21930 21931 Do not use floating-point double instructions. 21932 21933'-mmedia' 21934 21935 Use media instructions. 21936 21937'-mno-media' 21938 21939 Do not use media instructions. 21940 21941'-mmuladd' 21942 21943 Use multiply and add/subtract instructions. 21944 21945'-mno-muladd' 21946 21947 Do not use multiply and add/subtract instructions. 21948 21949'-mfdpic' 21950 21951 Select the FDPIC ABI, which uses function descriptors to represent 21952 pointers to functions. Without any PIC/PIE-related options, it 21953 implies '-fPIE'. With '-fpic' or '-fpie', it assumes GOT entries 21954 and small data are within a 12-bit range from the GOT base address; 21955 with '-fPIC' or '-fPIE', GOT offsets are computed with 32 bits. 21956 With a 'bfin-elf' target, this option implies '-msim'. 21957 21958'-minline-plt' 21959 21960 Enable inlining of PLT entries in function calls to functions that 21961 are not known to bind locally. It has no effect without '-mfdpic'. 21962 It's enabled by default if optimizing for speed and compiling for 21963 shared libraries (i.e., '-fPIC' or '-fpic'), or when an 21964 optimization option such as '-O3' or above is present in the 21965 command line. 21966 21967'-mTLS' 21968 21969 Assume a large TLS segment when generating thread-local code. 21970 21971'-mtls' 21972 21973 Do not assume a large TLS segment when generating thread-local 21974 code. 21975 21976'-mgprel-ro' 21977 21978 Enable the use of 'GPREL' relocations in the FDPIC ABI for data 21979 that is known to be in read-only sections. It's enabled by 21980 default, except for '-fpic' or '-fpie': even though it may help 21981 make the global offset table smaller, it trades 1 instruction for 21982 4. With '-fPIC' or '-fPIE', it trades 3 instructions for 4, one of 21983 which may be shared by multiple symbols, and it avoids the need for 21984 a GOT entry for the referenced symbol, so it's more likely to be a 21985 win. If it is not, '-mno-gprel-ro' can be used to disable it. 21986 21987'-multilib-library-pic' 21988 21989 Link with the (library, not FD) pic libraries. It's implied by 21990 '-mlibrary-pic', as well as by '-fPIC' and '-fpic' without 21991 '-mfdpic'. You should never have to use it explicitly. 21992 21993'-mlinked-fp' 21994 21995 Follow the EABI requirement of always creating a frame pointer 21996 whenever a stack frame is allocated. This option is enabled by 21997 default and can be disabled with '-mno-linked-fp'. 21998 21999'-mlong-calls' 22000 22001 Use indirect addressing to call functions outside the current 22002 compilation unit. This allows the functions to be placed anywhere 22003 within the 32-bit address space. 22004 22005'-malign-labels' 22006 22007 Try to align labels to an 8-byte boundary by inserting NOPs into 22008 the previous packet. This option only has an effect when VLIW 22009 packing is enabled. It doesn't create new packets; it merely adds 22010 NOPs to existing ones. 22011 22012'-mlibrary-pic' 22013 22014 Generate position-independent EABI code. 22015 22016'-macc-4' 22017 22018 Use only the first four media accumulator registers. 22019 22020'-macc-8' 22021 22022 Use all eight media accumulator registers. 22023 22024'-mpack' 22025 22026 Pack VLIW instructions. 22027 22028'-mno-pack' 22029 22030 Do not pack VLIW instructions. 22031 22032'-mno-eflags' 22033 22034 Do not mark ABI switches in e_flags. 22035 22036'-mcond-move' 22037 22038 Enable the use of conditional-move instructions (default). 22039 22040 This switch is mainly for debugging the compiler and will likely be 22041 removed in a future version. 22042 22043'-mno-cond-move' 22044 22045 Disable the use of conditional-move instructions. 22046 22047 This switch is mainly for debugging the compiler and will likely be 22048 removed in a future version. 22049 22050'-mscc' 22051 22052 Enable the use of conditional set instructions (default). 22053 22054 This switch is mainly for debugging the compiler and will likely be 22055 removed in a future version. 22056 22057'-mno-scc' 22058 22059 Disable the use of conditional set instructions. 22060 22061 This switch is mainly for debugging the compiler and will likely be 22062 removed in a future version. 22063 22064'-mcond-exec' 22065 22066 Enable the use of conditional execution (default). 22067 22068 This switch is mainly for debugging the compiler and will likely be 22069 removed in a future version. 22070 22071'-mno-cond-exec' 22072 22073 Disable the use of conditional execution. 22074 22075 This switch is mainly for debugging the compiler and will likely be 22076 removed in a future version. 22077 22078'-mvliw-branch' 22079 22080 Run a pass to pack branches into VLIW instructions (default). 22081 22082 This switch is mainly for debugging the compiler and will likely be 22083 removed in a future version. 22084 22085'-mno-vliw-branch' 22086 22087 Do not run a pass to pack branches into VLIW instructions. 22088 22089 This switch is mainly for debugging the compiler and will likely be 22090 removed in a future version. 22091 22092'-mmulti-cond-exec' 22093 22094 Enable optimization of '&&' and '||' in conditional execution 22095 (default). 22096 22097 This switch is mainly for debugging the compiler and will likely be 22098 removed in a future version. 22099 22100'-mno-multi-cond-exec' 22101 22102 Disable optimization of '&&' and '||' in conditional execution. 22103 22104 This switch is mainly for debugging the compiler and will likely be 22105 removed in a future version. 22106 22107'-mnested-cond-exec' 22108 22109 Enable nested conditional execution optimizations (default). 22110 22111 This switch is mainly for debugging the compiler and will likely be 22112 removed in a future version. 22113 22114'-mno-nested-cond-exec' 22115 22116 Disable nested conditional execution optimizations. 22117 22118 This switch is mainly for debugging the compiler and will likely be 22119 removed in a future version. 22120 22121'-moptimize-membar' 22122 22123 This switch removes redundant 'membar' instructions from the 22124 compiler-generated code. It is enabled by default. 22125 22126'-mno-optimize-membar' 22127 22128 This switch disables the automatic removal of redundant 'membar' 22129 instructions from the generated code. 22130 22131'-mtomcat-stats' 22132 22133 Cause gas to print out tomcat statistics. 22134 22135'-mcpu=CPU' 22136 22137 Select the processor type for which to generate code. Possible 22138 values are 'frv', 'fr550', 'tomcat', 'fr500', 'fr450', 'fr405', 22139 'fr400', 'fr300' and 'simple'. 22140 22141 22142File: gcc.info, Node: GNU/Linux Options, Next: H8/300 Options, Prev: FRV Options, Up: Submodel Options 22143 221443.19.18 GNU/Linux Options 22145------------------------- 22146 22147These '-m' options are defined for GNU/Linux targets: 22148 22149'-mglibc' 22150 Use the GNU C library. This is the default except on 22151 '*-*-linux-*uclibc*', '*-*-linux-*musl*' and '*-*-linux-*android*' 22152 targets. 22153 22154'-muclibc' 22155 Use uClibc C library. This is the default on '*-*-linux-*uclibc*' 22156 targets. 22157 22158'-mmusl' 22159 Use the musl C library. This is the default on '*-*-linux-*musl*' 22160 targets. 22161 22162'-mbionic' 22163 Use Bionic C library. This is the default on '*-*-linux-*android*' 22164 targets. 22165 22166'-mandroid' 22167 Compile code compatible with Android platform. This is the default 22168 on '*-*-linux-*android*' targets. 22169 22170 When compiling, this option enables '-mbionic', '-fPIC', 22171 '-fno-exceptions' and '-fno-rtti' by default. When linking, this 22172 option makes the GCC driver pass Android-specific options to the 22173 linker. Finally, this option causes the preprocessor macro 22174 '__ANDROID__' to be defined. 22175 22176'-tno-android-cc' 22177 Disable compilation effects of '-mandroid', i.e., do not enable 22178 '-mbionic', '-fPIC', '-fno-exceptions' and '-fno-rtti' by default. 22179 22180'-tno-android-ld' 22181 Disable linking effects of '-mandroid', i.e., pass standard Linux 22182 linking options to the linker. 22183 22184 22185File: gcc.info, Node: H8/300 Options, Next: HPPA Options, Prev: GNU/Linux Options, Up: Submodel Options 22186 221873.19.19 H8/300 Options 22188---------------------- 22189 22190These '-m' options are defined for the H8/300 implementations: 22191 22192'-mrelax' 22193 Shorten some address references at link time, when possible; uses 22194 the linker option '-relax'. *Note 'ld' and the H8/300: (ld)H8/300, 22195 for a fuller description. 22196 22197'-mh' 22198 Generate code for the H8/300H. 22199 22200'-ms' 22201 Generate code for the H8S. 22202 22203'-mn' 22204 Generate code for the H8S and H8/300H in the normal mode. This 22205 switch must be used either with '-mh' or '-ms'. 22206 22207'-ms2600' 22208 Generate code for the H8S/2600. This switch must be used with 22209 '-ms'. 22210 22211'-mexr' 22212 Extended registers are stored on stack before execution of function 22213 with monitor attribute. Default option is '-mexr'. This option is 22214 valid only for H8S targets. 22215 22216'-mno-exr' 22217 Extended registers are not stored on stack before execution of 22218 function with monitor attribute. Default option is '-mno-exr'. 22219 This option is valid only for H8S targets. 22220 22221'-mint32' 22222 Make 'int' data 32 bits by default. 22223 22224'-malign-300' 22225 On the H8/300H and H8S, use the same alignment rules as for the 22226 H8/300. The default for the H8/300H and H8S is to align longs and 22227 floats on 4-byte boundaries. '-malign-300' causes them to be 22228 aligned on 2-byte boundaries. This option has no effect on the 22229 H8/300. 22230 22231 22232File: gcc.info, Node: HPPA Options, Next: IA-64 Options, Prev: H8/300 Options, Up: Submodel Options 22233 222343.19.20 HPPA Options 22235-------------------- 22236 22237These '-m' options are defined for the HPPA family of computers: 22238 22239'-march=ARCHITECTURE-TYPE' 22240 Generate code for the specified architecture. The choices for 22241 ARCHITECTURE-TYPE are '1.0' for PA 1.0, '1.1' for PA 1.1, and '2.0' 22242 for PA 2.0 processors. Refer to '/usr/lib/sched.models' on an 22243 HP-UX system to determine the proper architecture option for your 22244 machine. Code compiled for lower numbered architectures runs on 22245 higher numbered architectures, but not the other way around. 22246 22247'-mpa-risc-1-0' 22248'-mpa-risc-1-1' 22249'-mpa-risc-2-0' 22250 Synonyms for '-march=1.0', '-march=1.1', and '-march=2.0' 22251 respectively. 22252 22253'-mcaller-copies' 22254 The caller copies function arguments passed by hidden reference. 22255 This option should be used with care as it is not compatible with 22256 the default 32-bit runtime. However, only aggregates larger than 22257 eight bytes are passed by hidden reference and the option provides 22258 better compatibility with OpenMP. 22259 22260'-mjump-in-delay' 22261 This option is ignored and provided for compatibility purposes 22262 only. 22263 22264'-mdisable-fpregs' 22265 Prevent floating-point registers from being used in any manner. 22266 This is necessary for compiling kernels that perform lazy context 22267 switching of floating-point registers. If you use this option and 22268 attempt to perform floating-point operations, the compiler aborts. 22269 22270'-mdisable-indexing' 22271 Prevent the compiler from using indexing address modes. This 22272 avoids some rather obscure problems when compiling MIG generated 22273 code under MACH. 22274 22275'-mno-space-regs' 22276 Generate code that assumes the target has no space registers. This 22277 allows GCC to generate faster indirect calls and use unscaled index 22278 address modes. 22279 22280 Such code is suitable for level 0 PA systems and kernels. 22281 22282'-mfast-indirect-calls' 22283 Generate code that assumes calls never cross space boundaries. 22284 This allows GCC to emit code that performs faster indirect calls. 22285 22286 This option does not work in the presence of shared libraries or 22287 nested functions. 22288 22289'-mfixed-range=REGISTER-RANGE' 22290 Generate code treating the given register range as fixed registers. 22291 A fixed register is one that the register allocator cannot use. 22292 This is useful when compiling kernel code. A register range is 22293 specified as two registers separated by a dash. Multiple register 22294 ranges can be specified separated by a comma. 22295 22296'-mlong-load-store' 22297 Generate 3-instruction load and store sequences as sometimes 22298 required by the HP-UX 10 linker. This is equivalent to the '+k' 22299 option to the HP compilers. 22300 22301'-mportable-runtime' 22302 Use the portable calling conventions proposed by HP for ELF 22303 systems. 22304 22305'-mgas' 22306 Enable the use of assembler directives only GAS understands. 22307 22308'-mschedule=CPU-TYPE' 22309 Schedule code according to the constraints for the machine type 22310 CPU-TYPE. The choices for CPU-TYPE are '700' '7100', '7100LC', 22311 '7200', '7300' and '8000'. Refer to '/usr/lib/sched.models' on an 22312 HP-UX system to determine the proper scheduling option for your 22313 machine. The default scheduling is '8000'. 22314 22315'-mlinker-opt' 22316 Enable the optimization pass in the HP-UX linker. Note this makes 22317 symbolic debugging impossible. It also triggers a bug in the HP-UX 22318 8 and HP-UX 9 linkers in which they give bogus error messages when 22319 linking some programs. 22320 22321'-msoft-float' 22322 Generate output containing library calls for floating point. 22323 *Warning:* the requisite libraries are not available for all HPPA 22324 targets. Normally the facilities of the machine's usual C compiler 22325 are used, but this cannot be done directly in cross-compilation. 22326 You must make your own arrangements to provide suitable library 22327 functions for cross-compilation. 22328 22329 '-msoft-float' changes the calling convention in the output file; 22330 therefore, it is only useful if you compile _all_ of a program with 22331 this option. In particular, you need to compile 'libgcc.a', the 22332 library that comes with GCC, with '-msoft-float' in order for this 22333 to work. 22334 22335'-msio' 22336 Generate the predefine, '_SIO', for server IO. The default is 22337 '-mwsio'. This generates the predefines, '__hp9000s700', 22338 '__hp9000s700__' and '_WSIO', for workstation IO. These options 22339 are available under HP-UX and HI-UX. 22340 22341'-mgnu-ld' 22342 Use options specific to GNU 'ld'. This passes '-shared' to 'ld' 22343 when building a shared library. It is the default when GCC is 22344 configured, explicitly or implicitly, with the GNU linker. This 22345 option does not affect which 'ld' is called; it only changes what 22346 parameters are passed to that 'ld'. The 'ld' that is called is 22347 determined by the '--with-ld' configure option, GCC's program 22348 search path, and finally by the user's 'PATH'. The linker used by 22349 GCC can be printed using 'which `gcc -print-prog-name=ld`'. This 22350 option is only available on the 64-bit HP-UX GCC, i.e. configured 22351 with 'hppa*64*-*-hpux*'. 22352 22353'-mhp-ld' 22354 Use options specific to HP 'ld'. This passes '-b' to 'ld' when 22355 building a shared library and passes '+Accept TypeMismatch' to 'ld' 22356 on all links. It is the default when GCC is configured, explicitly 22357 or implicitly, with the HP linker. This option does not affect 22358 which 'ld' is called; it only changes what parameters are passed to 22359 that 'ld'. The 'ld' that is called is determined by the 22360 '--with-ld' configure option, GCC's program search path, and 22361 finally by the user's 'PATH'. The linker used by GCC can be 22362 printed using 'which `gcc -print-prog-name=ld`'. This option is 22363 only available on the 64-bit HP-UX GCC, i.e. configured with 22364 'hppa*64*-*-hpux*'. 22365 22366'-mlong-calls' 22367 Generate code that uses long call sequences. This ensures that a 22368 call is always able to reach linker generated stubs. The default 22369 is to generate long calls only when the distance from the call site 22370 to the beginning of the function or translation unit, as the case 22371 may be, exceeds a predefined limit set by the branch type being 22372 used. The limits for normal calls are 7,600,000 and 240,000 bytes, 22373 respectively for the PA 2.0 and PA 1.X architectures. Sibcalls are 22374 always limited at 240,000 bytes. 22375 22376 Distances are measured from the beginning of functions when using 22377 the '-ffunction-sections' option, or when using the '-mgas' and 22378 '-mno-portable-runtime' options together under HP-UX with the SOM 22379 linker. 22380 22381 It is normally not desirable to use this option as it degrades 22382 performance. However, it may be useful in large applications, 22383 particularly when partial linking is used to build the application. 22384 22385 The types of long calls used depends on the capabilities of the 22386 assembler and linker, and the type of code being generated. The 22387 impact on systems that support long absolute calls, and long pic 22388 symbol-difference or pc-relative calls should be relatively small. 22389 However, an indirect call is used on 32-bit ELF systems in pic code 22390 and it is quite long. 22391 22392'-munix=UNIX-STD' 22393 Generate compiler predefines and select a startfile for the 22394 specified UNIX standard. The choices for UNIX-STD are '93', '95' 22395 and '98'. '93' is supported on all HP-UX versions. '95' is 22396 available on HP-UX 10.10 and later. '98' is available on HP-UX 22397 11.11 and later. The default values are '93' for HP-UX 10.00, '95' 22398 for HP-UX 10.10 though to 11.00, and '98' for HP-UX 11.11 and 22399 later. 22400 22401 '-munix=93' provides the same predefines as GCC 3.3 and 3.4. 22402 '-munix=95' provides additional predefines for 'XOPEN_UNIX' and 22403 '_XOPEN_SOURCE_EXTENDED', and the startfile 'unix95.o'. 22404 '-munix=98' provides additional predefines for '_XOPEN_UNIX', 22405 '_XOPEN_SOURCE_EXTENDED', '_INCLUDE__STDC_A1_SOURCE' and 22406 '_INCLUDE_XOPEN_SOURCE_500', and the startfile 'unix98.o'. 22407 22408 It is _important_ to note that this option changes the interfaces 22409 for various library routines. It also affects the operational 22410 behavior of the C library. Thus, _extreme_ care is needed in using 22411 this option. 22412 22413 Library code that is intended to operate with more than one UNIX 22414 standard must test, set and restore the variable 22415 '__xpg4_extended_mask' as appropriate. Most GNU software doesn't 22416 provide this capability. 22417 22418'-nolibdld' 22419 Suppress the generation of link options to search libdld.sl when 22420 the '-static' option is specified on HP-UX 10 and later. 22421 22422'-static' 22423 The HP-UX implementation of setlocale in libc has a dependency on 22424 libdld.sl. There isn't an archive version of libdld.sl. Thus, 22425 when the '-static' option is specified, special link options are 22426 needed to resolve this dependency. 22427 22428 On HP-UX 10 and later, the GCC driver adds the necessary options to 22429 link with libdld.sl when the '-static' option is specified. This 22430 causes the resulting binary to be dynamic. On the 64-bit port, the 22431 linkers generate dynamic binaries by default in any case. The 22432 '-nolibdld' option can be used to prevent the GCC driver from 22433 adding these link options. 22434 22435'-threads' 22436 Add support for multithreading with the "dce thread" library under 22437 HP-UX. This option sets flags for both the preprocessor and 22438 linker. 22439 22440 22441File: gcc.info, Node: IA-64 Options, Next: LM32 Options, Prev: HPPA Options, Up: Submodel Options 22442 224433.19.21 IA-64 Options 22444--------------------- 22445 22446These are the '-m' options defined for the Intel IA-64 architecture. 22447 22448'-mbig-endian' 22449 Generate code for a big-endian target. This is the default for 22450 HP-UX. 22451 22452'-mlittle-endian' 22453 Generate code for a little-endian target. This is the default for 22454 AIX5 and GNU/Linux. 22455 22456'-mgnu-as' 22457'-mno-gnu-as' 22458 Generate (or don't) code for the GNU assembler. This is the 22459 default. 22460 22461'-mgnu-ld' 22462'-mno-gnu-ld' 22463 Generate (or don't) code for the GNU linker. This is the default. 22464 22465'-mno-pic' 22466 Generate code that does not use a global pointer register. The 22467 result is not position independent code, and violates the IA-64 22468 ABI. 22469 22470'-mvolatile-asm-stop' 22471'-mno-volatile-asm-stop' 22472 Generate (or don't) a stop bit immediately before and after 22473 volatile asm statements. 22474 22475'-mregister-names' 22476'-mno-register-names' 22477 Generate (or don't) 'in', 'loc', and 'out' register names for the 22478 stacked registers. This may make assembler output more readable. 22479 22480'-mno-sdata' 22481'-msdata' 22482 Disable (or enable) optimizations that use the small data section. 22483 This may be useful for working around optimizer bugs. 22484 22485'-mconstant-gp' 22486 Generate code that uses a single constant global pointer value. 22487 This is useful when compiling kernel code. 22488 22489'-mauto-pic' 22490 Generate code that is self-relocatable. This implies 22491 '-mconstant-gp'. This is useful when compiling firmware code. 22492 22493'-minline-float-divide-min-latency' 22494 Generate code for inline divides of floating-point values using the 22495 minimum latency algorithm. 22496 22497'-minline-float-divide-max-throughput' 22498 Generate code for inline divides of floating-point values using the 22499 maximum throughput algorithm. 22500 22501'-mno-inline-float-divide' 22502 Do not generate inline code for divides of floating-point values. 22503 22504'-minline-int-divide-min-latency' 22505 Generate code for inline divides of integer values using the 22506 minimum latency algorithm. 22507 22508'-minline-int-divide-max-throughput' 22509 Generate code for inline divides of integer values using the 22510 maximum throughput algorithm. 22511 22512'-mno-inline-int-divide' 22513 Do not generate inline code for divides of integer values. 22514 22515'-minline-sqrt-min-latency' 22516 Generate code for inline square roots using the minimum latency 22517 algorithm. 22518 22519'-minline-sqrt-max-throughput' 22520 Generate code for inline square roots using the maximum throughput 22521 algorithm. 22522 22523'-mno-inline-sqrt' 22524 Do not generate inline code for 'sqrt'. 22525 22526'-mfused-madd' 22527'-mno-fused-madd' 22528 Do (don't) generate code that uses the fused multiply/add or 22529 multiply/subtract instructions. The default is to use these 22530 instructions. 22531 22532'-mno-dwarf2-asm' 22533'-mdwarf2-asm' 22534 Don't (or do) generate assembler code for the DWARF line number 22535 debugging info. This may be useful when not using the GNU 22536 assembler. 22537 22538'-mearly-stop-bits' 22539'-mno-early-stop-bits' 22540 Allow stop bits to be placed earlier than immediately preceding the 22541 instruction that triggered the stop bit. This can improve 22542 instruction scheduling, but does not always do so. 22543 22544'-mfixed-range=REGISTER-RANGE' 22545 Generate code treating the given register range as fixed registers. 22546 A fixed register is one that the register allocator cannot use. 22547 This is useful when compiling kernel code. A register range is 22548 specified as two registers separated by a dash. Multiple register 22549 ranges can be specified separated by a comma. 22550 22551'-mtls-size=TLS-SIZE' 22552 Specify bit size of immediate TLS offsets. Valid values are 14, 22553 22, and 64. 22554 22555'-mtune=CPU-TYPE' 22556 Tune the instruction scheduling for a particular CPU, Valid values 22557 are 'itanium', 'itanium1', 'merced', 'itanium2', and 'mckinley'. 22558 22559'-milp32' 22560'-mlp64' 22561 Generate code for a 32-bit or 64-bit environment. The 32-bit 22562 environment sets int, long and pointer to 32 bits. The 64-bit 22563 environment sets int to 32 bits and long and pointer to 64 bits. 22564 These are HP-UX specific flags. 22565 22566'-mno-sched-br-data-spec' 22567'-msched-br-data-spec' 22568 (Dis/En)able data speculative scheduling before reload. This 22569 results in generation of 'ld.a' instructions and the corresponding 22570 check instructions ('ld.c' / 'chk.a'). The default setting is 22571 disabled. 22572 22573'-msched-ar-data-spec' 22574'-mno-sched-ar-data-spec' 22575 (En/Dis)able data speculative scheduling after reload. This 22576 results in generation of 'ld.a' instructions and the corresponding 22577 check instructions ('ld.c' / 'chk.a'). The default setting is 22578 enabled. 22579 22580'-mno-sched-control-spec' 22581'-msched-control-spec' 22582 (Dis/En)able control speculative scheduling. This feature is 22583 available only during region scheduling (i.e. before reload). This 22584 results in generation of the 'ld.s' instructions and the 22585 corresponding check instructions 'chk.s'. The default setting is 22586 disabled. 22587 22588'-msched-br-in-data-spec' 22589'-mno-sched-br-in-data-spec' 22590 (En/Dis)able speculative scheduling of the instructions that are 22591 dependent on the data speculative loads before reload. This is 22592 effective only with '-msched-br-data-spec' enabled. The default 22593 setting is enabled. 22594 22595'-msched-ar-in-data-spec' 22596'-mno-sched-ar-in-data-spec' 22597 (En/Dis)able speculative scheduling of the instructions that are 22598 dependent on the data speculative loads after reload. This is 22599 effective only with '-msched-ar-data-spec' enabled. The default 22600 setting is enabled. 22601 22602'-msched-in-control-spec' 22603'-mno-sched-in-control-spec' 22604 (En/Dis)able speculative scheduling of the instructions that are 22605 dependent on the control speculative loads. This is effective only 22606 with '-msched-control-spec' enabled. The default setting is 22607 enabled. 22608 22609'-mno-sched-prefer-non-data-spec-insns' 22610'-msched-prefer-non-data-spec-insns' 22611 If enabled, data-speculative instructions are chosen for schedule 22612 only if there are no other choices at the moment. This makes the 22613 use of the data speculation much more conservative. The default 22614 setting is disabled. 22615 22616'-mno-sched-prefer-non-control-spec-insns' 22617'-msched-prefer-non-control-spec-insns' 22618 If enabled, control-speculative instructions are chosen for 22619 schedule only if there are no other choices at the moment. This 22620 makes the use of the control speculation much more conservative. 22621 The default setting is disabled. 22622 22623'-mno-sched-count-spec-in-critical-path' 22624'-msched-count-spec-in-critical-path' 22625 If enabled, speculative dependencies are considered during 22626 computation of the instructions priorities. This makes the use of 22627 the speculation a bit more conservative. The default setting is 22628 disabled. 22629 22630'-msched-spec-ldc' 22631 Use a simple data speculation check. This option is on by default. 22632 22633'-msched-control-spec-ldc' 22634 Use a simple check for control speculation. This option is on by 22635 default. 22636 22637'-msched-stop-bits-after-every-cycle' 22638 Place a stop bit after every cycle when scheduling. This option is 22639 on by default. 22640 22641'-msched-fp-mem-deps-zero-cost' 22642 Assume that floating-point stores and loads are not likely to cause 22643 a conflict when placed into the same instruction group. This 22644 option is disabled by default. 22645 22646'-msel-sched-dont-check-control-spec' 22647 Generate checks for control speculation in selective scheduling. 22648 This flag is disabled by default. 22649 22650'-msched-max-memory-insns=MAX-INSNS' 22651 Limit on the number of memory insns per instruction group, giving 22652 lower priority to subsequent memory insns attempting to schedule in 22653 the same instruction group. Frequently useful to prevent cache 22654 bank conflicts. The default value is 1. 22655 22656'-msched-max-memory-insns-hard-limit' 22657 Makes the limit specified by 'msched-max-memory-insns' a hard 22658 limit, disallowing more than that number in an instruction group. 22659 Otherwise, the limit is "soft", meaning that non-memory operations 22660 are preferred when the limit is reached, but memory operations may 22661 still be scheduled. 22662 22663 22664File: gcc.info, Node: LM32 Options, Next: LoongArch Options, Prev: IA-64 Options, Up: Submodel Options 22665 226663.19.22 LM32 Options 22667-------------------- 22668 22669These '-m' options are defined for the LatticeMico32 architecture: 22670 22671'-mbarrel-shift-enabled' 22672 Enable barrel-shift instructions. 22673 22674'-mdivide-enabled' 22675 Enable divide and modulus instructions. 22676 22677'-mmultiply-enabled' 22678 Enable multiply instructions. 22679 22680'-msign-extend-enabled' 22681 Enable sign extend instructions. 22682 22683'-muser-enabled' 22684 Enable user-defined instructions. 22685 22686 22687File: gcc.info, Node: LoongArch Options, Next: M32C Options, Prev: LM32 Options, Up: Submodel Options 22688 226893.19.23 LoongArch Options 22690------------------------- 22691 22692These command-line options are defined for LoongArch targets: 22693 22694'-march=CPU-TYPE' 22695 Generate instructions for the machine type CPU-TYPE. In contrast 22696 to '-mtune=CPU-TYPE', which merely tunes the generated code for the 22697 specified CPU-TYPE, '-march=CPU-TYPE' allows GCC to generate code 22698 that may not run at all on processors other than the one indicated. 22699 Specifying '-march=CPU-TYPE' implies '-mtune=CPU-TYPE', except 22700 where noted otherwise. 22701 22702 The choices for CPU-TYPE are: 22703 22704 'native' 22705 This selects the CPU to generate code for at compilation time 22706 by determining the processor type of the compiling machine. 22707 Using '-march=native' enables all instruction subsets 22708 supported by the local machine (hence the result might not run 22709 on different machines). Using '-mtune=native' produces code 22710 optimized for the local machine under the constraints of the 22711 selected instruction set. 22712 'loongarch64' 22713 A generic CPU with 64-bit extensions. 22714 'la464' 22715 LoongArch LA464 CPU with LBT, LSX, LASX, LVZ. 22716 22717'-mtune=CPU-TYPE' 22718 Optimize the output for the given processor, specified by 22719 microarchitecture name. 22720 22721'-mabi=BASE-ABI-TYPE' 22722 Generate code for the specified calling convention. BASE-ABI-TYPE 22723 can be one of: 22724 'lp64d' 22725 Uses 64-bit general purpose registers and 32/64-bit 22726 floating-point registers for parameter passing. Data model is 22727 LP64, where 'int' is 32 bits, while 'long int' and pointers 22728 are 64 bits. 22729 'lp64f' 22730 Uses 64-bit general purpose registers and 32-bit 22731 floating-point registers for parameter passing. Data model is 22732 LP64, where 'int' is 32 bits, while 'long int' and pointers 22733 are 64 bits. 22734 'lp64s' 22735 Uses 64-bit general purpose registers and no floating-point 22736 registers for parameter passing. Data model is LP64, where 22737 'int' is 32 bits, while 'long int' and pointers are 64 bits. 22738 22739'-mfpu=FPU-TYPE' 22740 Generate code for the specified FPU type, which can be one of: 22741 '64' 22742 Allow the use of hardware floating-point instructions for 22743 32-bit and 64-bit operations. 22744 '32' 22745 Allow the use of hardware floating-point instructions for 22746 32-bit operations. 22747 'none' 22748 '0' 22749 Prevent the use of hardware floating-point instructions. 22750 22751'-msoft-float' 22752 Force '-mfpu=none' and prevents the use of floating-point registers 22753 for parameter passing. This option may change the target ABI. 22754 22755'-msingle-float' 22756 Force '-mfpu=32' and allow the use of 32-bit floating-point 22757 registers for parameter passing. This option may change the target 22758 ABI. 22759 22760'-mdouble-float' 22761 Force '-mfpu=64' and allow the use of 32/64-bit floating-point 22762 registers for parameter passing. This option may change the target 22763 ABI. 22764 22765'-mbranch-cost=N' 22766 Set the cost of branches to roughly N instructions. 22767 22768'-mcheck-zero-division' 22769'-mno-check-zero-divison' 22770 Trap (do not trap) on integer division by zero. The default is 22771 '-mcheck-zero-division' for '-O0' or '-Og', and 22772 '-mno-check-zero-division' for other optimization levels. 22773 22774'-mcond-move-int' 22775'-mno-cond-move-int' 22776 Conditional moves for integral data in general-purpose registers 22777 are enabled (disabled). The default is '-mcond-move-int'. 22778 22779'-mcond-move-float' 22780'-mno-cond-move-float' 22781 Conditional moves for floating-point registers are enabled 22782 (disabled). The default is '-mcond-move-float'. 22783 22784'-mmemcpy' 22785'-mno-memcpy' 22786 Force (do not force) the use of 'memcpy' for non-trivial block 22787 moves. The default is '-mno-memcpy', which allows GCC to inline 22788 most constant-sized copies. Setting optimization level to '-Os' 22789 also forces the use of 'memcpy', but '-mno-memcpy' may override 22790 this behavior if explicitly specified, regardless of the order 22791 these options on the command line. 22792 22793'-mstrict-align' 22794'-mno-strict-align' 22795 Avoid or allow generating memory accesses that may not be aligned 22796 on a natural object boundary as described in the architecture 22797 specification. The default is '-mno-strict-align'. 22798 22799'-msmall-data-limit=NUMBER' 22800 Put global and static data smaller than NUMBER bytes into a special 22801 section (on some targets). The default value is 0. 22802 22803'-mmax-inline-memcpy-size=N' 22804 Inline all block moves (such as calls to 'memcpy' or structure 22805 copies) less than or equal to N bytes. The default value of N is 22806 1024. 22807 22808'-mcmodel=CODE-MODEL' 22809 Set the code model to one of: 22810 'tiny-static' 22811 * local symbol and global strong symbol: The data section 22812 must be within +/-2MiB addressing space. The text 22813 section must be within +/-128MiB addressing space. 22814 * global weak symbol: The got table must be within +/-2GiB 22815 addressing space. 22816 22817 'tiny' 22818 * local symbol: The data section must be within +/-2MiB 22819 addressing space. The text section must be within 22820 +/-128MiB addressing space. 22821 * global symbol: The got table must be within +/-2GiB 22822 addressing space. 22823 22824 'normal' 22825 * local symbol: The data section must be within +/-2GiB 22826 addressing space. The text section must be within 22827 +/-128MiB addressing space. 22828 * global symbol: The got table must be within +/-2GiB 22829 addressing space. 22830 22831 'large' 22832 * local symbol: The data section must be within +/-2GiB 22833 addressing space. The text section must be within 22834 +/-128GiB addressing space. 22835 * global symbol: The got table must be within +/-2GiB 22836 addressing space. 22837 22838 'extreme(Not implemented yet)' 22839 * local symbol: The data and text section must be within 22840 +/-8EiB addressing space. 22841 * global symbol: The data got table must be within +/-8EiB 22842 addressing space. 22843 The default code model is 'normal'. 22844 22845 22846File: gcc.info, Node: M32C Options, Next: M32R/D Options, Prev: LoongArch Options, Up: Submodel Options 22847 228483.19.24 M32C Options 22849-------------------- 22850 22851'-mcpu=NAME' 22852 Select the CPU for which code is generated. NAME may be one of 22853 'r8c' for the R8C/Tiny series, 'm16c' for the M16C (up to /60) 22854 series, 'm32cm' for the M16C/80 series, or 'm32c' for the M32C/80 22855 series. 22856 22857'-msim' 22858 Specifies that the program will be run on the simulator. This 22859 causes an alternate runtime library to be linked in which supports, 22860 for example, file I/O. You must not use this option when 22861 generating programs that will run on real hardware; you must 22862 provide your own runtime library for whatever I/O functions are 22863 needed. 22864 22865'-memregs=NUMBER' 22866 Specifies the number of memory-based pseudo-registers GCC uses 22867 during code generation. These pseudo-registers are used like real 22868 registers, so there is a tradeoff between GCC's ability to fit the 22869 code into available registers, and the performance penalty of using 22870 memory instead of registers. Note that all modules in a program 22871 must be compiled with the same value for this option. Because of 22872 that, you must not use this option with GCC's default runtime 22873 libraries. 22874 22875 22876File: gcc.info, Node: M32R/D Options, Next: M680x0 Options, Prev: M32C Options, Up: Submodel Options 22877 228783.19.25 M32R/D Options 22879---------------------- 22880 22881These '-m' options are defined for Renesas M32R/D architectures: 22882 22883'-m32r2' 22884 Generate code for the M32R/2. 22885 22886'-m32rx' 22887 Generate code for the M32R/X. 22888 22889'-m32r' 22890 Generate code for the M32R. This is the default. 22891 22892'-mmodel=small' 22893 Assume all objects live in the lower 16MB of memory (so that their 22894 addresses can be loaded with the 'ld24' instruction), and assume 22895 all subroutines are reachable with the 'bl' instruction. This is 22896 the default. 22897 22898 The addressability of a particular object can be set with the 22899 'model' attribute. 22900 22901'-mmodel=medium' 22902 Assume objects may be anywhere in the 32-bit address space (the 22903 compiler generates 'seth/add3' instructions to load their 22904 addresses), and assume all subroutines are reachable with the 'bl' 22905 instruction. 22906 22907'-mmodel=large' 22908 Assume objects may be anywhere in the 32-bit address space (the 22909 compiler generates 'seth/add3' instructions to load their 22910 addresses), and assume subroutines may not be reachable with the 22911 'bl' instruction (the compiler generates the much slower 22912 'seth/add3/jl' instruction sequence). 22913 22914'-msdata=none' 22915 Disable use of the small data area. Variables are put into one of 22916 '.data', '.bss', or '.rodata' (unless the 'section' attribute has 22917 been specified). This is the default. 22918 22919 The small data area consists of sections '.sdata' and '.sbss'. 22920 Objects may be explicitly put in the small data area with the 22921 'section' attribute using one of these sections. 22922 22923'-msdata=sdata' 22924 Put small global and static data in the small data area, but do not 22925 generate special code to reference them. 22926 22927'-msdata=use' 22928 Put small global and static data in the small data area, and 22929 generate special instructions to reference them. 22930 22931'-G NUM' 22932 Put global and static objects less than or equal to NUM bytes into 22933 the small data or BSS sections instead of the normal data or BSS 22934 sections. The default value of NUM is 8. The '-msdata' option 22935 must be set to one of 'sdata' or 'use' for this option to have any 22936 effect. 22937 22938 All modules should be compiled with the same '-G NUM' value. 22939 Compiling with different values of NUM may or may not work; if it 22940 doesn't the linker gives an error message--incorrect code is not 22941 generated. 22942 22943'-mdebug' 22944 Makes the M32R-specific code in the compiler display some 22945 statistics that might help in debugging programs. 22946 22947'-malign-loops' 22948 Align all loops to a 32-byte boundary. 22949 22950'-mno-align-loops' 22951 Do not enforce a 32-byte alignment for loops. This is the default. 22952 22953'-missue-rate=NUMBER' 22954 Issue NUMBER instructions per cycle. NUMBER can only be 1 or 2. 22955 22956'-mbranch-cost=NUMBER' 22957 NUMBER can only be 1 or 2. If it is 1 then branches are preferred 22958 over conditional code, if it is 2, then the opposite applies. 22959 22960'-mflush-trap=NUMBER' 22961 Specifies the trap number to use to flush the cache. The default 22962 is 12. Valid numbers are between 0 and 15 inclusive. 22963 22964'-mno-flush-trap' 22965 Specifies that the cache cannot be flushed by using a trap. 22966 22967'-mflush-func=NAME' 22968 Specifies the name of the operating system function to call to 22969 flush the cache. The default is '_flush_cache', but a function 22970 call is only used if a trap is not available. 22971 22972'-mno-flush-func' 22973 Indicates that there is no OS function for flushing the cache. 22974 22975 22976File: gcc.info, Node: M680x0 Options, Next: MCore Options, Prev: M32R/D Options, Up: Submodel Options 22977 229783.19.26 M680x0 Options 22979---------------------- 22980 22981These are the '-m' options defined for M680x0 and ColdFire processors. 22982The default settings depend on which architecture was selected when the 22983compiler was configured; the defaults for the most common choices are 22984given below. 22985 22986'-march=ARCH' 22987 Generate code for a specific M680x0 or ColdFire instruction set 22988 architecture. Permissible values of ARCH for M680x0 architectures 22989 are: '68000', '68010', '68020', '68030', '68040', '68060' and 22990 'cpu32'. ColdFire architectures are selected according to 22991 Freescale's ISA classification and the permissible values are: 22992 'isaa', 'isaaplus', 'isab' and 'isac'. 22993 22994 GCC defines a macro '__mcfARCH__' whenever it is generating code 22995 for a ColdFire target. The ARCH in this macro is one of the 22996 '-march' arguments given above. 22997 22998 When used together, '-march' and '-mtune' select code that runs on 22999 a family of similar processors but that is optimized for a 23000 particular microarchitecture. 23001 23002'-mcpu=CPU' 23003 Generate code for a specific M680x0 or ColdFire processor. The 23004 M680x0 CPUs are: '68000', '68010', '68020', '68030', '68040', 23005 '68060', '68302', '68332' and 'cpu32'. The ColdFire CPUs are given 23006 by the table below, which also classifies the CPUs into families: 23007 23008 *Family* *'-mcpu' arguments* 23009 -------------------------------------------------------------------------- 23010 '51' '51' '51ac' '51ag' '51cn' '51em' '51je' '51jf' '51jg' 23011 '51jm' '51mm' '51qe' '51qm' 23012 '5206' '5202' '5204' '5206' 23013 '5206e' '5206e' 23014 '5208' '5207' '5208' 23015 '5211a' '5210a' '5211a' 23016 '5213' '5211' '5212' '5213' 23017 '5216' '5214' '5216' 23018 '52235' '52230' '52231' '52232' '52233' '52234' '52235' 23019 '5225' '5224' '5225' 23020 '52259' '52252' '52254' '52255' '52256' '52258' '52259' 23021 '5235' '5232' '5233' '5234' '5235' '523x' 23022 '5249' '5249' 23023 '5250' '5250' 23024 '5271' '5270' '5271' 23025 '5272' '5272' 23026 '5275' '5274' '5275' 23027 '5282' '5280' '5281' '5282' '528x' 23028 '53017' '53011' '53012' '53013' '53014' '53015' '53016' '53017' 23029 '5307' '5307' 23030 '5329' '5327' '5328' '5329' '532x' 23031 '5373' '5372' '5373' '537x' 23032 '5407' '5407' 23033 '5475' '5470' '5471' '5472' '5473' '5474' '5475' '547x' '5480' 23034 '5481' '5482' '5483' '5484' '5485' 23035 23036 '-mcpu=CPU' overrides '-march=ARCH' if ARCH is compatible with CPU. 23037 Other combinations of '-mcpu' and '-march' are rejected. 23038 23039 GCC defines the macro '__mcf_cpu_CPU' when ColdFire target CPU is 23040 selected. It also defines '__mcf_family_FAMILY', where the value 23041 of FAMILY is given by the table above. 23042 23043'-mtune=TUNE' 23044 Tune the code for a particular microarchitecture within the 23045 constraints set by '-march' and '-mcpu'. The M680x0 23046 microarchitectures are: '68000', '68010', '68020', '68030', 23047 '68040', '68060' and 'cpu32'. The ColdFire microarchitectures are: 23048 'cfv1', 'cfv2', 'cfv3', 'cfv4' and 'cfv4e'. 23049 23050 You can also use '-mtune=68020-40' for code that needs to run 23051 relatively well on 68020, 68030 and 68040 targets. 23052 '-mtune=68020-60' is similar but includes 68060 targets as well. 23053 These two options select the same tuning decisions as '-m68020-40' 23054 and '-m68020-60' respectively. 23055 23056 GCC defines the macros '__mcARCH' and '__mcARCH__' when tuning for 23057 680x0 architecture ARCH. It also defines 'mcARCH' unless either 23058 '-ansi' or a non-GNU '-std' option is used. If GCC is tuning for a 23059 range of architectures, as selected by '-mtune=68020-40' or 23060 '-mtune=68020-60', it defines the macros for every architecture in 23061 the range. 23062 23063 GCC also defines the macro '__mUARCH__' when tuning for ColdFire 23064 microarchitecture UARCH, where UARCH is one of the arguments given 23065 above. 23066 23067'-m68000' 23068'-mc68000' 23069 Generate output for a 68000. This is the default when the compiler 23070 is configured for 68000-based systems. It is equivalent to 23071 '-march=68000'. 23072 23073 Use this option for microcontrollers with a 68000 or EC000 core, 23074 including the 68008, 68302, 68306, 68307, 68322, 68328 and 68356. 23075 23076'-m68010' 23077 Generate output for a 68010. This is the default when the compiler 23078 is configured for 68010-based systems. It is equivalent to 23079 '-march=68010'. 23080 23081'-m68020' 23082'-mc68020' 23083 Generate output for a 68020. This is the default when the compiler 23084 is configured for 68020-based systems. It is equivalent to 23085 '-march=68020'. 23086 23087'-m68030' 23088 Generate output for a 68030. This is the default when the compiler 23089 is configured for 68030-based systems. It is equivalent to 23090 '-march=68030'. 23091 23092'-m68040' 23093 Generate output for a 68040. This is the default when the compiler 23094 is configured for 68040-based systems. It is equivalent to 23095 '-march=68040'. 23096 23097 This option inhibits the use of 68881/68882 instructions that have 23098 to be emulated by software on the 68040. Use this option if your 23099 68040 does not have code to emulate those instructions. 23100 23101'-m68060' 23102 Generate output for a 68060. This is the default when the compiler 23103 is configured for 68060-based systems. It is equivalent to 23104 '-march=68060'. 23105 23106 This option inhibits the use of 68020 and 68881/68882 instructions 23107 that have to be emulated by software on the 68060. Use this option 23108 if your 68060 does not have code to emulate those instructions. 23109 23110'-mcpu32' 23111 Generate output for a CPU32. This is the default when the compiler 23112 is configured for CPU32-based systems. It is equivalent to 23113 '-march=cpu32'. 23114 23115 Use this option for microcontrollers with a CPU32 or CPU32+ core, 23116 including the 68330, 68331, 68332, 68333, 68334, 68336, 68340, 23117 68341, 68349 and 68360. 23118 23119'-m5200' 23120 Generate output for a 520X ColdFire CPU. This is the default when 23121 the compiler is configured for 520X-based systems. It is 23122 equivalent to '-mcpu=5206', and is now deprecated in favor of that 23123 option. 23124 23125 Use this option for microcontroller with a 5200 core, including the 23126 MCF5202, MCF5203, MCF5204 and MCF5206. 23127 23128'-m5206e' 23129 Generate output for a 5206e ColdFire CPU. The option is now 23130 deprecated in favor of the equivalent '-mcpu=5206e'. 23131 23132'-m528x' 23133 Generate output for a member of the ColdFire 528X family. The 23134 option is now deprecated in favor of the equivalent '-mcpu=528x'. 23135 23136'-m5307' 23137 Generate output for a ColdFire 5307 CPU. The option is now 23138 deprecated in favor of the equivalent '-mcpu=5307'. 23139 23140'-m5407' 23141 Generate output for a ColdFire 5407 CPU. The option is now 23142 deprecated in favor of the equivalent '-mcpu=5407'. 23143 23144'-mcfv4e' 23145 Generate output for a ColdFire V4e family CPU (e.g. 547x/548x). 23146 This includes use of hardware floating-point instructions. The 23147 option is equivalent to '-mcpu=547x', and is now deprecated in 23148 favor of that option. 23149 23150'-m68020-40' 23151 Generate output for a 68040, without using any of the new 23152 instructions. This results in code that can run relatively 23153 efficiently on either a 68020/68881 or a 68030 or a 68040. The 23154 generated code does use the 68881 instructions that are emulated on 23155 the 68040. 23156 23157 The option is equivalent to '-march=68020' '-mtune=68020-40'. 23158 23159'-m68020-60' 23160 Generate output for a 68060, without using any of the new 23161 instructions. This results in code that can run relatively 23162 efficiently on either a 68020/68881 or a 68030 or a 68040. The 23163 generated code does use the 68881 instructions that are emulated on 23164 the 68060. 23165 23166 The option is equivalent to '-march=68020' '-mtune=68020-60'. 23167 23168'-mhard-float' 23169'-m68881' 23170 Generate floating-point instructions. This is the default for 23171 68020 and above, and for ColdFire devices that have an FPU. It 23172 defines the macro '__HAVE_68881__' on M680x0 targets and 23173 '__mcffpu__' on ColdFire targets. 23174 23175'-msoft-float' 23176 Do not generate floating-point instructions; use library calls 23177 instead. This is the default for 68000, 68010, and 68832 targets. 23178 It is also the default for ColdFire devices that have no FPU. 23179 23180'-mdiv' 23181'-mno-div' 23182 Generate (do not generate) ColdFire hardware divide and remainder 23183 instructions. If '-march' is used without '-mcpu', the default is 23184 "on" for ColdFire architectures and "off" for M680x0 architectures. 23185 Otherwise, the default is taken from the target CPU (either the 23186 default CPU, or the one specified by '-mcpu'). For example, the 23187 default is "off" for '-mcpu=5206' and "on" for '-mcpu=5206e'. 23188 23189 GCC defines the macro '__mcfhwdiv__' when this option is enabled. 23190 23191'-mshort' 23192 Consider type 'int' to be 16 bits wide, like 'short int'. 23193 Additionally, parameters passed on the stack are also aligned to a 23194 16-bit boundary even on targets whose API mandates promotion to 23195 32-bit. 23196 23197'-mno-short' 23198 Do not consider type 'int' to be 16 bits wide. This is the 23199 default. 23200 23201'-mnobitfield' 23202'-mno-bitfield' 23203 Do not use the bit-field instructions. The '-m68000', '-mcpu32' 23204 and '-m5200' options imply '-mnobitfield'. 23205 23206'-mbitfield' 23207 Do use the bit-field instructions. The '-m68020' option implies 23208 '-mbitfield'. This is the default if you use a configuration 23209 designed for a 68020. 23210 23211'-mrtd' 23212 Use a different function-calling convention, in which functions 23213 that take a fixed number of arguments return with the 'rtd' 23214 instruction, which pops their arguments while returning. This 23215 saves one instruction in the caller since there is no need to pop 23216 the arguments there. 23217 23218 This calling convention is incompatible with the one normally used 23219 on Unix, so you cannot use it if you need to call libraries 23220 compiled with the Unix compiler. 23221 23222 Also, you must provide function prototypes for all functions that 23223 take variable numbers of arguments (including 'printf'); otherwise 23224 incorrect code is generated for calls to those functions. 23225 23226 In addition, seriously incorrect code results if you call a 23227 function with too many arguments. (Normally, extra arguments are 23228 harmlessly ignored.) 23229 23230 The 'rtd' instruction is supported by the 68010, 68020, 68030, 23231 68040, 68060 and CPU32 processors, but not by the 68000 or 5200. 23232 23233 The default is '-mno-rtd'. 23234 23235'-malign-int' 23236'-mno-align-int' 23237 Control whether GCC aligns 'int', 'long', 'long long', 'float', 23238 'double', and 'long double' variables on a 32-bit boundary 23239 ('-malign-int') or a 16-bit boundary ('-mno-align-int'). Aligning 23240 variables on 32-bit boundaries produces code that runs somewhat 23241 faster on processors with 32-bit busses at the expense of more 23242 memory. 23243 23244 *Warning:* if you use the '-malign-int' switch, GCC aligns 23245 structures containing the above types differently than most 23246 published application binary interface specifications for the m68k. 23247 23248 Use the pc-relative addressing mode of the 68000 directly, instead 23249 of using a global offset table. At present, this option implies 23250 '-fpic', allowing at most a 16-bit offset for pc-relative 23251 addressing. '-fPIC' is not presently supported with '-mpcrel', 23252 though this could be supported for 68020 and higher processors. 23253 23254'-mno-strict-align' 23255'-mstrict-align' 23256 Do not (do) assume that unaligned memory references are handled by 23257 the system. 23258 23259'-msep-data' 23260 Generate code that allows the data segment to be located in a 23261 different area of memory from the text segment. This allows for 23262 execute-in-place in an environment without virtual memory 23263 management. This option implies '-fPIC'. 23264 23265'-mno-sep-data' 23266 Generate code that assumes that the data segment follows the text 23267 segment. This is the default. 23268 23269'-mid-shared-library' 23270 Generate code that supports shared libraries via the library ID 23271 method. This allows for execute-in-place and shared libraries in 23272 an environment without virtual memory management. This option 23273 implies '-fPIC'. 23274 23275'-mno-id-shared-library' 23276 Generate code that doesn't assume ID-based shared libraries are 23277 being used. This is the default. 23278 23279'-mshared-library-id=n' 23280 Specifies the identification number of the ID-based shared library 23281 being compiled. Specifying a value of 0 generates more compact 23282 code; specifying other values forces the allocation of that number 23283 to the current library, but is no more space- or time-efficient 23284 than omitting this option. 23285 23286'-mxgot' 23287'-mno-xgot' 23288 When generating position-independent code for ColdFire, generate 23289 code that works if the GOT has more than 8192 entries. This code 23290 is larger and slower than code generated without this option. On 23291 M680x0 processors, this option is not needed; '-fPIC' suffices. 23292 23293 GCC normally uses a single instruction to load values from the GOT. 23294 While this is relatively efficient, it only works if the GOT is 23295 smaller than about 64k. Anything larger causes the linker to 23296 report an error such as: 23297 23298 relocation truncated to fit: R_68K_GOT16O foobar 23299 23300 If this happens, you should recompile your code with '-mxgot'. It 23301 should then work with very large GOTs. However, code generated 23302 with '-mxgot' is less efficient, since it takes 4 instructions to 23303 fetch the value of a global symbol. 23304 23305 Note that some linkers, including newer versions of the GNU linker, 23306 can create multiple GOTs and sort GOT entries. If you have such a 23307 linker, you should only need to use '-mxgot' when compiling a 23308 single object file that accesses more than 8192 GOT entries. Very 23309 few do. 23310 23311 These options have no effect unless GCC is generating 23312 position-independent code. 23313 23314'-mlong-jump-table-offsets' 23315 Use 32-bit offsets in 'switch' tables. The default is to use 23316 16-bit offsets. 23317 23318 23319File: gcc.info, Node: MCore Options, Next: MeP Options, Prev: M680x0 Options, Up: Submodel Options 23320 233213.19.27 MCore Options 23322--------------------- 23323 23324These are the '-m' options defined for the Motorola M*Core processors. 23325 23326'-mhardlit' 23327'-mno-hardlit' 23328 Inline constants into the code stream if it can be done in two 23329 instructions or less. 23330 23331'-mdiv' 23332'-mno-div' 23333 Use the divide instruction. (Enabled by default). 23334 23335'-mrelax-immediate' 23336'-mno-relax-immediate' 23337 Allow arbitrary-sized immediates in bit operations. 23338 23339'-mwide-bitfields' 23340'-mno-wide-bitfields' 23341 Always treat bit-fields as 'int'-sized. 23342 23343'-m4byte-functions' 23344'-mno-4byte-functions' 23345 Force all functions to be aligned to a 4-byte boundary. 23346 23347'-mcallgraph-data' 23348'-mno-callgraph-data' 23349 Emit callgraph information. 23350 23351'-mslow-bytes' 23352'-mno-slow-bytes' 23353 Prefer word access when reading byte quantities. 23354 23355'-mlittle-endian' 23356'-mbig-endian' 23357 Generate code for a little-endian target. 23358 23359'-m210' 23360'-m340' 23361 Generate code for the 210 processor. 23362 23363'-mno-lsim' 23364 Assume that runtime support has been provided and so omit the 23365 simulator library ('libsim.a)' from the linker command line. 23366 23367'-mstack-increment=SIZE' 23368 Set the maximum amount for a single stack increment operation. 23369 Large values can increase the speed of programs that contain 23370 functions that need a large amount of stack space, but they can 23371 also trigger a segmentation fault if the stack is extended too 23372 much. The default value is 0x1000. 23373 23374 23375File: gcc.info, Node: MeP Options, Next: MicroBlaze Options, Prev: MCore Options, Up: Submodel Options 23376 233773.19.28 MeP Options 23378------------------- 23379 23380'-mabsdiff' 23381 Enables the 'abs' instruction, which is the absolute difference 23382 between two registers. 23383 23384'-mall-opts' 23385 Enables all the optional instructions--average, multiply, divide, 23386 bit operations, leading zero, absolute difference, min/max, clip, 23387 and saturation. 23388 23389'-maverage' 23390 Enables the 'ave' instruction, which computes the average of two 23391 registers. 23392 23393'-mbased=N' 23394 Variables of size N bytes or smaller are placed in the '.based' 23395 section by default. Based variables use the '$tp' register as a 23396 base register, and there is a 128-byte limit to the '.based' 23397 section. 23398 23399'-mbitops' 23400 Enables the bit operation instructions--bit test ('btstm'), set 23401 ('bsetm'), clear ('bclrm'), invert ('bnotm'), and test-and-set 23402 ('tas'). 23403 23404'-mc=NAME' 23405 Selects which section constant data is placed in. NAME may be 23406 'tiny', 'near', or 'far'. 23407 23408'-mclip' 23409 Enables the 'clip' instruction. Note that '-mclip' is not useful 23410 unless you also provide '-mminmax'. 23411 23412'-mconfig=NAME' 23413 Selects one of the built-in core configurations. Each MeP chip has 23414 one or more modules in it; each module has a core CPU and a variety 23415 of coprocessors, optional instructions, and peripherals. The 23416 'MeP-Integrator' tool, not part of GCC, provides these 23417 configurations through this option; using this option is the same 23418 as using all the corresponding command-line options. The default 23419 configuration is 'default'. 23420 23421'-mcop' 23422 Enables the coprocessor instructions. By default, this is a 32-bit 23423 coprocessor. Note that the coprocessor is normally enabled via the 23424 '-mconfig=' option. 23425 23426'-mcop32' 23427 Enables the 32-bit coprocessor's instructions. 23428 23429'-mcop64' 23430 Enables the 64-bit coprocessor's instructions. 23431 23432'-mivc2' 23433 Enables IVC2 scheduling. IVC2 is a 64-bit VLIW coprocessor. 23434 23435'-mdc' 23436 Causes constant variables to be placed in the '.near' section. 23437 23438'-mdiv' 23439 Enables the 'div' and 'divu' instructions. 23440 23441'-meb' 23442 Generate big-endian code. 23443 23444'-mel' 23445 Generate little-endian code. 23446 23447'-mio-volatile' 23448 Tells the compiler that any variable marked with the 'io' attribute 23449 is to be considered volatile. 23450 23451'-ml' 23452 Causes variables to be assigned to the '.far' section by default. 23453 23454'-mleadz' 23455 Enables the 'leadz' (leading zero) instruction. 23456 23457'-mm' 23458 Causes variables to be assigned to the '.near' section by default. 23459 23460'-mminmax' 23461 Enables the 'min' and 'max' instructions. 23462 23463'-mmult' 23464 Enables the multiplication and multiply-accumulate instructions. 23465 23466'-mno-opts' 23467 Disables all the optional instructions enabled by '-mall-opts'. 23468 23469'-mrepeat' 23470 Enables the 'repeat' and 'erepeat' instructions, used for 23471 low-overhead looping. 23472 23473'-ms' 23474 Causes all variables to default to the '.tiny' section. Note that 23475 there is a 65536-byte limit to this section. Accesses to these 23476 variables use the '%gp' base register. 23477 23478'-msatur' 23479 Enables the saturation instructions. Note that the compiler does 23480 not currently generate these itself, but this option is included 23481 for compatibility with other tools, like 'as'. 23482 23483'-msdram' 23484 Link the SDRAM-based runtime instead of the default ROM-based 23485 runtime. 23486 23487'-msim' 23488 Link the simulator run-time libraries. 23489 23490'-msimnovec' 23491 Link the simulator runtime libraries, excluding built-in support 23492 for reset and exception vectors and tables. 23493 23494'-mtf' 23495 Causes all functions to default to the '.far' section. Without 23496 this option, functions default to the '.near' section. 23497 23498'-mtiny=N' 23499 Variables that are N bytes or smaller are allocated to the '.tiny' 23500 section. These variables use the '$gp' base register. The default 23501 for this option is 4, but note that there's a 65536-byte limit to 23502 the '.tiny' section. 23503 23504 23505File: gcc.info, Node: MicroBlaze Options, Next: MIPS Options, Prev: MeP Options, Up: Submodel Options 23506 235073.19.29 MicroBlaze Options 23508-------------------------- 23509 23510'-msoft-float' 23511 Use software emulation for floating point (default). 23512 23513'-mhard-float' 23514 Use hardware floating-point instructions. 23515 23516'-mmemcpy' 23517 Do not optimize block moves, use 'memcpy'. 23518 23519'-mno-clearbss' 23520 This option is deprecated. Use '-fno-zero-initialized-in-bss' 23521 instead. 23522 23523'-mcpu=CPU-TYPE' 23524 Use features of, and schedule code for, the given CPU. Supported 23525 values are in the format 'vX.YY.Z', where X is a major version, YY 23526 is the minor version, and Z is compatibility code. Example values 23527 are 'v3.00.a', 'v4.00.b', 'v5.00.a', 'v5.00.b', 'v6.00.a'. 23528 23529'-mxl-soft-mul' 23530 Use software multiply emulation (default). 23531 23532'-mxl-soft-div' 23533 Use software emulation for divides (default). 23534 23535'-mxl-barrel-shift' 23536 Use the hardware barrel shifter. 23537 23538'-mxl-pattern-compare' 23539 Use pattern compare instructions. 23540 23541'-msmall-divides' 23542 Use table lookup optimization for small signed integer divisions. 23543 23544'-mxl-stack-check' 23545 This option is deprecated. Use '-fstack-check' instead. 23546 23547'-mxl-gp-opt' 23548 Use GP-relative '.sdata'/'.sbss' sections. 23549 23550'-mxl-multiply-high' 23551 Use multiply high instructions for high part of 32x32 multiply. 23552 23553'-mxl-float-convert' 23554 Use hardware floating-point conversion instructions. 23555 23556'-mxl-float-sqrt' 23557 Use hardware floating-point square root instruction. 23558 23559'-mbig-endian' 23560 Generate code for a big-endian target. 23561 23562'-mlittle-endian' 23563 Generate code for a little-endian target. 23564 23565'-mxl-reorder' 23566 Use reorder instructions (swap and byte reversed load/store). 23567 23568'-mxl-mode-APP-MODEL' 23569 Select application model APP-MODEL. Valid models are 23570 'executable' 23571 normal executable (default), uses startup code 'crt0.o'. 23572 23573 'xmdstub' 23574 for use with Xilinx Microprocessor Debugger (XMD) based 23575 software intrusive debug agent called xmdstub. This uses 23576 startup file 'crt1.o' and sets the start address of the 23577 program to 0x800. 23578 23579 'bootstrap' 23580 for applications that are loaded using a bootloader. This 23581 model uses startup file 'crt2.o' which does not contain a 23582 processor reset vector handler. This is suitable for 23583 transferring control on a processor reset to the bootloader 23584 rather than the application. 23585 23586 'novectors' 23587 for applications that do not require any of the MicroBlaze 23588 vectors. This option may be useful for applications running 23589 within a monitoring application. This model uses 'crt3.o' as 23590 a startup file. 23591 23592 Option '-xl-mode-APP-MODEL' is a deprecated alias for 23593 '-mxl-mode-APP-MODEL'. 23594 23595'-mpic-data-is-text-relative' 23596 Assume that the displacement between the text and data segments is 23597 fixed at static link time. This allows data to be referenced by 23598 offset from start of text address instead of GOT since PC-relative 23599 addressing is not supported. 23600 23601 23602File: gcc.info, Node: MIPS Options, Next: MMIX Options, Prev: MicroBlaze Options, Up: Submodel Options 23603 236043.19.30 MIPS Options 23605-------------------- 23606 23607'-EB' 23608 Generate big-endian code. 23609 23610'-EL' 23611 Generate little-endian code. This is the default for 'mips*el-*-*' 23612 configurations. 23613 23614'-march=ARCH' 23615 Generate code that runs on ARCH, which can be the name of a generic 23616 MIPS ISA, or the name of a particular processor. The ISA names 23617 are: 'mips1', 'mips2', 'mips3', 'mips4', 'mips32', 'mips32r2', 23618 'mips32r3', 'mips32r5', 'mips32r6', 'mips64', 'mips64r2', 23619 'mips64r3', 'mips64r5' and 'mips64r6'. The processor names are: 23620 '4kc', '4km', '4kp', '4ksc', '4kec', '4kem', '4kep', '4ksd', '5kc', 23621 '5kf', '20kc', '24kc', '24kf2_1', '24kf1_1', '24kec', '24kef2_1', 23622 '24kef1_1', '34kc', '34kf2_1', '34kf1_1', '34kn', '74kc', 23623 '74kf2_1', '74kf1_1', '74kf3_2', '1004kc', '1004kf2_1', 23624 '1004kf1_1', 'i6400', 'i6500', 'interaptiv', 'loongson2e', 23625 'loongson2f', 'loongson3a', 'gs464', 'gs464e', 'gs264e', 'm4k', 23626 'm14k', 'm14kc', 'm14ke', 'm14kec', 'm5100', 'm5101', 'octeon', 23627 'octeon+', 'octeon2', 'octeon3', 'orion', 'p5600', 'p6600', 23628 'r2000', 'r3000', 'r3900', 'r4000', 'r4400', 'r4600', 'r4650', 23629 'r4700', 'r5900', 'r6000', 'r8000', 'rm7000', 'rm9000', 'r10000', 23630 'r12000', 'r14000', 'r16000', 'sb1', 'sr71000', 'vr4100', 'vr4111', 23631 'vr4120', 'vr4130', 'vr4300', 'vr5000', 'vr5400', 'vr5500', 'xlr' 23632 and 'xlp'. The special value 'from-abi' selects the most 23633 compatible architecture for the selected ABI (that is, 'mips1' for 23634 32-bit ABIs and 'mips3' for 64-bit ABIs). 23635 23636 The native Linux/GNU toolchain also supports the value 'native', 23637 which selects the best architecture option for the host processor. 23638 '-march=native' has no effect if GCC does not recognize the 23639 processor. 23640 23641 In processor names, a final '000' can be abbreviated as 'k' (for 23642 example, '-march=r2k'). Prefixes are optional, and 'vr' may be 23643 written 'r'. 23644 23645 Names of the form 'Nf2_1' refer to processors with FPUs clocked at 23646 half the rate of the core, names of the form 'Nf1_1' refer to 23647 processors with FPUs clocked at the same rate as the core, and 23648 names of the form 'Nf3_2' refer to processors with FPUs clocked a 23649 ratio of 3:2 with respect to the core. For compatibility reasons, 23650 'Nf' is accepted as a synonym for 'Nf2_1' while 'Nx' and 'Bfx' are 23651 accepted as synonyms for 'Nf1_1'. 23652 23653 GCC defines two macros based on the value of this option. The 23654 first is '_MIPS_ARCH', which gives the name of target architecture, 23655 as a string. The second has the form '_MIPS_ARCH_FOO', where FOO 23656 is the capitalized value of '_MIPS_ARCH'. For example, 23657 '-march=r2000' sets '_MIPS_ARCH' to '"r2000"' and defines the macro 23658 '_MIPS_ARCH_R2000'. 23659 23660 Note that the '_MIPS_ARCH' macro uses the processor names given 23661 above. In other words, it has the full prefix and does not 23662 abbreviate '000' as 'k'. In the case of 'from-abi', the macro 23663 names the resolved architecture (either '"mips1"' or '"mips3"'). 23664 It names the default architecture when no '-march' option is given. 23665 23666'-mtune=ARCH' 23667 Optimize for ARCH. Among other things, this option controls the 23668 way instructions are scheduled, and the perceived cost of 23669 arithmetic operations. The list of ARCH values is the same as for 23670 '-march'. 23671 23672 When this option is not used, GCC optimizes for the processor 23673 specified by '-march'. By using '-march' and '-mtune' together, it 23674 is possible to generate code that runs on a family of processors, 23675 but optimize the code for one particular member of that family. 23676 23677 '-mtune' defines the macros '_MIPS_TUNE' and '_MIPS_TUNE_FOO', 23678 which work in the same way as the '-march' ones described above. 23679 23680'-mips1' 23681 Equivalent to '-march=mips1'. 23682 23683'-mips2' 23684 Equivalent to '-march=mips2'. 23685 23686'-mips3' 23687 Equivalent to '-march=mips3'. 23688 23689'-mips4' 23690 Equivalent to '-march=mips4'. 23691 23692'-mips32' 23693 Equivalent to '-march=mips32'. 23694 23695'-mips32r3' 23696 Equivalent to '-march=mips32r3'. 23697 23698'-mips32r5' 23699 Equivalent to '-march=mips32r5'. 23700 23701'-mips32r6' 23702 Equivalent to '-march=mips32r6'. 23703 23704'-mips64' 23705 Equivalent to '-march=mips64'. 23706 23707'-mips64r2' 23708 Equivalent to '-march=mips64r2'. 23709 23710'-mips64r3' 23711 Equivalent to '-march=mips64r3'. 23712 23713'-mips64r5' 23714 Equivalent to '-march=mips64r5'. 23715 23716'-mips64r6' 23717 Equivalent to '-march=mips64r6'. 23718 23719'-mips16' 23720'-mno-mips16' 23721 Generate (do not generate) MIPS16 code. If GCC is targeting a 23722 MIPS32 or MIPS64 architecture, it makes use of the MIPS16e ASE. 23723 23724 MIPS16 code generation can also be controlled on a per-function 23725 basis by means of 'mips16' and 'nomips16' attributes. *Note 23726 Function Attributes::, for more information. 23727 23728'-mflip-mips16' 23729 Generate MIPS16 code on alternating functions. This option is 23730 provided for regression testing of mixed MIPS16/non-MIPS16 code 23731 generation, and is not intended for ordinary use in compiling user 23732 code. 23733 23734'-minterlink-compressed' 23735'-mno-interlink-compressed' 23736 Require (do not require) that code using the standard 23737 (uncompressed) MIPS ISA be link-compatible with MIPS16 and 23738 microMIPS code, and vice versa. 23739 23740 For example, code using the standard ISA encoding cannot jump 23741 directly to MIPS16 or microMIPS code; it must either use a call or 23742 an indirect jump. '-minterlink-compressed' therefore disables 23743 direct jumps unless GCC knows that the target of the jump is not 23744 compressed. 23745 23746'-minterlink-mips16' 23747'-mno-interlink-mips16' 23748 Aliases of '-minterlink-compressed' and 23749 '-mno-interlink-compressed'. These options predate the microMIPS 23750 ASE and are retained for backwards compatibility. 23751 23752'-mabi=32' 23753'-mabi=o64' 23754'-mabi=n32' 23755'-mabi=64' 23756'-mabi=eabi' 23757 Generate code for the given ABI. 23758 23759 Note that the EABI has a 32-bit and a 64-bit variant. GCC normally 23760 generates 64-bit code when you select a 64-bit architecture, but 23761 you can use '-mgp32' to get 32-bit code instead. 23762 23763 For information about the O64 ABI, see 23764 <https://gcc.gnu.org/projects/mipso64-abi.html>. 23765 23766 GCC supports a variant of the o32 ABI in which floating-point 23767 registers are 64 rather than 32 bits wide. You can select this 23768 combination with '-mabi=32' '-mfp64'. This ABI relies on the 23769 'mthc1' and 'mfhc1' instructions and is therefore only supported 23770 for MIPS32R2, MIPS32R3 and MIPS32R5 processors. 23771 23772 The register assignments for arguments and return values remain the 23773 same, but each scalar value is passed in a single 64-bit register 23774 rather than a pair of 32-bit registers. For example, scalar 23775 floating-point values are returned in '$f0' only, not a '$f0'/'$f1' 23776 pair. The set of call-saved registers also remains the same in 23777 that the even-numbered double-precision registers are saved. 23778 23779 Two additional variants of the o32 ABI are supported to enable a 23780 transition from 32-bit to 64-bit registers. These are FPXX 23781 ('-mfpxx') and FP64A ('-mfp64' '-mno-odd-spreg'). The FPXX 23782 extension mandates that all code must execute correctly when run 23783 using 32-bit or 64-bit registers. The code can be interlinked with 23784 either FP32 or FP64, but not both. The FP64A extension is similar 23785 to the FP64 extension but forbids the use of odd-numbered 23786 single-precision registers. This can be used in conjunction with 23787 the 'FRE' mode of FPUs in MIPS32R5 processors and allows both FP32 23788 and FP64A code to interlink and run in the same process without 23789 changing FPU modes. 23790 23791'-mabicalls' 23792'-mno-abicalls' 23793 Generate (do not generate) code that is suitable for SVR4-style 23794 dynamic objects. '-mabicalls' is the default for SVR4-based 23795 systems. 23796 23797'-mshared' 23798'-mno-shared' 23799 Generate (do not generate) code that is fully position-independent, 23800 and that can therefore be linked into shared libraries. This 23801 option only affects '-mabicalls'. 23802 23803 All '-mabicalls' code has traditionally been position-independent, 23804 regardless of options like '-fPIC' and '-fpic'. However, as an 23805 extension, the GNU toolchain allows executables to use absolute 23806 accesses for locally-binding symbols. It can also use shorter GP 23807 initialization sequences and generate direct calls to 23808 locally-defined functions. This mode is selected by '-mno-shared'. 23809 23810 '-mno-shared' depends on binutils 2.16 or higher and generates 23811 objects that can only be linked by the GNU linker. However, the 23812 option does not affect the ABI of the final executable; it only 23813 affects the ABI of relocatable objects. Using '-mno-shared' 23814 generally makes executables both smaller and quicker. 23815 23816 '-mshared' is the default. 23817 23818'-mplt' 23819'-mno-plt' 23820 Assume (do not assume) that the static and dynamic linkers support 23821 PLTs and copy relocations. This option only affects '-mno-shared 23822 -mabicalls'. For the n64 ABI, this option has no effect without 23823 '-msym32'. 23824 23825 You can make '-mplt' the default by configuring GCC with 23826 '--with-mips-plt'. The default is '-mno-plt' otherwise. 23827 23828'-mxgot' 23829'-mno-xgot' 23830 Lift (do not lift) the usual restrictions on the size of the global 23831 offset table. 23832 23833 GCC normally uses a single instruction to load values from the GOT. 23834 While this is relatively efficient, it only works if the GOT is 23835 smaller than about 64k. Anything larger causes the linker to 23836 report an error such as: 23837 23838 relocation truncated to fit: R_MIPS_GOT16 foobar 23839 23840 If this happens, you should recompile your code with '-mxgot'. 23841 This works with very large GOTs, although the code is also less 23842 efficient, since it takes three instructions to fetch the value of 23843 a global symbol. 23844 23845 Note that some linkers can create multiple GOTs. If you have such 23846 a linker, you should only need to use '-mxgot' when a single object 23847 file accesses more than 64k's worth of GOT entries. Very few do. 23848 23849 These options have no effect unless GCC is generating position 23850 independent code. 23851 23852'-mgp32' 23853 Assume that general-purpose registers are 32 bits wide. 23854 23855'-mgp64' 23856 Assume that general-purpose registers are 64 bits wide. 23857 23858'-mfp32' 23859 Assume that floating-point registers are 32 bits wide. 23860 23861'-mfp64' 23862 Assume that floating-point registers are 64 bits wide. 23863 23864'-mfpxx' 23865 Do not assume the width of floating-point registers. 23866 23867'-mhard-float' 23868 Use floating-point coprocessor instructions. 23869 23870'-msoft-float' 23871 Do not use floating-point coprocessor instructions. Implement 23872 floating-point calculations using library calls instead. 23873 23874'-mno-float' 23875 Equivalent to '-msoft-float', but additionally asserts that the 23876 program being compiled does not perform any floating-point 23877 operations. This option is presently supported only by some 23878 bare-metal MIPS configurations, where it may select a special set 23879 of libraries that lack all floating-point support (including, for 23880 example, the floating-point 'printf' formats). If code compiled 23881 with '-mno-float' accidentally contains floating-point operations, 23882 it is likely to suffer a link-time or run-time failure. 23883 23884'-msingle-float' 23885 Assume that the floating-point coprocessor only supports 23886 single-precision operations. 23887 23888'-mdouble-float' 23889 Assume that the floating-point coprocessor supports 23890 double-precision operations. This is the default. 23891 23892'-modd-spreg' 23893'-mno-odd-spreg' 23894 Enable the use of odd-numbered single-precision floating-point 23895 registers for the o32 ABI. This is the default for processors that 23896 are known to support these registers. When using the o32 FPXX ABI, 23897 '-mno-odd-spreg' is set by default. 23898 23899'-mabs=2008' 23900'-mabs=legacy' 23901 These options control the treatment of the special not-a-number 23902 (NaN) IEEE 754 floating-point data with the 'abs.fmt' and 'neg.fmt' 23903 machine instructions. 23904 23905 By default or when '-mabs=legacy' is used the legacy treatment is 23906 selected. In this case these instructions are considered 23907 arithmetic and avoided where correct operation is required and the 23908 input operand might be a NaN. A longer sequence of instructions 23909 that manipulate the sign bit of floating-point datum manually is 23910 used instead unless the '-ffinite-math-only' option has also been 23911 specified. 23912 23913 The '-mabs=2008' option selects the IEEE 754-2008 treatment. In 23914 this case these instructions are considered non-arithmetic and 23915 therefore operating correctly in all cases, including in particular 23916 where the input operand is a NaN. These instructions are therefore 23917 always used for the respective operations. 23918 23919'-mnan=2008' 23920'-mnan=legacy' 23921 These options control the encoding of the special not-a-number 23922 (NaN) IEEE 754 floating-point data. 23923 23924 The '-mnan=legacy' option selects the legacy encoding. In this 23925 case quiet NaNs (qNaNs) are denoted by the first bit of their 23926 trailing significand field being 0, whereas signaling NaNs (sNaNs) 23927 are denoted by the first bit of their trailing significand field 23928 being 1. 23929 23930 The '-mnan=2008' option selects the IEEE 754-2008 encoding. In 23931 this case qNaNs are denoted by the first bit of their trailing 23932 significand field being 1, whereas sNaNs are denoted by the first 23933 bit of their trailing significand field being 0. 23934 23935 The default is '-mnan=legacy' unless GCC has been configured with 23936 '--with-nan=2008'. 23937 23938'-mllsc' 23939'-mno-llsc' 23940 Use (do not use) 'll', 'sc', and 'sync' instructions to implement 23941 atomic memory built-in functions. When neither option is 23942 specified, GCC uses the instructions if the target architecture 23943 supports them. 23944 23945 '-mllsc' is useful if the runtime environment can emulate the 23946 instructions and '-mno-llsc' can be useful when compiling for 23947 nonstandard ISAs. You can make either option the default by 23948 configuring GCC with '--with-llsc' and '--without-llsc' 23949 respectively. '--with-llsc' is the default for some 23950 configurations; see the installation documentation for details. 23951 23952'-mdsp' 23953'-mno-dsp' 23954 Use (do not use) revision 1 of the MIPS DSP ASE. *Note MIPS DSP 23955 Built-in Functions::. This option defines the preprocessor macro 23956 '__mips_dsp'. It also defines '__mips_dsp_rev' to 1. 23957 23958'-mdspr2' 23959'-mno-dspr2' 23960 Use (do not use) revision 2 of the MIPS DSP ASE. *Note MIPS DSP 23961 Built-in Functions::. This option defines the preprocessor macros 23962 '__mips_dsp' and '__mips_dspr2'. It also defines '__mips_dsp_rev' 23963 to 2. 23964 23965'-msmartmips' 23966'-mno-smartmips' 23967 Use (do not use) the MIPS SmartMIPS ASE. 23968 23969'-mpaired-single' 23970'-mno-paired-single' 23971 Use (do not use) paired-single floating-point instructions. *Note 23972 MIPS Paired-Single Support::. This option requires hardware 23973 floating-point support to be enabled. 23974 23975'-mdmx' 23976'-mno-mdmx' 23977 Use (do not use) MIPS Digital Media Extension instructions. This 23978 option can only be used when generating 64-bit code and requires 23979 hardware floating-point support to be enabled. 23980 23981'-mips3d' 23982'-mno-mips3d' 23983 Use (do not use) the MIPS-3D ASE. *Note MIPS-3D Built-in 23984 Functions::. The option '-mips3d' implies '-mpaired-single'. 23985 23986'-mmicromips' 23987'-mno-micromips' 23988 Generate (do not generate) microMIPS code. 23989 23990 MicroMIPS code generation can also be controlled on a per-function 23991 basis by means of 'micromips' and 'nomicromips' attributes. *Note 23992 Function Attributes::, for more information. 23993 23994'-mmt' 23995'-mno-mt' 23996 Use (do not use) MT Multithreading instructions. 23997 23998'-mmcu' 23999'-mno-mcu' 24000 Use (do not use) the MIPS MCU ASE instructions. 24001 24002'-meva' 24003'-mno-eva' 24004 Use (do not use) the MIPS Enhanced Virtual Addressing instructions. 24005 24006'-mvirt' 24007'-mno-virt' 24008 Use (do not use) the MIPS Virtualization (VZ) instructions. 24009 24010'-mxpa' 24011'-mno-xpa' 24012 Use (do not use) the MIPS eXtended Physical Address (XPA) 24013 instructions. 24014 24015'-mcrc' 24016'-mno-crc' 24017 Use (do not use) the MIPS Cyclic Redundancy Check (CRC) 24018 instructions. 24019 24020'-mginv' 24021'-mno-ginv' 24022 Use (do not use) the MIPS Global INValidate (GINV) instructions. 24023 24024'-mloongson-mmi' 24025'-mno-loongson-mmi' 24026 Use (do not use) the MIPS Loongson MultiMedia extensions 24027 Instructions (MMI). 24028 24029'-mloongson-ext' 24030'-mno-loongson-ext' 24031 Use (do not use) the MIPS Loongson EXTensions (EXT) instructions. 24032 24033'-mloongson-ext2' 24034'-mno-loongson-ext2' 24035 Use (do not use) the MIPS Loongson EXTensions r2 (EXT2) 24036 instructions. 24037 24038'-mlong64' 24039 Force 'long' types to be 64 bits wide. See '-mlong32' for an 24040 explanation of the default and the way that the pointer size is 24041 determined. 24042 24043'-mlong32' 24044 Force 'long', 'int', and pointer types to be 32 bits wide. 24045 24046 The default size of 'int's, 'long's and pointers depends on the 24047 ABI. All the supported ABIs use 32-bit 'int's. The n64 ABI uses 24048 64-bit 'long's, as does the 64-bit EABI; the others use 32-bit 24049 'long's. Pointers are the same size as 'long's, or the same size 24050 as integer registers, whichever is smaller. 24051 24052'-msym32' 24053'-mno-sym32' 24054 Assume (do not assume) that all symbols have 32-bit values, 24055 regardless of the selected ABI. This option is useful in 24056 combination with '-mabi=64' and '-mno-abicalls' because it allows 24057 GCC to generate shorter and faster references to symbolic 24058 addresses. 24059 24060'-G NUM' 24061 Put definitions of externally-visible data in a small data section 24062 if that data is no bigger than NUM bytes. GCC can then generate 24063 more efficient accesses to the data; see '-mgpopt' for details. 24064 24065 The default '-G' option depends on the configuration. 24066 24067'-mlocal-sdata' 24068'-mno-local-sdata' 24069 Extend (do not extend) the '-G' behavior to local data too, such as 24070 to static variables in C. '-mlocal-sdata' is the default for all 24071 configurations. 24072 24073 If the linker complains that an application is using too much small 24074 data, you might want to try rebuilding the less 24075 performance-critical parts with '-mno-local-sdata'. You might also 24076 want to build large libraries with '-mno-local-sdata', so that the 24077 libraries leave more room for the main program. 24078 24079'-mextern-sdata' 24080'-mno-extern-sdata' 24081 Assume (do not assume) that externally-defined data is in a small 24082 data section if the size of that data is within the '-G' limit. 24083 '-mextern-sdata' is the default for all configurations. 24084 24085 If you compile a module MOD with '-mextern-sdata' '-G NUM' 24086 '-mgpopt', and MOD references a variable VAR that is no bigger than 24087 NUM bytes, you must make sure that VAR is placed in a small data 24088 section. If VAR is defined by another module, you must either 24089 compile that module with a high-enough '-G' setting or attach a 24090 'section' attribute to VAR's definition. If VAR is common, you 24091 must link the application with a high-enough '-G' setting. 24092 24093 The easiest way of satisfying these restrictions is to compile and 24094 link every module with the same '-G' option. However, you may wish 24095 to build a library that supports several different small data 24096 limits. You can do this by compiling the library with the highest 24097 supported '-G' setting and additionally using '-mno-extern-sdata' 24098 to stop the library from making assumptions about 24099 externally-defined data. 24100 24101'-mgpopt' 24102'-mno-gpopt' 24103 Use (do not use) GP-relative accesses for symbols that are known to 24104 be in a small data section; see '-G', '-mlocal-sdata' and 24105 '-mextern-sdata'. '-mgpopt' is the default for all configurations. 24106 24107 '-mno-gpopt' is useful for cases where the '$gp' register might not 24108 hold the value of '_gp'. For example, if the code is part of a 24109 library that might be used in a boot monitor, programs that call 24110 boot monitor routines pass an unknown value in '$gp'. (In such 24111 situations, the boot monitor itself is usually compiled with 24112 '-G0'.) 24113 24114 '-mno-gpopt' implies '-mno-local-sdata' and '-mno-extern-sdata'. 24115 24116'-membedded-data' 24117'-mno-embedded-data' 24118 Allocate variables to the read-only data section first if possible, 24119 then next in the small data section if possible, otherwise in data. 24120 This gives slightly slower code than the default, but reduces the 24121 amount of RAM required when executing, and thus may be preferred 24122 for some embedded systems. 24123 24124'-muninit-const-in-rodata' 24125'-mno-uninit-const-in-rodata' 24126 Put uninitialized 'const' variables in the read-only data section. 24127 This option is only meaningful in conjunction with 24128 '-membedded-data'. 24129 24130'-mcode-readable=SETTING' 24131 Specify whether GCC may generate code that reads from executable 24132 sections. There are three possible settings: 24133 24134 '-mcode-readable=yes' 24135 Instructions may freely access executable sections. This is 24136 the default setting. 24137 24138 '-mcode-readable=pcrel' 24139 MIPS16 PC-relative load instructions can access executable 24140 sections, but other instructions must not do so. This option 24141 is useful on 4KSc and 4KSd processors when the code TLBs have 24142 the Read Inhibit bit set. It is also useful on processors 24143 that can be configured to have a dual instruction/data SRAM 24144 interface and that, like the M4K, automatically redirect 24145 PC-relative loads to the instruction RAM. 24146 24147 '-mcode-readable=no' 24148 Instructions must not access executable sections. This option 24149 can be useful on targets that are configured to have a dual 24150 instruction/data SRAM interface but that (unlike the M4K) do 24151 not automatically redirect PC-relative loads to the 24152 instruction RAM. 24153 24154'-msplit-addresses' 24155'-mno-split-addresses' 24156 Enable (disable) use of the '%hi()' and '%lo()' assembler 24157 relocation operators. This option has been superseded by 24158 '-mexplicit-relocs' but is retained for backwards compatibility. 24159 24160'-mexplicit-relocs' 24161'-mno-explicit-relocs' 24162 Use (do not use) assembler relocation operators when dealing with 24163 symbolic addresses. The alternative, selected by 24164 '-mno-explicit-relocs', is to use assembler macros instead. 24165 24166 '-mexplicit-relocs' is the default if GCC was configured to use an 24167 assembler that supports relocation operators. 24168 24169'-mcheck-zero-division' 24170'-mno-check-zero-division' 24171 Trap (do not trap) on integer division by zero. 24172 24173 The default is '-mcheck-zero-division'. 24174 24175'-mdivide-traps' 24176'-mdivide-breaks' 24177 MIPS systems check for division by zero by generating either a 24178 conditional trap or a break instruction. Using traps results in 24179 smaller code, but is only supported on MIPS II and later. Also, 24180 some versions of the Linux kernel have a bug that prevents trap 24181 from generating the proper signal ('SIGFPE'). Use '-mdivide-traps' 24182 to allow conditional traps on architectures that support them and 24183 '-mdivide-breaks' to force the use of breaks. 24184 24185 The default is usually '-mdivide-traps', but this can be overridden 24186 at configure time using '--with-divide=breaks'. Divide-by-zero 24187 checks can be completely disabled using '-mno-check-zero-division'. 24188 24189'-mload-store-pairs' 24190'-mno-load-store-pairs' 24191 Enable (disable) an optimization that pairs consecutive load or 24192 store instructions to enable load/store bonding. This option is 24193 enabled by default but only takes effect when the selected 24194 architecture is known to support bonding. 24195 24196'-munaligned-access' 24197'-mno-unaligned-access' 24198 Enable (disable) direct unaligned access for MIPS Release 6. 24199 MIPSr6 requires load/store unaligned-access support, by hardware or 24200 trap&emulate. So '-mno-unaligned-access' may be needed by kernel. 24201 24202'-mmemcpy' 24203'-mno-memcpy' 24204 Force (do not force) the use of 'memcpy' for non-trivial block 24205 moves. The default is '-mno-memcpy', which allows GCC to inline 24206 most constant-sized copies. 24207 24208'-mlong-calls' 24209'-mno-long-calls' 24210 Disable (do not disable) use of the 'jal' instruction. Calling 24211 functions using 'jal' is more efficient but requires the caller and 24212 callee to be in the same 256 megabyte segment. 24213 24214 This option has no effect on abicalls code. The default is 24215 '-mno-long-calls'. 24216 24217'-mmad' 24218'-mno-mad' 24219 Enable (disable) use of the 'mad', 'madu' and 'mul' instructions, 24220 as provided by the R4650 ISA. 24221 24222'-mimadd' 24223'-mno-imadd' 24224 Enable (disable) use of the 'madd' and 'msub' integer instructions. 24225 The default is '-mimadd' on architectures that support 'madd' and 24226 'msub' except for the 74k architecture where it was found to 24227 generate slower code. 24228 24229'-mfused-madd' 24230'-mno-fused-madd' 24231 Enable (disable) use of the floating-point multiply-accumulate 24232 instructions, when they are available. The default is 24233 '-mfused-madd'. 24234 24235 On the R8000 CPU when multiply-accumulate instructions are used, 24236 the intermediate product is calculated to infinite precision and is 24237 not subject to the FCSR Flush to Zero bit. This may be undesirable 24238 in some circumstances. On other processors the result is 24239 numerically identical to the equivalent computation using separate 24240 multiply, add, subtract and negate instructions. 24241 24242'-nocpp' 24243 Tell the MIPS assembler to not run its preprocessor over user 24244 assembler files (with a '.s' suffix) when assembling them. 24245 24246'-mfix-24k' 24247'-mno-fix-24k' 24248 Work around the 24K E48 (lost data on stores during refill) errata. 24249 The workarounds are implemented by the assembler rather than by 24250 GCC. 24251 24252'-mfix-r4000' 24253'-mno-fix-r4000' 24254 Work around certain R4000 CPU errata: 24255 - A double-word or a variable shift may give an incorrect result 24256 if executed immediately after starting an integer division. 24257 - A double-word or a variable shift may give an incorrect result 24258 if executed while an integer multiplication is in progress. 24259 - An integer division may give an incorrect result if started in 24260 a delay slot of a taken branch or a jump. 24261 24262'-mfix-r4400' 24263'-mno-fix-r4400' 24264 Work around certain R4400 CPU errata: 24265 - A double-word or a variable shift may give an incorrect result 24266 if executed immediately after starting an integer division. 24267 24268'-mfix-r10000' 24269'-mno-fix-r10000' 24270 Work around certain R10000 errata: 24271 - 'll'/'sc' sequences may not behave atomically on revisions 24272 prior to 3.0. They may deadlock on revisions 2.6 and earlier. 24273 24274 This option can only be used if the target architecture supports 24275 branch-likely instructions. '-mfix-r10000' is the default when 24276 '-march=r10000' is used; '-mno-fix-r10000' is the default 24277 otherwise. 24278 24279'-mfix-r5900' 24280'-mno-fix-r5900' 24281 Do not attempt to schedule the preceding instruction into the delay 24282 slot of a branch instruction placed at the end of a short loop of 24283 six instructions or fewer and always schedule a 'nop' instruction 24284 there instead. The short loop bug under certain conditions causes 24285 loops to execute only once or twice, due to a hardware bug in the 24286 R5900 chip. The workaround is implemented by the assembler rather 24287 than by GCC. 24288 24289'-mfix-rm7000' 24290'-mno-fix-rm7000' 24291 Work around the RM7000 'dmult'/'dmultu' errata. The workarounds 24292 are implemented by the assembler rather than by GCC. 24293 24294'-mfix-vr4120' 24295'-mno-fix-vr4120' 24296 Work around certain VR4120 errata: 24297 - 'dmultu' does not always produce the correct result. 24298 - 'div' and 'ddiv' do not always produce the correct result if 24299 one of the operands is negative. 24300 The workarounds for the division errata rely on special functions 24301 in 'libgcc.a'. At present, these functions are only provided by 24302 the 'mips64vr*-elf' configurations. 24303 24304 Other VR4120 errata require a NOP to be inserted between certain 24305 pairs of instructions. These errata are handled by the assembler, 24306 not by GCC itself. 24307 24308'-mfix-vr4130' 24309 Work around the VR4130 'mflo'/'mfhi' errata. The workarounds are 24310 implemented by the assembler rather than by GCC, although GCC 24311 avoids using 'mflo' and 'mfhi' if the VR4130 'macc', 'macchi', 24312 'dmacc' and 'dmacchi' instructions are available instead. 24313 24314'-mfix-sb1' 24315'-mno-fix-sb1' 24316 Work around certain SB-1 CPU core errata. (This flag currently 24317 works around the SB-1 revision 2 "F1" and "F2" floating-point 24318 errata.) 24319 24320'-mr10k-cache-barrier=SETTING' 24321 Specify whether GCC should insert cache barriers to avoid the side 24322 effects of speculation on R10K processors. 24323 24324 In common with many processors, the R10K tries to predict the 24325 outcome of a conditional branch and speculatively executes 24326 instructions from the "taken" branch. It later aborts these 24327 instructions if the predicted outcome is wrong. However, on the 24328 R10K, even aborted instructions can have side effects. 24329 24330 This problem only affects kernel stores and, depending on the 24331 system, kernel loads. As an example, a speculatively-executed 24332 store may load the target memory into cache and mark the cache line 24333 as dirty, even if the store itself is later aborted. If a DMA 24334 operation writes to the same area of memory before the "dirty" line 24335 is flushed, the cached data overwrites the DMA-ed data. See the 24336 R10K processor manual for a full description, including other 24337 potential problems. 24338 24339 One workaround is to insert cache barrier instructions before every 24340 memory access that might be speculatively executed and that might 24341 have side effects even if aborted. '-mr10k-cache-barrier=SETTING' 24342 controls GCC's implementation of this workaround. It assumes that 24343 aborted accesses to any byte in the following regions does not have 24344 side effects: 24345 24346 1. the memory occupied by the current function's stack frame; 24347 24348 2. the memory occupied by an incoming stack argument; 24349 24350 3. the memory occupied by an object with a link-time-constant 24351 address. 24352 24353 It is the kernel's responsibility to ensure that speculative 24354 accesses to these regions are indeed safe. 24355 24356 If the input program contains a function declaration such as: 24357 24358 void foo (void); 24359 24360 then the implementation of 'foo' must allow 'j foo' and 'jal foo' 24361 to be executed speculatively. GCC honors this restriction for 24362 functions it compiles itself. It expects non-GCC functions (such 24363 as hand-written assembly code) to do the same. 24364 24365 The option has three forms: 24366 24367 '-mr10k-cache-barrier=load-store' 24368 Insert a cache barrier before a load or store that might be 24369 speculatively executed and that might have side effects even 24370 if aborted. 24371 24372 '-mr10k-cache-barrier=store' 24373 Insert a cache barrier before a store that might be 24374 speculatively executed and that might have side effects even 24375 if aborted. 24376 24377 '-mr10k-cache-barrier=none' 24378 Disable the insertion of cache barriers. This is the default 24379 setting. 24380 24381'-mflush-func=FUNC' 24382'-mno-flush-func' 24383 Specifies the function to call to flush the I and D caches, or to 24384 not call any such function. If called, the function must take the 24385 same arguments as the common '_flush_func', that is, the address of 24386 the memory range for which the cache is being flushed, the size of 24387 the memory range, and the number 3 (to flush both caches). The 24388 default depends on the target GCC was configured for, but commonly 24389 is either '_flush_func' or '__cpu_flush'. 24390 24391'mbranch-cost=NUM' 24392 Set the cost of branches to roughly NUM "simple" instructions. 24393 This cost is only a heuristic and is not guaranteed to produce 24394 consistent results across releases. A zero cost redundantly 24395 selects the default, which is based on the '-mtune' setting. 24396 24397'-mbranch-likely' 24398'-mno-branch-likely' 24399 Enable or disable use of Branch Likely instructions, regardless of 24400 the default for the selected architecture. By default, Branch 24401 Likely instructions may be generated if they are supported by the 24402 selected architecture. An exception is for the MIPS32 and MIPS64 24403 architectures and processors that implement those architectures; 24404 for those, Branch Likely instructions are not be generated by 24405 default because the MIPS32 and MIPS64 architectures specifically 24406 deprecate their use. 24407 24408'-mcompact-branches=never' 24409'-mcompact-branches=optimal' 24410'-mcompact-branches=always' 24411 These options control which form of branches will be generated. 24412 The default is '-mcompact-branches=optimal'. 24413 24414 The '-mcompact-branches=never' option ensures that compact branch 24415 instructions will never be generated. 24416 24417 The '-mcompact-branches=always' option ensures that a compact 24418 branch instruction will be generated if available. If a compact 24419 branch instruction is not available, a delay slot form of the 24420 branch will be used instead. 24421 24422 This option is supported from MIPS Release 6 onwards. 24423 24424 The '-mcompact-branches=optimal' option will cause a delay slot 24425 branch to be used if one is available in the current ISA and the 24426 delay slot is successfully filled. If the delay slot is not 24427 filled, a compact branch will be chosen if one is available. 24428 24429'-mfp-exceptions' 24430'-mno-fp-exceptions' 24431 Specifies whether FP exceptions are enabled. This affects how FP 24432 instructions are scheduled for some processors. The default is 24433 that FP exceptions are enabled. 24434 24435 For instance, on the SB-1, if FP exceptions are disabled, and we 24436 are emitting 64-bit code, then we can use both FP pipes. 24437 Otherwise, we can only use one FP pipe. 24438 24439'-mvr4130-align' 24440'-mno-vr4130-align' 24441 The VR4130 pipeline is two-way superscalar, but can only issue two 24442 instructions together if the first one is 8-byte aligned. When 24443 this option is enabled, GCC aligns pairs of instructions that it 24444 thinks should execute in parallel. 24445 24446 This option only has an effect when optimizing for the VR4130. It 24447 normally makes code faster, but at the expense of making it bigger. 24448 It is enabled by default at optimization level '-O3'. 24449 24450'-msynci' 24451'-mno-synci' 24452 Enable (disable) generation of 'synci' instructions on 24453 architectures that support it. The 'synci' instructions (if 24454 enabled) are generated when '__builtin___clear_cache' is compiled. 24455 24456 This option defaults to '-mno-synci', but the default can be 24457 overridden by configuring GCC with '--with-synci'. 24458 24459 When compiling code for single processor systems, it is generally 24460 safe to use 'synci'. However, on many multi-core (SMP) systems, it 24461 does not invalidate the instruction caches on all cores and may 24462 lead to undefined behavior. 24463 24464'-mrelax-pic-calls' 24465'-mno-relax-pic-calls' 24466 Try to turn PIC calls that are normally dispatched via register 24467 '$25' into direct calls. This is only possible if the linker can 24468 resolve the destination at link time and if the destination is 24469 within range for a direct call. 24470 24471 '-mrelax-pic-calls' is the default if GCC was configured to use an 24472 assembler and a linker that support the '.reloc' assembly directive 24473 and '-mexplicit-relocs' is in effect. With '-mno-explicit-relocs', 24474 this optimization can be performed by the assembler and the linker 24475 alone without help from the compiler. 24476 24477'-mmcount-ra-address' 24478'-mno-mcount-ra-address' 24479 Emit (do not emit) code that allows '_mcount' to modify the calling 24480 function's return address. When enabled, this option extends the 24481 usual '_mcount' interface with a new RA-ADDRESS parameter, which 24482 has type 'intptr_t *' and is passed in register '$12'. '_mcount' 24483 can then modify the return address by doing both of the following: 24484 * Returning the new address in register '$31'. 24485 * Storing the new address in '*RA-ADDRESS', if RA-ADDRESS is 24486 nonnull. 24487 24488 The default is '-mno-mcount-ra-address'. 24489 24490'-mframe-header-opt' 24491'-mno-frame-header-opt' 24492 Enable (disable) frame header optimization in the o32 ABI. When 24493 using the o32 ABI, calling functions will allocate 16 bytes on the 24494 stack for the called function to write out register arguments. 24495 When enabled, this optimization will suppress the allocation of the 24496 frame header if it can be determined that it is unused. 24497 24498 This optimization is off by default at all optimization levels. 24499 24500'-mlxc1-sxc1' 24501'-mno-lxc1-sxc1' 24502 When applicable, enable (disable) the generation of 'lwxc1', 24503 'swxc1', 'ldxc1', 'sdxc1' instructions. Enabled by default. 24504 24505'-mmadd4' 24506'-mno-madd4' 24507 When applicable, enable (disable) the generation of 4-operand 24508 'madd.s', 'madd.d' and related instructions. Enabled by default. 24509 24510 24511File: gcc.info, Node: MMIX Options, Next: MN10300 Options, Prev: MIPS Options, Up: Submodel Options 24512 245133.19.31 MMIX Options 24514-------------------- 24515 24516These options are defined for the MMIX: 24517 24518'-mlibfuncs' 24519'-mno-libfuncs' 24520 Specify that intrinsic library functions are being compiled, 24521 passing all values in registers, no matter the size. 24522 24523'-mepsilon' 24524'-mno-epsilon' 24525 Generate floating-point comparison instructions that compare with 24526 respect to the 'rE' epsilon register. 24527 24528'-mabi=mmixware' 24529'-mabi=gnu' 24530 Generate code that passes function parameters and return values 24531 that (in the called function) are seen as registers '$0' and up, as 24532 opposed to the GNU ABI which uses global registers '$231' and up. 24533 24534'-mzero-extend' 24535'-mno-zero-extend' 24536 When reading data from memory in sizes shorter than 64 bits, use 24537 (do not use) zero-extending load instructions by default, rather 24538 than sign-extending ones. 24539 24540'-mknuthdiv' 24541'-mno-knuthdiv' 24542 Make the result of a division yielding a remainder have the same 24543 sign as the divisor. With the default, '-mno-knuthdiv', the sign 24544 of the remainder follows the sign of the dividend. Both methods 24545 are arithmetically valid, the latter being almost exclusively used. 24546 24547'-mtoplevel-symbols' 24548'-mno-toplevel-symbols' 24549 Prepend (do not prepend) a ':' to all global symbols, so the 24550 assembly code can be used with the 'PREFIX' assembly directive. 24551 24552'-melf' 24553 Generate an executable in the ELF format, rather than the default 24554 'mmo' format used by the 'mmix' simulator. 24555 24556'-mbranch-predict' 24557'-mno-branch-predict' 24558 Use (do not use) the probable-branch instructions, when static 24559 branch prediction indicates a probable branch. 24560 24561'-mbase-addresses' 24562'-mno-base-addresses' 24563 Generate (do not generate) code that uses _base addresses_. Using 24564 a base address automatically generates a request (handled by the 24565 assembler and the linker) for a constant to be set up in a global 24566 register. The register is used for one or more base address 24567 requests within the range 0 to 255 from the value held in the 24568 register. The generally leads to short and fast code, but the 24569 number of different data items that can be addressed is limited. 24570 This means that a program that uses lots of static data may require 24571 '-mno-base-addresses'. 24572 24573'-msingle-exit' 24574'-mno-single-exit' 24575 Force (do not force) generated code to have a single exit point in 24576 each function. 24577 24578 24579File: gcc.info, Node: MN10300 Options, Next: Moxie Options, Prev: MMIX Options, Up: Submodel Options 24580 245813.19.32 MN10300 Options 24582----------------------- 24583 24584These '-m' options are defined for Matsushita MN10300 architectures: 24585 24586'-mmult-bug' 24587 Generate code to avoid bugs in the multiply instructions for the 24588 MN10300 processors. This is the default. 24589 24590'-mno-mult-bug' 24591 Do not generate code to avoid bugs in the multiply instructions for 24592 the MN10300 processors. 24593 24594'-mam33' 24595 Generate code using features specific to the AM33 processor. 24596 24597'-mno-am33' 24598 Do not generate code using features specific to the AM33 processor. 24599 This is the default. 24600 24601'-mam33-2' 24602 Generate code using features specific to the AM33/2.0 processor. 24603 24604'-mam34' 24605 Generate code using features specific to the AM34 processor. 24606 24607'-mtune=CPU-TYPE' 24608 Use the timing characteristics of the indicated CPU type when 24609 scheduling instructions. This does not change the targeted 24610 processor type. The CPU type must be one of 'mn10300', 'am33', 24611 'am33-2' or 'am34'. 24612 24613'-mreturn-pointer-on-d0' 24614 When generating a function that returns a pointer, return the 24615 pointer in both 'a0' and 'd0'. Otherwise, the pointer is returned 24616 only in 'a0', and attempts to call such functions without a 24617 prototype result in errors. Note that this option is on by 24618 default; use '-mno-return-pointer-on-d0' to disable it. 24619 24620'-mno-crt0' 24621 Do not link in the C run-time initialization object file. 24622 24623'-mrelax' 24624 Indicate to the linker that it should perform a relaxation 24625 optimization pass to shorten branches, calls and absolute memory 24626 addresses. This option only has an effect when used on the command 24627 line for the final link step. 24628 24629 This option makes symbolic debugging impossible. 24630 24631'-mliw' 24632 Allow the compiler to generate _Long Instruction Word_ instructions 24633 if the target is the 'AM33' or later. This is the default. This 24634 option defines the preprocessor macro '__LIW__'. 24635 24636'-mno-liw' 24637 Do not allow the compiler to generate _Long Instruction Word_ 24638 instructions. This option defines the preprocessor macro 24639 '__NO_LIW__'. 24640 24641'-msetlb' 24642 Allow the compiler to generate the _SETLB_ and _Lcc_ instructions 24643 if the target is the 'AM33' or later. This is the default. This 24644 option defines the preprocessor macro '__SETLB__'. 24645 24646'-mno-setlb' 24647 Do not allow the compiler to generate _SETLB_ or _Lcc_ 24648 instructions. This option defines the preprocessor macro 24649 '__NO_SETLB__'. 24650 24651 24652File: gcc.info, Node: Moxie Options, Next: MSP430 Options, Prev: MN10300 Options, Up: Submodel Options 24653 246543.19.33 Moxie Options 24655--------------------- 24656 24657'-meb' 24658 Generate big-endian code. This is the default for 'moxie-*-*' 24659 configurations. 24660 24661'-mel' 24662 Generate little-endian code. 24663 24664'-mmul.x' 24665 Generate mul.x and umul.x instructions. This is the default for 24666 'moxiebox-*-*' configurations. 24667 24668'-mno-crt0' 24669 Do not link in the C run-time initialization object file. 24670 24671 24672File: gcc.info, Node: MSP430 Options, Next: NDS32 Options, Prev: Moxie Options, Up: Submodel Options 24673 246743.19.34 MSP430 Options 24675---------------------- 24676 24677These options are defined for the MSP430: 24678 24679'-masm-hex' 24680 Force assembly output to always use hex constants. Normally such 24681 constants are signed decimals, but this option is available for 24682 testsuite and/or aesthetic purposes. 24683 24684'-mmcu=' 24685 Select the MCU to target. This is used to create a C preprocessor 24686 symbol based upon the MCU name, converted to upper case and pre- 24687 and post-fixed with '__'. This in turn is used by the 'msp430.h' 24688 header file to select an MCU-specific supplementary header file. 24689 24690 The option also sets the ISA to use. If the MCU name is one that 24691 is known to only support the 430 ISA then that is selected, 24692 otherwise the 430X ISA is selected. A generic MCU name of 'msp430' 24693 can also be used to select the 430 ISA. Similarly the generic 24694 'msp430x' MCU name selects the 430X ISA. 24695 24696 In addition an MCU-specific linker script is added to the linker 24697 command line. The script's name is the name of the MCU with '.ld' 24698 appended. Thus specifying '-mmcu=xxx' on the 'gcc' command line 24699 defines the C preprocessor symbol '__XXX__' and cause the linker to 24700 search for a script called 'xxx.ld'. 24701 24702 The ISA and hardware multiply supported for the different MCUs is 24703 hard-coded into GCC. However, an external 'devices.csv' file can be 24704 used to extend device support beyond those that have been 24705 hard-coded. 24706 24707 GCC searches for the 'devices.csv' file using the following methods 24708 in the given precedence order, where the first method takes 24709 precendence over the second which takes precedence over the third. 24710 24711 Include path specified with '-I' and '-L' 24712 'devices.csv' will be searched for in each of the directories 24713 specified by include paths and linker library search paths. 24714 Path specified by the environment variable 'MSP430_GCC_INCLUDE_DIR' 24715 Define the value of the global environment variable 24716 'MSP430_GCC_INCLUDE_DIR' to the full path to the directory 24717 containing devices.csv, and GCC will search this directory for 24718 devices.csv. If devices.csv is found, this directory will 24719 also be registered as an include path, and linker library 24720 path. Header files and linker scripts in this directory can 24721 therefore be used without manually specifying '-I' and '-L' on 24722 the command line. 24723 The 'msp430-elf{,bare}/include/devices' directory 24724 Finally, GCC will examine 'msp430-elf{,bare}/include/devices' 24725 from the toolchain root directory. This directory does not 24726 exist in a default installation, but if the user has created 24727 it and copied 'devices.csv' there, then the MCU data will be 24728 read. As above, this directory will also be registered as an 24729 include path, and linker library path. 24730 24731 If none of the above search methods find 'devices.csv', then the 24732 hard-coded MCU data is used. 24733 24734'-mwarn-mcu' 24735'-mno-warn-mcu' 24736 This option enables or disables warnings about conflicts between 24737 the MCU name specified by the '-mmcu' option and the ISA set by the 24738 '-mcpu' option and/or the hardware multiply support set by the 24739 '-mhwmult' option. It also toggles warnings about unrecognized MCU 24740 names. This option is on by default. 24741 24742'-mcpu=' 24743 Specifies the ISA to use. Accepted values are 'msp430', 'msp430x' 24744 and 'msp430xv2'. This option is deprecated. The '-mmcu=' option 24745 should be used to select the ISA. 24746 24747'-msim' 24748 Link to the simulator runtime libraries and linker script. 24749 Overrides any scripts that would be selected by the '-mmcu=' 24750 option. 24751 24752'-mlarge' 24753 Use large-model addressing (20-bit pointers, 20-bit 'size_t'). 24754 24755'-msmall' 24756 Use small-model addressing (16-bit pointers, 16-bit 'size_t'). 24757 24758'-mrelax' 24759 This option is passed to the assembler and linker, and allows the 24760 linker to perform certain optimizations that cannot be done until 24761 the final link. 24762 24763'mhwmult=' 24764 Describes the type of hardware multiply supported by the target. 24765 Accepted values are 'none' for no hardware multiply, '16bit' for 24766 the original 16-bit-only multiply supported by early MCUs. '32bit' 24767 for the 16/32-bit multiply supported by later MCUs and 'f5series' 24768 for the 16/32-bit multiply supported by F5-series MCUs. A value of 24769 'auto' can also be given. This tells GCC to deduce the hardware 24770 multiply support based upon the MCU name provided by the '-mmcu' 24771 option. If no '-mmcu' option is specified or if the MCU name is 24772 not recognized then no hardware multiply support is assumed. 24773 'auto' is the default setting. 24774 24775 Hardware multiplies are normally performed by calling a library 24776 routine. This saves space in the generated code. When compiling 24777 at '-O3' or higher however the hardware multiplier is invoked 24778 inline. This makes for bigger, but faster code. 24779 24780 The hardware multiply routines disable interrupts whilst running 24781 and restore the previous interrupt state when they finish. This 24782 makes them safe to use inside interrupt handlers as well as in 24783 normal code. 24784 24785'-minrt' 24786 Enable the use of a minimum runtime environment - no static 24787 initializers or constructors. This is intended for 24788 memory-constrained devices. The compiler includes special symbols 24789 in some objects that tell the linker and runtime which code 24790 fragments are required. 24791 24792'-mtiny-printf' 24793 Enable reduced code size 'printf' and 'puts' library functions. 24794 The 'tiny' implementations of these functions are not reentrant, so 24795 must be used with caution in multi-threaded applications. 24796 24797 Support for streams has been removed and the string to be printed 24798 will always be sent to stdout via the 'write' syscall. The string 24799 is not buffered before it is sent to write. 24800 24801 This option requires Newlib Nano IO, so GCC must be configured with 24802 '--enable-newlib-nano-formatted-io'. 24803 24804'-mmax-inline-shift=' 24805 This option takes an integer between 0 and 64 inclusive, and sets 24806 the maximum number of inline shift instructions which should be 24807 emitted to perform a shift operation by a constant amount. When 24808 this value needs to be exceeded, an mspabi helper function is used 24809 instead. The default value is 4. 24810 24811 This only affects cases where a shift by multiple positions cannot 24812 be completed with a single instruction (e.g. all shifts >1 on the 24813 430 ISA). 24814 24815 Shifts of a 32-bit value are at least twice as costly, so the value 24816 passed for this option is divided by 2 and the resulting value used 24817 instead. 24818 24819'-mcode-region=' 24820'-mdata-region=' 24821 These options tell the compiler where to place functions and data 24822 that do not have one of the 'lower', 'upper', 'either' or 'section' 24823 attributes. Possible values are 'lower', 'upper', 'either' or 24824 'any'. The first three behave like the corresponding attribute. 24825 The fourth possible value - 'any' - is the default. It leaves 24826 placement entirely up to the linker script and how it assigns the 24827 standard sections ('.text', '.data', etc) to the memory regions. 24828 24829'-msilicon-errata=' 24830 This option passes on a request to assembler to enable the fixes 24831 for the named silicon errata. 24832 24833'-msilicon-errata-warn=' 24834 This option passes on a request to the assembler to enable warning 24835 messages when a silicon errata might need to be applied. 24836 24837'-mwarn-devices-csv' 24838'-mno-warn-devices-csv' 24839 Warn if 'devices.csv' is not found or there are problem parsing it 24840 (default: on). 24841 24842 24843File: gcc.info, Node: NDS32 Options, Next: Nios II Options, Prev: MSP430 Options, Up: Submodel Options 24844 248453.19.35 NDS32 Options 24846--------------------- 24847 24848These options are defined for NDS32 implementations: 24849 24850'-mbig-endian' 24851 Generate code in big-endian mode. 24852 24853'-mlittle-endian' 24854 Generate code in little-endian mode. 24855 24856'-mreduced-regs' 24857 Use reduced-set registers for register allocation. 24858 24859'-mfull-regs' 24860 Use full-set registers for register allocation. 24861 24862'-mcmov' 24863 Generate conditional move instructions. 24864 24865'-mno-cmov' 24866 Do not generate conditional move instructions. 24867 24868'-mext-perf' 24869 Generate performance extension instructions. 24870 24871'-mno-ext-perf' 24872 Do not generate performance extension instructions. 24873 24874'-mext-perf2' 24875 Generate performance extension 2 instructions. 24876 24877'-mno-ext-perf2' 24878 Do not generate performance extension 2 instructions. 24879 24880'-mext-string' 24881 Generate string extension instructions. 24882 24883'-mno-ext-string' 24884 Do not generate string extension instructions. 24885 24886'-mv3push' 24887 Generate v3 push25/pop25 instructions. 24888 24889'-mno-v3push' 24890 Do not generate v3 push25/pop25 instructions. 24891 24892'-m16-bit' 24893 Generate 16-bit instructions. 24894 24895'-mno-16-bit' 24896 Do not generate 16-bit instructions. 24897 24898'-misr-vector-size=NUM' 24899 Specify the size of each interrupt vector, which must be 4 or 16. 24900 24901'-mcache-block-size=NUM' 24902 Specify the size of each cache block, which must be a power of 2 24903 between 4 and 512. 24904 24905'-march=ARCH' 24906 Specify the name of the target architecture. 24907 24908'-mcmodel=CODE-MODEL' 24909 Set the code model to one of 24910 'small' 24911 All the data and read-only data segments must be within 512KB 24912 addressing space. The text segment must be within 16MB 24913 addressing space. 24914 'medium' 24915 The data segment must be within 512KB while the read-only data 24916 segment can be within 4GB addressing space. The text segment 24917 should be still within 16MB addressing space. 24918 'large' 24919 All the text and data segments can be within 4GB addressing 24920 space. 24921 24922'-mctor-dtor' 24923 Enable constructor/destructor feature. 24924 24925'-mrelax' 24926 Guide linker to relax instructions. 24927 24928 24929File: gcc.info, Node: Nios II Options, Next: Nvidia PTX Options, Prev: NDS32 Options, Up: Submodel Options 24930 249313.19.36 Nios II Options 24932----------------------- 24933 24934These are the options defined for the Altera Nios II processor. 24935 24936'-G NUM' 24937 Put global and static objects less than or equal to NUM bytes into 24938 the small data or BSS sections instead of the normal data or BSS 24939 sections. The default value of NUM is 8. 24940 24941'-mgpopt=OPTION' 24942'-mgpopt' 24943'-mno-gpopt' 24944 Generate (do not generate) GP-relative accesses. The following 24945 OPTION names are recognized: 24946 24947 'none' 24948 Do not generate GP-relative accesses. 24949 24950 'local' 24951 Generate GP-relative accesses for small data objects that are 24952 not external, weak, or uninitialized common symbols. Also use 24953 GP-relative addressing for objects that have been explicitly 24954 placed in a small data section via a 'section' attribute. 24955 24956 'global' 24957 As for 'local', but also generate GP-relative accesses for 24958 small data objects that are external, weak, or common. If you 24959 use this option, you must ensure that all parts of your 24960 program (including libraries) are compiled with the same '-G' 24961 setting. 24962 24963 'data' 24964 Generate GP-relative accesses for all data objects in the 24965 program. If you use this option, the entire data and BSS 24966 segments of your program must fit in 64K of memory and you 24967 must use an appropriate linker script to allocate them within 24968 the addressable range of the global pointer. 24969 24970 'all' 24971 Generate GP-relative addresses for function pointers as well 24972 as data pointers. If you use this option, the entire text, 24973 data, and BSS segments of your program must fit in 64K of 24974 memory and you must use an appropriate linker script to 24975 allocate them within the addressable range of the global 24976 pointer. 24977 24978 '-mgpopt' is equivalent to '-mgpopt=local', and '-mno-gpopt' is 24979 equivalent to '-mgpopt=none'. 24980 24981 The default is '-mgpopt' except when '-fpic' or '-fPIC' is 24982 specified to generate position-independent code. Note that the 24983 Nios II ABI does not permit GP-relative accesses from shared 24984 libraries. 24985 24986 You may need to specify '-mno-gpopt' explicitly when building 24987 programs that include large amounts of small data, including large 24988 GOT data sections. In this case, the 16-bit offset for GP-relative 24989 addressing may not be large enough to allow access to the entire 24990 small data section. 24991 24992'-mgprel-sec=REGEXP' 24993 This option specifies additional section names that can be accessed 24994 via GP-relative addressing. It is most useful in conjunction with 24995 'section' attributes on variable declarations (*note Common 24996 Variable Attributes::) and a custom linker script. The REGEXP is a 24997 POSIX Extended Regular Expression. 24998 24999 This option does not affect the behavior of the '-G' option, and 25000 the specified sections are in addition to the standard '.sdata' and 25001 '.sbss' small-data sections that are recognized by '-mgpopt'. 25002 25003'-mr0rel-sec=REGEXP' 25004 This option specifies names of sections that can be accessed via a 25005 16-bit offset from 'r0'; that is, in the low 32K or high 32K of the 25006 32-bit address space. It is most useful in conjunction with 25007 'section' attributes on variable declarations (*note Common 25008 Variable Attributes::) and a custom linker script. The REGEXP is a 25009 POSIX Extended Regular Expression. 25010 25011 In contrast to the use of GP-relative addressing for small data, 25012 zero-based addressing is never generated by default and there are 25013 no conventional section names used in standard linker scripts for 25014 sections in the low or high areas of memory. 25015 25016'-mel' 25017'-meb' 25018 Generate little-endian (default) or big-endian (experimental) code, 25019 respectively. 25020 25021'-march=ARCH' 25022 This specifies the name of the target Nios II architecture. GCC 25023 uses this name to determine what kind of instructions it can emit 25024 when generating assembly code. Permissible names are: 'r1', 'r2'. 25025 25026 The preprocessor macro '__nios2_arch__' is available to programs, 25027 with value 1 or 2, indicating the targeted ISA level. 25028 25029'-mbypass-cache' 25030'-mno-bypass-cache' 25031 Force all load and store instructions to always bypass cache by 25032 using I/O variants of the instructions. The default is not to 25033 bypass the cache. 25034 25035'-mno-cache-volatile' 25036'-mcache-volatile' 25037 Volatile memory access bypass the cache using the I/O variants of 25038 the load and store instructions. The default is not to bypass the 25039 cache. 25040 25041'-mno-fast-sw-div' 25042'-mfast-sw-div' 25043 Do not use table-based fast divide for small numbers. The default 25044 is to use the fast divide at '-O3' and above. 25045 25046'-mno-hw-mul' 25047'-mhw-mul' 25048'-mno-hw-mulx' 25049'-mhw-mulx' 25050'-mno-hw-div' 25051'-mhw-div' 25052 Enable or disable emitting 'mul', 'mulx' and 'div' family of 25053 instructions by the compiler. The default is to emit 'mul' and not 25054 emit 'div' and 'mulx'. 25055 25056'-mbmx' 25057'-mno-bmx' 25058'-mcdx' 25059'-mno-cdx' 25060 Enable or disable generation of Nios II R2 BMX (bit manipulation) 25061 and CDX (code density) instructions. Enabling these instructions 25062 also requires '-march=r2'. Since these instructions are optional 25063 extensions to the R2 architecture, the default is not to emit them. 25064 25065'-mcustom-INSN=N' 25066'-mno-custom-INSN' 25067 Each '-mcustom-INSN=N' option enables use of a custom instruction 25068 with encoding N when generating code that uses INSN. For example, 25069 '-mcustom-fadds=253' generates custom instruction 253 for 25070 single-precision floating-point add operations instead of the 25071 default behavior of using a library call. 25072 25073 The following values of INSN are supported. Except as otherwise 25074 noted, floating-point operations are expected to be implemented 25075 with normal IEEE 754 semantics and correspond directly to the C 25076 operators or the equivalent GCC built-in functions (*note Other 25077 Builtins::). 25078 25079 Single-precision floating point: 25080 25081 'fadds', 'fsubs', 'fdivs', 'fmuls' 25082 Binary arithmetic operations. 25083 25084 'fnegs' 25085 Unary negation. 25086 25087 'fabss' 25088 Unary absolute value. 25089 25090 'fcmpeqs', 'fcmpges', 'fcmpgts', 'fcmples', 'fcmplts', 'fcmpnes' 25091 Comparison operations. 25092 25093 'fmins', 'fmaxs' 25094 Floating-point minimum and maximum. These instructions are 25095 only generated if '-ffinite-math-only' is specified. 25096 25097 'fsqrts' 25098 Unary square root operation. 25099 25100 'fcoss', 'fsins', 'ftans', 'fatans', 'fexps', 'flogs' 25101 Floating-point trigonometric and exponential functions. These 25102 instructions are only generated if 25103 '-funsafe-math-optimizations' is also specified. 25104 25105 Double-precision floating point: 25106 25107 'faddd', 'fsubd', 'fdivd', 'fmuld' 25108 Binary arithmetic operations. 25109 25110 'fnegd' 25111 Unary negation. 25112 25113 'fabsd' 25114 Unary absolute value. 25115 25116 'fcmpeqd', 'fcmpged', 'fcmpgtd', 'fcmpled', 'fcmpltd', 'fcmpned' 25117 Comparison operations. 25118 25119 'fmind', 'fmaxd' 25120 Double-precision minimum and maximum. These instructions are 25121 only generated if '-ffinite-math-only' is specified. 25122 25123 'fsqrtd' 25124 Unary square root operation. 25125 25126 'fcosd', 'fsind', 'ftand', 'fatand', 'fexpd', 'flogd' 25127 Double-precision trigonometric and exponential functions. 25128 These instructions are only generated if 25129 '-funsafe-math-optimizations' is also specified. 25130 25131 Conversions: 25132 'fextsd' 25133 Conversion from single precision to double precision. 25134 25135 'ftruncds' 25136 Conversion from double precision to single precision. 25137 25138 'fixsi', 'fixsu', 'fixdi', 'fixdu' 25139 Conversion from floating point to signed or unsigned integer 25140 types, with truncation towards zero. 25141 25142 'round' 25143 Conversion from single-precision floating point to signed 25144 integer, rounding to the nearest integer and ties away from 25145 zero. This corresponds to the '__builtin_lroundf' function 25146 when '-fno-math-errno' is used. 25147 25148 'floatis', 'floatus', 'floatid', 'floatud' 25149 Conversion from signed or unsigned integer types to 25150 floating-point types. 25151 25152 In addition, all of the following transfer instructions for 25153 internal registers X and Y must be provided to use any of the 25154 double-precision floating-point instructions. Custom instructions 25155 taking two double-precision source operands expect the first 25156 operand in the 64-bit register X. The other operand (or only 25157 operand of a unary operation) is given to the custom arithmetic 25158 instruction with the least significant half in source register SRC1 25159 and the most significant half in SRC2. A custom instruction that 25160 returns a double-precision result returns the most significant 32 25161 bits in the destination register and the other half in 32-bit 25162 register Y. GCC automatically generates the necessary code 25163 sequences to write register X and/or read register Y when 25164 double-precision floating-point instructions are used. 25165 25166 'fwrx' 25167 Write SRC1 into the least significant half of X and SRC2 into 25168 the most significant half of X. 25169 25170 'fwry' 25171 Write SRC1 into Y. 25172 25173 'frdxhi', 'frdxlo' 25174 Read the most or least (respectively) significant half of X 25175 and store it in DEST. 25176 25177 'frdy' 25178 Read the value of Y and store it into DEST. 25179 25180 Note that you can gain more local control over generation of Nios 25181 II custom instructions by using the 'target("custom-INSN=N")' and 25182 'target("no-custom-INSN")' function attributes (*note Function 25183 Attributes::) or pragmas (*note Function Specific Option 25184 Pragmas::). 25185 25186'-mcustom-fpu-cfg=NAME' 25187 25188 This option enables a predefined, named set of custom instruction 25189 encodings (see '-mcustom-INSN' above). Currently, the following 25190 sets are defined: 25191 25192 '-mcustom-fpu-cfg=60-1' is equivalent to: 25193 -mcustom-fmuls=252 25194 -mcustom-fadds=253 25195 -mcustom-fsubs=254 25196 -fsingle-precision-constant 25197 25198 '-mcustom-fpu-cfg=60-2' is equivalent to: 25199 -mcustom-fmuls=252 25200 -mcustom-fadds=253 25201 -mcustom-fsubs=254 25202 -mcustom-fdivs=255 25203 -fsingle-precision-constant 25204 25205 '-mcustom-fpu-cfg=72-3' is equivalent to: 25206 -mcustom-floatus=243 25207 -mcustom-fixsi=244 25208 -mcustom-floatis=245 25209 -mcustom-fcmpgts=246 25210 -mcustom-fcmples=249 25211 -mcustom-fcmpeqs=250 25212 -mcustom-fcmpnes=251 25213 -mcustom-fmuls=252 25214 -mcustom-fadds=253 25215 -mcustom-fsubs=254 25216 -mcustom-fdivs=255 25217 -fsingle-precision-constant 25218 25219 '-mcustom-fpu-cfg=fph2' is equivalent to: 25220 -mcustom-fabss=224 25221 -mcustom-fnegs=225 25222 -mcustom-fcmpnes=226 25223 -mcustom-fcmpeqs=227 25224 -mcustom-fcmpges=228 25225 -mcustom-fcmpgts=229 25226 -mcustom-fcmples=230 25227 -mcustom-fcmplts=231 25228 -mcustom-fmaxs=232 25229 -mcustom-fmins=233 25230 -mcustom-round=248 25231 -mcustom-fixsi=249 25232 -mcustom-floatis=250 25233 -mcustom-fsqrts=251 25234 -mcustom-fmuls=252 25235 -mcustom-fadds=253 25236 -mcustom-fsubs=254 25237 -mcustom-fdivs=255 25238 25239 Custom instruction assignments given by individual '-mcustom-INSN=' 25240 options override those given by '-mcustom-fpu-cfg=', regardless of 25241 the order of the options on the command line. 25242 25243 Note that you can gain more local control over selection of a FPU 25244 configuration by using the 'target("custom-fpu-cfg=NAME")' function 25245 attribute (*note Function Attributes::) or pragma (*note Function 25246 Specific Option Pragmas::). 25247 25248 The name FPH2 is an abbreviation for _Nios II Floating Point 25249 Hardware 2 Component_. Please note that the custom instructions 25250 enabled by '-mcustom-fmins=233' and '-mcustom-fmaxs=234' are only 25251 generated if '-ffinite-math-only' is specified. The custom 25252 instruction enabled by '-mcustom-round=248' is only generated if 25253 '-fno-math-errno' is specified. In contrast to the other 25254 configurations, '-fsingle-precision-constant' is not set. 25255 25256 These additional '-m' options are available for the Altera Nios II ELF 25257(bare-metal) target: 25258 25259'-mhal' 25260 Link with HAL BSP. This suppresses linking with the GCC-provided C 25261 runtime startup and termination code, and is typically used in 25262 conjunction with '-msys-crt0=' to specify the location of the 25263 alternate startup code provided by the HAL BSP. 25264 25265'-msmallc' 25266 Link with a limited version of the C library, '-lsmallc', rather 25267 than Newlib. 25268 25269'-msys-crt0=STARTFILE' 25270 STARTFILE is the file name of the startfile (crt0) to use when 25271 linking. This option is only useful in conjunction with '-mhal'. 25272 25273'-msys-lib=SYSTEMLIB' 25274 SYSTEMLIB is the library name of the library that provides 25275 low-level system calls required by the C library, e.g. 'read' and 25276 'write'. This option is typically used to link with a library 25277 provided by a HAL BSP. 25278 25279 25280File: gcc.info, Node: Nvidia PTX Options, Next: OpenRISC Options, Prev: Nios II Options, Up: Submodel Options 25281 252823.19.37 Nvidia PTX Options 25283-------------------------- 25284 25285These options are defined for Nvidia PTX: 25286 25287'-m64' 25288 Ignored, but preserved for backward compatibility. Only 64-bit ABI 25289 is supported. 25290 25291'-march=ARCHITECTURE-STRING' 25292 Generate code for the specified PTX ISA target architecture (e.g. 25293 'sm_35'). Valid architecture strings are 'sm_30', 'sm_35', 25294 'sm_53', 'sm_70', 'sm_75' and 'sm_80'. The default target 25295 architecture is sm_30. 25296 25297 This option sets the value of the preprocessor macro '__PTX_SM__'; 25298 for instance, for 'sm_35', it has the value '350'. 25299 25300'-misa=ARCHITECTURE-STRING' 25301 Alias of '-march='. 25302 25303'-march-map=ARCHITECTURE-STRING' 25304 Select the closest available '-march=' value that is not more 25305 capable. For instance, for '-march-map=sm_50' select 25306 '-march=sm_35', and for '-march-map=sm_53' select '-march=sm_53'. 25307 25308'-mptx=VERSION-STRING' 25309 Generate code for the specified PTX ISA version (e.g. '7.0'). 25310 Valid version strings include '3.1', '6.0', '6.3', and '7.0'. The 25311 default PTX ISA version is 6.0, unless a higher version is required 25312 for specified PTX ISA target architecture via option '-march='. 25313 25314 This option sets the values of the preprocessor macros 25315 '__PTX_ISA_VERSION_MAJOR__' and '__PTX_ISA_VERSION_MINOR__'; for 25316 instance, for '3.1' the macros have the values '3' and '1', 25317 respectively. 25318 25319'-mmainkernel' 25320 Link in code for a __main kernel. This is for stand-alone instead 25321 of offloading execution. 25322 25323'-moptimize' 25324 Apply partitioned execution optimizations. This is the default 25325 when any level of optimization is selected. 25326 25327'-msoft-stack' 25328 Generate code that does not use '.local' memory directly for stack 25329 storage. Instead, a per-warp stack pointer is maintained 25330 explicitly. This enables variable-length stack allocation (with 25331 variable-length arrays or 'alloca'), and when global memory is used 25332 for underlying storage, makes it possible to access automatic 25333 variables from other threads, or with atomic instructions. This 25334 code generation variant is used for OpenMP offloading, but the 25335 option is exposed on its own for the purpose of testing the 25336 compiler; to generate code suitable for linking into programs using 25337 OpenMP offloading, use option '-mgomp'. 25338 25339'-muniform-simt' 25340 Switch to code generation variant that allows to execute all 25341 threads in each warp, while maintaining memory state and side 25342 effects as if only one thread in each warp was active outside of 25343 OpenMP SIMD regions. All atomic operations and calls to runtime 25344 (malloc, free, vprintf) are conditionally executed (iff current 25345 lane index equals the master lane index), and the register being 25346 assigned is copied via a shuffle instruction from the master lane. 25347 Outside of SIMD regions lane 0 is the master; inside, each thread 25348 sees itself as the master. Shared memory array 'int __nvptx_uni[]' 25349 stores all-zeros or all-ones bitmasks for each warp, indicating 25350 current mode (0 outside of SIMD regions). Each thread can 25351 bitwise-and the bitmask at position 'tid.y' with current lane index 25352 to compute the master lane index. 25353 25354'-mgomp' 25355 Generate code for use in OpenMP offloading: enables '-msoft-stack' 25356 and '-muniform-simt' options, and selects corresponding multilib 25357 variant. 25358 25359 25360File: gcc.info, Node: OpenRISC Options, Next: PDP-11 Options, Prev: Nvidia PTX Options, Up: Submodel Options 25361 253623.19.38 OpenRISC Options 25363------------------------ 25364 25365These options are defined for OpenRISC: 25366 25367'-mboard=NAME' 25368 Configure a board specific runtime. This will be passed to the 25369 linker for newlib board library linking. The default is 'or1ksim'. 25370 25371'-mnewlib' 25372 This option is ignored; it is for compatibility purposes only. 25373 This used to select linker and preprocessor options for use with 25374 newlib. 25375 25376'-msoft-div' 25377'-mhard-div' 25378 Select software or hardware divide ('l.div', 'l.divu') 25379 instructions. This default is hardware divide. 25380 25381'-msoft-mul' 25382'-mhard-mul' 25383 Select software or hardware multiply ('l.mul', 'l.muli') 25384 instructions. This default is hardware multiply. 25385 25386'-msoft-float' 25387'-mhard-float' 25388 Select software or hardware for floating point operations. The 25389 default is software. 25390 25391'-mdouble-float' 25392 When '-mhard-float' is selected, enables generation of 25393 double-precision floating point instructions. By default functions 25394 from 'libgcc' are used to perform double-precision floating point 25395 operations. 25396 25397'-munordered-float' 25398 When '-mhard-float' is selected, enables generation of unordered 25399 floating point compare and set flag ('lf.sfun*') instructions. By 25400 default functions from 'libgcc' are used to perform unordered 25401 floating point compare and set flag operations. 25402 25403'-mcmov' 25404 Enable generation of conditional move ('l.cmov') instructions. By 25405 default the equivalent will be generated using set and branch. 25406 25407'-mror' 25408 Enable generation of rotate right ('l.ror') instructions. By 25409 default functions from 'libgcc' are used to perform rotate right 25410 operations. 25411 25412'-mrori' 25413 Enable generation of rotate right with immediate ('l.rori') 25414 instructions. By default functions from 'libgcc' are used to 25415 perform rotate right with immediate operations. 25416 25417'-msext' 25418 Enable generation of sign extension ('l.ext*') instructions. By 25419 default memory loads are used to perform sign extension. 25420 25421'-msfimm' 25422 Enable generation of compare and set flag with immediate ('l.sf*i') 25423 instructions. By default extra instructions will be generated to 25424 store the immediate to a register first. 25425 25426'-mshftimm' 25427 Enable generation of shift with immediate ('l.srai', 'l.srli', 25428 'l.slli') instructions. By default extra instructions will be 25429 generated to store the immediate to a register first. 25430 25431'-mcmodel=small' 25432 Generate OpenRISC code for the small model: The GOT is limited to 25433 64k. This is the default model. 25434 25435'-mcmodel=large' 25436 Generate OpenRISC code for the large model: The GOT may grow up to 25437 4G in size. 25438 25439 25440File: gcc.info, Node: PDP-11 Options, Next: picoChip Options, Prev: OpenRISC Options, Up: Submodel Options 25441 254423.19.39 PDP-11 Options 25443---------------------- 25444 25445These options are defined for the PDP-11: 25446 25447'-mfpu' 25448 Use hardware FPP floating point. This is the default. (FIS 25449 floating point on the PDP-11/40 is not supported.) Implies -m45. 25450 25451'-msoft-float' 25452 Do not use hardware floating point. 25453 25454'-mac0' 25455 Return floating-point results in ac0 (fr0 in Unix assembler 25456 syntax). 25457 25458'-mno-ac0' 25459 Return floating-point results in memory. This is the default. 25460 25461'-m40' 25462 Generate code for a PDP-11/40. Implies -msoft-float -mno-split. 25463 25464'-m45' 25465 Generate code for a PDP-11/45. This is the default. 25466 25467'-m10' 25468 Generate code for a PDP-11/10. Implies -msoft-float -mno-split. 25469 25470'-mint16' 25471'-mno-int32' 25472 Use 16-bit 'int'. This is the default. 25473 25474'-mint32' 25475'-mno-int16' 25476 Use 32-bit 'int'. 25477 25478'-msplit' 25479 Target has split instruction and data space. Implies -m45. 25480 25481'-munix-asm' 25482 Use Unix assembler syntax. 25483 25484'-mdec-asm' 25485 Use DEC assembler syntax. 25486 25487'-mgnu-asm' 25488 Use GNU assembler syntax. This is the default. 25489 25490'-mlra' 25491 Use the new LRA register allocator. By default, the old "reload" 25492 allocator is used. 25493 25494 25495File: gcc.info, Node: picoChip Options, Next: PowerPC Options, Prev: PDP-11 Options, Up: Submodel Options 25496 254973.19.40 picoChip Options 25498------------------------ 25499 25500These '-m' options are defined for picoChip implementations: 25501 25502'-mae=AE_TYPE' 25503 Set the instruction set, register set, and instruction scheduling 25504 parameters for array element type AE_TYPE. Supported values for 25505 AE_TYPE are 'ANY', 'MUL', and 'MAC'. 25506 25507 '-mae=ANY' selects a completely generic AE type. Code generated 25508 with this option runs on any of the other AE types. The code is 25509 not as efficient as it would be if compiled for a specific AE type, 25510 and some types of operation (e.g., multiplication) do not work 25511 properly on all types of AE. 25512 25513 '-mae=MUL' selects a MUL AE type. This is the most useful AE type 25514 for compiled code, and is the default. 25515 25516 '-mae=MAC' selects a DSP-style MAC AE. Code compiled with this 25517 option may suffer from poor performance of byte (char) 25518 manipulation, since the DSP AE does not provide hardware support 25519 for byte load/stores. 25520 25521'-msymbol-as-address' 25522 Enable the compiler to directly use a symbol name as an address in 25523 a load/store instruction, without first loading it into a register. 25524 Typically, the use of this option generates larger programs, which 25525 run faster than when the option isn't used. However, the results 25526 vary from program to program, so it is left as a user option, 25527 rather than being permanently enabled. 25528 25529'-mno-inefficient-warnings' 25530 Disables warnings about the generation of inefficient code. These 25531 warnings can be generated, for example, when compiling code that 25532 performs byte-level memory operations on the MAC AE type. The MAC 25533 AE has no hardware support for byte-level memory operations, so all 25534 byte load/stores must be synthesized from word load/store 25535 operations. This is inefficient and a warning is generated to 25536 indicate that you should rewrite the code to avoid byte operations, 25537 or to target an AE type that has the necessary hardware support. 25538 This option disables these warnings. 25539 25540 25541File: gcc.info, Node: PowerPC Options, Next: PRU Options, Prev: picoChip Options, Up: Submodel Options 25542 255433.19.41 PowerPC Options 25544----------------------- 25545 25546These are listed under *Note RS/6000 and PowerPC Options::. 25547 25548 25549File: gcc.info, Node: PRU Options, Next: RISC-V Options, Prev: PowerPC Options, Up: Submodel Options 25550 255513.19.42 PRU Options 25552------------------- 25553 25554These command-line options are defined for PRU target: 25555 25556'-minrt' 25557 Link with a minimum runtime environment, with no support for static 25558 initializers and constructors. Using this option can significantly 25559 reduce the size of the final ELF binary. Beware that the compiler 25560 could still generate code with static initializers and 25561 constructors. It is up to the programmer to ensure that the source 25562 program will not use those features. 25563 25564'-mmcu=MCU' 25565 Specify the PRU MCU variant to use. Check Newlib for the exact 25566 list of supported MCUs. 25567 25568'-mno-relax' 25569 Make GCC pass the '--no-relax' command-line option to the linker 25570 instead of the '--relax' option. 25571 25572'-mloop' 25573 Allow (or do not allow) GCC to use the LOOP instruction. 25574 25575'-mabi=VARIANT' 25576 Specify the ABI variant to output code for. '-mabi=ti' selects the 25577 unmodified TI ABI while '-mabi=gnu' selects a GNU variant that 25578 copes more naturally with certain GCC assumptions. These are the 25579 differences: 25580 25581 'Function Pointer Size' 25582 TI ABI specifies that function (code) pointers are 16-bit, 25583 whereas GNU supports only 32-bit data and code pointers. 25584 25585 'Optional Return Value Pointer' 25586 Function return values larger than 64 bits are passed by using 25587 a hidden pointer as the first argument of the function. TI 25588 ABI, though, mandates that the pointer can be NULL in case the 25589 caller is not using the returned value. GNU always passes and 25590 expects a valid return value pointer. 25591 25592 The current '-mabi=ti' implementation simply raises a compile error 25593 when any of the above code constructs is detected. As a 25594 consequence the standard C library cannot be built and it is 25595 omitted when linking with '-mabi=ti'. 25596 25597 Relaxation is a GNU feature and for safety reasons is disabled when 25598 using '-mabi=ti'. The TI toolchain does not emit relocations for 25599 QBBx instructions, so the GNU linker cannot adjust them when 25600 shortening adjacent LDI32 pseudo instructions. 25601 25602 25603File: gcc.info, Node: RISC-V Options, Next: RL78 Options, Prev: PRU Options, Up: Submodel Options 25604 256053.19.43 RISC-V Options 25606---------------------- 25607 25608These command-line options are defined for RISC-V targets: 25609 25610'-mbranch-cost=N' 25611 Set the cost of branches to roughly N instructions. 25612 25613'-mplt' 25614'-mno-plt' 25615 When generating PIC code, do or don't allow the use of PLTs. 25616 Ignored for non-PIC. The default is '-mplt'. 25617 25618'-mabi=ABI-STRING' 25619 Specify integer and floating-point calling convention. ABI-STRING 25620 contains two parts: the size of integer types and the registers 25621 used for floating-point types. For example '-march=rv64ifd 25622 -mabi=lp64d' means that 'long' and pointers are 64-bit (implicitly 25623 defining 'int' to be 32-bit), and that floating-point values up to 25624 64 bits wide are passed in F registers. Contrast this with 25625 '-march=rv64ifd -mabi=lp64f', which still allows the compiler to 25626 generate code that uses the F and D extensions but only allows 25627 floating-point values up to 32 bits long to be passed in registers; 25628 or '-march=rv64ifd -mabi=lp64', in which no floating-point 25629 arguments will be passed in registers. 25630 25631 The default for this argument is system dependent, users who want a 25632 specific calling convention should specify one explicitly. The 25633 valid calling conventions are: 'ilp32', 'ilp32f', 'ilp32d', 'lp64', 25634 'lp64f', and 'lp64d'. Some calling conventions are impossible to 25635 implement on some ISAs: for example, '-march=rv32if -mabi=ilp32d' 25636 is invalid because the ABI requires 64-bit values be passed in F 25637 registers, but F registers are only 32 bits wide. There is also 25638 the 'ilp32e' ABI that can only be used with the 'rv32e' 25639 architecture. This ABI is not well specified at present, and is 25640 subject to change. 25641 25642'-mfdiv' 25643'-mno-fdiv' 25644 Do or don't use hardware floating-point divide and square root 25645 instructions. This requires the F or D extensions for 25646 floating-point registers. The default is to use them if the 25647 specified architecture has these instructions. 25648 25649'-mdiv' 25650'-mno-div' 25651 Do or don't use hardware instructions for integer division. This 25652 requires the M extension. The default is to use them if the 25653 specified architecture has these instructions. 25654 25655'-misa-spec=ISA-SPEC-STRING' 25656 Specify the version of the RISC-V Unprivileged (formerly 25657 User-Level) ISA specification to produce code conforming to. The 25658 possibilities for ISA-SPEC-STRING are: 25659 '2.2' 25660 Produce code conforming to version 2.2. 25661 '20190608' 25662 Produce code conforming to version 20190608. 25663 '20191213' 25664 Produce code conforming to version 20191213. 25665 The default is '-misa-spec=20191213' unless GCC has been configured 25666 with '--with-isa-spec=' specifying a different default version. 25667 25668'-march=ISA-STRING' 25669 Generate code for given RISC-V ISA (e.g. 'rv64im'). ISA strings 25670 must be lower-case. Examples include 'rv64i', 'rv32g', 'rv32e', 25671 and 'rv32imaf'. 25672 25673 When '-march=' is not specified, use the setting from '-mcpu'. 25674 25675 If both '-march' and '-mcpu=' are not specified, the default for 25676 this argument is system dependent, users who want a specific 25677 architecture extensions should specify one explicitly. 25678 25679'-mcpu=PROCESSOR-STRING' 25680 Use architecture of and optimize the output for the given 25681 processor, specified by particular CPU name. Permissible values 25682 for this option are: 'sifive-e20', 'sifive-e21', 'sifive-e24', 25683 'sifive-e31', 'sifive-e34', 'sifive-e76', 'sifive-s21', 25684 'sifive-s51', 'sifive-s54', 'sifive-s76', 'sifive-u54', and 25685 'sifive-u74'. 25686 25687'-mtune=PROCESSOR-STRING' 25688 Optimize the output for the given processor, specified by 25689 microarchitecture or particular CPU name. Permissible values for 25690 this option are: 'rocket', 'sifive-3-series', 'sifive-5-series', 25691 'sifive-7-series', 'size', and all valid options for '-mcpu='. 25692 25693 When '-mtune=' is not specified, use the setting from '-mcpu', the 25694 default is 'rocket' if both are not specified. 25695 25696 The 'size' choice is not intended for use by end-users. This is 25697 used when '-Os' is specified. It overrides the instruction cost 25698 info provided by '-mtune=', but does not override the pipeline 25699 info. This helps reduce code size while still giving good 25700 performance. 25701 25702'-mpreferred-stack-boundary=NUM' 25703 Attempt to keep the stack boundary aligned to a 2 raised to NUM 25704 byte boundary. If '-mpreferred-stack-boundary' is not specified, 25705 the default is 4 (16 bytes or 128-bits). 25706 25707 *Warning:* If you use this switch, then you must build all modules 25708 with the same value, including any libraries. This includes the 25709 system libraries and startup modules. 25710 25711'-msmall-data-limit=N' 25712 Put global and static data smaller than N bytes into a special 25713 section (on some targets). 25714 25715'-msave-restore' 25716'-mno-save-restore' 25717 Do or don't use smaller but slower prologue and epilogue code that 25718 uses library function calls. The default is to use fast inline 25719 prologues and epilogues. 25720 25721'-mshorten-memrefs' 25722'-mno-shorten-memrefs' 25723 Do or do not attempt to make more use of compressed load/store 25724 instructions by replacing a load/store of 'base register + large 25725 offset' with a new load/store of 'new base + small offset'. If the 25726 new base gets stored in a compressed register, then the new 25727 load/store can be compressed. Currently targets 32-bit integer 25728 load/stores only. 25729 25730'-mstrict-align' 25731'-mno-strict-align' 25732 Do not or do generate unaligned memory accesses. The default is 25733 set depending on whether the processor we are optimizing for 25734 supports fast unaligned access or not. 25735 25736'-mcmodel=medlow' 25737 Generate code for the medium-low code model. The program and its 25738 statically defined symbols must lie within a single 2 GiB address 25739 range and must lie between absolute addresses -2 GiB and +2 GiB. 25740 Programs can be statically or dynamically linked. This is the 25741 default code model. 25742 25743'-mcmodel=medany' 25744 Generate code for the medium-any code model. The program and its 25745 statically defined symbols must be within any single 2 GiB address 25746 range. Programs can be statically or dynamically linked. 25747 25748 The code generated by the medium-any code model is 25749 position-independent, but is not guaranteed to function correctly 25750 when linked into position-independent executables or libraries. 25751 25752'-mexplicit-relocs' 25753'-mno-exlicit-relocs' 25754 Use or do not use assembler relocation operators when dealing with 25755 symbolic addresses. The alternative is to use assembler macros 25756 instead, which may limit optimization. 25757 25758'-mrelax' 25759'-mno-relax' 25760 Take advantage of linker relaxations to reduce the number of 25761 instructions required to materialize symbol addresses. The default 25762 is to take advantage of linker relaxations. 25763 25764'-memit-attribute' 25765'-mno-emit-attribute' 25766 Emit (do not emit) RISC-V attribute to record extra information 25767 into ELF objects. This feature requires at least binutils 2.32. 25768 25769'-malign-data=TYPE' 25770 Control how GCC aligns variables and constants of array, structure, 25771 or union types. Supported values for TYPE are 'xlen' which uses x 25772 register width as the alignment value, and 'natural' which uses 25773 natural alignment. 'xlen' is the default. 25774 25775'-mbig-endian' 25776 Generate big-endian code. This is the default when GCC is 25777 configured for a 'riscv64be-*-*' or 'riscv32be-*-*' target. 25778 25779'-mlittle-endian' 25780 Generate little-endian code. This is the default when GCC is 25781 configured for a 'riscv64-*-*' or 'riscv32-*-*' but not a 25782 'riscv64be-*-*' or 'riscv32be-*-*' target. 25783 25784'-mstack-protector-guard=GUARD' 25785'-mstack-protector-guard-reg=REG' 25786'-mstack-protector-guard-offset=OFFSET' 25787 Generate stack protection code using canary at GUARD. Supported 25788 locations are 'global' for a global canary or 'tls' for per-thread 25789 canary in the TLS block. 25790 25791 With the latter choice the options 25792 '-mstack-protector-guard-reg=REG' and 25793 '-mstack-protector-guard-offset=OFFSET' furthermore specify which 25794 register to use as base register for reading the canary, and from 25795 what offset from that base register. There is no default register 25796 or offset as this is entirely for use within the Linux kernel. 25797 25798 25799File: gcc.info, Node: RL78 Options, Next: RS/6000 and PowerPC Options, Prev: RISC-V Options, Up: Submodel Options 25800 258013.19.44 RL78 Options 25802-------------------- 25803 25804'-msim' 25805 Links in additional target libraries to support operation within a 25806 simulator. 25807 25808'-mmul=none' 25809'-mmul=g10' 25810'-mmul=g13' 25811'-mmul=g14' 25812'-mmul=rl78' 25813 Specifies the type of hardware multiplication and division support 25814 to be used. The simplest is 'none', which uses software for both 25815 multiplication and division. This is the default. The 'g13' value 25816 is for the hardware multiply/divide peripheral found on the 25817 RL78/G13 (S2 core) targets. The 'g14' value selects the use of the 25818 multiplication and division instructions supported by the RL78/G14 25819 (S3 core) parts. The value 'rl78' is an alias for 'g14' and the 25820 value 'mg10' is an alias for 'none'. 25821 25822 In addition a C preprocessor macro is defined, based upon the 25823 setting of this option. Possible values are: '__RL78_MUL_NONE__', 25824 '__RL78_MUL_G13__' or '__RL78_MUL_G14__'. 25825 25826'-mcpu=g10' 25827'-mcpu=g13' 25828'-mcpu=g14' 25829'-mcpu=rl78' 25830 Specifies the RL78 core to target. The default is the G14 core, 25831 also known as an S3 core or just RL78. The G13 or S2 core does not 25832 have multiply or divide instructions, instead it uses a hardware 25833 peripheral for these operations. The G10 or S1 core does not have 25834 register banks, so it uses a different calling convention. 25835 25836 If this option is set it also selects the type of hardware multiply 25837 support to use, unless this is overridden by an explicit 25838 '-mmul=none' option on the command line. Thus specifying 25839 '-mcpu=g13' enables the use of the G13 hardware multiply peripheral 25840 and specifying '-mcpu=g10' disables the use of hardware 25841 multiplications altogether. 25842 25843 Note, although the RL78/G14 core is the default target, specifying 25844 '-mcpu=g14' or '-mcpu=rl78' on the command line does change the 25845 behavior of the toolchain since it also enables G14 hardware 25846 multiply support. If these options are not specified on the 25847 command line then software multiplication routines will be used 25848 even though the code targets the RL78 core. This is for backwards 25849 compatibility with older toolchains which did not have hardware 25850 multiply and divide support. 25851 25852 In addition a C preprocessor macro is defined, based upon the 25853 setting of this option. Possible values are: '__RL78_G10__', 25854 '__RL78_G13__' or '__RL78_G14__'. 25855 25856'-mg10' 25857'-mg13' 25858'-mg14' 25859'-mrl78' 25860 These are aliases for the corresponding '-mcpu=' option. They are 25861 provided for backwards compatibility. 25862 25863'-mallregs' 25864 Allow the compiler to use all of the available registers. By 25865 default registers 'r24..r31' are reserved for use in interrupt 25866 handlers. With this option enabled these registers can be used in 25867 ordinary functions as well. 25868 25869'-m64bit-doubles' 25870'-m32bit-doubles' 25871 Make the 'double' data type be 64 bits ('-m64bit-doubles') or 32 25872 bits ('-m32bit-doubles') in size. The default is 25873 '-m32bit-doubles'. 25874 25875'-msave-mduc-in-interrupts' 25876'-mno-save-mduc-in-interrupts' 25877 Specifies that interrupt handler functions should preserve the MDUC 25878 registers. This is only necessary if normal code might use the 25879 MDUC registers, for example because it performs multiplication and 25880 division operations. The default is to ignore the MDUC registers 25881 as this makes the interrupt handlers faster. The target option 25882 -mg13 needs to be passed for this to work as this feature is only 25883 available on the G13 target (S2 core). The MDUC registers will 25884 only be saved if the interrupt handler performs a multiplication or 25885 division operation or it calls another function. 25886 25887 25888File: gcc.info, Node: RS/6000 and PowerPC Options, Next: RX Options, Prev: RL78 Options, Up: Submodel Options 25889 258903.19.45 IBM RS/6000 and PowerPC Options 25891--------------------------------------- 25892 25893These '-m' options are defined for the IBM RS/6000 and PowerPC: 25894'-mpowerpc-gpopt' 25895'-mno-powerpc-gpopt' 25896'-mpowerpc-gfxopt' 25897'-mno-powerpc-gfxopt' 25898'-mpowerpc64' 25899'-mno-powerpc64' 25900'-mmfcrf' 25901'-mno-mfcrf' 25902'-mpopcntb' 25903'-mno-popcntb' 25904'-mpopcntd' 25905'-mno-popcntd' 25906'-mfprnd' 25907'-mno-fprnd' 25908'-mcmpb' 25909'-mno-cmpb' 25910'-mhard-dfp' 25911'-mno-hard-dfp' 25912 You use these options to specify which instructions are available 25913 on the processor you are using. The default value of these options 25914 is determined when configuring GCC. Specifying the 25915 '-mcpu=CPU_TYPE' overrides the specification of these options. We 25916 recommend you use the '-mcpu=CPU_TYPE' option rather than the 25917 options listed above. 25918 25919 Specifying '-mpowerpc-gpopt' allows GCC to use the optional PowerPC 25920 architecture instructions in the General Purpose group, including 25921 floating-point square root. Specifying '-mpowerpc-gfxopt' allows 25922 GCC to use the optional PowerPC architecture instructions in the 25923 Graphics group, including floating-point select. 25924 25925 The '-mmfcrf' option allows GCC to generate the move from condition 25926 register field instruction implemented on the POWER4 processor and 25927 other processors that support the PowerPC V2.01 architecture. The 25928 '-mpopcntb' option allows GCC to generate the popcount and 25929 double-precision FP reciprocal estimate instruction implemented on 25930 the POWER5 processor and other processors that support the PowerPC 25931 V2.02 architecture. The '-mpopcntd' option allows GCC to generate 25932 the popcount instruction implemented on the POWER7 processor and 25933 other processors that support the PowerPC V2.06 architecture. The 25934 '-mfprnd' option allows GCC to generate the FP round to integer 25935 instructions implemented on the POWER5+ processor and other 25936 processors that support the PowerPC V2.03 architecture. The 25937 '-mcmpb' option allows GCC to generate the compare bytes 25938 instruction implemented on the POWER6 processor and other 25939 processors that support the PowerPC V2.05 architecture. The 25940 '-mhard-dfp' option allows GCC to generate the decimal 25941 floating-point instructions implemented on some POWER processors. 25942 25943 The '-mpowerpc64' option allows GCC to generate the additional 25944 64-bit instructions that are found in the full PowerPC64 25945 architecture and to treat GPRs as 64-bit, doubleword quantities. 25946 GCC defaults to '-mno-powerpc64'. 25947 25948'-mcpu=CPU_TYPE' 25949 Set architecture type, register usage, and instruction scheduling 25950 parameters for machine type CPU_TYPE. Supported values for 25951 CPU_TYPE are '401', '403', '405', '405fp', '440', '440fp', '464', 25952 '464fp', '476', '476fp', '505', '601', '602', '603', '603e', '604', 25953 '604e', '620', '630', '740', '7400', '7450', '750', '801', '821', 25954 '823', '860', '970', '8540', 'a2', 'e300c2', 'e300c3', 'e500mc', 25955 'e500mc64', 'e5500', 'e6500', 'ec603e', 'G3', 'G4', 'G5', 'titan', 25956 'power3', 'power4', 'power5', 'power5+', 'power6', 'power6x', 25957 'power7', 'power8', 'power9', 'power10', 'powerpc', 'powerpc64', 25958 'powerpc64le', 'rs64', and 'native'. 25959 25960 '-mcpu=powerpc', '-mcpu=powerpc64', and '-mcpu=powerpc64le' specify 25961 pure 32-bit PowerPC (either endian), 64-bit big endian PowerPC and 25962 64-bit little endian PowerPC architecture machine types, with an 25963 appropriate, generic processor model assumed for scheduling 25964 purposes. 25965 25966 Specifying 'native' as cpu type detects and selects the 25967 architecture option that corresponds to the host processor of the 25968 system performing the compilation. '-mcpu=native' has no effect if 25969 GCC does not recognize the processor. 25970 25971 The other options specify a specific processor. Code generated 25972 under those options runs best on that processor, and may not run at 25973 all on others. 25974 25975 The '-mcpu' options automatically enable or disable the following 25976 options: 25977 25978 -maltivec -mfprnd -mhard-float -mmfcrf -mmultiple 25979 -mpopcntb -mpopcntd -mpowerpc64 25980 -mpowerpc-gpopt -mpowerpc-gfxopt 25981 -mmulhw -mdlmzb -mmfpgpr -mvsx 25982 -mcrypto -mhtm -mpower8-fusion -mpower8-vector 25983 -mquad-memory -mquad-memory-atomic -mfloat128 25984 -mfloat128-hardware -mprefixed -mpcrel -mmma 25985 -mrop-protect 25986 25987 The particular options set for any particular CPU varies between 25988 compiler versions, depending on what setting seems to produce 25989 optimal code for that CPU; it doesn't necessarily reflect the 25990 actual hardware's capabilities. If you wish to set an individual 25991 option to a particular value, you may specify it after the '-mcpu' 25992 option, like '-mcpu=970 -mno-altivec'. 25993 25994 On AIX, the '-maltivec' and '-mpowerpc64' options are not enabled 25995 or disabled by the '-mcpu' option at present because AIX does not 25996 have full support for these options. You may still enable or 25997 disable them individually if you're sure it'll work in your 25998 environment. 25999 26000'-mtune=CPU_TYPE' 26001 Set the instruction scheduling parameters for machine type 26002 CPU_TYPE, but do not set the architecture type or register usage, 26003 as '-mcpu=CPU_TYPE' does. The same values for CPU_TYPE are used 26004 for '-mtune' as for '-mcpu'. If both are specified, the code 26005 generated uses the architecture and registers set by '-mcpu', but 26006 the scheduling parameters set by '-mtune'. 26007 26008'-mcmodel=small' 26009 Generate PowerPC64 code for the small model: The TOC is limited to 26010 64k. 26011 26012'-mcmodel=medium' 26013 Generate PowerPC64 code for the medium model: The TOC and other 26014 static data may be up to a total of 4G in size. This is the 26015 default for 64-bit Linux. 26016 26017'-mcmodel=large' 26018 Generate PowerPC64 code for the large model: The TOC may be up to 26019 4G in size. Other data and code is only limited by the 64-bit 26020 address space. 26021 26022'-maltivec' 26023'-mno-altivec' 26024 Generate code that uses (does not use) AltiVec instructions, and 26025 also enable the use of built-in functions that allow more direct 26026 access to the AltiVec instruction set. You may also need to set 26027 '-mabi=altivec' to adjust the current ABI with AltiVec ABI 26028 enhancements. 26029 26030 When '-maltivec' is used, the element order for AltiVec intrinsics 26031 such as 'vec_splat', 'vec_extract', and 'vec_insert' match array 26032 element order corresponding to the endianness of the target. That 26033 is, element zero identifies the leftmost element in a vector 26034 register when targeting a big-endian platform, and identifies the 26035 rightmost element in a vector register when targeting a 26036 little-endian platform. 26037 26038'-mvrsave' 26039'-mno-vrsave' 26040 Generate VRSAVE instructions when generating AltiVec code. 26041 26042'-msecure-plt' 26043 Generate code that allows 'ld' and 'ld.so' to build executables and 26044 shared libraries with non-executable '.plt' and '.got' sections. 26045 This is a PowerPC 32-bit SYSV ABI option. 26046 26047'-mbss-plt' 26048 Generate code that uses a BSS '.plt' section that 'ld.so' fills in, 26049 and requires '.plt' and '.got' sections that are both writable and 26050 executable. This is a PowerPC 32-bit SYSV ABI option. 26051 26052'-misel' 26053'-mno-isel' 26054 This switch enables or disables the generation of ISEL 26055 instructions. 26056 26057'-mvsx' 26058'-mno-vsx' 26059 Generate code that uses (does not use) vector/scalar (VSX) 26060 instructions, and also enable the use of built-in functions that 26061 allow more direct access to the VSX instruction set. 26062 26063'-mcrypto' 26064'-mno-crypto' 26065 Enable the use (disable) of the built-in functions that allow 26066 direct access to the cryptographic instructions that were added in 26067 version 2.07 of the PowerPC ISA. 26068 26069'-mhtm' 26070'-mno-htm' 26071 Enable (disable) the use of the built-in functions that allow 26072 direct access to the Hardware Transactional Memory (HTM) 26073 instructions that were added in version 2.07 of the PowerPC ISA. 26074 26075'-mpower8-fusion' 26076'-mno-power8-fusion' 26077 Generate code that keeps (does not keeps) some integer operations 26078 adjacent so that the instructions can be fused together on power8 26079 and later processors. 26080 26081'-mpower8-vector' 26082'-mno-power8-vector' 26083 Generate code that uses (does not use) the vector and scalar 26084 instructions that were added in version 2.07 of the PowerPC ISA. 26085 Also enable the use of built-in functions that allow more direct 26086 access to the vector instructions. 26087 26088'-mquad-memory' 26089'-mno-quad-memory' 26090 Generate code that uses (does not use) the non-atomic quad word 26091 memory instructions. The '-mquad-memory' option requires use of 26092 64-bit mode. 26093 26094'-mquad-memory-atomic' 26095'-mno-quad-memory-atomic' 26096 Generate code that uses (does not use) the atomic quad word memory 26097 instructions. The '-mquad-memory-atomic' option requires use of 26098 64-bit mode. 26099 26100'-mfloat128' 26101'-mno-float128' 26102 Enable/disable the __FLOAT128 keyword for IEEE 128-bit floating 26103 point and use either software emulation for IEEE 128-bit floating 26104 point or hardware instructions. 26105 26106 The VSX instruction set ('-mvsx') must be enabled to use the IEEE 26107 128-bit floating point support. The IEEE 128-bit floating point is 26108 only supported on Linux. 26109 26110 The default for '-mfloat128' is enabled on PowerPC Linux systems 26111 using the VSX instruction set, and disabled on other systems. 26112 26113 If you use the ISA 3.0 instruction set ('-mpower9-vector' or 26114 '-mcpu=power9') on a 64-bit system, the IEEE 128-bit floating point 26115 support will also enable the generation of ISA 3.0 IEEE 128-bit 26116 floating point instructions. Otherwise, if you do not specify to 26117 generate ISA 3.0 instructions or you are targeting a 32-bit big 26118 endian system, IEEE 128-bit floating point will be done with 26119 software emulation. 26120 26121'-mfloat128-hardware' 26122'-mno-float128-hardware' 26123 Enable/disable using ISA 3.0 hardware instructions to support the 26124 __FLOAT128 data type. 26125 26126 The default for '-mfloat128-hardware' is enabled on PowerPC Linux 26127 systems using the ISA 3.0 instruction set, and disabled on other 26128 systems. 26129 26130'-m32' 26131'-m64' 26132 Generate code for 32-bit or 64-bit environments of Darwin and SVR4 26133 targets (including GNU/Linux). The 32-bit environment sets int, 26134 long and pointer to 32 bits and generates code that runs on any 26135 PowerPC variant. The 64-bit environment sets int to 32 bits and 26136 long and pointer to 64 bits, and generates code for PowerPC64, as 26137 for '-mpowerpc64'. 26138 26139'-mfull-toc' 26140'-mno-fp-in-toc' 26141'-mno-sum-in-toc' 26142'-mminimal-toc' 26143 Modify generation of the TOC (Table Of Contents), which is created 26144 for every executable file. The '-mfull-toc' option is selected by 26145 default. In that case, GCC allocates at least one TOC entry for 26146 each unique non-automatic variable reference in your program. GCC 26147 also places floating-point constants in the TOC. However, only 26148 16,384 entries are available in the TOC. 26149 26150 If you receive a linker error message that saying you have 26151 overflowed the available TOC space, you can reduce the amount of 26152 TOC space used with the '-mno-fp-in-toc' and '-mno-sum-in-toc' 26153 options. '-mno-fp-in-toc' prevents GCC from putting floating-point 26154 constants in the TOC and '-mno-sum-in-toc' forces GCC to generate 26155 code to calculate the sum of an address and a constant at run time 26156 instead of putting that sum into the TOC. You may specify one or 26157 both of these options. Each causes GCC to produce very slightly 26158 slower and larger code at the expense of conserving TOC space. 26159 26160 If you still run out of space in the TOC even when you specify both 26161 of these options, specify '-mminimal-toc' instead. This option 26162 causes GCC to make only one TOC entry for every file. When you 26163 specify this option, GCC produces code that is slower and larger 26164 but which uses extremely little TOC space. You may wish to use 26165 this option only on files that contain less frequently-executed 26166 code. 26167 26168'-maix64' 26169'-maix32' 26170 Enable 64-bit AIX ABI and calling convention: 64-bit pointers, 26171 64-bit 'long' type, and the infrastructure needed to support them. 26172 Specifying '-maix64' implies '-mpowerpc64', while '-maix32' 26173 disables the 64-bit ABI and implies '-mno-powerpc64'. GCC defaults 26174 to '-maix32'. 26175 26176'-mxl-compat' 26177'-mno-xl-compat' 26178 Produce code that conforms more closely to IBM XL compiler 26179 semantics when using AIX-compatible ABI. Pass floating-point 26180 arguments to prototyped functions beyond the register save area 26181 (RSA) on the stack in addition to argument FPRs. Do not assume 26182 that most significant double in 128-bit long double value is 26183 properly rounded when comparing values and converting to double. 26184 Use XL symbol names for long double support routines. 26185 26186 The AIX calling convention was extended but not initially 26187 documented to handle an obscure K&R C case of calling a function 26188 that takes the address of its arguments with fewer arguments than 26189 declared. IBM XL compilers access floating-point arguments that do 26190 not fit in the RSA from the stack when a subroutine is compiled 26191 without optimization. Because always storing floating-point 26192 arguments on the stack is inefficient and rarely needed, this 26193 option is not enabled by default and only is necessary when calling 26194 subroutines compiled by IBM XL compilers without optimization. 26195 26196'-mpe' 26197 Support "IBM RS/6000 SP" "Parallel Environment" (PE). Link an 26198 application written to use message passing with special startup 26199 code to enable the application to run. The system must have PE 26200 installed in the standard location ('/usr/lpp/ppe.poe/'), or the 26201 'specs' file must be overridden with the '-specs=' option to 26202 specify the appropriate directory location. The Parallel 26203 Environment does not support threads, so the '-mpe' option and the 26204 '-pthread' option are incompatible. 26205 26206'-malign-natural' 26207'-malign-power' 26208 On AIX, 32-bit Darwin, and 64-bit PowerPC GNU/Linux, the option 26209 '-malign-natural' overrides the ABI-defined alignment of larger 26210 types, such as floating-point doubles, on their natural size-based 26211 boundary. The option '-malign-power' instructs GCC to follow the 26212 ABI-specified alignment rules. GCC defaults to the standard 26213 alignment defined in the ABI. 26214 26215 On 64-bit Darwin, natural alignment is the default, and 26216 '-malign-power' is not supported. 26217 26218'-msoft-float' 26219'-mhard-float' 26220 Generate code that does not use (uses) the floating-point register 26221 set. Software floating-point emulation is provided if you use the 26222 '-msoft-float' option, and pass the option to GCC when linking. 26223 26224'-mmultiple' 26225'-mno-multiple' 26226 Generate code that uses (does not use) the load multiple word 26227 instructions and the store multiple word instructions. These 26228 instructions are generated by default on POWER systems, and not 26229 generated on PowerPC systems. Do not use '-mmultiple' on 26230 little-endian PowerPC systems, since those instructions do not work 26231 when the processor is in little-endian mode. The exceptions are 26232 PPC740 and PPC750 which permit these instructions in little-endian 26233 mode. 26234 26235'-mupdate' 26236'-mno-update' 26237 Generate code that uses (does not use) the load or store 26238 instructions that update the base register to the address of the 26239 calculated memory location. These instructions are generated by 26240 default. If you use '-mno-update', there is a small window between 26241 the time that the stack pointer is updated and the address of the 26242 previous frame is stored, which means code that walks the stack 26243 frame across interrupts or signals may get corrupted data. 26244 26245'-mavoid-indexed-addresses' 26246'-mno-avoid-indexed-addresses' 26247 Generate code that tries to avoid (not avoid) the use of indexed 26248 load or store instructions. These instructions can incur a 26249 performance penalty on Power6 processors in certain situations, 26250 such as when stepping through large arrays that cross a 16M 26251 boundary. This option is enabled by default when targeting Power6 26252 and disabled otherwise. 26253 26254'-mfused-madd' 26255'-mno-fused-madd' 26256 Generate code that uses (does not use) the floating-point multiply 26257 and accumulate instructions. These instructions are generated by 26258 default if hardware floating point is used. The machine-dependent 26259 '-mfused-madd' option is now mapped to the machine-independent 26260 '-ffp-contract=fast' option, and '-mno-fused-madd' is mapped to 26261 '-ffp-contract=off'. 26262 26263'-mmulhw' 26264'-mno-mulhw' 26265 Generate code that uses (does not use) the half-word multiply and 26266 multiply-accumulate instructions on the IBM 405, 440, 464 and 476 26267 processors. These instructions are generated by default when 26268 targeting those processors. 26269 26270'-mdlmzb' 26271'-mno-dlmzb' 26272 Generate code that uses (does not use) the string-search 'dlmzb' 26273 instruction on the IBM 405, 440, 464 and 476 processors. This 26274 instruction is generated by default when targeting those 26275 processors. 26276 26277'-mno-bit-align' 26278'-mbit-align' 26279 On System V.4 and embedded PowerPC systems do not (do) force 26280 structures and unions that contain bit-fields to be aligned to the 26281 base type of the bit-field. 26282 26283 For example, by default a structure containing nothing but 8 26284 'unsigned' bit-fields of length 1 is aligned to a 4-byte boundary 26285 and has a size of 4 bytes. By using '-mno-bit-align', the 26286 structure is aligned to a 1-byte boundary and is 1 byte in size. 26287 26288'-mno-strict-align' 26289'-mstrict-align' 26290 On System V.4 and embedded PowerPC systems do not (do) assume that 26291 unaligned memory references are handled by the system. 26292 26293'-mrelocatable' 26294'-mno-relocatable' 26295 Generate code that allows (does not allow) a static executable to 26296 be relocated to a different address at run time. A simple embedded 26297 PowerPC system loader should relocate the entire contents of 26298 '.got2' and 4-byte locations listed in the '.fixup' section, a 26299 table of 32-bit addresses generated by this option. For this to 26300 work, all objects linked together must be compiled with 26301 '-mrelocatable' or '-mrelocatable-lib'. '-mrelocatable' code 26302 aligns the stack to an 8-byte boundary. 26303 26304'-mrelocatable-lib' 26305'-mno-relocatable-lib' 26306 Like '-mrelocatable', '-mrelocatable-lib' generates a '.fixup' 26307 section to allow static executables to be relocated at run time, 26308 but '-mrelocatable-lib' does not use the smaller stack alignment of 26309 '-mrelocatable'. Objects compiled with '-mrelocatable-lib' may be 26310 linked with objects compiled with any combination of the 26311 '-mrelocatable' options. 26312 26313'-mno-toc' 26314'-mtoc' 26315 On System V.4 and embedded PowerPC systems do not (do) assume that 26316 register 2 contains a pointer to a global area pointing to the 26317 addresses used in the program. 26318 26319'-mlittle' 26320'-mlittle-endian' 26321 On System V.4 and embedded PowerPC systems compile code for the 26322 processor in little-endian mode. The '-mlittle-endian' option is 26323 the same as '-mlittle'. 26324 26325'-mbig' 26326'-mbig-endian' 26327 On System V.4 and embedded PowerPC systems compile code for the 26328 processor in big-endian mode. The '-mbig-endian' option is the 26329 same as '-mbig'. 26330 26331'-mdynamic-no-pic' 26332 On Darwin and Mac OS X systems, compile code so that it is not 26333 relocatable, but that its external references are relocatable. The 26334 resulting code is suitable for applications, but not shared 26335 libraries. 26336 26337'-msingle-pic-base' 26338 Treat the register used for PIC addressing as read-only, rather 26339 than loading it in the prologue for each function. The runtime 26340 system is responsible for initializing this register with an 26341 appropriate value before execution begins. 26342 26343'-mprioritize-restricted-insns=PRIORITY' 26344 This option controls the priority that is assigned to dispatch-slot 26345 restricted instructions during the second scheduling pass. The 26346 argument PRIORITY takes the value '0', '1', or '2' to assign no, 26347 highest, or second-highest (respectively) priority to dispatch-slot 26348 restricted instructions. 26349 26350'-msched-costly-dep=DEPENDENCE_TYPE' 26351 This option controls which dependences are considered costly by the 26352 target during instruction scheduling. The argument DEPENDENCE_TYPE 26353 takes one of the following values: 26354 26355 'no' 26356 No dependence is costly. 26357 26358 'all' 26359 All dependences are costly. 26360 26361 'true_store_to_load' 26362 A true dependence from store to load is costly. 26363 26364 'store_to_load' 26365 Any dependence from store to load is costly. 26366 26367 NUMBER 26368 Any dependence for which the latency is greater than or equal 26369 to NUMBER is costly. 26370 26371'-minsert-sched-nops=SCHEME' 26372 This option controls which NOP insertion scheme is used during the 26373 second scheduling pass. The argument SCHEME takes one of the 26374 following values: 26375 26376 'no' 26377 Don't insert NOPs. 26378 26379 'pad' 26380 Pad with NOPs any dispatch group that has vacant issue slots, 26381 according to the scheduler's grouping. 26382 26383 'regroup_exact' 26384 Insert NOPs to force costly dependent insns into separate 26385 groups. Insert exactly as many NOPs as needed to force an 26386 insn to a new group, according to the estimated processor 26387 grouping. 26388 26389 NUMBER 26390 Insert NOPs to force costly dependent insns into separate 26391 groups. Insert NUMBER NOPs to force an insn to a new group. 26392 26393'-mcall-sysv' 26394 On System V.4 and embedded PowerPC systems compile code using 26395 calling conventions that adhere to the March 1995 draft of the 26396 System V Application Binary Interface, PowerPC processor 26397 supplement. This is the default unless you configured GCC using 26398 'powerpc-*-eabiaix'. 26399 26400'-mcall-sysv-eabi' 26401'-mcall-eabi' 26402 Specify both '-mcall-sysv' and '-meabi' options. 26403 26404'-mcall-sysv-noeabi' 26405 Specify both '-mcall-sysv' and '-mno-eabi' options. 26406 26407'-mcall-aixdesc' 26408 On System V.4 and embedded PowerPC systems compile code for the AIX 26409 operating system. 26410 26411'-mcall-linux' 26412 On System V.4 and embedded PowerPC systems compile code for the 26413 Linux-based GNU system. 26414 26415'-mcall-freebsd' 26416 On System V.4 and embedded PowerPC systems compile code for the 26417 FreeBSD operating system. 26418 26419'-mcall-netbsd' 26420 On System V.4 and embedded PowerPC systems compile code for the 26421 NetBSD operating system. 26422 26423'-mcall-openbsd' 26424 On System V.4 and embedded PowerPC systems compile code for the 26425 OpenBSD operating system. 26426 26427'-mtraceback=TRACEBACK_TYPE' 26428 Select the type of traceback table. Valid values for 26429 TRACEBACK_TYPE are 'full', 'part', and 'no'. 26430 26431'-maix-struct-return' 26432 Return all structures in memory (as specified by the AIX ABI). 26433 26434'-msvr4-struct-return' 26435 Return structures smaller than 8 bytes in registers (as specified 26436 by the SVR4 ABI). 26437 26438'-mabi=ABI-TYPE' 26439 Extend the current ABI with a particular extension, or remove such 26440 extension. Valid values are: 'altivec', 'no-altivec', 26441 'ibmlongdouble', 'ieeelongdouble', 'elfv1', 'elfv2', and for AIX: 26442 'vec-extabi', 'vec-default'. 26443 26444'-mabi=ibmlongdouble' 26445 Change the current ABI to use IBM extended-precision long double. 26446 This is not likely to work if your system defaults to using IEEE 26447 extended-precision long double. If you change the long double type 26448 from IEEE extended-precision, the compiler will issue a warning 26449 unless you use the '-Wno-psabi' option. Requires 26450 '-mlong-double-128' to be enabled. 26451 26452'-mabi=ieeelongdouble' 26453 Change the current ABI to use IEEE extended-precision long double. 26454 This is not likely to work if your system defaults to using IBM 26455 extended-precision long double. If you change the long double type 26456 from IBM extended-precision, the compiler will issue a warning 26457 unless you use the '-Wno-psabi' option. Requires 26458 '-mlong-double-128' to be enabled. 26459 26460'-mabi=elfv1' 26461 Change the current ABI to use the ELFv1 ABI. This is the default 26462 ABI for big-endian PowerPC 64-bit Linux. Overriding the default 26463 ABI requires special system support and is likely to fail in 26464 spectacular ways. 26465 26466'-mabi=elfv2' 26467 Change the current ABI to use the ELFv2 ABI. This is the default 26468 ABI for little-endian PowerPC 64-bit Linux. Overriding the default 26469 ABI requires special system support and is likely to fail in 26470 spectacular ways. 26471 26472'-mgnu-attribute' 26473'-mno-gnu-attribute' 26474 Emit .gnu_attribute assembly directives to set tag/value pairs in a 26475 .gnu.attributes section that specify ABI variations in function 26476 parameters or return values. 26477 26478'-mprototype' 26479'-mno-prototype' 26480 On System V.4 and embedded PowerPC systems assume that all calls to 26481 variable argument functions are properly prototyped. Otherwise, 26482 the compiler must insert an instruction before every non-prototyped 26483 call to set or clear bit 6 of the condition code register ('CR') to 26484 indicate whether floating-point values are passed in the 26485 floating-point registers in case the function takes variable 26486 arguments. With '-mprototype', only calls to prototyped variable 26487 argument functions set or clear the bit. 26488 26489'-msim' 26490 On embedded PowerPC systems, assume that the startup module is 26491 called 'sim-crt0.o' and that the standard C libraries are 26492 'libsim.a' and 'libc.a'. This is the default for 26493 'powerpc-*-eabisim' configurations. 26494 26495'-mmvme' 26496 On embedded PowerPC systems, assume that the startup module is 26497 called 'crt0.o' and the standard C libraries are 'libmvme.a' and 26498 'libc.a'. 26499 26500'-mads' 26501 On embedded PowerPC systems, assume that the startup module is 26502 called 'crt0.o' and the standard C libraries are 'libads.a' and 26503 'libc.a'. 26504 26505'-myellowknife' 26506 On embedded PowerPC systems, assume that the startup module is 26507 called 'crt0.o' and the standard C libraries are 'libyk.a' and 26508 'libc.a'. 26509 26510'-mvxworks' 26511 On System V.4 and embedded PowerPC systems, specify that you are 26512 compiling for a VxWorks system. 26513 26514'-memb' 26515 On embedded PowerPC systems, set the 'PPC_EMB' bit in the ELF flags 26516 header to indicate that 'eabi' extended relocations are used. 26517 26518'-meabi' 26519'-mno-eabi' 26520 On System V.4 and embedded PowerPC systems do (do not) adhere to 26521 the Embedded Applications Binary Interface (EABI), which is a set 26522 of modifications to the System V.4 specifications. Selecting 26523 '-meabi' means that the stack is aligned to an 8-byte boundary, a 26524 function '__eabi' is called from 'main' to set up the EABI 26525 environment, and the '-msdata' option can use both 'r2' and 'r13' 26526 to point to two separate small data areas. Selecting '-mno-eabi' 26527 means that the stack is aligned to a 16-byte boundary, no EABI 26528 initialization function is called from 'main', and the '-msdata' 26529 option only uses 'r13' to point to a single small data area. The 26530 '-meabi' option is on by default if you configured GCC using one of 26531 the 'powerpc*-*-eabi*' options. 26532 26533'-msdata=eabi' 26534 On System V.4 and embedded PowerPC systems, put small initialized 26535 'const' global and static data in the '.sdata2' section, which is 26536 pointed to by register 'r2'. Put small initialized non-'const' 26537 global and static data in the '.sdata' section, which is pointed to 26538 by register 'r13'. Put small uninitialized global and static data 26539 in the '.sbss' section, which is adjacent to the '.sdata' section. 26540 The '-msdata=eabi' option is incompatible with the '-mrelocatable' 26541 option. The '-msdata=eabi' option also sets the '-memb' option. 26542 26543'-msdata=sysv' 26544 On System V.4 and embedded PowerPC systems, put small global and 26545 static data in the '.sdata' section, which is pointed to by 26546 register 'r13'. Put small uninitialized global and static data in 26547 the '.sbss' section, which is adjacent to the '.sdata' section. 26548 The '-msdata=sysv' option is incompatible with the '-mrelocatable' 26549 option. 26550 26551'-msdata=default' 26552'-msdata' 26553 On System V.4 and embedded PowerPC systems, if '-meabi' is used, 26554 compile code the same as '-msdata=eabi', otherwise compile code the 26555 same as '-msdata=sysv'. 26556 26557'-msdata=data' 26558 On System V.4 and embedded PowerPC systems, put small global data 26559 in the '.sdata' section. Put small uninitialized global data in 26560 the '.sbss' section. Do not use register 'r13' to address small 26561 data however. This is the default behavior unless other '-msdata' 26562 options are used. 26563 26564'-msdata=none' 26565'-mno-sdata' 26566 On embedded PowerPC systems, put all initialized global and static 26567 data in the '.data' section, and all uninitialized data in the 26568 '.bss' section. 26569 26570'-mreadonly-in-sdata' 26571 Put read-only objects in the '.sdata' section as well. This is the 26572 default. 26573 26574'-mblock-move-inline-limit=NUM' 26575 Inline all block moves (such as calls to 'memcpy' or structure 26576 copies) less than or equal to NUM bytes. The minimum value for NUM 26577 is 32 bytes on 32-bit targets and 64 bytes on 64-bit targets. The 26578 default value is target-specific. 26579 26580'-mblock-compare-inline-limit=NUM' 26581 Generate non-looping inline code for all block compares (such as 26582 calls to 'memcmp' or structure compares) less than or equal to NUM 26583 bytes. If NUM is 0, all inline expansion (non-loop and loop) of 26584 block compare is disabled. The default value is target-specific. 26585 26586'-mblock-compare-inline-loop-limit=NUM' 26587 Generate an inline expansion using loop code for all block compares 26588 that are less than or equal to NUM bytes, but greater than the 26589 limit for non-loop inline block compare expansion. If the block 26590 length is not constant, at most NUM bytes will be compared before 26591 'memcmp' is called to compare the remainder of the block. The 26592 default value is target-specific. 26593 26594'-mstring-compare-inline-limit=NUM' 26595 Compare at most NUM string bytes with inline code. If the 26596 difference or end of string is not found at the end of the inline 26597 compare a call to 'strcmp' or 'strncmp' will take care of the rest 26598 of the comparison. The default is 64 bytes. 26599 26600'-G NUM' 26601 On embedded PowerPC systems, put global and static items less than 26602 or equal to NUM bytes into the small data or BSS sections instead 26603 of the normal data or BSS section. By default, NUM is 8. The '-G 26604 NUM' switch is also passed to the linker. All modules should be 26605 compiled with the same '-G NUM' value. 26606 26607'-mregnames' 26608'-mno-regnames' 26609 On System V.4 and embedded PowerPC systems do (do not) emit 26610 register names in the assembly language output using symbolic 26611 forms. 26612 26613'-mlongcall' 26614'-mno-longcall' 26615 By default assume that all calls are far away so that a longer and 26616 more expensive calling sequence is required. This is required for 26617 calls farther than 32 megabytes (33,554,432 bytes) from the current 26618 location. A short call is generated if the compiler knows the call 26619 cannot be that far away. This setting can be overridden by the 26620 'shortcall' function attribute, or by '#pragma longcall(0)'. 26621 26622 Some linkers are capable of detecting out-of-range calls and 26623 generating glue code on the fly. On these systems, long calls are 26624 unnecessary and generate slower code. As of this writing, the AIX 26625 linker can do this, as can the GNU linker for PowerPC/64. It is 26626 planned to add this feature to the GNU linker for 32-bit PowerPC 26627 systems as well. 26628 26629 On PowerPC64 ELFv2 and 32-bit PowerPC systems with newer GNU 26630 linkers, GCC can generate long calls using an inline PLT call 26631 sequence (see '-mpltseq'). PowerPC with '-mbss-plt' and PowerPC64 26632 ELFv1 (big-endian) do not support inline PLT calls. 26633 26634 On Darwin/PPC systems, '#pragma longcall' generates 'jbsr callee, 26635 L42', plus a "branch island" (glue code). The two target addresses 26636 represent the callee and the branch island. The Darwin/PPC linker 26637 prefers the first address and generates a 'bl callee' if the PPC 26638 'bl' instruction reaches the callee directly; otherwise, the linker 26639 generates 'bl L42' to call the branch island. The branch island is 26640 appended to the body of the calling function; it computes the full 26641 32-bit address of the callee and jumps to it. 26642 26643 On Mach-O (Darwin) systems, this option directs the compiler emit 26644 to the glue for every direct call, and the Darwin linker decides 26645 whether to use or discard it. 26646 26647 In the future, GCC may ignore all longcall specifications when the 26648 linker is known to generate glue. 26649 26650'-mpltseq' 26651'-mno-pltseq' 26652 Implement (do not implement) -fno-plt and long calls using an 26653 inline PLT call sequence that supports lazy linking and long calls 26654 to functions in dlopen'd shared libraries. Inline PLT calls are 26655 only supported on PowerPC64 ELFv2 and 32-bit PowerPC systems with 26656 newer GNU linkers, and are enabled by default if the support is 26657 detected when configuring GCC, and, in the case of 32-bit PowerPC, 26658 if GCC is configured with '--enable-secureplt'. '-mpltseq' code 26659 and '-mbss-plt' 32-bit PowerPC relocatable objects may not be 26660 linked together. 26661 26662'-mtls-markers' 26663'-mno-tls-markers' 26664 Mark (do not mark) calls to '__tls_get_addr' with a relocation 26665 specifying the function argument. The relocation allows the linker 26666 to reliably associate function call with argument setup 26667 instructions for TLS optimization, which in turn allows GCC to 26668 better schedule the sequence. 26669 26670'-mrecip' 26671'-mno-recip' 26672 This option enables use of the reciprocal estimate and reciprocal 26673 square root estimate instructions with additional Newton-Raphson 26674 steps to increase precision instead of doing a divide or square 26675 root and divide for floating-point arguments. You should use the 26676 '-ffast-math' option when using '-mrecip' (or at least 26677 '-funsafe-math-optimizations', '-ffinite-math-only', 26678 '-freciprocal-math' and '-fno-trapping-math'). Note that while the 26679 throughput of the sequence is generally higher than the throughput 26680 of the non-reciprocal instruction, the precision of the sequence 26681 can be decreased by up to 2 ulp (i.e. the inverse of 1.0 equals 26682 0.99999994) for reciprocal square roots. 26683 26684'-mrecip=OPT' 26685 This option controls which reciprocal estimate instructions may be 26686 used. OPT is a comma-separated list of options, which may be 26687 preceded by a '!' to invert the option: 26688 26689 'all' 26690 Enable all estimate instructions. 26691 26692 'default' 26693 Enable the default instructions, equivalent to '-mrecip'. 26694 26695 'none' 26696 Disable all estimate instructions, equivalent to '-mno-recip'. 26697 26698 'div' 26699 Enable the reciprocal approximation instructions for both 26700 single and double precision. 26701 26702 'divf' 26703 Enable the single-precision reciprocal approximation 26704 instructions. 26705 26706 'divd' 26707 Enable the double-precision reciprocal approximation 26708 instructions. 26709 26710 'rsqrt' 26711 Enable the reciprocal square root approximation instructions 26712 for both single and double precision. 26713 26714 'rsqrtf' 26715 Enable the single-precision reciprocal square root 26716 approximation instructions. 26717 26718 'rsqrtd' 26719 Enable the double-precision reciprocal square root 26720 approximation instructions. 26721 26722 So, for example, '-mrecip=all,!rsqrtd' enables all of the 26723 reciprocal estimate instructions, except for the 'FRSQRTE', 26724 'XSRSQRTEDP', and 'XVRSQRTEDP' instructions which handle the 26725 double-precision reciprocal square root calculations. 26726 26727'-mrecip-precision' 26728'-mno-recip-precision' 26729 Assume (do not assume) that the reciprocal estimate instructions 26730 provide higher-precision estimates than is mandated by the PowerPC 26731 ABI. Selecting '-mcpu=power6', '-mcpu=power7' or '-mcpu=power8' 26732 automatically selects '-mrecip-precision'. The double-precision 26733 square root estimate instructions are not generated by default on 26734 low-precision machines, since they do not provide an estimate that 26735 converges after three steps. 26736 26737'-mveclibabi=TYPE' 26738 Specifies the ABI type to use for vectorizing intrinsics using an 26739 external library. The only type supported at present is 'mass', 26740 which specifies to use IBM's Mathematical Acceleration Subsystem 26741 (MASS) libraries for vectorizing intrinsics using external 26742 libraries. GCC currently emits calls to 'acosd2', 'acosf4', 26743 'acoshd2', 'acoshf4', 'asind2', 'asinf4', 'asinhd2', 'asinhf4', 26744 'atan2d2', 'atan2f4', 'atand2', 'atanf4', 'atanhd2', 'atanhf4', 26745 'cbrtd2', 'cbrtf4', 'cosd2', 'cosf4', 'coshd2', 'coshf4', 'erfcd2', 26746 'erfcf4', 'erfd2', 'erff4', 'exp2d2', 'exp2f4', 'expd2', 'expf4', 26747 'expm1d2', 'expm1f4', 'hypotd2', 'hypotf4', 'lgammad2', 'lgammaf4', 26748 'log10d2', 'log10f4', 'log1pd2', 'log1pf4', 'log2d2', 'log2f4', 26749 'logd2', 'logf4', 'powd2', 'powf4', 'sind2', 'sinf4', 'sinhd2', 26750 'sinhf4', 'sqrtd2', 'sqrtf4', 'tand2', 'tanf4', 'tanhd2', and 26751 'tanhf4' when generating code for power7. Both '-ftree-vectorize' 26752 and '-funsafe-math-optimizations' must also be enabled. The MASS 26753 libraries must be specified at link time. 26754 26755'-mfriz' 26756'-mno-friz' 26757 Generate (do not generate) the 'friz' instruction when the 26758 '-funsafe-math-optimizations' option is used to optimize rounding 26759 of floating-point values to 64-bit integer and back to floating 26760 point. The 'friz' instruction does not return the same value if 26761 the floating-point number is too large to fit in an integer. 26762 26763'-mpointers-to-nested-functions' 26764'-mno-pointers-to-nested-functions' 26765 Generate (do not generate) code to load up the static chain 26766 register ('r11') when calling through a pointer on AIX and 64-bit 26767 Linux systems where a function pointer points to a 3-word 26768 descriptor giving the function address, TOC value to be loaded in 26769 register 'r2', and static chain value to be loaded in register 26770 'r11'. The '-mpointers-to-nested-functions' is on by default. You 26771 cannot call through pointers to nested functions or pointers to 26772 functions compiled in other languages that use the static chain if 26773 you use '-mno-pointers-to-nested-functions'. 26774 26775'-msave-toc-indirect' 26776'-mno-save-toc-indirect' 26777 Generate (do not generate) code to save the TOC value in the 26778 reserved stack location in the function prologue if the function 26779 calls through a pointer on AIX and 64-bit Linux systems. If the 26780 TOC value is not saved in the prologue, it is saved just before the 26781 call through the pointer. The '-mno-save-toc-indirect' option is 26782 the default. 26783 26784'-mcompat-align-parm' 26785'-mno-compat-align-parm' 26786 Generate (do not generate) code to pass structure parameters with a 26787 maximum alignment of 64 bits, for compatibility with older versions 26788 of GCC. 26789 26790 Older versions of GCC (prior to 4.9.0) incorrectly did not align a 26791 structure parameter on a 128-bit boundary when that structure 26792 contained a member requiring 128-bit alignment. This is corrected 26793 in more recent versions of GCC. This option may be used to generate 26794 code that is compatible with functions compiled with older versions 26795 of GCC. 26796 26797 The '-mno-compat-align-parm' option is the default. 26798 26799'-mstack-protector-guard=GUARD' 26800'-mstack-protector-guard-reg=REG' 26801'-mstack-protector-guard-offset=OFFSET' 26802'-mstack-protector-guard-symbol=SYMBOL' 26803 Generate stack protection code using canary at GUARD. Supported 26804 locations are 'global' for global canary or 'tls' for per-thread 26805 canary in the TLS block (the default with GNU libc version 2.4 or 26806 later). 26807 26808 With the latter choice the options 26809 '-mstack-protector-guard-reg=REG' and 26810 '-mstack-protector-guard-offset=OFFSET' furthermore specify which 26811 register to use as base register for reading the canary, and from 26812 what offset from that base register. The default for those is as 26813 specified in the relevant ABI. 26814 '-mstack-protector-guard-symbol=SYMBOL' overrides the offset with a 26815 symbol reference to a canary in the TLS block. 26816 26817'-mpcrel' 26818'-mno-pcrel' 26819 Generate (do not generate) pc-relative addressing. The '-mpcrel' 26820 option requires that the medium code model ('-mcmodel=medium') and 26821 prefixed addressing ('-mprefixed') options are enabled. 26822 26823'-mprefixed' 26824'-mno-prefixed' 26825 Generate (do not generate) addressing modes using prefixed load and 26826 store instructions. The '-mprefixed' option requires that the 26827 option '-mcpu=power10' (or later) is enabled. 26828 26829'-mmma' 26830'-mno-mma' 26831 Generate (do not generate) the MMA instructions. The '-mma' option 26832 requires that the option '-mcpu=power10' (or later) is enabled. 26833 26834'-mrop-protect' 26835'-mno-rop-protect' 26836 Generate (do not generate) ROP protection instructions when the 26837 target processor supports them. Currently this option disables the 26838 shrink-wrap optimization ('-fshrink-wrap'). 26839 26840'-mprivileged' 26841'-mno-privileged' 26842 Generate (do not generate) code that will run in privileged state. 26843 26844'-mblock-ops-unaligned-vsx' 26845'-mno-block-ops-unaligned-vsx' 26846 Generate (do not generate) unaligned vsx loads and stores for 26847 inline expansion of 'memcpy' and 'memmove'. 26848 26849 26850File: gcc.info, Node: RX Options, Next: S/390 and zSeries Options, Prev: RS/6000 and PowerPC Options, Up: Submodel Options 26851 268523.19.46 RX Options 26853------------------ 26854 26855These command-line options are defined for RX targets: 26856 26857'-m64bit-doubles' 26858'-m32bit-doubles' 26859 Make the 'double' data type be 64 bits ('-m64bit-doubles') or 32 26860 bits ('-m32bit-doubles') in size. The default is 26861 '-m32bit-doubles'. _Note_ RX floating-point hardware only works on 26862 32-bit values, which is why the default is '-m32bit-doubles'. 26863 26864'-fpu' 26865'-nofpu' 26866 Enables ('-fpu') or disables ('-nofpu') the use of RX 26867 floating-point hardware. The default is enabled for the RX600 26868 series and disabled for the RX200 series. 26869 26870 Floating-point instructions are only generated for 32-bit 26871 floating-point values, however, so the FPU hardware is not used for 26872 doubles if the '-m64bit-doubles' option is used. 26873 26874 _Note_ If the '-fpu' option is enabled then 26875 '-funsafe-math-optimizations' is also enabled automatically. This 26876 is because the RX FPU instructions are themselves unsafe. 26877 26878'-mcpu=NAME' 26879 Selects the type of RX CPU to be targeted. Currently three types 26880 are supported, the generic 'RX600' and 'RX200' series hardware and 26881 the specific 'RX610' CPU. The default is 'RX600'. 26882 26883 The only difference between 'RX600' and 'RX610' is that the 'RX610' 26884 does not support the 'MVTIPL' instruction. 26885 26886 The 'RX200' series does not have a hardware floating-point unit and 26887 so '-nofpu' is enabled by default when this type is selected. 26888 26889'-mbig-endian-data' 26890'-mlittle-endian-data' 26891 Store data (but not code) in the big-endian format. The default is 26892 '-mlittle-endian-data', i.e. to store data in the little-endian 26893 format. 26894 26895'-msmall-data-limit=N' 26896 Specifies the maximum size in bytes of global and static variables 26897 which can be placed into the small data area. Using the small data 26898 area can lead to smaller and faster code, but the size of area is 26899 limited and it is up to the programmer to ensure that the area does 26900 not overflow. Also when the small data area is used one of the 26901 RX's registers (usually 'r13') is reserved for use pointing to this 26902 area, so it is no longer available for use by the compiler. This 26903 could result in slower and/or larger code if variables are pushed 26904 onto the stack instead of being held in this register. 26905 26906 Note, common variables (variables that have not been initialized) 26907 and constants are not placed into the small data area as they are 26908 assigned to other sections in the output executable. 26909 26910 The default value is zero, which disables this feature. Note, this 26911 feature is not enabled by default with higher optimization levels 26912 ('-O2' etc) because of the potentially detrimental effects of 26913 reserving a register. It is up to the programmer to experiment and 26914 discover whether this feature is of benefit to their program. See 26915 the description of the '-mpid' option for a description of how the 26916 actual register to hold the small data area pointer is chosen. 26917 26918'-msim' 26919'-mno-sim' 26920 Use the simulator runtime. The default is to use the libgloss 26921 board-specific runtime. 26922 26923'-mas100-syntax' 26924'-mno-as100-syntax' 26925 When generating assembler output use a syntax that is compatible 26926 with Renesas's AS100 assembler. This syntax can also be handled by 26927 the GAS assembler, but it has some restrictions so it is not 26928 generated by default. 26929 26930'-mmax-constant-size=N' 26931 Specifies the maximum size, in bytes, of a constant that can be 26932 used as an operand in a RX instruction. Although the RX 26933 instruction set does allow constants of up to 4 bytes in length to 26934 be used in instructions, a longer value equates to a longer 26935 instruction. Thus in some circumstances it can be beneficial to 26936 restrict the size of constants that are used in instructions. 26937 Constants that are too big are instead placed into a constant pool 26938 and referenced via register indirection. 26939 26940 The value N can be between 0 and 4. A value of 0 (the default) or 26941 4 means that constants of any size are allowed. 26942 26943'-mrelax' 26944 Enable linker relaxation. Linker relaxation is a process whereby 26945 the linker attempts to reduce the size of a program by finding 26946 shorter versions of various instructions. Disabled by default. 26947 26948'-mint-register=N' 26949 Specify the number of registers to reserve for fast interrupt 26950 handler functions. The value N can be between 0 and 4. A value of 26951 1 means that register 'r13' is reserved for the exclusive use of 26952 fast interrupt handlers. A value of 2 reserves 'r13' and 'r12'. A 26953 value of 3 reserves 'r13', 'r12' and 'r11', and a value of 4 26954 reserves 'r13' through 'r10'. A value of 0, the default, does not 26955 reserve any registers. 26956 26957'-msave-acc-in-interrupts' 26958 Specifies that interrupt handler functions should preserve the 26959 accumulator register. This is only necessary if normal code might 26960 use the accumulator register, for example because it performs 26961 64-bit multiplications. The default is to ignore the accumulator 26962 as this makes the interrupt handlers faster. 26963 26964'-mpid' 26965'-mno-pid' 26966 Enables the generation of position independent data. When enabled 26967 any access to constant data is done via an offset from a base 26968 address held in a register. This allows the location of constant 26969 data to be determined at run time without requiring the executable 26970 to be relocated, which is a benefit to embedded applications with 26971 tight memory constraints. Data that can be modified is not 26972 affected by this option. 26973 26974 Note, using this feature reserves a register, usually 'r13', for 26975 the constant data base address. This can result in slower and/or 26976 larger code, especially in complicated functions. 26977 26978 The actual register chosen to hold the constant data base address 26979 depends upon whether the '-msmall-data-limit' and/or the 26980 '-mint-register' command-line options are enabled. Starting with 26981 register 'r13' and proceeding downwards, registers are allocated 26982 first to satisfy the requirements of '-mint-register', then '-mpid' 26983 and finally '-msmall-data-limit'. Thus it is possible for the 26984 small data area register to be 'r8' if both '-mint-register=4' and 26985 '-mpid' are specified on the command line. 26986 26987 By default this feature is not enabled. The default can be 26988 restored via the '-mno-pid' command-line option. 26989 26990'-mno-warn-multiple-fast-interrupts' 26991'-mwarn-multiple-fast-interrupts' 26992 Prevents GCC from issuing a warning message if it finds more than 26993 one fast interrupt handler when it is compiling a file. The 26994 default is to issue a warning for each extra fast interrupt handler 26995 found, as the RX only supports one such interrupt. 26996 26997'-mallow-string-insns' 26998'-mno-allow-string-insns' 26999 Enables or disables the use of the string manipulation instructions 27000 'SMOVF', 'SCMPU', 'SMOVB', 'SMOVU', 'SUNTIL' 'SWHILE' and also the 27001 'RMPA' instruction. These instructions may prefetch data, which is 27002 not safe to do if accessing an I/O register. (See section 12.2.7 27003 of the RX62N Group User's Manual for more information). 27004 27005 The default is to allow these instructions, but it is not possible 27006 for GCC to reliably detect all circumstances where a string 27007 instruction might be used to access an I/O register, so their use 27008 cannot be disabled automatically. Instead it is reliant upon the 27009 programmer to use the '-mno-allow-string-insns' option if their 27010 program accesses I/O space. 27011 27012 When the instructions are enabled GCC defines the C preprocessor 27013 symbol '__RX_ALLOW_STRING_INSNS__', otherwise it defines the symbol 27014 '__RX_DISALLOW_STRING_INSNS__'. 27015 27016'-mjsr' 27017'-mno-jsr' 27018 Use only (or not only) 'JSR' instructions to access functions. 27019 This option can be used when code size exceeds the range of 'BSR' 27020 instructions. Note that '-mno-jsr' does not mean to not use 'JSR' 27021 but instead means that any type of branch may be used. 27022 27023 _Note:_ The generic GCC command-line option '-ffixed-REG' has special 27024significance to the RX port when used with the 'interrupt' function 27025attribute. This attribute indicates a function intended to process fast 27026interrupts. GCC ensures that it only uses the registers 'r10', 'r11', 27027'r12' and/or 'r13' and only provided that the normal use of the 27028corresponding registers have been restricted via the '-ffixed-REG' or 27029'-mint-register' command-line options. 27030 27031 27032File: gcc.info, Node: S/390 and zSeries Options, Next: Score Options, Prev: RX Options, Up: Submodel Options 27033 270343.19.47 S/390 and zSeries Options 27035--------------------------------- 27036 27037These are the '-m' options defined for the S/390 and zSeries 27038architecture. 27039 27040'-mhard-float' 27041'-msoft-float' 27042 Use (do not use) the hardware floating-point instructions and 27043 registers for floating-point operations. When '-msoft-float' is 27044 specified, functions in 'libgcc.a' are used to perform 27045 floating-point operations. When '-mhard-float' is specified, the 27046 compiler generates IEEE floating-point instructions. This is the 27047 default. 27048 27049'-mhard-dfp' 27050'-mno-hard-dfp' 27051 Use (do not use) the hardware decimal-floating-point instructions 27052 for decimal-floating-point operations. When '-mno-hard-dfp' is 27053 specified, functions in 'libgcc.a' are used to perform 27054 decimal-floating-point operations. When '-mhard-dfp' is specified, 27055 the compiler generates decimal-floating-point hardware 27056 instructions. This is the default for '-march=z9-ec' or higher. 27057 27058'-mlong-double-64' 27059'-mlong-double-128' 27060 These switches control the size of 'long double' type. A size of 27061 64 bits makes the 'long double' type equivalent to the 'double' 27062 type. This is the default. 27063 27064'-mbackchain' 27065'-mno-backchain' 27066 Store (do not store) the address of the caller's frame as backchain 27067 pointer into the callee's stack frame. A backchain may be needed 27068 to allow debugging using tools that do not understand DWARF call 27069 frame information. When '-mno-packed-stack' is in effect, the 27070 backchain pointer is stored at the bottom of the stack frame; when 27071 '-mpacked-stack' is in effect, the backchain is placed into the 27072 topmost word of the 96/160 byte register save area. 27073 27074 In general, code compiled with '-mbackchain' is call-compatible 27075 with code compiled with '-mno-backchain'; however, use of the 27076 backchain for debugging purposes usually requires that the whole 27077 binary is built with '-mbackchain'. Note that the combination of 27078 '-mbackchain', '-mpacked-stack' and '-mhard-float' is not 27079 supported. In order to build a linux kernel use '-msoft-float'. 27080 27081 The default is to not maintain the backchain. 27082 27083'-mpacked-stack' 27084'-mno-packed-stack' 27085 Use (do not use) the packed stack layout. When '-mno-packed-stack' 27086 is specified, the compiler uses the all fields of the 96/160 byte 27087 register save area only for their default purpose; unused fields 27088 still take up stack space. When '-mpacked-stack' is specified, 27089 register save slots are densely packed at the top of the register 27090 save area; unused space is reused for other purposes, allowing for 27091 more efficient use of the available stack space. However, when 27092 '-mbackchain' is also in effect, the topmost word of the save area 27093 is always used to store the backchain, and the return address 27094 register is always saved two words below the backchain. 27095 27096 As long as the stack frame backchain is not used, code generated 27097 with '-mpacked-stack' is call-compatible with code generated with 27098 '-mno-packed-stack'. Note that some non-FSF releases of GCC 2.95 27099 for S/390 or zSeries generated code that uses the stack frame 27100 backchain at run time, not just for debugging purposes. Such code 27101 is not call-compatible with code compiled with '-mpacked-stack'. 27102 Also, note that the combination of '-mbackchain', '-mpacked-stack' 27103 and '-mhard-float' is not supported. In order to build a linux 27104 kernel use '-msoft-float'. 27105 27106 The default is to not use the packed stack layout. 27107 27108'-msmall-exec' 27109'-mno-small-exec' 27110 Generate (or do not generate) code using the 'bras' instruction to 27111 do subroutine calls. This only works reliably if the total 27112 executable size does not exceed 64k. The default is to use the 27113 'basr' instruction instead, which does not have this limitation. 27114 27115'-m64' 27116'-m31' 27117 When '-m31' is specified, generate code compliant to the GNU/Linux 27118 for S/390 ABI. When '-m64' is specified, generate code compliant 27119 to the GNU/Linux for zSeries ABI. This allows GCC in particular to 27120 generate 64-bit instructions. For the 's390' targets, the default 27121 is '-m31', while the 's390x' targets default to '-m64'. 27122 27123'-mzarch' 27124'-mesa' 27125 When '-mzarch' is specified, generate code using the instructions 27126 available on z/Architecture. When '-mesa' is specified, generate 27127 code using the instructions available on ESA/390. Note that 27128 '-mesa' is not possible with '-m64'. When generating code 27129 compliant to the GNU/Linux for S/390 ABI, the default is '-mesa'. 27130 When generating code compliant to the GNU/Linux for zSeries ABI, 27131 the default is '-mzarch'. 27132 27133'-mhtm' 27134'-mno-htm' 27135 The '-mhtm' option enables a set of builtins making use of 27136 instructions available with the transactional execution facility 27137 introduced with the IBM zEnterprise EC12 machine generation *note 27138 S/390 System z Built-in Functions::. '-mhtm' is enabled by default 27139 when using '-march=zEC12'. 27140 27141'-mvx' 27142'-mno-vx' 27143 When '-mvx' is specified, generate code using the instructions 27144 available with the vector extension facility introduced with the 27145 IBM z13 machine generation. This option changes the ABI for some 27146 vector type values with regard to alignment and calling 27147 conventions. In case vector type values are being used in an 27148 ABI-relevant context a GAS '.gnu_attribute' command will be added 27149 to mark the resulting binary with the ABI used. '-mvx' is enabled 27150 by default when using '-march=z13'. 27151 27152'-mzvector' 27153'-mno-zvector' 27154 The '-mzvector' option enables vector language extensions and 27155 builtins using instructions available with the vector extension 27156 facility introduced with the IBM z13 machine generation. This 27157 option adds support for 'vector' to be used as a keyword to define 27158 vector type variables and arguments. 'vector' is only available 27159 when GNU extensions are enabled. It will not be expanded when 27160 requesting strict standard compliance e.g. with '-std=c99'. In 27161 addition to the GCC low-level builtins '-mzvector' enables a set of 27162 builtins added for compatibility with AltiVec-style implementations 27163 like Power and Cell. In order to make use of these builtins the 27164 header file 'vecintrin.h' needs to be included. '-mzvector' is 27165 disabled by default. 27166 27167'-mmvcle' 27168'-mno-mvcle' 27169 Generate (or do not generate) code using the 'mvcle' instruction to 27170 perform block moves. When '-mno-mvcle' is specified, use a 'mvc' 27171 loop instead. This is the default unless optimizing for size. 27172 27173'-mdebug' 27174'-mno-debug' 27175 Print (or do not print) additional debug information when 27176 compiling. The default is to not print debug information. 27177 27178'-march=CPU-TYPE' 27179 Generate code that runs on CPU-TYPE, which is the name of a system 27180 representing a certain processor type. Possible values for 27181 CPU-TYPE are 'z900'/'arch5', 'z990'/'arch6', 'z9-109', 27182 'z9-ec'/'arch7', 'z10'/'arch8', 'z196'/'arch9', 'zEC12', 27183 'z13'/'arch11', 'z14'/'arch12', 'z15'/'arch13', 'z16'/'arch14', and 27184 'native'. 27185 27186 The default is '-march=z900'. 27187 27188 Specifying 'native' as cpu type can be used to select the best 27189 architecture option for the host processor. '-march=native' has no 27190 effect if GCC does not recognize the processor. 27191 27192'-mtune=CPU-TYPE' 27193 Tune to CPU-TYPE everything applicable about the generated code, 27194 except for the ABI and the set of available instructions. The list 27195 of CPU-TYPE values is the same as for '-march'. The default is the 27196 value used for '-march'. 27197 27198'-mtpf-trace' 27199'-mno-tpf-trace' 27200 Generate code that adds (does not add) in TPF OS specific branches 27201 to trace routines in the operating system. This option is off by 27202 default, even when compiling for the TPF OS. 27203 27204'-mtpf-trace-skip' 27205'-mno-tpf-trace-skip' 27206 Generate code that changes (does not change) the default branch 27207 targets enabled by '-mtpf-trace' to point to specialized trace 27208 routines providing the ability of selectively skipping function 27209 trace entries for the TPF OS. This option is off by default, even 27210 when compiling for the TPF OS and specifying '-mtpf-trace'. 27211 27212'-mfused-madd' 27213'-mno-fused-madd' 27214 Generate code that uses (does not use) the floating-point multiply 27215 and accumulate instructions. These instructions are generated by 27216 default if hardware floating point is used. 27217 27218'-mwarn-framesize=FRAMESIZE' 27219 Emit a warning if the current function exceeds the given frame 27220 size. Because this is a compile-time check it doesn't need to be a 27221 real problem when the program runs. It is intended to identify 27222 functions that most probably cause a stack overflow. It is useful 27223 to be used in an environment with limited stack size e.g. the linux 27224 kernel. 27225 27226'-mwarn-dynamicstack' 27227 Emit a warning if the function calls 'alloca' or uses 27228 dynamically-sized arrays. This is generally a bad idea with a 27229 limited stack size. 27230 27231'-mstack-guard=STACK-GUARD' 27232'-mstack-size=STACK-SIZE' 27233 If these options are provided the S/390 back end emits additional 27234 instructions in the function prologue that trigger a trap if the 27235 stack size is STACK-GUARD bytes above the STACK-SIZE (remember that 27236 the stack on S/390 grows downward). If the STACK-GUARD option is 27237 omitted the smallest power of 2 larger than the frame size of the 27238 compiled function is chosen. These options are intended to be used 27239 to help debugging stack overflow problems. The additionally 27240 emitted code causes only little overhead and hence can also be used 27241 in production-like systems without greater performance degradation. 27242 The given values have to be exact powers of 2 and STACK-SIZE has to 27243 be greater than STACK-GUARD without exceeding 64k. In order to be 27244 efficient the extra code makes the assumption that the stack starts 27245 at an address aligned to the value given by STACK-SIZE. The 27246 STACK-GUARD option can only be used in conjunction with STACK-SIZE. 27247 27248'-mhotpatch=PRE-HALFWORDS,POST-HALFWORDS' 27249 If the hotpatch option is enabled, a "hot-patching" function 27250 prologue is generated for all functions in the compilation unit. 27251 The funtion label is prepended with the given number of two-byte 27252 NOP instructions (PRE-HALFWORDS, maximum 1000000). After the 27253 label, 2 * POST-HALFWORDS bytes are appended, using the largest NOP 27254 like instructions the architecture allows (maximum 1000000). 27255 27256 If both arguments are zero, hotpatching is disabled. 27257 27258 This option can be overridden for individual functions with the 27259 'hotpatch' attribute. 27260 27261 27262File: gcc.info, Node: Score Options, Next: SH Options, Prev: S/390 and zSeries Options, Up: Submodel Options 27263 272643.19.48 Score Options 27265--------------------- 27266 27267These options are defined for Score implementations: 27268 27269'-meb' 27270 Compile code for big-endian mode. This is the default. 27271 27272'-mel' 27273 Compile code for little-endian mode. 27274 27275'-mnhwloop' 27276 Disable generation of 'bcnz' instructions. 27277 27278'-muls' 27279 Enable generation of unaligned load and store instructions. 27280 27281'-mmac' 27282 Enable the use of multiply-accumulate instructions. Disabled by 27283 default. 27284 27285'-mscore5' 27286 Specify the SCORE5 as the target architecture. 27287 27288'-mscore5u' 27289 Specify the SCORE5U of the target architecture. 27290 27291'-mscore7' 27292 Specify the SCORE7 as the target architecture. This is the 27293 default. 27294 27295'-mscore7d' 27296 Specify the SCORE7D as the target architecture. 27297 27298 27299File: gcc.info, Node: SH Options, Next: Solaris 2 Options, Prev: Score Options, Up: Submodel Options 27300 273013.19.49 SH Options 27302------------------ 27303 27304These '-m' options are defined for the SH implementations: 27305 27306'-m1' 27307 Generate code for the SH1. 27308 27309'-m2' 27310 Generate code for the SH2. 27311 27312'-m2e' 27313 Generate code for the SH2e. 27314 27315'-m2a-nofpu' 27316 Generate code for the SH2a without FPU, or for a SH2a-FPU in such a 27317 way that the floating-point unit is not used. 27318 27319'-m2a-single-only' 27320 Generate code for the SH2a-FPU, in such a way that no 27321 double-precision floating-point operations are used. 27322 27323'-m2a-single' 27324 Generate code for the SH2a-FPU assuming the floating-point unit is 27325 in single-precision mode by default. 27326 27327'-m2a' 27328 Generate code for the SH2a-FPU assuming the floating-point unit is 27329 in double-precision mode by default. 27330 27331'-m3' 27332 Generate code for the SH3. 27333 27334'-m3e' 27335 Generate code for the SH3e. 27336 27337'-m4-nofpu' 27338 Generate code for the SH4 without a floating-point unit. 27339 27340'-m4-single-only' 27341 Generate code for the SH4 with a floating-point unit that only 27342 supports single-precision arithmetic. 27343 27344'-m4-single' 27345 Generate code for the SH4 assuming the floating-point unit is in 27346 single-precision mode by default. 27347 27348'-m4' 27349 Generate code for the SH4. 27350 27351'-m4-100' 27352 Generate code for SH4-100. 27353 27354'-m4-100-nofpu' 27355 Generate code for SH4-100 in such a way that the floating-point 27356 unit is not used. 27357 27358'-m4-100-single' 27359 Generate code for SH4-100 assuming the floating-point unit is in 27360 single-precision mode by default. 27361 27362'-m4-100-single-only' 27363 Generate code for SH4-100 in such a way that no double-precision 27364 floating-point operations are used. 27365 27366'-m4-200' 27367 Generate code for SH4-200. 27368 27369'-m4-200-nofpu' 27370 Generate code for SH4-200 without in such a way that the 27371 floating-point unit is not used. 27372 27373'-m4-200-single' 27374 Generate code for SH4-200 assuming the floating-point unit is in 27375 single-precision mode by default. 27376 27377'-m4-200-single-only' 27378 Generate code for SH4-200 in such a way that no double-precision 27379 floating-point operations are used. 27380 27381'-m4-300' 27382 Generate code for SH4-300. 27383 27384'-m4-300-nofpu' 27385 Generate code for SH4-300 without in such a way that the 27386 floating-point unit is not used. 27387 27388'-m4-300-single' 27389 Generate code for SH4-300 in such a way that no double-precision 27390 floating-point operations are used. 27391 27392'-m4-300-single-only' 27393 Generate code for SH4-300 in such a way that no double-precision 27394 floating-point operations are used. 27395 27396'-m4-340' 27397 Generate code for SH4-340 (no MMU, no FPU). 27398 27399'-m4-500' 27400 Generate code for SH4-500 (no FPU). Passes '-isa=sh4-nofpu' to the 27401 assembler. 27402 27403'-m4a-nofpu' 27404 Generate code for the SH4al-dsp, or for a SH4a in such a way that 27405 the floating-point unit is not used. 27406 27407'-m4a-single-only' 27408 Generate code for the SH4a, in such a way that no double-precision 27409 floating-point operations are used. 27410 27411'-m4a-single' 27412 Generate code for the SH4a assuming the floating-point unit is in 27413 single-precision mode by default. 27414 27415'-m4a' 27416 Generate code for the SH4a. 27417 27418'-m4al' 27419 Same as '-m4a-nofpu', except that it implicitly passes '-dsp' to 27420 the assembler. GCC doesn't generate any DSP instructions at the 27421 moment. 27422 27423'-mb' 27424 Compile code for the processor in big-endian mode. 27425 27426'-ml' 27427 Compile code for the processor in little-endian mode. 27428 27429'-mdalign' 27430 Align doubles at 64-bit boundaries. Note that this changes the 27431 calling conventions, and thus some functions from the standard C 27432 library do not work unless you recompile it first with '-mdalign'. 27433 27434'-mrelax' 27435 Shorten some address references at link time, when possible; uses 27436 the linker option '-relax'. 27437 27438'-mbigtable' 27439 Use 32-bit offsets in 'switch' tables. The default is to use 27440 16-bit offsets. 27441 27442'-mbitops' 27443 Enable the use of bit manipulation instructions on SH2A. 27444 27445'-mfmovd' 27446 Enable the use of the instruction 'fmovd'. Check '-mdalign' for 27447 alignment constraints. 27448 27449'-mrenesas' 27450 Comply with the calling conventions defined by Renesas. 27451 27452'-mno-renesas' 27453 Comply with the calling conventions defined for GCC before the 27454 Renesas conventions were available. This option is the default for 27455 all targets of the SH toolchain. 27456 27457'-mnomacsave' 27458 Mark the 'MAC' register as call-clobbered, even if '-mrenesas' is 27459 given. 27460 27461'-mieee' 27462'-mno-ieee' 27463 Control the IEEE compliance of floating-point comparisons, which 27464 affects the handling of cases where the result of a comparison is 27465 unordered. By default '-mieee' is implicitly enabled. If 27466 '-ffinite-math-only' is enabled '-mno-ieee' is implicitly set, 27467 which results in faster floating-point greater-equal and less-equal 27468 comparisons. The implicit settings can be overridden by specifying 27469 either '-mieee' or '-mno-ieee'. 27470 27471'-minline-ic_invalidate' 27472 Inline code to invalidate instruction cache entries after setting 27473 up nested function trampolines. This option has no effect if 27474 '-musermode' is in effect and the selected code generation option 27475 (e.g. '-m4') does not allow the use of the 'icbi' instruction. If 27476 the selected code generation option does not allow the use of the 27477 'icbi' instruction, and '-musermode' is not in effect, the inlined 27478 code manipulates the instruction cache address array directly with 27479 an associative write. This not only requires privileged mode at 27480 run time, but it also fails if the cache line had been mapped via 27481 the TLB and has become unmapped. 27482 27483'-misize' 27484 Dump instruction size and location in the assembly code. 27485 27486'-mpadstruct' 27487 This option is deprecated. It pads structures to multiple of 4 27488 bytes, which is incompatible with the SH ABI. 27489 27490'-matomic-model=MODEL' 27491 Sets the model of atomic operations and additional parameters as a 27492 comma separated list. For details on the atomic built-in functions 27493 see *note __atomic Builtins::. The following models and parameters 27494 are supported: 27495 27496 'none' 27497 Disable compiler generated atomic sequences and emit library 27498 calls for atomic operations. This is the default if the 27499 target is not 'sh*-*-linux*'. 27500 27501 'soft-gusa' 27502 Generate GNU/Linux compatible gUSA software atomic sequences 27503 for the atomic built-in functions. The generated atomic 27504 sequences require additional support from the 27505 interrupt/exception handling code of the system and are only 27506 suitable for SH3* and SH4* single-core systems. This option 27507 is enabled by default when the target is 'sh*-*-linux*' and 27508 SH3* or SH4*. When the target is SH4A, this option also 27509 partially utilizes the hardware atomic instructions 'movli.l' 27510 and 'movco.l' to create more efficient code, unless 'strict' 27511 is specified. 27512 27513 'soft-tcb' 27514 Generate software atomic sequences that use a variable in the 27515 thread control block. This is a variation of the gUSA 27516 sequences which can also be used on SH1* and SH2* targets. 27517 The generated atomic sequences require additional support from 27518 the interrupt/exception handling code of the system and are 27519 only suitable for single-core systems. When using this model, 27520 the 'gbr-offset=' parameter has to be specified as well. 27521 27522 'soft-imask' 27523 Generate software atomic sequences that temporarily disable 27524 interrupts by setting 'SR.IMASK = 1111'. This model works 27525 only when the program runs in privileged mode and is only 27526 suitable for single-core systems. Additional support from the 27527 interrupt/exception handling code of the system is not 27528 required. This model is enabled by default when the target is 27529 'sh*-*-linux*' and SH1* or SH2*. 27530 27531 'hard-llcs' 27532 Generate hardware atomic sequences using the 'movli.l' and 27533 'movco.l' instructions only. This is only available on SH4A 27534 and is suitable for multi-core systems. Since the hardware 27535 instructions support only 32 bit atomic variables access to 8 27536 or 16 bit variables is emulated with 32 bit accesses. Code 27537 compiled with this option is also compatible with other 27538 software atomic model interrupt/exception handling systems if 27539 executed on an SH4A system. Additional support from the 27540 interrupt/exception handling code of the system is not 27541 required for this model. 27542 27543 'gbr-offset=' 27544 This parameter specifies the offset in bytes of the variable 27545 in the thread control block structure that should be used by 27546 the generated atomic sequences when the 'soft-tcb' model has 27547 been selected. For other models this parameter is ignored. 27548 The specified value must be an integer multiple of four and in 27549 the range 0-1020. 27550 27551 'strict' 27552 This parameter prevents mixed usage of multiple atomic models, 27553 even if they are compatible, and makes the compiler generate 27554 atomic sequences of the specified model only. 27555 27556'-mtas' 27557 Generate the 'tas.b' opcode for '__atomic_test_and_set'. Notice 27558 that depending on the particular hardware and software 27559 configuration this can degrade overall performance due to the 27560 operand cache line flushes that are implied by the 'tas.b' 27561 instruction. On multi-core SH4A processors the 'tas.b' instruction 27562 must be used with caution since it can result in data corruption 27563 for certain cache configurations. 27564 27565'-mprefergot' 27566 When generating position-independent code, emit function calls 27567 using the Global Offset Table instead of the Procedure Linkage 27568 Table. 27569 27570'-musermode' 27571'-mno-usermode' 27572 Don't allow (allow) the compiler generating privileged mode code. 27573 Specifying '-musermode' also implies '-mno-inline-ic_invalidate' if 27574 the inlined code would not work in user mode. '-musermode' is the 27575 default when the target is 'sh*-*-linux*'. If the target is SH1* 27576 or SH2* '-musermode' has no effect, since there is no user mode. 27577 27578'-multcost=NUMBER' 27579 Set the cost to assume for a multiply insn. 27580 27581'-mdiv=STRATEGY' 27582 Set the division strategy to be used for integer division 27583 operations. STRATEGY can be one of: 27584 27585 'call-div1' 27586 Calls a library function that uses the single-step division 27587 instruction 'div1' to perform the operation. Division by zero 27588 calculates an unspecified result and does not trap. This is 27589 the default except for SH4, SH2A and SHcompact. 27590 27591 'call-fp' 27592 Calls a library function that performs the operation in double 27593 precision floating point. Division by zero causes a 27594 floating-point exception. This is the default for SHcompact 27595 with FPU. Specifying this for targets that do not have a 27596 double precision FPU defaults to 'call-div1'. 27597 27598 'call-table' 27599 Calls a library function that uses a lookup table for small 27600 divisors and the 'div1' instruction with case distinction for 27601 larger divisors. Division by zero calculates an unspecified 27602 result and does not trap. This is the default for SH4. 27603 Specifying this for targets that do not have dynamic shift 27604 instructions defaults to 'call-div1'. 27605 27606 When a division strategy has not been specified the default 27607 strategy is selected based on the current target. For SH2A the 27608 default strategy is to use the 'divs' and 'divu' instructions 27609 instead of library function calls. 27610 27611'-maccumulate-outgoing-args' 27612 Reserve space once for outgoing arguments in the function prologue 27613 rather than around each call. Generally beneficial for performance 27614 and size. Also needed for unwinding to avoid changing the stack 27615 frame around conditional code. 27616 27617'-mdivsi3_libfunc=NAME' 27618 Set the name of the library function used for 32-bit signed 27619 division to NAME. This only affects the name used in the 'call' 27620 division strategies, and the compiler still expects the same sets 27621 of input/output/clobbered registers as if this option were not 27622 present. 27623 27624'-mfixed-range=REGISTER-RANGE' 27625 Generate code treating the given register range as fixed registers. 27626 A fixed register is one that the register allocator cannot use. 27627 This is useful when compiling kernel code. A register range is 27628 specified as two registers separated by a dash. Multiple register 27629 ranges can be specified separated by a comma. 27630 27631'-mbranch-cost=NUM' 27632 Assume NUM to be the cost for a branch instruction. Higher numbers 27633 make the compiler try to generate more branch-free code if 27634 possible. If not specified the value is selected depending on the 27635 processor type that is being compiled for. 27636 27637'-mzdcbranch' 27638'-mno-zdcbranch' 27639 Assume (do not assume) that zero displacement conditional branch 27640 instructions 'bt' and 'bf' are fast. If '-mzdcbranch' is 27641 specified, the compiler prefers zero displacement branch code 27642 sequences. This is enabled by default when generating code for SH4 27643 and SH4A. It can be explicitly disabled by specifying 27644 '-mno-zdcbranch'. 27645 27646'-mcbranch-force-delay-slot' 27647 Force the usage of delay slots for conditional branches, which 27648 stuffs the delay slot with a 'nop' if a suitable instruction cannot 27649 be found. By default this option is disabled. It can be enabled 27650 to work around hardware bugs as found in the original SH7055. 27651 27652'-mfused-madd' 27653'-mno-fused-madd' 27654 Generate code that uses (does not use) the floating-point multiply 27655 and accumulate instructions. These instructions are generated by 27656 default if hardware floating point is used. The machine-dependent 27657 '-mfused-madd' option is now mapped to the machine-independent 27658 '-ffp-contract=fast' option, and '-mno-fused-madd' is mapped to 27659 '-ffp-contract=off'. 27660 27661'-mfsca' 27662'-mno-fsca' 27663 Allow or disallow the compiler to emit the 'fsca' instruction for 27664 sine and cosine approximations. The option '-mfsca' must be used 27665 in combination with '-funsafe-math-optimizations'. It is enabled 27666 by default when generating code for SH4A. Using '-mno-fsca' 27667 disables sine and cosine approximations even if 27668 '-funsafe-math-optimizations' is in effect. 27669 27670'-mfsrra' 27671'-mno-fsrra' 27672 Allow or disallow the compiler to emit the 'fsrra' instruction for 27673 reciprocal square root approximations. The option '-mfsrra' must 27674 be used in combination with '-funsafe-math-optimizations' and 27675 '-ffinite-math-only'. It is enabled by default when generating 27676 code for SH4A. Using '-mno-fsrra' disables reciprocal square root 27677 approximations even if '-funsafe-math-optimizations' and 27678 '-ffinite-math-only' are in effect. 27679 27680'-mpretend-cmove' 27681 Prefer zero-displacement conditional branches for conditional move 27682 instruction patterns. This can result in faster code on the SH4 27683 processor. 27684 27685'-mfdpic' 27686 Generate code using the FDPIC ABI. 27687 27688 27689File: gcc.info, Node: Solaris 2 Options, Next: SPARC Options, Prev: SH Options, Up: Submodel Options 27690 276913.19.50 Solaris 2 Options 27692------------------------- 27693 27694These '-m' options are supported on Solaris 2: 27695 27696'-mclear-hwcap' 27697 '-mclear-hwcap' tells the compiler to remove the hardware 27698 capabilities generated by the Solaris assembler. This is only 27699 necessary when object files use ISA extensions not supported by the 27700 current machine, but check at runtime whether or not to use them. 27701 27702'-mimpure-text' 27703 '-mimpure-text', used in addition to '-shared', tells the compiler 27704 to not pass '-z text' to the linker when linking a shared object. 27705 Using this option, you can link position-dependent code into a 27706 shared object. 27707 27708 '-mimpure-text' suppresses the "relocations remain against 27709 allocatable but non-writable sections" linker error message. 27710 However, the necessary relocations trigger copy-on-write, and the 27711 shared object is not actually shared across processes. Instead of 27712 using '-mimpure-text', you should compile all source code with 27713 '-fpic' or '-fPIC'. 27714 27715 These switches are supported in addition to the above on Solaris 2: 27716 27717'-pthreads' 27718 This is a synonym for '-pthread'. 27719 27720 27721File: gcc.info, Node: SPARC Options, Next: System V Options, Prev: Solaris 2 Options, Up: Submodel Options 27722 277233.19.51 SPARC Options 27724--------------------- 27725 27726These '-m' options are supported on the SPARC: 27727 27728'-mno-app-regs' 27729'-mapp-regs' 27730 Specify '-mapp-regs' to generate output using the global registers 27731 2 through 4, which the SPARC SVR4 ABI reserves for applications. 27732 Like the global register 1, each global register 2 through 4 is 27733 then treated as an allocable register that is clobbered by function 27734 calls. This is the default. 27735 27736 To be fully SVR4 ABI-compliant at the cost of some performance 27737 loss, specify '-mno-app-regs'. You should compile libraries and 27738 system software with this option. 27739 27740'-mflat' 27741'-mno-flat' 27742 With '-mflat', the compiler does not generate save/restore 27743 instructions and uses a "flat" or single register window model. 27744 This model is compatible with the regular register window model. 27745 The local registers and the input registers (0-5) are still treated 27746 as "call-saved" registers and are saved on the stack as needed. 27747 27748 With '-mno-flat' (the default), the compiler generates save/restore 27749 instructions (except for leaf functions). This is the normal 27750 operating mode. 27751 27752'-mfpu' 27753'-mhard-float' 27754 Generate output containing floating-point instructions. This is 27755 the default. 27756 27757'-mno-fpu' 27758'-msoft-float' 27759 Generate output containing library calls for floating point. 27760 *Warning:* the requisite libraries are not available for all SPARC 27761 targets. Normally the facilities of the machine's usual C compiler 27762 are used, but this cannot be done directly in cross-compilation. 27763 You must make your own arrangements to provide suitable library 27764 functions for cross-compilation. The embedded targets 27765 'sparc-*-aout' and 'sparclite-*-*' do provide software 27766 floating-point support. 27767 27768 '-msoft-float' changes the calling convention in the output file; 27769 therefore, it is only useful if you compile _all_ of a program with 27770 this option. In particular, you need to compile 'libgcc.a', the 27771 library that comes with GCC, with '-msoft-float' in order for this 27772 to work. 27773 27774'-mhard-quad-float' 27775 Generate output containing quad-word (long double) floating-point 27776 instructions. 27777 27778'-msoft-quad-float' 27779 Generate output containing library calls for quad-word (long 27780 double) floating-point instructions. The functions called are 27781 those specified in the SPARC ABI. This is the default. 27782 27783 As of this writing, there are no SPARC implementations that have 27784 hardware support for the quad-word floating-point instructions. 27785 They all invoke a trap handler for one of these instructions, and 27786 then the trap handler emulates the effect of the instruction. 27787 Because of the trap handler overhead, this is much slower than 27788 calling the ABI library routines. Thus the '-msoft-quad-float' 27789 option is the default. 27790 27791'-mno-unaligned-doubles' 27792'-munaligned-doubles' 27793 Assume that doubles have 8-byte alignment. This is the default. 27794 27795 With '-munaligned-doubles', GCC assumes that doubles have 8-byte 27796 alignment only if they are contained in another type, or if they 27797 have an absolute address. Otherwise, it assumes they have 4-byte 27798 alignment. Specifying this option avoids some rare compatibility 27799 problems with code generated by other compilers. It is not the 27800 default because it results in a performance loss, especially for 27801 floating-point code. 27802 27803'-muser-mode' 27804'-mno-user-mode' 27805 Do not generate code that can only run in supervisor mode. This is 27806 relevant only for the 'casa' instruction emitted for the LEON3 27807 processor. This is the default. 27808 27809'-mfaster-structs' 27810'-mno-faster-structs' 27811 With '-mfaster-structs', the compiler assumes that structures 27812 should have 8-byte alignment. This enables the use of pairs of 27813 'ldd' and 'std' instructions for copies in structure assignment, in 27814 place of twice as many 'ld' and 'st' pairs. However, the use of 27815 this changed alignment directly violates the SPARC ABI. Thus, it's 27816 intended only for use on targets where the developer acknowledges 27817 that their resulting code is not directly in line with the rules of 27818 the ABI. 27819 27820'-mstd-struct-return' 27821'-mno-std-struct-return' 27822 With '-mstd-struct-return', the compiler generates checking code in 27823 functions returning structures or unions to detect size mismatches 27824 between the two sides of function calls, as per the 32-bit ABI. 27825 27826 The default is '-mno-std-struct-return'. This option has no effect 27827 in 64-bit mode. 27828 27829'-mlra' 27830'-mno-lra' 27831 Enable Local Register Allocation. This is the default for SPARC 27832 since GCC 7 so '-mno-lra' needs to be passed to get old Reload. 27833 27834'-mcpu=CPU_TYPE' 27835 Set the instruction set, register set, and instruction scheduling 27836 parameters for machine type CPU_TYPE. Supported values for 27837 CPU_TYPE are 'v7', 'cypress', 'v8', 'supersparc', 'hypersparc', 27838 'leon', 'leon3', 'leon3v7', 'leon5', 'sparclite', 'f930', 'f934', 27839 'sparclite86x', 'sparclet', 'tsc701', 'v9', 'ultrasparc', 27840 'ultrasparc3', 'niagara', 'niagara2', 'niagara3', 'niagara4', 27841 'niagara7' and 'm8'. 27842 27843 Native Solaris and GNU/Linux toolchains also support the value 27844 'native', which selects the best architecture option for the host 27845 processor. '-mcpu=native' has no effect if GCC does not recognize 27846 the processor. 27847 27848 Default instruction scheduling parameters are used for values that 27849 select an architecture and not an implementation. These are 'v7', 27850 'v8', 'sparclite', 'sparclet', 'v9'. 27851 27852 Here is a list of each supported architecture and their supported 27853 implementations. 27854 27855 v7 27856 cypress, leon3v7 27857 27858 v8 27859 supersparc, hypersparc, leon, leon3, leon5 27860 27861 sparclite 27862 f930, f934, sparclite86x 27863 27864 sparclet 27865 tsc701 27866 27867 v9 27868 ultrasparc, ultrasparc3, niagara, niagara2, niagara3, 27869 niagara4, niagara7, m8 27870 27871 By default (unless configured otherwise), GCC generates code for 27872 the V7 variant of the SPARC architecture. With '-mcpu=cypress', 27873 the compiler additionally optimizes it for the Cypress CY7C602 27874 chip, as used in the SPARCStation/SPARCServer 3xx series. This is 27875 also appropriate for the older SPARCStation 1, 2, IPX etc. 27876 27877 With '-mcpu=v8', GCC generates code for the V8 variant of the SPARC 27878 architecture. The only difference from V7 code is that the 27879 compiler emits the integer multiply and integer divide instructions 27880 which exist in SPARC-V8 but not in SPARC-V7. With 27881 '-mcpu=supersparc', the compiler additionally optimizes it for the 27882 SuperSPARC chip, as used in the SPARCStation 10, 1000 and 2000 27883 series. 27884 27885 With '-mcpu=sparclite', GCC generates code for the SPARClite 27886 variant of the SPARC architecture. This adds the integer multiply, 27887 integer divide step and scan ('ffs') instructions which exist in 27888 SPARClite but not in SPARC-V7. With '-mcpu=f930', the compiler 27889 additionally optimizes it for the Fujitsu MB86930 chip, which is 27890 the original SPARClite, with no FPU. With '-mcpu=f934', the 27891 compiler additionally optimizes it for the Fujitsu MB86934 chip, 27892 which is the more recent SPARClite with FPU. 27893 27894 With '-mcpu=sparclet', GCC generates code for the SPARClet variant 27895 of the SPARC architecture. This adds the integer multiply, 27896 multiply/accumulate, integer divide step and scan ('ffs') 27897 instructions which exist in SPARClet but not in SPARC-V7. With 27898 '-mcpu=tsc701', the compiler additionally optimizes it for the 27899 TEMIC SPARClet chip. 27900 27901 With '-mcpu=v9', GCC generates code for the V9 variant of the SPARC 27902 architecture. This adds 64-bit integer and floating-point move 27903 instructions, 3 additional floating-point condition code registers 27904 and conditional move instructions. With '-mcpu=ultrasparc', the 27905 compiler additionally optimizes it for the Sun UltraSPARC I/II/IIi 27906 chips. With '-mcpu=ultrasparc3', the compiler additionally 27907 optimizes it for the Sun UltraSPARC III/III+/IIIi/IIIi+/IV/IV+ 27908 chips. With '-mcpu=niagara', the compiler additionally optimizes 27909 it for Sun UltraSPARC T1 chips. With '-mcpu=niagara2', the 27910 compiler additionally optimizes it for Sun UltraSPARC T2 chips. 27911 With '-mcpu=niagara3', the compiler additionally optimizes it for 27912 Sun UltraSPARC T3 chips. With '-mcpu=niagara4', the compiler 27913 additionally optimizes it for Sun UltraSPARC T4 chips. With 27914 '-mcpu=niagara7', the compiler additionally optimizes it for Oracle 27915 SPARC M7 chips. With '-mcpu=m8', the compiler additionally 27916 optimizes it for Oracle M8 chips. 27917 27918'-mtune=CPU_TYPE' 27919 Set the instruction scheduling parameters for machine type 27920 CPU_TYPE, but do not set the instruction set or register set that 27921 the option '-mcpu=CPU_TYPE' does. 27922 27923 The same values for '-mcpu=CPU_TYPE' can be used for 27924 '-mtune=CPU_TYPE', but the only useful values are those that select 27925 a particular CPU implementation. Those are 'cypress', 27926 'supersparc', 'hypersparc', 'leon', 'leon3', 'leon3v7', 'leon5', 27927 'f930', 'f934', 'sparclite86x', 'tsc701', 'ultrasparc', 27928 'ultrasparc3', 'niagara', 'niagara2', 'niagara3', 'niagara4', 27929 'niagara7' and 'm8'. With native Solaris and GNU/Linux toolchains, 27930 'native' can also be used. 27931 27932'-mv8plus' 27933'-mno-v8plus' 27934 With '-mv8plus', GCC generates code for the SPARC-V8+ ABI. The 27935 difference from the V8 ABI is that the global and out registers are 27936 considered 64 bits wide. This is enabled by default on Solaris in 27937 32-bit mode for all SPARC-V9 processors. 27938 27939'-mvis' 27940'-mno-vis' 27941 With '-mvis', GCC generates code that takes advantage of the 27942 UltraSPARC Visual Instruction Set extensions. The default is 27943 '-mno-vis'. 27944 27945'-mvis2' 27946'-mno-vis2' 27947 With '-mvis2', GCC generates code that takes advantage of version 27948 2.0 of the UltraSPARC Visual Instruction Set extensions. The 27949 default is '-mvis2' when targeting a cpu that supports such 27950 instructions, such as UltraSPARC-III and later. Setting '-mvis2' 27951 also sets '-mvis'. 27952 27953'-mvis3' 27954'-mno-vis3' 27955 With '-mvis3', GCC generates code that takes advantage of version 27956 3.0 of the UltraSPARC Visual Instruction Set extensions. The 27957 default is '-mvis3' when targeting a cpu that supports such 27958 instructions, such as niagara-3 and later. Setting '-mvis3' also 27959 sets '-mvis2' and '-mvis'. 27960 27961'-mvis4' 27962'-mno-vis4' 27963 With '-mvis4', GCC generates code that takes advantage of version 27964 4.0 of the UltraSPARC Visual Instruction Set extensions. The 27965 default is '-mvis4' when targeting a cpu that supports such 27966 instructions, such as niagara-7 and later. Setting '-mvis4' also 27967 sets '-mvis3', '-mvis2' and '-mvis'. 27968 27969'-mvis4b' 27970'-mno-vis4b' 27971 With '-mvis4b', GCC generates code that takes advantage of version 27972 4.0 of the UltraSPARC Visual Instruction Set extensions, plus the 27973 additional VIS instructions introduced in the Oracle SPARC 27974 Architecture 2017. The default is '-mvis4b' when targeting a cpu 27975 that supports such instructions, such as m8 and later. Setting 27976 '-mvis4b' also sets '-mvis4', '-mvis3', '-mvis2' and '-mvis'. 27977 27978'-mcbcond' 27979'-mno-cbcond' 27980 With '-mcbcond', GCC generates code that takes advantage of the 27981 UltraSPARC Compare-and-Branch-on-Condition instructions. The 27982 default is '-mcbcond' when targeting a CPU that supports such 27983 instructions, such as Niagara-4 and later. 27984 27985'-mfmaf' 27986'-mno-fmaf' 27987 With '-mfmaf', GCC generates code that takes advantage of the 27988 UltraSPARC Fused Multiply-Add Floating-point instructions. The 27989 default is '-mfmaf' when targeting a CPU that supports such 27990 instructions, such as Niagara-3 and later. 27991 27992'-mfsmuld' 27993'-mno-fsmuld' 27994 With '-mfsmuld', GCC generates code that takes advantage of the 27995 Floating-point Multiply Single to Double (FsMULd) instruction. The 27996 default is '-mfsmuld' when targeting a CPU supporting the 27997 architecture versions V8 or V9 with FPU except '-mcpu=leon'. 27998 27999'-mpopc' 28000'-mno-popc' 28001 With '-mpopc', GCC generates code that takes advantage of the 28002 UltraSPARC Population Count instruction. The default is '-mpopc' 28003 when targeting a CPU that supports such an instruction, such as 28004 Niagara-2 and later. 28005 28006'-msubxc' 28007'-mno-subxc' 28008 With '-msubxc', GCC generates code that takes advantage of the 28009 UltraSPARC Subtract-Extended-with-Carry instruction. The default 28010 is '-msubxc' when targeting a CPU that supports such an 28011 instruction, such as Niagara-7 and later. 28012 28013'-mfix-at697f' 28014 Enable the documented workaround for the single erratum of the 28015 Atmel AT697F processor (which corresponds to erratum #13 of the 28016 AT697E processor). 28017 28018'-mfix-ut699' 28019 Enable the documented workarounds for the floating-point errata and 28020 the data cache nullify errata of the UT699 processor. 28021 28022'-mfix-ut700' 28023 Enable the documented workaround for the back-to-back store errata 28024 of the UT699E/UT700 processor. 28025 28026'-mfix-gr712rc' 28027 Enable the documented workaround for the back-to-back store errata 28028 of the GR712RC processor. 28029 28030 These '-m' options are supported in addition to the above on SPARC-V9 28031processors in 64-bit environments: 28032 28033'-m32' 28034'-m64' 28035 Generate code for a 32-bit or 64-bit environment. The 32-bit 28036 environment sets int, long and pointer to 32 bits. The 64-bit 28037 environment sets int to 32 bits and long and pointer to 64 bits. 28038 28039'-mcmodel=WHICH' 28040 Set the code model to one of 28041 28042 'medlow' 28043 The Medium/Low code model: 64-bit addresses, programs must be 28044 linked in the low 32 bits of memory. Programs can be 28045 statically or dynamically linked. 28046 28047 'medmid' 28048 The Medium/Middle code model: 64-bit addresses, programs must 28049 be linked in the low 44 bits of memory, the text and data 28050 segments must be less than 2GB in size and the data segment 28051 must be located within 2GB of the text segment. 28052 28053 'medany' 28054 The Medium/Anywhere code model: 64-bit addresses, programs may 28055 be linked anywhere in memory, the text and data segments must 28056 be less than 2GB in size and the data segment must be located 28057 within 2GB of the text segment. 28058 28059 'embmedany' 28060 The Medium/Anywhere code model for embedded systems: 64-bit 28061 addresses, the text and data segments must be less than 2GB in 28062 size, both starting anywhere in memory (determined at link 28063 time). The global register %g4 points to the base of the data 28064 segment. Programs are statically linked and PIC is not 28065 supported. 28066 28067'-mmemory-model=MEM-MODEL' 28068 Set the memory model in force on the processor to one of 28069 28070 'default' 28071 The default memory model for the processor and operating 28072 system. 28073 28074 'rmo' 28075 Relaxed Memory Order 28076 28077 'pso' 28078 Partial Store Order 28079 28080 'tso' 28081 Total Store Order 28082 28083 'sc' 28084 Sequential Consistency 28085 28086 These memory models are formally defined in Appendix D of the 28087 SPARC-V9 architecture manual, as set in the processor's 'PSTATE.MM' 28088 field. 28089 28090'-mstack-bias' 28091'-mno-stack-bias' 28092 With '-mstack-bias', GCC assumes that the stack pointer, and frame 28093 pointer if present, are offset by -2047 which must be added back 28094 when making stack frame references. This is the default in 64-bit 28095 mode. Otherwise, assume no such offset is present. 28096 28097 28098File: gcc.info, Node: System V Options, Next: TILE-Gx Options, Prev: SPARC Options, Up: Submodel Options 28099 281003.19.52 Options for System V 28101---------------------------- 28102 28103These additional options are available on System V Release 4 for 28104compatibility with other compilers on those systems: 28105 28106'-G' 28107 Create a shared object. It is recommended that '-symbolic' or 28108 '-shared' be used instead. 28109 28110'-Qy' 28111 Identify the versions of each tool used by the compiler, in a 28112 '.ident' assembler directive in the output. 28113 28114'-Qn' 28115 Refrain from adding '.ident' directives to the output file (this is 28116 the default). 28117 28118'-YP,DIRS' 28119 Search the directories DIRS, and no others, for libraries specified 28120 with '-l'. 28121 28122'-Ym,DIR' 28123 Look in the directory DIR to find the M4 preprocessor. The 28124 assembler uses this option. 28125 28126 28127File: gcc.info, Node: TILE-Gx Options, Next: TILEPro Options, Prev: System V Options, Up: Submodel Options 28128 281293.19.53 TILE-Gx Options 28130----------------------- 28131 28132These '-m' options are supported on the TILE-Gx: 28133 28134'-mcmodel=small' 28135 Generate code for the small model. The distance for direct calls 28136 is limited to 500M in either direction. PC-relative addresses are 28137 32 bits. Absolute addresses support the full address range. 28138 28139'-mcmodel=large' 28140 Generate code for the large model. There is no limitation on call 28141 distance, pc-relative addresses, or absolute addresses. 28142 28143'-mcpu=NAME' 28144 Selects the type of CPU to be targeted. Currently the only 28145 supported type is 'tilegx'. 28146 28147'-m32' 28148'-m64' 28149 Generate code for a 32-bit or 64-bit environment. The 32-bit 28150 environment sets int, long, and pointer to 32 bits. The 64-bit 28151 environment sets int to 32 bits and long and pointer to 64 bits. 28152 28153'-mbig-endian' 28154'-mlittle-endian' 28155 Generate code in big/little endian mode, respectively. 28156 28157 28158File: gcc.info, Node: TILEPro Options, Next: V850 Options, Prev: TILE-Gx Options, Up: Submodel Options 28159 281603.19.54 TILEPro Options 28161----------------------- 28162 28163These '-m' options are supported on the TILEPro: 28164 28165'-mcpu=NAME' 28166 Selects the type of CPU to be targeted. Currently the only 28167 supported type is 'tilepro'. 28168 28169'-m32' 28170 Generate code for a 32-bit environment, which sets int, long, and 28171 pointer to 32 bits. This is the only supported behavior so the 28172 flag is essentially ignored. 28173 28174 28175File: gcc.info, Node: V850 Options, Next: VAX Options, Prev: TILEPro Options, Up: Submodel Options 28176 281773.19.55 V850 Options 28178-------------------- 28179 28180These '-m' options are defined for V850 implementations: 28181 28182'-mlong-calls' 28183'-mno-long-calls' 28184 Treat all calls as being far away (near). If calls are assumed to 28185 be far away, the compiler always loads the function's address into 28186 a register, and calls indirect through the pointer. 28187 28188'-mno-ep' 28189'-mep' 28190 Do not optimize (do optimize) basic blocks that use the same index 28191 pointer 4 or more times to copy pointer into the 'ep' register, and 28192 use the shorter 'sld' and 'sst' instructions. The '-mep' option is 28193 on by default if you optimize. 28194 28195'-mno-prolog-function' 28196'-mprolog-function' 28197 Do not use (do use) external functions to save and restore 28198 registers at the prologue and epilogue of a function. The external 28199 functions are slower, but use less code space if more than one 28200 function saves the same number of registers. The 28201 '-mprolog-function' option is on by default if you optimize. 28202 28203'-mspace' 28204 Try to make the code as small as possible. At present, this just 28205 turns on the '-mep' and '-mprolog-function' options. 28206 28207'-mtda=N' 28208 Put static or global variables whose size is N bytes or less into 28209 the tiny data area that register 'ep' points to. The tiny data 28210 area can hold up to 256 bytes in total (128 bytes for byte 28211 references). 28212 28213'-msda=N' 28214 Put static or global variables whose size is N bytes or less into 28215 the small data area that register 'gp' points to. The small data 28216 area can hold up to 64 kilobytes. 28217 28218'-mzda=N' 28219 Put static or global variables whose size is N bytes or less into 28220 the first 32 kilobytes of memory. 28221 28222'-mv850' 28223 Specify that the target processor is the V850. 28224 28225'-mv850e3v5' 28226 Specify that the target processor is the V850E3V5. The 28227 preprocessor constant '__v850e3v5__' is defined if this option is 28228 used. 28229 28230'-mv850e2v4' 28231 Specify that the target processor is the V850E3V5. This is an 28232 alias for the '-mv850e3v5' option. 28233 28234'-mv850e2v3' 28235 Specify that the target processor is the V850E2V3. The 28236 preprocessor constant '__v850e2v3__' is defined if this option is 28237 used. 28238 28239'-mv850e2' 28240 Specify that the target processor is the V850E2. The preprocessor 28241 constant '__v850e2__' is defined if this option is used. 28242 28243'-mv850e1' 28244 Specify that the target processor is the V850E1. The preprocessor 28245 constants '__v850e1__' and '__v850e__' are defined if this option 28246 is used. 28247 28248'-mv850es' 28249 Specify that the target processor is the V850ES. This is an alias 28250 for the '-mv850e1' option. 28251 28252'-mv850e' 28253 Specify that the target processor is the V850E. The preprocessor 28254 constant '__v850e__' is defined if this option is used. 28255 28256 If neither '-mv850' nor '-mv850e' nor '-mv850e1' nor '-mv850e2' nor 28257 '-mv850e2v3' nor '-mv850e3v5' are defined then a default target 28258 processor is chosen and the relevant '__v850*__' preprocessor 28259 constant is defined. 28260 28261 The preprocessor constants '__v850' and '__v851__' are always 28262 defined, regardless of which processor variant is the target. 28263 28264'-mdisable-callt' 28265'-mno-disable-callt' 28266 This option suppresses generation of the 'CALLT' instruction for 28267 the v850e, v850e1, v850e2, v850e2v3 and v850e3v5 flavors of the 28268 v850 architecture. 28269 28270 This option is enabled by default when the RH850 ABI is in use (see 28271 '-mrh850-abi'), and disabled by default when the GCC ABI is in use. 28272 If 'CALLT' instructions are being generated then the C preprocessor 28273 symbol '__V850_CALLT__' is defined. 28274 28275'-mrelax' 28276'-mno-relax' 28277 Pass on (or do not pass on) the '-mrelax' command-line option to 28278 the assembler. 28279 28280'-mlong-jumps' 28281'-mno-long-jumps' 28282 Disable (or re-enable) the generation of PC-relative jump 28283 instructions. 28284 28285'-msoft-float' 28286'-mhard-float' 28287 Disable (or re-enable) the generation of hardware floating point 28288 instructions. This option is only significant when the target 28289 architecture is 'V850E2V3' or higher. If hardware floating point 28290 instructions are being generated then the C preprocessor symbol 28291 '__FPU_OK__' is defined, otherwise the symbol '__NO_FPU__' is 28292 defined. 28293 28294'-mloop' 28295 Enables the use of the e3v5 LOOP instruction. The use of this 28296 instruction is not enabled by default when the e3v5 architecture is 28297 selected because its use is still experimental. 28298 28299'-mrh850-abi' 28300'-mghs' 28301 Enables support for the RH850 version of the V850 ABI. This is the 28302 default. With this version of the ABI the following rules apply: 28303 28304 * Integer sized structures and unions are returned via a memory 28305 pointer rather than a register. 28306 28307 * Large structures and unions (more than 8 bytes in size) are 28308 passed by value. 28309 28310 * Functions are aligned to 16-bit boundaries. 28311 28312 * The '-m8byte-align' command-line option is supported. 28313 28314 * The '-mdisable-callt' command-line option is enabled by 28315 default. The '-mno-disable-callt' command-line option is not 28316 supported. 28317 28318 When this version of the ABI is enabled the C preprocessor symbol 28319 '__V850_RH850_ABI__' is defined. 28320 28321'-mgcc-abi' 28322 Enables support for the old GCC version of the V850 ABI. With this 28323 version of the ABI the following rules apply: 28324 28325 * Integer sized structures and unions are returned in register 28326 'r10'. 28327 28328 * Large structures and unions (more than 8 bytes in size) are 28329 passed by reference. 28330 28331 * Functions are aligned to 32-bit boundaries, unless optimizing 28332 for size. 28333 28334 * The '-m8byte-align' command-line option is not supported. 28335 28336 * The '-mdisable-callt' command-line option is supported but not 28337 enabled by default. 28338 28339 When this version of the ABI is enabled the C preprocessor symbol 28340 '__V850_GCC_ABI__' is defined. 28341 28342'-m8byte-align' 28343'-mno-8byte-align' 28344 Enables support for 'double' and 'long long' types to be aligned on 28345 8-byte boundaries. The default is to restrict the alignment of all 28346 objects to at most 4-bytes. When '-m8byte-align' is in effect the 28347 C preprocessor symbol '__V850_8BYTE_ALIGN__' is defined. 28348 28349'-mbig-switch' 28350 Generate code suitable for big switch tables. Use this option only 28351 if the assembler/linker complain about out of range branches within 28352 a switch table. 28353 28354'-mapp-regs' 28355 This option causes r2 and r5 to be used in the code generated by 28356 the compiler. This setting is the default. 28357 28358'-mno-app-regs' 28359 This option causes r2 and r5 to be treated as fixed registers. 28360 28361 28362File: gcc.info, Node: VAX Options, Next: Visium Options, Prev: V850 Options, Up: Submodel Options 28363 283643.19.56 VAX Options 28365------------------- 28366 28367These '-m' options are defined for the VAX: 28368 28369'-munix' 28370 Do not output certain jump instructions ('aobleq' and so on) that 28371 the Unix assembler for the VAX cannot handle across long ranges. 28372 28373'-mgnu' 28374 Do output those jump instructions, on the assumption that the GNU 28375 assembler is being used. 28376 28377'-mg' 28378 Output code for G-format floating-point numbers instead of 28379 D-format. 28380 28381'-mlra' 28382'-mno-lra' 28383 Enable Local Register Allocation. This is still experimental for 28384 the VAX, so by default the compiler uses standard reload. 28385 28386 28387File: gcc.info, Node: Visium Options, Next: VMS Options, Prev: VAX Options, Up: Submodel Options 28388 283893.19.57 Visium Options 28390---------------------- 28391 28392'-mdebug' 28393 A program which performs file I/O and is destined to run on an MCM 28394 target should be linked with this option. It causes the libraries 28395 libc.a and libdebug.a to be linked. The program should be run on 28396 the target under the control of the GDB remote debugging stub. 28397 28398'-msim' 28399 A program which performs file I/O and is destined to run on the 28400 simulator should be linked with option. This causes libraries 28401 libc.a and libsim.a to be linked. 28402 28403'-mfpu' 28404'-mhard-float' 28405 Generate code containing floating-point instructions. This is the 28406 default. 28407 28408'-mno-fpu' 28409'-msoft-float' 28410 Generate code containing library calls for floating-point. 28411 28412 '-msoft-float' changes the calling convention in the output file; 28413 therefore, it is only useful if you compile _all_ of a program with 28414 this option. In particular, you need to compile 'libgcc.a', the 28415 library that comes with GCC, with '-msoft-float' in order for this 28416 to work. 28417 28418'-mcpu=CPU_TYPE' 28419 Set the instruction set, register set, and instruction scheduling 28420 parameters for machine type CPU_TYPE. Supported values for 28421 CPU_TYPE are 'mcm', 'gr5' and 'gr6'. 28422 28423 'mcm' is a synonym of 'gr5' present for backward compatibility. 28424 28425 By default (unless configured otherwise), GCC generates code for 28426 the GR5 variant of the Visium architecture. 28427 28428 With '-mcpu=gr6', GCC generates code for the GR6 variant of the 28429 Visium architecture. The only difference from GR5 code is that the 28430 compiler will generate block move instructions. 28431 28432'-mtune=CPU_TYPE' 28433 Set the instruction scheduling parameters for machine type 28434 CPU_TYPE, but do not set the instruction set or register set that 28435 the option '-mcpu=CPU_TYPE' would. 28436 28437'-msv-mode' 28438 Generate code for the supervisor mode, where there are no 28439 restrictions on the access to general registers. This is the 28440 default. 28441 28442'-muser-mode' 28443 Generate code for the user mode, where the access to some general 28444 registers is forbidden: on the GR5, registers r24 to r31 cannot be 28445 accessed in this mode; on the GR6, only registers r29 to r31 are 28446 affected. 28447 28448 28449File: gcc.info, Node: VMS Options, Next: VxWorks Options, Prev: Visium Options, Up: Submodel Options 28450 284513.19.58 VMS Options 28452------------------- 28453 28454These '-m' options are defined for the VMS implementations: 28455 28456'-mvms-return-codes' 28457 Return VMS condition codes from 'main'. The default is to return 28458 POSIX-style condition (e.g. error) codes. 28459 28460'-mdebug-main=PREFIX' 28461 Flag the first routine whose name starts with PREFIX as the main 28462 routine for the debugger. 28463 28464'-mmalloc64' 28465 Default to 64-bit memory allocation routines. 28466 28467'-mpointer-size=SIZE' 28468 Set the default size of pointers. Possible options for SIZE are 28469 '32' or 'short' for 32 bit pointers, '64' or 'long' for 64 bit 28470 pointers, and 'no' for supporting only 32 bit pointers. The later 28471 option disables 'pragma pointer_size'. 28472 28473 28474File: gcc.info, Node: VxWorks Options, Next: x86 Options, Prev: VMS Options, Up: Submodel Options 28475 284763.19.59 VxWorks Options 28477----------------------- 28478 28479The options in this section are defined for all VxWorks targets. 28480Options specific to the target hardware are listed with the other 28481options for that target. 28482 28483'-mrtp' 28484 GCC can generate code for both VxWorks kernels and real time 28485 processes (RTPs). This option switches from the former to the 28486 latter. It also defines the preprocessor macro '__RTP__'. 28487 28488'-non-static' 28489 Link an RTP executable against shared libraries rather than static 28490 libraries. The options '-static' and '-shared' can also be used 28491 for RTPs (*note Link Options::); '-static' is the default. 28492 28493'-Bstatic' 28494'-Bdynamic' 28495 These options are passed down to the linker. They are defined for 28496 compatibility with Diab. 28497 28498'-Xbind-lazy' 28499 Enable lazy binding of function calls. This option is equivalent 28500 to '-Wl,-z,now' and is defined for compatibility with Diab. 28501 28502'-Xbind-now' 28503 Disable lazy binding of function calls. This option is the default 28504 and is defined for compatibility with Diab. 28505 28506 28507File: gcc.info, Node: x86 Options, Next: x86 Windows Options, Prev: VxWorks Options, Up: Submodel Options 28508 285093.19.60 x86 Options 28510------------------- 28511 28512These '-m' options are defined for the x86 family of computers. 28513 28514'-march=CPU-TYPE' 28515 Generate instructions for the machine type CPU-TYPE. In contrast 28516 to '-mtune=CPU-TYPE', which merely tunes the generated code for the 28517 specified CPU-TYPE, '-march=CPU-TYPE' allows GCC to generate code 28518 that may not run at all on processors other than the one indicated. 28519 Specifying '-march=CPU-TYPE' implies '-mtune=CPU-TYPE', except 28520 where noted otherwise. 28521 28522 The choices for CPU-TYPE are: 28523 28524 'native' 28525 This selects the CPU to generate code for at compilation time 28526 by determining the processor type of the compiling machine. 28527 Using '-march=native' enables all instruction subsets 28528 supported by the local machine (hence the result might not run 28529 on different machines). Using '-mtune=native' produces code 28530 optimized for the local machine under the constraints of the 28531 selected instruction set. 28532 28533 'x86-64' 28534 A generic CPU with 64-bit extensions. 28535 28536 'x86-64-v2' 28537 'x86-64-v3' 28538 'x86-64-v4' 28539 These choices for CPU-TYPE select the corresponding 28540 micro-architecture level from the x86-64 psABI. On ABIs other 28541 than the x86-64 psABI they select the same CPU features as the 28542 x86-64 psABI documents for the particular micro-architecture 28543 level. 28544 28545 Since these CPU-TYPE values do not have a corresponding 28546 '-mtune' setting, using '-march' with these values enables 28547 generic tuning. Specific tuning can be enabled using the 28548 '-mtune=OTHER-CPU-TYPE' option with an appropriate 28549 OTHER-CPU-TYPE value. 28550 28551 'i386' 28552 Original Intel i386 CPU. 28553 28554 'i486' 28555 Intel i486 CPU. (No scheduling is implemented for this chip.) 28556 28557 'i586' 28558 'pentium' 28559 Intel Pentium CPU with no MMX support. 28560 28561 'lakemont' 28562 Intel Lakemont MCU, based on Intel Pentium CPU. 28563 28564 'pentium-mmx' 28565 Intel Pentium MMX CPU, based on Pentium core with MMX 28566 instruction set support. 28567 28568 'pentiumpro' 28569 Intel Pentium Pro CPU. 28570 28571 'i686' 28572 When used with '-march', the Pentium Pro instruction set is 28573 used, so the code runs on all i686 family chips. When used 28574 with '-mtune', it has the same meaning as 'generic'. 28575 28576 'pentium2' 28577 Intel Pentium II CPU, based on Pentium Pro core with MMX and 28578 FXSR instruction set support. 28579 28580 'pentium3' 28581 'pentium3m' 28582 Intel Pentium III CPU, based on Pentium Pro core with MMX, 28583 FXSR and SSE instruction set support. 28584 28585 'pentium-m' 28586 Intel Pentium M; low-power version of Intel Pentium III CPU 28587 with MMX, SSE, SSE2 and FXSR instruction set support. Used by 28588 Centrino notebooks. 28589 28590 'pentium4' 28591 'pentium4m' 28592 Intel Pentium 4 CPU with MMX, SSE, SSE2 and FXSR instruction 28593 set support. 28594 28595 'prescott' 28596 Improved version of Intel Pentium 4 CPU with MMX, SSE, SSE2, 28597 SSE3 and FXSR instruction set support. 28598 28599 'nocona' 28600 Improved version of Intel Pentium 4 CPU with 64-bit 28601 extensions, MMX, SSE, SSE2, SSE3 and FXSR instruction set 28602 support. 28603 28604 'core2' 28605 Intel Core 2 CPU with 64-bit extensions, MMX, SSE, SSE2, SSE3, 28606 SSSE3, CX16, SAHF and FXSR instruction set support. 28607 28608 'nehalem' 28609 Intel Nehalem CPU with 64-bit extensions, MMX, SSE, SSE2, 28610 SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, CX16, SAHF and FXSR 28611 instruction set support. 28612 28613 'westmere' 28614 Intel Westmere CPU with 64-bit extensions, MMX, SSE, SSE2, 28615 SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, CX16, SAHF, FXSR and 28616 PCLMUL instruction set support. 28617 28618 'sandybridge' 28619 Intel Sandy Bridge CPU with 64-bit extensions, MMX, SSE, SSE2, 28620 SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, CX16, SAHF, FXSR, AVX, 28621 XSAVE and PCLMUL instruction set support. 28622 28623 'ivybridge' 28624 Intel Ivy Bridge CPU with 64-bit extensions, MMX, SSE, SSE2, 28625 SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, CX16, SAHF, FXSR, AVX, 28626 XSAVE, PCLMUL, FSGSBASE, RDRND and F16C instruction set 28627 support. 28628 28629 'haswell' 28630 Intel Haswell CPU with 64-bit extensions, MOVBE, MMX, SSE, 28631 SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, CX16, SAHF, FXSR, 28632 AVX, XSAVE, PCLMUL, FSGSBASE, RDRND, F16C, AVX2, BMI, BMI2, 28633 LZCNT, FMA, MOVBE and HLE instruction set support. 28634 28635 'broadwell' 28636 Intel Broadwell CPU with 64-bit extensions, MOVBE, MMX, SSE, 28637 SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, CX16, SAHF, FXSR, 28638 AVX, XSAVE, PCLMUL, FSGSBASE, RDRND, F16C, AVX2, BMI, BMI2, 28639 LZCNT, FMA, MOVBE, HLE, RDSEED, ADCX and PREFETCHW instruction 28640 set support. 28641 28642 'skylake' 28643 Intel Skylake CPU with 64-bit extensions, MOVBE, MMX, SSE, 28644 SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, CX16, SAHF, FXSR, 28645 AVX, XSAVE, PCLMUL, FSGSBASE, RDRND, F16C, AVX2, BMI, BMI2, 28646 LZCNT, FMA, MOVBE, HLE, RDSEED, ADCX, PREFETCHW, AES, 28647 CLFLUSHOPT, XSAVEC, XSAVES and SGX instruction set support. 28648 28649 'bonnell' 28650 Intel Bonnell CPU with 64-bit extensions, MOVBE, MMX, SSE, 28651 SSE2, SSE3 and SSSE3 instruction set support. 28652 28653 'silvermont' 28654 Intel Silvermont CPU with 64-bit extensions, MOVBE, MMX, SSE, 28655 SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, CX16, SAHF, FXSR, 28656 PCLMUL, PREFETCHW and RDRND instruction set support. 28657 28658 'goldmont' 28659 Intel Goldmont CPU with 64-bit extensions, MOVBE, MMX, SSE, 28660 SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, CX16, SAHF, FXSR, 28661 PCLMUL, PREFETCHW, RDRND, AES, SHA, RDSEED, XSAVE, XSAVEC, 28662 XSAVES, XSAVEOPT, CLFLUSHOPT and FSGSBASE instruction set 28663 support. 28664 28665 'goldmont-plus' 28666 Intel Goldmont Plus CPU with 64-bit extensions, MOVBE, MMX, 28667 SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, CX16, SAHF, 28668 FXSR, PCLMUL, PREFETCHW, RDRND, AES, SHA, RDSEED, XSAVE, 28669 XSAVEC, XSAVES, XSAVEOPT, CLFLUSHOPT, FSGSBASE, PTWRITE, RDPID 28670 and SGX instruction set support. 28671 28672 'tremont' 28673 Intel Tremont CPU with 64-bit extensions, MOVBE, MMX, SSE, 28674 SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, CX16, SAHF, FXSR, 28675 PCLMUL, PREFETCHW, RDRND, AES, SHA, RDSEED, XSAVE, XSAVEC, 28676 XSAVES, XSAVEOPT, CLFLUSHOPT, FSGSBASE, PTWRITE, RDPID, SGX, 28677 CLWB, GFNI-SSE, MOVDIRI, MOVDIR64B, CLDEMOTE and WAITPKG 28678 instruction set support. 28679 28680 'knl' 28681 Intel Knight's Landing CPU with 64-bit extensions, MOVBE, MMX, 28682 SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, CX16, SAHF, 28683 FXSR, AVX, XSAVE, PCLMUL, FSGSBASE, RDRND, F16C, AVX2, BMI, 28684 BMI2, LZCNT, FMA, MOVBE, HLE, RDSEED, ADCX, PREFETCHW, 28685 AVX512PF, AVX512ER, AVX512F, AVX512CD and PREFETCHWT1 28686 instruction set support. 28687 28688 'knm' 28689 Intel Knights Mill CPU with 64-bit extensions, MOVBE, MMX, 28690 SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, CX16, SAHF, 28691 FXSR, AVX, XSAVE, PCLMUL, FSGSBASE, RDRND, F16C, AVX2, BMI, 28692 BMI2, LZCNT, FMA, MOVBE, HLE, RDSEED, ADCX, PREFETCHW, 28693 AVX512PF, AVX512ER, AVX512F, AVX512CD and PREFETCHWT1, 28694 AVX5124VNNIW, AVX5124FMAPS and AVX512VPOPCNTDQ instruction set 28695 support. 28696 28697 'skylake-avx512' 28698 Intel Skylake Server CPU with 64-bit extensions, MOVBE, MMX, 28699 SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, CX16, SAHF, 28700 FXSR, AVX, XSAVE, PCLMUL, FSGSBASE, RDRND, F16C, AVX2, BMI, 28701 BMI2, LZCNT, FMA, MOVBE, HLE, RDSEED, ADCX, PREFETCHW, AES, 28702 CLFLUSHOPT, XSAVEC, XSAVES, SGX, AVX512F, CLWB, AVX512VL, 28703 AVX512BW, AVX512DQ and AVX512CD instruction set support. 28704 28705 'cannonlake' 28706 Intel Cannonlake Server CPU with 64-bit extensions, MOVBE, 28707 MMX, SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, CX16, 28708 SAHF, FXSR, AVX, XSAVE, PCLMUL, FSGSBASE, RDRND, F16C, AVX2, 28709 BMI, BMI2, LZCNT, FMA, MOVBE, HLE, RDSEED, ADCX, PREFETCHW, 28710 AES, CLFLUSHOPT, XSAVEC, XSAVES, SGX, AVX512F, AVX512VL, 28711 AVX512BW, AVX512DQ, AVX512CD, PKU, AVX512VBMI, AVX512IFMA and 28712 SHA instruction set support. 28713 28714 'icelake-client' 28715 Intel Icelake Client CPU with 64-bit extensions, MOVBE, MMX, 28716 SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, CX16, SAHF, 28717 FXSR, AVX, XSAVE, PCLMUL, FSGSBASE, RDRND, F16C, AVX2, BMI, 28718 BMI2, LZCNT, FMA, MOVBE, HLE, RDSEED, ADCX, PREFETCHW, AES, 28719 CLFLUSHOPT, XSAVEC, XSAVES, SGX, AVX512F, AVX512VL, AVX512BW, 28720 AVX512DQ, AVX512CD, PKU, AVX512VBMI, AVX512IFMA, SHA, 28721 AVX512VNNI, GFNI, VAES, AVX512VBMI2 , VPCLMULQDQ, 28722 AVX512BITALG, RDPID and AVX512VPOPCNTDQ instruction set 28723 support. 28724 28725 'icelake-server' 28726 Intel Icelake Server CPU with 64-bit extensions, MOVBE, MMX, 28727 SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, CX16, SAHF, 28728 FXSR, AVX, XSAVE, PCLMUL, FSGSBASE, RDRND, F16C, AVX2, BMI, 28729 BMI2, LZCNT, FMA, MOVBE, HLE, RDSEED, ADCX, PREFETCHW, AES, 28730 CLFLUSHOPT, XSAVEC, XSAVES, SGX, AVX512F, AVX512VL, AVX512BW, 28731 AVX512DQ, AVX512CD, PKU, AVX512VBMI, AVX512IFMA, SHA, 28732 AVX512VNNI, GFNI, VAES, AVX512VBMI2 , VPCLMULQDQ, 28733 AVX512BITALG, RDPID, AVX512VPOPCNTDQ, PCONFIG, WBNOINVD and 28734 CLWB instruction set support. 28735 28736 'cascadelake' 28737 Intel Cascadelake CPU with 64-bit extensions, MOVBE, MMX, SSE, 28738 SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, CX16, SAHF, FXSR, 28739 AVX, XSAVE, PCLMUL, FSGSBASE, RDRND, F16C, AVX2, BMI, BMI2, 28740 LZCNT, FMA, MOVBE, HLE, RDSEED, ADCX, PREFETCHW, AES, 28741 CLFLUSHOPT, XSAVEC, XSAVES, SGX, AVX512F, CLWB, AVX512VL, 28742 AVX512BW, AVX512DQ, AVX512CD and AVX512VNNI instruction set 28743 support. 28744 28745 'cooperlake' 28746 Intel cooperlake CPU with 64-bit extensions, MOVBE, MMX, SSE, 28747 SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, CX16, SAHF, FXSR, 28748 AVX, XSAVE, PCLMUL, FSGSBASE, RDRND, F16C, AVX2, BMI, BMI2, 28749 LZCNT, FMA, MOVBE, HLE, RDSEED, ADCX, PREFETCHW, AES, 28750 CLFLUSHOPT, XSAVEC, XSAVES, SGX, AVX512F, CLWB, AVX512VL, 28751 AVX512BW, AVX512DQ, AVX512CD, AVX512VNNI and AVX512BF16 28752 instruction set support. 28753 28754 'tigerlake' 28755 Intel Tigerlake CPU with 64-bit extensions, MOVBE, MMX, SSE, 28756 SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, CX16, SAHF, FXSR, 28757 AVX, XSAVE, PCLMUL, FSGSBASE, RDRND, F16C, AVX2, BMI, BMI2, 28758 LZCNT, FMA, MOVBE, HLE, RDSEED, ADCX, PREFETCHW, AES, 28759 CLFLUSHOPT, XSAVEC, XSAVES, SGX, AVX512F, AVX512VL, AVX512BW, 28760 AVX512DQ, AVX512CD PKU, AVX512VBMI, AVX512IFMA, SHA, 28761 AVX512VNNI, GFNI, VAES, AVX512VBMI2, VPCLMULQDQ, AVX512BITALG, 28762 RDPID, AVX512VPOPCNTDQ, MOVDIRI, MOVDIR64B, CLWB, 28763 AVX512VP2INTERSECT and KEYLOCKER instruction set support. 28764 28765 'sapphirerapids' 28766 Intel sapphirerapids CPU with 64-bit extensions, MOVBE, MMX, 28767 SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, CX16, SAHF, 28768 FXSR, AVX, XSAVE, PCLMUL, FSGSBASE, RDRND, F16C, AVX2, BMI, 28769 BMI2, LZCNT, FMA, MOVBE, HLE, RDSEED, ADCX, PREFETCHW, AES, 28770 CLFLUSHOPT, XSAVEC, XSAVES, SGX, AVX512F, AVX512VL, AVX512BW, 28771 AVX512DQ, AVX512CD, PKU, AVX512VBMI, AVX512IFMA, SHA, 28772 AVX512VNNI, GFNI, VAES, AVX512VBMI2, VPCLMULQDQ, AVX512BITALG, 28773 RDPID, AVX512VPOPCNTDQ, PCONFIG, WBNOINVD, CLWB, MOVDIRI, 28774 MOVDIR64B, ENQCMD, CLDEMOTE, PTWRITE, WAITPKG, SERIALIZE, 28775 TSXLDTRK, UINTR, AMX-BF16, AMX-TILE, AMX-INT8, AVX-VNNI, 28776 AVX512-FP16 and AVX512BF16 instruction set support. 28777 28778 'alderlake' 28779 Intel Alderlake CPU with 64-bit extensions, MOVBE, MMX, SSE, 28780 SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, POPCNT, AES, PREFETCHW, 28781 PCLMUL, RDRND, XSAVE, XSAVEC, XSAVES, XSAVEOPT, FSGSBASE, 28782 PTWRITE, RDPID, SGX, GFNI-SSE, CLWB, MOVDIRI, MOVDIR64B, 28783 CLDEMOTE, WAITPKG, ADCX, AVX, AVX2, BMI, BMI2, F16C, FMA, 28784 LZCNT, PCONFIG, PKU, VAES, VPCLMULQDQ, SERIALIZE, HRESET, KL, 28785 WIDEKL and AVX-VNNI instruction set support. 28786 28787 'rocketlake' 28788 Intel Rocketlake CPU with 64-bit extensions, MOVBE, MMX, SSE, 28789 SSE2, SSE3, SSSE3 , SSE4.1, SSE4.2, POPCNT, CX16, SAHF, FXSR, 28790 AVX, XSAVE, PCLMUL, FSGSBASE, RDRND, F16C, AVX2, BMI, BMI2, 28791 LZCNT, FMA, MOVBE, HLE, RDSEED, ADCX, PREFETCHW, AES, 28792 CLFLUSHOPT, XSAVEC, XSAVES, AVX512F, AVX512VL, AVX512BW, 28793 AVX512DQ, AVX512CD PKU, AVX512VBMI, AVX512IFMA, SHA, 28794 AVX512VNNI, GFNI, VAES, AVX512VBMI2, VPCLMULQDQ, AVX512BITALG, 28795 RDPID and AVX512VPOPCNTDQ instruction set support. 28796 28797 'k6' 28798 AMD K6 CPU with MMX instruction set support. 28799 28800 'k6-2' 28801 'k6-3' 28802 Improved versions of AMD K6 CPU with MMX and 3DNow! 28803 instruction set support. 28804 28805 'athlon' 28806 'athlon-tbird' 28807 AMD Athlon CPU with MMX, 3dNOW!, enhanced 3DNow! and SSE 28808 prefetch instructions support. 28809 28810 'athlon-4' 28811 'athlon-xp' 28812 'athlon-mp' 28813 Improved AMD Athlon CPU with MMX, 3DNow!, enhanced 3DNow! and 28814 full SSE instruction set support. 28815 28816 'k8' 28817 'opteron' 28818 'athlon64' 28819 'athlon-fx' 28820 Processors based on the AMD K8 core with x86-64 instruction 28821 set support, including the AMD Opteron, Athlon 64, and Athlon 28822 64 FX processors. (This supersets MMX, SSE, SSE2, 3DNow!, 28823 enhanced 3DNow! and 64-bit instruction set extensions.) 28824 28825 'k8-sse3' 28826 'opteron-sse3' 28827 'athlon64-sse3' 28828 Improved versions of AMD K8 cores with SSE3 instruction set 28829 support. 28830 28831 'amdfam10' 28832 'barcelona' 28833 CPUs based on AMD Family 10h cores with x86-64 instruction set 28834 support. (This supersets MMX, SSE, SSE2, SSE3, SSE4A, 3DNow!, 28835 enhanced 3DNow!, ABM and 64-bit instruction set extensions.) 28836 28837 'bdver1' 28838 CPUs based on AMD Family 15h cores with x86-64 instruction set 28839 support. (This supersets FMA4, AVX, XOP, LWP, AES, PCLMUL, 28840 CX16, MMX, SSE, SSE2, SSE3, SSE4A, SSSE3, SSE4.1, SSE4.2, ABM 28841 and 64-bit instruction set extensions.) 28842 28843 'bdver2' 28844 AMD Family 15h core based CPUs with x86-64 instruction set 28845 support. (This supersets BMI, TBM, F16C, FMA, FMA4, AVX, XOP, 28846 LWP, AES, PCLMUL, CX16, MMX, SSE, SSE2, SSE3, SSE4A, SSSE3, 28847 SSE4.1, SSE4.2, ABM and 64-bit instruction set extensions.) 28848 28849 'bdver3' 28850 AMD Family 15h core based CPUs with x86-64 instruction set 28851 support. (This supersets BMI, TBM, F16C, FMA, FMA4, FSGSBASE, 28852 AVX, XOP, LWP, AES, PCLMUL, CX16, MMX, SSE, SSE2, SSE3, SSE4A, 28853 SSSE3, SSE4.1, SSE4.2, ABM and 64-bit instruction set 28854 extensions.) 28855 28856 'bdver4' 28857 AMD Family 15h core based CPUs with x86-64 instruction set 28858 support. (This supersets BMI, BMI2, TBM, F16C, FMA, FMA4, 28859 FSGSBASE, AVX, AVX2, XOP, LWP, AES, PCLMUL, CX16, MOVBE, MMX, 28860 SSE, SSE2, SSE3, SSE4A, SSSE3, SSE4.1, SSE4.2, ABM and 64-bit 28861 instruction set extensions.) 28862 28863 'znver1' 28864 AMD Family 17h core based CPUs with x86-64 instruction set 28865 support. (This supersets BMI, BMI2, F16C, FMA, FSGSBASE, AVX, 28866 AVX2, ADCX, RDSEED, MWAITX, SHA, CLZERO, AES, PCLMUL, CX16, 28867 MOVBE, MMX, SSE, SSE2, SSE3, SSE4A, SSSE3, SSE4.1, SSE4.2, 28868 ABM, XSAVEC, XSAVES, CLFLUSHOPT, POPCNT, and 64-bit 28869 instruction set extensions.) 28870 28871 'znver2' 28872 AMD Family 17h core based CPUs with x86-64 instruction set 28873 support. (This supersets BMI, BMI2, CLWB, F16C, FMA, 28874 FSGSBASE, AVX, AVX2, ADCX, RDSEED, MWAITX, SHA, CLZERO, AES, 28875 PCLMUL, CX16, MOVBE, MMX, SSE, SSE2, SSE3, SSE4A, SSSE3, 28876 SSE4.1, SSE4.2, ABM, XSAVEC, XSAVES, CLFLUSHOPT, POPCNT, 28877 RDPID, WBNOINVD, and 64-bit instruction set extensions.) 28878 28879 'znver3' 28880 AMD Family 19h core based CPUs with x86-64 instruction set 28881 support. (This supersets BMI, BMI2, CLWB, F16C, FMA, 28882 FSGSBASE, AVX, AVX2, ADCX, RDSEED, MWAITX, SHA, CLZERO, AES, 28883 PCLMUL, CX16, MOVBE, MMX, SSE, SSE2, SSE3, SSE4A, SSSE3, 28884 SSE4.1, SSE4.2, ABM, XSAVEC, XSAVES, CLFLUSHOPT, POPCNT, 28885 RDPID, WBNOINVD, PKU, VPCLMULQDQ, VAES, and 64-bit instruction 28886 set extensions.) 28887 28888 'znver4' 28889 AMD Family 19h core based CPUs with x86-64 instruction set 28890 support. (This supersets BMI, BMI2, CLWB, F16C, FMA, 28891 FSGSBASE, AVX, AVX2, ADCX, RDSEED, MWAITX, SHA, CLZERO, AES, 28892 PCLMUL, CX16, MOVBE, MMX, SSE, SSE2, SSE3, SSE4A, SSSE3, 28893 SSE4.1, SSE4.2, ABM, XSAVEC, XSAVES, CLFLUSHOPT, POPCNT, 28894 RDPID, WBNOINVD, PKU, VPCLMULQDQ, VAES, AVX512F, AVX512DQ, 28895 AVX512IFMA, AVX512CD, AVX512BW, AVX512VL, AVX512BF16, 28896 AVX512VBMI, AVX512VBMI2, AVX512VNNI, AVX512BITALG, 28897 AVX512VPOPCNTDQ, GFNI and 64-bit instruction set extensions.) 28898 28899 'btver1' 28900 CPUs based on AMD Family 14h cores with x86-64 instruction set 28901 support. (This supersets MMX, SSE, SSE2, SSE3, SSSE3, SSE4A, 28902 CX16, ABM and 64-bit instruction set extensions.) 28903 28904 'btver2' 28905 CPUs based on AMD Family 16h cores with x86-64 instruction set 28906 support. This includes MOVBE, F16C, BMI, AVX, PCLMUL, AES, 28907 SSE4.2, SSE4.1, CX16, ABM, SSE4A, SSSE3, SSE3, SSE2, SSE, MMX 28908 and 64-bit instruction set extensions. 28909 28910 'winchip-c6' 28911 IDT WinChip C6 CPU, dealt in same way as i486 with additional 28912 MMX instruction set support. 28913 28914 'winchip2' 28915 IDT WinChip 2 CPU, dealt in same way as i486 with additional 28916 MMX and 3DNow! instruction set support. 28917 28918 'c3' 28919 VIA C3 CPU with MMX and 3DNow! instruction set support. (No 28920 scheduling is implemented for this chip.) 28921 28922 'c3-2' 28923 VIA C3-2 (Nehemiah/C5XL) CPU with MMX and SSE instruction set 28924 support. (No scheduling is implemented for this chip.) 28925 28926 'c7' 28927 VIA C7 (Esther) CPU with MMX, SSE, SSE2 and SSE3 instruction 28928 set support. (No scheduling is implemented for this chip.) 28929 28930 'samuel-2' 28931 VIA Eden Samuel 2 CPU with MMX and 3DNow! instruction set 28932 support. (No scheduling is implemented for this chip.) 28933 28934 'nehemiah' 28935 VIA Eden Nehemiah CPU with MMX and SSE instruction set 28936 support. (No scheduling is implemented for this chip.) 28937 28938 'esther' 28939 VIA Eden Esther CPU with MMX, SSE, SSE2 and SSE3 instruction 28940 set support. (No scheduling is implemented for this chip.) 28941 28942 'eden-x2' 28943 VIA Eden X2 CPU with x86-64, MMX, SSE, SSE2 and SSE3 28944 instruction set support. (No scheduling is implemented for 28945 this chip.) 28946 28947 'eden-x4' 28948 VIA Eden X4 CPU with x86-64, MMX, SSE, SSE2, SSE3, SSSE3, 28949 SSE4.1, SSE4.2, AVX and AVX2 instruction set support. (No 28950 scheduling is implemented for this chip.) 28951 28952 'nano' 28953 Generic VIA Nano CPU with x86-64, MMX, SSE, SSE2, SSE3 and 28954 SSSE3 instruction set support. (No scheduling is implemented 28955 for this chip.) 28956 28957 'nano-1000' 28958 VIA Nano 1xxx CPU with x86-64, MMX, SSE, SSE2, SSE3 and SSSE3 28959 instruction set support. (No scheduling is implemented for 28960 this chip.) 28961 28962 'nano-2000' 28963 VIA Nano 2xxx CPU with x86-64, MMX, SSE, SSE2, SSE3 and SSSE3 28964 instruction set support. (No scheduling is implemented for 28965 this chip.) 28966 28967 'nano-3000' 28968 VIA Nano 3xxx CPU with x86-64, MMX, SSE, SSE2, SSE3, SSSE3 and 28969 SSE4.1 instruction set support. (No scheduling is implemented 28970 for this chip.) 28971 28972 'nano-x2' 28973 VIA Nano Dual Core CPU with x86-64, MMX, SSE, SSE2, SSE3, 28974 SSSE3 and SSE4.1 instruction set support. (No scheduling is 28975 implemented for this chip.) 28976 28977 'nano-x4' 28978 VIA Nano Quad Core CPU with x86-64, MMX, SSE, SSE2, SSE3, 28979 SSSE3 and SSE4.1 instruction set support. (No scheduling is 28980 implemented for this chip.) 28981 28982 'geode' 28983 AMD Geode embedded processor with MMX and 3DNow! instruction 28984 set support. 28985 28986'-mtune=CPU-TYPE' 28987 Tune to CPU-TYPE everything applicable about the generated code, 28988 except for the ABI and the set of available instructions. While 28989 picking a specific CPU-TYPE schedules things appropriately for that 28990 particular chip, the compiler does not generate any code that 28991 cannot run on the default machine type unless you use a 28992 '-march=CPU-TYPE' option. For example, if GCC is configured for 28993 i686-pc-linux-gnu then '-mtune=pentium4' generates code that is 28994 tuned for Pentium 4 but still runs on i686 machines. 28995 28996 The choices for CPU-TYPE are the same as for '-march'. In 28997 addition, '-mtune' supports 2 extra choices for CPU-TYPE: 28998 28999 'generic' 29000 Produce code optimized for the most common IA32/AMD64/EM64T 29001 processors. If you know the CPU on which your code will run, 29002 then you should use the corresponding '-mtune' or '-march' 29003 option instead of '-mtune=generic'. But, if you do not know 29004 exactly what CPU users of your application will have, then you 29005 should use this option. 29006 29007 As new processors are deployed in the marketplace, the 29008 behavior of this option will change. Therefore, if you 29009 upgrade to a newer version of GCC, code generation controlled 29010 by this option will change to reflect the processors that are 29011 most common at the time that version of GCC is released. 29012 29013 There is no '-march=generic' option because '-march' indicates 29014 the instruction set the compiler can use, and there is no 29015 generic instruction set applicable to all processors. In 29016 contrast, '-mtune' indicates the processor (or, in this case, 29017 collection of processors) for which the code is optimized. 29018 29019 'intel' 29020 Produce code optimized for the most current Intel processors, 29021 which are Haswell and Silvermont for this version of GCC. If 29022 you know the CPU on which your code will run, then you should 29023 use the corresponding '-mtune' or '-march' option instead of 29024 '-mtune=intel'. But, if you want your application performs 29025 better on both Haswell and Silvermont, then you should use 29026 this option. 29027 29028 As new Intel processors are deployed in the marketplace, the 29029 behavior of this option will change. Therefore, if you 29030 upgrade to a newer version of GCC, code generation controlled 29031 by this option will change to reflect the most current Intel 29032 processors at the time that version of GCC is released. 29033 29034 There is no '-march=intel' option because '-march' indicates 29035 the instruction set the compiler can use, and there is no 29036 common instruction set applicable to all processors. In 29037 contrast, '-mtune' indicates the processor (or, in this case, 29038 collection of processors) for which the code is optimized. 29039 29040'-mcpu=CPU-TYPE' 29041 A deprecated synonym for '-mtune'. 29042 29043'-mfpmath=UNIT' 29044 Generate floating-point arithmetic for selected unit UNIT. The 29045 choices for UNIT are: 29046 29047 '387' 29048 Use the standard 387 floating-point coprocessor present on the 29049 majority of chips and emulated otherwise. Code compiled with 29050 this option runs almost everywhere. The temporary results are 29051 computed in 80-bit precision instead of the precision 29052 specified by the type, resulting in slightly different results 29053 compared to most of other chips. See '-ffloat-store' for more 29054 detailed description. 29055 29056 This is the default choice for non-Darwin x86-32 targets. 29057 29058 'sse' 29059 Use scalar floating-point instructions present in the SSE 29060 instruction set. This instruction set is supported by Pentium 29061 III and newer chips, and in the AMD line by Athlon-4, Athlon 29062 XP and Athlon MP chips. The earlier version of the SSE 29063 instruction set supports only single-precision arithmetic, 29064 thus the double and extended-precision arithmetic are still 29065 done using 387. A later version, present only in Pentium 4 29066 and AMD x86-64 chips, supports double-precision arithmetic 29067 too. 29068 29069 For the x86-32 compiler, you must use '-march=CPU-TYPE', 29070 '-msse' or '-msse2' switches to enable SSE extensions and make 29071 this option effective. For the x86-64 compiler, these 29072 extensions are enabled by default. 29073 29074 The resulting code should be considerably faster in the 29075 majority of cases and avoid the numerical instability problems 29076 of 387 code, but may break some existing code that expects 29077 temporaries to be 80 bits. 29078 29079 This is the default choice for the x86-64 compiler, Darwin 29080 x86-32 targets, and the default choice for x86-32 targets with 29081 the SSE2 instruction set when '-ffast-math' is enabled. 29082 29083 'sse,387' 29084 'sse+387' 29085 'both' 29086 Attempt to utilize both instruction sets at once. This 29087 effectively doubles the amount of available registers, and on 29088 chips with separate execution units for 387 and SSE the 29089 execution resources too. Use this option with care, as it is 29090 still experimental, because the GCC register allocator does 29091 not model separate functional units well, resulting in 29092 unstable performance. 29093 29094'-masm=DIALECT' 29095 Output assembly instructions using selected DIALECT. Also affects 29096 which dialect is used for basic 'asm' (*note Basic Asm::) and 29097 extended 'asm' (*note Extended Asm::). Supported choices (in 29098 dialect order) are 'att' or 'intel'. The default is 'att'. Darwin 29099 does not support 'intel'. 29100 29101'-mieee-fp' 29102'-mno-ieee-fp' 29103 Control whether or not the compiler uses IEEE floating-point 29104 comparisons. These correctly handle the case where the result of a 29105 comparison is unordered. 29106 29107'-m80387' 29108'-mhard-float' 29109 Generate output containing 80387 instructions for floating point. 29110 29111'-mno-80387' 29112'-msoft-float' 29113 Generate output containing library calls for floating point. 29114 29115 *Warning:* the requisite libraries are not part of GCC. Normally 29116 the facilities of the machine's usual C compiler are used, but this 29117 cannot be done directly in cross-compilation. You must make your 29118 own arrangements to provide suitable library functions for 29119 cross-compilation. 29120 29121 On machines where a function returns floating-point results in the 29122 80387 register stack, some floating-point opcodes may be emitted 29123 even if '-msoft-float' is used. 29124 29125'-mno-fp-ret-in-387' 29126 Do not use the FPU registers for return values of functions. 29127 29128 The usual calling convention has functions return values of types 29129 'float' and 'double' in an FPU register, even if there is no FPU. 29130 The idea is that the operating system should emulate an FPU. 29131 29132 The option '-mno-fp-ret-in-387' causes such values to be returned 29133 in ordinary CPU registers instead. 29134 29135'-mno-fancy-math-387' 29136 Some 387 emulators do not support the 'sin', 'cos' and 'sqrt' 29137 instructions for the 387. Specify this option to avoid generating 29138 those instructions. This option is overridden when '-march' 29139 indicates that the target CPU always has an FPU and so the 29140 instruction does not need emulation. These instructions are not 29141 generated unless you also use the '-funsafe-math-optimizations' 29142 switch. 29143 29144'-malign-double' 29145'-mno-align-double' 29146 Control whether GCC aligns 'double', 'long double', and 'long long' 29147 variables on a two-word boundary or a one-word boundary. Aligning 29148 'double' variables on a two-word boundary produces code that runs 29149 somewhat faster on a Pentium at the expense of more memory. 29150 29151 On x86-64, '-malign-double' is enabled by default. 29152 29153 *Warning:* if you use the '-malign-double' switch, structures 29154 containing the above types are aligned differently than the 29155 published application binary interface specifications for the 29156 x86-32 and are not binary compatible with structures in code 29157 compiled without that switch. 29158 29159'-m96bit-long-double' 29160'-m128bit-long-double' 29161 These switches control the size of 'long double' type. The x86-32 29162 application binary interface specifies the size to be 96 bits, so 29163 '-m96bit-long-double' is the default in 32-bit mode. 29164 29165 Modern architectures (Pentium and newer) prefer 'long double' to be 29166 aligned to an 8- or 16-byte boundary. In arrays or structures 29167 conforming to the ABI, this is not possible. So specifying 29168 '-m128bit-long-double' aligns 'long double' to a 16-byte boundary 29169 by padding the 'long double' with an additional 32-bit zero. 29170 29171 In the x86-64 compiler, '-m128bit-long-double' is the default 29172 choice as its ABI specifies that 'long double' is aligned on 29173 16-byte boundary. 29174 29175 Notice that neither of these options enable any extra precision 29176 over the x87 standard of 80 bits for a 'long double'. 29177 29178 *Warning:* if you override the default value for your target ABI, 29179 this changes the size of structures and arrays containing 'long 29180 double' variables, as well as modifying the function calling 29181 convention for functions taking 'long double'. Hence they are not 29182 binary-compatible with code compiled without that switch. 29183 29184'-mlong-double-64' 29185'-mlong-double-80' 29186'-mlong-double-128' 29187 These switches control the size of 'long double' type. A size of 29188 64 bits makes the 'long double' type equivalent to the 'double' 29189 type. This is the default for 32-bit Bionic C library. A size of 29190 128 bits makes the 'long double' type equivalent to the 29191 '__float128' type. This is the default for 64-bit Bionic C 29192 library. 29193 29194 *Warning:* if you override the default value for your target ABI, 29195 this changes the size of structures and arrays containing 'long 29196 double' variables, as well as modifying the function calling 29197 convention for functions taking 'long double'. Hence they are not 29198 binary-compatible with code compiled without that switch. 29199 29200'-malign-data=TYPE' 29201 Control how GCC aligns variables. Supported values for TYPE are 29202 'compat' uses increased alignment value compatible uses GCC 4.8 and 29203 earlier, 'abi' uses alignment value as specified by the psABI, and 29204 'cacheline' uses increased alignment value to match the cache line 29205 size. 'compat' is the default. 29206 29207'-mlarge-data-threshold=THRESHOLD' 29208 When '-mcmodel=medium' is specified, data objects larger than 29209 THRESHOLD are placed in the large data section. This value must be 29210 the same across all objects linked into the binary, and defaults to 29211 65535. 29212 29213'-mrtd' 29214 Use a different function-calling convention, in which functions 29215 that take a fixed number of arguments return with the 'ret NUM' 29216 instruction, which pops their arguments while returning. This 29217 saves one instruction in the caller since there is no need to pop 29218 the arguments there. 29219 29220 You can specify that an individual function is called with this 29221 calling sequence with the function attribute 'stdcall'. You can 29222 also override the '-mrtd' option by using the function attribute 29223 'cdecl'. *Note Function Attributes::. 29224 29225 *Warning:* this calling convention is incompatible with the one 29226 normally used on Unix, so you cannot use it if you need to call 29227 libraries compiled with the Unix compiler. 29228 29229 Also, you must provide function prototypes for all functions that 29230 take variable numbers of arguments (including 'printf'); otherwise 29231 incorrect code is generated for calls to those functions. 29232 29233 In addition, seriously incorrect code results if you call a 29234 function with too many arguments. (Normally, extra arguments are 29235 harmlessly ignored.) 29236 29237'-mregparm=NUM' 29238 Control how many registers are used to pass integer arguments. By 29239 default, no registers are used to pass arguments, and at most 3 29240 registers can be used. You can control this behavior for a 29241 specific function by using the function attribute 'regparm'. *Note 29242 Function Attributes::. 29243 29244 *Warning:* if you use this switch, and NUM is nonzero, then you 29245 must build all modules with the same value, including any 29246 libraries. This includes the system libraries and startup modules. 29247 29248'-msseregparm' 29249 Use SSE register passing conventions for float and double arguments 29250 and return values. You can control this behavior for a specific 29251 function by using the function attribute 'sseregparm'. *Note 29252 Function Attributes::. 29253 29254 *Warning:* if you use this switch then you must build all modules 29255 with the same value, including any libraries. This includes the 29256 system libraries and startup modules. 29257 29258'-mvect8-ret-in-mem' 29259 Return 8-byte vectors in memory instead of MMX registers. This is 29260 the default on VxWorks to match the ABI of the Sun Studio compilers 29261 until version 12. _Only_ use this option if you need to remain 29262 compatible with existing code produced by those previous compiler 29263 versions or older versions of GCC. 29264 29265'-mpc32' 29266'-mpc64' 29267'-mpc80' 29268 29269 Set 80387 floating-point precision to 32, 64 or 80 bits. When 29270 '-mpc32' is specified, the significands of results of 29271 floating-point operations are rounded to 24 bits (single 29272 precision); '-mpc64' rounds the significands of results of 29273 floating-point operations to 53 bits (double precision) and 29274 '-mpc80' rounds the significands of results of floating-point 29275 operations to 64 bits (extended double precision), which is the 29276 default. When this option is used, floating-point operations in 29277 higher precisions are not available to the programmer without 29278 setting the FPU control word explicitly. 29279 29280 Setting the rounding of floating-point operations to less than the 29281 default 80 bits can speed some programs by 2% or more. Note that 29282 some mathematical libraries assume that extended-precision (80-bit) 29283 floating-point operations are enabled by default; routines in such 29284 libraries could suffer significant loss of accuracy, typically 29285 through so-called "catastrophic cancellation", when this option is 29286 used to set the precision to less than extended precision. 29287 29288'-mstackrealign' 29289 Realign the stack at entry. On the x86, the '-mstackrealign' 29290 option generates an alternate prologue and epilogue that realigns 29291 the run-time stack if necessary. This supports mixing legacy codes 29292 that keep 4-byte stack alignment with modern codes that keep 29293 16-byte stack alignment for SSE compatibility. See also the 29294 attribute 'force_align_arg_pointer', applicable to individual 29295 functions. 29296 29297'-mpreferred-stack-boundary=NUM' 29298 Attempt to keep the stack boundary aligned to a 2 raised to NUM 29299 byte boundary. If '-mpreferred-stack-boundary' is not specified, 29300 the default is 4 (16 bytes or 128 bits). 29301 29302 *Warning:* When generating code for the x86-64 architecture with 29303 SSE extensions disabled, '-mpreferred-stack-boundary=3' can be used 29304 to keep the stack boundary aligned to 8 byte boundary. Since 29305 x86-64 ABI require 16 byte stack alignment, this is ABI 29306 incompatible and intended to be used in controlled environment 29307 where stack space is important limitation. This option leads to 29308 wrong code when functions compiled with 16 byte stack alignment 29309 (such as functions from a standard library) are called with 29310 misaligned stack. In this case, SSE instructions may lead to 29311 misaligned memory access traps. In addition, variable arguments 29312 are handled incorrectly for 16 byte aligned objects (including x87 29313 long double and __int128), leading to wrong results. You must 29314 build all modules with '-mpreferred-stack-boundary=3', including 29315 any libraries. This includes the system libraries and startup 29316 modules. 29317 29318'-mincoming-stack-boundary=NUM' 29319 Assume the incoming stack is aligned to a 2 raised to NUM byte 29320 boundary. If '-mincoming-stack-boundary' is not specified, the one 29321 specified by '-mpreferred-stack-boundary' is used. 29322 29323 On Pentium and Pentium Pro, 'double' and 'long double' values 29324 should be aligned to an 8-byte boundary (see '-malign-double') or 29325 suffer significant run time performance penalties. On Pentium III, 29326 the Streaming SIMD Extension (SSE) data type '__m128' may not work 29327 properly if it is not 16-byte aligned. 29328 29329 To ensure proper alignment of this values on the stack, the stack 29330 boundary must be as aligned as that required by any value stored on 29331 the stack. Further, every function must be generated such that it 29332 keeps the stack aligned. Thus calling a function compiled with a 29333 higher preferred stack boundary from a function compiled with a 29334 lower preferred stack boundary most likely misaligns the stack. It 29335 is recommended that libraries that use callbacks always use the 29336 default setting. 29337 29338 This extra alignment does consume extra stack space, and generally 29339 increases code size. Code that is sensitive to stack space usage, 29340 such as embedded systems and operating system kernels, may want to 29341 reduce the preferred alignment to '-mpreferred-stack-boundary=2'. 29342 29343'-mmmx' 29344'-msse' 29345'-msse2' 29346'-msse3' 29347'-mssse3' 29348'-msse4' 29349'-msse4a' 29350'-msse4.1' 29351'-msse4.2' 29352'-mavx' 29353'-mavx2' 29354'-mavx512f' 29355'-mavx512pf' 29356'-mavx512er' 29357'-mavx512cd' 29358'-mavx512vl' 29359'-mavx512bw' 29360'-mavx512dq' 29361'-mavx512ifma' 29362'-mavx512vbmi' 29363'-msha' 29364'-maes' 29365'-mpclmul' 29366'-mclflushopt' 29367'-mclwb' 29368'-mfsgsbase' 29369'-mptwrite' 29370'-mrdrnd' 29371'-mf16c' 29372'-mfma' 29373'-mpconfig' 29374'-mwbnoinvd' 29375'-mfma4' 29376'-mprfchw' 29377'-mrdpid' 29378'-mprefetchwt1' 29379'-mrdseed' 29380'-msgx' 29381'-mxop' 29382'-mlwp' 29383'-m3dnow' 29384'-m3dnowa' 29385'-mpopcnt' 29386'-mabm' 29387'-madx' 29388'-mbmi' 29389'-mbmi2' 29390'-mlzcnt' 29391'-mfxsr' 29392'-mxsave' 29393'-mxsaveopt' 29394'-mxsavec' 29395'-mxsaves' 29396'-mrtm' 29397'-mhle' 29398'-mtbm' 29399'-mmwaitx' 29400'-mclzero' 29401'-mpku' 29402'-mavx512vbmi2' 29403'-mavx512bf16' 29404'-mavx512fp16' 29405'-mgfni' 29406'-mvaes' 29407'-mwaitpkg' 29408'-mvpclmulqdq' 29409'-mavx512bitalg' 29410'-mmovdiri' 29411'-mmovdir64b' 29412'-menqcmd' 29413'-muintr' 29414'-mtsxldtrk' 29415'-mavx512vpopcntdq' 29416'-mavx512vp2intersect' 29417'-mavx5124fmaps' 29418'-mavx512vnni' 29419'-mavxvnni' 29420'-mavx5124vnniw' 29421'-mcldemote' 29422'-mserialize' 29423'-mamx-tile' 29424'-mamx-int8' 29425'-mamx-bf16' 29426'-mhreset' 29427'-mkl' 29428'-mwidekl' 29429 These switches enable the use of instructions in the MMX, SSE, 29430 SSE2, SSE3, SSSE3, SSE4, SSE4A, SSE4.1, SSE4.2, AVX, AVX2, AVX512F, 29431 AVX512PF, AVX512ER, AVX512CD, AVX512VL, AVX512BW, AVX512DQ, 29432 AVX512IFMA, AVX512VBMI, SHA, AES, PCLMUL, CLFLUSHOPT, CLWB, 29433 FSGSBASE, PTWRITE, RDRND, F16C, FMA, PCONFIG, WBNOINVD, FMA4, 29434 PREFETCHW, RDPID, PREFETCHWT1, RDSEED, SGX, XOP, LWP, 3DNow!, 29435 enhanced 3DNow!, POPCNT, ABM, ADX, BMI, BMI2, LZCNT, FXSR, XSAVE, 29436 XSAVEOPT, XSAVEC, XSAVES, RTM, HLE, TBM, MWAITX, CLZERO, PKU, 29437 AVX512VBMI2, GFNI, VAES, WAITPKG, VPCLMULQDQ, AVX512BITALG, 29438 MOVDIRI, MOVDIR64B, AVX512BF16, ENQCMD, AVX512VPOPCNTDQ, 29439 AVX5124FMAPS, AVX512VNNI, AVX5124VNNIW, SERIALIZE, UINTR, HRESET, 29440 AMXTILE, AMXINT8, AMXBF16, KL, WIDEKL, AVXVNNI, AVX512-FP16 or 29441 CLDEMOTE extended instruction sets. Each has a corresponding 29442 '-mno-' option to disable use of these instructions. 29443 29444 These extensions are also available as built-in functions: see 29445 *note x86 Built-in Functions::, for details of the functions 29446 enabled and disabled by these switches. 29447 29448 To generate SSE/SSE2 instructions automatically from floating-point 29449 code (as opposed to 387 instructions), see '-mfpmath=sse'. 29450 29451 GCC depresses SSEx instructions when '-mavx' is used. Instead, it 29452 generates new AVX instructions or AVX equivalence for all SSEx 29453 instructions when needed. 29454 29455 These options enable GCC to use these extended instructions in 29456 generated code, even without '-mfpmath=sse'. Applications that 29457 perform run-time CPU detection must compile separate files for each 29458 supported architecture, using the appropriate flags. In 29459 particular, the file containing the CPU detection code should be 29460 compiled without these options. 29461 29462'-mdump-tune-features' 29463 This option instructs GCC to dump the names of the x86 performance 29464 tuning features and default settings. The names can be used in 29465 '-mtune-ctrl=FEATURE-LIST'. 29466 29467'-mtune-ctrl=FEATURE-LIST' 29468 This option is used to do fine grain control of x86 code generation 29469 features. FEATURE-LIST is a comma separated list of FEATURE names. 29470 See also '-mdump-tune-features'. When specified, the FEATURE is 29471 turned on if it is not preceded with '^', otherwise, it is turned 29472 off. '-mtune-ctrl=FEATURE-LIST' is intended to be used by GCC 29473 developers. Using it may lead to code paths not covered by testing 29474 and can potentially result in compiler ICEs or runtime errors. 29475 29476'-mno-default' 29477 This option instructs GCC to turn off all tunable features. See 29478 also '-mtune-ctrl=FEATURE-LIST' and '-mdump-tune-features'. 29479 29480'-mcld' 29481 This option instructs GCC to emit a 'cld' instruction in the 29482 prologue of functions that use string instructions. String 29483 instructions depend on the DF flag to select between autoincrement 29484 or autodecrement mode. While the ABI specifies the DF flag to be 29485 cleared on function entry, some operating systems violate this 29486 specification by not clearing the DF flag in their exception 29487 dispatchers. The exception handler can be invoked with the DF flag 29488 set, which leads to wrong direction mode when string instructions 29489 are used. This option can be enabled by default on 32-bit x86 29490 targets by configuring GCC with the '--enable-cld' configure 29491 option. Generation of 'cld' instructions can be suppressed with 29492 the '-mno-cld' compiler option in this case. 29493 29494'-mvzeroupper' 29495 This option instructs GCC to emit a 'vzeroupper' instruction before 29496 a transfer of control flow out of the function to minimize the AVX 29497 to SSE transition penalty as well as remove unnecessary 'zeroupper' 29498 intrinsics. 29499 29500'-mprefer-avx128' 29501 This option instructs GCC to use 128-bit AVX instructions instead 29502 of 256-bit AVX instructions in the auto-vectorizer. 29503 29504'-mprefer-vector-width=OPT' 29505 This option instructs GCC to use OPT-bit vector width in 29506 instructions instead of default on the selected platform. 29507 29508'-mmove-max=BITS' 29509 This option instructs GCC to set the maximum number of bits can be 29510 moved from memory to memory efficiently to BITS. The valid BITS 29511 are 128, 256 and 512. 29512 29513'-mstore-max=BITS' 29514 This option instructs GCC to set the maximum number of bits can be 29515 stored to memory efficiently to BITS. The valid BITS are 128, 256 29516 and 512. 29517 29518 'none' 29519 No extra limitations applied to GCC other than defined by the 29520 selected platform. 29521 29522 '128' 29523 Prefer 128-bit vector width for instructions. 29524 29525 '256' 29526 Prefer 256-bit vector width for instructions. 29527 29528 '512' 29529 Prefer 512-bit vector width for instructions. 29530 29531'-mcx16' 29532 This option enables GCC to generate 'CMPXCHG16B' instructions in 29533 64-bit code to implement compare-and-exchange operations on 16-byte 29534 aligned 128-bit objects. This is useful for atomic updates of data 29535 structures exceeding one machine word in size. The compiler uses 29536 this instruction to implement *note __sync Builtins::. However, 29537 for *note __atomic Builtins:: operating on 128-bit integers, a 29538 library call is always used. 29539 29540'-msahf' 29541 This option enables generation of 'SAHF' instructions in 64-bit 29542 code. Early Intel Pentium 4 CPUs with Intel 64 support, prior to 29543 the introduction of Pentium 4 G1 step in December 2005, lacked the 29544 'LAHF' and 'SAHF' instructions which are supported by AMD64. These 29545 are load and store instructions, respectively, for certain status 29546 flags. In 64-bit mode, the 'SAHF' instruction is used to optimize 29547 'fmod', 'drem', and 'remainder' built-in functions; see *note Other 29548 Builtins:: for details. 29549 29550'-mmovbe' 29551 This option enables use of the 'movbe' instruction to implement 29552 '__builtin_bswap32' and '__builtin_bswap64'. 29553 29554'-mshstk' 29555 The '-mshstk' option enables shadow stack built-in functions from 29556 x86 Control-flow Enforcement Technology (CET). 29557 29558'-mcrc32' 29559 This option enables built-in functions '__builtin_ia32_crc32qi', 29560 '__builtin_ia32_crc32hi', '__builtin_ia32_crc32si' and 29561 '__builtin_ia32_crc32di' to generate the 'crc32' machine 29562 instruction. 29563 29564'-mmwait' 29565 This option enables built-in functions '__builtin_ia32_monitor', 29566 and '__builtin_ia32_mwait' to generate the 'monitor' and 'mwait' 29567 machine instructions. 29568 29569'-mrecip' 29570 This option enables use of 'RCPSS' and 'RSQRTSS' instructions (and 29571 their vectorized variants 'RCPPS' and 'RSQRTPS') with an additional 29572 Newton-Raphson step to increase precision instead of 'DIVSS' and 29573 'SQRTSS' (and their vectorized variants) for single-precision 29574 floating-point arguments. These instructions are generated only 29575 when '-funsafe-math-optimizations' is enabled together with 29576 '-ffinite-math-only' and '-fno-trapping-math'. Note that while the 29577 throughput of the sequence is higher than the throughput of the 29578 non-reciprocal instruction, the precision of the sequence can be 29579 decreased by up to 2 ulp (i.e. the inverse of 1.0 equals 29580 0.99999994). 29581 29582 Note that GCC implements '1.0f/sqrtf(X)' in terms of 'RSQRTSS' (or 29583 'RSQRTPS') already with '-ffast-math' (or the above option 29584 combination), and doesn't need '-mrecip'. 29585 29586 Also note that GCC emits the above sequence with additional 29587 Newton-Raphson step for vectorized single-float division and 29588 vectorized 'sqrtf(X)' already with '-ffast-math' (or the above 29589 option combination), and doesn't need '-mrecip'. 29590 29591'-mrecip=OPT' 29592 This option controls which reciprocal estimate instructions may be 29593 used. OPT is a comma-separated list of options, which may be 29594 preceded by a '!' to invert the option: 29595 29596 'all' 29597 Enable all estimate instructions. 29598 29599 'default' 29600 Enable the default instructions, equivalent to '-mrecip'. 29601 29602 'none' 29603 Disable all estimate instructions, equivalent to '-mno-recip'. 29604 29605 'div' 29606 Enable the approximation for scalar division. 29607 29608 'vec-div' 29609 Enable the approximation for vectorized division. 29610 29611 'sqrt' 29612 Enable the approximation for scalar square root. 29613 29614 'vec-sqrt' 29615 Enable the approximation for vectorized square root. 29616 29617 So, for example, '-mrecip=all,!sqrt' enables all of the reciprocal 29618 approximations, except for square root. 29619 29620'-mveclibabi=TYPE' 29621 Specifies the ABI type to use for vectorizing intrinsics using an 29622 external library. Supported values for TYPE are 'svml' for the 29623 Intel short vector math library and 'acml' for the AMD math core 29624 library. To use this option, both '-ftree-vectorize' and 29625 '-funsafe-math-optimizations' have to be enabled, and an SVML or 29626 ACML ABI-compatible library must be specified at link time. 29627 29628 GCC currently emits calls to 'vmldExp2', 'vmldLn2', 'vmldLog102', 29629 'vmldPow2', 'vmldTanh2', 'vmldTan2', 'vmldAtan2', 'vmldAtanh2', 29630 'vmldCbrt2', 'vmldSinh2', 'vmldSin2', 'vmldAsinh2', 'vmldAsin2', 29631 'vmldCosh2', 'vmldCos2', 'vmldAcosh2', 'vmldAcos2', 'vmlsExp4', 29632 'vmlsLn4', 'vmlsLog104', 'vmlsPow4', 'vmlsTanh4', 'vmlsTan4', 29633 'vmlsAtan4', 'vmlsAtanh4', 'vmlsCbrt4', 'vmlsSinh4', 'vmlsSin4', 29634 'vmlsAsinh4', 'vmlsAsin4', 'vmlsCosh4', 'vmlsCos4', 'vmlsAcosh4' 29635 and 'vmlsAcos4' for corresponding function type when 29636 '-mveclibabi=svml' is used, and '__vrd2_sin', '__vrd2_cos', 29637 '__vrd2_exp', '__vrd2_log', '__vrd2_log2', '__vrd2_log10', 29638 '__vrs4_sinf', '__vrs4_cosf', '__vrs4_expf', '__vrs4_logf', 29639 '__vrs4_log2f', '__vrs4_log10f' and '__vrs4_powf' for the 29640 corresponding function type when '-mveclibabi=acml' is used. 29641 29642'-mabi=NAME' 29643 Generate code for the specified calling convention. Permissible 29644 values are 'sysv' for the ABI used on GNU/Linux and other systems, 29645 and 'ms' for the Microsoft ABI. The default is to use the Microsoft 29646 ABI when targeting Microsoft Windows and the SysV ABI on all other 29647 systems. You can control this behavior for specific functions by 29648 using the function attributes 'ms_abi' and 'sysv_abi'. *Note 29649 Function Attributes::. 29650 29651'-mforce-indirect-call' 29652 Force all calls to functions to be indirect. This is useful when 29653 using Intel Processor Trace where it generates more precise timing 29654 information for function calls. 29655 29656'-mmanual-endbr' 29657 Insert ENDBR instruction at function entry only via the 'cf_check' 29658 function attribute. This is useful when used with the option 29659 '-fcf-protection=branch' to control ENDBR insertion at the function 29660 entry. 29661 29662'-mcall-ms2sysv-xlogues' 29663 Due to differences in 64-bit ABIs, any Microsoft ABI function that 29664 calls a System V ABI function must consider RSI, RDI and XMM6-15 as 29665 clobbered. By default, the code for saving and restoring these 29666 registers is emitted inline, resulting in fairly lengthy prologues 29667 and epilogues. Using '-mcall-ms2sysv-xlogues' emits prologues and 29668 epilogues that use stubs in the static portion of libgcc to perform 29669 these saves and restores, thus reducing function size at the cost 29670 of a few extra instructions. 29671 29672'-mtls-dialect=TYPE' 29673 Generate code to access thread-local storage using the 'gnu' or 29674 'gnu2' conventions. 'gnu' is the conservative default; 'gnu2' is 29675 more efficient, but it may add compile- and run-time requirements 29676 that cannot be satisfied on all systems. 29677 29678'-mpush-args' 29679'-mno-push-args' 29680 Use PUSH operations to store outgoing parameters. This method is 29681 shorter and usually equally fast as method using SUB/MOV operations 29682 and is enabled by default. In some cases disabling it may improve 29683 performance because of improved scheduling and reduced 29684 dependencies. 29685 29686'-maccumulate-outgoing-args' 29687 If enabled, the maximum amount of space required for outgoing 29688 arguments is computed in the function prologue. This is faster on 29689 most modern CPUs because of reduced dependencies, improved 29690 scheduling and reduced stack usage when the preferred stack 29691 boundary is not equal to 2. The drawback is a notable increase in 29692 code size. This switch implies '-mno-push-args'. 29693 29694'-mthreads' 29695 Support thread-safe exception handling on MinGW. Programs that rely 29696 on thread-safe exception handling must compile and link all code 29697 with the '-mthreads' option. When compiling, '-mthreads' defines 29698 '-D_MT'; when linking, it links in a special thread helper library 29699 '-lmingwthrd' which cleans up per-thread exception-handling data. 29700 29701'-mms-bitfields' 29702'-mno-ms-bitfields' 29703 29704 Enable/disable bit-field layout compatible with the native 29705 Microsoft Windows compiler. 29706 29707 If 'packed' is used on a structure, or if bit-fields are used, it 29708 may be that the Microsoft ABI lays out the structure differently 29709 than the way GCC normally does. Particularly when moving packed 29710 data between functions compiled with GCC and the native Microsoft 29711 compiler (either via function call or as data in a file), it may be 29712 necessary to access either format. 29713 29714 This option is enabled by default for Microsoft Windows targets. 29715 This behavior can also be controlled locally by use of variable or 29716 type attributes. For more information, see *note x86 Variable 29717 Attributes:: and *note x86 Type Attributes::. 29718 29719 The Microsoft structure layout algorithm is fairly simple with the 29720 exception of the bit-field packing. The padding and alignment of 29721 members of structures and whether a bit-field can straddle a 29722 storage-unit boundary are determine by these rules: 29723 29724 1. Structure members are stored sequentially in the order in 29725 which they are declared: the first member has the lowest 29726 memory address and the last member the highest. 29727 29728 2. Every data object has an alignment requirement. The alignment 29729 requirement for all data except structures, unions, and arrays 29730 is either the size of the object or the current packing size 29731 (specified with either the 'aligned' attribute or the 'pack' 29732 pragma), whichever is less. For structures, unions, and 29733 arrays, the alignment requirement is the largest alignment 29734 requirement of its members. Every object is allocated an 29735 offset so that: 29736 29737 offset % alignment_requirement == 0 29738 29739 3. Adjacent bit-fields are packed into the same 1-, 2-, or 4-byte 29740 allocation unit if the integral types are the same size and if 29741 the next bit-field fits into the current allocation unit 29742 without crossing the boundary imposed by the common alignment 29743 requirements of the bit-fields. 29744 29745 MSVC interprets zero-length bit-fields in the following ways: 29746 29747 1. If a zero-length bit-field is inserted between two bit-fields 29748 that are normally coalesced, the bit-fields are not coalesced. 29749 29750 For example: 29751 29752 struct 29753 { 29754 unsigned long bf_1 : 12; 29755 unsigned long : 0; 29756 unsigned long bf_2 : 12; 29757 } t1; 29758 29759 The size of 't1' is 8 bytes with the zero-length bit-field. 29760 If the zero-length bit-field were removed, 't1''s size would 29761 be 4 bytes. 29762 29763 2. If a zero-length bit-field is inserted after a bit-field, 29764 'foo', and the alignment of the zero-length bit-field is 29765 greater than the member that follows it, 'bar', 'bar' is 29766 aligned as the type of the zero-length bit-field. 29767 29768 For example: 29769 29770 struct 29771 { 29772 char foo : 4; 29773 short : 0; 29774 char bar; 29775 } t2; 29776 29777 struct 29778 { 29779 char foo : 4; 29780 short : 0; 29781 double bar; 29782 } t3; 29783 29784 For 't2', 'bar' is placed at offset 2, rather than offset 1. 29785 Accordingly, the size of 't2' is 4. For 't3', the zero-length 29786 bit-field does not affect the alignment of 'bar' or, as a 29787 result, the size of the structure. 29788 29789 Taking this into account, it is important to note the 29790 following: 29791 29792 1. If a zero-length bit-field follows a normal bit-field, 29793 the type of the zero-length bit-field may affect the 29794 alignment of the structure as whole. For example, 't2' 29795 has a size of 4 bytes, since the zero-length bit-field 29796 follows a normal bit-field, and is of type short. 29797 29798 2. Even if a zero-length bit-field is not followed by a 29799 normal bit-field, it may still affect the alignment of 29800 the structure: 29801 29802 struct 29803 { 29804 char foo : 6; 29805 long : 0; 29806 } t4; 29807 29808 Here, 't4' takes up 4 bytes. 29809 29810 3. Zero-length bit-fields following non-bit-field members are 29811 ignored: 29812 29813 struct 29814 { 29815 char foo; 29816 long : 0; 29817 char bar; 29818 } t5; 29819 29820 Here, 't5' takes up 2 bytes. 29821 29822'-mno-align-stringops' 29823 Do not align the destination of inlined string operations. This 29824 switch reduces code size and improves performance in case the 29825 destination is already aligned, but GCC doesn't know about it. 29826 29827'-minline-all-stringops' 29828 By default GCC inlines string operations only when the destination 29829 is known to be aligned to least a 4-byte boundary. This enables 29830 more inlining and increases code size, but may improve performance 29831 of code that depends on fast 'memcpy' and 'memset' for short 29832 lengths. The option enables inline expansion of 'strlen' for all 29833 pointer alignments. 29834 29835'-minline-stringops-dynamically' 29836 For string operations of unknown size, use run-time checks with 29837 inline code for small blocks and a library call for large blocks. 29838 29839'-mstringop-strategy=ALG' 29840 Override the internal decision heuristic for the particular 29841 algorithm to use for inlining string operations. The allowed 29842 values for ALG are: 29843 29844 'rep_byte' 29845 'rep_4byte' 29846 'rep_8byte' 29847 Expand using i386 'rep' prefix of the specified size. 29848 29849 'byte_loop' 29850 'loop' 29851 'unrolled_loop' 29852 Expand into an inline loop. 29853 29854 'libcall' 29855 Always use a library call. 29856 29857'-mmemcpy-strategy=STRATEGY' 29858 Override the internal decision heuristic to decide if 29859 '__builtin_memcpy' should be inlined and what inline algorithm to 29860 use when the expected size of the copy operation is known. 29861 STRATEGY is a comma-separated list of ALG:MAX_SIZE:DEST_ALIGN 29862 triplets. ALG is specified in '-mstringop-strategy', MAX_SIZE 29863 specifies the max byte size with which inline algorithm ALG is 29864 allowed. For the last triplet, the MAX_SIZE must be '-1'. The 29865 MAX_SIZE of the triplets in the list must be specified in 29866 increasing order. The minimal byte size for ALG is '0' for the 29867 first triplet and 'MAX_SIZE + 1' of the preceding range. 29868 29869'-mmemset-strategy=STRATEGY' 29870 The option is similar to '-mmemcpy-strategy=' except that it is to 29871 control '__builtin_memset' expansion. 29872 29873'-momit-leaf-frame-pointer' 29874 Don't keep the frame pointer in a register for leaf functions. 29875 This avoids the instructions to save, set up, and restore frame 29876 pointers and makes an extra register available in leaf functions. 29877 The option '-fomit-leaf-frame-pointer' removes the frame pointer 29878 for leaf functions, which might make debugging harder. 29879 29880'-mtls-direct-seg-refs' 29881'-mno-tls-direct-seg-refs' 29882 Controls whether TLS variables may be accessed with offsets from 29883 the TLS segment register ('%gs' for 32-bit, '%fs' for 64-bit), or 29884 whether the thread base pointer must be added. Whether or not this 29885 is valid depends on the operating system, and whether it maps the 29886 segment to cover the entire TLS area. 29887 29888 For systems that use the GNU C Library, the default is on. 29889 29890'-msse2avx' 29891'-mno-sse2avx' 29892 Specify that the assembler should encode SSE instructions with VEX 29893 prefix. The option '-mavx' turns this on by default. 29894 29895'-mfentry' 29896'-mno-fentry' 29897 If profiling is active ('-pg'), put the profiling counter call 29898 before the prologue. Note: On x86 architectures the attribute 29899 'ms_hook_prologue' isn't possible at the moment for '-mfentry' and 29900 '-pg'. 29901 29902'-mrecord-mcount' 29903'-mno-record-mcount' 29904 If profiling is active ('-pg'), generate a __mcount_loc section 29905 that contains pointers to each profiling call. This is useful for 29906 automatically patching and out calls. 29907 29908'-mnop-mcount' 29909'-mno-nop-mcount' 29910 If profiling is active ('-pg'), generate the calls to the profiling 29911 functions as NOPs. This is useful when they should be patched in 29912 later dynamically. This is likely only useful together with 29913 '-mrecord-mcount'. 29914 29915'-minstrument-return=TYPE' 29916 Instrument function exit in -pg -mfentry instrumented functions 29917 with call to specified function. This only instruments true 29918 returns ending with ret, but not sibling calls ending with jump. 29919 Valid types are NONE to not instrument, CALL to generate a call to 29920 __return__, or NOP5 to generate a 5 byte nop. 29921 29922'-mrecord-return' 29923'-mno-record-return' 29924 Generate a __return_loc section pointing to all return 29925 instrumentation code. 29926 29927'-mfentry-name=NAME' 29928 Set name of __fentry__ symbol called at function entry for -pg 29929 -mfentry functions. 29930 29931'-mfentry-section=NAME' 29932 Set name of section to record -mrecord-mcount calls (default 29933 __mcount_loc). 29934 29935'-mskip-rax-setup' 29936'-mno-skip-rax-setup' 29937 When generating code for the x86-64 architecture with SSE 29938 extensions disabled, '-mskip-rax-setup' can be used to skip setting 29939 up RAX register when there are no variable arguments passed in 29940 vector registers. 29941 29942 *Warning:* Since RAX register is used to avoid unnecessarily saving 29943 vector registers on stack when passing variable arguments, the 29944 impacts of this option are callees may waste some stack space, 29945 misbehave or jump to a random location. GCC 4.4 or newer don't 29946 have those issues, regardless the RAX register value. 29947 29948'-m8bit-idiv' 29949'-mno-8bit-idiv' 29950 On some processors, like Intel Atom, 8-bit unsigned integer divide 29951 is much faster than 32-bit/64-bit integer divide. This option 29952 generates a run-time check. If both dividend and divisor are 29953 within range of 0 to 255, 8-bit unsigned integer divide is used 29954 instead of 32-bit/64-bit integer divide. 29955 29956'-mavx256-split-unaligned-load' 29957'-mavx256-split-unaligned-store' 29958 Split 32-byte AVX unaligned load and store. 29959 29960'-mstack-protector-guard=GUARD' 29961'-mstack-protector-guard-reg=REG' 29962'-mstack-protector-guard-offset=OFFSET' 29963 Generate stack protection code using canary at GUARD. Supported 29964 locations are 'global' for global canary or 'tls' for per-thread 29965 canary in the TLS block (the default). This option has effect only 29966 when '-fstack-protector' or '-fstack-protector-all' is specified. 29967 29968 With the latter choice the options 29969 '-mstack-protector-guard-reg=REG' and 29970 '-mstack-protector-guard-offset=OFFSET' furthermore specify which 29971 segment register ('%fs' or '%gs') to use as base register for 29972 reading the canary, and from what offset from that base register. 29973 The default for those is as specified in the relevant ABI. 29974 29975'-mgeneral-regs-only' 29976 Generate code that uses only the general-purpose registers. This 29977 prevents the compiler from using floating-point, vector, mask and 29978 bound registers. 29979 29980'-mrelax-cmpxchg-loop' 29981 Relax cmpxchg loop by emitting an early load and compare before 29982 cmpxchg, execute pause if load value is not expected. This reduces 29983 excessive cachline bouncing when and works for all atomic logic 29984 fetch builtins that generates compare and swap loop. 29985 29986'-mindirect-branch=CHOICE' 29987 Convert indirect call and jump with CHOICE. The default is 'keep', 29988 which keeps indirect call and jump unmodified. 'thunk' converts 29989 indirect call and jump to call and return thunk. 'thunk-inline' 29990 converts indirect call and jump to inlined call and return thunk. 29991 'thunk-extern' converts indirect call and jump to external call and 29992 return thunk provided in a separate object file. You can control 29993 this behavior for a specific function by using the function 29994 attribute 'indirect_branch'. *Note Function Attributes::. 29995 29996 Note that '-mcmodel=large' is incompatible with 29997 '-mindirect-branch=thunk' and '-mindirect-branch=thunk-extern' 29998 since the thunk function may not be reachable in the large code 29999 model. 30000 30001 Note that '-mindirect-branch=thunk-extern' is compatible with 30002 '-fcf-protection=branch' since the external thunk can be made to 30003 enable control-flow check. 30004 30005'-mfunction-return=CHOICE' 30006 Convert function return with CHOICE. The default is 'keep', which 30007 keeps function return unmodified. 'thunk' converts function return 30008 to call and return thunk. 'thunk-inline' converts function return 30009 to inlined call and return thunk. 'thunk-extern' converts function 30010 return to external call and return thunk provided in a separate 30011 object file. You can control this behavior for a specific function 30012 by using the function attribute 'function_return'. *Note Function 30013 Attributes::. 30014 30015 Note that '-mindirect-return=thunk-extern' is compatible with 30016 '-fcf-protection=branch' since the external thunk can be made to 30017 enable control-flow check. 30018 30019 Note that '-mcmodel=large' is incompatible with 30020 '-mfunction-return=thunk' and '-mfunction-return=thunk-extern' 30021 since the thunk function may not be reachable in the large code 30022 model. 30023 30024'-mindirect-branch-register' 30025 Force indirect call and jump via register. 30026 30027'-mharden-sls=CHOICE' 30028 Generate code to mitigate against straight line speculation (SLS) 30029 with CHOICE. The default is 'none' which disables all SLS 30030 hardening. 'return' enables SLS hardening for function returns. 30031 'indirect-jmp' enables SLS hardening for indirect jumps. 'all' 30032 enables all SLS hardening. 30033 30034'-mindirect-branch-cs-prefix' 30035 Add CS prefix to call and jmp to indirect thunk with branch target 30036 in r8-r15 registers so that the call and jmp instruction length is 30037 6 bytes to allow them to be replaced with 'lfence; call *%r8-r15' 30038 or 'lfence; jmp *%r8-r15' at run-time. 30039 30040 These '-m' switches are supported in addition to the above on x86-64 30041processors in 64-bit environments. 30042 30043'-m32' 30044'-m64' 30045'-mx32' 30046'-m16' 30047'-miamcu' 30048 Generate code for a 16-bit, 32-bit or 64-bit environment. The 30049 '-m32' option sets 'int', 'long', and pointer types to 32 bits, and 30050 generates code that runs on any i386 system. 30051 30052 The '-m64' option sets 'int' to 32 bits and 'long' and pointer 30053 types to 64 bits, and generates code for the x86-64 architecture. 30054 For Darwin only the '-m64' option also turns off the '-fno-pic' and 30055 '-mdynamic-no-pic' options. 30056 30057 The '-mx32' option sets 'int', 'long', and pointer types to 32 30058 bits, and generates code for the x86-64 architecture. 30059 30060 The '-m16' option is the same as '-m32', except for that it outputs 30061 the '.code16gcc' assembly directive at the beginning of the 30062 assembly output so that the binary can run in 16-bit mode. 30063 30064 The '-miamcu' option generates code which conforms to Intel MCU 30065 psABI. It requires the '-m32' option to be turned on. 30066 30067'-mno-red-zone' 30068 Do not use a so-called "red zone" for x86-64 code. The red zone is 30069 mandated by the x86-64 ABI; it is a 128-byte area beyond the 30070 location of the stack pointer that is not modified by signal or 30071 interrupt handlers and therefore can be used for temporary data 30072 without adjusting the stack pointer. The flag '-mno-red-zone' 30073 disables this red zone. 30074 30075'-mcmodel=small' 30076 Generate code for the small code model: the program and its symbols 30077 must be linked in the lower 2 GB of the address space. Pointers 30078 are 64 bits. Programs can be statically or dynamically linked. 30079 This is the default code model. 30080 30081'-mcmodel=kernel' 30082 Generate code for the kernel code model. The kernel runs in the 30083 negative 2 GB of the address space. This model has to be used for 30084 Linux kernel code. 30085 30086'-mcmodel=medium' 30087 Generate code for the medium model: the program is linked in the 30088 lower 2 GB of the address space. Small symbols are also placed 30089 there. Symbols with sizes larger than '-mlarge-data-threshold' are 30090 put into large data or BSS sections and can be located above 2GB. 30091 Programs can be statically or dynamically linked. 30092 30093'-mcmodel=large' 30094 Generate code for the large model. This model makes no assumptions 30095 about addresses and sizes of sections. 30096 30097'-maddress-mode=long' 30098 Generate code for long address mode. This is only supported for 30099 64-bit and x32 environments. It is the default address mode for 30100 64-bit environments. 30101 30102'-maddress-mode=short' 30103 Generate code for short address mode. This is only supported for 30104 32-bit and x32 environments. It is the default address mode for 30105 32-bit and x32 environments. 30106 30107'-mneeded' 30108'-mno-needed' 30109 Emit GNU_PROPERTY_X86_ISA_1_NEEDED GNU property for Linux target to 30110 indicate the micro-architecture ISA level required to execute the 30111 binary. 30112 30113'-mno-direct-extern-access' 30114 Without '-fpic' nor '-fPIC', always use the GOT pointer to access 30115 external symbols. With '-fpic' or '-fPIC', treat access to 30116 protected symbols as local symbols. The default is 30117 '-mdirect-extern-access'. 30118 30119 *Warning:* shared libraries compiled with 30120 '-mno-direct-extern-access' and executable compiled with 30121 '-mdirect-extern-access' may not be binary compatible if protected 30122 symbols are used in shared libraries and executable. 30123 30124 30125File: gcc.info, Node: x86 Windows Options, Next: Xstormy16 Options, Prev: x86 Options, Up: Submodel Options 30126 301273.19.61 x86 Windows Options 30128--------------------------- 30129 30130These additional options are available for Microsoft Windows targets: 30131 30132'-mconsole' 30133 This option specifies that a console application is to be 30134 generated, by instructing the linker to set the PE header subsystem 30135 type required for console applications. This option is available 30136 for Cygwin and MinGW targets and is enabled by default on those 30137 targets. 30138 30139'-mdll' 30140 This option is available for Cygwin and MinGW targets. It 30141 specifies that a DLL--a dynamic link library--is to be generated, 30142 enabling the selection of the required runtime startup object and 30143 entry point. 30144 30145'-mnop-fun-dllimport' 30146 This option is available for Cygwin and MinGW targets. It 30147 specifies that the 'dllimport' attribute should be ignored. 30148 30149'-mthreads' 30150 This option is available for MinGW targets. It specifies that 30151 MinGW-specific thread support is to be used. 30152 30153'-municode' 30154 This option is available for MinGW-w64 targets. It causes the 30155 'UNICODE' preprocessor macro to be predefined, and chooses 30156 Unicode-capable runtime startup code. 30157 30158'-mwin32' 30159 This option is available for Cygwin and MinGW targets. It 30160 specifies that the typical Microsoft Windows predefined macros are 30161 to be set in the pre-processor, but does not influence the choice 30162 of runtime library/startup code. 30163 30164'-mwindows' 30165 This option is available for Cygwin and MinGW targets. It 30166 specifies that a GUI application is to be generated by instructing 30167 the linker to set the PE header subsystem type appropriately. 30168 30169'-fno-set-stack-executable' 30170 This option is available for MinGW targets. It specifies that the 30171 executable flag for the stack used by nested functions isn't set. 30172 This is necessary for binaries running in kernel mode of Microsoft 30173 Windows, as there the User32 API, which is used to set executable 30174 privileges, isn't available. 30175 30176'-fwritable-relocated-rdata' 30177 This option is available for MinGW and Cygwin targets. It 30178 specifies that relocated-data in read-only section is put into the 30179 '.data' section. This is a necessary for older runtimes not 30180 supporting modification of '.rdata' sections for pseudo-relocation. 30181 30182'-mpe-aligned-commons' 30183 This option is available for Cygwin and MinGW targets. It 30184 specifies that the GNU extension to the PE file format that permits 30185 the correct alignment of COMMON variables should be used when 30186 generating code. It is enabled by default if GCC detects that the 30187 target assembler found during configuration supports the feature. 30188 30189 See also under *note x86 Options:: for standard options. 30190 30191 30192File: gcc.info, Node: Xstormy16 Options, Next: Xtensa Options, Prev: x86 Windows Options, Up: Submodel Options 30193 301943.19.62 Xstormy16 Options 30195------------------------- 30196 30197These options are defined for Xstormy16: 30198 30199'-msim' 30200 Choose startup files and linker script suitable for the simulator. 30201 30202 30203File: gcc.info, Node: Xtensa Options, Next: zSeries Options, Prev: Xstormy16 Options, Up: Submodel Options 30204 302053.19.63 Xtensa Options 30206---------------------- 30207 30208These options are supported for Xtensa targets: 30209 30210'-mconst16' 30211'-mno-const16' 30212 Enable or disable use of 'CONST16' instructions for loading 30213 constant values. The 'CONST16' instruction is currently not a 30214 standard option from Tensilica. When enabled, 'CONST16' 30215 instructions are always used in place of the standard 'L32R' 30216 instructions. The use of 'CONST16' is enabled by default only if 30217 the 'L32R' instruction is not available. 30218 30219'-mfused-madd' 30220'-mno-fused-madd' 30221 Enable or disable use of fused multiply/add and multiply/subtract 30222 instructions in the floating-point option. This has no effect if 30223 the floating-point option is not also enabled. Disabling fused 30224 multiply/add and multiply/subtract instructions forces the compiler 30225 to use separate instructions for the multiply and add/subtract 30226 operations. This may be desirable in some cases where strict IEEE 30227 754-compliant results are required: the fused multiply add/subtract 30228 instructions do not round the intermediate result, thereby 30229 producing results with _more_ bits of precision than specified by 30230 the IEEE standard. Disabling fused multiply add/subtract 30231 instructions also ensures that the program output is not sensitive 30232 to the compiler's ability to combine multiply and add/subtract 30233 operations. 30234 30235'-mserialize-volatile' 30236'-mno-serialize-volatile' 30237 When this option is enabled, GCC inserts 'MEMW' instructions before 30238 'volatile' memory references to guarantee sequential consistency. 30239 The default is '-mserialize-volatile'. Use 30240 '-mno-serialize-volatile' to omit the 'MEMW' instructions. 30241 30242'-mforce-no-pic' 30243 For targets, like GNU/Linux, where all user-mode Xtensa code must 30244 be position-independent code (PIC), this option disables PIC for 30245 compiling kernel code. 30246 30247'-mtext-section-literals' 30248'-mno-text-section-literals' 30249 These options control the treatment of literal pools. The default 30250 is '-mno-text-section-literals', which places literals in a 30251 separate section in the output file. This allows the literal pool 30252 to be placed in a data RAM/ROM, and it also allows the linker to 30253 combine literal pools from separate object files to remove 30254 redundant literals and improve code size. With 30255 '-mtext-section-literals', the literals are interspersed in the 30256 text section in order to keep them as close as possible to their 30257 references. This may be necessary for large assembly files. 30258 Literals for each function are placed right before that function. 30259 30260'-mauto-litpools' 30261'-mno-auto-litpools' 30262 These options control the treatment of literal pools. The default 30263 is '-mno-auto-litpools', which places literals in a separate 30264 section in the output file unless '-mtext-section-literals' is 30265 used. With '-mauto-litpools' the literals are interspersed in the 30266 text section by the assembler. Compiler does not produce explicit 30267 '.literal' directives and loads literals into registers with 'MOVI' 30268 instructions instead of 'L32R' to let the assembler do relaxation 30269 and place literals as necessary. This option allows assembler to 30270 create several literal pools per function and assemble very big 30271 functions, which may not be possible with 30272 '-mtext-section-literals'. 30273 30274'-mtarget-align' 30275'-mno-target-align' 30276 When this option is enabled, GCC instructs the assembler to 30277 automatically align instructions to reduce branch penalties at the 30278 expense of some code density. The assembler attempts to widen 30279 density instructions to align branch targets and the instructions 30280 following call instructions. If there are not enough preceding 30281 safe density instructions to align a target, no widening is 30282 performed. The default is '-mtarget-align'. These options do not 30283 affect the treatment of auto-aligned instructions like 'LOOP', 30284 which the assembler always aligns, either by widening density 30285 instructions or by inserting NOP instructions. 30286 30287'-mlongcalls' 30288'-mno-longcalls' 30289 When this option is enabled, GCC instructs the assembler to 30290 translate direct calls to indirect calls unless it can determine 30291 that the target of a direct call is in the range allowed by the 30292 call instruction. This translation typically occurs for calls to 30293 functions in other source files. Specifically, the assembler 30294 translates a direct 'CALL' instruction into an 'L32R' followed by a 30295 'CALLX' instruction. The default is '-mno-longcalls'. This option 30296 should be used in programs where the call target can potentially be 30297 out of range. This option is implemented in the assembler, not the 30298 compiler, so the assembly code generated by GCC still shows direct 30299 call instructions--look at the disassembled object code to see the 30300 actual instructions. Note that the assembler uses an indirect call 30301 for every cross-file call, not just those that really are out of 30302 range. 30303 30304'-mabi=NAME' 30305 Generate code for the specified ABI. Permissible values are: 30306 'call0', 'windowed'. Default ABI is chosen by the Xtensa core 30307 configuration. 30308 30309'-mabi=call0' 30310 When this option is enabled function parameters are passed in 30311 registers 'a2' through 'a7', registers 'a12' through 'a15' are 30312 caller-saved, and register 'a15' may be used as a frame pointer. 30313 When this version of the ABI is enabled the C preprocessor symbol 30314 '__XTENSA_CALL0_ABI__' is defined. 30315 30316'-mabi=windowed' 30317 When this option is enabled function parameters are passed in 30318 registers 'a10' through 'a15', and called function rotates register 30319 window by 8 registers on entry so that its arguments are found in 30320 registers 'a2' through 'a7'. Register 'a7' may be used as a frame 30321 pointer. Register window is rotated 8 registers back upon return. 30322 When this version of the ABI is enabled the C preprocessor symbol 30323 '__XTENSA_WINDOWED_ABI__' is defined. 30324 30325 30326File: gcc.info, Node: zSeries Options, Prev: Xtensa Options, Up: Submodel Options 30327 303283.19.64 zSeries Options 30329----------------------- 30330 30331These are listed under *Note S/390 and zSeries Options::. 30332 30333 30334File: gcc.info, Node: Spec Files, Next: Environment Variables, Prev: Submodel Options, Up: Invoking GCC 30335 303363.20 Specifying Subprocesses and the Switches to Pass to Them 30337============================================================= 30338 30339'gcc' is a driver program. It performs its job by invoking a sequence 30340of other programs to do the work of compiling, assembling and linking. 30341GCC interprets its command-line parameters and uses these to deduce 30342which programs it should invoke, and which command-line options it ought 30343to place on their command lines. This behavior is controlled by "spec 30344strings". In most cases there is one spec string for each program that 30345GCC can invoke, but a few programs have multiple spec strings to control 30346their behavior. The spec strings built into GCC can be overridden by 30347using the '-specs=' command-line switch to specify a spec file. 30348 30349 "Spec files" are plain-text files that are used to construct spec 30350strings. They consist of a sequence of directives separated by blank 30351lines. The type of directive is determined by the first non-whitespace 30352character on the line, which can be one of the following: 30353 30354'%COMMAND' 30355 Issues a COMMAND to the spec file processor. The commands that can 30356 appear here are: 30357 30358 '%include <FILE>' 30359 Search for FILE and insert its text at the current point in 30360 the specs file. 30361 30362 '%include_noerr <FILE>' 30363 Just like '%include', but do not generate an error message if 30364 the include file cannot be found. 30365 30366 '%rename OLD_NAME NEW_NAME' 30367 Rename the spec string OLD_NAME to NEW_NAME. 30368 30369'*[SPEC_NAME]:' 30370 This tells the compiler to create, override or delete the named 30371 spec string. All lines after this directive up to the next 30372 directive or blank line are considered to be the text for the spec 30373 string. If this results in an empty string then the spec is 30374 deleted. (Or, if the spec did not exist, then nothing happens.) 30375 Otherwise, if the spec does not currently exist a new spec is 30376 created. If the spec does exist then its contents are overridden 30377 by the text of this directive, unless the first character of that 30378 text is the '+' character, in which case the text is appended to 30379 the spec. 30380 30381'[SUFFIX]:' 30382 Creates a new '[SUFFIX] spec' pair. All lines after this directive 30383 and up to the next directive or blank line are considered to make 30384 up the spec string for the indicated suffix. When the compiler 30385 encounters an input file with the named suffix, it processes the 30386 spec string in order to work out how to compile that file. For 30387 example: 30388 30389 .ZZ: 30390 z-compile -input %i 30391 30392 This says that any input file whose name ends in '.ZZ' should be 30393 passed to the program 'z-compile', which should be invoked with the 30394 command-line switch '-input' and with the result of performing the 30395 '%i' substitution. (See below.) 30396 30397 As an alternative to providing a spec string, the text following a 30398 suffix directive can be one of the following: 30399 30400 '@LANGUAGE' 30401 This says that the suffix is an alias for a known LANGUAGE. 30402 This is similar to using the '-x' command-line switch to GCC 30403 to specify a language explicitly. For example: 30404 30405 .ZZ: 30406 @c++ 30407 30408 Says that .ZZ files are, in fact, C++ source files. 30409 30410 '#NAME' 30411 This causes an error messages saying: 30412 30413 NAME compiler not installed on this system. 30414 30415 GCC already has an extensive list of suffixes built into it. This 30416 directive adds an entry to the end of the list of suffixes, but 30417 since the list is searched from the end backwards, it is 30418 effectively possible to override earlier entries using this 30419 technique. 30420 30421 GCC has the following spec strings built into it. Spec files can 30422override these strings or create their own. Note that individual 30423targets can also add their own spec strings to this list. 30424 30425 asm Options to pass to the assembler 30426 asm_final Options to pass to the assembler post-processor 30427 cpp Options to pass to the C preprocessor 30428 cc1 Options to pass to the C compiler 30429 cc1plus Options to pass to the C++ compiler 30430 endfile Object files to include at the end of the link 30431 link Options to pass to the linker 30432 lib Libraries to include on the command line to the linker 30433 libgcc Decides which GCC support library to pass to the linker 30434 linker Sets the name of the linker 30435 predefines Defines to be passed to the C preprocessor 30436 signed_char Defines to pass to CPP to say whether char is signed 30437 by default 30438 startfile Object files to include at the start of the link 30439 30440 Here is a small example of a spec file: 30441 30442 %rename lib old_lib 30443 30444 *lib: 30445 --start-group -lgcc -lc -leval1 --end-group %(old_lib) 30446 30447 This example renames the spec called 'lib' to 'old_lib' and then 30448overrides the previous definition of 'lib' with a new one. The new 30449definition adds in some extra command-line options before including the 30450text of the old definition. 30451 30452 "Spec strings" are a list of command-line options to be passed to their 30453corresponding program. In addition, the spec strings can contain 30454'%'-prefixed sequences to substitute variable text or to conditionally 30455insert text into the command line. Using these constructs it is 30456possible to generate quite complex command lines. 30457 30458 Here is a table of all defined '%'-sequences for spec strings. Note 30459that spaces are not generated automatically around the results of 30460expanding these sequences. Therefore you can concatenate them together 30461or combine them with constant text in a single argument. 30462 30463'%%' 30464 Substitute one '%' into the program name or argument. 30465 30466'%"' 30467 Substitute an empty argument. 30468 30469'%i' 30470 Substitute the name of the input file being processed. 30471 30472'%b' 30473 Substitute the basename for outputs related with the input file 30474 being processed. This is often the substring up to (and not 30475 including) the last period and not including the directory but, 30476 unless %w is active, it expands to the basename for auxiliary 30477 outputs, which may be influenced by an explicit output name, and by 30478 various other options that control how auxiliary outputs are named. 30479 30480'%B' 30481 This is the same as '%b', but include the file suffix (text after 30482 the last period). Without %w, it expands to the basename for dump 30483 outputs. 30484 30485'%d' 30486 Marks the argument containing or following the '%d' as a temporary 30487 file name, so that that file is deleted if GCC exits successfully. 30488 Unlike '%g', this contributes no text to the argument. 30489 30490'%gSUFFIX' 30491 Substitute a file name that has suffix SUFFIX and is chosen once 30492 per compilation, and mark the argument in the same way as '%d'. To 30493 reduce exposure to denial-of-service attacks, the file name is now 30494 chosen in a way that is hard to predict even when previously chosen 30495 file names are known. For example, '%g.s ... %g.o ... %g.s' might 30496 turn into 'ccUVUUAU.s ccXYAXZ12.o ccUVUUAU.s'. SUFFIX matches the 30497 regexp '[.A-Za-z]*' or the special string '%O', which is treated 30498 exactly as if '%O' had been preprocessed. Previously, '%g' was 30499 simply substituted with a file name chosen once per compilation, 30500 without regard to any appended suffix (which was therefore treated 30501 just like ordinary text), making such attacks more likely to 30502 succeed. 30503 30504'%uSUFFIX' 30505 Like '%g', but generates a new temporary file name each time it 30506 appears instead of once per compilation. 30507 30508'%USUFFIX' 30509 Substitutes the last file name generated with '%uSUFFIX', 30510 generating a new one if there is no such last file name. In the 30511 absence of any '%uSUFFIX', this is just like '%gSUFFIX', except 30512 they don't share the same suffix _space_, so '%g.s ... %U.s ... 30513 %g.s ... %U.s' involves the generation of two distinct file names, 30514 one for each '%g.s' and another for each '%U.s'. Previously, '%U' 30515 was simply substituted with a file name chosen for the previous 30516 '%u', without regard to any appended suffix. 30517 30518'%jSUFFIX' 30519 Substitutes the name of the 'HOST_BIT_BUCKET', if any, and if it is 30520 writable, and if '-save-temps' is not used; otherwise, substitute 30521 the name of a temporary file, just like '%u'. This temporary file 30522 is not meant for communication between processes, but rather as a 30523 junk disposal mechanism. 30524 30525'%|SUFFIX' 30526'%mSUFFIX' 30527 Like '%g', except if '-pipe' is in effect. In that case '%|' 30528 substitutes a single dash and '%m' substitutes nothing at all. 30529 These are the two most common ways to instruct a program that it 30530 should read from standard input or write to standard output. If 30531 you need something more elaborate you can use an '%{pipe:'X'}' 30532 construct: see for example 'gcc/fortran/lang-specs.h'. 30533 30534'%.SUFFIX' 30535 Substitutes .SUFFIX for the suffixes of a matched switch's args 30536 when it is subsequently output with '%*'. SUFFIX is terminated by 30537 the next space or %. 30538 30539'%w' 30540 Marks the argument containing or following the '%w' as the 30541 designated output file of this compilation. This puts the argument 30542 into the sequence of arguments that '%o' substitutes. 30543 30544'%V' 30545 Indicates that this compilation produces no output file. 30546 30547'%o' 30548 Substitutes the names of all the output files, with spaces 30549 automatically placed around them. You should write spaces around 30550 the '%o' as well or the results are undefined. '%o' is for use in 30551 the specs for running the linker. Input files whose names have no 30552 recognized suffix are not compiled at all, but they are included 30553 among the output files, so they are linked. 30554 30555'%O' 30556 Substitutes the suffix for object files. Note that this is handled 30557 specially when it immediately follows '%g, %u, or %U', because of 30558 the need for those to form complete file names. The handling is 30559 such that '%O' is treated exactly as if it had already been 30560 substituted, except that '%g, %u, and %U' do not currently support 30561 additional SUFFIX characters following '%O' as they do following, 30562 for example, '.o'. 30563 30564'%I' 30565 Substitute any of '-iprefix' (made from 'GCC_EXEC_PREFIX'), 30566 '-isysroot' (made from 'TARGET_SYSTEM_ROOT'), '-isystem' (made from 30567 'COMPILER_PATH' and '-B' options) and '-imultilib' as necessary. 30568 30569'%s' 30570 Current argument is the name of a library or startup file of some 30571 sort. Search for that file in a standard list of directories and 30572 substitute the full name found. The current working directory is 30573 included in the list of directories scanned. 30574 30575'%T' 30576 Current argument is the name of a linker script. Search for that 30577 file in the current list of directories to scan for libraries. If 30578 the file is located insert a '--script' option into the command 30579 line followed by the full path name found. If the file is not 30580 found then generate an error message. Note: the current working 30581 directory is not searched. 30582 30583'%eSTR' 30584 Print STR as an error message. STR is terminated by a newline. 30585 Use this when inconsistent options are detected. 30586 30587'%nSTR' 30588 Print STR as a notice. STR is terminated by a newline. 30589 30590'%(NAME)' 30591 Substitute the contents of spec string NAME at this point. 30592 30593'%x{OPTION}' 30594 Accumulate an option for '%X'. 30595 30596'%X' 30597 Output the accumulated linker options specified by a '%x' spec 30598 string. 30599 30600'%Y' 30601 Output the accumulated assembler options specified by '-Wa'. 30602 30603'%Z' 30604 Output the accumulated preprocessor options specified by '-Wp'. 30605 30606'%M' 30607 Output 'multilib_os_dir'. 30608 30609'%R' 30610 Output the concatenation of 'target_system_root' and 30611 'target_sysroot_suffix'. 30612 30613'%a' 30614 Process the 'asm' spec. This is used to compute the switches to be 30615 passed to the assembler. 30616 30617'%A' 30618 Process the 'asm_final' spec. This is a spec string for passing 30619 switches to an assembler post-processor, if such a program is 30620 needed. 30621 30622'%l' 30623 Process the 'link' spec. This is the spec for computing the 30624 command line passed to the linker. Typically it makes use of the 30625 '%L %G %S %D and %E' sequences. 30626 30627'%D' 30628 Dump out a '-L' option for each directory that GCC believes might 30629 contain startup files. If the target supports multilibs then the 30630 current multilib directory is prepended to each of these paths. 30631 30632'%L' 30633 Process the 'lib' spec. This is a spec string for deciding which 30634 libraries are included on the command line to the linker. 30635 30636'%G' 30637 Process the 'libgcc' spec. This is a spec string for deciding 30638 which GCC support library is included on the command line to the 30639 linker. 30640 30641'%S' 30642 Process the 'startfile' spec. This is a spec for deciding which 30643 object files are the first ones passed to the linker. Typically 30644 this might be a file named 'crt0.o'. 30645 30646'%E' 30647 Process the 'endfile' spec. This is a spec string that specifies 30648 the last object files that are passed to the linker. 30649 30650'%C' 30651 Process the 'cpp' spec. This is used to construct the arguments to 30652 be passed to the C preprocessor. 30653 30654'%1' 30655 Process the 'cc1' spec. This is used to construct the options to 30656 be passed to the actual C compiler ('cc1'). 30657 30658'%2' 30659 Process the 'cc1plus' spec. This is used to construct the options 30660 to be passed to the actual C++ compiler ('cc1plus'). 30661 30662'%*' 30663 Substitute the variable part of a matched option. See below. Note 30664 that each comma in the substituted string is replaced by a single 30665 space. 30666 30667'%<S' 30668 Remove all occurrences of '-S' from the command line. Note--this 30669 command is position dependent. '%' commands in the spec string 30670 before this one see '-S', '%' commands in the spec string after 30671 this one do not. 30672 30673'%<S*' 30674 Similar to '%<S', but match all switches beginning with '-S'. 30675 30676'%>S' 30677 Similar to '%<S', but keep '-S' in the GCC command line. 30678 30679'%:FUNCTION(ARGS)' 30680 Call the named function FUNCTION, passing it ARGS. ARGS is first 30681 processed as a nested spec string, then split into an argument 30682 vector in the usual fashion. The function returns a string which 30683 is processed as if it had appeared literally as part of the current 30684 spec. 30685 30686 The following built-in spec functions are provided: 30687 30688 'getenv' 30689 The 'getenv' spec function takes two arguments: an environment 30690 variable name and a string. If the environment variable is 30691 not defined, a fatal error is issued. Otherwise, the return 30692 value is the value of the environment variable concatenated 30693 with the string. For example, if 'TOPDIR' is defined as 30694 '/path/to/top', then: 30695 30696 %:getenv(TOPDIR /include) 30697 30698 expands to '/path/to/top/include'. 30699 30700 'if-exists' 30701 The 'if-exists' spec function takes one argument, an absolute 30702 pathname to a file. If the file exists, 'if-exists' returns 30703 the pathname. Here is a small example of its usage: 30704 30705 *startfile: 30706 crt0%O%s %:if-exists(crti%O%s) crtbegin%O%s 30707 30708 'if-exists-else' 30709 The 'if-exists-else' spec function is similar to the 30710 'if-exists' spec function, except that it takes two arguments. 30711 The first argument is an absolute pathname to a file. If the 30712 file exists, 'if-exists-else' returns the pathname. If it 30713 does not exist, it returns the second argument. This way, 30714 'if-exists-else' can be used to select one file or another, 30715 based on the existence of the first. Here is a small example 30716 of its usage: 30717 30718 *startfile: 30719 crt0%O%s %:if-exists(crti%O%s) \ 30720 %:if-exists-else(crtbeginT%O%s crtbegin%O%s) 30721 30722 'if-exists-then-else' 30723 The 'if-exists-then-else' spec function takes at least two 30724 arguments and an optional third one. The first argument is an 30725 absolute pathname to a file. If the file exists, the function 30726 returns the second argument. If the file does not exist, the 30727 function returns the third argument if there is one, or NULL 30728 otherwise. This can be used to expand one text, or optionally 30729 another, based on the existence of a file. Here is a small 30730 example of its usage: 30731 30732 -l%:if-exists-then-else(%:getenv(VSB_DIR rtnet.h) rtnet net) 30733 30734 'sanitize' 30735 The 'sanitize' spec function takes no arguments. It returns 30736 non-NULL if any address, thread or undefined behavior 30737 sanitizers are active. 30738 30739 %{%:sanitize(address):-funwind-tables} 30740 30741 'replace-outfile' 30742 The 'replace-outfile' spec function takes two arguments. It 30743 looks for the first argument in the outfiles array and 30744 replaces it with the second argument. Here is a small example 30745 of its usage: 30746 30747 %{fgnu-runtime:%:replace-outfile(-lobjc -lobjc-gnu)} 30748 30749 'remove-outfile' 30750 The 'remove-outfile' spec function takes one argument. It 30751 looks for the first argument in the outfiles array and removes 30752 it. Here is a small example its usage: 30753 30754 %:remove-outfile(-lm) 30755 30756 'version-compare' 30757 The 'version-compare' spec function takes four or five 30758 arguments of the following form: 30759 30760 <comparison-op> <arg1> [<arg2>] <switch> <result> 30761 30762 It returns 'result' if the comparison evaluates to true, and 30763 NULL if it doesn't. The supported 'comparison-op' values are: 30764 30765 '>=' 30766 True if 'switch' is a later (or same) version than 'arg1' 30767 30768 '!>' 30769 Opposite of '>=' 30770 30771 '<' 30772 True if 'switch' is an earlier version than 'arg1' 30773 30774 '!<' 30775 Opposite of '<' 30776 30777 '><' 30778 True if 'switch' is 'arg1' or later, and earlier than 30779 'arg2' 30780 30781 '<>' 30782 True if 'switch' is earlier than 'arg1', or is 'arg2' or 30783 later 30784 30785 If the 'switch' is not present at all, the condition is false 30786 unless the first character of the 'comparison-op' is '!'. 30787 30788 %:version-compare(>= 10.3 mmacosx-version-min= -lmx) 30789 30790 The above example would add '-lmx' if 30791 '-mmacosx-version-min=10.3.9' was passed. 30792 30793 'include' 30794 The 'include' spec function behaves much like '%include', with 30795 the advantage that it can be nested inside a spec and thus be 30796 conditionalized. It takes one argument, the filename, and 30797 looks for it in the startfile path. It always returns NULL. 30798 30799 %{static-libasan|static:%:include(libsanitizer.spec)%(link_libasan)} 30800 30801 'pass-through-libs' 30802 The 'pass-through-libs' spec function takes any number of 30803 arguments. It finds any '-l' options and any non-options 30804 ending in '.a' (which it assumes are the names of linker input 30805 library archive files) and returns a result containing all the 30806 found arguments each prepended by '-plugin-opt=-pass-through=' 30807 and joined by spaces. This list is intended to be passed to 30808 the LTO linker plugin. 30809 30810 %:pass-through-libs(%G %L %G) 30811 30812 'print-asm-header' 30813 The 'print-asm-header' function takes no arguments and simply 30814 prints a banner like: 30815 30816 Assembler options 30817 ================= 30818 30819 Use "-Wa,OPTION" to pass "OPTION" to the assembler. 30820 30821 It is used to separate compiler options from assembler options 30822 in the '--target-help' output. 30823 30824 'gt' 30825 The 'gt' spec function takes two or more arguments. It 30826 returns '""' (the empty string) if the second-to-last argument 30827 is greater than the last argument, and NULL otherwise. The 30828 following example inserts the 'link_gomp' spec if the last 30829 '-ftree-parallelize-loops=' option given on the command line 30830 is greater than 1: 30831 30832 %{%:gt(%{ftree-parallelize-loops=*:%*} 1):%:include(libgomp.spec)%(link_gomp)} 30833 30834 'debug-level-gt' 30835 The 'debug-level-gt' spec function takes one argument and 30836 returns '""' (the empty string) if 'debug_info_level' is 30837 greater than the specified number, and NULL otherwise. 30838 30839 %{%:debug-level-gt(0):%{gdwarf*:--gdwarf2}} 30840 30841'%{S}' 30842 Substitutes the '-S' switch, if that switch is given to GCC. If 30843 that switch is not specified, this substitutes nothing. Note that 30844 the leading dash is omitted when specifying this option, and it is 30845 automatically inserted if the substitution is performed. Thus the 30846 spec string '%{foo}' matches the command-line option '-foo' and 30847 outputs the command-line option '-foo'. 30848 30849'%W{S}' 30850 Like %{'S'} but mark last argument supplied within as a file to be 30851 deleted on failure. 30852 30853'%@{S}' 30854 Like %{'S'} but puts the result into a 'FILE' and substitutes 30855 '@FILE' if an '@file' argument has been supplied. 30856 30857'%{S*}' 30858 Substitutes all the switches specified to GCC whose names start 30859 with '-S', but which also take an argument. This is used for 30860 switches like '-o', '-D', '-I', etc. GCC considers '-o foo' as 30861 being one switch whose name starts with 'o'. %{o*} substitutes 30862 this text, including the space. Thus two arguments are generated. 30863 30864'%{S*&T*}' 30865 Like %{'S'*}, but preserve order of 'S' and 'T' options (the order 30866 of 'S' and 'T' in the spec is not significant). There can be any 30867 number of ampersand-separated variables; for each the wild card is 30868 optional. Useful for CPP as '%{D*&U*&A*}'. 30869 30870'%{S:X}' 30871 Substitutes 'X', if the '-S' switch is given to GCC. 30872 30873'%{!S:X}' 30874 Substitutes 'X', if the '-S' switch is _not_ given to GCC. 30875 30876'%{S*:X}' 30877 Substitutes 'X' if one or more switches whose names start with '-S' 30878 are specified to GCC. Normally 'X' is substituted only once, no 30879 matter how many such switches appeared. However, if '%*' appears 30880 somewhere in 'X', then 'X' is substituted once for each matching 30881 switch, with the '%*' replaced by the part of that switch matching 30882 the '*'. 30883 30884 If '%*' appears as the last part of a spec sequence then a space is 30885 added after the end of the last substitution. If there is more 30886 text in the sequence, however, then a space is not generated. This 30887 allows the '%*' substitution to be used as part of a larger string. 30888 For example, a spec string like this: 30889 30890 %{mcu=*:--script=%*/memory.ld} 30891 30892 when matching an option like '-mcu=newchip' produces: 30893 30894 --script=newchip/memory.ld 30895 30896'%{.S:X}' 30897 Substitutes 'X', if processing a file with suffix 'S'. 30898 30899'%{!.S:X}' 30900 Substitutes 'X', if _not_ processing a file with suffix 'S'. 30901 30902'%{,S:X}' 30903 Substitutes 'X', if processing a file for language 'S'. 30904 30905'%{!,S:X}' 30906 Substitutes 'X', if not processing a file for language 'S'. 30907 30908'%{S|P:X}' 30909 Substitutes 'X' if either '-S' or '-P' is given to GCC. This may 30910 be combined with '!', '.', ',', and '*' sequences as well, although 30911 they have a stronger binding than the '|'. If '%*' appears in 'X', 30912 all of the alternatives must be starred, and only the first 30913 matching alternative is substituted. 30914 30915 For example, a spec string like this: 30916 30917 %{.c:-foo} %{!.c:-bar} %{.c|d:-baz} %{!.c|d:-boggle} 30918 30919 outputs the following command-line options from the following input 30920 command-line options: 30921 30922 fred.c -foo -baz 30923 jim.d -bar -boggle 30924 -d fred.c -foo -baz -boggle 30925 -d jim.d -bar -baz -boggle 30926 30927'%{%:FUNCTION(ARGS):X}' 30928 30929 Call function named FUNCTION with args ARGS. If the function 30930 returns non-NULL, then 'X' is substituted, if it returns NULL, it 30931 isn't substituted. 30932 30933'%{S:X; T:Y; :D}' 30934 30935 If 'S' is given to GCC, substitutes 'X'; else if 'T' is given to 30936 GCC, substitutes 'Y'; else substitutes 'D'. There can be as many 30937 clauses as you need. This may be combined with '.', ',', '!', '|', 30938 and '*' as needed. 30939 30940 The switch matching text 'S' in a '%{S}', '%{S:X}' or similar construct 30941can use a backslash to ignore the special meaning of the character 30942following it, thus allowing literal matching of a character that is 30943otherwise specially treated. For example, '%{std=iso9899\:1999:X}' 30944substitutes 'X' if the '-std=iso9899:1999' option is given. 30945 30946 The conditional text 'X' in a '%{S:X}' or similar construct may contain 30947other nested '%' constructs or spaces, or even newlines. They are 30948processed as usual, as described above. Trailing white space in 'X' is 30949ignored. White space may also appear anywhere on the left side of the 30950colon in these constructs, except between '.' or '*' and the 30951corresponding word. 30952 30953 The '-O', '-f', '-m', and '-W' switches are handled specifically in 30954these constructs. If another value of '-O' or the negated form of a 30955'-f', '-m', or '-W' switch is found later in the command line, the 30956earlier switch value is ignored, except with {'S'*} where 'S' is just 30957one letter, which passes all matching options. 30958 30959 The character '|' at the beginning of the predicate text is used to 30960indicate that a command should be piped to the following command, but 30961only if '-pipe' is specified. 30962 30963 It is built into GCC which switches take arguments and which do not. 30964(You might think it would be useful to generalize this to allow each 30965compiler's spec to say which switches take arguments. But this cannot 30966be done in a consistent fashion. GCC cannot even decide which input 30967files have been specified without knowing which switches take arguments, 30968and it must know which input files to compile in order to tell which 30969compilers to run). 30970 30971 GCC also knows implicitly that arguments starting in '-l' are to be 30972treated as compiler output files, and passed to the linker in their 30973proper position among the other output files. 30974 30975 30976File: gcc.info, Node: Environment Variables, Next: Precompiled Headers, Prev: Spec Files, Up: Invoking GCC 30977 309783.21 Environment Variables Affecting GCC 30979======================================== 30980 30981This section describes several environment variables that affect how GCC 30982operates. Some of them work by specifying directories or prefixes to 30983use when searching for various kinds of files. Some are used to specify 30984other aspects of the compilation environment. 30985 30986 Note that you can also specify places to search using options such as 30987'-B', '-I' and '-L' (*note Directory Options::). These take precedence 30988over places specified using environment variables, which in turn take 30989precedence over those specified by the configuration of GCC. *Note 30990Controlling the Compilation Driver 'gcc': (gccint)Driver. 30991 30992'LANG' 30993'LC_CTYPE' 30994'LC_MESSAGES' 30995'LC_ALL' 30996 These environment variables control the way that GCC uses 30997 localization information which allows GCC to work with different 30998 national conventions. GCC inspects the locale categories 30999 'LC_CTYPE' and 'LC_MESSAGES' if it has been configured to do so. 31000 These locale categories can be set to any value supported by your 31001 installation. A typical value is 'en_GB.UTF-8' for English in the 31002 United Kingdom encoded in UTF-8. 31003 31004 The 'LC_CTYPE' environment variable specifies character 31005 classification. GCC uses it to determine the character boundaries 31006 in a string; this is needed for some multibyte encodings that 31007 contain quote and escape characters that are otherwise interpreted 31008 as a string end or escape. 31009 31010 The 'LC_MESSAGES' environment variable specifies the language to 31011 use in diagnostic messages. 31012 31013 If the 'LC_ALL' environment variable is set, it overrides the value 31014 of 'LC_CTYPE' and 'LC_MESSAGES'; otherwise, 'LC_CTYPE' and 31015 'LC_MESSAGES' default to the value of the 'LANG' environment 31016 variable. If none of these variables are set, GCC defaults to 31017 traditional C English behavior. 31018 31019'TMPDIR' 31020 If 'TMPDIR' is set, it specifies the directory to use for temporary 31021 files. GCC uses temporary files to hold the output of one stage of 31022 compilation which is to be used as input to the next stage: for 31023 example, the output of the preprocessor, which is the input to the 31024 compiler proper. 31025 31026'GCC_COMPARE_DEBUG' 31027 Setting 'GCC_COMPARE_DEBUG' is nearly equivalent to passing 31028 '-fcompare-debug' to the compiler driver. See the documentation of 31029 this option for more details. 31030 31031'GCC_EXEC_PREFIX' 31032 If 'GCC_EXEC_PREFIX' is set, it specifies a prefix to use in the 31033 names of the subprograms executed by the compiler. No slash is 31034 added when this prefix is combined with the name of a subprogram, 31035 but you can specify a prefix that ends with a slash if you wish. 31036 31037 If 'GCC_EXEC_PREFIX' is not set, GCC attempts to figure out an 31038 appropriate prefix to use based on the pathname it is invoked with. 31039 31040 If GCC cannot find the subprogram using the specified prefix, it 31041 tries looking in the usual places for the subprogram. 31042 31043 The default value of 'GCC_EXEC_PREFIX' is 'PREFIX/lib/gcc/' where 31044 PREFIX is the prefix to the installed compiler. In many cases 31045 PREFIX is the value of 'prefix' when you ran the 'configure' 31046 script. 31047 31048 Other prefixes specified with '-B' take precedence over this 31049 prefix. 31050 31051 This prefix is also used for finding files such as 'crt0.o' that 31052 are used for linking. 31053 31054 In addition, the prefix is used in an unusual way in finding the 31055 directories to search for header files. For each of the standard 31056 directories whose name normally begins with '/usr/local/lib/gcc' 31057 (more precisely, with the value of 'GCC_INCLUDE_DIR'), GCC tries 31058 replacing that beginning with the specified prefix to produce an 31059 alternate directory name. Thus, with '-Bfoo/', GCC searches 31060 'foo/bar' just before it searches the standard directory 31061 '/usr/local/lib/bar'. If a standard directory begins with the 31062 configured PREFIX then the value of PREFIX is replaced by 31063 'GCC_EXEC_PREFIX' when looking for header files. 31064 31065'COMPILER_PATH' 31066 The value of 'COMPILER_PATH' is a colon-separated list of 31067 directories, much like 'PATH'. GCC tries the directories thus 31068 specified when searching for subprograms, if it cannot find the 31069 subprograms using 'GCC_EXEC_PREFIX'. 31070 31071'LIBRARY_PATH' 31072 The value of 'LIBRARY_PATH' is a colon-separated list of 31073 directories, much like 'PATH'. When configured as a native 31074 compiler, GCC tries the directories thus specified when searching 31075 for special linker files, if it cannot find them using 31076 'GCC_EXEC_PREFIX'. Linking using GCC also uses these directories 31077 when searching for ordinary libraries for the '-l' option (but 31078 directories specified with '-L' come first). 31079 31080'LANG' 31081 This variable is used to pass locale information to the compiler. 31082 One way in which this information is used is to determine the 31083 character set to be used when character literals, string literals 31084 and comments are parsed in C and C++. When the compiler is 31085 configured to allow multibyte characters, the following values for 31086 'LANG' are recognized: 31087 31088 'C-JIS' 31089 Recognize JIS characters. 31090 'C-SJIS' 31091 Recognize SJIS characters. 31092 'C-EUCJP' 31093 Recognize EUCJP characters. 31094 31095 If 'LANG' is not defined, or if it has some other value, then the 31096 compiler uses 'mblen' and 'mbtowc' as defined by the default locale 31097 to recognize and translate multibyte characters. 31098 31099'GCC_EXTRA_DIAGNOSTIC_OUTPUT' 31100 If 'GCC_EXTRA_DIAGNOSTIC_OUTPUT' is set to one of the following 31101 values, then additional text will be emitted to stderr when fix-it 31102 hints are emitted. '-fdiagnostics-parseable-fixits' and 31103 '-fno-diagnostics-parseable-fixits' take precedence over this 31104 environment variable. 31105 31106 'fixits-v1' 31107 Emit parseable fix-it hints, equivalent to 31108 '-fdiagnostics-parseable-fixits'. In particular, columns are 31109 expressed as a count of bytes, starting at byte 1 for the 31110 initial column. 31111 31112 'fixits-v2' 31113 As 'fixits-v1', but columns are expressed as display columns, 31114 as per '-fdiagnostics-column-unit=display'. 31115 31116Some additional environment variables affect the behavior of the 31117preprocessor. 31118 31119'CPATH' 31120'C_INCLUDE_PATH' 31121'CPLUS_INCLUDE_PATH' 31122'OBJC_INCLUDE_PATH' 31123 Each variable's value is a list of directories separated by a 31124 special character, much like 'PATH', in which to look for header 31125 files. The special character, 'PATH_SEPARATOR', is 31126 target-dependent and determined at GCC build time. For Microsoft 31127 Windows-based targets it is a semicolon, and for almost all other 31128 targets it is a colon. 31129 31130 'CPATH' specifies a list of directories to be searched as if 31131 specified with '-I', but after any paths given with '-I' options on 31132 the command line. This environment variable is used regardless of 31133 which language is being preprocessed. 31134 31135 The remaining environment variables apply only when preprocessing 31136 the particular language indicated. Each specifies a list of 31137 directories to be searched as if specified with '-isystem', but 31138 after any paths given with '-isystem' options on the command line. 31139 31140 In all these variables, an empty element instructs the compiler to 31141 search its current working directory. Empty elements can appear at 31142 the beginning or end of a path. For instance, if the value of 31143 'CPATH' is ':/special/include', that has the same effect as 31144 '-I. -I/special/include'. 31145 31146'DEPENDENCIES_OUTPUT' 31147 If this variable is set, its value specifies how to output 31148 dependencies for Make based on the non-system header files 31149 processed by the compiler. System header files are ignored in the 31150 dependency output. 31151 31152 The value of 'DEPENDENCIES_OUTPUT' can be just a file name, in 31153 which case the Make rules are written to that file, guessing the 31154 target name from the source file name. Or the value can have the 31155 form 'FILE TARGET', in which case the rules are written to file 31156 FILE using TARGET as the target name. 31157 31158 In other words, this environment variable is equivalent to 31159 combining the options '-MM' and '-MF' (*note Preprocessor 31160 Options::), with an optional '-MT' switch too. 31161 31162'SUNPRO_DEPENDENCIES' 31163 This variable is the same as 'DEPENDENCIES_OUTPUT' (see above), 31164 except that system header files are not ignored, so it implies '-M' 31165 rather than '-MM'. However, the dependence on the main input file 31166 is omitted. *Note Preprocessor Options::. 31167 31168'SOURCE_DATE_EPOCH' 31169 If this variable is set, its value specifies a UNIX timestamp to be 31170 used in replacement of the current date and time in the '__DATE__' 31171 and '__TIME__' macros, so that the embedded timestamps become 31172 reproducible. 31173 31174 The value of 'SOURCE_DATE_EPOCH' must be a UNIX timestamp, defined 31175 as the number of seconds (excluding leap seconds) since 01 Jan 1970 31176 00:00:00 represented in ASCII; identical to the output of 'date 31177 +%s' on GNU/Linux and other systems that support the '%s' extension 31178 in the 'date' command. 31179 31180 The value should be a known timestamp such as the last modification 31181 time of the source or package and it should be set by the build 31182 process. 31183 31184 31185File: gcc.info, Node: Precompiled Headers, Next: C++ Modules, Prev: Environment Variables, Up: Invoking GCC 31186 311873.22 Using Precompiled Headers 31188============================== 31189 31190Often large projects have many header files that are included in every 31191source file. The time the compiler takes to process these header files 31192over and over again can account for nearly all of the time required to 31193build the project. To make builds faster, GCC allows you to 31194"precompile" a header file. 31195 31196 To create a precompiled header file, simply compile it as you would any 31197other file, if necessary using the '-x' option to make the driver treat 31198it as a C or C++ header file. You may want to use a tool like 'make' to 31199keep the precompiled header up-to-date when the headers it contains 31200change. 31201 31202 A precompiled header file is searched for when '#include' is seen in 31203the compilation. As it searches for the included file (*note Search 31204Path: (cpp)Search Path.) the compiler looks for a precompiled header in 31205each directory just before it looks for the include file in that 31206directory. The name searched for is the name specified in the 31207'#include' with '.gch' appended. If the precompiled header file cannot 31208be used, it is ignored. 31209 31210 For instance, if you have '#include "all.h"', and you have 'all.h.gch' 31211in the same directory as 'all.h', then the precompiled header file is 31212used if possible, and the original header is used otherwise. 31213 31214 Alternatively, you might decide to put the precompiled header file in a 31215directory and use '-I' to ensure that directory is searched before (or 31216instead of) the directory containing the original header. Then, if you 31217want to check that the precompiled header file is always used, you can 31218put a file of the same name as the original header in this directory 31219containing an '#error' command. 31220 31221 This also works with '-include'. So yet another way to use precompiled 31222headers, good for projects not designed with precompiled header files in 31223mind, is to simply take most of the header files used by a project, 31224include them from another header file, precompile that header file, and 31225'-include' the precompiled header. If the header files have guards 31226against multiple inclusion, they are skipped because they've already 31227been included (in the precompiled header). 31228 31229 If you need to precompile the same header file for different languages, 31230targets, or compiler options, you can instead make a _directory_ named 31231like 'all.h.gch', and put each precompiled header in the directory, 31232perhaps using '-o'. It doesn't matter what you call the files in the 31233directory; every precompiled header in the directory is considered. The 31234first precompiled header encountered in the directory that is valid for 31235this compilation is used; they're searched in no particular order. 31236 31237 There are many other possibilities, limited only by your imagination, 31238good sense, and the constraints of your build system. 31239 31240 A precompiled header file can be used only when these conditions apply: 31241 31242 * Only one precompiled header can be used in a particular 31243 compilation. 31244 31245 * A precompiled header cannot be used once the first C token is seen. 31246 You can have preprocessor directives before a precompiled header; 31247 you cannot include a precompiled header from inside another header. 31248 31249 * The precompiled header file must be produced for the same language 31250 as the current compilation. You cannot use a C precompiled header 31251 for a C++ compilation. 31252 31253 * The precompiled header file must have been produced by the same 31254 compiler binary as the current compilation is using. 31255 31256 * Any macros defined before the precompiled header is included must 31257 either be defined in the same way as when the precompiled header 31258 was generated, or must not affect the precompiled header, which 31259 usually means that they don't appear in the precompiled header at 31260 all. 31261 31262 The '-D' option is one way to define a macro before a precompiled 31263 header is included; using a '#define' can also do it. There are 31264 also some options that define macros implicitly, like '-O' and 31265 '-Wdeprecated'; the same rule applies to macros defined this way. 31266 31267 * If debugging information is output when using the precompiled 31268 header, using '-g' or similar, the same kind of debugging 31269 information must have been output when building the precompiled 31270 header. However, a precompiled header built using '-g' can be used 31271 in a compilation when no debugging information is being output. 31272 31273 * The same '-m' options must generally be used when building and 31274 using the precompiled header. *Note Submodel Options::, for any 31275 cases where this rule is relaxed. 31276 31277 * Each of the following options must be the same when building and 31278 using the precompiled header: 31279 31280 -fexceptions 31281 31282 * Some other command-line options starting with '-f', '-p', or '-O' 31283 must be defined in the same way as when the precompiled header was 31284 generated. At present, it's not clear which options are safe to 31285 change and which are not; the safest choice is to use exactly the 31286 same options when generating and using the precompiled header. The 31287 following are known to be safe: 31288 31289 -fmessage-length= -fpreprocessed -fsched-interblock 31290 -fsched-spec -fsched-spec-load -fsched-spec-load-dangerous 31291 -fsched-verbose=NUMBER -fschedule-insns -fvisibility= 31292 -pedantic-errors 31293 31294 * Address space layout randomization (ASLR) can lead to not binary 31295 identical PCH files. If you rely on stable PCH file contents 31296 disable ASLR when generating PCH files. 31297 31298 For all of these except the last, the compiler automatically ignores 31299the precompiled header if the conditions aren't met. If you find an 31300option combination that doesn't work and doesn't cause the precompiled 31301header to be ignored, please consider filing a bug report, see *note 31302Bugs::. 31303 31304 If you do use differing options when generating and using the 31305precompiled header, the actual behavior is a mixture of the behavior for 31306the options. For instance, if you use '-g' to generate the precompiled 31307header but not when using it, you may or may not get debugging 31308information for routines in the precompiled header. 31309 31310 31311File: gcc.info, Node: C++ Modules, Prev: Precompiled Headers, Up: Invoking GCC 31312 313133.23 C++ Modules 31314================ 31315 31316Modules are a C++20 language feature. As the name suggests, they 31317provides a modular compilation system, intending to provide both faster 31318builds and better library isolation. The "Merging Modules" paper 31319<https://wg21.link/p1103>, provides the easiest to read set of changes 31320to the standard, although it does not capture later changes. 31321 31322 _G++'s modules support is not complete._ Other than bugs, the known 31323missing pieces are: 31324 31325_Private Module Fragment_ 31326 The Private Module Fragment is recognized, but an error is emitted. 31327 31328_Partition definition visibility rules_ 31329 Entities may be defined in implementation partitions, and those 31330 definitions are not available outside of the module. This is not 31331 implemented, and the definitions are available to extra-module use. 31332 31333_Textual merging of reachable GM entities_ 31334 Entities may be multiply defined across different header-units. 31335 These must be de-duplicated, and this is implemented across 31336 imports, or when an import redefines a textually-defined entity. 31337 However the reverse is not implemented--textually redefining an 31338 entity that has been defined in an imported header-unit. A 31339 redefinition error is emitted. 31340 31341_Translation-Unit local referencing rules_ 31342 Papers p1815 (<https://wg21.link/p1815>) and p2003 31343 (<https://wg21.link/p2003>) add limitations on which entities an 31344 exported region may reference (for instance, the entities an 31345 exported template definition may reference). These are not fully 31346 implemented. 31347 31348_Language-linkage module attachment_ 31349 Declarations with explicit language linkage ('extern "C"' or 31350 'extern "C++"') are attached to the global module, even when in the 31351 purview of a named module. This is not implemented. Such 31352 declarations will be attached to the module, if any, in which they 31353 are declared. 31354 31355_Standard Library Header Units_ 31356 The Standard Library is not provided as importable header units. 31357 If you want to import such units, you must explicitly build them 31358 first. If you do not do this with care, you may have multiple 31359 declarations, which the module machinery must merge--compiler 31360 resource usage can be affected by how you partition header files 31361 into header units. 31362 31363 Modular compilation is _not_ enabled with just the '-std=c++20' option. 31364You must explicitly enable it with the '-fmodules-ts' option. It is 31365independent of the language version selected, although in pre-C++20 31366versions, it is of course an extension. 31367 31368 No new source file suffixes are required or supported. If you wish to 31369use a non-standard suffix (*note Overall Options::), you also need to 31370provide a '-x c++' option too.(1) 31371 31372 Compiling a module interface unit produces an additional output (to the 31373assembly or object file), called a Compiled Module Interface (CMI). This 31374encodes the exported declarations of the module. Importing a module 31375reads in the CMI. The import graph is a Directed Acyclic Graph (DAG). 31376You must build imports before the importer. 31377 31378 Header files may themselves be compiled to header units, which are a 31379transitional ability aiming at faster compilation. The 31380'-fmodule-header' option is used to enable this, and implies the 31381'-fmodules-ts' option. These CMIs are named by the fully resolved 31382underlying header file, and thus may be a complete pathname containing 31383subdirectories. If the header file is found at an absolute pathname, 31384the CMI location is still relative to a CMI root directory. 31385 31386 As header files often have no suffix, you commonly have to specify a 31387'-x' option to tell the compiler the source is a header file. You may 31388use '-x c++-header', '-x c++-user-header' or '-x c++-system-header'. 31389When used in conjunction with '-fmodules-ts', these all imply an 31390appropriate '-fmodule-header' option. The latter two variants use the 31391user or system include path to search for the file specified. This 31392allows you to, for instance, compile standard library header files as 31393header units, without needing to know exactly where they are installed. 31394Specifying the language as one of these variants also inhibits output of 31395the object file, as header files have no associated object file. 31396 31397 The '-fmodule-only' option disables generation of the associated object 31398file for compiling a module interface. Only the CMI is generated. This 31399option is implied when using the '-fmodule-header' option. 31400 31401 The '-flang-info-include-translate' and 31402'-flang-info-include-translate-not' options notes whether include 31403translation occurs or not. With no argument, the first will note all 31404include translation. The second will note all non-translations of 31405include files not known to intentionally be textual. With an argument, 31406queries about include translation of a header files with that particular 31407trailing pathname are noted. You may repeat this form to cover several 31408different header files. This option may be helpful in determining 31409whether include translation is happening--if it is working correctly, it 31410behaves as if it isn't there at all. 31411 31412 The '-flang-info-module-cmi' option can be used to determine where the 31413compiler is reading a CMI from. Without the option, the compiler is 31414silent when such a read is successful. This option has an optional 31415argument, which will restrict the notification to just the set of named 31416modules or header units specified. 31417 31418 The '-Winvalid-imported-macros' option causes all imported macros to be 31419resolved at the end of compilation. Without this, imported macros are 31420only resolved when expanded or (re)defined. This option detects 31421conflicting import definitions for all macros. 31422 31423 For details of the '-fmodule-mapper' family of options, *note C++ 31424Module Mapper::. 31425 31426* Menu: 31427 31428* C++ Module Mapper:: Module Mapper 31429* C++ Module Preprocessing:: Module Preprocessing 31430* C++ Compiled Module Interface:: Compiled Module Interface 31431 31432 ---------- Footnotes ---------- 31433 31434 (1) Some users like to distinguish module interface files with a new 31435suffix, such as naming the source 'module.cppm', which involves teaching 31436all tools about the new suffix. A different scheme, such as naming 31437'module-m.cpp' would be less invasive. 31438 31439 31440File: gcc.info, Node: C++ Module Mapper, Next: C++ Module Preprocessing, Up: C++ Modules 31441 314423.23.1 Module Mapper 31443-------------------- 31444 31445A module mapper provides a server or file that the compiler queries to 31446determine the mapping between module names and CMI files. It is also 31447used to build CMIs on demand. _Mapper functionality is in its infancy 31448and is intended for experimentation with build system interactions._ 31449 31450 You can specify a mapper with the '-fmodule-mapper=VAL' option or 31451'CXX_MODULE_MAPPER' environment variable. The value may have one of the 31452following forms: 31453 31454'[HOSTNAME]:PORT[?IDENT]' 31455 An optional hostname and a numeric port number to connect to. If 31456 the hostname is omitted, the loopback address is used. If the 31457 hostname corresponds to multiple IPV6 addresses, these are tried in 31458 turn, until one is successful. If your host lacks IPv6, this form 31459 is non-functional. If you must use IPv4 use 31460 '-fmodule-mapper='|ncat IPV4HOST PORT''. 31461 31462'=SOCKET[?IDENT]' 31463 A local domain socket. If your host lacks local domain sockets, 31464 this form is non-functional. 31465 31466'|PROGRAM[?IDENT] [ARGS...]' 31467 A program to spawn, and communicate with on its stdin/stdout 31468 streams. Your PATH environment variable is searched for the 31469 program. Arguments are separated by space characters, (it is not 31470 possible for one of the arguments delivered to the program to 31471 contain a space). An exception is if PROGRAM begins with @. In 31472 that case PROGRAM (sans @) is looked for in the compiler's internal 31473 binary directory. Thus the sample mapper-server can be specified 31474 with '@g++-mapper-server'. 31475 31476'<>[?IDENT]' 31477'<>INOUT[?IDENT]' 31478'<IN>OUT[?IDENT]' 31479 Named pipes or file descriptors to communicate over. The first 31480 form, '<>', communicates over stdin and stdout. The other forms 31481 allow you to specify a file descriptor or name a pipe. A numeric 31482 value is interpreted as a file descriptor, otherwise named pipe is 31483 opened. The second form specifies a bidirectional pipe and the 31484 last form allows specifying two independent pipes. Using file 31485 descriptors directly in this manner is fragile in general, as it 31486 can require the cooperation of intermediate processes. In 31487 particular using stdin & stdout is fraught with danger as other 31488 compiler options might also cause the compiler to read stdin or 31489 write stdout, and it can have unfortunate interactions with signal 31490 delivery from the terminal. 31491 31492'FILE[?IDENT]' 31493 A mapping file consisting of space-separated module-name, filename 31494 pairs, one per line. Only the mappings for the direct imports and 31495 any module export name need be provided. If other mappings are 31496 provided, they override those stored in any imported CMI files. A 31497 repository root may be specified in the mapping file by using 31498 '$root' as the module name in the first active line. Use of this 31499 option will disable any default module->CMI name mapping. 31500 31501 As shown, an optional IDENT may suffix the first word of the option, 31502indicated by a '?' prefix. The value is used in the initial handshake 31503with the module server, or to specify a prefix on mapping file lines. 31504In the server case, the main source file name is used if no IDENT is 31505specified. In the file case, all non-blank lines are significant, 31506unless a value is specified, in which case only lines beginning with 31507IDENT are significant. The IDENT must be separated by whitespace from 31508the module name. Be aware that '<', '>', '?', and '|' characters are 31509often significant to the shell, and therefore may need quoting. 31510 31511 The mapper is connected to or loaded lazily, when the first module 31512mapping is required. The networking protocols are only supported on 31513hosts that provide networking. If no mapper is specified a default is 31514provided. 31515 31516 A project-specific mapper is expected to be provided by the build 31517system that invokes the compiler. It is not expected that a 31518general-purpose server is provided for all compilations. As such, the 31519server will know the build configuration, the compiler it invoked, and 31520the environment (such as working directory) in which that is operating. 31521As it may parallelize builds, several compilations may connect to the 31522same socket. 31523 31524 The default mapper generates CMI files in a 'gcm.cache' directory. CMI 31525files have a '.gcm' suffix. The module unit name is used directly to 31526provide the basename. Header units construct a relative path using the 31527underlying header file name. If the path is already relative, a ',' 31528directory is prepended. Internal '..' components are translated to 31529',,'. No attempt is made to canonicalize these filenames beyond that 31530done by the preprocessor's include search algorithm, as in general it is 31531ambiguous when symbolic links are present. 31532 31533 The mapper protocol was published as "A Module Mapper" 31534<https://wg21.link/p1184>. The implementation is provided by 'libcody', 31535<https://github.com/urnathan/libcody>, which specifies the canonical 31536protocol definition. A proof of concept server implementation embedded 31537in 'make' was described in "Make Me A Module", 31538<https://wg21.link/p1602>. 31539 31540 31541File: gcc.info, Node: C++ Module Preprocessing, Next: C++ Compiled Module Interface, Prev: C++ Module Mapper, Up: C++ Modules 31542 315433.23.2 Module Preprocessing 31544--------------------------- 31545 31546Modules affect preprocessing because of header units and include 31547translation. Some uses of the preprocessor as a separate step either do 31548not produce a correct output, or require CMIs to be available. 31549 31550 Header units import macros. These macros can affect later conditional 31551inclusion, which therefore can cascade to differing import sets. When 31552preprocessing, it is necessary to load the CMI. If a header unit is 31553unavailable, the preprocessor issues a warning and continue (when not 31554just preprocessing, an error is emitted). Detecting such imports 31555requires preprocessor tokenization of the input stream to phase 4 (macro 31556expansion). 31557 31558 Include translation converts '#include', '#include_next' and '#import' 31559directives to internal 'import' declarations. Whether a particular 31560directive is translated is controlled by the module mapper. Header unit 31561names are canonicalized during preprocessing. 31562 31563 Dependency information can be emitted for macro import, extending the 31564functionality of '-MD' and '-MMD' options. Detection of import 31565declarations also requires phase 4 preprocessing, and thus requires full 31566preprocessing (or compilation). 31567 31568 The '-M', '-MM' and '-E -fdirectives-only' options halt preprocessing 31569before phase 4. 31570 31571 The '-save-temps' option uses '-fdirectives-only' for preprocessing, 31572and preserve the macro definitions in the preprocessed output. Usually 31573you also want to use this option when explicitly preprocessing a 31574header-unit, or consuming such preprocessed output: 31575 31576 g++ -fmodules-ts -E -fdirectives-only my-header.hh -o my-header.ii 31577 g++ -x c++-header -fmodules-ts -fpreprocessed -fdirectives-only my-header.ii 31578 31579 31580File: gcc.info, Node: C++ Compiled Module Interface, Prev: C++ Module Preprocessing, Up: C++ Modules 31581 315823.23.3 Compiled Module Interface 31583-------------------------------- 31584 31585CMIs are an additional artifact when compiling named module interfaces, 31586partitions or header units. These are read when importing. CMI 31587contents are implementation-specific, and in GCC's case tied to the 31588compiler version. Consider them a rebuildable cache artifact, not a 31589distributable object. 31590 31591 When creating an output CMI, any missing directory components are 31592created in a manner that is safe for concurrent builds creating 31593multiple, different, CMIs within a common subdirectory tree. 31594 31595 CMI contents are written to a temporary file, which is then atomically 31596renamed. Observers either see old contents (if there is an existing 31597file), or complete new contents. They do not observe the CMI during its 31598creation. This is unlike object file writing, which may be observed by 31599an external process. 31600 31601 CMIs are read in lazily, if the host OS provides 'mmap' functionality. 31602Generally blocks are read when name lookup or template instantiation 31603occurs. To inhibit this, the '-fno-module-lazy' option may be used. 31604 31605 The '--param lazy-modules=N' parameter controls the limit on the number 31606of concurrently open module files during lazy loading. Should more 31607modules be imported, an LRU algorithm is used to determine which files 31608to close--until that file is needed again. This limit may be exceeded 31609with deep module dependency hierarchies. With large code bases there 31610may be more imports than the process limit of file descriptors. By 31611default, the limit is a few less than the per-process file descriptor 31612hard limit, if that is determinable.(1) 31613 31614 GCC CMIs use ELF32 as an architecture-neutral encapsulation mechanism. 31615You may use 'readelf' to inspect them, although section contents are 31616largely undecipherable. There is a section named '.gnu.c++.README', 31617which contains human-readable text. Other than the first line, each 31618line consists of 'TAG: value' tuples. 31619 31620 > readelf -p.gnu.c++.README gcm.cache/foo.gcm 31621 31622 String dump of section '.gnu.c++.README': 31623 [ 0] GNU C++ primary module interface 31624 [ 21] compiler: 11.0.0 20201116 (experimental) [c++-modules revision 20201116-0454] 31625 [ 6f] version: 2020/11/16-04:54 31626 [ 89] module: foo 31627 [ 95] source: c_b.ii 31628 [ a4] dialect: C++20/coroutines 31629 [ be] cwd: /data/users/nathans/modules/obj/x86_64/gcc 31630 [ ee] repository: gcm.cache 31631 [ 104] buildtime: 2020/11/16 15:03:21 UTC 31632 [ 127] localtime: 2020/11/16 07:03:21 PST 31633 [ 14a] export: foo:part1 foo-part1.gcm 31634 31635 Amongst other things, this lists the source that was built, C++ dialect 31636used and imports of the module.(2) The timestamp is the same value as 31637that provided by the '__DATE__' & '__TIME__' macros, and may be 31638explicitly specified with the environment variable 'SOURCE_DATE_EPOCH'. 31639For further details *note Environment Variables::. 31640 31641 A set of related CMIs may be copied, provided the relative pathnames 31642are preserved. 31643 31644 The '.gnu.c++.README' contents do not affect CMI integrity, and it may 31645be removed or altered. The section numbering of the sections whose 31646names do not begin with '.gnu.c++.', or are not the string section is 31647significant and must not be altered. 31648 31649 ---------- Footnotes ---------- 31650 31651 (1) Where applicable the soft limit is incremented as needed towards 31652the hard limit. 31653 31654 (2) The precise contents of this output may change. 31655 31656 31657File: gcc.info, Node: C Implementation, Next: C++ Implementation, Prev: Invoking GCC, Up: Top 31658 316594 C Implementation-Defined Behavior 31660*********************************** 31661 31662A conforming implementation of ISO C is required to document its choice 31663of behavior in each of the areas that are designated "implementation 31664defined". The following lists all such areas, along with the section 31665numbers from the ISO/IEC 9899:1990, ISO/IEC 9899:1999 and ISO/IEC 316669899:2011 standards. Some areas are only implementation-defined in one 31667version of the standard. 31668 31669 Some choices depend on the externally determined ABI for the platform 31670(including standard character encodings) which GCC follows; these are 31671listed as "determined by ABI" below. *Note Binary Compatibility: 31672Compatibility, and <https://gcc.gnu.org/readings.html>. Some choices 31673are documented in the preprocessor manual. *Note Implementation-defined 31674behavior: (cpp)Implementation-defined behavior. Some choices are made 31675by the library and operating system (or other environment when compiling 31676for a freestanding environment); refer to their documentation for 31677details. 31678 31679* Menu: 31680 31681* Translation implementation:: 31682* Environment implementation:: 31683* Identifiers implementation:: 31684* Characters implementation:: 31685* Integers implementation:: 31686* Floating point implementation:: 31687* Arrays and pointers implementation:: 31688* Hints implementation:: 31689* Structures unions enumerations and bit-fields implementation:: 31690* Qualifiers implementation:: 31691* Declarators implementation:: 31692* Statements implementation:: 31693* Preprocessing directives implementation:: 31694* Library functions implementation:: 31695* Architecture implementation:: 31696* Locale-specific behavior implementation:: 31697 31698 31699File: gcc.info, Node: Translation implementation, Next: Environment implementation, Up: C Implementation 31700 317014.1 Translation 31702=============== 31703 31704 * 'How a diagnostic is identified (C90 3.7, C99 and C11 3.10, C90, 31705 C99 and C11 5.1.1.3).' 31706 31707 Diagnostics consist of all the output sent to stderr by GCC. 31708 31709 * 'Whether each nonempty sequence of white-space characters other 31710 than new-line is retained or replaced by one space character in 31711 translation phase 3 (C90, C99 and C11 5.1.1.2).' 31712 31713 *Note Implementation-defined behavior: (cpp)Implementation-defined 31714 behavior. 31715 31716 31717File: gcc.info, Node: Environment implementation, Next: Identifiers implementation, Prev: Translation implementation, Up: C Implementation 31718 317194.2 Environment 31720=============== 31721 31722The behavior of most of these points are dependent on the implementation 31723of the C library, and are not defined by GCC itself. 31724 31725 * 'The mapping between physical source file multibyte characters and 31726 the source character set in translation phase 1 (C90, C99 and C11 31727 5.1.1.2).' 31728 31729 *Note Implementation-defined behavior: (cpp)Implementation-defined 31730 behavior. 31731 31732 31733File: gcc.info, Node: Identifiers implementation, Next: Characters implementation, Prev: Environment implementation, Up: C Implementation 31734 317354.3 Identifiers 31736=============== 31737 31738 * 'Which additional multibyte characters may appear in identifiers 31739 and their correspondence to universal character names (C99 and C11 31740 6.4.2).' 31741 31742 *Note Implementation-defined behavior: (cpp)Implementation-defined 31743 behavior. 31744 31745 * 'The number of significant initial characters in an identifier (C90 31746 6.1.2, C90, C99 and C11 5.2.4.1, C99 and C11 6.4.2).' 31747 31748 For internal names, all characters are significant. For external 31749 names, the number of significant characters are defined by the 31750 linker; for almost all targets, all characters are significant. 31751 31752 * 'Whether case distinctions are significant in an identifier with 31753 external linkage (C90 6.1.2).' 31754 31755 This is a property of the linker. C99 and C11 require that case 31756 distinctions are always significant in identifiers with external 31757 linkage and systems without this property are not supported by GCC. 31758 31759 31760File: gcc.info, Node: Characters implementation, Next: Integers implementation, Prev: Identifiers implementation, Up: C Implementation 31761 317624.4 Characters 31763============== 31764 31765 * 'The number of bits in a byte (C90 3.4, C99 and C11 3.6).' 31766 31767 Determined by ABI. 31768 31769 * 'The values of the members of the execution character set (C90, C99 31770 and C11 5.2.1).' 31771 31772 Determined by ABI. 31773 31774 * 'The unique value of the member of the execution character set 31775 produced for each of the standard alphabetic escape sequences (C90, 31776 C99 and C11 5.2.2).' 31777 31778 Determined by ABI. 31779 31780 * 'The value of a 'char' object into which has been stored any 31781 character other than a member of the basic execution character set 31782 (C90 6.1.2.5, C99 and C11 6.2.5).' 31783 31784 Determined by ABI. 31785 31786 * 'Which of 'signed char' or 'unsigned char' has the same range, 31787 representation, and behavior as "plain" 'char' (C90 6.1.2.5, C90 31788 6.2.1.1, C99 and C11 6.2.5, C99 and C11 6.3.1.1).' 31789 31790 Determined by ABI. The options '-funsigned-char' and 31791 '-fsigned-char' change the default. *Note Options Controlling C 31792 Dialect: C Dialect Options. 31793 31794 * 'The mapping of members of the source character set (in character 31795 constants and string literals) to members of the execution 31796 character set (C90 6.1.3.4, C99 and C11 6.4.4.4, C90, C99 and C11 31797 5.1.1.2).' 31798 31799 Determined by ABI. 31800 31801 * 'The value of an integer character constant containing more than 31802 one character or containing a character or escape sequence that 31803 does not map to a single-byte execution character (C90 6.1.3.4, C99 31804 and C11 6.4.4.4).' 31805 31806 *Note Implementation-defined behavior: (cpp)Implementation-defined 31807 behavior. 31808 31809 * 'The value of a wide character constant containing more than one 31810 multibyte character or a single multibyte character that maps to 31811 multiple members of the extended execution character set, or 31812 containing a multibyte character or escape sequence not represented 31813 in the extended execution character set (C90 6.1.3.4, C99 and C11 31814 6.4.4.4).' 31815 31816 *Note Implementation-defined behavior: (cpp)Implementation-defined 31817 behavior. 31818 31819 * 'The current locale used to convert a wide character constant 31820 consisting of a single multibyte character that maps to a member of 31821 the extended execution character set into a corresponding wide 31822 character code (C90 6.1.3.4, C99 and C11 6.4.4.4).' 31823 31824 *Note Implementation-defined behavior: (cpp)Implementation-defined 31825 behavior. 31826 31827 * 'Whether differently-prefixed wide string literal tokens can be 31828 concatenated and, if so, the treatment of the resulting multibyte 31829 character sequence (C11 6.4.5).' 31830 31831 Such tokens may not be concatenated. 31832 31833 * 'The current locale used to convert a wide string literal into 31834 corresponding wide character codes (C90 6.1.4, C99 and C11 6.4.5).' 31835 31836 *Note Implementation-defined behavior: (cpp)Implementation-defined 31837 behavior. 31838 31839 * 'The value of a string literal containing a multibyte character or 31840 escape sequence not represented in the execution character set (C90 31841 6.1.4, C99 and C11 6.4.5).' 31842 31843 *Note Implementation-defined behavior: (cpp)Implementation-defined 31844 behavior. 31845 31846 * 'The encoding of any of 'wchar_t', 'char16_t', and 'char32_t' where 31847 the corresponding standard encoding macro ('__STDC_ISO_10646__', 31848 '__STDC_UTF_16__', or '__STDC_UTF_32__') is not defined (C11 31849 6.10.8.2).' 31850 31851 *Note Implementation-defined behavior: (cpp)Implementation-defined 31852 behavior. 'char16_t' and 'char32_t' literals are always encoded in 31853 UTF-16 and UTF-32 respectively. 31854 31855 31856File: gcc.info, Node: Integers implementation, Next: Floating point implementation, Prev: Characters implementation, Up: C Implementation 31857 318584.5 Integers 31859============ 31860 31861 * 'Any extended integer types that exist in the implementation (C99 31862 and C11 6.2.5).' 31863 31864 GCC does not support any extended integer types. 31865 31866 * 'Whether signed integer types are represented using sign and 31867 magnitude, two's complement, or one's complement, and whether the 31868 extraordinary value is a trap representation or an ordinary value 31869 (C99 and C11 6.2.6.2).' 31870 31871 GCC supports only two's complement integer types, and all bit 31872 patterns are ordinary values. 31873 31874 * 'The rank of any extended integer type relative to another extended 31875 integer type with the same precision (C99 and C11 6.3.1.1).' 31876 31877 GCC does not support any extended integer types. 31878 31879 * 'The result of, or the signal raised by, converting an integer to a 31880 signed integer type when the value cannot be represented in an 31881 object of that type (C90 6.2.1.2, C99 and C11 6.3.1.3).' 31882 31883 For conversion to a type of width N, the value is reduced modulo 31884 2^N to be within range of the type; no signal is raised. 31885 31886 * 'The results of some bitwise operations on signed integers (C90 31887 6.3, C99 and C11 6.5).' 31888 31889 Bitwise operators act on the representation of the value including 31890 both the sign and value bits, where the sign bit is considered 31891 immediately above the highest-value value bit. Signed '>>' acts on 31892 negative numbers by sign extension. 31893 31894 As an extension to the C language, GCC does not use the latitude 31895 given in C99 and C11 only to treat certain aspects of signed '<<' 31896 as undefined. However, '-fsanitize=shift' (and 31897 '-fsanitize=undefined') will diagnose such cases. They are also 31898 diagnosed where constant expressions are required. 31899 31900 * 'The sign of the remainder on integer division (C90 6.3.5).' 31901 31902 GCC always follows the C99 and C11 requirement that the result of 31903 division is truncated towards zero. 31904 31905 31906File: gcc.info, Node: Floating point implementation, Next: Arrays and pointers implementation, Prev: Integers implementation, Up: C Implementation 31907 319084.6 Floating Point 31909================== 31910 31911 * 'The accuracy of the floating-point operations and of the library 31912 functions in '<math.h>' and '<complex.h>' that return 31913 floating-point results (C90, C99 and C11 5.2.4.2.2).' 31914 31915 The accuracy is unknown. 31916 31917 * 'The rounding behaviors characterized by non-standard values of 31918 'FLT_ROUNDS' (C90, C99 and C11 5.2.4.2.2).' 31919 31920 GCC does not use such values. 31921 31922 * 'The evaluation methods characterized by non-standard negative 31923 values of 'FLT_EVAL_METHOD' (C99 and C11 5.2.4.2.2).' 31924 31925 GCC does not use such values. 31926 31927 * 'The direction of rounding when an integer is converted to a 31928 floating-point number that cannot exactly represent the original 31929 value (C90 6.2.1.3, C99 and C11 6.3.1.4).' 31930 31931 C99 Annex F is followed. 31932 31933 * 'The direction of rounding when a floating-point number is 31934 converted to a narrower floating-point number (C90 6.2.1.4, C99 and 31935 C11 6.3.1.5).' 31936 31937 C99 Annex F is followed. 31938 31939 * 'How the nearest representable value or the larger or smaller 31940 representable value immediately adjacent to the nearest 31941 representable value is chosen for certain floating constants (C90 31942 6.1.3.1, C99 and C11 6.4.4.2).' 31943 31944 C99 Annex F is followed. 31945 31946 * 'Whether and how floating expressions are contracted when not 31947 disallowed by the 'FP_CONTRACT' pragma (C99 and C11 6.5).' 31948 31949 Expressions are currently only contracted if '-ffp-contract=fast', 31950 '-funsafe-math-optimizations' or '-ffast-math' are used. This is 31951 subject to change. 31952 31953 * 'The default state for the 'FENV_ACCESS' pragma (C99 and C11 31954 7.6.1).' 31955 31956 This pragma is not implemented, but the default is to "off" unless 31957 '-frounding-math' is used in which case it is "on". 31958 31959 * 'Additional floating-point exceptions, rounding modes, 31960 environments, and classifications, and their macro names (C99 and 31961 C11 7.6, C99 and C11 7.12).' 31962 31963 This is dependent on the implementation of the C library, and is 31964 not defined by GCC itself. 31965 31966 * 'The default state for the 'FP_CONTRACT' pragma (C99 and C11 31967 7.12.2).' 31968 31969 This pragma is not implemented. Expressions are currently only 31970 contracted if '-ffp-contract=fast', '-funsafe-math-optimizations' 31971 or '-ffast-math' are used. This is subject to change. 31972 31973 * 'Whether the "inexact" floating-point exception can be raised when 31974 the rounded result actually does equal the mathematical result in 31975 an IEC 60559 conformant implementation (C99 F.9).' 31976 31977 This is dependent on the implementation of the C library, and is 31978 not defined by GCC itself. 31979 31980 * 'Whether the "underflow" (and "inexact") floating-point exception 31981 can be raised when a result is tiny but not inexact in an IEC 60559 31982 conformant implementation (C99 F.9).' 31983 31984 This is dependent on the implementation of the C library, and is 31985 not defined by GCC itself. 31986 31987 31988File: gcc.info, Node: Arrays and pointers implementation, Next: Hints implementation, Prev: Floating point implementation, Up: C Implementation 31989 319904.7 Arrays and Pointers 31991======================= 31992 31993 * 'The result of converting a pointer to an integer or vice versa 31994 (C90 6.3.4, C99 and C11 6.3.2.3).' 31995 31996 A cast from pointer to integer discards most-significant bits if 31997 the pointer representation is larger than the integer type, 31998 sign-extends(1) if the pointer representation is smaller than the 31999 integer type, otherwise the bits are unchanged. 32000 32001 A cast from integer to pointer discards most-significant bits if 32002 the pointer representation is smaller than the integer type, 32003 extends according to the signedness of the integer type if the 32004 pointer representation is larger than the integer type, otherwise 32005 the bits are unchanged. 32006 32007 When casting from pointer to integer and back again, the resulting 32008 pointer must reference the same object as the original pointer, 32009 otherwise the behavior is undefined. That is, one may not use 32010 integer arithmetic to avoid the undefined behavior of pointer 32011 arithmetic as proscribed in C99 and C11 6.5.6/8. 32012 32013 * 'The size of the result of subtracting two pointers to elements of 32014 the same array (C90 6.3.6, C99 and C11 6.5.6).' 32015 32016 The value is as specified in the standard and the type is 32017 determined by the ABI. 32018 32019 ---------- Footnotes ---------- 32020 32021 (1) Future versions of GCC may zero-extend, or use a target-defined 32022'ptr_extend' pattern. Do not rely on sign extension. 32023 32024 32025File: gcc.info, Node: Hints implementation, Next: Structures unions enumerations and bit-fields implementation, Prev: Arrays and pointers implementation, Up: C Implementation 32026 320274.8 Hints 32028========= 32029 32030 * 'The extent to which suggestions made by using the 'register' 32031 storage-class specifier are effective (C90 6.5.1, C99 and C11 32032 6.7.1).' 32033 32034 The 'register' specifier affects code generation only in these 32035 ways: 32036 32037 * When used as part of the register variable extension, see 32038 *note Explicit Register Variables::. 32039 32040 * When '-O0' is in use, the compiler allocates distinct stack 32041 memory for all variables that do not have the 'register' 32042 storage-class specifier; if 'register' is specified, the 32043 variable may have a shorter lifespan than the code would 32044 indicate and may never be placed in memory. 32045 32046 * On some rare x86 targets, 'setjmp' doesn't save the registers 32047 in all circumstances. In those cases, GCC doesn't allocate 32048 any variables in registers unless they are marked 'register'. 32049 32050 * 'The extent to which suggestions made by using the inline function 32051 specifier are effective (C99 and C11 6.7.4).' 32052 32053 GCC will not inline any functions if the '-fno-inline' option is 32054 used or if '-O0' is used. Otherwise, GCC may still be unable to 32055 inline a function for many reasons; the '-Winline' option may be 32056 used to determine if a function has not been inlined and why not. 32057 32058 32059File: gcc.info, Node: Structures unions enumerations and bit-fields implementation, Next: Qualifiers implementation, Prev: Hints implementation, Up: C Implementation 32060 320614.9 Structures, Unions, Enumerations, and Bit-Fields 32062==================================================== 32063 32064 * 'A member of a union object is accessed using a member of a 32065 different type (C90 6.3.2.3).' 32066 32067 The relevant bytes of the representation of the object are treated 32068 as an object of the type used for the access. *Note 32069 Type-punning::. This may be a trap representation. 32070 32071 * 'Whether a "plain" 'int' bit-field is treated as a 'signed int' 32072 bit-field or as an 'unsigned int' bit-field (C90 6.5.2, C90 32073 6.5.2.1, C99 and C11 6.7.2, C99 and C11 6.7.2.1).' 32074 32075 By default it is treated as 'signed int' but this may be changed by 32076 the '-funsigned-bitfields' option. 32077 32078 * 'Allowable bit-field types other than '_Bool', 'signed int', and 32079 'unsigned int' (C99 and C11 6.7.2.1).' 32080 32081 Other integer types, such as 'long int', and enumerated types are 32082 permitted even in strictly conforming mode. 32083 32084 * 'Whether atomic types are permitted for bit-fields (C11 6.7.2.1).' 32085 32086 Atomic types are not permitted for bit-fields. 32087 32088 * 'Whether a bit-field can straddle a storage-unit boundary (C90 32089 6.5.2.1, C99 and C11 6.7.2.1).' 32090 32091 Determined by ABI. 32092 32093 * 'The order of allocation of bit-fields within a unit (C90 6.5.2.1, 32094 C99 and C11 6.7.2.1).' 32095 32096 Determined by ABI. 32097 32098 * 'The alignment of non-bit-field members of structures (C90 6.5.2.1, 32099 C99 and C11 6.7.2.1).' 32100 32101 Determined by ABI. 32102 32103 * 'The integer type compatible with each enumerated type (C90 32104 6.5.2.2, C99 and C11 6.7.2.2).' 32105 32106 Normally, the type is 'unsigned int' if there are no negative 32107 values in the enumeration, otherwise 'int'. If '-fshort-enums' is 32108 specified, then if there are negative values it is the first of 32109 'signed char', 'short' and 'int' that can represent all the values, 32110 otherwise it is the first of 'unsigned char', 'unsigned short' and 32111 'unsigned int' that can represent all the values. 32112 32113 On some targets, '-fshort-enums' is the default; this is determined 32114 by the ABI. 32115 32116 32117File: gcc.info, Node: Qualifiers implementation, Next: Declarators implementation, Prev: Structures unions enumerations and bit-fields implementation, Up: C Implementation 32118 321194.10 Qualifiers 32120=============== 32121 32122 * 'What constitutes an access to an object that has 32123 volatile-qualified type (C90 6.5.3, C99 and C11 6.7.3).' 32124 32125 Such an object is normally accessed by pointers and used for 32126 accessing hardware. In most expressions, it is intuitively obvious 32127 what is a read and what is a write. For example 32128 32129 volatile int *dst = SOMEVALUE; 32130 volatile int *src = SOMEOTHERVALUE; 32131 *dst = *src; 32132 32133 will cause a read of the volatile object pointed to by SRC and 32134 store the value into the volatile object pointed to by DST. There 32135 is no guarantee that these reads and writes are atomic, especially 32136 for objects larger than 'int'. 32137 32138 However, if the volatile storage is not being modified, and the 32139 value of the volatile storage is not used, then the situation is 32140 less obvious. For example 32141 32142 volatile int *src = SOMEVALUE; 32143 *src; 32144 32145 According to the C standard, such an expression is an rvalue whose 32146 type is the unqualified version of its original type, i.e. 'int'. 32147 Whether GCC interprets this as a read of the volatile object being 32148 pointed to or only as a request to evaluate the expression for its 32149 side effects depends on this type. 32150 32151 If it is a scalar type, or on most targets an aggregate type whose 32152 only member object is of a scalar type, or a union type whose 32153 member objects are of scalar types, the expression is interpreted 32154 by GCC as a read of the volatile object; in the other cases, the 32155 expression is only evaluated for its side effects. 32156 32157 When an object of an aggregate type, with the same size and 32158 alignment as a scalar type 'S', is the subject of a volatile access 32159 by an assignment expression or an atomic function, the access to it 32160 is performed as if the object's declared type were 'volatile S'. 32161 32162 32163File: gcc.info, Node: Declarators implementation, Next: Statements implementation, Prev: Qualifiers implementation, Up: C Implementation 32164 321654.11 Declarators 32166================ 32167 32168 * 'The maximum number of declarators that may modify an arithmetic, 32169 structure or union type (C90 6.5.4).' 32170 32171 GCC is only limited by available memory. 32172 32173 32174File: gcc.info, Node: Statements implementation, Next: Preprocessing directives implementation, Prev: Declarators implementation, Up: C Implementation 32175 321764.12 Statements 32177=============== 32178 32179 * 'The maximum number of 'case' values in a 'switch' statement (C90 32180 6.6.4.2).' 32181 32182 GCC is only limited by available memory. 32183 32184 32185File: gcc.info, Node: Preprocessing directives implementation, Next: Library functions implementation, Prev: Statements implementation, Up: C Implementation 32186 321874.13 Preprocessing Directives 32188============================= 32189 32190*Note Implementation-defined behavior: (cpp)Implementation-defined 32191behavior, for details of these aspects of implementation-defined 32192behavior. 32193 32194 * 'The locations within '#pragma' directives where header name 32195 preprocessing tokens are recognized (C11 6.4, C11 6.4.7).' 32196 32197 * 'How sequences in both forms of header names are mapped to headers 32198 or external source file names (C90 6.1.7, C99 and C11 6.4.7).' 32199 32200 * 'Whether the value of a character constant in a constant expression 32201 that controls conditional inclusion matches the value of the same 32202 character constant in the execution character set (C90 6.8.1, C99 32203 and C11 6.10.1).' 32204 32205 * 'Whether the value of a single-character character constant in a 32206 constant expression that controls conditional inclusion may have a 32207 negative value (C90 6.8.1, C99 and C11 6.10.1).' 32208 32209 * 'The places that are searched for an included '<>' delimited 32210 header, and how the places are specified or the header is 32211 identified (C90 6.8.2, C99 and C11 6.10.2).' 32212 32213 * 'How the named source file is searched for in an included '""' 32214 delimited header (C90 6.8.2, C99 and C11 6.10.2).' 32215 32216 * 'The method by which preprocessing tokens (possibly resulting from 32217 macro expansion) in a '#include' directive are combined into a 32218 header name (C90 6.8.2, C99 and C11 6.10.2).' 32219 32220 * 'The nesting limit for '#include' processing (C90 6.8.2, C99 and 32221 C11 6.10.2).' 32222 32223 * 'Whether the '#' operator inserts a '\' character before the '\' 32224 character that begins a universal character name in a character 32225 constant or string literal (C99 and C11 6.10.3.2).' 32226 32227 * 'The behavior on each recognized non-'STDC #pragma' directive (C90 32228 6.8.6, C99 and C11 6.10.6).' 32229 32230 *Note Pragmas: (cpp)Pragmas, for details of pragmas accepted by GCC 32231 on all targets. *Note Pragmas Accepted by GCC: Pragmas, for 32232 details of target-specific pragmas. 32233 32234 * 'The definitions for '__DATE__' and '__TIME__' when respectively, 32235 the date and time of translation are not available (C90 6.8.8, C99 32236 6.10.8, C11 6.10.8.1).' 32237 32238 32239File: gcc.info, Node: Library functions implementation, Next: Architecture implementation, Prev: Preprocessing directives implementation, Up: C Implementation 32240 322414.14 Library Functions 32242====================== 32243 32244The behavior of most of these points are dependent on the implementation 32245of the C library, and are not defined by GCC itself. 32246 32247 * 'The null pointer constant to which the macro 'NULL' expands (C90 32248 7.1.6, C99 7.17, C11 7.19).' 32249 32250 In '<stddef.h>', 'NULL' expands to '((void *)0)'. GCC does not 32251 provide the other headers which define 'NULL' and some library 32252 implementations may use other definitions in those headers. 32253 32254 32255File: gcc.info, Node: Architecture implementation, Next: Locale-specific behavior implementation, Prev: Library functions implementation, Up: C Implementation 32256 322574.15 Architecture 32258================= 32259 32260 * 'The values or expressions assigned to the macros specified in the 32261 headers '<float.h>', '<limits.h>', and '<stdint.h>' (C90, C99 and 32262 C11 5.2.4.2, C99 7.18.2, C99 7.18.3, C11 7.20.2, C11 7.20.3).' 32263 32264 Determined by ABI. 32265 32266 * 'The result of attempting to indirectly access an object with 32267 automatic or thread storage duration from a thread other than the 32268 one with which it is associated (C11 6.2.4).' 32269 32270 Such accesses are supported, subject to the same requirements for 32271 synchronization for concurrent accesses as for concurrent accesses 32272 to any object. 32273 32274 * 'The number, order, and encoding of bytes in any object (when not 32275 explicitly specified in this International Standard) (C99 and C11 32276 6.2.6.1).' 32277 32278 Determined by ABI. 32279 32280 * 'Whether any extended alignments are supported and the contexts in 32281 which they are supported (C11 6.2.8).' 32282 32283 Extended alignments up to 2^{28} (bytes) are supported for objects 32284 of automatic storage duration. Alignments supported for objects of 32285 static and thread storage duration are determined by the ABI. 32286 32287 * 'Valid alignment values other than those returned by an _Alignof 32288 expression for fundamental types, if any (C11 6.2.8).' 32289 32290 Valid alignments are powers of 2 up to and including 2^{28}. 32291 32292 * 'The value of the result of the 'sizeof' and '_Alignof' operators 32293 (C90 6.3.3.4, C99 and C11 6.5.3.4).' 32294 32295 Determined by ABI. 32296 32297 32298File: gcc.info, Node: Locale-specific behavior implementation, Prev: Architecture implementation, Up: C Implementation 32299 323004.16 Locale-Specific Behavior 32301============================= 32302 32303The behavior of these points are dependent on the implementation of the 32304C library, and are not defined by GCC itself. 32305 32306 32307File: gcc.info, Node: C++ Implementation, Next: C Extensions, Prev: C Implementation, Up: Top 32308 323095 C++ Implementation-Defined Behavior 32310************************************* 32311 32312A conforming implementation of ISO C++ is required to document its 32313choice of behavior in each of the areas that are designated 32314"implementation defined". The following lists all such areas, along 32315with the section numbers from the ISO/IEC 14882:1998 and ISO/IEC 3231614882:2003 standards. Some areas are only implementation-defined in one 32317version of the standard. 32318 32319 Some choices depend on the externally determined ABI for the platform 32320(including standard character encodings) which GCC follows; these are 32321listed as "determined by ABI" below. *Note Binary Compatibility: 32322Compatibility, and <https://gcc.gnu.org/readings.html>. Some choices 32323are documented in the preprocessor manual. *Note Implementation-defined 32324behavior: (cpp)Implementation-defined behavior. Some choices are 32325documented in the corresponding document for the C language. *Note C 32326Implementation::. Some choices are made by the library and operating 32327system (or other environment when compiling for a freestanding 32328environment); refer to their documentation for details. 32329 32330* Menu: 32331 32332* Conditionally-supported behavior:: 32333* Exception handling:: 32334 32335 32336File: gcc.info, Node: Conditionally-supported behavior, Next: Exception handling, Up: C++ Implementation 32337 323385.1 Conditionally-Supported Behavior 32339==================================== 32340 32341'Each implementation shall include documentation that identifies all 32342conditionally-supported constructs that it does not support (C++0x 323431.4).' 32344 32345 * 'Whether an argument of class type with a non-trivial copy 32346 constructor or destructor can be passed to ... (C++0x 5.2.2).' 32347 32348 Such argument passing is supported, using the same 32349 pass-by-invisible-reference approach used for normal function 32350 arguments of such types. 32351 32352 32353File: gcc.info, Node: Exception handling, Prev: Conditionally-supported behavior, Up: C++ Implementation 32354 323555.2 Exception Handling 32356====================== 32357 32358 * 'In the situation where no matching handler is found, it is 32359 implementation-defined whether or not the stack is unwound before 32360 std::terminate() is called (C++98 15.5.1).' 32361 32362 The stack is not unwound before std::terminate is called. 32363 32364 c Copyright (C) 1988-2022 Free Software Foundation, Inc. 32365 32366 32367File: gcc.info, Node: C Extensions, Next: C++ Extensions, Prev: C++ Implementation, Up: Top 32368 323696 Extensions to the C Language Family 32370************************************* 32371 32372GNU C provides several language features not found in ISO standard C. 32373(The '-pedantic' option directs GCC to print a warning message if any of 32374these features is used.) To test for the availability of these features 32375in conditional compilation, check for a predefined macro '__GNUC__', 32376which is always defined under GCC. 32377 32378 These extensions are available in C and Objective-C. Most of them are 32379also available in C++. *Note Extensions to the C++ Language: C++ 32380Extensions, for extensions that apply _only_ to C++. 32381 32382 Some features that are in ISO C99 but not C90 or C++ are also, as 32383extensions, accepted by GCC in C90 mode and in C++. 32384 32385* Menu: 32386 32387* Statement Exprs:: Putting statements and declarations inside expressions. 32388* Local Labels:: Labels local to a block. 32389* Labels as Values:: Getting pointers to labels, and computed gotos. 32390* Nested Functions:: Nested function in GNU C. 32391* Nonlocal Gotos:: Nonlocal gotos. 32392* Constructing Calls:: Dispatching a call to another function. 32393* Typeof:: 'typeof': referring to the type of an expression. 32394* Conditionals:: Omitting the middle operand of a '?:' expression. 32395* __int128:: 128-bit integers--'__int128'. 32396* Long Long:: Double-word integers--'long long int'. 32397* Complex:: Data types for complex numbers. 32398* Floating Types:: Additional Floating Types. 32399* Half-Precision:: Half-Precision Floating Point. 32400* Decimal Float:: Decimal Floating Types. 32401* Hex Floats:: Hexadecimal floating-point constants. 32402* Fixed-Point:: Fixed-Point Types. 32403* Named Address Spaces::Named address spaces. 32404* Zero Length:: Zero-length arrays. 32405* Empty Structures:: Structures with no members. 32406* Variable Length:: Arrays whose length is computed at run time. 32407* Variadic Macros:: Macros with a variable number of arguments. 32408* Escaped Newlines:: Slightly looser rules for escaped newlines. 32409* Subscripting:: Any array can be subscripted, even if not an lvalue. 32410* Pointer Arith:: Arithmetic on 'void'-pointers and function pointers. 32411* Variadic Pointer Args:: Pointer arguments to variadic functions. 32412* Pointers to Arrays:: Pointers to arrays with qualifiers work as expected. 32413* Initializers:: Non-constant initializers. 32414* Compound Literals:: Compound literals give structures, unions 32415 or arrays as values. 32416* Designated Inits:: Labeling elements of initializers. 32417* Case Ranges:: 'case 1 ... 9' and such. 32418* Cast to Union:: Casting to union type from any member of the union. 32419* Mixed Labels and Declarations:: Mixing declarations, labels and code. 32420* Function Attributes:: Declaring that functions have no side effects, 32421 or that they can never return. 32422* Variable Attributes:: Specifying attributes of variables. 32423* Type Attributes:: Specifying attributes of types. 32424* Label Attributes:: Specifying attributes on labels. 32425* Enumerator Attributes:: Specifying attributes on enumerators. 32426* Statement Attributes:: Specifying attributes on statements. 32427* Attribute Syntax:: Formal syntax for attributes. 32428* Function Prototypes:: Prototype declarations and old-style definitions. 32429* C++ Comments:: C++ comments are recognized. 32430* Dollar Signs:: Dollar sign is allowed in identifiers. 32431* Character Escapes:: '\e' stands for the character <ESC>. 32432* Alignment:: Determining the alignment of a function, type or variable. 32433* Inline:: Defining inline functions (as fast as macros). 32434* Volatiles:: What constitutes an access to a volatile object. 32435* Using Assembly Language with C:: Instructions and extensions for interfacing C with assembler. 32436* Alternate Keywords:: '__const__', '__asm__', etc., for header files. 32437* Incomplete Enums:: 'enum foo;', with details to follow. 32438* Function Names:: Printable strings which are the name of the current 32439 function. 32440* Return Address:: Getting the return or frame address of a function. 32441* Vector Extensions:: Using vector instructions through built-in functions. 32442* Offsetof:: Special syntax for implementing 'offsetof'. 32443* __sync Builtins:: Legacy built-in functions for atomic memory access. 32444* __atomic Builtins:: Atomic built-in functions with memory model. 32445* Integer Overflow Builtins:: Built-in functions to perform arithmetics and 32446 arithmetic overflow checking. 32447* x86 specific memory model extensions for transactional memory:: x86 memory models. 32448* Object Size Checking:: Built-in functions for limited buffer overflow 32449 checking. 32450* Other Builtins:: Other built-in functions. 32451* Target Builtins:: Built-in functions specific to particular targets. 32452* Target Format Checks:: Format checks specific to particular targets. 32453* Pragmas:: Pragmas accepted by GCC. 32454* Unnamed Fields:: Unnamed struct/union fields within structs/unions. 32455* Thread-Local:: Per-thread variables. 32456* Binary constants:: Binary constants using the '0b' prefix. 32457 32458 32459File: gcc.info, Node: Statement Exprs, Next: Local Labels, Up: C Extensions 32460 324616.1 Statements and Declarations in Expressions 32462============================================== 32463 32464A compound statement enclosed in parentheses may appear as an expression 32465in GNU C. This allows you to use loops, switches, and local variables 32466within an expression. 32467 32468 Recall that a compound statement is a sequence of statements surrounded 32469by braces; in this construct, parentheses go around the braces. For 32470example: 32471 32472 ({ int y = foo (); int z; 32473 if (y > 0) z = y; 32474 else z = - y; 32475 z; }) 32476 32477is a valid (though slightly more complex than necessary) expression for 32478the absolute value of 'foo ()'. 32479 32480 The last thing in the compound statement should be an expression 32481followed by a semicolon; the value of this subexpression serves as the 32482value of the entire construct. (If you use some other kind of statement 32483last within the braces, the construct has type 'void', and thus 32484effectively no value.) 32485 32486 This feature is especially useful in making macro definitions "safe" 32487(so that they evaluate each operand exactly once). For example, the 32488"maximum" function is commonly defined as a macro in standard C as 32489follows: 32490 32491 #define max(a,b) ((a) > (b) ? (a) : (b)) 32492 32493But this definition computes either A or B twice, with bad results if 32494the operand has side effects. In GNU C, if you know the type of the 32495operands (here taken as 'int'), you can avoid this problem by defining 32496the macro as follows: 32497 32498 #define maxint(a,b) \ 32499 ({int _a = (a), _b = (b); _a > _b ? _a : _b; }) 32500 32501 Note that introducing variable declarations (as we do in 'maxint') can 32502cause variable shadowing, so while this example using the 'max' macro 32503produces correct results: 32504 int _a = 1, _b = 2, c; 32505 c = max (_a, _b); 32506this example using maxint will not: 32507 int _a = 1, _b = 2, c; 32508 c = maxint (_a, _b); 32509 32510 This problem may for instance occur when we use this pattern 32511recursively, like so: 32512 32513 #define maxint3(a, b, c) \ 32514 ({int _a = (a), _b = (b), _c = (c); maxint (maxint (_a, _b), _c); }) 32515 32516 Embedded statements are not allowed in constant expressions, such as 32517the value of an enumeration constant, the width of a bit-field, or the 32518initial value of a static variable. 32519 32520 If you don't know the type of the operand, you can still do this, but 32521you must use 'typeof' or '__auto_type' (*note Typeof::). 32522 32523 In G++, the result value of a statement expression undergoes array and 32524function pointer decay, and is returned by value to the enclosing 32525expression. For instance, if 'A' is a class, then 32526 32527 A a; 32528 32529 ({a;}).Foo () 32530 32531constructs a temporary 'A' object to hold the result of the statement 32532expression, and that is used to invoke 'Foo'. Therefore the 'this' 32533pointer observed by 'Foo' is not the address of 'a'. 32534 32535 In a statement expression, any temporaries created within a statement 32536are destroyed at that statement's end. This makes statement expressions 32537inside macros slightly different from function calls. In the latter 32538case temporaries introduced during argument evaluation are destroyed at 32539the end of the statement that includes the function call. In the 32540statement expression case they are destroyed during the statement 32541expression. For instance, 32542 32543 #define macro(a) ({__typeof__(a) b = (a); b + 3; }) 32544 template<typename T> T function(T a) { T b = a; return b + 3; } 32545 32546 void foo () 32547 { 32548 macro (X ()); 32549 function (X ()); 32550 } 32551 32552has different places where temporaries are destroyed. For the 'macro' 32553case, the temporary 'X' is destroyed just after the initialization of 32554'b'. In the 'function' case that temporary is destroyed when the 32555function returns. 32556 32557 These considerations mean that it is probably a bad idea to use 32558statement expressions of this form in header files that are designed to 32559work with C++. (Note that some versions of the GNU C Library contained 32560header files using statement expressions that lead to precisely this 32561bug.) 32562 32563 Jumping into a statement expression with 'goto' or using a 'switch' 32564statement outside the statement expression with a 'case' or 'default' 32565label inside the statement expression is not permitted. Jumping into a 32566statement expression with a computed 'goto' (*note Labels as Values::) 32567has undefined behavior. Jumping out of a statement expression is 32568permitted, but if the statement expression is part of a larger 32569expression then it is unspecified which other subexpressions of that 32570expression have been evaluated except where the language definition 32571requires certain subexpressions to be evaluated before or after the 32572statement expression. A 'break' or 'continue' statement inside of a 32573statement expression used in 'while', 'do' or 'for' loop or 'switch' 32574statement condition or 'for' statement init or increment expressions 32575jumps to an outer loop or 'switch' statement if any (otherwise it is an 32576error), rather than to the loop or 'switch' statement in whose condition 32577or init or increment expression it appears. In any case, as with a 32578function call, the evaluation of a statement expression is not 32579interleaved with the evaluation of other parts of the containing 32580expression. For example, 32581 32582 foo (), (({ bar1 (); goto a; 0; }) + bar2 ()), baz(); 32583 32584calls 'foo' and 'bar1' and does not call 'baz' but may or may not call 32585'bar2'. If 'bar2' is called, it is called after 'foo' and before 32586'bar1'. 32587 32588 32589File: gcc.info, Node: Local Labels, Next: Labels as Values, Prev: Statement Exprs, Up: C Extensions 32590 325916.2 Locally Declared Labels 32592=========================== 32593 32594GCC allows you to declare "local labels" in any nested block scope. A 32595local label is just like an ordinary label, but you can only reference 32596it (with a 'goto' statement, or by taking its address) within the block 32597in which it is declared. 32598 32599 A local label declaration looks like this: 32600 32601 __label__ LABEL; 32602 32603or 32604 32605 __label__ LABEL1, LABEL2, /* ... */; 32606 32607 Local label declarations must come at the beginning of the block, 32608before any ordinary declarations or statements. 32609 32610 The label declaration defines the label _name_, but does not define the 32611label itself. You must do this in the usual way, with 'LABEL:', within 32612the statements of the statement expression. 32613 32614 The local label feature is useful for complex macros. If a macro 32615contains nested loops, a 'goto' can be useful for breaking out of them. 32616However, an ordinary label whose scope is the whole function cannot be 32617used: if the macro can be expanded several times in one function, the 32618label is multiply defined in that function. A local label avoids this 32619problem. For example: 32620 32621 #define SEARCH(value, array, target) \ 32622 do { \ 32623 __label__ found; \ 32624 typeof (target) _SEARCH_target = (target); \ 32625 typeof (*(array)) *_SEARCH_array = (array); \ 32626 int i, j; \ 32627 int value; \ 32628 for (i = 0; i < max; i++) \ 32629 for (j = 0; j < max; j++) \ 32630 if (_SEARCH_array[i][j] == _SEARCH_target) \ 32631 { (value) = i; goto found; } \ 32632 (value) = -1; \ 32633 found:; \ 32634 } while (0) 32635 32636 This could also be written using a statement expression: 32637 32638 #define SEARCH(array, target) \ 32639 ({ \ 32640 __label__ found; \ 32641 typeof (target) _SEARCH_target = (target); \ 32642 typeof (*(array)) *_SEARCH_array = (array); \ 32643 int i, j; \ 32644 int value; \ 32645 for (i = 0; i < max; i++) \ 32646 for (j = 0; j < max; j++) \ 32647 if (_SEARCH_array[i][j] == _SEARCH_target) \ 32648 { value = i; goto found; } \ 32649 value = -1; \ 32650 found: \ 32651 value; \ 32652 }) 32653 32654 Local label declarations also make the labels they declare visible to 32655nested functions, if there are any. *Note Nested Functions::, for 32656details. 32657 32658 32659File: gcc.info, Node: Labels as Values, Next: Nested Functions, Prev: Local Labels, Up: C Extensions 32660 326616.3 Labels as Values 32662==================== 32663 32664You can get the address of a label defined in the current function (or a 32665containing function) with the unary operator '&&'. The value has type 32666'void *'. This value is a constant and can be used wherever a constant 32667of that type is valid. For example: 32668 32669 void *ptr; 32670 /* ... */ 32671 ptr = &&foo; 32672 32673 To use these values, you need to be able to jump to one. This is done 32674with the computed goto statement(1), 'goto *EXP;'. For example, 32675 32676 goto *ptr; 32677 32678Any expression of type 'void *' is allowed. 32679 32680 One way of using these constants is in initializing a static array that 32681serves as a jump table: 32682 32683 static void *array[] = { &&foo, &&bar, &&hack }; 32684 32685Then you can select a label with indexing, like this: 32686 32687 goto *array[i]; 32688 32689Note that this does not check whether the subscript is in bounds--array 32690indexing in C never does that. 32691 32692 Such an array of label values serves a purpose much like that of the 32693'switch' statement. The 'switch' statement is cleaner, so use that 32694rather than an array unless the problem does not fit a 'switch' 32695statement very well. 32696 32697 Another use of label values is in an interpreter for threaded code. 32698The labels within the interpreter function can be stored in the threaded 32699code for super-fast dispatching. 32700 32701 You may not use this mechanism to jump to code in a different function. 32702If you do that, totally unpredictable things happen. The best way to 32703avoid this is to store the label address only in automatic variables and 32704never pass it as an argument. 32705 32706 An alternate way to write the above example is 32707 32708 static const int array[] = { &&foo - &&foo, &&bar - &&foo, 32709 &&hack - &&foo }; 32710 goto *(&&foo + array[i]); 32711 32712This is more friendly to code living in shared libraries, as it reduces 32713the number of dynamic relocations that are needed, and by consequence, 32714allows the data to be read-only. This alternative with label 32715differences is not supported for the AVR target, please use the first 32716approach for AVR programs. 32717 32718 The '&&foo' expressions for the same label might have different values 32719if the containing function is inlined or cloned. If a program relies on 32720them being always the same, '__attribute__((__noinline__,__noclone__))' 32721should be used to prevent inlining and cloning. If '&&foo' is used in a 32722static variable initializer, inlining and cloning is forbidden. 32723 32724 ---------- Footnotes ---------- 32725 32726 (1) The analogous feature in Fortran is called an assigned goto, but 32727that name seems inappropriate in C, where one can do more than simply 32728store label addresses in label variables. 32729 32730 32731File: gcc.info, Node: Nested Functions, Next: Nonlocal Gotos, Prev: Labels as Values, Up: C Extensions 32732 327336.4 Nested Functions 32734==================== 32735 32736A "nested function" is a function defined inside another function. 32737Nested functions are supported as an extension in GNU C, but are not 32738supported by GNU C++. 32739 32740 The nested function's name is local to the block where it is defined. 32741For example, here we define a nested function named 'square', and call 32742it twice: 32743 32744 foo (double a, double b) 32745 { 32746 double square (double z) { return z * z; } 32747 32748 return square (a) + square (b); 32749 } 32750 32751 The nested function can access all the variables of the containing 32752function that are visible at the point of its definition. This is 32753called "lexical scoping". For example, here we show a nested function 32754which uses an inherited variable named 'offset': 32755 32756 bar (int *array, int offset, int size) 32757 { 32758 int access (int *array, int index) 32759 { return array[index + offset]; } 32760 int i; 32761 /* ... */ 32762 for (i = 0; i < size; i++) 32763 /* ... */ access (array, i) /* ... */ 32764 } 32765 32766 Nested function definitions are permitted within functions in the 32767places where variable definitions are allowed; that is, in any block, 32768mixed with the other declarations and statements in the block. 32769 32770 It is possible to call the nested function from outside the scope of 32771its name by storing its address or passing the address to another 32772function: 32773 32774 hack (int *array, int size) 32775 { 32776 void store (int index, int value) 32777 { array[index] = value; } 32778 32779 intermediate (store, size); 32780 } 32781 32782 Here, the function 'intermediate' receives the address of 'store' as an 32783argument. If 'intermediate' calls 'store', the arguments given to 32784'store' are used to store into 'array'. But this technique works only 32785so long as the containing function ('hack', in this example) does not 32786exit. 32787 32788 If you try to call the nested function through its address after the 32789containing function exits, all hell breaks loose. If you try to call it 32790after a containing scope level exits, and if it refers to some of the 32791variables that are no longer in scope, you may be lucky, but it's not 32792wise to take the risk. If, however, the nested function does not refer 32793to anything that has gone out of scope, you should be safe. 32794 32795 GCC implements taking the address of a nested function using a 32796technique called "trampolines". This technique was described in 32797'Lexical Closures for C++' (Thomas M. Breuel, USENIX C++ Conference 32798Proceedings, October 17-21, 1988). 32799 32800 A nested function can jump to a label inherited from a containing 32801function, provided the label is explicitly declared in the containing 32802function (*note Local Labels::). Such a jump returns instantly to the 32803containing function, exiting the nested function that did the 'goto' and 32804any intermediate functions as well. Here is an example: 32805 32806 bar (int *array, int offset, int size) 32807 { 32808 __label__ failure; 32809 int access (int *array, int index) 32810 { 32811 if (index > size) 32812 goto failure; 32813 return array[index + offset]; 32814 } 32815 int i; 32816 /* ... */ 32817 for (i = 0; i < size; i++) 32818 /* ... */ access (array, i) /* ... */ 32819 /* ... */ 32820 return 0; 32821 32822 /* Control comes here from 'access' 32823 if it detects an error. */ 32824 failure: 32825 return -1; 32826 } 32827 32828 A nested function always has no linkage. Declaring one with 'extern' 32829or 'static' is erroneous. If you need to declare the nested function 32830before its definition, use 'auto' (which is otherwise meaningless for 32831function declarations). 32832 32833 bar (int *array, int offset, int size) 32834 { 32835 __label__ failure; 32836 auto int access (int *, int); 32837 /* ... */ 32838 int access (int *array, int index) 32839 { 32840 if (index > size) 32841 goto failure; 32842 return array[index + offset]; 32843 } 32844 /* ... */ 32845 } 32846 32847 32848File: gcc.info, Node: Nonlocal Gotos, Next: Constructing Calls, Prev: Nested Functions, Up: C Extensions 32849 328506.5 Nonlocal Gotos 32851================== 32852 32853GCC provides the built-in functions '__builtin_setjmp' and 32854'__builtin_longjmp' which are similar to, but not interchangeable with, 32855the C library functions 'setjmp' and 'longjmp'. The built-in versions 32856are used internally by GCC's libraries to implement exception handling 32857on some targets. You should use the standard C library functions 32858declared in '<setjmp.h>' in user code instead of the builtins. 32859 32860 The built-in versions of these functions use GCC's normal mechanisms to 32861save and restore registers using the stack on function entry and exit. 32862The jump buffer argument BUF holds only the information needed to 32863restore the stack frame, rather than the entire set of saved register 32864values. 32865 32866 An important caveat is that GCC arranges to save and restore only those 32867registers known to the specific architecture variant being compiled for. 32868This can make '__builtin_setjmp' and '__builtin_longjmp' more efficient 32869than their library counterparts in some cases, but it can also cause 32870incorrect and mysterious behavior when mixing with code that uses the 32871full register set. 32872 32873 You should declare the jump buffer argument BUF to the built-in 32874functions as: 32875 32876 #include <stdint.h> 32877 intptr_t BUF[5]; 32878 32879 -- Built-in Function: int __builtin_setjmp (intptr_t *BUF) 32880 This function saves the current stack context in BUF. 32881 '__builtin_setjmp' returns 0 when returning directly, and 1 when 32882 returning from '__builtin_longjmp' using the same BUF. 32883 32884 -- Built-in Function: void __builtin_longjmp (intptr_t *BUF, int VAL) 32885 This function restores the stack context in BUF, saved by a 32886 previous call to '__builtin_setjmp'. After '__builtin_longjmp' is 32887 finished, the program resumes execution as if the matching 32888 '__builtin_setjmp' returns the value VAL, which must be 1. 32889 32890 Because '__builtin_longjmp' depends on the function return 32891 mechanism to restore the stack context, it cannot be called from 32892 the same function calling '__builtin_setjmp' to initialize BUF. It 32893 can only be called from a function called (directly or indirectly) 32894 from the function calling '__builtin_setjmp'. 32895 32896 32897File: gcc.info, Node: Constructing Calls, Next: Typeof, Prev: Nonlocal Gotos, Up: C Extensions 32898 328996.6 Constructing Function Calls 32900=============================== 32901 32902Using the built-in functions described below, you can record the 32903arguments a function received, and call another function with the same 32904arguments, without knowing the number or types of the arguments. 32905 32906 You can also record the return value of that function call, and later 32907return that value, without knowing what data type the function tried to 32908return (as long as your caller expects that data type). 32909 32910 However, these built-in functions may interact badly with some 32911sophisticated features or other extensions of the language. It is, 32912therefore, not recommended to use them outside very simple functions 32913acting as mere forwarders for their arguments. 32914 32915 -- Built-in Function: void * __builtin_apply_args () 32916 This built-in function returns a pointer to data describing how to 32917 perform a call with the same arguments as are passed to the current 32918 function. 32919 32920 The function saves the arg pointer register, structure value 32921 address, and all registers that might be used to pass arguments to 32922 a function into a block of memory allocated on the stack. Then it 32923 returns the address of that block. 32924 32925 -- Built-in Function: void * __builtin_apply (void (*FUNCTION)(), void 32926 *ARGUMENTS, size_t SIZE) 32927 This built-in function invokes FUNCTION with a copy of the 32928 parameters described by ARGUMENTS and SIZE. 32929 32930 The value of ARGUMENTS should be the value returned by 32931 '__builtin_apply_args'. The argument SIZE specifies the size of 32932 the stack argument data, in bytes. 32933 32934 This function returns a pointer to data describing how to return 32935 whatever value is returned by FUNCTION. The data is saved in a 32936 block of memory allocated on the stack. 32937 32938 It is not always simple to compute the proper value for SIZE. The 32939 value is used by '__builtin_apply' to compute the amount of data 32940 that should be pushed on the stack and copied from the incoming 32941 argument area. 32942 32943 -- Built-in Function: void __builtin_return (void *RESULT) 32944 This built-in function returns the value described by RESULT from 32945 the containing function. You should specify, for RESULT, a value 32946 returned by '__builtin_apply'. 32947 32948 -- Built-in Function: __builtin_va_arg_pack () 32949 This built-in function represents all anonymous arguments of an 32950 inline function. It can be used only in inline functions that are 32951 always inlined, never compiled as a separate function, such as 32952 those using '__attribute__ ((__always_inline__))' or '__attribute__ 32953 ((__gnu_inline__))' extern inline functions. It must be only 32954 passed as last argument to some other function with variable 32955 arguments. This is useful for writing small wrapper inlines for 32956 variable argument functions, when using preprocessor macros is 32957 undesirable. For example: 32958 extern int myprintf (FILE *f, const char *format, ...); 32959 extern inline __attribute__ ((__gnu_inline__)) int 32960 myprintf (FILE *f, const char *format, ...) 32961 { 32962 int r = fprintf (f, "myprintf: "); 32963 if (r < 0) 32964 return r; 32965 int s = fprintf (f, format, __builtin_va_arg_pack ()); 32966 if (s < 0) 32967 return s; 32968 return r + s; 32969 } 32970 32971 -- Built-in Function: int __builtin_va_arg_pack_len () 32972 This built-in function returns the number of anonymous arguments of 32973 an inline function. It can be used only in inline functions that 32974 are always inlined, never compiled as a separate function, such as 32975 those using '__attribute__ ((__always_inline__))' or '__attribute__ 32976 ((__gnu_inline__))' extern inline functions. For example following 32977 does link- or run-time checking of open arguments for optimized 32978 code: 32979 #ifdef __OPTIMIZE__ 32980 extern inline __attribute__((__gnu_inline__)) int 32981 myopen (const char *path, int oflag, ...) 32982 { 32983 if (__builtin_va_arg_pack_len () > 1) 32984 warn_open_too_many_arguments (); 32985 32986 if (__builtin_constant_p (oflag)) 32987 { 32988 if ((oflag & O_CREAT) != 0 && __builtin_va_arg_pack_len () < 1) 32989 { 32990 warn_open_missing_mode (); 32991 return __open_2 (path, oflag); 32992 } 32993 return open (path, oflag, __builtin_va_arg_pack ()); 32994 } 32995 32996 if (__builtin_va_arg_pack_len () < 1) 32997 return __open_2 (path, oflag); 32998 32999 return open (path, oflag, __builtin_va_arg_pack ()); 33000 } 33001 #endif 33002 33003 33004File: gcc.info, Node: Typeof, Next: Conditionals, Prev: Constructing Calls, Up: C Extensions 33005 330066.7 Referring to a Type with 'typeof' 33007===================================== 33008 33009Another way to refer to the type of an expression is with 'typeof'. The 33010syntax of using of this keyword looks like 'sizeof', but the construct 33011acts semantically like a type name defined with 'typedef'. 33012 33013 There are two ways of writing the argument to 'typeof': with an 33014expression or with a type. Here is an example with an expression: 33015 33016 typeof (x[0](1)) 33017 33018This assumes that 'x' is an array of pointers to functions; the type 33019described is that of the values of the functions. 33020 33021 Here is an example with a typename as the argument: 33022 33023 typeof (int *) 33024 33025Here the type described is that of pointers to 'int'. 33026 33027 If you are writing a header file that must work when included in ISO C 33028programs, write '__typeof__' instead of 'typeof'. *Note Alternate 33029Keywords::. 33030 33031 A 'typeof' construct can be used anywhere a typedef name can be used. 33032For example, you can use it in a declaration, in a cast, or inside of 33033'sizeof' or 'typeof'. 33034 33035 The operand of 'typeof' is evaluated for its side effects if and only 33036if it is an expression of variably modified type or the name of such a 33037type. 33038 33039 'typeof' is often useful in conjunction with statement expressions 33040(*note Statement Exprs::). Here is how the two together can be used to 33041define a safe "maximum" macro which operates on any arithmetic type and 33042evaluates each of its arguments exactly once: 33043 33044 #define max(a,b) \ 33045 ({ typeof (a) _a = (a); \ 33046 typeof (b) _b = (b); \ 33047 _a > _b ? _a : _b; }) 33048 33049 The reason for using names that start with underscores for the local 33050variables is to avoid conflicts with variable names that occur within 33051the expressions that are substituted for 'a' and 'b'. Eventually we 33052hope to design a new form of declaration syntax that allows you to 33053declare variables whose scopes start only after their initializers; this 33054will be a more reliable way to prevent such conflicts. 33055 33056Some more examples of the use of 'typeof': 33057 33058 * This declares 'y' with the type of what 'x' points to. 33059 33060 typeof (*x) y; 33061 33062 * This declares 'y' as an array of such values. 33063 33064 typeof (*x) y[4]; 33065 33066 * This declares 'y' as an array of pointers to characters: 33067 33068 typeof (typeof (char *)[4]) y; 33069 33070 It is equivalent to the following traditional C declaration: 33071 33072 char *y[4]; 33073 33074 To see the meaning of the declaration using 'typeof', and why it 33075 might be a useful way to write, rewrite it with these macros: 33076 33077 #define pointer(T) typeof(T *) 33078 #define array(T, N) typeof(T [N]) 33079 33080 Now the declaration can be rewritten this way: 33081 33082 array (pointer (char), 4) y; 33083 33084 Thus, 'array (pointer (char), 4)' is the type of arrays of 4 33085 pointers to 'char'. 33086 33087 In GNU C, but not GNU C++, you may also declare the type of a variable 33088as '__auto_type'. In that case, the declaration must declare only one 33089variable, whose declarator must just be an identifier, the declaration 33090must be initialized, and the type of the variable is determined by the 33091initializer; the name of the variable is not in scope until after the 33092initializer. (In C++, you should use C++11 'auto' for this purpose.) 33093Using '__auto_type', the "maximum" macro above could be written as: 33094 33095 #define max(a,b) \ 33096 ({ __auto_type _a = (a); \ 33097 __auto_type _b = (b); \ 33098 _a > _b ? _a : _b; }) 33099 33100 Using '__auto_type' instead of 'typeof' has two advantages: 33101 33102 * Each argument to the macro appears only once in the expansion of 33103 the macro. This prevents the size of the macro expansion growing 33104 exponentially when calls to such macros are nested inside arguments 33105 of such macros. 33106 33107 * If the argument to the macro has variably modified type, it is 33108 evaluated only once when using '__auto_type', but twice if 'typeof' 33109 is used. 33110 33111 33112File: gcc.info, Node: Conditionals, Next: __int128, Prev: Typeof, Up: C Extensions 33113 331146.8 Conditionals with Omitted Operands 33115====================================== 33116 33117The middle operand in a conditional expression may be omitted. Then if 33118the first operand is nonzero, its value is the value of the conditional 33119expression. 33120 33121 Therefore, the expression 33122 33123 x ? : y 33124 33125has the value of 'x' if that is nonzero; otherwise, the value of 'y'. 33126 33127 This example is perfectly equivalent to 33128 33129 x ? x : y 33130 33131In this simple case, the ability to omit the middle operand is not 33132especially useful. When it becomes useful is when the first operand 33133does, or may (if it is a macro argument), contain a side effect. Then 33134repeating the operand in the middle would perform the side effect twice. 33135Omitting the middle operand uses the value already computed without the 33136undesirable effects of recomputing it. 33137 33138 33139File: gcc.info, Node: __int128, Next: Long Long, Prev: Conditionals, Up: C Extensions 33140 331416.9 128-bit Integers 33142==================== 33143 33144As an extension the integer scalar type '__int128' is supported for 33145targets which have an integer mode wide enough to hold 128 bits. Simply 33146write '__int128' for a signed 128-bit integer, or 'unsigned __int128' 33147for an unsigned 128-bit integer. There is no support in GCC for 33148expressing an integer constant of type '__int128' for targets with 'long 33149long' integer less than 128 bits wide. 33150 33151 33152File: gcc.info, Node: Long Long, Next: Complex, Prev: __int128, Up: C Extensions 33153 331546.10 Double-Word Integers 33155========================= 33156 33157ISO C99 and ISO C++11 support data types for integers that are at least 3315864 bits wide, and as an extension GCC supports them in C90 and C++98 33159modes. Simply write 'long long int' for a signed integer, or 'unsigned 33160long long int' for an unsigned integer. To make an integer constant of 33161type 'long long int', add the suffix 'LL' to the integer. To make an 33162integer constant of type 'unsigned long long int', add the suffix 'ULL' 33163to the integer. 33164 33165 You can use these types in arithmetic like any other integer types. 33166Addition, subtraction, and bitwise boolean operations on these types are 33167open-coded on all types of machines. Multiplication is open-coded if 33168the machine supports a fullword-to-doubleword widening multiply 33169instruction. Division and shifts are open-coded only on machines that 33170provide special support. The operations that are not open-coded use 33171special library routines that come with GCC. 33172 33173 There may be pitfalls when you use 'long long' types for function 33174arguments without function prototypes. If a function expects type 'int' 33175for its argument, and you pass a value of type 'long long int', 33176confusion results because the caller and the subroutine disagree about 33177the number of bytes for the argument. Likewise, if the function expects 33178'long long int' and you pass 'int'. The best way to avoid such problems 33179is to use prototypes. 33180 33181 33182File: gcc.info, Node: Complex, Next: Floating Types, Prev: Long Long, Up: C Extensions 33183 331846.11 Complex Numbers 33185==================== 33186 33187ISO C99 supports complex floating data types, and as an extension GCC 33188supports them in C90 mode and in C++. GCC also supports complex integer 33189data types which are not part of ISO C99. You can declare complex types 33190using the keyword '_Complex'. As an extension, the older GNU keyword 33191'__complex__' is also supported. 33192 33193 For example, '_Complex double x;' declares 'x' as a variable whose real 33194part and imaginary part are both of type 'double'. '_Complex short int 33195y;' declares 'y' to have real and imaginary parts of type 'short int'; 33196this is not likely to be useful, but it shows that the set of complex 33197types is complete. 33198 33199 To write a constant with a complex data type, use the suffix 'i' or 'j' 33200(either one; they are equivalent). For example, '2.5fi' has type 33201'_Complex float' and '3i' has type '_Complex int'. Such a constant 33202always has a pure imaginary value, but you can form any complex value 33203you like by adding one to a real constant. This is a GNU extension; if 33204you have an ISO C99 conforming C library (such as the GNU C Library), 33205and want to construct complex constants of floating type, you should 33206include '<complex.h>' and use the macros 'I' or '_Complex_I' instead. 33207 33208 The ISO C++14 library also defines the 'i' suffix, so C++14 code that 33209includes the '<complex>' header cannot use 'i' for the GNU extension. 33210The 'j' suffix still has the GNU meaning. 33211 33212 GCC can handle both implicit and explicit casts between the '_Complex' 33213types and other '_Complex' types as casting both the real and imaginary 33214parts to the scalar type. GCC can handle implicit and explicit casts 33215from a scalar type to a '_Complex' type and where the imaginary part 33216will be considered zero. The C front-end can handle implicit and 33217explicit casts from a '_Complex' type to a scalar type where the 33218imaginary part will be ignored. In C++ code, this cast is considered 33219illformed and G++ will error out. 33220 33221 GCC provides a built-in function '__builtin_complex' will can be used 33222to construct a complex value. 33223 33224 GCC has a few extensions which can be used to extract the real and the 33225imaginary part of the complex-valued expression. Note these expressions 33226are lvalues if the EXP is an lvalue. These expressions operands have 33227the type of a complex type which might get prompoted to a complex type 33228from a scalar type. E.g. '__real__ (int)X' is the same as casting to 33229'_Complex int' before '__real__' is done. 33230 33231Expression Description 33232-------------------------------------------------------------------------- 33233'__real__ EXP' Extract the real part of EXP. 33234'__imag__ EXP' Extract the imaginary part of EXP. 33235 33236 For values of floating point, you should use the ISO C99 functions, 33237declared in '<complex.h>' and also provided as built-in functions by 33238GCC. 33239 33240Expression float double long double 33241--------------------------------------------------------------------------- 33242'__real__ EXP' 'crealf' 'creal' 'creall' 33243'__imag__ EXP' 'cimagf' 'cimag' 'cimagl' 33244 33245 The operator '~' performs complex conjugation when used on a value with 33246a complex type. This is a GNU extension; for values of floating type, 33247you should use the ISO C99 functions 'conjf', 'conj' and 'conjl', 33248declared in '<complex.h>' and also provided as built-in functions by 33249GCC. Note unlike the '__real__' and '__imag__' operators, this operator 33250will not do an implicit cast to the complex type because the '~' is 33251already a normal operator. 33252 33253 GCC can allocate complex automatic variables in a noncontiguous 33254fashion; it's even possible for the real part to be in a register while 33255the imaginary part is on the stack (or vice versa). Only the DWARF 33256debug info format can represent this, so use of DWARF is recommended. 33257If you are using the stabs debug info format, GCC describes a 33258noncontiguous complex variable as if it were two separate variables of 33259noncomplex type. If the variable's actual name is 'foo', the two 33260fictitious variables are named 'foo$real' and 'foo$imag'. You can 33261examine and set these two fictitious variables with your debugger. 33262 33263 -- Built-in Function: TYPE __builtin_complex (REAL, IMAG) 33264 33265 The built-in function '__builtin_complex' is provided for use in 33266 implementing the ISO C11 macros 'CMPLXF', 'CMPLX' and 'CMPLXL'. 33267 REAL and IMAG must have the same type, a real binary floating-point 33268 type, and the result has the corresponding complex type with real 33269 and imaginary parts REAL and IMAG. Unlike 'REAL + I * IMAG', this 33270 works even when infinities, NaNs and negative zeros are involved. 33271 33272 33273File: gcc.info, Node: Floating Types, Next: Half-Precision, Prev: Complex, Up: C Extensions 33274 332756.12 Additional Floating Types 33276============================== 33277 33278ISO/IEC TS 18661-3:2015 defines C support for additional floating types 33279'_FloatN' and '_FloatNx', and GCC supports these type names; the set of 33280types supported depends on the target architecture. These types are not 33281supported when compiling C++. Constants with these types use suffixes 33282'fN' or 'FN' and 'fNx' or 'FNx'. These type names can be used together 33283with '_Complex' to declare complex types. 33284 33285 As an extension, GNU C and GNU C++ support additional floating types, 33286which are not supported by all targets. 33287 * '__float128' is available on i386, x86_64, IA-64, and hppa HP-UX, 33288 as well as on PowerPC GNU/Linux targets that enable the vector 33289 scalar (VSX) instruction set. '__float128' supports the 128-bit 33290 floating type. On i386, x86_64, PowerPC, and IA-64 other than 33291 HP-UX, '__float128' is an alias for '_Float128'. On hppa and IA-64 33292 HP-UX, '__float128' is an alias for 'long double'. 33293 33294 * '__float80' is available on the i386, x86_64, and IA-64 targets, 33295 and supports the 80-bit ('XFmode') floating type. It is an alias 33296 for the type name '_Float64x' on these targets. 33297 33298 * '__ibm128' is available on PowerPC targets, and provides access to 33299 the IBM extended double format which is the current format used for 33300 'long double'. When 'long double' transitions to '__float128' on 33301 PowerPC in the future, '__ibm128' will remain for use in 33302 conversions between the two types. 33303 33304 Support for these additional types includes the arithmetic operators: 33305add, subtract, multiply, divide; unary arithmetic operators; relational 33306operators; equality operators; and conversions to and from integer and 33307other floating types. Use a suffix 'w' or 'W' in a literal constant of 33308type '__float80' or type '__ibm128'. Use a suffix 'q' or 'Q' for 33309'_float128'. 33310 33311 In order to use '_Float128', '__float128', and '__ibm128' on PowerPC 33312Linux systems, you must use the '-mfloat128' option. It is expected in 33313future versions of GCC that '_Float128' and '__float128' will be enabled 33314automatically. 33315 33316 The '_Float128' type is supported on all systems where '__float128' is 33317supported or where 'long double' has the IEEE binary128 format. The 33318'_Float64x' type is supported on all systems where '__float128' is 33319supported. The '_Float32' type is supported on all systems supporting 33320IEEE binary32; the '_Float64' and '_Float32x' types are supported on all 33321systems supporting IEEE binary64. The '_Float16' type is supported on 33322AArch64 systems by default, on ARM systems when the IEEE format for 3332316-bit floating-point types is selected with '-mfp16-format=ieee' and, 33324for both C and C++, on x86 systems with SSE2 enabled. GCC does not 33325currently support '_Float128x' on any systems. 33326 33327 On the i386, x86_64, IA-64, and HP-UX targets, you can declare complex 33328types using the corresponding internal complex type, 'XCmode' for 33329'__float80' type and 'TCmode' for '__float128' type: 33330 33331 typedef _Complex float __attribute__((mode(TC))) _Complex128; 33332 typedef _Complex float __attribute__((mode(XC))) _Complex80; 33333 33334 On the PowerPC Linux VSX targets, you can declare complex types using 33335the corresponding internal complex type, 'KCmode' for '__float128' type 33336and 'ICmode' for '__ibm128' type: 33337 33338 typedef _Complex float __attribute__((mode(KC))) _Complex_float128; 33339 typedef _Complex float __attribute__((mode(IC))) _Complex_ibm128; 33340 33341 33342File: gcc.info, Node: Half-Precision, Next: Decimal Float, Prev: Floating Types, Up: C Extensions 33343 333446.13 Half-Precision Floating Point 33345================================== 33346 33347On ARM and AArch64 targets, GCC supports half-precision (16-bit) 33348floating point via the '__fp16' type defined in the ARM C Language 33349Extensions. On ARM systems, you must enable this type explicitly with 33350the '-mfp16-format' command-line option in order to use it. On x86 33351targets with SSE2 enabled, GCC supports half-precision (16-bit) floating 33352point via the '_Float16' type. For C++, x86 provides a builtin type 33353named '_Float16' which contains same data format as C. 33354 33355 ARM targets support two incompatible representations for half-precision 33356floating-point values. You must choose one of the representations and 33357use it consistently in your program. 33358 33359 Specifying '-mfp16-format=ieee' selects the IEEE 754-2008 format. This 33360format can represent normalized values in the range of 2^{-14} to 65504. 33361There are 11 bits of significand precision, approximately 3 decimal 33362digits. 33363 33364 Specifying '-mfp16-format=alternative' selects the ARM alternative 33365format. This representation is similar to the IEEE format, but does not 33366support infinities or NaNs. Instead, the range of exponents is 33367extended, so that this format can represent normalized values in the 33368range of 2^{-14} to 131008. 33369 33370 The GCC port for AArch64 only supports the IEEE 754-2008 format, and 33371does not require use of the '-mfp16-format' command-line option. 33372 33373 The '__fp16' type may only be used as an argument to intrinsics defined 33374in '<arm_fp16.h>', or as a storage format. For purposes of arithmetic 33375and other operations, '__fp16' values in C or C++ expressions are 33376automatically promoted to 'float'. 33377 33378 The ARM target provides hardware support for conversions between 33379'__fp16' and 'float' values as an extension to VFP and NEON (Advanced 33380SIMD), and from ARMv8-A provides hardware support for conversions 33381between '__fp16' and 'double' values. GCC generates code using these 33382hardware instructions if you compile with options to select an FPU that 33383provides them; for example, '-mfpu=neon-fp16 -mfloat-abi=softfp', in 33384addition to the '-mfp16-format' option to select a half-precision 33385format. 33386 33387 Language-level support for the '__fp16' data type is independent of 33388whether GCC generates code using hardware floating-point instructions. 33389In cases where hardware support is not specified, GCC implements 33390conversions between '__fp16' and other types as library calls. 33391 33392 It is recommended that portable code use the '_Float16' type defined by 33393ISO/IEC TS 18661-3:2015. *Note Floating Types::. 33394 33395 On x86 targets with SSE2 enabled, without '-mavx512fp16', all 33396operations will be emulated by software emulation and the 'float' 33397instructions. The default behavior for 'FLT_EVAL_METHOD' is to keep the 33398intermediate result of the operation as 32-bit precision. This may lead 33399to inconsistent behavior between software emulation and AVX512-FP16 33400instructions. Using '-fexcess-precision=16' will force round back after 33401each operation. 33402 33403 Using '-mavx512fp16' will generate AVX512-FP16 instructions instead of 33404software emulation. The default behavior of 'FLT_EVAL_METHOD' is to 33405round after each operation. The same is true with 33406'-fexcess-precision=standard' and '-mfpmath=sse'. If there is no 33407'-mfpmath=sse', '-fexcess-precision=standard' alone does the same thing 33408as before, It is useful for code that does not have '_Float16' and runs 33409on the x87 FPU. 33410 33411 33412File: gcc.info, Node: Decimal Float, Next: Hex Floats, Prev: Half-Precision, Up: C Extensions 33413 334146.14 Decimal Floating Types 33415=========================== 33416 33417As an extension, GNU C supports decimal floating types as defined in the 33418N1312 draft of ISO/IEC WDTR24732. Support for decimal floating types in 33419GCC will evolve as the draft technical report changes. Calling 33420conventions for any target might also change. Not all targets support 33421decimal floating types. 33422 33423 The decimal floating types are '_Decimal32', '_Decimal64', and 33424'_Decimal128'. They use a radix of ten, unlike the floating types 33425'float', 'double', and 'long double' whose radix is not specified by the 33426C standard but is usually two. 33427 33428 Support for decimal floating types includes the arithmetic operators 33429add, subtract, multiply, divide; unary arithmetic operators; relational 33430operators; equality operators; and conversions to and from integer and 33431other floating types. Use a suffix 'df' or 'DF' in a literal constant 33432of type '_Decimal32', 'dd' or 'DD' for '_Decimal64', and 'dl' or 'DL' 33433for '_Decimal128'. 33434 33435 GCC support of decimal float as specified by the draft technical report 33436is incomplete: 33437 33438 * When the value of a decimal floating type cannot be represented in 33439 the integer type to which it is being converted, the result is 33440 undefined rather than the result value specified by the draft 33441 technical report. 33442 33443 * GCC does not provide the C library functionality associated with 33444 'math.h', 'fenv.h', 'stdio.h', 'stdlib.h', and 'wchar.h', which 33445 must come from a separate C library implementation. Because of 33446 this the GNU C compiler does not define macro '__STDC_DEC_FP__' to 33447 indicate that the implementation conforms to the technical report. 33448 33449 Types '_Decimal32', '_Decimal64', and '_Decimal128' are supported by 33450the DWARF debug information format. 33451 33452 33453File: gcc.info, Node: Hex Floats, Next: Fixed-Point, Prev: Decimal Float, Up: C Extensions 33454 334556.15 Hex Floats 33456=============== 33457 33458ISO C99 and ISO C++17 support floating-point numbers written not only in 33459the usual decimal notation, such as '1.55e1', but also numbers such as 33460'0x1.fp3' written in hexadecimal format. As a GNU extension, GCC 33461supports this in C90 mode (except in some cases when strictly 33462conforming) and in C++98, C++11 and C++14 modes. In that format the 33463'0x' hex introducer and the 'p' or 'P' exponent field are mandatory. 33464The exponent is a decimal number that indicates the power of 2 by which 33465the significant part is multiplied. Thus '0x1.f' is 1 15/16, 'p3' 33466multiplies it by 8, and the value of '0x1.fp3' is the same as '1.55e1'. 33467 33468 Unlike for floating-point numbers in the decimal notation the exponent 33469is always required in the hexadecimal notation. Otherwise the compiler 33470would not be able to resolve the ambiguity of, e.g., '0x1.f'. This 33471could mean '1.0f' or '1.9375' since 'f' is also the extension for 33472floating-point constants of type 'float'. 33473 33474 33475File: gcc.info, Node: Fixed-Point, Next: Named Address Spaces, Prev: Hex Floats, Up: C Extensions 33476 334776.16 Fixed-Point Types 33478====================== 33479 33480As an extension, GNU C supports fixed-point types as defined in the 33481N1169 draft of ISO/IEC DTR 18037. Support for fixed-point types in GCC 33482will evolve as the draft technical report changes. Calling conventions 33483for any target might also change. Not all targets support fixed-point 33484types. 33485 33486 The fixed-point types are 'short _Fract', '_Fract', 'long _Fract', 33487'long long _Fract', 'unsigned short _Fract', 'unsigned _Fract', 33488'unsigned long _Fract', 'unsigned long long _Fract', '_Sat short 33489_Fract', '_Sat _Fract', '_Sat long _Fract', '_Sat long long _Fract', 33490'_Sat unsigned short _Fract', '_Sat unsigned _Fract', '_Sat unsigned 33491long _Fract', '_Sat unsigned long long _Fract', 'short _Accum', 33492'_Accum', 'long _Accum', 'long long _Accum', 'unsigned short _Accum', 33493'unsigned _Accum', 'unsigned long _Accum', 'unsigned long long _Accum', 33494'_Sat short _Accum', '_Sat _Accum', '_Sat long _Accum', '_Sat long long 33495_Accum', '_Sat unsigned short _Accum', '_Sat unsigned _Accum', '_Sat 33496unsigned long _Accum', '_Sat unsigned long long _Accum'. 33497 33498 Fixed-point data values contain fractional and optional integral parts. 33499The format of fixed-point data varies and depends on the target machine. 33500 33501 Support for fixed-point types includes: 33502 * prefix and postfix increment and decrement operators ('++', '--') 33503 * unary arithmetic operators ('+', '-', '!') 33504 * binary arithmetic operators ('+', '-', '*', '/') 33505 * binary shift operators ('<<', '>>') 33506 * relational operators ('<', '<=', '>=', '>') 33507 * equality operators ('==', '!=') 33508 * assignment operators ('+=', '-=', '*=', '/=', '<<=', '>>=') 33509 * conversions to and from integer, floating-point, or fixed-point 33510 types 33511 33512 Use a suffix in a fixed-point literal constant: 33513 * 'hr' or 'HR' for 'short _Fract' and '_Sat short _Fract' 33514 * 'r' or 'R' for '_Fract' and '_Sat _Fract' 33515 * 'lr' or 'LR' for 'long _Fract' and '_Sat long _Fract' 33516 * 'llr' or 'LLR' for 'long long _Fract' and '_Sat long long _Fract' 33517 * 'uhr' or 'UHR' for 'unsigned short _Fract' and '_Sat unsigned short 33518 _Fract' 33519 * 'ur' or 'UR' for 'unsigned _Fract' and '_Sat unsigned _Fract' 33520 * 'ulr' or 'ULR' for 'unsigned long _Fract' and '_Sat unsigned long 33521 _Fract' 33522 * 'ullr' or 'ULLR' for 'unsigned long long _Fract' and '_Sat unsigned 33523 long long _Fract' 33524 * 'hk' or 'HK' for 'short _Accum' and '_Sat short _Accum' 33525 * 'k' or 'K' for '_Accum' and '_Sat _Accum' 33526 * 'lk' or 'LK' for 'long _Accum' and '_Sat long _Accum' 33527 * 'llk' or 'LLK' for 'long long _Accum' and '_Sat long long _Accum' 33528 * 'uhk' or 'UHK' for 'unsigned short _Accum' and '_Sat unsigned short 33529 _Accum' 33530 * 'uk' or 'UK' for 'unsigned _Accum' and '_Sat unsigned _Accum' 33531 * 'ulk' or 'ULK' for 'unsigned long _Accum' and '_Sat unsigned long 33532 _Accum' 33533 * 'ullk' or 'ULLK' for 'unsigned long long _Accum' and '_Sat unsigned 33534 long long _Accum' 33535 33536 GCC support of fixed-point types as specified by the draft technical 33537report is incomplete: 33538 33539 * Pragmas to control overflow and rounding behaviors are not 33540 implemented. 33541 33542 Fixed-point types are supported by the DWARF debug information format. 33543 33544 33545File: gcc.info, Node: Named Address Spaces, Next: Zero Length, Prev: Fixed-Point, Up: C Extensions 33546 335476.17 Named Address Spaces 33548========================= 33549 33550As an extension, GNU C supports named address spaces as defined in the 33551N1275 draft of ISO/IEC DTR 18037. Support for named address spaces in 33552GCC will evolve as the draft technical report changes. Calling 33553conventions for any target might also change. At present, only the AVR, 33554M32C, PRU, RL78, and x86 targets support address spaces other than the 33555generic address space. 33556 33557 Address space identifiers may be used exactly like any other C type 33558qualifier (e.g., 'const' or 'volatile'). See the N1275 document for 33559more details. 33560 335616.17.1 AVR Named Address Spaces 33562------------------------------- 33563 33564On the AVR target, there are several address spaces that can be used in 33565order to put read-only data into the flash memory and access that data 33566by means of the special instructions 'LPM' or 'ELPM' needed to read from 33567flash. 33568 33569 Devices belonging to 'avrtiny' and 'avrxmega3' can access flash memory 33570by means of 'LD*' instructions because the flash memory is mapped into 33571the RAM address space. There is _no need_ for language extensions like 33572'__flash' or attribute *note 'progmem': AVR Variable Attributes. The 33573default linker description files for these devices cater for that 33574feature and '.rodata' stays in flash: The compiler just generates 'LD*' 33575instructions, and the linker script adds core specific offsets to all 33576'.rodata' symbols: '0x4000' in the case of 'avrtiny' and '0x8000' in the 33577case of 'avrxmega3'. See *note AVR Options:: for a list of respective 33578devices. 33579 33580 For devices not in 'avrtiny' or 'avrxmega3', any data including 33581read-only data is located in RAM (the generic address space) because 33582flash memory is not visible in the RAM address space. In order to 33583locate read-only data in flash memory _and_ to generate the right 33584instructions to access this data without using (inline) assembler code, 33585special address spaces are needed. 33586 33587'__flash' 33588 The '__flash' qualifier locates data in the '.progmem.data' 33589 section. Data is read using the 'LPM' instruction. Pointers to 33590 this address space are 16 bits wide. 33591 33592'__flash1' 33593'__flash2' 33594'__flash3' 33595'__flash4' 33596'__flash5' 33597 These are 16-bit address spaces locating data in section 33598 '.progmemN.data' where N refers to address space '__flashN'. The 33599 compiler sets the 'RAMPZ' segment register appropriately before 33600 reading data by means of the 'ELPM' instruction. 33601 33602'__memx' 33603 This is a 24-bit address space that linearizes flash and RAM: If 33604 the high bit of the address is set, data is read from RAM using the 33605 lower two bytes as RAM address. If the high bit of the address is 33606 clear, data is read from flash with 'RAMPZ' set according to the 33607 high byte of the address. *Note '__builtin_avr_flash_segment': AVR 33608 Built-in Functions. 33609 33610 Objects in this address space are located in '.progmemx.data'. 33611 33612 Example 33613 33614 char my_read (const __flash char ** p) 33615 { 33616 /* p is a pointer to RAM that points to a pointer to flash. 33617 The first indirection of p reads that flash pointer 33618 from RAM and the second indirection reads a char from this 33619 flash address. */ 33620 33621 return **p; 33622 } 33623 33624 /* Locate array[] in flash memory */ 33625 const __flash int array[] = { 3, 5, 7, 11, 13, 17, 19 }; 33626 33627 int i = 1; 33628 33629 int main (void) 33630 { 33631 /* Return 17 by reading from flash memory */ 33632 return array[array[i]]; 33633 } 33634 33635For each named address space supported by avr-gcc there is an equally 33636named but uppercase built-in macro defined. The purpose is to 33637facilitate testing if respective address space support is available or 33638not: 33639 33640 #ifdef __FLASH 33641 const __flash int var = 1; 33642 33643 int read_var (void) 33644 { 33645 return var; 33646 } 33647 #else 33648 #include <avr/pgmspace.h> /* From AVR-LibC */ 33649 33650 const int var PROGMEM = 1; 33651 33652 int read_var (void) 33653 { 33654 return (int) pgm_read_word (&var); 33655 } 33656 #endif /* __FLASH */ 33657 33658Notice that attribute *note 'progmem': AVR Variable Attributes. locates 33659data in flash but accesses to these data read from generic address 33660space, i.e. from RAM, so that you need special accessors like 33661'pgm_read_byte' from AVR-LibC (http://nongnu.org/avr-libc/user-manual/) 33662together with attribute 'progmem'. 33663 33664Limitations and caveats 33665 33666 * Reading across the 64 KiB section boundary of the '__flash' or 33667 '__flashN' address spaces shows undefined behavior. The only 33668 address space that supports reading across the 64 KiB flash segment 33669 boundaries is '__memx'. 33670 33671 * If you use one of the '__flashN' address spaces you must arrange 33672 your linker script to locate the '.progmemN.data' sections 33673 according to your needs. 33674 33675 * Any data or pointers to the non-generic address spaces must be 33676 qualified as 'const', i.e. as read-only data. This still applies 33677 if the data in one of these address spaces like software version 33678 number or calibration lookup table are intended to be changed after 33679 load time by, say, a boot loader. In this case the right 33680 qualification is 'const' 'volatile' so that the compiler must not 33681 optimize away known values or insert them as immediates into 33682 operands of instructions. 33683 33684 * The following code initializes a variable 'pfoo' located in static 33685 storage with a 24-bit address: 33686 extern const __memx char foo; 33687 const __memx void *pfoo = &foo; 33688 33689 * On the reduced Tiny devices like ATtiny40, no address spaces are 33690 supported. Just use vanilla C / C++ code without overhead as 33691 outlined above. Attribute 'progmem' is supported but works 33692 differently, see *note AVR Variable Attributes::. 33693 336946.17.2 M32C Named Address Spaces 33695-------------------------------- 33696 33697On the M32C target, with the R8C and M16C CPU variants, variables 33698qualified with '__far' are accessed using 32-bit addresses in order to 33699access memory beyond the first 64 Ki bytes. If '__far' is used with the 33700M32CM or M32C CPU variants, it has no effect. 33701 337026.17.3 PRU Named Address Spaces 33703------------------------------- 33704 33705On the PRU target, variables qualified with '__regio_symbol' are aliases 33706used to access the special I/O CPU registers. They must be declared as 33707'extern' because such variables will not be allocated in any data 33708memory. They must also be marked as 'volatile', and can only be 32-bit 33709integer types. The only names those variables can have are '__R30' and 33710'__R31', representing respectively the 'R30' and 'R31' special I/O CPU 33711registers. Hence the following example is the only valid usage of 33712'__regio_symbol': 33713 33714 extern volatile __regio_symbol uint32_t __R30; 33715 extern volatile __regio_symbol uint32_t __R31; 33716 337176.17.4 RL78 Named Address Spaces 33718-------------------------------- 33719 33720On the RL78 target, variables qualified with '__far' are accessed with 3372132-bit pointers (20-bit addresses) rather than the default 16-bit 33722addresses. Non-far variables are assumed to appear in the topmost 3372364 KiB of the address space. 33724 337256.17.5 x86 Named Address Spaces 33726------------------------------- 33727 33728On the x86 target, variables may be declared as being relative to the 33729'%fs' or '%gs' segments. 33730 33731'__seg_fs' 33732'__seg_gs' 33733 The object is accessed with the respective segment override prefix. 33734 33735 The respective segment base must be set via some method specific to 33736 the operating system. Rather than require an expensive system call 33737 to retrieve the segment base, these address spaces are not 33738 considered to be subspaces of the generic (flat) address space. 33739 This means that explicit casts are required to convert pointers 33740 between these address spaces and the generic address space. In 33741 practice the application should cast to 'uintptr_t' and apply the 33742 segment base offset that it installed previously. 33743 33744 The preprocessor symbols '__SEG_FS' and '__SEG_GS' are defined when 33745 these address spaces are supported. 33746 33747 33748File: gcc.info, Node: Zero Length, Next: Empty Structures, Prev: Named Address Spaces, Up: C Extensions 33749 337506.18 Arrays of Length Zero 33751========================== 33752 33753Declaring zero-length arrays is allowed in GNU C as an extension. A 33754zero-length array can be useful as the last element of a structure that 33755is really a header for a variable-length object: 33756 33757 struct line { 33758 int length; 33759 char contents[0]; 33760 }; 33761 33762 struct line *thisline = (struct line *) 33763 malloc (sizeof (struct line) + this_length); 33764 thisline->length = this_length; 33765 33766 Although the size of a zero-length array is zero, an array member of 33767this kind may increase the size of the enclosing type as a result of 33768tail padding. The offset of a zero-length array member from the 33769beginning of the enclosing structure is the same as the offset of an 33770array with one or more elements of the same type. The alignment of a 33771zero-length array is the same as the alignment of its elements. 33772 33773 Declaring zero-length arrays in other contexts, including as interior 33774members of structure objects or as non-member objects, is discouraged. 33775Accessing elements of zero-length arrays declared in such contexts is 33776undefined and may be diagnosed. 33777 33778 In the absence of the zero-length array extension, in ISO C90 the 33779'contents' array in the example above would typically be declared to 33780have a single element. Unlike a zero-length array which only 33781contributes to the size of the enclosing structure for the purposes of 33782alignment, a one-element array always occupies at least as much space as 33783a single object of the type. Although using one-element arrays this way 33784is discouraged, GCC handles accesses to trailing one-element array 33785members analogously to zero-length arrays. 33786 33787 The preferred mechanism to declare variable-length types like 'struct 33788line' above is the ISO C99 "flexible array member", with slightly 33789different syntax and semantics: 33790 33791 * Flexible array members are written as 'contents[]' without the '0'. 33792 33793 * Flexible array members have incomplete type, and so the 'sizeof' 33794 operator may not be applied. As a quirk of the original 33795 implementation of zero-length arrays, 'sizeof' evaluates to zero. 33796 33797 * Flexible array members may only appear as the last member of a 33798 'struct' that is otherwise non-empty. 33799 33800 * A structure containing a flexible array member, or a union 33801 containing such a structure (possibly recursively), may not be a 33802 member of a structure or an element of an array. (However, these 33803 uses are permitted by GCC as extensions.) 33804 33805 Non-empty initialization of zero-length arrays is treated like any case 33806where there are more initializer elements than the array holds, in that 33807a suitable warning about "excess elements in array" is given, and the 33808excess elements (all of them, in this case) are ignored. 33809 33810 GCC allows static initialization of flexible array members. This is 33811equivalent to defining a new structure containing the original structure 33812followed by an array of sufficient size to contain the data. E.g. in 33813the following, 'f1' is constructed as if it were declared like 'f2'. 33814 33815 struct f1 { 33816 int x; int y[]; 33817 } f1 = { 1, { 2, 3, 4 } }; 33818 33819 struct f2 { 33820 struct f1 f1; int data[3]; 33821 } f2 = { { 1 }, { 2, 3, 4 } }; 33822 33823The convenience of this extension is that 'f1' has the desired type, 33824eliminating the need to consistently refer to 'f2.f1'. 33825 33826 This has symmetry with normal static arrays, in that an array of 33827unknown size is also written with '[]'. 33828 33829 Of course, this extension only makes sense if the extra data comes at 33830the end of a top-level object, as otherwise we would be overwriting data 33831at subsequent offsets. To avoid undue complication and confusion with 33832initialization of deeply nested arrays, we simply disallow any non-empty 33833initialization except when the structure is the top-level object. For 33834example: 33835 33836 struct foo { int x; int y[]; }; 33837 struct bar { struct foo z; }; 33838 33839 struct foo a = { 1, { 2, 3, 4 } }; // Valid. 33840 struct bar b = { { 1, { 2, 3, 4 } } }; // Invalid. 33841 struct bar c = { { 1, { } } }; // Valid. 33842 struct foo d[1] = { { 1, { 2, 3, 4 } } }; // Invalid. 33843 33844 33845File: gcc.info, Node: Empty Structures, Next: Variable Length, Prev: Zero Length, Up: C Extensions 33846 338476.19 Structures with No Members 33848=============================== 33849 33850GCC permits a C structure to have no members: 33851 33852 struct empty { 33853 }; 33854 33855 The structure has size zero. In C++, empty structures are part of the 33856language. G++ treats empty structures as if they had a single member of 33857type 'char'. 33858 33859 33860File: gcc.info, Node: Variable Length, Next: Variadic Macros, Prev: Empty Structures, Up: C Extensions 33861 338626.20 Arrays of Variable Length 33863============================== 33864 33865Variable-length automatic arrays are allowed in ISO C99, and as an 33866extension GCC accepts them in C90 mode and in C++. These arrays are 33867declared like any other automatic arrays, but with a length that is not 33868a constant expression. The storage is allocated at the point of 33869declaration and deallocated when the block scope containing the 33870declaration exits. For example: 33871 33872 FILE * 33873 concat_fopen (char *s1, char *s2, char *mode) 33874 { 33875 char str[strlen (s1) + strlen (s2) + 1]; 33876 strcpy (str, s1); 33877 strcat (str, s2); 33878 return fopen (str, mode); 33879 } 33880 33881 Jumping or breaking out of the scope of the array name deallocates the 33882storage. Jumping into the scope is not allowed; you get an error 33883message for it. 33884 33885 As an extension, GCC accepts variable-length arrays as a member of a 33886structure or a union. For example: 33887 33888 void 33889 foo (int n) 33890 { 33891 struct S { int x[n]; }; 33892 } 33893 33894 You can use the function 'alloca' to get an effect much like 33895variable-length arrays. The function 'alloca' is available in many 33896other C implementations (but not in all). On the other hand, 33897variable-length arrays are more elegant. 33898 33899 There are other differences between these two methods. Space allocated 33900with 'alloca' exists until the containing _function_ returns. The space 33901for a variable-length array is deallocated as soon as the array name's 33902scope ends, unless you also use 'alloca' in this scope. 33903 33904 You can also use variable-length arrays as arguments to functions: 33905 33906 struct entry 33907 tester (int len, char data[len][len]) 33908 { 33909 /* ... */ 33910 } 33911 33912 The length of an array is computed once when the storage is allocated 33913and is remembered for the scope of the array in case you access it with 33914'sizeof'. 33915 33916 If you want to pass the array first and the length afterward, you can 33917use a forward declaration in the parameter list--another GNU extension. 33918 33919 struct entry 33920 tester (int len; char data[len][len], int len) 33921 { 33922 /* ... */ 33923 } 33924 33925 The 'int len' before the semicolon is a "parameter forward 33926declaration", and it serves the purpose of making the name 'len' known 33927when the declaration of 'data' is parsed. 33928 33929 You can write any number of such parameter forward declarations in the 33930parameter list. They can be separated by commas or semicolons, but the 33931last one must end with a semicolon, which is followed by the "real" 33932parameter declarations. Each forward declaration must match a "real" 33933declaration in parameter name and data type. ISO C99 does not support 33934parameter forward declarations. 33935 33936 33937File: gcc.info, Node: Variadic Macros, Next: Escaped Newlines, Prev: Variable Length, Up: C Extensions 33938 339396.21 Macros with a Variable Number of Arguments. 33940================================================ 33941 33942In the ISO C standard of 1999, a macro can be declared to accept a 33943variable number of arguments much as a function can. The syntax for 33944defining the macro is similar to that of a function. Here is an 33945example: 33946 33947 #define debug(format, ...) fprintf (stderr, format, __VA_ARGS__) 33948 33949Here '...' is a "variable argument". In the invocation of such a macro, 33950it represents the zero or more tokens until the closing parenthesis that 33951ends the invocation, including any commas. This set of tokens replaces 33952the identifier '__VA_ARGS__' in the macro body wherever it appears. See 33953the CPP manual for more information. 33954 33955 GCC has long supported variadic macros, and used a different syntax 33956that allowed you to give a name to the variable arguments just like any 33957other argument. Here is an example: 33958 33959 #define debug(format, args...) fprintf (stderr, format, args) 33960 33961This is in all ways equivalent to the ISO C example above, but arguably 33962more readable and descriptive. 33963 33964 GNU CPP has two further variadic macro extensions, and permits them to 33965be used with either of the above forms of macro definition. 33966 33967 In standard C, you are not allowed to leave the variable argument out 33968entirely; but you are allowed to pass an empty argument. For example, 33969this invocation is invalid in ISO C, because there is no comma after the 33970string: 33971 33972 debug ("A message") 33973 33974 GNU CPP permits you to completely omit the variable arguments in this 33975way. In the above examples, the compiler would complain, though since 33976the expansion of the macro still has the extra comma after the format 33977string. 33978 33979 To help solve this problem, CPP behaves specially for variable 33980arguments used with the token paste operator, '##'. If instead you 33981write 33982 33983 #define debug(format, ...) fprintf (stderr, format, ## __VA_ARGS__) 33984 33985and if the variable arguments are omitted or empty, the '##' operator 33986causes the preprocessor to remove the comma before it. If you do 33987provide some variable arguments in your macro invocation, GNU CPP does 33988not complain about the paste operation and instead places the variable 33989arguments after the comma. Just like any other pasted macro argument, 33990these arguments are not macro expanded. 33991 33992 33993File: gcc.info, Node: Escaped Newlines, Next: Subscripting, Prev: Variadic Macros, Up: C Extensions 33994 339956.22 Slightly Looser Rules for Escaped Newlines 33996=============================================== 33997 33998The preprocessor treatment of escaped newlines is more relaxed than that 33999specified by the C90 standard, which requires the newline to immediately 34000follow a backslash. GCC's implementation allows whitespace in the form 34001of spaces, horizontal and vertical tabs, and form feeds between the 34002backslash and the subsequent newline. The preprocessor issues a 34003warning, but treats it as a valid escaped newline and combines the two 34004lines to form a single logical line. This works within comments and 34005tokens, as well as between tokens. Comments are _not_ treated as 34006whitespace for the purposes of this relaxation, since they have not yet 34007been replaced with spaces. 34008 34009 34010File: gcc.info, Node: Subscripting, Next: Pointer Arith, Prev: Escaped Newlines, Up: C Extensions 34011 340126.23 Non-Lvalue Arrays May Have Subscripts 34013========================================== 34014 34015In ISO C99, arrays that are not lvalues still decay to pointers, and may 34016be subscripted, although they may not be modified or used after the next 34017sequence point and the unary '&' operator may not be applied to them. 34018As an extension, GNU C allows such arrays to be subscripted in C90 mode, 34019though otherwise they do not decay to pointers outside C99 mode. For 34020example, this is valid in GNU C though not valid in C90: 34021 34022 struct foo {int a[4];}; 34023 34024 struct foo f(); 34025 34026 bar (int index) 34027 { 34028 return f().a[index]; 34029 } 34030 34031 34032File: gcc.info, Node: Pointer Arith, Next: Variadic Pointer Args, Prev: Subscripting, Up: C Extensions 34033 340346.24 Arithmetic on 'void'- and Function-Pointers 34035================================================ 34036 34037In GNU C, addition and subtraction operations are supported on pointers 34038to 'void' and on pointers to functions. This is done by treating the 34039size of a 'void' or of a function as 1. 34040 34041 A consequence of this is that 'sizeof' is also allowed on 'void' and on 34042function types, and returns 1. 34043 34044 The option '-Wpointer-arith' requests a warning if these extensions are 34045used. 34046 34047 34048File: gcc.info, Node: Variadic Pointer Args, Next: Pointers to Arrays, Prev: Pointer Arith, Up: C Extensions 34049 340506.25 Pointer Arguments in Variadic Functions 34051============================================ 34052 34053Standard C requires that pointer types used with 'va_arg' in functions 34054with variable argument lists either must be compatible with that of the 34055actual argument, or that one type must be a pointer to 'void' and the 34056other a pointer to a character type. GNU C implements the POSIX XSI 34057extension that additionally permits the use of 'va_arg' with a pointer 34058type to receive arguments of any other pointer type. 34059 34060 In particular, in GNU C 'va_arg (ap, void *)' can safely be used to 34061consume an argument of any pointer type. 34062 34063 34064File: gcc.info, Node: Pointers to Arrays, Next: Initializers, Prev: Variadic Pointer Args, Up: C Extensions 34065 340666.26 Pointers to Arrays with Qualifiers Work as Expected 34067======================================================== 34068 34069In GNU C, pointers to arrays with qualifiers work similar to pointers to 34070other qualified types. For example, a value of type 'int (*)[5]' can be 34071used to initialize a variable of type 'const int (*)[5]'. These types 34072are incompatible in ISO C because the 'const' qualifier is formally 34073attached to the element type of the array and not the array itself. 34074 34075 extern void 34076 transpose (int N, int M, double out[M][N], const double in[N][M]); 34077 double x[3][2]; 34078 double y[2][3]; 34079 ... 34080 transpose(3, 2, y, x); 34081 34082 34083File: gcc.info, Node: Initializers, Next: Compound Literals, Prev: Pointers to Arrays, Up: C Extensions 34084 340856.27 Non-Constant Initializers 34086============================== 34087 34088As in standard C++ and ISO C99, the elements of an aggregate initializer 34089for an automatic variable are not required to be constant expressions in 34090GNU C. Here is an example of an initializer with run-time varying 34091elements: 34092 34093 foo (float f, float g) 34094 { 34095 float beat_freqs[2] = { f-g, f+g }; 34096 /* ... */ 34097 } 34098 34099 34100File: gcc.info, Node: Compound Literals, Next: Designated Inits, Prev: Initializers, Up: C Extensions 34101 341026.28 Compound Literals 34103====================== 34104 34105A compound literal looks like a cast of a brace-enclosed aggregate 34106initializer list. Its value is an object of the type specified in the 34107cast, containing the elements specified in the initializer. Unlike the 34108result of a cast, a compound literal is an lvalue. ISO C99 and later 34109support compound literals. As an extension, GCC supports compound 34110literals also in C90 mode and in C++, although as explained below, the 34111C++ semantics are somewhat different. 34112 34113 Usually, the specified type of a compound literal is a structure. 34114Assume that 'struct foo' and 'structure' are declared as shown: 34115 34116 struct foo {int a; char b[2];} structure; 34117 34118Here is an example of constructing a 'struct foo' with a compound 34119literal: 34120 34121 structure = ((struct foo) {x + y, 'a', 0}); 34122 34123This is equivalent to writing the following: 34124 34125 { 34126 struct foo temp = {x + y, 'a', 0}; 34127 structure = temp; 34128 } 34129 34130 You can also construct an array, though this is dangerous in C++, as 34131explained below. If all the elements of the compound literal are (made 34132up of) simple constant expressions suitable for use in initializers of 34133objects of static storage duration, then the compound literal can be 34134coerced to a pointer to its first element and used in such an 34135initializer, as shown here: 34136 34137 char **foo = (char *[]) { "x", "y", "z" }; 34138 34139 Compound literals for scalar types and union types are also allowed. 34140In the following example the variable 'i' is initialized to the value 34141'2', the result of incrementing the unnamed object created by the 34142compound literal. 34143 34144 int i = ++(int) { 1 }; 34145 34146 As a GNU extension, GCC allows initialization of objects with static 34147storage duration by compound literals (which is not possible in ISO C99 34148because the initializer is not a constant). It is handled as if the 34149object were initialized only with the brace-enclosed list if the types 34150of the compound literal and the object match. The elements of the 34151compound literal must be constant. If the object being initialized has 34152array type of unknown size, the size is determined by the size of the 34153compound literal. 34154 34155 static struct foo x = (struct foo) {1, 'a', 'b'}; 34156 static int y[] = (int []) {1, 2, 3}; 34157 static int z[] = (int [3]) {1}; 34158 34159The above lines are equivalent to the following: 34160 static struct foo x = {1, 'a', 'b'}; 34161 static int y[] = {1, 2, 3}; 34162 static int z[] = {1, 0, 0}; 34163 34164 In C, a compound literal designates an unnamed object with static or 34165automatic storage duration. In C++, a compound literal designates a 34166temporary object that only lives until the end of its full-expression. 34167As a result, well-defined C code that takes the address of a subobject 34168of a compound literal can be undefined in C++, so G++ rejects the 34169conversion of a temporary array to a pointer. For instance, if the 34170array compound literal example above appeared inside a function, any 34171subsequent use of 'foo' in C++ would have undefined behavior because the 34172lifetime of the array ends after the declaration of 'foo'. 34173 34174 As an optimization, G++ sometimes gives array compound literals longer 34175lifetimes: when the array either appears outside a function or has a 34176'const'-qualified type. If 'foo' and its initializer had elements of 34177type 'char *const' rather than 'char *', or if 'foo' were a global 34178variable, the array would have static storage duration. But it is 34179probably safest just to avoid the use of array compound literals in C++ 34180code. 34181 34182 34183File: gcc.info, Node: Designated Inits, Next: Case Ranges, Prev: Compound Literals, Up: C Extensions 34184 341856.29 Designated Initializers 34186============================ 34187 34188Standard C90 requires the elements of an initializer to appear in a 34189fixed order, the same as the order of the elements in the array or 34190structure being initialized. 34191 34192 In ISO C99 you can give the elements in any order, specifying the array 34193indices or structure field names they apply to, and GNU C allows this as 34194an extension in C90 mode as well. This extension is not implemented in 34195GNU C++. 34196 34197 To specify an array index, write '[INDEX] =' before the element value. 34198For example, 34199 34200 int a[6] = { [4] = 29, [2] = 15 }; 34201 34202is equivalent to 34203 34204 int a[6] = { 0, 0, 15, 0, 29, 0 }; 34205 34206The index values must be constant expressions, even if the array being 34207initialized is automatic. 34208 34209 An alternative syntax for this that has been obsolete since GCC 2.5 but 34210GCC still accepts is to write '[INDEX]' before the element value, with 34211no '='. 34212 34213 To initialize a range of elements to the same value, write '[FIRST ... 34214LAST] = VALUE'. This is a GNU extension. For example, 34215 34216 int widths[] = { [0 ... 9] = 1, [10 ... 99] = 2, [100] = 3 }; 34217 34218If the value in it has side effects, the side effects happen only once, 34219not for each initialized field by the range initializer. 34220 34221Note that the length of the array is the highest value specified plus 34222one. 34223 34224 In a structure initializer, specify the name of a field to initialize 34225with '.FIELDNAME =' before the element value. For example, given the 34226following structure, 34227 34228 struct point { int x, y; }; 34229 34230the following initialization 34231 34232 struct point p = { .y = yvalue, .x = xvalue }; 34233 34234is equivalent to 34235 34236 struct point p = { xvalue, yvalue }; 34237 34238 Another syntax that has the same meaning, obsolete since GCC 2.5, is 34239'FIELDNAME:', as shown here: 34240 34241 struct point p = { y: yvalue, x: xvalue }; 34242 34243 Omitted fields are implicitly initialized the same as for objects that 34244have static storage duration. 34245 34246 The '[INDEX]' or '.FIELDNAME' is known as a "designator". You can also 34247use a designator (or the obsolete colon syntax) when initializing a 34248union, to specify which element of the union should be used. For 34249example, 34250 34251 union foo { int i; double d; }; 34252 34253 union foo f = { .d = 4 }; 34254 34255converts 4 to a 'double' to store it in the union using the second 34256element. By contrast, casting 4 to type 'union foo' stores it into the 34257union as the integer 'i', since it is an integer. *Note Cast to 34258Union::. 34259 34260 You can combine this technique of naming elements with ordinary C 34261initialization of successive elements. Each initializer element that 34262does not have a designator applies to the next consecutive element of 34263the array or structure. For example, 34264 34265 int a[6] = { [1] = v1, v2, [4] = v4 }; 34266 34267is equivalent to 34268 34269 int a[6] = { 0, v1, v2, 0, v4, 0 }; 34270 34271 Labeling the elements of an array initializer is especially useful when 34272the indices are characters or belong to an 'enum' type. For example: 34273 34274 int whitespace[256] 34275 = { [' '] = 1, ['\t'] = 1, ['\h'] = 1, 34276 ['\f'] = 1, ['\n'] = 1, ['\r'] = 1 }; 34277 34278 You can also write a series of '.FIELDNAME' and '[INDEX]' designators 34279before an '=' to specify a nested subobject to initialize; the list is 34280taken relative to the subobject corresponding to the closest surrounding 34281brace pair. For example, with the 'struct point' declaration above: 34282 34283 struct point ptarray[10] = { [2].y = yv2, [2].x = xv2, [0].x = xv0 }; 34284 34285 If the same field is initialized multiple times, or overlapping fields 34286of a union are initialized, the value from the last initialization is 34287used. When a field of a union is itself a structure, the entire 34288structure from the last field initialized is used. If any previous 34289initializer has side effect, it is unspecified whether the side effect 34290happens or not. Currently, GCC discards the side-effecting initializer 34291expressions and issues a warning. 34292 34293 34294File: gcc.info, Node: Case Ranges, Next: Cast to Union, Prev: Designated Inits, Up: C Extensions 34295 342966.30 Case Ranges 34297================ 34298 34299You can specify a range of consecutive values in a single 'case' label, 34300like this: 34301 34302 case LOW ... HIGH: 34303 34304This has the same effect as the proper number of individual 'case' 34305labels, one for each integer value from LOW to HIGH, inclusive. 34306 34307 This feature is especially useful for ranges of ASCII character codes: 34308 34309 case 'A' ... 'Z': 34310 34311 *Be careful:* Write spaces around the '...', for otherwise it may be 34312parsed wrong when you use it with integer values. For example, write 34313this: 34314 34315 case 1 ... 5: 34316 34317rather than this: 34318 34319 case 1...5: 34320 34321 34322File: gcc.info, Node: Cast to Union, Next: Mixed Labels and Declarations, Prev: Case Ranges, Up: C Extensions 34323 343246.31 Cast to a Union Type 34325========================= 34326 34327A cast to a union type is a C extension not available in C++. It looks 34328just like ordinary casts with the constraint that the type specified is 34329a union type. You can specify the type either with the 'union' keyword 34330or with a 'typedef' name that refers to a union. The result of a cast 34331to a union is a temporary rvalue of the union type with a member whose 34332type matches that of the operand initialized to the value of the 34333operand. The effect of a cast to a union is similar to a compound 34334literal except that it yields an rvalue like standard casts do. *Note 34335Compound Literals::. 34336 34337 Expressions that may be cast to the union type are those whose type 34338matches at least one of the members of the union. Thus, given the 34339following union and variables: 34340 34341 union foo { int i; double d; }; 34342 int x; 34343 double y; 34344 union foo z; 34345 34346both 'x' and 'y' can be cast to type 'union foo' and the following 34347assignments 34348 z = (union foo) x; 34349 z = (union foo) y; 34350 are shorthand equivalents of these 34351 z = (union foo) { .i = x }; 34352 z = (union foo) { .d = y }; 34353 34354 However, '(union foo) FLT_MAX;' is not a valid cast because the union 34355has no member of type 'float'. 34356 34357 Using the cast as the right-hand side of an assignment to a variable of 34358union type is equivalent to storing in a member of the union with the 34359same type 34360 34361 union foo u; 34362 /* ... */ 34363 u = (union foo) x == u.i = x 34364 u = (union foo) y == u.d = y 34365 34366 You can also use the union cast as a function argument: 34367 34368 void hack (union foo); 34369 /* ... */ 34370 hack ((union foo) x); 34371 34372 34373File: gcc.info, Node: Mixed Labels and Declarations, Next: Function Attributes, Prev: Cast to Union, Up: C Extensions 34374 343756.32 Mixed Declarations, Labels and Code 34376======================================== 34377 34378ISO C99 and ISO C++ allow declarations and code to be freely mixed 34379within compound statements. ISO C2X allows labels to be placed before 34380declarations and at the end of a compound statement. As an extension, 34381GNU C also allows all this in C90 mode. For example, you could do: 34382 34383 int i; 34384 /* ... */ 34385 i++; 34386 int j = i + 2; 34387 34388 Each identifier is visible from where it is declared until the end of 34389the enclosing block. 34390 34391 34392File: gcc.info, Node: Function Attributes, Next: Variable Attributes, Prev: Mixed Labels and Declarations, Up: C Extensions 34393 343946.33 Declaring Attributes of Functions 34395====================================== 34396 34397In GNU C and C++, you can use function attributes to specify certain 34398function properties that may help the compiler optimize calls or check 34399code more carefully for correctness. For example, you can use 34400attributes to specify that a function never returns ('noreturn'), 34401returns a value depending only on the values of its arguments ('const'), 34402or has 'printf'-style arguments ('format'). 34403 34404 You can also use attributes to control memory placement, code 34405generation options or call/return conventions within the function being 34406annotated. Many of these attributes are target-specific. For example, 34407many targets support attributes for defining interrupt handler 34408functions, which typically must follow special register usage and return 34409conventions. Such attributes are described in the subsection for each 34410target. However, a considerable number of attributes are supported by 34411most, if not all targets. Those are described in the *note Common 34412Function Attributes:: section. 34413 34414 Function attributes are introduced by the '__attribute__' keyword in 34415the declaration of a function, followed by an attribute specification 34416enclosed in double parentheses. You can specify multiple attributes in 34417a declaration by separating them by commas within the double parentheses 34418or by immediately following one attribute specification with another. 34419*Note Attribute Syntax::, for the exact rules on attribute syntax and 34420placement. Compatible attribute specifications on distinct declarations 34421of the same function are merged. An attribute specification that is not 34422compatible with attributes already applied to a declaration of the same 34423function is ignored with a warning. 34424 34425 Some function attributes take one or more arguments that refer to the 34426function's parameters by their positions within the function parameter 34427list. Such attribute arguments are referred to as "positional 34428arguments". Unless specified otherwise, positional arguments that 34429specify properties of parameters with pointer types can also specify the 34430same properties of the implicit C++ 'this' argument in non-static member 34431functions, and of parameters of reference to a pointer type. For 34432ordinary functions, position one refers to the first parameter on the 34433list. In C++ non-static member functions, position one refers to the 34434implicit 'this' pointer. The same restrictions and effects apply to 34435function attributes used with ordinary functions or C++ member 34436functions. 34437 34438 GCC also supports attributes on variable declarations (*note Variable 34439Attributes::), labels (*note Label Attributes::), enumerators (*note 34440Enumerator Attributes::), statements (*note Statement Attributes::), 34441types (*note Type Attributes::), and on field declarations (for 34442'tainted_args'). 34443 34444 There is some overlap between the purposes of attributes and pragmas 34445(*note Pragmas Accepted by GCC: Pragmas.). It has been found convenient 34446to use '__attribute__' to achieve a natural attachment of attributes to 34447their corresponding declarations, whereas '#pragma' is of use for 34448compatibility with other compilers or constructs that do not naturally 34449form part of the grammar. 34450 34451 In addition to the attributes documented here, GCC plugins may provide 34452their own attributes. 34453 34454* Menu: 34455 34456* Common Function Attributes:: 34457* AArch64 Function Attributes:: 34458* AMD GCN Function Attributes:: 34459* ARC Function Attributes:: 34460* ARM Function Attributes:: 34461* AVR Function Attributes:: 34462* Blackfin Function Attributes:: 34463* BPF Function Attributes:: 34464* CR16 Function Attributes:: 34465* C-SKY Function Attributes:: 34466* Epiphany Function Attributes:: 34467* H8/300 Function Attributes:: 34468* IA-64 Function Attributes:: 34469* M32C Function Attributes:: 34470* M32R/D Function Attributes:: 34471* m68k Function Attributes:: 34472* MCORE Function Attributes:: 34473* MeP Function Attributes:: 34474* MicroBlaze Function Attributes:: 34475* Microsoft Windows Function Attributes:: 34476* MIPS Function Attributes:: 34477* MSP430 Function Attributes:: 34478* NDS32 Function Attributes:: 34479* Nios II Function Attributes:: 34480* Nvidia PTX Function Attributes:: 34481* PowerPC Function Attributes:: 34482* RISC-V Function Attributes:: 34483* RL78 Function Attributes:: 34484* RX Function Attributes:: 34485* S/390 Function Attributes:: 34486* SH Function Attributes:: 34487* Symbian OS Function Attributes:: 34488* V850 Function Attributes:: 34489* Visium Function Attributes:: 34490* x86 Function Attributes:: 34491* Xstormy16 Function Attributes:: 34492 34493 34494File: gcc.info, Node: Common Function Attributes, Next: AArch64 Function Attributes, Up: Function Attributes 34495 344966.33.1 Common Function Attributes 34497--------------------------------- 34498 34499The following attributes are supported on most targets. 34500 34501'access (ACCESS-MODE, REF-INDEX)' 34502'access (ACCESS-MODE, REF-INDEX, SIZE-INDEX)' 34503 34504 The 'access' attribute enables the detection of invalid or unsafe 34505 accesses by functions to which they apply or their callers, as well 34506 as write-only accesses to objects that are never read from. Such 34507 accesses may be diagnosed by warnings such as 34508 '-Wstringop-overflow', '-Wuninitialized', '-Wunused', and others. 34509 34510 The 'access' attribute specifies that a function to whose 34511 by-reference arguments the attribute applies accesses the 34512 referenced object according to ACCESS-MODE. The ACCESS-MODE 34513 argument is required and must be one of four names: 'read_only', 34514 'read_write', 'write_only', or 'none'. The remaining two are 34515 positional arguments. 34516 34517 The required REF-INDEX positional argument denotes a function 34518 argument of pointer (or in C++, reference) type that is subject to 34519 the access. The same pointer argument can be referenced by at most 34520 one distinct 'access' attribute. 34521 34522 The optional SIZE-INDEX positional argument denotes a function 34523 argument of integer type that specifies the maximum size of the 34524 access. The size is the number of elements of the type referenced 34525 by REF-INDEX, or the number of bytes when the pointer type is 34526 'void*'. When no SIZE-INDEX argument is specified, the pointer 34527 argument must be either null or point to a space that is suitably 34528 aligned and large for at least one object of the referenced type 34529 (this implies that a past-the-end pointer is not a valid argument). 34530 The actual size of the access may be less but it must not be more. 34531 34532 The 'read_only' access mode specifies that the pointer to which it 34533 applies is used to read the referenced object but not write to it. 34534 Unless the argument specifying the size of the access denoted by 34535 SIZE-INDEX is zero, the referenced object must be initialized. The 34536 mode implies a stronger guarantee than the 'const' qualifier which, 34537 when cast away from a pointer, does not prevent the pointed-to 34538 object from being modified. Examples of the use of the 'read_only' 34539 access mode is the argument to the 'puts' function, or the second 34540 and third arguments to the 'memcpy' function. 34541 34542 __attribute__ ((access (read_only, 1))) int puts (const char*); 34543 __attribute__ ((access (read_only, 2, 3))) void* memcpy (void*, const void*, size_t); 34544 34545 The 'read_write' access mode applies to arguments of pointer types 34546 without the 'const' qualifier. It specifies that the pointer to 34547 which it applies is used to both read and write the referenced 34548 object. Unless the argument specifying the size of the access 34549 denoted by SIZE-INDEX is zero, the object referenced by the pointer 34550 must be initialized. An example of the use of the 'read_write' 34551 access mode is the first argument to the 'strcat' function. 34552 34553 __attribute__ ((access (read_write, 1), access (read_only, 2))) char* strcat (char*, const char*); 34554 34555 The 'write_only' access mode applies to arguments of pointer types 34556 without the 'const' qualifier. It specifies that the pointer to 34557 which it applies is used to write to the referenced object but not 34558 read from it. The object referenced by the pointer need not be 34559 initialized. An example of the use of the 'write_only' access mode 34560 is the first argument to the 'strcpy' function, or the first two 34561 arguments to the 'fgets' function. 34562 34563 __attribute__ ((access (write_only, 1), access (read_only, 2))) char* strcpy (char*, const char*); 34564 __attribute__ ((access (write_only, 1, 2), access (read_write, 3))) int fgets (char*, int, FILE*); 34565 34566 The access mode 'none' specifies that the pointer to which it 34567 applies is not used to access the referenced object at all. Unless 34568 the pointer is null the pointed-to object must exist and have at 34569 least the size as denoted by the SIZE-INDEX argument. When the 34570 optional SIZE-INDEX argument is omitted for an argument of 'void*' 34571 type the actual pointer agument is ignored. The referenced object 34572 need not be initialized. The mode is intended to be used as a 34573 means to help validate the expected object size, for example in 34574 functions that call '__builtin_object_size'. *Note Object Size 34575 Checking::. 34576 34577 Note that the 'access' attribute merely specifies how an object 34578 referenced by the pointer argument can be accessed; it does not 34579 imply that an access *will* happen. Also, the 'access' attribute 34580 does not imply the attribute 'nonnull'; it may be appropriate to 34581 add both attributes at the declaration of a function that 34582 unconditionally manipulates a buffer via a pointer argument. See 34583 the 'nonnull' attribute for more information and caveats. 34584 34585'alias ("TARGET")' 34586 The 'alias' attribute causes the declaration to be emitted as an 34587 alias for another symbol, which must have been previously declared 34588 with the same type, and for variables, also the same size and 34589 alignment. Declaring an alias with a different type than the 34590 target is undefined and may be diagnosed. As an example, the 34591 following declarations: 34592 34593 void __f () { /* Do something. */; } 34594 void f () __attribute__ ((weak, alias ("__f"))); 34595 34596 define 'f' to be a weak alias for '__f'. In C++, the mangled name 34597 for the target must be used. It is an error if '__f' is not 34598 defined in the same translation unit. 34599 34600 This attribute requires assembler and object file support, and may 34601 not be available on all targets. 34602 34603'aligned' 34604'aligned (ALIGNMENT)' 34605 The 'aligned' attribute specifies a minimum alignment for the first 34606 instruction of the function, measured in bytes. When specified, 34607 ALIGNMENT must be an integer constant power of 2. Specifying no 34608 ALIGNMENT argument implies the ideal alignment for the target. The 34609 '__alignof__' operator can be used to determine what that is (*note 34610 Alignment::). The attribute has no effect when a definition for 34611 the function is not provided in the same translation unit. 34612 34613 The attribute cannot be used to decrease the alignment of a 34614 function previously declared with a more restrictive alignment; 34615 only to increase it. Attempts to do otherwise are diagnosed. Some 34616 targets specify a minimum default alignment for functions that is 34617 greater than 1. On such targets, specifying a less restrictive 34618 alignment is silently ignored. Using the attribute overrides the 34619 effect of the '-falign-functions' (*note Optimize Options::) option 34620 for this function. 34621 34622 Note that the effectiveness of 'aligned' attributes may be limited 34623 by inherent limitations in the system linker and/or object file 34624 format. On some systems, the linker is only able to arrange for 34625 functions to be aligned up to a certain maximum alignment. (For 34626 some linkers, the maximum supported alignment may be very very 34627 small.) See your linker documentation for further information. 34628 34629 The 'aligned' attribute can also be used for variables and fields 34630 (*note Variable Attributes::.) 34631 34632'alloc_align (POSITION)' 34633 The 'alloc_align' attribute may be applied to a function that 34634 returns a pointer and takes at least one argument of an integer or 34635 enumerated type. It indicates that the returned pointer is aligned 34636 on a boundary given by the function argument at POSITION. 34637 Meaningful alignments are powers of 2 greater than one. GCC uses 34638 this information to improve pointer alignment analysis. 34639 34640 The function parameter denoting the allocated alignment is 34641 specified by one constant integer argument whose number is the 34642 argument of the attribute. Argument numbering starts at one. 34643 34644 For instance, 34645 34646 void* my_memalign (size_t, size_t) __attribute__ ((alloc_align (1))); 34647 34648 declares that 'my_memalign' returns memory with minimum alignment 34649 given by parameter 1. 34650 34651'alloc_size (POSITION)' 34652'alloc_size (POSITION-1, POSITION-2)' 34653 The 'alloc_size' attribute may be applied to a function that 34654 returns a pointer and takes at least one argument of an integer or 34655 enumerated type. It indicates that the returned pointer points to 34656 memory whose size is given by the function argument at POSITION-1, 34657 or by the product of the arguments at POSITION-1 and POSITION-2. 34658 Meaningful sizes are positive values less than 'PTRDIFF_MAX'. GCC 34659 uses this information to improve the results of 34660 '__builtin_object_size'. 34661 34662 The function parameter(s) denoting the allocated size are specified 34663 by one or two integer arguments supplied to the attribute. The 34664 allocated size is either the value of the single function argument 34665 specified or the product of the two function arguments specified. 34666 Argument numbering starts at one for ordinary functions, and at two 34667 for C++ non-static member functions. 34668 34669 For instance, 34670 34671 void* my_calloc (size_t, size_t) __attribute__ ((alloc_size (1, 2))); 34672 void* my_realloc (void*, size_t) __attribute__ ((alloc_size (2))); 34673 34674 declares that 'my_calloc' returns memory of the size given by the 34675 product of parameter 1 and 2 and that 'my_realloc' returns memory 34676 of the size given by parameter 2. 34677 34678'always_inline' 34679 Generally, functions are not inlined unless optimization is 34680 specified. For functions declared inline, this attribute inlines 34681 the function independent of any restrictions that otherwise apply 34682 to inlining. Failure to inline such a function is diagnosed as an 34683 error. Note that if such a function is called indirectly the 34684 compiler may or may not inline it depending on optimization level 34685 and a failure to inline an indirect call may or may not be 34686 diagnosed. 34687 34688'artificial' 34689 This attribute is useful for small inline wrappers that if possible 34690 should appear during debugging as a unit. Depending on the debug 34691 info format it either means marking the function as artificial or 34692 using the caller location for all instructions within the inlined 34693 body. 34694 34695'assume_aligned (ALIGNMENT)' 34696'assume_aligned (ALIGNMENT, OFFSET)' 34697 The 'assume_aligned' attribute may be applied to a function that 34698 returns a pointer. It indicates that the returned pointer is 34699 aligned on a boundary given by ALIGNMENT. If the attribute has two 34700 arguments, the second argument is misalignment OFFSET. Meaningful 34701 values of ALIGNMENT are powers of 2 greater than one. Meaningful 34702 values of OFFSET are greater than zero and less than ALIGNMENT. 34703 34704 For instance 34705 34706 void* my_alloc1 (size_t) __attribute__((assume_aligned (16))); 34707 void* my_alloc2 (size_t) __attribute__((assume_aligned (32, 8))); 34708 34709 declares that 'my_alloc1' returns 16-byte aligned pointers and that 34710 'my_alloc2' returns a pointer whose value modulo 32 is equal to 8. 34711 34712'cold' 34713 The 'cold' attribute on functions is used to inform the compiler 34714 that the function is unlikely to be executed. The function is 34715 optimized for size rather than speed and on many targets it is 34716 placed into a special subsection of the text section so all cold 34717 functions appear close together, improving code locality of 34718 non-cold parts of program. The paths leading to calls of cold 34719 functions within code are marked as unlikely by the branch 34720 prediction mechanism. It is thus useful to mark functions used to 34721 handle unlikely conditions, such as 'perror', as cold to improve 34722 optimization of hot functions that do call marked functions in rare 34723 occasions. 34724 34725 When profile feedback is available, via '-fprofile-use', cold 34726 functions are automatically detected and this attribute is ignored. 34727 34728'const' 34729 Calls to functions whose return value is not affected by changes to 34730 the observable state of the program and that have no observable 34731 effects on such state other than to return a value may lend 34732 themselves to optimizations such as common subexpression 34733 elimination. Declaring such functions with the 'const' attribute 34734 allows GCC to avoid emitting some calls in repeated invocations of 34735 the function with the same argument values. 34736 34737 For example, 34738 34739 int square (int) __attribute__ ((const)); 34740 34741 tells GCC that subsequent calls to function 'square' with the same 34742 argument value can be replaced by the result of the first call 34743 regardless of the statements in between. 34744 34745 The 'const' attribute prohibits a function from reading objects 34746 that affect its return value between successive invocations. 34747 However, functions declared with the attribute can safely read 34748 objects that do not change their return value, such as non-volatile 34749 constants. 34750 34751 The 'const' attribute imposes greater restrictions on a function's 34752 definition than the similar 'pure' attribute. Declaring the same 34753 function with both the 'const' and the 'pure' attribute is 34754 diagnosed. Because a const function cannot have any observable 34755 side effects it does not make sense for it to return 'void'. 34756 Declaring such a function is diagnosed. 34757 34758 Note that a function that has pointer arguments and examines the 34759 data pointed to must _not_ be declared 'const' if the pointed-to 34760 data might change between successive invocations of the function. 34761 In general, since a function cannot distinguish data that might 34762 change from data that cannot, const functions should never take 34763 pointer or, in C++, reference arguments. Likewise, a function that 34764 calls a non-const function usually must not be const itself. 34765 34766'constructor' 34767'destructor' 34768'constructor (PRIORITY)' 34769'destructor (PRIORITY)' 34770 The 'constructor' attribute causes the function to be called 34771 automatically before execution enters 'main ()'. Similarly, the 34772 'destructor' attribute causes the function to be called 34773 automatically after 'main ()' completes or 'exit ()' is called. 34774 Functions with these attributes are useful for initializing data 34775 that is used implicitly during the execution of the program. 34776 34777 On some targets the attributes also accept an integer argument to 34778 specify a priority to control the order in which constructor and 34779 destructor functions are run. A constructor with a smaller 34780 priority number runs before a constructor with a larger priority 34781 number; the opposite relationship holds for destructors. Note that 34782 priorities 0-100 are reserved. So, if you have a constructor that 34783 allocates a resource and a destructor that deallocates the same 34784 resource, both functions typically have the same priority. The 34785 priorities for constructor and destructor functions are the same as 34786 those specified for namespace-scope C++ objects (*note C++ 34787 Attributes::). However, at present, the order in which 34788 constructors for C++ objects with static storage duration and 34789 functions decorated with attribute 'constructor' are invoked is 34790 unspecified. In mixed declarations, attribute 'init_priority' can 34791 be used to impose a specific ordering. 34792 34793 Using the argument forms of the 'constructor' and 'destructor' 34794 attributes on targets where the feature is not supported is 34795 rejected with an error. 34796 34797'copy' 34798'copy (FUNCTION)' 34799 The 'copy' attribute applies the set of attributes with which 34800 FUNCTION has been declared to the declaration of the function to 34801 which the attribute is applied. The attribute is designed for 34802 libraries that define aliases or function resolvers that are 34803 expected to specify the same set of attributes as their targets. 34804 The 'copy' attribute can be used with functions, variables, or 34805 types. However, the kind of symbol to which the attribute is 34806 applied (either function or variable) must match the kind of symbol 34807 to which the argument refers. The 'copy' attribute copies only 34808 syntactic and semantic attributes but not attributes that affect a 34809 symbol's linkage or visibility such as 'alias', 'visibility', or 34810 'weak'. The 'deprecated' and 'target_clones' attribute are also 34811 not copied. *Note Common Type Attributes::. *Note Common Variable 34812 Attributes::. 34813 34814 For example, the STRONGALIAS macro below makes use of the 'alias' 34815 and 'copy' attributes to define an alias named ALLOC for function 34816 ALLOCATE declared with attributes ALLOC_SIZE, MALLOC, and NOTHROW. 34817 Thanks to the '__typeof__' operator the alias has the same type as 34818 the target function. As a result of the 'copy' attribute the alias 34819 also shares the same attributes as the target. 34820 34821 #define StrongAlias(TargetFunc, AliasDecl) \ 34822 extern __typeof__ (TargetFunc) AliasDecl \ 34823 __attribute__ ((alias (#TargetFunc), copy (TargetFunc))); 34824 34825 extern __attribute__ ((alloc_size (1), malloc, nothrow)) 34826 void* allocate (size_t); 34827 StrongAlias (allocate, alloc); 34828 34829'deprecated' 34830'deprecated (MSG)' 34831 The 'deprecated' attribute results in a warning if the function is 34832 used anywhere in the source file. This is useful when identifying 34833 functions that are expected to be removed in a future version of a 34834 program. The warning also includes the location of the declaration 34835 of the deprecated function, to enable users to easily find further 34836 information about why the function is deprecated, or what they 34837 should do instead. Note that the warnings only occurs for uses: 34838 34839 int old_fn () __attribute__ ((deprecated)); 34840 int old_fn (); 34841 int (*fn_ptr)() = old_fn; 34842 34843 results in a warning on line 3 but not line 2. The optional MSG 34844 argument, which must be a string, is printed in the warning if 34845 present. 34846 34847 The 'deprecated' attribute can also be used for variables and types 34848 (*note Variable Attributes::, *note Type Attributes::.) 34849 34850 The message attached to the attribute is affected by the setting of 34851 the '-fmessage-length' option. 34852 34853'unavailable' 34854'unavailable (MSG)' 34855 The 'unavailable' attribute results in an error if the function is 34856 used anywhere in the source file. This is useful when identifying 34857 functions that have been removed from a particular variation of an 34858 interface. Other than emitting an error rather than a warning, the 34859 'unavailable' attribute behaves in the same manner as 'deprecated'. 34860 34861 The 'unavailable' attribute can also be used for variables and 34862 types (*note Variable Attributes::, *note Type Attributes::.) 34863 34864'error ("MESSAGE")' 34865'warning ("MESSAGE")' 34866 If the 'error' or 'warning' attribute is used on a function 34867 declaration and a call to such a function is not eliminated through 34868 dead code elimination or other optimizations, an error or warning 34869 (respectively) that includes MESSAGE is diagnosed. This is useful 34870 for compile-time checking, especially together with 34871 '__builtin_constant_p' and inline functions where checking the 34872 inline function arguments is not possible through 'extern char 34873 [(condition) ? 1 : -1];' tricks. 34874 34875 While it is possible to leave the function undefined and thus 34876 invoke a link failure (to define the function with a message in 34877 '.gnu.warning*' section), when using these attributes the problem 34878 is diagnosed earlier and with exact location of the call even in 34879 presence of inline functions or when not emitting debugging 34880 information. 34881 34882'externally_visible' 34883 This attribute, attached to a global variable or function, 34884 nullifies the effect of the '-fwhole-program' command-line option, 34885 so the object remains visible outside the current compilation unit. 34886 34887 If '-fwhole-program' is used together with '-flto' and 'gold' is 34888 used as the linker plugin, 'externally_visible' attributes are 34889 automatically added to functions (not variable yet due to a current 34890 'gold' issue) that are accessed outside of LTO objects according to 34891 resolution file produced by 'gold'. For other linkers that cannot 34892 generate resolution file, explicit 'externally_visible' attributes 34893 are still necessary. 34894 34895'flatten' 34896 Generally, inlining into a function is limited. For a function 34897 marked with this attribute, every call inside this function is 34898 inlined, if possible. Functions declared with attribute 'noinline' 34899 and similar are not inlined. Whether the function itself is 34900 considered for inlining depends on its size and the current 34901 inlining parameters. 34902 34903'format (ARCHETYPE, STRING-INDEX, FIRST-TO-CHECK)' 34904 The 'format' attribute specifies that a function takes 'printf', 34905 'scanf', 'strftime' or 'strfmon' style arguments that should be 34906 type-checked against a format string. For example, the 34907 declaration: 34908 34909 extern int 34910 my_printf (void *my_object, const char *my_format, ...) 34911 __attribute__ ((format (printf, 2, 3))); 34912 34913 causes the compiler to check the arguments in calls to 'my_printf' 34914 for consistency with the 'printf' style format string argument 34915 'my_format'. 34916 34917 The parameter ARCHETYPE determines how the format string is 34918 interpreted, and should be 'printf', 'scanf', 'strftime', 34919 'gnu_printf', 'gnu_scanf', 'gnu_strftime' or 'strfmon'. (You can 34920 also use '__printf__', '__scanf__', '__strftime__' or 34921 '__strfmon__'.) On MinGW targets, 'ms_printf', 'ms_scanf', and 34922 'ms_strftime' are also present. ARCHETYPE values such as 'printf' 34923 refer to the formats accepted by the system's C runtime library, 34924 while values prefixed with 'gnu_' always refer to the formats 34925 accepted by the GNU C Library. On Microsoft Windows targets, 34926 values prefixed with 'ms_' refer to the formats accepted by the 34927 'msvcrt.dll' library. The parameter STRING-INDEX specifies which 34928 argument is the format string argument (starting from 1), while 34929 FIRST-TO-CHECK is the number of the first argument to check against 34930 the format string. For functions where the arguments are not 34931 available to be checked (such as 'vprintf'), specify the third 34932 parameter as zero. In this case the compiler only checks the 34933 format string for consistency. For 'strftime' formats, the third 34934 parameter is required to be zero. Since non-static C++ methods 34935 have an implicit 'this' argument, the arguments of such methods 34936 should be counted from two, not one, when giving values for 34937 STRING-INDEX and FIRST-TO-CHECK. 34938 34939 In the example above, the format string ('my_format') is the second 34940 argument of the function 'my_print', and the arguments to check 34941 start with the third argument, so the correct parameters for the 34942 format attribute are 2 and 3. 34943 34944 The 'format' attribute allows you to identify your own functions 34945 that take format strings as arguments, so that GCC can check the 34946 calls to these functions for errors. The compiler always (unless 34947 '-ffreestanding' or '-fno-builtin' is used) checks formats for the 34948 standard library functions 'printf', 'fprintf', 'sprintf', 'scanf', 34949 'fscanf', 'sscanf', 'strftime', 'vprintf', 'vfprintf' and 34950 'vsprintf' whenever such warnings are requested (using '-Wformat'), 34951 so there is no need to modify the header file 'stdio.h'. In C99 34952 mode, the functions 'snprintf', 'vsnprintf', 'vscanf', 'vfscanf' 34953 and 'vsscanf' are also checked. Except in strictly conforming C 34954 standard modes, the X/Open function 'strfmon' is also checked as 34955 are 'printf_unlocked' and 'fprintf_unlocked'. *Note Options 34956 Controlling C Dialect: C Dialect Options. 34957 34958 For Objective-C dialects, 'NSString' (or '__NSString__') is 34959 recognized in the same context. Declarations including these 34960 format attributes are parsed for correct syntax, however the result 34961 of checking of such format strings is not yet defined, and is not 34962 carried out by this version of the compiler. 34963 34964 The target may also provide additional types of format checks. 34965 *Note Format Checks Specific to Particular Target Machines: Target 34966 Format Checks. 34967 34968'format_arg (STRING-INDEX)' 34969 The 'format_arg' attribute specifies that a function takes one or 34970 more format strings for a 'printf', 'scanf', 'strftime' or 34971 'strfmon' style function and modifies it (for example, to translate 34972 it into another language), so the result can be passed to a 34973 'printf', 'scanf', 'strftime' or 'strfmon' style function (with the 34974 remaining arguments to the format function the same as they would 34975 have been for the unmodified string). Multiple 'format_arg' 34976 attributes may be applied to the same function, each designating a 34977 distinct parameter as a format string. For example, the 34978 declaration: 34979 34980 extern char * 34981 my_dgettext (char *my_domain, const char *my_format) 34982 __attribute__ ((format_arg (2))); 34983 34984 causes the compiler to check the arguments in calls to a 'printf', 34985 'scanf', 'strftime' or 'strfmon' type function, whose format string 34986 argument is a call to the 'my_dgettext' function, for consistency 34987 with the format string argument 'my_format'. If the 'format_arg' 34988 attribute had not been specified, all the compiler could tell in 34989 such calls to format functions would be that the format string 34990 argument is not constant; this would generate a warning when 34991 '-Wformat-nonliteral' is used, but the calls could not be checked 34992 without the attribute. 34993 34994 In calls to a function declared with more than one 'format_arg' 34995 attribute, each with a distinct argument value, the corresponding 34996 actual function arguments are checked against all format strings 34997 designated by the attributes. This capability is designed to 34998 support the GNU 'ngettext' family of functions. 34999 35000 The parameter STRING-INDEX specifies which argument is the format 35001 string argument (starting from one). Since non-static C++ methods 35002 have an implicit 'this' argument, the arguments of such methods 35003 should be counted from two. 35004 35005 The 'format_arg' attribute allows you to identify your own 35006 functions that modify format strings, so that GCC can check the 35007 calls to 'printf', 'scanf', 'strftime' or 'strfmon' type function 35008 whose operands are a call to one of your own function. The 35009 compiler always treats 'gettext', 'dgettext', and 'dcgettext' in 35010 this manner except when strict ISO C support is requested by 35011 '-ansi' or an appropriate '-std' option, or '-ffreestanding' or 35012 '-fno-builtin' is used. *Note Options Controlling C Dialect: C 35013 Dialect Options. 35014 35015 For Objective-C dialects, the 'format-arg' attribute may refer to 35016 an 'NSString' reference for compatibility with the 'format' 35017 attribute above. 35018 35019 The target may also allow additional types in 'format-arg' 35020 attributes. *Note Format Checks Specific to Particular Target 35021 Machines: Target Format Checks. 35022 35023'gnu_inline' 35024 This attribute should be used with a function that is also declared 35025 with the 'inline' keyword. It directs GCC to treat the function as 35026 if it were defined in gnu90 mode even when compiling in C99 or 35027 gnu99 mode. 35028 35029 If the function is declared 'extern', then this definition of the 35030 function is used only for inlining. In no case is the function 35031 compiled as a standalone function, not even if you take its address 35032 explicitly. Such an address becomes an external reference, as if 35033 you had only declared the function, and had not defined it. This 35034 has almost the effect of a macro. The way to use this is to put a 35035 function definition in a header file with this attribute, and put 35036 another copy of the function, without 'extern', in a library file. 35037 The definition in the header file causes most calls to the function 35038 to be inlined. If any uses of the function remain, they refer to 35039 the single copy in the library. Note that the two definitions of 35040 the functions need not be precisely the same, although if they do 35041 not have the same effect your program may behave oddly. 35042 35043 In C, if the function is neither 'extern' nor 'static', then the 35044 function is compiled as a standalone function, as well as being 35045 inlined where possible. 35046 35047 This is how GCC traditionally handled functions declared 'inline'. 35048 Since ISO C99 specifies a different semantics for 'inline', this 35049 function attribute is provided as a transition measure and as a 35050 useful feature in its own right. This attribute is available in 35051 GCC 4.1.3 and later. It is available if either of the preprocessor 35052 macros '__GNUC_GNU_INLINE__' or '__GNUC_STDC_INLINE__' are defined. 35053 *Note An Inline Function is As Fast As a Macro: Inline. 35054 35055 In C++, this attribute does not depend on 'extern' in any way, but 35056 it still requires the 'inline' keyword to enable its special 35057 behavior. 35058 35059'hot' 35060 The 'hot' attribute on a function is used to inform the compiler 35061 that the function is a hot spot of the compiled program. The 35062 function is optimized more aggressively and on many targets it is 35063 placed into a special subsection of the text section so all hot 35064 functions appear close together, improving locality. 35065 35066 When profile feedback is available, via '-fprofile-use', hot 35067 functions are automatically detected and this attribute is ignored. 35068 35069'ifunc ("RESOLVER")' 35070 The 'ifunc' attribute is used to mark a function as an indirect 35071 function using the STT_GNU_IFUNC symbol type extension to the ELF 35072 standard. This allows the resolution of the symbol value to be 35073 determined dynamically at load time, and an optimized version of 35074 the routine to be selected for the particular processor or other 35075 system characteristics determined then. To use this attribute, 35076 first define the implementation functions available, and a resolver 35077 function that returns a pointer to the selected implementation 35078 function. The implementation functions' declarations must match 35079 the API of the function being implemented. The resolver should be 35080 declared to be a function taking no arguments and returning a 35081 pointer to a function of the same type as the implementation. For 35082 example: 35083 35084 void *my_memcpy (void *dst, const void *src, size_t len) 35085 { 35086 ... 35087 return dst; 35088 } 35089 35090 static void * (*resolve_memcpy (void))(void *, const void *, size_t) 35091 { 35092 return my_memcpy; // we will just always select this routine 35093 } 35094 35095 The exported header file declaring the function the user calls 35096 would contain: 35097 35098 extern void *memcpy (void *, const void *, size_t); 35099 35100 allowing the user to call 'memcpy' as a regular function, unaware 35101 of the actual implementation. Finally, the indirect function needs 35102 to be defined in the same translation unit as the resolver 35103 function: 35104 35105 void *memcpy (void *, const void *, size_t) 35106 __attribute__ ((ifunc ("resolve_memcpy"))); 35107 35108 In C++, the 'ifunc' attribute takes a string that is the mangled 35109 name of the resolver function. A C++ resolver for a non-static 35110 member function of class 'C' should be declared to return a pointer 35111 to a non-member function taking pointer to 'C' as the first 35112 argument, followed by the same arguments as of the implementation 35113 function. G++ checks the signatures of the two functions and 35114 issues a '-Wattribute-alias' warning for mismatches. To suppress a 35115 warning for the necessary cast from a pointer to the implementation 35116 member function to the type of the corresponding non-member 35117 function use the '-Wno-pmf-conversions' option. For example: 35118 35119 class S 35120 { 35121 private: 35122 int debug_impl (int); 35123 int optimized_impl (int); 35124 35125 typedef int Func (S*, int); 35126 35127 static Func* resolver (); 35128 public: 35129 35130 int interface (int); 35131 }; 35132 35133 int S::debug_impl (int) { /* ... */ } 35134 int S::optimized_impl (int) { /* ... */ } 35135 35136 S::Func* S::resolver () 35137 { 35138 int (S::*pimpl) (int) 35139 = getenv ("DEBUG") ? &S::debug_impl : &S::optimized_impl; 35140 35141 // Cast triggers -Wno-pmf-conversions. 35142 return reinterpret_cast<Func*>(pimpl); 35143 } 35144 35145 int S::interface (int) __attribute__ ((ifunc ("_ZN1S8resolverEv"))); 35146 35147 Indirect functions cannot be weak. Binutils version 2.20.1 or 35148 higher and GNU C Library version 2.11.1 are required to use this 35149 feature. 35150 35151'interrupt' 35152'interrupt_handler' 35153 Many GCC back ends support attributes to indicate that a function 35154 is an interrupt handler, which tells the compiler to generate 35155 function entry and exit sequences that differ from those from 35156 regular functions. The exact syntax and behavior are 35157 target-specific; refer to the following subsections for details. 35158 35159'leaf' 35160 Calls to external functions with this attribute must return to the 35161 current compilation unit only by return or by exception handling. 35162 In particular, a leaf function is not allowed to invoke callback 35163 functions passed to it from the current compilation unit, directly 35164 call functions exported by the unit, or 'longjmp' into the unit. 35165 Leaf functions might still call functions from other compilation 35166 units and thus they are not necessarily leaf in the sense that they 35167 contain no function calls at all. 35168 35169 The attribute is intended for library functions to improve dataflow 35170 analysis. The compiler takes the hint that any data not escaping 35171 the current compilation unit cannot be used or modified by the leaf 35172 function. For example, the 'sin' function is a leaf function, but 35173 'qsort' is not. 35174 35175 Note that leaf functions might indirectly run a signal handler 35176 defined in the current compilation unit that uses static variables. 35177 Similarly, when lazy symbol resolution is in effect, leaf functions 35178 might invoke indirect functions whose resolver function or 35179 implementation function is defined in the current compilation unit 35180 and uses static variables. There is no standard-compliant way to 35181 write such a signal handler, resolver function, or implementation 35182 function, and the best that you can do is to remove the 'leaf' 35183 attribute or mark all such static variables 'volatile'. Lastly, 35184 for ELF-based systems that support symbol interposition, care 35185 should be taken that functions defined in the current compilation 35186 unit do not unexpectedly interpose other symbols based on the 35187 defined standards mode and defined feature test macros; otherwise 35188 an inadvertent callback would be added. 35189 35190 The attribute has no effect on functions defined within the current 35191 compilation unit. This is to allow easy merging of multiple 35192 compilation units into one, for example, by using the link-time 35193 optimization. For this reason the attribute is not allowed on 35194 types to annotate indirect calls. 35195 35196'malloc' 35197'malloc (DEALLOCATOR)' 35198'malloc (DEALLOCATOR, PTR-INDEX)' 35199 Attribute 'malloc' indicates that a function is 'malloc'-like, 35200 i.e., that the pointer P returned by the function cannot alias any 35201 other pointer valid when the function returns, and moreover no 35202 pointers to valid objects occur in any storage addressed by P. In 35203 addition, the GCC predicts that a function with the attribute 35204 returns non-null in most cases. 35205 35206 Independently, the form of the attribute with one or two arguments 35207 associates 'deallocator' as a suitable deallocation function for 35208 pointers returned from the 'malloc'-like function. PTR-INDEX 35209 denotes the positional argument to which when the pointer is passed 35210 in calls to 'deallocator' has the effect of deallocating it. 35211 35212 Using the attribute with no arguments is designed to improve 35213 optimization by relying on the aliasing property it implies. 35214 Functions like 'malloc' and 'calloc' have this property because 35215 they return a pointer to uninitialized or zeroed-out, newly 35216 obtained storage. However, functions like 'realloc' do not have 35217 this property, as they may return pointers to storage containing 35218 pointers to existing objects. Additionally, since all such 35219 functions are assumed to return null only infrequently, callers can 35220 be optimized based on that assumption. 35221 35222 Associating a function with a DEALLOCATOR helps detect calls to 35223 mismatched allocation and deallocation functions and diagnose them 35224 under the control of options such as '-Wmismatched-dealloc'. It 35225 also makes it possible to diagnose attempts to deallocate objects 35226 that were not allocated dynamically, by '-Wfree-nonheap-object'. 35227 To indicate that an allocation function both satisifies the 35228 nonaliasing property and has a deallocator associated with it, both 35229 the plain form of the attribute and the one with the DEALLOCATOR 35230 argument must be used. The same function can be both an allocator 35231 and a deallocator. Since inlining one of the associated functions 35232 but not the other could result in apparent mismatches, this form of 35233 attribute 'malloc' is not accepted on inline functions. For the 35234 same reason, using the attribute prevents both the allocation and 35235 deallocation functions from being expanded inline. 35236 35237 For example, besides stating that the functions return pointers 35238 that do not alias any others, the following declarations make 35239 'fclose' a suitable deallocator for pointers returned from all 35240 functions except 'popen', and 'pclose' as the only suitable 35241 deallocator for pointers returned from 'popen'. The deallocator 35242 functions must be declared before they can be referenced in the 35243 attribute. 35244 35245 int fclose (FILE*); 35246 int pclose (FILE*); 35247 35248 __attribute__ ((malloc, malloc (fclose, 1))) 35249 FILE* fdopen (int, const char*); 35250 __attribute__ ((malloc, malloc (fclose, 1))) 35251 FILE* fopen (const char*, const char*); 35252 __attribute__ ((malloc, malloc (fclose, 1))) 35253 FILE* fmemopen(void *, size_t, const char *); 35254 __attribute__ ((malloc, malloc (pclose, 1))) 35255 FILE* popen (const char*, const char*); 35256 __attribute__ ((malloc, malloc (fclose, 1))) 35257 FILE* tmpfile (void); 35258 35259 The warnings guarded by '-fanalyzer' respect allocation and 35260 deallocation pairs marked with the 'malloc'. In particular: 35261 35262 * The analyzer will emit a '-Wanalyzer-mismatching-deallocation' 35263 diagnostic if there is an execution path in which the result 35264 of an allocation call is passed to a different deallocator. 35265 35266 * The analyzer will emit a '-Wanalyzer-double-free' diagnostic 35267 if there is an execution path in which a value is passed more 35268 than once to a deallocation call. 35269 35270 * The analyzer will consider the possibility that an allocation 35271 function could fail and return NULL. It will emit 35272 '-Wanalyzer-possible-null-dereference' and 35273 '-Wanalyzer-possible-null-argument' diagnostics if there are 35274 execution paths in which an unchecked result of an allocation 35275 call is dereferenced or passed to a function requiring a 35276 non-null argument. If the allocator always returns non-null, 35277 use '__attribute__ ((returns_nonnull))' to suppress these 35278 warnings. For example: 35279 char *xstrdup (const char *) 35280 __attribute__((malloc (free), returns_nonnull)); 35281 35282 * The analyzer will emit a '-Wanalyzer-use-after-free' 35283 diagnostic if there is an execution path in which the memory 35284 passed by pointer to a deallocation call is used after the 35285 deallocation. 35286 35287 * The analyzer will emit a '-Wanalyzer-malloc-leak' diagnostic 35288 if there is an execution path in which the result of an 35289 allocation call is leaked (without being passed to the 35290 deallocation function). 35291 35292 * The analyzer will emit a '-Wanalyzer-free-of-non-heap' 35293 diagnostic if a deallocation function is used on a global or 35294 on-stack variable. 35295 35296 The analyzer assumes that deallocators can gracefully handle the 35297 'NULL' pointer. If this is not the case, the deallocator can be 35298 marked with '__attribute__((nonnull))' so that '-fanalyzer' can 35299 emit a '-Wanalyzer-possible-null-argument' diagnostic for code 35300 paths in which the deallocator is called with NULL. 35301 35302'no_icf' 35303 This function attribute prevents a functions from being merged with 35304 another semantically equivalent function. 35305 35306'no_instrument_function' 35307 If any of '-finstrument-functions', '-p', or '-pg' are given, 35308 profiling function calls are generated at entry and exit of most 35309 user-compiled functions. Functions with this attribute are not so 35310 instrumented. 35311 35312'no_profile_instrument_function' 35313 The 'no_profile_instrument_function' attribute on functions is used 35314 to inform the compiler that it should not process any profile 35315 feedback based optimization code instrumentation. 35316 35317'no_reorder' 35318 Do not reorder functions or variables marked 'no_reorder' against 35319 each other or top level assembler statements the executable. The 35320 actual order in the program will depend on the linker command line. 35321 Static variables marked like this are also not removed. This has a 35322 similar effect as the '-fno-toplevel-reorder' option, but only 35323 applies to the marked symbols. 35324 35325'no_sanitize ("SANITIZE_OPTION")' 35326 The 'no_sanitize' attribute on functions is used to inform the 35327 compiler that it should not do sanitization of any option mentioned 35328 in SANITIZE_OPTION. A list of values acceptable by the 35329 '-fsanitize' option can be provided. 35330 35331 void __attribute__ ((no_sanitize ("alignment", "object-size"))) 35332 f () { /* Do something. */; } 35333 void __attribute__ ((no_sanitize ("alignment,object-size"))) 35334 g () { /* Do something. */; } 35335 35336'no_sanitize_address' 35337'no_address_safety_analysis' 35338 The 'no_sanitize_address' attribute on functions is used to inform 35339 the compiler that it should not instrument memory accesses in the 35340 function when compiling with the '-fsanitize=address' option. The 35341 'no_address_safety_analysis' is a deprecated alias of the 35342 'no_sanitize_address' attribute, new code should use 35343 'no_sanitize_address'. 35344 35345'no_sanitize_thread' 35346 The 'no_sanitize_thread' attribute on functions is used to inform 35347 the compiler that it should not instrument memory accesses in the 35348 function when compiling with the '-fsanitize=thread' option. 35349 35350'no_sanitize_undefined' 35351 The 'no_sanitize_undefined' attribute on functions is used to 35352 inform the compiler that it should not check for undefined behavior 35353 in the function when compiling with the '-fsanitize=undefined' 35354 option. 35355 35356'no_sanitize_coverage' 35357 The 'no_sanitize_coverage' attribute on functions is used to inform 35358 the compiler that it should not do coverage-guided fuzzing code 35359 instrumentation ('-fsanitize-coverage'). 35360 35361'no_split_stack' 35362 If '-fsplit-stack' is given, functions have a small prologue which 35363 decides whether to split the stack. Functions with the 35364 'no_split_stack' attribute do not have that prologue, and thus may 35365 run with only a small amount of stack space available. 35366 35367'no_stack_limit' 35368 This attribute locally overrides the '-fstack-limit-register' and 35369 '-fstack-limit-symbol' command-line options; it has the effect of 35370 disabling stack limit checking in the function it applies to. 35371 35372'noclone' 35373 This function attribute prevents a function from being considered 35374 for cloning--a mechanism that produces specialized copies of 35375 functions and which is (currently) performed by interprocedural 35376 constant propagation. 35377 35378'noinline' 35379 This function attribute prevents a function from being considered 35380 for inlining. If the function does not have side effects, there 35381 are optimizations other than inlining that cause function calls to 35382 be optimized away, although the function call is live. To keep 35383 such calls from being optimized away, put 35384 asm (""); 35385 35386 (*note Extended Asm::) in the called function, to serve as a 35387 special side effect. 35388 35389'noipa' 35390 Disable interprocedural optimizations between the function with 35391 this attribute and its callers, as if the body of the function is 35392 not available when optimizing callers and the callers are 35393 unavailable when optimizing the body. This attribute implies 35394 'noinline', 'noclone' and 'no_icf' attributes. However, this 35395 attribute is not equivalent to a combination of other attributes, 35396 because its purpose is to suppress existing and future 35397 optimizations employing interprocedural analysis, including those 35398 that do not have an attribute suitable for disabling them 35399 individually. This attribute is supported mainly for the purpose 35400 of testing the compiler. 35401 35402'nonnull' 35403'nonnull (ARG-INDEX, ...)' 35404 The 'nonnull' attribute may be applied to a function that takes at 35405 least one argument of a pointer type. It indicates that the 35406 referenced arguments must be non-null pointers. For instance, the 35407 declaration: 35408 35409 extern void * 35410 my_memcpy (void *dest, const void *src, size_t len) 35411 __attribute__((nonnull (1, 2))); 35412 35413 informs the compiler that, in calls to 'my_memcpy', arguments DEST 35414 and SRC must be non-null. 35415 35416 The attribute has an effect both on functions calls and function 35417 definitions. 35418 35419 For function calls: 35420 * If the compiler determines that a null pointer is passed in an 35421 argument slot marked as non-null, and the '-Wnonnull' option 35422 is enabled, a warning is issued. *Note Warning Options::. 35423 * The '-fisolate-erroneous-paths-attribute' option can be 35424 specified to have GCC transform calls with null arguments to 35425 non-null functions into traps. *Note Optimize Options::. 35426 * The compiler may also perform optimizations based on the 35427 knowledge that certain function arguments cannot be null. 35428 These optimizations can be disabled by the 35429 '-fno-delete-null-pointer-checks' option. *Note Optimize 35430 Options::. 35431 35432 For function definitions: 35433 * If the compiler determines that a function parameter that is 35434 marked with nonnull is compared with null, and 35435 '-Wnonnull-compare' option is enabled, a warning is issued. 35436 *Note Warning Options::. 35437 * The compiler may also perform optimizations based on the 35438 knowledge that 'nonnul' parameters cannot be null. This can 35439 currently not be disabled other than by removing the nonnull 35440 attribute. 35441 35442 If no ARG-INDEX is given to the 'nonnull' attribute, all pointer 35443 arguments are marked as non-null. To illustrate, the following 35444 declaration is equivalent to the previous example: 35445 35446 extern void * 35447 my_memcpy (void *dest, const void *src, size_t len) 35448 __attribute__((nonnull)); 35449 35450'noplt' 35451 The 'noplt' attribute is the counterpart to option '-fno-plt'. 35452 Calls to functions marked with this attribute in 35453 position-independent code do not use the PLT. 35454 35455 /* Externally defined function foo. */ 35456 int foo () __attribute__ ((noplt)); 35457 35458 int 35459 main (/* ... */) 35460 { 35461 /* ... */ 35462 foo (); 35463 /* ... */ 35464 } 35465 35466 The 'noplt' attribute on function 'foo' tells the compiler to 35467 assume that the function 'foo' is externally defined and that the 35468 call to 'foo' must avoid the PLT in position-independent code. 35469 35470 In position-dependent code, a few targets also convert calls to 35471 functions that are marked to not use the PLT to use the GOT 35472 instead. 35473 35474'noreturn' 35475 A few standard library functions, such as 'abort' and 'exit', 35476 cannot return. GCC knows this automatically. Some programs define 35477 their own functions that never return. You can declare them 35478 'noreturn' to tell the compiler this fact. For example, 35479 35480 void fatal () __attribute__ ((noreturn)); 35481 35482 void 35483 fatal (/* ... */) 35484 { 35485 /* ... */ /* Print error message. */ /* ... */ 35486 exit (1); 35487 } 35488 35489 The 'noreturn' keyword tells the compiler to assume that 'fatal' 35490 cannot return. It can then optimize without regard to what would 35491 happen if 'fatal' ever did return. This makes slightly better 35492 code. More importantly, it helps avoid spurious warnings of 35493 uninitialized variables. 35494 35495 The 'noreturn' keyword does not affect the exceptional path when 35496 that applies: a 'noreturn'-marked function may still return to the 35497 caller by throwing an exception or calling 'longjmp'. 35498 35499 In order to preserve backtraces, GCC will never turn calls to 35500 'noreturn' functions into tail calls. 35501 35502 Do not assume that registers saved by the calling function are 35503 restored before calling the 'noreturn' function. 35504 35505 It does not make sense for a 'noreturn' function to have a return 35506 type other than 'void'. 35507 35508'nothrow' 35509 The 'nothrow' attribute is used to inform the compiler that a 35510 function cannot throw an exception. For example, most functions in 35511 the standard C library can be guaranteed not to throw an exception 35512 with the notable exceptions of 'qsort' and 'bsearch' that take 35513 function pointer arguments. 35514 35515'optimize (LEVEL, ...)' 35516'optimize (STRING, ...)' 35517 The 'optimize' attribute is used to specify that a function is to 35518 be compiled with different optimization options than specified on 35519 the command line. The optimize attribute arguments of a function 35520 behave behave as if appended to the command-line. 35521 35522 Valid arguments are constant non-negative integers and strings. 35523 Each numeric argument specifies an optimization LEVEL. Each STRING 35524 argument consists of one or more comma-separated substrings. Each 35525 substring that begins with the letter 'O' refers to an optimization 35526 option such as '-O0' or '-Os'. Other substrings are taken as 35527 suffixes to the '-f' prefix jointly forming the name of an 35528 optimization option. *Note Optimize Options::. 35529 35530 '#pragma GCC optimize' can be used to set optimization options for 35531 more than one function. *Note Function Specific Option Pragmas::, 35532 for details about the pragma. 35533 35534 Providing multiple strings as arguments separated by commas to 35535 specify multiple options is equivalent to separating the option 35536 suffixes with a comma (',') within a single string. Spaces are not 35537 permitted within the strings. 35538 35539 Not every optimization option that starts with the -F prefix 35540 specified by the attribute necessarily has an effect on the 35541 function. The 'optimize' attribute should be used for debugging 35542 purposes only. It is not suitable in production code. 35543 35544'patchable_function_entry' 35545 In case the target's text segment can be made writable at run time 35546 by any means, padding the function entry with a number of NOPs can 35547 be used to provide a universal tool for instrumentation. 35548 35549 The 'patchable_function_entry' function attribute can be used to 35550 change the number of NOPs to any desired value. The two-value 35551 syntax is the same as for the command-line switch 35552 '-fpatchable-function-entry=N,M', generating N NOPs, with the 35553 function entry point before the Mth NOP instruction. M defaults to 35554 0 if omitted e.g. function entry point is before the first NOP. 35555 35556 If patchable function entries are enabled globally using the 35557 command-line option '-fpatchable-function-entry=N,M', then you must 35558 disable instrumentation on all functions that are part of the 35559 instrumentation framework with the attribute 35560 'patchable_function_entry (0)' to prevent recursion. 35561 35562'pure' 35563 35564 Calls to functions that have no observable effects on the state of 35565 the program other than to return a value may lend themselves to 35566 optimizations such as common subexpression elimination. Declaring 35567 such functions with the 'pure' attribute allows GCC to avoid 35568 emitting some calls in repeated invocations of the function with 35569 the same argument values. 35570 35571 The 'pure' attribute prohibits a function from modifying the state 35572 of the program that is observable by means other than inspecting 35573 the function's return value. However, functions declared with the 35574 'pure' attribute can safely read any non-volatile objects, and 35575 modify the value of objects in a way that does not affect their 35576 return value or the observable state of the program. 35577 35578 For example, 35579 35580 int hash (char *) __attribute__ ((pure)); 35581 35582 tells GCC that subsequent calls to the function 'hash' with the 35583 same string can be replaced by the result of the first call 35584 provided the state of the program observable by 'hash', including 35585 the contents of the array itself, does not change in between. Even 35586 though 'hash' takes a non-const pointer argument it must not modify 35587 the array it points to, or any other object whose value the rest of 35588 the program may depend on. However, the caller may safely change 35589 the contents of the array between successive calls to the function 35590 (doing so disables the optimization). The restriction also applies 35591 to member objects referenced by the 'this' pointer in C++ 35592 non-static member functions. 35593 35594 Some common examples of pure functions are 'strlen' or 'memcmp'. 35595 Interesting non-pure functions are functions with infinite loops or 35596 those depending on volatile memory or other system resource, that 35597 may change between consecutive calls (such as the standard C 'feof' 35598 function in a multithreading environment). 35599 35600 The 'pure' attribute imposes similar but looser restrictions on a 35601 function's definition than the 'const' attribute: 'pure' allows the 35602 function to read any non-volatile memory, even if it changes in 35603 between successive invocations of the function. Declaring the same 35604 function with both the 'pure' and the 'const' attribute is 35605 diagnosed. Because a pure function cannot have any observable side 35606 effects it does not make sense for such a function to return 35607 'void'. Declaring such a function is diagnosed. 35608 35609'returns_nonnull' 35610 The 'returns_nonnull' attribute specifies that the function return 35611 value should be a non-null pointer. For instance, the declaration: 35612 35613 extern void * 35614 mymalloc (size_t len) __attribute__((returns_nonnull)); 35615 35616 lets the compiler optimize callers based on the knowledge that the 35617 return value will never be null. 35618 35619'returns_twice' 35620 The 'returns_twice' attribute tells the compiler that a function 35621 may return more than one time. The compiler ensures that all 35622 registers are dead before calling such a function and emits a 35623 warning about the variables that may be clobbered after the second 35624 return from the function. Examples of such functions are 'setjmp' 35625 and 'vfork'. The 'longjmp'-like counterpart of such function, if 35626 any, might need to be marked with the 'noreturn' attribute. 35627 35628'section ("SECTION-NAME")' 35629 Normally, the compiler places the code it generates in the 'text' 35630 section. Sometimes, however, you need additional sections, or you 35631 need certain particular functions to appear in special sections. 35632 The 'section' attribute specifies that a function lives in a 35633 particular section. For example, the declaration: 35634 35635 extern void foobar (void) __attribute__ ((section ("bar"))); 35636 35637 puts the function 'foobar' in the 'bar' section. 35638 35639 Some file formats do not support arbitrary sections so the 35640 'section' attribute is not available on all platforms. If you need 35641 to map the entire contents of a module to a particular section, 35642 consider using the facilities of the linker instead. 35643 35644'sentinel' 35645'sentinel (POSITION)' 35646 This function attribute indicates that an argument in a call to the 35647 function is expected to be an explicit 'NULL'. The attribute is 35648 only valid on variadic functions. By default, the sentinel is 35649 expected to be the last argument of the function call. If the 35650 optional POSITION argument is specified to the attribute, the 35651 sentinel must be located at POSITION counting backwards from the 35652 end of the argument list. 35653 35654 __attribute__ ((sentinel)) 35655 is equivalent to 35656 __attribute__ ((sentinel(0))) 35657 35658 The attribute is automatically set with a position of 0 for the 35659 built-in functions 'execl' and 'execlp'. The built-in function 35660 'execle' has the attribute set with a position of 1. 35661 35662 A valid 'NULL' in this context is defined as zero with any object 35663 pointer type. If your system defines the 'NULL' macro with an 35664 integer type then you need to add an explicit cast. During 35665 installation GCC replaces the system '<stddef.h>' header with a 35666 copy that redefines NULL appropriately. 35667 35668 The warnings for missing or incorrect sentinels are enabled with 35669 '-Wformat'. 35670 35671'simd' 35672'simd("MASK")' 35673 This attribute enables creation of one or more function versions 35674 that can process multiple arguments using SIMD instructions from a 35675 single invocation. Specifying this attribute allows compiler to 35676 assume that such versions are available at link time (provided in 35677 the same or another translation unit). Generated versions are 35678 target-dependent and described in the corresponding Vector ABI 35679 document. For x86_64 target this document can be found 35680 here (https://sourceware.org/glibc/wiki/libmvec?action=AttachFile&do=view&target=VectorABI.txt). 35681 35682 The optional argument MASK may have the value 'notinbranch' or 35683 'inbranch', and instructs the compiler to generate non-masked or 35684 masked clones correspondingly. By default, all clones are 35685 generated. 35686 35687 If the attribute is specified and '#pragma omp declare simd' is 35688 present on a declaration and the '-fopenmp' or '-fopenmp-simd' 35689 switch is specified, then the attribute is ignored. 35690 35691'stack_protect' 35692 This attribute adds stack protection code to the function if flags 35693 '-fstack-protector', '-fstack-protector-strong' or 35694 '-fstack-protector-explicit' are set. 35695 35696'no_stack_protector' 35697 This attribute prevents stack protection code for the function. 35698 35699'target (STRING, ...)' 35700 Multiple target back ends implement the 'target' attribute to 35701 specify that a function is to be compiled with different target 35702 options than specified on the command line. The original target 35703 command-line options are ignored. One or more strings can be 35704 provided as arguments. Each string consists of one or more 35705 comma-separated suffixes to the '-m' prefix jointly forming the 35706 name of a machine-dependent option. *Note Machine-Dependent 35707 Options: Submodel Options. 35708 35709 The 'target' attribute can be used for instance to have a function 35710 compiled with a different ISA (instruction set architecture) than 35711 the default. '#pragma GCC target' can be used to specify 35712 target-specific options for more than one function. *Note Function 35713 Specific Option Pragmas::, for details about the pragma. 35714 35715 For instance, on an x86, you could declare one function with the 35716 'target("sse4.1,arch=core2")' attribute and another with 35717 'target("sse4a,arch=amdfam10")'. This is equivalent to compiling 35718 the first function with '-msse4.1' and '-march=core2' options, and 35719 the second function with '-msse4a' and '-march=amdfam10' options. 35720 It is up to you to make sure that a function is only invoked on a 35721 machine that supports the particular ISA it is compiled for (for 35722 example by using 'cpuid' on x86 to determine what feature bits and 35723 architecture family are used). 35724 35725 int core2_func (void) __attribute__ ((__target__ ("arch=core2"))); 35726 int sse3_func (void) __attribute__ ((__target__ ("sse3"))); 35727 35728 Providing multiple strings as arguments separated by commas to 35729 specify multiple options is equivalent to separating the option 35730 suffixes with a comma (',') within a single string. Spaces are not 35731 permitted within the strings. 35732 35733 The options supported are specific to each target; refer to *note 35734 x86 Function Attributes::, *note PowerPC Function Attributes::, 35735 *note ARM Function Attributes::, *note AArch64 Function 35736 Attributes::, *note Nios II Function Attributes::, and *note S/390 35737 Function Attributes:: for details. 35738 35739'symver ("NAME2@NODENAME")' 35740 On ELF targets this attribute creates a symbol version. The NAME2 35741 part of the parameter is the actual name of the symbol by which it 35742 will be externally referenced. The 'nodename' portion should be 35743 the name of a node specified in the version script supplied to the 35744 linker when building a shared library. Versioned symbol must be 35745 defined and must be exported with default visibility. 35746 35747 __attribute__ ((__symver__ ("foo@VERS_1"))) int 35748 foo_v1 (void) 35749 { 35750 } 35751 35752 Will produce a '.symver foo_v1, foo@VERS_1' directive in the 35753 assembler output. 35754 35755 One can also define multiple version for a given symbol (starting 35756 from binutils 2.35). 35757 35758 __attribute__ ((__symver__ ("foo@VERS_2"), __symver__ ("foo@VERS_3"))) 35759 int symver_foo_v1 (void) 35760 { 35761 } 35762 35763 This example creates a symbol name 'symver_foo_v1' which will be 35764 version 'VERS_2' and 'VERS_3' of 'foo'. 35765 35766 If you have an older release of binutils, then symbol alias needs 35767 to be used: 35768 35769 __attribute__ ((__symver__ ("foo@VERS_2"))) 35770 int foo_v1 (void) 35771 { 35772 return 0; 35773 } 35774 35775 __attribute__ ((__symver__ ("foo@VERS_3"))) 35776 __attribute__ ((alias ("foo_v1"))) 35777 int symver_foo_v1 (void); 35778 35779 Finally if the parameter is '"NAME2@@NODENAME"' then in addition to 35780 creating a symbol version (as if '"NAME2@NODENAME"' was used) the 35781 version will be also used to resolve NAME2 by the linker. 35782 35783'tainted_args' 35784 The 'tainted_args' attribute is used to specify that a function is 35785 called in a way that requires sanitization of its arguments, such 35786 as a system call in an operating system kernel. Such a function 35787 can be considered part of the "attack surface" of the program. The 35788 attribute can be used both on function declarations, and on field 35789 declarations containing function pointers. In the latter case, any 35790 function used as an initializer of such a callback field will be 35791 treated as being called with tainted arguments. 35792 35793 The analyzer will pay particular attention to such functions when 35794 both '-fanalyzer' and '-fanalyzer-checker=taint' are supplied, 35795 potentially issuing warnings guarded by 35796 '-Wanalyzer-tainted-allocation-size', 35797 '-Wanalyzer-tainted-array-index', '-Wanalyzer-tainted-divisor', 35798 '-Wanalyzer-tainted-offset', and '-Wanalyzer-tainted-size'. 35799 35800'target_clones (OPTIONS)' 35801 The 'target_clones' attribute is used to specify that a function be 35802 cloned into multiple versions compiled with different target 35803 options than specified on the command line. The supported options 35804 and restrictions are the same as for 'target' attribute. 35805 35806 For instance, on an x86, you could compile a function with 35807 'target_clones("sse4.1,avx")'. GCC creates two function clones, 35808 one compiled with '-msse4.1' and another with '-mavx'. 35809 35810 On a PowerPC, you can compile a function with 35811 'target_clones("cpu=power9,default")'. GCC will create two 35812 function clones, one compiled with '-mcpu=power9' and another with 35813 the default options. GCC must be configured to use GLIBC 2.23 or 35814 newer in order to use the 'target_clones' attribute. 35815 35816 It also creates a resolver function (see the 'ifunc' attribute 35817 above) that dynamically selects a clone suitable for current 35818 architecture. The resolver is created only if there is a usage of 35819 a function with 'target_clones' attribute. 35820 35821 Note that any subsequent call of a function without 'target_clone' 35822 from a 'target_clone' caller will not lead to copying (target 35823 clone) of the called function. If you want to enforce such 35824 behaviour, we recommend declaring the calling function with the 35825 'flatten' attribute? 35826 35827'unused' 35828 This attribute, attached to a function, means that the function is 35829 meant to be possibly unused. GCC does not produce a warning for 35830 this function. 35831 35832'used' 35833 This attribute, attached to a function, means that code must be 35834 emitted for the function even if it appears that the function is 35835 not referenced. This is useful, for example, when the function is 35836 referenced only in inline assembly. 35837 35838 When applied to a member function of a C++ class template, the 35839 attribute also means that the function is instantiated if the class 35840 itself is instantiated. 35841 35842'retain' 35843 For ELF targets that support the GNU or FreeBSD OSABIs, this 35844 attribute will save the function from linker garbage collection. 35845 To support this behavior, functions that have not been placed in 35846 specific sections (e.g. by the 'section' attribute, or the 35847 '-ffunction-sections' option), will be placed in new, unique 35848 sections. 35849 35850 This additional functionality requires Binutils version 2.36 or 35851 later. 35852 35853'visibility ("VISIBILITY_TYPE")' 35854 This attribute affects the linkage of the declaration to which it 35855 is attached. It can be applied to variables (*note Common Variable 35856 Attributes::) and types (*note Common Type Attributes::) as well as 35857 functions. 35858 35859 There are four supported VISIBILITY_TYPE values: default, hidden, 35860 protected or internal visibility. 35861 35862 void __attribute__ ((visibility ("protected"))) 35863 f () { /* Do something. */; } 35864 int i __attribute__ ((visibility ("hidden"))); 35865 35866 The possible values of VISIBILITY_TYPE correspond to the visibility 35867 settings in the ELF gABI. 35868 35869 'default' 35870 Default visibility is the normal case for the object file 35871 format. This value is available for the visibility attribute 35872 to override other options that may change the assumed 35873 visibility of entities. 35874 35875 On ELF, default visibility means that the declaration is 35876 visible to other modules and, in shared libraries, means that 35877 the declared entity may be overridden. 35878 35879 On Darwin, default visibility means that the declaration is 35880 visible to other modules. 35881 35882 Default visibility corresponds to "external linkage" in the 35883 language. 35884 35885 'hidden' 35886 Hidden visibility indicates that the entity declared has a new 35887 form of linkage, which we call "hidden linkage". Two 35888 declarations of an object with hidden linkage refer to the 35889 same object if they are in the same shared object. 35890 35891 'internal' 35892 Internal visibility is like hidden visibility, but with 35893 additional processor specific semantics. Unless otherwise 35894 specified by the psABI, GCC defines internal visibility to 35895 mean that a function is _never_ called from another module. 35896 Compare this with hidden functions which, while they cannot be 35897 referenced directly by other modules, can be referenced 35898 indirectly via function pointers. By indicating that a 35899 function cannot be called from outside the module, GCC may for 35900 instance omit the load of a PIC register since it is known 35901 that the calling function loaded the correct value. 35902 35903 'protected' 35904 Protected visibility is like default visibility except that it 35905 indicates that references within the defining module bind to 35906 the definition in that module. That is, the declared entity 35907 cannot be overridden by another module. 35908 35909 All visibilities are supported on many, but not all, ELF targets 35910 (supported when the assembler supports the '.visibility' 35911 pseudo-op). Default visibility is supported everywhere. Hidden 35912 visibility is supported on Darwin targets. 35913 35914 The visibility attribute should be applied only to declarations 35915 that would otherwise have external linkage. The attribute should 35916 be applied consistently, so that the same entity should not be 35917 declared with different settings of the attribute. 35918 35919 In C++, the visibility attribute applies to types as well as 35920 functions and objects, because in C++ types have linkage. A class 35921 must not have greater visibility than its non-static data member 35922 types and bases, and class members default to the visibility of 35923 their class. Also, a declaration without explicit visibility is 35924 limited to the visibility of its type. 35925 35926 In C++, you can mark member functions and static member variables 35927 of a class with the visibility attribute. This is useful if you 35928 know a particular method or static member variable should only be 35929 used from one shared object; then you can mark it hidden while the 35930 rest of the class has default visibility. Care must be taken to 35931 avoid breaking the One Definition Rule; for example, it is usually 35932 not useful to mark an inline method as hidden without marking the 35933 whole class as hidden. 35934 35935 A C++ namespace declaration can also have the visibility attribute. 35936 35937 namespace nspace1 __attribute__ ((visibility ("protected"))) 35938 { /* Do something. */; } 35939 35940 This attribute applies only to the particular namespace body, not 35941 to other definitions of the same namespace; it is equivalent to 35942 using '#pragma GCC visibility' before and after the namespace 35943 definition (*note Visibility Pragmas::). 35944 35945 In C++, if a template argument has limited visibility, this 35946 restriction is implicitly propagated to the template instantiation. 35947 Otherwise, template instantiations and specializations default to 35948 the visibility of their template. 35949 35950 If both the template and enclosing class have explicit visibility, 35951 the visibility from the template is used. 35952 35953'warn_unused_result' 35954 The 'warn_unused_result' attribute causes a warning to be emitted 35955 if a caller of the function with this attribute does not use its 35956 return value. This is useful for functions where not checking the 35957 result is either a security problem or always a bug, such as 35958 'realloc'. 35959 35960 int fn () __attribute__ ((warn_unused_result)); 35961 int foo () 35962 { 35963 if (fn () < 0) return -1; 35964 fn (); 35965 return 0; 35966 } 35967 35968 results in warning on line 5. 35969 35970'weak' 35971 The 'weak' attribute causes a declaration of an external symbol to 35972 be emitted as a weak symbol rather than a global. This is 35973 primarily useful in defining library functions that can be 35974 overridden in user code, though it can also be used with 35975 non-function declarations. The overriding symbol must have the 35976 same type as the weak symbol. In addition, if it designates a 35977 variable it must also have the same size and alignment as the weak 35978 symbol. Weak symbols are supported for ELF targets, and also for 35979 a.out targets when using the GNU assembler and linker. 35980 35981'weakref' 35982'weakref ("TARGET")' 35983 The 'weakref' attribute marks a declaration as a weak reference. 35984 Without arguments, it should be accompanied by an 'alias' attribute 35985 naming the target symbol. Alternatively, TARGET may be given as an 35986 argument to 'weakref' itself, naming the target definition of the 35987 alias. The TARGET must have the same type as the declaration. In 35988 addition, if it designates a variable it must also have the same 35989 size and alignment as the declaration. In either form of the 35990 declaration 'weakref' implicitly marks the declared symbol as 35991 'weak'. Without a TARGET given as an argument to 'weakref' or to 35992 'alias', 'weakref' is equivalent to 'weak' (in that case the 35993 declaration may be 'extern'). 35994 35995 /* Given the declaration: */ 35996 extern int y (void); 35997 35998 /* the following... */ 35999 static int x (void) __attribute__ ((weakref ("y"))); 36000 36001 /* is equivalent to... */ 36002 static int x (void) __attribute__ ((weakref, alias ("y"))); 36003 36004 /* or, alternatively, to... */ 36005 static int x (void) __attribute__ ((weakref)); 36006 static int x (void) __attribute__ ((alias ("y"))); 36007 36008 A weak reference is an alias that does not by itself require a 36009 definition to be given for the target symbol. If the target symbol 36010 is only referenced through weak references, then it becomes a 36011 'weak' undefined symbol. If it is directly referenced, however, 36012 then such strong references prevail, and a definition is required 36013 for the symbol, not necessarily in the same translation unit. 36014 36015 The effect is equivalent to moving all references to the alias to a 36016 separate translation unit, renaming the alias to the aliased 36017 symbol, declaring it as weak, compiling the two separate 36018 translation units and performing a link with relocatable output 36019 (i.e. 'ld -r') on them. 36020 36021 A declaration to which 'weakref' is attached and that is associated 36022 with a named 'target' must be 'static'. 36023 36024'zero_call_used_regs ("CHOICE")' 36025 36026 The 'zero_call_used_regs' attribute causes the compiler to zero a 36027 subset of all call-used registers(1) at function return. This is 36028 used to increase program security by either mitigating 36029 Return-Oriented Programming (ROP) attacks or preventing information 36030 leakage through registers. 36031 36032 In order to satisfy users with different security needs and control 36033 the run-time overhead at the same time, the CHOICE parameter 36034 provides a flexible way to choose the subset of the call-used 36035 registers to be zeroed. The three basic values of CHOICE are: 36036 36037 * 'skip' doesn't zero any call-used registers. 36038 36039 * 'used' only zeros call-used registers that are used in the 36040 function. A "used" register is one whose content has been set 36041 or referenced in the function. 36042 36043 * 'all' zeros all call-used registers. 36044 36045 In addition to these three basic choices, it is possible to modify 36046 'used' or 'all' as follows: 36047 36048 * Adding '-gpr' restricts the zeroing to general-purpose 36049 registers. 36050 36051 * Adding '-arg' restricts the zeroing to registers that can 36052 sometimes be used to pass function arguments. This includes 36053 all argument registers defined by the platform's calling 36054 conversion, regardless of whether the function uses those 36055 registers for function arguments or not. 36056 36057 The modifiers can be used individually or together. If they are 36058 used together, they must appear in the order above. 36059 36060 The full list of CHOICEs is therefore: 36061 36062 'skip' 36063 doesn't zero any call-used register. 36064 36065 'used' 36066 only zeros call-used registers that are used in the function. 36067 36068 'used-gpr' 36069 only zeros call-used general purpose registers that are used 36070 in the function. 36071 36072 'used-arg' 36073 only zeros call-used registers that are used in the function 36074 and pass arguments. 36075 36076 'used-gpr-arg' 36077 only zeros call-used general purpose registers that are used 36078 in the function and pass arguments. 36079 36080 'all' 36081 zeros all call-used registers. 36082 36083 'all-gpr' 36084 zeros all call-used general purpose registers. 36085 36086 'all-arg' 36087 zeros all call-used registers that pass arguments. 36088 36089 'all-gpr-arg' 36090 zeros all call-used general purpose registers that pass 36091 arguments. 36092 36093 Of this list, 'used-arg', 'used-gpr-arg', 'all-arg', and 36094 'all-gpr-arg' are mainly used for ROP mitigation. 36095 36096 The default for the attribute is controlled by 36097 '-fzero-call-used-regs'. 36098 36099 ---------- Footnotes ---------- 36100 36101 (1) A "call-used" register is a register whose contents can be 36102changed by a function call; therefore, a caller cannot assume that the 36103register has the same contents on return from the function as it had 36104before calling the function. Such registers are also called 36105"call-clobbered", "caller-saved", or "volatile". 36106 36107 36108File: gcc.info, Node: AArch64 Function Attributes, Next: AMD GCN Function Attributes, Prev: Common Function Attributes, Up: Function Attributes 36109 361106.33.2 AArch64 Function Attributes 36111---------------------------------- 36112 36113The following target-specific function attributes are available for the 36114AArch64 target. For the most part, these options mirror the behavior of 36115similar command-line options (*note AArch64 Options::), but on a 36116per-function basis. 36117 36118'general-regs-only' 36119 Indicates that no floating-point or Advanced SIMD registers should 36120 be used when generating code for this function. If the function 36121 explicitly uses floating-point code, then the compiler gives an 36122 error. This is the same behavior as that of the command-line 36123 option '-mgeneral-regs-only'. 36124 36125'fix-cortex-a53-835769' 36126 Indicates that the workaround for the Cortex-A53 erratum 835769 36127 should be applied to this function. To explicitly disable the 36128 workaround for this function specify the negated form: 36129 'no-fix-cortex-a53-835769'. This corresponds to the behavior of 36130 the command line options '-mfix-cortex-a53-835769' and 36131 '-mno-fix-cortex-a53-835769'. 36132 36133'cmodel=' 36134 Indicates that code should be generated for a particular code model 36135 for this function. The behavior and permissible arguments are the 36136 same as for the command line option '-mcmodel='. 36137 36138'strict-align' 36139'no-strict-align' 36140 'strict-align' indicates that the compiler should not assume that 36141 unaligned memory references are handled by the system. To allow 36142 the compiler to assume that aligned memory references are handled 36143 by the system, the inverse attribute 'no-strict-align' can be 36144 specified. The behavior is same as for the command-line option 36145 '-mstrict-align' and '-mno-strict-align'. 36146 36147'omit-leaf-frame-pointer' 36148 Indicates that the frame pointer should be omitted for a leaf 36149 function call. To keep the frame pointer, the inverse attribute 36150 'no-omit-leaf-frame-pointer' can be specified. These attributes 36151 have the same behavior as the command-line options 36152 '-momit-leaf-frame-pointer' and '-mno-omit-leaf-frame-pointer'. 36153 36154'tls-dialect=' 36155 Specifies the TLS dialect to use for this function. The behavior 36156 and permissible arguments are the same as for the command-line 36157 option '-mtls-dialect='. 36158 36159'arch=' 36160 Specifies the architecture version and architectural extensions to 36161 use for this function. The behavior and permissible arguments are 36162 the same as for the '-march=' command-line option. 36163 36164'tune=' 36165 Specifies the core for which to tune the performance of this 36166 function. The behavior and permissible arguments are the same as 36167 for the '-mtune=' command-line option. 36168 36169'cpu=' 36170 Specifies the core for which to tune the performance of this 36171 function and also whose architectural features to use. The 36172 behavior and valid arguments are the same as for the '-mcpu=' 36173 command-line option. 36174 36175'sign-return-address' 36176 Select the function scope on which return address signing will be 36177 applied. The behavior and permissible arguments are the same as 36178 for the command-line option '-msign-return-address='. The default 36179 value is 'none'. This attribute is deprecated. The 36180 'branch-protection' attribute should be used instead. 36181 36182'branch-protection' 36183 Select the function scope on which branch protection will be 36184 applied. The behavior and permissible arguments are the same as 36185 for the command-line option '-mbranch-protection='. The default 36186 value is 'none'. 36187 36188'outline-atomics' 36189 Enable or disable calls to out-of-line helpers to implement atomic 36190 operations. This corresponds to the behavior of the command line 36191 options '-moutline-atomics' and '-mno-outline-atomics'. 36192 36193 The above target attributes can be specified as follows: 36194 36195 __attribute__((target("ATTR-STRING"))) 36196 int 36197 f (int a) 36198 { 36199 return a + 5; 36200 } 36201 36202 where 'ATTR-STRING' is one of the attribute strings specified above. 36203 36204 Additionally, the architectural extension string may be specified on 36205its own. This can be used to turn on and off particular architectural 36206extensions without having to specify a particular architecture version 36207or core. Example: 36208 36209 __attribute__((target("+crc+nocrypto"))) 36210 int 36211 foo (int a) 36212 { 36213 return a + 5; 36214 } 36215 36216 In this example 'target("+crc+nocrypto")' enables the 'crc' extension 36217and disables the 'crypto' extension for the function 'foo' without 36218modifying an existing '-march=' or '-mcpu' option. 36219 36220 Multiple target function attributes can be specified by separating them 36221with a comma. For example: 36222 __attribute__((target("arch=armv8-a+crc+crypto,tune=cortex-a53"))) 36223 int 36224 foo (int a) 36225 { 36226 return a + 5; 36227 } 36228 36229 is valid and compiles function 'foo' for ARMv8-A with 'crc' and 36230'crypto' extensions and tunes it for 'cortex-a53'. 36231 362326.33.2.1 Inlining rules 36233....................... 36234 36235Specifying target attributes on individual functions or performing 36236link-time optimization across translation units compiled with different 36237target options can affect function inlining rules: 36238 36239 In particular, a caller function can inline a callee function only if 36240the architectural features available to the callee are a subset of the 36241features available to the caller. For example: A function 'foo' 36242compiled with '-march=armv8-a+crc', or tagged with the equivalent 36243'arch=armv8-a+crc' attribute, can inline a function 'bar' compiled with 36244'-march=armv8-a+nocrc' because the all the architectural features that 36245function 'bar' requires are available to function 'foo'. Conversely, 36246function 'bar' cannot inline function 'foo'. 36247 36248 Additionally inlining a function compiled with '-mstrict-align' into a 36249function compiled without '-mstrict-align' is not allowed. However, 36250inlining a function compiled without '-mstrict-align' into a function 36251compiled with '-mstrict-align' is allowed. 36252 36253 Note that CPU tuning options and attributes such as the '-mcpu=', 36254'-mtune=' do not inhibit inlining unless the CPU specified by the 36255'-mcpu=' option or the 'cpu=' attribute conflicts with the architectural 36256feature rules specified above. 36257 36258 36259File: gcc.info, Node: AMD GCN Function Attributes, Next: ARC Function Attributes, Prev: AArch64 Function Attributes, Up: Function Attributes 36260 362616.33.3 AMD GCN Function Attributes 36262---------------------------------- 36263 36264These function attributes are supported by the AMD GCN back end: 36265 36266'amdgpu_hsa_kernel' 36267 This attribute indicates that the corresponding function should be 36268 compiled as a kernel function, that is an entry point that can be 36269 invoked from the host via the HSA runtime library. By default 36270 functions are only callable only from other GCN functions. 36271 36272 This attribute is implicitly applied to any function named 'main', 36273 using default parameters. 36274 36275 Kernel functions may return an integer value, which will be written 36276 to a conventional place within the HSA "kernargs" region. 36277 36278 The attribute parameters configure what values are passed into the 36279 kernel function by the GPU drivers, via the initial register state. 36280 Some values are used by the compiler, and therefore forced on. 36281 Enabling other options may break assumptions in the compiler and/or 36282 run-time libraries. 36283 36284 'private_segment_buffer' 36285 Set 'enable_sgpr_private_segment_buffer' flag. Always on 36286 (required to locate the stack). 36287 36288 'dispatch_ptr' 36289 Set 'enable_sgpr_dispatch_ptr' flag. Always on (required to 36290 locate the launch dimensions). 36291 36292 'queue_ptr' 36293 Set 'enable_sgpr_queue_ptr' flag. Always on (required to 36294 convert address spaces). 36295 36296 'kernarg_segment_ptr' 36297 Set 'enable_sgpr_kernarg_segment_ptr' flag. Always on 36298 (required to locate the kernel arguments, "kernargs"). 36299 36300 'dispatch_id' 36301 Set 'enable_sgpr_dispatch_id' flag. 36302 36303 'flat_scratch_init' 36304 Set 'enable_sgpr_flat_scratch_init' flag. 36305 36306 'private_segment_size' 36307 Set 'enable_sgpr_private_segment_size' flag. 36308 36309 'grid_workgroup_count_X' 36310 Set 'enable_sgpr_grid_workgroup_count_x' flag. Always on 36311 (required to use OpenACC/OpenMP). 36312 36313 'grid_workgroup_count_Y' 36314 Set 'enable_sgpr_grid_workgroup_count_y' flag. 36315 36316 'grid_workgroup_count_Z' 36317 Set 'enable_sgpr_grid_workgroup_count_z' flag. 36318 36319 'workgroup_id_X' 36320 Set 'enable_sgpr_workgroup_id_x' flag. 36321 36322 'workgroup_id_Y' 36323 Set 'enable_sgpr_workgroup_id_y' flag. 36324 36325 'workgroup_id_Z' 36326 Set 'enable_sgpr_workgroup_id_z' flag. 36327 36328 'workgroup_info' 36329 Set 'enable_sgpr_workgroup_info' flag. 36330 36331 'private_segment_wave_offset' 36332 Set 'enable_sgpr_private_segment_wave_byte_offset' flag. 36333 Always on (required to locate the stack). 36334 36335 'work_item_id_X' 36336 Set 'enable_vgpr_workitem_id' parameter. Always on (can't be 36337 disabled). 36338 36339 'work_item_id_Y' 36340 Set 'enable_vgpr_workitem_id' parameter. Always on (required 36341 to enable vectorization.) 36342 36343 'work_item_id_Z' 36344 Set 'enable_vgpr_workitem_id' parameter. Always on (required 36345 to use OpenACC/OpenMP). 36346 36347 36348File: gcc.info, Node: ARC Function Attributes, Next: ARM Function Attributes, Prev: AMD GCN Function Attributes, Up: Function Attributes 36349 363506.33.4 ARC Function Attributes 36351------------------------------ 36352 36353These function attributes are supported by the ARC back end: 36354 36355'interrupt' 36356 Use this attribute to indicate that the specified function is an 36357 interrupt handler. The compiler generates function entry and exit 36358 sequences suitable for use in an interrupt handler when this 36359 attribute is present. 36360 36361 On the ARC, you must specify the kind of interrupt to be handled in 36362 a parameter to the interrupt attribute like this: 36363 36364 void f () __attribute__ ((interrupt ("ilink1"))); 36365 36366 Permissible values for this parameter are: 'ilink1' and 'ilink2' 36367 for ARCv1 architecture, and 'ilink' and 'firq' for ARCv2 36368 architecture. 36369 36370'long_call' 36371'medium_call' 36372'short_call' 36373 These attributes specify how a particular function is called. 36374 These attributes override the '-mlong-calls' and '-mmedium-calls' 36375 (*note ARC Options::) command-line switches and '#pragma 36376 long_calls' settings. 36377 36378 For ARC, a function marked with the 'long_call' attribute is always 36379 called using register-indirect jump-and-link instructions, thereby 36380 enabling the called function to be placed anywhere within the 36381 32-bit address space. A function marked with the 'medium_call' 36382 attribute will always be close enough to be called with an 36383 unconditional branch-and-link instruction, which has a 25-bit 36384 offset from the call site. A function marked with the 'short_call' 36385 attribute will always be close enough to be called with a 36386 conditional branch-and-link instruction, which has a 21-bit offset 36387 from the call site. 36388 36389'jli_always' 36390 Forces a particular function to be called using 'jli' instruction. 36391 The 'jli' instruction makes use of a table stored into '.jlitab' 36392 section, which holds the location of the functions which are 36393 addressed using this instruction. 36394 36395'jli_fixed' 36396 Identical like the above one, but the location of the function in 36397 the 'jli' table is known and given as an attribute parameter. 36398 36399'secure_call' 36400 This attribute allows one to mark secure-code functions that are 36401 callable from normal mode. The location of the secure call 36402 function into the 'sjli' table needs to be passed as argument. 36403 36404'naked' 36405 This attribute allows the compiler to construct the requisite 36406 function declaration, while allowing the body of the function to be 36407 assembly code. The specified function will not have 36408 prologue/epilogue sequences generated by the compiler. Only basic 36409 'asm' statements can safely be included in naked functions (*note 36410 Basic Asm::). While using extended 'asm' or a mixture of basic 36411 'asm' and C code may appear to work, they cannot be depended upon 36412 to work reliably and are not supported. 36413 36414 36415File: gcc.info, Node: ARM Function Attributes, Next: AVR Function Attributes, Prev: ARC Function Attributes, Up: Function Attributes 36416 364176.33.5 ARM Function Attributes 36418------------------------------ 36419 36420These function attributes are supported for ARM targets: 36421 36422'general-regs-only' 36423 Indicates that no floating-point or Advanced SIMD registers should 36424 be used when generating code for this function. If the function 36425 explicitly uses floating-point code, then the compiler gives an 36426 error. This is the same behavior as that of the command-line 36427 option '-mgeneral-regs-only'. 36428 36429'interrupt' 36430 Use this attribute to indicate that the specified function is an 36431 interrupt handler. The compiler generates function entry and exit 36432 sequences suitable for use in an interrupt handler when this 36433 attribute is present. 36434 36435 You can specify the kind of interrupt to be handled by adding an 36436 optional parameter to the interrupt attribute like this: 36437 36438 void f () __attribute__ ((interrupt ("IRQ"))); 36439 36440 Permissible values for this parameter are: 'IRQ', 'FIQ', 'SWI', 36441 'ABORT' and 'UNDEF'. 36442 36443 On ARMv7-M the interrupt type is ignored, and the attribute means 36444 the function may be called with a word-aligned stack pointer. 36445 36446'isr' 36447 Use this attribute on ARM to write Interrupt Service Routines. 36448 This is an alias to the 'interrupt' attribute above. 36449 36450'long_call' 36451'short_call' 36452 These attributes specify how a particular function is called. 36453 These attributes override the '-mlong-calls' (*note ARM Options::) 36454 command-line switch and '#pragma long_calls' settings. For ARM, 36455 the 'long_call' attribute indicates that the function might be far 36456 away from the call site and require a different (more expensive) 36457 calling sequence. The 'short_call' attribute always places the 36458 offset to the function from the call site into the 'BL' instruction 36459 directly. 36460 36461'naked' 36462 This attribute allows the compiler to construct the requisite 36463 function declaration, while allowing the body of the function to be 36464 assembly code. The specified function will not have 36465 prologue/epilogue sequences generated by the compiler. Only basic 36466 'asm' statements can safely be included in naked functions (*note 36467 Basic Asm::). While using extended 'asm' or a mixture of basic 36468 'asm' and C code may appear to work, they cannot be depended upon 36469 to work reliably and are not supported. 36470 36471'pcs' 36472 36473 The 'pcs' attribute can be used to control the calling convention 36474 used for a function on ARM. The attribute takes an argument that 36475 specifies the calling convention to use. 36476 36477 When compiling using the AAPCS ABI (or a variant of it) then valid 36478 values for the argument are '"aapcs"' and '"aapcs-vfp"'. In order 36479 to use a variant other than '"aapcs"' then the compiler must be 36480 permitted to use the appropriate co-processor registers (i.e., the 36481 VFP registers must be available in order to use '"aapcs-vfp"'). 36482 For example, 36483 36484 /* Argument passed in r0, and result returned in r0+r1. */ 36485 double f2d (float) __attribute__((pcs("aapcs"))); 36486 36487 Variadic functions always use the '"aapcs"' calling convention and 36488 the compiler rejects attempts to specify an alternative. 36489 36490'target (OPTIONS)' 36491 As discussed in *note Common Function Attributes::, this attribute 36492 allows specification of target-specific compilation options. 36493 36494 On ARM, the following options are allowed: 36495 36496 'thumb' 36497 Force code generation in the Thumb (T16/T32) ISA, depending on 36498 the architecture level. 36499 36500 'arm' 36501 Force code generation in the ARM (A32) ISA. 36502 36503 Functions from different modes can be inlined in the caller's 36504 mode. 36505 36506 'fpu=' 36507 Specifies the fpu for which to tune the performance of this 36508 function. The behavior and permissible arguments are the same 36509 as for the '-mfpu=' command-line option. 36510 36511 'arch=' 36512 Specifies the architecture version and architectural 36513 extensions to use for this function. The behavior and 36514 permissible arguments are the same as for the '-march=' 36515 command-line option. 36516 36517 The above target attributes can be specified as follows: 36518 36519 __attribute__((target("arch=armv8-a+crc"))) 36520 int 36521 f (int a) 36522 { 36523 return a + 5; 36524 } 36525 36526 Additionally, the architectural extension string may be 36527 specified on its own. This can be used to turn on and off 36528 particular architectural extensions without having to specify 36529 a particular architecture version or core. Example: 36530 36531 __attribute__((target("+crc+nocrypto"))) 36532 int 36533 foo (int a) 36534 { 36535 return a + 5; 36536 } 36537 36538 In this example 'target("+crc+nocrypto")' enables the 'crc' 36539 extension and disables the 'crypto' extension for the function 36540 'foo' without modifying an existing '-march=' or '-mcpu' 36541 option. 36542 36543 36544File: gcc.info, Node: AVR Function Attributes, Next: Blackfin Function Attributes, Prev: ARM Function Attributes, Up: Function Attributes 36545 365466.33.6 AVR Function Attributes 36547------------------------------ 36548 36549These function attributes are supported by the AVR back end: 36550 36551'interrupt' 36552 Use this attribute to indicate that the specified function is an 36553 interrupt handler. The compiler generates function entry and exit 36554 sequences suitable for use in an interrupt handler when this 36555 attribute is present. 36556 36557 On the AVR, the hardware globally disables interrupts when an 36558 interrupt is executed. The first instruction of an interrupt 36559 handler declared with this attribute is a 'SEI' instruction to 36560 re-enable interrupts. See also the 'signal' function attribute 36561 that does not insert a 'SEI' instruction. If both 'signal' and 36562 'interrupt' are specified for the same function, 'signal' is 36563 silently ignored. 36564 36565'naked' 36566 This attribute allows the compiler to construct the requisite 36567 function declaration, while allowing the body of the function to be 36568 assembly code. The specified function will not have 36569 prologue/epilogue sequences generated by the compiler. Only basic 36570 'asm' statements can safely be included in naked functions (*note 36571 Basic Asm::). While using extended 'asm' or a mixture of basic 36572 'asm' and C code may appear to work, they cannot be depended upon 36573 to work reliably and are not supported. 36574 36575'no_gccisr' 36576 Do not use '__gcc_isr' pseudo instructions in a function with the 36577 'interrupt' or 'signal' attribute aka. interrupt service routine 36578 (ISR). Use this attribute if the preamble of the ISR prologue 36579 should always read 36580 push __zero_reg__ 36581 push __tmp_reg__ 36582 in __tmp_reg__, __SREG__ 36583 push __tmp_reg__ 36584 clr __zero_reg__ 36585 and accordingly for the postamble of the epilogue -- no matter 36586 whether the mentioned registers are actually used in the ISR or 36587 not. Situations where you might want to use this attribute 36588 include: 36589 * Code that (effectively) clobbers bits of 'SREG' other than the 36590 'I'-flag by writing to the memory location of 'SREG'. 36591 * Code that uses inline assembler to jump to a different 36592 function which expects (parts of) the prologue code as 36593 outlined above to be present. 36594 To disable '__gcc_isr' generation for the whole compilation unit, 36595 there is option '-mno-gas-isr-prologues', *note AVR Options::. 36596 36597'OS_main' 36598'OS_task' 36599 On AVR, functions with the 'OS_main' or 'OS_task' attribute do not 36600 save/restore any call-saved register in their prologue/epilogue. 36601 36602 The 'OS_main' attribute can be used when there _is guarantee_ that 36603 interrupts are disabled at the time when the function is entered. 36604 This saves resources when the stack pointer has to be changed to 36605 set up a frame for local variables. 36606 36607 The 'OS_task' attribute can be used when there is _no guarantee_ 36608 that interrupts are disabled at that time when the function is 36609 entered like for, e.g. task functions in a multi-threading 36610 operating system. In that case, changing the stack pointer 36611 register is guarded by save/clear/restore of the global interrupt 36612 enable flag. 36613 36614 The differences to the 'naked' function attribute are: 36615 * 'naked' functions do not have a return instruction whereas 36616 'OS_main' and 'OS_task' functions have a 'RET' or 'RETI' 36617 return instruction. 36618 * 'naked' functions do not set up a frame for local variables or 36619 a frame pointer whereas 'OS_main' and 'OS_task' do this as 36620 needed. 36621 36622'signal' 36623 Use this attribute on the AVR to indicate that the specified 36624 function is an interrupt handler. The compiler generates function 36625 entry and exit sequences suitable for use in an interrupt handler 36626 when this attribute is present. 36627 36628 See also the 'interrupt' function attribute. 36629 36630 The AVR hardware globally disables interrupts when an interrupt is 36631 executed. Interrupt handler functions defined with the 'signal' 36632 attribute do not re-enable interrupts. It is save to enable 36633 interrupts in a 'signal' handler. This "save" only applies to the 36634 code generated by the compiler and not to the IRQ layout of the 36635 application which is responsibility of the application. 36636 36637 If both 'signal' and 'interrupt' are specified for the same 36638 function, 'signal' is silently ignored. 36639 36640 36641File: gcc.info, Node: Blackfin Function Attributes, Next: BPF Function Attributes, Prev: AVR Function Attributes, Up: Function Attributes 36642 366436.33.7 Blackfin Function Attributes 36644----------------------------------- 36645 36646These function attributes are supported by the Blackfin back end: 36647 36648'exception_handler' 36649 Use this attribute on the Blackfin to indicate that the specified 36650 function is an exception handler. The compiler generates function 36651 entry and exit sequences suitable for use in an exception handler 36652 when this attribute is present. 36653 36654'interrupt_handler' 36655 Use this attribute to indicate that the specified function is an 36656 interrupt handler. The compiler generates function entry and exit 36657 sequences suitable for use in an interrupt handler when this 36658 attribute is present. 36659 36660'kspisusp' 36661 When used together with 'interrupt_handler', 'exception_handler' or 36662 'nmi_handler', code is generated to load the stack pointer from the 36663 USP register in the function prologue. 36664 36665'l1_text' 36666 This attribute specifies a function to be placed into L1 36667 Instruction SRAM. The function is put into a specific section 36668 named '.l1.text'. With '-mfdpic', function calls with a such 36669 function as the callee or caller uses inlined PLT. 36670 36671'l2' 36672 This attribute specifies a function to be placed into L2 SRAM. The 36673 function is put into a specific section named '.l2.text'. With 36674 '-mfdpic', callers of such functions use an inlined PLT. 36675 36676'longcall' 36677'shortcall' 36678 The 'longcall' attribute indicates that the function might be far 36679 away from the call site and require a different (more expensive) 36680 calling sequence. The 'shortcall' attribute indicates that the 36681 function is always close enough for the shorter calling sequence to 36682 be used. These attributes override the '-mlongcall' switch. 36683 36684'nesting' 36685 Use this attribute together with 'interrupt_handler', 36686 'exception_handler' or 'nmi_handler' to indicate that the function 36687 entry code should enable nested interrupts or exceptions. 36688 36689'nmi_handler' 36690 Use this attribute on the Blackfin to indicate that the specified 36691 function is an NMI handler. The compiler generates function entry 36692 and exit sequences suitable for use in an NMI handler when this 36693 attribute is present. 36694 36695'saveall' 36696 Use this attribute to indicate that all registers except the stack 36697 pointer should be saved in the prologue regardless of whether they 36698 are used or not. 36699 36700 36701File: gcc.info, Node: BPF Function Attributes, Next: CR16 Function Attributes, Prev: Blackfin Function Attributes, Up: Function Attributes 36702 367036.33.8 BPF Function Attributes 36704------------------------------ 36705 36706These function attributes are supported by the BPF back end: 36707 36708'kernel_helper' 36709 use this attribute to indicate the specified function declaration 36710 is a kernel helper. The helper function is passed as an argument 36711 to the attribute. Example: 36712 36713 int bpf_probe_read (void *dst, int size, const void *unsafe_ptr) 36714 __attribute__ ((kernel_helper (4))); 36715 36716 36717File: gcc.info, Node: CR16 Function Attributes, Next: C-SKY Function Attributes, Prev: BPF Function Attributes, Up: Function Attributes 36718 367196.33.9 CR16 Function Attributes 36720------------------------------- 36721 36722These function attributes are supported by the CR16 back end: 36723 36724'interrupt' 36725 Use this attribute to indicate that the specified function is an 36726 interrupt handler. The compiler generates function entry and exit 36727 sequences suitable for use in an interrupt handler when this 36728 attribute is present. 36729 36730 36731File: gcc.info, Node: C-SKY Function Attributes, Next: Epiphany Function Attributes, Prev: CR16 Function Attributes, Up: Function Attributes 36732 367336.33.10 C-SKY Function Attributes 36734--------------------------------- 36735 36736These function attributes are supported by the C-SKY back end: 36737 36738'interrupt' 36739'isr' 36740 Use these attributes to indicate that the specified function is an 36741 interrupt handler. The compiler generates function entry and exit 36742 sequences suitable for use in an interrupt handler when either of 36743 these attributes are present. 36744 36745 Use of these options requires the '-mistack' command-line option to 36746 enable support for the necessary interrupt stack instructions. 36747 They are ignored with a warning otherwise. *Note C-SKY Options::. 36748 36749'naked' 36750 This attribute allows the compiler to construct the requisite 36751 function declaration, while allowing the body of the function to be 36752 assembly code. The specified function will not have 36753 prologue/epilogue sequences generated by the compiler. Only basic 36754 'asm' statements can safely be included in naked functions (*note 36755 Basic Asm::). While using extended 'asm' or a mixture of basic 36756 'asm' and C code may appear to work, they cannot be depended upon 36757 to work reliably and are not supported. 36758 36759 36760File: gcc.info, Node: Epiphany Function Attributes, Next: H8/300 Function Attributes, Prev: C-SKY Function Attributes, Up: Function Attributes 36761 367626.33.11 Epiphany Function Attributes 36763------------------------------------ 36764 36765These function attributes are supported by the Epiphany back end: 36766 36767'disinterrupt' 36768 This attribute causes the compiler to emit instructions to disable 36769 interrupts for the duration of the given function. 36770 36771'forwarder_section' 36772 This attribute modifies the behavior of an interrupt handler. The 36773 interrupt handler may be in external memory which cannot be reached 36774 by a branch instruction, so generate a local memory trampoline to 36775 transfer control. The single parameter identifies the section 36776 where the trampoline is placed. 36777 36778'interrupt' 36779 Use this attribute to indicate that the specified function is an 36780 interrupt handler. The compiler generates function entry and exit 36781 sequences suitable for use in an interrupt handler when this 36782 attribute is present. It may also generate a special section with 36783 code to initialize the interrupt vector table. 36784 36785 On Epiphany targets one or more optional parameters can be added 36786 like this: 36787 36788 void __attribute__ ((interrupt ("dma0, dma1"))) universal_dma_handler (); 36789 36790 Permissible values for these parameters are: 'reset', 36791 'software_exception', 'page_miss', 'timer0', 'timer1', 'message', 36792 'dma0', 'dma1', 'wand' and 'swi'. Multiple parameters indicate 36793 that multiple entries in the interrupt vector table should be 36794 initialized for this function, i.e. for each parameter NAME, a jump 36795 to the function is emitted in the section ivt_entry_NAME. The 36796 parameter(s) may be omitted entirely, in which case no interrupt 36797 vector table entry is provided. 36798 36799 Note that interrupts are enabled inside the function unless the 36800 'disinterrupt' attribute is also specified. 36801 36802 The following examples are all valid uses of these attributes on 36803 Epiphany targets: 36804 void __attribute__ ((interrupt)) universal_handler (); 36805 void __attribute__ ((interrupt ("dma1"))) dma1_handler (); 36806 void __attribute__ ((interrupt ("dma0, dma1"))) 36807 universal_dma_handler (); 36808 void __attribute__ ((interrupt ("timer0"), disinterrupt)) 36809 fast_timer_handler (); 36810 void __attribute__ ((interrupt ("dma0, dma1"), 36811 forwarder_section ("tramp"))) 36812 external_dma_handler (); 36813 36814'long_call' 36815'short_call' 36816 These attributes specify how a particular function is called. 36817 These attributes override the '-mlong-calls' (*note Adapteva 36818 Epiphany Options::) command-line switch and '#pragma long_calls' 36819 settings. 36820 36821 36822File: gcc.info, Node: H8/300 Function Attributes, Next: IA-64 Function Attributes, Prev: Epiphany Function Attributes, Up: Function Attributes 36823 368246.33.12 H8/300 Function Attributes 36825---------------------------------- 36826 36827These function attributes are available for H8/300 targets: 36828 36829'function_vector' 36830 Use this attribute on the H8/300, H8/300H, and H8S to indicate that 36831 the specified function should be called through the function 36832 vector. Calling a function through the function vector reduces 36833 code size; however, the function vector has a limited size (maximum 36834 128 entries on the H8/300 and 64 entries on the H8/300H and H8S) 36835 and shares space with the interrupt vector. 36836 36837'interrupt_handler' 36838 Use this attribute on the H8/300, H8/300H, and H8S to indicate that 36839 the specified function is an interrupt handler. The compiler 36840 generates function entry and exit sequences suitable for use in an 36841 interrupt handler when this attribute is present. 36842 36843'saveall' 36844 Use this attribute on the H8/300, H8/300H, and H8S to indicate that 36845 all registers except the stack pointer should be saved in the 36846 prologue regardless of whether they are used or not. 36847 36848 36849File: gcc.info, Node: IA-64 Function Attributes, Next: M32C Function Attributes, Prev: H8/300 Function Attributes, Up: Function Attributes 36850 368516.33.13 IA-64 Function Attributes 36852--------------------------------- 36853 36854These function attributes are supported on IA-64 targets: 36855 36856'syscall_linkage' 36857 This attribute is used to modify the IA-64 calling convention by 36858 marking all input registers as live at all function exits. This 36859 makes it possible to restart a system call after an interrupt 36860 without having to save/restore the input registers. This also 36861 prevents kernel data from leaking into application code. 36862 36863'version_id' 36864 This IA-64 HP-UX attribute, attached to a global variable or 36865 function, renames a symbol to contain a version string, thus 36866 allowing for function level versioning. HP-UX system header files 36867 may use function level versioning for some system calls. 36868 36869 extern int foo () __attribute__((version_id ("20040821"))); 36870 36871 Calls to 'foo' are mapped to calls to 'foo{20040821}'. 36872 36873 36874File: gcc.info, Node: M32C Function Attributes, Next: M32R/D Function Attributes, Prev: IA-64 Function Attributes, Up: Function Attributes 36875 368766.33.14 M32C Function Attributes 36877-------------------------------- 36878 36879These function attributes are supported by the M32C back end: 36880 36881'bank_switch' 36882 When added to an interrupt handler with the M32C port, causes the 36883 prologue and epilogue to use bank switching to preserve the 36884 registers rather than saving them on the stack. 36885 36886'fast_interrupt' 36887 Use this attribute on the M32C port to indicate that the specified 36888 function is a fast interrupt handler. This is just like the 36889 'interrupt' attribute, except that 'freit' is used to return 36890 instead of 'reit'. 36891 36892'function_vector' 36893 On M16C/M32C targets, the 'function_vector' attribute declares a 36894 special page subroutine call function. Use of this attribute 36895 reduces the code size by 2 bytes for each call generated to the 36896 subroutine. The argument to the attribute is the vector number 36897 entry from the special page vector table which contains the 16 36898 low-order bits of the subroutine's entry address. Each vector 36899 table has special page number (18 to 255) that is used in 'jsrs' 36900 instructions. Jump addresses of the routines are generated by 36901 adding 0x0F0000 (in case of M16C targets) or 0xFF0000 (in case of 36902 M32C targets), to the 2-byte addresses set in the vector table. 36903 Therefore you need to ensure that all the special page vector 36904 routines should get mapped within the address range 0x0F0000 to 36905 0x0FFFFF (for M16C) and 0xFF0000 to 0xFFFFFF (for M32C). 36906 36907 In the following example 2 bytes are saved for each call to 36908 function 'foo'. 36909 36910 void foo (void) __attribute__((function_vector(0x18))); 36911 void foo (void) 36912 { 36913 } 36914 36915 void bar (void) 36916 { 36917 foo(); 36918 } 36919 36920 If functions are defined in one file and are called in another 36921 file, then be sure to write this declaration in both files. 36922 36923 This attribute is ignored for R8C target. 36924 36925'interrupt' 36926 Use this attribute to indicate that the specified function is an 36927 interrupt handler. The compiler generates function entry and exit 36928 sequences suitable for use in an interrupt handler when this 36929 attribute is present. 36930 36931 36932File: gcc.info, Node: M32R/D Function Attributes, Next: m68k Function Attributes, Prev: M32C Function Attributes, Up: Function Attributes 36933 369346.33.15 M32R/D Function Attributes 36935---------------------------------- 36936 36937These function attributes are supported by the M32R/D back end: 36938 36939'interrupt' 36940 Use this attribute to indicate that the specified function is an 36941 interrupt handler. The compiler generates function entry and exit 36942 sequences suitable for use in an interrupt handler when this 36943 attribute is present. 36944 36945'model (MODEL-NAME)' 36946 36947 On the M32R/D, use this attribute to set the addressability of an 36948 object, and of the code generated for a function. The identifier 36949 MODEL-NAME is one of 'small', 'medium', or 'large', representing 36950 each of the code models. 36951 36952 Small model objects live in the lower 16MB of memory (so that their 36953 addresses can be loaded with the 'ld24' instruction), and are 36954 callable with the 'bl' instruction. 36955 36956 Medium model objects may live anywhere in the 32-bit address space 36957 (the compiler generates 'seth/add3' instructions to load their 36958 addresses), and are callable with the 'bl' instruction. 36959 36960 Large model objects may live anywhere in the 32-bit address space 36961 (the compiler generates 'seth/add3' instructions to load their 36962 addresses), and may not be reachable with the 'bl' instruction (the 36963 compiler generates the much slower 'seth/add3/jl' instruction 36964 sequence). 36965 36966 36967File: gcc.info, Node: m68k Function Attributes, Next: MCORE Function Attributes, Prev: M32R/D Function Attributes, Up: Function Attributes 36968 369696.33.16 m68k Function Attributes 36970-------------------------------- 36971 36972These function attributes are supported by the m68k back end: 36973 36974'interrupt' 36975'interrupt_handler' 36976 Use this attribute to indicate that the specified function is an 36977 interrupt handler. The compiler generates function entry and exit 36978 sequences suitable for use in an interrupt handler when this 36979 attribute is present. Either name may be used. 36980 36981'interrupt_thread' 36982 Use this attribute on fido, a subarchitecture of the m68k, to 36983 indicate that the specified function is an interrupt handler that 36984 is designed to run as a thread. The compiler omits generate 36985 prologue/epilogue sequences and replaces the return instruction 36986 with a 'sleep' instruction. This attribute is available only on 36987 fido. 36988 36989 36990File: gcc.info, Node: MCORE Function Attributes, Next: MeP Function Attributes, Prev: m68k Function Attributes, Up: Function Attributes 36991 369926.33.17 MCORE Function Attributes 36993--------------------------------- 36994 36995These function attributes are supported by the MCORE back end: 36996 36997'naked' 36998 This attribute allows the compiler to construct the requisite 36999 function declaration, while allowing the body of the function to be 37000 assembly code. The specified function will not have 37001 prologue/epilogue sequences generated by the compiler. Only basic 37002 'asm' statements can safely be included in naked functions (*note 37003 Basic Asm::). While using extended 'asm' or a mixture of basic 37004 'asm' and C code may appear to work, they cannot be depended upon 37005 to work reliably and are not supported. 37006 37007 37008File: gcc.info, Node: MeP Function Attributes, Next: MicroBlaze Function Attributes, Prev: MCORE Function Attributes, Up: Function Attributes 37009 370106.33.18 MeP Function Attributes 37011------------------------------- 37012 37013These function attributes are supported by the MeP back end: 37014 37015'disinterrupt' 37016 On MeP targets, this attribute causes the compiler to emit 37017 instructions to disable interrupts for the duration of the given 37018 function. 37019 37020'interrupt' 37021 Use this attribute to indicate that the specified function is an 37022 interrupt handler. The compiler generates function entry and exit 37023 sequences suitable for use in an interrupt handler when this 37024 attribute is present. 37025 37026'near' 37027 This attribute causes the compiler to assume the called function is 37028 close enough to use the normal calling convention, overriding the 37029 '-mtf' command-line option. 37030 37031'far' 37032 On MeP targets this causes the compiler to use a calling convention 37033 that assumes the called function is too far away for the built-in 37034 addressing modes. 37035 37036'vliw' 37037 The 'vliw' attribute tells the compiler to emit instructions in 37038 VLIW mode instead of core mode. Note that this attribute is not 37039 allowed unless a VLIW coprocessor has been configured and enabled 37040 through command-line options. 37041 37042 37043File: gcc.info, Node: MicroBlaze Function Attributes, Next: Microsoft Windows Function Attributes, Prev: MeP Function Attributes, Up: Function Attributes 37044 370456.33.19 MicroBlaze Function Attributes 37046-------------------------------------- 37047 37048These function attributes are supported on MicroBlaze targets: 37049 37050'save_volatiles' 37051 Use this attribute to indicate that the function is an interrupt 37052 handler. All volatile registers (in addition to non-volatile 37053 registers) are saved in the function prologue. If the function is 37054 a leaf function, only volatiles used by the function are saved. A 37055 normal function return is generated instead of a return from 37056 interrupt. 37057 37058'break_handler' 37059 Use this attribute to indicate that the specified function is a 37060 break handler. The compiler generates function entry and exit 37061 sequences suitable for use in an break handler when this attribute 37062 is present. The return from 'break_handler' is done through the 37063 'rtbd' instead of 'rtsd'. 37064 37065 void f () __attribute__ ((break_handler)); 37066 37067'interrupt_handler' 37068'fast_interrupt' 37069 These attributes indicate that the specified function is an 37070 interrupt handler. Use the 'fast_interrupt' attribute to indicate 37071 handlers used in low-latency interrupt mode, and 37072 'interrupt_handler' for interrupts that do not use low-latency 37073 handlers. In both cases, GCC emits appropriate prologue code and 37074 generates a return from the handler using 'rtid' instead of 'rtsd'. 37075 37076 37077File: gcc.info, Node: Microsoft Windows Function Attributes, Next: MIPS Function Attributes, Prev: MicroBlaze Function Attributes, Up: Function Attributes 37078 370796.33.20 Microsoft Windows Function Attributes 37080--------------------------------------------- 37081 37082The following attributes are available on Microsoft Windows and Symbian 37083OS targets. 37084 37085'dllexport' 37086 On Microsoft Windows targets and Symbian OS targets the 'dllexport' 37087 attribute causes the compiler to provide a global pointer to a 37088 pointer in a DLL, so that it can be referenced with the 'dllimport' 37089 attribute. On Microsoft Windows targets, the pointer name is 37090 formed by combining '_imp__' and the function or variable name. 37091 37092 You can use '__declspec(dllexport)' as a synonym for '__attribute__ 37093 ((dllexport))' for compatibility with other compilers. 37094 37095 On systems that support the 'visibility' attribute, this attribute 37096 also implies "default" visibility. It is an error to explicitly 37097 specify any other visibility. 37098 37099 GCC's default behavior is to emit all inline functions with the 37100 'dllexport' attribute. Since this can cause object file-size 37101 bloat, you can use '-fno-keep-inline-dllexport', which tells GCC to 37102 ignore the attribute for inlined functions unless the 37103 '-fkeep-inline-functions' flag is used instead. 37104 37105 The attribute is ignored for undefined symbols. 37106 37107 When applied to C++ classes, the attribute marks defined 37108 non-inlined member functions and static data members as exports. 37109 Static consts initialized in-class are not marked unless they are 37110 also defined out-of-class. 37111 37112 For Microsoft Windows targets there are alternative methods for 37113 including the symbol in the DLL's export table such as using a 37114 '.def' file with an 'EXPORTS' section or, with GNU ld, using the 37115 '--export-all' linker flag. 37116 37117'dllimport' 37118 On Microsoft Windows and Symbian OS targets, the 'dllimport' 37119 attribute causes the compiler to reference a function or variable 37120 via a global pointer to a pointer that is set up by the DLL 37121 exporting the symbol. The attribute implies 'extern'. On 37122 Microsoft Windows targets, the pointer name is formed by combining 37123 '_imp__' and the function or variable name. 37124 37125 You can use '__declspec(dllimport)' as a synonym for '__attribute__ 37126 ((dllimport))' for compatibility with other compilers. 37127 37128 On systems that support the 'visibility' attribute, this attribute 37129 also implies "default" visibility. It is an error to explicitly 37130 specify any other visibility. 37131 37132 Currently, the attribute is ignored for inlined functions. If the 37133 attribute is applied to a symbol _definition_, an error is 37134 reported. If a symbol previously declared 'dllimport' is later 37135 defined, the attribute is ignored in subsequent references, and a 37136 warning is emitted. The attribute is also overridden by a 37137 subsequent declaration as 'dllexport'. 37138 37139 When applied to C++ classes, the attribute marks non-inlined member 37140 functions and static data members as imports. However, the 37141 attribute is ignored for virtual methods to allow creation of 37142 vtables using thunks. 37143 37144 On the SH Symbian OS target the 'dllimport' attribute also has 37145 another affect--it can cause the vtable and run-time type 37146 information for a class to be exported. This happens when the 37147 class has a dllimported constructor or a non-inline, non-pure 37148 virtual function and, for either of those two conditions, the class 37149 also has an inline constructor or destructor and has a key function 37150 that is defined in the current translation unit. 37151 37152 For Microsoft Windows targets the use of the 'dllimport' attribute 37153 on functions is not necessary, but provides a small performance 37154 benefit by eliminating a thunk in the DLL. The use of the 37155 'dllimport' attribute on imported variables can be avoided by 37156 passing the '--enable-auto-import' switch to the GNU linker. As 37157 with functions, using the attribute for a variable eliminates a 37158 thunk in the DLL. 37159 37160 One drawback to using this attribute is that a pointer to a 37161 _variable_ marked as 'dllimport' cannot be used as a constant 37162 address. However, a pointer to a _function_ with the 'dllimport' 37163 attribute can be used as a constant initializer; in this case, the 37164 address of a stub function in the import lib is referenced. On 37165 Microsoft Windows targets, the attribute can be disabled for 37166 functions by setting the '-mnop-fun-dllimport' flag. 37167 37168 37169File: gcc.info, Node: MIPS Function Attributes, Next: MSP430 Function Attributes, Prev: Microsoft Windows Function Attributes, Up: Function Attributes 37170 371716.33.21 MIPS Function Attributes 37172-------------------------------- 37173 37174These function attributes are supported by the MIPS back end: 37175 37176'interrupt' 37177 Use this attribute to indicate that the specified function is an 37178 interrupt handler. The compiler generates function entry and exit 37179 sequences suitable for use in an interrupt handler when this 37180 attribute is present. An optional argument is supported for the 37181 interrupt attribute which allows the interrupt mode to be 37182 described. By default GCC assumes the external interrupt 37183 controller (EIC) mode is in use, this can be explicitly set using 37184 'eic'. When interrupts are non-masked then the requested Interrupt 37185 Priority Level (IPL) is copied to the current IPL which has the 37186 effect of only enabling higher priority interrupts. To use 37187 vectored interrupt mode use the argument 37188 'vector=[sw0|sw1|hw0|hw1|hw2|hw3|hw4|hw5]', this will change the 37189 behavior of the non-masked interrupt support and GCC will arrange 37190 to mask all interrupts from sw0 up to and including the specified 37191 interrupt vector. 37192 37193 You can use the following attributes to modify the behavior of an 37194 interrupt handler: 37195 'use_shadow_register_set' 37196 Assume that the handler uses a shadow register set, instead of 37197 the main general-purpose registers. An optional argument 37198 'intstack' is supported to indicate that the shadow register 37199 set contains a valid stack pointer. 37200 37201 'keep_interrupts_masked' 37202 Keep interrupts masked for the whole function. Without this 37203 attribute, GCC tries to reenable interrupts for as much of the 37204 function as it can. 37205 37206 'use_debug_exception_return' 37207 Return using the 'deret' instruction. Interrupt handlers that 37208 don't have this attribute return using 'eret' instead. 37209 37210 You can use any combination of these attributes, as shown below: 37211 void __attribute__ ((interrupt)) v0 (); 37212 void __attribute__ ((interrupt, use_shadow_register_set)) v1 (); 37213 void __attribute__ ((interrupt, keep_interrupts_masked)) v2 (); 37214 void __attribute__ ((interrupt, use_debug_exception_return)) v3 (); 37215 void __attribute__ ((interrupt, use_shadow_register_set, 37216 keep_interrupts_masked)) v4 (); 37217 void __attribute__ ((interrupt, use_shadow_register_set, 37218 use_debug_exception_return)) v5 (); 37219 void __attribute__ ((interrupt, keep_interrupts_masked, 37220 use_debug_exception_return)) v6 (); 37221 void __attribute__ ((interrupt, use_shadow_register_set, 37222 keep_interrupts_masked, 37223 use_debug_exception_return)) v7 (); 37224 void __attribute__ ((interrupt("eic"))) v8 (); 37225 void __attribute__ ((interrupt("vector=hw3"))) v9 (); 37226 37227'long_call' 37228'short_call' 37229'near' 37230'far' 37231 These attributes specify how a particular function is called on 37232 MIPS. The attributes override the '-mlong-calls' (*note MIPS 37233 Options::) command-line switch. The 'long_call' and 'far' 37234 attributes are synonyms, and cause the compiler to always call the 37235 function by first loading its address into a register, and then 37236 using the contents of that register. The 'short_call' and 'near' 37237 attributes are synonyms, and have the opposite effect; they specify 37238 that non-PIC calls should be made using the more efficient 'jal' 37239 instruction. 37240 37241'mips16' 37242'nomips16' 37243 37244 On MIPS targets, you can use the 'mips16' and 'nomips16' function 37245 attributes to locally select or turn off MIPS16 code generation. A 37246 function with the 'mips16' attribute is emitted as MIPS16 code, 37247 while MIPS16 code generation is disabled for functions with the 37248 'nomips16' attribute. These attributes override the '-mips16' and 37249 '-mno-mips16' options on the command line (*note MIPS Options::). 37250 37251 When compiling files containing mixed MIPS16 and non-MIPS16 code, 37252 the preprocessor symbol '__mips16' reflects the setting on the 37253 command line, not that within individual functions. Mixed MIPS16 37254 and non-MIPS16 code may interact badly with some GCC extensions 37255 such as '__builtin_apply' (*note Constructing Calls::). 37256 37257'micromips, MIPS' 37258'nomicromips, MIPS' 37259 37260 On MIPS targets, you can use the 'micromips' and 'nomicromips' 37261 function attributes to locally select or turn off microMIPS code 37262 generation. A function with the 'micromips' attribute is emitted 37263 as microMIPS code, while microMIPS code generation is disabled for 37264 functions with the 'nomicromips' attribute. These attributes 37265 override the '-mmicromips' and '-mno-micromips' options on the 37266 command line (*note MIPS Options::). 37267 37268 When compiling files containing mixed microMIPS and non-microMIPS 37269 code, the preprocessor symbol '__mips_micromips' reflects the 37270 setting on the command line, not that within individual functions. 37271 Mixed microMIPS and non-microMIPS code may interact badly with some 37272 GCC extensions such as '__builtin_apply' (*note Constructing 37273 Calls::). 37274 37275'nocompression' 37276 On MIPS targets, you can use the 'nocompression' function attribute 37277 to locally turn off MIPS16 and microMIPS code generation. This 37278 attribute overrides the '-mips16' and '-mmicromips' options on the 37279 command line (*note MIPS Options::). 37280 37281 37282File: gcc.info, Node: MSP430 Function Attributes, Next: NDS32 Function Attributes, Prev: MIPS Function Attributes, Up: Function Attributes 37283 372846.33.22 MSP430 Function Attributes 37285---------------------------------- 37286 37287These function attributes are supported by the MSP430 back end: 37288 37289'critical' 37290 Critical functions disable interrupts upon entry and restore the 37291 previous interrupt state upon exit. Critical functions cannot also 37292 have the 'naked', 'reentrant' or 'interrupt' attributes. 37293 37294 The MSP430 hardware ensures that interrupts are disabled on entry 37295 to 'interrupt' functions, and restores the previous interrupt state 37296 on exit. The 'critical' attribute is therefore redundant on 37297 'interrupt' functions. 37298 37299'interrupt' 37300 Use this attribute to indicate that the specified function is an 37301 interrupt handler. The compiler generates function entry and exit 37302 sequences suitable for use in an interrupt handler when this 37303 attribute is present. 37304 37305 You can provide an argument to the interrupt attribute which 37306 specifies a name or number. If the argument is a number it 37307 indicates the slot in the interrupt vector table (0 - 31) to which 37308 this handler should be assigned. If the argument is a name it is 37309 treated as a symbolic name for the vector slot. These names should 37310 match up with appropriate entries in the linker script. By default 37311 the names 'watchdog' for vector 26, 'nmi' for vector 30 and 'reset' 37312 for vector 31 are recognized. 37313 37314'naked' 37315 This attribute allows the compiler to construct the requisite 37316 function declaration, while allowing the body of the function to be 37317 assembly code. The specified function will not have 37318 prologue/epilogue sequences generated by the compiler. Only basic 37319 'asm' statements can safely be included in naked functions (*note 37320 Basic Asm::). While using extended 'asm' or a mixture of basic 37321 'asm' and C code may appear to work, they cannot be depended upon 37322 to work reliably and are not supported. 37323 37324'reentrant' 37325 Reentrant functions disable interrupts upon entry and enable them 37326 upon exit. Reentrant functions cannot also have the 'naked' or 37327 'critical' attributes. They can have the 'interrupt' attribute. 37328 37329'wakeup' 37330 This attribute only applies to interrupt functions. It is silently 37331 ignored if applied to a non-interrupt function. A wakeup interrupt 37332 function will rouse the processor from any low-power state that it 37333 might be in when the function exits. 37334 37335'lower' 37336'upper' 37337'either' 37338 On the MSP430 target these attributes can be used to specify 37339 whether the function or variable should be placed into low memory, 37340 high memory, or the placement should be left to the linker to 37341 decide. The attributes are only significant if compiling for the 37342 MSP430X architecture in the large memory model. 37343 37344 The attributes work in conjunction with a linker script that has 37345 been augmented to specify where to place sections with a '.lower' 37346 and a '.upper' prefix. So, for example, as well as placing the 37347 '.data' section, the script also specifies the placement of a 37348 '.lower.data' and a '.upper.data' section. The intention is that 37349 'lower' sections are placed into a small but easier to access 37350 memory region and the upper sections are placed into a larger, but 37351 slower to access, region. 37352 37353 The 'either' attribute is special. It tells the linker to place 37354 the object into the corresponding 'lower' section if there is room 37355 for it. If there is insufficient room then the object is placed 37356 into the corresponding 'upper' section instead. Note that the 37357 placement algorithm is not very sophisticated. It does not attempt 37358 to find an optimal packing of the 'lower' sections. It just makes 37359 one pass over the objects and does the best that it can. Using the 37360 '-ffunction-sections' and '-fdata-sections' command-line options 37361 can help the packing, however, since they produce smaller, easier 37362 to pack regions. 37363 37364 37365File: gcc.info, Node: NDS32 Function Attributes, Next: Nios II Function Attributes, Prev: MSP430 Function Attributes, Up: Function Attributes 37366 373676.33.23 NDS32 Function Attributes 37368--------------------------------- 37369 37370These function attributes are supported by the NDS32 back end: 37371 37372'exception' 37373 Use this attribute on the NDS32 target to indicate that the 37374 specified function is an exception handler. The compiler will 37375 generate corresponding sections for use in an exception handler. 37376 37377'interrupt' 37378 On NDS32 target, this attribute indicates that the specified 37379 function is an interrupt handler. The compiler generates 37380 corresponding sections for use in an interrupt handler. You can 37381 use the following attributes to modify the behavior: 37382 'nested' 37383 This interrupt service routine is interruptible. 37384 'not_nested' 37385 This interrupt service routine is not interruptible. 37386 'nested_ready' 37387 This interrupt service routine is interruptible after 37388 'PSW.GIE' (global interrupt enable) is set. This allows 37389 interrupt service routine to finish some short critical code 37390 before enabling interrupts. 37391 'save_all' 37392 The system will help save all registers into stack before 37393 entering interrupt handler. 37394 'partial_save' 37395 The system will help save caller registers into stack before 37396 entering interrupt handler. 37397 37398'naked' 37399 This attribute allows the compiler to construct the requisite 37400 function declaration, while allowing the body of the function to be 37401 assembly code. The specified function will not have 37402 prologue/epilogue sequences generated by the compiler. Only basic 37403 'asm' statements can safely be included in naked functions (*note 37404 Basic Asm::). While using extended 'asm' or a mixture of basic 37405 'asm' and C code may appear to work, they cannot be depended upon 37406 to work reliably and are not supported. 37407 37408'reset' 37409 Use this attribute on the NDS32 target to indicate that the 37410 specified function is a reset handler. The compiler will generate 37411 corresponding sections for use in a reset handler. You can use the 37412 following attributes to provide extra exception handling: 37413 'nmi' 37414 Provide a user-defined function to handle NMI exception. 37415 'warm' 37416 Provide a user-defined function to handle warm reset 37417 exception. 37418 37419 37420File: gcc.info, Node: Nios II Function Attributes, Next: Nvidia PTX Function Attributes, Prev: NDS32 Function Attributes, Up: Function Attributes 37421 374226.33.24 Nios II Function Attributes 37423----------------------------------- 37424 37425These function attributes are supported by the Nios II back end: 37426 37427'target (OPTIONS)' 37428 As discussed in *note Common Function Attributes::, this attribute 37429 allows specification of target-specific compilation options. 37430 37431 When compiling for Nios II, the following options are allowed: 37432 37433 'custom-INSN=N' 37434 'no-custom-INSN' 37435 Each 'custom-INSN=N' attribute locally enables use of a custom 37436 instruction with encoding N when generating code that uses 37437 INSN. Similarly, 'no-custom-INSN' locally inhibits use of the 37438 custom instruction INSN. These target attributes correspond 37439 to the '-mcustom-INSN=N' and '-mno-custom-INSN' command-line 37440 options, and support the same set of INSN keywords. *Note 37441 Nios II Options::, for more information. 37442 37443 'custom-fpu-cfg=NAME' 37444 This attribute corresponds to the '-mcustom-fpu-cfg=NAME' 37445 command-line option, to select a predefined set of custom 37446 instructions named NAME. *Note Nios II Options::, for more 37447 information. 37448 37449 37450File: gcc.info, Node: Nvidia PTX Function Attributes, Next: PowerPC Function Attributes, Prev: Nios II Function Attributes, Up: Function Attributes 37451 374526.33.25 Nvidia PTX Function Attributes 37453-------------------------------------- 37454 37455These function attributes are supported by the Nvidia PTX back end: 37456 37457'kernel' 37458 This attribute indicates that the corresponding function should be 37459 compiled as a kernel function, which can be invoked from the host 37460 via the CUDA RT library. By default functions are only callable 37461 only from other PTX functions. 37462 37463 Kernel functions must have 'void' return type. 37464 37465 37466File: gcc.info, Node: PowerPC Function Attributes, Next: RISC-V Function Attributes, Prev: Nvidia PTX Function Attributes, Up: Function Attributes 37467 374686.33.26 PowerPC Function Attributes 37469----------------------------------- 37470 37471These function attributes are supported by the PowerPC back end: 37472 37473'longcall' 37474'shortcall' 37475 The 'longcall' attribute indicates that the function might be far 37476 away from the call site and require a different (more expensive) 37477 calling sequence. The 'shortcall' attribute indicates that the 37478 function is always close enough for the shorter calling sequence to 37479 be used. These attributes override both the '-mlongcall' switch 37480 and the '#pragma longcall' setting. 37481 37482 *Note RS/6000 and PowerPC Options::, for more information on 37483 whether long calls are necessary. 37484 37485'target (OPTIONS)' 37486 As discussed in *note Common Function Attributes::, this attribute 37487 allows specification of target-specific compilation options. 37488 37489 On the PowerPC, the following options are allowed: 37490 37491 'altivec' 37492 'no-altivec' 37493 Generate code that uses (does not use) AltiVec instructions. 37494 In 32-bit code, you cannot enable AltiVec instructions unless 37495 '-mabi=altivec' is used on the command line. 37496 37497 'cmpb' 37498 'no-cmpb' 37499 Generate code that uses (does not use) the compare bytes 37500 instruction implemented on the POWER6 processor and other 37501 processors that support the PowerPC V2.05 architecture. 37502 37503 'dlmzb' 37504 'no-dlmzb' 37505 Generate code that uses (does not use) the string-search 37506 'dlmzb' instruction on the IBM 405, 440, 464 and 476 37507 processors. This instruction is generated by default when 37508 targeting those processors. 37509 37510 'fprnd' 37511 'no-fprnd' 37512 Generate code that uses (does not use) the FP round to integer 37513 instructions implemented on the POWER5+ processor and other 37514 processors that support the PowerPC V2.03 architecture. 37515 37516 'hard-dfp' 37517 'no-hard-dfp' 37518 Generate code that uses (does not use) the decimal 37519 floating-point instructions implemented on some POWER 37520 processors. 37521 37522 'isel' 37523 'no-isel' 37524 Generate code that uses (does not use) ISEL instruction. 37525 37526 'mfcrf' 37527 'no-mfcrf' 37528 Generate code that uses (does not use) the move from condition 37529 register field instruction implemented on the POWER4 processor 37530 and other processors that support the PowerPC V2.01 37531 architecture. 37532 37533 'mulhw' 37534 'no-mulhw' 37535 Generate code that uses (does not use) the half-word multiply 37536 and multiply-accumulate instructions on the IBM 405, 440, 464 37537 and 476 processors. These instructions are generated by 37538 default when targeting those processors. 37539 37540 'multiple' 37541 'no-multiple' 37542 Generate code that uses (does not use) the load multiple word 37543 instructions and the store multiple word instructions. 37544 37545 'update' 37546 'no-update' 37547 Generate code that uses (does not use) the load or store 37548 instructions that update the base register to the address of 37549 the calculated memory location. 37550 37551 'popcntb' 37552 'no-popcntb' 37553 Generate code that uses (does not use) the popcount and 37554 double-precision FP reciprocal estimate instruction 37555 implemented on the POWER5 processor and other processors that 37556 support the PowerPC V2.02 architecture. 37557 37558 'popcntd' 37559 'no-popcntd' 37560 Generate code that uses (does not use) the popcount 37561 instruction implemented on the POWER7 processor and other 37562 processors that support the PowerPC V2.06 architecture. 37563 37564 'powerpc-gfxopt' 37565 'no-powerpc-gfxopt' 37566 Generate code that uses (does not use) the optional PowerPC 37567 architecture instructions in the Graphics group, including 37568 floating-point select. 37569 37570 'powerpc-gpopt' 37571 'no-powerpc-gpopt' 37572 Generate code that uses (does not use) the optional PowerPC 37573 architecture instructions in the General Purpose group, 37574 including floating-point square root. 37575 37576 'recip-precision' 37577 'no-recip-precision' 37578 Assume (do not assume) that the reciprocal estimate 37579 instructions provide higher-precision estimates than is 37580 mandated by the PowerPC ABI. 37581 37582 'string' 37583 'no-string' 37584 Generate code that uses (does not use) the load string 37585 instructions and the store string word instructions to save 37586 multiple registers and do small block moves. 37587 37588 'vsx' 37589 'no-vsx' 37590 Generate code that uses (does not use) vector/scalar (VSX) 37591 instructions, and also enable the use of built-in functions 37592 that allow more direct access to the VSX instruction set. In 37593 32-bit code, you cannot enable VSX or AltiVec instructions 37594 unless '-mabi=altivec' is used on the command line. 37595 37596 'friz' 37597 'no-friz' 37598 Generate (do not generate) the 'friz' instruction when the 37599 '-funsafe-math-optimizations' option is used to optimize 37600 rounding a floating-point value to 64-bit integer and back to 37601 floating point. The 'friz' instruction does not return the 37602 same value if the floating-point number is too large to fit in 37603 an integer. 37604 37605 'avoid-indexed-addresses' 37606 'no-avoid-indexed-addresses' 37607 Generate code that tries to avoid (not avoid) the use of 37608 indexed load or store instructions. 37609 37610 'paired' 37611 'no-paired' 37612 Generate code that uses (does not use) the generation of 37613 PAIRED simd instructions. 37614 37615 'longcall' 37616 'no-longcall' 37617 Generate code that assumes (does not assume) that all calls 37618 are far away so that a longer more expensive calling sequence 37619 is required. 37620 37621 'cpu=CPU' 37622 Specify the architecture to generate code for when compiling 37623 the function. If you select the 'target("cpu=power7")' 37624 attribute when generating 32-bit code, VSX and AltiVec 37625 instructions are not generated unless you use the 37626 '-mabi=altivec' option on the command line. 37627 37628 'tune=TUNE' 37629 Specify the architecture to tune for when compiling the 37630 function. If you do not specify the 'target("tune=TUNE")' 37631 attribute and you do specify the 'target("cpu=CPU")' 37632 attribute, compilation tunes for the CPU architecture, and not 37633 the default tuning specified on the command line. 37634 37635 On the PowerPC, the inliner does not inline a function that has 37636 different target options than the caller, unless the callee has a 37637 subset of the target options of the caller. 37638 37639 37640File: gcc.info, Node: RISC-V Function Attributes, Next: RL78 Function Attributes, Prev: PowerPC Function Attributes, Up: Function Attributes 37641 376426.33.27 RISC-V Function Attributes 37643---------------------------------- 37644 37645These function attributes are supported by the RISC-V back end: 37646 37647'naked' 37648 This attribute allows the compiler to construct the requisite 37649 function declaration, while allowing the body of the function to be 37650 assembly code. The specified function will not have 37651 prologue/epilogue sequences generated by the compiler. Only basic 37652 'asm' statements can safely be included in naked functions (*note 37653 Basic Asm::). While using extended 'asm' or a mixture of basic 37654 'asm' and C code may appear to work, they cannot be depended upon 37655 to work reliably and are not supported. 37656 37657'interrupt' 37658 Use this attribute to indicate that the specified function is an 37659 interrupt handler. The compiler generates function entry and exit 37660 sequences suitable for use in an interrupt handler when this 37661 attribute is present. 37662 37663 You can specify the kind of interrupt to be handled by adding an 37664 optional parameter to the interrupt attribute like this: 37665 37666 void f (void) __attribute__ ((interrupt ("user"))); 37667 37668 Permissible values for this parameter are 'user', 'supervisor', and 37669 'machine'. If there is no parameter, then it defaults to 37670 'machine'. 37671 37672 37673File: gcc.info, Node: RL78 Function Attributes, Next: RX Function Attributes, Prev: RISC-V Function Attributes, Up: Function Attributes 37674 376756.33.28 RL78 Function Attributes 37676-------------------------------- 37677 37678These function attributes are supported by the RL78 back end: 37679 37680'interrupt' 37681'brk_interrupt' 37682 These attributes indicate that the specified function is an 37683 interrupt handler. The compiler generates function entry and exit 37684 sequences suitable for use in an interrupt handler when this 37685 attribute is present. 37686 37687 Use 'brk_interrupt' instead of 'interrupt' for handlers intended to 37688 be used with the 'BRK' opcode (i.e. those that must end with 'RETB' 37689 instead of 'RETI'). 37690 37691'naked' 37692 This attribute allows the compiler to construct the requisite 37693 function declaration, while allowing the body of the function to be 37694 assembly code. The specified function will not have 37695 prologue/epilogue sequences generated by the compiler. Only basic 37696 'asm' statements can safely be included in naked functions (*note 37697 Basic Asm::). While using extended 'asm' or a mixture of basic 37698 'asm' and C code may appear to work, they cannot be depended upon 37699 to work reliably and are not supported. 37700 37701 37702File: gcc.info, Node: RX Function Attributes, Next: S/390 Function Attributes, Prev: RL78 Function Attributes, Up: Function Attributes 37703 377046.33.29 RX Function Attributes 37705------------------------------ 37706 37707These function attributes are supported by the RX back end: 37708 37709'fast_interrupt' 37710 Use this attribute on the RX port to indicate that the specified 37711 function is a fast interrupt handler. This is just like the 37712 'interrupt' attribute, except that 'freit' is used to return 37713 instead of 'reit'. 37714 37715'interrupt' 37716 Use this attribute to indicate that the specified function is an 37717 interrupt handler. The compiler generates function entry and exit 37718 sequences suitable for use in an interrupt handler when this 37719 attribute is present. 37720 37721 On RX and RL78 targets, you may specify one or more vector numbers 37722 as arguments to the attribute, as well as naming an alternate table 37723 name. Parameters are handled sequentially, so one handler can be 37724 assigned to multiple entries in multiple tables. One may also pass 37725 the magic string '"$default"' which causes the function to be used 37726 for any unfilled slots in the current table. 37727 37728 This example shows a simple assignment of a function to one vector 37729 in the default table (note that preprocessor macros may be used for 37730 chip-specific symbolic vector names): 37731 void __attribute__ ((interrupt (5))) txd1_handler (); 37732 37733 This example assigns a function to two slots in the default table 37734 (using preprocessor macros defined elsewhere) and makes it the 37735 default for the 'dct' table: 37736 void __attribute__ ((interrupt (RXD1_VECT,RXD2_VECT,"dct","$default"))) 37737 txd1_handler (); 37738 37739'naked' 37740 This attribute allows the compiler to construct the requisite 37741 function declaration, while allowing the body of the function to be 37742 assembly code. The specified function will not have 37743 prologue/epilogue sequences generated by the compiler. Only basic 37744 'asm' statements can safely be included in naked functions (*note 37745 Basic Asm::). While using extended 'asm' or a mixture of basic 37746 'asm' and C code may appear to work, they cannot be depended upon 37747 to work reliably and are not supported. 37748 37749'vector' 37750 This RX attribute is similar to the 'interrupt' attribute, 37751 including its parameters, but does not make the function an 37752 interrupt-handler type function (i.e. it retains the normal C 37753 function calling ABI). See the 'interrupt' attribute for a 37754 description of its arguments. 37755 37756 37757File: gcc.info, Node: S/390 Function Attributes, Next: SH Function Attributes, Prev: RX Function Attributes, Up: Function Attributes 37758 377596.33.30 S/390 Function Attributes 37760--------------------------------- 37761 37762These function attributes are supported on the S/390: 37763 37764'hotpatch (HALFWORDS-BEFORE-FUNCTION-LABEL,HALFWORDS-AFTER-FUNCTION-LABEL)' 37765 37766 On S/390 System z targets, you can use this function attribute to 37767 make GCC generate a "hot-patching" function prologue. If the 37768 '-mhotpatch=' command-line option is used at the same time, the 37769 'hotpatch' attribute takes precedence. The first of the two 37770 arguments specifies the number of halfwords to be added before the 37771 function label. A second argument can be used to specify the 37772 number of halfwords to be added after the function label. For both 37773 arguments the maximum allowed value is 1000000. 37774 37775 If both arguments are zero, hotpatching is disabled. 37776 37777'target (OPTIONS)' 37778 As discussed in *note Common Function Attributes::, this attribute 37779 allows specification of target-specific compilation options. 37780 37781 On S/390, the following options are supported: 37782 37783 'arch=' 37784 'tune=' 37785 'stack-guard=' 37786 'stack-size=' 37787 'branch-cost=' 37788 'warn-framesize=' 37789 'backchain' 37790 'no-backchain' 37791 'hard-dfp' 37792 'no-hard-dfp' 37793 'hard-float' 37794 'soft-float' 37795 'htm' 37796 'no-htm' 37797 'vx' 37798 'no-vx' 37799 'packed-stack' 37800 'no-packed-stack' 37801 'small-exec' 37802 'no-small-exec' 37803 'mvcle' 37804 'no-mvcle' 37805 'warn-dynamicstack' 37806 'no-warn-dynamicstack' 37807 37808 The options work exactly like the S/390 specific command line 37809 options (without the prefix '-m') except that they do not change 37810 any feature macros. For example, 37811 37812 target("no-vx") 37813 37814 does not undefine the '__VEC__' macro. 37815 37816 37817File: gcc.info, Node: SH Function Attributes, Next: Symbian OS Function Attributes, Prev: S/390 Function Attributes, Up: Function Attributes 37818 378196.33.31 SH Function Attributes 37820------------------------------ 37821 37822These function attributes are supported on the SH family of processors: 37823 37824'function_vector' 37825 On SH2A targets, this attribute declares a function to be called 37826 using the TBR relative addressing mode. The argument to this 37827 attribute is the entry number of the same function in a vector 37828 table containing all the TBR relative addressable functions. For 37829 correct operation the TBR must be setup accordingly to point to the 37830 start of the vector table before any functions with this attribute 37831 are invoked. Usually a good place to do the initialization is the 37832 startup routine. The TBR relative vector table can have at max 256 37833 function entries. The jumps to these functions are generated using 37834 a SH2A specific, non delayed branch instruction JSR/N @(disp8,TBR). 37835 You must use GAS and GLD from GNU binutils version 2.7 or later for 37836 this attribute to work correctly. 37837 37838 In an application, for a function being called once, this attribute 37839 saves at least 8 bytes of code; and if other successive calls are 37840 being made to the same function, it saves 2 bytes of code per each 37841 of these calls. 37842 37843'interrupt_handler' 37844 Use this attribute to indicate that the specified function is an 37845 interrupt handler. The compiler generates function entry and exit 37846 sequences suitable for use in an interrupt handler when this 37847 attribute is present. 37848 37849'nosave_low_regs' 37850 Use this attribute on SH targets to indicate that an 37851 'interrupt_handler' function should not save and restore registers 37852 R0..R7. This can be used on SH3* and SH4* targets that have a 37853 second R0..R7 register bank for non-reentrant interrupt handlers. 37854 37855'renesas' 37856 On SH targets this attribute specifies that the function or struct 37857 follows the Renesas ABI. 37858 37859'resbank' 37860 On the SH2A target, this attribute enables the high-speed register 37861 saving and restoration using a register bank for 37862 'interrupt_handler' routines. Saving to the bank is performed 37863 automatically after the CPU accepts an interrupt that uses a 37864 register bank. 37865 37866 The nineteen 32-bit registers comprising general register R0 to 37867 R14, control register GBR, and system registers MACH, MACL, and PR 37868 and the vector table address offset are saved into a register bank. 37869 Register banks are stacked in first-in last-out (FILO) sequence. 37870 Restoration from the bank is executed by issuing a RESBANK 37871 instruction. 37872 37873'sp_switch' 37874 Use this attribute on the SH to indicate an 'interrupt_handler' 37875 function should switch to an alternate stack. It expects a string 37876 argument that names a global variable holding the address of the 37877 alternate stack. 37878 37879 void *alt_stack; 37880 void f () __attribute__ ((interrupt_handler, 37881 sp_switch ("alt_stack"))); 37882 37883'trap_exit' 37884 Use this attribute on the SH for an 'interrupt_handler' to return 37885 using 'trapa' instead of 'rte'. This attribute expects an integer 37886 argument specifying the trap number to be used. 37887 37888'trapa_handler' 37889 On SH targets this function attribute is similar to 37890 'interrupt_handler' but it does not save and restore all registers. 37891 37892 37893File: gcc.info, Node: Symbian OS Function Attributes, Next: V850 Function Attributes, Prev: SH Function Attributes, Up: Function Attributes 37894 378956.33.32 Symbian OS Function Attributes 37896-------------------------------------- 37897 37898*Note Microsoft Windows Function Attributes::, for discussion of the 37899'dllexport' and 'dllimport' attributes. 37900 37901 37902File: gcc.info, Node: V850 Function Attributes, Next: Visium Function Attributes, Prev: Symbian OS Function Attributes, Up: Function Attributes 37903 379046.33.33 V850 Function Attributes 37905-------------------------------- 37906 37907The V850 back end supports these function attributes: 37908 37909'interrupt' 37910'interrupt_handler' 37911 Use these attributes to indicate that the specified function is an 37912 interrupt handler. The compiler generates function entry and exit 37913 sequences suitable for use in an interrupt handler when either 37914 attribute is present. 37915 37916 37917File: gcc.info, Node: Visium Function Attributes, Next: x86 Function Attributes, Prev: V850 Function Attributes, Up: Function Attributes 37918 379196.33.34 Visium Function Attributes 37920---------------------------------- 37921 37922These function attributes are supported by the Visium back end: 37923 37924'interrupt' 37925 Use this attribute to indicate that the specified function is an 37926 interrupt handler. The compiler generates function entry and exit 37927 sequences suitable for use in an interrupt handler when this 37928 attribute is present. 37929 37930 37931File: gcc.info, Node: x86 Function Attributes, Next: Xstormy16 Function Attributes, Prev: Visium Function Attributes, Up: Function Attributes 37932 379336.33.35 x86 Function Attributes 37934------------------------------- 37935 37936These function attributes are supported by the x86 back end: 37937 37938'cdecl' 37939 On the x86-32 targets, the 'cdecl' attribute causes the compiler to 37940 assume that the calling function pops off the stack space used to 37941 pass arguments. This is useful to override the effects of the 37942 '-mrtd' switch. 37943 37944'fastcall' 37945 On x86-32 targets, the 'fastcall' attribute causes the compiler to 37946 pass the first argument (if of integral type) in the register ECX 37947 and the second argument (if of integral type) in the register EDX. 37948 Subsequent and other typed arguments are passed on the stack. The 37949 called function pops the arguments off the stack. If the number of 37950 arguments is variable all arguments are pushed on the stack. 37951 37952'thiscall' 37953 On x86-32 targets, the 'thiscall' attribute causes the compiler to 37954 pass the first argument (if of integral type) in the register ECX. 37955 Subsequent and other typed arguments are passed on the stack. The 37956 called function pops the arguments off the stack. If the number of 37957 arguments is variable all arguments are pushed on the stack. The 37958 'thiscall' attribute is intended for C++ non-static member 37959 functions. As a GCC extension, this calling convention can be used 37960 for C functions and for static member methods. 37961 37962'ms_abi' 37963'sysv_abi' 37964 37965 On 32-bit and 64-bit x86 targets, you can use an ABI attribute to 37966 indicate which calling convention should be used for a function. 37967 The 'ms_abi' attribute tells the compiler to use the Microsoft ABI, 37968 while the 'sysv_abi' attribute tells the compiler to use the System 37969 V ELF ABI, which is used on GNU/Linux and other systems. The 37970 default is to use the Microsoft ABI when targeting Windows. On all 37971 other systems, the default is the System V ELF ABI. 37972 37973 Note, the 'ms_abi' attribute for Microsoft Windows 64-bit targets 37974 currently requires the '-maccumulate-outgoing-args' option. 37975 37976'callee_pop_aggregate_return (NUMBER)' 37977 37978 On x86-32 targets, you can use this attribute to control how 37979 aggregates are returned in memory. If the caller is responsible 37980 for popping the hidden pointer together with the rest of the 37981 arguments, specify NUMBER equal to zero. If callee is responsible 37982 for popping the hidden pointer, specify NUMBER equal to one. 37983 37984 The default x86-32 ABI assumes that the callee pops the stack for 37985 hidden pointer. However, on x86-32 Microsoft Windows targets, the 37986 compiler assumes that the caller pops the stack for hidden pointer. 37987 37988'ms_hook_prologue' 37989 37990 On 32-bit and 64-bit x86 targets, you can use this function 37991 attribute to make GCC generate the "hot-patching" function prologue 37992 used in Win32 API functions in Microsoft Windows XP Service Pack 2 37993 and newer. 37994 37995'naked' 37996 This attribute allows the compiler to construct the requisite 37997 function declaration, while allowing the body of the function to be 37998 assembly code. The specified function will not have 37999 prologue/epilogue sequences generated by the compiler. Only basic 38000 'asm' statements can safely be included in naked functions (*note 38001 Basic Asm::). While using extended 'asm' or a mixture of basic 38002 'asm' and C code may appear to work, they cannot be depended upon 38003 to work reliably and are not supported. 38004 38005'regparm (NUMBER)' 38006 On x86-32 targets, the 'regparm' attribute causes the compiler to 38007 pass arguments number one to NUMBER if they are of integral type in 38008 registers EAX, EDX, and ECX instead of on the stack. Functions 38009 that take a variable number of arguments continue to be passed all 38010 of their arguments on the stack. 38011 38012 Beware that on some ELF systems this attribute is unsuitable for 38013 global functions in shared libraries with lazy binding (which is 38014 the default). Lazy binding sends the first call via resolving code 38015 in the loader, which might assume EAX, EDX and ECX can be 38016 clobbered, as per the standard calling conventions. Solaris 8 is 38017 affected by this. Systems with the GNU C Library version 2.1 or 38018 higher and FreeBSD are believed to be safe since the loaders there 38019 save EAX, EDX and ECX. (Lazy binding can be disabled with the 38020 linker or the loader if desired, to avoid the problem.) 38021 38022'sseregparm' 38023 On x86-32 targets with SSE support, the 'sseregparm' attribute 38024 causes the compiler to pass up to 3 floating-point arguments in SSE 38025 registers instead of on the stack. Functions that take a variable 38026 number of arguments continue to pass all of their floating-point 38027 arguments on the stack. 38028 38029'force_align_arg_pointer' 38030 On x86 targets, the 'force_align_arg_pointer' attribute may be 38031 applied to individual function definitions, generating an alternate 38032 prologue and epilogue that realigns the run-time stack if 38033 necessary. This supports mixing legacy codes that run with a 38034 4-byte aligned stack with modern codes that keep a 16-byte stack 38035 for SSE compatibility. 38036 38037'stdcall' 38038 On x86-32 targets, the 'stdcall' attribute causes the compiler to 38039 assume that the called function pops off the stack space used to 38040 pass arguments, unless it takes a variable number of arguments. 38041 38042'no_caller_saved_registers' 38043 Use this attribute to indicate that the specified function has no 38044 caller-saved registers. That is, all registers are callee-saved. 38045 For example, this attribute can be used for a function called from 38046 an interrupt handler. The compiler generates proper function entry 38047 and exit sequences to save and restore any modified registers, 38048 except for the EFLAGS register. Since GCC doesn't preserve SSE, 38049 MMX nor x87 states, the GCC option '-mgeneral-regs-only' should be 38050 used to compile functions with 'no_caller_saved_registers' 38051 attribute. 38052 38053'interrupt' 38054 Use this attribute to indicate that the specified function is an 38055 interrupt handler or an exception handler (depending on parameters 38056 passed to the function, explained further). The compiler generates 38057 function entry and exit sequences suitable for use in an interrupt 38058 handler when this attribute is present. The 'IRET' instruction, 38059 instead of the 'RET' instruction, is used to return from interrupt 38060 handlers. All registers, except for the EFLAGS register which is 38061 restored by the 'IRET' instruction, are preserved by the compiler. 38062 Since GCC doesn't preserve SSE, MMX nor x87 states, the GCC option 38063 '-mgeneral-regs-only' should be used to compile interrupt and 38064 exception handlers. 38065 38066 Any interruptible-without-stack-switch code must be compiled with 38067 '-mno-red-zone' since interrupt handlers can and will, because of 38068 the hardware design, touch the red zone. 38069 38070 An interrupt handler must be declared with a mandatory pointer 38071 argument: 38072 38073 struct interrupt_frame; 38074 38075 __attribute__ ((interrupt)) 38076 void 38077 f (struct interrupt_frame *frame) 38078 { 38079 } 38080 38081 and you must define 'struct interrupt_frame' as described in the 38082 processor's manual. 38083 38084 Exception handlers differ from interrupt handlers because the 38085 system pushes an error code on the stack. An exception handler 38086 declaration is similar to that for an interrupt handler, but with a 38087 different mandatory function signature. The compiler arranges to 38088 pop the error code off the stack before the 'IRET' instruction. 38089 38090 #ifdef __x86_64__ 38091 typedef unsigned long long int uword_t; 38092 #else 38093 typedef unsigned int uword_t; 38094 #endif 38095 38096 struct interrupt_frame; 38097 38098 __attribute__ ((interrupt)) 38099 void 38100 f (struct interrupt_frame *frame, uword_t error_code) 38101 { 38102 ... 38103 } 38104 38105 Exception handlers should only be used for exceptions that push an 38106 error code; you should use an interrupt handler in other cases. 38107 The system will crash if the wrong kind of handler is used. 38108 38109'target (OPTIONS)' 38110 As discussed in *note Common Function Attributes::, this attribute 38111 allows specification of target-specific compilation options. 38112 38113 On the x86, the following options are allowed: 38114 '3dnow' 38115 'no-3dnow' 38116 Enable/disable the generation of the 3DNow! instructions. 38117 38118 '3dnowa' 38119 'no-3dnowa' 38120 Enable/disable the generation of the enhanced 3DNow! 38121 instructions. 38122 38123 'abm' 38124 'no-abm' 38125 Enable/disable the generation of the advanced bit 38126 instructions. 38127 38128 'adx' 38129 'no-adx' 38130 Enable/disable the generation of the ADX instructions. 38131 38132 'aes' 38133 'no-aes' 38134 Enable/disable the generation of the AES instructions. 38135 38136 'avx' 38137 'no-avx' 38138 Enable/disable the generation of the AVX instructions. 38139 38140 'avx2' 38141 'no-avx2' 38142 Enable/disable the generation of the AVX2 instructions. 38143 38144 'avx5124fmaps' 38145 'no-avx5124fmaps' 38146 Enable/disable the generation of the AVX5124FMAPS 38147 instructions. 38148 38149 'avx5124vnniw' 38150 'no-avx5124vnniw' 38151 Enable/disable the generation of the AVX5124VNNIW 38152 instructions. 38153 38154 'avx512bitalg' 38155 'no-avx512bitalg' 38156 Enable/disable the generation of the AVX512BITALG 38157 instructions. 38158 38159 'avx512bw' 38160 'no-avx512bw' 38161 Enable/disable the generation of the AVX512BW instructions. 38162 38163 'avx512cd' 38164 'no-avx512cd' 38165 Enable/disable the generation of the AVX512CD instructions. 38166 38167 'avx512dq' 38168 'no-avx512dq' 38169 Enable/disable the generation of the AVX512DQ instructions. 38170 38171 'avx512er' 38172 'no-avx512er' 38173 Enable/disable the generation of the AVX512ER instructions. 38174 38175 'avx512f' 38176 'no-avx512f' 38177 Enable/disable the generation of the AVX512F instructions. 38178 38179 'avx512ifma' 38180 'no-avx512ifma' 38181 Enable/disable the generation of the AVX512IFMA instructions. 38182 38183 'avx512pf' 38184 'no-avx512pf' 38185 Enable/disable the generation of the AVX512PF instructions. 38186 38187 'avx512vbmi' 38188 'no-avx512vbmi' 38189 Enable/disable the generation of the AVX512VBMI instructions. 38190 38191 'avx512vbmi2' 38192 'no-avx512vbmi2' 38193 Enable/disable the generation of the AVX512VBMI2 instructions. 38194 38195 'avx512vl' 38196 'no-avx512vl' 38197 Enable/disable the generation of the AVX512VL instructions. 38198 38199 'avx512vnni' 38200 'no-avx512vnni' 38201 Enable/disable the generation of the AVX512VNNI instructions. 38202 38203 'avx512vpopcntdq' 38204 'no-avx512vpopcntdq' 38205 Enable/disable the generation of the AVX512VPOPCNTDQ 38206 instructions. 38207 38208 'bmi' 38209 'no-bmi' 38210 Enable/disable the generation of the BMI instructions. 38211 38212 'bmi2' 38213 'no-bmi2' 38214 Enable/disable the generation of the BMI2 instructions. 38215 38216 'cldemote' 38217 'no-cldemote' 38218 Enable/disable the generation of the CLDEMOTE instructions. 38219 38220 'clflushopt' 38221 'no-clflushopt' 38222 Enable/disable the generation of the CLFLUSHOPT instructions. 38223 38224 'clwb' 38225 'no-clwb' 38226 Enable/disable the generation of the CLWB instructions. 38227 38228 'clzero' 38229 'no-clzero' 38230 Enable/disable the generation of the CLZERO instructions. 38231 38232 'crc32' 38233 'no-crc32' 38234 Enable/disable the generation of the CRC32 instructions. 38235 38236 'cx16' 38237 'no-cx16' 38238 Enable/disable the generation of the CMPXCHG16B instructions. 38239 38240 'default' 38241 *Note Function Multiversioning::, where it is used to specify 38242 the default function version. 38243 38244 'f16c' 38245 'no-f16c' 38246 Enable/disable the generation of the F16C instructions. 38247 38248 'fma' 38249 'no-fma' 38250 Enable/disable the generation of the FMA instructions. 38251 38252 'fma4' 38253 'no-fma4' 38254 Enable/disable the generation of the FMA4 instructions. 38255 38256 'fsgsbase' 38257 'no-fsgsbase' 38258 Enable/disable the generation of the FSGSBASE instructions. 38259 38260 'fxsr' 38261 'no-fxsr' 38262 Enable/disable the generation of the FXSR instructions. 38263 38264 'gfni' 38265 'no-gfni' 38266 Enable/disable the generation of the GFNI instructions. 38267 38268 'hle' 38269 'no-hle' 38270 Enable/disable the generation of the HLE instruction prefixes. 38271 38272 'lwp' 38273 'no-lwp' 38274 Enable/disable the generation of the LWP instructions. 38275 38276 'lzcnt' 38277 'no-lzcnt' 38278 Enable/disable the generation of the LZCNT instructions. 38279 38280 'mmx' 38281 'no-mmx' 38282 Enable/disable the generation of the MMX instructions. 38283 38284 'movbe' 38285 'no-movbe' 38286 Enable/disable the generation of the MOVBE instructions. 38287 38288 'movdir64b' 38289 'no-movdir64b' 38290 Enable/disable the generation of the MOVDIR64B instructions. 38291 38292 'movdiri' 38293 'no-movdiri' 38294 Enable/disable the generation of the MOVDIRI instructions. 38295 38296 'mwait' 38297 'no-mwait' 38298 Enable/disable the generation of the MWAIT and MONITOR 38299 instructions. 38300 38301 'mwaitx' 38302 'no-mwaitx' 38303 Enable/disable the generation of the MWAITX instructions. 38304 38305 'pclmul' 38306 'no-pclmul' 38307 Enable/disable the generation of the PCLMUL instructions. 38308 38309 'pconfig' 38310 'no-pconfig' 38311 Enable/disable the generation of the PCONFIG instructions. 38312 38313 'pku' 38314 'no-pku' 38315 Enable/disable the generation of the PKU instructions. 38316 38317 'popcnt' 38318 'no-popcnt' 38319 Enable/disable the generation of the POPCNT instruction. 38320 38321 'prefetchwt1' 38322 'no-prefetchwt1' 38323 Enable/disable the generation of the PREFETCHWT1 instructions. 38324 38325 'prfchw' 38326 'no-prfchw' 38327 Enable/disable the generation of the PREFETCHW instruction. 38328 38329 'ptwrite' 38330 'no-ptwrite' 38331 Enable/disable the generation of the PTWRITE instructions. 38332 38333 'rdpid' 38334 'no-rdpid' 38335 Enable/disable the generation of the RDPID instructions. 38336 38337 'rdrnd' 38338 'no-rdrnd' 38339 Enable/disable the generation of the RDRND instructions. 38340 38341 'rdseed' 38342 'no-rdseed' 38343 Enable/disable the generation of the RDSEED instructions. 38344 38345 'rtm' 38346 'no-rtm' 38347 Enable/disable the generation of the RTM instructions. 38348 38349 'sahf' 38350 'no-sahf' 38351 Enable/disable the generation of the SAHF instructions. 38352 38353 'sgx' 38354 'no-sgx' 38355 Enable/disable the generation of the SGX instructions. 38356 38357 'sha' 38358 'no-sha' 38359 Enable/disable the generation of the SHA instructions. 38360 38361 'shstk' 38362 'no-shstk' 38363 Enable/disable the shadow stack built-in functions from CET. 38364 38365 'sse' 38366 'no-sse' 38367 Enable/disable the generation of the SSE instructions. 38368 38369 'sse2' 38370 'no-sse2' 38371 Enable/disable the generation of the SSE2 instructions. 38372 38373 'sse3' 38374 'no-sse3' 38375 Enable/disable the generation of the SSE3 instructions. 38376 38377 'sse4' 38378 'no-sse4' 38379 Enable/disable the generation of the SSE4 instructions (both 38380 SSE4.1 and SSE4.2). 38381 38382 'sse4.1' 38383 'no-sse4.1' 38384 Enable/disable the generation of the SSE4.1 instructions. 38385 38386 'sse4.2' 38387 'no-sse4.2' 38388 Enable/disable the generation of the SSE4.2 instructions. 38389 38390 'sse4a' 38391 'no-sse4a' 38392 Enable/disable the generation of the SSE4A instructions. 38393 38394 'ssse3' 38395 'no-ssse3' 38396 Enable/disable the generation of the SSSE3 instructions. 38397 38398 'tbm' 38399 'no-tbm' 38400 Enable/disable the generation of the TBM instructions. 38401 38402 'vaes' 38403 'no-vaes' 38404 Enable/disable the generation of the VAES instructions. 38405 38406 'vpclmulqdq' 38407 'no-vpclmulqdq' 38408 Enable/disable the generation of the VPCLMULQDQ instructions. 38409 38410 'waitpkg' 38411 'no-waitpkg' 38412 Enable/disable the generation of the WAITPKG instructions. 38413 38414 'wbnoinvd' 38415 'no-wbnoinvd' 38416 Enable/disable the generation of the WBNOINVD instructions. 38417 38418 'xop' 38419 'no-xop' 38420 Enable/disable the generation of the XOP instructions. 38421 38422 'xsave' 38423 'no-xsave' 38424 Enable/disable the generation of the XSAVE instructions. 38425 38426 'xsavec' 38427 'no-xsavec' 38428 Enable/disable the generation of the XSAVEC instructions. 38429 38430 'xsaveopt' 38431 'no-xsaveopt' 38432 Enable/disable the generation of the XSAVEOPT instructions. 38433 38434 'xsaves' 38435 'no-xsaves' 38436 Enable/disable the generation of the XSAVES instructions. 38437 38438 'amx-tile' 38439 'no-amx-tile' 38440 Enable/disable the generation of the AMX-TILE instructions. 38441 38442 'amx-int8' 38443 'no-amx-int8' 38444 Enable/disable the generation of the AMX-INT8 instructions. 38445 38446 'amx-bf16' 38447 'no-amx-bf16' 38448 Enable/disable the generation of the AMX-BF16 instructions. 38449 38450 'uintr' 38451 'no-uintr' 38452 Enable/disable the generation of the UINTR instructions. 38453 38454 'hreset' 38455 'no-hreset' 38456 Enable/disable the generation of the HRESET instruction. 38457 38458 'kl' 38459 'no-kl' 38460 Enable/disable the generation of the KEYLOCKER instructions. 38461 38462 'widekl' 38463 'no-widekl' 38464 Enable/disable the generation of the WIDEKL instructions. 38465 38466 'avxvnni' 38467 'no-avxvnni' 38468 Enable/disable the generation of the AVXVNNI instructions. 38469 38470 'cld' 38471 'no-cld' 38472 Enable/disable the generation of the CLD before string moves. 38473 38474 'fancy-math-387' 38475 'no-fancy-math-387' 38476 Enable/disable the generation of the 'sin', 'cos', and 'sqrt' 38477 instructions on the 387 floating-point unit. 38478 38479 'ieee-fp' 38480 'no-ieee-fp' 38481 Enable/disable the generation of floating point that depends 38482 on IEEE arithmetic. 38483 38484 'inline-all-stringops' 38485 'no-inline-all-stringops' 38486 Enable/disable inlining of string operations. 38487 38488 'inline-stringops-dynamically' 38489 'no-inline-stringops-dynamically' 38490 Enable/disable the generation of the inline code to do small 38491 string operations and calling the library routines for large 38492 operations. 38493 38494 'align-stringops' 38495 'no-align-stringops' 38496 Do/do not align destination of inlined string operations. 38497 38498 'recip' 38499 'no-recip' 38500 Enable/disable the generation of RCPSS, RCPPS, RSQRTSS and 38501 RSQRTPS instructions followed an additional Newton-Raphson 38502 step instead of doing a floating-point division. 38503 38504 'general-regs-only' 38505 Generate code which uses only the general registers. 38506 38507 'arch=ARCH' 38508 Specify the architecture to generate code for in compiling the 38509 function. 38510 38511 'tune=TUNE' 38512 Specify the architecture to tune for in compiling the 38513 function. 38514 38515 'fpmath=FPMATH' 38516 Specify which floating-point unit to use. You must specify 38517 the 'target("fpmath=sse,387")' option as 38518 'target("fpmath=sse+387")' because the comma would separate 38519 different options. 38520 38521 'prefer-vector-width=OPT' 38522 On x86 targets, the 'prefer-vector-width' attribute informs 38523 the compiler to use OPT-bit vector width in instructions 38524 instead of the default on the selected platform. 38525 38526 Valid OPT values are: 38527 38528 'none' 38529 No extra limitations applied to GCC other than defined by 38530 the selected platform. 38531 38532 '128' 38533 Prefer 128-bit vector width for instructions. 38534 38535 '256' 38536 Prefer 256-bit vector width for instructions. 38537 38538 '512' 38539 Prefer 512-bit vector width for instructions. 38540 38541 On the x86, the inliner does not inline a function that has 38542 different target options than the caller, unless the callee 38543 has a subset of the target options of the caller. For example 38544 a function declared with 'target("sse3")' can inline a 38545 function with 'target("sse2")', since '-msse3' implies 38546 '-msse2'. 38547 38548'indirect_branch("CHOICE")' 38549 On x86 targets, the 'indirect_branch' attribute causes the compiler 38550 to convert indirect call and jump with CHOICE. 'keep' keeps 38551 indirect call and jump unmodified. 'thunk' converts indirect call 38552 and jump to call and return thunk. 'thunk-inline' converts 38553 indirect call and jump to inlined call and return thunk. 38554 'thunk-extern' converts indirect call and jump to external call and 38555 return thunk provided in a separate object file. 38556 38557'function_return("CHOICE")' 38558 On x86 targets, the 'function_return' attribute causes the compiler 38559 to convert function return with CHOICE. 'keep' keeps function 38560 return unmodified. 'thunk' converts function return to call and 38561 return thunk. 'thunk-inline' converts function return to inlined 38562 call and return thunk. 'thunk-extern' converts function return to 38563 external call and return thunk provided in a separate object file. 38564 38565'nocf_check' 38566 The 'nocf_check' attribute on a function is used to inform the 38567 compiler that the function's prologue should not be instrumented 38568 when compiled with the '-fcf-protection=branch' option. The 38569 compiler assumes that the function's address is a valid target for 38570 a control-flow transfer. 38571 38572 The 'nocf_check' attribute on a type of pointer to function is used 38573 to inform the compiler that a call through the pointer should not 38574 be instrumented when compiled with the '-fcf-protection=branch' 38575 option. The compiler assumes that the function's address from the 38576 pointer is a valid target for a control-flow transfer. A direct 38577 function call through a function name is assumed to be a safe call 38578 thus direct calls are not instrumented by the compiler. 38579 38580 The 'nocf_check' attribute is applied to an object's type. In case 38581 of assignment of a function address or a function pointer to 38582 another pointer, the attribute is not carried over from the 38583 right-hand object's type; the type of left-hand object stays 38584 unchanged. The compiler checks for 'nocf_check' attribute mismatch 38585 and reports a warning in case of mismatch. 38586 38587 { 38588 int foo (void) __attribute__(nocf_check); 38589 void (*foo1)(void) __attribute__(nocf_check); 38590 void (*foo2)(void); 38591 38592 /* foo's address is assumed to be valid. */ 38593 int 38594 foo (void) 38595 38596 /* This call site is not checked for control-flow 38597 validity. */ 38598 (*foo1)(); 38599 38600 /* A warning is issued about attribute mismatch. */ 38601 foo1 = foo2; 38602 38603 /* This call site is still not checked. */ 38604 (*foo1)(); 38605 38606 /* This call site is checked. */ 38607 (*foo2)(); 38608 38609 /* A warning is issued about attribute mismatch. */ 38610 foo2 = foo1; 38611 38612 /* This call site is still checked. */ 38613 (*foo2)(); 38614 38615 return 0; 38616 } 38617 38618'cf_check' 38619 38620 The 'cf_check' attribute on a function is used to inform the 38621 compiler that ENDBR instruction should be placed at the function 38622 entry when '-fcf-protection=branch' is enabled. 38623 38624'indirect_return' 38625 38626 The 'indirect_return' attribute can be applied to a function, as 38627 well as variable or type of function pointer to inform the compiler 38628 that the function may return via indirect branch. 38629 38630'fentry_name("NAME")' 38631 On x86 targets, the 'fentry_name' attribute sets the function to 38632 call on function entry when function instrumentation is enabled 38633 with '-pg -mfentry'. When NAME is nop then a 5 byte nop sequence 38634 is generated. 38635 38636'fentry_section("NAME")' 38637 On x86 targets, the 'fentry_section' attribute sets the name of the 38638 section to record function entry instrumentation calls in when 38639 enabled with '-pg -mrecord-mcount' 38640 38641'nodirect_extern_access' 38642 This attribute, attached to a global variable or function, is the 38643 counterpart to option '-mno-direct-extern-access'. 38644 38645 38646File: gcc.info, Node: Xstormy16 Function Attributes, Prev: x86 Function Attributes, Up: Function Attributes 38647 386486.33.36 Xstormy16 Function Attributes 38649------------------------------------- 38650 38651These function attributes are supported by the Xstormy16 back end: 38652 38653'interrupt' 38654 Use this attribute to indicate that the specified function is an 38655 interrupt handler. The compiler generates function entry and exit 38656 sequences suitable for use in an interrupt handler when this 38657 attribute is present. 38658 38659 38660File: gcc.info, Node: Variable Attributes, Next: Type Attributes, Prev: Function Attributes, Up: C Extensions 38661 386626.34 Specifying Attributes of Variables 38663======================================= 38664 38665The keyword '__attribute__' allows you to specify special properties of 38666variables, function parameters, or structure, union, and, in C++, class 38667members. This '__attribute__' keyword is followed by an attribute 38668specification enclosed in double parentheses. Some attributes are 38669currently defined generically for variables. Other attributes are 38670defined for variables on particular target systems. Other attributes 38671are available for functions (*note Function Attributes::), labels (*note 38672Label Attributes::), enumerators (*note Enumerator Attributes::), 38673statements (*note Statement Attributes::), and for types (*note Type 38674Attributes::). Other front ends might define more attributes (*note 38675Extensions to the C++ Language: C++ Extensions.). 38676 38677 *Note Attribute Syntax::, for details of the exact syntax for using 38678attributes. 38679 38680* Menu: 38681 38682* Common Variable Attributes:: 38683* ARC Variable Attributes:: 38684* AVR Variable Attributes:: 38685* Blackfin Variable Attributes:: 38686* H8/300 Variable Attributes:: 38687* IA-64 Variable Attributes:: 38688* M32R/D Variable Attributes:: 38689* MeP Variable Attributes:: 38690* Microsoft Windows Variable Attributes:: 38691* MSP430 Variable Attributes:: 38692* Nvidia PTX Variable Attributes:: 38693* PowerPC Variable Attributes:: 38694* RL78 Variable Attributes:: 38695* V850 Variable Attributes:: 38696* x86 Variable Attributes:: 38697* Xstormy16 Variable Attributes:: 38698 38699 38700File: gcc.info, Node: Common Variable Attributes, Next: ARC Variable Attributes, Up: Variable Attributes 38701 387026.34.1 Common Variable Attributes 38703--------------------------------- 38704 38705The following attributes are supported on most targets. 38706 38707'alias ("TARGET")' 38708 The 'alias' variable attribute causes the declaration to be emitted 38709 as an alias for another symbol known as an "alias target". Except 38710 for top-level qualifiers the alias target must have the same type 38711 as the alias. For instance, the following 38712 38713 int var_target; 38714 extern int __attribute__ ((alias ("var_target"))) var_alias; 38715 38716 defines 'var_alias' to be an alias for the 'var_target' variable. 38717 38718 It is an error if the alias target is not defined in the same 38719 translation unit as the alias. 38720 38721 Note that in the absence of the attribute GCC assumes that distinct 38722 declarations with external linkage denote distinct objects. Using 38723 both the alias and the alias target to access the same object is 38724 undefined in a translation unit without a declaration of the alias 38725 with the attribute. 38726 38727 This attribute requires assembler and object file support, and may 38728 not be available on all targets. 38729 38730'aligned' 38731'aligned (ALIGNMENT)' 38732 The 'aligned' attribute specifies a minimum alignment for the 38733 variable or structure field, measured in bytes. When specified, 38734 ALIGNMENT must be an integer constant power of 2. Specifying no 38735 ALIGNMENT argument implies the maximum alignment for the target, 38736 which is often, but by no means always, 8 or 16 bytes. 38737 38738 For example, the declaration: 38739 38740 int x __attribute__ ((aligned (16))) = 0; 38741 38742 causes the compiler to allocate the global variable 'x' on a 38743 16-byte boundary. On a 68040, this could be used in conjunction 38744 with an 'asm' expression to access the 'move16' instruction which 38745 requires 16-byte aligned operands. 38746 38747 You can also specify the alignment of structure fields. For 38748 example, to create a double-word aligned 'int' pair, you could 38749 write: 38750 38751 struct foo { int x[2] __attribute__ ((aligned (8))); }; 38752 38753 This is an alternative to creating a union with a 'double' member, 38754 which forces the union to be double-word aligned. 38755 38756 As in the preceding examples, you can explicitly specify the 38757 alignment (in bytes) that you wish the compiler to use for a given 38758 variable or structure field. Alternatively, you can leave out the 38759 alignment factor and just ask the compiler to align a variable or 38760 field to the default alignment for the target architecture you are 38761 compiling for. The default alignment is sufficient for all scalar 38762 types, but may not be enough for all vector types on a target that 38763 supports vector operations. The default alignment is fixed for a 38764 particular target ABI. 38765 38766 GCC also provides a target specific macro '__BIGGEST_ALIGNMENT__', 38767 which is the largest alignment ever used for any data type on the 38768 target machine you are compiling for. For example, you could 38769 write: 38770 38771 short array[3] __attribute__ ((aligned (__BIGGEST_ALIGNMENT__))); 38772 38773 The compiler automatically sets the alignment for the declared 38774 variable or field to '__BIGGEST_ALIGNMENT__'. Doing this can often 38775 make copy operations more efficient, because the compiler can use 38776 whatever instructions copy the biggest chunks of memory when 38777 performing copies to or from the variables or fields that you have 38778 aligned this way. Note that the value of '__BIGGEST_ALIGNMENT__' 38779 may change depending on command-line options. 38780 38781 When used on a struct, or struct member, the 'aligned' attribute 38782 can only increase the alignment; in order to decrease it, the 38783 'packed' attribute must be specified as well. When used as part of 38784 a typedef, the 'aligned' attribute can both increase and decrease 38785 alignment, and specifying the 'packed' attribute generates a 38786 warning. 38787 38788 Note that the effectiveness of 'aligned' attributes for static 38789 variables may be limited by inherent limitations in the system 38790 linker and/or object file format. On some systems, the linker is 38791 only able to arrange for variables to be aligned up to a certain 38792 maximum alignment. (For some linkers, the maximum supported 38793 alignment may be very very small.) If your linker is only able to 38794 align variables up to a maximum of 8-byte alignment, then 38795 specifying 'aligned(16)' in an '__attribute__' still only provides 38796 you with 8-byte alignment. See your linker documentation for 38797 further information. 38798 38799 Stack variables are not affected by linker restrictions; GCC can 38800 properly align them on any target. 38801 38802 The 'aligned' attribute can also be used for functions (*note 38803 Common Function Attributes::.) 38804 38805'warn_if_not_aligned (ALIGNMENT)' 38806 This attribute specifies a threshold for the structure field, 38807 measured in bytes. If the structure field is aligned below the 38808 threshold, a warning will be issued. For example, the declaration: 38809 38810 struct foo 38811 { 38812 int i1; 38813 int i2; 38814 unsigned long long x __attribute__ ((warn_if_not_aligned (16))); 38815 }; 38816 38817 causes the compiler to issue an warning on 'struct foo', like 38818 'warning: alignment 8 of 'struct foo' is less than 16'. The 38819 compiler also issues a warning, like 'warning: 'x' offset 8 in 38820 'struct foo' isn't aligned to 16', when the structure field has the 38821 misaligned offset: 38822 38823 struct __attribute__ ((aligned (16))) foo 38824 { 38825 int i1; 38826 int i2; 38827 unsigned long long x __attribute__ ((warn_if_not_aligned (16))); 38828 }; 38829 38830 This warning can be disabled by '-Wno-if-not-aligned'. The 38831 'warn_if_not_aligned' attribute can also be used for types (*note 38832 Common Type Attributes::.) 38833 38834'alloc_size (POSITION)' 38835'alloc_size (POSITION-1, POSITION-2)' 38836 The 'alloc_size' variable attribute may be applied to the 38837 declaration of a pointer to a function that returns a pointer and 38838 takes at least one argument of an integer type. It indicates that 38839 the returned pointer points to an object whose size is given by the 38840 function argument at POSITION, or by the product of the arguments 38841 at POSITION-1 and POSITION-2. Meaningful sizes are positive values 38842 less than 'PTRDIFF_MAX'. Other sizes are diagnosed when detected. 38843 GCC uses this information to improve the results of 38844 '__builtin_object_size'. 38845 38846 For instance, the following declarations 38847 38848 typedef __attribute__ ((alloc_size (1, 2))) void* 38849 (*calloc_ptr) (size_t, size_t); 38850 typedef __attribute__ ((alloc_size (1))) void* 38851 (*malloc_ptr) (size_t); 38852 38853 specify that 'calloc_ptr' is a pointer of a function that, like the 38854 standard C function 'calloc', returns an object whose size is given 38855 by the product of arguments 1 and 2, and similarly, that 38856 'malloc_ptr', like the standard C function 'malloc', returns an 38857 object whose size is given by argument 1 to the function. 38858 38859'cleanup (CLEANUP_FUNCTION)' 38860 The 'cleanup' attribute runs a function when the variable goes out 38861 of scope. This attribute can only be applied to auto function 38862 scope variables; it may not be applied to parameters or variables 38863 with static storage duration. The function must take one 38864 parameter, a pointer to a type compatible with the variable. The 38865 return value of the function (if any) is ignored. 38866 38867 If '-fexceptions' is enabled, then CLEANUP_FUNCTION is run during 38868 the stack unwinding that happens during the processing of the 38869 exception. Note that the 'cleanup' attribute does not allow the 38870 exception to be caught, only to perform an action. It is undefined 38871 what happens if CLEANUP_FUNCTION does not return normally. 38872 38873'common' 38874'nocommon' 38875 The 'common' attribute requests GCC to place a variable in "common" 38876 storage. The 'nocommon' attribute requests the opposite--to 38877 allocate space for it directly. 38878 38879 These attributes override the default chosen by the '-fno-common' 38880 and '-fcommon' flags respectively. 38881 38882'copy' 38883'copy (VARIABLE)' 38884 The 'copy' attribute applies the set of attributes with which 38885 VARIABLE has been declared to the declaration of the variable to 38886 which the attribute is applied. The attribute is designed for 38887 libraries that define aliases that are expected to specify the same 38888 set of attributes as the aliased symbols. The 'copy' attribute can 38889 be used with variables, functions or types. However, the kind of 38890 symbol to which the attribute is applied (either varible or 38891 function) must match the kind of symbol to which the argument 38892 refers. The 'copy' attribute copies only syntactic and semantic 38893 attributes but not attributes that affect a symbol's linkage or 38894 visibility such as 'alias', 'visibility', or 'weak'. The 38895 'deprecated' attribute is also not copied. *Note Common Function 38896 Attributes::. *Note Common Type Attributes::. 38897 38898'deprecated' 38899'deprecated (MSG)' 38900 The 'deprecated' attribute results in a warning if the variable is 38901 used anywhere in the source file. This is useful when identifying 38902 variables that are expected to be removed in a future version of a 38903 program. The warning also includes the location of the declaration 38904 of the deprecated variable, to enable users to easily find further 38905 information about why the variable is deprecated, or what they 38906 should do instead. Note that the warning only occurs for uses: 38907 38908 extern int old_var __attribute__ ((deprecated)); 38909 extern int old_var; 38910 int new_fn () { return old_var; } 38911 38912 results in a warning on line 3 but not line 2. The optional MSG 38913 argument, which must be a string, is printed in the warning if 38914 present. 38915 38916 The 'deprecated' attribute can also be used for functions and types 38917 (*note Common Function Attributes::, *note Common Type 38918 Attributes::). 38919 38920 The message attached to the attribute is affected by the setting of 38921 the '-fmessage-length' option. 38922 38923'unavailable' 38924'unavailable (MSG)' 38925 The 'unavailable' attribute indicates that the variable so marked 38926 is not available, if it is used anywhere in the source file. It 38927 behaves in the same manner as the 'deprecated' attribute except 38928 that the compiler will emit an error rather than a warning. 38929 38930 It is expected that items marked as 'deprecated' will eventually be 38931 withdrawn from interfaces, and then become unavailable. This 38932 attribute allows for marking them appropriately. 38933 38934 The 'unavailable' attribute can also be used for functions and 38935 types (*note Common Function Attributes::, *note Common Type 38936 Attributes::). 38937 38938'mode (MODE)' 38939 This attribute specifies the data type for the 38940 declaration--whichever type corresponds to the mode MODE. This in 38941 effect lets you request an integer or floating-point type according 38942 to its width. 38943 38944 *Note (gccint)Machine Modes::, for a list of the possible keywords 38945 for MODE. You may also specify a mode of 'byte' or '__byte__' to 38946 indicate the mode corresponding to a one-byte integer, 'word' or 38947 '__word__' for the mode of a one-word integer, and 'pointer' or 38948 '__pointer__' for the mode used to represent pointers. 38949 38950'nonstring' 38951 The 'nonstring' variable attribute specifies that an object or 38952 member declaration with type array of 'char', 'signed char', or 38953 'unsigned char', or pointer to such a type is intended to store 38954 character arrays that do not necessarily contain a terminating 38955 'NUL'. This is useful in detecting uses of such arrays or pointers 38956 with functions that expect 'NUL'-terminated strings, and to avoid 38957 warnings when such an array or pointer is used as an argument to a 38958 bounded string manipulation function such as 'strncpy'. For 38959 example, without the attribute, GCC will issue a warning for the 38960 'strncpy' call below because it may truncate the copy without 38961 appending the terminating 'NUL' character. Using the attribute 38962 makes it possible to suppress the warning. However, when the array 38963 is declared with the attribute the call to 'strlen' is diagnosed 38964 because when the array doesn't contain a 'NUL'-terminated string 38965 the call is undefined. To copy, compare, of search non-string 38966 character arrays use the 'memcpy', 'memcmp', 'memchr', and other 38967 functions that operate on arrays of bytes. In addition, calling 38968 'strnlen' and 'strndup' with such arrays is safe provided a 38969 suitable bound is specified, and not diagnosed. 38970 38971 struct Data 38972 { 38973 char name [32] __attribute__ ((nonstring)); 38974 }; 38975 38976 int f (struct Data *pd, const char *s) 38977 { 38978 strncpy (pd->name, s, sizeof pd->name); 38979 ... 38980 return strlen (pd->name); // unsafe, gets a warning 38981 } 38982 38983'packed' 38984 The 'packed' attribute specifies that a structure member should 38985 have the smallest possible alignment--one bit for a bit-field and 38986 one byte otherwise, unless a larger value is specified with the 38987 'aligned' attribute. The attribute does not apply to non-member 38988 objects. 38989 38990 For example in the structure below, the member array 'x' is packed 38991 so that it immediately follows 'a' with no intervening padding: 38992 38993 struct foo 38994 { 38995 char a; 38996 int x[2] __attribute__ ((packed)); 38997 }; 38998 38999 _Note:_ The 4.1, 4.2 and 4.3 series of GCC ignore the 'packed' 39000 attribute on bit-fields of type 'char'. This has been fixed in GCC 39001 4.4 but the change can lead to differences in the structure layout. 39002 See the documentation of '-Wpacked-bitfield-compat' for more 39003 information. 39004 39005'section ("SECTION-NAME")' 39006 Normally, the compiler places the objects it generates in sections 39007 like 'data' and 'bss'. Sometimes, however, you need additional 39008 sections, or you need certain particular variables to appear in 39009 special sections, for example to map to special hardware. The 39010 'section' attribute specifies that a variable (or function) lives 39011 in a particular section. For example, this small program uses 39012 several specific section names: 39013 39014 struct duart a __attribute__ ((section ("DUART_A"))) = { 0 }; 39015 struct duart b __attribute__ ((section ("DUART_B"))) = { 0 }; 39016 char stack[10000] __attribute__ ((section ("STACK"))) = { 0 }; 39017 int init_data __attribute__ ((section ("INITDATA"))); 39018 39019 main() 39020 { 39021 /* Initialize stack pointer */ 39022 init_sp (stack + sizeof (stack)); 39023 39024 /* Initialize initialized data */ 39025 memcpy (&init_data, &data, &edata - &data); 39026 39027 /* Turn on the serial ports */ 39028 init_duart (&a); 39029 init_duart (&b); 39030 } 39031 39032 Use the 'section' attribute with _global_ variables and not _local_ 39033 variables, as shown in the example. 39034 39035 You may use the 'section' attribute with initialized or 39036 uninitialized global variables but the linker requires each object 39037 be defined once, with the exception that uninitialized variables 39038 tentatively go in the 'common' (or 'bss') section and can be 39039 multiply "defined". Using the 'section' attribute changes what 39040 section the variable goes into and may cause the linker to issue an 39041 error if an uninitialized variable has multiple definitions. You 39042 can force a variable to be initialized with the '-fno-common' flag 39043 or the 'nocommon' attribute. 39044 39045 Some file formats do not support arbitrary sections so the 39046 'section' attribute is not available on all platforms. If you need 39047 to map the entire contents of a module to a particular section, 39048 consider using the facilities of the linker instead. 39049 39050'tls_model ("TLS_MODEL")' 39051 The 'tls_model' attribute sets thread-local storage model (*note 39052 Thread-Local::) of a particular '__thread' variable, overriding 39053 '-ftls-model=' command-line switch on a per-variable basis. The 39054 TLS_MODEL argument should be one of 'global-dynamic', 39055 'local-dynamic', 'initial-exec' or 'local-exec'. 39056 39057 Not all targets support this attribute. 39058 39059'unused' 39060 This attribute, attached to a variable or structure field, means 39061 that the variable or field is meant to be possibly unused. GCC 39062 does not produce a warning for this variable or field. 39063 39064'used' 39065 This attribute, attached to a variable with static storage, means 39066 that the variable must be emitted even if it appears that the 39067 variable is not referenced. 39068 39069 When applied to a static data member of a C++ class template, the 39070 attribute also means that the member is instantiated if the class 39071 itself is instantiated. 39072 39073'retain' 39074 For ELF targets that support the GNU or FreeBSD OSABIs, this 39075 attribute will save the variable from linker garbage collection. 39076 To support this behavior, variables that have not been placed in 39077 specific sections (e.g. by the 'section' attribute, or the 39078 '-fdata-sections' option), will be placed in new, unique sections. 39079 39080 This additional functionality requires Binutils version 2.36 or 39081 later. 39082 39083'uninitialized' 39084 This attribute, attached to a variable with automatic storage, 39085 means that the variable should not be automatically initialized by 39086 the compiler when the option '-ftrivial-auto-var-init' presents. 39087 39088 With the option '-ftrivial-auto-var-init', all the automatic 39089 variables that do not have explicit initializers will be 39090 initialized by the compiler. These additional compiler 39091 initializations might incur run-time overhead, sometimes 39092 dramatically. This attribute can be used to mark some variables to 39093 be excluded from such automatical initialization in order to reduce 39094 runtime overhead. 39095 39096 This attribute has no effect when the option 39097 '-ftrivial-auto-var-init' does not present. 39098 39099'vector_size (BYTES)' 39100 This attribute specifies the vector size for the type of the 39101 declared variable, measured in bytes. The type to which it applies 39102 is known as the "base type". The BYTES argument must be a positive 39103 power-of-two multiple of the base type size. For example, the 39104 declaration: 39105 39106 int foo __attribute__ ((vector_size (16))); 39107 39108 causes the compiler to set the mode for 'foo', to be 16 bytes, 39109 divided into 'int' sized units. Assuming a 32-bit 'int', 'foo''s 39110 type is a vector of four units of four bytes each, and the 39111 corresponding mode of 'foo' is 'V4SI'. *Note Vector Extensions::, 39112 for details of manipulating vector variables. 39113 39114 This attribute is only applicable to integral and floating scalars, 39115 although arrays, pointers, and function return values are allowed 39116 in conjunction with this construct. 39117 39118 Aggregates with this attribute are invalid, even if they are of the 39119 same size as a corresponding scalar. For example, the declaration: 39120 39121 struct S { int a; }; 39122 struct S __attribute__ ((vector_size (16))) foo; 39123 39124 is invalid even if the size of the structure is the same as the 39125 size of the 'int'. 39126 39127'visibility ("VISIBILITY_TYPE")' 39128 This attribute affects the linkage of the declaration to which it 39129 is attached. The 'visibility' attribute is described in *note 39130 Common Function Attributes::. 39131 39132'weak' 39133 The 'weak' attribute is described in *note Common Function 39134 Attributes::. 39135 39136'noinit' 39137 Any data with the 'noinit' attribute will not be initialized by the 39138 C runtime startup code, or the program loader. Not initializing 39139 data in this way can reduce program startup times. 39140 39141 This attribute is specific to ELF targets and relies on the linker 39142 script to place sections with the '.noinit' prefix in the right 39143 location. 39144 39145'persistent' 39146 Any data with the 'persistent' attribute will not be initialized by 39147 the C runtime startup code, but will be initialized by the program 39148 loader. This enables the value of the variable to 'persist' 39149 between processor resets. 39150 39151 This attribute is specific to ELF targets and relies on the linker 39152 script to place the sections with the '.persistent' prefix in the 39153 right location. Specifically, some type of non-volatile, writeable 39154 memory is required. 39155 39156'objc_nullability (NULLABILITY KIND) (Objective-C and Objective-C++ only)' 39157 This attribute applies to pointer variables only. It allows 39158 marking the pointer with one of four possible values describing the 39159 conditions under which the pointer might have a 'nil' value. In 39160 most cases, the attribute is intended to be an internal 39161 representation for property and method nullability (specified by 39162 language keywords); it is not recommended to use it directly. 39163 39164 When NULLABILITY KIND is '"unspecified"' or '0', nothing is known 39165 about the conditions in which the pointer might be 'nil'. Making 39166 this state specific serves to avoid false positives in diagnostics. 39167 39168 When NULLABILITY KIND is '"nonnull"' or '1', the pointer has no 39169 meaning if it is 'nil' and thus the compiler is free to emit 39170 diagnostics if it can be determined that the value will be 'nil'. 39171 39172 When NULLABILITY KIND is '"nullable"' or '2', the pointer might be 39173 'nil' and carry meaning as such. 39174 39175 When NULLABILITY KIND is '"resettable"' or '3' (used only in the 39176 context of property attribute lists) this describes the case in 39177 which a property setter may take the value 'nil' (which perhaps 39178 causes the property to be reset in some manner to a default) but 39179 for which the property getter will never validly return 'nil'. 39180 39181 39182File: gcc.info, Node: ARC Variable Attributes, Next: AVR Variable Attributes, Prev: Common Variable Attributes, Up: Variable Attributes 39183 391846.34.2 ARC Variable Attributes 39185------------------------------ 39186 39187'aux' 39188 The 'aux' attribute is used to directly access the ARC's auxiliary 39189 register space from C. The auxilirary register number is given via 39190 attribute argument. 39191 39192 39193File: gcc.info, Node: AVR Variable Attributes, Next: Blackfin Variable Attributes, Prev: ARC Variable Attributes, Up: Variable Attributes 39194 391956.34.3 AVR Variable Attributes 39196------------------------------ 39197 39198'progmem' 39199 The 'progmem' attribute is used on the AVR to place read-only data 39200 in the non-volatile program memory (flash). The 'progmem' 39201 attribute accomplishes this by putting respective variables into a 39202 section whose name starts with '.progmem'. 39203 39204 This attribute works similar to the 'section' attribute but adds 39205 additional checking. 39206 39207 * Ordinary AVR cores with 32 general purpose registers: 39208 'progmem' affects the location of the data but not how this 39209 data is accessed. In order to read data located with the 39210 'progmem' attribute (inline) assembler must be used. 39211 /* Use custom macros from AVR-LibC (http://nongnu.org/avr-libc/user-manual/) */ 39212 #include <avr/pgmspace.h> 39213 39214 /* Locate var in flash memory */ 39215 const int var[2] PROGMEM = { 1, 2 }; 39216 39217 int read_var (int i) 39218 { 39219 /* Access var[] by accessor macro from avr/pgmspace.h */ 39220 return (int) pgm_read_word (& var[i]); 39221 } 39222 39223 AVR is a Harvard architecture processor and data and read-only 39224 data normally resides in the data memory (RAM). 39225 39226 See also the *note AVR Named Address Spaces:: section for an 39227 alternate way to locate and access data in flash memory. 39228 39229 * AVR cores with flash memory visible in the RAM address range: 39230 On such devices, there is no need for attribute 'progmem' or 39231 *note '__flash': AVR Named Address Spaces. qualifier at all. 39232 Just use standard C / C++. The compiler will generate 'LD*' 39233 instructions. As flash memory is visible in the RAM address 39234 range, and the default linker script does _not_ locate 39235 '.rodata' in RAM, no special features are needed in order not 39236 to waste RAM for read-only data or to read from flash. You 39237 might even get slightly better performance by avoiding 39238 'progmem' and '__flash'. This applies to devices from 39239 families 'avrtiny' and 'avrxmega3', see *note AVR Options:: 39240 for an overview. 39241 39242 * Reduced AVR Tiny cores like ATtiny40: 39243 The compiler adds '0x4000' to the addresses of objects and 39244 declarations in 'progmem' and locates the objects in flash 39245 memory, namely in section '.progmem.data'. The offset is 39246 needed because the flash memory is visible in the RAM address 39247 space starting at address '0x4000'. 39248 39249 Data in 'progmem' can be accessed by means of ordinary C code, 39250 no special functions or macros are needed. 39251 39252 /* var is located in flash memory */ 39253 extern const int var[2] __attribute__((progmem)); 39254 39255 int read_var (int i) 39256 { 39257 return var[i]; 39258 } 39259 39260 Please notice that on these devices, there is no need for 39261 'progmem' at all. 39262 39263'io' 39264'io (ADDR)' 39265 Variables with the 'io' attribute are used to address memory-mapped 39266 peripherals in the io address range. If an address is specified, 39267 the variable is assigned that address, and the value is interpreted 39268 as an address in the data address space. Example: 39269 39270 volatile int porta __attribute__((io (0x22))); 39271 39272 The address specified in the address in the data address range. 39273 39274 Otherwise, the variable it is not assigned an address, but the 39275 compiler will still use in/out instructions where applicable, 39276 assuming some other module assigns an address in the io address 39277 range. Example: 39278 39279 extern volatile int porta __attribute__((io)); 39280 39281'io_low' 39282'io_low (ADDR)' 39283 This is like the 'io' attribute, but additionally it informs the 39284 compiler that the object lies in the lower half of the I/O area, 39285 allowing the use of 'cbi', 'sbi', 'sbic' and 'sbis' instructions. 39286 39287'address' 39288'address (ADDR)' 39289 Variables with the 'address' attribute are used to address 39290 memory-mapped peripherals that may lie outside the io address 39291 range. 39292 39293 volatile int porta __attribute__((address (0x600))); 39294 39295'absdata' 39296 Variables in static storage and with the 'absdata' attribute can be 39297 accessed by the 'LDS' and 'STS' instructions which take absolute 39298 addresses. 39299 39300 * This attribute is only supported for the reduced AVR Tiny core 39301 like ATtiny40. 39302 39303 * You must make sure that respective data is located in the 39304 address range '0x40'...'0xbf' accessible by 'LDS' and 'STS'. 39305 One way to achieve this as an appropriate linker description 39306 file. 39307 39308 * If the location does not fit the address range of 'LDS' and 39309 'STS', there is currently (Binutils 2.26) just an unspecific 39310 warning like 39311 'module.cc:(.text+0x1c): warning: internal error: out of 39312 range error' 39313 39314 See also the '-mabsdata' *note command-line option: AVR Options. 39315 39316 39317File: gcc.info, Node: Blackfin Variable Attributes, Next: H8/300 Variable Attributes, Prev: AVR Variable Attributes, Up: Variable Attributes 39318 393196.34.4 Blackfin Variable Attributes 39320----------------------------------- 39321 39322Three attributes are currently defined for the Blackfin. 39323 39324'l1_data' 39325'l1_data_A' 39326'l1_data_B' 39327 Use these attributes on the Blackfin to place the variable into L1 39328 Data SRAM. Variables with 'l1_data' attribute are put into the 39329 specific section named '.l1.data'. Those with 'l1_data_A' 39330 attribute are put into the specific section named '.l1.data.A'. 39331 Those with 'l1_data_B' attribute are put into the specific section 39332 named '.l1.data.B'. 39333 39334'l2' 39335 Use this attribute on the Blackfin to place the variable into L2 39336 SRAM. Variables with 'l2' attribute are put into the specific 39337 section named '.l2.data'. 39338 39339 39340File: gcc.info, Node: H8/300 Variable Attributes, Next: IA-64 Variable Attributes, Prev: Blackfin Variable Attributes, Up: Variable Attributes 39341 393426.34.5 H8/300 Variable Attributes 39343--------------------------------- 39344 39345These variable attributes are available for H8/300 targets: 39346 39347'eightbit_data' 39348 Use this attribute on the H8/300, H8/300H, and H8S to indicate that 39349 the specified variable should be placed into the eight-bit data 39350 section. The compiler generates more efficient code for certain 39351 operations on data in the eight-bit data area. Note the eight-bit 39352 data area is limited to 256 bytes of data. 39353 39354 You must use GAS and GLD from GNU binutils version 2.7 or later for 39355 this attribute to work correctly. 39356 39357'tiny_data' 39358 Use this attribute on the H8/300H and H8S to indicate that the 39359 specified variable should be placed into the tiny data section. 39360 The compiler generates more efficient code for loads and stores on 39361 data in the tiny data section. Note the tiny data area is limited 39362 to slightly under 32KB of data. 39363 39364 39365File: gcc.info, Node: IA-64 Variable Attributes, Next: M32R/D Variable Attributes, Prev: H8/300 Variable Attributes, Up: Variable Attributes 39366 393676.34.6 IA-64 Variable Attributes 39368-------------------------------- 39369 39370The IA-64 back end supports the following variable attribute: 39371 39372'model (MODEL-NAME)' 39373 39374 On IA-64, use this attribute to set the addressability of an 39375 object. At present, the only supported identifier for MODEL-NAME 39376 is 'small', indicating addressability via "small" (22-bit) 39377 addresses (so that their addresses can be loaded with the 'addl' 39378 instruction). Caveat: such addressing is by definition not 39379 position independent and hence this attribute must not be used for 39380 objects defined by shared libraries. 39381 39382 39383File: gcc.info, Node: M32R/D Variable Attributes, Next: MeP Variable Attributes, Prev: IA-64 Variable Attributes, Up: Variable Attributes 39384 393856.34.7 M32R/D Variable Attributes 39386--------------------------------- 39387 39388One attribute is currently defined for the M32R/D. 39389 39390'model (MODEL-NAME)' 39391 Use this attribute on the M32R/D to set the addressability of an 39392 object. The identifier MODEL-NAME is one of 'small', 'medium', or 39393 'large', representing each of the code models. 39394 39395 Small model objects live in the lower 16MB of memory (so that their 39396 addresses can be loaded with the 'ld24' instruction). 39397 39398 Medium and large model objects may live anywhere in the 32-bit 39399 address space (the compiler generates 'seth/add3' instructions to 39400 load their addresses). 39401 39402 39403File: gcc.info, Node: MeP Variable Attributes, Next: Microsoft Windows Variable Attributes, Prev: M32R/D Variable Attributes, Up: Variable Attributes 39404 394056.34.8 MeP Variable Attributes 39406------------------------------ 39407 39408The MeP target has a number of addressing modes and busses. The 'near' 39409space spans the standard memory space's first 16 megabytes (24 bits). 39410The 'far' space spans the entire 32-bit memory space. The 'based' space 39411is a 128-byte region in the memory space that is addressed relative to 39412the '$tp' register. The 'tiny' space is a 65536-byte region relative to 39413the '$gp' register. In addition to these memory regions, the MeP target 39414has a separate 16-bit control bus which is specified with 'cb' 39415attributes. 39416 39417'based' 39418 Any variable with the 'based' attribute is assigned to the '.based' 39419 section, and is accessed with relative to the '$tp' register. 39420 39421'tiny' 39422 Likewise, the 'tiny' attribute assigned variables to the '.tiny' 39423 section, relative to the '$gp' register. 39424 39425'near' 39426 Variables with the 'near' attribute are assumed to have addresses 39427 that fit in a 24-bit addressing mode. This is the default for 39428 large variables ('-mtiny=4' is the default) but this attribute can 39429 override '-mtiny=' for small variables, or override '-ml'. 39430 39431'far' 39432 Variables with the 'far' attribute are addressed using a full 39433 32-bit address. Since this covers the entire memory space, this 39434 allows modules to make no assumptions about where variables might 39435 be stored. 39436 39437'io' 39438'io (ADDR)' 39439 Variables with the 'io' attribute are used to address memory-mapped 39440 peripherals. If an address is specified, the variable is assigned 39441 that address, else it is not assigned an address (it is assumed 39442 some other module assigns an address). Example: 39443 39444 int timer_count __attribute__((io(0x123))); 39445 39446'cb' 39447'cb (ADDR)' 39448 Variables with the 'cb' attribute are used to access the control 39449 bus, using special instructions. 'addr' indicates the control bus 39450 address. Example: 39451 39452 int cpu_clock __attribute__((cb(0x123))); 39453 39454 39455File: gcc.info, Node: Microsoft Windows Variable Attributes, Next: MSP430 Variable Attributes, Prev: MeP Variable Attributes, Up: Variable Attributes 39456 394576.34.9 Microsoft Windows Variable Attributes 39458-------------------------------------------- 39459 39460You can use these attributes on Microsoft Windows targets. *note x86 39461Variable Attributes:: for additional Windows compatibility attributes 39462available on all x86 targets. 39463 39464'dllimport' 39465'dllexport' 39466 The 'dllimport' and 'dllexport' attributes are described in *note 39467 Microsoft Windows Function Attributes::. 39468 39469'selectany' 39470 The 'selectany' attribute causes an initialized global variable to 39471 have link-once semantics. When multiple definitions of the 39472 variable are encountered by the linker, the first is selected and 39473 the remainder are discarded. Following usage by the Microsoft 39474 compiler, the linker is told _not_ to warn about size or content 39475 differences of the multiple definitions. 39476 39477 Although the primary usage of this attribute is for POD types, the 39478 attribute can also be applied to global C++ objects that are 39479 initialized by a constructor. In this case, the static 39480 initialization and destruction code for the object is emitted in 39481 each translation defining the object, but the calls to the 39482 constructor and destructor are protected by a link-once guard 39483 variable. 39484 39485 The 'selectany' attribute is only available on Microsoft Windows 39486 targets. You can use '__declspec (selectany)' as a synonym for 39487 '__attribute__ ((selectany))' for compatibility with other 39488 compilers. 39489 39490'shared' 39491 On Microsoft Windows, in addition to putting variable definitions 39492 in a named section, the section can also be shared among all 39493 running copies of an executable or DLL. For example, this small 39494 program defines shared data by putting it in a named section 39495 'shared' and marking the section shareable: 39496 39497 int foo __attribute__((section ("shared"), shared)) = 0; 39498 39499 int 39500 main() 39501 { 39502 /* Read and write foo. All running 39503 copies see the same value. */ 39504 return 0; 39505 } 39506 39507 You may only use the 'shared' attribute along with 'section' 39508 attribute with a fully-initialized global definition because of the 39509 way linkers work. See 'section' attribute for more information. 39510 39511 The 'shared' attribute is only available on Microsoft Windows. 39512 39513 39514File: gcc.info, Node: MSP430 Variable Attributes, Next: Nvidia PTX Variable Attributes, Prev: Microsoft Windows Variable Attributes, Up: Variable Attributes 39515 395166.34.10 MSP430 Variable Attributes 39517---------------------------------- 39518 39519'upper' 39520'either' 39521 These attributes are the same as the MSP430 function attributes of 39522 the same name (*note MSP430 Function Attributes::). 39523 39524'lower' 39525 This option behaves mostly the same as the MSP430 function 39526 attribute of the same name (*note MSP430 Function Attributes::), 39527 but it has some additional functionality. 39528 39529 If '-mdata-region='{'upper,either,none'} has been passed, or the 39530 'section' attribute is applied to a variable, the compiler will 39531 generate 430X instructions to handle it. This is because the 39532 compiler has to assume that the variable could get placed in the 39533 upper memory region (above address 0xFFFF). Marking the variable 39534 with the 'lower' attribute informs the compiler that the variable 39535 will be placed in lower memory so it is safe to use 430 39536 instructions to handle it. 39537 39538 In the case of the 'section' attribute, the section name given will 39539 be used, and the '.lower' prefix will not be added. 39540 39541 39542File: gcc.info, Node: Nvidia PTX Variable Attributes, Next: PowerPC Variable Attributes, Prev: MSP430 Variable Attributes, Up: Variable Attributes 39543 395446.34.11 Nvidia PTX Variable Attributes 39545-------------------------------------- 39546 39547These variable attributes are supported by the Nvidia PTX back end: 39548 39549'shared' 39550 Use this attribute to place a variable in the '.shared' memory 39551 space. This memory space is private to each cooperative thread 39552 array; only threads within one thread block refer to the same 39553 instance of the variable. The runtime does not initialize 39554 variables in this memory space. 39555 39556 39557File: gcc.info, Node: PowerPC Variable Attributes, Next: RL78 Variable Attributes, Prev: Nvidia PTX Variable Attributes, Up: Variable Attributes 39558 395596.34.12 PowerPC Variable Attributes 39560----------------------------------- 39561 39562Three attributes currently are defined for PowerPC configurations: 39563'altivec', 'ms_struct' and 'gcc_struct'. 39564 39565 For full documentation of the struct attributes please see the 39566documentation in *note x86 Variable Attributes::. 39567 39568 For documentation of 'altivec' attribute please see the documentation 39569in *note PowerPC Type Attributes::. 39570 39571 39572File: gcc.info, Node: RL78 Variable Attributes, Next: V850 Variable Attributes, Prev: PowerPC Variable Attributes, Up: Variable Attributes 39573 395746.34.13 RL78 Variable Attributes 39575-------------------------------- 39576 39577The RL78 back end supports the 'saddr' variable attribute. This 39578specifies placement of the corresponding variable in the SADDR area, 39579which can be accessed more efficiently than the default memory region. 39580 39581 39582File: gcc.info, Node: V850 Variable Attributes, Next: x86 Variable Attributes, Prev: RL78 Variable Attributes, Up: Variable Attributes 39583 395846.34.14 V850 Variable Attributes 39585-------------------------------- 39586 39587These variable attributes are supported by the V850 back end: 39588 39589'sda' 39590 Use this attribute to explicitly place a variable in the small data 39591 area, which can hold up to 64 kilobytes. 39592 39593'tda' 39594 Use this attribute to explicitly place a variable in the tiny data 39595 area, which can hold up to 256 bytes in total. 39596 39597'zda' 39598 Use this attribute to explicitly place a variable in the first 32 39599 kilobytes of memory. 39600 39601 39602File: gcc.info, Node: x86 Variable Attributes, Next: Xstormy16 Variable Attributes, Prev: V850 Variable Attributes, Up: Variable Attributes 39603 396046.34.15 x86 Variable Attributes 39605------------------------------- 39606 39607Two attributes are currently defined for x86 configurations: 'ms_struct' 39608and 'gcc_struct'. 39609 39610'ms_struct' 39611'gcc_struct' 39612 39613 If 'packed' is used on a structure, or if bit-fields are used, it 39614 may be that the Microsoft ABI lays out the structure differently 39615 than the way GCC normally does. Particularly when moving packed 39616 data between functions compiled with GCC and the native Microsoft 39617 compiler (either via function call or as data in a file), it may be 39618 necessary to access either format. 39619 39620 The 'ms_struct' and 'gcc_struct' attributes correspond to the 39621 '-mms-bitfields' and '-mno-ms-bitfields' command-line options, 39622 respectively; see *note x86 Options::, for details of how structure 39623 layout is affected. *Note x86 Type Attributes::, for information 39624 about the corresponding attributes on types. 39625 39626 39627File: gcc.info, Node: Xstormy16 Variable Attributes, Prev: x86 Variable Attributes, Up: Variable Attributes 39628 396296.34.16 Xstormy16 Variable Attributes 39630------------------------------------- 39631 39632One attribute is currently defined for xstormy16 configurations: 39633'below100'. 39634 39635'below100' 39636 39637 If a variable has the 'below100' attribute ('BELOW100' is allowed 39638 also), GCC places the variable in the first 0x100 bytes of memory 39639 and use special opcodes to access it. Such variables are placed in 39640 either the '.bss_below100' section or the '.data_below100' section. 39641 39642 39643File: gcc.info, Node: Type Attributes, Next: Label Attributes, Prev: Variable Attributes, Up: C Extensions 39644 396456.35 Specifying Attributes of Types 39646=================================== 39647 39648The keyword '__attribute__' allows you to specify various special 39649properties of types. Some type attributes apply only to structure and 39650union types, and in C++, also class types, while others can apply to any 39651type defined via a 'typedef' declaration. Unless otherwise specified, 39652the same restrictions and effects apply to attributes regardless of 39653whether a type is a trivial structure or a C++ class with user-defined 39654constructors, destructors, or a copy assignment. 39655 39656 Other attributes are defined for functions (*note Function 39657Attributes::), labels (*note Label Attributes::), enumerators (*note 39658Enumerator Attributes::), statements (*note Statement Attributes::), and 39659for variables (*note Variable Attributes::). 39660 39661 The '__attribute__' keyword is followed by an attribute specification 39662enclosed in double parentheses. 39663 39664 You may specify type attributes in an enum, struct or union type 39665declaration or definition by placing them immediately after the 39666'struct', 'union' or 'enum' keyword. You can also place them just past 39667the closing curly brace of the definition, but this is less preferred 39668because logically the type should be fully defined at the closing brace. 39669 39670 You can also include type attributes in a 'typedef' declaration. *Note 39671Attribute Syntax::, for details of the exact syntax for using 39672attributes. 39673 39674* Menu: 39675 39676* Common Type Attributes:: 39677* ARC Type Attributes:: 39678* ARM Type Attributes:: 39679* BPF Type Attributes:: 39680* MeP Type Attributes:: 39681* PowerPC Type Attributes:: 39682* x86 Type Attributes:: 39683 39684 39685File: gcc.info, Node: Common Type Attributes, Next: ARC Type Attributes, Up: Type Attributes 39686 396876.35.1 Common Type Attributes 39688----------------------------- 39689 39690The following type attributes are supported on most targets. 39691 39692'aligned' 39693'aligned (ALIGNMENT)' 39694 The 'aligned' attribute specifies a minimum alignment (in bytes) 39695 for variables of the specified type. When specified, ALIGNMENT 39696 must be a power of 2. Specifying no ALIGNMENT argument implies the 39697 maximum alignment for the target, which is often, but by no means 39698 always, 8 or 16 bytes. For example, the declarations: 39699 39700 struct __attribute__ ((aligned (8))) S { short f[3]; }; 39701 typedef int more_aligned_int __attribute__ ((aligned (8))); 39702 39703 force the compiler to ensure (as far as it can) that each variable 39704 whose type is 'struct S' or 'more_aligned_int' is allocated and 39705 aligned _at least_ on a 8-byte boundary. On a SPARC, having all 39706 variables of type 'struct S' aligned to 8-byte boundaries allows 39707 the compiler to use the 'ldd' and 'std' (doubleword load and store) 39708 instructions when copying one variable of type 'struct S' to 39709 another, thus improving run-time efficiency. 39710 39711 Note that the alignment of any given 'struct' or 'union' type is 39712 required by the ISO C standard to be at least a perfect multiple of 39713 the lowest common multiple of the alignments of all of the members 39714 of the 'struct' or 'union' in question. This means that you _can_ 39715 effectively adjust the alignment of a 'struct' or 'union' type by 39716 attaching an 'aligned' attribute to any one of the members of such 39717 a type, but the notation illustrated in the example above is a more 39718 obvious, intuitive, and readable way to request the compiler to 39719 adjust the alignment of an entire 'struct' or 'union' type. 39720 39721 As in the preceding example, you can explicitly specify the 39722 alignment (in bytes) that you wish the compiler to use for a given 39723 'struct' or 'union' type. Alternatively, you can leave out the 39724 alignment factor and just ask the compiler to align a type to the 39725 maximum useful alignment for the target machine you are compiling 39726 for. For example, you could write: 39727 39728 struct __attribute__ ((aligned)) S { short f[3]; }; 39729 39730 Whenever you leave out the alignment factor in an 'aligned' 39731 attribute specification, the compiler automatically sets the 39732 alignment for the type to the largest alignment that is ever used 39733 for any data type on the target machine you are compiling for. 39734 Doing this can often make copy operations more efficient, because 39735 the compiler can use whatever instructions copy the biggest chunks 39736 of memory when performing copies to or from the variables that have 39737 types that you have aligned this way. 39738 39739 In the example above, if the size of each 'short' is 2 bytes, then 39740 the size of the entire 'struct S' type is 6 bytes. The smallest 39741 power of two that is greater than or equal to that is 8, so the 39742 compiler sets the alignment for the entire 'struct S' type to 8 39743 bytes. 39744 39745 Note that although you can ask the compiler to select a 39746 time-efficient alignment for a given type and then declare only 39747 individual stand-alone objects of that type, the compiler's ability 39748 to select a time-efficient alignment is primarily useful only when 39749 you plan to create arrays of variables having the relevant 39750 (efficiently aligned) type. If you declare or use arrays of 39751 variables of an efficiently-aligned type, then it is likely that 39752 your program also does pointer arithmetic (or subscripting, which 39753 amounts to the same thing) on pointers to the relevant type, and 39754 the code that the compiler generates for these pointer arithmetic 39755 operations is often more efficient for efficiently-aligned types 39756 than for other types. 39757 39758 Note that the effectiveness of 'aligned' attributes may be limited 39759 by inherent limitations in your linker. On many systems, the 39760 linker is only able to arrange for variables to be aligned up to a 39761 certain maximum alignment. (For some linkers, the maximum 39762 supported alignment may be very very small.) If your linker is 39763 only able to align variables up to a maximum of 8-byte alignment, 39764 then specifying 'aligned (16)' in an '__attribute__' still only 39765 provides you with 8-byte alignment. See your linker documentation 39766 for further information. 39767 39768 When used on a struct, or struct member, the 'aligned' attribute 39769 can only increase the alignment; in order to decrease it, the 39770 'packed' attribute must be specified as well. When used as part of 39771 a typedef, the 'aligned' attribute can both increase and decrease 39772 alignment, and specifying the 'packed' attribute generates a 39773 warning. 39774 39775'warn_if_not_aligned (ALIGNMENT)' 39776 This attribute specifies a threshold for the structure field, 39777 measured in bytes. If the structure field is aligned below the 39778 threshold, a warning will be issued. For example, the declaration: 39779 39780 typedef unsigned long long __u64 39781 __attribute__((aligned (4), warn_if_not_aligned (8))); 39782 39783 struct foo 39784 { 39785 int i1; 39786 int i2; 39787 __u64 x; 39788 }; 39789 39790 causes the compiler to issue an warning on 'struct foo', like 39791 'warning: alignment 4 of 'struct foo' is less than 8'. It is used 39792 to define 'struct foo' in such a way that 'struct foo' has the same 39793 layout and the structure field 'x' has the same alignment when 39794 '__u64' is aligned at either 4 or 8 bytes. Align 'struct foo' to 8 39795 bytes: 39796 39797 struct __attribute__ ((aligned (8))) foo 39798 { 39799 int i1; 39800 int i2; 39801 __u64 x; 39802 }; 39803 39804 silences the warning. The compiler also issues a warning, like 39805 'warning: 'x' offset 12 in 'struct foo' isn't aligned to 8', when 39806 the structure field has the misaligned offset: 39807 39808 struct __attribute__ ((aligned (8))) foo 39809 { 39810 int i1; 39811 int i2; 39812 int i3; 39813 __u64 x; 39814 }; 39815 39816 This warning can be disabled by '-Wno-if-not-aligned'. 39817 39818'alloc_size (POSITION)' 39819'alloc_size (POSITION-1, POSITION-2)' 39820 The 'alloc_size' type attribute may be applied to the definition of 39821 a type of a function that returns a pointer and takes at least one 39822 argument of an integer type. It indicates that the returned 39823 pointer points to an object whose size is given by the function 39824 argument at POSITION-1, or by the product of the arguments at 39825 POSITION-1 and POSITION-2. Meaningful sizes are positive values 39826 less than 'PTRDIFF_MAX'. Other sizes are disagnosed when detected. 39827 GCC uses this information to improve the results of 39828 '__builtin_object_size'. 39829 39830 For instance, the following declarations 39831 39832 typedef __attribute__ ((alloc_size (1, 2))) void* 39833 calloc_type (size_t, size_t); 39834 typedef __attribute__ ((alloc_size (1))) void* 39835 malloc_type (size_t); 39836 39837 specify that 'calloc_type' is a type of a function that, like the 39838 standard C function 'calloc', returns an object whose size is given 39839 by the product of arguments 1 and 2, and that 'malloc_type', like 39840 the standard C function 'malloc', returns an object whose size is 39841 given by argument 1 to the function. 39842 39843'copy' 39844'copy (EXPRESSION)' 39845 The 'copy' attribute applies the set of attributes with which the 39846 type of the EXPRESSION has been declared to the declaration of the 39847 type to which the attribute is applied. The attribute is designed 39848 for libraries that define aliases that are expected to specify the 39849 same set of attributes as the aliased symbols. The 'copy' 39850 attribute can be used with types, variables, or functions. 39851 However, the kind of symbol to which the attribute is applied 39852 (either varible or function) must match the kind of symbol to which 39853 the argument refers. The 'copy' attribute copies only syntactic 39854 and semantic attributes but not attributes that affect a symbol's 39855 linkage or visibility such as 'alias', 'visibility', or 'weak'. 39856 The 'deprecated' attribute is also not copied. *Note Common 39857 Function Attributes::. *Note Common Variable Attributes::. 39858 39859 For example, suppose 'struct A' below is defined in some third 39860 party library header to have the alignment requirement 'N' and to 39861 force a warning whenever a variable of the type is not so aligned 39862 due to attribute 'packed'. Specifying the 'copy' attribute on the 39863 definition on the unrelated 'struct B' has the effect of copying 39864 all relevant attributes from the type referenced by the pointer 39865 expression to 'struct B'. 39866 39867 struct __attribute__ ((aligned (N), warn_if_not_aligned (N))) 39868 A { /* ... */ }; 39869 struct __attribute__ ((copy ( (struct A *)0)) B { /* ... */ }; 39870 39871'deprecated' 39872'deprecated (MSG)' 39873 The 'deprecated' attribute results in a warning if the type is used 39874 anywhere in the source file. This is useful when identifying types 39875 that are expected to be removed in a future version of a program. 39876 If possible, the warning also includes the location of the 39877 declaration of the deprecated type, to enable users to easily find 39878 further information about why the type is deprecated, or what they 39879 should do instead. Note that the warnings only occur for uses and 39880 then only if the type is being applied to an identifier that itself 39881 is not being declared as deprecated. 39882 39883 typedef int T1 __attribute__ ((deprecated)); 39884 T1 x; 39885 typedef T1 T2; 39886 T2 y; 39887 typedef T1 T3 __attribute__ ((deprecated)); 39888 T3 z __attribute__ ((deprecated)); 39889 39890 results in a warning on line 2 and 3 but not lines 4, 5, or 6. No 39891 warning is issued for line 4 because T2 is not explicitly 39892 deprecated. Line 5 has no warning because T3 is explicitly 39893 deprecated. Similarly for line 6. The optional MSG argument, 39894 which must be a string, is printed in the warning if present. 39895 Control characters in the string will be replaced with escape 39896 sequences, and if the '-fmessage-length' option is set to 0 (its 39897 default value) then any newline characters will be ignored. 39898 39899 The 'deprecated' attribute can also be used for functions and 39900 variables (*note Function Attributes::, *note Variable 39901 Attributes::.) 39902 39903 The message attached to the attribute is affected by the setting of 39904 the '-fmessage-length' option. 39905 39906'unavailable' 39907'unavailable (MSG)' 39908 The 'unavailable' attribute behaves in the same manner as the 39909 'deprecated' one, but emits an error rather than a warning. It is 39910 used to indicate that a (perhaps previously 'deprecated') type is 39911 no longer usable. 39912 39913 The 'unavailable' attribute can also be used for functions and 39914 variables (*note Function Attributes::, *note Variable 39915 Attributes::.) 39916 39917'designated_init' 39918 This attribute may only be applied to structure types. It 39919 indicates that any initialization of an object of this type must 39920 use designated initializers rather than positional initializers. 39921 The intent of this attribute is to allow the programmer to indicate 39922 that a structure's layout may change, and that therefore relying on 39923 positional initialization will result in future breakage. 39924 39925 GCC emits warnings based on this attribute by default; use 39926 '-Wno-designated-init' to suppress them. 39927 39928'may_alias' 39929 Accesses through pointers to types with this attribute are not 39930 subject to type-based alias analysis, but are instead assumed to be 39931 able to alias any other type of objects. In the context of section 39932 6.5 paragraph 7 of the C99 standard, an lvalue expression 39933 dereferencing such a pointer is treated like having a character 39934 type. See '-fstrict-aliasing' for more information on aliasing 39935 issues. This extension exists to support some vector APIs, in 39936 which pointers to one vector type are permitted to alias pointers 39937 to a different vector type. 39938 39939 Note that an object of a type with this attribute does not have any 39940 special semantics. 39941 39942 Example of use: 39943 39944 typedef short __attribute__ ((__may_alias__)) short_a; 39945 39946 int 39947 main (void) 39948 { 39949 int a = 0x12345678; 39950 short_a *b = (short_a *) &a; 39951 39952 b[1] = 0; 39953 39954 if (a == 0x12345678) 39955 abort(); 39956 39957 exit(0); 39958 } 39959 39960 If you replaced 'short_a' with 'short' in the variable declaration, 39961 the above program would abort when compiled with 39962 '-fstrict-aliasing', which is on by default at '-O2' or above. 39963 39964'mode (MODE)' 39965 This attribute specifies the data type for the 39966 declaration--whichever type corresponds to the mode MODE. This in 39967 effect lets you request an integer or floating-point type according 39968 to its width. 39969 39970 *Note (gccint)Machine Modes::, for a list of the possible keywords 39971 for MODE. You may also specify a mode of 'byte' or '__byte__' to 39972 indicate the mode corresponding to a one-byte integer, 'word' or 39973 '__word__' for the mode of a one-word integer, and 'pointer' or 39974 '__pointer__' for the mode used to represent pointers. 39975 39976'packed' 39977 This attribute, attached to a 'struct', 'union', or C++ 'class' 39978 type definition, specifies that each of its members (other than 39979 zero-width bit-fields) is placed to minimize the memory required. 39980 This is equivalent to specifying the 'packed' attribute on each of 39981 the members. 39982 39983 When attached to an 'enum' definition, the 'packed' attribute 39984 indicates that the smallest integral type should be used. 39985 Specifying the '-fshort-enums' flag on the command line is 39986 equivalent to specifying the 'packed' attribute on all 'enum' 39987 definitions. 39988 39989 In the following example 'struct my_packed_struct''s members are 39990 packed closely together, but the internal layout of its 's' member 39991 is not packed--to do that, 'struct my_unpacked_struct' needs to be 39992 packed too. 39993 39994 struct my_unpacked_struct 39995 { 39996 char c; 39997 int i; 39998 }; 39999 40000 struct __attribute__ ((__packed__)) my_packed_struct 40001 { 40002 char c; 40003 int i; 40004 struct my_unpacked_struct s; 40005 }; 40006 40007 You may only specify the 'packed' attribute on the definition of an 40008 'enum', 'struct', 'union', or 'class', not on a 'typedef' that does 40009 not also define the enumerated type, structure, union, or class. 40010 40011'scalar_storage_order ("ENDIANNESS")' 40012 When attached to a 'union' or a 'struct', this attribute sets the 40013 storage order, aka endianness, of the scalar fields of the type, as 40014 well as the array fields whose component is scalar. The supported 40015 endiannesses are 'big-endian' and 'little-endian'. The attribute 40016 has no effects on fields which are themselves a 'union', a 'struct' 40017 or an array whose component is a 'union' or a 'struct', and it is 40018 possible for these fields to have a different scalar storage order 40019 than the enclosing type. 40020 40021 Note that neither pointer nor vector fields are considered scalar 40022 fields in this context, so the attribute has no effects on these 40023 fields. 40024 40025 This attribute is supported only for targets that use a uniform 40026 default scalar storage order (fortunately, most of them), i.e. 40027 targets that store the scalars either all in big-endian or all in 40028 little-endian. 40029 40030 Additional restrictions are enforced for types with the reverse 40031 scalar storage order with regard to the scalar storage order of the 40032 target: 40033 40034 * Taking the address of a scalar field of a 'union' or a 40035 'struct' with reverse scalar storage order is not permitted 40036 and yields an error. 40037 * Taking the address of an array field, whose component is 40038 scalar, of a 'union' or a 'struct' with reverse scalar storage 40039 order is permitted but yields a warning, unless 40040 '-Wno-scalar-storage-order' is specified. 40041 * Taking the address of a 'union' or a 'struct' with reverse 40042 scalar storage order is permitted. 40043 40044 These restrictions exist because the storage order attribute is 40045 lost when the address of a scalar or the address of an array with 40046 scalar component is taken, so storing indirectly through this 40047 address generally does not work. The second case is nevertheless 40048 allowed to be able to perform a block copy from or to the array. 40049 40050 Moreover, the use of type punning or aliasing to toggle the storage 40051 order is not supported; that is to say, if a given scalar object 40052 can be accessed through distinct types that assign a different 40053 storage order to it, then the behavior is undefined. 40054 40055'transparent_union' 40056 40057 This attribute, attached to a 'union' type definition, indicates 40058 that any function parameter having that union type causes calls to 40059 that function to be treated in a special way. 40060 40061 First, the argument corresponding to a transparent union type can 40062 be of any type in the union; no cast is required. Also, if the 40063 union contains a pointer type, the corresponding argument can be a 40064 null pointer constant or a void pointer expression; and if the 40065 union contains a void pointer type, the corresponding argument can 40066 be any pointer expression. If the union member type is a pointer, 40067 qualifiers like 'const' on the referenced type must be respected, 40068 just as with normal pointer conversions. 40069 40070 Second, the argument is passed to the function using the calling 40071 conventions of the first member of the transparent union, not the 40072 calling conventions of the union itself. All members of the union 40073 must have the same machine representation; this is necessary for 40074 this argument passing to work properly. 40075 40076 Transparent unions are designed for library functions that have 40077 multiple interfaces for compatibility reasons. For example, 40078 suppose the 'wait' function must accept either a value of type 'int 40079 *' to comply with POSIX, or a value of type 'union wait *' to 40080 comply with the 4.1BSD interface. If 'wait''s parameter were 'void 40081 *', 'wait' would accept both kinds of arguments, but it would also 40082 accept any other pointer type and this would make argument type 40083 checking less useful. Instead, '<sys/wait.h>' might define the 40084 interface as follows: 40085 40086 typedef union __attribute__ ((__transparent_union__)) 40087 { 40088 int *__ip; 40089 union wait *__up; 40090 } wait_status_ptr_t; 40091 40092 pid_t wait (wait_status_ptr_t); 40093 40094 This interface allows either 'int *' or 'union wait *' arguments to 40095 be passed, using the 'int *' calling convention. The program can 40096 call 'wait' with arguments of either type: 40097 40098 int w1 () { int w; return wait (&w); } 40099 int w2 () { union wait w; return wait (&w); } 40100 40101 With this interface, 'wait''s implementation might look like this: 40102 40103 pid_t wait (wait_status_ptr_t p) 40104 { 40105 return waitpid (-1, p.__ip, 0); 40106 } 40107 40108'unused' 40109 When attached to a type (including a 'union' or a 'struct'), this 40110 attribute means that variables of that type are meant to appear 40111 possibly unused. GCC does not produce a warning for any variables 40112 of that type, even if the variable appears to do nothing. This is 40113 often the case with lock or thread classes, which are usually 40114 defined and then not referenced, but contain constructors and 40115 destructors that have nontrivial bookkeeping functions. 40116 40117'vector_size (BYTES)' 40118 This attribute specifies the vector size for the type, measured in 40119 bytes. The type to which it applies is known as the "base type". 40120 The BYTES argument must be a positive power-of-two multiple of the 40121 base type size. For example, the following declarations: 40122 40123 typedef __attribute__ ((vector_size (32))) int int_vec32_t ; 40124 typedef __attribute__ ((vector_size (32))) int* int_vec32_ptr_t; 40125 typedef __attribute__ ((vector_size (32))) int int_vec32_arr3_t[3]; 40126 40127 define 'int_vec32_t' to be a 32-byte vector type composed of 'int' 40128 sized units. With 'int' having a size of 4 bytes, the type defines 40129 a vector of eight units, four bytes each. The mode of variables of 40130 type 'int_vec32_t' is 'V8SI'. 'int_vec32_ptr_t' is then defined to 40131 be a pointer to such a vector type, and 'int_vec32_arr3_t' to be an 40132 array of three such vectors. *Note Vector Extensions::, for 40133 details of manipulating objects of vector types. 40134 40135 This attribute is only applicable to integral and floating scalar 40136 types. In function declarations the attribute applies to the 40137 function return type. 40138 40139 For example, the following: 40140 __attribute__ ((vector_size (16))) float get_flt_vec16 (void); 40141 declares 'get_flt_vec16' to be a function returning a 16-byte 40142 vector with the base type 'float'. 40143 40144'visibility' 40145 In C++, attribute visibility (*note Function Attributes::) can also 40146 be applied to class, struct, union and enum types. Unlike other 40147 type attributes, the attribute must appear between the initial 40148 keyword and the name of the type; it cannot appear after the body 40149 of the type. 40150 40151 Note that the type visibility is applied to vague linkage entities 40152 associated with the class (vtable, typeinfo node, etc.). In 40153 particular, if a class is thrown as an exception in one shared 40154 object and caught in another, the class must have default 40155 visibility. Otherwise the two shared objects are unable to use the 40156 same typeinfo node and exception handling will break. 40157 40158'objc_root_class (Objective-C and Objective-C++ only)' 40159 This attribute marks a class as being a root class, and thus allows 40160 the compiler to elide any warnings about a missing superclass and 40161 to make additional checks for mandatory methods as needed. 40162 40163 To specify multiple attributes, separate them by commas within the 40164double parentheses: for example, '__attribute__ ((aligned (16), 40165packed))'. 40166 40167 40168File: gcc.info, Node: ARC Type Attributes, Next: ARM Type Attributes, Prev: Common Type Attributes, Up: Type Attributes 40169 401706.35.2 ARC Type Attributes 40171-------------------------- 40172 40173Declaring objects with 'uncached' allows you to exclude data-cache 40174participation in load and store operations on those objects without 40175involving the additional semantic implications of 'volatile'. The '.di' 40176instruction suffix is used for all loads and stores of data declared 40177'uncached'. 40178 40179 40180File: gcc.info, Node: ARM Type Attributes, Next: BPF Type Attributes, Prev: ARC Type Attributes, Up: Type Attributes 40181 401826.35.3 ARM Type Attributes 40183-------------------------- 40184 40185On those ARM targets that support 'dllimport' (such as Symbian OS), you 40186can use the 'notshared' attribute to indicate that the virtual table and 40187other similar data for a class should not be exported from a DLL. For 40188example: 40189 40190 class __declspec(notshared) C { 40191 public: 40192 __declspec(dllimport) C(); 40193 virtual void f(); 40194 } 40195 40196 __declspec(dllexport) 40197 C::C() {} 40198 40199In this code, 'C::C' is exported from the current DLL, but the virtual 40200table for 'C' is not exported. (You can use '__attribute__' instead of 40201'__declspec' if you prefer, but most Symbian OS code uses '__declspec'.) 40202 40203 40204File: gcc.info, Node: BPF Type Attributes, Next: MeP Type Attributes, Prev: ARM Type Attributes, Up: Type Attributes 40205 402066.35.4 BPF Type Attributes 40207-------------------------- 40208 40209BPF Compile Once - Run Everywhere (CO-RE) support. When attached to a 40210'struct' or 'union' type definition, indicates that CO-RE relocation 40211information should be generated for any access to a variable of that 40212type. The behavior is equivalent to the programmer manually wrapping 40213every such access with '__builtin_preserve_access_index'. 40214 40215 40216File: gcc.info, Node: MeP Type Attributes, Next: PowerPC Type Attributes, Prev: BPF Type Attributes, Up: Type Attributes 40217 402186.35.5 MeP Type Attributes 40219-------------------------- 40220 40221Many of the MeP variable attributes may be applied to types as well. 40222Specifically, the 'based', 'tiny', 'near', and 'far' attributes may be 40223applied to either. The 'io' and 'cb' attributes may not be applied to 40224types. 40225 40226 40227File: gcc.info, Node: PowerPC Type Attributes, Next: x86 Type Attributes, Prev: MeP Type Attributes, Up: Type Attributes 40228 402296.35.6 PowerPC Type Attributes 40230------------------------------ 40231 40232Three attributes currently are defined for PowerPC configurations: 40233'altivec', 'ms_struct' and 'gcc_struct'. 40234 40235 For full documentation of the 'ms_struct' and 'gcc_struct' attributes 40236please see the documentation in *note x86 Type Attributes::. 40237 40238 The 'altivec' attribute allows one to declare AltiVec vector data types 40239supported by the AltiVec Programming Interface Manual. The attribute 40240requires an argument to specify one of three vector types: 'vector__', 40241'pixel__' (always followed by unsigned short), and 'bool__' (always 40242followed by unsigned). 40243 40244 __attribute__((altivec(vector__))) 40245 __attribute__((altivec(pixel__))) unsigned short 40246 __attribute__((altivec(bool__))) unsigned 40247 40248 These attributes mainly are intended to support the '__vector', 40249'__pixel', and '__bool' AltiVec keywords. 40250 40251 40252File: gcc.info, Node: x86 Type Attributes, Prev: PowerPC Type Attributes, Up: Type Attributes 40253 402546.35.7 x86 Type Attributes 40255-------------------------- 40256 40257Two attributes are currently defined for x86 configurations: 'ms_struct' 40258and 'gcc_struct'. 40259 40260'ms_struct' 40261'gcc_struct' 40262 40263 If 'packed' is used on a structure, or if bit-fields are used it 40264 may be that the Microsoft ABI packs them differently than GCC 40265 normally packs them. Particularly when moving packed data between 40266 functions compiled with GCC and the native Microsoft compiler 40267 (either via function call or as data in a file), it may be 40268 necessary to access either format. 40269 40270 The 'ms_struct' and 'gcc_struct' attributes correspond to the 40271 '-mms-bitfields' and '-mno-ms-bitfields' command-line options, 40272 respectively; see *note x86 Options::, for details of how structure 40273 layout is affected. *Note x86 Variable Attributes::, for 40274 information about the corresponding attributes on variables. 40275 40276 40277File: gcc.info, Node: Label Attributes, Next: Enumerator Attributes, Prev: Type Attributes, Up: C Extensions 40278 402796.36 Label Attributes 40280===================== 40281 40282GCC allows attributes to be set on C labels. *Note Attribute Syntax::, 40283for details of the exact syntax for using attributes. Other attributes 40284are available for functions (*note Function Attributes::), variables 40285(*note Variable Attributes::), enumerators (*note Enumerator 40286Attributes::), statements (*note Statement Attributes::), and for types 40287(*note Type Attributes::). A label attribute followed by a declaration 40288appertains to the label and not the declaration. 40289 40290 This example uses the 'cold' label attribute to indicate the 40291'ErrorHandling' branch is unlikely to be taken and that the 40292'ErrorHandling' label is unused: 40293 40294 40295 asm goto ("some asm" : : : : NoError); 40296 40297 /* This branch (the fall-through from the asm) is less commonly used */ 40298 ErrorHandling: 40299 __attribute__((cold, unused)); /* Semi-colon is required here */ 40300 printf("error\n"); 40301 return 0; 40302 40303 NoError: 40304 printf("no error\n"); 40305 return 1; 40306 40307'unused' 40308 This feature is intended for program-generated code that may 40309 contain unused labels, but which is compiled with '-Wall'. It is 40310 not normally appropriate to use in it human-written code, though it 40311 could be useful in cases where the code that jumps to the label is 40312 contained within an '#ifdef' conditional. 40313 40314'hot' 40315 The 'hot' attribute on a label is used to inform the compiler that 40316 the path following the label is more likely than paths that are not 40317 so annotated. This attribute is used in cases where 40318 '__builtin_expect' cannot be used, for instance with computed goto 40319 or 'asm goto'. 40320 40321'cold' 40322 The 'cold' attribute on labels is used to inform the compiler that 40323 the path following the label is unlikely to be executed. This 40324 attribute is used in cases where '__builtin_expect' cannot be used, 40325 for instance with computed goto or 'asm goto'. 40326 40327 40328File: gcc.info, Node: Enumerator Attributes, Next: Statement Attributes, Prev: Label Attributes, Up: C Extensions 40329 403306.37 Enumerator Attributes 40331========================== 40332 40333GCC allows attributes to be set on enumerators. *Note Attribute 40334Syntax::, for details of the exact syntax for using attributes. Other 40335attributes are available for functions (*note Function Attributes::), 40336variables (*note Variable Attributes::), labels (*note Label 40337Attributes::), statements (*note Statement Attributes::), and for types 40338(*note Type Attributes::). 40339 40340 This example uses the 'deprecated' enumerator attribute to indicate the 40341'oldval' enumerator is deprecated: 40342 40343 enum E { 40344 oldval __attribute__((deprecated)), 40345 newval 40346 }; 40347 40348 int 40349 fn (void) 40350 { 40351 return oldval; 40352 } 40353 40354'deprecated' 40355 The 'deprecated' attribute results in a warning if the enumerator 40356 is used anywhere in the source file. This is useful when 40357 identifying enumerators that are expected to be removed in a future 40358 version of a program. The warning also includes the location of 40359 the declaration of the deprecated enumerator, to enable users to 40360 easily find further information about why the enumerator is 40361 deprecated, or what they should do instead. Note that the warnings 40362 only occurs for uses. 40363 40364'unavailable' 40365 The 'unavailable' attribute results in an error if the enumerator 40366 is used anywhere in the source file. In other respects it behaves 40367 in the same manner as the 'deprecated' attribute. 40368 40369 40370File: gcc.info, Node: Statement Attributes, Next: Attribute Syntax, Prev: Enumerator Attributes, Up: C Extensions 40371 403726.38 Statement Attributes 40373========================= 40374 40375GCC allows attributes to be set on null statements. *Note Attribute 40376Syntax::, for details of the exact syntax for using attributes. Other 40377attributes are available for functions (*note Function Attributes::), 40378variables (*note Variable Attributes::), labels (*note Label 40379Attributes::), enumerators (*note Enumerator Attributes::), and for 40380types (*note Type Attributes::). 40381 40382 This example uses the 'fallthrough' statement attribute to indicate 40383that the '-Wimplicit-fallthrough' warning should not be emitted: 40384 40385 switch (cond) 40386 { 40387 case 1: 40388 bar (1); 40389 __attribute__((fallthrough)); 40390 case 2: 40391 ... 40392 } 40393 40394'fallthrough' 40395 The 'fallthrough' attribute with a null statement serves as a 40396 fallthrough statement. It hints to the compiler that a statement 40397 that falls through to another case label, or user-defined label in 40398 a switch statement is intentional and thus the 40399 '-Wimplicit-fallthrough' warning must not trigger. The fallthrough 40400 attribute may appear at most once in each attribute list, and may 40401 not be mixed with other attributes. It can only be used in a 40402 switch statement (the compiler will issue an error otherwise), 40403 after a preceding statement and before a logically succeeding case 40404 label, or user-defined label. 40405 40406 40407File: gcc.info, Node: Attribute Syntax, Next: Function Prototypes, Prev: Statement Attributes, Up: C Extensions 40408 404096.39 Attribute Syntax 40410===================== 40411 40412This section describes the syntax with which '__attribute__' may be 40413used, and the constructs to which attribute specifiers bind, for the C 40414language. Some details may vary for C++ and Objective-C. Because of 40415infelicities in the grammar for attributes, some forms described here 40416may not be successfully parsed in all cases. 40417 40418 There are some problems with the semantics of attributes in C++. For 40419example, there are no manglings for attributes, although they may affect 40420code generation, so problems may arise when attributed types are used in 40421conjunction with templates or overloading. Similarly, 'typeid' does not 40422distinguish between types with different attributes. Support for 40423attributes in C++ may be restricted in future to attributes on 40424declarations only, but not on nested declarators. 40425 40426 *Note Function Attributes::, for details of the semantics of attributes 40427applying to functions. *Note Variable Attributes::, for details of the 40428semantics of attributes applying to variables. *Note Type Attributes::, 40429for details of the semantics of attributes applying to structure, union 40430and enumerated types. *Note Label Attributes::, for details of the 40431semantics of attributes applying to labels. *Note Enumerator 40432Attributes::, for details of the semantics of attributes applying to 40433enumerators. *Note Statement Attributes::, for details of the semantics 40434of attributes applying to statements. 40435 40436 An "attribute specifier" is of the form '__attribute__ 40437((ATTRIBUTE-LIST))'. An "attribute list" is a possibly empty 40438comma-separated sequence of "attributes", where each attribute is one of 40439the following: 40440 40441 * Empty. Empty attributes are ignored. 40442 40443 * An attribute name (which may be an identifier such as 'unused', or 40444 a reserved word such as 'const'). 40445 40446 * An attribute name followed by a parenthesized list of parameters 40447 for the attribute. These parameters take one of the following 40448 forms: 40449 40450 * An identifier. For example, 'mode' attributes use this form. 40451 40452 * An identifier followed by a comma and a non-empty 40453 comma-separated list of expressions. For example, 'format' 40454 attributes use this form. 40455 40456 * A possibly empty comma-separated list of expressions. For 40457 example, 'format_arg' attributes use this form with the list 40458 being a single integer constant expression, and 'alias' 40459 attributes use this form with the list being a single string 40460 constant. 40461 40462 An "attribute specifier list" is a sequence of one or more attribute 40463specifiers, not separated by any other tokens. 40464 40465 You may optionally specify attribute names with '__' preceding and 40466following the name. This allows you to use them in header files without 40467being concerned about a possible macro of the same name. For example, 40468you may use the attribute name '__noreturn__' instead of 'noreturn'. 40469 40470Label Attributes 40471................ 40472 40473In GNU C, an attribute specifier list may appear after the colon 40474following a label, other than a 'case' or 'default' label. GNU C++ only 40475permits attributes on labels if the attribute specifier is immediately 40476followed by a semicolon (i.e., the label applies to an empty statement). 40477If the semicolon is missing, C++ label attributes are ambiguous, as it 40478is permissible for a declaration, which could begin with an attribute 40479list, to be labelled in C++. Declarations cannot be labelled in C90 or 40480C99, so the ambiguity does not arise there. 40481 40482Enumerator Attributes 40483..................... 40484 40485In GNU C, an attribute specifier list may appear as part of an 40486enumerator. The attribute goes after the enumeration constant, before 40487'=', if present. The optional attribute in the enumerator appertains to 40488the enumeration constant. It is not possible to place the attribute 40489after the constant expression, if present. 40490 40491Statement Attributes 40492.................... 40493 40494In GNU C, an attribute specifier list may appear as part of a null 40495statement. The attribute goes before the semicolon. 40496 40497Type Attributes 40498............... 40499 40500An attribute specifier list may appear as part of a 'struct', 'union' or 40501'enum' specifier. It may go either immediately after the 'struct', 40502'union' or 'enum' keyword, or after the closing brace. The former 40503syntax is preferred. Where attribute specifiers follow the closing 40504brace, they are considered to relate to the structure, union or 40505enumerated type defined, not to any enclosing declaration the type 40506specifier appears in, and the type defined is not complete until after 40507the attribute specifiers. 40508 40509All other attributes 40510.................... 40511 40512Otherwise, an attribute specifier appears as part of a declaration, 40513counting declarations of unnamed parameters and type names, and relates 40514to that declaration (which may be nested in another declaration, for 40515example in the case of a parameter declaration), or to a particular 40516declarator within a declaration. Where an attribute specifier is 40517applied to a parameter declared as a function or an array, it should 40518apply to the function or array rather than the pointer to which the 40519parameter is implicitly converted, but this is not yet correctly 40520implemented. 40521 40522 Any list of specifiers and qualifiers at the start of a declaration may 40523contain attribute specifiers, whether or not such a list may in that 40524context contain storage class specifiers. (Some attributes, however, 40525are essentially in the nature of storage class specifiers, and only make 40526sense where storage class specifiers may be used; for example, 40527'section'.) There is one necessary limitation to this syntax: the first 40528old-style parameter declaration in a function definition cannot begin 40529with an attribute specifier, because such an attribute applies to the 40530function instead by syntax described below (which, however, is not yet 40531implemented in this case). In some other cases, attribute specifiers 40532are permitted by this grammar but not yet supported by the compiler. 40533All attribute specifiers in this place relate to the declaration as a 40534whole. In the obsolescent usage where a type of 'int' is implied by the 40535absence of type specifiers, such a list of specifiers and qualifiers may 40536be an attribute specifier list with no other specifiers or qualifiers. 40537 40538 At present, the first parameter in a function prototype must have some 40539type specifier that is not an attribute specifier; this resolves an 40540ambiguity in the interpretation of 'void f(int (__attribute__((foo)) 40541x))', but is subject to change. At present, if the parentheses of a 40542function declarator contain only attributes then those attributes are 40543ignored, rather than yielding an error or warning or implying a single 40544parameter of type int, but this is subject to change. 40545 40546 An attribute specifier list may appear immediately before a declarator 40547(other than the first) in a comma-separated list of declarators in a 40548declaration of more than one identifier using a single list of 40549specifiers and qualifiers. Such attribute specifiers apply only to the 40550identifier before whose declarator they appear. For example, in 40551 40552 __attribute__((noreturn)) void d0 (void), 40553 __attribute__((format(printf, 1, 2))) d1 (const char *, ...), 40554 d2 (void); 40555 40556the 'noreturn' attribute applies to all the functions declared; the 40557'format' attribute only applies to 'd1'. 40558 40559 An attribute specifier list may appear immediately before the comma, 40560'=' or semicolon terminating the declaration of an identifier other than 40561a function definition. Such attribute specifiers apply to the declared 40562object or function. Where an assembler name for an object or function 40563is specified (*note Asm Labels::), the attribute must follow the 'asm' 40564specification. 40565 40566 An attribute specifier list may, in future, be permitted to appear 40567after the declarator in a function definition (before any old-style 40568parameter declarations or the function body). 40569 40570 Attribute specifiers may be mixed with type qualifiers appearing inside 40571the '[]' of a parameter array declarator, in the C99 construct by which 40572such qualifiers are applied to the pointer to which the array is 40573implicitly converted. Such attribute specifiers apply to the pointer, 40574not to the array, but at present this is not implemented and they are 40575ignored. 40576 40577 An attribute specifier list may appear at the start of a nested 40578declarator. At present, there are some limitations in this usage: the 40579attributes correctly apply to the declarator, but for most individual 40580attributes the semantics this implies are not implemented. When 40581attribute specifiers follow the '*' of a pointer declarator, they may be 40582mixed with any type qualifiers present. The following describes the 40583formal semantics of this syntax. It makes the most sense if you are 40584familiar with the formal specification of declarators in the ISO C 40585standard. 40586 40587 Consider (as in C99 subclause 6.7.5 paragraph 4) a declaration 'T D1', 40588where 'T' contains declaration specifiers that specify a type TYPE (such 40589as 'int') and 'D1' is a declarator that contains an identifier IDENT. 40590The type specified for IDENT for derived declarators whose type does not 40591include an attribute specifier is as in the ISO C standard. 40592 40593 If 'D1' has the form '( ATTRIBUTE-SPECIFIER-LIST D )', and the 40594declaration 'T D' specifies the type "DERIVED-DECLARATOR-TYPE-LIST TYPE" 40595for IDENT, then 'T D1' specifies the type "DERIVED-DECLARATOR-TYPE-LIST 40596ATTRIBUTE-SPECIFIER-LIST TYPE" for IDENT. 40597 40598 If 'D1' has the form '* TYPE-QUALIFIER-AND-ATTRIBUTE-SPECIFIER-LIST D', 40599and the declaration 'T D' specifies the type 40600"DERIVED-DECLARATOR-TYPE-LIST TYPE" for IDENT, then 'T D1' specifies the 40601type "DERIVED-DECLARATOR-TYPE-LIST 40602TYPE-QUALIFIER-AND-ATTRIBUTE-SPECIFIER-LIST pointer to TYPE" for IDENT. 40603 40604 For example, 40605 40606 void (__attribute__((noreturn)) ****f) (void); 40607 40608specifies the type "pointer to pointer to pointer to pointer to 40609non-returning function returning 'void'". As another example, 40610 40611 char *__attribute__((aligned(8))) *f; 40612 40613specifies the type "pointer to 8-byte-aligned pointer to 'char'". Note 40614again that this does not work with most attributes; for example, the 40615usage of 'aligned' and 'noreturn' attributes given above is not yet 40616supported. 40617 40618 For compatibility with existing code written for compiler versions that 40619did not implement attributes on nested declarators, some laxity is 40620allowed in the placing of attributes. If an attribute that only applies 40621to types is applied to a declaration, it is treated as applying to the 40622type of that declaration. If an attribute that only applies to 40623declarations is applied to the type of a declaration, it is treated as 40624applying to that declaration; and, for compatibility with code placing 40625the attributes immediately before the identifier declared, such an 40626attribute applied to a function return type is treated as applying to 40627the function type, and such an attribute applied to an array element 40628type is treated as applying to the array type. If an attribute that 40629only applies to function types is applied to a pointer-to-function type, 40630it is treated as applying to the pointer target type; if such an 40631attribute is applied to a function return type that is not a 40632pointer-to-function type, it is treated as applying to the function 40633type. 40634 40635 40636File: gcc.info, Node: Function Prototypes, Next: C++ Comments, Prev: Attribute Syntax, Up: C Extensions 40637 406386.40 Prototypes and Old-Style Function Definitions 40639================================================== 40640 40641GNU C extends ISO C to allow a function prototype to override a later 40642old-style non-prototype definition. Consider the following example: 40643 40644 /* Use prototypes unless the compiler is old-fashioned. */ 40645 #ifdef __STDC__ 40646 #define P(x) x 40647 #else 40648 #define P(x) () 40649 #endif 40650 40651 /* Prototype function declaration. */ 40652 int isroot P((uid_t)); 40653 40654 /* Old-style function definition. */ 40655 int 40656 isroot (x) /* ??? lossage here ??? */ 40657 uid_t x; 40658 { 40659 return x == 0; 40660 } 40661 40662 Suppose the type 'uid_t' happens to be 'short'. ISO C does not allow 40663this example, because subword arguments in old-style non-prototype 40664definitions are promoted. Therefore in this example the function 40665definition's argument is really an 'int', which does not match the 40666prototype argument type of 'short'. 40667 40668 This restriction of ISO C makes it hard to write code that is portable 40669to traditional C compilers, because the programmer does not know whether 40670the 'uid_t' type is 'short', 'int', or 'long'. Therefore, in cases like 40671these GNU C allows a prototype to override a later old-style definition. 40672More precisely, in GNU C, a function prototype argument type overrides 40673the argument type specified by a later old-style definition if the 40674former type is the same as the latter type before promotion. Thus in 40675GNU C the above example is equivalent to the following: 40676 40677 int isroot (uid_t); 40678 40679 int 40680 isroot (uid_t x) 40681 { 40682 return x == 0; 40683 } 40684 40685GNU C++ does not support old-style function definitions, so this 40686extension is irrelevant. 40687 40688 40689File: gcc.info, Node: C++ Comments, Next: Dollar Signs, Prev: Function Prototypes, Up: C Extensions 40690 406916.41 C++ Style Comments 40692======================= 40693 40694In GNU C, you may use C++ style comments, which start with '//' and 40695continue until the end of the line. Many other C implementations allow 40696such comments, and they are included in the 1999 C standard. However, 40697C++ style comments are not recognized if you specify an '-std' option 40698specifying a version of ISO C before C99, or '-ansi' (equivalent to 40699'-std=c90'). 40700 40701 40702File: gcc.info, Node: Dollar Signs, Next: Character Escapes, Prev: C++ Comments, Up: C Extensions 40703 407046.42 Dollar Signs in Identifier Names 40705===================================== 40706 40707In GNU C, you may normally use dollar signs in identifier names. This 40708is because many traditional C implementations allow such identifiers. 40709However, dollar signs in identifiers are not supported on a few target 40710machines, typically because the target assembler does not allow them. 40711 40712 40713File: gcc.info, Node: Character Escapes, Next: Alignment, Prev: Dollar Signs, Up: C Extensions 40714 407156.43 The Character <ESC> in Constants 40716===================================== 40717 40718You can use the sequence '\e' in a string or character constant to stand 40719for the ASCII character <ESC>. 40720 40721 40722File: gcc.info, Node: Alignment, Next: Inline, Prev: Character Escapes, Up: C Extensions 40723 407246.44 Determining the Alignment of Functions, Types or Variables 40725=============================================================== 40726 40727The keyword '__alignof__' determines the alignment requirement of a 40728function, object, or a type, or the minimum alignment usually required 40729by a type. Its syntax is just like 'sizeof' and C11 '_Alignof'. 40730 40731 For example, if the target machine requires a 'double' value to be 40732aligned on an 8-byte boundary, then '__alignof__ (double)' is 8. This 40733is true on many RISC machines. On more traditional machine designs, 40734'__alignof__ (double)' is 4 or even 2. 40735 40736 Some machines never actually require alignment; they allow references 40737to any data type even at an odd address. For these machines, 40738'__alignof__' reports the smallest alignment that GCC gives the data 40739type, usually as mandated by the target ABI. 40740 40741 If the operand of '__alignof__' is an lvalue rather than a type, its 40742value is the required alignment for its type, taking into account any 40743minimum alignment specified by attribute 'aligned' (*note Common 40744Variable Attributes::). For example, after this declaration: 40745 40746 struct foo { int x; char y; } foo1; 40747 40748the value of '__alignof__ (foo1.y)' is 1, even though its actual 40749alignment is probably 2 or 4, the same as '__alignof__ (int)'. It is an 40750error to ask for the alignment of an incomplete type other than 'void'. 40751 40752 If the operand of the '__alignof__' expression is a function, the 40753expression evaluates to the alignment of the function which may be 40754specified by attribute 'aligned' (*note Common Function Attributes::). 40755 40756 40757File: gcc.info, Node: Inline, Next: Volatiles, Prev: Alignment, Up: C Extensions 40758 407596.45 An Inline Function is As Fast As a Macro 40760============================================= 40761 40762By declaring a function inline, you can direct GCC to make calls to that 40763function faster. One way GCC can achieve this is to integrate that 40764function's code into the code for its callers. This makes execution 40765faster by eliminating the function-call overhead; in addition, if any of 40766the actual argument values are constant, their known values may permit 40767simplifications at compile time so that not all of the inline function's 40768code needs to be included. The effect on code size is less predictable; 40769object code may be larger or smaller with function inlining, depending 40770on the particular case. You can also direct GCC to try to integrate all 40771"simple enough" functions into their callers with the option 40772'-finline-functions'. 40773 40774 GCC implements three different semantics of declaring a function 40775inline. One is available with '-std=gnu89' or '-fgnu89-inline' or when 40776'gnu_inline' attribute is present on all inline declarations, another 40777when '-std=c99', '-std=gnu99' or an option for a later C version is used 40778(without '-fgnu89-inline'), and the third is used when compiling C++. 40779 40780 To declare a function inline, use the 'inline' keyword in its 40781declaration, like this: 40782 40783 static inline int 40784 inc (int *a) 40785 { 40786 return (*a)++; 40787 } 40788 40789 If you are writing a header file to be included in ISO C90 programs, 40790write '__inline__' instead of 'inline'. *Note Alternate Keywords::. 40791 40792 The three types of inlining behave similarly in two important cases: 40793when the 'inline' keyword is used on a 'static' function, like the 40794example above, and when a function is first declared without using the 40795'inline' keyword and then is defined with 'inline', like this: 40796 40797 extern int inc (int *a); 40798 inline int 40799 inc (int *a) 40800 { 40801 return (*a)++; 40802 } 40803 40804 In both of these common cases, the program behaves the same as if you 40805had not used the 'inline' keyword, except for its speed. 40806 40807 When a function is both inline and 'static', if all calls to the 40808function are integrated into the caller, and the function's address is 40809never used, then the function's own assembler code is never referenced. 40810In this case, GCC does not actually output assembler code for the 40811function, unless you specify the option '-fkeep-inline-functions'. If 40812there is a nonintegrated call, then the function is compiled to 40813assembler code as usual. The function must also be compiled as usual if 40814the program refers to its address, because that cannot be inlined. 40815 40816 Note that certain usages in a function definition can make it 40817unsuitable for inline substitution. Among these usages are: variadic 40818functions, use of 'alloca', use of computed goto (*note Labels as 40819Values::), use of nonlocal goto, use of nested functions, use of 40820'setjmp', use of '__builtin_longjmp' and use of '__builtin_return' or 40821'__builtin_apply_args'. Using '-Winline' warns when a function marked 40822'inline' could not be substituted, and gives the reason for the failure. 40823 40824 As required by ISO C++, GCC considers member functions defined within 40825the body of a class to be marked inline even if they are not explicitly 40826declared with the 'inline' keyword. You can override this with 40827'-fno-default-inline'; *note Options Controlling C++ Dialect: C++ 40828Dialect Options. 40829 40830 GCC does not inline any functions when not optimizing unless you 40831specify the 'always_inline' attribute for the function, like this: 40832 40833 /* Prototype. */ 40834 inline void foo (const char) __attribute__((always_inline)); 40835 40836 The remainder of this section is specific to GNU C90 inlining. 40837 40838 When an inline function is not 'static', then the compiler must assume 40839that there may be calls from other source files; since a global symbol 40840can be defined only once in any program, the function must not be 40841defined in the other source files, so the calls therein cannot be 40842integrated. Therefore, a non-'static' inline function is always 40843compiled on its own in the usual fashion. 40844 40845 If you specify both 'inline' and 'extern' in the function definition, 40846then the definition is used only for inlining. In no case is the 40847function compiled on its own, not even if you refer to its address 40848explicitly. Such an address becomes an external reference, as if you 40849had only declared the function, and had not defined it. 40850 40851 This combination of 'inline' and 'extern' has almost the effect of a 40852macro. The way to use it is to put a function definition in a header 40853file with these keywords, and put another copy of the definition 40854(lacking 'inline' and 'extern') in a library file. The definition in 40855the header file causes most calls to the function to be inlined. If any 40856uses of the function remain, they refer to the single copy in the 40857library. 40858 40859 40860File: gcc.info, Node: Volatiles, Next: Using Assembly Language with C, Prev: Inline, Up: C Extensions 40861 408626.46 When is a Volatile Object Accessed? 40863======================================== 40864 40865C has the concept of volatile objects. These are normally accessed by 40866pointers and used for accessing hardware or inter-thread communication. 40867The standard encourages compilers to refrain from optimizations 40868concerning accesses to volatile objects, but leaves it implementation 40869defined as to what constitutes a volatile access. The minimum 40870requirement is that at a sequence point all previous accesses to 40871volatile objects have stabilized and no subsequent accesses have 40872occurred. Thus an implementation is free to reorder and combine 40873volatile accesses that occur between sequence points, but cannot do so 40874for accesses across a sequence point. The use of volatile does not 40875allow you to violate the restriction on updating objects multiple times 40876between two sequence points. 40877 40878 Accesses to non-volatile objects are not ordered with respect to 40879volatile accesses. You cannot use a volatile object as a memory barrier 40880to order a sequence of writes to non-volatile memory. For instance: 40881 40882 int *ptr = SOMETHING; 40883 volatile int vobj; 40884 *ptr = SOMETHING; 40885 vobj = 1; 40886 40887Unless *PTR and VOBJ can be aliased, it is not guaranteed that the write 40888to *PTR occurs by the time the update of VOBJ happens. If you need this 40889guarantee, you must use a stronger memory barrier such as: 40890 40891 int *ptr = SOMETHING; 40892 volatile int vobj; 40893 *ptr = SOMETHING; 40894 asm volatile ("" : : : "memory"); 40895 vobj = 1; 40896 40897 A scalar volatile object is read when it is accessed in a void context: 40898 40899 volatile int *src = SOMEVALUE; 40900 *src; 40901 40902 Such expressions are rvalues, and GCC implements this as a read of the 40903volatile object being pointed to. 40904 40905 Assignments are also expressions and have an rvalue. However when 40906assigning to a scalar volatile, the volatile object is not reread, 40907regardless of whether the assignment expression's rvalue is used or not. 40908If the assignment's rvalue is used, the value is that assigned to the 40909volatile object. For instance, there is no read of VOBJ in all the 40910following cases: 40911 40912 int obj; 40913 volatile int vobj; 40914 vobj = SOMETHING; 40915 obj = vobj = SOMETHING; 40916 obj ? vobj = ONETHING : vobj = ANOTHERTHING; 40917 obj = (SOMETHING, vobj = ANOTHERTHING); 40918 40919 If you need to read the volatile object after an assignment has 40920occurred, you must use a separate expression with an intervening 40921sequence point. 40922 40923 As bit-fields are not individually addressable, volatile bit-fields may 40924be implicitly read when written to, or when adjacent bit-fields are 40925accessed. Bit-field operations may be optimized such that adjacent 40926bit-fields are only partially accessed, if they straddle a storage unit 40927boundary. For these reasons it is unwise to use volatile bit-fields to 40928access hardware. 40929 40930 40931File: gcc.info, Node: Using Assembly Language with C, Next: Alternate Keywords, Prev: Volatiles, Up: C Extensions 40932 409336.47 How to Use Inline Assembly Language in C Code 40934================================================== 40935 40936The 'asm' keyword allows you to embed assembler instructions within C 40937code. GCC provides two forms of inline 'asm' statements. A "basic 40938'asm'" statement is one with no operands (*note Basic Asm::), while an 40939"extended 'asm'" statement (*note Extended Asm::) includes one or more 40940operands. The extended form is preferred for mixing C and assembly 40941language within a function, but to include assembly language at top 40942level you must use basic 'asm'. 40943 40944 You can also use the 'asm' keyword to override the assembler name for a 40945C symbol, or to place a C variable in a specific register. 40946 40947* Menu: 40948 40949* Basic Asm:: Inline assembler without operands. 40950* Extended Asm:: Inline assembler with operands. 40951* Constraints:: Constraints for 'asm' operands 40952* Asm Labels:: Specifying the assembler name to use for a C symbol. 40953* Explicit Register Variables:: Defining variables residing in specified 40954 registers. 40955* Size of an asm:: How GCC calculates the size of an 'asm' block. 40956 40957 40958File: gcc.info, Node: Basic Asm, Next: Extended Asm, Up: Using Assembly Language with C 40959 409606.47.1 Basic Asm -- Assembler Instructions Without Operands 40961----------------------------------------------------------- 40962 40963A basic 'asm' statement has the following syntax: 40964 40965 asm ASM-QUALIFIERS ( ASSEMBLERINSTRUCTIONS ) 40966 40967 For the C language, the 'asm' keyword is a GNU extension. When writing 40968C code that can be compiled with '-ansi' and the '-std' options that 40969select C dialects without GNU extensions, use '__asm__' instead of 'asm' 40970(*note Alternate Keywords::). For the C++ language, 'asm' is a standard 40971keyword, but '__asm__' can be used for code compiled with '-fno-asm'. 40972 40973Qualifiers 40974.......... 40975 40976'volatile' 40977 The optional 'volatile' qualifier has no effect. All basic 'asm' 40978 blocks are implicitly volatile. 40979 40980'inline' 40981 If you use the 'inline' qualifier, then for inlining purposes the 40982 size of the 'asm' statement is taken as the smallest size possible 40983 (*note Size of an asm::). 40984 40985Parameters 40986.......... 40987 40988ASSEMBLERINSTRUCTIONS 40989 This is a literal string that specifies the assembler code. The 40990 string can contain any instructions recognized by the assembler, 40991 including directives. GCC does not parse the assembler 40992 instructions themselves and does not know what they mean or even 40993 whether they are valid assembler input. 40994 40995 You may place multiple assembler instructions together in a single 40996 'asm' string, separated by the characters normally used in assembly 40997 code for the system. A combination that works in most places is a 40998 newline to break the line, plus a tab character (written as 40999 '\n\t'). Some assemblers allow semicolons as a line separator. 41000 However, note that some assembler dialects use semicolons to start 41001 a comment. 41002 41003Remarks 41004....... 41005 41006Using extended 'asm' (*note Extended Asm::) typically produces smaller, 41007safer, and more efficient code, and in most cases it is a better 41008solution than basic 'asm'. However, there are two situations where only 41009basic 'asm' can be used: 41010 41011 * Extended 'asm' statements have to be inside a C function, so to 41012 write inline assembly language at file scope ("top-level"), outside 41013 of C functions, you must use basic 'asm'. You can use this 41014 technique to emit assembler directives, define assembly language 41015 macros that can be invoked elsewhere in the file, or write entire 41016 functions in assembly language. Basic 'asm' statements outside of 41017 functions may not use any qualifiers. 41018 41019 * Functions declared with the 'naked' attribute also require basic 41020 'asm' (*note Function Attributes::). 41021 41022 Safely accessing C data and calling functions from basic 'asm' is more 41023complex than it may appear. To access C data, it is better to use 41024extended 'asm'. 41025 41026 Do not expect a sequence of 'asm' statements to remain perfectly 41027consecutive after compilation. If certain instructions need to remain 41028consecutive in the output, put them in a single multi-instruction 'asm' 41029statement. Note that GCC's optimizers can move 'asm' statements 41030relative to other code, including across jumps. 41031 41032 'asm' statements may not perform jumps into other 'asm' statements. 41033GCC does not know about these jumps, and therefore cannot take account 41034of them when deciding how to optimize. Jumps from 'asm' to C labels are 41035only supported in extended 'asm'. 41036 41037 Under certain circumstances, GCC may duplicate (or remove duplicates 41038of) your assembly code when optimizing. This can lead to unexpected 41039duplicate symbol errors during compilation if your assembly code defines 41040symbols or labels. 41041 41042 *Warning:* The C standards do not specify semantics for 'asm', making 41043it a potential source of incompatibilities between compilers. These 41044incompatibilities may not produce compiler warnings/errors. 41045 41046 GCC does not parse basic 'asm''s ASSEMBLERINSTRUCTIONS, which means 41047there is no way to communicate to the compiler what is happening inside 41048them. GCC has no visibility of symbols in the 'asm' and may discard 41049them as unreferenced. It also does not know about side effects of the 41050assembler code, such as modifications to memory or registers. Unlike 41051some compilers, GCC assumes that no changes to general purpose registers 41052occur. This assumption may change in a future release. 41053 41054 To avoid complications from future changes to the semantics and the 41055compatibility issues between compilers, consider replacing basic 'asm' 41056with extended 'asm'. See How to convert from basic asm to extended asm 41057(https://gcc.gnu.org/wiki/ConvertBasicAsmToExtended) for information 41058about how to perform this conversion. 41059 41060 The compiler copies the assembler instructions in a basic 'asm' 41061verbatim to the assembly language output file, without processing 41062dialects or any of the '%' operators that are available with extended 41063'asm'. This results in minor differences between basic 'asm' strings 41064and extended 'asm' templates. For example, to refer to registers you 41065might use '%eax' in basic 'asm' and '%%eax' in extended 'asm'. 41066 41067 On targets such as x86 that support multiple assembler dialects, all 41068basic 'asm' blocks use the assembler dialect specified by the '-masm' 41069command-line option (*note x86 Options::). Basic 'asm' provides no 41070mechanism to provide different assembler strings for different dialects. 41071 41072 For basic 'asm' with non-empty assembler string GCC assumes the 41073assembler block does not change any general purpose registers, but it 41074may read or write any globally accessible variable. 41075 41076 Here is an example of basic 'asm' for i386: 41077 41078 /* Note that this code will not compile with -masm=intel */ 41079 #define DebugBreak() asm("int $3") 41080 41081 41082File: gcc.info, Node: Extended Asm, Next: Constraints, Prev: Basic Asm, Up: Using Assembly Language with C 41083 410846.47.2 Extended Asm - Assembler Instructions with C Expression Operands 41085----------------------------------------------------------------------- 41086 41087With extended 'asm' you can read and write C variables from assembler 41088and perform jumps from assembler code to C labels. Extended 'asm' 41089syntax uses colons (':') to delimit the operand parameters after the 41090assembler template: 41091 41092 asm ASM-QUALIFIERS ( ASSEMBLERTEMPLATE 41093 : OUTPUTOPERANDS 41094 [ : INPUTOPERANDS 41095 [ : CLOBBERS ] ]) 41096 41097 asm ASM-QUALIFIERS ( ASSEMBLERTEMPLATE 41098 : OUTPUTOPERANDS 41099 : INPUTOPERANDS 41100 : CLOBBERS 41101 : GOTOLABELS) 41102 where in the last form, ASM-QUALIFIERS contains 'goto' (and in the 41103first form, not). 41104 41105 The 'asm' keyword is a GNU extension. When writing code that can be 41106compiled with '-ansi' and the various '-std' options, use '__asm__' 41107instead of 'asm' (*note Alternate Keywords::). 41108 41109Qualifiers 41110.......... 41111 41112'volatile' 41113 The typical use of extended 'asm' statements is to manipulate input 41114 values to produce output values. However, your 'asm' statements 41115 may also produce side effects. If so, you may need to use the 41116 'volatile' qualifier to disable certain optimizations. *Note 41117 Volatile::. 41118 41119'inline' 41120 If you use the 'inline' qualifier, then for inlining purposes the 41121 size of the 'asm' statement is taken as the smallest size possible 41122 (*note Size of an asm::). 41123 41124'goto' 41125 This qualifier informs the compiler that the 'asm' statement may 41126 perform a jump to one of the labels listed in the GOTOLABELS. 41127 *Note GotoLabels::. 41128 41129Parameters 41130.......... 41131 41132ASSEMBLERTEMPLATE 41133 This is a literal string that is the template for the assembler 41134 code. It is a combination of fixed text and tokens that refer to 41135 the input, output, and goto parameters. *Note AssemblerTemplate::. 41136 41137OUTPUTOPERANDS 41138 A comma-separated list of the C variables modified by the 41139 instructions in the ASSEMBLERTEMPLATE. An empty list is permitted. 41140 *Note OutputOperands::. 41141 41142INPUTOPERANDS 41143 A comma-separated list of C expressions read by the instructions in 41144 the ASSEMBLERTEMPLATE. An empty list is permitted. *Note 41145 InputOperands::. 41146 41147CLOBBERS 41148 A comma-separated list of registers or other values changed by the 41149 ASSEMBLERTEMPLATE, beyond those listed as outputs. An empty list 41150 is permitted. *Note Clobbers and Scratch Registers::. 41151 41152GOTOLABELS 41153 When you are using the 'goto' form of 'asm', this section contains 41154 the list of all C labels to which the code in the ASSEMBLERTEMPLATE 41155 may jump. *Note GotoLabels::. 41156 41157 'asm' statements may not perform jumps into other 'asm' statements, 41158 only to the listed GOTOLABELS. GCC's optimizers do not know about 41159 other jumps; therefore they cannot take account of them when 41160 deciding how to optimize. 41161 41162 The total number of input + output + goto operands is limited to 30. 41163 41164Remarks 41165....... 41166 41167The 'asm' statement allows you to include assembly instructions directly 41168within C code. This may help you to maximize performance in 41169time-sensitive code or to access assembly instructions that are not 41170readily available to C programs. 41171 41172 Note that extended 'asm' statements must be inside a function. Only 41173basic 'asm' may be outside functions (*note Basic Asm::). Functions 41174declared with the 'naked' attribute also require basic 'asm' (*note 41175Function Attributes::). 41176 41177 While the uses of 'asm' are many and varied, it may help to think of an 41178'asm' statement as a series of low-level instructions that convert input 41179parameters to output parameters. So a simple (if not particularly 41180useful) example for i386 using 'asm' might look like this: 41181 41182 int src = 1; 41183 int dst; 41184 41185 asm ("mov %1, %0\n\t" 41186 "add $1, %0" 41187 : "=r" (dst) 41188 : "r" (src)); 41189 41190 printf("%d\n", dst); 41191 41192 This code copies 'src' to 'dst' and add 1 to 'dst'. 41193 411946.47.2.1 Volatile 41195................. 41196 41197GCC's optimizers sometimes discard 'asm' statements if they determine 41198there is no need for the output variables. Also, the optimizers may 41199move code out of loops if they believe that the code will always return 41200the same result (i.e. none of its input values change between calls). 41201Using the 'volatile' qualifier disables these optimizations. 'asm' 41202statements that have no output operands and 'asm goto' statements, are 41203implicitly volatile. 41204 41205 This i386 code demonstrates a case that does not use (or require) the 41206'volatile' qualifier. If it is performing assertion checking, this code 41207uses 'asm' to perform the validation. Otherwise, 'dwRes' is 41208unreferenced by any code. As a result, the optimizers can discard the 41209'asm' statement, which in turn removes the need for the entire 'DoCheck' 41210routine. By omitting the 'volatile' qualifier when it isn't needed you 41211allow the optimizers to produce the most efficient code possible. 41212 41213 void DoCheck(uint32_t dwSomeValue) 41214 { 41215 uint32_t dwRes; 41216 41217 // Assumes dwSomeValue is not zero. 41218 asm ("bsfl %1,%0" 41219 : "=r" (dwRes) 41220 : "r" (dwSomeValue) 41221 : "cc"); 41222 41223 assert(dwRes > 3); 41224 } 41225 41226 The next example shows a case where the optimizers can recognize that 41227the input ('dwSomeValue') never changes during the execution of the 41228function and can therefore move the 'asm' outside the loop to produce 41229more efficient code. Again, using the 'volatile' qualifier disables 41230this type of optimization. 41231 41232 void do_print(uint32_t dwSomeValue) 41233 { 41234 uint32_t dwRes; 41235 41236 for (uint32_t x=0; x < 5; x++) 41237 { 41238 // Assumes dwSomeValue is not zero. 41239 asm ("bsfl %1,%0" 41240 : "=r" (dwRes) 41241 : "r" (dwSomeValue) 41242 : "cc"); 41243 41244 printf("%u: %u %u\n", x, dwSomeValue, dwRes); 41245 } 41246 } 41247 41248 The following example demonstrates a case where you need to use the 41249'volatile' qualifier. It uses the x86 'rdtsc' instruction, which reads 41250the computer's time-stamp counter. Without the 'volatile' qualifier, 41251the optimizers might assume that the 'asm' block will always return the 41252same value and therefore optimize away the second call. 41253 41254 uint64_t msr; 41255 41256 asm volatile ( "rdtsc\n\t" // Returns the time in EDX:EAX. 41257 "shl $32, %%rdx\n\t" // Shift the upper bits left. 41258 "or %%rdx, %0" // 'Or' in the lower bits. 41259 : "=a" (msr) 41260 : 41261 : "rdx"); 41262 41263 printf("msr: %llx\n", msr); 41264 41265 // Do other work... 41266 41267 // Reprint the timestamp 41268 asm volatile ( "rdtsc\n\t" // Returns the time in EDX:EAX. 41269 "shl $32, %%rdx\n\t" // Shift the upper bits left. 41270 "or %%rdx, %0" // 'Or' in the lower bits. 41271 : "=a" (msr) 41272 : 41273 : "rdx"); 41274 41275 printf("msr: %llx\n", msr); 41276 41277 GCC's optimizers do not treat this code like the non-volatile code in 41278the earlier examples. They do not move it out of loops or omit it on 41279the assumption that the result from a previous call is still valid. 41280 41281 Note that the compiler can move even 'volatile asm' instructions 41282relative to other code, including across jump instructions. For 41283example, on many targets there is a system register that controls the 41284rounding mode of floating-point operations. Setting it with a 'volatile 41285asm' statement, as in the following PowerPC example, does not work 41286reliably. 41287 41288 asm volatile("mtfsf 255, %0" : : "f" (fpenv)); 41289 sum = x + y; 41290 41291 The compiler may move the addition back before the 'volatile asm' 41292statement. To make it work as expected, add an artificial dependency to 41293the 'asm' by referencing a variable in the subsequent code, for example: 41294 41295 asm volatile ("mtfsf 255,%1" : "=X" (sum) : "f" (fpenv)); 41296 sum = x + y; 41297 41298 Under certain circumstances, GCC may duplicate (or remove duplicates 41299of) your assembly code when optimizing. This can lead to unexpected 41300duplicate symbol errors during compilation if your 'asm' code defines 41301symbols or labels. Using '%=' (*note AssemblerTemplate::) may help 41302resolve this problem. 41303 413046.47.2.2 Assembler Template 41305........................... 41306 41307An assembler template is a literal string containing assembler 41308instructions. The compiler replaces tokens in the template that refer 41309to inputs, outputs, and goto labels, and then outputs the resulting 41310string to the assembler. The string can contain any instructions 41311recognized by the assembler, including directives. GCC does not parse 41312the assembler instructions themselves and does not know what they mean 41313or even whether they are valid assembler input. However, it does count 41314the statements (*note Size of an asm::). 41315 41316 You may place multiple assembler instructions together in a single 41317'asm' string, separated by the characters normally used in assembly code 41318for the system. A combination that works in most places is a newline to 41319break the line, plus a tab character to move to the instruction field 41320(written as '\n\t'). Some assemblers allow semicolons as a line 41321separator. However, note that some assembler dialects use semicolons to 41322start a comment. 41323 41324 Do not expect a sequence of 'asm' statements to remain perfectly 41325consecutive after compilation, even when you are using the 'volatile' 41326qualifier. If certain instructions need to remain consecutive in the 41327output, put them in a single multi-instruction 'asm' statement. 41328 41329 Accessing data from C programs without using input/output operands 41330(such as by using global symbols directly from the assembler template) 41331may not work as expected. Similarly, calling functions directly from an 41332assembler template requires a detailed understanding of the target 41333assembler and ABI. 41334 41335 Since GCC does not parse the assembler template, it has no visibility 41336of any symbols it references. This may result in GCC discarding those 41337symbols as unreferenced unless they are also listed as input, output, or 41338goto operands. 41339 41340Special format strings 41341...................... 41342 41343In addition to the tokens described by the input, output, and goto 41344operands, these tokens have special meanings in the assembler template: 41345 41346'%%' 41347 Outputs a single '%' into the assembler code. 41348 41349'%=' 41350 Outputs a number that is unique to each instance of the 'asm' 41351 statement in the entire compilation. This option is useful when 41352 creating local labels and referring to them multiple times in a 41353 single template that generates multiple assembler instructions. 41354 41355'%{' 41356'%|' 41357'%}' 41358 Outputs '{', '|', and '}' characters (respectively) into the 41359 assembler code. When unescaped, these characters have special 41360 meaning to indicate multiple assembler dialects, as described 41361 below. 41362 41363Multiple assembler dialects in 'asm' templates 41364.............................................. 41365 41366On targets such as x86, GCC supports multiple assembler dialects. The 41367'-masm' option controls which dialect GCC uses as its default for inline 41368assembler. The target-specific documentation for the '-masm' option 41369contains the list of supported dialects, as well as the default dialect 41370if the option is not specified. This information may be important to 41371understand, since assembler code that works correctly when compiled 41372using one dialect will likely fail if compiled using another. *Note x86 41373Options::. 41374 41375 If your code needs to support multiple assembler dialects (for example, 41376if you are writing public headers that need to support a variety of 41377compilation options), use constructs of this form: 41378 41379 { dialect0 | dialect1 | dialect2... } 41380 41381 This construct outputs 'dialect0' when using dialect #0 to compile the 41382code, 'dialect1' for dialect #1, etc. If there are fewer alternatives 41383within the braces than the number of dialects the compiler supports, the 41384construct outputs nothing. 41385 41386 For example, if an x86 compiler supports two dialects ('att', 'intel'), 41387an assembler template such as this: 41388 41389 "bt{l %[Offset],%[Base] | %[Base],%[Offset]}; jc %l2" 41390 41391is equivalent to one of 41392 41393 "btl %[Offset],%[Base] ; jc %l2" /* att dialect */ 41394 "bt %[Base],%[Offset]; jc %l2" /* intel dialect */ 41395 41396 Using that same compiler, this code: 41397 41398 "xchg{l}\t{%%}ebx, %1" 41399 41400corresponds to either 41401 41402 "xchgl\t%%ebx, %1" /* att dialect */ 41403 "xchg\tebx, %1" /* intel dialect */ 41404 41405 There is no support for nesting dialect alternatives. 41406 414076.47.2.3 Output Operands 41408........................ 41409 41410An 'asm' statement has zero or more output operands indicating the names 41411of C variables modified by the assembler code. 41412 41413 In this i386 example, 'old' (referred to in the template string as 41414'%0') and '*Base' (as '%1') are outputs and 'Offset' ('%2') is an input: 41415 41416 bool old; 41417 41418 __asm__ ("btsl %2,%1\n\t" // Turn on zero-based bit #Offset in Base. 41419 "sbb %0,%0" // Use the CF to calculate old. 41420 : "=r" (old), "+rm" (*Base) 41421 : "Ir" (Offset) 41422 : "cc"); 41423 41424 return old; 41425 41426 Operands are separated by commas. Each operand has this format: 41427 41428 [ [ASMSYMBOLICNAME] ] CONSTRAINT (CVARIABLENAME) 41429 41430ASMSYMBOLICNAME 41431 Specifies a symbolic name for the operand. Reference the name in 41432 the assembler template by enclosing it in square brackets (i.e. 41433 '%[Value]'). The scope of the name is the 'asm' statement that 41434 contains the definition. Any valid C variable name is acceptable, 41435 including names already defined in the surrounding code. No two 41436 operands within the same 'asm' statement can use the same symbolic 41437 name. 41438 41439 When not using an ASMSYMBOLICNAME, use the (zero-based) position of 41440 the operand in the list of operands in the assembler template. For 41441 example if there are three output operands, use '%0' in the 41442 template to refer to the first, '%1' for the second, and '%2' for 41443 the third. 41444 41445CONSTRAINT 41446 A string constant specifying constraints on the placement of the 41447 operand; *Note Constraints::, for details. 41448 41449 Output constraints must begin with either '=' (a variable 41450 overwriting an existing value) or '+' (when reading and writing). 41451 When using '=', do not assume the location contains the existing 41452 value on entry to the 'asm', except when the operand is tied to an 41453 input; *note Input Operands: InputOperands. 41454 41455 After the prefix, there must be one or more additional constraints 41456 (*note Constraints::) that describe where the value resides. 41457 Common constraints include 'r' for register and 'm' for memory. 41458 When you list more than one possible location (for example, 41459 '"=rm"'), the compiler chooses the most efficient one based on the 41460 current context. If you list as many alternates as the 'asm' 41461 statement allows, you permit the optimizers to produce the best 41462 possible code. If you must use a specific register, but your 41463 Machine Constraints do not provide sufficient control to select the 41464 specific register you want, local register variables may provide a 41465 solution (*note Local Register Variables::). 41466 41467CVARIABLENAME 41468 Specifies a C lvalue expression to hold the output, typically a 41469 variable name. The enclosing parentheses are a required part of 41470 the syntax. 41471 41472 When the compiler selects the registers to use to represent the output 41473operands, it does not use any of the clobbered registers (*note Clobbers 41474and Scratch Registers::). 41475 41476 Output operand expressions must be lvalues. The compiler cannot check 41477whether the operands have data types that are reasonable for the 41478instruction being executed. For output expressions that are not 41479directly addressable (for example a bit-field), the constraint must 41480allow a register. In that case, GCC uses the register as the output of 41481the 'asm', and then stores that register into the output. 41482 41483 Operands using the '+' constraint modifier count as two operands (that 41484is, both as input and output) towards the total maximum of 30 operands 41485per 'asm' statement. 41486 41487 Use the '&' constraint modifier (*note Modifiers::) on all output 41488operands that must not overlap an input. Otherwise, GCC may allocate 41489the output operand in the same register as an unrelated input operand, 41490on the assumption that the assembler code consumes its inputs before 41491producing outputs. This assumption may be false if the assembler code 41492actually consists of more than one instruction. 41493 41494 The same problem can occur if one output parameter (A) allows a 41495register constraint and another output parameter (B) allows a memory 41496constraint. The code generated by GCC to access the memory address in B 41497can contain registers which _might_ be shared by A, and GCC considers 41498those registers to be inputs to the asm. As above, GCC assumes that 41499such input registers are consumed before any outputs are written. This 41500assumption may result in incorrect behavior if the 'asm' statement 41501writes to A before using B. Combining the '&' modifier with the 41502register constraint on A ensures that modifying A does not affect the 41503address referenced by B. Otherwise, the location of B is undefined if A 41504is modified before using B. 41505 41506 'asm' supports operand modifiers on operands (for example '%k2' instead 41507of simply '%2'). Typically these qualifiers are hardware dependent. 41508The list of supported modifiers for x86 is found at *note x86 Operand 41509modifiers: x86Operandmodifiers. 41510 41511 If the C code that follows the 'asm' makes no use of any of the output 41512operands, use 'volatile' for the 'asm' statement to prevent the 41513optimizers from discarding the 'asm' statement as unneeded (see *note 41514Volatile::). 41515 41516 This code makes no use of the optional ASMSYMBOLICNAME. Therefore it 41517references the first output operand as '%0' (were there a second, it 41518would be '%1', etc). The number of the first input operand is one 41519greater than that of the last output operand. In this i386 example, 41520that makes 'Mask' referenced as '%1': 41521 41522 uint32_t Mask = 1234; 41523 uint32_t Index; 41524 41525 asm ("bsfl %1, %0" 41526 : "=r" (Index) 41527 : "r" (Mask) 41528 : "cc"); 41529 41530 That code overwrites the variable 'Index' ('='), placing the value in a 41531register ('r'). Using the generic 'r' constraint instead of a 41532constraint for a specific register allows the compiler to pick the 41533register to use, which can result in more efficient code. This may not 41534be possible if an assembler instruction requires a specific register. 41535 41536 The following i386 example uses the ASMSYMBOLICNAME syntax. It 41537produces the same result as the code above, but some may consider it 41538more readable or more maintainable since reordering index numbers is not 41539necessary when adding or removing operands. The names 'aIndex' and 41540'aMask' are only used in this example to emphasize which names get used 41541where. It is acceptable to reuse the names 'Index' and 'Mask'. 41542 41543 uint32_t Mask = 1234; 41544 uint32_t Index; 41545 41546 asm ("bsfl %[aMask], %[aIndex]" 41547 : [aIndex] "=r" (Index) 41548 : [aMask] "r" (Mask) 41549 : "cc"); 41550 41551 Here are some more examples of output operands. 41552 41553 uint32_t c = 1; 41554 uint32_t d; 41555 uint32_t *e = &c; 41556 41557 asm ("mov %[e], %[d]" 41558 : [d] "=rm" (d) 41559 : [e] "rm" (*e)); 41560 41561 Here, 'd' may either be in a register or in memory. Since the compiler 41562might already have the current value of the 'uint32_t' location pointed 41563to by 'e' in a register, you can enable it to choose the best location 41564for 'd' by specifying both constraints. 41565 415666.47.2.4 Flag Output Operands 41567............................. 41568 41569Some targets have a special register that holds the "flags" for the 41570result of an operation or comparison. Normally, the contents of that 41571register are either unmodifed by the asm, or the 'asm' statement is 41572considered to clobber the contents. 41573 41574 On some targets, a special form of output operand exists by which 41575conditions in the flags register may be outputs of the asm. The set of 41576conditions supported are target specific, but the general rule is that 41577the output variable must be a scalar integer, and the value is boolean. 41578When supported, the target defines the preprocessor symbol 41579'__GCC_ASM_FLAG_OUTPUTS__'. 41580 41581 Because of the special nature of the flag output operands, the 41582constraint may not include alternatives. 41583 41584 Most often, the target has only one flags register, and thus is an 41585implied operand of many instructions. In this case, the operand should 41586not be referenced within the assembler template via '%0' etc, as there's 41587no corresponding text in the assembly language. 41588 41589ARM 41590AArch64 41591 The flag output constraints for the ARM family are of the form 41592 '=@ccCOND' where COND is one of the standard conditions defined in 41593 the ARM ARM for 'ConditionHolds'. 41594 41595 'eq' 41596 Z flag set, or equal 41597 'ne' 41598 Z flag clear or not equal 41599 'cs' 41600 'hs' 41601 C flag set or unsigned greater than equal 41602 'cc' 41603 'lo' 41604 C flag clear or unsigned less than 41605 'mi' 41606 N flag set or "minus" 41607 'pl' 41608 N flag clear or "plus" 41609 'vs' 41610 V flag set or signed overflow 41611 'vc' 41612 V flag clear 41613 'hi' 41614 unsigned greater than 41615 'ls' 41616 unsigned less than equal 41617 'ge' 41618 signed greater than equal 41619 'lt' 41620 signed less than 41621 'gt' 41622 signed greater than 41623 'le' 41624 signed less than equal 41625 41626 The flag output constraints are not supported in thumb1 mode. 41627 41628x86 family 41629 The flag output constraints for the x86 family are of the form 41630 '=@ccCOND' where COND is one of the standard conditions defined in 41631 the ISA manual for 'jCC' or 'setCC'. 41632 41633 'a' 41634 "above" or unsigned greater than 41635 'ae' 41636 "above or equal" or unsigned greater than or equal 41637 'b' 41638 "below" or unsigned less than 41639 'be' 41640 "below or equal" or unsigned less than or equal 41641 'c' 41642 carry flag set 41643 'e' 41644 'z' 41645 "equal" or zero flag set 41646 'g' 41647 signed greater than 41648 'ge' 41649 signed greater than or equal 41650 'l' 41651 signed less than 41652 'le' 41653 signed less than or equal 41654 'o' 41655 overflow flag set 41656 'p' 41657 parity flag set 41658 's' 41659 sign flag set 41660 'na' 41661 'nae' 41662 'nb' 41663 'nbe' 41664 'nc' 41665 'ne' 41666 'ng' 41667 'nge' 41668 'nl' 41669 'nle' 41670 'no' 41671 'np' 41672 'ns' 41673 'nz' 41674 "not" FLAG, or inverted versions of those above 41675 416766.47.2.5 Input Operands 41677....................... 41678 41679Input operands make values from C variables and expressions available to 41680the assembly code. 41681 41682 Operands are separated by commas. Each operand has this format: 41683 41684 [ [ASMSYMBOLICNAME] ] CONSTRAINT (CEXPRESSION) 41685 41686ASMSYMBOLICNAME 41687 Specifies a symbolic name for the operand. Reference the name in 41688 the assembler template by enclosing it in square brackets (i.e. 41689 '%[Value]'). The scope of the name is the 'asm' statement that 41690 contains the definition. Any valid C variable name is acceptable, 41691 including names already defined in the surrounding code. No two 41692 operands within the same 'asm' statement can use the same symbolic 41693 name. 41694 41695 When not using an ASMSYMBOLICNAME, use the (zero-based) position of 41696 the operand in the list of operands in the assembler template. For 41697 example if there are two output operands and three inputs, use '%2' 41698 in the template to refer to the first input operand, '%3' for the 41699 second, and '%4' for the third. 41700 41701CONSTRAINT 41702 A string constant specifying constraints on the placement of the 41703 operand; *Note Constraints::, for details. 41704 41705 Input constraint strings may not begin with either '=' or '+'. 41706 When you list more than one possible location (for example, 41707 '"irm"'), the compiler chooses the most efficient one based on the 41708 current context. If you must use a specific register, but your 41709 Machine Constraints do not provide sufficient control to select the 41710 specific register you want, local register variables may provide a 41711 solution (*note Local Register Variables::). 41712 41713 Input constraints can also be digits (for example, '"0"'). This 41714 indicates that the specified input must be in the same place as the 41715 output constraint at the (zero-based) index in the output 41716 constraint list. When using ASMSYMBOLICNAME syntax for the output 41717 operands, you may use these names (enclosed in brackets '[]') 41718 instead of digits. 41719 41720CEXPRESSION 41721 This is the C variable or expression being passed to the 'asm' 41722 statement as input. The enclosing parentheses are a required part 41723 of the syntax. 41724 41725 When the compiler selects the registers to use to represent the input 41726operands, it does not use any of the clobbered registers (*note Clobbers 41727and Scratch Registers::). 41728 41729 If there are no output operands but there are input operands, place two 41730consecutive colons where the output operands would go: 41731 41732 __asm__ ("some instructions" 41733 : /* No outputs. */ 41734 : "r" (Offset / 8)); 41735 41736 *Warning:* Do _not_ modify the contents of input-only operands (except 41737for inputs tied to outputs). The compiler assumes that on exit from the 41738'asm' statement these operands contain the same values as they had 41739before executing the statement. It is _not_ possible to use clobbers to 41740inform the compiler that the values in these inputs are changing. One 41741common work-around is to tie the changing input variable to an output 41742variable that never gets used. Note, however, that if the code that 41743follows the 'asm' statement makes no use of any of the output operands, 41744the GCC optimizers may discard the 'asm' statement as unneeded (see 41745*note Volatile::). 41746 41747 'asm' supports operand modifiers on operands (for example '%k2' instead 41748of simply '%2'). Typically these qualifiers are hardware dependent. 41749The list of supported modifiers for x86 is found at *note x86 Operand 41750modifiers: x86Operandmodifiers. 41751 41752 In this example using the fictitious 'combine' instruction, the 41753constraint '"0"' for input operand 1 says that it must occupy the same 41754location as output operand 0. Only input operands may use numbers in 41755constraints, and they must each refer to an output operand. Only a 41756number (or the symbolic assembler name) in the constraint can guarantee 41757that one operand is in the same place as another. The mere fact that 41758'foo' is the value of both operands is not enough to guarantee that they 41759are in the same place in the generated assembler code. 41760 41761 asm ("combine %2, %0" 41762 : "=r" (foo) 41763 : "0" (foo), "g" (bar)); 41764 41765 Here is an example using symbolic names. 41766 41767 asm ("cmoveq %1, %2, %[result]" 41768 : [result] "=r"(result) 41769 : "r" (test), "r" (new), "[result]" (old)); 41770 417716.47.2.6 Clobbers and Scratch Registers 41772....................................... 41773 41774While the compiler is aware of changes to entries listed in the output 41775operands, the inline 'asm' code may modify more than just the outputs. 41776For example, calculations may require additional registers, or the 41777processor may overwrite a register as a side effect of a particular 41778assembler instruction. In order to inform the compiler of these 41779changes, list them in the clobber list. Clobber list items are either 41780register names or the special clobbers (listed below). Each clobber 41781list item is a string constant enclosed in double quotes and separated 41782by commas. 41783 41784 Clobber descriptions may not in any way overlap with an input or output 41785operand. For example, you may not have an operand describing a register 41786class with one member when listing that register in the clobber list. 41787Variables declared to live in specific registers (*note Explicit 41788Register Variables::) and used as 'asm' input or output operands must 41789have no part mentioned in the clobber description. In particular, there 41790is no way to specify that input operands get modified without also 41791specifying them as output operands. 41792 41793 When the compiler selects which registers to use to represent input and 41794output operands, it does not use any of the clobbered registers. As a 41795result, clobbered registers are available for any use in the assembler 41796code. 41797 41798 Another restriction is that the clobber list should not contain the 41799stack pointer register. This is because the compiler requires the value 41800of the stack pointer to be the same after an 'asm' statement as it was 41801on entry to the statement. However, previous versions of GCC did not 41802enforce this rule and allowed the stack pointer to appear in the list, 41803with unclear semantics. This behavior is deprecated and listing the 41804stack pointer may become an error in future versions of GCC. 41805 41806 Here is a realistic example for the VAX showing the use of clobbered 41807registers: 41808 41809 asm volatile ("movc3 %0, %1, %2" 41810 : /* No outputs. */ 41811 : "g" (from), "g" (to), "g" (count) 41812 : "r0", "r1", "r2", "r3", "r4", "r5", "memory"); 41813 41814 Also, there are two special clobber arguments: 41815 41816'"cc"' 41817 The '"cc"' clobber indicates that the assembler code modifies the 41818 flags register. On some machines, GCC represents the condition 41819 codes as a specific hardware register; '"cc"' serves to name this 41820 register. On other machines, condition code handling is different, 41821 and specifying '"cc"' has no effect. But it is valid no matter 41822 what the target. 41823 41824'"memory"' 41825 The '"memory"' clobber tells the compiler that the assembly code 41826 performs memory reads or writes to items other than those listed in 41827 the input and output operands (for example, accessing the memory 41828 pointed to by one of the input parameters). To ensure memory 41829 contains correct values, GCC may need to flush specific register 41830 values to memory before executing the 'asm'. Further, the compiler 41831 does not assume that any values read from memory before an 'asm' 41832 remain unchanged after that 'asm'; it reloads them as needed. 41833 Using the '"memory"' clobber effectively forms a read/write memory 41834 barrier for the compiler. 41835 41836 Note that this clobber does not prevent the _processor_ from doing 41837 speculative reads past the 'asm' statement. To prevent that, you 41838 need processor-specific fence instructions. 41839 41840 Flushing registers to memory has performance implications and may be an 41841issue for time-sensitive code. You can provide better information to 41842GCC to avoid this, as shown in the following examples. At a minimum, 41843aliasing rules allow GCC to know what memory _doesn't_ need to be 41844flushed. 41845 41846 Here is a fictitious sum of squares instruction, that takes two 41847pointers to floating point values in memory and produces a floating 41848point register output. Notice that 'x', and 'y' both appear twice in 41849the 'asm' parameters, once to specify memory accessed, and once to 41850specify a base register used by the 'asm'. You won't normally be 41851wasting a register by doing this as GCC can use the same register for 41852both purposes. However, it would be foolish to use both '%1' and '%3' 41853for 'x' in this 'asm' and expect them to be the same. In fact, '%3' may 41854well not be a register. It might be a symbolic memory reference to the 41855object pointed to by 'x'. 41856 41857 asm ("sumsq %0, %1, %2" 41858 : "+f" (result) 41859 : "r" (x), "r" (y), "m" (*x), "m" (*y)); 41860 41861 Here is a fictitious '*z++ = *x++ * *y++' instruction. Notice that the 41862'x', 'y' and 'z' pointer registers must be specified as input/output 41863because the 'asm' modifies them. 41864 41865 asm ("vecmul %0, %1, %2" 41866 : "+r" (z), "+r" (x), "+r" (y), "=m" (*z) 41867 : "m" (*x), "m" (*y)); 41868 41869 An x86 example where the string memory argument is of unknown length. 41870 41871 asm("repne scasb" 41872 : "=c" (count), "+D" (p) 41873 : "m" (*(const char (*)[]) p), "0" (-1), "a" (0)); 41874 41875 If you know the above will only be reading a ten byte array then you 41876could instead use a memory input like: '"m" (*(const char (*)[10]) p)'. 41877 41878 Here is an example of a PowerPC vector scale implemented in assembly, 41879complete with vector and condition code clobbers, and some initialized 41880offset registers that are unchanged by the 'asm'. 41881 41882 void 41883 dscal (size_t n, double *x, double alpha) 41884 { 41885 asm ("/* lots of asm here */" 41886 : "+m" (*(double (*)[n]) x), "+&r" (n), "+b" (x) 41887 : "d" (alpha), "b" (32), "b" (48), "b" (64), 41888 "b" (80), "b" (96), "b" (112) 41889 : "cr0", 41890 "vs32","vs33","vs34","vs35","vs36","vs37","vs38","vs39", 41891 "vs40","vs41","vs42","vs43","vs44","vs45","vs46","vs47"); 41892 } 41893 41894 Rather than allocating fixed registers via clobbers to provide scratch 41895registers for an 'asm' statement, an alternative is to define a variable 41896and make it an early-clobber output as with 'a2' and 'a3' in the example 41897below. This gives the compiler register allocator more freedom. You 41898can also define a variable and make it an output tied to an input as 41899with 'a0' and 'a1', tied respectively to 'ap' and 'lda'. Of course, 41900with tied outputs your 'asm' can't use the input value after modifying 41901the output register since they are one and the same register. What's 41902more, if you omit the early-clobber on the output, it is possible that 41903GCC might allocate the same register to another of the inputs if GCC 41904could prove they had the same value on entry to the 'asm'. This is why 41905'a1' has an early-clobber. Its tied input, 'lda' might conceivably be 41906known to have the value 16 and without an early-clobber share the same 41907register as '%11'. On the other hand, 'ap' can't be the same as any of 41908the other inputs, so an early-clobber on 'a0' is not needed. It is also 41909not desirable in this case. An early-clobber on 'a0' would cause GCC to 41910allocate a separate register for the '"m" (*(const double (*)[]) ap)' 41911input. Note that tying an input to an output is the way to set up an 41912initialized temporary register modified by an 'asm' statement. An input 41913not tied to an output is assumed by GCC to be unchanged, for example 41914'"b" (16)' below sets up '%11' to 16, and GCC might use that register in 41915following code if the value 16 happened to be needed. You can even use 41916a normal 'asm' output for a scratch if all inputs that might share the 41917same register are consumed before the scratch is used. The VSX 41918registers clobbered by the 'asm' statement could have used this 41919technique except for GCC's limit on the number of 'asm' parameters. 41920 41921 static void 41922 dgemv_kernel_4x4 (long n, const double *ap, long lda, 41923 const double *x, double *y, double alpha) 41924 { 41925 double *a0; 41926 double *a1; 41927 double *a2; 41928 double *a3; 41929 41930 __asm__ 41931 ( 41932 /* lots of asm here */ 41933 "#n=%1 ap=%8=%12 lda=%13 x=%7=%10 y=%0=%2 alpha=%9 o16=%11\n" 41934 "#a0=%3 a1=%4 a2=%5 a3=%6" 41935 : 41936 "+m" (*(double (*)[n]) y), 41937 "+&r" (n), // 1 41938 "+b" (y), // 2 41939 "=b" (a0), // 3 41940 "=&b" (a1), // 4 41941 "=&b" (a2), // 5 41942 "=&b" (a3) // 6 41943 : 41944 "m" (*(const double (*)[n]) x), 41945 "m" (*(const double (*)[]) ap), 41946 "d" (alpha), // 9 41947 "r" (x), // 10 41948 "b" (16), // 11 41949 "3" (ap), // 12 41950 "4" (lda) // 13 41951 : 41952 "cr0", 41953 "vs32","vs33","vs34","vs35","vs36","vs37", 41954 "vs40","vs41","vs42","vs43","vs44","vs45","vs46","vs47" 41955 ); 41956 } 41957 419586.47.2.7 Goto Labels 41959.................... 41960 41961'asm goto' allows assembly code to jump to one or more C labels. The 41962GOTOLABELS section in an 'asm goto' statement contains a comma-separated 41963list of all C labels to which the assembler code may jump. GCC assumes 41964that 'asm' execution falls through to the next statement (if this is not 41965the case, consider using the '__builtin_unreachable' intrinsic after the 41966'asm' statement). Optimization of 'asm goto' may be improved by using 41967the 'hot' and 'cold' label attributes (*note Label Attributes::). 41968 41969 If the assembler code does modify anything, use the '"memory"' clobber 41970to force the optimizers to flush all register values to memory and 41971reload them if necessary after the 'asm' statement. 41972 41973 Also note that an 'asm goto' statement is always implicitly considered 41974volatile. 41975 41976 Be careful when you set output operands inside 'asm goto' only on some 41977possible control flow paths. If you don't set up the output on given 41978path and never use it on this path, it is okay. Otherwise, you should 41979use '+' constraint modifier meaning that the operand is input and output 41980one. With this modifier you will have the correct values on all 41981possible paths from the 'asm goto'. 41982 41983 To reference a label in the assembler template, prefix it with '%l' 41984(lowercase 'L') followed by its (zero-based) position in GOTOLABELS plus 41985the number of input and output operands. Output operand with constraint 41986modifier '+' is counted as two operands because it is considered as one 41987output and one input operand. For example, if the 'asm' has three 41988inputs, one output operand with constraint modifier '+' and one output 41989operand with constraint modifier '=' and references two labels, refer to 41990the first label as '%l6' and the second as '%l7'). 41991 41992 Alternately, you can reference labels using the actual C label name 41993enclosed in brackets. For example, to reference a label named 'carry', 41994you can use '%l[carry]'. The label must still be listed in the 41995GOTOLABELS section when using this approach. It is better to use the 41996named references for labels as in this case you can avoid counting input 41997and output operands and special treatment of output operands with 41998constraint modifier '+'. 41999 42000 Here is an example of 'asm goto' for i386: 42001 42002 asm goto ( 42003 "btl %1, %0\n\t" 42004 "jc %l2" 42005 : /* No outputs. */ 42006 : "r" (p1), "r" (p2) 42007 : "cc" 42008 : carry); 42009 42010 return 0; 42011 42012 carry: 42013 return 1; 42014 42015 The following example shows an 'asm goto' that uses a memory clobber. 42016 42017 int frob(int x) 42018 { 42019 int y; 42020 asm goto ("frob %%r5, %1; jc %l[error]; mov (%2), %%r5" 42021 : /* No outputs. */ 42022 : "r"(x), "r"(&y) 42023 : "r5", "memory" 42024 : error); 42025 return y; 42026 error: 42027 return -1; 42028 } 42029 42030 The following example shows an 'asm goto' that uses an output. 42031 42032 int foo(int count) 42033 { 42034 asm goto ("dec %0; jb %l[stop]" 42035 : "+r" (count) 42036 : 42037 : 42038 : stop); 42039 return count; 42040 stop: 42041 return 0; 42042 } 42043 42044 The following artificial example shows an 'asm goto' that sets up an 42045output only on one path inside the 'asm goto'. Usage of constraint 42046modifier '=' instead of '+' would be wrong as 'factor' is used on all 42047paths from the 'asm goto'. 42048 42049 int foo(int inp) 42050 { 42051 int factor = 0; 42052 asm goto ("cmp %1, 10; jb %l[lab]; mov 2, %0" 42053 : "+r" (factor) 42054 : "r" (inp) 42055 : 42056 : lab); 42057 lab: 42058 return inp * factor; /* return 2 * inp or 0 if inp < 10 */ 42059 } 42060 420616.47.2.8 x86 Operand Modifiers 42062.............................. 42063 42064References to input, output, and goto operands in the assembler template 42065of extended 'asm' statements can use modifiers to affect the way the 42066operands are formatted in the code output to the assembler. For 42067example, the following code uses the 'h' and 'b' modifiers for x86: 42068 42069 uint16_t num; 42070 asm volatile ("xchg %h0, %b0" : "+a" (num) ); 42071 42072These modifiers generate this assembler code: 42073 42074 xchg %ah, %al 42075 42076 The rest of this discussion uses the following code for illustrative 42077purposes. 42078 42079 int main() 42080 { 42081 int iInt = 1; 42082 42083 top: 42084 42085 asm volatile goto ("some assembler instructions here" 42086 : /* No outputs. */ 42087 : "q" (iInt), "X" (sizeof(unsigned char) + 1), "i" (42) 42088 : /* No clobbers. */ 42089 : top); 42090 } 42091 42092 With no modifiers, this is what the output from the operands would be 42093for the 'att' and 'intel' dialects of assembler: 42094 42095Operand 'att' 'intel' 42096----------------------------------- 42097'%0' '%eax' 'eax' 42098'%1' '$2' '2' 42099'%3' '$.L3' 'OFFSET 42100 FLAT:.L3' 42101'%4' '$8' '8' 42102'%5' '%xmm0''xmm0' 42103'%7' '$0' '0' 42104 42105 The table below shows the list of supported modifiers and their 42106effects. 42107 42108Modifier Description Operand 'att' 'intel' 42109------------------------------------------------------------------------------------ 42110'A' Print an absolute memory reference. '%A0' '*%rax' 'rax' 42111'b' Print the QImode name of the register. '%b0' '%al' 'al' 42112'B' print the opcode suffix of b. '%B0' 'b' 42113'c' Require a constant operand and print the '%c1' '2' '2' 42114 constant expression with no punctuation. 42115'd' print duplicated register operand for AVX '%d5' '%xmm0, 'xmm0, 42116 instruction. %xmm0' xmm0' 42117'E' Print the address in Double Integer '%E1' '%(rax)''[rax]' 42118 (DImode) mode (8 bytes) when the target is 42119 64-bit. Otherwise mode is unspecified 42120 (VOIDmode). 42121'g' Print the V16SFmode name of the register. '%g0' '%zmm0' 'zmm0' 42122'h' Print the QImode name for a "high" '%h0' '%ah' 'ah' 42123 register. 42124'H' Add 8 bytes to an offsettable memory '%H0' '8(%rax)''8[rax]' 42125 reference. Useful when accessing the high 42126 8 bytes of SSE values. For a memref in 42127 (%rax), it generates 42128'k' Print the SImode name of the register. '%k0' '%eax' 'eax' 42129'l' Print the label name with no punctuation. '%l3' '.L3' '.L3' 42130'L' print the opcode suffix of l. '%L0' 'l' 42131'N' print maskz. '%N7' '{z}' '{z}' 42132'p' Print raw symbol name (without '%p2' '42' '42' 42133 syntax-specific prefixes). 42134'P' If used for a function, print the PLT 42135 suffix and generate PIC code. For 42136 example, emit 'foo@PLT' instead of 'foo' 42137 for the function foo(). If used for a 42138 constant, drop all syntax-specific 42139 prefixes and issue the bare constant. See 42140 'p' above. 42141'q' Print the DImode name of the register. '%q0' '%rax' 'rax' 42142'Q' print the opcode suffix of q. '%Q0' 'q' 42143'R' print embedded rounding and sae. '%R4' '{rn-sae},', 42144 ' {rn-sae}' 42145'r' print only sae. '%r4' '{sae}, ', 42146 ' {sae}' 42147's' print a shift double count, followed by '%s1' '$2, '2, ' 42148 the assemblers argument delimiterprint the ' 42149 opcode suffix of s. 42150'S' print the opcode suffix of s. '%S0' 's' 42151't' print the V8SFmode name of the register. '%t5' '%ymm0' 'ymm0' 42152'T' print the opcode suffix of t. '%T0' 't' 42153'V' print naked full integer register name '%V0' 'eax' 'eax' 42154 without %. 42155'w' Print the HImode name of the register. '%w0' '%ax' 'ax' 42156'W' print the opcode suffix of w. '%W0' 'w' 42157'x' print the V4SFmode name of the register. '%x5' '%xmm0' 'xmm0' 42158'y' print "st(0)" instead of "st" as a '%y6' '%st(0)''st(0)' 42159 register. 42160'z' Print the opcode suffix for the size of '%z0' 'l' 42161 the current integer operand (one of 42162 'b'/'w'/'l'/'q'). 42163'Z' Like 'z', with special suffixes for x87 42164 instructions. 42165 421666.47.2.9 x86 Floating-Point 'asm' Operands 42167.......................................... 42168 42169On x86 targets, there are several rules on the usage of stack-like 42170registers in the operands of an 'asm'. These rules apply only to the 42171operands that are stack-like registers: 42172 42173 1. Given a set of input registers that die in an 'asm', it is 42174 necessary to know which are implicitly popped by the 'asm', and 42175 which must be explicitly popped by GCC. 42176 42177 An input register that is implicitly popped by the 'asm' must be 42178 explicitly clobbered, unless it is constrained to match an output 42179 operand. 42180 42181 2. For any input register that is implicitly popped by an 'asm', it is 42182 necessary to know how to adjust the stack to compensate for the 42183 pop. If any non-popped input is closer to the top of the reg-stack 42184 than the implicitly popped register, it would not be possible to 42185 know what the stack looked like--it's not clear how the rest of the 42186 stack "slides up". 42187 42188 All implicitly popped input registers must be closer to the top of 42189 the reg-stack than any input that is not implicitly popped. 42190 42191 It is possible that if an input dies in an 'asm', the compiler 42192 might use the input register for an output reload. Consider this 42193 example: 42194 42195 asm ("foo" : "=t" (a) : "f" (b)); 42196 42197 This code says that input 'b' is not popped by the 'asm', and that 42198 the 'asm' pushes a result onto the reg-stack, i.e., the stack is 42199 one deeper after the 'asm' than it was before. But, it is possible 42200 that reload may think that it can use the same register for both 42201 the input and the output. 42202 42203 To prevent this from happening, if any input operand uses the 'f' 42204 constraint, all output register constraints must use the '&' 42205 early-clobber modifier. 42206 42207 The example above is correctly written as: 42208 42209 asm ("foo" : "=&t" (a) : "f" (b)); 42210 42211 3. Some operands need to be in particular places on the stack. All 42212 output operands fall in this category--GCC has no other way to know 42213 which registers the outputs appear in unless you indicate this in 42214 the constraints. 42215 42216 Output operands must specifically indicate which register an output 42217 appears in after an 'asm'. '=f' is not allowed: the operand 42218 constraints must select a class with a single register. 42219 42220 4. Output operands may not be "inserted" between existing stack 42221 registers. Since no 387 opcode uses a read/write operand, all 42222 output operands are dead before the 'asm', and are pushed by the 42223 'asm'. It makes no sense to push anywhere but the top of the 42224 reg-stack. 42225 42226 Output operands must start at the top of the reg-stack: output 42227 operands may not "skip" a register. 42228 42229 5. Some 'asm' statements may need extra stack space for internal 42230 calculations. This can be guaranteed by clobbering stack registers 42231 unrelated to the inputs and outputs. 42232 42233 This 'asm' takes one input, which is internally popped, and produces 42234two outputs. 42235 42236 asm ("fsincos" : "=t" (cos), "=u" (sin) : "0" (inp)); 42237 42238This 'asm' takes two inputs, which are popped by the 'fyl2xp1' opcode, 42239and replaces them with one output. The 'st(1)' clobber is necessary for 42240the compiler to know that 'fyl2xp1' pops both inputs. 42241 42242 asm ("fyl2xp1" : "=t" (result) : "0" (x), "u" (y) : "st(1)"); 42243 422446.47.2.10 MSP430 Operand Modifiers 42245.................................. 42246 42247The list below describes the supported modifiers and their effects for 42248MSP430. 42249 42250ModifierDescription 42251-------------------------------------------------------------------------- 42252'A' Select low 16-bits of the constant/register/memory operand. 42253'B' Select high 16-bits of the constant/register/memory operand. 42254'C' Select bits 32-47 of the constant/register/memory operand. 42255'D' Select bits 48-63 of the constant/register/memory operand. 42256'H' Equivalent to 'B' (for backwards compatibility). 42257'I' Print the inverse (logical 'NOT') of the constant value. 42258'J' Print an integer without a '#' prefix. 42259'L' Equivalent to 'A' (for backwards compatibility). 42260'O' Offset of the current frame from the top of the stack. 42261'Q' Use the 'A' instruction postfix. 42262'R' Inverse of condition code, for unsigned comparisons. 42263'W' Subtract 16 from the constant value. 42264'X' Use the 'X' instruction postfix. 42265'Y' Subtract 4 from the constant value. 42266'Z' Subtract 1 from the constant value. 42267'b' Append '.B', '.W' or '.A' to the instruction, depending on the 42268 mode. 42269'd' Offset 1 byte of a memory reference or constant value. 42270'e' Offset 3 bytes of a memory reference or constant value. 42271'f' Offset 5 bytes of a memory reference or constant value. 42272'g' Offset 7 bytes of a memory reference or constant value. 42273'p' Print the value of 2, raised to the power of the given 42274 constant. Used to select the specified bit position. 42275'r' Inverse of condition code, for signed comparisons. 42276'x' Equivialent to 'X', but only for pointers. 42277 42278 42279File: gcc.info, Node: Constraints, Next: Asm Labels, Prev: Extended Asm, Up: Using Assembly Language with C 42280 422816.47.3 Constraints for 'asm' Operands 42282------------------------------------- 42283 42284Here are specific details on what constraint letters you can use with 42285'asm' operands. Constraints can say whether an operand may be in a 42286register, and which kinds of register; whether the operand can be a 42287memory reference, and which kinds of address; whether the operand may be 42288an immediate constant, and which possible values it may have. 42289Constraints can also require two operands to match. Side-effects aren't 42290allowed in operands of inline 'asm', unless '<' or '>' constraints are 42291used, because there is no guarantee that the side effects will happen 42292exactly once in an instruction that can update the addressing register. 42293 42294* Menu: 42295 42296* Simple Constraints:: Basic use of constraints. 42297* Multi-Alternative:: When an insn has two alternative constraint-patterns. 42298* Modifiers:: More precise control over effects of constraints. 42299* Machine Constraints:: Special constraints for some particular machines. 42300 42301 42302File: gcc.info, Node: Simple Constraints, Next: Multi-Alternative, Up: Constraints 42303 423046.47.3.1 Simple Constraints 42305........................... 42306 42307The simplest kind of constraint is a string full of letters, each of 42308which describes one kind of operand that is permitted. Here are the 42309letters that are allowed: 42310 42311whitespace 42312 Whitespace characters are ignored and can be inserted at any 42313 position except the first. This enables each alternative for 42314 different operands to be visually aligned in the machine 42315 description even if they have different number of constraints and 42316 modifiers. 42317 42318'm' 42319 A memory operand is allowed, with any kind of address that the 42320 machine supports in general. Note that the letter used for the 42321 general memory constraint can be re-defined by a back end using the 42322 'TARGET_MEM_CONSTRAINT' macro. 42323 42324'o' 42325 A memory operand is allowed, but only if the address is 42326 "offsettable". This means that adding a small integer (actually, 42327 the width in bytes of the operand, as determined by its machine 42328 mode) may be added to the address and the result is also a valid 42329 memory address. 42330 42331 For example, an address which is constant is offsettable; so is an 42332 address that is the sum of a register and a constant (as long as a 42333 slightly larger constant is also within the range of 42334 address-offsets supported by the machine); but an autoincrement or 42335 autodecrement address is not offsettable. More complicated 42336 indirect/indexed addresses may or may not be offsettable depending 42337 on the other addressing modes that the machine supports. 42338 42339 Note that in an output operand which can be matched by another 42340 operand, the constraint letter 'o' is valid only when accompanied 42341 by both '<' (if the target machine has predecrement addressing) and 42342 '>' (if the target machine has preincrement addressing). 42343 42344'V' 42345 A memory operand that is not offsettable. In other words, anything 42346 that would fit the 'm' constraint but not the 'o' constraint. 42347 42348'<' 42349 A memory operand with autodecrement addressing (either predecrement 42350 or postdecrement) is allowed. In inline 'asm' this constraint is 42351 only allowed if the operand is used exactly once in an instruction 42352 that can handle the side effects. Not using an operand with '<' in 42353 constraint string in the inline 'asm' pattern at all or using it in 42354 multiple instructions isn't valid, because the side effects 42355 wouldn't be performed or would be performed more than once. 42356 Furthermore, on some targets the operand with '<' in constraint 42357 string must be accompanied by special instruction suffixes like 42358 '%U0' instruction suffix on PowerPC or '%P0' on IA-64. 42359 42360'>' 42361 A memory operand with autoincrement addressing (either preincrement 42362 or postincrement) is allowed. In inline 'asm' the same 42363 restrictions as for '<' apply. 42364 42365'r' 42366 A register operand is allowed provided that it is in a general 42367 register. 42368 42369'i' 42370 An immediate integer operand (one with constant value) is allowed. 42371 This includes symbolic constants whose values will be known only at 42372 assembly time or later. 42373 42374'n' 42375 An immediate integer operand with a known numeric value is allowed. 42376 Many systems cannot support assembly-time constants for operands 42377 less than a word wide. Constraints for these operands should use 42378 'n' rather than 'i'. 42379 42380'I', 'J', 'K', ... 'P' 42381 Other letters in the range 'I' through 'P' may be defined in a 42382 machine-dependent fashion to permit immediate integer operands with 42383 explicit integer values in specified ranges. For example, on the 42384 68000, 'I' is defined to stand for the range of values 1 to 8. 42385 This is the range permitted as a shift count in the shift 42386 instructions. 42387 42388'E' 42389 An immediate floating operand (expression code 'const_double') is 42390 allowed, but only if the target floating point format is the same 42391 as that of the host machine (on which the compiler is running). 42392 42393'F' 42394 An immediate floating operand (expression code 'const_double' or 42395 'const_vector') is allowed. 42396 42397'G', 'H' 42398 'G' and 'H' may be defined in a machine-dependent fashion to permit 42399 immediate floating operands in particular ranges of values. 42400 42401's' 42402 An immediate integer operand whose value is not an explicit integer 42403 is allowed. 42404 42405 This might appear strange; if an insn allows a constant operand 42406 with a value not known at compile time, it certainly must allow any 42407 known value. So why use 's' instead of 'i'? Sometimes it allows 42408 better code to be generated. 42409 42410 For example, on the 68000 in a fullword instruction it is possible 42411 to use an immediate operand; but if the immediate value is between 42412 -128 and 127, better code results from loading the value into a 42413 register and using the register. This is because the load into the 42414 register can be done with a 'moveq' instruction. We arrange for 42415 this to happen by defining the letter 'K' to mean "any integer 42416 outside the range -128 to 127", and then specifying 'Ks' in the 42417 operand constraints. 42418 42419'g' 42420 Any register, memory or immediate integer operand is allowed, 42421 except for registers that are not general registers. 42422 42423'X' 42424 Any operand whatsoever is allowed. 42425 42426'0', '1', '2', ... '9' 42427 An operand that matches the specified operand number is allowed. 42428 If a digit is used together with letters within the same 42429 alternative, the digit should come last. 42430 42431 This number is allowed to be more than a single digit. If multiple 42432 digits are encountered consecutively, they are interpreted as a 42433 single decimal integer. There is scant chance for ambiguity, since 42434 to-date it has never been desirable that '10' be interpreted as 42435 matching either operand 1 _or_ operand 0. Should this be desired, 42436 one can use multiple alternatives instead. 42437 42438 This is called a "matching constraint" and what it really means is 42439 that the assembler has only a single operand that fills two roles 42440 which 'asm' distinguishes. For example, an add instruction uses 42441 two input operands and an output operand, but on most CISC machines 42442 an add instruction really has only two operands, one of them an 42443 input-output operand: 42444 42445 addl #35,r12 42446 42447 Matching constraints are used in these circumstances. More 42448 precisely, the two operands that match must include one input-only 42449 operand and one output-only operand. Moreover, the digit must be a 42450 smaller number than the number of the operand that uses it in the 42451 constraint. 42452 42453'p' 42454 An operand that is a valid memory address is allowed. This is for 42455 "load address" and "push address" instructions. 42456 42457 'p' in the constraint must be accompanied by 'address_operand' as 42458 the predicate in the 'match_operand'. This predicate interprets 42459 the mode specified in the 'match_operand' as the mode of the memory 42460 reference for which the address would be valid. 42461 42462OTHER-LETTERS 42463 Other letters can be defined in machine-dependent fashion to stand 42464 for particular classes of registers or other arbitrary operand 42465 types. 'd', 'a' and 'f' are defined on the 68000/68020 to stand 42466 for data, address and floating point registers. 42467 42468 42469File: gcc.info, Node: Multi-Alternative, Next: Modifiers, Prev: Simple Constraints, Up: Constraints 42470 424716.47.3.2 Multiple Alternative Constraints 42472......................................... 42473 42474Sometimes a single instruction has multiple alternative sets of possible 42475operands. For example, on the 68000, a logical-or instruction can 42476combine register or an immediate value into memory, or it can combine 42477any kind of operand into a register; but it cannot combine one memory 42478location into another. 42479 42480 These constraints are represented as multiple alternatives. An 42481alternative can be described by a series of letters for each operand. 42482The overall constraint for an operand is made from the letters for this 42483operand from the first alternative, a comma, the letters for this 42484operand from the second alternative, a comma, and so on until the last 42485alternative. All operands for a single instruction must have the same 42486number of alternatives. 42487 42488 So the first alternative for the 68000's logical-or could be written as 42489'"+m" (output) : "ir" (input)'. The second could be '"+r" (output): 42490"irm" (input)'. However, the fact that two memory locations cannot be 42491used in a single instruction prevents simply using '"+rm" (output) : 42492"irm" (input)'. Using multi-alternatives, this might be written as 42493'"+m,r" (output) : "ir,irm" (input)'. This describes all the available 42494alternatives to the compiler, allowing it to choose the most efficient 42495one for the current conditions. 42496 42497 There is no way within the template to determine which alternative was 42498chosen. However you may be able to wrap your 'asm' statements with 42499builtins such as '__builtin_constant_p' to achieve the desired results. 42500 42501 42502File: gcc.info, Node: Modifiers, Next: Machine Constraints, Prev: Multi-Alternative, Up: Constraints 42503 425046.47.3.3 Constraint Modifier Characters 42505....................................... 42506 42507Here are constraint modifier characters. 42508 42509'=' 42510 Means that this operand is written to by this instruction: the 42511 previous value is discarded and replaced by new data. 42512 42513'+' 42514 Means that this operand is both read and written by the 42515 instruction. 42516 42517 When the compiler fixes up the operands to satisfy the constraints, 42518 it needs to know which operands are read by the instruction and 42519 which are written by it. '=' identifies an operand which is only 42520 written; '+' identifies an operand that is both read and written; 42521 all other operands are assumed to only be read. 42522 42523 If you specify '=' or '+' in a constraint, you put it in the first 42524 character of the constraint string. 42525 42526'&' 42527 Means (in a particular alternative) that this operand is an 42528 "earlyclobber" operand, which is written before the instruction is 42529 finished using the input operands. Therefore, this operand may not 42530 lie in a register that is read by the instruction or as part of any 42531 memory address. 42532 42533 '&' applies only to the alternative in which it is written. In 42534 constraints with multiple alternatives, sometimes one alternative 42535 requires '&' while others do not. See, for example, the 'movdf' 42536 insn of the 68000. 42537 42538 An operand which is read by the instruction can be tied to an 42539 earlyclobber operand if its only use as an input occurs before the 42540 early result is written. Adding alternatives of this form often 42541 allows GCC to produce better code when only some of the read 42542 operands can be affected by the earlyclobber. See, for example, 42543 the 'mulsi3' insn of the ARM. 42544 42545 Furthermore, if the "earlyclobber" operand is also a read/write 42546 operand, then that operand is written only after it's used. 42547 42548 '&' does not obviate the need to write '=' or '+'. As 42549 "earlyclobber" operands are always written, a read-only 42550 "earlyclobber" operand is ill-formed and will be rejected by the 42551 compiler. 42552 42553'%' 42554 Declares the instruction to be commutative for this operand and the 42555 following operand. This means that the compiler may interchange 42556 the two operands if that is the cheapest way to make all operands 42557 fit the constraints. '%' applies to all alternatives and must 42558 appear as the first character in the constraint. Only read-only 42559 operands can use '%'. 42560 42561 GCC can only handle one commutative pair in an asm; if you use 42562 more, the compiler may fail. Note that you need not use the 42563 modifier if the two alternatives are strictly identical; this would 42564 only waste time in the reload pass. 42565 42566 42567File: gcc.info, Node: Machine Constraints, Prev: Modifiers, Up: Constraints 42568 425696.47.3.4 Constraints for Particular Machines 42570............................................ 42571 42572Whenever possible, you should use the general-purpose constraint letters 42573in 'asm' arguments, since they will convey meaning more readily to 42574people reading your code. Failing that, use the constraint letters that 42575usually have very similar meanings across architectures. The most 42576commonly used constraints are 'm' and 'r' (for memory and 42577general-purpose registers respectively; *note Simple Constraints::), and 42578'I', usually the letter indicating the most common immediate-constant 42579format. 42580 42581 Each architecture defines additional constraints. These constraints 42582are used by the compiler itself for instruction generation, as well as 42583for 'asm' statements; therefore, some of the constraints are not 42584particularly useful for 'asm'. Here is a summary of some of the 42585machine-dependent constraints available on some particular machines; it 42586includes both constraints that are useful for 'asm' and constraints that 42587aren't. The compiler source file mentioned in the table heading for 42588each architecture is the definitive reference for the meanings of that 42589architecture's constraints. 42590 42591_AArch64 family--'config/aarch64/constraints.md'_ 42592 'k' 42593 The stack pointer register ('SP') 42594 42595 'w' 42596 Floating point register, Advanced SIMD vector register or SVE 42597 vector register 42598 42599 'x' 42600 Like 'w', but restricted to registers 0 to 15 inclusive. 42601 42602 'y' 42603 Like 'w', but restricted to registers 0 to 7 inclusive. 42604 42605 'Upl' 42606 One of the low eight SVE predicate registers ('P0' to 'P7') 42607 42608 'Upa' 42609 Any of the SVE predicate registers ('P0' to 'P15') 42610 42611 'I' 42612 Integer constant that is valid as an immediate operand in an 42613 'ADD' instruction 42614 42615 'J' 42616 Integer constant that is valid as an immediate operand in a 42617 'SUB' instruction (once negated) 42618 42619 'K' 42620 Integer constant that can be used with a 32-bit logical 42621 instruction 42622 42623 'L' 42624 Integer constant that can be used with a 64-bit logical 42625 instruction 42626 42627 'M' 42628 Integer constant that is valid as an immediate operand in a 42629 32-bit 'MOV' pseudo instruction. The 'MOV' may be assembled 42630 to one of several different machine instructions depending on 42631 the value 42632 42633 'N' 42634 Integer constant that is valid as an immediate operand in a 42635 64-bit 'MOV' pseudo instruction 42636 42637 'S' 42638 An absolute symbolic address or a label reference 42639 42640 'Y' 42641 Floating point constant zero 42642 42643 'Z' 42644 Integer constant zero 42645 42646 'Ush' 42647 The high part (bits 12 and upwards) of the pc-relative address 42648 of a symbol within 4GB of the instruction 42649 42650 'Q' 42651 A memory address which uses a single base register with no 42652 offset 42653 42654 'Ump' 42655 A memory address suitable for a load/store pair instruction in 42656 SI, DI, SF and DF modes 42657 42658_AMD GCN --'config/gcn/constraints.md'_ 42659 'I' 42660 Immediate integer in the range -16 to 64 42661 42662 'J' 42663 Immediate 16-bit signed integer 42664 42665 'Kf' 42666 Immediate constant -1 42667 42668 'L' 42669 Immediate 15-bit unsigned integer 42670 42671 'A' 42672 Immediate constant that can be inlined in an instruction 42673 encoding: integer -16..64, or float 0.0, +/-0.5, +/-1.0, 42674 +/-2.0, +/-4.0, 1.0/(2.0*PI) 42675 42676 'B' 42677 Immediate 32-bit signed integer that can be attached to an 42678 instruction encoding 42679 42680 'C' 42681 Immediate 32-bit integer in range -16..4294967295 (i.e. 42682 32-bit unsigned integer or 'A' constraint) 42683 42684 'DA' 42685 Immediate 64-bit constant that can be split into two 'A' 42686 constants 42687 42688 'DB' 42689 Immediate 64-bit constant that can be split into two 'B' 42690 constants 42691 42692 'U' 42693 Any 'unspec' 42694 42695 'Y' 42696 Any 'symbol_ref' or 'label_ref' 42697 42698 'v' 42699 VGPR register 42700 42701 'Sg' 42702 SGPR register 42703 42704 'SD' 42705 SGPR registers valid for instruction destinations, including 42706 VCC, M0 and EXEC 42707 42708 'SS' 42709 SGPR registers valid for instruction sources, including VCC, 42710 M0, EXEC and SCC 42711 42712 'Sm' 42713 SGPR registers valid as a source for scalar memory 42714 instructions (excludes M0 and EXEC) 42715 42716 'Sv' 42717 SGPR registers valid as a source or destination for vector 42718 instructions (excludes EXEC) 42719 42720 'ca' 42721 All condition registers: SCC, VCCZ, EXECZ 42722 42723 'cs' 42724 Scalar condition register: SCC 42725 42726 'cV' 42727 Vector condition register: VCC, VCC_LO, VCC_HI 42728 42729 'e' 42730 EXEC register (EXEC_LO and EXEC_HI) 42731 42732 'RB' 42733 Memory operand with address space suitable for 'buffer_*' 42734 instructions 42735 42736 'RF' 42737 Memory operand with address space suitable for 'flat_*' 42738 instructions 42739 42740 'RS' 42741 Memory operand with address space suitable for 's_*' 42742 instructions 42743 42744 'RL' 42745 Memory operand with address space suitable for 'ds_*' LDS 42746 instructions 42747 42748 'RG' 42749 Memory operand with address space suitable for 'ds_*' GDS 42750 instructions 42751 42752 'RD' 42753 Memory operand with address space suitable for any 'ds_*' 42754 instructions 42755 42756 'RM' 42757 Memory operand with address space suitable for 'global_*' 42758 instructions 42759 42760_ARC --'config/arc/constraints.md'_ 42761 'q' 42762 Registers usable in ARCompact 16-bit instructions: 'r0'-'r3', 42763 'r12'-'r15'. This constraint can only match when the '-mq' 42764 option is in effect. 42765 42766 'e' 42767 Registers usable as base-regs of memory addresses in ARCompact 42768 16-bit memory instructions: 'r0'-'r3', 'r12'-'r15', 'sp'. 42769 This constraint can only match when the '-mq' option is in 42770 effect. 42771 'D' 42772 ARC FPX (dpfp) 64-bit registers. 'D0', 'D1'. 42773 42774 'I' 42775 A signed 12-bit integer constant. 42776 42777 'Cal' 42778 constant for arithmetic/logical operations. This might be any 42779 constant that can be put into a long immediate by the assmbler 42780 or linker without involving a PIC relocation. 42781 42782 'K' 42783 A 3-bit unsigned integer constant. 42784 42785 'L' 42786 A 6-bit unsigned integer constant. 42787 42788 'CnL' 42789 One's complement of a 6-bit unsigned integer constant. 42790 42791 'CmL' 42792 Two's complement of a 6-bit unsigned integer constant. 42793 42794 'M' 42795 A 5-bit unsigned integer constant. 42796 42797 'O' 42798 A 7-bit unsigned integer constant. 42799 42800 'P' 42801 A 8-bit unsigned integer constant. 42802 42803 'H' 42804 Any const_double value. 42805 42806_ARM family--'config/arm/constraints.md'_ 42807 42808 'h' 42809 In Thumb state, the core registers 'r8'-'r15'. 42810 42811 'k' 42812 The stack pointer register. 42813 42814 'l' 42815 In Thumb State the core registers 'r0'-'r7'. In ARM state 42816 this is an alias for the 'r' constraint. 42817 42818 't' 42819 VFP floating-point registers 's0'-'s31'. Used for 32 bit 42820 values. 42821 42822 'w' 42823 VFP floating-point registers 'd0'-'d31' and the appropriate 42824 subset 'd0'-'d15' based on command line options. Used for 64 42825 bit values only. Not valid for Thumb1. 42826 42827 'y' 42828 The iWMMX co-processor registers. 42829 42830 'z' 42831 The iWMMX GR registers. 42832 42833 'G' 42834 The floating-point constant 0.0 42835 42836 'I' 42837 Integer that is valid as an immediate operand in a data 42838 processing instruction. That is, an integer in the range 0 to 42839 255 rotated by a multiple of 2 42840 42841 'J' 42842 Integer in the range -4095 to 4095 42843 42844 'K' 42845 Integer that satisfies constraint 'I' when inverted (ones 42846 complement) 42847 42848 'L' 42849 Integer that satisfies constraint 'I' when negated (twos 42850 complement) 42851 42852 'M' 42853 Integer in the range 0 to 32 42854 42855 'Q' 42856 A memory reference where the exact address is in a single 42857 register (''m'' is preferable for 'asm' statements) 42858 42859 'R' 42860 An item in the constant pool 42861 42862 'S' 42863 A symbol in the text segment of the current file 42864 42865 'Uv' 42866 A memory reference suitable for VFP load/store insns 42867 (reg+constant offset) 42868 42869 'Uy' 42870 A memory reference suitable for iWMMXt load/store 42871 instructions. 42872 42873 'Uq' 42874 A memory reference suitable for the ARMv4 ldrsb instruction. 42875 42876_AVR family--'config/avr/constraints.md'_ 42877 'l' 42878 Registers from r0 to r15 42879 42880 'a' 42881 Registers from r16 to r23 42882 42883 'd' 42884 Registers from r16 to r31 42885 42886 'w' 42887 Registers from r24 to r31. These registers can be used in 42888 'adiw' command 42889 42890 'e' 42891 Pointer register (r26-r31) 42892 42893 'b' 42894 Base pointer register (r28-r31) 42895 42896 'q' 42897 Stack pointer register (SPH:SPL) 42898 42899 't' 42900 Temporary register r0 42901 42902 'x' 42903 Register pair X (r27:r26) 42904 42905 'y' 42906 Register pair Y (r29:r28) 42907 42908 'z' 42909 Register pair Z (r31:r30) 42910 42911 'I' 42912 Constant greater than -1, less than 64 42913 42914 'J' 42915 Constant greater than -64, less than 1 42916 42917 'K' 42918 Constant integer 2 42919 42920 'L' 42921 Constant integer 0 42922 42923 'M' 42924 Constant that fits in 8 bits 42925 42926 'N' 42927 Constant integer -1 42928 42929 'O' 42930 Constant integer 8, 16, or 24 42931 42932 'P' 42933 Constant integer 1 42934 42935 'G' 42936 A floating point constant 0.0 42937 42938 'Q' 42939 A memory address based on Y or Z pointer with displacement. 42940 42941_Blackfin family--'config/bfin/constraints.md'_ 42942 'a' 42943 P register 42944 42945 'd' 42946 D register 42947 42948 'z' 42949 A call clobbered P register. 42950 42951 'qN' 42952 A single register. If N is in the range 0 to 7, the 42953 corresponding D register. If it is 'A', then the register P0. 42954 42955 'D' 42956 Even-numbered D register 42957 42958 'W' 42959 Odd-numbered D register 42960 42961 'e' 42962 Accumulator register. 42963 42964 'A' 42965 Even-numbered accumulator register. 42966 42967 'B' 42968 Odd-numbered accumulator register. 42969 42970 'b' 42971 I register 42972 42973 'v' 42974 B register 42975 42976 'f' 42977 M register 42978 42979 'c' 42980 Registers used for circular buffering, i.e. I, B, or L 42981 registers. 42982 42983 'C' 42984 The CC register. 42985 42986 't' 42987 LT0 or LT1. 42988 42989 'k' 42990 LC0 or LC1. 42991 42992 'u' 42993 LB0 or LB1. 42994 42995 'x' 42996 Any D, P, B, M, I or L register. 42997 42998 'y' 42999 Additional registers typically used only in prologues and 43000 epilogues: RETS, RETN, RETI, RETX, RETE, ASTAT, SEQSTAT and 43001 USP. 43002 43003 'w' 43004 Any register except accumulators or CC. 43005 43006 'Ksh' 43007 Signed 16 bit integer (in the range -32768 to 32767) 43008 43009 'Kuh' 43010 Unsigned 16 bit integer (in the range 0 to 65535) 43011 43012 'Ks7' 43013 Signed 7 bit integer (in the range -64 to 63) 43014 43015 'Ku7' 43016 Unsigned 7 bit integer (in the range 0 to 127) 43017 43018 'Ku5' 43019 Unsigned 5 bit integer (in the range 0 to 31) 43020 43021 'Ks4' 43022 Signed 4 bit integer (in the range -8 to 7) 43023 43024 'Ks3' 43025 Signed 3 bit integer (in the range -3 to 4) 43026 43027 'Ku3' 43028 Unsigned 3 bit integer (in the range 0 to 7) 43029 43030 'PN' 43031 Constant N, where N is a single-digit constant in the range 0 43032 to 4. 43033 43034 'PA' 43035 An integer equal to one of the MACFLAG_XXX constants that is 43036 suitable for use with either accumulator. 43037 43038 'PB' 43039 An integer equal to one of the MACFLAG_XXX constants that is 43040 suitable for use only with accumulator A1. 43041 43042 'M1' 43043 Constant 255. 43044 43045 'M2' 43046 Constant 65535. 43047 43048 'J' 43049 An integer constant with exactly a single bit set. 43050 43051 'L' 43052 An integer constant with all bits set except exactly one. 43053 43054 'H' 43055 43056 'Q' 43057 Any SYMBOL_REF. 43058 43059_CR16 Architecture--'config/cr16/cr16.h'_ 43060 43061 'b' 43062 Registers from r0 to r14 (registers without stack pointer) 43063 43064 't' 43065 Register from r0 to r11 (all 16-bit registers) 43066 43067 'p' 43068 Register from r12 to r15 (all 32-bit registers) 43069 43070 'I' 43071 Signed constant that fits in 4 bits 43072 43073 'J' 43074 Signed constant that fits in 5 bits 43075 43076 'K' 43077 Signed constant that fits in 6 bits 43078 43079 'L' 43080 Unsigned constant that fits in 4 bits 43081 43082 'M' 43083 Signed constant that fits in 32 bits 43084 43085 'N' 43086 Check for 64 bits wide constants for add/sub instructions 43087 43088 'G' 43089 Floating point constant that is legal for store immediate 43090 43091_C-SKY--'config/csky/constraints.md'_ 43092 43093 'a' 43094 The mini registers r0 - r7. 43095 43096 'b' 43097 The low registers r0 - r15. 43098 43099 'c' 43100 C register. 43101 43102 'y' 43103 HI and LO registers. 43104 43105 'l' 43106 LO register. 43107 43108 'h' 43109 HI register. 43110 43111 'v' 43112 Vector registers. 43113 43114 'z' 43115 Stack pointer register (SP). 43116 43117 'Q' 43118 A memory address which uses a base register with a short 43119 offset or with a index register with its scale. 43120 43121 'W' 43122 A memory address which uses a base register with a index 43123 register with its scale. 43124 43125_Epiphany--'config/epiphany/constraints.md'_ 43126 'U16' 43127 An unsigned 16-bit constant. 43128 43129 'K' 43130 An unsigned 5-bit constant. 43131 43132 'L' 43133 A signed 11-bit constant. 43134 43135 'Cm1' 43136 A signed 11-bit constant added to -1. Can only match when the 43137 '-m1reg-REG' option is active. 43138 43139 'Cl1' 43140 Left-shift of -1, i.e., a bit mask with a block of leading 43141 ones, the rest being a block of trailing zeroes. Can only 43142 match when the '-m1reg-REG' option is active. 43143 43144 'Cr1' 43145 Right-shift of -1, i.e., a bit mask with a trailing block of 43146 ones, the rest being zeroes. Or to put it another way, one 43147 less than a power of two. Can only match when the 43148 '-m1reg-REG' option is active. 43149 43150 'Cal' 43151 Constant for arithmetic/logical operations. This is like 'i', 43152 except that for position independent code, no symbols / 43153 expressions needing relocations are allowed. 43154 43155 'Csy' 43156 Symbolic constant for call/jump instruction. 43157 43158 'Rcs' 43159 The register class usable in short insns. This is a register 43160 class constraint, and can thus drive register allocation. 43161 This constraint won't match unless '-mprefer-short-insn-regs' 43162 is in effect. 43163 43164 'Rsc' 43165 The the register class of registers that can be used to hold a 43166 sibcall call address. I.e., a caller-saved register. 43167 43168 'Rct' 43169 Core control register class. 43170 43171 'Rgs' 43172 The register group usable in short insns. This constraint 43173 does not use a register class, so that it only passively 43174 matches suitable registers, and doesn't drive register 43175 allocation. 43176 43177 'Rra' 43178 Matches the return address if it can be replaced with the link 43179 register. 43180 43181 'Rcc' 43182 Matches the integer condition code register. 43183 43184 'Sra' 43185 Matches the return address if it is in a stack slot. 43186 43187 'Cfm' 43188 Matches control register values to switch fp mode, which are 43189 encapsulated in 'UNSPEC_FP_MODE'. 43190 43191_FRV--'config/frv/frv.h'_ 43192 'a' 43193 Register in the class 'ACC_REGS' ('acc0' to 'acc7'). 43194 43195 'b' 43196 Register in the class 'EVEN_ACC_REGS' ('acc0' to 'acc7'). 43197 43198 'c' 43199 Register in the class 'CC_REGS' ('fcc0' to 'fcc3' and 'icc0' 43200 to 'icc3'). 43201 43202 'd' 43203 Register in the class 'GPR_REGS' ('gr0' to 'gr63'). 43204 43205 'e' 43206 Register in the class 'EVEN_REGS' ('gr0' to 'gr63'). Odd 43207 registers are excluded not in the class but through the use of 43208 a machine mode larger than 4 bytes. 43209 43210 'f' 43211 Register in the class 'FPR_REGS' ('fr0' to 'fr63'). 43212 43213 'h' 43214 Register in the class 'FEVEN_REGS' ('fr0' to 'fr63'). Odd 43215 registers are excluded not in the class but through the use of 43216 a machine mode larger than 4 bytes. 43217 43218 'l' 43219 Register in the class 'LR_REG' (the 'lr' register). 43220 43221 'q' 43222 Register in the class 'QUAD_REGS' ('gr2' to 'gr63'). Register 43223 numbers not divisible by 4 are excluded not in the class but 43224 through the use of a machine mode larger than 8 bytes. 43225 43226 't' 43227 Register in the class 'ICC_REGS' ('icc0' to 'icc3'). 43228 43229 'u' 43230 Register in the class 'FCC_REGS' ('fcc0' to 'fcc3'). 43231 43232 'v' 43233 Register in the class 'ICR_REGS' ('cc4' to 'cc7'). 43234 43235 'w' 43236 Register in the class 'FCR_REGS' ('cc0' to 'cc3'). 43237 43238 'x' 43239 Register in the class 'QUAD_FPR_REGS' ('fr0' to 'fr63'). 43240 Register numbers not divisible by 4 are excluded not in the 43241 class but through the use of a machine mode larger than 8 43242 bytes. 43243 43244 'z' 43245 Register in the class 'SPR_REGS' ('lcr' and 'lr'). 43246 43247 'A' 43248 Register in the class 'QUAD_ACC_REGS' ('acc0' to 'acc7'). 43249 43250 'B' 43251 Register in the class 'ACCG_REGS' ('accg0' to 'accg7'). 43252 43253 'C' 43254 Register in the class 'CR_REGS' ('cc0' to 'cc7'). 43255 43256 'G' 43257 Floating point constant zero 43258 43259 'I' 43260 6-bit signed integer constant 43261 43262 'J' 43263 10-bit signed integer constant 43264 43265 'L' 43266 16-bit signed integer constant 43267 43268 'M' 43269 16-bit unsigned integer constant 43270 43271 'N' 43272 12-bit signed integer constant that is negative--i.e. in the 43273 range of -2048 to -1 43274 43275 'O' 43276 Constant zero 43277 43278 'P' 43279 12-bit signed integer constant that is greater than zero--i.e. 43280 in the range of 1 to 2047. 43281 43282_FT32--'config/ft32/constraints.md'_ 43283 'A' 43284 An absolute address 43285 43286 'B' 43287 An offset address 43288 43289 'W' 43290 A register indirect memory operand 43291 43292 'e' 43293 An offset address. 43294 43295 'f' 43296 An offset address. 43297 43298 'O' 43299 The constant zero or one 43300 43301 'I' 43302 A 16-bit signed constant (-32768 ... 32767) 43303 43304 'w' 43305 A bitfield mask suitable for bext or bins 43306 43307 'x' 43308 An inverted bitfield mask suitable for bext or bins 43309 43310 'L' 43311 A 16-bit unsigned constant, multiple of 4 (0 ... 65532) 43312 43313 'S' 43314 A 20-bit signed constant (-524288 ... 524287) 43315 43316 'b' 43317 A constant for a bitfield width (1 ... 16) 43318 43319 'KA' 43320 A 10-bit signed constant (-512 ... 511) 43321 43322_Hewlett-Packard PA-RISC--'config/pa/pa.h'_ 43323 'a' 43324 General register 1 43325 43326 'f' 43327 Floating point register 43328 43329 'q' 43330 Shift amount register 43331 43332 'x' 43333 Floating point register (deprecated) 43334 43335 'y' 43336 Upper floating point register (32-bit), floating point 43337 register (64-bit) 43338 43339 'Z' 43340 Any register 43341 43342 'I' 43343 Signed 11-bit integer constant 43344 43345 'J' 43346 Signed 14-bit integer constant 43347 43348 'K' 43349 Integer constant that can be deposited with a 'zdepi' 43350 instruction 43351 43352 'L' 43353 Signed 5-bit integer constant 43354 43355 'M' 43356 Integer constant 0 43357 43358 'N' 43359 Integer constant that can be loaded with a 'ldil' instruction 43360 43361 'O' 43362 Integer constant whose value plus one is a power of 2 43363 43364 'P' 43365 Integer constant that can be used for 'and' operations in 43366 'depi' and 'extru' instructions 43367 43368 'S' 43369 Integer constant 31 43370 43371 'U' 43372 Integer constant 63 43373 43374 'G' 43375 Floating-point constant 0.0 43376 43377 'A' 43378 A 'lo_sum' data-linkage-table memory operand 43379 43380 'Q' 43381 A memory operand that can be used as the destination operand 43382 of an integer store instruction 43383 43384 'R' 43385 A scaled or unscaled indexed memory operand 43386 43387 'T' 43388 A memory operand for floating-point loads and stores 43389 43390 'W' 43391 A register indirect memory operand 43392 43393_Intel IA-64--'config/ia64/ia64.h'_ 43394 'a' 43395 General register 'r0' to 'r3' for 'addl' instruction 43396 43397 'b' 43398 Branch register 43399 43400 'c' 43401 Predicate register ('c' as in "conditional") 43402 43403 'd' 43404 Application register residing in M-unit 43405 43406 'e' 43407 Application register residing in I-unit 43408 43409 'f' 43410 Floating-point register 43411 43412 'm' 43413 Memory operand. If used together with '<' or '>', the operand 43414 can have postincrement and postdecrement which require 43415 printing with '%Pn' on IA-64. 43416 43417 'G' 43418 Floating-point constant 0.0 or 1.0 43419 43420 'I' 43421 14-bit signed integer constant 43422 43423 'J' 43424 22-bit signed integer constant 43425 43426 'K' 43427 8-bit signed integer constant for logical instructions 43428 43429 'L' 43430 8-bit adjusted signed integer constant for compare pseudo-ops 43431 43432 'M' 43433 6-bit unsigned integer constant for shift counts 43434 43435 'N' 43436 9-bit signed integer constant for load and store 43437 postincrements 43438 43439 'O' 43440 The constant zero 43441 43442 'P' 43443 0 or -1 for 'dep' instruction 43444 43445 'Q' 43446 Non-volatile memory for floating-point loads and stores 43447 43448 'R' 43449 Integer constant in the range 1 to 4 for 'shladd' instruction 43450 43451 'S' 43452 Memory operand except postincrement and postdecrement. This 43453 is now roughly the same as 'm' when not used together with '<' 43454 or '>'. 43455 43456_M32C--'config/m32c/m32c.cc'_ 43457 'Rsp' 43458 'Rfb' 43459 'Rsb' 43460 '$sp', '$fb', '$sb'. 43461 43462 'Rcr' 43463 Any control register, when they're 16 bits wide (nothing if 43464 control registers are 24 bits wide) 43465 43466 'Rcl' 43467 Any control register, when they're 24 bits wide. 43468 43469 'R0w' 43470 'R1w' 43471 'R2w' 43472 'R3w' 43473 $r0, $r1, $r2, $r3. 43474 43475 'R02' 43476 $r0 or $r2, or $r2r0 for 32 bit values. 43477 43478 'R13' 43479 $r1 or $r3, or $r3r1 for 32 bit values. 43480 43481 'Rdi' 43482 A register that can hold a 64 bit value. 43483 43484 'Rhl' 43485 $r0 or $r1 (registers with addressable high/low bytes) 43486 43487 'R23' 43488 $r2 or $r3 43489 43490 'Raa' 43491 Address registers 43492 43493 'Raw' 43494 Address registers when they're 16 bits wide. 43495 43496 'Ral' 43497 Address registers when they're 24 bits wide. 43498 43499 'Rqi' 43500 Registers that can hold QI values. 43501 43502 'Rad' 43503 Registers that can be used with displacements ($a0, $a1, $sb). 43504 43505 'Rsi' 43506 Registers that can hold 32 bit values. 43507 43508 'Rhi' 43509 Registers that can hold 16 bit values. 43510 43511 'Rhc' 43512 Registers chat can hold 16 bit values, including all control 43513 registers. 43514 43515 'Rra' 43516 $r0 through R1, plus $a0 and $a1. 43517 43518 'Rfl' 43519 The flags register. 43520 43521 'Rmm' 43522 The memory-based pseudo-registers $mem0 through $mem15. 43523 43524 'Rpi' 43525 Registers that can hold pointers (16 bit registers for r8c, 43526 m16c; 24 bit registers for m32cm, m32c). 43527 43528 'Rpa' 43529 Matches multiple registers in a PARALLEL to form a larger 43530 register. Used to match function return values. 43531 43532 'Is3' 43533 -8 ... 7 43534 43535 'IS1' 43536 -128 ... 127 43537 43538 'IS2' 43539 -32768 ... 32767 43540 43541 'IU2' 43542 0 ... 65535 43543 43544 'In4' 43545 -8 ... -1 or 1 ... 8 43546 43547 'In5' 43548 -16 ... -1 or 1 ... 16 43549 43550 'In6' 43551 -32 ... -1 or 1 ... 32 43552 43553 'IM2' 43554 -65536 ... -1 43555 43556 'Ilb' 43557 An 8 bit value with exactly one bit set. 43558 43559 'Ilw' 43560 A 16 bit value with exactly one bit set. 43561 43562 'Sd' 43563 The common src/dest memory addressing modes. 43564 43565 'Sa' 43566 Memory addressed using $a0 or $a1. 43567 43568 'Si' 43569 Memory addressed with immediate addresses. 43570 43571 'Ss' 43572 Memory addressed using the stack pointer ($sp). 43573 43574 'Sf' 43575 Memory addressed using the frame base register ($fb). 43576 43577 'Ss' 43578 Memory addressed using the small base register ($sb). 43579 43580 'S1' 43581 $r1h 43582 43583_LoongArch--'config/loongarch/constraints.md'_ 43584 'f' 43585 A floating-point register (if available). 43586 'k' 43587 A memory operand whose address is formed by a base register 43588 and (optionally scaled) index register. 43589 'l' 43590 A signed 16-bit constant. 43591 'm' 43592 A memory operand whose address is formed by a base register 43593 and offset that is suitable for use in instructions with the 43594 same addressing mode as 'st.w' and 'ld.w'. 43595 'I' 43596 A signed 12-bit constant (for arithmetic instructions). 43597 'K' 43598 An unsigned 12-bit constant (for logic instructions). 43599 'ZB' 43600 An address that is held in a general-purpose register. The 43601 offset is zero. 43602 'ZC' 43603 A memory operand whose address is formed by a base register 43604 and offset that is suitable for use in instructions with the 43605 same addressing mode as 'll.w' and 'sc.w'. 43606 43607_MicroBlaze--'config/microblaze/constraints.md'_ 43608 'd' 43609 A general register ('r0' to 'r31'). 43610 43611 'z' 43612 A status register ('rmsr', '$fcc1' to '$fcc7'). 43613 43614_MIPS--'config/mips/constraints.md'_ 43615 'd' 43616 A general-purpose register. This is equivalent to 'r' unless 43617 generating MIPS16 code, in which case the MIPS16 register set 43618 is used. 43619 43620 'f' 43621 A floating-point register (if available). 43622 43623 'h' 43624 Formerly the 'hi' register. This constraint is no longer 43625 supported. 43626 43627 'l' 43628 The 'lo' register. Use this register to store values that are 43629 no bigger than a word. 43630 43631 'x' 43632 The concatenated 'hi' and 'lo' registers. Use this register 43633 to store doubleword values. 43634 43635 'c' 43636 A register suitable for use in an indirect jump. This will 43637 always be '$25' for '-mabicalls'. 43638 43639 'v' 43640 Register '$3'. Do not use this constraint in new code; it is 43641 retained only for compatibility with glibc. 43642 43643 'y' 43644 Equivalent to 'r'; retained for backwards compatibility. 43645 43646 'z' 43647 A floating-point condition code register. 43648 43649 'I' 43650 A signed 16-bit constant (for arithmetic instructions). 43651 43652 'J' 43653 Integer zero. 43654 43655 'K' 43656 An unsigned 16-bit constant (for logic instructions). 43657 43658 'L' 43659 A signed 32-bit constant in which the lower 16 bits are zero. 43660 Such constants can be loaded using 'lui'. 43661 43662 'M' 43663 A constant that cannot be loaded using 'lui', 'addiu' or 43664 'ori'. 43665 43666 'N' 43667 A constant in the range -65535 to -1 (inclusive). 43668 43669 'O' 43670 A signed 15-bit constant. 43671 43672 'P' 43673 A constant in the range 1 to 65535 (inclusive). 43674 43675 'G' 43676 Floating-point zero. 43677 43678 'R' 43679 An address that can be used in a non-macro load or store. 43680 43681 'ZC' 43682 A memory operand whose address is formed by a base register 43683 and offset that is suitable for use in instructions with the 43684 same addressing mode as 'll' and 'sc'. 43685 43686 'ZD' 43687 An address suitable for a 'prefetch' instruction, or for any 43688 other instruction with the same addressing mode as 'prefetch'. 43689 43690_Motorola 680x0--'config/m68k/constraints.md'_ 43691 'a' 43692 Address register 43693 43694 'd' 43695 Data register 43696 43697 'f' 43698 68881 floating-point register, if available 43699 43700 'I' 43701 Integer in the range 1 to 8 43702 43703 'J' 43704 16-bit signed number 43705 43706 'K' 43707 Signed number whose magnitude is greater than 0x80 43708 43709 'L' 43710 Integer in the range -8 to -1 43711 43712 'M' 43713 Signed number whose magnitude is greater than 0x100 43714 43715 'N' 43716 Range 24 to 31, rotatert:SI 8 to 1 expressed as rotate 43717 43718 'O' 43719 16 (for rotate using swap) 43720 43721 'P' 43722 Range 8 to 15, rotatert:HI 8 to 1 expressed as rotate 43723 43724 'R' 43725 Numbers that mov3q can handle 43726 43727 'G' 43728 Floating point constant that is not a 68881 constant 43729 43730 'S' 43731 Operands that satisfy 'm' when -mpcrel is in effect 43732 43733 'T' 43734 Operands that satisfy 's' when -mpcrel is not in effect 43735 43736 'Q' 43737 Address register indirect addressing mode 43738 43739 'U' 43740 Register offset addressing 43741 43742 'W' 43743 const_call_operand 43744 43745 'Cs' 43746 symbol_ref or const 43747 43748 'Ci' 43749 const_int 43750 43751 'C0' 43752 const_int 0 43753 43754 'Cj' 43755 Range of signed numbers that don't fit in 16 bits 43756 43757 'Cmvq' 43758 Integers valid for mvq 43759 43760 'Capsw' 43761 Integers valid for a moveq followed by a swap 43762 43763 'Cmvz' 43764 Integers valid for mvz 43765 43766 'Cmvs' 43767 Integers valid for mvs 43768 43769 'Ap' 43770 push_operand 43771 43772 'Ac' 43773 Non-register operands allowed in clr 43774 43775_Moxie--'config/moxie/constraints.md'_ 43776 'A' 43777 An absolute address 43778 43779 'B' 43780 An offset address 43781 43782 'W' 43783 A register indirect memory operand 43784 43785 'I' 43786 A constant in the range of 0 to 255. 43787 43788 'N' 43789 A constant in the range of 0 to -255. 43790 43791_MSP430-'config/msp430/constraints.md'_ 43792 43793 'R12' 43794 Register R12. 43795 43796 'R13' 43797 Register R13. 43798 43799 'K' 43800 Integer constant 1. 43801 43802 'L' 43803 Integer constant -1^20..1^19. 43804 43805 'M' 43806 Integer constant 1-4. 43807 43808 'Ya' 43809 Memory references which do not require an extended MOVX 43810 instruction. 43811 43812 'Yl' 43813 Memory reference, labels only. 43814 43815 'Ys' 43816 Memory reference, stack only. 43817 43818_NDS32--'config/nds32/constraints.md'_ 43819 'w' 43820 LOW register class $r0 to $r7 constraint for V3/V3M ISA. 43821 'l' 43822 LOW register class $r0 to $r7. 43823 'd' 43824 MIDDLE register class $r0 to $r11, $r16 to $r19. 43825 'h' 43826 HIGH register class $r12 to $r14, $r20 to $r31. 43827 't' 43828 Temporary assist register $ta (i.e. $r15). 43829 'k' 43830 Stack register $sp. 43831 'Iu03' 43832 Unsigned immediate 3-bit value. 43833 'In03' 43834 Negative immediate 3-bit value in the range of -7-0. 43835 'Iu04' 43836 Unsigned immediate 4-bit value. 43837 'Is05' 43838 Signed immediate 5-bit value. 43839 'Iu05' 43840 Unsigned immediate 5-bit value. 43841 'In05' 43842 Negative immediate 5-bit value in the range of -31-0. 43843 'Ip05' 43844 Unsigned immediate 5-bit value for movpi45 instruction with 43845 range 16-47. 43846 'Iu06' 43847 Unsigned immediate 6-bit value constraint for addri36.sp 43848 instruction. 43849 'Iu08' 43850 Unsigned immediate 8-bit value. 43851 'Iu09' 43852 Unsigned immediate 9-bit value. 43853 'Is10' 43854 Signed immediate 10-bit value. 43855 'Is11' 43856 Signed immediate 11-bit value. 43857 'Is15' 43858 Signed immediate 15-bit value. 43859 'Iu15' 43860 Unsigned immediate 15-bit value. 43861 'Ic15' 43862 A constant which is not in the range of imm15u but ok for bclr 43863 instruction. 43864 'Ie15' 43865 A constant which is not in the range of imm15u but ok for bset 43866 instruction. 43867 'It15' 43868 A constant which is not in the range of imm15u but ok for btgl 43869 instruction. 43870 'Ii15' 43871 A constant whose compliment value is in the range of imm15u 43872 and ok for bitci instruction. 43873 'Is16' 43874 Signed immediate 16-bit value. 43875 'Is17' 43876 Signed immediate 17-bit value. 43877 'Is19' 43878 Signed immediate 19-bit value. 43879 'Is20' 43880 Signed immediate 20-bit value. 43881 'Ihig' 43882 The immediate value that can be simply set high 20-bit. 43883 'Izeb' 43884 The immediate value 0xff. 43885 'Izeh' 43886 The immediate value 0xffff. 43887 'Ixls' 43888 The immediate value 0x01. 43889 'Ix11' 43890 The immediate value 0x7ff. 43891 'Ibms' 43892 The immediate value with power of 2. 43893 'Ifex' 43894 The immediate value with power of 2 minus 1. 43895 'U33' 43896 Memory constraint for 333 format. 43897 'U45' 43898 Memory constraint for 45 format. 43899 'U37' 43900 Memory constraint for 37 format. 43901 43902_Nios II family--'config/nios2/constraints.md'_ 43903 43904 'I' 43905 Integer that is valid as an immediate operand in an 43906 instruction taking a signed 16-bit number. Range -32768 to 43907 32767. 43908 43909 'J' 43910 Integer that is valid as an immediate operand in an 43911 instruction taking an unsigned 16-bit number. Range 0 to 43912 65535. 43913 43914 'K' 43915 Integer that is valid as an immediate operand in an 43916 instruction taking only the upper 16-bits of a 32-bit number. 43917 Range 32-bit numbers with the lower 16-bits being 0. 43918 43919 'L' 43920 Integer that is valid as an immediate operand for a shift 43921 instruction. Range 0 to 31. 43922 43923 'M' 43924 Integer that is valid as an immediate operand for only the 43925 value 0. Can be used in conjunction with the format modifier 43926 'z' to use 'r0' instead of '0' in the assembly output. 43927 43928 'N' 43929 Integer that is valid as an immediate operand for a custom 43930 instruction opcode. Range 0 to 255. 43931 43932 'P' 43933 An immediate operand for R2 andchi/andci instructions. 43934 43935 'S' 43936 Matches immediates which are addresses in the small data 43937 section and therefore can be added to 'gp' as a 16-bit 43938 immediate to re-create their 32-bit value. 43939 43940 'U' 43941 Matches constants suitable as an operand for the rdprs and 43942 cache instructions. 43943 43944 'v' 43945 A memory operand suitable for Nios II R2 load/store exclusive 43946 instructions. 43947 43948 'w' 43949 A memory operand suitable for load/store IO and cache 43950 instructions. 43951 43952_OpenRISC--'config/or1k/constraints.md'_ 43953 'I' 43954 Integer that is valid as an immediate operand in an 43955 instruction taking a signed 16-bit number. Range -32768 to 43956 32767. 43957 43958 'K' 43959 Integer that is valid as an immediate operand in an 43960 instruction taking an unsigned 16-bit number. Range 0 to 43961 65535. 43962 43963 'M' 43964 Signed 16-bit constant shifted left 16 bits. (Used with 43965 'l.movhi') 43966 43967 'O' 43968 Zero 43969 43970_PDP-11--'config/pdp11/constraints.md'_ 43971 'a' 43972 Floating point registers AC0 through AC3. These can be loaded 43973 from/to memory with a single instruction. 43974 43975 'd' 43976 Odd numbered general registers (R1, R3, R5). These are used 43977 for 16-bit multiply operations. 43978 43979 'D' 43980 A memory reference that is encoded within the opcode, but not 43981 auto-increment or auto-decrement. 43982 43983 'f' 43984 Any of the floating point registers (AC0 through AC5). 43985 43986 'G' 43987 Floating point constant 0. 43988 43989 'h' 43990 Floating point registers AC4 and AC5. These cannot be loaded 43991 from/to memory with a single instruction. 43992 43993 'I' 43994 An integer constant that fits in 16 bits. 43995 43996 'J' 43997 An integer constant whose low order 16 bits are zero. 43998 43999 'K' 44000 An integer constant that does not meet the constraints for 44001 codes 'I' or 'J'. 44002 44003 'L' 44004 The integer constant 1. 44005 44006 'M' 44007 The integer constant -1. 44008 44009 'N' 44010 The integer constant 0. 44011 44012 'O' 44013 Integer constants 0 through 3; shifts by these amounts are 44014 handled as multiple single-bit shifts rather than a single 44015 variable-length shift. 44016 44017 'Q' 44018 A memory reference which requires an additional word (address 44019 or offset) after the opcode. 44020 44021 'R' 44022 A memory reference that is encoded within the opcode. 44023 44024_PowerPC and IBM RS6000--'config/rs6000/constraints.md'_ 44025 'r' 44026 A general purpose register (GPR), 'r0'...'r31'. 44027 44028 'b' 44029 A base register. Like 'r', but 'r0' is not allowed, so 44030 'r1'...'r31'. 44031 44032 'f' 44033 A floating point register (FPR), 'f0'...'f31'. 44034 44035 'd' 44036 A floating point register. This is the same as 'f' nowadays; 44037 historically 'f' was for single-precision and 'd' was for 44038 double-precision floating point. 44039 44040 'v' 44041 An Altivec vector register (VR), 'v0'...'v31'. 44042 44043 'wa' 44044 A VSX register (VSR), 'vs0'...'vs63'. This is either an FPR 44045 ('vs0'...'vs31' are 'f0'...'f31') or a VR ('vs32'...'vs63' are 44046 'v0'...'v31'). 44047 44048 When using 'wa', you should use the '%x' output modifier, so 44049 that the correct register number is printed. For example: 44050 44051 asm ("xvadddp %x0,%x1,%x2" 44052 : "=wa" (v1) 44053 : "wa" (v2), "wa" (v3)); 44054 44055 You should not use '%x' for 'v' operands: 44056 44057 asm ("xsaddqp %0,%1,%2" 44058 : "=v" (v1) 44059 : "v" (v2), "v" (v3)); 44060 44061 'c' 44062 The count register, 'ctr'. 44063 44064 'l' 44065 The link register, 'lr'. 44066 44067 'x' 44068 Condition register field 0, 'cr0'. 44069 44070 'y' 44071 Any condition register field, 'cr0'...'cr7'. 44072 44073 'I' 44074 A signed 16-bit constant. 44075 44076 'J' 44077 An unsigned 16-bit constant shifted left 16 bits (use 'L' 44078 instead for 'SImode' constants). 44079 44080 'K' 44081 An unsigned 16-bit constant. 44082 44083 'L' 44084 A signed 16-bit constant shifted left 16 bits. 44085 44086 'eI' 44087 A signed 34-bit integer constant if prefixed instructions are 44088 supported. 44089 44090 'eP' 44091 A scalar floating point constant or a vector constant that can 44092 be loaded to a VSX register with one prefixed instruction. 44093 44094 'eQ' 44095 An IEEE 128-bit constant that can be loaded into a VSX 44096 register with the 'lxvkq' instruction. 44097 44098 'm' 44099 A memory operand. Normally, 'm' does not allow addresses that 44100 update the base register. If the '<' or '>' constraint is 44101 also used, they are allowed and therefore on PowerPC targets 44102 in that case it is only safe to use 'm<>' in an 'asm' 44103 statement if that 'asm' statement accesses the operand exactly 44104 once. The 'asm' statement must also use '%U<OPNO>' as a 44105 placeholder for the "update" flag in the corresponding load or 44106 store instruction. For example: 44107 44108 asm ("st%U0 %1,%0" : "=m<>" (mem) : "r" (val)); 44109 44110 is correct but: 44111 44112 asm ("st %1,%0" : "=m<>" (mem) : "r" (val)); 44113 44114 is not. 44115 44116 'Q' 44117 A memory operand addressed by just a base register. 44118 44119 'Z' 44120 A memory operand accessed with indexed or indirect addressing. 44121 44122 'a' 44123 An indexed or indirect address. 44124 44125_PRU--'config/pru/constraints.md'_ 44126 'I' 44127 An unsigned 8-bit integer constant. 44128 44129 'J' 44130 An unsigned 16-bit integer constant. 44131 44132 'L' 44133 An unsigned 5-bit integer constant (for shift counts). 44134 44135 'T' 44136 A text segment (program memory) constant label. 44137 44138 'Z' 44139 Integer constant zero. 44140 44141_RL78--'config/rl78/constraints.md'_ 44142 44143 'Int3' 44144 An integer constant in the range 1 ... 7. 44145 'Int8' 44146 An integer constant in the range 0 ... 255. 44147 'J' 44148 An integer constant in the range -255 ... 0 44149 'K' 44150 The integer constant 1. 44151 'L' 44152 The integer constant -1. 44153 'M' 44154 The integer constant 0. 44155 'N' 44156 The integer constant 2. 44157 'O' 44158 The integer constant -2. 44159 'P' 44160 An integer constant in the range 1 ... 15. 44161 'Qbi' 44162 The built-in compare types-eq, ne, gtu, ltu, geu, and leu. 44163 'Qsc' 44164 The synthetic compare types-gt, lt, ge, and le. 44165 'Wab' 44166 A memory reference with an absolute address. 44167 'Wbc' 44168 A memory reference using 'BC' as a base register, with an 44169 optional offset. 44170 'Wca' 44171 A memory reference using 'AX', 'BC', 'DE', or 'HL' for the 44172 address, for calls. 44173 'Wcv' 44174 A memory reference using any 16-bit register pair for the 44175 address, for calls. 44176 'Wd2' 44177 A memory reference using 'DE' as a base register, with an 44178 optional offset. 44179 'Wde' 44180 A memory reference using 'DE' as a base register, without any 44181 offset. 44182 'Wfr' 44183 Any memory reference to an address in the far address space. 44184 'Wh1' 44185 A memory reference using 'HL' as a base register, with an 44186 optional one-byte offset. 44187 'Whb' 44188 A memory reference using 'HL' as a base register, with 'B' or 44189 'C' as the index register. 44190 'Whl' 44191 A memory reference using 'HL' as a base register, without any 44192 offset. 44193 'Ws1' 44194 A memory reference using 'SP' as a base register, with an 44195 optional one-byte offset. 44196 'Y' 44197 Any memory reference to an address in the near address space. 44198 'A' 44199 The 'AX' register. 44200 'B' 44201 The 'BC' register. 44202 'D' 44203 The 'DE' register. 44204 'R' 44205 'A' through 'L' registers. 44206 'S' 44207 The 'SP' register. 44208 'T' 44209 The 'HL' register. 44210 'Z08W' 44211 The 16-bit 'R8' register. 44212 'Z10W' 44213 The 16-bit 'R10' register. 44214 'Zint' 44215 The registers reserved for interrupts ('R24' to 'R31'). 44216 'a' 44217 The 'A' register. 44218 'b' 44219 The 'B' register. 44220 'c' 44221 The 'C' register. 44222 'd' 44223 The 'D' register. 44224 'e' 44225 The 'E' register. 44226 'h' 44227 The 'H' register. 44228 'l' 44229 The 'L' register. 44230 'v' 44231 The virtual registers. 44232 'w' 44233 The 'PSW' register. 44234 'x' 44235 The 'X' register. 44236 44237_RISC-V--'config/riscv/constraints.md'_ 44238 44239 'f' 44240 A floating-point register (if available). 44241 44242 'I' 44243 An I-type 12-bit signed immediate. 44244 44245 'J' 44246 Integer zero. 44247 44248 'K' 44249 A 5-bit unsigned immediate for CSR access instructions. 44250 44251 'A' 44252 An address that is held in a general-purpose register. 44253 44254 'S' 44255 A constraint that matches an absolute symbolic address. 44256 44257_RX--'config/rx/constraints.md'_ 44258 'Q' 44259 An address which does not involve register indirect addressing 44260 or pre/post increment/decrement addressing. 44261 44262 'Symbol' 44263 A symbol reference. 44264 44265 'Int08' 44266 A constant in the range -256 to 255, inclusive. 44267 44268 'Sint08' 44269 A constant in the range -128 to 127, inclusive. 44270 44271 'Sint16' 44272 A constant in the range -32768 to 32767, inclusive. 44273 44274 'Sint24' 44275 A constant in the range -8388608 to 8388607, inclusive. 44276 44277 'Uint04' 44278 A constant in the range 0 to 15, inclusive. 44279 44280_S/390 and zSeries--'config/s390/s390.h'_ 44281 'a' 44282 Address register (general purpose register except r0) 44283 44284 'c' 44285 Condition code register 44286 44287 'd' 44288 Data register (arbitrary general purpose register) 44289 44290 'f' 44291 Floating-point register 44292 44293 'I' 44294 Unsigned 8-bit constant (0-255) 44295 44296 'J' 44297 Unsigned 12-bit constant (0-4095) 44298 44299 'K' 44300 Signed 16-bit constant (-32768-32767) 44301 44302 'L' 44303 Value appropriate as displacement. 44304 '(0..4095)' 44305 for short displacement 44306 '(-524288..524287)' 44307 for long displacement 44308 44309 'M' 44310 Constant integer with a value of 0x7fffffff. 44311 44312 'N' 44313 Multiple letter constraint followed by 4 parameter letters. 44314 '0..9:' 44315 number of the part counting from most to least 44316 significant 44317 'H,Q:' 44318 mode of the part 44319 'D,S,H:' 44320 mode of the containing operand 44321 '0,F:' 44322 value of the other parts (F--all bits set) 44323 The constraint matches if the specified part of a constant has 44324 a value different from its other parts. 44325 44326 'Q' 44327 Memory reference without index register and with short 44328 displacement. 44329 44330 'R' 44331 Memory reference with index register and short displacement. 44332 44333 'S' 44334 Memory reference without index register but with long 44335 displacement. 44336 44337 'T' 44338 Memory reference with index register and long displacement. 44339 44340 'U' 44341 Pointer with short displacement. 44342 44343 'W' 44344 Pointer with long displacement. 44345 44346 'Y' 44347 Shift count operand. 44348 44349_SPARC--'config/sparc/sparc.h'_ 44350 'f' 44351 Floating-point register on the SPARC-V8 architecture and lower 44352 floating-point register on the SPARC-V9 architecture. 44353 44354 'e' 44355 Floating-point register. It is equivalent to 'f' on the 44356 SPARC-V8 architecture and contains both lower and upper 44357 floating-point registers on the SPARC-V9 architecture. 44358 44359 'c' 44360 Floating-point condition code register. 44361 44362 'd' 44363 Lower floating-point register. It is only valid on the 44364 SPARC-V9 architecture when the Visual Instruction Set is 44365 available. 44366 44367 'b' 44368 Floating-point register. It is only valid on the SPARC-V9 44369 architecture when the Visual Instruction Set is available. 44370 44371 'h' 44372 64-bit global or out register for the SPARC-V8+ architecture. 44373 44374 'C' 44375 The constant all-ones, for floating-point. 44376 44377 'A' 44378 Signed 5-bit constant 44379 44380 'D' 44381 A vector constant 44382 44383 'I' 44384 Signed 13-bit constant 44385 44386 'J' 44387 Zero 44388 44389 'K' 44390 32-bit constant with the low 12 bits clear (a constant that 44391 can be loaded with the 'sethi' instruction) 44392 44393 'L' 44394 A constant in the range supported by 'movcc' instructions 44395 (11-bit signed immediate) 44396 44397 'M' 44398 A constant in the range supported by 'movrcc' instructions 44399 (10-bit signed immediate) 44400 44401 'N' 44402 Same as 'K', except that it verifies that bits that are not in 44403 the lower 32-bit range are all zero. Must be used instead of 44404 'K' for modes wider than 'SImode' 44405 44406 'O' 44407 The constant 4096 44408 44409 'G' 44410 Floating-point zero 44411 44412 'H' 44413 Signed 13-bit constant, sign-extended to 32 or 64 bits 44414 44415 'P' 44416 The constant -1 44417 44418 'Q' 44419 Floating-point constant whose integral representation can be 44420 moved into an integer register using a single sethi 44421 instruction 44422 44423 'R' 44424 Floating-point constant whose integral representation can be 44425 moved into an integer register using a single mov instruction 44426 44427 'S' 44428 Floating-point constant whose integral representation can be 44429 moved into an integer register using a high/lo_sum instruction 44430 sequence 44431 44432 'T' 44433 Memory address aligned to an 8-byte boundary 44434 44435 'U' 44436 Even register 44437 44438 'W' 44439 Memory address for 'e' constraint registers 44440 44441 'w' 44442 Memory address with only a base register 44443 44444 'Y' 44445 Vector zero 44446 44447_TI C6X family--'config/c6x/constraints.md'_ 44448 'a' 44449 Register file A (A0-A31). 44450 44451 'b' 44452 Register file B (B0-B31). 44453 44454 'A' 44455 Predicate registers in register file A (A0-A2 on C64X and 44456 higher, A1 and A2 otherwise). 44457 44458 'B' 44459 Predicate registers in register file B (B0-B2). 44460 44461 'C' 44462 A call-used register in register file B (B0-B9, B16-B31). 44463 44464 'Da' 44465 Register file A, excluding predicate registers (A3-A31, plus 44466 A0 if not C64X or higher). 44467 44468 'Db' 44469 Register file B, excluding predicate registers (B3-B31). 44470 44471 'Iu4' 44472 Integer constant in the range 0 ... 15. 44473 44474 'Iu5' 44475 Integer constant in the range 0 ... 31. 44476 44477 'In5' 44478 Integer constant in the range -31 ... 0. 44479 44480 'Is5' 44481 Integer constant in the range -16 ... 15. 44482 44483 'I5x' 44484 Integer constant that can be the operand of an ADDA or a SUBA 44485 insn. 44486 44487 'IuB' 44488 Integer constant in the range 0 ... 65535. 44489 44490 'IsB' 44491 Integer constant in the range -32768 ... 32767. 44492 44493 'IsC' 44494 Integer constant in the range -2^{20} ... 2^{20} - 1. 44495 44496 'Jc' 44497 Integer constant that is a valid mask for the clr instruction. 44498 44499 'Js' 44500 Integer constant that is a valid mask for the set instruction. 44501 44502 'Q' 44503 Memory location with A base register. 44504 44505 'R' 44506 Memory location with B base register. 44507 44508 'Z' 44509 Register B14 (aka DP). 44510 44511_TILE-Gx--'config/tilegx/constraints.md'_ 44512 'R00' 44513 'R01' 44514 'R02' 44515 'R03' 44516 'R04' 44517 'R05' 44518 'R06' 44519 'R07' 44520 'R08' 44521 'R09' 44522 'R10' 44523 Each of these represents a register constraint for an 44524 individual register, from r0 to r10. 44525 44526 'I' 44527 Signed 8-bit integer constant. 44528 44529 'J' 44530 Signed 16-bit integer constant. 44531 44532 'K' 44533 Unsigned 16-bit integer constant. 44534 44535 'L' 44536 Integer constant that fits in one signed byte when incremented 44537 by one (-129 ... 126). 44538 44539 'm' 44540 Memory operand. If used together with '<' or '>', the operand 44541 can have postincrement which requires printing with '%In' and 44542 '%in' on TILE-Gx. For example: 44543 44544 asm ("st_add %I0,%1,%i0" : "=m<>" (*mem) : "r" (val)); 44545 44546 'M' 44547 A bit mask suitable for the BFINS instruction. 44548 44549 'N' 44550 Integer constant that is a byte tiled out eight times. 44551 44552 'O' 44553 The integer zero constant. 44554 44555 'P' 44556 Integer constant that is a sign-extended byte tiled out as 44557 four shorts. 44558 44559 'Q' 44560 Integer constant that fits in one signed byte when incremented 44561 (-129 ... 126), but excluding -1. 44562 44563 'S' 44564 Integer constant that has all 1 bits consecutive and starting 44565 at bit 0. 44566 44567 'T' 44568 A 16-bit fragment of a got, tls, or pc-relative reference. 44569 44570 'U' 44571 Memory operand except postincrement. This is roughly the same 44572 as 'm' when not used together with '<' or '>'. 44573 44574 'W' 44575 An 8-element vector constant with identical elements. 44576 44577 'Y' 44578 A 4-element vector constant with identical elements. 44579 44580 'Z0' 44581 The integer constant 0xffffffff. 44582 44583 'Z1' 44584 The integer constant 0xffffffff00000000. 44585 44586_TILEPro--'config/tilepro/constraints.md'_ 44587 'R00' 44588 'R01' 44589 'R02' 44590 'R03' 44591 'R04' 44592 'R05' 44593 'R06' 44594 'R07' 44595 'R08' 44596 'R09' 44597 'R10' 44598 Each of these represents a register constraint for an 44599 individual register, from r0 to r10. 44600 44601 'I' 44602 Signed 8-bit integer constant. 44603 44604 'J' 44605 Signed 16-bit integer constant. 44606 44607 'K' 44608 Nonzero integer constant with low 16 bits zero. 44609 44610 'L' 44611 Integer constant that fits in one signed byte when incremented 44612 by one (-129 ... 126). 44613 44614 'm' 44615 Memory operand. If used together with '<' or '>', the operand 44616 can have postincrement which requires printing with '%In' and 44617 '%in' on TILEPro. For example: 44618 44619 asm ("swadd %I0,%1,%i0" : "=m<>" (mem) : "r" (val)); 44620 44621 'M' 44622 A bit mask suitable for the MM instruction. 44623 44624 'N' 44625 Integer constant that is a byte tiled out four times. 44626 44627 'O' 44628 The integer zero constant. 44629 44630 'P' 44631 Integer constant that is a sign-extended byte tiled out as two 44632 shorts. 44633 44634 'Q' 44635 Integer constant that fits in one signed byte when incremented 44636 (-129 ... 126), but excluding -1. 44637 44638 'T' 44639 A symbolic operand, or a 16-bit fragment of a got, tls, or 44640 pc-relative reference. 44641 44642 'U' 44643 Memory operand except postincrement. This is roughly the same 44644 as 'm' when not used together with '<' or '>'. 44645 44646 'W' 44647 A 4-element vector constant with identical elements. 44648 44649 'Y' 44650 A 2-element vector constant with identical elements. 44651 44652_Visium--'config/visium/constraints.md'_ 44653 'b' 44654 EAM register 'mdb' 44655 44656 'c' 44657 EAM register 'mdc' 44658 44659 'f' 44660 Floating point register 44661 44662 'l' 44663 General register, but not 'r29', 'r30' and 'r31' 44664 44665 't' 44666 Register 'r1' 44667 44668 'u' 44669 Register 'r2' 44670 44671 'v' 44672 Register 'r3' 44673 44674 'G' 44675 Floating-point constant 0.0 44676 44677 'J' 44678 Integer constant in the range 0 .. 65535 (16-bit immediate) 44679 44680 'K' 44681 Integer constant in the range 1 .. 31 (5-bit immediate) 44682 44683 'L' 44684 Integer constant in the range -65535 .. -1 (16-bit negative 44685 immediate) 44686 44687 'M' 44688 Integer constant -1 44689 44690 'O' 44691 Integer constant 0 44692 44693 'P' 44694 Integer constant 32 44695 44696_x86 family--'config/i386/constraints.md'_ 44697 'R' 44698 Legacy register--the eight integer registers available on all 44699 i386 processors ('a', 'b', 'c', 'd', 'si', 'di', 'bp', 'sp'). 44700 44701 'q' 44702 Any register accessible as 'Rl'. In 32-bit mode, 'a', 'b', 44703 'c', and 'd'; in 64-bit mode, any integer register. 44704 44705 'Q' 44706 Any register accessible as 'Rh': 'a', 'b', 'c', and 'd'. 44707 44708 'a' 44709 The 'a' register. 44710 44711 'b' 44712 The 'b' register. 44713 44714 'c' 44715 The 'c' register. 44716 44717 'd' 44718 The 'd' register. 44719 44720 'S' 44721 The 'si' register. 44722 44723 'D' 44724 The 'di' register. 44725 44726 'A' 44727 The 'a' and 'd' registers. This class is used for 44728 instructions that return double word results in the 'ax:dx' 44729 register pair. Single word values will be allocated either in 44730 'ax' or 'dx'. For example on i386 the following implements 44731 'rdtsc': 44732 44733 unsigned long long rdtsc (void) 44734 { 44735 unsigned long long tick; 44736 __asm__ __volatile__("rdtsc":"=A"(tick)); 44737 return tick; 44738 } 44739 44740 This is not correct on x86-64 as it would allocate tick in 44741 either 'ax' or 'dx'. You have to use the following variant 44742 instead: 44743 44744 unsigned long long rdtsc (void) 44745 { 44746 unsigned int tickl, tickh; 44747 __asm__ __volatile__("rdtsc":"=a"(tickl),"=d"(tickh)); 44748 return ((unsigned long long)tickh << 32)|tickl; 44749 } 44750 44751 'U' 44752 The call-clobbered integer registers. 44753 44754 'f' 44755 Any 80387 floating-point (stack) register. 44756 44757 't' 44758 Top of 80387 floating-point stack ('%st(0)'). 44759 44760 'u' 44761 Second from top of 80387 floating-point stack ('%st(1)'). 44762 44763 'y' 44764 Any MMX register. 44765 44766 'x' 44767 Any SSE register. 44768 44769 'v' 44770 Any EVEX encodable SSE register ('%xmm0-%xmm31'). 44771 44772 'Yz' 44773 First SSE register ('%xmm0'). 44774 44775 'I' 44776 Integer constant in the range 0 ... 31, for 32-bit shifts. 44777 44778 'J' 44779 Integer constant in the range 0 ... 63, for 64-bit shifts. 44780 44781 'K' 44782 Signed 8-bit integer constant. 44783 44784 'L' 44785 '0xFF' or '0xFFFF', for andsi as a zero-extending move. 44786 44787 'M' 44788 0, 1, 2, or 3 (shifts for the 'lea' instruction). 44789 44790 'N' 44791 Unsigned 8-bit integer constant (for 'in' and 'out' 44792 instructions). 44793 44794 'G' 44795 Standard 80387 floating point constant. 44796 44797 'C' 44798 SSE constant zero operand. 44799 44800 'e' 44801 32-bit signed integer constant, or a symbolic reference known 44802 to fit that range (for immediate operands in sign-extending 44803 x86-64 instructions). 44804 44805 'We' 44806 32-bit signed integer constant, or a symbolic reference known 44807 to fit that range (for sign-extending conversion operations 44808 that require non-'VOIDmode' immediate operands). 44809 44810 'Wz' 44811 32-bit unsigned integer constant, or a symbolic reference 44812 known to fit that range (for zero-extending conversion 44813 operations that require non-'VOIDmode' immediate operands). 44814 44815 'Wd' 44816 128-bit integer constant where both the high and low 64-bit 44817 word satisfy the 'e' constraint. 44818 44819 'Z' 44820 32-bit unsigned integer constant, or a symbolic reference 44821 known to fit that range (for immediate operands in 44822 zero-extending x86-64 instructions). 44823 44824 'Tv' 44825 VSIB address operand. 44826 44827 'Ts' 44828 Address operand without segment register. 44829 44830_Xstormy16--'config/stormy16/stormy16.h'_ 44831 'a' 44832 Register r0. 44833 44834 'b' 44835 Register r1. 44836 44837 'c' 44838 Register r2. 44839 44840 'd' 44841 Register r8. 44842 44843 'e' 44844 Registers r0 through r7. 44845 44846 't' 44847 Registers r0 and r1. 44848 44849 'y' 44850 The carry register. 44851 44852 'z' 44853 Registers r8 and r9. 44854 44855 'I' 44856 A constant between 0 and 3 inclusive. 44857 44858 'J' 44859 A constant that has exactly one bit set. 44860 44861 'K' 44862 A constant that has exactly one bit clear. 44863 44864 'L' 44865 A constant between 0 and 255 inclusive. 44866 44867 'M' 44868 A constant between -255 and 0 inclusive. 44869 44870 'N' 44871 A constant between -3 and 0 inclusive. 44872 44873 'O' 44874 A constant between 1 and 4 inclusive. 44875 44876 'P' 44877 A constant between -4 and -1 inclusive. 44878 44879 'Q' 44880 A memory reference that is a stack push. 44881 44882 'R' 44883 A memory reference that is a stack pop. 44884 44885 'S' 44886 A memory reference that refers to a constant address of known 44887 value. 44888 44889 'T' 44890 The register indicated by Rx (not implemented yet). 44891 44892 'U' 44893 A constant that is not between 2 and 15 inclusive. 44894 44895 'Z' 44896 The constant 0. 44897 44898_Xtensa--'config/xtensa/constraints.md'_ 44899 'a' 44900 General-purpose 32-bit register 44901 44902 'b' 44903 One-bit boolean register 44904 44905 'A' 44906 MAC16 40-bit accumulator register 44907 44908 'I' 44909 Signed 12-bit integer constant, for use in MOVI instructions 44910 44911 'J' 44912 Signed 8-bit integer constant, for use in ADDI instructions 44913 44914 'K' 44915 Integer constant valid for BccI instructions 44916 44917 'L' 44918 Unsigned constant valid for BccUI instructions 44919 44920 44921File: gcc.info, Node: Asm Labels, Next: Explicit Register Variables, Prev: Constraints, Up: Using Assembly Language with C 44922 449236.47.4 Controlling Names Used in Assembler Code 44924----------------------------------------------- 44925 44926You can specify the name to be used in the assembler code for a C 44927function or variable by writing the 'asm' (or '__asm__') keyword after 44928the declarator. It is up to you to make sure that the assembler names 44929you choose do not conflict with any other assembler symbols, or 44930reference registers. 44931 44932Assembler names for data: 44933......................... 44934 44935This sample shows how to specify the assembler name for data: 44936 44937 int foo asm ("myfoo") = 2; 44938 44939This specifies that the name to be used for the variable 'foo' in the 44940assembler code should be 'myfoo' rather than the usual '_foo'. 44941 44942 On systems where an underscore is normally prepended to the name of a C 44943variable, this feature allows you to define names for the linker that do 44944not start with an underscore. 44945 44946 GCC does not support using this feature with a non-static local 44947variable since such variables do not have assembler names. If you are 44948trying to put the variable in a particular register, see *note Explicit 44949Register Variables::. 44950 44951Assembler names for functions: 44952.............................. 44953 44954To specify the assembler name for functions, write a declaration for the 44955function before its definition and put 'asm' there, like this: 44956 44957 int func (int x, int y) asm ("MYFUNC"); 44958 44959 int func (int x, int y) 44960 { 44961 /* ... */ 44962 44963This specifies that the name to be used for the function 'func' in the 44964assembler code should be 'MYFUNC'. 44965 44966 44967File: gcc.info, Node: Explicit Register Variables, Next: Size of an asm, Prev: Asm Labels, Up: Using Assembly Language with C 44968 449696.47.5 Variables in Specified Registers 44970--------------------------------------- 44971 44972GNU C allows you to associate specific hardware registers with C 44973variables. In almost all cases, allowing the compiler to assign 44974registers produces the best code. However under certain unusual 44975circumstances, more precise control over the variable storage is 44976required. 44977 44978 Both global and local variables can be associated with a register. The 44979consequences of performing this association are very different between 44980the two, as explained in the sections below. 44981 44982* Menu: 44983 44984* Global Register Variables:: Variables declared at global scope. 44985* Local Register Variables:: Variables declared within a function. 44986 44987 44988File: gcc.info, Node: Global Register Variables, Next: Local Register Variables, Up: Explicit Register Variables 44989 449906.47.5.1 Defining Global Register Variables 44991........................................... 44992 44993You can define a global register variable and associate it with a 44994specified register like this: 44995 44996 register int *foo asm ("r12"); 44997 44998Here 'r12' is the name of the register that should be used. Note that 44999this is the same syntax used for defining local register variables, but 45000for a global variable the declaration appears outside a function. The 45001'register' keyword is required, and cannot be combined with 'static'. 45002The register name must be a valid register name for the target platform. 45003 45004 Do not use type qualifiers such as 'const' and 'volatile', as the 45005outcome may be contrary to expectations. In particular, using the 45006'volatile' qualifier does not fully prevent the compiler from optimizing 45007accesses to the register. 45008 45009 Registers are a scarce resource on most systems and allowing the 45010compiler to manage their usage usually results in the best code. 45011However, under special circumstances it can make sense to reserve some 45012globally. For example this may be useful in programs such as 45013programming language interpreters that have a couple of global variables 45014that are accessed very often. 45015 45016 After defining a global register variable, for the current compilation 45017unit: 45018 45019 * If the register is a call-saved register, call ABI is affected: the 45020 register will not be restored in function epilogue sequences after 45021 the variable has been assigned. Therefore, functions cannot safely 45022 return to callers that assume standard ABI. 45023 * Conversely, if the register is a call-clobbered register, making 45024 calls to functions that use standard ABI may lose contents of the 45025 variable. Such calls may be created by the compiler even if none 45026 are evident in the original program, for example when libgcc 45027 functions are used to make up for unavailable instructions. 45028 * Accesses to the variable may be optimized as usual and the register 45029 remains available for allocation and use in any computations, 45030 provided that observable values of the variable are not affected. 45031 * If the variable is referenced in inline assembly, the type of 45032 access must be provided to the compiler via constraints (*note 45033 Constraints::). Accesses from basic asms are not supported. 45034 45035 Note that these points _only_ apply to code that is compiled with the 45036definition. The behavior of code that is merely linked in (for example 45037code from libraries) is not affected. 45038 45039 If you want to recompile source files that do not actually use your 45040global register variable so they do not use the specified register for 45041any other purpose, you need not actually add the global register 45042declaration to their source code. It suffices to specify the compiler 45043option '-ffixed-REG' (*note Code Gen Options::) to reserve the register. 45044 45045Declaring the variable 45046...................... 45047 45048Global register variables cannot have initial values, because an 45049executable file has no means to supply initial contents for a register. 45050 45051 When selecting a register, choose one that is normally saved and 45052restored by function calls on your machine. This ensures that code 45053which is unaware of this reservation (such as library routines) will 45054restore it before returning. 45055 45056 On machines with register windows, be sure to choose a global register 45057that is not affected magically by the function call mechanism. 45058 45059Using the variable 45060.................. 45061 45062When calling routines that are not aware of the reservation, be cautious 45063if those routines call back into code which uses them. As an example, 45064if you call the system library version of 'qsort', it may clobber your 45065registers during execution, but (if you have selected appropriate 45066registers) it will restore them before returning. However it will _not_ 45067restore them before calling 'qsort''s comparison function. As a result, 45068global values will not reliably be available to the comparison function 45069unless the 'qsort' function itself is rebuilt. 45070 45071 Similarly, it is not safe to access the global register variables from 45072signal handlers or from more than one thread of control. Unless you 45073recompile them specially for the task at hand, the system library 45074routines may temporarily use the register for other things. 45075Furthermore, since the register is not reserved exclusively for the 45076variable, accessing it from handlers of asynchronous signals may observe 45077unrelated temporary values residing in the register. 45078 45079 On most machines, 'longjmp' restores to each global register variable 45080the value it had at the time of the 'setjmp'. On some machines, 45081however, 'longjmp' does not change the value of global register 45082variables. To be portable, the function that called 'setjmp' should 45083make other arrangements to save the values of the global register 45084variables, and to restore them in a 'longjmp'. This way, the same thing 45085happens regardless of what 'longjmp' does. 45086 45087 45088File: gcc.info, Node: Local Register Variables, Prev: Global Register Variables, Up: Explicit Register Variables 45089 450906.47.5.2 Specifying Registers for Local Variables 45091................................................. 45092 45093You can define a local register variable and associate it with a 45094specified register like this: 45095 45096 register int *foo asm ("r12"); 45097 45098Here 'r12' is the name of the register that should be used. Note that 45099this is the same syntax used for defining global register variables, but 45100for a local variable the declaration appears within a function. The 45101'register' keyword is required, and cannot be combined with 'static'. 45102The register name must be a valid register name for the target platform. 45103 45104 Do not use type qualifiers such as 'const' and 'volatile', as the 45105outcome may be contrary to expectations. In particular, when the 45106'const' qualifier is used, the compiler may substitute the variable with 45107its initializer in 'asm' statements, which may cause the corresponding 45108operand to appear in a different register. 45109 45110 As with global register variables, it is recommended that you choose a 45111register that is normally saved and restored by function calls on your 45112machine, so that calls to library routines will not clobber it. 45113 45114 The only supported use for this feature is to specify registers for 45115input and output operands when calling Extended 'asm' (*note Extended 45116Asm::). This may be necessary if the constraints for a particular 45117machine don't provide sufficient control to select the desired register. 45118To force an operand into a register, create a local variable and specify 45119the register name after the variable's declaration. Then use the local 45120variable for the 'asm' operand and specify any constraint letter that 45121matches the register: 45122 45123 register int *p1 asm ("r0") = ...; 45124 register int *p2 asm ("r1") = ...; 45125 register int *result asm ("r0"); 45126 asm ("sysint" : "=r" (result) : "0" (p1), "r" (p2)); 45127 45128 _Warning:_ In the above example, be aware that a register (for example 45129'r0') can be call-clobbered by subsequent code, including function calls 45130and library calls for arithmetic operators on other variables (for 45131example the initialization of 'p2'). In this case, use temporary 45132variables for expressions between the register assignments: 45133 45134 int t1 = ...; 45135 register int *p1 asm ("r0") = ...; 45136 register int *p2 asm ("r1") = t1; 45137 register int *result asm ("r0"); 45138 asm ("sysint" : "=r" (result) : "0" (p1), "r" (p2)); 45139 45140 Defining a register variable does not reserve the register. Other than 45141when invoking the Extended 'asm', the contents of the specified register 45142are not guaranteed. For this reason, the following uses are explicitly 45143_not_ supported. If they appear to work, it is only happenstance, and 45144may stop working as intended due to (seemingly) unrelated changes in 45145surrounding code, or even minor changes in the optimization of a future 45146version of gcc: 45147 45148 * Passing parameters to or from Basic 'asm' 45149 * Passing parameters to or from Extended 'asm' without using input or 45150 output operands. 45151 * Passing parameters to or from routines written in assembler (or 45152 other languages) using non-standard calling conventions. 45153 45154 Some developers use Local Register Variables in an attempt to improve 45155gcc's allocation of registers, especially in large functions. In this 45156case the register name is essentially a hint to the register allocator. 45157While in some instances this can generate better code, improvements are 45158subject to the whims of the allocator/optimizers. Since there are no 45159guarantees that your improvements won't be lost, this usage of Local 45160Register Variables is discouraged. 45161 45162 On the MIPS platform, there is related use for local register variables 45163with slightly different characteristics (*note Defining coprocessor 45164specifics for MIPS targets: (gccint)MIPS Coprocessors.). 45165 45166 45167File: gcc.info, Node: Size of an asm, Prev: Explicit Register Variables, Up: Using Assembly Language with C 45168 451696.47.6 Size of an 'asm' 45170----------------------- 45171 45172Some targets require that GCC track the size of each instruction used in 45173order to generate correct code. Because the final length of the code 45174produced by an 'asm' statement is only known by the assembler, GCC must 45175make an estimate as to how big it will be. It does this by counting the 45176number of instructions in the pattern of the 'asm' and multiplying that 45177by the length of the longest instruction supported by that processor. 45178(When working out the number of instructions, it assumes that any 45179occurrence of a newline or of whatever statement separator character is 45180supported by the assembler -- typically ';' -- indicates the end of an 45181instruction.) 45182 45183 Normally, GCC's estimate is adequate to ensure that correct code is 45184generated, but it is possible to confuse the compiler if you use pseudo 45185instructions or assembler macros that expand into multiple real 45186instructions, or if you use assembler directives that expand to more 45187space in the object file than is needed for a single instruction. If 45188this happens then the assembler may produce a diagnostic saying that a 45189label is unreachable. 45190 45191 This size is also used for inlining decisions. If you use 'asm inline' 45192instead of just 'asm', then for inlining purposes the size of the asm is 45193taken as the minimum size, ignoring how many instructions GCC thinks it 45194is. 45195 45196 45197File: gcc.info, Node: Alternate Keywords, Next: Incomplete Enums, Prev: Using Assembly Language with C, Up: C Extensions 45198 451996.48 Alternate Keywords 45200======================= 45201 45202'-ansi' and the various '-std' options disable certain keywords. This 45203causes trouble when you want to use GNU C extensions, or a 45204general-purpose header file that should be usable by all programs, 45205including ISO C programs. The keywords 'asm', 'typeof' and 'inline' are 45206not available in programs compiled with '-ansi' or '-std' (although 45207'inline' can be used in a program compiled with '-std=c99' or a later 45208standard). The ISO C99 keyword 'restrict' is only available when 45209'-std=gnu99' (which will eventually be the default) or '-std=c99' (or 45210the equivalent '-std=iso9899:1999'), or an option for a later standard 45211version, is used. 45212 45213 The way to solve these problems is to put '__' at the beginning and end 45214of each problematical keyword. For example, use '__asm__' instead of 45215'asm', and '__inline__' instead of 'inline'. 45216 45217 Other C compilers won't accept these alternative keywords; if you want 45218to compile with another compiler, you can define the alternate keywords 45219as macros to replace them with the customary keywords. It looks like 45220this: 45221 45222 #ifndef __GNUC__ 45223 #define __asm__ asm 45224 #endif 45225 45226 '-pedantic' and other options cause warnings for many GNU C extensions. 45227You can prevent such warnings within one expression by writing 45228'__extension__' before the expression. '__extension__' has no effect 45229aside from this. 45230 45231 45232File: gcc.info, Node: Incomplete Enums, Next: Function Names, Prev: Alternate Keywords, Up: C Extensions 45233 452346.49 Incomplete 'enum' Types 45235============================ 45236 45237You can define an 'enum' tag without specifying its possible values. 45238This results in an incomplete type, much like what you get if you write 45239'struct foo' without describing the elements. A later declaration that 45240does specify the possible values completes the type. 45241 45242 You cannot allocate variables or storage using the type while it is 45243incomplete. However, you can work with pointers to that type. 45244 45245 This extension may not be very useful, but it makes the handling of 45246'enum' more consistent with the way 'struct' and 'union' are handled. 45247 45248 This extension is not supported by GNU C++. 45249 45250 45251File: gcc.info, Node: Function Names, Next: Return Address, Prev: Incomplete Enums, Up: C Extensions 45252 452536.50 Function Names as Strings 45254============================== 45255 45256GCC provides three magic constants that hold the name of the current 45257function as a string. In C++11 and later modes, all three are treated 45258as constant expressions and can be used in 'constexpr' constexts. The 45259first of these constants is '__func__', which is part of the C99 45260standard: 45261 45262 The identifier '__func__' is implicitly declared by the translator as 45263if, immediately following the opening brace of each function definition, 45264the declaration 45265 45266 static const char __func__[] = "function-name"; 45267 45268appeared, where function-name is the name of the lexically-enclosing 45269function. This name is the unadorned name of the function. As an 45270extension, at file (or, in C++, namespace scope), '__func__' evaluates 45271to the empty string. 45272 45273 '__FUNCTION__' is another name for '__func__', provided for backward 45274compatibility with old versions of GCC. 45275 45276 In C, '__PRETTY_FUNCTION__' is yet another name for '__func__', except 45277that at file scope (or, in C++, namespace scope), it evaluates to the 45278string '"top level"'. In addition, in C++, '__PRETTY_FUNCTION__' 45279contains the signature of the function as well as its bare name. For 45280example, this program: 45281 45282 extern "C" int printf (const char *, ...); 45283 45284 class a { 45285 public: 45286 void sub (int i) 45287 { 45288 printf ("__FUNCTION__ = %s\n", __FUNCTION__); 45289 printf ("__PRETTY_FUNCTION__ = %s\n", __PRETTY_FUNCTION__); 45290 } 45291 }; 45292 45293 int 45294 main (void) 45295 { 45296 a ax; 45297 ax.sub (0); 45298 return 0; 45299 } 45300 45301gives this output: 45302 45303 __FUNCTION__ = sub 45304 __PRETTY_FUNCTION__ = void a::sub(int) 45305 45306 These identifiers are variables, not preprocessor macros, and may not 45307be used to initialize 'char' arrays or be concatenated with string 45308literals. 45309 45310 45311File: gcc.info, Node: Return Address, Next: Vector Extensions, Prev: Function Names, Up: C Extensions 45312 453136.51 Getting the Return or Frame Address of a Function 45314====================================================== 45315 45316These functions may be used to get information about the callers of a 45317function. 45318 45319 -- Built-in Function: void * __builtin_return_address (unsigned int 45320 LEVEL) 45321 This function returns the return address of the current function, 45322 or of one of its callers. The LEVEL argument is number of frames 45323 to scan up the call stack. A value of '0' yields the return 45324 address of the current function, a value of '1' yields the return 45325 address of the caller of the current function, and so forth. When 45326 inlining the expected behavior is that the function returns the 45327 address of the function that is returned to. To work around this 45328 behavior use the 'noinline' function attribute. 45329 45330 The LEVEL argument must be a constant integer. 45331 45332 On some machines it may be impossible to determine the return 45333 address of any function other than the current one; in such cases, 45334 or when the top of the stack has been reached, this function 45335 returns an unspecified value. In addition, 45336 '__builtin_frame_address' may be used to determine if the top of 45337 the stack has been reached. 45338 45339 Additional post-processing of the returned value may be needed, see 45340 '__builtin_extract_return_addr'. 45341 45342 The stored representation of the return address in memory may be 45343 different from the address returned by '__builtin_return_address'. 45344 For example, on AArch64 the stored address may be mangled with 45345 return address signing whereas the address returned by 45346 '__builtin_return_address' is not. 45347 45348 Calling this function with a nonzero argument can have 45349 unpredictable effects, including crashing the calling program. As 45350 a result, calls that are considered unsafe are diagnosed when the 45351 '-Wframe-address' option is in effect. Such calls should only be 45352 made in debugging situations. 45353 45354 On targets where code addresses are representable as 'void *', 45355 void *addr = __builtin_extract_return_addr (__builtin_return_address (0)); 45356 gives the code address where the current function would return. 45357 For example, such an address may be used with 'dladdr' or other 45358 interfaces that work with code addresses. 45359 45360 -- Built-in Function: void * __builtin_extract_return_addr (void *ADDR) 45361 The address as returned by '__builtin_return_address' may have to 45362 be fed through this function to get the actual encoded address. 45363 For example, on the 31-bit S/390 platform the highest bit has to be 45364 masked out, or on SPARC platforms an offset has to be added for the 45365 true next instruction to be executed. 45366 45367 If no fixup is needed, this function simply passes through ADDR. 45368 45369 -- Built-in Function: void * __builtin_frob_return_addr (void *ADDR) 45370 This function does the reverse of '__builtin_extract_return_addr'. 45371 45372 -- Built-in Function: void * __builtin_frame_address (unsigned int 45373 LEVEL) 45374 This function is similar to '__builtin_return_address', but it 45375 returns the address of the function frame rather than the return 45376 address of the function. Calling '__builtin_frame_address' with a 45377 value of '0' yields the frame address of the current function, a 45378 value of '1' yields the frame address of the caller of the current 45379 function, and so forth. 45380 45381 The frame is the area on the stack that holds local variables and 45382 saved registers. The frame address is normally the address of the 45383 first word pushed on to the stack by the function. However, the 45384 exact definition depends upon the processor and the calling 45385 convention. If the processor has a dedicated frame pointer 45386 register, and the function has a frame, then 45387 '__builtin_frame_address' returns the value of the frame pointer 45388 register. 45389 45390 On some machines it may be impossible to determine the frame 45391 address of any function other than the current one; in such cases, 45392 or when the top of the stack has been reached, this function 45393 returns '0' if the first frame pointer is properly initialized by 45394 the startup code. 45395 45396 Calling this function with a nonzero argument can have 45397 unpredictable effects, including crashing the calling program. As 45398 a result, calls that are considered unsafe are diagnosed when the 45399 '-Wframe-address' option is in effect. Such calls should only be 45400 made in debugging situations. 45401 45402 45403File: gcc.info, Node: Vector Extensions, Next: Offsetof, Prev: Return Address, Up: C Extensions 45404 454056.52 Using Vector Instructions through Built-in Functions 45406========================================================= 45407 45408On some targets, the instruction set contains SIMD vector instructions 45409which operate on multiple values contained in one large register at the 45410same time. For example, on the x86 the MMX, 3DNow! and SSE extensions 45411can be used this way. 45412 45413 The first step in using these extensions is to provide the necessary 45414data types. This should be done using an appropriate 'typedef': 45415 45416 typedef int v4si __attribute__ ((vector_size (16))); 45417 45418The 'int' type specifies the "base type", while the attribute specifies 45419the vector size for the variable, measured in bytes. For example, the 45420declaration above causes the compiler to set the mode for the 'v4si' 45421type to be 16 bytes wide and divided into 'int' sized units. For a 4542232-bit 'int' this means a vector of 4 units of 4 bytes, and the 45423corresponding mode of 'foo' is V4SI. 45424 45425 The 'vector_size' attribute is only applicable to integral and floating 45426scalars, although arrays, pointers, and function return values are 45427allowed in conjunction with this construct. Only sizes that are 45428positive power-of-two multiples of the base type size are currently 45429allowed. 45430 45431 All the basic integer types can be used as base types, both as signed 45432and as unsigned: 'char', 'short', 'int', 'long', 'long long'. In 45433addition, 'float' and 'double' can be used to build floating-point 45434vector types. 45435 45436 Specifying a combination that is not valid for the current architecture 45437causes GCC to synthesize the instructions using a narrower mode. For 45438example, if you specify a variable of type 'V4SI' and your architecture 45439does not allow for this specific SIMD type, GCC produces code that uses 454404 'SIs'. 45441 45442 The types defined in this manner can be used with a subset of normal C 45443operations. Currently, GCC allows using the following operators on 45444these types: '+, -, *, /, unary minus, ^, |, &, ~, %'. 45445 45446 The operations behave like C++ 'valarrays'. Addition is defined as the 45447addition of the corresponding elements of the operands. For example, in 45448the code below, each of the 4 elements in A is added to the 45449corresponding 4 elements in B and the resulting vector is stored in C. 45450 45451 typedef int v4si __attribute__ ((vector_size (16))); 45452 45453 v4si a, b, c; 45454 45455 c = a + b; 45456 45457 Subtraction, multiplication, division, and the logical operations 45458operate in a similar manner. Likewise, the result of using the unary 45459minus or complement operators on a vector type is a vector whose 45460elements are the negative or complemented values of the corresponding 45461elements in the operand. 45462 45463 It is possible to use shifting operators '<<', '>>' on integer-type 45464vectors. The operation is defined as following: '{a0, a1, ..., an} >> 45465{b0, b1, ..., bn} == {a0 >> b0, a1 >> b1, ..., an >> bn}'. Vector 45466operands must have the same number of elements. 45467 45468 For convenience, it is allowed to use a binary vector operation where 45469one operand is a scalar. In that case the compiler transforms the 45470scalar operand into a vector where each element is the scalar from the 45471operation. The transformation happens only if the scalar could be 45472safely converted to the vector-element type. Consider the following 45473code. 45474 45475 typedef int v4si __attribute__ ((vector_size (16))); 45476 45477 v4si a, b, c; 45478 long l; 45479 45480 a = b + 1; /* a = b + {1,1,1,1}; */ 45481 a = 2 * b; /* a = {2,2,2,2} * b; */ 45482 45483 a = l + a; /* Error, cannot convert long to int. */ 45484 45485 Vectors can be subscripted as if the vector were an array with the same 45486number of elements and base type. Out of bound accesses invoke 45487undefined behavior at run time. Warnings for out of bound accesses for 45488vector subscription can be enabled with '-Warray-bounds'. 45489 45490 Vector comparison is supported with standard comparison operators: '==, 45491!=, <, <=, >, >='. Comparison operands can be vector expressions of 45492integer-type or real-type. Comparison between integer-type vectors and 45493real-type vectors are not supported. The result of the comparison is a 45494vector of the same width and number of elements as the comparison 45495operands with a signed integral element type. 45496 45497 Vectors are compared element-wise producing 0 when comparison is false 45498and -1 (constant of the appropriate type where all bits are set) 45499otherwise. Consider the following example. 45500 45501 typedef int v4si __attribute__ ((vector_size (16))); 45502 45503 v4si a = {1,2,3,4}; 45504 v4si b = {3,2,1,4}; 45505 v4si c; 45506 45507 c = a > b; /* The result would be {0, 0,-1, 0} */ 45508 c = a == b; /* The result would be {0,-1, 0,-1} */ 45509 45510 In C++, the ternary operator '?:' is available. 'a?b:c', where 'b' and 45511'c' are vectors of the same type and 'a' is an integer vector with the 45512same number of elements of the same size as 'b' and 'c', computes all 45513three arguments and creates a vector '{a[0]?b[0]:c[0], a[1]?b[1]:c[1], 45514...}'. Note that unlike in OpenCL, 'a' is thus interpreted as 'a != 0' 45515and not 'a < 0'. As in the case of binary operations, this syntax is 45516also accepted when one of 'b' or 'c' is a scalar that is then 45517transformed into a vector. If both 'b' and 'c' are scalars and the type 45518of 'true?b:c' has the same size as the element type of 'a', then 'b' and 45519'c' are converted to a vector type whose elements have this type and 45520with the same number of elements as 'a'. 45521 45522 In C++, the logic operators '!, &&, ||' are available for vectors. 45523'!v' is equivalent to 'v == 0', 'a && b' is equivalent to 'a!=0 & b!=0' 45524and 'a || b' is equivalent to 'a!=0 | b!=0'. For mixed operations 45525between a scalar 's' and a vector 'v', 's && v' is equivalent to 45526's?v!=0:0' (the evaluation is short-circuit) and 'v && s' is equivalent 45527to 'v!=0 & (s?-1:0)'. 45528 45529 Vector shuffling is available using functions '__builtin_shuffle (vec, 45530mask)' and '__builtin_shuffle (vec0, vec1, mask)'. Both functions 45531construct a permutation of elements from one or two vectors and return a 45532vector of the same type as the input vector(s). The MASK is an integral 45533vector with the same width (W) and element count (N) as the output 45534vector. 45535 45536 The elements of the input vectors are numbered in memory ordering of 45537VEC0 beginning at 0 and VEC1 beginning at N. The elements of MASK are 45538considered modulo N in the single-operand case and modulo 2*N in the 45539two-operand case. 45540 45541 Consider the following example, 45542 45543 typedef int v4si __attribute__ ((vector_size (16))); 45544 45545 v4si a = {1,2,3,4}; 45546 v4si b = {5,6,7,8}; 45547 v4si mask1 = {0,1,1,3}; 45548 v4si mask2 = {0,4,2,5}; 45549 v4si res; 45550 45551 res = __builtin_shuffle (a, mask1); /* res is {1,2,2,4} */ 45552 res = __builtin_shuffle (a, b, mask2); /* res is {1,5,3,6} */ 45553 45554 Note that '__builtin_shuffle' is intentionally semantically compatible 45555with the OpenCL 'shuffle' and 'shuffle2' functions. 45556 45557 You can declare variables and use them in function calls and returns, 45558as well as in assignments and some casts. You can specify a vector type 45559as a return type for a function. Vector types can also be used as 45560function arguments. It is possible to cast from one vector type to 45561another, provided they are of the same size (in fact, you can also cast 45562vectors to and from other datatypes of the same size). 45563 45564 You cannot operate between vectors of different lengths or different 45565signedness without a cast. 45566 45567 Vector shuffling is available using the '__builtin_shufflevector (vec1, 45568vec2, index...)' function. VEC1 and VEC2 must be expressions with 45569vector type with a compatible element type. The result of 45570'__builtin_shufflevector' is a vector with the same element type as VEC1 45571and VEC2 but that has an element count equal to the number of indices 45572specified. 45573 45574 The INDEX arguments are a list of integers that specify the elements 45575indices of the first two vectors that should be extracted and returned 45576in a new vector. These element indices are numbered sequentially 45577starting with the first vector, continuing into the second vector. An 45578index of -1 can be used to indicate that the corresponding element in 45579the returned vector is a don't care and can be freely chosen to 45580optimized the generated code sequence performing the shuffle operation. 45581 45582 Consider the following example, 45583 typedef int v4si __attribute__ ((vector_size (16))); 45584 typedef int v8si __attribute__ ((vector_size (32))); 45585 45586 v8si a = {1,-2,3,-4,5,-6,7,-8}; 45587 v4si b = __builtin_shufflevector (a, a, 0, 2, 4, 6); /* b is {1,3,5,7} */ 45588 v4si c = {-2,-4,-6,-8}; 45589 v8si d = __builtin_shufflevector (c, b, 4, 0, 5, 1, 6, 2, 7, 3); /* d is a */ 45590 45591 Vector conversion is available using the '__builtin_convertvector (vec, 45592vectype)' function. VEC must be an expression with integral or floating 45593vector type and VECTYPE an integral or floating vector type with the 45594same number of elements. The result has VECTYPE type and value of a C 45595cast of every element of VEC to the element type of VECTYPE. 45596 45597 Consider the following example, 45598 typedef int v4si __attribute__ ((vector_size (16))); 45599 typedef float v4sf __attribute__ ((vector_size (16))); 45600 typedef double v4df __attribute__ ((vector_size (32))); 45601 typedef unsigned long long v4di __attribute__ ((vector_size (32))); 45602 45603 v4si a = {1,-2,3,-4}; 45604 v4sf b = {1.5f,-2.5f,3.f,7.f}; 45605 v4di c = {1ULL,5ULL,0ULL,10ULL}; 45606 v4sf d = __builtin_convertvector (a, v4sf); /* d is {1.f,-2.f,3.f,-4.f} */ 45607 /* Equivalent of: 45608 v4sf d = { (float)a[0], (float)a[1], (float)a[2], (float)a[3] }; */ 45609 v4df e = __builtin_convertvector (a, v4df); /* e is {1.,-2.,3.,-4.} */ 45610 v4df f = __builtin_convertvector (b, v4df); /* f is {1.5,-2.5,3.,7.} */ 45611 v4si g = __builtin_convertvector (f, v4si); /* g is {1,-2,3,7} */ 45612 v4si h = __builtin_convertvector (c, v4si); /* h is {1,5,0,10} */ 45613 45614 Sometimes it is desirable to write code using a mix of generic vector 45615operations (for clarity) and machine-specific vector intrinsics (to 45616access vector instructions that are not exposed via generic built-ins). 45617On x86, intrinsic functions for integer vectors typically use the same 45618vector type '__m128i' irrespective of how they interpret the vector, 45619making it necessary to cast their arguments and return values from/to 45620other vector types. In C, you can make use of a 'union' type: 45621 #include <immintrin.h> 45622 45623 typedef unsigned char u8x16 __attribute__ ((vector_size (16))); 45624 typedef unsigned int u32x4 __attribute__ ((vector_size (16))); 45625 45626 typedef union { 45627 __m128i mm; 45628 u8x16 u8; 45629 u32x4 u32; 45630 } v128; 45631 45632for variables that can be used with both built-in operators and x86 45633intrinsics: 45634 45635 v128 x, y = { 0 }; 45636 memcpy (&x, ptr, sizeof x); 45637 y.u8 += 0x80; 45638 x.mm = _mm_adds_epu8 (x.mm, y.mm); 45639 x.u32 &= 0xffffff; 45640 45641 /* Instead of a variable, a compound literal may be used to pass the 45642 return value of an intrinsic call to a function expecting the union: */ 45643 v128 foo (v128); 45644 x = foo ((v128) {_mm_adds_epu8 (x.mm, y.mm)}); 45645 45646 45647File: gcc.info, Node: Offsetof, Next: __sync Builtins, Prev: Vector Extensions, Up: C Extensions 45648 456496.53 Support for 'offsetof' 45650=========================== 45651 45652GCC implements for both C and C++ a syntactic extension to implement the 45653'offsetof' macro. 45654 45655 primary: 45656 "__builtin_offsetof" "(" typename "," offsetof_member_designator ")" 45657 45658 offsetof_member_designator: 45659 identifier 45660 | offsetof_member_designator "." identifier 45661 | offsetof_member_designator "[" expr "]" 45662 45663 This extension is sufficient such that 45664 45665 #define offsetof(TYPE, MEMBER) __builtin_offsetof (TYPE, MEMBER) 45666 45667is a suitable definition of the 'offsetof' macro. In C++, TYPE may be 45668dependent. In either case, MEMBER may consist of a single identifier, 45669or a sequence of member accesses and array references. 45670 45671 45672File: gcc.info, Node: __sync Builtins, Next: __atomic Builtins, Prev: Offsetof, Up: C Extensions 45673 456746.54 Legacy '__sync' Built-in Functions for Atomic Memory Access 45675================================================================ 45676 45677The following built-in functions are intended to be compatible with 45678those described in the 'Intel Itanium Processor-specific Application 45679Binary Interface', section 7.4. As such, they depart from normal GCC 45680practice by not using the '__builtin_' prefix and also by being 45681overloaded so that they work on multiple types. 45682 45683 The definition given in the Intel documentation allows only for the use 45684of the types 'int', 'long', 'long long' or their unsigned counterparts. 45685GCC allows any scalar type that is 1, 2, 4 or 8 bytes in size other than 45686the C type '_Bool' or the C++ type 'bool'. Operations on pointer 45687arguments are performed as if the operands were of the 'uintptr_t' type. 45688That is, they are not scaled by the size of the type to which the 45689pointer points. 45690 45691 These functions are implemented in terms of the '__atomic' builtins 45692(*note __atomic Builtins::). They should not be used for new code which 45693should use the '__atomic' builtins instead. 45694 45695 Not all operations are supported by all target processors. If a 45696particular operation cannot be implemented on the target processor, a 45697warning is generated and a call to an external function is generated. 45698The external function carries the same name as the built-in version, 45699with an additional suffix '_N' where N is the size of the data type. 45700 45701 In most cases, these built-in functions are considered a "full 45702barrier". That is, no memory operand is moved across the operation, 45703either forward or backward. Further, instructions are issued as 45704necessary to prevent the processor from speculating loads across the 45705operation and from queuing stores after the operation. 45706 45707 All of the routines are described in the Intel documentation to take 45708"an optional list of variables protected by the memory barrier". It's 45709not clear what is meant by that; it could mean that _only_ the listed 45710variables are protected, or it could mean a list of additional variables 45711to be protected. The list is ignored by GCC which treats it as empty. 45712GCC interprets an empty list as meaning that all globally accessible 45713variables should be protected. 45714 45715'TYPE __sync_fetch_and_add (TYPE *ptr, TYPE value, ...)' 45716'TYPE __sync_fetch_and_sub (TYPE *ptr, TYPE value, ...)' 45717'TYPE __sync_fetch_and_or (TYPE *ptr, TYPE value, ...)' 45718'TYPE __sync_fetch_and_and (TYPE *ptr, TYPE value, ...)' 45719'TYPE __sync_fetch_and_xor (TYPE *ptr, TYPE value, ...)' 45720'TYPE __sync_fetch_and_nand (TYPE *ptr, TYPE value, ...)' 45721 These built-in functions perform the operation suggested by the 45722 name, and returns the value that had previously been in memory. 45723 That is, operations on integer operands have the following 45724 semantics. Operations on pointer arguments are performed as if the 45725 operands were of the 'uintptr_t' type. That is, they are not 45726 scaled by the size of the type to which the pointer points. 45727 45728 { tmp = *ptr; *ptr OP= value; return tmp; } 45729 { tmp = *ptr; *ptr = ~(tmp & value); return tmp; } // nand 45730 45731 The object pointed to by the first argument must be of integer or 45732 pointer type. It must not be a boolean type. 45733 45734 _Note:_ GCC 4.4 and later implement '__sync_fetch_and_nand' as 45735 '*ptr = ~(tmp & value)' instead of '*ptr = ~tmp & value'. 45736 45737'TYPE __sync_add_and_fetch (TYPE *ptr, TYPE value, ...)' 45738'TYPE __sync_sub_and_fetch (TYPE *ptr, TYPE value, ...)' 45739'TYPE __sync_or_and_fetch (TYPE *ptr, TYPE value, ...)' 45740'TYPE __sync_and_and_fetch (TYPE *ptr, TYPE value, ...)' 45741'TYPE __sync_xor_and_fetch (TYPE *ptr, TYPE value, ...)' 45742'TYPE __sync_nand_and_fetch (TYPE *ptr, TYPE value, ...)' 45743 These built-in functions perform the operation suggested by the 45744 name, and return the new value. That is, operations on integer 45745 operands have the following semantics. Operations on pointer 45746 operands are performed as if the operand's type were 'uintptr_t'. 45747 45748 { *ptr OP= value; return *ptr; } 45749 { *ptr = ~(*ptr & value); return *ptr; } // nand 45750 45751 The same constraints on arguments apply as for the corresponding 45752 '__sync_op_and_fetch' built-in functions. 45753 45754 _Note:_ GCC 4.4 and later implement '__sync_nand_and_fetch' as 45755 '*ptr = ~(*ptr & value)' instead of '*ptr = ~*ptr & value'. 45756 45757'bool __sync_bool_compare_and_swap (TYPE *ptr, TYPE oldval, TYPE newval, ...)' 45758'TYPE __sync_val_compare_and_swap (TYPE *ptr, TYPE oldval, TYPE newval, ...)' 45759 These built-in functions perform an atomic compare and swap. That 45760 is, if the current value of '*PTR' is OLDVAL, then write NEWVAL 45761 into '*PTR'. 45762 45763 The "bool" version returns 'true' if the comparison is successful 45764 and NEWVAL is written. The "val" version returns the contents of 45765 '*PTR' before the operation. 45766 45767'__sync_synchronize (...)' 45768 This built-in function issues a full memory barrier. 45769 45770'TYPE __sync_lock_test_and_set (TYPE *ptr, TYPE value, ...)' 45771 This built-in function, as described by Intel, is not a traditional 45772 test-and-set operation, but rather an atomic exchange operation. 45773 It writes VALUE into '*PTR', and returns the previous contents of 45774 '*PTR'. 45775 45776 Many targets have only minimal support for such locks, and do not 45777 support a full exchange operation. In this case, a target may 45778 support reduced functionality here by which the _only_ valid value 45779 to store is the immediate constant 1. The exact value actually 45780 stored in '*PTR' is implementation defined. 45781 45782 This built-in function is not a full barrier, but rather an 45783 "acquire barrier". This means that references after the operation 45784 cannot move to (or be speculated to) before the operation, but 45785 previous memory stores may not be globally visible yet, and 45786 previous memory loads may not yet be satisfied. 45787 45788'void __sync_lock_release (TYPE *ptr, ...)' 45789 This built-in function releases the lock acquired by 45790 '__sync_lock_test_and_set'. Normally this means writing the 45791 constant 0 to '*PTR'. 45792 45793 This built-in function is not a full barrier, but rather a "release 45794 barrier". This means that all previous memory stores are globally 45795 visible, and all previous memory loads have been satisfied, but 45796 following memory reads are not prevented from being speculated to 45797 before the barrier. 45798 45799 45800File: gcc.info, Node: __atomic Builtins, Next: Integer Overflow Builtins, Prev: __sync Builtins, Up: C Extensions 45801 458026.55 Built-in Functions for Memory Model Aware Atomic Operations 45803================================================================ 45804 45805The following built-in functions approximately match the requirements 45806for the C++11 memory model. They are all identified by being prefixed 45807with '__atomic' and most are overloaded so that they work with multiple 45808types. 45809 45810 These functions are intended to replace the legacy '__sync' builtins. 45811The main difference is that the memory order that is requested is a 45812parameter to the functions. New code should always use the '__atomic' 45813builtins rather than the '__sync' builtins. 45814 45815 Note that the '__atomic' builtins assume that programs will conform to 45816the C++11 memory model. In particular, they assume that programs are 45817free of data races. See the C++11 standard for detailed requirements. 45818 45819 The '__atomic' builtins can be used with any integral scalar or pointer 45820type that is 1, 2, 4, or 8 bytes in length. 16-byte integral types are 45821also allowed if '__int128' (*note __int128::) is supported by the 45822architecture. 45823 45824 The four non-arithmetic functions (load, store, exchange, and 45825compare_exchange) all have a generic version as well. This generic 45826version works on any data type. It uses the lock-free built-in function 45827if the specific data type size makes that possible; otherwise, an 45828external call is left to be resolved at run time. This external call is 45829the same format with the addition of a 'size_t' parameter inserted as 45830the first parameter indicating the size of the object being pointed to. 45831All objects must be the same size. 45832 45833 There are 6 different memory orders that can be specified. These map 45834to the C++11 memory orders with the same names, see the C++11 standard 45835or the GCC wiki on atomic synchronization 45836(https://gcc.gnu.org/wiki/Atomic/GCCMM/AtomicSync) for detailed 45837definitions. Individual targets may also support additional memory 45838orders for use on specific architectures. Refer to the target 45839documentation for details of these. 45840 45841 An atomic operation can both constrain code motion and be mapped to 45842hardware instructions for synchronization between threads (e.g., a 45843fence). To which extent this happens is controlled by the memory 45844orders, which are listed here in approximately ascending order of 45845strength. The description of each memory order is only meant to roughly 45846illustrate the effects and is not a specification; see the C++11 memory 45847model for precise semantics. 45848 45849'__ATOMIC_RELAXED' 45850 Implies no inter-thread ordering constraints. 45851'__ATOMIC_CONSUME' 45852 This is currently implemented using the stronger '__ATOMIC_ACQUIRE' 45853 memory order because of a deficiency in C++11's semantics for 45854 'memory_order_consume'. 45855'__ATOMIC_ACQUIRE' 45856 Creates an inter-thread happens-before constraint from the release 45857 (or stronger) semantic store to this acquire load. Can prevent 45858 hoisting of code to before the operation. 45859'__ATOMIC_RELEASE' 45860 Creates an inter-thread happens-before constraint to acquire (or 45861 stronger) semantic loads that read from this release store. Can 45862 prevent sinking of code to after the operation. 45863'__ATOMIC_ACQ_REL' 45864 Combines the effects of both '__ATOMIC_ACQUIRE' and 45865 '__ATOMIC_RELEASE'. 45866'__ATOMIC_SEQ_CST' 45867 Enforces total ordering with all other '__ATOMIC_SEQ_CST' 45868 operations. 45869 45870 Note that in the C++11 memory model, _fences_ (e.g., 45871'__atomic_thread_fence') take effect in combination with other atomic 45872operations on specific memory locations (e.g., atomic loads); operations 45873on specific memory locations do not necessarily affect other operations 45874in the same way. 45875 45876 Target architectures are encouraged to provide their own patterns for 45877each of the atomic built-in functions. If no target is provided, the 45878original non-memory model set of '__sync' atomic built-in functions are 45879used, along with any required synchronization fences surrounding it in 45880order to achieve the proper behavior. Execution in this case is subject 45881to the same restrictions as those built-in functions. 45882 45883 If there is no pattern or mechanism to provide a lock-free instruction 45884sequence, a call is made to an external routine with the same parameters 45885to be resolved at run time. 45886 45887 When implementing patterns for these built-in functions, the memory 45888order parameter can be ignored as long as the pattern implements the 45889most restrictive '__ATOMIC_SEQ_CST' memory order. Any of the other 45890memory orders execute correctly with this memory order but they may not 45891execute as efficiently as they could with a more appropriate 45892implementation of the relaxed requirements. 45893 45894 Note that the C++11 standard allows for the memory order parameter to 45895be determined at run time rather than at compile time. These built-in 45896functions map any run-time value to '__ATOMIC_SEQ_CST' rather than 45897invoke a runtime library call or inline a switch statement. This is 45898standard compliant, safe, and the simplest approach for now. 45899 45900 The memory order parameter is a signed int, but only the lower 16 bits 45901are reserved for the memory order. The remainder of the signed int is 45902reserved for target use and should be 0. Use of the predefined atomic 45903values ensures proper usage. 45904 45905 -- Built-in Function: TYPE __atomic_load_n (TYPE *ptr, int memorder) 45906 This built-in function implements an atomic load operation. It 45907 returns the contents of '*PTR'. 45908 45909 The valid memory order variants are '__ATOMIC_RELAXED', 45910 '__ATOMIC_SEQ_CST', '__ATOMIC_ACQUIRE', and '__ATOMIC_CONSUME'. 45911 45912 -- Built-in Function: void __atomic_load (TYPE *ptr, TYPE *ret, int 45913 memorder) 45914 This is the generic version of an atomic load. It returns the 45915 contents of '*PTR' in '*RET'. 45916 45917 -- Built-in Function: void __atomic_store_n (TYPE *ptr, TYPE val, int 45918 memorder) 45919 This built-in function implements an atomic store operation. It 45920 writes 'VAL' into '*PTR'. 45921 45922 The valid memory order variants are '__ATOMIC_RELAXED', 45923 '__ATOMIC_SEQ_CST', and '__ATOMIC_RELEASE'. 45924 45925 -- Built-in Function: void __atomic_store (TYPE *ptr, TYPE *val, int 45926 memorder) 45927 This is the generic version of an atomic store. It stores the 45928 value of '*VAL' into '*PTR'. 45929 45930 -- Built-in Function: TYPE __atomic_exchange_n (TYPE *ptr, TYPE val, 45931 int memorder) 45932 This built-in function implements an atomic exchange operation. It 45933 writes VAL into '*PTR', and returns the previous contents of 45934 '*PTR'. 45935 45936 All memory order variants are valid. 45937 45938 -- Built-in Function: void __atomic_exchange (TYPE *ptr, TYPE *val, 45939 TYPE *ret, int memorder) 45940 This is the generic version of an atomic exchange. It stores the 45941 contents of '*VAL' into '*PTR'. The original value of '*PTR' is 45942 copied into '*RET'. 45943 45944 -- Built-in Function: bool __atomic_compare_exchange_n (TYPE *ptr, TYPE 45945 *expected, TYPE desired, bool weak, int success_memorder, int 45946 failure_memorder) 45947 This built-in function implements an atomic compare and exchange 45948 operation. This compares the contents of '*PTR' with the contents 45949 of '*EXPECTED'. If equal, the operation is a _read-modify-write_ 45950 operation that writes DESIRED into '*PTR'. If they are not equal, 45951 the operation is a _read_ and the current contents of '*PTR' are 45952 written into '*EXPECTED'. WEAK is 'true' for weak 45953 compare_exchange, which may fail spuriously, and 'false' for the 45954 strong variation, which never fails spuriously. Many targets only 45955 offer the strong variation and ignore the parameter. When in 45956 doubt, use the strong variation. 45957 45958 If DESIRED is written into '*PTR' then 'true' is returned and 45959 memory is affected according to the memory order specified by 45960 SUCCESS_MEMORDER. There are no restrictions on what memory order 45961 can be used here. 45962 45963 Otherwise, 'false' is returned and memory is affected according to 45964 FAILURE_MEMORDER. This memory order cannot be '__ATOMIC_RELEASE' 45965 nor '__ATOMIC_ACQ_REL'. It also cannot be a stronger order than 45966 that specified by SUCCESS_MEMORDER. 45967 45968 -- Built-in Function: bool __atomic_compare_exchange (TYPE *ptr, TYPE 45969 *expected, TYPE *desired, bool weak, int success_memorder, int 45970 failure_memorder) 45971 This built-in function implements the generic version of 45972 '__atomic_compare_exchange'. The function is virtually identical 45973 to '__atomic_compare_exchange_n', except the desired value is also 45974 a pointer. 45975 45976 -- Built-in Function: TYPE __atomic_add_fetch (TYPE *ptr, TYPE val, int 45977 memorder) 45978 -- Built-in Function: TYPE __atomic_sub_fetch (TYPE *ptr, TYPE val, int 45979 memorder) 45980 -- Built-in Function: TYPE __atomic_and_fetch (TYPE *ptr, TYPE val, int 45981 memorder) 45982 -- Built-in Function: TYPE __atomic_xor_fetch (TYPE *ptr, TYPE val, int 45983 memorder) 45984 -- Built-in Function: TYPE __atomic_or_fetch (TYPE *ptr, TYPE val, int 45985 memorder) 45986 -- Built-in Function: TYPE __atomic_nand_fetch (TYPE *ptr, TYPE val, 45987 int memorder) 45988 These built-in functions perform the operation suggested by the 45989 name, and return the result of the operation. Operations on 45990 pointer arguments are performed as if the operands were of the 45991 'uintptr_t' type. That is, they are not scaled by the size of the 45992 type to which the pointer points. 45993 45994 { *ptr OP= val; return *ptr; } 45995 { *ptr = ~(*ptr & val); return *ptr; } // nand 45996 45997 The object pointed to by the first argument must be of integer or 45998 pointer type. It must not be a boolean type. All memory orders 45999 are valid. 46000 46001 -- Built-in Function: TYPE __atomic_fetch_add (TYPE *ptr, TYPE val, int 46002 memorder) 46003 -- Built-in Function: TYPE __atomic_fetch_sub (TYPE *ptr, TYPE val, int 46004 memorder) 46005 -- Built-in Function: TYPE __atomic_fetch_and (TYPE *ptr, TYPE val, int 46006 memorder) 46007 -- Built-in Function: TYPE __atomic_fetch_xor (TYPE *ptr, TYPE val, int 46008 memorder) 46009 -- Built-in Function: TYPE __atomic_fetch_or (TYPE *ptr, TYPE val, int 46010 memorder) 46011 -- Built-in Function: TYPE __atomic_fetch_nand (TYPE *ptr, TYPE val, 46012 int memorder) 46013 These built-in functions perform the operation suggested by the 46014 name, and return the value that had previously been in '*PTR'. 46015 Operations on pointer arguments are performed as if the operands 46016 were of the 'uintptr_t' type. That is, they are not scaled by the 46017 size of the type to which the pointer points. 46018 46019 { tmp = *ptr; *ptr OP= val; return tmp; } 46020 { tmp = *ptr; *ptr = ~(*ptr & val); return tmp; } // nand 46021 46022 The same constraints on arguments apply as for the corresponding 46023 '__atomic_op_fetch' built-in functions. All memory orders are 46024 valid. 46025 46026 -- Built-in Function: bool __atomic_test_and_set (void *ptr, int 46027 memorder) 46028 46029 This built-in function performs an atomic test-and-set operation on 46030 the byte at '*PTR'. The byte is set to some implementation defined 46031 nonzero "set" value and the return value is 'true' if and only if 46032 the previous contents were "set". It should be only used for 46033 operands of type 'bool' or 'char'. For other types only part of 46034 the value may be set. 46035 46036 All memory orders are valid. 46037 46038 -- Built-in Function: void __atomic_clear (bool *ptr, int memorder) 46039 46040 This built-in function performs an atomic clear operation on 46041 '*PTR'. After the operation, '*PTR' contains 0. It should be only 46042 used for operands of type 'bool' or 'char' and in conjunction with 46043 '__atomic_test_and_set'. For other types it may only clear 46044 partially. If the type is not 'bool' prefer using 46045 '__atomic_store'. 46046 46047 The valid memory order variants are '__ATOMIC_RELAXED', 46048 '__ATOMIC_SEQ_CST', and '__ATOMIC_RELEASE'. 46049 46050 -- Built-in Function: void __atomic_thread_fence (int memorder) 46051 46052 This built-in function acts as a synchronization fence between 46053 threads based on the specified memory order. 46054 46055 All memory orders are valid. 46056 46057 -- Built-in Function: void __atomic_signal_fence (int memorder) 46058 46059 This built-in function acts as a synchronization fence between a 46060 thread and signal handlers based in the same thread. 46061 46062 All memory orders are valid. 46063 46064 -- Built-in Function: bool __atomic_always_lock_free (size_t size, void 46065 *ptr) 46066 46067 This built-in function returns 'true' if objects of SIZE bytes 46068 always generate lock-free atomic instructions for the target 46069 architecture. SIZE must resolve to a compile-time constant and the 46070 result also resolves to a compile-time constant. 46071 46072 PTR is an optional pointer to the object that may be used to 46073 determine alignment. A value of 0 indicates typical alignment 46074 should be used. The compiler may also ignore this parameter. 46075 46076 if (__atomic_always_lock_free (sizeof (long long), 0)) 46077 46078 -- Built-in Function: bool __atomic_is_lock_free (size_t size, void 46079 *ptr) 46080 46081 This built-in function returns 'true' if objects of SIZE bytes 46082 always generate lock-free atomic instructions for the target 46083 architecture. If the built-in function is not known to be 46084 lock-free, a call is made to a runtime routine named 46085 '__atomic_is_lock_free'. 46086 46087 PTR is an optional pointer to the object that may be used to 46088 determine alignment. A value of 0 indicates typical alignment 46089 should be used. The compiler may also ignore this parameter. 46090 46091 46092File: gcc.info, Node: Integer Overflow Builtins, Next: x86 specific memory model extensions for transactional memory, Prev: __atomic Builtins, Up: C Extensions 46093 460946.56 Built-in Functions to Perform Arithmetic with Overflow Checking 46095==================================================================== 46096 46097The following built-in functions allow performing simple arithmetic 46098operations together with checking whether the operations overflowed. 46099 46100 -- Built-in Function: bool __builtin_add_overflow (TYPE1 a, TYPE2 b, 46101 TYPE3 *res) 46102 -- Built-in Function: bool __builtin_sadd_overflow (int a, int b, int 46103 *res) 46104 -- Built-in Function: bool __builtin_saddl_overflow (long int a, long 46105 int b, long int *res) 46106 -- Built-in Function: bool __builtin_saddll_overflow (long long int a, 46107 long long int b, long long int *res) 46108 -- Built-in Function: bool __builtin_uadd_overflow (unsigned int a, 46109 unsigned int b, unsigned int *res) 46110 -- Built-in Function: bool __builtin_uaddl_overflow (unsigned long int 46111 a, unsigned long int b, unsigned long int *res) 46112 -- Built-in Function: bool __builtin_uaddll_overflow (unsigned long 46113 long int a, unsigned long long int b, unsigned long long int 46114 *res) 46115 46116 These built-in functions promote the first two operands into 46117 infinite precision signed type and perform addition on those 46118 promoted operands. The result is then cast to the type the third 46119 pointer argument points to and stored there. If the stored result 46120 is equal to the infinite precision result, the built-in functions 46121 return 'false', otherwise they return 'true'. As the addition is 46122 performed in infinite signed precision, these built-in functions 46123 have fully defined behavior for all argument values. 46124 46125 The first built-in function allows arbitrary integral types for 46126 operands and the result type must be pointer to some integral type 46127 other than enumerated or boolean type, the rest of the built-in 46128 functions have explicit integer types. 46129 46130 The compiler will attempt to use hardware instructions to implement 46131 these built-in functions where possible, like conditional jump on 46132 overflow after addition, conditional jump on carry etc. 46133 46134 -- Built-in Function: bool __builtin_sub_overflow (TYPE1 a, TYPE2 b, 46135 TYPE3 *res) 46136 -- Built-in Function: bool __builtin_ssub_overflow (int a, int b, int 46137 *res) 46138 -- Built-in Function: bool __builtin_ssubl_overflow (long int a, long 46139 int b, long int *res) 46140 -- Built-in Function: bool __builtin_ssubll_overflow (long long int a, 46141 long long int b, long long int *res) 46142 -- Built-in Function: bool __builtin_usub_overflow (unsigned int a, 46143 unsigned int b, unsigned int *res) 46144 -- Built-in Function: bool __builtin_usubl_overflow (unsigned long int 46145 a, unsigned long int b, unsigned long int *res) 46146 -- Built-in Function: bool __builtin_usubll_overflow (unsigned long 46147 long int a, unsigned long long int b, unsigned long long int 46148 *res) 46149 46150 These built-in functions are similar to the add overflow checking 46151 built-in functions above, except they perform subtraction, subtract 46152 the second argument from the first one, instead of addition. 46153 46154 -- Built-in Function: bool __builtin_mul_overflow (TYPE1 a, TYPE2 b, 46155 TYPE3 *res) 46156 -- Built-in Function: bool __builtin_smul_overflow (int a, int b, int 46157 *res) 46158 -- Built-in Function: bool __builtin_smull_overflow (long int a, long 46159 int b, long int *res) 46160 -- Built-in Function: bool __builtin_smulll_overflow (long long int a, 46161 long long int b, long long int *res) 46162 -- Built-in Function: bool __builtin_umul_overflow (unsigned int a, 46163 unsigned int b, unsigned int *res) 46164 -- Built-in Function: bool __builtin_umull_overflow (unsigned long int 46165 a, unsigned long int b, unsigned long int *res) 46166 -- Built-in Function: bool __builtin_umulll_overflow (unsigned long 46167 long int a, unsigned long long int b, unsigned long long int 46168 *res) 46169 46170 These built-in functions are similar to the add overflow checking 46171 built-in functions above, except they perform multiplication, 46172 instead of addition. 46173 46174 The following built-in functions allow checking if simple arithmetic 46175operation would overflow. 46176 46177 -- Built-in Function: bool __builtin_add_overflow_p (TYPE1 a, TYPE2 b, 46178 TYPE3 c) 46179 -- Built-in Function: bool __builtin_sub_overflow_p (TYPE1 a, TYPE2 b, 46180 TYPE3 c) 46181 -- Built-in Function: bool __builtin_mul_overflow_p (TYPE1 a, TYPE2 b, 46182 TYPE3 c) 46183 46184 These built-in functions are similar to '__builtin_add_overflow', 46185 '__builtin_sub_overflow', or '__builtin_mul_overflow', except that 46186 they don't store the result of the arithmetic operation anywhere 46187 and the last argument is not a pointer, but some expression with 46188 integral type other than enumerated or boolean type. 46189 46190 The built-in functions promote the first two operands into infinite 46191 precision signed type and perform addition on those promoted 46192 operands. The result is then cast to the type of the third 46193 argument. If the cast result is equal to the infinite precision 46194 result, the built-in functions return 'false', otherwise they 46195 return 'true'. The value of the third argument is ignored, just 46196 the side effects in the third argument are evaluated, and no 46197 integral argument promotions are performed on the last argument. 46198 If the third argument is a bit-field, the type used for the result 46199 cast has the precision and signedness of the given bit-field, 46200 rather than precision and signedness of the underlying type. 46201 46202 For example, the following macro can be used to portably check, at 46203 compile-time, whether or not adding two constant integers will 46204 overflow, and perform the addition only when it is known to be safe 46205 and not to trigger a '-Woverflow' warning. 46206 46207 #define INT_ADD_OVERFLOW_P(a, b) \ 46208 __builtin_add_overflow_p (a, b, (__typeof__ ((a) + (b))) 0) 46209 46210 enum { 46211 A = INT_MAX, B = 3, 46212 C = INT_ADD_OVERFLOW_P (A, B) ? 0 : A + B, 46213 D = __builtin_add_overflow_p (1, SCHAR_MAX, (signed char) 0) 46214 }; 46215 46216 The compiler will attempt to use hardware instructions to implement 46217 these built-in functions where possible, like conditional jump on 46218 overflow after addition, conditional jump on carry etc. 46219 46220 46221File: gcc.info, Node: x86 specific memory model extensions for transactional memory, Next: Object Size Checking, Prev: Integer Overflow Builtins, Up: C Extensions 46222 462236.57 x86-Specific Memory Model Extensions for Transactional Memory 46224================================================================== 46225 46226The x86 architecture supports additional memory ordering flags to mark 46227critical sections for hardware lock elision. These must be specified in 46228addition to an existing memory order to atomic intrinsics. 46229 46230'__ATOMIC_HLE_ACQUIRE' 46231 Start lock elision on a lock variable. Memory order must be 46232 '__ATOMIC_ACQUIRE' or stronger. 46233'__ATOMIC_HLE_RELEASE' 46234 End lock elision on a lock variable. Memory order must be 46235 '__ATOMIC_RELEASE' or stronger. 46236 46237 When a lock acquire fails, it is required for good performance to abort 46238the transaction quickly. This can be done with a '_mm_pause'. 46239 46240 #include <immintrin.h> // For _mm_pause 46241 46242 int lockvar; 46243 46244 /* Acquire lock with lock elision */ 46245 while (__atomic_exchange_n(&lockvar, 1, __ATOMIC_ACQUIRE|__ATOMIC_HLE_ACQUIRE)) 46246 _mm_pause(); /* Abort failed transaction */ 46247 ... 46248 /* Free lock with lock elision */ 46249 __atomic_store_n(&lockvar, 0, __ATOMIC_RELEASE|__ATOMIC_HLE_RELEASE); 46250 46251 46252File: gcc.info, Node: Object Size Checking, Next: Other Builtins, Prev: x86 specific memory model extensions for transactional memory, Up: C Extensions 46253 462546.58 Object Size Checking Built-in Functions 46255============================================ 46256 46257GCC implements a limited buffer overflow protection mechanism that can 46258prevent some buffer overflow attacks by determining the sizes of objects 46259into which data is about to be written and preventing the writes when 46260the size isn't sufficient. The built-in functions described below yield 46261the best results when used together and when optimization is enabled. 46262For example, to detect object sizes across function boundaries or to 46263follow pointer assignments through non-trivial control flow they rely on 46264various optimization passes enabled with '-O2'. However, to a limited 46265extent, they can be used without optimization as well. 46266 46267 -- Built-in Function: size_t __builtin_object_size (const void * PTR, 46268 int TYPE) 46269 is a built-in construct that returns a constant number of bytes 46270 from PTR to the end of the object PTR pointer points to (if known 46271 at compile time). To determine the sizes of dynamically allocated 46272 objects the function relies on the allocation functions called to 46273 obtain the storage to be declared with the 'alloc_size' attribute 46274 (*note Common Function Attributes::). '__builtin_object_size' 46275 never evaluates its arguments for side effects. If there are any 46276 side effects in them, it returns '(size_t) -1' for TYPE 0 or 1 and 46277 '(size_t) 0' for TYPE 2 or 3. If there are multiple objects PTR 46278 can point to and all of them are known at compile time, the 46279 returned number is the maximum of remaining byte counts in those 46280 objects if TYPE & 2 is 0 and minimum if nonzero. If it is not 46281 possible to determine which objects PTR points to at compile time, 46282 '__builtin_object_size' should return '(size_t) -1' for TYPE 0 or 1 46283 and '(size_t) 0' for TYPE 2 or 3. 46284 46285 TYPE is an integer constant from 0 to 3. If the least significant 46286 bit is clear, objects are whole variables, if it is set, a closest 46287 surrounding subobject is considered the object a pointer points to. 46288 The second bit determines if maximum or minimum of remaining bytes 46289 is computed. 46290 46291 struct V { char buf1[10]; int b; char buf2[10]; } var; 46292 char *p = &var.buf1[1], *q = &var.b; 46293 46294 /* Here the object p points to is var. */ 46295 assert (__builtin_object_size (p, 0) == sizeof (var) - 1); 46296 /* The subobject p points to is var.buf1. */ 46297 assert (__builtin_object_size (p, 1) == sizeof (var.buf1) - 1); 46298 /* The object q points to is var. */ 46299 assert (__builtin_object_size (q, 0) 46300 == (char *) (&var + 1) - (char *) &var.b); 46301 /* The subobject q points to is var.b. */ 46302 assert (__builtin_object_size (q, 1) == sizeof (var.b)); 46303 46304 -- Built-in Function: size_t __builtin_dynamic_object_size (const void 46305 * PTR, int TYPE) 46306 is similar to '__builtin_object_size' in that it returns a number 46307 of bytes from PTR to the end of the object PTR pointer points to, 46308 except that the size returned may not be a constant. This results 46309 in successful evaluation of object size estimates in a wider range 46310 of use cases and can be more precise than '__builtin_object_size', 46311 but it incurs a performance penalty since it may add a runtime 46312 overhead on size computation. Semantics of TYPE as well as return 46313 values in case it is not possible to determine which objects PTR 46314 points to at compile time are the same as in the case of 46315 '__builtin_object_size'. 46316 46317 There are built-in functions added for many common string operation 46318functions, e.g., for 'memcpy' '__builtin___memcpy_chk' built-in is 46319provided. This built-in has an additional last argument, which is the 46320number of bytes remaining in the object the DEST argument points to or 46321'(size_t) -1' if the size is not known. 46322 46323 The built-in functions are optimized into the normal string functions 46324like 'memcpy' if the last argument is '(size_t) -1' or if it is known at 46325compile time that the destination object will not be overflowed. If the 46326compiler can determine at compile time that the object will always be 46327overflowed, it issues a warning. 46328 46329 The intended use can be e.g. 46330 46331 #undef memcpy 46332 #define bos0(dest) __builtin_object_size (dest, 0) 46333 #define memcpy(dest, src, n) \ 46334 __builtin___memcpy_chk (dest, src, n, bos0 (dest)) 46335 46336 char *volatile p; 46337 char buf[10]; 46338 /* It is unknown what object p points to, so this is optimized 46339 into plain memcpy - no checking is possible. */ 46340 memcpy (p, "abcde", n); 46341 /* Destination is known and length too. It is known at compile 46342 time there will be no overflow. */ 46343 memcpy (&buf[5], "abcde", 5); 46344 /* Destination is known, but the length is not known at compile time. 46345 This will result in __memcpy_chk call that can check for overflow 46346 at run time. */ 46347 memcpy (&buf[5], "abcde", n); 46348 /* Destination is known and it is known at compile time there will 46349 be overflow. There will be a warning and __memcpy_chk call that 46350 will abort the program at run time. */ 46351 memcpy (&buf[6], "abcde", 5); 46352 46353 Such built-in functions are provided for 'memcpy', 'mempcpy', 46354'memmove', 'memset', 'strcpy', 'stpcpy', 'strncpy', 'strcat' and 46355'strncat'. 46356 46357 There are also checking built-in functions for formatted output 46358functions. 46359 int __builtin___sprintf_chk (char *s, int flag, size_t os, const char *fmt, ...); 46360 int __builtin___snprintf_chk (char *s, size_t maxlen, int flag, size_t os, 46361 const char *fmt, ...); 46362 int __builtin___vsprintf_chk (char *s, int flag, size_t os, const char *fmt, 46363 va_list ap); 46364 int __builtin___vsnprintf_chk (char *s, size_t maxlen, int flag, size_t os, 46365 const char *fmt, va_list ap); 46366 46367 The added FLAG argument is passed unchanged to '__sprintf_chk' etc. 46368functions and can contain implementation specific flags on what 46369additional security measures the checking function might take, such as 46370handling '%n' differently. 46371 46372 The OS argument is the object size S points to, like in the other 46373built-in functions. There is a small difference in the behavior though, 46374if OS is '(size_t) -1', the built-in functions are optimized into the 46375non-checking functions only if FLAG is 0, otherwise the checking 46376function is called with OS argument set to '(size_t) -1'. 46377 46378 In addition to this, there are checking built-in functions 46379'__builtin___printf_chk', '__builtin___vprintf_chk', 46380'__builtin___fprintf_chk' and '__builtin___vfprintf_chk'. These have 46381just one additional argument, FLAG, right before format string FMT. If 46382the compiler is able to optimize them to 'fputc' etc. functions, it 46383does, otherwise the checking function is called and the FLAG argument 46384passed to it. 46385 46386 46387File: gcc.info, Node: Other Builtins, Next: Target Builtins, Prev: Object Size Checking, Up: C Extensions 46388 463896.59 Other Built-in Functions Provided by GCC 46390============================================= 46391 46392GCC provides a large number of built-in functions other than the ones 46393mentioned above. Some of these are for internal use in the processing 46394of exceptions or variable-length argument lists and are not documented 46395here because they may change from time to time; we do not recommend 46396general use of these functions. 46397 46398 The remaining functions are provided for optimization purposes. 46399 46400 With the exception of built-ins that have library equivalents such as 46401the standard C library functions discussed below, or that expand to 46402library calls, GCC built-in functions are always expanded inline and 46403thus do not have corresponding entry points and their address cannot be 46404obtained. Attempting to use them in an expression other than a function 46405call results in a compile-time error. 46406 46407 GCC includes built-in versions of many of the functions in the standard 46408C library. These functions come in two forms: one whose names start 46409with the '__builtin_' prefix, and the other without. Both forms have 46410the same type (including prototype), the same address (when their 46411address is taken), and the same meaning as the C library functions even 46412if you specify the '-fno-builtin' option *note C Dialect Options::). 46413Many of these functions are only optimized in certain cases; if they are 46414not optimized in a particular case, a call to the library function is 46415emitted. 46416 46417 Outside strict ISO C mode ('-ansi', '-std=c90', '-std=c99' or 46418'-std=c11'), the functions '_exit', 'alloca', 'bcmp', 'bzero', 46419'dcgettext', 'dgettext', 'dremf', 'dreml', 'drem', 'exp10f', 'exp10l', 46420'exp10', 'ffsll', 'ffsl', 'ffs', 'fprintf_unlocked', 'fputs_unlocked', 46421'gammaf', 'gammal', 'gamma', 'gammaf_r', 'gammal_r', 'gamma_r', 46422'gettext', 'index', 'isascii', 'j0f', 'j0l', 'j0', 'j1f', 'j1l', 'j1', 46423'jnf', 'jnl', 'jn', 'lgammaf_r', 'lgammal_r', 'lgamma_r', 'mempcpy', 46424'pow10f', 'pow10l', 'pow10', 'printf_unlocked', 'rindex', 'roundeven', 46425'roundevenf', 'roundevenl', 'scalbf', 'scalbl', 'scalb', 'signbit', 46426'signbitf', 'signbitl', 'signbitd32', 'signbitd64', 'signbitd128', 46427'significandf', 'significandl', 'significand', 'sincosf', 'sincosl', 46428'sincos', 'stpcpy', 'stpncpy', 'strcasecmp', 'strdup', 'strfmon', 46429'strncasecmp', 'strndup', 'strnlen', 'toascii', 'y0f', 'y0l', 'y0', 46430'y1f', 'y1l', 'y1', 'ynf', 'ynl' and 'yn' may be handled as built-in 46431functions. All these functions have corresponding versions prefixed 46432with '__builtin_', which may be used even in strict C90 mode. 46433 46434 The ISO C99 functions '_Exit', 'acoshf', 'acoshl', 'acosh', 'asinhf', 46435'asinhl', 'asinh', 'atanhf', 'atanhl', 'atanh', 'cabsf', 'cabsl', 46436'cabs', 'cacosf', 'cacoshf', 'cacoshl', 'cacosh', 'cacosl', 'cacos', 46437'cargf', 'cargl', 'carg', 'casinf', 'casinhf', 'casinhl', 'casinh', 46438'casinl', 'casin', 'catanf', 'catanhf', 'catanhl', 'catanh', 'catanl', 46439'catan', 'cbrtf', 'cbrtl', 'cbrt', 'ccosf', 'ccoshf', 'ccoshl', 'ccosh', 46440'ccosl', 'ccos', 'cexpf', 'cexpl', 'cexp', 'cimagf', 'cimagl', 'cimag', 46441'clogf', 'clogl', 'clog', 'conjf', 'conjl', 'conj', 'copysignf', 46442'copysignl', 'copysign', 'cpowf', 'cpowl', 'cpow', 'cprojf', 'cprojl', 46443'cproj', 'crealf', 'creall', 'creal', 'csinf', 'csinhf', 'csinhl', 46444'csinh', 'csinl', 'csin', 'csqrtf', 'csqrtl', 'csqrt', 'ctanf', 46445'ctanhf', 'ctanhl', 'ctanh', 'ctanl', 'ctan', 'erfcf', 'erfcl', 'erfc', 46446'erff', 'erfl', 'erf', 'exp2f', 'exp2l', 'exp2', 'expm1f', 'expm1l', 46447'expm1', 'fdimf', 'fdiml', 'fdim', 'fmaf', 'fmal', 'fmaxf', 'fmaxl', 46448'fmax', 'fma', 'fminf', 'fminl', 'fmin', 'hypotf', 'hypotl', 'hypot', 46449'ilogbf', 'ilogbl', 'ilogb', 'imaxabs', 'isblank', 'iswblank', 46450'lgammaf', 'lgammal', 'lgamma', 'llabs', 'llrintf', 'llrintl', 'llrint', 46451'llroundf', 'llroundl', 'llround', 'log1pf', 'log1pl', 'log1p', 'log2f', 46452'log2l', 'log2', 'logbf', 'logbl', 'logb', 'lrintf', 'lrintl', 'lrint', 46453'lroundf', 'lroundl', 'lround', 'nearbyintf', 'nearbyintl', 'nearbyint', 46454'nextafterf', 'nextafterl', 'nextafter', 'nexttowardf', 'nexttowardl', 46455'nexttoward', 'remainderf', 'remainderl', 'remainder', 'remquof', 46456'remquol', 'remquo', 'rintf', 'rintl', 'rint', 'roundf', 'roundl', 46457'round', 'scalblnf', 'scalblnl', 'scalbln', 'scalbnf', 'scalbnl', 46458'scalbn', 'snprintf', 'tgammaf', 'tgammal', 'tgamma', 'truncf', 46459'truncl', 'trunc', 'vfscanf', 'vscanf', 'vsnprintf' and 'vsscanf' are 46460handled as built-in functions except in strict ISO C90 mode ('-ansi' or 46461'-std=c90'). 46462 46463 There are also built-in versions of the ISO C99 functions 'acosf', 46464'acosl', 'asinf', 'asinl', 'atan2f', 'atan2l', 'atanf', 'atanl', 46465'ceilf', 'ceill', 'cosf', 'coshf', 'coshl', 'cosl', 'expf', 'expl', 46466'fabsf', 'fabsl', 'floorf', 'floorl', 'fmodf', 'fmodl', 'frexpf', 46467'frexpl', 'ldexpf', 'ldexpl', 'log10f', 'log10l', 'logf', 'logl', 46468'modfl', 'modff', 'powf', 'powl', 'sinf', 'sinhf', 'sinhl', 'sinl', 46469'sqrtf', 'sqrtl', 'tanf', 'tanhf', 'tanhl' and 'tanl' that are 46470recognized in any mode since ISO C90 reserves these names for the 46471purpose to which ISO C99 puts them. All these functions have 46472corresponding versions prefixed with '__builtin_'. 46473 46474 There are also built-in functions '__builtin_fabsfN', 46475'__builtin_fabsfNx', '__builtin_copysignfN' and '__builtin_copysignfNx', 46476corresponding to the TS 18661-3 functions 'fabsfN', 'fabsfNx', 46477'copysignfN' and 'copysignfNx', for supported types '_FloatN' and 46478'_FloatNx'. 46479 46480 There are also GNU extension functions 'clog10', 'clog10f' and 46481'clog10l' which names are reserved by ISO C99 for future use. All these 46482functions have versions prefixed with '__builtin_'. 46483 46484 The ISO C94 functions 'iswalnum', 'iswalpha', 'iswcntrl', 'iswdigit', 46485'iswgraph', 'iswlower', 'iswprint', 'iswpunct', 'iswspace', 'iswupper', 46486'iswxdigit', 'towlower' and 'towupper' are handled as built-in functions 46487except in strict ISO C90 mode ('-ansi' or '-std=c90'). 46488 46489 The ISO C90 functions 'abort', 'abs', 'acos', 'asin', 'atan2', 'atan', 46490'calloc', 'ceil', 'cosh', 'cos', 'exit', 'exp', 'fabs', 'floor', 'fmod', 46491'fprintf', 'fputs', 'free', 'frexp', 'fscanf', 'isalnum', 'isalpha', 46492'iscntrl', 'isdigit', 'isgraph', 'islower', 'isprint', 'ispunct', 46493'isspace', 'isupper', 'isxdigit', 'tolower', 'toupper', 'labs', 'ldexp', 46494'log10', 'log', 'malloc', 'memchr', 'memcmp', 'memcpy', 'memset', 46495'modf', 'pow', 'printf', 'putchar', 'puts', 'realloc', 'scanf', 'sinh', 46496'sin', 'snprintf', 'sprintf', 'sqrt', 'sscanf', 'strcat', 'strchr', 46497'strcmp', 'strcpy', 'strcspn', 'strlen', 'strncat', 'strncmp', 46498'strncpy', 'strpbrk', 'strrchr', 'strspn', 'strstr', 'tanh', 'tan', 46499'vfprintf', 'vprintf' and 'vsprintf' are all recognized as built-in 46500functions unless '-fno-builtin' is specified (or '-fno-builtin-FUNCTION' 46501is specified for an individual function). All of these functions have 46502corresponding versions prefixed with '__builtin_'. 46503 46504 GCC provides built-in versions of the ISO C99 floating-point comparison 46505macros that avoid raising exceptions for unordered operands. They have 46506the same names as the standard macros ( 'isgreater', 'isgreaterequal', 46507'isless', 'islessequal', 'islessgreater', and 'isunordered') , with 46508'__builtin_' prefixed. We intend for a library implementor to be able 46509to simply '#define' each standard macro to its built-in equivalent. In 46510the same fashion, GCC provides 'fpclassify', 'isfinite', 'isinf_sign', 46511'isnormal' and 'signbit' built-ins used with '__builtin_' prefixed. The 46512'isinf' and 'isnan' built-in functions appear both with and without the 46513'__builtin_' prefix. 46514 46515 GCC provides built-in versions of the ISO C99 floating-point rounding 46516and exceptions handling functions 'fegetround', 'feclearexcept' and 46517'feraiseexcept'. They may not be available for all targets, and because 46518they need close interaction with libc internal values, they may not be 46519available for all target libcs, but in all cases they will gracefully 46520fallback to libc calls. These built-in functions appear both with and 46521without the '__builtin_' prefix. 46522 46523 -- Built-in Function: void *__builtin_alloca (size_t size) 46524 The '__builtin_alloca' function must be called at block scope. The 46525 function allocates an object SIZE bytes large on the stack of the 46526 calling function. The object is aligned on the default stack 46527 alignment boundary for the target determined by the 46528 '__BIGGEST_ALIGNMENT__' macro. The '__builtin_alloca' function 46529 returns a pointer to the first byte of the allocated object. The 46530 lifetime of the allocated object ends just before the calling 46531 function returns to its caller. This is so even when 46532 '__builtin_alloca' is called within a nested block. 46533 46534 For example, the following function allocates eight objects of 'n' 46535 bytes each on the stack, storing a pointer to each in consecutive 46536 elements of the array 'a'. It then passes the array to function 46537 'g' which can safely use the storage pointed to by each of the 46538 array elements. 46539 46540 void f (unsigned n) 46541 { 46542 void *a [8]; 46543 for (int i = 0; i != 8; ++i) 46544 a [i] = __builtin_alloca (n); 46545 46546 g (a, n); // safe 46547 } 46548 46549 Since the '__builtin_alloca' function doesn't validate its argument 46550 it is the responsibility of its caller to make sure the argument 46551 doesn't cause it to exceed the stack size limit. The 46552 '__builtin_alloca' function is provided to make it possible to 46553 allocate on the stack arrays of bytes with an upper bound that may 46554 be computed at run time. Since C99 Variable Length Arrays offer 46555 similar functionality under a portable, more convenient, and safer 46556 interface they are recommended instead, in both C99 and C++ 46557 programs where GCC provides them as an extension. *Note Variable 46558 Length::, for details. 46559 46560 -- Built-in Function: void *__builtin_alloca_with_align (size_t size, 46561 size_t alignment) 46562 The '__builtin_alloca_with_align' function must be called at block 46563 scope. The function allocates an object SIZE bytes large on the 46564 stack of the calling function. The allocated object is aligned on 46565 the boundary specified by the argument ALIGNMENT whose unit is 46566 given in bits (not bytes). The SIZE argument must be positive and 46567 not exceed the stack size limit. The ALIGNMENT argument must be a 46568 constant integer expression that evaluates to a power of 2 greater 46569 than or equal to 'CHAR_BIT' and less than some unspecified maximum. 46570 Invocations with other values are rejected with an error indicating 46571 the valid bounds. The function returns a pointer to the first byte 46572 of the allocated object. The lifetime of the allocated object ends 46573 at the end of the block in which the function was called. The 46574 allocated storage is released no later than just before the calling 46575 function returns to its caller, but may be released at the end of 46576 the block in which the function was called. 46577 46578 For example, in the following function the call to 'g' is unsafe 46579 because when 'overalign' is non-zero, the space allocated by 46580 '__builtin_alloca_with_align' may have been released at the end of 46581 the 'if' statement in which it was called. 46582 46583 void f (unsigned n, bool overalign) 46584 { 46585 void *p; 46586 if (overalign) 46587 p = __builtin_alloca_with_align (n, 64 /* bits */); 46588 else 46589 p = __builtin_alloc (n); 46590 46591 g (p, n); // unsafe 46592 } 46593 46594 Since the '__builtin_alloca_with_align' function doesn't validate 46595 its SIZE argument it is the responsibility of its caller to make 46596 sure the argument doesn't cause it to exceed the stack size limit. 46597 The '__builtin_alloca_with_align' function is provided to make it 46598 possible to allocate on the stack overaligned arrays of bytes with 46599 an upper bound that may be computed at run time. Since C99 46600 Variable Length Arrays offer the same functionality under a 46601 portable, more convenient, and safer interface they are recommended 46602 instead, in both C99 and C++ programs where GCC provides them as an 46603 extension. *Note Variable Length::, for details. 46604 46605 -- Built-in Function: void *__builtin_alloca_with_align_and_max (size_t 46606 size, size_t alignment, size_t max_size) 46607 Similar to '__builtin_alloca_with_align' but takes an extra 46608 argument specifying an upper bound for SIZE in case its value 46609 cannot be computed at compile time, for use by '-fstack-usage', 46610 '-Wstack-usage' and '-Walloca-larger-than'. MAX_SIZE must be a 46611 constant integer expression, it has no effect on code generation 46612 and no attempt is made to check its compatibility with SIZE. 46613 46614 -- Built-in Function: bool __builtin_has_attribute (TYPE-OR-EXPRESSION, 46615 ATTRIBUTE) 46616 The '__builtin_has_attribute' function evaluates to an integer 46617 constant expression equal to 'true' if the symbol or type 46618 referenced by the TYPE-OR-EXPRESSION argument has been declared 46619 with the ATTRIBUTE referenced by the second argument. For an 46620 TYPE-OR-EXPRESSION argument that does not reference a symbol, since 46621 attributes do not apply to expressions the built-in consider the 46622 type of the argument. Neither argument is evaluated. The 46623 TYPE-OR-EXPRESSION argument is subject to the same restrictions as 46624 the argument to 'typeof' (*note Typeof::). The ATTRIBUTE argument 46625 is an attribute name optionally followed by a comma-separated list 46626 of arguments enclosed in parentheses. Both forms of attribute 46627 names--with and without double leading and trailing 46628 underscores--are recognized. *Note Attribute Syntax::, for 46629 details. When no attribute arguments are specified for an 46630 attribute that expects one or more arguments the function returns 46631 'true' if TYPE-OR-EXPRESSION has been declared with the attribute 46632 regardless of the attribute argument values. Arguments provided 46633 for an attribute that expects some are validated and matched up to 46634 the provided number. The function returns 'true' if all provided 46635 arguments match. For example, the first call to the function below 46636 evaluates to 'true' because 'x' is declared with the 'aligned' 46637 attribute but the second call evaluates to 'false' because 'x' is 46638 declared 'aligned (8)' and not 'aligned (4)'. 46639 46640 __attribute__ ((aligned (8))) int x; 46641 _Static_assert (__builtin_has_attribute (x, aligned), "aligned"); 46642 _Static_assert (!__builtin_has_attribute (x, aligned (4)), "aligned (4)"); 46643 46644 Due to a limitation the '__builtin_has_attribute' function returns 46645 'false' for the 'mode' attribute even if the type or variable 46646 referenced by the TYPE-OR-EXPRESSION argument was declared with 46647 one. The function is also not supported with labels, and in C with 46648 enumerators. 46649 46650 Note that unlike the '__has_attribute' preprocessor operator which 46651 is suitable for use in '#if' preprocessing directives 46652 '__builtin_has_attribute' is an intrinsic function that is not 46653 recognized in such contexts. 46654 46655 -- Built-in Function: TYPE __builtin_speculation_safe_value (TYPE val, 46656 TYPE failval) 46657 46658 This built-in function can be used to help mitigate against unsafe 46659 speculative execution. TYPE may be any integral type or any 46660 pointer type. 46661 46662 1. If the CPU is not speculatively executing the code, then VAL 46663 is returned. 46664 2. If the CPU is executing speculatively then either: 46665 * The function may cause execution to pause until it is 46666 known that the code is no-longer being executed 46667 speculatively (in which case VAL can be returned, as 46668 above); or 46669 * The function may use target-dependent speculation 46670 tracking state to cause FAILVAL to be returned when it is 46671 known that speculative execution has incorrectly 46672 predicted a conditional branch operation. 46673 46674 The second argument, FAILVAL, is optional and defaults to zero if 46675 omitted. 46676 46677 GCC defines the preprocessor macro 46678 '__HAVE_BUILTIN_SPECULATION_SAFE_VALUE' for targets that have been 46679 updated to support this builtin. 46680 46681 The built-in function can be used where a variable appears to be 46682 used in a safe way, but the CPU, due to speculative execution may 46683 temporarily ignore the bounds checks. Consider, for example, the 46684 following function: 46685 46686 int array[500]; 46687 int f (unsigned untrusted_index) 46688 { 46689 if (untrusted_index < 500) 46690 return array[untrusted_index]; 46691 return 0; 46692 } 46693 46694 If the function is called repeatedly with 'untrusted_index' less 46695 than the limit of 500, then a branch predictor will learn that the 46696 block of code that returns a value stored in 'array' will be 46697 executed. If the function is subsequently called with an 46698 out-of-range value it will still try to execute that block of code 46699 first until the CPU determines that the prediction was incorrect 46700 (the CPU will unwind any incorrect operations at that point). 46701 However, depending on how the result of the function is used, it 46702 might be possible to leave traces in the cache that can reveal what 46703 was stored at the out-of-bounds location. The built-in function 46704 can be used to provide some protection against leaking data in this 46705 way by changing the code to: 46706 46707 int array[500]; 46708 int f (unsigned untrusted_index) 46709 { 46710 if (untrusted_index < 500) 46711 return array[__builtin_speculation_safe_value (untrusted_index)]; 46712 return 0; 46713 } 46714 46715 The built-in function will either cause execution to stall until 46716 the conditional branch has been fully resolved, or it may permit 46717 speculative execution to continue, but using 0 instead of 46718 'untrusted_value' if that exceeds the limit. 46719 46720 If accessing any memory location is potentially unsafe when 46721 speculative execution is incorrect, then the code can be rewritten 46722 as 46723 46724 int array[500]; 46725 int f (unsigned untrusted_index) 46726 { 46727 if (untrusted_index < 500) 46728 return *__builtin_speculation_safe_value (&array[untrusted_index], NULL); 46729 return 0; 46730 } 46731 46732 which will cause a 'NULL' pointer to be used for the unsafe case. 46733 46734 -- Built-in Function: int __builtin_types_compatible_p (TYPE1, TYPE2) 46735 46736 You can use the built-in function '__builtin_types_compatible_p' to 46737 determine whether two types are the same. 46738 46739 This built-in function returns 1 if the unqualified versions of the 46740 types TYPE1 and TYPE2 (which are types, not expressions) are 46741 compatible, 0 otherwise. The result of this built-in function can 46742 be used in integer constant expressions. 46743 46744 This built-in function ignores top level qualifiers (e.g., 'const', 46745 'volatile'). For example, 'int' is equivalent to 'const int'. 46746 46747 The type 'int[]' and 'int[5]' are compatible. On the other hand, 46748 'int' and 'char *' are not compatible, even if the size of their 46749 types, on the particular architecture are the same. Also, the 46750 amount of pointer indirection is taken into account when 46751 determining similarity. Consequently, 'short *' is not similar to 46752 'short **'. Furthermore, two types that are typedefed are 46753 considered compatible if their underlying types are compatible. 46754 46755 An 'enum' type is not considered to be compatible with another 46756 'enum' type even if both are compatible with the same integer type; 46757 this is what the C standard specifies. For example, 'enum {foo, 46758 bar}' is not similar to 'enum {hot, dog}'. 46759 46760 You typically use this function in code whose execution varies 46761 depending on the arguments' types. For example: 46762 46763 #define foo(x) \ 46764 ({ \ 46765 typeof (x) tmp = (x); \ 46766 if (__builtin_types_compatible_p (typeof (x), long double)) \ 46767 tmp = foo_long_double (tmp); \ 46768 else if (__builtin_types_compatible_p (typeof (x), double)) \ 46769 tmp = foo_double (tmp); \ 46770 else if (__builtin_types_compatible_p (typeof (x), float)) \ 46771 tmp = foo_float (tmp); \ 46772 else \ 46773 abort (); \ 46774 tmp; \ 46775 }) 46776 46777 _Note:_ This construct is only available for C. 46778 46779 -- Built-in Function: TYPE __builtin_call_with_static_chain (CALL_EXP, 46780 POINTER_EXP) 46781 46782 The CALL_EXP expression must be a function call, and the 46783 POINTER_EXP expression must be a pointer. The POINTER_EXP is 46784 passed to the function call in the target's static chain location. 46785 The result of builtin is the result of the function call. 46786 46787 _Note:_ This builtin is only available for C. This builtin can be 46788 used to call Go closures from C. 46789 46790 -- Built-in Function: TYPE __builtin_choose_expr (CONST_EXP, EXP1, 46791 EXP2) 46792 46793 You can use the built-in function '__builtin_choose_expr' to 46794 evaluate code depending on the value of a constant expression. 46795 This built-in function returns EXP1 if CONST_EXP, which is an 46796 integer constant expression, is nonzero. Otherwise it returns 46797 EXP2. 46798 46799 This built-in function is analogous to the '? :' operator in C, 46800 except that the expression returned has its type unaltered by 46801 promotion rules. Also, the built-in function does not evaluate the 46802 expression that is not chosen. For example, if CONST_EXP evaluates 46803 to 'true', EXP2 is not evaluated even if it has side effects. 46804 46805 This built-in function can return an lvalue if the chosen argument 46806 is an lvalue. 46807 46808 If EXP1 is returned, the return type is the same as EXP1's type. 46809 Similarly, if EXP2 is returned, its return type is the same as 46810 EXP2. 46811 46812 Example: 46813 46814 #define foo(x) \ 46815 __builtin_choose_expr ( \ 46816 __builtin_types_compatible_p (typeof (x), double), \ 46817 foo_double (x), \ 46818 __builtin_choose_expr ( \ 46819 __builtin_types_compatible_p (typeof (x), float), \ 46820 foo_float (x), \ 46821 /* The void expression results in a compile-time error \ 46822 when assigning the result to something. */ \ 46823 (void)0)) 46824 46825 _Note:_ This construct is only available for C. Furthermore, the 46826 unused expression (EXP1 or EXP2 depending on the value of 46827 CONST_EXP) may still generate syntax errors. This may change in 46828 future revisions. 46829 46830 -- Built-in Function: TYPE __builtin_tgmath (FUNCTIONS, ARGUMENTS) 46831 46832 The built-in function '__builtin_tgmath', available only for C and 46833 Objective-C, calls a function determined according to the rules of 46834 '<tgmath.h>' macros. It is intended to be used in implementations 46835 of that header, so that expansions of macros from that header only 46836 expand each of their arguments once, to avoid problems when calls 46837 to such macros are nested inside the arguments of other calls to 46838 such macros; in addition, it results in better diagnostics for 46839 invalid calls to '<tgmath.h>' macros than implementations using 46840 other GNU C language features. For example, the 'pow' type-generic 46841 macro might be defined as: 46842 46843 #define pow(a, b) __builtin_tgmath (powf, pow, powl, \ 46844 cpowf, cpow, cpowl, a, b) 46845 46846 The arguments to '__builtin_tgmath' are at least two pointers to 46847 functions, followed by the arguments to the type-generic macro 46848 (which will be passed as arguments to the selected function). All 46849 the pointers to functions must be pointers to prototyped functions, 46850 none of which may have variable arguments, and all of which must 46851 have the same number of parameters; the number of parameters of the 46852 first function determines how many arguments to '__builtin_tgmath' 46853 are interpreted as function pointers, and how many as the arguments 46854 to the called function. 46855 46856 The types of the specified functions must all be different, but 46857 related to each other in the same way as a set of functions that 46858 may be selected between by a macro in '<tgmath.h>'. This means 46859 that the functions are parameterized by a floating-point type T, 46860 different for each such function. The function return types may 46861 all be the same type, or they may be T for each function, or they 46862 may be the real type corresponding to T for each function (if some 46863 of the types T are complex). Likewise, for each parameter 46864 position, the type of the parameter in that position may always be 46865 the same type, or may be T for each function (this case must apply 46866 for at least one parameter position), or may be the real type 46867 corresponding to T for each function. 46868 46869 The standard rules for '<tgmath.h>' macros are used to find a 46870 common type U from the types of the arguments for parameters whose 46871 types vary between the functions; complex integer types (a GNU 46872 extension) are treated like '_Complex double' for this purpose (or 46873 '_Complex _Float64' if all the function return types are the same 46874 '_FloatN' or '_FloatNx' type). If the function return types vary, 46875 or are all the same integer type, the function called is the one 46876 for which T is U, and it is an error if there is no such function. 46877 If the function return types are all the same floating-point type, 46878 the type-generic macro is taken to be one of those from TS 18661 46879 that rounds the result to a narrower type; if there is a function 46880 for which T is U, it is called, and otherwise the first function, 46881 if any, for which T has at least the range and precision of U is 46882 called, and it is an error if there is no such function. 46883 46884 -- Built-in Function: int __builtin_constant_p (EXP) 46885 You can use the built-in function '__builtin_constant_p' to 46886 determine if a value is known to be constant at compile time and 46887 hence that GCC can perform constant-folding on expressions 46888 involving that value. The argument of the function is the value to 46889 test. The function returns the integer 1 if the argument is known 46890 to be a compile-time constant and 0 if it is not known to be a 46891 compile-time constant. A return of 0 does not indicate that the 46892 value is _not_ a constant, but merely that GCC cannot prove it is a 46893 constant with the specified value of the '-O' option. 46894 46895 You typically use this function in an embedded application where 46896 memory is a critical resource. If you have some complex 46897 calculation, you may want it to be folded if it involves constants, 46898 but need to call a function if it does not. For example: 46899 46900 #define Scale_Value(X) \ 46901 (__builtin_constant_p (X) \ 46902 ? ((X) * SCALE + OFFSET) : Scale (X)) 46903 46904 You may use this built-in function in either a macro or an inline 46905 function. However, if you use it in an inlined function and pass 46906 an argument of the function as the argument to the built-in, GCC 46907 never returns 1 when you call the inline function with a string 46908 constant or compound literal (*note Compound Literals::) and does 46909 not return 1 when you pass a constant numeric value to the inline 46910 function unless you specify the '-O' option. 46911 46912 You may also use '__builtin_constant_p' in initializers for static 46913 data. For instance, you can write 46914 46915 static const int table[] = { 46916 __builtin_constant_p (EXPRESSION) ? (EXPRESSION) : -1, 46917 /* ... */ 46918 }; 46919 46920 This is an acceptable initializer even if EXPRESSION is not a 46921 constant expression, including the case where 46922 '__builtin_constant_p' returns 1 because EXPRESSION can be folded 46923 to a constant but EXPRESSION contains operands that are not 46924 otherwise permitted in a static initializer (for example, '0 && foo 46925 ()'). GCC must be more conservative about evaluating the built-in 46926 in this case, because it has no opportunity to perform 46927 optimization. 46928 46929 -- Built-in Function: bool __builtin_is_constant_evaluated (void) 46930 The '__builtin_is_constant_evaluated' function is available only in 46931 C++. The built-in is intended to be used by implementations of the 46932 'std::is_constant_evaluated' C++ function. Programs should make 46933 use of the latter function rather than invoking the built-in 46934 directly. 46935 46936 The main use case of the built-in is to determine whether a 46937 'constexpr' function is being called in a 'constexpr' context. A 46938 call to the function evaluates to a core constant expression with 46939 the value 'true' if and only if it occurs within the evaluation of 46940 an expression or conversion that is manifestly constant-evaluated 46941 as defined in the C++ standard. Manifestly constant-evaluated 46942 contexts include constant-expressions, the conditions of 'constexpr 46943 if' statements, constraint-expressions, and initializers of 46944 variables usable in constant expressions. For more details refer 46945 to the latest revision of the C++ standard. 46946 46947 -- Built-in Function: void __builtin_clear_padding (PTR) 46948 The built-in function '__builtin_clear_padding' function clears 46949 padding bits inside of the object representation of object pointed 46950 by PTR, which has to be a pointer. The value representation of the 46951 object is not affected. The type of the object is assumed to be 46952 the type the pointer points to. Inside of a union, the only 46953 cleared bits are bits that are padding bits for all the union 46954 members. 46955 46956 This built-in-function is useful if the padding bits of an object 46957 might have intederminate values and the object representation needs 46958 to be bitwise compared to some other object, for example for atomic 46959 operations. 46960 46961 For C++, PTR argument type should be pointer to trivially-copyable 46962 type, unless the argument is address of a variable or parameter, 46963 because otherwise it isn't known if the type isn't just a base 46964 class whose padding bits are reused or laid out differently in a 46965 derived class. 46966 46967 -- Built-in Function: TYPE __builtin_bit_cast (TYPE, ARG) 46968 The '__builtin_bit_cast' function is available only in C++. The 46969 built-in is intended to be used by implementations of the 46970 'std::bit_cast' C++ template function. Programs should make use of 46971 the latter function rather than invoking the built-in directly. 46972 46973 This built-in function allows reinterpreting the bits of the ARG 46974 argument as if it had type TYPE. TYPE and the type of the ARG 46975 argument need to be trivially copyable types with the same size. 46976 When manifestly constant-evaluated, it performs extra diagnostics 46977 required for 'std::bit_cast' and returns a constant expression if 46978 ARG is a constant expression. For more details refer to the latest 46979 revision of the C++ standard. 46980 46981 -- Built-in Function: long __builtin_expect (long EXP, long C) 46982 You may use '__builtin_expect' to provide the compiler with branch 46983 prediction information. In general, you should prefer to use 46984 actual profile feedback for this ('-fprofile-arcs'), as programmers 46985 are notoriously bad at predicting how their programs actually 46986 perform. However, there are applications in which this data is 46987 hard to collect. 46988 46989 The return value is the value of EXP, which should be an integral 46990 expression. The semantics of the built-in are that it is expected 46991 that EXP == C. For example: 46992 46993 if (__builtin_expect (x, 0)) 46994 foo (); 46995 46996 indicates that we do not expect to call 'foo', since we expect 'x' 46997 to be zero. Since you are limited to integral expressions for EXP, 46998 you should use constructions such as 46999 47000 if (__builtin_expect (ptr != NULL, 1)) 47001 foo (*ptr); 47002 47003 when testing pointer or floating-point values. 47004 47005 For the purposes of branch prediction optimizations, the 47006 probability that a '__builtin_expect' expression is 'true' is 47007 controlled by GCC's 'builtin-expect-probability' parameter, which 47008 defaults to 90%. 47009 47010 You can also use '__builtin_expect_with_probability' to explicitly 47011 assign a probability value to individual expressions. If the 47012 built-in is used in a loop construct, the provided probability will 47013 influence the expected number of iterations made by loop 47014 optimizations. 47015 47016 -- Built-in Function: long __builtin_expect_with_probability 47017 (long EXP, long C, double PROBABILITY) 47018 47019 This function has the same semantics as '__builtin_expect', but the 47020 caller provides the expected probability that EXP == C. The last 47021 argument, PROBABILITY, is a floating-point value in the range 0.0 47022 to 1.0, inclusive. The PROBABILITY argument must be constant 47023 floating-point expression. 47024 47025 -- Built-in Function: void __builtin_trap (void) 47026 This function causes the program to exit abnormally. GCC 47027 implements this function by using a target-dependent mechanism 47028 (such as intentionally executing an illegal instruction) or by 47029 calling 'abort'. The mechanism used may vary from release to 47030 release so you should not rely on any particular implementation. 47031 47032 -- Built-in Function: void __builtin_unreachable (void) 47033 If control flow reaches the point of the '__builtin_unreachable', 47034 the program is undefined. It is useful in situations where the 47035 compiler cannot deduce the unreachability of the code. 47036 47037 One such case is immediately following an 'asm' statement that 47038 either never terminates, or one that transfers control elsewhere 47039 and never returns. In this example, without the 47040 '__builtin_unreachable', GCC issues a warning that control reaches 47041 the end of a non-void function. It also generates code to return 47042 after the 'asm'. 47043 47044 int f (int c, int v) 47045 { 47046 if (c) 47047 { 47048 return v; 47049 } 47050 else 47051 { 47052 asm("jmp error_handler"); 47053 __builtin_unreachable (); 47054 } 47055 } 47056 47057 Because the 'asm' statement unconditionally transfers control out 47058 of the function, control never reaches the end of the function 47059 body. The '__builtin_unreachable' is in fact unreachable and 47060 communicates this fact to the compiler. 47061 47062 Another use for '__builtin_unreachable' is following a call a 47063 function that never returns but that is not declared 47064 '__attribute__((noreturn))', as in this example: 47065 47066 void function_that_never_returns (void); 47067 47068 int g (int c) 47069 { 47070 if (c) 47071 { 47072 return 1; 47073 } 47074 else 47075 { 47076 function_that_never_returns (); 47077 __builtin_unreachable (); 47078 } 47079 } 47080 47081 -- Built-in Function: TYPE __builtin_assoc_barrier (TYPE EXPR) 47082 This built-in inhibits re-association of the floating-point 47083 expression EXPR with expressions consuming the return value of the 47084 built-in. The expression EXPR itself can be reordered, and the 47085 whole expression EXPR can be reordered with operands after the 47086 barrier. The barrier is only relevant when '-fassociative-math' is 47087 active, since otherwise floating-point is not treated as 47088 associative. 47089 47090 float x0 = a + b - b; 47091 float x1 = __builtin_assoc_barrier(a + b) - b; 47092 47093 means that, with '-fassociative-math', 'x0' can be optimized to 'x0 47094 = a' but 'x1' cannot. 47095 47096 -- Built-in Function: void * __builtin_assume_aligned (const void *EXP, 47097 size_t ALIGN, ...) 47098 This function returns its first argument, and allows the compiler 47099 to assume that the returned pointer is at least ALIGN bytes 47100 aligned. This built-in can have either two or three arguments, if 47101 it has three, the third argument should have integer type, and if 47102 it is nonzero means misalignment offset. For example: 47103 47104 void *x = __builtin_assume_aligned (arg, 16); 47105 47106 means that the compiler can assume 'x', set to 'arg', is at least 47107 16-byte aligned, while: 47108 47109 void *x = __builtin_assume_aligned (arg, 32, 8); 47110 47111 means that the compiler can assume for 'x', set to 'arg', that 47112 '(char *) x - 8' is 32-byte aligned. 47113 47114 -- Built-in Function: int __builtin_LINE () 47115 This function is the equivalent of the preprocessor '__LINE__' 47116 macro and returns a constant integer expression that evaluates to 47117 the line number of the invocation of the built-in. When used as a 47118 C++ default argument for a function F, it returns the line number 47119 of the call to F. 47120 47121 -- Built-in Function: const char * __builtin_FUNCTION () 47122 This function is the equivalent of the '__FUNCTION__' symbol and 47123 returns an address constant pointing to the name of the function 47124 from which the built-in was invoked, or the empty string if the 47125 invocation is not at function scope. When used as a C++ default 47126 argument for a function F, it returns the name of F's caller or the 47127 empty string if the call was not made at function scope. 47128 47129 -- Built-in Function: const char * __builtin_FILE () 47130 This function is the equivalent of the preprocessor '__FILE__' 47131 macro and returns an address constant pointing to the file name 47132 containing the invocation of the built-in, or the empty string if 47133 the invocation is not at function scope. When used as a C++ 47134 default argument for a function F, it returns the file name of the 47135 call to F or the empty string if the call was not made at function 47136 scope. 47137 47138 For example, in the following, each call to function 'foo' will 47139 print a line similar to '"file.c:123: foo: message"' with the name 47140 of the file and the line number of the 'printf' call, the name of 47141 the function 'foo', followed by the word 'message'. 47142 47143 const char* 47144 function (const char *func = __builtin_FUNCTION ()) 47145 { 47146 return func; 47147 } 47148 47149 void foo (void) 47150 { 47151 printf ("%s:%i: %s: message\n", file (), line (), function ()); 47152 } 47153 47154 -- Built-in Function: void __builtin___clear_cache (void *BEGIN, void 47155 *END) 47156 This function is used to flush the processor's instruction cache 47157 for the region of memory between BEGIN inclusive and END exclusive. 47158 Some targets require that the instruction cache be flushed, after 47159 modifying memory containing code, in order to obtain deterministic 47160 behavior. 47161 47162 If the target does not require instruction cache flushes, 47163 '__builtin___clear_cache' has no effect. Otherwise either 47164 instructions are emitted in-line to clear the instruction cache or 47165 a call to the '__clear_cache' function in libgcc is made. 47166 47167 -- Built-in Function: void __builtin_prefetch (const void *ADDR, ...) 47168 This function is used to minimize cache-miss latency by moving data 47169 into a cache before it is accessed. You can insert calls to 47170 '__builtin_prefetch' into code for which you know addresses of data 47171 in memory that is likely to be accessed soon. If the target 47172 supports them, data prefetch instructions are generated. If the 47173 prefetch is done early enough before the access then the data will 47174 be in the cache by the time it is accessed. 47175 47176 The value of ADDR is the address of the memory to prefetch. There 47177 are two optional arguments, RW and LOCALITY. The value of RW is a 47178 compile-time constant one or zero; one means that the prefetch is 47179 preparing for a write to the memory address and zero, the default, 47180 means that the prefetch is preparing for a read. The value 47181 LOCALITY must be a compile-time constant integer between zero and 47182 three. A value of zero means that the data has no temporal 47183 locality, so it need not be left in the cache after the access. A 47184 value of three means that the data has a high degree of temporal 47185 locality and should be left in all levels of cache possible. 47186 Values of one and two mean, respectively, a low or moderate degree 47187 of temporal locality. The default is three. 47188 47189 for (i = 0; i < n; i++) 47190 { 47191 a[i] = a[i] + b[i]; 47192 __builtin_prefetch (&a[i+j], 1, 1); 47193 __builtin_prefetch (&b[i+j], 0, 1); 47194 /* ... */ 47195 } 47196 47197 Data prefetch does not generate faults if ADDR is invalid, but the 47198 address expression itself must be valid. For example, a prefetch 47199 of 'p->next' does not fault if 'p->next' is not a valid address, 47200 but evaluation faults if 'p' is not a valid address. 47201 47202 If the target does not support data prefetch, the address 47203 expression is evaluated if it includes side effects but no other 47204 code is generated and GCC does not issue a warning. 47205 47206 -- Built-in Function: size_t __builtin_object_size (const void * PTR, 47207 int TYPE) 47208 Returns the size of an object pointed to by PTR. *Note Object Size 47209 Checking::, for a detailed description of the function. 47210 47211 -- Built-in Function: double __builtin_huge_val (void) 47212 Returns a positive infinity, if supported by the floating-point 47213 format, else 'DBL_MAX'. This function is suitable for implementing 47214 the ISO C macro 'HUGE_VAL'. 47215 47216 -- Built-in Function: float __builtin_huge_valf (void) 47217 Similar to '__builtin_huge_val', except the return type is 'float'. 47218 47219 -- Built-in Function: long double __builtin_huge_vall (void) 47220 Similar to '__builtin_huge_val', except the return type is 'long 47221 double'. 47222 47223 -- Built-in Function: _FloatN __builtin_huge_valfN (void) 47224 Similar to '__builtin_huge_val', except the return type is 47225 '_FloatN'. 47226 47227 -- Built-in Function: _FloatNx __builtin_huge_valfNx (void) 47228 Similar to '__builtin_huge_val', except the return type is 47229 '_FloatNx'. 47230 47231 -- Built-in Function: int __builtin_fpclassify (int, int, int, int, 47232 int, ...) 47233 This built-in implements the C99 fpclassify functionality. The 47234 first five int arguments should be the target library's notion of 47235 the possible FP classes and are used for return values. They must 47236 be constant values and they must appear in this order: 'FP_NAN', 47237 'FP_INFINITE', 'FP_NORMAL', 'FP_SUBNORMAL' and 'FP_ZERO'. The 47238 ellipsis is for exactly one floating-point value to classify. GCC 47239 treats the last argument as type-generic, which means it does not 47240 do default promotion from float to double. 47241 47242 -- Built-in Function: double __builtin_inf (void) 47243 Similar to '__builtin_huge_val', except a warning is generated if 47244 the target floating-point format does not support infinities. 47245 47246 -- Built-in Function: _Decimal32 __builtin_infd32 (void) 47247 Similar to '__builtin_inf', except the return type is '_Decimal32'. 47248 47249 -- Built-in Function: _Decimal64 __builtin_infd64 (void) 47250 Similar to '__builtin_inf', except the return type is '_Decimal64'. 47251 47252 -- Built-in Function: _Decimal128 __builtin_infd128 (void) 47253 Similar to '__builtin_inf', except the return type is 47254 '_Decimal128'. 47255 47256 -- Built-in Function: float __builtin_inff (void) 47257 Similar to '__builtin_inf', except the return type is 'float'. 47258 This function is suitable for implementing the ISO C99 macro 47259 'INFINITY'. 47260 47261 -- Built-in Function: long double __builtin_infl (void) 47262 Similar to '__builtin_inf', except the return type is 'long 47263 double'. 47264 47265 -- Built-in Function: _FloatN __builtin_inffN (void) 47266 Similar to '__builtin_inf', except the return type is '_FloatN'. 47267 47268 -- Built-in Function: _FloatN __builtin_inffNx (void) 47269 Similar to '__builtin_inf', except the return type is '_FloatNx'. 47270 47271 -- Built-in Function: int __builtin_isinf_sign (...) 47272 Similar to 'isinf', except the return value is -1 for an argument 47273 of '-Inf' and 1 for an argument of '+Inf'. Note while the 47274 parameter list is an ellipsis, this function only accepts exactly 47275 one floating-point argument. GCC treats this parameter as 47276 type-generic, which means it does not do default promotion from 47277 float to double. 47278 47279 -- Built-in Function: double __builtin_nan (const char *str) 47280 This is an implementation of the ISO C99 function 'nan'. 47281 47282 Since ISO C99 defines this function in terms of 'strtod', which we 47283 do not implement, a description of the parsing is in order. The 47284 string is parsed as by 'strtol'; that is, the base is recognized by 47285 leading '0' or '0x' prefixes. The number parsed is placed in the 47286 significand such that the least significant bit of the number is at 47287 the least significant bit of the significand. The number is 47288 truncated to fit the significand field provided. The significand 47289 is forced to be a quiet NaN. 47290 47291 This function, if given a string literal all of which would have 47292 been consumed by 'strtol', is evaluated early enough that it is 47293 considered a compile-time constant. 47294 47295 -- Built-in Function: _Decimal32 __builtin_nand32 (const char *str) 47296 Similar to '__builtin_nan', except the return type is '_Decimal32'. 47297 47298 -- Built-in Function: _Decimal64 __builtin_nand64 (const char *str) 47299 Similar to '__builtin_nan', except the return type is '_Decimal64'. 47300 47301 -- Built-in Function: _Decimal128 __builtin_nand128 (const char *str) 47302 Similar to '__builtin_nan', except the return type is 47303 '_Decimal128'. 47304 47305 -- Built-in Function: float __builtin_nanf (const char *str) 47306 Similar to '__builtin_nan', except the return type is 'float'. 47307 47308 -- Built-in Function: long double __builtin_nanl (const char *str) 47309 Similar to '__builtin_nan', except the return type is 'long 47310 double'. 47311 47312 -- Built-in Function: _FloatN __builtin_nanfN (const char *str) 47313 Similar to '__builtin_nan', except the return type is '_FloatN'. 47314 47315 -- Built-in Function: _FloatNx __builtin_nanfNx (const char *str) 47316 Similar to '__builtin_nan', except the return type is '_FloatNx'. 47317 47318 -- Built-in Function: double __builtin_nans (const char *str) 47319 Similar to '__builtin_nan', except the significand is forced to be 47320 a signaling NaN. The 'nans' function is proposed by WG14 N965. 47321 47322 -- Built-in Function: _Decimal32 __builtin_nansd32 (const char *str) 47323 Similar to '__builtin_nans', except the return type is 47324 '_Decimal32'. 47325 47326 -- Built-in Function: _Decimal64 __builtin_nansd64 (const char *str) 47327 Similar to '__builtin_nans', except the return type is 47328 '_Decimal64'. 47329 47330 -- Built-in Function: _Decimal128 __builtin_nansd128 (const char *str) 47331 Similar to '__builtin_nans', except the return type is 47332 '_Decimal128'. 47333 47334 -- Built-in Function: float __builtin_nansf (const char *str) 47335 Similar to '__builtin_nans', except the return type is 'float'. 47336 47337 -- Built-in Function: long double __builtin_nansl (const char *str) 47338 Similar to '__builtin_nans', except the return type is 'long 47339 double'. 47340 47341 -- Built-in Function: _FloatN __builtin_nansfN (const char *str) 47342 Similar to '__builtin_nans', except the return type is '_FloatN'. 47343 47344 -- Built-in Function: _FloatNx __builtin_nansfNx (const char *str) 47345 Similar to '__builtin_nans', except the return type is '_FloatNx'. 47346 47347 -- Built-in Function: int __builtin_ffs (int x) 47348 Returns one plus the index of the least significant 1-bit of X, or 47349 if X is zero, returns zero. 47350 47351 -- Built-in Function: int __builtin_clz (unsigned int x) 47352 Returns the number of leading 0-bits in X, starting at the most 47353 significant bit position. If X is 0, the result is undefined. 47354 47355 -- Built-in Function: int __builtin_ctz (unsigned int x) 47356 Returns the number of trailing 0-bits in X, starting at the least 47357 significant bit position. If X is 0, the result is undefined. 47358 47359 -- Built-in Function: int __builtin_clrsb (int x) 47360 Returns the number of leading redundant sign bits in X, i.e. the 47361 number of bits following the most significant bit that are 47362 identical to it. There are no special cases for 0 or other values. 47363 47364 -- Built-in Function: int __builtin_popcount (unsigned int x) 47365 Returns the number of 1-bits in X. 47366 47367 -- Built-in Function: int __builtin_parity (unsigned int x) 47368 Returns the parity of X, i.e. the number of 1-bits in X modulo 2. 47369 47370 -- Built-in Function: int __builtin_ffsl (long) 47371 Similar to '__builtin_ffs', except the argument type is 'long'. 47372 47373 -- Built-in Function: int __builtin_clzl (unsigned long) 47374 Similar to '__builtin_clz', except the argument type is 'unsigned 47375 long'. 47376 47377 -- Built-in Function: int __builtin_ctzl (unsigned long) 47378 Similar to '__builtin_ctz', except the argument type is 'unsigned 47379 long'. 47380 47381 -- Built-in Function: int __builtin_clrsbl (long) 47382 Similar to '__builtin_clrsb', except the argument type is 'long'. 47383 47384 -- Built-in Function: int __builtin_popcountl (unsigned long) 47385 Similar to '__builtin_popcount', except the argument type is 47386 'unsigned long'. 47387 47388 -- Built-in Function: int __builtin_parityl (unsigned long) 47389 Similar to '__builtin_parity', except the argument type is 47390 'unsigned long'. 47391 47392 -- Built-in Function: int __builtin_ffsll (long long) 47393 Similar to '__builtin_ffs', except the argument type is 'long 47394 long'. 47395 47396 -- Built-in Function: int __builtin_clzll (unsigned long long) 47397 Similar to '__builtin_clz', except the argument type is 'unsigned 47398 long long'. 47399 47400 -- Built-in Function: int __builtin_ctzll (unsigned long long) 47401 Similar to '__builtin_ctz', except the argument type is 'unsigned 47402 long long'. 47403 47404 -- Built-in Function: int __builtin_clrsbll (long long) 47405 Similar to '__builtin_clrsb', except the argument type is 'long 47406 long'. 47407 47408 -- Built-in Function: int __builtin_popcountll (unsigned long long) 47409 Similar to '__builtin_popcount', except the argument type is 47410 'unsigned long long'. 47411 47412 -- Built-in Function: int __builtin_parityll (unsigned long long) 47413 Similar to '__builtin_parity', except the argument type is 47414 'unsigned long long'. 47415 47416 -- Built-in Function: double __builtin_powi (double, int) 47417 Returns the first argument raised to the power of the second. 47418 Unlike the 'pow' function no guarantees about precision and 47419 rounding are made. 47420 47421 -- Built-in Function: float __builtin_powif (float, int) 47422 Similar to '__builtin_powi', except the argument and return types 47423 are 'float'. 47424 47425 -- Built-in Function: long double __builtin_powil (long double, int) 47426 Similar to '__builtin_powi', except the argument and return types 47427 are 'long double'. 47428 47429 -- Built-in Function: uint16_t __builtin_bswap16 (uint16_t x) 47430 Returns X with the order of the bytes reversed; for example, 47431 '0xaabb' becomes '0xbbaa'. Byte here always means exactly 8 bits. 47432 47433 -- Built-in Function: uint32_t __builtin_bswap32 (uint32_t x) 47434 Similar to '__builtin_bswap16', except the argument and return 47435 types are 32-bit. 47436 47437 -- Built-in Function: uint64_t __builtin_bswap64 (uint64_t x) 47438 Similar to '__builtin_bswap32', except the argument and return 47439 types are 64-bit. 47440 47441 -- Built-in Function: uint128_t __builtin_bswap128 (uint128_t x) 47442 Similar to '__builtin_bswap64', except the argument and return 47443 types are 128-bit. Only supported on targets when 128-bit types 47444 are supported. 47445 47446 -- Built-in Function: Pmode __builtin_extend_pointer (void * x) 47447 On targets where the user visible pointer size is smaller than the 47448 size of an actual hardware address this function returns the 47449 extended user pointer. Targets where this is true included ILP32 47450 mode on x86_64 or Aarch64. This function is mainly useful when 47451 writing inline assembly code. 47452 47453 -- Built-in Function: int __builtin_goacc_parlevel_id (int x) 47454 Returns the openacc gang, worker or vector id depending on whether 47455 X is 0, 1 or 2. 47456 47457 -- Built-in Function: int __builtin_goacc_parlevel_size (int x) 47458 Returns the openacc gang, worker or vector size depending on 47459 whether X is 0, 1 or 2. 47460 47461 47462File: gcc.info, Node: Target Builtins, Next: Target Format Checks, Prev: Other Builtins, Up: C Extensions 47463 474646.60 Built-in Functions Specific to Particular Target Machines 47465============================================================== 47466 47467On some target machines, GCC supports many built-in functions specific 47468to those machines. Generally these generate calls to specific machine 47469instructions, but allow the compiler to schedule those calls. 47470 47471* Menu: 47472 47473* AArch64 Built-in Functions:: 47474* Alpha Built-in Functions:: 47475* Altera Nios II Built-in Functions:: 47476* ARC Built-in Functions:: 47477* ARC SIMD Built-in Functions:: 47478* ARM iWMMXt Built-in Functions:: 47479* ARM C Language Extensions (ACLE):: 47480* ARM Floating Point Status and Control Intrinsics:: 47481* ARM ARMv8-M Security Extensions:: 47482* AVR Built-in Functions:: 47483* Blackfin Built-in Functions:: 47484* BPF Built-in Functions:: 47485* FR-V Built-in Functions:: 47486* MIPS DSP Built-in Functions:: 47487* MIPS Paired-Single Support:: 47488* MIPS Loongson Built-in Functions:: 47489* MIPS SIMD Architecture (MSA) Support:: 47490* Other MIPS Built-in Functions:: 47491* MSP430 Built-in Functions:: 47492* NDS32 Built-in Functions:: 47493* picoChip Built-in Functions:: 47494* Basic PowerPC Built-in Functions:: 47495* PowerPC AltiVec/VSX Built-in Functions:: 47496* PowerPC Hardware Transactional Memory Built-in Functions:: 47497* PowerPC Atomic Memory Operation Functions:: 47498* PowerPC Matrix-Multiply Assist Built-in Functions:: 47499* PRU Built-in Functions:: 47500* RISC-V Built-in Functions:: 47501* RX Built-in Functions:: 47502* S/390 System z Built-in Functions:: 47503* SH Built-in Functions:: 47504* SPARC VIS Built-in Functions:: 47505* TI C6X Built-in Functions:: 47506* TILE-Gx Built-in Functions:: 47507* TILEPro Built-in Functions:: 47508* x86 Built-in Functions:: 47509* x86 transactional memory intrinsics:: 47510* x86 control-flow protection intrinsics:: 47511 47512 47513File: gcc.info, Node: AArch64 Built-in Functions, Next: Alpha Built-in Functions, Up: Target Builtins 47514 475156.60.1 AArch64 Built-in Functions 47516--------------------------------- 47517 47518These built-in functions are available for the AArch64 family of 47519processors. 47520 unsigned int __builtin_aarch64_get_fpcr (); 47521 void __builtin_aarch64_set_fpcr (unsigned int); 47522 unsigned int __builtin_aarch64_get_fpsr (); 47523 void __builtin_aarch64_set_fpsr (unsigned int); 47524 47525 unsigned long long __builtin_aarch64_get_fpcr64 (); 47526 void __builtin_aarch64_set_fpcr64 (unsigned long long); 47527 unsigned long long __builtin_aarch64_get_fpsr64 (); 47528 void __builtin_aarch64_set_fpsr64 (unsigned long long); 47529 47530 47531File: gcc.info, Node: Alpha Built-in Functions, Next: Altera Nios II Built-in Functions, Prev: AArch64 Built-in Functions, Up: Target Builtins 47532 475336.60.2 Alpha Built-in Functions 47534------------------------------- 47535 47536These built-in functions are available for the Alpha family of 47537processors, depending on the command-line switches used. 47538 47539 The following built-in functions are always available. They all 47540generate the machine instruction that is part of the name. 47541 47542 long __builtin_alpha_implver (void); 47543 long __builtin_alpha_rpcc (void); 47544 long __builtin_alpha_amask (long); 47545 long __builtin_alpha_cmpbge (long, long); 47546 long __builtin_alpha_extbl (long, long); 47547 long __builtin_alpha_extwl (long, long); 47548 long __builtin_alpha_extll (long, long); 47549 long __builtin_alpha_extql (long, long); 47550 long __builtin_alpha_extwh (long, long); 47551 long __builtin_alpha_extlh (long, long); 47552 long __builtin_alpha_extqh (long, long); 47553 long __builtin_alpha_insbl (long, long); 47554 long __builtin_alpha_inswl (long, long); 47555 long __builtin_alpha_insll (long, long); 47556 long __builtin_alpha_insql (long, long); 47557 long __builtin_alpha_inswh (long, long); 47558 long __builtin_alpha_inslh (long, long); 47559 long __builtin_alpha_insqh (long, long); 47560 long __builtin_alpha_mskbl (long, long); 47561 long __builtin_alpha_mskwl (long, long); 47562 long __builtin_alpha_mskll (long, long); 47563 long __builtin_alpha_mskql (long, long); 47564 long __builtin_alpha_mskwh (long, long); 47565 long __builtin_alpha_msklh (long, long); 47566 long __builtin_alpha_mskqh (long, long); 47567 long __builtin_alpha_umulh (long, long); 47568 long __builtin_alpha_zap (long, long); 47569 long __builtin_alpha_zapnot (long, long); 47570 47571 The following built-in functions are always with '-mmax' or '-mcpu=CPU' 47572where CPU is 'pca56' or later. They all generate the machine 47573instruction that is part of the name. 47574 47575 long __builtin_alpha_pklb (long); 47576 long __builtin_alpha_pkwb (long); 47577 long __builtin_alpha_unpkbl (long); 47578 long __builtin_alpha_unpkbw (long); 47579 long __builtin_alpha_minub8 (long, long); 47580 long __builtin_alpha_minsb8 (long, long); 47581 long __builtin_alpha_minuw4 (long, long); 47582 long __builtin_alpha_minsw4 (long, long); 47583 long __builtin_alpha_maxub8 (long, long); 47584 long __builtin_alpha_maxsb8 (long, long); 47585 long __builtin_alpha_maxuw4 (long, long); 47586 long __builtin_alpha_maxsw4 (long, long); 47587 long __builtin_alpha_perr (long, long); 47588 47589 The following built-in functions are always with '-mcix' or '-mcpu=CPU' 47590where CPU is 'ev67' or later. They all generate the machine instruction 47591that is part of the name. 47592 47593 long __builtin_alpha_cttz (long); 47594 long __builtin_alpha_ctlz (long); 47595 long __builtin_alpha_ctpop (long); 47596 47597 The following built-in functions are available on systems that use the 47598OSF/1 PALcode. Normally they invoke the 'rduniq' and 'wruniq' PAL 47599calls, but when invoked with '-mtls-kernel', they invoke 'rdval' and 47600'wrval'. 47601 47602 void *__builtin_thread_pointer (void); 47603 void __builtin_set_thread_pointer (void *); 47604 47605 47606File: gcc.info, Node: Altera Nios II Built-in Functions, Next: ARC Built-in Functions, Prev: Alpha Built-in Functions, Up: Target Builtins 47607 476086.60.3 Altera Nios II Built-in Functions 47609---------------------------------------- 47610 47611These built-in functions are available for the Altera Nios II family of 47612processors. 47613 47614 The following built-in functions are always available. They all 47615generate the machine instruction that is part of the name. 47616 47617 int __builtin_ldbio (volatile const void *); 47618 int __builtin_ldbuio (volatile const void *); 47619 int __builtin_ldhio (volatile const void *); 47620 int __builtin_ldhuio (volatile const void *); 47621 int __builtin_ldwio (volatile const void *); 47622 void __builtin_stbio (volatile void *, int); 47623 void __builtin_sthio (volatile void *, int); 47624 void __builtin_stwio (volatile void *, int); 47625 void __builtin_sync (void); 47626 int __builtin_rdctl (int); 47627 int __builtin_rdprs (int, int); 47628 void __builtin_wrctl (int, int); 47629 void __builtin_flushd (volatile void *); 47630 void __builtin_flushda (volatile void *); 47631 int __builtin_wrpie (int); 47632 void __builtin_eni (int); 47633 int __builtin_ldex (volatile const void *); 47634 int __builtin_stex (volatile void *, int); 47635 int __builtin_ldsex (volatile const void *); 47636 int __builtin_stsex (volatile void *, int); 47637 47638 The following built-in functions are always available. They all 47639generate a Nios II Custom Instruction. The name of the function 47640represents the types that the function takes and returns. The letter 47641before the 'n' is the return type or void if absent. The 'n' represents 47642the first parameter to all the custom instructions, the custom 47643instruction number. The two letters after the 'n' represent the up to 47644two parameters to the function. 47645 47646 The letters represent the following data types: 47647'<no letter>' 47648 'void' for return type and no parameter for parameter types. 47649 47650'i' 47651 'int' for return type and parameter type 47652 47653'f' 47654 'float' for return type and parameter type 47655 47656'p' 47657 'void *' for return type and parameter type 47658 47659 And the function names are: 47660 void __builtin_custom_n (void); 47661 void __builtin_custom_ni (int); 47662 void __builtin_custom_nf (float); 47663 void __builtin_custom_np (void *); 47664 void __builtin_custom_nii (int, int); 47665 void __builtin_custom_nif (int, float); 47666 void __builtin_custom_nip (int, void *); 47667 void __builtin_custom_nfi (float, int); 47668 void __builtin_custom_nff (float, float); 47669 void __builtin_custom_nfp (float, void *); 47670 void __builtin_custom_npi (void *, int); 47671 void __builtin_custom_npf (void *, float); 47672 void __builtin_custom_npp (void *, void *); 47673 int __builtin_custom_in (void); 47674 int __builtin_custom_ini (int); 47675 int __builtin_custom_inf (float); 47676 int __builtin_custom_inp (void *); 47677 int __builtin_custom_inii (int, int); 47678 int __builtin_custom_inif (int, float); 47679 int __builtin_custom_inip (int, void *); 47680 int __builtin_custom_infi (float, int); 47681 int __builtin_custom_inff (float, float); 47682 int __builtin_custom_infp (float, void *); 47683 int __builtin_custom_inpi (void *, int); 47684 int __builtin_custom_inpf (void *, float); 47685 int __builtin_custom_inpp (void *, void *); 47686 float __builtin_custom_fn (void); 47687 float __builtin_custom_fni (int); 47688 float __builtin_custom_fnf (float); 47689 float __builtin_custom_fnp (void *); 47690 float __builtin_custom_fnii (int, int); 47691 float __builtin_custom_fnif (int, float); 47692 float __builtin_custom_fnip (int, void *); 47693 float __builtin_custom_fnfi (float, int); 47694 float __builtin_custom_fnff (float, float); 47695 float __builtin_custom_fnfp (float, void *); 47696 float __builtin_custom_fnpi (void *, int); 47697 float __builtin_custom_fnpf (void *, float); 47698 float __builtin_custom_fnpp (void *, void *); 47699 void * __builtin_custom_pn (void); 47700 void * __builtin_custom_pni (int); 47701 void * __builtin_custom_pnf (float); 47702 void * __builtin_custom_pnp (void *); 47703 void * __builtin_custom_pnii (int, int); 47704 void * __builtin_custom_pnif (int, float); 47705 void * __builtin_custom_pnip (int, void *); 47706 void * __builtin_custom_pnfi (float, int); 47707 void * __builtin_custom_pnff (float, float); 47708 void * __builtin_custom_pnfp (float, void *); 47709 void * __builtin_custom_pnpi (void *, int); 47710 void * __builtin_custom_pnpf (void *, float); 47711 void * __builtin_custom_pnpp (void *, void *); 47712 47713 47714File: gcc.info, Node: ARC Built-in Functions, Next: ARC SIMD Built-in Functions, Prev: Altera Nios II Built-in Functions, Up: Target Builtins 47715 477166.60.4 ARC Built-in Functions 47717----------------------------- 47718 47719The following built-in functions are provided for ARC targets. The 47720built-ins generate the corresponding assembly instructions. In the 47721examples given below, the generated code often requires an operand or 47722result to be in a register. Where necessary further code will be 47723generated to ensure this is true, but for brevity this is not described 47724in each case. 47725 47726 _Note:_ Using a built-in to generate an instruction not supported by a 47727target may cause problems. At present the compiler is not guaranteed to 47728detect such misuse, and as a result an internal compiler error may be 47729generated. 47730 47731 -- Built-in Function: int __builtin_arc_aligned (void *VAL, int 47732 ALIGNVAL) 47733 Return 1 if VAL is known to have the byte alignment given by 47734 ALIGNVAL, otherwise return 0. Note that this is different from 47735 __alignof__(*(char *)VAL) >= alignval 47736 because __alignof__ sees only the type of the dereference, whereas 47737 __builtin_arc_align uses alignment information from the pointer as 47738 well as from the pointed-to type. The information available will 47739 depend on optimization level. 47740 47741 -- Built-in Function: void __builtin_arc_brk (void) 47742 Generates 47743 brk 47744 47745 -- Built-in Function: unsigned int __builtin_arc_core_read (unsigned 47746 int REGNO) 47747 The operand is the number of a register to be read. Generates: 47748 mov DEST, rREGNO 47749 where the value in DEST will be the result returned from the 47750 built-in. 47751 47752 -- Built-in Function: void __builtin_arc_core_write (unsigned int 47753 REGNO, unsigned int VAL) 47754 The first operand is the number of a register to be written, the 47755 second operand is a compile time constant to write into that 47756 register. Generates: 47757 mov rREGNO, VAL 47758 47759 -- Built-in Function: int __builtin_arc_divaw (int A, int B) 47760 Only available if either '-mcpu=ARC700' or '-meA' is set. 47761 Generates: 47762 divaw DEST, A, B 47763 where the value in DEST will be the result returned from the 47764 built-in. 47765 47766 -- Built-in Function: void __builtin_arc_flag (unsigned int A) 47767 Generates 47768 flag A 47769 47770 -- Built-in Function: unsigned int __builtin_arc_lr (unsigned int AUXR) 47771 The operand, AUXV, is the address of an auxiliary register and must 47772 be a compile time constant. Generates: 47773 lr DEST, [AUXR] 47774 Where the value in DEST will be the result returned from the 47775 built-in. 47776 47777 -- Built-in Function: void __builtin_arc_mul64 (int A, int B) 47778 Only available with '-mmul64'. Generates: 47779 mul64 A, B 47780 47781 -- Built-in Function: void __builtin_arc_mulu64 (unsigned int A, 47782 unsigned int B) 47783 Only available with '-mmul64'. Generates: 47784 mulu64 A, B 47785 47786 -- Built-in Function: void __builtin_arc_nop (void) 47787 Generates: 47788 nop 47789 47790 -- Built-in Function: int __builtin_arc_norm (int SRC) 47791 Only valid if the 'norm' instruction is available through the 47792 '-mnorm' option or by default with '-mcpu=ARC700'. Generates: 47793 norm DEST, SRC 47794 Where the value in DEST will be the result returned from the 47795 built-in. 47796 47797 -- Built-in Function: short int __builtin_arc_normw (short int SRC) 47798 Only valid if the 'normw' instruction is available through the 47799 '-mnorm' option or by default with '-mcpu=ARC700'. Generates: 47800 normw DEST, SRC 47801 Where the value in DEST will be the result returned from the 47802 built-in. 47803 47804 -- Built-in Function: void __builtin_arc_rtie (void) 47805 Generates: 47806 rtie 47807 47808 -- Built-in Function: void __builtin_arc_sleep (int A 47809 Generates: 47810 sleep A 47811 47812 -- Built-in Function: void __builtin_arc_sr (unsigned int VAL, unsigned 47813 int AUXR) 47814 The first argument, VAL, is a compile time constant to be written 47815 to the register, the second argument, AUXR, is the address of an 47816 auxiliary register. Generates: 47817 sr VAL, [AUXR] 47818 47819 -- Built-in Function: int __builtin_arc_swap (int SRC) 47820 Only valid with '-mswap'. Generates: 47821 swap DEST, SRC 47822 Where the value in DEST will be the result returned from the 47823 built-in. 47824 47825 -- Built-in Function: void __builtin_arc_swi (void) 47826 Generates: 47827 swi 47828 47829 -- Built-in Function: void __builtin_arc_sync (void) 47830 Only available with '-mcpu=ARC700'. Generates: 47831 sync 47832 47833 -- Built-in Function: void __builtin_arc_trap_s (unsigned int C) 47834 Only available with '-mcpu=ARC700'. Generates: 47835 trap_s C 47836 47837 -- Built-in Function: void __builtin_arc_unimp_s (void) 47838 Only available with '-mcpu=ARC700'. Generates: 47839 unimp_s 47840 47841 The instructions generated by the following builtins are not considered 47842as candidates for scheduling. They are not moved around by the compiler 47843during scheduling, and thus can be expected to appear where they are put 47844in the C code: 47845 __builtin_arc_brk() 47846 __builtin_arc_core_read() 47847 __builtin_arc_core_write() 47848 __builtin_arc_flag() 47849 __builtin_arc_lr() 47850 __builtin_arc_sleep() 47851 __builtin_arc_sr() 47852 __builtin_arc_swi() 47853 47854 47855File: gcc.info, Node: ARC SIMD Built-in Functions, Next: ARM iWMMXt Built-in Functions, Prev: ARC Built-in Functions, Up: Target Builtins 47856 478576.60.5 ARC SIMD Built-in Functions 47858---------------------------------- 47859 47860SIMD builtins provided by the compiler can be used to generate the 47861vector instructions. This section describes the available builtins and 47862their usage in programs. With the '-msimd' option, the compiler 47863provides 128-bit vector types, which can be specified using the 47864'vector_size' attribute. The header file 'arc-simd.h' can be included 47865to use the following predefined types: 47866 typedef int __v4si __attribute__((vector_size(16))); 47867 typedef short __v8hi __attribute__((vector_size(16))); 47868 47869 These types can be used to define 128-bit variables. The built-in 47870functions listed in the following section can be used on these variables 47871to generate the vector operations. 47872 47873 For all builtins, '__builtin_arc_SOMEINSN', the header file 47874'arc-simd.h' also provides equivalent macros called '_SOMEINSN' that can 47875be used for programming ease and improved readability. The following 47876macros for DMA control are also provided: 47877 #define _setup_dma_in_channel_reg _vdiwr 47878 #define _setup_dma_out_channel_reg _vdowr 47879 47880 The following is a complete list of all the SIMD built-ins provided for 47881ARC, grouped by calling signature. 47882 47883 The following take two '__v8hi' arguments and return a '__v8hi' result: 47884 __v8hi __builtin_arc_vaddaw (__v8hi, __v8hi); 47885 __v8hi __builtin_arc_vaddw (__v8hi, __v8hi); 47886 __v8hi __builtin_arc_vand (__v8hi, __v8hi); 47887 __v8hi __builtin_arc_vandaw (__v8hi, __v8hi); 47888 __v8hi __builtin_arc_vavb (__v8hi, __v8hi); 47889 __v8hi __builtin_arc_vavrb (__v8hi, __v8hi); 47890 __v8hi __builtin_arc_vbic (__v8hi, __v8hi); 47891 __v8hi __builtin_arc_vbicaw (__v8hi, __v8hi); 47892 __v8hi __builtin_arc_vdifaw (__v8hi, __v8hi); 47893 __v8hi __builtin_arc_vdifw (__v8hi, __v8hi); 47894 __v8hi __builtin_arc_veqw (__v8hi, __v8hi); 47895 __v8hi __builtin_arc_vh264f (__v8hi, __v8hi); 47896 __v8hi __builtin_arc_vh264ft (__v8hi, __v8hi); 47897 __v8hi __builtin_arc_vh264fw (__v8hi, __v8hi); 47898 __v8hi __builtin_arc_vlew (__v8hi, __v8hi); 47899 __v8hi __builtin_arc_vltw (__v8hi, __v8hi); 47900 __v8hi __builtin_arc_vmaxaw (__v8hi, __v8hi); 47901 __v8hi __builtin_arc_vmaxw (__v8hi, __v8hi); 47902 __v8hi __builtin_arc_vminaw (__v8hi, __v8hi); 47903 __v8hi __builtin_arc_vminw (__v8hi, __v8hi); 47904 __v8hi __builtin_arc_vmr1aw (__v8hi, __v8hi); 47905 __v8hi __builtin_arc_vmr1w (__v8hi, __v8hi); 47906 __v8hi __builtin_arc_vmr2aw (__v8hi, __v8hi); 47907 __v8hi __builtin_arc_vmr2w (__v8hi, __v8hi); 47908 __v8hi __builtin_arc_vmr3aw (__v8hi, __v8hi); 47909 __v8hi __builtin_arc_vmr3w (__v8hi, __v8hi); 47910 __v8hi __builtin_arc_vmr4aw (__v8hi, __v8hi); 47911 __v8hi __builtin_arc_vmr4w (__v8hi, __v8hi); 47912 __v8hi __builtin_arc_vmr5aw (__v8hi, __v8hi); 47913 __v8hi __builtin_arc_vmr5w (__v8hi, __v8hi); 47914 __v8hi __builtin_arc_vmr6aw (__v8hi, __v8hi); 47915 __v8hi __builtin_arc_vmr6w (__v8hi, __v8hi); 47916 __v8hi __builtin_arc_vmr7aw (__v8hi, __v8hi); 47917 __v8hi __builtin_arc_vmr7w (__v8hi, __v8hi); 47918 __v8hi __builtin_arc_vmrb (__v8hi, __v8hi); 47919 __v8hi __builtin_arc_vmulaw (__v8hi, __v8hi); 47920 __v8hi __builtin_arc_vmulfaw (__v8hi, __v8hi); 47921 __v8hi __builtin_arc_vmulfw (__v8hi, __v8hi); 47922 __v8hi __builtin_arc_vmulw (__v8hi, __v8hi); 47923 __v8hi __builtin_arc_vnew (__v8hi, __v8hi); 47924 __v8hi __builtin_arc_vor (__v8hi, __v8hi); 47925 __v8hi __builtin_arc_vsubaw (__v8hi, __v8hi); 47926 __v8hi __builtin_arc_vsubw (__v8hi, __v8hi); 47927 __v8hi __builtin_arc_vsummw (__v8hi, __v8hi); 47928 __v8hi __builtin_arc_vvc1f (__v8hi, __v8hi); 47929 __v8hi __builtin_arc_vvc1ft (__v8hi, __v8hi); 47930 __v8hi __builtin_arc_vxor (__v8hi, __v8hi); 47931 __v8hi __builtin_arc_vxoraw (__v8hi, __v8hi); 47932 47933 The following take one '__v8hi' and one 'int' argument and return a 47934'__v8hi' result: 47935 47936 __v8hi __builtin_arc_vbaddw (__v8hi, int); 47937 __v8hi __builtin_arc_vbmaxw (__v8hi, int); 47938 __v8hi __builtin_arc_vbminw (__v8hi, int); 47939 __v8hi __builtin_arc_vbmulaw (__v8hi, int); 47940 __v8hi __builtin_arc_vbmulfw (__v8hi, int); 47941 __v8hi __builtin_arc_vbmulw (__v8hi, int); 47942 __v8hi __builtin_arc_vbrsubw (__v8hi, int); 47943 __v8hi __builtin_arc_vbsubw (__v8hi, int); 47944 47945 The following take one '__v8hi' argument and one 'int' argument which 47946must be a 3-bit compile time constant indicating a register number 47947I0-I7. They return a '__v8hi' result. 47948 __v8hi __builtin_arc_vasrw (__v8hi, const int); 47949 __v8hi __builtin_arc_vsr8 (__v8hi, const int); 47950 __v8hi __builtin_arc_vsr8aw (__v8hi, const int); 47951 47952 The following take one '__v8hi' argument and one 'int' argument which 47953must be a 6-bit compile time constant. They return a '__v8hi' result. 47954 __v8hi __builtin_arc_vasrpwbi (__v8hi, const int); 47955 __v8hi __builtin_arc_vasrrpwbi (__v8hi, const int); 47956 __v8hi __builtin_arc_vasrrwi (__v8hi, const int); 47957 __v8hi __builtin_arc_vasrsrwi (__v8hi, const int); 47958 __v8hi __builtin_arc_vasrwi (__v8hi, const int); 47959 __v8hi __builtin_arc_vsr8awi (__v8hi, const int); 47960 __v8hi __builtin_arc_vsr8i (__v8hi, const int); 47961 47962 The following take one '__v8hi' argument and one 'int' argument which 47963must be a 8-bit compile time constant. They return a '__v8hi' result. 47964 __v8hi __builtin_arc_vd6tapf (__v8hi, const int); 47965 __v8hi __builtin_arc_vmvaw (__v8hi, const int); 47966 __v8hi __builtin_arc_vmvw (__v8hi, const int); 47967 __v8hi __builtin_arc_vmvzw (__v8hi, const int); 47968 47969 The following take two 'int' arguments, the second of which which must 47970be a 8-bit compile time constant. They return a '__v8hi' result: 47971 __v8hi __builtin_arc_vmovaw (int, const int); 47972 __v8hi __builtin_arc_vmovw (int, const int); 47973 __v8hi __builtin_arc_vmovzw (int, const int); 47974 47975 The following take a single '__v8hi' argument and return a '__v8hi' 47976result: 47977 __v8hi __builtin_arc_vabsaw (__v8hi); 47978 __v8hi __builtin_arc_vabsw (__v8hi); 47979 __v8hi __builtin_arc_vaddsuw (__v8hi); 47980 __v8hi __builtin_arc_vexch1 (__v8hi); 47981 __v8hi __builtin_arc_vexch2 (__v8hi); 47982 __v8hi __builtin_arc_vexch4 (__v8hi); 47983 __v8hi __builtin_arc_vsignw (__v8hi); 47984 __v8hi __builtin_arc_vupbaw (__v8hi); 47985 __v8hi __builtin_arc_vupbw (__v8hi); 47986 __v8hi __builtin_arc_vupsbaw (__v8hi); 47987 __v8hi __builtin_arc_vupsbw (__v8hi); 47988 47989 The following take two 'int' arguments and return no result: 47990 void __builtin_arc_vdirun (int, int); 47991 void __builtin_arc_vdorun (int, int); 47992 47993 The following take two 'int' arguments and return no result. The first 47994argument must a 3-bit compile time constant indicating one of the 47995DR0-DR7 DMA setup channels: 47996 void __builtin_arc_vdiwr (const int, int); 47997 void __builtin_arc_vdowr (const int, int); 47998 47999 The following take an 'int' argument and return no result: 48000 void __builtin_arc_vendrec (int); 48001 void __builtin_arc_vrec (int); 48002 void __builtin_arc_vrecrun (int); 48003 void __builtin_arc_vrun (int); 48004 48005 The following take a '__v8hi' argument and two 'int' arguments and 48006return a '__v8hi' result. The second argument must be a 3-bit compile 48007time constants, indicating one the registers I0-I7, and the third 48008argument must be an 8-bit compile time constant. 48009 48010 _Note:_ Although the equivalent hardware instructions do not take an 48011SIMD register as an operand, these builtins overwrite the relevant bits 48012of the '__v8hi' register provided as the first argument with the value 48013loaded from the '[Ib, u8]' location in the SDM. 48014 48015 __v8hi __builtin_arc_vld32 (__v8hi, const int, const int); 48016 __v8hi __builtin_arc_vld32wh (__v8hi, const int, const int); 48017 __v8hi __builtin_arc_vld32wl (__v8hi, const int, const int); 48018 __v8hi __builtin_arc_vld64 (__v8hi, const int, const int); 48019 48020 The following take two 'int' arguments and return a '__v8hi' result. 48021The first argument must be a 3-bit compile time constants, indicating 48022one the registers I0-I7, and the second argument must be an 8-bit 48023compile time constant. 48024 48025 __v8hi __builtin_arc_vld128 (const int, const int); 48026 __v8hi __builtin_arc_vld64w (const int, const int); 48027 48028 The following take a '__v8hi' argument and two 'int' arguments and 48029return no result. The second argument must be a 3-bit compile time 48030constants, indicating one the registers I0-I7, and the third argument 48031must be an 8-bit compile time constant. 48032 48033 void __builtin_arc_vst128 (__v8hi, const int, const int); 48034 void __builtin_arc_vst64 (__v8hi, const int, const int); 48035 48036 The following take a '__v8hi' argument and three 'int' arguments and 48037return no result. The second argument must be a 3-bit compile-time 48038constant, identifying the 16-bit sub-register to be stored, the third 48039argument must be a 3-bit compile time constants, indicating one the 48040registers I0-I7, and the fourth argument must be an 8-bit compile time 48041constant. 48042 48043 void __builtin_arc_vst16_n (__v8hi, const int, const int, const int); 48044 void __builtin_arc_vst32_n (__v8hi, const int, const int, const int); 48045 48046 48047File: gcc.info, Node: ARM iWMMXt Built-in Functions, Next: ARM C Language Extensions (ACLE), Prev: ARC SIMD Built-in Functions, Up: Target Builtins 48048 480496.60.6 ARM iWMMXt Built-in Functions 48050------------------------------------ 48051 48052These built-in functions are available for the ARM family of processors 48053when the '-mcpu=iwmmxt' switch is used: 48054 48055 typedef int v2si __attribute__ ((vector_size (8))); 48056 typedef short v4hi __attribute__ ((vector_size (8))); 48057 typedef char v8qi __attribute__ ((vector_size (8))); 48058 48059 int __builtin_arm_getwcgr0 (void); 48060 void __builtin_arm_setwcgr0 (int); 48061 int __builtin_arm_getwcgr1 (void); 48062 void __builtin_arm_setwcgr1 (int); 48063 int __builtin_arm_getwcgr2 (void); 48064 void __builtin_arm_setwcgr2 (int); 48065 int __builtin_arm_getwcgr3 (void); 48066 void __builtin_arm_setwcgr3 (int); 48067 int __builtin_arm_textrmsb (v8qi, int); 48068 int __builtin_arm_textrmsh (v4hi, int); 48069 int __builtin_arm_textrmsw (v2si, int); 48070 int __builtin_arm_textrmub (v8qi, int); 48071 int __builtin_arm_textrmuh (v4hi, int); 48072 int __builtin_arm_textrmuw (v2si, int); 48073 v8qi __builtin_arm_tinsrb (v8qi, int, int); 48074 v4hi __builtin_arm_tinsrh (v4hi, int, int); 48075 v2si __builtin_arm_tinsrw (v2si, int, int); 48076 long long __builtin_arm_tmia (long long, int, int); 48077 long long __builtin_arm_tmiabb (long long, int, int); 48078 long long __builtin_arm_tmiabt (long long, int, int); 48079 long long __builtin_arm_tmiaph (long long, int, int); 48080 long long __builtin_arm_tmiatb (long long, int, int); 48081 long long __builtin_arm_tmiatt (long long, int, int); 48082 int __builtin_arm_tmovmskb (v8qi); 48083 int __builtin_arm_tmovmskh (v4hi); 48084 int __builtin_arm_tmovmskw (v2si); 48085 long long __builtin_arm_waccb (v8qi); 48086 long long __builtin_arm_wacch (v4hi); 48087 long long __builtin_arm_waccw (v2si); 48088 v8qi __builtin_arm_waddb (v8qi, v8qi); 48089 v8qi __builtin_arm_waddbss (v8qi, v8qi); 48090 v8qi __builtin_arm_waddbus (v8qi, v8qi); 48091 v4hi __builtin_arm_waddh (v4hi, v4hi); 48092 v4hi __builtin_arm_waddhss (v4hi, v4hi); 48093 v4hi __builtin_arm_waddhus (v4hi, v4hi); 48094 v2si __builtin_arm_waddw (v2si, v2si); 48095 v2si __builtin_arm_waddwss (v2si, v2si); 48096 v2si __builtin_arm_waddwus (v2si, v2si); 48097 v8qi __builtin_arm_walign (v8qi, v8qi, int); 48098 long long __builtin_arm_wand(long long, long long); 48099 long long __builtin_arm_wandn (long long, long long); 48100 v8qi __builtin_arm_wavg2b (v8qi, v8qi); 48101 v8qi __builtin_arm_wavg2br (v8qi, v8qi); 48102 v4hi __builtin_arm_wavg2h (v4hi, v4hi); 48103 v4hi __builtin_arm_wavg2hr (v4hi, v4hi); 48104 v8qi __builtin_arm_wcmpeqb (v8qi, v8qi); 48105 v4hi __builtin_arm_wcmpeqh (v4hi, v4hi); 48106 v2si __builtin_arm_wcmpeqw (v2si, v2si); 48107 v8qi __builtin_arm_wcmpgtsb (v8qi, v8qi); 48108 v4hi __builtin_arm_wcmpgtsh (v4hi, v4hi); 48109 v2si __builtin_arm_wcmpgtsw (v2si, v2si); 48110 v8qi __builtin_arm_wcmpgtub (v8qi, v8qi); 48111 v4hi __builtin_arm_wcmpgtuh (v4hi, v4hi); 48112 v2si __builtin_arm_wcmpgtuw (v2si, v2si); 48113 long long __builtin_arm_wmacs (long long, v4hi, v4hi); 48114 long long __builtin_arm_wmacsz (v4hi, v4hi); 48115 long long __builtin_arm_wmacu (long long, v4hi, v4hi); 48116 long long __builtin_arm_wmacuz (v4hi, v4hi); 48117 v4hi __builtin_arm_wmadds (v4hi, v4hi); 48118 v4hi __builtin_arm_wmaddu (v4hi, v4hi); 48119 v8qi __builtin_arm_wmaxsb (v8qi, v8qi); 48120 v4hi __builtin_arm_wmaxsh (v4hi, v4hi); 48121 v2si __builtin_arm_wmaxsw (v2si, v2si); 48122 v8qi __builtin_arm_wmaxub (v8qi, v8qi); 48123 v4hi __builtin_arm_wmaxuh (v4hi, v4hi); 48124 v2si __builtin_arm_wmaxuw (v2si, v2si); 48125 v8qi __builtin_arm_wminsb (v8qi, v8qi); 48126 v4hi __builtin_arm_wminsh (v4hi, v4hi); 48127 v2si __builtin_arm_wminsw (v2si, v2si); 48128 v8qi __builtin_arm_wminub (v8qi, v8qi); 48129 v4hi __builtin_arm_wminuh (v4hi, v4hi); 48130 v2si __builtin_arm_wminuw (v2si, v2si); 48131 v4hi __builtin_arm_wmulsm (v4hi, v4hi); 48132 v4hi __builtin_arm_wmulul (v4hi, v4hi); 48133 v4hi __builtin_arm_wmulum (v4hi, v4hi); 48134 long long __builtin_arm_wor (long long, long long); 48135 v2si __builtin_arm_wpackdss (long long, long long); 48136 v2si __builtin_arm_wpackdus (long long, long long); 48137 v8qi __builtin_arm_wpackhss (v4hi, v4hi); 48138 v8qi __builtin_arm_wpackhus (v4hi, v4hi); 48139 v4hi __builtin_arm_wpackwss (v2si, v2si); 48140 v4hi __builtin_arm_wpackwus (v2si, v2si); 48141 long long __builtin_arm_wrord (long long, long long); 48142 long long __builtin_arm_wrordi (long long, int); 48143 v4hi __builtin_arm_wrorh (v4hi, long long); 48144 v4hi __builtin_arm_wrorhi (v4hi, int); 48145 v2si __builtin_arm_wrorw (v2si, long long); 48146 v2si __builtin_arm_wrorwi (v2si, int); 48147 v2si __builtin_arm_wsadb (v2si, v8qi, v8qi); 48148 v2si __builtin_arm_wsadbz (v8qi, v8qi); 48149 v2si __builtin_arm_wsadh (v2si, v4hi, v4hi); 48150 v2si __builtin_arm_wsadhz (v4hi, v4hi); 48151 v4hi __builtin_arm_wshufh (v4hi, int); 48152 long long __builtin_arm_wslld (long long, long long); 48153 long long __builtin_arm_wslldi (long long, int); 48154 v4hi __builtin_arm_wsllh (v4hi, long long); 48155 v4hi __builtin_arm_wsllhi (v4hi, int); 48156 v2si __builtin_arm_wsllw (v2si, long long); 48157 v2si __builtin_arm_wsllwi (v2si, int); 48158 long long __builtin_arm_wsrad (long long, long long); 48159 long long __builtin_arm_wsradi (long long, int); 48160 v4hi __builtin_arm_wsrah (v4hi, long long); 48161 v4hi __builtin_arm_wsrahi (v4hi, int); 48162 v2si __builtin_arm_wsraw (v2si, long long); 48163 v2si __builtin_arm_wsrawi (v2si, int); 48164 long long __builtin_arm_wsrld (long long, long long); 48165 long long __builtin_arm_wsrldi (long long, int); 48166 v4hi __builtin_arm_wsrlh (v4hi, long long); 48167 v4hi __builtin_arm_wsrlhi (v4hi, int); 48168 v2si __builtin_arm_wsrlw (v2si, long long); 48169 v2si __builtin_arm_wsrlwi (v2si, int); 48170 v8qi __builtin_arm_wsubb (v8qi, v8qi); 48171 v8qi __builtin_arm_wsubbss (v8qi, v8qi); 48172 v8qi __builtin_arm_wsubbus (v8qi, v8qi); 48173 v4hi __builtin_arm_wsubh (v4hi, v4hi); 48174 v4hi __builtin_arm_wsubhss (v4hi, v4hi); 48175 v4hi __builtin_arm_wsubhus (v4hi, v4hi); 48176 v2si __builtin_arm_wsubw (v2si, v2si); 48177 v2si __builtin_arm_wsubwss (v2si, v2si); 48178 v2si __builtin_arm_wsubwus (v2si, v2si); 48179 v4hi __builtin_arm_wunpckehsb (v8qi); 48180 v2si __builtin_arm_wunpckehsh (v4hi); 48181 long long __builtin_arm_wunpckehsw (v2si); 48182 v4hi __builtin_arm_wunpckehub (v8qi); 48183 v2si __builtin_arm_wunpckehuh (v4hi); 48184 long long __builtin_arm_wunpckehuw (v2si); 48185 v4hi __builtin_arm_wunpckelsb (v8qi); 48186 v2si __builtin_arm_wunpckelsh (v4hi); 48187 long long __builtin_arm_wunpckelsw (v2si); 48188 v4hi __builtin_arm_wunpckelub (v8qi); 48189 v2si __builtin_arm_wunpckeluh (v4hi); 48190 long long __builtin_arm_wunpckeluw (v2si); 48191 v8qi __builtin_arm_wunpckihb (v8qi, v8qi); 48192 v4hi __builtin_arm_wunpckihh (v4hi, v4hi); 48193 v2si __builtin_arm_wunpckihw (v2si, v2si); 48194 v8qi __builtin_arm_wunpckilb (v8qi, v8qi); 48195 v4hi __builtin_arm_wunpckilh (v4hi, v4hi); 48196 v2si __builtin_arm_wunpckilw (v2si, v2si); 48197 long long __builtin_arm_wxor (long long, long long); 48198 long long __builtin_arm_wzero (); 48199 48200 48201File: gcc.info, Node: ARM C Language Extensions (ACLE), Next: ARM Floating Point Status and Control Intrinsics, Prev: ARM iWMMXt Built-in Functions, Up: Target Builtins 48202 482036.60.7 ARM C Language Extensions (ACLE) 48204--------------------------------------- 48205 48206GCC implements extensions for C as described in the ARM C Language 48207Extensions (ACLE) specification, which can be found at 48208<https://developer.arm.com/documentation/ihi0053/latest/>. 48209 48210 As a part of ACLE, GCC implements extensions for Advanced SIMD as 48211described in the ARM C Language Extensions Specification. The complete 48212list of Advanced SIMD intrinsics can be found at 48213<https://developer.arm.com/documentation/ihi0073/latest/>. The built-in 48214intrinsics for the Advanced SIMD extension are available when NEON is 48215enabled. 48216 48217 Currently, ARM and AArch64 back ends do not support ACLE 2.0 fully. 48218Both back ends support CRC32 intrinsics and the ARM back end supports 48219the Coprocessor intrinsics, all from 'arm_acle.h'. The ARM back end's 4822016-bit floating-point Advanced SIMD intrinsics currently comply to ACLE 48221v1.1. AArch64's back end does not have support for 16-bit floating 48222point Advanced SIMD intrinsics yet. 48223 48224 See *note ARM Options:: and *note AArch64 Options:: for more 48225information on the availability of extensions. 48226 48227 48228File: gcc.info, Node: ARM Floating Point Status and Control Intrinsics, Next: ARM ARMv8-M Security Extensions, Prev: ARM C Language Extensions (ACLE), Up: Target Builtins 48229 482306.60.8 ARM Floating Point Status and Control Intrinsics 48231------------------------------------------------------- 48232 48233These built-in functions are available for the ARM family of processors 48234with floating-point unit. 48235 48236 unsigned int __builtin_arm_get_fpscr (); 48237 void __builtin_arm_set_fpscr (unsigned int); 48238 48239 48240File: gcc.info, Node: ARM ARMv8-M Security Extensions, Next: AVR Built-in Functions, Prev: ARM Floating Point Status and Control Intrinsics, Up: Target Builtins 48241 482426.60.9 ARM ARMv8-M Security Extensions 48243-------------------------------------- 48244 48245GCC implements the ARMv8-M Security Extensions as described in the 48246ARMv8-M Security Extensions: Requirements on Development Tools 48247Engineering Specification, which can be found at 48248<https://developer.arm.com/documentation/ecm0359818/latest/>. 48249 48250 As part of the Security Extensions GCC implements two new function 48251attributes: 'cmse_nonsecure_entry' and 'cmse_nonsecure_call'. 48252 48253 As part of the Security Extensions GCC implements the intrinsics below. 48254FPTR is used here to mean any function pointer type. 48255 48256 cmse_address_info_t cmse_TT (void *); 48257 cmse_address_info_t cmse_TT_fptr (FPTR); 48258 cmse_address_info_t cmse_TTT (void *); 48259 cmse_address_info_t cmse_TTT_fptr (FPTR); 48260 cmse_address_info_t cmse_TTA (void *); 48261 cmse_address_info_t cmse_TTA_fptr (FPTR); 48262 cmse_address_info_t cmse_TTAT (void *); 48263 cmse_address_info_t cmse_TTAT_fptr (FPTR); 48264 void * cmse_check_address_range (void *, size_t, int); 48265 typeof(p) cmse_nsfptr_create (FPTR p); 48266 intptr_t cmse_is_nsfptr (FPTR); 48267 int cmse_nonsecure_caller (void); 48268 48269 48270File: gcc.info, Node: AVR Built-in Functions, Next: Blackfin Built-in Functions, Prev: ARM ARMv8-M Security Extensions, Up: Target Builtins 48271 482726.60.10 AVR Built-in Functions 48273------------------------------ 48274 48275For each built-in function for AVR, there is an equally named, uppercase 48276built-in macro defined. That way users can easily query if or if not a 48277specific built-in is implemented or not. For example, if 48278'__builtin_avr_nop' is available the macro '__BUILTIN_AVR_NOP' is 48279defined to '1' and undefined otherwise. 48280 48281'void __builtin_avr_nop (void)' 48282'void __builtin_avr_sei (void)' 48283'void __builtin_avr_cli (void)' 48284'void __builtin_avr_sleep (void)' 48285'void __builtin_avr_wdr (void)' 48286'unsigned char __builtin_avr_swap (unsigned char)' 48287'unsigned int __builtin_avr_fmul (unsigned char, unsigned char)' 48288'int __builtin_avr_fmuls (char, char)' 48289'int __builtin_avr_fmulsu (char, unsigned char)' 48290 These built-in functions map to the respective machine instruction, 48291 i.e. 'nop', 'sei', 'cli', 'sleep', 'wdr', 'swap', 'fmul', 'fmuls' 48292 resp. 'fmulsu'. The three 'fmul*' built-ins are implemented as 48293 library call if no hardware multiplier is available. 48294 48295'void __builtin_avr_delay_cycles (unsigned long ticks)' 48296 Delay execution for TICKS cycles. Note that this built-in does not 48297 take into account the effect of interrupts that might increase 48298 delay time. TICKS must be a compile-time integer constant; delays 48299 with a variable number of cycles are not supported. 48300 48301'char __builtin_avr_flash_segment (const __memx void*)' 48302 This built-in takes a byte address to the 24-bit *note address 48303 space: AVR Named Address Spaces. '__memx' and returns the number of 48304 the flash segment (the 64 KiB chunk) where the address points to. 48305 Counting starts at '0'. If the address does not point to flash 48306 memory, return '-1'. 48307 48308'uint8_t __builtin_avr_insert_bits (uint32_t map, uint8_t bits, uint8_t val)' 48309 Insert bits from BITS into VAL and return the resulting value. The 48310 nibbles of MAP determine how the insertion is performed: Let X be 48311 the N-th nibble of MAP 48312 1. If X is '0xf', then the N-th bit of VAL is returned unaltered. 48313 48314 2. If X is in the range 0...7, then the N-th result bit is set to 48315 the X-th bit of BITS 48316 48317 3. If X is in the range 8...'0xe', then the N-th result bit is 48318 undefined. 48319 48320 One typical use case for this built-in is adjusting input and 48321 output values to non-contiguous port layouts. Some examples: 48322 48323 // same as val, bits is unused 48324 __builtin_avr_insert_bits (0xffffffff, bits, val); 48325 48326 // same as bits, val is unused 48327 __builtin_avr_insert_bits (0x76543210, bits, val); 48328 48329 // same as rotating bits by 4 48330 __builtin_avr_insert_bits (0x32107654, bits, 0); 48331 48332 // high nibble of result is the high nibble of val 48333 // low nibble of result is the low nibble of bits 48334 __builtin_avr_insert_bits (0xffff3210, bits, val); 48335 48336 // reverse the bit order of bits 48337 __builtin_avr_insert_bits (0x01234567, bits, 0); 48338 48339'void __builtin_avr_nops (unsigned count)' 48340 Insert COUNT 'NOP' instructions. The number of instructions must 48341 be a compile-time integer constant. 48342 48343There are many more AVR-specific built-in functions that are used to 48344implement the ISO/IEC TR 18037 "Embedded C" fixed-point functions of 48345section 7.18a.6. You don't need to use these built-ins directly. 48346Instead, use the declarations as supplied by the 'stdfix.h' header with 48347GNU-C99: 48348 48349 #include <stdfix.h> 48350 48351 // Re-interpret the bit representation of unsigned 16-bit 48352 // integer UVAL as Q-format 0.16 value. 48353 unsigned fract get_bits (uint_ur_t uval) 48354 { 48355 return urbits (uval); 48356 } 48357 48358 48359File: gcc.info, Node: Blackfin Built-in Functions, Next: BPF Built-in Functions, Prev: AVR Built-in Functions, Up: Target Builtins 48360 483616.60.11 Blackfin Built-in Functions 48362----------------------------------- 48363 48364Currently, there are two Blackfin-specific built-in functions. These 48365are used for generating 'CSYNC' and 'SSYNC' machine insns without using 48366inline assembly; by using these built-in functions the compiler can 48367automatically add workarounds for hardware errata involving these 48368instructions. These functions are named as follows: 48369 48370 void __builtin_bfin_csync (void); 48371 void __builtin_bfin_ssync (void); 48372 48373 48374File: gcc.info, Node: BPF Built-in Functions, Next: FR-V Built-in Functions, Prev: Blackfin Built-in Functions, Up: Target Builtins 48375 483766.60.12 BPF Built-in Functions 48377------------------------------ 48378 48379The following built-in functions are available for eBPF targets. 48380 48381 -- Built-in Function: unsigned long long __builtin_bpf_load_byte 48382 (unsigned long long OFFSET) 48383 Load a byte from the 'struct sk_buff' packet data pointed by the 48384 register '%r6' and return it. 48385 48386 -- Built-in Function: unsigned long long __builtin_bpf_load_half 48387 (unsigned long long OFFSET) 48388 Load 16-bits from the 'struct sk_buff' packet data pointed by the 48389 register '%r6' and return it. 48390 48391 -- Built-in Function: unsigned long long __builtin_bpf_load_word 48392 (unsigned long long OFFSET) 48393 Load 32-bits from the 'struct sk_buff' packet data pointed by the 48394 register '%r6' and return it. 48395 48396 -- Built-in Function: void * __builtin_preserve_access_index (EXPR) 48397 BPF Compile Once-Run Everywhere (CO-RE) support. Instruct GCC to 48398 generate CO-RE relocation records for any accesses to aggregate 48399 data structures (struct, union, array types) in EXPR. This builtin 48400 is otherwise transparent, the return value is whatever EXPR 48401 evaluates to. It is also overloaded: EXPR may be of any type (not 48402 necessarily a pointer), the return type is the same. Has no effect 48403 if '-mco-re' is not in effect (either specified or implied). 48404 48405 48406File: gcc.info, Node: FR-V Built-in Functions, Next: MIPS DSP Built-in Functions, Prev: BPF Built-in Functions, Up: Target Builtins 48407 484086.60.13 FR-V Built-in Functions 48409------------------------------- 48410 48411GCC provides many FR-V-specific built-in functions. In general, these 48412functions are intended to be compatible with those described by 'FR-V 48413Family, Softune C/C++ Compiler Manual (V6), Fujitsu Semiconductor'. The 48414two exceptions are '__MDUNPACKH' and '__MBTOHE', the GCC forms of which 48415pass 128-bit values by pointer rather than by value. 48416 48417 Most of the functions are named after specific FR-V instructions. Such 48418functions are said to be "directly mapped" and are summarized here in 48419tabular form. 48420 48421* Menu: 48422 48423* Argument Types:: 48424* Directly-mapped Integer Functions:: 48425* Directly-mapped Media Functions:: 48426* Raw read/write Functions:: 48427* Other Built-in Functions:: 48428 48429 48430File: gcc.info, Node: Argument Types, Next: Directly-mapped Integer Functions, Up: FR-V Built-in Functions 48431 484326.60.13.1 Argument Types 48433........................ 48434 48435The arguments to the built-in functions can be divided into three 48436groups: register numbers, compile-time constants and run-time values. 48437In order to make this classification clear at a glance, the arguments 48438and return values are given the following pseudo types: 48439 48440Pseudo type Real C type Constant? Description 48441---------------------------------------------------------------------------- 48442'uh' 'unsigned short' No an unsigned halfword 48443'uw1' 'unsigned int' No an unsigned word 48444'sw1' 'int' No a signed word 48445'uw2' 'unsigned long long' No an unsigned doubleword 48446'sw2' 'long long' No a signed doubleword 48447'const' 'int' Yes an integer constant 48448'acc' 'int' Yes an ACC register number 48449'iacc' 'int' Yes an IACC register number 48450 48451 These pseudo types are not defined by GCC, they are simply a notational 48452convenience used in this manual. 48453 48454 Arguments of type 'uh', 'uw1', 'sw1', 'uw2' and 'sw2' are evaluated at 48455run time. They correspond to register operands in the underlying FR-V 48456instructions. 48457 48458 'const' arguments represent immediate operands in the underlying FR-V 48459instructions. They must be compile-time constants. 48460 48461 'acc' arguments are evaluated at compile time and specify the number of 48462an accumulator register. For example, an 'acc' argument of 2 selects 48463the ACC2 register. 48464 48465 'iacc' arguments are similar to 'acc' arguments but specify the number 48466of an IACC register. See *note Other Built-in Functions:: for more 48467details. 48468 48469 48470File: gcc.info, Node: Directly-mapped Integer Functions, Next: Directly-mapped Media Functions, Prev: Argument Types, Up: FR-V Built-in Functions 48471 484726.60.13.2 Directly-Mapped Integer Functions 48473........................................... 48474 48475The functions listed below map directly to FR-V I-type instructions. 48476 48477Function prototype Example usage Assembly output 48478--------------------------------------------------------------------------- 48479'sw1 __ADDSS (sw1, sw1)' 'C = __ADDSS (A, B)' 'ADDSS A,B,C' 48480'sw1 __SCAN (sw1, sw1)' 'C = __SCAN (A, B)' 'SCAN A,B,C' 48481'sw1 __SCUTSS (sw1)' 'B = __SCUTSS (A)' 'SCUTSS A,B' 48482'sw1 __SLASS (sw1, sw1)' 'C = __SLASS (A, B)' 'SLASS A,B,C' 48483'void __SMASS (sw1, sw1)' '__SMASS (A, B)' 'SMASS A,B' 48484'void __SMSSS (sw1, sw1)' '__SMSSS (A, B)' 'SMSSS A,B' 48485'void __SMU (sw1, sw1)' '__SMU (A, B)' 'SMU A,B' 48486'sw2 __SMUL (sw1, sw1)' 'C = __SMUL (A, B)' 'SMUL A,B,C' 48487'sw1 __SUBSS (sw1, sw1)' 'C = __SUBSS (A, B)' 'SUBSS A,B,C' 48488'uw2 __UMUL (uw1, uw1)' 'C = __UMUL (A, B)' 'UMUL A,B,C' 48489 48490 48491File: gcc.info, Node: Directly-mapped Media Functions, Next: Raw read/write Functions, Prev: Directly-mapped Integer Functions, Up: FR-V Built-in Functions 48492 484936.60.13.3 Directly-Mapped Media Functions 48494......................................... 48495 48496The functions listed below map directly to FR-V M-type instructions. 48497 48498Function prototype Example usage Assembly output 48499--------------------------------------------------------------------------- 48500'uw1 __MABSHS (sw1)' 'B = __MABSHS (A)' 'MABSHS A,B' 48501'void __MADDACCS (acc, acc)' '__MADDACCS (B, A)' 'MADDACCS A,B' 48502'sw1 __MADDHSS (sw1, sw1)' 'C = __MADDHSS (A, 'MADDHSS A,B,C' 48503 B)' 48504'uw1 __MADDHUS (uw1, uw1)' 'C = __MADDHUS (A, 'MADDHUS A,B,C' 48505 B)' 48506'uw1 __MAND (uw1, uw1)' 'C = __MAND (A, B)' 'MAND A,B,C' 48507'void __MASACCS (acc, acc)' '__MASACCS (B, A)' 'MASACCS A,B' 48508'uw1 __MAVEH (uw1, uw1)' 'C = __MAVEH (A, B)' 'MAVEH A,B,C' 48509'uw2 __MBTOH (uw1)' 'B = __MBTOH (A)' 'MBTOH A,B' 48510'void __MBTOHE (uw1 *, uw1)' '__MBTOHE (&B, A)' 'MBTOHE A,B' 48511'void __MCLRACC (acc)' '__MCLRACC (A)' 'MCLRACC A' 48512'void __MCLRACCA (void)' '__MCLRACCA ()' 'MCLRACCA' 48513'uw1 __Mcop1 (uw1, uw1)' 'C = __Mcop1 (A, B)' 'Mcop1 A,B,C' 48514'uw1 __Mcop2 (uw1, uw1)' 'C = __Mcop2 (A, B)' 'Mcop2 A,B,C' 48515'uw1 __MCPLHI (uw2, const)' 'C = __MCPLHI (A, B)' 'MCPLHI A,#B,C' 48516'uw1 __MCPLI (uw2, const)' 'C = __MCPLI (A, B)' 'MCPLI A,#B,C' 48517'void __MCPXIS (acc, sw1, '__MCPXIS (C, A, B)' 'MCPXIS A,B,C' 48518sw1)' 48519'void __MCPXIU (acc, uw1, '__MCPXIU (C, A, B)' 'MCPXIU A,B,C' 48520uw1)' 48521'void __MCPXRS (acc, sw1, '__MCPXRS (C, A, B)' 'MCPXRS A,B,C' 48522sw1)' 48523'void __MCPXRU (acc, uw1, '__MCPXRU (C, A, B)' 'MCPXRU A,B,C' 48524uw1)' 48525'uw1 __MCUT (acc, uw1)' 'C = __MCUT (A, B)' 'MCUT A,B,C' 48526'uw1 __MCUTSS (acc, sw1)' 'C = __MCUTSS (A, B)' 'MCUTSS A,B,C' 48527'void __MDADDACCS (acc, acc)' '__MDADDACCS (B, A)' 'MDADDACCS A,B' 48528'void __MDASACCS (acc, acc)' '__MDASACCS (B, A)' 'MDASACCS A,B' 48529'uw2 __MDCUTSSI (acc, const)' 'C = __MDCUTSSI (A, 'MDCUTSSI 48530 B)' A,#B,C' 48531'uw2 __MDPACKH (uw2, uw2)' 'C = __MDPACKH (A, 'MDPACKH A,B,C' 48532 B)' 48533'uw2 __MDROTLI (uw2, const)' 'C = __MDROTLI (A, 'MDROTLI 48534 B)' A,#B,C' 48535'void __MDSUBACCS (acc, acc)' '__MDSUBACCS (B, A)' 'MDSUBACCS A,B' 48536'void __MDUNPACKH (uw1 *, '__MDUNPACKH (&B, A)' 'MDUNPACKH A,B' 48537uw2)' 48538'uw2 __MEXPDHD (uw1, const)' 'C = __MEXPDHD (A, 'MEXPDHD 48539 B)' A,#B,C' 48540'uw1 __MEXPDHW (uw1, const)' 'C = __MEXPDHW (A, 'MEXPDHW 48541 B)' A,#B,C' 48542'uw1 __MHDSETH (uw1, const)' 'C = __MHDSETH (A, 'MHDSETH 48543 B)' A,#B,C' 48544'sw1 __MHDSETS (const)' 'B = __MHDSETS (A)' 'MHDSETS #A,B' 48545'uw1 __MHSETHIH (uw1, const)' 'B = __MHSETHIH (B, 'MHSETHIH #A,B' 48546 A)' 48547'sw1 __MHSETHIS (sw1, const)' 'B = __MHSETHIS (B, 'MHSETHIS #A,B' 48548 A)' 48549'uw1 __MHSETLOH (uw1, const)' 'B = __MHSETLOH (B, 'MHSETLOH #A,B' 48550 A)' 48551'sw1 __MHSETLOS (sw1, const)' 'B = __MHSETLOS (B, 'MHSETLOS #A,B' 48552 A)' 48553'uw1 __MHTOB (uw2)' 'B = __MHTOB (A)' 'MHTOB A,B' 48554'void __MMACHS (acc, sw1, '__MMACHS (C, A, B)' 'MMACHS A,B,C' 48555sw1)' 48556'void __MMACHU (acc, uw1, '__MMACHU (C, A, B)' 'MMACHU A,B,C' 48557uw1)' 48558'void __MMRDHS (acc, sw1, '__MMRDHS (C, A, B)' 'MMRDHS A,B,C' 48559sw1)' 48560'void __MMRDHU (acc, uw1, '__MMRDHU (C, A, B)' 'MMRDHU A,B,C' 48561uw1)' 48562'void __MMULHS (acc, sw1, '__MMULHS (C, A, B)' 'MMULHS A,B,C' 48563sw1)' 48564'void __MMULHU (acc, uw1, '__MMULHU (C, A, B)' 'MMULHU A,B,C' 48565uw1)' 48566'void __MMULXHS (acc, sw1, '__MMULXHS (C, A, B)' 'MMULXHS A,B,C' 48567sw1)' 48568'void __MMULXHU (acc, uw1, '__MMULXHU (C, A, B)' 'MMULXHU A,B,C' 48569uw1)' 48570'uw1 __MNOT (uw1)' 'B = __MNOT (A)' 'MNOT A,B' 48571'uw1 __MOR (uw1, uw1)' 'C = __MOR (A, B)' 'MOR A,B,C' 48572'uw1 __MPACKH (uh, uh)' 'C = __MPACKH (A, B)' 'MPACKH A,B,C' 48573'sw2 __MQADDHSS (sw2, sw2)' 'C = __MQADDHSS (A, 'MQADDHSS 48574 B)' A,B,C' 48575'uw2 __MQADDHUS (uw2, uw2)' 'C = __MQADDHUS (A, 'MQADDHUS 48576 B)' A,B,C' 48577'void __MQCPXIS (acc, sw2, '__MQCPXIS (C, A, B)' 'MQCPXIS A,B,C' 48578sw2)' 48579'void __MQCPXIU (acc, uw2, '__MQCPXIU (C, A, B)' 'MQCPXIU A,B,C' 48580uw2)' 48581'void __MQCPXRS (acc, sw2, '__MQCPXRS (C, A, B)' 'MQCPXRS A,B,C' 48582sw2)' 48583'void __MQCPXRU (acc, uw2, '__MQCPXRU (C, A, B)' 'MQCPXRU A,B,C' 48584uw2)' 48585'sw2 __MQLCLRHS (sw2, sw2)' 'C = __MQLCLRHS (A, 'MQLCLRHS 48586 B)' A,B,C' 48587'sw2 __MQLMTHS (sw2, sw2)' 'C = __MQLMTHS (A, 'MQLMTHS A,B,C' 48588 B)' 48589'void __MQMACHS (acc, sw2, '__MQMACHS (C, A, B)' 'MQMACHS A,B,C' 48590sw2)' 48591'void __MQMACHU (acc, uw2, '__MQMACHU (C, A, B)' 'MQMACHU A,B,C' 48592uw2)' 48593'void __MQMACXHS (acc, sw2, '__MQMACXHS (C, A, 'MQMACXHS 48594sw2)' B)' A,B,C' 48595'void __MQMULHS (acc, sw2, '__MQMULHS (C, A, B)' 'MQMULHS A,B,C' 48596sw2)' 48597'void __MQMULHU (acc, uw2, '__MQMULHU (C, A, B)' 'MQMULHU A,B,C' 48598uw2)' 48599'void __MQMULXHS (acc, sw2, '__MQMULXHS (C, A, 'MQMULXHS 48600sw2)' B)' A,B,C' 48601'void __MQMULXHU (acc, uw2, '__MQMULXHU (C, A, 'MQMULXHU 48602uw2)' B)' A,B,C' 48603'sw2 __MQSATHS (sw2, sw2)' 'C = __MQSATHS (A, 'MQSATHS A,B,C' 48604 B)' 48605'uw2 __MQSLLHI (uw2, int)' 'C = __MQSLLHI (A, 'MQSLLHI A,B,C' 48606 B)' 48607'sw2 __MQSRAHI (sw2, int)' 'C = __MQSRAHI (A, 'MQSRAHI A,B,C' 48608 B)' 48609'sw2 __MQSUBHSS (sw2, sw2)' 'C = __MQSUBHSS (A, 'MQSUBHSS 48610 B)' A,B,C' 48611'uw2 __MQSUBHUS (uw2, uw2)' 'C = __MQSUBHUS (A, 'MQSUBHUS 48612 B)' A,B,C' 48613'void __MQXMACHS (acc, sw2, '__MQXMACHS (C, A, 'MQXMACHS 48614sw2)' B)' A,B,C' 48615'void __MQXMACXHS (acc, sw2, '__MQXMACXHS (C, A, 'MQXMACXHS 48616sw2)' B)' A,B,C' 48617'uw1 __MRDACC (acc)' 'B = __MRDACC (A)' 'MRDACC A,B' 48618'uw1 __MRDACCG (acc)' 'B = __MRDACCG (A)' 'MRDACCG A,B' 48619'uw1 __MROTLI (uw1, const)' 'C = __MROTLI (A, B)' 'MROTLI A,#B,C' 48620'uw1 __MROTRI (uw1, const)' 'C = __MROTRI (A, B)' 'MROTRI A,#B,C' 48621'sw1 __MSATHS (sw1, sw1)' 'C = __MSATHS (A, B)' 'MSATHS A,B,C' 48622'uw1 __MSATHU (uw1, uw1)' 'C = __MSATHU (A, B)' 'MSATHU A,B,C' 48623'uw1 __MSLLHI (uw1, const)' 'C = __MSLLHI (A, B)' 'MSLLHI A,#B,C' 48624'sw1 __MSRAHI (sw1, const)' 'C = __MSRAHI (A, B)' 'MSRAHI A,#B,C' 48625'uw1 __MSRLHI (uw1, const)' 'C = __MSRLHI (A, B)' 'MSRLHI A,#B,C' 48626'void __MSUBACCS (acc, acc)' '__MSUBACCS (B, A)' 'MSUBACCS A,B' 48627'sw1 __MSUBHSS (sw1, sw1)' 'C = __MSUBHSS (A, 'MSUBHSS A,B,C' 48628 B)' 48629'uw1 __MSUBHUS (uw1, uw1)' 'C = __MSUBHUS (A, 'MSUBHUS A,B,C' 48630 B)' 48631'void __MTRAP (void)' '__MTRAP ()' 'MTRAP' 48632'uw2 __MUNPACKH (uw1)' 'B = __MUNPACKH (A)' 'MUNPACKH A,B' 48633'uw1 __MWCUT (uw2, uw1)' 'C = __MWCUT (A, B)' 'MWCUT A,B,C' 48634'void __MWTACC (acc, uw1)' '__MWTACC (B, A)' 'MWTACC A,B' 48635'void __MWTACCG (acc, uw1)' '__MWTACCG (B, A)' 'MWTACCG A,B' 48636'uw1 __MXOR (uw1, uw1)' 'C = __MXOR (A, B)' 'MXOR A,B,C' 48637 48638 48639File: gcc.info, Node: Raw read/write Functions, Next: Other Built-in Functions, Prev: Directly-mapped Media Functions, Up: FR-V Built-in Functions 48640 486416.60.13.4 Raw Read/Write Functions 48642.................................. 48643 48644This sections describes built-in functions related to read and write 48645instructions to access memory. These functions generate 'membar' 48646instructions to flush the I/O load and stores where appropriate, as 48647described in Fujitsu's manual described above. 48648 48649'unsigned char __builtin_read8 (void *DATA)' 48650'unsigned short __builtin_read16 (void *DATA)' 48651'unsigned long __builtin_read32 (void *DATA)' 48652'unsigned long long __builtin_read64 (void *DATA)' 48653 48654'void __builtin_write8 (void *DATA, unsigned char DATUM)' 48655'void __builtin_write16 (void *DATA, unsigned short DATUM)' 48656'void __builtin_write32 (void *DATA, unsigned long DATUM)' 48657'void __builtin_write64 (void *DATA, unsigned long long DATUM)' 48658 48659 48660File: gcc.info, Node: Other Built-in Functions, Prev: Raw read/write Functions, Up: FR-V Built-in Functions 48661 486626.60.13.5 Other Built-in Functions 48663.................................. 48664 48665This section describes built-in functions that are not named after a 48666specific FR-V instruction. 48667 48668'sw2 __IACCreadll (iacc REG)' 48669 Return the full 64-bit value of IACC0. The REG argument is 48670 reserved for future expansion and must be 0. 48671 48672'sw1 __IACCreadl (iacc REG)' 48673 Return the value of IACC0H if REG is 0 and IACC0L if REG is 1. 48674 Other values of REG are rejected as invalid. 48675 48676'void __IACCsetll (iacc REG, sw2 X)' 48677 Set the full 64-bit value of IACC0 to X. The REG argument is 48678 reserved for future expansion and must be 0. 48679 48680'void __IACCsetl (iacc REG, sw1 X)' 48681 Set IACC0H to X if REG is 0 and IACC0L to X if REG is 1. Other 48682 values of REG are rejected as invalid. 48683 48684'void __data_prefetch0 (const void *X)' 48685 Use the 'dcpl' instruction to load the contents of address X into 48686 the data cache. 48687 48688'void __data_prefetch (const void *X)' 48689 Use the 'nldub' instruction to load the contents of address X into 48690 the data cache. The instruction is issued in slot I1. 48691 48692 48693File: gcc.info, Node: MIPS DSP Built-in Functions, Next: MIPS Paired-Single Support, Prev: FR-V Built-in Functions, Up: Target Builtins 48694 486956.60.14 MIPS DSP Built-in Functions 48696----------------------------------- 48697 48698The MIPS DSP Application-Specific Extension (ASE) includes new 48699instructions that are designed to improve the performance of DSP and 48700media applications. It provides instructions that operate on packed 487018-bit/16-bit integer data, Q7, Q15 and Q31 fractional data. 48702 48703 GCC supports MIPS DSP operations using both the generic vector 48704extensions (*note Vector Extensions::) and a collection of MIPS-specific 48705built-in functions. Both kinds of support are enabled by the '-mdsp' 48706command-line option. 48707 48708 Revision 2 of the ASE was introduced in the second half of 2006. This 48709revision adds extra instructions to the original ASE, but is otherwise 48710backwards-compatible with it. You can select revision 2 using the 48711command-line option '-mdspr2'; this option implies '-mdsp'. 48712 48713 The SCOUNT and POS bits of the DSP control register are global. The 48714WRDSP, EXTPDP, EXTPDPV and MTHLIP instructions modify the SCOUNT and POS 48715bits. During optimization, the compiler does not delete these 48716instructions and it does not delete calls to functions containing these 48717instructions. 48718 48719 At present, GCC only provides support for operations on 32-bit vectors. 48720The vector type associated with 8-bit integer data is usually called 48721'v4i8', the vector type associated with Q7 is usually called 'v4q7', the 48722vector type associated with 16-bit integer data is usually called 48723'v2i16', and the vector type associated with Q15 is usually called 48724'v2q15'. They can be defined in C as follows: 48725 48726 typedef signed char v4i8 __attribute__ ((vector_size(4))); 48727 typedef signed char v4q7 __attribute__ ((vector_size(4))); 48728 typedef short v2i16 __attribute__ ((vector_size(4))); 48729 typedef short v2q15 __attribute__ ((vector_size(4))); 48730 48731 'v4i8', 'v4q7', 'v2i16' and 'v2q15' values are initialized in the same 48732way as aggregates. For example: 48733 48734 v4i8 a = {1, 2, 3, 4}; 48735 v4i8 b; 48736 b = (v4i8) {5, 6, 7, 8}; 48737 48738 v2q15 c = {0x0fcb, 0x3a75}; 48739 v2q15 d; 48740 d = (v2q15) {0.1234 * 0x1.0p15, 0.4567 * 0x1.0p15}; 48741 48742 _Note:_ The CPU's endianness determines the order in which values are 48743packed. On little-endian targets, the first value is the least 48744significant and the last value is the most significant. The opposite 48745order applies to big-endian targets. For example, the code above sets 48746the lowest byte of 'a' to '1' on little-endian targets and '4' on 48747big-endian targets. 48748 48749 _Note:_ Q7, Q15 and Q31 values must be initialized with their integer 48750representation. As shown in this example, the integer representation of 48751a Q7 value can be obtained by multiplying the fractional value by 48752'0x1.0p7'. The equivalent for Q15 values is to multiply by '0x1.0p15'. 48753The equivalent for Q31 values is to multiply by '0x1.0p31'. 48754 48755 The table below lists the 'v4i8' and 'v2q15' operations for which 48756hardware support exists. 'a' and 'b' are 'v4i8' values, and 'c' and 'd' 48757are 'v2q15' values. 48758 48759C code MIPS instruction 48760-------------------------------------------------------------------------- 48761'a + b' 'addu.qb' 48762'c + d' 'addq.ph' 48763'a - b' 'subu.qb' 48764'c - d' 'subq.ph' 48765 48766 The table below lists the 'v2i16' operation for which hardware support 48767exists for the DSP ASE REV 2. 'e' and 'f' are 'v2i16' values. 48768 48769C code MIPS instruction 48770-------------------------------------------------------------------------- 48771'e * f' 'mul.ph' 48772 48773 It is easier to describe the DSP built-in functions if we first define 48774the following types: 48775 48776 typedef int q31; 48777 typedef int i32; 48778 typedef unsigned int ui32; 48779 typedef long long a64; 48780 48781 'q31' and 'i32' are actually the same as 'int', but we use 'q31' to 48782indicate a Q31 fractional value and 'i32' to indicate a 32-bit integer 48783value. Similarly, 'a64' is the same as 'long long', but we use 'a64' to 48784indicate values that are placed in one of the four DSP accumulators 48785('$ac0', '$ac1', '$ac2' or '$ac3'). 48786 48787 Also, some built-in functions prefer or require immediate numbers as 48788parameters, because the corresponding DSP instructions accept both 48789immediate numbers and register operands, or accept immediate numbers 48790only. The immediate parameters are listed as follows. 48791 48792 imm0_3: 0 to 3. 48793 imm0_7: 0 to 7. 48794 imm0_15: 0 to 15. 48795 imm0_31: 0 to 31. 48796 imm0_63: 0 to 63. 48797 imm0_255: 0 to 255. 48798 imm_n32_31: -32 to 31. 48799 imm_n512_511: -512 to 511. 48800 48801 The following built-in functions map directly to a particular MIPS DSP 48802instruction. Please refer to the architecture specification for details 48803on what each instruction does. 48804 48805 v2q15 __builtin_mips_addq_ph (v2q15, v2q15); 48806 v2q15 __builtin_mips_addq_s_ph (v2q15, v2q15); 48807 q31 __builtin_mips_addq_s_w (q31, q31); 48808 v4i8 __builtin_mips_addu_qb (v4i8, v4i8); 48809 v4i8 __builtin_mips_addu_s_qb (v4i8, v4i8); 48810 v2q15 __builtin_mips_subq_ph (v2q15, v2q15); 48811 v2q15 __builtin_mips_subq_s_ph (v2q15, v2q15); 48812 q31 __builtin_mips_subq_s_w (q31, q31); 48813 v4i8 __builtin_mips_subu_qb (v4i8, v4i8); 48814 v4i8 __builtin_mips_subu_s_qb (v4i8, v4i8); 48815 i32 __builtin_mips_addsc (i32, i32); 48816 i32 __builtin_mips_addwc (i32, i32); 48817 i32 __builtin_mips_modsub (i32, i32); 48818 i32 __builtin_mips_raddu_w_qb (v4i8); 48819 v2q15 __builtin_mips_absq_s_ph (v2q15); 48820 q31 __builtin_mips_absq_s_w (q31); 48821 v4i8 __builtin_mips_precrq_qb_ph (v2q15, v2q15); 48822 v2q15 __builtin_mips_precrq_ph_w (q31, q31); 48823 v2q15 __builtin_mips_precrq_rs_ph_w (q31, q31); 48824 v4i8 __builtin_mips_precrqu_s_qb_ph (v2q15, v2q15); 48825 q31 __builtin_mips_preceq_w_phl (v2q15); 48826 q31 __builtin_mips_preceq_w_phr (v2q15); 48827 v2q15 __builtin_mips_precequ_ph_qbl (v4i8); 48828 v2q15 __builtin_mips_precequ_ph_qbr (v4i8); 48829 v2q15 __builtin_mips_precequ_ph_qbla (v4i8); 48830 v2q15 __builtin_mips_precequ_ph_qbra (v4i8); 48831 v2q15 __builtin_mips_preceu_ph_qbl (v4i8); 48832 v2q15 __builtin_mips_preceu_ph_qbr (v4i8); 48833 v2q15 __builtin_mips_preceu_ph_qbla (v4i8); 48834 v2q15 __builtin_mips_preceu_ph_qbra (v4i8); 48835 v4i8 __builtin_mips_shll_qb (v4i8, imm0_7); 48836 v4i8 __builtin_mips_shll_qb (v4i8, i32); 48837 v2q15 __builtin_mips_shll_ph (v2q15, imm0_15); 48838 v2q15 __builtin_mips_shll_ph (v2q15, i32); 48839 v2q15 __builtin_mips_shll_s_ph (v2q15, imm0_15); 48840 v2q15 __builtin_mips_shll_s_ph (v2q15, i32); 48841 q31 __builtin_mips_shll_s_w (q31, imm0_31); 48842 q31 __builtin_mips_shll_s_w (q31, i32); 48843 v4i8 __builtin_mips_shrl_qb (v4i8, imm0_7); 48844 v4i8 __builtin_mips_shrl_qb (v4i8, i32); 48845 v2q15 __builtin_mips_shra_ph (v2q15, imm0_15); 48846 v2q15 __builtin_mips_shra_ph (v2q15, i32); 48847 v2q15 __builtin_mips_shra_r_ph (v2q15, imm0_15); 48848 v2q15 __builtin_mips_shra_r_ph (v2q15, i32); 48849 q31 __builtin_mips_shra_r_w (q31, imm0_31); 48850 q31 __builtin_mips_shra_r_w (q31, i32); 48851 v2q15 __builtin_mips_muleu_s_ph_qbl (v4i8, v2q15); 48852 v2q15 __builtin_mips_muleu_s_ph_qbr (v4i8, v2q15); 48853 v2q15 __builtin_mips_mulq_rs_ph (v2q15, v2q15); 48854 q31 __builtin_mips_muleq_s_w_phl (v2q15, v2q15); 48855 q31 __builtin_mips_muleq_s_w_phr (v2q15, v2q15); 48856 a64 __builtin_mips_dpau_h_qbl (a64, v4i8, v4i8); 48857 a64 __builtin_mips_dpau_h_qbr (a64, v4i8, v4i8); 48858 a64 __builtin_mips_dpsu_h_qbl (a64, v4i8, v4i8); 48859 a64 __builtin_mips_dpsu_h_qbr (a64, v4i8, v4i8); 48860 a64 __builtin_mips_dpaq_s_w_ph (a64, v2q15, v2q15); 48861 a64 __builtin_mips_dpaq_sa_l_w (a64, q31, q31); 48862 a64 __builtin_mips_dpsq_s_w_ph (a64, v2q15, v2q15); 48863 a64 __builtin_mips_dpsq_sa_l_w (a64, q31, q31); 48864 a64 __builtin_mips_mulsaq_s_w_ph (a64, v2q15, v2q15); 48865 a64 __builtin_mips_maq_s_w_phl (a64, v2q15, v2q15); 48866 a64 __builtin_mips_maq_s_w_phr (a64, v2q15, v2q15); 48867 a64 __builtin_mips_maq_sa_w_phl (a64, v2q15, v2q15); 48868 a64 __builtin_mips_maq_sa_w_phr (a64, v2q15, v2q15); 48869 i32 __builtin_mips_bitrev (i32); 48870 i32 __builtin_mips_insv (i32, i32); 48871 v4i8 __builtin_mips_repl_qb (imm0_255); 48872 v4i8 __builtin_mips_repl_qb (i32); 48873 v2q15 __builtin_mips_repl_ph (imm_n512_511); 48874 v2q15 __builtin_mips_repl_ph (i32); 48875 void __builtin_mips_cmpu_eq_qb (v4i8, v4i8); 48876 void __builtin_mips_cmpu_lt_qb (v4i8, v4i8); 48877 void __builtin_mips_cmpu_le_qb (v4i8, v4i8); 48878 i32 __builtin_mips_cmpgu_eq_qb (v4i8, v4i8); 48879 i32 __builtin_mips_cmpgu_lt_qb (v4i8, v4i8); 48880 i32 __builtin_mips_cmpgu_le_qb (v4i8, v4i8); 48881 void __builtin_mips_cmp_eq_ph (v2q15, v2q15); 48882 void __builtin_mips_cmp_lt_ph (v2q15, v2q15); 48883 void __builtin_mips_cmp_le_ph (v2q15, v2q15); 48884 v4i8 __builtin_mips_pick_qb (v4i8, v4i8); 48885 v2q15 __builtin_mips_pick_ph (v2q15, v2q15); 48886 v2q15 __builtin_mips_packrl_ph (v2q15, v2q15); 48887 i32 __builtin_mips_extr_w (a64, imm0_31); 48888 i32 __builtin_mips_extr_w (a64, i32); 48889 i32 __builtin_mips_extr_r_w (a64, imm0_31); 48890 i32 __builtin_mips_extr_s_h (a64, i32); 48891 i32 __builtin_mips_extr_rs_w (a64, imm0_31); 48892 i32 __builtin_mips_extr_rs_w (a64, i32); 48893 i32 __builtin_mips_extr_s_h (a64, imm0_31); 48894 i32 __builtin_mips_extr_r_w (a64, i32); 48895 i32 __builtin_mips_extp (a64, imm0_31); 48896 i32 __builtin_mips_extp (a64, i32); 48897 i32 __builtin_mips_extpdp (a64, imm0_31); 48898 i32 __builtin_mips_extpdp (a64, i32); 48899 a64 __builtin_mips_shilo (a64, imm_n32_31); 48900 a64 __builtin_mips_shilo (a64, i32); 48901 a64 __builtin_mips_mthlip (a64, i32); 48902 void __builtin_mips_wrdsp (i32, imm0_63); 48903 i32 __builtin_mips_rddsp (imm0_63); 48904 i32 __builtin_mips_lbux (void *, i32); 48905 i32 __builtin_mips_lhx (void *, i32); 48906 i32 __builtin_mips_lwx (void *, i32); 48907 a64 __builtin_mips_ldx (void *, i32); /* MIPS64 only */ 48908 i32 __builtin_mips_bposge32 (void); 48909 a64 __builtin_mips_madd (a64, i32, i32); 48910 a64 __builtin_mips_maddu (a64, ui32, ui32); 48911 a64 __builtin_mips_msub (a64, i32, i32); 48912 a64 __builtin_mips_msubu (a64, ui32, ui32); 48913 a64 __builtin_mips_mult (i32, i32); 48914 a64 __builtin_mips_multu (ui32, ui32); 48915 48916 The following built-in functions map directly to a particular MIPS DSP 48917REV 2 instruction. Please refer to the architecture specification for 48918details on what each instruction does. 48919 48920 v4q7 __builtin_mips_absq_s_qb (v4q7); 48921 v2i16 __builtin_mips_addu_ph (v2i16, v2i16); 48922 v2i16 __builtin_mips_addu_s_ph (v2i16, v2i16); 48923 v4i8 __builtin_mips_adduh_qb (v4i8, v4i8); 48924 v4i8 __builtin_mips_adduh_r_qb (v4i8, v4i8); 48925 i32 __builtin_mips_append (i32, i32, imm0_31); 48926 i32 __builtin_mips_balign (i32, i32, imm0_3); 48927 i32 __builtin_mips_cmpgdu_eq_qb (v4i8, v4i8); 48928 i32 __builtin_mips_cmpgdu_lt_qb (v4i8, v4i8); 48929 i32 __builtin_mips_cmpgdu_le_qb (v4i8, v4i8); 48930 a64 __builtin_mips_dpa_w_ph (a64, v2i16, v2i16); 48931 a64 __builtin_mips_dps_w_ph (a64, v2i16, v2i16); 48932 v2i16 __builtin_mips_mul_ph (v2i16, v2i16); 48933 v2i16 __builtin_mips_mul_s_ph (v2i16, v2i16); 48934 q31 __builtin_mips_mulq_rs_w (q31, q31); 48935 v2q15 __builtin_mips_mulq_s_ph (v2q15, v2q15); 48936 q31 __builtin_mips_mulq_s_w (q31, q31); 48937 a64 __builtin_mips_mulsa_w_ph (a64, v2i16, v2i16); 48938 v4i8 __builtin_mips_precr_qb_ph (v2i16, v2i16); 48939 v2i16 __builtin_mips_precr_sra_ph_w (i32, i32, imm0_31); 48940 v2i16 __builtin_mips_precr_sra_r_ph_w (i32, i32, imm0_31); 48941 i32 __builtin_mips_prepend (i32, i32, imm0_31); 48942 v4i8 __builtin_mips_shra_qb (v4i8, imm0_7); 48943 v4i8 __builtin_mips_shra_r_qb (v4i8, imm0_7); 48944 v4i8 __builtin_mips_shra_qb (v4i8, i32); 48945 v4i8 __builtin_mips_shra_r_qb (v4i8, i32); 48946 v2i16 __builtin_mips_shrl_ph (v2i16, imm0_15); 48947 v2i16 __builtin_mips_shrl_ph (v2i16, i32); 48948 v2i16 __builtin_mips_subu_ph (v2i16, v2i16); 48949 v2i16 __builtin_mips_subu_s_ph (v2i16, v2i16); 48950 v4i8 __builtin_mips_subuh_qb (v4i8, v4i8); 48951 v4i8 __builtin_mips_subuh_r_qb (v4i8, v4i8); 48952 v2q15 __builtin_mips_addqh_ph (v2q15, v2q15); 48953 v2q15 __builtin_mips_addqh_r_ph (v2q15, v2q15); 48954 q31 __builtin_mips_addqh_w (q31, q31); 48955 q31 __builtin_mips_addqh_r_w (q31, q31); 48956 v2q15 __builtin_mips_subqh_ph (v2q15, v2q15); 48957 v2q15 __builtin_mips_subqh_r_ph (v2q15, v2q15); 48958 q31 __builtin_mips_subqh_w (q31, q31); 48959 q31 __builtin_mips_subqh_r_w (q31, q31); 48960 a64 __builtin_mips_dpax_w_ph (a64, v2i16, v2i16); 48961 a64 __builtin_mips_dpsx_w_ph (a64, v2i16, v2i16); 48962 a64 __builtin_mips_dpaqx_s_w_ph (a64, v2q15, v2q15); 48963 a64 __builtin_mips_dpaqx_sa_w_ph (a64, v2q15, v2q15); 48964 a64 __builtin_mips_dpsqx_s_w_ph (a64, v2q15, v2q15); 48965 a64 __builtin_mips_dpsqx_sa_w_ph (a64, v2q15, v2q15); 48966 48967 48968File: gcc.info, Node: MIPS Paired-Single Support, Next: MIPS Loongson Built-in Functions, Prev: MIPS DSP Built-in Functions, Up: Target Builtins 48969 489706.60.15 MIPS Paired-Single Support 48971---------------------------------- 48972 48973The MIPS64 architecture includes a number of instructions that operate 48974on pairs of single-precision floating-point values. Each pair is packed 48975into a 64-bit floating-point register, with one element being designated 48976the "upper half" and the other being designated the "lower half". 48977 48978 GCC supports paired-single operations using both the generic vector 48979extensions (*note Vector Extensions::) and a collection of MIPS-specific 48980built-in functions. Both kinds of support are enabled by the 48981'-mpaired-single' command-line option. 48982 48983 The vector type associated with paired-single values is usually called 48984'v2sf'. It can be defined in C as follows: 48985 48986 typedef float v2sf __attribute__ ((vector_size (8))); 48987 48988 'v2sf' values are initialized in the same way as aggregates. For 48989example: 48990 48991 v2sf a = {1.5, 9.1}; 48992 v2sf b; 48993 float e, f; 48994 b = (v2sf) {e, f}; 48995 48996 _Note:_ The CPU's endianness determines which value is stored in the 48997upper half of a register and which value is stored in the lower half. 48998On little-endian targets, the first value is the lower one and the 48999second value is the upper one. The opposite order applies to big-endian 49000targets. For example, the code above sets the lower half of 'a' to 49001'1.5' on little-endian targets and '9.1' on big-endian targets. 49002 49003 49004File: gcc.info, Node: MIPS Loongson Built-in Functions, Next: MIPS SIMD Architecture (MSA) Support, Prev: MIPS Paired-Single Support, Up: Target Builtins 49005 490066.60.16 MIPS Loongson Built-in Functions 49007---------------------------------------- 49008 49009GCC provides intrinsics to access the SIMD instructions provided by the 49010ST Microelectronics Loongson-2E and -2F processors. These intrinsics, 49011available after inclusion of the 'loongson.h' header file, operate on 49012the following 64-bit vector types: 49013 49014 * 'uint8x8_t', a vector of eight unsigned 8-bit integers; 49015 * 'uint16x4_t', a vector of four unsigned 16-bit integers; 49016 * 'uint32x2_t', a vector of two unsigned 32-bit integers; 49017 * 'int8x8_t', a vector of eight signed 8-bit integers; 49018 * 'int16x4_t', a vector of four signed 16-bit integers; 49019 * 'int32x2_t', a vector of two signed 32-bit integers. 49020 49021 The intrinsics provided are listed below; each is named after the 49022machine instruction to which it corresponds, with suffixes added as 49023appropriate to distinguish intrinsics that expand to the same machine 49024instruction yet have different argument types. Refer to the 49025architecture documentation for a description of the functionality of 49026each instruction. 49027 49028 int16x4_t packsswh (int32x2_t s, int32x2_t t); 49029 int8x8_t packsshb (int16x4_t s, int16x4_t t); 49030 uint8x8_t packushb (uint16x4_t s, uint16x4_t t); 49031 uint32x2_t paddw_u (uint32x2_t s, uint32x2_t t); 49032 uint16x4_t paddh_u (uint16x4_t s, uint16x4_t t); 49033 uint8x8_t paddb_u (uint8x8_t s, uint8x8_t t); 49034 int32x2_t paddw_s (int32x2_t s, int32x2_t t); 49035 int16x4_t paddh_s (int16x4_t s, int16x4_t t); 49036 int8x8_t paddb_s (int8x8_t s, int8x8_t t); 49037 uint64_t paddd_u (uint64_t s, uint64_t t); 49038 int64_t paddd_s (int64_t s, int64_t t); 49039 int16x4_t paddsh (int16x4_t s, int16x4_t t); 49040 int8x8_t paddsb (int8x8_t s, int8x8_t t); 49041 uint16x4_t paddush (uint16x4_t s, uint16x4_t t); 49042 uint8x8_t paddusb (uint8x8_t s, uint8x8_t t); 49043 uint64_t pandn_ud (uint64_t s, uint64_t t); 49044 uint32x2_t pandn_uw (uint32x2_t s, uint32x2_t t); 49045 uint16x4_t pandn_uh (uint16x4_t s, uint16x4_t t); 49046 uint8x8_t pandn_ub (uint8x8_t s, uint8x8_t t); 49047 int64_t pandn_sd (int64_t s, int64_t t); 49048 int32x2_t pandn_sw (int32x2_t s, int32x2_t t); 49049 int16x4_t pandn_sh (int16x4_t s, int16x4_t t); 49050 int8x8_t pandn_sb (int8x8_t s, int8x8_t t); 49051 uint16x4_t pavgh (uint16x4_t s, uint16x4_t t); 49052 uint8x8_t pavgb (uint8x8_t s, uint8x8_t t); 49053 uint32x2_t pcmpeqw_u (uint32x2_t s, uint32x2_t t); 49054 uint16x4_t pcmpeqh_u (uint16x4_t s, uint16x4_t t); 49055 uint8x8_t pcmpeqb_u (uint8x8_t s, uint8x8_t t); 49056 int32x2_t pcmpeqw_s (int32x2_t s, int32x2_t t); 49057 int16x4_t pcmpeqh_s (int16x4_t s, int16x4_t t); 49058 int8x8_t pcmpeqb_s (int8x8_t s, int8x8_t t); 49059 uint32x2_t pcmpgtw_u (uint32x2_t s, uint32x2_t t); 49060 uint16x4_t pcmpgth_u (uint16x4_t s, uint16x4_t t); 49061 uint8x8_t pcmpgtb_u (uint8x8_t s, uint8x8_t t); 49062 int32x2_t pcmpgtw_s (int32x2_t s, int32x2_t t); 49063 int16x4_t pcmpgth_s (int16x4_t s, int16x4_t t); 49064 int8x8_t pcmpgtb_s (int8x8_t s, int8x8_t t); 49065 uint16x4_t pextrh_u (uint16x4_t s, int field); 49066 int16x4_t pextrh_s (int16x4_t s, int field); 49067 uint16x4_t pinsrh_0_u (uint16x4_t s, uint16x4_t t); 49068 uint16x4_t pinsrh_1_u (uint16x4_t s, uint16x4_t t); 49069 uint16x4_t pinsrh_2_u (uint16x4_t s, uint16x4_t t); 49070 uint16x4_t pinsrh_3_u (uint16x4_t s, uint16x4_t t); 49071 int16x4_t pinsrh_0_s (int16x4_t s, int16x4_t t); 49072 int16x4_t pinsrh_1_s (int16x4_t s, int16x4_t t); 49073 int16x4_t pinsrh_2_s (int16x4_t s, int16x4_t t); 49074 int16x4_t pinsrh_3_s (int16x4_t s, int16x4_t t); 49075 int32x2_t pmaddhw (int16x4_t s, int16x4_t t); 49076 int16x4_t pmaxsh (int16x4_t s, int16x4_t t); 49077 uint8x8_t pmaxub (uint8x8_t s, uint8x8_t t); 49078 int16x4_t pminsh (int16x4_t s, int16x4_t t); 49079 uint8x8_t pminub (uint8x8_t s, uint8x8_t t); 49080 uint8x8_t pmovmskb_u (uint8x8_t s); 49081 int8x8_t pmovmskb_s (int8x8_t s); 49082 uint16x4_t pmulhuh (uint16x4_t s, uint16x4_t t); 49083 int16x4_t pmulhh (int16x4_t s, int16x4_t t); 49084 int16x4_t pmullh (int16x4_t s, int16x4_t t); 49085 int64_t pmuluw (uint32x2_t s, uint32x2_t t); 49086 uint8x8_t pasubub (uint8x8_t s, uint8x8_t t); 49087 uint16x4_t biadd (uint8x8_t s); 49088 uint16x4_t psadbh (uint8x8_t s, uint8x8_t t); 49089 uint16x4_t pshufh_u (uint16x4_t dest, uint16x4_t s, uint8_t order); 49090 int16x4_t pshufh_s (int16x4_t dest, int16x4_t s, uint8_t order); 49091 uint16x4_t psllh_u (uint16x4_t s, uint8_t amount); 49092 int16x4_t psllh_s (int16x4_t s, uint8_t amount); 49093 uint32x2_t psllw_u (uint32x2_t s, uint8_t amount); 49094 int32x2_t psllw_s (int32x2_t s, uint8_t amount); 49095 uint16x4_t psrlh_u (uint16x4_t s, uint8_t amount); 49096 int16x4_t psrlh_s (int16x4_t s, uint8_t amount); 49097 uint32x2_t psrlw_u (uint32x2_t s, uint8_t amount); 49098 int32x2_t psrlw_s (int32x2_t s, uint8_t amount); 49099 uint16x4_t psrah_u (uint16x4_t s, uint8_t amount); 49100 int16x4_t psrah_s (int16x4_t s, uint8_t amount); 49101 uint32x2_t psraw_u (uint32x2_t s, uint8_t amount); 49102 int32x2_t psraw_s (int32x2_t s, uint8_t amount); 49103 uint32x2_t psubw_u (uint32x2_t s, uint32x2_t t); 49104 uint16x4_t psubh_u (uint16x4_t s, uint16x4_t t); 49105 uint8x8_t psubb_u (uint8x8_t s, uint8x8_t t); 49106 int32x2_t psubw_s (int32x2_t s, int32x2_t t); 49107 int16x4_t psubh_s (int16x4_t s, int16x4_t t); 49108 int8x8_t psubb_s (int8x8_t s, int8x8_t t); 49109 uint64_t psubd_u (uint64_t s, uint64_t t); 49110 int64_t psubd_s (int64_t s, int64_t t); 49111 int16x4_t psubsh (int16x4_t s, int16x4_t t); 49112 int8x8_t psubsb (int8x8_t s, int8x8_t t); 49113 uint16x4_t psubush (uint16x4_t s, uint16x4_t t); 49114 uint8x8_t psubusb (uint8x8_t s, uint8x8_t t); 49115 uint32x2_t punpckhwd_u (uint32x2_t s, uint32x2_t t); 49116 uint16x4_t punpckhhw_u (uint16x4_t s, uint16x4_t t); 49117 uint8x8_t punpckhbh_u (uint8x8_t s, uint8x8_t t); 49118 int32x2_t punpckhwd_s (int32x2_t s, int32x2_t t); 49119 int16x4_t punpckhhw_s (int16x4_t s, int16x4_t t); 49120 int8x8_t punpckhbh_s (int8x8_t s, int8x8_t t); 49121 uint32x2_t punpcklwd_u (uint32x2_t s, uint32x2_t t); 49122 uint16x4_t punpcklhw_u (uint16x4_t s, uint16x4_t t); 49123 uint8x8_t punpcklbh_u (uint8x8_t s, uint8x8_t t); 49124 int32x2_t punpcklwd_s (int32x2_t s, int32x2_t t); 49125 int16x4_t punpcklhw_s (int16x4_t s, int16x4_t t); 49126 int8x8_t punpcklbh_s (int8x8_t s, int8x8_t t); 49127 49128* Menu: 49129 49130* Paired-Single Arithmetic:: 49131* Paired-Single Built-in Functions:: 49132* MIPS-3D Built-in Functions:: 49133 49134 49135File: gcc.info, Node: Paired-Single Arithmetic, Next: Paired-Single Built-in Functions, Up: MIPS Loongson Built-in Functions 49136 491376.60.16.1 Paired-Single Arithmetic 49138.................................. 49139 49140The table below lists the 'v2sf' operations for which hardware support 49141exists. 'a', 'b' and 'c' are 'v2sf' values and 'x' is an integral 49142value. 49143 49144C code MIPS instruction 49145-------------------------------------------------------------------------- 49146'a + b' 'add.ps' 49147'a - b' 'sub.ps' 49148'-a' 'neg.ps' 49149'a * b' 'mul.ps' 49150'a * b + c' 'madd.ps' 49151'a * b - c' 'msub.ps' 49152'-(a * b + c)' 'nmadd.ps' 49153'-(a * b - c)' 'nmsub.ps' 49154'x ? a : b' 'movn.ps'/'movz.ps' 49155 49156 Note that the multiply-accumulate instructions can be disabled using 49157the command-line option '-mno-fused-madd'. 49158 49159 49160File: gcc.info, Node: Paired-Single Built-in Functions, Next: MIPS-3D Built-in Functions, Prev: Paired-Single Arithmetic, Up: MIPS Loongson Built-in Functions 49161 491626.60.16.2 Paired-Single Built-in Functions 49163.......................................... 49164 49165The following paired-single functions map directly to a particular MIPS 49166instruction. Please refer to the architecture specification for details 49167on what each instruction does. 49168 49169'v2sf __builtin_mips_pll_ps (v2sf, v2sf)' 49170 Pair lower lower ('pll.ps'). 49171 49172'v2sf __builtin_mips_pul_ps (v2sf, v2sf)' 49173 Pair upper lower ('pul.ps'). 49174 49175'v2sf __builtin_mips_plu_ps (v2sf, v2sf)' 49176 Pair lower upper ('plu.ps'). 49177 49178'v2sf __builtin_mips_puu_ps (v2sf, v2sf)' 49179 Pair upper upper ('puu.ps'). 49180 49181'v2sf __builtin_mips_cvt_ps_s (float, float)' 49182 Convert pair to paired single ('cvt.ps.s'). 49183 49184'float __builtin_mips_cvt_s_pl (v2sf)' 49185 Convert pair lower to single ('cvt.s.pl'). 49186 49187'float __builtin_mips_cvt_s_pu (v2sf)' 49188 Convert pair upper to single ('cvt.s.pu'). 49189 49190'v2sf __builtin_mips_abs_ps (v2sf)' 49191 Absolute value ('abs.ps'). 49192 49193'v2sf __builtin_mips_alnv_ps (v2sf, v2sf, int)' 49194 Align variable ('alnv.ps'). 49195 49196 _Note:_ The value of the third parameter must be 0 or 4 modulo 8, 49197 otherwise the result is unpredictable. Please read the instruction 49198 description for details. 49199 49200 The following multi-instruction functions are also available. In each 49201case, COND can be any of the 16 floating-point conditions: 'f', 'un', 49202'eq', 'ueq', 'olt', 'ult', 'ole', 'ule', 'sf', 'ngle', 'seq', 'ngl', 49203'lt', 'nge', 'le' or 'ngt'. 49204 49205'v2sf __builtin_mips_movt_c_COND_ps (v2sf A, v2sf B, v2sf C, v2sf D)' 49206'v2sf __builtin_mips_movf_c_COND_ps (v2sf A, v2sf B, v2sf C, v2sf D)' 49207 Conditional move based on floating-point comparison ('c.COND.ps', 49208 'movt.ps'/'movf.ps'). 49209 49210 The 'movt' functions return the value X computed by: 49211 49212 c.COND.ps CC,A,B 49213 mov.ps X,C 49214 movt.ps X,D,CC 49215 49216 The 'movf' functions are similar but use 'movf.ps' instead of 49217 'movt.ps'. 49218 49219'int __builtin_mips_upper_c_COND_ps (v2sf A, v2sf B)' 49220'int __builtin_mips_lower_c_COND_ps (v2sf A, v2sf B)' 49221 Comparison of two paired-single values ('c.COND.ps', 49222 'bc1t'/'bc1f'). 49223 49224 These functions compare A and B using 'c.COND.ps' and return either 49225 the upper or lower half of the result. For example: 49226 49227 v2sf a, b; 49228 if (__builtin_mips_upper_c_eq_ps (a, b)) 49229 upper_halves_are_equal (); 49230 else 49231 upper_halves_are_unequal (); 49232 49233 if (__builtin_mips_lower_c_eq_ps (a, b)) 49234 lower_halves_are_equal (); 49235 else 49236 lower_halves_are_unequal (); 49237 49238 49239File: gcc.info, Node: MIPS-3D Built-in Functions, Prev: Paired-Single Built-in Functions, Up: MIPS Loongson Built-in Functions 49240 492416.60.16.3 MIPS-3D Built-in Functions 49242.................................... 49243 49244The MIPS-3D Application-Specific Extension (ASE) includes additional 49245paired-single instructions that are designed to improve the performance 49246of 3D graphics operations. Support for these instructions is controlled 49247by the '-mips3d' command-line option. 49248 49249 The functions listed below map directly to a particular MIPS-3D 49250instruction. Please refer to the architecture specification for more 49251details on what each instruction does. 49252 49253'v2sf __builtin_mips_addr_ps (v2sf, v2sf)' 49254 Reduction add ('addr.ps'). 49255 49256'v2sf __builtin_mips_mulr_ps (v2sf, v2sf)' 49257 Reduction multiply ('mulr.ps'). 49258 49259'v2sf __builtin_mips_cvt_pw_ps (v2sf)' 49260 Convert paired single to paired word ('cvt.pw.ps'). 49261 49262'v2sf __builtin_mips_cvt_ps_pw (v2sf)' 49263 Convert paired word to paired single ('cvt.ps.pw'). 49264 49265'float __builtin_mips_recip1_s (float)' 49266'double __builtin_mips_recip1_d (double)' 49267'v2sf __builtin_mips_recip1_ps (v2sf)' 49268 Reduced-precision reciprocal (sequence step 1) ('recip1.FMT'). 49269 49270'float __builtin_mips_recip2_s (float, float)' 49271'double __builtin_mips_recip2_d (double, double)' 49272'v2sf __builtin_mips_recip2_ps (v2sf, v2sf)' 49273 Reduced-precision reciprocal (sequence step 2) ('recip2.FMT'). 49274 49275'float __builtin_mips_rsqrt1_s (float)' 49276'double __builtin_mips_rsqrt1_d (double)' 49277'v2sf __builtin_mips_rsqrt1_ps (v2sf)' 49278 Reduced-precision reciprocal square root (sequence step 1) 49279 ('rsqrt1.FMT'). 49280 49281'float __builtin_mips_rsqrt2_s (float, float)' 49282'double __builtin_mips_rsqrt2_d (double, double)' 49283'v2sf __builtin_mips_rsqrt2_ps (v2sf, v2sf)' 49284 Reduced-precision reciprocal square root (sequence step 2) 49285 ('rsqrt2.FMT'). 49286 49287 The following multi-instruction functions are also available. In each 49288case, COND can be any of the 16 floating-point conditions: 'f', 'un', 49289'eq', 'ueq', 'olt', 'ult', 'ole', 'ule', 'sf', 'ngle', 'seq', 'ngl', 49290'lt', 'nge', 'le' or 'ngt'. 49291 49292'int __builtin_mips_cabs_COND_s (float A, float B)' 49293'int __builtin_mips_cabs_COND_d (double A, double B)' 49294 Absolute comparison of two scalar values ('cabs.COND.FMT', 49295 'bc1t'/'bc1f'). 49296 49297 These functions compare A and B using 'cabs.COND.s' or 49298 'cabs.COND.d' and return the result as a boolean value. For 49299 example: 49300 49301 float a, b; 49302 if (__builtin_mips_cabs_eq_s (a, b)) 49303 true (); 49304 else 49305 false (); 49306 49307'int __builtin_mips_upper_cabs_COND_ps (v2sf A, v2sf B)' 49308'int __builtin_mips_lower_cabs_COND_ps (v2sf A, v2sf B)' 49309 Absolute comparison of two paired-single values ('cabs.COND.ps', 49310 'bc1t'/'bc1f'). 49311 49312 These functions compare A and B using 'cabs.COND.ps' and return 49313 either the upper or lower half of the result. For example: 49314 49315 v2sf a, b; 49316 if (__builtin_mips_upper_cabs_eq_ps (a, b)) 49317 upper_halves_are_equal (); 49318 else 49319 upper_halves_are_unequal (); 49320 49321 if (__builtin_mips_lower_cabs_eq_ps (a, b)) 49322 lower_halves_are_equal (); 49323 else 49324 lower_halves_are_unequal (); 49325 49326'v2sf __builtin_mips_movt_cabs_COND_ps (v2sf A, v2sf B, v2sf C, v2sf D)' 49327'v2sf __builtin_mips_movf_cabs_COND_ps (v2sf A, v2sf B, v2sf C, v2sf D)' 49328 Conditional move based on absolute comparison ('cabs.COND.ps', 49329 'movt.ps'/'movf.ps'). 49330 49331 The 'movt' functions return the value X computed by: 49332 49333 cabs.COND.ps CC,A,B 49334 mov.ps X,C 49335 movt.ps X,D,CC 49336 49337 The 'movf' functions are similar but use 'movf.ps' instead of 49338 'movt.ps'. 49339 49340'int __builtin_mips_any_c_COND_ps (v2sf A, v2sf B)' 49341'int __builtin_mips_all_c_COND_ps (v2sf A, v2sf B)' 49342'int __builtin_mips_any_cabs_COND_ps (v2sf A, v2sf B)' 49343'int __builtin_mips_all_cabs_COND_ps (v2sf A, v2sf B)' 49344 Comparison of two paired-single values ('c.COND.ps'/'cabs.COND.ps', 49345 'bc1any2t'/'bc1any2f'). 49346 49347 These functions compare A and B using 'c.COND.ps' or 49348 'cabs.COND.ps'. The 'any' forms return 'true' if either result is 49349 'true' and the 'all' forms return 'true' if both results are 49350 'true'. For example: 49351 49352 v2sf a, b; 49353 if (__builtin_mips_any_c_eq_ps (a, b)) 49354 one_is_true (); 49355 else 49356 both_are_false (); 49357 49358 if (__builtin_mips_all_c_eq_ps (a, b)) 49359 both_are_true (); 49360 else 49361 one_is_false (); 49362 49363'int __builtin_mips_any_c_COND_4s (v2sf A, v2sf B, v2sf C, v2sf D)' 49364'int __builtin_mips_all_c_COND_4s (v2sf A, v2sf B, v2sf C, v2sf D)' 49365'int __builtin_mips_any_cabs_COND_4s (v2sf A, v2sf B, v2sf C, v2sf D)' 49366'int __builtin_mips_all_cabs_COND_4s (v2sf A, v2sf B, v2sf C, v2sf D)' 49367 Comparison of four paired-single values 49368 ('c.COND.ps'/'cabs.COND.ps', 'bc1any4t'/'bc1any4f'). 49369 49370 These functions use 'c.COND.ps' or 'cabs.COND.ps' to compare A with 49371 B and to compare C with D. The 'any' forms return 'true' if any of 49372 the four results are 'true' and the 'all' forms return 'true' if 49373 all four results are 'true'. For example: 49374 49375 v2sf a, b, c, d; 49376 if (__builtin_mips_any_c_eq_4s (a, b, c, d)) 49377 some_are_true (); 49378 else 49379 all_are_false (); 49380 49381 if (__builtin_mips_all_c_eq_4s (a, b, c, d)) 49382 all_are_true (); 49383 else 49384 some_are_false (); 49385 49386 49387File: gcc.info, Node: MIPS SIMD Architecture (MSA) Support, Next: Other MIPS Built-in Functions, Prev: MIPS Loongson Built-in Functions, Up: Target Builtins 49388 493896.60.17 MIPS SIMD Architecture (MSA) Support 49390-------------------------------------------- 49391 49392* Menu: 49393 49394* MIPS SIMD Architecture Built-in Functions:: 49395 49396GCC provides intrinsics to access the SIMD instructions provided by the 49397MSA MIPS SIMD Architecture. The interface is made available by 49398including '<msa.h>' and using '-mmsa -mhard-float -mfp64 -mnan=2008'. 49399For each '__builtin_msa_*', there is a shortened name of the intrinsic, 49400'__msa_*'. 49401 49402 MSA implements 128-bit wide vector registers, operating on 8-, 16-, 32- 49403and 64-bit integer, 16- and 32-bit fixed-point, or 32- and 64-bit 49404floating point data elements. The following vectors typedefs are 49405included in 'msa.h': 49406 * 'v16i8', a vector of sixteen signed 8-bit integers; 49407 * 'v16u8', a vector of sixteen unsigned 8-bit integers; 49408 * 'v8i16', a vector of eight signed 16-bit integers; 49409 * 'v8u16', a vector of eight unsigned 16-bit integers; 49410 * 'v4i32', a vector of four signed 32-bit integers; 49411 * 'v4u32', a vector of four unsigned 32-bit integers; 49412 * 'v2i64', a vector of two signed 64-bit integers; 49413 * 'v2u64', a vector of two unsigned 64-bit integers; 49414 * 'v4f32', a vector of four 32-bit floats; 49415 * 'v2f64', a vector of two 64-bit doubles. 49416 49417 Instructions and corresponding built-ins may have additional 49418restrictions and/or input/output values manipulated: 49419 * 'imm0_1', an integer literal in range 0 to 1; 49420 * 'imm0_3', an integer literal in range 0 to 3; 49421 * 'imm0_7', an integer literal in range 0 to 7; 49422 * 'imm0_15', an integer literal in range 0 to 15; 49423 * 'imm0_31', an integer literal in range 0 to 31; 49424 * 'imm0_63', an integer literal in range 0 to 63; 49425 * 'imm0_255', an integer literal in range 0 to 255; 49426 * 'imm_n16_15', an integer literal in range -16 to 15; 49427 * 'imm_n512_511', an integer literal in range -512 to 511; 49428 * 'imm_n1024_1022', an integer literal in range -512 to 511 left 49429 shifted by 1 bit, i.e., -1024, -1022, ..., 1020, 1022; 49430 * 'imm_n2048_2044', an integer literal in range -512 to 511 left 49431 shifted by 2 bits, i.e., -2048, -2044, ..., 2040, 2044; 49432 * 'imm_n4096_4088', an integer literal in range -512 to 511 left 49433 shifted by 3 bits, i.e., -4096, -4088, ..., 4080, 4088; 49434 * 'imm1_4', an integer literal in range 1 to 4; 49435 * 'i32, i64, u32, u64, f32, f64', defined as follows: 49436 49437 { 49438 typedef int i32; 49439 #if __LONG_MAX__ == __LONG_LONG_MAX__ 49440 typedef long i64; 49441 #else 49442 typedef long long i64; 49443 #endif 49444 49445 typedef unsigned int u32; 49446 #if __LONG_MAX__ == __LONG_LONG_MAX__ 49447 typedef unsigned long u64; 49448 #else 49449 typedef unsigned long long u64; 49450 #endif 49451 49452 typedef double f64; 49453 typedef float f32; 49454 } 49455 49456 49457File: gcc.info, Node: MIPS SIMD Architecture Built-in Functions, Up: MIPS SIMD Architecture (MSA) Support 49458 494596.60.17.1 MIPS SIMD Architecture Built-in Functions 49460................................................... 49461 49462The intrinsics provided are listed below; each is named after the 49463machine instruction. 49464 49465 v16i8 __builtin_msa_add_a_b (v16i8, v16i8); 49466 v8i16 __builtin_msa_add_a_h (v8i16, v8i16); 49467 v4i32 __builtin_msa_add_a_w (v4i32, v4i32); 49468 v2i64 __builtin_msa_add_a_d (v2i64, v2i64); 49469 49470 v16i8 __builtin_msa_adds_a_b (v16i8, v16i8); 49471 v8i16 __builtin_msa_adds_a_h (v8i16, v8i16); 49472 v4i32 __builtin_msa_adds_a_w (v4i32, v4i32); 49473 v2i64 __builtin_msa_adds_a_d (v2i64, v2i64); 49474 49475 v16i8 __builtin_msa_adds_s_b (v16i8, v16i8); 49476 v8i16 __builtin_msa_adds_s_h (v8i16, v8i16); 49477 v4i32 __builtin_msa_adds_s_w (v4i32, v4i32); 49478 v2i64 __builtin_msa_adds_s_d (v2i64, v2i64); 49479 49480 v16u8 __builtin_msa_adds_u_b (v16u8, v16u8); 49481 v8u16 __builtin_msa_adds_u_h (v8u16, v8u16); 49482 v4u32 __builtin_msa_adds_u_w (v4u32, v4u32); 49483 v2u64 __builtin_msa_adds_u_d (v2u64, v2u64); 49484 49485 v16i8 __builtin_msa_addv_b (v16i8, v16i8); 49486 v8i16 __builtin_msa_addv_h (v8i16, v8i16); 49487 v4i32 __builtin_msa_addv_w (v4i32, v4i32); 49488 v2i64 __builtin_msa_addv_d (v2i64, v2i64); 49489 49490 v16i8 __builtin_msa_addvi_b (v16i8, imm0_31); 49491 v8i16 __builtin_msa_addvi_h (v8i16, imm0_31); 49492 v4i32 __builtin_msa_addvi_w (v4i32, imm0_31); 49493 v2i64 __builtin_msa_addvi_d (v2i64, imm0_31); 49494 49495 v16u8 __builtin_msa_and_v (v16u8, v16u8); 49496 49497 v16u8 __builtin_msa_andi_b (v16u8, imm0_255); 49498 49499 v16i8 __builtin_msa_asub_s_b (v16i8, v16i8); 49500 v8i16 __builtin_msa_asub_s_h (v8i16, v8i16); 49501 v4i32 __builtin_msa_asub_s_w (v4i32, v4i32); 49502 v2i64 __builtin_msa_asub_s_d (v2i64, v2i64); 49503 49504 v16u8 __builtin_msa_asub_u_b (v16u8, v16u8); 49505 v8u16 __builtin_msa_asub_u_h (v8u16, v8u16); 49506 v4u32 __builtin_msa_asub_u_w (v4u32, v4u32); 49507 v2u64 __builtin_msa_asub_u_d (v2u64, v2u64); 49508 49509 v16i8 __builtin_msa_ave_s_b (v16i8, v16i8); 49510 v8i16 __builtin_msa_ave_s_h (v8i16, v8i16); 49511 v4i32 __builtin_msa_ave_s_w (v4i32, v4i32); 49512 v2i64 __builtin_msa_ave_s_d (v2i64, v2i64); 49513 49514 v16u8 __builtin_msa_ave_u_b (v16u8, v16u8); 49515 v8u16 __builtin_msa_ave_u_h (v8u16, v8u16); 49516 v4u32 __builtin_msa_ave_u_w (v4u32, v4u32); 49517 v2u64 __builtin_msa_ave_u_d (v2u64, v2u64); 49518 49519 v16i8 __builtin_msa_aver_s_b (v16i8, v16i8); 49520 v8i16 __builtin_msa_aver_s_h (v8i16, v8i16); 49521 v4i32 __builtin_msa_aver_s_w (v4i32, v4i32); 49522 v2i64 __builtin_msa_aver_s_d (v2i64, v2i64); 49523 49524 v16u8 __builtin_msa_aver_u_b (v16u8, v16u8); 49525 v8u16 __builtin_msa_aver_u_h (v8u16, v8u16); 49526 v4u32 __builtin_msa_aver_u_w (v4u32, v4u32); 49527 v2u64 __builtin_msa_aver_u_d (v2u64, v2u64); 49528 49529 v16u8 __builtin_msa_bclr_b (v16u8, v16u8); 49530 v8u16 __builtin_msa_bclr_h (v8u16, v8u16); 49531 v4u32 __builtin_msa_bclr_w (v4u32, v4u32); 49532 v2u64 __builtin_msa_bclr_d (v2u64, v2u64); 49533 49534 v16u8 __builtin_msa_bclri_b (v16u8, imm0_7); 49535 v8u16 __builtin_msa_bclri_h (v8u16, imm0_15); 49536 v4u32 __builtin_msa_bclri_w (v4u32, imm0_31); 49537 v2u64 __builtin_msa_bclri_d (v2u64, imm0_63); 49538 49539 v16u8 __builtin_msa_binsl_b (v16u8, v16u8, v16u8); 49540 v8u16 __builtin_msa_binsl_h (v8u16, v8u16, v8u16); 49541 v4u32 __builtin_msa_binsl_w (v4u32, v4u32, v4u32); 49542 v2u64 __builtin_msa_binsl_d (v2u64, v2u64, v2u64); 49543 49544 v16u8 __builtin_msa_binsli_b (v16u8, v16u8, imm0_7); 49545 v8u16 __builtin_msa_binsli_h (v8u16, v8u16, imm0_15); 49546 v4u32 __builtin_msa_binsli_w (v4u32, v4u32, imm0_31); 49547 v2u64 __builtin_msa_binsli_d (v2u64, v2u64, imm0_63); 49548 49549 v16u8 __builtin_msa_binsr_b (v16u8, v16u8, v16u8); 49550 v8u16 __builtin_msa_binsr_h (v8u16, v8u16, v8u16); 49551 v4u32 __builtin_msa_binsr_w (v4u32, v4u32, v4u32); 49552 v2u64 __builtin_msa_binsr_d (v2u64, v2u64, v2u64); 49553 49554 v16u8 __builtin_msa_binsri_b (v16u8, v16u8, imm0_7); 49555 v8u16 __builtin_msa_binsri_h (v8u16, v8u16, imm0_15); 49556 v4u32 __builtin_msa_binsri_w (v4u32, v4u32, imm0_31); 49557 v2u64 __builtin_msa_binsri_d (v2u64, v2u64, imm0_63); 49558 49559 v16u8 __builtin_msa_bmnz_v (v16u8, v16u8, v16u8); 49560 49561 v16u8 __builtin_msa_bmnzi_b (v16u8, v16u8, imm0_255); 49562 49563 v16u8 __builtin_msa_bmz_v (v16u8, v16u8, v16u8); 49564 49565 v16u8 __builtin_msa_bmzi_b (v16u8, v16u8, imm0_255); 49566 49567 v16u8 __builtin_msa_bneg_b (v16u8, v16u8); 49568 v8u16 __builtin_msa_bneg_h (v8u16, v8u16); 49569 v4u32 __builtin_msa_bneg_w (v4u32, v4u32); 49570 v2u64 __builtin_msa_bneg_d (v2u64, v2u64); 49571 49572 v16u8 __builtin_msa_bnegi_b (v16u8, imm0_7); 49573 v8u16 __builtin_msa_bnegi_h (v8u16, imm0_15); 49574 v4u32 __builtin_msa_bnegi_w (v4u32, imm0_31); 49575 v2u64 __builtin_msa_bnegi_d (v2u64, imm0_63); 49576 49577 i32 __builtin_msa_bnz_b (v16u8); 49578 i32 __builtin_msa_bnz_h (v8u16); 49579 i32 __builtin_msa_bnz_w (v4u32); 49580 i32 __builtin_msa_bnz_d (v2u64); 49581 49582 i32 __builtin_msa_bnz_v (v16u8); 49583 49584 v16u8 __builtin_msa_bsel_v (v16u8, v16u8, v16u8); 49585 49586 v16u8 __builtin_msa_bseli_b (v16u8, v16u8, imm0_255); 49587 49588 v16u8 __builtin_msa_bset_b (v16u8, v16u8); 49589 v8u16 __builtin_msa_bset_h (v8u16, v8u16); 49590 v4u32 __builtin_msa_bset_w (v4u32, v4u32); 49591 v2u64 __builtin_msa_bset_d (v2u64, v2u64); 49592 49593 v16u8 __builtin_msa_bseti_b (v16u8, imm0_7); 49594 v8u16 __builtin_msa_bseti_h (v8u16, imm0_15); 49595 v4u32 __builtin_msa_bseti_w (v4u32, imm0_31); 49596 v2u64 __builtin_msa_bseti_d (v2u64, imm0_63); 49597 49598 i32 __builtin_msa_bz_b (v16u8); 49599 i32 __builtin_msa_bz_h (v8u16); 49600 i32 __builtin_msa_bz_w (v4u32); 49601 i32 __builtin_msa_bz_d (v2u64); 49602 49603 i32 __builtin_msa_bz_v (v16u8); 49604 49605 v16i8 __builtin_msa_ceq_b (v16i8, v16i8); 49606 v8i16 __builtin_msa_ceq_h (v8i16, v8i16); 49607 v4i32 __builtin_msa_ceq_w (v4i32, v4i32); 49608 v2i64 __builtin_msa_ceq_d (v2i64, v2i64); 49609 49610 v16i8 __builtin_msa_ceqi_b (v16i8, imm_n16_15); 49611 v8i16 __builtin_msa_ceqi_h (v8i16, imm_n16_15); 49612 v4i32 __builtin_msa_ceqi_w (v4i32, imm_n16_15); 49613 v2i64 __builtin_msa_ceqi_d (v2i64, imm_n16_15); 49614 49615 i32 __builtin_msa_cfcmsa (imm0_31); 49616 49617 v16i8 __builtin_msa_cle_s_b (v16i8, v16i8); 49618 v8i16 __builtin_msa_cle_s_h (v8i16, v8i16); 49619 v4i32 __builtin_msa_cle_s_w (v4i32, v4i32); 49620 v2i64 __builtin_msa_cle_s_d (v2i64, v2i64); 49621 49622 v16i8 __builtin_msa_cle_u_b (v16u8, v16u8); 49623 v8i16 __builtin_msa_cle_u_h (v8u16, v8u16); 49624 v4i32 __builtin_msa_cle_u_w (v4u32, v4u32); 49625 v2i64 __builtin_msa_cle_u_d (v2u64, v2u64); 49626 49627 v16i8 __builtin_msa_clei_s_b (v16i8, imm_n16_15); 49628 v8i16 __builtin_msa_clei_s_h (v8i16, imm_n16_15); 49629 v4i32 __builtin_msa_clei_s_w (v4i32, imm_n16_15); 49630 v2i64 __builtin_msa_clei_s_d (v2i64, imm_n16_15); 49631 49632 v16i8 __builtin_msa_clei_u_b (v16u8, imm0_31); 49633 v8i16 __builtin_msa_clei_u_h (v8u16, imm0_31); 49634 v4i32 __builtin_msa_clei_u_w (v4u32, imm0_31); 49635 v2i64 __builtin_msa_clei_u_d (v2u64, imm0_31); 49636 49637 v16i8 __builtin_msa_clt_s_b (v16i8, v16i8); 49638 v8i16 __builtin_msa_clt_s_h (v8i16, v8i16); 49639 v4i32 __builtin_msa_clt_s_w (v4i32, v4i32); 49640 v2i64 __builtin_msa_clt_s_d (v2i64, v2i64); 49641 49642 v16i8 __builtin_msa_clt_u_b (v16u8, v16u8); 49643 v8i16 __builtin_msa_clt_u_h (v8u16, v8u16); 49644 v4i32 __builtin_msa_clt_u_w (v4u32, v4u32); 49645 v2i64 __builtin_msa_clt_u_d (v2u64, v2u64); 49646 49647 v16i8 __builtin_msa_clti_s_b (v16i8, imm_n16_15); 49648 v8i16 __builtin_msa_clti_s_h (v8i16, imm_n16_15); 49649 v4i32 __builtin_msa_clti_s_w (v4i32, imm_n16_15); 49650 v2i64 __builtin_msa_clti_s_d (v2i64, imm_n16_15); 49651 49652 v16i8 __builtin_msa_clti_u_b (v16u8, imm0_31); 49653 v8i16 __builtin_msa_clti_u_h (v8u16, imm0_31); 49654 v4i32 __builtin_msa_clti_u_w (v4u32, imm0_31); 49655 v2i64 __builtin_msa_clti_u_d (v2u64, imm0_31); 49656 49657 i32 __builtin_msa_copy_s_b (v16i8, imm0_15); 49658 i32 __builtin_msa_copy_s_h (v8i16, imm0_7); 49659 i32 __builtin_msa_copy_s_w (v4i32, imm0_3); 49660 i64 __builtin_msa_copy_s_d (v2i64, imm0_1); 49661 49662 u32 __builtin_msa_copy_u_b (v16i8, imm0_15); 49663 u32 __builtin_msa_copy_u_h (v8i16, imm0_7); 49664 u32 __builtin_msa_copy_u_w (v4i32, imm0_3); 49665 u64 __builtin_msa_copy_u_d (v2i64, imm0_1); 49666 49667 void __builtin_msa_ctcmsa (imm0_31, i32); 49668 49669 v16i8 __builtin_msa_div_s_b (v16i8, v16i8); 49670 v8i16 __builtin_msa_div_s_h (v8i16, v8i16); 49671 v4i32 __builtin_msa_div_s_w (v4i32, v4i32); 49672 v2i64 __builtin_msa_div_s_d (v2i64, v2i64); 49673 49674 v16u8 __builtin_msa_div_u_b (v16u8, v16u8); 49675 v8u16 __builtin_msa_div_u_h (v8u16, v8u16); 49676 v4u32 __builtin_msa_div_u_w (v4u32, v4u32); 49677 v2u64 __builtin_msa_div_u_d (v2u64, v2u64); 49678 49679 v8i16 __builtin_msa_dotp_s_h (v16i8, v16i8); 49680 v4i32 __builtin_msa_dotp_s_w (v8i16, v8i16); 49681 v2i64 __builtin_msa_dotp_s_d (v4i32, v4i32); 49682 49683 v8u16 __builtin_msa_dotp_u_h (v16u8, v16u8); 49684 v4u32 __builtin_msa_dotp_u_w (v8u16, v8u16); 49685 v2u64 __builtin_msa_dotp_u_d (v4u32, v4u32); 49686 49687 v8i16 __builtin_msa_dpadd_s_h (v8i16, v16i8, v16i8); 49688 v4i32 __builtin_msa_dpadd_s_w (v4i32, v8i16, v8i16); 49689 v2i64 __builtin_msa_dpadd_s_d (v2i64, v4i32, v4i32); 49690 49691 v8u16 __builtin_msa_dpadd_u_h (v8u16, v16u8, v16u8); 49692 v4u32 __builtin_msa_dpadd_u_w (v4u32, v8u16, v8u16); 49693 v2u64 __builtin_msa_dpadd_u_d (v2u64, v4u32, v4u32); 49694 49695 v8i16 __builtin_msa_dpsub_s_h (v8i16, v16i8, v16i8); 49696 v4i32 __builtin_msa_dpsub_s_w (v4i32, v8i16, v8i16); 49697 v2i64 __builtin_msa_dpsub_s_d (v2i64, v4i32, v4i32); 49698 49699 v8i16 __builtin_msa_dpsub_u_h (v8i16, v16u8, v16u8); 49700 v4i32 __builtin_msa_dpsub_u_w (v4i32, v8u16, v8u16); 49701 v2i64 __builtin_msa_dpsub_u_d (v2i64, v4u32, v4u32); 49702 49703 v4f32 __builtin_msa_fadd_w (v4f32, v4f32); 49704 v2f64 __builtin_msa_fadd_d (v2f64, v2f64); 49705 49706 v4i32 __builtin_msa_fcaf_w (v4f32, v4f32); 49707 v2i64 __builtin_msa_fcaf_d (v2f64, v2f64); 49708 49709 v4i32 __builtin_msa_fceq_w (v4f32, v4f32); 49710 v2i64 __builtin_msa_fceq_d (v2f64, v2f64); 49711 49712 v4i32 __builtin_msa_fclass_w (v4f32); 49713 v2i64 __builtin_msa_fclass_d (v2f64); 49714 49715 v4i32 __builtin_msa_fcle_w (v4f32, v4f32); 49716 v2i64 __builtin_msa_fcle_d (v2f64, v2f64); 49717 49718 v4i32 __builtin_msa_fclt_w (v4f32, v4f32); 49719 v2i64 __builtin_msa_fclt_d (v2f64, v2f64); 49720 49721 v4i32 __builtin_msa_fcne_w (v4f32, v4f32); 49722 v2i64 __builtin_msa_fcne_d (v2f64, v2f64); 49723 49724 v4i32 __builtin_msa_fcor_w (v4f32, v4f32); 49725 v2i64 __builtin_msa_fcor_d (v2f64, v2f64); 49726 49727 v4i32 __builtin_msa_fcueq_w (v4f32, v4f32); 49728 v2i64 __builtin_msa_fcueq_d (v2f64, v2f64); 49729 49730 v4i32 __builtin_msa_fcule_w (v4f32, v4f32); 49731 v2i64 __builtin_msa_fcule_d (v2f64, v2f64); 49732 49733 v4i32 __builtin_msa_fcult_w (v4f32, v4f32); 49734 v2i64 __builtin_msa_fcult_d (v2f64, v2f64); 49735 49736 v4i32 __builtin_msa_fcun_w (v4f32, v4f32); 49737 v2i64 __builtin_msa_fcun_d (v2f64, v2f64); 49738 49739 v4i32 __builtin_msa_fcune_w (v4f32, v4f32); 49740 v2i64 __builtin_msa_fcune_d (v2f64, v2f64); 49741 49742 v4f32 __builtin_msa_fdiv_w (v4f32, v4f32); 49743 v2f64 __builtin_msa_fdiv_d (v2f64, v2f64); 49744 49745 v8i16 __builtin_msa_fexdo_h (v4f32, v4f32); 49746 v4f32 __builtin_msa_fexdo_w (v2f64, v2f64); 49747 49748 v4f32 __builtin_msa_fexp2_w (v4f32, v4i32); 49749 v2f64 __builtin_msa_fexp2_d (v2f64, v2i64); 49750 49751 v4f32 __builtin_msa_fexupl_w (v8i16); 49752 v2f64 __builtin_msa_fexupl_d (v4f32); 49753 49754 v4f32 __builtin_msa_fexupr_w (v8i16); 49755 v2f64 __builtin_msa_fexupr_d (v4f32); 49756 49757 v4f32 __builtin_msa_ffint_s_w (v4i32); 49758 v2f64 __builtin_msa_ffint_s_d (v2i64); 49759 49760 v4f32 __builtin_msa_ffint_u_w (v4u32); 49761 v2f64 __builtin_msa_ffint_u_d (v2u64); 49762 49763 v4f32 __builtin_msa_ffql_w (v8i16); 49764 v2f64 __builtin_msa_ffql_d (v4i32); 49765 49766 v4f32 __builtin_msa_ffqr_w (v8i16); 49767 v2f64 __builtin_msa_ffqr_d (v4i32); 49768 49769 v16i8 __builtin_msa_fill_b (i32); 49770 v8i16 __builtin_msa_fill_h (i32); 49771 v4i32 __builtin_msa_fill_w (i32); 49772 v2i64 __builtin_msa_fill_d (i64); 49773 49774 v4f32 __builtin_msa_flog2_w (v4f32); 49775 v2f64 __builtin_msa_flog2_d (v2f64); 49776 49777 v4f32 __builtin_msa_fmadd_w (v4f32, v4f32, v4f32); 49778 v2f64 __builtin_msa_fmadd_d (v2f64, v2f64, v2f64); 49779 49780 v4f32 __builtin_msa_fmax_w (v4f32, v4f32); 49781 v2f64 __builtin_msa_fmax_d (v2f64, v2f64); 49782 49783 v4f32 __builtin_msa_fmax_a_w (v4f32, v4f32); 49784 v2f64 __builtin_msa_fmax_a_d (v2f64, v2f64); 49785 49786 v4f32 __builtin_msa_fmin_w (v4f32, v4f32); 49787 v2f64 __builtin_msa_fmin_d (v2f64, v2f64); 49788 49789 v4f32 __builtin_msa_fmin_a_w (v4f32, v4f32); 49790 v2f64 __builtin_msa_fmin_a_d (v2f64, v2f64); 49791 49792 v4f32 __builtin_msa_fmsub_w (v4f32, v4f32, v4f32); 49793 v2f64 __builtin_msa_fmsub_d (v2f64, v2f64, v2f64); 49794 49795 v4f32 __builtin_msa_fmul_w (v4f32, v4f32); 49796 v2f64 __builtin_msa_fmul_d (v2f64, v2f64); 49797 49798 v4f32 __builtin_msa_frint_w (v4f32); 49799 v2f64 __builtin_msa_frint_d (v2f64); 49800 49801 v4f32 __builtin_msa_frcp_w (v4f32); 49802 v2f64 __builtin_msa_frcp_d (v2f64); 49803 49804 v4f32 __builtin_msa_frsqrt_w (v4f32); 49805 v2f64 __builtin_msa_frsqrt_d (v2f64); 49806 49807 v4i32 __builtin_msa_fsaf_w (v4f32, v4f32); 49808 v2i64 __builtin_msa_fsaf_d (v2f64, v2f64); 49809 49810 v4i32 __builtin_msa_fseq_w (v4f32, v4f32); 49811 v2i64 __builtin_msa_fseq_d (v2f64, v2f64); 49812 49813 v4i32 __builtin_msa_fsle_w (v4f32, v4f32); 49814 v2i64 __builtin_msa_fsle_d (v2f64, v2f64); 49815 49816 v4i32 __builtin_msa_fslt_w (v4f32, v4f32); 49817 v2i64 __builtin_msa_fslt_d (v2f64, v2f64); 49818 49819 v4i32 __builtin_msa_fsne_w (v4f32, v4f32); 49820 v2i64 __builtin_msa_fsne_d (v2f64, v2f64); 49821 49822 v4i32 __builtin_msa_fsor_w (v4f32, v4f32); 49823 v2i64 __builtin_msa_fsor_d (v2f64, v2f64); 49824 49825 v4f32 __builtin_msa_fsqrt_w (v4f32); 49826 v2f64 __builtin_msa_fsqrt_d (v2f64); 49827 49828 v4f32 __builtin_msa_fsub_w (v4f32, v4f32); 49829 v2f64 __builtin_msa_fsub_d (v2f64, v2f64); 49830 49831 v4i32 __builtin_msa_fsueq_w (v4f32, v4f32); 49832 v2i64 __builtin_msa_fsueq_d (v2f64, v2f64); 49833 49834 v4i32 __builtin_msa_fsule_w (v4f32, v4f32); 49835 v2i64 __builtin_msa_fsule_d (v2f64, v2f64); 49836 49837 v4i32 __builtin_msa_fsult_w (v4f32, v4f32); 49838 v2i64 __builtin_msa_fsult_d (v2f64, v2f64); 49839 49840 v4i32 __builtin_msa_fsun_w (v4f32, v4f32); 49841 v2i64 __builtin_msa_fsun_d (v2f64, v2f64); 49842 49843 v4i32 __builtin_msa_fsune_w (v4f32, v4f32); 49844 v2i64 __builtin_msa_fsune_d (v2f64, v2f64); 49845 49846 v4i32 __builtin_msa_ftint_s_w (v4f32); 49847 v2i64 __builtin_msa_ftint_s_d (v2f64); 49848 49849 v4u32 __builtin_msa_ftint_u_w (v4f32); 49850 v2u64 __builtin_msa_ftint_u_d (v2f64); 49851 49852 v8i16 __builtin_msa_ftq_h (v4f32, v4f32); 49853 v4i32 __builtin_msa_ftq_w (v2f64, v2f64); 49854 49855 v4i32 __builtin_msa_ftrunc_s_w (v4f32); 49856 v2i64 __builtin_msa_ftrunc_s_d (v2f64); 49857 49858 v4u32 __builtin_msa_ftrunc_u_w (v4f32); 49859 v2u64 __builtin_msa_ftrunc_u_d (v2f64); 49860 49861 v8i16 __builtin_msa_hadd_s_h (v16i8, v16i8); 49862 v4i32 __builtin_msa_hadd_s_w (v8i16, v8i16); 49863 v2i64 __builtin_msa_hadd_s_d (v4i32, v4i32); 49864 49865 v8u16 __builtin_msa_hadd_u_h (v16u8, v16u8); 49866 v4u32 __builtin_msa_hadd_u_w (v8u16, v8u16); 49867 v2u64 __builtin_msa_hadd_u_d (v4u32, v4u32); 49868 49869 v8i16 __builtin_msa_hsub_s_h (v16i8, v16i8); 49870 v4i32 __builtin_msa_hsub_s_w (v8i16, v8i16); 49871 v2i64 __builtin_msa_hsub_s_d (v4i32, v4i32); 49872 49873 v8i16 __builtin_msa_hsub_u_h (v16u8, v16u8); 49874 v4i32 __builtin_msa_hsub_u_w (v8u16, v8u16); 49875 v2i64 __builtin_msa_hsub_u_d (v4u32, v4u32); 49876 49877 v16i8 __builtin_msa_ilvev_b (v16i8, v16i8); 49878 v8i16 __builtin_msa_ilvev_h (v8i16, v8i16); 49879 v4i32 __builtin_msa_ilvev_w (v4i32, v4i32); 49880 v2i64 __builtin_msa_ilvev_d (v2i64, v2i64); 49881 49882 v16i8 __builtin_msa_ilvl_b (v16i8, v16i8); 49883 v8i16 __builtin_msa_ilvl_h (v8i16, v8i16); 49884 v4i32 __builtin_msa_ilvl_w (v4i32, v4i32); 49885 v2i64 __builtin_msa_ilvl_d (v2i64, v2i64); 49886 49887 v16i8 __builtin_msa_ilvod_b (v16i8, v16i8); 49888 v8i16 __builtin_msa_ilvod_h (v8i16, v8i16); 49889 v4i32 __builtin_msa_ilvod_w (v4i32, v4i32); 49890 v2i64 __builtin_msa_ilvod_d (v2i64, v2i64); 49891 49892 v16i8 __builtin_msa_ilvr_b (v16i8, v16i8); 49893 v8i16 __builtin_msa_ilvr_h (v8i16, v8i16); 49894 v4i32 __builtin_msa_ilvr_w (v4i32, v4i32); 49895 v2i64 __builtin_msa_ilvr_d (v2i64, v2i64); 49896 49897 v16i8 __builtin_msa_insert_b (v16i8, imm0_15, i32); 49898 v8i16 __builtin_msa_insert_h (v8i16, imm0_7, i32); 49899 v4i32 __builtin_msa_insert_w (v4i32, imm0_3, i32); 49900 v2i64 __builtin_msa_insert_d (v2i64, imm0_1, i64); 49901 49902 v16i8 __builtin_msa_insve_b (v16i8, imm0_15, v16i8); 49903 v8i16 __builtin_msa_insve_h (v8i16, imm0_7, v8i16); 49904 v4i32 __builtin_msa_insve_w (v4i32, imm0_3, v4i32); 49905 v2i64 __builtin_msa_insve_d (v2i64, imm0_1, v2i64); 49906 49907 v16i8 __builtin_msa_ld_b (const void *, imm_n512_511); 49908 v8i16 __builtin_msa_ld_h (const void *, imm_n1024_1022); 49909 v4i32 __builtin_msa_ld_w (const void *, imm_n2048_2044); 49910 v2i64 __builtin_msa_ld_d (const void *, imm_n4096_4088); 49911 49912 v16i8 __builtin_msa_ldi_b (imm_n512_511); 49913 v8i16 __builtin_msa_ldi_h (imm_n512_511); 49914 v4i32 __builtin_msa_ldi_w (imm_n512_511); 49915 v2i64 __builtin_msa_ldi_d (imm_n512_511); 49916 49917 v8i16 __builtin_msa_madd_q_h (v8i16, v8i16, v8i16); 49918 v4i32 __builtin_msa_madd_q_w (v4i32, v4i32, v4i32); 49919 49920 v8i16 __builtin_msa_maddr_q_h (v8i16, v8i16, v8i16); 49921 v4i32 __builtin_msa_maddr_q_w (v4i32, v4i32, v4i32); 49922 49923 v16i8 __builtin_msa_maddv_b (v16i8, v16i8, v16i8); 49924 v8i16 __builtin_msa_maddv_h (v8i16, v8i16, v8i16); 49925 v4i32 __builtin_msa_maddv_w (v4i32, v4i32, v4i32); 49926 v2i64 __builtin_msa_maddv_d (v2i64, v2i64, v2i64); 49927 49928 v16i8 __builtin_msa_max_a_b (v16i8, v16i8); 49929 v8i16 __builtin_msa_max_a_h (v8i16, v8i16); 49930 v4i32 __builtin_msa_max_a_w (v4i32, v4i32); 49931 v2i64 __builtin_msa_max_a_d (v2i64, v2i64); 49932 49933 v16i8 __builtin_msa_max_s_b (v16i8, v16i8); 49934 v8i16 __builtin_msa_max_s_h (v8i16, v8i16); 49935 v4i32 __builtin_msa_max_s_w (v4i32, v4i32); 49936 v2i64 __builtin_msa_max_s_d (v2i64, v2i64); 49937 49938 v16u8 __builtin_msa_max_u_b (v16u8, v16u8); 49939 v8u16 __builtin_msa_max_u_h (v8u16, v8u16); 49940 v4u32 __builtin_msa_max_u_w (v4u32, v4u32); 49941 v2u64 __builtin_msa_max_u_d (v2u64, v2u64); 49942 49943 v16i8 __builtin_msa_maxi_s_b (v16i8, imm_n16_15); 49944 v8i16 __builtin_msa_maxi_s_h (v8i16, imm_n16_15); 49945 v4i32 __builtin_msa_maxi_s_w (v4i32, imm_n16_15); 49946 v2i64 __builtin_msa_maxi_s_d (v2i64, imm_n16_15); 49947 49948 v16u8 __builtin_msa_maxi_u_b (v16u8, imm0_31); 49949 v8u16 __builtin_msa_maxi_u_h (v8u16, imm0_31); 49950 v4u32 __builtin_msa_maxi_u_w (v4u32, imm0_31); 49951 v2u64 __builtin_msa_maxi_u_d (v2u64, imm0_31); 49952 49953 v16i8 __builtin_msa_min_a_b (v16i8, v16i8); 49954 v8i16 __builtin_msa_min_a_h (v8i16, v8i16); 49955 v4i32 __builtin_msa_min_a_w (v4i32, v4i32); 49956 v2i64 __builtin_msa_min_a_d (v2i64, v2i64); 49957 49958 v16i8 __builtin_msa_min_s_b (v16i8, v16i8); 49959 v8i16 __builtin_msa_min_s_h (v8i16, v8i16); 49960 v4i32 __builtin_msa_min_s_w (v4i32, v4i32); 49961 v2i64 __builtin_msa_min_s_d (v2i64, v2i64); 49962 49963 v16u8 __builtin_msa_min_u_b (v16u8, v16u8); 49964 v8u16 __builtin_msa_min_u_h (v8u16, v8u16); 49965 v4u32 __builtin_msa_min_u_w (v4u32, v4u32); 49966 v2u64 __builtin_msa_min_u_d (v2u64, v2u64); 49967 49968 v16i8 __builtin_msa_mini_s_b (v16i8, imm_n16_15); 49969 v8i16 __builtin_msa_mini_s_h (v8i16, imm_n16_15); 49970 v4i32 __builtin_msa_mini_s_w (v4i32, imm_n16_15); 49971 v2i64 __builtin_msa_mini_s_d (v2i64, imm_n16_15); 49972 49973 v16u8 __builtin_msa_mini_u_b (v16u8, imm0_31); 49974 v8u16 __builtin_msa_mini_u_h (v8u16, imm0_31); 49975 v4u32 __builtin_msa_mini_u_w (v4u32, imm0_31); 49976 v2u64 __builtin_msa_mini_u_d (v2u64, imm0_31); 49977 49978 v16i8 __builtin_msa_mod_s_b (v16i8, v16i8); 49979 v8i16 __builtin_msa_mod_s_h (v8i16, v8i16); 49980 v4i32 __builtin_msa_mod_s_w (v4i32, v4i32); 49981 v2i64 __builtin_msa_mod_s_d (v2i64, v2i64); 49982 49983 v16u8 __builtin_msa_mod_u_b (v16u8, v16u8); 49984 v8u16 __builtin_msa_mod_u_h (v8u16, v8u16); 49985 v4u32 __builtin_msa_mod_u_w (v4u32, v4u32); 49986 v2u64 __builtin_msa_mod_u_d (v2u64, v2u64); 49987 49988 v16i8 __builtin_msa_move_v (v16i8); 49989 49990 v8i16 __builtin_msa_msub_q_h (v8i16, v8i16, v8i16); 49991 v4i32 __builtin_msa_msub_q_w (v4i32, v4i32, v4i32); 49992 49993 v8i16 __builtin_msa_msubr_q_h (v8i16, v8i16, v8i16); 49994 v4i32 __builtin_msa_msubr_q_w (v4i32, v4i32, v4i32); 49995 49996 v16i8 __builtin_msa_msubv_b (v16i8, v16i8, v16i8); 49997 v8i16 __builtin_msa_msubv_h (v8i16, v8i16, v8i16); 49998 v4i32 __builtin_msa_msubv_w (v4i32, v4i32, v4i32); 49999 v2i64 __builtin_msa_msubv_d (v2i64, v2i64, v2i64); 50000 50001 v8i16 __builtin_msa_mul_q_h (v8i16, v8i16); 50002 v4i32 __builtin_msa_mul_q_w (v4i32, v4i32); 50003 50004 v8i16 __builtin_msa_mulr_q_h (v8i16, v8i16); 50005 v4i32 __builtin_msa_mulr_q_w (v4i32, v4i32); 50006 50007 v16i8 __builtin_msa_mulv_b (v16i8, v16i8); 50008 v8i16 __builtin_msa_mulv_h (v8i16, v8i16); 50009 v4i32 __builtin_msa_mulv_w (v4i32, v4i32); 50010 v2i64 __builtin_msa_mulv_d (v2i64, v2i64); 50011 50012 v16i8 __builtin_msa_nloc_b (v16i8); 50013 v8i16 __builtin_msa_nloc_h (v8i16); 50014 v4i32 __builtin_msa_nloc_w (v4i32); 50015 v2i64 __builtin_msa_nloc_d (v2i64); 50016 50017 v16i8 __builtin_msa_nlzc_b (v16i8); 50018 v8i16 __builtin_msa_nlzc_h (v8i16); 50019 v4i32 __builtin_msa_nlzc_w (v4i32); 50020 v2i64 __builtin_msa_nlzc_d (v2i64); 50021 50022 v16u8 __builtin_msa_nor_v (v16u8, v16u8); 50023 50024 v16u8 __builtin_msa_nori_b (v16u8, imm0_255); 50025 50026 v16u8 __builtin_msa_or_v (v16u8, v16u8); 50027 50028 v16u8 __builtin_msa_ori_b (v16u8, imm0_255); 50029 50030 v16i8 __builtin_msa_pckev_b (v16i8, v16i8); 50031 v8i16 __builtin_msa_pckev_h (v8i16, v8i16); 50032 v4i32 __builtin_msa_pckev_w (v4i32, v4i32); 50033 v2i64 __builtin_msa_pckev_d (v2i64, v2i64); 50034 50035 v16i8 __builtin_msa_pckod_b (v16i8, v16i8); 50036 v8i16 __builtin_msa_pckod_h (v8i16, v8i16); 50037 v4i32 __builtin_msa_pckod_w (v4i32, v4i32); 50038 v2i64 __builtin_msa_pckod_d (v2i64, v2i64); 50039 50040 v16i8 __builtin_msa_pcnt_b (v16i8); 50041 v8i16 __builtin_msa_pcnt_h (v8i16); 50042 v4i32 __builtin_msa_pcnt_w (v4i32); 50043 v2i64 __builtin_msa_pcnt_d (v2i64); 50044 50045 v16i8 __builtin_msa_sat_s_b (v16i8, imm0_7); 50046 v8i16 __builtin_msa_sat_s_h (v8i16, imm0_15); 50047 v4i32 __builtin_msa_sat_s_w (v4i32, imm0_31); 50048 v2i64 __builtin_msa_sat_s_d (v2i64, imm0_63); 50049 50050 v16u8 __builtin_msa_sat_u_b (v16u8, imm0_7); 50051 v8u16 __builtin_msa_sat_u_h (v8u16, imm0_15); 50052 v4u32 __builtin_msa_sat_u_w (v4u32, imm0_31); 50053 v2u64 __builtin_msa_sat_u_d (v2u64, imm0_63); 50054 50055 v16i8 __builtin_msa_shf_b (v16i8, imm0_255); 50056 v8i16 __builtin_msa_shf_h (v8i16, imm0_255); 50057 v4i32 __builtin_msa_shf_w (v4i32, imm0_255); 50058 50059 v16i8 __builtin_msa_sld_b (v16i8, v16i8, i32); 50060 v8i16 __builtin_msa_sld_h (v8i16, v8i16, i32); 50061 v4i32 __builtin_msa_sld_w (v4i32, v4i32, i32); 50062 v2i64 __builtin_msa_sld_d (v2i64, v2i64, i32); 50063 50064 v16i8 __builtin_msa_sldi_b (v16i8, v16i8, imm0_15); 50065 v8i16 __builtin_msa_sldi_h (v8i16, v8i16, imm0_7); 50066 v4i32 __builtin_msa_sldi_w (v4i32, v4i32, imm0_3); 50067 v2i64 __builtin_msa_sldi_d (v2i64, v2i64, imm0_1); 50068 50069 v16i8 __builtin_msa_sll_b (v16i8, v16i8); 50070 v8i16 __builtin_msa_sll_h (v8i16, v8i16); 50071 v4i32 __builtin_msa_sll_w (v4i32, v4i32); 50072 v2i64 __builtin_msa_sll_d (v2i64, v2i64); 50073 50074 v16i8 __builtin_msa_slli_b (v16i8, imm0_7); 50075 v8i16 __builtin_msa_slli_h (v8i16, imm0_15); 50076 v4i32 __builtin_msa_slli_w (v4i32, imm0_31); 50077 v2i64 __builtin_msa_slli_d (v2i64, imm0_63); 50078 50079 v16i8 __builtin_msa_splat_b (v16i8, i32); 50080 v8i16 __builtin_msa_splat_h (v8i16, i32); 50081 v4i32 __builtin_msa_splat_w (v4i32, i32); 50082 v2i64 __builtin_msa_splat_d (v2i64, i32); 50083 50084 v16i8 __builtin_msa_splati_b (v16i8, imm0_15); 50085 v8i16 __builtin_msa_splati_h (v8i16, imm0_7); 50086 v4i32 __builtin_msa_splati_w (v4i32, imm0_3); 50087 v2i64 __builtin_msa_splati_d (v2i64, imm0_1); 50088 50089 v16i8 __builtin_msa_sra_b (v16i8, v16i8); 50090 v8i16 __builtin_msa_sra_h (v8i16, v8i16); 50091 v4i32 __builtin_msa_sra_w (v4i32, v4i32); 50092 v2i64 __builtin_msa_sra_d (v2i64, v2i64); 50093 50094 v16i8 __builtin_msa_srai_b (v16i8, imm0_7); 50095 v8i16 __builtin_msa_srai_h (v8i16, imm0_15); 50096 v4i32 __builtin_msa_srai_w (v4i32, imm0_31); 50097 v2i64 __builtin_msa_srai_d (v2i64, imm0_63); 50098 50099 v16i8 __builtin_msa_srar_b (v16i8, v16i8); 50100 v8i16 __builtin_msa_srar_h (v8i16, v8i16); 50101 v4i32 __builtin_msa_srar_w (v4i32, v4i32); 50102 v2i64 __builtin_msa_srar_d (v2i64, v2i64); 50103 50104 v16i8 __builtin_msa_srari_b (v16i8, imm0_7); 50105 v8i16 __builtin_msa_srari_h (v8i16, imm0_15); 50106 v4i32 __builtin_msa_srari_w (v4i32, imm0_31); 50107 v2i64 __builtin_msa_srari_d (v2i64, imm0_63); 50108 50109 v16i8 __builtin_msa_srl_b (v16i8, v16i8); 50110 v8i16 __builtin_msa_srl_h (v8i16, v8i16); 50111 v4i32 __builtin_msa_srl_w (v4i32, v4i32); 50112 v2i64 __builtin_msa_srl_d (v2i64, v2i64); 50113 50114 v16i8 __builtin_msa_srli_b (v16i8, imm0_7); 50115 v8i16 __builtin_msa_srli_h (v8i16, imm0_15); 50116 v4i32 __builtin_msa_srli_w (v4i32, imm0_31); 50117 v2i64 __builtin_msa_srli_d (v2i64, imm0_63); 50118 50119 v16i8 __builtin_msa_srlr_b (v16i8, v16i8); 50120 v8i16 __builtin_msa_srlr_h (v8i16, v8i16); 50121 v4i32 __builtin_msa_srlr_w (v4i32, v4i32); 50122 v2i64 __builtin_msa_srlr_d (v2i64, v2i64); 50123 50124 v16i8 __builtin_msa_srlri_b (v16i8, imm0_7); 50125 v8i16 __builtin_msa_srlri_h (v8i16, imm0_15); 50126 v4i32 __builtin_msa_srlri_w (v4i32, imm0_31); 50127 v2i64 __builtin_msa_srlri_d (v2i64, imm0_63); 50128 50129 void __builtin_msa_st_b (v16i8, void *, imm_n512_511); 50130 void __builtin_msa_st_h (v8i16, void *, imm_n1024_1022); 50131 void __builtin_msa_st_w (v4i32, void *, imm_n2048_2044); 50132 void __builtin_msa_st_d (v2i64, void *, imm_n4096_4088); 50133 50134 v16i8 __builtin_msa_subs_s_b (v16i8, v16i8); 50135 v8i16 __builtin_msa_subs_s_h (v8i16, v8i16); 50136 v4i32 __builtin_msa_subs_s_w (v4i32, v4i32); 50137 v2i64 __builtin_msa_subs_s_d (v2i64, v2i64); 50138 50139 v16u8 __builtin_msa_subs_u_b (v16u8, v16u8); 50140 v8u16 __builtin_msa_subs_u_h (v8u16, v8u16); 50141 v4u32 __builtin_msa_subs_u_w (v4u32, v4u32); 50142 v2u64 __builtin_msa_subs_u_d (v2u64, v2u64); 50143 50144 v16u8 __builtin_msa_subsus_u_b (v16u8, v16i8); 50145 v8u16 __builtin_msa_subsus_u_h (v8u16, v8i16); 50146 v4u32 __builtin_msa_subsus_u_w (v4u32, v4i32); 50147 v2u64 __builtin_msa_subsus_u_d (v2u64, v2i64); 50148 50149 v16i8 __builtin_msa_subsuu_s_b (v16u8, v16u8); 50150 v8i16 __builtin_msa_subsuu_s_h (v8u16, v8u16); 50151 v4i32 __builtin_msa_subsuu_s_w (v4u32, v4u32); 50152 v2i64 __builtin_msa_subsuu_s_d (v2u64, v2u64); 50153 50154 v16i8 __builtin_msa_subv_b (v16i8, v16i8); 50155 v8i16 __builtin_msa_subv_h (v8i16, v8i16); 50156 v4i32 __builtin_msa_subv_w (v4i32, v4i32); 50157 v2i64 __builtin_msa_subv_d (v2i64, v2i64); 50158 50159 v16i8 __builtin_msa_subvi_b (v16i8, imm0_31); 50160 v8i16 __builtin_msa_subvi_h (v8i16, imm0_31); 50161 v4i32 __builtin_msa_subvi_w (v4i32, imm0_31); 50162 v2i64 __builtin_msa_subvi_d (v2i64, imm0_31); 50163 50164 v16i8 __builtin_msa_vshf_b (v16i8, v16i8, v16i8); 50165 v8i16 __builtin_msa_vshf_h (v8i16, v8i16, v8i16); 50166 v4i32 __builtin_msa_vshf_w (v4i32, v4i32, v4i32); 50167 v2i64 __builtin_msa_vshf_d (v2i64, v2i64, v2i64); 50168 50169 v16u8 __builtin_msa_xor_v (v16u8, v16u8); 50170 50171 v16u8 __builtin_msa_xori_b (v16u8, imm0_255); 50172 50173 50174File: gcc.info, Node: Other MIPS Built-in Functions, Next: MSP430 Built-in Functions, Prev: MIPS SIMD Architecture (MSA) Support, Up: Target Builtins 50175 501766.60.18 Other MIPS Built-in Functions 50177------------------------------------- 50178 50179GCC provides other MIPS-specific built-in functions: 50180 50181'void __builtin_mips_cache (int OP, const volatile void *ADDR)' 50182 Insert a 'cache' instruction with operands OP and ADDR. GCC 50183 defines the preprocessor macro '___GCC_HAVE_BUILTIN_MIPS_CACHE' 50184 when this function is available. 50185 50186'unsigned int __builtin_mips_get_fcsr (void)' 50187'void __builtin_mips_set_fcsr (unsigned int VALUE)' 50188 Get and set the contents of the floating-point control and status 50189 register (FPU control register 31). These functions are only 50190 available in hard-float code but can be called in both MIPS16 and 50191 non-MIPS16 contexts. 50192 50193 '__builtin_mips_set_fcsr' can be used to change any bit of the 50194 register except the condition codes, which GCC assumes are 50195 preserved. 50196 50197 50198File: gcc.info, Node: MSP430 Built-in Functions, Next: NDS32 Built-in Functions, Prev: Other MIPS Built-in Functions, Up: Target Builtins 50199 502006.60.19 MSP430 Built-in Functions 50201--------------------------------- 50202 50203GCC provides a couple of special builtin functions to aid in the writing 50204of interrupt handlers in C. 50205 50206'__bic_SR_register_on_exit (int MASK)' 50207 This clears the indicated bits in the saved copy of the status 50208 register currently residing on the stack. This only works inside 50209 interrupt handlers and the changes to the status register will only 50210 take affect once the handler returns. 50211 50212'__bis_SR_register_on_exit (int MASK)' 50213 This sets the indicated bits in the saved copy of the status 50214 register currently residing on the stack. This only works inside 50215 interrupt handlers and the changes to the status register will only 50216 take affect once the handler returns. 50217 50218'__delay_cycles (long long CYCLES)' 50219 This inserts an instruction sequence that takes exactly CYCLES 50220 cycles (between 0 and about 17E9) to complete. The inserted 50221 sequence may use jumps, loops, or no-ops, and does not interfere 50222 with any other instructions. Note that CYCLES must be a 50223 compile-time constant integer - that is, you must pass a number, 50224 not a variable that may be optimized to a constant later. The 50225 number of cycles delayed by this builtin is exact. 50226 50227 50228File: gcc.info, Node: NDS32 Built-in Functions, Next: picoChip Built-in Functions, Prev: MSP430 Built-in Functions, Up: Target Builtins 50229 502306.60.20 NDS32 Built-in Functions 50231-------------------------------- 50232 50233These built-in functions are available for the NDS32 target: 50234 50235 -- Built-in Function: void __builtin_nds32_isync (int *ADDR) 50236 Insert an ISYNC instruction into the instruction stream where ADDR 50237 is an instruction address for serialization. 50238 50239 -- Built-in Function: void __builtin_nds32_isb (void) 50240 Insert an ISB instruction into the instruction stream. 50241 50242 -- Built-in Function: int __builtin_nds32_mfsr (int SR) 50243 Return the content of a system register which is mapped by SR. 50244 50245 -- Built-in Function: int __builtin_nds32_mfusr (int USR) 50246 Return the content of a user space register which is mapped by USR. 50247 50248 -- Built-in Function: void __builtin_nds32_mtsr (int VALUE, int SR) 50249 Move the VALUE to a system register which is mapped by SR. 50250 50251 -- Built-in Function: void __builtin_nds32_mtusr (int VALUE, int USR) 50252 Move the VALUE to a user space register which is mapped by USR. 50253 50254 -- Built-in Function: void __builtin_nds32_setgie_en (void) 50255 Enable global interrupt. 50256 50257 -- Built-in Function: void __builtin_nds32_setgie_dis (void) 50258 Disable global interrupt. 50259 50260 50261File: gcc.info, Node: picoChip Built-in Functions, Next: Basic PowerPC Built-in Functions, Prev: NDS32 Built-in Functions, Up: Target Builtins 50262 502636.60.21 picoChip Built-in Functions 50264----------------------------------- 50265 50266GCC provides an interface to selected machine instructions from the 50267picoChip instruction set. 50268 50269'int __builtin_sbc (int VALUE)' 50270 Sign bit count. Return the number of consecutive bits in VALUE 50271 that have the same value as the sign bit. The result is the number 50272 of leading sign bits minus one, giving the number of redundant sign 50273 bits in VALUE. 50274 50275'int __builtin_byteswap (int VALUE)' 50276 Byte swap. Return the result of swapping the upper and lower bytes 50277 of VALUE. 50278 50279'int __builtin_brev (int VALUE)' 50280 Bit reversal. Return the result of reversing the bits in VALUE. 50281 Bit 15 is swapped with bit 0, bit 14 is swapped with bit 1, and so 50282 on. 50283 50284'int __builtin_adds (int X, int Y)' 50285 Saturating addition. Return the result of adding X and Y, storing 50286 the value 32767 if the result overflows. 50287 50288'int __builtin_subs (int X, int Y)' 50289 Saturating subtraction. Return the result of subtracting Y from X, 50290 storing the value -32768 if the result overflows. 50291 50292'void __builtin_halt (void)' 50293 Halt. The processor stops execution. This built-in is useful for 50294 implementing assertions. 50295 50296 50297File: gcc.info, Node: Basic PowerPC Built-in Functions, Next: PowerPC AltiVec/VSX Built-in Functions, Prev: picoChip Built-in Functions, Up: Target Builtins 50298 502996.60.22 Basic PowerPC Built-in Functions 50300---------------------------------------- 50301 50302* Menu: 50303 50304* Basic PowerPC Built-in Functions Available on all Configurations:: 50305* Basic PowerPC Built-in Functions Available on ISA 2.05:: 50306* Basic PowerPC Built-in Functions Available on ISA 2.06:: 50307* Basic PowerPC Built-in Functions Available on ISA 2.07:: 50308* Basic PowerPC Built-in Functions Available on ISA 3.0:: 50309* Basic PowerPC Built-in Functions Available on ISA 3.1:: 50310 50311This section describes PowerPC built-in functions that do not require 50312the inclusion of any special header files to declare prototypes or 50313provide macro definitions. The sections that follow describe additional 50314PowerPC built-in functions. 50315 50316 50317File: gcc.info, Node: Basic PowerPC Built-in Functions Available on all Configurations, Next: Basic PowerPC Built-in Functions Available on ISA 2.05, Up: Basic PowerPC Built-in Functions 50318 503196.60.22.1 Basic PowerPC Built-in Functions Available on all Configurations 50320.......................................................................... 50321 50322 -- Built-in Function: void __builtin_cpu_init (void) 50323 This function is a 'nop' on the PowerPC platform and is included 50324 solely to maintain API compatibility with the x86 builtins. 50325 50326 -- Built-in Function: int __builtin_cpu_is (const char *CPUNAME) 50327 This function returns a value of '1' if the run-time CPU is of type 50328 CPUNAME and returns '0' otherwise 50329 50330 The '__builtin_cpu_is' function requires GLIBC 2.23 or newer which 50331 exports the hardware capability bits. GCC defines the macro 50332 '__BUILTIN_CPU_SUPPORTS__' if the '__builtin_cpu_supports' built-in 50333 function is fully supported. 50334 50335 If GCC was configured to use a GLIBC before 2.23, the built-in 50336 function '__builtin_cpu_is' always returns a 0 and the compiler 50337 issues a warning. 50338 50339 The following CPU names can be detected: 50340 50341 'power10' 50342 IBM POWER10 Server CPU. 50343 'power9' 50344 IBM POWER9 Server CPU. 50345 'power8' 50346 IBM POWER8 Server CPU. 50347 'power7' 50348 IBM POWER7 Server CPU. 50349 'power6x' 50350 IBM POWER6 Server CPU (RAW mode). 50351 'power6' 50352 IBM POWER6 Server CPU (Architected mode). 50353 'power5+' 50354 IBM POWER5+ Server CPU. 50355 'power5' 50356 IBM POWER5 Server CPU. 50357 'ppc970' 50358 IBM 970 Server CPU (ie, Apple G5). 50359 'power4' 50360 IBM POWER4 Server CPU. 50361 'ppca2' 50362 IBM A2 64-bit Embedded CPU 50363 'ppc476' 50364 IBM PowerPC 476FP 32-bit Embedded CPU. 50365 'ppc464' 50366 IBM PowerPC 464 32-bit Embedded CPU. 50367 'ppc440' 50368 PowerPC 440 32-bit Embedded CPU. 50369 'ppc405' 50370 PowerPC 405 32-bit Embedded CPU. 50371 'ppc-cell-be' 50372 IBM PowerPC Cell Broadband Engine Architecture CPU. 50373 50374 Here is an example: 50375 #ifdef __BUILTIN_CPU_SUPPORTS__ 50376 if (__builtin_cpu_is ("power8")) 50377 { 50378 do_power8 (); // POWER8 specific implementation. 50379 } 50380 else 50381 #endif 50382 { 50383 do_generic (); // Generic implementation. 50384 } 50385 50386 -- Built-in Function: int __builtin_cpu_supports (const char *FEATURE) 50387 This function returns a value of '1' if the run-time CPU supports 50388 the HWCAP feature FEATURE and returns '0' otherwise. 50389 50390 The '__builtin_cpu_supports' function requires GLIBC 2.23 or newer 50391 which exports the hardware capability bits. GCC defines the macro 50392 '__BUILTIN_CPU_SUPPORTS__' if the '__builtin_cpu_supports' built-in 50393 function is fully supported. 50394 50395 If GCC was configured to use a GLIBC before 2.23, the built-in 50396 function '__builtin_cpu_supports' always returns a 0 and the 50397 compiler issues a warning. 50398 50399 The following features can be detected: 50400 50401 '4xxmac' 50402 4xx CPU has a Multiply Accumulator. 50403 'altivec' 50404 CPU has a SIMD/Vector Unit. 50405 'arch_2_05' 50406 CPU supports ISA 2.05 (eg, POWER6) 50407 'arch_2_06' 50408 CPU supports ISA 2.06 (eg, POWER7) 50409 'arch_2_07' 50410 CPU supports ISA 2.07 (eg, POWER8) 50411 'arch_3_00' 50412 CPU supports ISA 3.0 (eg, POWER9) 50413 'arch_3_1' 50414 CPU supports ISA 3.1 (eg, POWER10) 50415 'archpmu' 50416 CPU supports the set of compatible performance monitoring 50417 events. 50418 'booke' 50419 CPU supports the Embedded ISA category. 50420 'cellbe' 50421 CPU has a CELL broadband engine. 50422 'darn' 50423 CPU supports the 'darn' (deliver a random number) instruction. 50424 'dfp' 50425 CPU has a decimal floating point unit. 50426 'dscr' 50427 CPU supports the data stream control register. 50428 'ebb' 50429 CPU supports event base branching. 50430 'efpdouble' 50431 CPU has a SPE double precision floating point unit. 50432 'efpsingle' 50433 CPU has a SPE single precision floating point unit. 50434 'fpu' 50435 CPU has a floating point unit. 50436 'htm' 50437 CPU has hardware transaction memory instructions. 50438 'htm-nosc' 50439 Kernel aborts hardware transactions when a syscall is made. 50440 'htm-no-suspend' 50441 CPU supports hardware transaction memory but does not support 50442 the 'tsuspend.' instruction. 50443 'ic_snoop' 50444 CPU supports icache snooping capabilities. 50445 'ieee128' 50446 CPU supports 128-bit IEEE binary floating point instructions. 50447 'isel' 50448 CPU supports the integer select instruction. 50449 'mma' 50450 CPU supports the matrix-multiply assist instructions. 50451 'mmu' 50452 CPU has a memory management unit. 50453 'notb' 50454 CPU does not have a timebase (eg, 601 and 403gx). 50455 'pa6t' 50456 CPU supports the PA Semi 6T CORE ISA. 50457 'power4' 50458 CPU supports ISA 2.00 (eg, POWER4) 50459 'power5' 50460 CPU supports ISA 2.02 (eg, POWER5) 50461 'power5+' 50462 CPU supports ISA 2.03 (eg, POWER5+) 50463 'power6x' 50464 CPU supports ISA 2.05 (eg, POWER6) extended opcodes mffgpr and 50465 mftgpr. 50466 'ppc32' 50467 CPU supports 32-bit mode execution. 50468 'ppc601' 50469 CPU supports the old POWER ISA (eg, 601) 50470 'ppc64' 50471 CPU supports 64-bit mode execution. 50472 'ppcle' 50473 CPU supports a little-endian mode that uses address swizzling. 50474 'scv' 50475 Kernel supports system call vectored. 50476 'smt' 50477 CPU support simultaneous multi-threading. 50478 'spe' 50479 CPU has a signal processing extension unit. 50480 'tar' 50481 CPU supports the target address register. 50482 'true_le' 50483 CPU supports true little-endian mode. 50484 'ucache' 50485 CPU has unified I/D cache. 50486 'vcrypto' 50487 CPU supports the vector cryptography instructions. 50488 'vsx' 50489 CPU supports the vector-scalar extension. 50490 50491 Here is an example: 50492 #ifdef __BUILTIN_CPU_SUPPORTS__ 50493 if (__builtin_cpu_supports ("fpu")) 50494 { 50495 asm("fadd %0,%1,%2" : "=d"(dst) : "d"(src1), "d"(src2)); 50496 } 50497 else 50498 #endif 50499 { 50500 dst = __fadd (src1, src2); // Software FP addition function. 50501 } 50502 50503 The following built-in functions are also available on all PowerPC 50504processors: 50505 uint64_t __builtin_ppc_get_timebase (); 50506 unsigned long __builtin_ppc_mftb (); 50507 double __builtin_unpack_ibm128 (__ibm128, int); 50508 __ibm128 __builtin_pack_ibm128 (double, double); 50509 double __builtin_mffs (void); 50510 void __builtin_mtfsf (const int, double); 50511 void __builtin_mtfsb0 (const int); 50512 void __builtin_mtfsb1 (const int); 50513 void __builtin_set_fpscr_rn (int); 50514 50515 The '__builtin_ppc_get_timebase' and '__builtin_ppc_mftb' functions 50516generate instructions to read the Time Base Register. The 50517'__builtin_ppc_get_timebase' function may generate multiple instructions 50518and always returns the 64 bits of the Time Base Register. The 50519'__builtin_ppc_mftb' function always generates one instruction and 50520returns the Time Base Register value as an unsigned long, throwing away 50521the most significant word on 32-bit environments. The '__builtin_mffs' 50522return the value of the FPSCR register. Note, ISA 3.0 supports the 50523'__builtin_mffsl()' which permits software to read the control and 50524non-sticky status bits in the FSPCR without the higher latency 50525associated with accessing the sticky status bits. The '__builtin_mtfsf' 50526takes a constant 8-bit integer field mask and a double precision 50527floating point argument and generates the 'mtfsf' (extended mnemonic) 50528instruction to write new values to selected fields of the FPSCR. The 50529'__builtin_mtfsb0' and '__builtin_mtfsb1' take the bit to change as an 50530argument. The valid bit range is between 0 and 31. The builtins map to 50531the 'mtfsb0' and 'mtfsb1' instructions which take the argument and add 5053232. Hence these instructions only modify the FPSCR[32:63] bits by 50533changing the specified bit to a zero or one respectively. The 50534'__builtin_set_fpscr_rn' builtin allows changing both of the floating 50535point rounding mode bits. The argument is a 2-bit value. The argument 50536can either be a 'const int' or stored in a variable. The builtin uses 50537the ISA 3.0 instruction 'mffscrn' if available, otherwise it reads the 50538FPSCR, masks the current rounding mode bits out and OR's in the new 50539value. 50540 50541 50542File: gcc.info, Node: Basic PowerPC Built-in Functions Available on ISA 2.05, Next: Basic PowerPC Built-in Functions Available on ISA 2.06, Prev: Basic PowerPC Built-in Functions Available on all Configurations, Up: Basic PowerPC Built-in Functions 50543 505446.60.22.2 Basic PowerPC Built-in Functions Available on ISA 2.05 50545................................................................ 50546 50547The basic built-in functions described in this section are available on 50548the PowerPC family of processors starting with ISA 2.05 or later. 50549Unless specific options are explicitly disabled on the command line, 50550specifying option '-mcpu=power6' has the effect of enabling the 50551'-mpowerpc64', '-mpowerpc-gpopt', '-mpowerpc-gfxopt', '-mmfcrf', 50552'-mpopcntb', '-mfprnd', '-mcmpb', '-mhard-dfp', and '-mrecip-precision' 50553options. Specify the '-maltivec' option explicitly in combination with 50554the above options if desired. 50555 50556 The following functions require option '-mcmpb'. 50557 unsigned long long __builtin_cmpb (unsigned long long int, unsigned long long int); 50558 unsigned int __builtin_cmpb (unsigned int, unsigned int); 50559 50560 The '__builtin_cmpb' function performs a byte-wise compare on the 50561contents of its two arguments, returning the result of the byte-wise 50562comparison as the returned value. For each byte comparison, the 50563corresponding byte of the return value holds 0xff if the input bytes are 50564equal and 0 if the input bytes are not equal. If either of the 50565arguments to this built-in function is wider than 32 bits, the function 50566call expands into the form that expects 'unsigned long long int' 50567arguments which is only available on 64-bit targets. 50568 50569 The following built-in functions are available when hardware decimal 50570floating point ('-mhard-dfp') is available: 50571 void __builtin_set_fpscr_drn(int); 50572 _Decimal64 __builtin_ddedpd (int, _Decimal64); 50573 _Decimal128 __builtin_ddedpdq (int, _Decimal128); 50574 _Decimal64 __builtin_denbcd (int, _Decimal64); 50575 _Decimal128 __builtin_denbcdq (int, _Decimal128); 50576 _Decimal64 __builtin_diex (long long, _Decimal64); 50577 _Decimal128 _builtin_diexq (long long, _Decimal128); 50578 _Decimal64 __builtin_dscli (_Decimal64, int); 50579 _Decimal128 __builtin_dscliq (_Decimal128, int); 50580 _Decimal64 __builtin_dscri (_Decimal64, int); 50581 _Decimal128 __builtin_dscriq (_Decimal128, int); 50582 long long __builtin_dxex (_Decimal64); 50583 long long __builtin_dxexq (_Decimal128); 50584 _Decimal128 __builtin_pack_dec128 (unsigned long long, unsigned long long); 50585 unsigned long long __builtin_unpack_dec128 (_Decimal128, int); 50586 50587 The __builtin_set_fpscr_drn builtin allows changing the three decimal 50588 floating point rounding mode bits. The argument is a 3-bit value. The 50589 argument can either be a const int or the value can be stored in 50590 a variable. 50591 The builtin uses the ISA 3.0 instruction mffscdrn if available. 50592 Otherwise the builtin reads the FPSCR, masks the current decimal rounding 50593 mode bits out and OR's in the new value. 50594 50595 50596 The following functions require '-mhard-float', '-mpowerpc-gfxopt', and 50597'-mpopcntb' options. 50598 50599 double __builtin_recipdiv (double, double); 50600 float __builtin_recipdivf (float, float); 50601 double __builtin_rsqrt (double); 50602 float __builtin_rsqrtf (float); 50603 50604 The 'vec_rsqrt', '__builtin_rsqrt', and '__builtin_rsqrtf' functions 50605generate multiple instructions to implement the reciprocal sqrt 50606functionality using reciprocal sqrt estimate instructions. 50607 50608 The '__builtin_recipdiv', and '__builtin_recipdivf' functions generate 50609multiple instructions to implement division using the reciprocal 50610estimate instructions. 50611 50612 The following functions require '-mhard-float' and '-mmultiple' 50613options. 50614 50615 The '__builtin_unpack_longdouble' function takes a 'long double' 50616argument and a compile time constant of 0 or 1. If the constant is 0, 50617the first 'double' within the 'long double' is returned, otherwise the 50618second 'double' is returned. The '__builtin_unpack_longdouble' function 50619is only available if 'long double' uses the IBM extended double 50620representation. 50621 50622 The '__builtin_pack_longdouble' function takes two 'double' arguments 50623and returns a 'long double' value that combines the two arguments. The 50624'__builtin_pack_longdouble' function is only available if 'long double' 50625uses the IBM extended double representation. 50626 50627 The '__builtin_unpack_ibm128' function takes a '__ibm128' argument and 50628a compile time constant of 0 or 1. If the constant is 0, the first 50629'double' within the '__ibm128' is returned, otherwise the second 50630'double' is returned. 50631 50632 The '__builtin_pack_ibm128' function takes two 'double' arguments and 50633returns a '__ibm128' value that combines the two arguments. 50634 50635 Additional built-in functions are available for the 64-bit PowerPC 50636family of processors, for efficient use of 128-bit floating point 50637('__float128') values. 50638 50639 50640File: gcc.info, Node: Basic PowerPC Built-in Functions Available on ISA 2.06, Next: Basic PowerPC Built-in Functions Available on ISA 2.07, Prev: Basic PowerPC Built-in Functions Available on ISA 2.05, Up: Basic PowerPC Built-in Functions 50641 506426.60.22.3 Basic PowerPC Built-in Functions Available on ISA 2.06 50643................................................................ 50644 50645The basic built-in functions described in this section are available on 50646the PowerPC family of processors starting with ISA 2.05 or later. 50647Unless specific options are explicitly disabled on the command line, 50648specifying option '-mcpu=power7' has the effect of enabling all the same 50649options as for '-mcpu=power6' in addition to the '-maltivec', 50650'-mpopcntd', and '-mvsx' options. 50651 50652 The following basic built-in functions require '-mpopcntd': 50653 unsigned int __builtin_addg6s (unsigned int, unsigned int); 50654 long long __builtin_bpermd (long long, long long); 50655 unsigned int __builtin_cbcdtd (unsigned int); 50656 unsigned int __builtin_cdtbcd (unsigned int); 50657 long long __builtin_divde (long long, long long); 50658 unsigned long long __builtin_divdeu (unsigned long long, unsigned long long); 50659 int __builtin_divwe (int, int); 50660 unsigned int __builtin_divweu (unsigned int, unsigned int); 50661 vector __int128 __builtin_pack_vector_int128 (long long, long long); 50662 void __builtin_rs6000_speculation_barrier (void); 50663 long long __builtin_unpack_vector_int128 (vector __int128, signed char); 50664 50665 Of these, the '__builtin_divde' and '__builtin_divdeu' functions 50666require a 64-bit environment. 50667 50668 The following basic built-in functions, which are also supported on x86 50669targets, require '-mfloat128'. 50670 __float128 __builtin_fabsq (__float128); 50671 __float128 __builtin_copysignq (__float128, __float128); 50672 __float128 __builtin_infq (void); 50673 __float128 __builtin_huge_valq (void); 50674 __float128 __builtin_nanq (void); 50675 __float128 __builtin_nansq (void); 50676 50677 __float128 __builtin_sqrtf128 (__float128); 50678 __float128 __builtin_fmaf128 (__float128, __float128, __float128); 50679 50680 50681File: gcc.info, Node: Basic PowerPC Built-in Functions Available on ISA 2.07, Next: Basic PowerPC Built-in Functions Available on ISA 3.0, Prev: Basic PowerPC Built-in Functions Available on ISA 2.06, Up: Basic PowerPC Built-in Functions 50682 506836.60.22.4 Basic PowerPC Built-in Functions Available on ISA 2.07 50684................................................................ 50685 50686The basic built-in functions described in this section are available on 50687the PowerPC family of processors starting with ISA 2.07 or later. 50688Unless specific options are explicitly disabled on the command line, 50689specifying option '-mcpu=power8' has the effect of enabling all the same 50690options as for '-mcpu=power7' in addition to the '-mpower8-fusion', 50691'-mpower8-vector', '-mcrypto', '-mhtm', '-mquad-memory', and 50692'-mquad-memory-atomic' options. 50693 50694 This section intentionally empty. 50695 50696 50697File: gcc.info, Node: Basic PowerPC Built-in Functions Available on ISA 3.0, Next: Basic PowerPC Built-in Functions Available on ISA 3.1, Prev: Basic PowerPC Built-in Functions Available on ISA 2.07, Up: Basic PowerPC Built-in Functions 50698 506996.60.22.5 Basic PowerPC Built-in Functions Available on ISA 3.0 50700............................................................... 50701 50702The basic built-in functions described in this section are available on 50703the PowerPC family of processors starting with ISA 3.0 or later. Unless 50704specific options are explicitly disabled on the command line, specifying 50705option '-mcpu=power9' has the effect of enabling all the same options as 50706for '-mcpu=power8' in addition to the '-misel' option. 50707 50708 The following built-in functions are available on Linux 64-bit systems 50709that use the ISA 3.0 instruction set ('-mcpu=power9'): 50710 50711'__float128 __builtin_addf128_round_to_odd (__float128, __float128)' 50712 Perform a 128-bit IEEE floating point add using round to odd as the 50713 rounding mode. 50714 50715'__float128 __builtin_subf128_round_to_odd (__float128, __float128)' 50716 Perform a 128-bit IEEE floating point subtract using round to odd 50717 as the rounding mode. 50718 50719'__float128 __builtin_mulf128_round_to_odd (__float128, __float128)' 50720 Perform a 128-bit IEEE floating point multiply using round to odd 50721 as the rounding mode. 50722 50723'__float128 __builtin_divf128_round_to_odd (__float128, __float128)' 50724 Perform a 128-bit IEEE floating point divide using round to odd as 50725 the rounding mode. 50726 50727'__float128 __builtin_sqrtf128_round_to_odd (__float128)' 50728 Perform a 128-bit IEEE floating point square root using round to 50729 odd as the rounding mode. 50730 50731'__float128 __builtin_fmaf128_round_to_odd (__float128, __float128, __float128)' 50732 Perform a 128-bit IEEE floating point fused multiply and add 50733 operation using round to odd as the rounding mode. 50734 50735'double __builtin_truncf128_round_to_odd (__float128)' 50736 Convert a 128-bit IEEE floating point value to 'double' using round 50737 to odd as the rounding mode. 50738 50739 The following additional built-in functions are also available for the 50740PowerPC family of processors, starting with ISA 3.0 or later: 50741 long long __builtin_darn (void); 50742 long long __builtin_darn_raw (void); 50743 int __builtin_darn_32 (void); 50744 50745 The '__builtin_darn' and '__builtin_darn_raw' functions require a 5074664-bit environment supporting ISA 3.0 or later. The '__builtin_darn' 50747function provides a 64-bit conditioned random number. The 50748'__builtin_darn_raw' function provides a 64-bit raw random number. The 50749'__builtin_darn_32' function provides a 32-bit conditioned random 50750number. 50751 50752 The following additional built-in functions are also available for the 50753PowerPC family of processors, starting with ISA 3.0 or later: 50754 50755 int __builtin_byte_in_set (unsigned char u, unsigned long long set); 50756 int __builtin_byte_in_range (unsigned char u, unsigned int range); 50757 int __builtin_byte_in_either_range (unsigned char u, unsigned int ranges); 50758 50759 int __builtin_dfp_dtstsfi_lt (unsigned int comparison, _Decimal64 value); 50760 int __builtin_dfp_dtstsfi_lt (unsigned int comparison, _Decimal128 value); 50761 int __builtin_dfp_dtstsfi_lt_dd (unsigned int comparison, _Decimal64 value); 50762 int __builtin_dfp_dtstsfi_lt_td (unsigned int comparison, _Decimal128 value); 50763 50764 int __builtin_dfp_dtstsfi_gt (unsigned int comparison, _Decimal64 value); 50765 int __builtin_dfp_dtstsfi_gt (unsigned int comparison, _Decimal128 value); 50766 int __builtin_dfp_dtstsfi_gt_dd (unsigned int comparison, _Decimal64 value); 50767 int __builtin_dfp_dtstsfi_gt_td (unsigned int comparison, _Decimal128 value); 50768 50769 int __builtin_dfp_dtstsfi_eq (unsigned int comparison, _Decimal64 value); 50770 int __builtin_dfp_dtstsfi_eq (unsigned int comparison, _Decimal128 value); 50771 int __builtin_dfp_dtstsfi_eq_dd (unsigned int comparison, _Decimal64 value); 50772 int __builtin_dfp_dtstsfi_eq_td (unsigned int comparison, _Decimal128 value); 50773 50774 int __builtin_dfp_dtstsfi_ov (unsigned int comparison, _Decimal64 value); 50775 int __builtin_dfp_dtstsfi_ov (unsigned int comparison, _Decimal128 value); 50776 int __builtin_dfp_dtstsfi_ov_dd (unsigned int comparison, _Decimal64 value); 50777 int __builtin_dfp_dtstsfi_ov_td (unsigned int comparison, _Decimal128 value); 50778 50779 double __builtin_mffsl(void); 50780 50781 The '__builtin_byte_in_set' function requires a 64-bit environment 50782supporting ISA 3.0 or later. This function returns a non-zero value if 50783and only if its 'u' argument exactly equals one of the eight bytes 50784contained within its 64-bit 'set' argument. 50785 50786 The '__builtin_byte_in_range' and '__builtin_byte_in_either_range' 50787require an environment supporting ISA 3.0 or later. For these two 50788functions, the 'range' argument is encoded as 4 bytes, organized as 50789'hi_1:lo_1:hi_2:lo_2'. The '__builtin_byte_in_range' function returns a 50790non-zero value if and only if its 'u' argument is within the range 50791bounded between 'lo_2' and 'hi_2' inclusive. The 50792'__builtin_byte_in_either_range' function returns non-zero if and only 50793if its 'u' argument is within either the range bounded between 'lo_1' 50794and 'hi_1' inclusive or the range bounded between 'lo_2' and 'hi_2' 50795inclusive. 50796 50797 The '__builtin_dfp_dtstsfi_lt' function returns a non-zero value if and 50798only if the number of signficant digits of its 'value' argument is less 50799than its 'comparison' argument. The '__builtin_dfp_dtstsfi_lt_dd' and 50800'__builtin_dfp_dtstsfi_lt_td' functions behave similarly, but require 50801that the type of the 'value' argument be '__Decimal64' and 50802'__Decimal128' respectively. 50803 50804 The '__builtin_dfp_dtstsfi_gt' function returns a non-zero value if and 50805only if the number of signficant digits of its 'value' argument is 50806greater than its 'comparison' argument. The 50807'__builtin_dfp_dtstsfi_gt_dd' and '__builtin_dfp_dtstsfi_gt_td' 50808functions behave similarly, but require that the type of the 'value' 50809argument be '__Decimal64' and '__Decimal128' respectively. 50810 50811 The '__builtin_dfp_dtstsfi_eq' function returns a non-zero value if and 50812only if the number of signficant digits of its 'value' argument equals 50813its 'comparison' argument. The '__builtin_dfp_dtstsfi_eq_dd' and 50814'__builtin_dfp_dtstsfi_eq_td' functions behave similarly, but require 50815that the type of the 'value' argument be '__Decimal64' and 50816'__Decimal128' respectively. 50817 50818 The '__builtin_dfp_dtstsfi_ov' function returns a non-zero value if and 50819only if its 'value' argument has an undefined number of significant 50820digits, such as when 'value' is an encoding of 'NaN'. The 50821'__builtin_dfp_dtstsfi_ov_dd' and '__builtin_dfp_dtstsfi_ov_td' 50822functions behave similarly, but require that the type of the 'value' 50823argument be '__Decimal64' and '__Decimal128' respectively. 50824 50825 The '__builtin_mffsl' uses the ISA 3.0 'mffsl' instruction to read the 50826FPSCR. The instruction is a lower latency version of the 'mffs' 50827instruction. If the 'mffsl' instruction is not available, then the 50828builtin uses the older 'mffs' instruction to read the FPSCR. 50829 50830 50831File: gcc.info, Node: Basic PowerPC Built-in Functions Available on ISA 3.1, Prev: Basic PowerPC Built-in Functions Available on ISA 3.0, Up: Basic PowerPC Built-in Functions 50832 508336.60.22.6 Basic PowerPC Built-in Functions Available on ISA 3.1 50834............................................................... 50835 50836The basic built-in functions described in this section are available on 50837the PowerPC family of processors starting with ISA 3.1. Unless specific 50838options are explicitly disabled on the command line, specifying option 50839'-mcpu=power10' has the effect of enabling all the same options as for 50840'-mcpu=power9'. 50841 50842 The following built-in functions are available on Linux 64-bit systems 50843that use a future architecture instruction set ('-mcpu=power10'): 50844 50845unsigned long long 50846__builtin_cfuged (unsigned long long, unsigned long long) 50847 Perform a 64-bit centrifuge operation, as if implemented by the 50848'cfuged' instruction. 50849 50850unsigned long long 50851__builtin_cntlzdm (unsigned long long, unsigned long long) 50852 Perform a 64-bit count leading zeros operation under mask, as if 50853implemented by the 'cntlzdm' instruction. 50854 50855unsigned long long 50856__builtin_cnttzdm (unsigned long long, unsigned long long) 50857 Perform a 64-bit count trailing zeros operation under mask, as if 50858implemented by the 'cnttzdm' instruction. 50859 50860unsigned long long 50861__builtin_pdepd (unsigned long long, unsigned long long) 50862 Perform a 64-bit parallel bits deposit operation, as if implemented by 50863the 'pdepd' instruction. 50864 50865unsigned long long 50866__builtin_pextd (unsigned long long, unsigned long long) 50867 Perform a 64-bit parallel bits extract operation, as if implemented by 50868the 'pextd' instruction. 50869 50870vector signed __int128 vsx_xl_sext (signed long long, signed char *) 50871 50872vector signed __int128 vsx_xl_sext (signed long long, signed short *) 50873 50874vector signed __int128 vsx_xl_sext (signed long long, signed int *) 50875 50876vector signed __int128 vsx_xl_sext (signed long long, signed long long *) 50877 50878vector unsigned __int128 vsx_xl_zext (signed long long, unsigned char *) 50879 50880vector unsigned __int128 vsx_xl_zext (signed long long, unsigned short *) 50881 50882vector unsigned __int128 vsx_xl_zext (signed long long, unsigned int *) 50883 50884vector unsigned __int128 vsx_xl_zext (signed long long, unsigned long long *) 50885 50886 Load (and sign extend) to an __int128 vector, as if implemented by the 50887ISA 3.1 'lxvrbx', 'lxvrhx', 'lxvrwx', and 'lxvrdx' instructions. 50888 50889void vec_xst_trunc (vector signed __int128, signed long long, signed char *) 50890 50891void vec_xst_trunc (vector signed __int128, signed long long, signed short *) 50892 50893void vec_xst_trunc (vector signed __int128, signed long long, signed int *) 50894 50895void vec_xst_trunc (vector signed __int128, signed long long, signed long long *) 50896 50897void vec_xst_trunc (vector unsigned __int128, signed long long, unsigned char *) 50898 50899void vec_xst_trunc (vector unsigned __int128, signed long long, unsigned short *) 50900 50901void vec_xst_trunc (vector unsigned __int128, signed long long, unsigned int *) 50902 50903void vec_xst_trunc (vector unsigned __int128, signed long long, unsigned long long *) 50904 50905 Truncate and store the rightmost element of a vector, as if implemented 50906by the ISA 3.1 'stxvrbx', 'stxvrhx', 'stxvrwx', and 'stxvrdx' 50907instructions. 50908 50909 50910File: gcc.info, Node: PowerPC AltiVec/VSX Built-in Functions, Next: PowerPC Hardware Transactional Memory Built-in Functions, Prev: Basic PowerPC Built-in Functions, Up: Target Builtins 50911 509126.60.23 PowerPC AltiVec/VSX Built-in Functions 50913---------------------------------------------- 50914 50915GCC provides an interface for the PowerPC family of processors to access 50916the AltiVec operations described in Motorola's AltiVec Programming 50917Interface Manual. The interface is made available by including 50918'<altivec.h>' and using '-maltivec' and '-mabi=altivec'. The interface 50919supports the following vector types. 50920 50921 vector unsigned char 50922 vector signed char 50923 vector bool char 50924 50925 vector unsigned short 50926 vector signed short 50927 vector bool short 50928 vector pixel 50929 50930 vector unsigned int 50931 vector signed int 50932 vector bool int 50933 vector float 50934 50935 GCC's implementation of the high-level language interface available 50936from C and C++ code differs from Motorola's documentation in several 50937ways. 50938 50939 * A vector constant is a list of constant expressions within curly 50940 braces. 50941 50942 * A vector initializer requires no cast if the vector constant is of 50943 the same type as the variable it is initializing. 50944 50945 * If 'signed' or 'unsigned' is omitted, the signedness of the vector 50946 type is the default signedness of the base type. The default 50947 varies depending on the operating system, so a portable program 50948 should always specify the signedness. 50949 50950 * Compiling with '-maltivec' adds keywords '__vector', 'vector', 50951 '__pixel', 'pixel', '__bool' and 'bool'. When compiling ISO C, the 50952 context-sensitive substitution of the keywords 'vector', 'pixel' 50953 and 'bool' is disabled. To use them, you must include 50954 '<altivec.h>' instead. 50955 50956 * GCC allows using a 'typedef' name as the type specifier for a 50957 vector type, but only under the following circumstances: 50958 50959 * When using '__vector' instead of 'vector'; for example, 50960 50961 typedef signed short int16; 50962 __vector int16 data; 50963 50964 * When using 'vector' in keyword-and-predefine mode; for 50965 example, 50966 50967 typedef signed short int16; 50968 vector int16 data; 50969 50970 Note that keyword-and-predefine mode is enabled by disabling 50971 GNU extensions (e.g., by using '-std=c11') and including 50972 '<altivec.h>'. 50973 50974 * For C, overloaded functions are implemented with macros so the 50975 following does not work: 50976 50977 vec_add ((vector signed int){1, 2, 3, 4}, foo); 50978 50979 Since 'vec_add' is a macro, the vector constant in the example is 50980 treated as four separate arguments. Wrap the entire argument in 50981 parentheses for this to work. 50982 50983 _Note:_ Only the '<altivec.h>' interface is supported. Internally, GCC 50984uses built-in functions to achieve the functionality in the 50985aforementioned header file, but they are not supported and are subject 50986to change without notice. 50987 50988 GCC complies with the Power Vector Intrinsic Programming Reference 50989(PVIPR), which may be found at 50990<https://openpowerfoundation.org/?resource_lib=power-vector-intrinsic-programming-reference>. 50991Chapter 4 of this document fully documents the vector API interfaces 50992that must be provided by compliant compilers. Programmers should 50993preferentially use the interfaces described therein. However, 50994historically GCC has provided additional interfaces for access to vector 50995instructions. These are briefly described below. Where the PVIPR 50996provides a portable interface, other functions in GCC that provide the 50997same capabilities should be considered deprecated. 50998 50999 The PVIPR documents the following overloaded functions: 51000 51001'vec_abs' 'vec_absd' 'vec_abss' 51002'vec_add' 'vec_addc' 'vec_adde' 51003'vec_addec' 'vec_adds' 'vec_all_eq' 51004'vec_all_ge' 'vec_all_gt' 'vec_all_in' 51005'vec_all_le' 'vec_all_lt' 'vec_all_nan' 51006'vec_all_ne' 'vec_all_nge' 'vec_all_ngt' 51007'vec_all_nle' 'vec_all_nlt' 'vec_all_numeric' 51008'vec_and' 'vec_andc' 'vec_any_eq' 51009'vec_any_ge' 'vec_any_gt' 'vec_any_le' 51010'vec_any_lt' 'vec_any_nan' 'vec_any_ne' 51011'vec_any_nge' 'vec_any_ngt' 'vec_any_nle' 51012'vec_any_nlt' 'vec_any_numeric' 'vec_any_out' 51013'vec_avg' 'vec_bperm' 'vec_ceil' 51014'vec_cipher_be' 'vec_cipherlast_be' 'vec_cmpb' 51015'vec_cmpeq' 'vec_cmpge' 'vec_cmpgt' 51016'vec_cmple' 'vec_cmplt' 'vec_cmpne' 51017'vec_cmpnez' 'vec_cntlz' 'vec_cntlz_lsbb' 51018'vec_cnttz' 'vec_cnttz_lsbb' 'vec_cpsgn' 51019'vec_ctf' 'vec_cts' 'vec_ctu' 51020'vec_div' 'vec_double' 'vec_doublee' 51021'vec_doubleh' 'vec_doublel' 'vec_doubleo' 51022'vec_eqv' 'vec_expte' 'vec_extract' 51023'vec_extract_exp' 'vec_extract_fp32_from_shorth''vec_extract_fp32_from_shortl' 51024'vec_extract_sig' 'vec_extract_4b' 'vec_first_match_index' 51025'vec_first_match_or_eos_index''vec_first_mismatch_index''vec_first_mismatch_or_eos_index' 51026'vec_float' 'vec_float2' 'vec_floate' 51027'vec_floato' 'vec_floor' 'vec_gb' 51028'vec_insert' 'vec_insert_exp' 'vec_insert4b' 51029'vec_ld' 'vec_lde' 'vec_ldl' 51030'vec_loge' 'vec_madd' 'vec_madds' 51031'vec_max' 'vec_mergee' 'vec_mergeh' 51032'vec_mergel' 'vec_mergeo' 'vec_mfvscr' 51033'vec_min' 'vec_mradds' 'vec_msub' 51034'vec_msum' 'vec_msums' 'vec_mtvscr' 51035'vec_mul' 'vec_mule' 'vec_mulo' 51036'vec_nabs' 'vec_nand' 'vec_ncipher_be' 51037'vec_ncipherlast_be' 'vec_nearbyint' 'vec_neg' 51038'vec_nmadd' 'vec_nmsub' 'vec_nor' 51039'vec_or' 'vec_orc' 'vec_pack' 51040'vec_pack_to_short_fp32' 'vec_packpx' 'vec_packs' 51041'vec_packsu' 'vec_parity_lsbb' 'vec_perm' 51042'vec_permxor' 'vec_pmsum_be' 'vec_popcnt' 51043'vec_re' 'vec_recipdiv' 'vec_revb' 51044'vec_reve' 'vec_rint' 'vec_rl' 51045'vec_rlmi' 'vec_rlnm' 'vec_round' 51046'vec_rsqrt' 'vec_rsqrte' 'vec_sbox_be' 51047'vec_sel' 'vec_shasigma_be' 'vec_signed' 51048'vec_signed2' 'vec_signede' 'vec_signedo' 51049'vec_sl' 'vec_sld' 'vec_sldw' 51050'vec_sll' 'vec_slo' 'vec_slv' 51051'vec_splat' 'vec_splat_s8' 'vec_splat_s16' 51052'vec_splat_s32' 'vec_splat_u8' 'vec_splat_u16' 51053'vec_splat_u32' 'vec_splats' 'vec_sqrt' 51054'vec_sr' 'vec_sra' 'vec_srl' 51055'vec_sro' 'vec_srv' 'vec_st' 51056'vec_ste' 'vec_stl' 'vec_sub' 51057'vec_subc' 'vec_sube' 'vec_subec' 51058'vec_subs' 'vec_sum2s' 'vec_sum4s' 51059'vec_sums' 'vec_test_data_class' 'vec_trunc' 51060'vec_unpackh' 'vec_unpackl' 'vec_unsigned' 51061'vec_unsigned2' 'vec_unsignede' 'vec_unsignedo' 51062'vec_xl' 'vec_xl_be' 'vec_xl_len' 51063'vec_xl_len_r' 'vec_xor' 'vec_xst' 51064'vec_xst_be' 'vec_xst_len' 'vec_xst_len_r' 51065 51066 51067* Menu: 51068 51069* PowerPC AltiVec Built-in Functions on ISA 2.05:: 51070* PowerPC AltiVec Built-in Functions Available on ISA 2.06:: 51071* PowerPC AltiVec Built-in Functions Available on ISA 2.07:: 51072* PowerPC AltiVec Built-in Functions Available on ISA 3.0:: 51073* PowerPC AltiVec Built-in Functions Available on ISA 3.1:: 51074 51075 51076File: gcc.info, Node: PowerPC AltiVec Built-in Functions on ISA 2.05, Next: PowerPC AltiVec Built-in Functions Available on ISA 2.06, Up: PowerPC AltiVec/VSX Built-in Functions 51077 510786.60.23.1 PowerPC AltiVec Built-in Functions on ISA 2.05 51079........................................................ 51080 51081The following interfaces are supported for the generic and specific 51082AltiVec operations and the AltiVec predicates. In cases where there is 51083a direct mapping between generic and specific operations, only the 51084generic names are shown here, although the specific operations can also 51085be used. 51086 51087 Arguments that are documented as 'const int' require literal integral 51088values within the range required for that operation. 51089 51090 Only functions excluded from the PVIPR are listed here. 51091 51092 void vec_dss (const int); 51093 51094 void vec_dssall (void); 51095 51096 void vec_dst (const vector unsigned char *, int, const int); 51097 void vec_dst (const vector signed char *, int, const int); 51098 void vec_dst (const vector bool char *, int, const int); 51099 void vec_dst (const vector unsigned short *, int, const int); 51100 void vec_dst (const vector signed short *, int, const int); 51101 void vec_dst (const vector bool short *, int, const int); 51102 void vec_dst (const vector pixel *, int, const int); 51103 void vec_dst (const vector unsigned int *, int, const int); 51104 void vec_dst (const vector signed int *, int, const int); 51105 void vec_dst (const vector bool int *, int, const int); 51106 void vec_dst (const vector float *, int, const int); 51107 void vec_dst (const unsigned char *, int, const int); 51108 void vec_dst (const signed char *, int, const int); 51109 void vec_dst (const unsigned short *, int, const int); 51110 void vec_dst (const short *, int, const int); 51111 void vec_dst (const unsigned int *, int, const int); 51112 void vec_dst (const int *, int, const int); 51113 void vec_dst (const float *, int, const int); 51114 51115 void vec_dstst (const vector unsigned char *, int, const int); 51116 void vec_dstst (const vector signed char *, int, const int); 51117 void vec_dstst (const vector bool char *, int, const int); 51118 void vec_dstst (const vector unsigned short *, int, const int); 51119 void vec_dstst (const vector signed short *, int, const int); 51120 void vec_dstst (const vector bool short *, int, const int); 51121 void vec_dstst (const vector pixel *, int, const int); 51122 void vec_dstst (const vector unsigned int *, int, const int); 51123 void vec_dstst (const vector signed int *, int, const int); 51124 void vec_dstst (const vector bool int *, int, const int); 51125 void vec_dstst (const vector float *, int, const int); 51126 void vec_dstst (const unsigned char *, int, const int); 51127 void vec_dstst (const signed char *, int, const int); 51128 void vec_dstst (const unsigned short *, int, const int); 51129 void vec_dstst (const short *, int, const int); 51130 void vec_dstst (const unsigned int *, int, const int); 51131 void vec_dstst (const int *, int, const int); 51132 void vec_dstst (const unsigned long *, int, const int); 51133 void vec_dstst (const long *, int, const int); 51134 void vec_dstst (const float *, int, const int); 51135 51136 void vec_dststt (const vector unsigned char *, int, const int); 51137 void vec_dststt (const vector signed char *, int, const int); 51138 void vec_dststt (const vector bool char *, int, const int); 51139 void vec_dststt (const vector unsigned short *, int, const int); 51140 void vec_dststt (const vector signed short *, int, const int); 51141 void vec_dststt (const vector bool short *, int, const int); 51142 void vec_dststt (const vector pixel *, int, const int); 51143 void vec_dststt (const vector unsigned int *, int, const int); 51144 void vec_dststt (const vector signed int *, int, const int); 51145 void vec_dststt (const vector bool int *, int, const int); 51146 void vec_dststt (const vector float *, int, const int); 51147 void vec_dststt (const unsigned char *, int, const int); 51148 void vec_dststt (const signed char *, int, const int); 51149 void vec_dststt (const unsigned short *, int, const int); 51150 void vec_dststt (const short *, int, const int); 51151 void vec_dststt (const unsigned int *, int, const int); 51152 void vec_dststt (const int *, int, const int); 51153 void vec_dststt (const float *, int, const int); 51154 51155 void vec_dstt (const vector unsigned char *, int, const int); 51156 void vec_dstt (const vector signed char *, int, const int); 51157 void vec_dstt (const vector bool char *, int, const int); 51158 void vec_dstt (const vector unsigned short *, int, const int); 51159 void vec_dstt (const vector signed short *, int, const int); 51160 void vec_dstt (const vector bool short *, int, const int); 51161 void vec_dstt (const vector pixel *, int, const int); 51162 void vec_dstt (const vector unsigned int *, int, const int); 51163 void vec_dstt (const vector signed int *, int, const int); 51164 void vec_dstt (const vector bool int *, int, const int); 51165 void vec_dstt (const vector float *, int, const int); 51166 void vec_dstt (const unsigned char *, int, const int); 51167 void vec_dstt (const signed char *, int, const int); 51168 void vec_dstt (const unsigned short *, int, const int); 51169 void vec_dstt (const short *, int, const int); 51170 void vec_dstt (const unsigned int *, int, const int); 51171 void vec_dstt (const int *, int, const int); 51172 void vec_dstt (const float *, int, const int); 51173 51174 vector signed char vec_lvebx (int, char *); 51175 vector unsigned char vec_lvebx (int, unsigned char *); 51176 51177 vector signed short vec_lvehx (int, short *); 51178 vector unsigned short vec_lvehx (int, unsigned short *); 51179 51180 vector float vec_lvewx (int, float *); 51181 vector signed int vec_lvewx (int, int *); 51182 vector unsigned int vec_lvewx (int, unsigned int *); 51183 51184 vector unsigned char vec_lvsl (int, const unsigned char *); 51185 vector unsigned char vec_lvsl (int, const signed char *); 51186 vector unsigned char vec_lvsl (int, const unsigned short *); 51187 vector unsigned char vec_lvsl (int, const short *); 51188 vector unsigned char vec_lvsl (int, const unsigned int *); 51189 vector unsigned char vec_lvsl (int, const int *); 51190 vector unsigned char vec_lvsl (int, const float *); 51191 51192 vector unsigned char vec_lvsr (int, const unsigned char *); 51193 vector unsigned char vec_lvsr (int, const signed char *); 51194 vector unsigned char vec_lvsr (int, const unsigned short *); 51195 vector unsigned char vec_lvsr (int, const short *); 51196 vector unsigned char vec_lvsr (int, const unsigned int *); 51197 vector unsigned char vec_lvsr (int, const int *); 51198 vector unsigned char vec_lvsr (int, const float *); 51199 51200 void vec_stvebx (vector signed char, int, signed char *); 51201 void vec_stvebx (vector unsigned char, int, unsigned char *); 51202 void vec_stvebx (vector bool char, int, signed char *); 51203 void vec_stvebx (vector bool char, int, unsigned char *); 51204 51205 void vec_stvehx (vector signed short, int, short *); 51206 void vec_stvehx (vector unsigned short, int, unsigned short *); 51207 void vec_stvehx (vector bool short, int, short *); 51208 void vec_stvehx (vector bool short, int, unsigned short *); 51209 51210 void vec_stvewx (vector float, int, float *); 51211 void vec_stvewx (vector signed int, int, int *); 51212 void vec_stvewx (vector unsigned int, int, unsigned int *); 51213 void vec_stvewx (vector bool int, int, int *); 51214 void vec_stvewx (vector bool int, int, unsigned int *); 51215 51216 vector float vec_vaddfp (vector float, vector float); 51217 51218 vector signed char vec_vaddsbs (vector bool char, vector signed char); 51219 vector signed char vec_vaddsbs (vector signed char, vector bool char); 51220 vector signed char vec_vaddsbs (vector signed char, vector signed char); 51221 51222 vector signed short vec_vaddshs (vector bool short, vector signed short); 51223 vector signed short vec_vaddshs (vector signed short, vector bool short); 51224 vector signed short vec_vaddshs (vector signed short, vector signed short); 51225 51226 vector signed int vec_vaddsws (vector bool int, vector signed int); 51227 vector signed int vec_vaddsws (vector signed int, vector bool int); 51228 vector signed int vec_vaddsws (vector signed int, vector signed int); 51229 51230 vector signed char vec_vaddubm (vector bool char, vector signed char); 51231 vector signed char vec_vaddubm (vector signed char, vector bool char); 51232 vector signed char vec_vaddubm (vector signed char, vector signed char); 51233 vector unsigned char vec_vaddubm (vector bool char, vector unsigned char); 51234 vector unsigned char vec_vaddubm (vector unsigned char, vector bool char); 51235 vector unsigned char vec_vaddubm (vector unsigned char, vector unsigned char); 51236 51237 vector unsigned char vec_vaddubs (vector bool char, vector unsigned char); 51238 vector unsigned char vec_vaddubs (vector unsigned char, vector bool char); 51239 vector unsigned char vec_vaddubs (vector unsigned char, vector unsigned char); 51240 51241 vector signed short vec_vadduhm (vector bool short, vector signed short); 51242 vector signed short vec_vadduhm (vector signed short, vector bool short); 51243 vector signed short vec_vadduhm (vector signed short, vector signed short); 51244 vector unsigned short vec_vadduhm (vector bool short, vector unsigned short); 51245 vector unsigned short vec_vadduhm (vector unsigned short, vector bool short); 51246 vector unsigned short vec_vadduhm (vector unsigned short, vector unsigned short); 51247 51248 vector unsigned short vec_vadduhs (vector bool short, vector unsigned short); 51249 vector unsigned short vec_vadduhs (vector unsigned short, vector bool short); 51250 vector unsigned short vec_vadduhs (vector unsigned short, vector unsigned short); 51251 51252 vector signed int vec_vadduwm (vector bool int, vector signed int); 51253 vector signed int vec_vadduwm (vector signed int, vector bool int); 51254 vector signed int vec_vadduwm (vector signed int, vector signed int); 51255 vector unsigned int vec_vadduwm (vector bool int, vector unsigned int); 51256 vector unsigned int vec_vadduwm (vector unsigned int, vector bool int); 51257 vector unsigned int vec_vadduwm (vector unsigned int, vector unsigned int); 51258 51259 vector unsigned int vec_vadduws (vector bool int, vector unsigned int); 51260 vector unsigned int vec_vadduws (vector unsigned int, vector bool int); 51261 vector unsigned int vec_vadduws (vector unsigned int, vector unsigned int); 51262 51263 vector signed char vec_vavgsb (vector signed char, vector signed char); 51264 51265 vector signed short vec_vavgsh (vector signed short, vector signed short); 51266 51267 vector signed int vec_vavgsw (vector signed int, vector signed int); 51268 51269 vector unsigned char vec_vavgub (vector unsigned char, vector unsigned char); 51270 51271 vector unsigned short vec_vavguh (vector unsigned short, vector unsigned short); 51272 51273 vector unsigned int vec_vavguw (vector unsigned int, vector unsigned int); 51274 51275 vector float vec_vcfsx (vector signed int, const int); 51276 51277 vector float vec_vcfux (vector unsigned int, const int); 51278 51279 vector bool int vec_vcmpeqfp (vector float, vector float); 51280 51281 vector bool char vec_vcmpequb (vector signed char, vector signed char); 51282 vector bool char vec_vcmpequb (vector unsigned char, vector unsigned char); 51283 51284 vector bool short vec_vcmpequh (vector signed short, vector signed short); 51285 vector bool short vec_vcmpequh (vector unsigned short, vector unsigned short); 51286 51287 vector bool int vec_vcmpequw (vector signed int, vector signed int); 51288 vector bool int vec_vcmpequw (vector unsigned int, vector unsigned int); 51289 51290 vector bool int vec_vcmpgtfp (vector float, vector float); 51291 51292 vector bool char vec_vcmpgtsb (vector signed char, vector signed char); 51293 51294 vector bool short vec_vcmpgtsh (vector signed short, vector signed short); 51295 51296 vector bool int vec_vcmpgtsw (vector signed int, vector signed int); 51297 51298 vector bool char vec_vcmpgtub (vector unsigned char, vector unsigned char); 51299 51300 vector bool short vec_vcmpgtuh (vector unsigned short, vector unsigned short); 51301 51302 vector bool int vec_vcmpgtuw (vector unsigned int, vector unsigned int); 51303 51304 vector float vec_vmaxfp (vector float, vector float); 51305 51306 vector signed char vec_vmaxsb (vector bool char, vector signed char); 51307 vector signed char vec_vmaxsb (vector signed char, vector bool char); 51308 vector signed char vec_vmaxsb (vector signed char, vector signed char); 51309 51310 vector signed short vec_vmaxsh (vector bool short, vector signed short); 51311 vector signed short vec_vmaxsh (vector signed short, vector bool short); 51312 vector signed short vec_vmaxsh (vector signed short, vector signed short); 51313 51314 vector signed int vec_vmaxsw (vector bool int, vector signed int); 51315 vector signed int vec_vmaxsw (vector signed int, vector bool int); 51316 vector signed int vec_vmaxsw (vector signed int, vector signed int); 51317 51318 vector unsigned char vec_vmaxub (vector bool char, vector unsigned char); 51319 vector unsigned char vec_vmaxub (vector unsigned char, vector bool char); 51320 vector unsigned char vec_vmaxub (vector unsigned char, vector unsigned char); 51321 51322 vector unsigned short vec_vmaxuh (vector bool short, vector unsigned short); 51323 vector unsigned short vec_vmaxuh (vector unsigned short, vector bool short); 51324 vector unsigned short vec_vmaxuh (vector unsigned short, vector unsigned short); 51325 51326 vector unsigned int vec_vmaxuw (vector bool int, vector unsigned int); 51327 vector unsigned int vec_vmaxuw (vector unsigned int, vector bool int); 51328 vector unsigned int vec_vmaxuw (vector unsigned int, vector unsigned int); 51329 51330 vector float vec_vminfp (vector float, vector float); 51331 51332 vector signed char vec_vminsb (vector bool char, vector signed char); 51333 vector signed char vec_vminsb (vector signed char, vector bool char); 51334 vector signed char vec_vminsb (vector signed char, vector signed char); 51335 51336 vector signed short vec_vminsh (vector bool short, vector signed short); 51337 vector signed short vec_vminsh (vector signed short, vector bool short); 51338 vector signed short vec_vminsh (vector signed short, vector signed short); 51339 51340 vector signed int vec_vminsw (vector bool int, vector signed int); 51341 vector signed int vec_vminsw (vector signed int, vector bool int); 51342 vector signed int vec_vminsw (vector signed int, vector signed int); 51343 51344 vector unsigned char vec_vminub (vector bool char, vector unsigned char); 51345 vector unsigned char vec_vminub (vector unsigned char, vector bool char); 51346 vector unsigned char vec_vminub (vector unsigned char, vector unsigned char); 51347 51348 vector unsigned short vec_vminuh (vector bool short, vector unsigned short); 51349 vector unsigned short vec_vminuh (vector unsigned short, vector bool short); 51350 vector unsigned short vec_vminuh (vector unsigned short, vector unsigned short); 51351 51352 vector unsigned int vec_vminuw (vector bool int, vector unsigned int); 51353 vector unsigned int vec_vminuw (vector unsigned int, vector bool int); 51354 vector unsigned int vec_vminuw (vector unsigned int, vector unsigned int); 51355 51356 vector bool char vec_vmrghb (vector bool char, vector bool char); 51357 vector signed char vec_vmrghb (vector signed char, vector signed char); 51358 vector unsigned char vec_vmrghb (vector unsigned char, vector unsigned char); 51359 51360 vector bool short vec_vmrghh (vector bool short, vector bool short); 51361 vector signed short vec_vmrghh (vector signed short, vector signed short); 51362 vector unsigned short vec_vmrghh (vector unsigned short, vector unsigned short); 51363 vector pixel vec_vmrghh (vector pixel, vector pixel); 51364 51365 vector float vec_vmrghw (vector float, vector float); 51366 vector bool int vec_vmrghw (vector bool int, vector bool int); 51367 vector signed int vec_vmrghw (vector signed int, vector signed int); 51368 vector unsigned int vec_vmrghw (vector unsigned int, vector unsigned int); 51369 51370 vector bool char vec_vmrglb (vector bool char, vector bool char); 51371 vector signed char vec_vmrglb (vector signed char, vector signed char); 51372 vector unsigned char vec_vmrglb (vector unsigned char, vector unsigned char); 51373 51374 vector bool short vec_vmrglh (vector bool short, vector bool short); 51375 vector signed short vec_vmrglh (vector signed short, vector signed short); 51376 vector unsigned short vec_vmrglh (vector unsigned short, vector unsigned short); 51377 vector pixel vec_vmrglh (vector pixel, vector pixel); 51378 51379 vector float vec_vmrglw (vector float, vector float); 51380 vector signed int vec_vmrglw (vector signed int, vector signed int); 51381 vector unsigned int vec_vmrglw (vector unsigned int, vector unsigned int); 51382 vector bool int vec_vmrglw (vector bool int, vector bool int); 51383 51384 vector signed int vec_vmsummbm (vector signed char, vector unsigned char, 51385 vector signed int); 51386 51387 vector signed int vec_vmsumshm (vector signed short, vector signed short, 51388 vector signed int); 51389 51390 vector signed int vec_vmsumshs (vector signed short, vector signed short, 51391 vector signed int); 51392 51393 vector unsigned int vec_vmsumubm (vector unsigned char, vector unsigned char, 51394 vector unsigned int); 51395 51396 vector unsigned int vec_vmsumuhm (vector unsigned short, vector unsigned short, 51397 vector unsigned int); 51398 51399 vector unsigned int vec_vmsumuhs (vector unsigned short, vector unsigned short, 51400 vector unsigned int); 51401 51402 vector signed short vec_vmulesb (vector signed char, vector signed char); 51403 51404 vector signed int vec_vmulesh (vector signed short, vector signed short); 51405 51406 vector unsigned short vec_vmuleub (vector unsigned char, vector unsigned char); 51407 51408 vector unsigned int vec_vmuleuh (vector unsigned short, vector unsigned short); 51409 51410 vector signed short vec_vmulosb (vector signed char, vector signed char); 51411 51412 vector signed int vec_vmulosh (vector signed short, vector signed short); 51413 51414 vector unsigned short vec_vmuloub (vector unsigned char, vector unsigned char); 51415 51416 vector unsigned int vec_vmulouh (vector unsigned short, vector unsigned short); 51417 51418 vector signed char vec_vpkshss (vector signed short, vector signed short); 51419 51420 vector unsigned char vec_vpkshus (vector signed short, vector signed short); 51421 51422 vector signed short vec_vpkswss (vector signed int, vector signed int); 51423 51424 vector unsigned short vec_vpkswus (vector signed int, vector signed int); 51425 51426 vector bool char vec_vpkuhum (vector bool short, vector bool short); 51427 vector signed char vec_vpkuhum (vector signed short, vector signed short); 51428 vector unsigned char vec_vpkuhum (vector unsigned short, vector unsigned short); 51429 51430 vector unsigned char vec_vpkuhus (vector unsigned short, vector unsigned short); 51431 51432 vector bool short vec_vpkuwum (vector bool int, vector bool int); 51433 vector signed short vec_vpkuwum (vector signed int, vector signed int); 51434 vector unsigned short vec_vpkuwum (vector unsigned int, vector unsigned int); 51435 51436 vector unsigned short vec_vpkuwus (vector unsigned int, vector unsigned int); 51437 51438 vector signed char vec_vrlb (vector signed char, vector unsigned char); 51439 vector unsigned char vec_vrlb (vector unsigned char, vector unsigned char); 51440 51441 vector signed short vec_vrlh (vector signed short, vector unsigned short); 51442 vector unsigned short vec_vrlh (vector unsigned short, vector unsigned short); 51443 51444 vector signed int vec_vrlw (vector signed int, vector unsigned int); 51445 vector unsigned int vec_vrlw (vector unsigned int, vector unsigned int); 51446 51447 vector signed char vec_vslb (vector signed char, vector unsigned char); 51448 vector unsigned char vec_vslb (vector unsigned char, vector unsigned char); 51449 51450 vector signed short vec_vslh (vector signed short, vector unsigned short); 51451 vector unsigned short vec_vslh (vector unsigned short, vector unsigned short); 51452 51453 vector signed int vec_vslw (vector signed int, vector unsigned int); 51454 vector unsigned int vec_vslw (vector unsigned int, vector unsigned int); 51455 51456 vector signed char vec_vspltb (vector signed char, const int); 51457 vector unsigned char vec_vspltb (vector unsigned char, const int); 51458 vector bool char vec_vspltb (vector bool char, const int); 51459 51460 vector bool short vec_vsplth (vector bool short, const int); 51461 vector signed short vec_vsplth (vector signed short, const int); 51462 vector unsigned short vec_vsplth (vector unsigned short, const int); 51463 vector pixel vec_vsplth (vector pixel, const int); 51464 51465 vector float vec_vspltw (vector float, const int); 51466 vector signed int vec_vspltw (vector signed int, const int); 51467 vector unsigned int vec_vspltw (vector unsigned int, const int); 51468 vector bool int vec_vspltw (vector bool int, const int); 51469 51470 vector signed char vec_vsrab (vector signed char, vector unsigned char); 51471 vector unsigned char vec_vsrab (vector unsigned char, vector unsigned char); 51472 51473 vector signed short vec_vsrah (vector signed short, vector unsigned short); 51474 vector unsigned short vec_vsrah (vector unsigned short, vector unsigned short); 51475 51476 vector signed int vec_vsraw (vector signed int, vector unsigned int); 51477 vector unsigned int vec_vsraw (vector unsigned int, vector unsigned int); 51478 51479 vector signed char vec_vsrb (vector signed char, vector unsigned char); 51480 vector unsigned char vec_vsrb (vector unsigned char, vector unsigned char); 51481 51482 vector signed short vec_vsrh (vector signed short, vector unsigned short); 51483 vector unsigned short vec_vsrh (vector unsigned short, vector unsigned short); 51484 51485 vector signed int vec_vsrw (vector signed int, vector unsigned int); 51486 vector unsigned int vec_vsrw (vector unsigned int, vector unsigned int); 51487 51488 vector float vec_vsubfp (vector float, vector float); 51489 51490 vector signed char vec_vsubsbs (vector bool char, vector signed char); 51491 vector signed char vec_vsubsbs (vector signed char, vector bool char); 51492 vector signed char vec_vsubsbs (vector signed char, vector signed char); 51493 51494 vector signed short vec_vsubshs (vector bool short, vector signed short); 51495 vector signed short vec_vsubshs (vector signed short, vector bool short); 51496 vector signed short vec_vsubshs (vector signed short, vector signed short); 51497 51498 vector signed int vec_vsubsws (vector bool int, vector signed int); 51499 vector signed int vec_vsubsws (vector signed int, vector bool int); 51500 vector signed int vec_vsubsws (vector signed int, vector signed int); 51501 51502 vector signed char vec_vsububm (vector bool char, vector signed char); 51503 vector signed char vec_vsububm (vector signed char, vector bool char); 51504 vector signed char vec_vsububm (vector signed char, vector signed char); 51505 vector unsigned char vec_vsububm (vector bool char, vector unsigned char); 51506 vector unsigned char vec_vsububm (vector unsigned char, vector bool char); 51507 vector unsigned char vec_vsububm (vector unsigned char, vector unsigned char); 51508 51509 vector unsigned char vec_vsububs (vector bool char, vector unsigned char); 51510 vector unsigned char vec_vsububs (vector unsigned char, vector bool char); 51511 vector unsigned char vec_vsububs (vector unsigned char, vector unsigned char); 51512 51513 vector signed short vec_vsubuhm (vector bool short, vector signed short); 51514 vector signed short vec_vsubuhm (vector signed short, vector bool short); 51515 vector signed short vec_vsubuhm (vector signed short, vector signed short); 51516 vector unsigned short vec_vsubuhm (vector bool short, vector unsigned short); 51517 vector unsigned short vec_vsubuhm (vector unsigned short, vector bool short); 51518 vector unsigned short vec_vsubuhm (vector unsigned short, vector unsigned short); 51519 51520 vector unsigned short vec_vsubuhs (vector bool short, vector unsigned short); 51521 vector unsigned short vec_vsubuhs (vector unsigned short, vector bool short); 51522 vector unsigned short vec_vsubuhs (vector unsigned short, vector unsigned short); 51523 51524 vector signed int vec_vsubuwm (vector bool int, vector signed int); 51525 vector signed int vec_vsubuwm (vector signed int, vector bool int); 51526 vector signed int vec_vsubuwm (vector signed int, vector signed int); 51527 vector unsigned int vec_vsubuwm (vector bool int, vector unsigned int); 51528 vector unsigned int vec_vsubuwm (vector unsigned int, vector bool int); 51529 vector unsigned int vec_vsubuwm (vector unsigned int, vector unsigned int); 51530 51531 vector unsigned int vec_vsubuws (vector bool int, vector unsigned int); 51532 vector unsigned int vec_vsubuws (vector unsigned int, vector bool int); 51533 vector unsigned int vec_vsubuws (vector unsigned int, vector unsigned int); 51534 51535 vector signed int vec_vsum4sbs (vector signed char, vector signed int); 51536 51537 vector signed int vec_vsum4shs (vector signed short, vector signed int); 51538 51539 vector unsigned int vec_vsum4ubs (vector unsigned char, vector unsigned int); 51540 51541 vector unsigned int vec_vupkhpx (vector pixel); 51542 51543 vector bool short vec_vupkhsb (vector bool char); 51544 vector signed short vec_vupkhsb (vector signed char); 51545 51546 vector bool int vec_vupkhsh (vector bool short); 51547 vector signed int vec_vupkhsh (vector signed short); 51548 51549 vector unsigned int vec_vupklpx (vector pixel); 51550 51551 vector bool short vec_vupklsb (vector bool char); 51552 vector signed short vec_vupklsb (vector signed char); 51553 51554 vector bool int vec_vupklsh (vector bool short); 51555 vector signed int vec_vupklsh (vector signed short); 51556 51557 51558File: gcc.info, Node: PowerPC AltiVec Built-in Functions Available on ISA 2.06, Next: PowerPC AltiVec Built-in Functions Available on ISA 2.07, Prev: PowerPC AltiVec Built-in Functions on ISA 2.05, Up: PowerPC AltiVec/VSX Built-in Functions 51559 515606.60.23.2 PowerPC AltiVec Built-in Functions Available on ISA 2.06 51561.................................................................. 51562 51563The AltiVec built-in functions described in this section are available 51564on the PowerPC family of processors starting with ISA 2.06 or later. 51565These are normally enabled by adding '-mvsx' to the command line. 51566 51567 When '-mvsx' is used, the following additional vector types are 51568implemented. 51569 51570 vector unsigned __int128 51571 vector signed __int128 51572 vector unsigned long long int 51573 vector signed long long int 51574 vector double 51575 51576 The long long types are only implemented for 64-bit code generation. 51577 51578 Only functions excluded from the PVIPR are listed here. 51579 51580 void vec_dst (const unsigned long *, int, const int); 51581 void vec_dst (const long *, int, const int); 51582 51583 void vec_dststt (const unsigned long *, int, const int); 51584 void vec_dststt (const long *, int, const int); 51585 51586 void vec_dstt (const unsigned long *, int, const int); 51587 void vec_dstt (const long *, int, const int); 51588 51589 vector unsigned char vec_lvsl (int, const unsigned long *); 51590 vector unsigned char vec_lvsl (int, const long *); 51591 51592 vector unsigned char vec_lvsr (int, const unsigned long *); 51593 vector unsigned char vec_lvsr (int, const long *); 51594 51595 vector unsigned char vec_lvsl (int, const double *); 51596 vector unsigned char vec_lvsr (int, const double *); 51597 51598 vector double vec_vsx_ld (int, const vector double *); 51599 vector double vec_vsx_ld (int, const double *); 51600 vector float vec_vsx_ld (int, const vector float *); 51601 vector float vec_vsx_ld (int, const float *); 51602 vector bool int vec_vsx_ld (int, const vector bool int *); 51603 vector signed int vec_vsx_ld (int, const vector signed int *); 51604 vector signed int vec_vsx_ld (int, const int *); 51605 vector signed int vec_vsx_ld (int, const long *); 51606 vector unsigned int vec_vsx_ld (int, const vector unsigned int *); 51607 vector unsigned int vec_vsx_ld (int, const unsigned int *); 51608 vector unsigned int vec_vsx_ld (int, const unsigned long *); 51609 vector bool short vec_vsx_ld (int, const vector bool short *); 51610 vector pixel vec_vsx_ld (int, const vector pixel *); 51611 vector signed short vec_vsx_ld (int, const vector signed short *); 51612 vector signed short vec_vsx_ld (int, const short *); 51613 vector unsigned short vec_vsx_ld (int, const vector unsigned short *); 51614 vector unsigned short vec_vsx_ld (int, const unsigned short *); 51615 vector bool char vec_vsx_ld (int, const vector bool char *); 51616 vector signed char vec_vsx_ld (int, const vector signed char *); 51617 vector signed char vec_vsx_ld (int, const signed char *); 51618 vector unsigned char vec_vsx_ld (int, const vector unsigned char *); 51619 vector unsigned char vec_vsx_ld (int, const unsigned char *); 51620 51621 void vec_vsx_st (vector double, int, vector double *); 51622 void vec_vsx_st (vector double, int, double *); 51623 void vec_vsx_st (vector float, int, vector float *); 51624 void vec_vsx_st (vector float, int, float *); 51625 void vec_vsx_st (vector signed int, int, vector signed int *); 51626 void vec_vsx_st (vector signed int, int, int *); 51627 void vec_vsx_st (vector unsigned int, int, vector unsigned int *); 51628 void vec_vsx_st (vector unsigned int, int, unsigned int *); 51629 void vec_vsx_st (vector bool int, int, vector bool int *); 51630 void vec_vsx_st (vector bool int, int, unsigned int *); 51631 void vec_vsx_st (vector bool int, int, int *); 51632 void vec_vsx_st (vector signed short, int, vector signed short *); 51633 void vec_vsx_st (vector signed short, int, short *); 51634 void vec_vsx_st (vector unsigned short, int, vector unsigned short *); 51635 void vec_vsx_st (vector unsigned short, int, unsigned short *); 51636 void vec_vsx_st (vector bool short, int, vector bool short *); 51637 void vec_vsx_st (vector bool short, int, unsigned short *); 51638 void vec_vsx_st (vector pixel, int, vector pixel *); 51639 void vec_vsx_st (vector pixel, int, unsigned short *); 51640 void vec_vsx_st (vector pixel, int, short *); 51641 void vec_vsx_st (vector bool short, int, short *); 51642 void vec_vsx_st (vector signed char, int, vector signed char *); 51643 void vec_vsx_st (vector signed char, int, signed char *); 51644 void vec_vsx_st (vector unsigned char, int, vector unsigned char *); 51645 void vec_vsx_st (vector unsigned char, int, unsigned char *); 51646 void vec_vsx_st (vector bool char, int, vector bool char *); 51647 void vec_vsx_st (vector bool char, int, unsigned char *); 51648 void vec_vsx_st (vector bool char, int, signed char *); 51649 51650 vector double vec_xxpermdi (vector double, vector double, const int); 51651 vector float vec_xxpermdi (vector float, vector float, const int); 51652 vector long long vec_xxpermdi (vector long long, vector long long, const int); 51653 vector unsigned long long vec_xxpermdi (vector unsigned long long, 51654 vector unsigned long long, const int); 51655 vector int vec_xxpermdi (vector int, vector int, const int); 51656 vector unsigned int vec_xxpermdi (vector unsigned int, 51657 vector unsigned int, const int); 51658 vector short vec_xxpermdi (vector short, vector short, const int); 51659 vector unsigned short vec_xxpermdi (vector unsigned short, 51660 vector unsigned short, const int); 51661 vector signed char vec_xxpermdi (vector signed char, vector signed char, 51662 const int); 51663 vector unsigned char vec_xxpermdi (vector unsigned char, 51664 vector unsigned char, const int); 51665 51666 vector double vec_xxsldi (vector double, vector double, int); 51667 vector float vec_xxsldi (vector float, vector float, int); 51668 vector long long vec_xxsldi (vector long long, vector long long, int); 51669 vector unsigned long long vec_xxsldi (vector unsigned long long, 51670 vector unsigned long long, int); 51671 vector int vec_xxsldi (vector int, vector int, int); 51672 vector unsigned int vec_xxsldi (vector unsigned int, vector unsigned int, int); 51673 vector short vec_xxsldi (vector short, vector short, int); 51674 vector unsigned short vec_xxsldi (vector unsigned short, 51675 vector unsigned short, int); 51676 vector signed char vec_xxsldi (vector signed char, vector signed char, int); 51677 vector unsigned char vec_xxsldi (vector unsigned char, 51678 vector unsigned char, int); 51679 51680 Note that the 'vec_ld' and 'vec_st' built-in functions always generate 51681the AltiVec 'LVX' and 'STVX' instructions even if the VSX instruction 51682set is available. The 'vec_vsx_ld' and 'vec_vsx_st' built-in functions 51683always generate the VSX 'LXVD2X', 'LXVW4X', 'STXVD2X', and 'STXVW4X' 51684instructions. 51685 51686 51687File: gcc.info, Node: PowerPC AltiVec Built-in Functions Available on ISA 2.07, Next: PowerPC AltiVec Built-in Functions Available on ISA 3.0, Prev: PowerPC AltiVec Built-in Functions Available on ISA 2.06, Up: PowerPC AltiVec/VSX Built-in Functions 51688 516896.60.23.3 PowerPC AltiVec Built-in Functions Available on ISA 2.07 51690.................................................................. 51691 51692If the ISA 2.07 additions to the vector/scalar (power8-vector) 51693instruction set are available, the following additional functions are 51694available for both 32-bit and 64-bit targets. For 64-bit targets, you 51695can use VECTOR LONG instead of VECTOR LONG LONG, VECTOR BOOL LONG 51696instead of VECTOR BOOL LONG LONG, and VECTOR UNSIGNED LONG instead of 51697VECTOR UNSIGNED LONG LONG. 51698 51699 Only functions excluded from the PVIPR are listed here. 51700 51701 vector long long vec_vaddudm (vector long long, vector long long); 51702 vector long long vec_vaddudm (vector bool long long, vector long long); 51703 vector long long vec_vaddudm (vector long long, vector bool long long); 51704 vector unsigned long long vec_vaddudm (vector unsigned long long, 51705 vector unsigned long long); 51706 vector unsigned long long vec_vaddudm (vector bool unsigned long long, 51707 vector unsigned long long); 51708 vector unsigned long long vec_vaddudm (vector unsigned long long, 51709 vector bool unsigned long long); 51710 51711 vector long long vec_vclz (vector long long); 51712 vector unsigned long long vec_vclz (vector unsigned long long); 51713 vector int vec_vclz (vector int); 51714 vector unsigned int vec_vclz (vector int); 51715 vector short vec_vclz (vector short); 51716 vector unsigned short vec_vclz (vector unsigned short); 51717 vector signed char vec_vclz (vector signed char); 51718 vector unsigned char vec_vclz (vector unsigned char); 51719 51720 vector signed char vec_vclzb (vector signed char); 51721 vector unsigned char vec_vclzb (vector unsigned char); 51722 51723 vector long long vec_vclzd (vector long long); 51724 vector unsigned long long vec_vclzd (vector unsigned long long); 51725 51726 vector short vec_vclzh (vector short); 51727 vector unsigned short vec_vclzh (vector unsigned short); 51728 51729 vector int vec_vclzw (vector int); 51730 vector unsigned int vec_vclzw (vector int); 51731 51732 vector signed char vec_vgbbd (vector signed char); 51733 vector unsigned char vec_vgbbd (vector unsigned char); 51734 51735 vector long long vec_vmaxsd (vector long long, vector long long); 51736 51737 vector unsigned long long vec_vmaxud (vector unsigned long long, 51738 unsigned vector long long); 51739 51740 vector long long vec_vminsd (vector long long, vector long long); 51741 51742 vector unsigned long long vec_vminud (vector long long, vector long long); 51743 51744 vector int vec_vpksdss (vector long long, vector long long); 51745 vector unsigned int vec_vpksdss (vector long long, vector long long); 51746 51747 vector unsigned int vec_vpkudus (vector unsigned long long, 51748 vector unsigned long long); 51749 51750 vector int vec_vpkudum (vector long long, vector long long); 51751 vector unsigned int vec_vpkudum (vector unsigned long long, 51752 vector unsigned long long); 51753 vector bool int vec_vpkudum (vector bool long long, vector bool long long); 51754 51755 vector long long vec_vpopcnt (vector long long); 51756 vector unsigned long long vec_vpopcnt (vector unsigned long long); 51757 vector int vec_vpopcnt (vector int); 51758 vector unsigned int vec_vpopcnt (vector int); 51759 vector short vec_vpopcnt (vector short); 51760 vector unsigned short vec_vpopcnt (vector unsigned short); 51761 vector signed char vec_vpopcnt (vector signed char); 51762 vector unsigned char vec_vpopcnt (vector unsigned char); 51763 51764 vector signed char vec_vpopcntb (vector signed char); 51765 vector unsigned char vec_vpopcntb (vector unsigned char); 51766 51767 vector long long vec_vpopcntd (vector long long); 51768 vector unsigned long long vec_vpopcntd (vector unsigned long long); 51769 51770 vector short vec_vpopcnth (vector short); 51771 vector unsigned short vec_vpopcnth (vector unsigned short); 51772 51773 vector int vec_vpopcntw (vector int); 51774 vector unsigned int vec_vpopcntw (vector int); 51775 51776 vector long long vec_vrld (vector long long, vector unsigned long long); 51777 vector unsigned long long vec_vrld (vector unsigned long long, 51778 vector unsigned long long); 51779 51780 vector long long vec_vsld (vector long long, vector unsigned long long); 51781 vector long long vec_vsld (vector unsigned long long, 51782 vector unsigned long long); 51783 51784 vector long long vec_vsrad (vector long long, vector unsigned long long); 51785 vector unsigned long long vec_vsrad (vector unsigned long long, 51786 vector unsigned long long); 51787 51788 vector long long vec_vsrd (vector long long, vector unsigned long long); 51789 vector unsigned long long char vec_vsrd (vector unsigned long long, 51790 vector unsigned long long); 51791 51792 vector long long vec_vsubudm (vector long long, vector long long); 51793 vector long long vec_vsubudm (vector bool long long, vector long long); 51794 vector long long vec_vsubudm (vector long long, vector bool long long); 51795 vector unsigned long long vec_vsubudm (vector unsigned long long, 51796 vector unsigned long long); 51797 vector unsigned long long vec_vsubudm (vector bool long long, 51798 vector unsigned long long); 51799 vector unsigned long long vec_vsubudm (vector unsigned long long, 51800 vector bool long long); 51801 51802 vector long long vec_vupkhsw (vector int); 51803 vector unsigned long long vec_vupkhsw (vector unsigned int); 51804 51805 vector long long vec_vupklsw (vector int); 51806 vector unsigned long long vec_vupklsw (vector int); 51807 51808 If the ISA 2.07 additions to the vector/scalar (power8-vector) 51809instruction set are available, the following additional functions are 51810available for 64-bit targets. New vector types (VECTOR __INT128 and 51811VECTOR __UINT128) are available to hold the __INT128 and __UINT128 types 51812to use these builtins. 51813 51814 The normal vector extract, and set operations work on VECTOR __INT128 51815and VECTOR __UINT128 types, but the index value must be 0. 51816 51817 Only functions excluded from the PVIPR are listed here. 51818 51819 vector __int128 vec_vaddcuq (vector __int128, vector __int128); 51820 vector __uint128 vec_vaddcuq (vector __uint128, vector __uint128); 51821 51822 vector __int128 vec_vadduqm (vector __int128, vector __int128); 51823 vector __uint128 vec_vadduqm (vector __uint128, vector __uint128); 51824 51825 vector __int128 vec_vaddecuq (vector __int128, vector __int128, 51826 vector __int128); 51827 vector __uint128 vec_vaddecuq (vector __uint128, vector __uint128, 51828 vector __uint128); 51829 51830 vector __int128 vec_vaddeuqm (vector __int128, vector __int128, 51831 vector __int128); 51832 vector __uint128 vec_vaddeuqm (vector __uint128, vector __uint128, 51833 vector __uint128); 51834 51835 vector __int128 vec_vsubecuq (vector __int128, vector __int128, 51836 vector __int128); 51837 vector __uint128 vec_vsubecuq (vector __uint128, vector __uint128, 51838 vector __uint128); 51839 51840 vector __int128 vec_vsubeuqm (vector __int128, vector __int128, 51841 vector __int128); 51842 vector __uint128 vec_vsubeuqm (vector __uint128, vector __uint128, 51843 vector __uint128); 51844 51845 vector __int128 vec_vsubcuq (vector __int128, vector __int128); 51846 vector __uint128 vec_vsubcuq (vector __uint128, vector __uint128); 51847 51848 __int128 vec_vsubuqm (__int128, __int128); 51849 __uint128 vec_vsubuqm (__uint128, __uint128); 51850 51851 vector __int128 __builtin_bcdadd (vector __int128, vector __int128, const int); 51852 vector unsigned char __builtin_bcdadd (vector unsigned char, vector unsigned char, 51853 const int); 51854 int __builtin_bcdadd_lt (vector __int128, vector __int128, const int); 51855 int __builtin_bcdadd_lt (vector unsigned char, vector unsigned char, const int); 51856 int __builtin_bcdadd_eq (vector __int128, vector __int128, const int); 51857 int __builtin_bcdadd_eq (vector unsigned char, vector unsigned char, const int); 51858 int __builtin_bcdadd_gt (vector __int128, vector __int128, const int); 51859 int __builtin_bcdadd_gt (vector unsigned char, vector unsigned char, const int); 51860 int __builtin_bcdadd_ov (vector __int128, vector __int128, const int); 51861 int __builtin_bcdadd_ov (vector unsigned char, vector unsigned char, const int); 51862 51863 vector __int128 __builtin_bcdsub (vector __int128, vector __int128, const int); 51864 vector unsigned char __builtin_bcdsub (vector unsigned char, vector unsigned char, 51865 const int); 51866 int __builtin_bcdsub_lt (vector __int128, vector __int128, const int); 51867 int __builtin_bcdsub_lt (vector unsigned char, vector unsigned char, const int); 51868 int __builtin_bcdsub_eq (vector __int128, vector __int128, const int); 51869 int __builtin_bcdsub_eq (vector unsigned char, vector unsigned char, const int); 51870 int __builtin_bcdsub_gt (vector __int128, vector __int128, const int); 51871 int __builtin_bcdsub_gt (vector unsigned char, vector unsigned char, const int); 51872 int __builtin_bcdsub_ov (vector __int128, vector __int128, const int); 51873 int __builtin_bcdsub_ov (vector unsigned char, vector unsigned char, const int); 51874 51875 51876File: gcc.info, Node: PowerPC AltiVec Built-in Functions Available on ISA 3.0, Next: PowerPC AltiVec Built-in Functions Available on ISA 3.1, Prev: PowerPC AltiVec Built-in Functions Available on ISA 2.07, Up: PowerPC AltiVec/VSX Built-in Functions 51877 518786.60.23.4 PowerPC AltiVec Built-in Functions Available on ISA 3.0 51879................................................................. 51880 51881The following additional built-in functions are also available for the 51882PowerPC family of processors, starting with ISA 3.0 ('-mcpu=power9') or 51883later. 51884 51885 Only instructions excluded from the PVIPR are listed here. 51886 51887 unsigned int scalar_extract_exp (double source); 51888 unsigned long long int scalar_extract_exp (__ieee128 source); 51889 51890 unsigned long long int scalar_extract_sig (double source); 51891 unsigned __int128 scalar_extract_sig (__ieee128 source); 51892 51893 double scalar_insert_exp (unsigned long long int significand, 51894 unsigned long long int exponent); 51895 double scalar_insert_exp (double significand, unsigned long long int exponent); 51896 51897 ieee_128 scalar_insert_exp (unsigned __int128 significand, 51898 unsigned long long int exponent); 51899 ieee_128 scalar_insert_exp (ieee_128 significand, unsigned long long int exponent); 51900 51901 int scalar_cmp_exp_gt (double arg1, double arg2); 51902 int scalar_cmp_exp_lt (double arg1, double arg2); 51903 int scalar_cmp_exp_eq (double arg1, double arg2); 51904 int scalar_cmp_exp_unordered (double arg1, double arg2); 51905 51906 bool scalar_test_data_class (float source, const int condition); 51907 bool scalar_test_data_class (double source, const int condition); 51908 bool scalar_test_data_class (__ieee128 source, const int condition); 51909 51910 bool scalar_test_neg (float source); 51911 bool scalar_test_neg (double source); 51912 bool scalar_test_neg (__ieee128 source); 51913 51914 The 'scalar_extract_exp' and 'scalar_extract_sig' functions require a 5191564-bit environment supporting ISA 3.0 or later. The 51916'scalar_extract_exp' and 'scalar_extract_sig' built-in functions return 51917the significand and the biased exponent value respectively of their 51918'source' arguments. When supplied with a 64-bit 'source' argument, the 51919result returned by 'scalar_extract_sig' has the '0x0010000000000000' bit 51920set if the function's 'source' argument is in normalized form. 51921Otherwise, this bit is set to 0. When supplied with a 128-bit 'source' 51922argument, the '0x00010000000000000000000000000000' bit of the result is 51923treated similarly. Note that the sign of the significand is not 51924represented in the result returned from the 'scalar_extract_sig' 51925function. Use the 'scalar_test_neg' function to test the sign of its 51926'double' argument. 51927 51928 The 'scalar_insert_exp' functions require a 64-bit environment 51929supporting ISA 3.0 or later. When supplied with a 64-bit first 51930argument, the 'scalar_insert_exp' built-in function returns a 51931double-precision floating point value that is constructed by assembling 51932the values of its 'significand' and 'exponent' arguments. The sign of 51933the result is copied from the most significant bit of the 'significand' 51934argument. The significand and exponent components of the result are 51935composed of the least significant 11 bits of the 'exponent' argument and 51936the least significant 52 bits of the 'significand' argument 51937respectively. 51938 51939 When supplied with a 128-bit first argument, the 'scalar_insert_exp' 51940built-in function returns a quad-precision ieee floating point value. 51941The sign bit of the result is copied from the most significant bit of 51942the 'significand' argument. The significand and exponent components of 51943the result are composed of the least significant 15 bits of the 51944'exponent' argument and the least significant 112 bits of the 51945'significand' argument respectively. 51946 51947 The 'scalar_cmp_exp_gt', 'scalar_cmp_exp_lt', 'scalar_cmp_exp_eq', and 51948'scalar_cmp_exp_unordered' built-in functions return a non-zero value if 51949'arg1' is greater than, less than, equal to, or not comparable to 'arg2' 51950respectively. The arguments are not comparable if one or the other 51951equals NaN (not a number). 51952 51953 The 'scalar_test_data_class' built-in function returns 1 if any of the 51954condition tests enabled by the value of the 'condition' variable are 51955true, and 0 otherwise. The 'condition' argument must be a compile-time 51956constant integer with value not exceeding 127. The 'condition' argument 51957is encoded as a bitmask with each bit enabling the testing of a 51958different condition, as characterized by the following: 51959 0x40 Test for NaN 51960 0x20 Test for +Infinity 51961 0x10 Test for -Infinity 51962 0x08 Test for +Zero 51963 0x04 Test for -Zero 51964 0x02 Test for +Denormal 51965 0x01 Test for -Denormal 51966 51967 The 'scalar_test_neg' built-in function returns 1 if its 'source' 51968argument holds a negative value, 0 otherwise. 51969 51970 The following built-in functions are also available for the PowerPC 51971family of processors, starting with ISA 3.0 or later ('-mcpu=power9'). 51972These string functions are described separately in order to group the 51973descriptions closer to the function prototypes. 51974 51975 Only functions excluded from the PVIPR are listed here. 51976 51977 int vec_all_nez (vector signed char, vector signed char); 51978 int vec_all_nez (vector unsigned char, vector unsigned char); 51979 int vec_all_nez (vector signed short, vector signed short); 51980 int vec_all_nez (vector unsigned short, vector unsigned short); 51981 int vec_all_nez (vector signed int, vector signed int); 51982 int vec_all_nez (vector unsigned int, vector unsigned int); 51983 51984 int vec_any_eqz (vector signed char, vector signed char); 51985 int vec_any_eqz (vector unsigned char, vector unsigned char); 51986 int vec_any_eqz (vector signed short, vector signed short); 51987 int vec_any_eqz (vector unsigned short, vector unsigned short); 51988 int vec_any_eqz (vector signed int, vector signed int); 51989 int vec_any_eqz (vector unsigned int, vector unsigned int); 51990 51991 signed char vec_xlx (unsigned int index, vector signed char data); 51992 unsigned char vec_xlx (unsigned int index, vector unsigned char data); 51993 signed short vec_xlx (unsigned int index, vector signed short data); 51994 unsigned short vec_xlx (unsigned int index, vector unsigned short data); 51995 signed int vec_xlx (unsigned int index, vector signed int data); 51996 unsigned int vec_xlx (unsigned int index, vector unsigned int data); 51997 float vec_xlx (unsigned int index, vector float data); 51998 51999 signed char vec_xrx (unsigned int index, vector signed char data); 52000 unsigned char vec_xrx (unsigned int index, vector unsigned char data); 52001 signed short vec_xrx (unsigned int index, vector signed short data); 52002 unsigned short vec_xrx (unsigned int index, vector unsigned short data); 52003 signed int vec_xrx (unsigned int index, vector signed int data); 52004 unsigned int vec_xrx (unsigned int index, vector unsigned int data); 52005 float vec_xrx (unsigned int index, vector float data); 52006 52007 The 'vec_all_nez', 'vec_any_eqz', and 'vec_cmpnez' perform pairwise 52008comparisons between the elements at the same positions within their two 52009vector arguments. The 'vec_all_nez' function returns a non-zero value 52010if and only if all pairwise comparisons are not equal and no element of 52011either vector argument contains a zero. The 'vec_any_eqz' function 52012returns a non-zero value if and only if at least one pairwise comparison 52013is equal or if at least one element of either vector argument contains a 52014zero. The 'vec_cmpnez' function returns a vector of the same type as 52015its two arguments, within which each element consists of all ones to 52016denote that either the corresponding elements of the incoming arguments 52017are not equal or that at least one of the corresponding elements 52018contains zero. Otherwise, the element of the returned vector contains 52019all zeros. 52020 52021 The 'vec_xlx' and 'vec_xrx' functions extract the single element 52022selected by the 'index' argument from the vector represented by the 52023'data' argument. The 'index' argument always specifies a byte offset, 52024regardless of the size of the vector element. With 'vec_xlx', 'index' 52025is the offset of the first byte of the element to be extracted. With 52026'vec_xrx', 'index' represents the last byte of the element to be 52027extracted, measured from the right end of the vector. In other words, 52028the last byte of the element to be extracted is found at position '(15 - 52029index)'. There is no requirement that 'index' be a multiple of the 52030vector element size. However, if the size of the vector element added 52031to 'index' is greater than 15, the content of the returned value is 52032undefined. 52033 52034 The following functions are also available if the ISA 3.0 instruction 52035set additions ('-mcpu=power9') are available. 52036 52037 Only functions excluded from the PVIPR are listed here. 52038 52039 vector long long vec_vctz (vector long long); 52040 vector unsigned long long vec_vctz (vector unsigned long long); 52041 vector int vec_vctz (vector int); 52042 vector unsigned int vec_vctz (vector int); 52043 vector short vec_vctz (vector short); 52044 vector unsigned short vec_vctz (vector unsigned short); 52045 vector signed char vec_vctz (vector signed char); 52046 vector unsigned char vec_vctz (vector unsigned char); 52047 52048 vector signed char vec_vctzb (vector signed char); 52049 vector unsigned char vec_vctzb (vector unsigned char); 52050 52051 vector long long vec_vctzd (vector long long); 52052 vector unsigned long long vec_vctzd (vector unsigned long long); 52053 52054 vector short vec_vctzh (vector short); 52055 vector unsigned short vec_vctzh (vector unsigned short); 52056 52057 vector int vec_vctzw (vector int); 52058 vector unsigned int vec_vctzw (vector int); 52059 52060 vector int vec_vprtyb (vector int); 52061 vector unsigned int vec_vprtyb (vector unsigned int); 52062 vector long long vec_vprtyb (vector long long); 52063 vector unsigned long long vec_vprtyb (vector unsigned long long); 52064 52065 vector int vec_vprtybw (vector int); 52066 vector unsigned int vec_vprtybw (vector unsigned int); 52067 52068 vector long long vec_vprtybd (vector long long); 52069 vector unsigned long long vec_vprtybd (vector unsigned long long); 52070 52071 On 64-bit targets, if the ISA 3.0 additions ('-mcpu=power9') are 52072available: 52073 52074 vector long vec_vprtyb (vector long); 52075 vector unsigned long vec_vprtyb (vector unsigned long); 52076 vector __int128 vec_vprtyb (vector __int128); 52077 vector __uint128 vec_vprtyb (vector __uint128); 52078 52079 vector long vec_vprtybd (vector long); 52080 vector unsigned long vec_vprtybd (vector unsigned long); 52081 52082 vector __int128 vec_vprtybq (vector __int128); 52083 vector __uint128 vec_vprtybd (vector __uint128); 52084 52085 The following built-in functions are available for the PowerPC family 52086of processors, starting with ISA 3.0 or later ('-mcpu=power9'). 52087 52088 Only functions excluded from the PVIPR are listed here. 52089 52090 __vector unsigned char 52091 vec_absdb (__vector unsigned char arg1, __vector unsigned char arg2); 52092 __vector unsigned short 52093 vec_absdh (__vector unsigned short arg1, __vector unsigned short arg2); 52094 __vector unsigned int 52095 vec_absdw (__vector unsigned int arg1, __vector unsigned int arg2); 52096 52097 The 'vec_absd', 'vec_absdb', 'vec_absdh', and 'vec_absdw' built-in 52098functions each computes the absolute differences of the pairs of vector 52099elements supplied in its two vector arguments, placing the absolute 52100differences into the corresponding elements of the vector result. 52101 52102 The following built-in functions are available for the PowerPC family 52103of processors, starting with ISA 3.0 or later ('-mcpu=power9'): 52104 vector unsigned int vec_vrlnm (vector unsigned int, vector unsigned int); 52105 vector unsigned long long vec_vrlnm (vector unsigned long long, 52106 vector unsigned long long); 52107 52108 The result of 'vec_vrlnm' is obtained by rotating each element of the 52109first argument vector left and ANDing it with a mask. The second 52110argument vector contains the mask beginning in bits 11:15, the mask end 52111in bits 19:23, and the shift count in bits 27:31, of each element. 52112 52113 If the cryptographic instructions are enabled ('-mcrypto' or 52114'-mcpu=power8'), the following builtins are enabled. 52115 52116 Only functions excluded from the PVIPR are listed here. 52117 52118 vector unsigned long long __builtin_crypto_vsbox (vector unsigned long long); 52119 52120 vector unsigned long long __builtin_crypto_vcipher (vector unsigned long long, 52121 vector unsigned long long); 52122 52123 vector unsigned long long __builtin_crypto_vcipherlast 52124 (vector unsigned long long, 52125 vector unsigned long long); 52126 52127 vector unsigned long long __builtin_crypto_vncipher (vector unsigned long long, 52128 vector unsigned long long); 52129 52130 vector unsigned long long __builtin_crypto_vncipherlast (vector unsigned long long, 52131 vector unsigned long long); 52132 52133 vector unsigned char __builtin_crypto_vpermxor (vector unsigned char, 52134 vector unsigned char, 52135 vector unsigned char); 52136 52137 vector unsigned short __builtin_crypto_vpermxor (vector unsigned short, 52138 vector unsigned short, 52139 vector unsigned short); 52140 52141 vector unsigned int __builtin_crypto_vpermxor (vector unsigned int, 52142 vector unsigned int, 52143 vector unsigned int); 52144 52145 vector unsigned long long __builtin_crypto_vpermxor (vector unsigned long long, 52146 vector unsigned long long, 52147 vector unsigned long long); 52148 52149 vector unsigned char __builtin_crypto_vpmsumb (vector unsigned char, 52150 vector unsigned char); 52151 52152 vector unsigned short __builtin_crypto_vpmsumh (vector unsigned short, 52153 vector unsigned short); 52154 52155 vector unsigned int __builtin_crypto_vpmsumw (vector unsigned int, 52156 vector unsigned int); 52157 52158 vector unsigned long long __builtin_crypto_vpmsumd (vector unsigned long long, 52159 vector unsigned long long); 52160 52161 vector unsigned long long __builtin_crypto_vshasigmad (vector unsigned long long, 52162 int, int); 52163 52164 vector unsigned int __builtin_crypto_vshasigmaw (vector unsigned int, int, int); 52165 52166 The second argument to __BUILTIN_CRYPTO_VSHASIGMAD and 52167__BUILTIN_CRYPTO_VSHASIGMAW must be a constant integer that is 0 or 1. 52168The third argument to these built-in functions must be a constant 52169integer in the range of 0 to 15. 52170 52171 The following sign extension builtins are provided: 52172 52173 vector signed int vec_signexti (vector signed char a); 52174 vector signed long long vec_signextll (vector signed char a); 52175 vector signed int vec_signexti (vector signed short a); 52176 vector signed long long vec_signextll (vector signed short a); 52177 vector signed long long vec_signextll (vector signed int a); 52178 vector signed long long vec_signextq (vector signed long long a); 52179 52180 Each element of the result is produced by sign-extending the element of 52181the input vector that would fall in the least significant portion of the 52182result element. For example, a sign-extension of a vector signed char 52183to a vector signed long long will sign extend the rightmost byte of each 52184doubleword. 52185 52186 52187File: gcc.info, Node: PowerPC AltiVec Built-in Functions Available on ISA 3.1, Prev: PowerPC AltiVec Built-in Functions Available on ISA 3.0, Up: PowerPC AltiVec/VSX Built-in Functions 52188 521896.60.23.5 PowerPC AltiVec Built-in Functions Available on ISA 3.1 52190................................................................. 52191 52192The following additional built-in functions are also available for the 52193PowerPC family of processors, starting with ISA 3.1 ('-mcpu=power10'): 52194 52195vector unsigned long long int 52196vec_cfuge (vector unsigned long long int, vector unsigned long long int); 52197 Perform a vector centrifuge operation, as if implemented by the 52198'vcfuged' instruction. 52199 52200vector unsigned long long int 52201vec_cntlzm (vector unsigned long long int, vector unsigned long long int); 52202 Perform a vector count leading zeros under bit mask operation, as if 52203implemented by the 'vclzdm' instruction. 52204 52205vector unsigned long long int 52206vec_cnttzm (vector unsigned long long int, vector unsigned long long int); 52207 Perform a vector count trailing zeros under bit mask operation, as if 52208implemented by the 'vctzdm' instruction. 52209 52210vector signed char 52211vec_clrl (vector signed char a, unsigned int n); 52212vector unsigned char 52213vec_clrl (vector unsigned char a, unsigned int n); 52214 Clear the left-most '(16 - n)' bytes of vector argument 'a', as if 52215implemented by the 'vclrlb' instruction on a big-endian target and by 52216the 'vclrrb' instruction on a little-endian target. A value of 'n' that 52217is greater than 16 is treated as if it equaled 16. 52218 52219vector signed char 52220vec_clrr (vector signed char a, unsigned int n); 52221vector unsigned char 52222vec_clrr (vector unsigned char a, unsigned int n); 52223 Clear the right-most '(16 - n)' bytes of vector argument 'a', as if 52224implemented by the 'vclrrb' instruction on a big-endian target and by 52225the 'vclrlb' instruction on a little-endian target. A value of 'n' that 52226is greater than 16 is treated as if it equaled 16. 52227 52228vector unsigned long long int 52229vec_gnb (vector unsigned __int128, const unsigned char); 52230 Perform a 128-bit vector gather operation, as if implemented by the 52231'vgnb' instruction. The second argument must be a literal integer value 52232between 2 and 7 inclusive. 52233 52234 Vector Extract 52235 52236vector unsigned long long int 52237vec_extractl (vector unsigned char, vector unsigned char, unsigned int); 52238vector unsigned long long int 52239vec_extractl (vector unsigned short, vector unsigned short, unsigned int); 52240vector unsigned long long int 52241vec_extractl (vector unsigned int, vector unsigned int, unsigned int); 52242vector unsigned long long int 52243vec_extractl (vector unsigned long long, vector unsigned long long, unsigned int); 52244 Extract an element from two concatenated vectors starting at the given 52245byte index in natural-endian order, and place it zero-extended in 52246doubleword 1 of the result according to natural element order. If the 52247byte index is out of range for the data type, the intrinsic will be 52248rejected. For little-endian, this output will match the placement by 52249the hardware instruction, i.e., dword[0] in RTL notation. For 52250big-endian, an additional instruction is needed to move it from the 52251"left" doubleword to the "right" one. For little-endian, semantics 52252matching the 'vextdubvrx', 'vextduhvrx', 'vextduwvrx' instruction will 52253be generated, while for big-endian, semantics matching the 'vextdubvlx', 52254'vextduhvlx', 'vextduwvlx' instructions will be generated. Note that 52255some fairly anomalous results can be generated if the byte index is not 52256aligned on an element boundary for the element being extracted. This is 52257a limitation of the bi-endian vector programming model is consistent 52258with the limitation on 'vec_perm'. 52259 52260vector unsigned long long int 52261vec_extracth (vector unsigned char, vector unsigned char, unsigned int); 52262vector unsigned long long int 52263vec_extracth (vector unsigned short, vector unsigned short, 52264 unsigned int); 52265vector unsigned long long int 52266vec_extracth (vector unsigned int, vector unsigned int, unsigned int); 52267vector unsigned long long int 52268vec_extracth (vector unsigned long long, vector unsigned long long, 52269 unsigned int); 52270 Extract an element from two concatenated vectors starting at the given 52271byte index. The index is based on big endian order for a little endian 52272system. Similarly, the index is based on little endian order for a big 52273endian system. The extraced elements are zero-extended and put in 52274doubleword 1 according to natural element order. If the byte index is 52275out of range for the data type, the intrinsic will be rejected. For 52276little-endian, this output will match the placement by the hardware 52277instruction (vextdubvrx, vextduhvrx, vextduwvrx, vextddvrx) i.e., 52278dword[0] in RTL notation. For big-endian, an additional instruction is 52279needed to move it from the "left" doubleword to the "right" one. For 52280little-endian, semantics matching the 'vextdubvlx', 'vextduhvlx', 52281'vextduwvlx' instructions will be generated, while for big-endian, 52282semantics matching the 'vextdubvrx', 'vextduhvrx', 'vextduwvrx' 52283instructions will be generated. Note that some fairly anomalous results 52284can be generated if the byte index is not aligned on the element 52285boundary for the element being extracted. This is a limitation of the 52286bi-endian vector programming model consistent with the limitation on 52287'vec_perm'. 52288vector unsigned long long int 52289vec_pdep (vector unsigned long long int, vector unsigned long long int); 52290 Perform a vector parallel bits deposit operation, as if implemented by 52291the 'vpdepd' instruction. 52292 52293 Vector Insert 52294 52295vector unsigned char 52296vec_insertl (unsigned char, vector unsigned char, unsigned int); 52297vector unsigned short 52298vec_insertl (unsigned short, vector unsigned short, unsigned int); 52299vector unsigned int 52300vec_insertl (unsigned int, vector unsigned int, unsigned int); 52301vector unsigned long long 52302vec_insertl (unsigned long long, vector unsigned long long, 52303 unsigned int); 52304vector unsigned char 52305vec_insertl (vector unsigned char, vector unsigned char, unsigned int; 52306vector unsigned short 52307vec_insertl (vector unsigned short, vector unsigned short, 52308 unsigned int); 52309vector unsigned int 52310vec_insertl (vector unsigned int, vector unsigned int, unsigned int); 52311 52312 Let src be the first argument, when the first argument is a scalar, or 52313the rightmost element of the left doubleword of the first argument, when 52314the first argument is a vector. Insert the source into the destination 52315at the position given by the third argument, using natural element order 52316in the second argument. The rest of the second argument is unchanged. 52317If the byte index is greater than 14 for halfwords, greater than 12 for 52318words, or greater than 8 for doublewords the result is undefined. For 52319little-endian, the generated code will be semantically equivalent to 52320'vins[bhwd]rx' instructions. Similarly for big-endian it will be 52321semantically equivalent to 'vins[bhwd]lx'. Note that some fairly 52322anomalous results can be generated if the byte index is not aligned on 52323an element boundary for the type of element being inserted. 52324 52325vector unsigned char 52326vec_inserth (unsigned char, vector unsigned char, unsigned int); 52327vector unsigned short 52328vec_inserth (unsigned short, vector unsigned short, unsigned int); 52329vector unsigned int 52330vec_inserth (unsigned int, vector unsigned int, unsigned int); 52331vector unsigned long long 52332vec_inserth (unsigned long long, vector unsigned long long, 52333 unsigned int); 52334vector unsigned char 52335vec_inserth (vector unsigned char, vector unsigned char, unsigned int); 52336vector unsigned short 52337vec_inserth (vector unsigned short, vector unsigned short, 52338 unsigned int); 52339vector unsigned int 52340vec_inserth (vector unsigned int, vector unsigned int, unsigned int); 52341 52342 Let src be the first argument, when the first argument is a scalar, or 52343the rightmost element of the first argument, when the first argument is 52344a vector. Insert src into the second argument at the position 52345identified by the third argument, using opposite element order in the 52346second argument, and leaving the rest of the second argument unchanged. 52347If the byte index is greater than 14 for halfwords, 12 for words, or 8 52348for doublewords, the intrinsic will be rejected. Note that the 52349underlying hardware instruction uses the same register for the second 52350argument and the result. For little-endian, the code generation will be 52351semantically equivalent to 'vins[bhwd]lx', while for big-endian it will 52352be semantically equivalent to 'vins[bhwd]rx'. Note that some fairly 52353anomalous results can be generated if the byte index is not aligned on 52354an element boundary for the sort of element being inserted. 52355 52356 Vector Replace Element 52357vector signed int vec_replace_elt (vector signed int, signed int, 52358 const int); 52359vector unsigned int vec_replace_elt (vector unsigned int, 52360 unsigned int, const int); 52361vector float vec_replace_elt (vector float, float, const int); 52362vector signed long long vec_replace_elt (vector signed long long, 52363 signed long long, const int); 52364vector unsigned long long vec_replace_elt (vector unsigned long long, 52365 unsigned long long, const int); 52366vector double rec_replace_elt (vector double, double, const int); 52367 The third argument (constrained to [0,3]) identifies the natural-endian 52368element number of the first argument that will be replaced by the second 52369argument to produce the result. The other elements of the first 52370argument will remain unchanged in the result. 52371 52372 If it's desirable to insert a word at an unaligned position, use 52373vec_replace_unaligned instead. 52374 52375 Vector Replace Unaligned 52376vector unsigned char vec_replace_unaligned (vector unsigned char, 52377 signed int, const int); 52378vector unsigned char vec_replace_unaligned (vector unsigned char, 52379 unsigned int, const int); 52380vector unsigned char vec_replace_unaligned (vector unsigned char, 52381 float, const int); 52382vector unsigned char vec_replace_unaligned (vector unsigned char, 52383 signed long long, const int); 52384vector unsigned char vec_replace_unaligned (vector unsigned char, 52385 unsigned long long, const int); 52386vector unsigned char vec_replace_unaligned (vector unsigned char, 52387 double, const int); 52388 52389 The second argument replaces a portion of the first argument to produce 52390the result, with the rest of the first argument unchanged in the result. 52391The third argument identifies the byte index (using left-to-right, or 52392big-endian order) where the high-order byte of the second argument will 52393be placed, with the remaining bytes of the second argument placed 52394naturally "to the right" of the high-order byte. 52395 52396 The programmer is responsible for understanding the endianness issues 52397involved with the first argument and the result. 52398 52399 Vector Shift Left Double Bit Immediate 52400vector signed char vec_sldb (vector signed char, vector signed char, 52401 const unsigned int); 52402vector unsigned char vec_sldb (vector unsigned char, 52403 vector unsigned char, const unsigned int); 52404vector signed short vec_sldb (vector signed short, vector signed short, 52405 const unsigned int); 52406vector unsigned short vec_sldb (vector unsigned short, 52407 vector unsigned short, const unsigned int); 52408vector signed int vec_sldb (vector signed int, vector signed int, 52409 const unsigned int); 52410vector unsigned int vec_sldb (vector unsigned int, vector unsigned int, 52411 const unsigned int); 52412vector signed long long vec_sldb (vector signed long long, 52413 vector signed long long, const unsigned int); 52414vector unsigned long long vec_sldb (vector unsigned long long, 52415 vector unsigned long long, const unsigned int); 52416 52417 Shift the combined input vectors left by the amount specified by the 52418low-order three bits of the third argument, and return the leftmost 52419remaining 128 bits. Code using this instruction must be endian-aware. 52420 52421 Vector Shift Right Double Bit Immediate 52422 52423vector signed char vec_srdb (vector signed char, vector signed char, 52424 const unsigned int); 52425vector unsigned char vec_srdb (vector unsigned char, vector unsigned char, 52426 const unsigned int); 52427vector signed short vec_srdb (vector signed short, vector signed short, 52428 const unsigned int); 52429vector unsigned short vec_srdb (vector unsigned short, vector unsigned short, 52430 const unsigned int); 52431vector signed int vec_srdb (vector signed int, vector signed int, 52432 const unsigned int); 52433vector unsigned int vec_srdb (vector unsigned int, vector unsigned int, 52434 const unsigned int); 52435vector signed long long vec_srdb (vector signed long long, 52436 vector signed long long, const unsigned int); 52437vector unsigned long long vec_srdb (vector unsigned long long, 52438 vector unsigned long long, const unsigned int); 52439 52440 Shift the combined input vectors right by the amount specified by the 52441low-order three bits of the third argument, and return the remaining 128 52442bits. Code using this built-in must be endian-aware. 52443 52444 Vector Splat 52445 52446vector signed int vec_splati (const signed int); 52447vector float vec_splati (const float); 52448 52449 Splat a 32-bit immediate into a vector of words. 52450 52451vector double vec_splatid (const float); 52452 52453 Convert a single precision floating-point value to double-precision and 52454splat the result to a vector of double-precision floats. 52455 52456vector signed int vec_splati_ins (vector signed int, 52457 const unsigned int, const signed int); 52458vector unsigned int vec_splati_ins (vector unsigned int, 52459 const unsigned int, const unsigned int); 52460vector float vec_splati_ins (vector float, const unsigned int, 52461 const float); 52462 52463 Argument 2 must be either 0 or 1. Splat the value of argument 3 into 52464the word identified by argument 2 of each doubleword of argument 1 and 52465return the result. The other words of argument 1 are unchanged. 52466 52467 Vector Blend Variable 52468 52469vector signed char vec_blendv (vector signed char, vector signed char, 52470 vector unsigned char); 52471vector unsigned char vec_blendv (vector unsigned char, 52472 vector unsigned char, vector unsigned char); 52473vector signed short vec_blendv (vector signed short, 52474 vector signed short, vector unsigned short); 52475vector unsigned short vec_blendv (vector unsigned short, 52476 vector unsigned short, vector unsigned short); 52477vector signed int vec_blendv (vector signed int, vector signed int, 52478 vector unsigned int); 52479vector unsigned int vec_blendv (vector unsigned int, 52480 vector unsigned int, vector unsigned int); 52481vector signed long long vec_blendv (vector signed long long, 52482 vector signed long long, vector unsigned long long); 52483vector unsigned long long vec_blendv (vector unsigned long long, 52484 vector unsigned long long, vector unsigned long long); 52485vector float vec_blendv (vector float, vector float, 52486 vector unsigned int); 52487vector double vec_blendv (vector double, vector double, 52488 vector unsigned long long); 52489 52490 Blend the first and second argument vectors according to the sign bits 52491of the corresponding elements of the third argument vector. This is 52492similar to the 'vsel' and 'xxsel' instructions but for bigger elements. 52493 52494 Vector Permute Extended 52495 52496vector signed char vec_permx (vector signed char, vector signed char, 52497 vector unsigned char, const int); 52498vector unsigned char vec_permx (vector unsigned char, 52499 vector unsigned char, vector unsigned char, const int); 52500vector signed short vec_permx (vector signed short, 52501 vector signed short, vector unsigned char, const int); 52502vector unsigned short vec_permx (vector unsigned short, 52503 vector unsigned short, vector unsigned char, const int); 52504vector signed int vec_permx (vector signed int, vector signed int, 52505 vector unsigned char, const int); 52506vector unsigned int vec_permx (vector unsigned int, 52507 vector unsigned int, vector unsigned char, const int); 52508vector signed long long vec_permx (vector signed long long, 52509 vector signed long long, vector unsigned char, const int); 52510vector unsigned long long vec_permx (vector unsigned long long, 52511 vector unsigned long long, vector unsigned char, const int); 52512vector float (vector float, vector float, vector unsigned char, 52513 const int); 52514vector double (vector double, vector double, vector unsigned char, 52515 const int); 52516 52517 Perform a partial permute of the first two arguments, which form a 5251832-byte section of an emulated vector up to 256 bytes wide, using the 52519partial permute control vector in the third argument. The fourth 52520argument (constrained to values of 0-7) identifies which 32-byte section 52521of the emulated vector is contained in the first two arguments. 52522 52523vector unsigned long long int 52524vec_pext (vector unsigned long long int, vector unsigned long long int); 52525 Perform a vector parallel bit extract operation, as if implemented by 52526the 'vpextd' instruction. 52527 52528vector unsigned char vec_stril (vector unsigned char); 52529vector signed char vec_stril (vector signed char); 52530vector unsigned short vec_stril (vector unsigned short); 52531vector signed short vec_stril (vector signed short); 52532 Isolate the left-most non-zero elements of the incoming vector 52533argument, replacing all elements to the right of the left-most zero 52534element found within the argument with zero. The typical implementation 52535uses the 'vstribl' or 'vstrihl' instruction on big-endian targets and 52536uses the 'vstribr' or 'vstrihr' instruction on little-endian targets. 52537 52538int vec_stril_p (vector unsigned char); 52539int vec_stril_p (vector signed char); 52540int short vec_stril_p (vector unsigned short); 52541int vec_stril_p (vector signed short); 52542 Return a non-zero value if and only if the argument contains a zero 52543element. The typical implementation uses the 'vstribl.' or 'vstrihl.' 52544instruction on big-endian targets and uses the 'vstribr.' or 'vstrihr.' 52545instruction on little-endian targets. Choose this built-in to check for 52546presence of zero element if the same argument is also passed to 52547'vec_stril'. 52548 52549vector unsigned char vec_strir (vector unsigned char); 52550vector signed char vec_strir (vector signed char); 52551vector unsigned short vec_strir (vector unsigned short); 52552vector signed short vec_strir (vector signed short); 52553 Isolate the right-most non-zero elements of the incoming vector 52554argument, replacing all elements to the left of the right-most zero 52555element found within the argument with zero. The typical implementation 52556uses the 'vstribr' or 'vstrihr' instruction on big-endian targets and 52557uses the 'vstribl' or 'vstrihl' instruction on little-endian targets. 52558 52559int vec_strir_p (vector unsigned char); 52560int vec_strir_p (vector signed char); 52561int short vec_strir_p (vector unsigned short); 52562int vec_strir_p (vector signed short); 52563 Return a non-zero value if and only if the argument contains a zero 52564element. The typical implementation uses the 'vstribr.' or 'vstrihr.' 52565instruction on big-endian targets and uses the 'vstribl.' or 'vstrihl.' 52566instruction on little-endian targets. Choose this built-in to check for 52567presence of zero element if the same argument is also passed to 52568'vec_strir'. 52569 52570vector unsigned char 52571vec_ternarylogic (vector unsigned char, vector unsigned char, 52572 vector unsigned char, const unsigned int); 52573vector unsigned short 52574vec_ternarylogic (vector unsigned short, vector unsigned short, 52575 vector unsigned short, const unsigned int); 52576vector unsigned int 52577vec_ternarylogic (vector unsigned int, vector unsigned int, 52578 vector unsigned int, const unsigned int); 52579vector unsigned long long int 52580vec_ternarylogic (vector unsigned long long int, vector unsigned long long int, 52581 vector unsigned long long int, const unsigned int); 52582vector unsigned __int128 52583vec_ternarylogic (vector unsigned __int128, vector unsigned __int128, 52584 vector unsigned __int128, const unsigned int); 52585 Perform a 128-bit vector evaluate operation, as if implemented by the 52586'xxeval' instruction. The fourth argument must be a literal integer 52587value between 0 and 255 inclusive. 52588 52589vector unsigned char vec_genpcvm (vector unsigned char, const int); 52590vector unsigned short vec_genpcvm (vector unsigned short, const int); 52591vector unsigned int vec_genpcvm (vector unsigned int, const int); 52592vector unsigned int vec_genpcvm (vector unsigned long long int, 52593 const int); 52594 52595 Vector Integer Multiply/Divide/Modulo 52596 52597vector signed int 52598vec_mulh (vector signed int a, vector signed int b); 52599vector unsigned int 52600vec_mulh (vector unsigned int a, vector unsigned int b); 52601 52602 For each integer value 'i' from 0 to 3, do the following. The integer 52603value in word element 'i' of a is multiplied by the integer value in 52604word element 'i' of b. The high-order 32 bits of the 64-bit product are 52605placed into word element 'i' of the vector returned. 52606 52607vector signed long long 52608vec_mulh (vector signed long long a, vector signed long long b); 52609vector unsigned long long 52610vec_mulh (vector unsigned long long a, vector unsigned long long b); 52611 52612 For each integer value 'i' from 0 to 1, do the following. The integer 52613value in doubleword element 'i' of a is multiplied by the integer value 52614in doubleword element 'i' of b. The high-order 64 bits of the 128-bit 52615product are placed into doubleword element 'i' of the vector returned. 52616 52617vector unsigned long long 52618vec_mul (vector unsigned long long a, vector unsigned long long b); 52619vector signed long long 52620vec_mul (vector signed long long a, vector signed long long b); 52621 52622 For each integer value 'i' from 0 to 1, do the following. The integer 52623value in doubleword element 'i' of a is multiplied by the integer value 52624in doubleword element 'i' of b. The low-order 64 bits of the 128-bit 52625product are placed into doubleword element 'i' of the vector returned. 52626 52627vector signed int 52628vec_div (vector signed int a, vector signed int b); 52629vector unsigned int 52630vec_div (vector unsigned int a, vector unsigned int b); 52631 52632 For each integer value 'i' from 0 to 3, do the following. The integer 52633in word element 'i' of a is divided by the integer in word element 'i' 52634of b. The unique integer quotient is placed into the word element 'i' 52635of the vector returned. If an attempt is made to perform any of the 52636divisions <anything> �� 0 then the quotient is undefined. 52637 52638vector signed long long 52639vec_div (vector signed long long a, vector signed long long b); 52640vector unsigned long long 52641vec_div (vector unsigned long long a, vector unsigned long long b); 52642 52643 For each integer value 'i' from 0 to 1, do the following. The integer 52644in doubleword element 'i' of a is divided by the integer in doubleword 52645element 'i' of b. The unique integer quotient is placed into the 52646doubleword element 'i' of the vector returned. If an attempt is made to 52647perform any of the divisions 0x8000_0000_0000_0000 �� -1 or <anything> 52648�� 0 then the quotient is undefined. 52649 52650vector signed int 52651vec_dive (vector signed int a, vector signed int b); 52652vector unsigned int 52653vec_dive (vector unsigned int a, vector unsigned int b); 52654 52655 For each integer value 'i' from 0 to 3, do the following. The integer 52656in word element 'i' of a is shifted left by 32 bits, then divided by the 52657integer in word element 'i' of b. The unique integer quotient is placed 52658into the word element 'i' of the vector returned. If the quotient 52659cannot be represented in 32 bits, or if an attempt is made to perform 52660any of the divisions <anything> �� 0 then the quotient is undefined. 52661 52662vector signed long long 52663vec_dive (vector signed long long a, vector signed long long b); 52664vector unsigned long long 52665vec_dive (vector unsigned long long a, vector unsigned long long b); 52666 52667 For each integer value 'i' from 0 to 1, do the following. The integer 52668in doubleword element 'i' of a is shifted left by 64 bits, then divided 52669by the integer in doubleword element 'i' of b. The unique integer 52670quotient is placed into the doubleword element 'i' of the vector 52671returned. If the quotient cannot be represented in 64 bits, or if an 52672attempt is made to perform <anything> �� 0 then the quotient is 52673undefined. 52674 52675vector signed int 52676vec_mod (vector signed int a, vector signed int b); 52677vector unsigned int 52678vec_mod (vector unsigned int a, vector unsigned int b); 52679 52680 For each integer value 'i' from 0 to 3, do the following. The integer 52681in word element 'i' of a is divided by the integer in word element 'i' 52682of b. The unique integer remainder is placed into the word element 'i' 52683of the vector returned. If an attempt is made to perform any of the 52684divisions 0x8000_0000 �� -1 or <anything> �� 0 then the remainder is 52685undefined. 52686 52687vector signed long long 52688vec_mod (vector signed long long a, vector signed long long b); 52689vector unsigned long long 52690vec_mod (vector unsigned long long a, vector unsigned long long b); 52691 52692 For each integer value 'i' from 0 to 1, do the following. The integer 52693in doubleword element 'i' of a is divided by the integer in doubleword 52694element 'i' of b. The unique integer remainder is placed into the 52695doubleword element 'i' of the vector returned. If an attempt is made to 52696perform <anything> �� 0 then the remainder is undefined. 52697 52698 Generate PCV from specified Mask size, as if implemented by the 52699'xxgenpcvbm', 'xxgenpcvhm', 'xxgenpcvwm' instructions, where immediate 52700value is either 0, 1, 2 or 3. 52701 52702vector unsigned __int128 vec_rl (vector unsigned __int128 A, 52703 vector unsigned __int128 B); 52704vector signed __int128 vec_rl (vector signed __int128 A, 52705 vector unsigned __int128 B); 52706 52707 Result value: Each element of R is obtained by rotating the 52708corresponding element of A left by the number of bits specified by the 52709corresponding element of B. 52710 52711vector unsigned __int128 vec_rlmi (vector unsigned __int128, 52712 vector unsigned __int128, 52713 vector unsigned __int128); 52714vector signed __int128 vec_rlmi (vector signed __int128, 52715 vector signed __int128, 52716 vector unsigned __int128); 52717 52718 Returns the result of rotating the first input and inserting it under 52719mask into the second input. The first bit in the mask, the last bit in 52720the mask are obtained from the two 7-bit fields bits [108:115] and bits 52721[117:123] respectively of the second input. The shift is obtained from 52722the third input in the 7-bit field [125:131] where all bits counted from 52723zero at the left. 52724 52725vector unsigned __int128 vec_rlnm (vector unsigned __int128, 52726 vector unsigned __int128, 52727 vector unsigned __int128); 52728vector signed __int128 vec_rlnm (vector signed __int128, 52729 vector unsigned __int128, 52730 vector unsigned __int128); 52731 52732 Returns the result of rotating the first input and ANDing it with a 52733mask. The first bit in the mask and the last bit in the mask are 52734obtained from the two 7-bit fields bits [117:123] and bits [125:131] 52735respectively of the second input. The shift is obtained from the third 52736input in the 7-bit field bits [125:131] where all bits counted from zero 52737at the left. 52738 52739vector unsigned __int128 vec_sl(vector unsigned __int128 A, vector unsigned __int128 B); 52740vector signed __int128 vec_sl(vector signed __int128 A, vector unsigned __int128 B); 52741 52742 Result value: Each element of R is obtained by shifting the 52743corresponding element of A left by the number of bits specified by the 52744corresponding element of B. 52745 52746vector unsigned __int128 vec_sr(vector unsigned __int128 A, vector unsigned __int128 B); 52747vector signed __int128 vec_sr(vector signed __int128 A, vector unsigned __int128 B); 52748 52749 Result value: Each element of R is obtained by shifting the 52750corresponding element of A right by the number of bits specified by the 52751corresponding element of B. 52752 52753vector unsigned __int128 vec_sra(vector unsigned __int128 A, vector unsigned __int128 B); 52754vector signed __int128 vec_sra(vector signed __int128 A, vector unsigned __int128 B); 52755 52756 Result value: Each element of R is obtained by arithmetic shifting the 52757corresponding element of A right by the number of bits specified by the 52758corresponding element of B. 52759 52760vector unsigned __int128 vec_mule (vector unsigned long long, 52761 vector unsigned long long); 52762vector signed __int128 vec_mule (vector signed long long, 52763 vector signed long long); 52764 52765 Returns a vector containing a 128-bit integer result of multiplying the 52766even doubleword elements of the two inputs. 52767 52768vector unsigned __int128 vec_mulo (vector unsigned long long, 52769 vector unsigned long long); 52770vector signed __int128 vec_mulo (vector signed long long, 52771 vector signed long long); 52772 52773 Returns a vector containing a 128-bit integer result of multiplying the 52774odd doubleword elements of the two inputs. 52775 52776vector unsigned __int128 vec_div (vector unsigned __int128, 52777 vector unsigned __int128); 52778vector signed __int128 vec_div (vector signed __int128, 52779 vector signed __int128); 52780 52781 Returns the result of dividing the first operand by the second operand. 52782An attempt to divide any value by zero or to divide the most negative 52783signed 128-bit integer by negative one results in an undefined value. 52784 52785vector unsigned __int128 vec_dive (vector unsigned __int128, 52786 vector unsigned __int128); 52787vector signed __int128 vec_dive (vector signed __int128, 52788 vector signed __int128); 52789 52790 The result is produced by shifting the first input left by 128 bits and 52791dividing by the second. If an attempt is made to divide by zero or the 52792result is larger than 128 bits, the result is undefined. 52793 52794vector unsigned __int128 vec_mod (vector unsigned __int128, 52795 vector unsigned __int128); 52796vector signed __int128 vec_mod (vector signed __int128, 52797 vector signed __int128); 52798 52799 The result is the modulo result of dividing the first input by the 52800second input. 52801 52802 The following builtins perform 128-bit vector comparisons. The 52803'vec_all_xx', 'vec_any_xx', and 'vec_cmpxx', where 'xx' is one of the 52804operations 'eq, ne, gt, lt, ge, le' perform pairwise comparisons between 52805the elements at the same positions within their two vector arguments. 52806The 'vec_all_xx'function returns a non-zero value if and only if all 52807pairwise comparisons are true. The 'vec_any_xx' function returns a 52808non-zero value if and only if at least one pairwise comparison is true. 52809The 'vec_cmpxx'function returns a vector of the same type as its two 52810arguments, within which each element consists of all ones to denote that 52811specified logical comparison of the corresponding elements was true. 52812Otherwise, the element of the returned vector contains all zeros. 52813 52814 vector bool __int128 vec_cmpeq (vector signed __int128, vector signed __int128); 52815 vector bool __int128 vec_cmpeq (vector unsigned __int128, vector unsigned __int128); 52816 vector bool __int128 vec_cmpne (vector signed __int128, vector signed __int128); 52817 vector bool __int128 vec_cmpne (vector unsigned __int128, vector unsigned __int128); 52818 vector bool __int128 vec_cmpgt (vector signed __int128, vector signed __int128); 52819 vector bool __int128 vec_cmpgt (vector unsigned __int128, vector unsigned __int128); 52820 vector bool __int128 vec_cmplt (vector signed __int128, vector signed __int128); 52821 vector bool __int128 vec_cmplt (vector unsigned __int128, vector unsigned __int128); 52822 vector bool __int128 vec_cmpge (vector signed __int128, vector signed __int128); 52823 vector bool __int128 vec_cmpge (vector unsigned __int128, vector unsigned __int128); 52824 vector bool __int128 vec_cmple (vector signed __int128, vector signed __int128); 52825 vector bool __int128 vec_cmple (vector unsigned __int128, vector unsigned __int128); 52826 52827 int vec_all_eq (vector signed __int128, vector signed __int128); 52828 int vec_all_eq (vector unsigned __int128, vector unsigned __int128); 52829 int vec_all_ne (vector signed __int128, vector signed __int128); 52830 int vec_all_ne (vector unsigned __int128, vector unsigned __int128); 52831 int vec_all_gt (vector signed __int128, vector signed __int128); 52832 int vec_all_gt (vector unsigned __int128, vector unsigned __int128); 52833 int vec_all_lt (vector signed __int128, vector signed __int128); 52834 int vec_all_lt (vector unsigned __int128, vector unsigned __int128); 52835 int vec_all_ge (vector signed __int128, vector signed __int128); 52836 int vec_all_ge (vector unsigned __int128, vector unsigned __int128); 52837 int vec_all_le (vector signed __int128, vector signed __int128); 52838 int vec_all_le (vector unsigned __int128, vector unsigned __int128); 52839 52840 int vec_any_eq (vector signed __int128, vector signed __int128); 52841 int vec_any_eq (vector unsigned __int128, vector unsigned __int128); 52842 int vec_any_ne (vector signed __int128, vector signed __int128); 52843 int vec_any_ne (vector unsigned __int128, vector unsigned __int128); 52844 int vec_any_gt (vector signed __int128, vector signed __int128); 52845 int vec_any_gt (vector unsigned __int128, vector unsigned __int128); 52846 int vec_any_lt (vector signed __int128, vector signed __int128); 52847 int vec_any_lt (vector unsigned __int128, vector unsigned __int128); 52848 int vec_any_ge (vector signed __int128, vector signed __int128); 52849 int vec_any_ge (vector unsigned __int128, vector unsigned __int128); 52850 int vec_any_le (vector signed __int128, vector signed __int128); 52851 int vec_any_le (vector unsigned __int128, vector unsigned __int128); 52852 52853 52854File: gcc.info, Node: PowerPC Hardware Transactional Memory Built-in Functions, Next: PowerPC Atomic Memory Operation Functions, Prev: PowerPC AltiVec/VSX Built-in Functions, Up: Target Builtins 52855 528566.60.24 PowerPC Hardware Transactional Memory Built-in Functions 52857---------------------------------------------------------------- 52858 52859GCC provides two interfaces for accessing the Hardware Transactional 52860Memory (HTM) instructions available on some of the PowerPC family of 52861processors (eg, POWER8). The two interfaces come in a low level 52862interface, consisting of built-in functions specific to PowerPC and a 52863higher level interface consisting of inline functions that are common 52864between PowerPC and S/390. 52865 528666.60.24.1 PowerPC HTM Low Level Built-in Functions 52867.................................................. 52868 52869The following low level built-in functions are available with '-mhtm' or 52870'-mcpu=CPU' where CPU is 'power8' or later. They all generate the 52871machine instruction that is part of the name. 52872 52873 The HTM builtins (with the exception of '__builtin_tbegin') return the 52874full 4-bit condition register value set by their associated hardware 52875instruction. The header file 'htmintrin.h' defines some macros that can 52876be used to decipher the return value. The '__builtin_tbegin' builtin 52877returns a simple 'true' or 'false' value depending on whether a 52878transaction was successfully started or not. The arguments of the 52879builtins match exactly the type and order of the associated hardware 52880instruction's operands, except for the '__builtin_tcheck' builtin, which 52881does not take any input arguments. Refer to the ISA manual for a 52882description of each instruction's operands. 52883 52884 unsigned int __builtin_tbegin (unsigned int); 52885 unsigned int __builtin_tend (unsigned int); 52886 52887 unsigned int __builtin_tabort (unsigned int); 52888 unsigned int __builtin_tabortdc (unsigned int, unsigned int, unsigned int); 52889 unsigned int __builtin_tabortdci (unsigned int, unsigned int, int); 52890 unsigned int __builtin_tabortwc (unsigned int, unsigned int, unsigned int); 52891 unsigned int __builtin_tabortwci (unsigned int, unsigned int, int); 52892 52893 unsigned int __builtin_tcheck (void); 52894 unsigned int __builtin_treclaim (unsigned int); 52895 unsigned int __builtin_trechkpt (void); 52896 unsigned int __builtin_tsr (unsigned int); 52897 52898 In addition to the above HTM built-ins, we have added built-ins for 52899some common extended mnemonics of the HTM instructions: 52900 52901 unsigned int __builtin_tendall (void); 52902 unsigned int __builtin_tresume (void); 52903 unsigned int __builtin_tsuspend (void); 52904 52905 Note that the semantics of the above HTM builtins are required to mimic 52906the locking semantics used for critical sections. Builtins that are 52907used to create a new transaction or restart a suspended transaction must 52908have lock acquisition like semantics while those builtins that end or 52909suspend a transaction must have lock release like semantics. 52910Specifically, this must mimic lock semantics as specified by C++11, for 52911example: Lock acquisition is as-if an execution of 52912__atomic_exchange_n(&globallock,1,__ATOMIC_ACQUIRE) that returns 0, and 52913lock release is as-if an execution of 52914__atomic_store(&globallock,0,__ATOMIC_RELEASE), with globallock being an 52915implicit implementation-defined lock used for all transactions. The HTM 52916instructions associated with with the builtins inherently provide the 52917correct acquisition and release hardware barriers required. However, 52918the compiler must also be prohibited from moving loads and stores across 52919the builtins in a way that would violate their semantics. This has been 52920accomplished by adding memory barriers to the associated HTM 52921instructions (which is a conservative approach to provide acquire and 52922release semantics). Earlier versions of the compiler did not treat the 52923HTM instructions as memory barriers. A '__TM_FENCE__' macro has been 52924added, which can be used to determine whether the current compiler 52925treats HTM instructions as memory barriers or not. This allows the user 52926to explicitly add memory barriers to their code when using an older 52927version of the compiler. 52928 52929 The following set of built-in functions are available to gain access to 52930the HTM specific special purpose registers. 52931 52932 unsigned long __builtin_get_texasr (void); 52933 unsigned long __builtin_get_texasru (void); 52934 unsigned long __builtin_get_tfhar (void); 52935 unsigned long __builtin_get_tfiar (void); 52936 52937 void __builtin_set_texasr (unsigned long); 52938 void __builtin_set_texasru (unsigned long); 52939 void __builtin_set_tfhar (unsigned long); 52940 void __builtin_set_tfiar (unsigned long); 52941 52942 Example usage of these low level built-in functions may look like: 52943 52944 #include <htmintrin.h> 52945 52946 int num_retries = 10; 52947 52948 while (1) 52949 { 52950 if (__builtin_tbegin (0)) 52951 { 52952 /* Transaction State Initiated. */ 52953 if (is_locked (lock)) 52954 __builtin_tabort (0); 52955 ... transaction code... 52956 __builtin_tend (0); 52957 break; 52958 } 52959 else 52960 { 52961 /* Transaction State Failed. Use locks if the transaction 52962 failure is "persistent" or we've tried too many times. */ 52963 if (num_retries-- <= 0 52964 || _TEXASRU_FAILURE_PERSISTENT (__builtin_get_texasru ())) 52965 { 52966 acquire_lock (lock); 52967 ... non transactional fallback path... 52968 release_lock (lock); 52969 break; 52970 } 52971 } 52972 } 52973 52974 One final built-in function has been added that returns the value of 52975the 2-bit Transaction State field of the Machine Status Register (MSR) 52976as stored in 'CR0'. 52977 52978 unsigned long __builtin_ttest (void) 52979 52980 This built-in can be used to determine the current transaction state 52981using the following code example: 52982 52983 #include <htmintrin.h> 52984 52985 unsigned char tx_state = _HTM_STATE (__builtin_ttest ()); 52986 52987 if (tx_state == _HTM_TRANSACTIONAL) 52988 { 52989 /* Code to use in transactional state. */ 52990 } 52991 else if (tx_state == _HTM_NONTRANSACTIONAL) 52992 { 52993 /* Code to use in non-transactional state. */ 52994 } 52995 else if (tx_state == _HTM_SUSPENDED) 52996 { 52997 /* Code to use in transaction suspended state. */ 52998 } 52999 530006.60.24.2 PowerPC HTM High Level Inline Functions 53001................................................. 53002 53003The following high level HTM interface is made available by including 53004'<htmxlintrin.h>' and using '-mhtm' or '-mcpu=CPU' where CPU is 'power8' 53005or later. This interface is common between PowerPC and S/390, allowing 53006users to write one HTM source implementation that can be compiled and 53007executed on either system. 53008 53009 long __TM_simple_begin (void); 53010 long __TM_begin (void* const TM_buff); 53011 long __TM_end (void); 53012 void __TM_abort (void); 53013 void __TM_named_abort (unsigned char const code); 53014 void __TM_resume (void); 53015 void __TM_suspend (void); 53016 53017 long __TM_is_user_abort (void* const TM_buff); 53018 long __TM_is_named_user_abort (void* const TM_buff, unsigned char *code); 53019 long __TM_is_illegal (void* const TM_buff); 53020 long __TM_is_footprint_exceeded (void* const TM_buff); 53021 long __TM_nesting_depth (void* const TM_buff); 53022 long __TM_is_nested_too_deep(void* const TM_buff); 53023 long __TM_is_conflict(void* const TM_buff); 53024 long __TM_is_failure_persistent(void* const TM_buff); 53025 long __TM_failure_address(void* const TM_buff); 53026 long long __TM_failure_code(void* const TM_buff); 53027 53028 Using these common set of HTM inline functions, we can create a more 53029portable version of the HTM example in the previous section that will 53030work on either PowerPC or S/390: 53031 53032 #include <htmxlintrin.h> 53033 53034 int num_retries = 10; 53035 TM_buff_type TM_buff; 53036 53037 while (1) 53038 { 53039 if (__TM_begin (TM_buff) == _HTM_TBEGIN_STARTED) 53040 { 53041 /* Transaction State Initiated. */ 53042 if (is_locked (lock)) 53043 __TM_abort (); 53044 ... transaction code... 53045 __TM_end (); 53046 break; 53047 } 53048 else 53049 { 53050 /* Transaction State Failed. Use locks if the transaction 53051 failure is "persistent" or we've tried too many times. */ 53052 if (num_retries-- <= 0 53053 || __TM_is_failure_persistent (TM_buff)) 53054 { 53055 acquire_lock (lock); 53056 ... non transactional fallback path... 53057 release_lock (lock); 53058 break; 53059 } 53060 } 53061 } 53062 53063 53064File: gcc.info, Node: PowerPC Atomic Memory Operation Functions, Next: PowerPC Matrix-Multiply Assist Built-in Functions, Prev: PowerPC Hardware Transactional Memory Built-in Functions, Up: Target Builtins 53065 530666.60.25 PowerPC Atomic Memory Operation Functions 53067------------------------------------------------- 53068 53069ISA 3.0 of the PowerPC added new atomic memory operation (amo) 53070instructions. GCC provides support for these instructions in 64-bit 53071environments. All of the functions are declared in the include file 53072'amo.h'. 53073 53074 The functions supported are: 53075 53076 #include <amo.h> 53077 53078 uint32_t amo_lwat_add (uint32_t *, uint32_t); 53079 uint32_t amo_lwat_xor (uint32_t *, uint32_t); 53080 uint32_t amo_lwat_ior (uint32_t *, uint32_t); 53081 uint32_t amo_lwat_and (uint32_t *, uint32_t); 53082 uint32_t amo_lwat_umax (uint32_t *, uint32_t); 53083 uint32_t amo_lwat_umin (uint32_t *, uint32_t); 53084 uint32_t amo_lwat_swap (uint32_t *, uint32_t); 53085 53086 int32_t amo_lwat_sadd (int32_t *, int32_t); 53087 int32_t amo_lwat_smax (int32_t *, int32_t); 53088 int32_t amo_lwat_smin (int32_t *, int32_t); 53089 int32_t amo_lwat_sswap (int32_t *, int32_t); 53090 53091 uint64_t amo_ldat_add (uint64_t *, uint64_t); 53092 uint64_t amo_ldat_xor (uint64_t *, uint64_t); 53093 uint64_t amo_ldat_ior (uint64_t *, uint64_t); 53094 uint64_t amo_ldat_and (uint64_t *, uint64_t); 53095 uint64_t amo_ldat_umax (uint64_t *, uint64_t); 53096 uint64_t amo_ldat_umin (uint64_t *, uint64_t); 53097 uint64_t amo_ldat_swap (uint64_t *, uint64_t); 53098 53099 int64_t amo_ldat_sadd (int64_t *, int64_t); 53100 int64_t amo_ldat_smax (int64_t *, int64_t); 53101 int64_t amo_ldat_smin (int64_t *, int64_t); 53102 int64_t amo_ldat_sswap (int64_t *, int64_t); 53103 53104 void amo_stwat_add (uint32_t *, uint32_t); 53105 void amo_stwat_xor (uint32_t *, uint32_t); 53106 void amo_stwat_ior (uint32_t *, uint32_t); 53107 void amo_stwat_and (uint32_t *, uint32_t); 53108 void amo_stwat_umax (uint32_t *, uint32_t); 53109 void amo_stwat_umin (uint32_t *, uint32_t); 53110 53111 void amo_stwat_sadd (int32_t *, int32_t); 53112 void amo_stwat_smax (int32_t *, int32_t); 53113 void amo_stwat_smin (int32_t *, int32_t); 53114 53115 void amo_stdat_add (uint64_t *, uint64_t); 53116 void amo_stdat_xor (uint64_t *, uint64_t); 53117 void amo_stdat_ior (uint64_t *, uint64_t); 53118 void amo_stdat_and (uint64_t *, uint64_t); 53119 void amo_stdat_umax (uint64_t *, uint64_t); 53120 void amo_stdat_umin (uint64_t *, uint64_t); 53121 53122 void amo_stdat_sadd (int64_t *, int64_t); 53123 void amo_stdat_smax (int64_t *, int64_t); 53124 void amo_stdat_smin (int64_t *, int64_t); 53125 53126 53127File: gcc.info, Node: PowerPC Matrix-Multiply Assist Built-in Functions, Next: PRU Built-in Functions, Prev: PowerPC Atomic Memory Operation Functions, Up: Target Builtins 53128 531296.60.26 PowerPC Matrix-Multiply Assist Built-in Functions 53130--------------------------------------------------------- 53131 53132ISA 3.1 of the PowerPC added new Matrix-Multiply Assist (MMA) 53133instructions. GCC provides support for these instructions through the 53134following built-in functions which are enabled with the '-mmma' option. 53135The vec_t type below is defined to be a normal vector unsigned char 53136type. The uint2, uint4 and uint8 parameters are 2-bit, 4-bit and 8-bit 53137unsigned integer constants respectively. The compiler will verify that 53138they are constants and that their values are within range. 53139 53140 The built-in functions supported are: 53141 53142 void __builtin_mma_xvi4ger8 (__vector_quad *, vec_t, vec_t); 53143 void __builtin_mma_xvi8ger4 (__vector_quad *, vec_t, vec_t); 53144 void __builtin_mma_xvi16ger2 (__vector_quad *, vec_t, vec_t); 53145 void __builtin_mma_xvi16ger2s (__vector_quad *, vec_t, vec_t); 53146 void __builtin_mma_xvf16ger2 (__vector_quad *, vec_t, vec_t); 53147 void __builtin_mma_xvbf16ger2 (__vector_quad *, vec_t, vec_t); 53148 void __builtin_mma_xvf32ger (__vector_quad *, vec_t, vec_t); 53149 53150 void __builtin_mma_xvi4ger8pp (__vector_quad *, vec_t, vec_t); 53151 void __builtin_mma_xvi8ger4pp (__vector_quad *, vec_t, vec_t); 53152 void __builtin_mma_xvi8ger4spp(__vector_quad *, vec_t, vec_t); 53153 void __builtin_mma_xvi16ger2pp (__vector_quad *, vec_t, vec_t); 53154 void __builtin_mma_xvi16ger2spp (__vector_quad *, vec_t, vec_t); 53155 void __builtin_mma_xvf16ger2pp (__vector_quad *, vec_t, vec_t); 53156 void __builtin_mma_xvf16ger2pn (__vector_quad *, vec_t, vec_t); 53157 void __builtin_mma_xvf16ger2np (__vector_quad *, vec_t, vec_t); 53158 void __builtin_mma_xvf16ger2nn (__vector_quad *, vec_t, vec_t); 53159 void __builtin_mma_xvbf16ger2pp (__vector_quad *, vec_t, vec_t); 53160 void __builtin_mma_xvbf16ger2pn (__vector_quad *, vec_t, vec_t); 53161 void __builtin_mma_xvbf16ger2np (__vector_quad *, vec_t, vec_t); 53162 void __builtin_mma_xvbf16ger2nn (__vector_quad *, vec_t, vec_t); 53163 void __builtin_mma_xvf32gerpp (__vector_quad *, vec_t, vec_t); 53164 void __builtin_mma_xvf32gerpn (__vector_quad *, vec_t, vec_t); 53165 void __builtin_mma_xvf32gernp (__vector_quad *, vec_t, vec_t); 53166 void __builtin_mma_xvf32gernn (__vector_quad *, vec_t, vec_t); 53167 53168 void __builtin_mma_pmxvi4ger8 (__vector_quad *, vec_t, vec_t, uint4, uint4, uint8); 53169 void __builtin_mma_pmxvi4ger8pp (__vector_quad *, vec_t, vec_t, uint4, uint4, uint8); 53170 53171 void __builtin_mma_pmxvi8ger4 (__vector_quad *, vec_t, vec_t, uint4, uint4, uint4); 53172 void __builtin_mma_pmxvi8ger4pp (__vector_quad *, vec_t, vec_t, uint4, uint4, uint4); 53173 void __builtin_mma_pmxvi8ger4spp(__vector_quad *, vec_t, vec_t, uint4, uint4, uint4); 53174 53175 void __builtin_mma_pmxvi16ger2 (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2); 53176 void __builtin_mma_pmxvi16ger2s (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2); 53177 void __builtin_mma_pmxvf16ger2 (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2); 53178 void __builtin_mma_pmxvbf16ger2 (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2); 53179 53180 void __builtin_mma_pmxvi16ger2pp (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2); 53181 void __builtin_mma_pmxvi16ger2spp (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2); 53182 void __builtin_mma_pmxvf16ger2pp (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2); 53183 void __builtin_mma_pmxvf16ger2pn (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2); 53184 void __builtin_mma_pmxvf16ger2np (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2); 53185 void __builtin_mma_pmxvf16ger2nn (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2); 53186 void __builtin_mma_pmxvbf16ger2pp (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2); 53187 void __builtin_mma_pmxvbf16ger2pn (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2); 53188 void __builtin_mma_pmxvbf16ger2np (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2); 53189 void __builtin_mma_pmxvbf16ger2nn (__vector_quad *, vec_t, vec_t, uint4, uint4, uint2); 53190 53191 void __builtin_mma_pmxvf32ger (__vector_quad *, vec_t, vec_t, uint4, uint4); 53192 void __builtin_mma_pmxvf32gerpp (__vector_quad *, vec_t, vec_t, uint4, uint4); 53193 void __builtin_mma_pmxvf32gerpn (__vector_quad *, vec_t, vec_t, uint4, uint4); 53194 void __builtin_mma_pmxvf32gernp (__vector_quad *, vec_t, vec_t, uint4, uint4); 53195 void __builtin_mma_pmxvf32gernn (__vector_quad *, vec_t, vec_t, uint4, uint4); 53196 53197 void __builtin_mma_xvf64ger (__vector_quad *, __vector_pair, vec_t); 53198 void __builtin_mma_xvf64gerpp (__vector_quad *, __vector_pair, vec_t); 53199 void __builtin_mma_xvf64gerpn (__vector_quad *, __vector_pair, vec_t); 53200 void __builtin_mma_xvf64gernp (__vector_quad *, __vector_pair, vec_t); 53201 void __builtin_mma_xvf64gernn (__vector_quad *, __vector_pair, vec_t); 53202 53203 void __builtin_mma_pmxvf64ger (__vector_quad *, __vector_pair, vec_t, uint4, uint2); 53204 void __builtin_mma_pmxvf64gerpp (__vector_quad *, __vector_pair, vec_t, uint4, uint2); 53205 void __builtin_mma_pmxvf64gerpn (__vector_quad *, __vector_pair, vec_t, uint4, uint2); 53206 void __builtin_mma_pmxvf64gernp (__vector_quad *, __vector_pair, vec_t, uint4, uint2); 53207 void __builtin_mma_pmxvf64gernn (__vector_quad *, __vector_pair, vec_t, uint4, uint2); 53208 53209 void __builtin_mma_xxmtacc (__vector_quad *); 53210 void __builtin_mma_xxmfacc (__vector_quad *); 53211 void __builtin_mma_xxsetaccz (__vector_quad *); 53212 53213 void __builtin_mma_build_acc (__vector_quad *, vec_t, vec_t, vec_t, vec_t); 53214 void __builtin_mma_disassemble_acc (void *, __vector_quad *); 53215 53216 void __builtin_vsx_build_pair (__vector_pair *, vec_t, vec_t); 53217 void __builtin_vsx_disassemble_pair (void *, __vector_pair *); 53218 53219 vec_t __builtin_vsx_xvcvspbf16 (vec_t); 53220 vec_t __builtin_vsx_xvcvbf16spn (vec_t); 53221 53222 __vector_pair __builtin_vsx_lxvp (size_t, __vector_pair *); 53223 void __builtin_vsx_stxvp (__vector_pair, size_t, __vector_pair *); 53224 53225 53226File: gcc.info, Node: PRU Built-in Functions, Next: RISC-V Built-in Functions, Prev: PowerPC Matrix-Multiply Assist Built-in Functions, Up: Target Builtins 53227 532286.60.27 PRU Built-in Functions 53229------------------------------ 53230 53231GCC provides a couple of special builtin functions to aid in utilizing 53232special PRU instructions. 53233 53234 The built-in functions supported are: 53235 53236'__delay_cycles (long long CYCLES)' 53237 This inserts an instruction sequence that takes exactly CYCLES 53238 cycles (between 0 and 0xffffffff) to complete. The inserted 53239 sequence may use jumps, loops, or no-ops, and does not interfere 53240 with any other instructions. Note that CYCLES must be a 53241 compile-time constant integer - that is, you must pass a number, 53242 not a variable that may be optimized to a constant later. The 53243 number of cycles delayed by this builtin is exact. 53244 53245'__halt (void)' 53246 This inserts a HALT instruction to stop processor execution. 53247 53248'unsigned int __lmbd (unsigned int WORDVAL, unsigned int BITVAL)' 53249 This inserts LMBD instruction to calculate the left-most bit with 53250 value BITVAL in value WORDVAL. Only the least significant bit of 53251 BITVAL is taken into account. 53252 53253 53254File: gcc.info, Node: RISC-V Built-in Functions, Next: RX Built-in Functions, Prev: PRU Built-in Functions, Up: Target Builtins 53255 532566.60.28 RISC-V Built-in Functions 53257--------------------------------- 53258 53259These built-in functions are available for the RISC-V family of 53260processors. 53261 53262 -- Built-in Function: void * __builtin_thread_pointer (void) 53263 Returns the value that is currently set in the 'tp' register. 53264 53265 53266File: gcc.info, Node: RX Built-in Functions, Next: S/390 System z Built-in Functions, Prev: RISC-V Built-in Functions, Up: Target Builtins 53267 532686.60.29 RX Built-in Functions 53269----------------------------- 53270 53271GCC supports some of the RX instructions which cannot be expressed in 53272the C programming language via the use of built-in functions. The 53273following functions are supported: 53274 53275 -- Built-in Function: void __builtin_rx_brk (void) 53276 Generates the 'brk' machine instruction. 53277 53278 -- Built-in Function: void __builtin_rx_clrpsw (int) 53279 Generates the 'clrpsw' machine instruction to clear the specified 53280 bit in the processor status word. 53281 53282 -- Built-in Function: void __builtin_rx_int (int) 53283 Generates the 'int' machine instruction to generate an interrupt 53284 with the specified value. 53285 53286 -- Built-in Function: void __builtin_rx_machi (int, int) 53287 Generates the 'machi' machine instruction to add the result of 53288 multiplying the top 16 bits of the two arguments into the 53289 accumulator. 53290 53291 -- Built-in Function: void __builtin_rx_maclo (int, int) 53292 Generates the 'maclo' machine instruction to add the result of 53293 multiplying the bottom 16 bits of the two arguments into the 53294 accumulator. 53295 53296 -- Built-in Function: void __builtin_rx_mulhi (int, int) 53297 Generates the 'mulhi' machine instruction to place the result of 53298 multiplying the top 16 bits of the two arguments into the 53299 accumulator. 53300 53301 -- Built-in Function: void __builtin_rx_mullo (int, int) 53302 Generates the 'mullo' machine instruction to place the result of 53303 multiplying the bottom 16 bits of the two arguments into the 53304 accumulator. 53305 53306 -- Built-in Function: int __builtin_rx_mvfachi (void) 53307 Generates the 'mvfachi' machine instruction to read the top 32 bits 53308 of the accumulator. 53309 53310 -- Built-in Function: int __builtin_rx_mvfacmi (void) 53311 Generates the 'mvfacmi' machine instruction to read the middle 32 53312 bits of the accumulator. 53313 53314 -- Built-in Function: int __builtin_rx_mvfc (int) 53315 Generates the 'mvfc' machine instruction which reads the control 53316 register specified in its argument and returns its value. 53317 53318 -- Built-in Function: void __builtin_rx_mvtachi (int) 53319 Generates the 'mvtachi' machine instruction to set the top 32 bits 53320 of the accumulator. 53321 53322 -- Built-in Function: void __builtin_rx_mvtaclo (int) 53323 Generates the 'mvtaclo' machine instruction to set the bottom 32 53324 bits of the accumulator. 53325 53326 -- Built-in Function: void __builtin_rx_mvtc (int reg, int val) 53327 Generates the 'mvtc' machine instruction which sets control 53328 register number 'reg' to 'val'. 53329 53330 -- Built-in Function: void __builtin_rx_mvtipl (int) 53331 Generates the 'mvtipl' machine instruction set the interrupt 53332 priority level. 53333 53334 -- Built-in Function: void __builtin_rx_racw (int) 53335 Generates the 'racw' machine instruction to round the accumulator 53336 according to the specified mode. 53337 53338 -- Built-in Function: int __builtin_rx_revw (int) 53339 Generates the 'revw' machine instruction which swaps the bytes in 53340 the argument so that bits 0-7 now occupy bits 8-15 and vice versa, 53341 and also bits 16-23 occupy bits 24-31 and vice versa. 53342 53343 -- Built-in Function: void __builtin_rx_rmpa (void) 53344 Generates the 'rmpa' machine instruction which initiates a repeated 53345 multiply and accumulate sequence. 53346 53347 -- Built-in Function: void __builtin_rx_round (float) 53348 Generates the 'round' machine instruction which returns the 53349 floating-point argument rounded according to the current rounding 53350 mode set in the floating-point status word register. 53351 53352 -- Built-in Function: int __builtin_rx_sat (int) 53353 Generates the 'sat' machine instruction which returns the saturated 53354 value of the argument. 53355 53356 -- Built-in Function: void __builtin_rx_setpsw (int) 53357 Generates the 'setpsw' machine instruction to set the specified bit 53358 in the processor status word. 53359 53360 -- Built-in Function: void __builtin_rx_wait (void) 53361 Generates the 'wait' machine instruction. 53362 53363 53364File: gcc.info, Node: S/390 System z Built-in Functions, Next: SH Built-in Functions, Prev: RX Built-in Functions, Up: Target Builtins 53365 533666.60.30 S/390 System z Built-in Functions 53367----------------------------------------- 53368 53369 -- Built-in Function: int __builtin_tbegin (void*) 53370 Generates the 'tbegin' machine instruction starting a 53371 non-constrained hardware transaction. If the parameter is non-NULL 53372 the memory area is used to store the transaction diagnostic buffer 53373 and will be passed as first operand to 'tbegin'. This buffer can 53374 be defined using the 'struct __htm_tdb' C struct defined in 53375 'htmintrin.h' and must reside on a double-word boundary. The 53376 second tbegin operand is set to '0xff0c'. This enables 53377 save/restore of all GPRs and disables aborts for FPR and AR 53378 manipulations inside the transaction body. The condition code set 53379 by the tbegin instruction is returned as integer value. The tbegin 53380 instruction by definition overwrites the content of all FPRs. The 53381 compiler will generate code which saves and restores the FPRs. For 53382 soft-float code it is recommended to used the '*_nofloat' variant. 53383 In order to prevent a TDB from being written it is required to pass 53384 a constant zero value as parameter. Passing a zero value through a 53385 variable is not sufficient. Although modifications of access 53386 registers inside the transaction will not trigger an transaction 53387 abort it is not supported to actually modify them. Access 53388 registers do not get saved when entering a transaction. They will 53389 have undefined state when reaching the abort code. 53390 53391 Macros for the possible return codes of tbegin are defined in the 53392'htmintrin.h' header file: 53393 53394'_HTM_TBEGIN_STARTED' 53395 'tbegin' has been executed as part of normal processing. The 53396 transaction body is supposed to be executed. 53397'_HTM_TBEGIN_INDETERMINATE' 53398 The transaction was aborted due to an indeterminate condition which 53399 might be persistent. 53400'_HTM_TBEGIN_TRANSIENT' 53401 The transaction aborted due to a transient failure. The 53402 transaction should be re-executed in that case. 53403'_HTM_TBEGIN_PERSISTENT' 53404 The transaction aborted due to a persistent failure. Re-execution 53405 under same circumstances will not be productive. 53406 53407 -- Macro: _HTM_FIRST_USER_ABORT_CODE 53408 The '_HTM_FIRST_USER_ABORT_CODE' defined in 'htmintrin.h' specifies 53409 the first abort code which can be used for '__builtin_tabort'. 53410 Values below this threshold are reserved for machine use. 53411 53412 -- Data type: struct __htm_tdb 53413 The 'struct __htm_tdb' defined in 'htmintrin.h' describes the 53414 structure of the transaction diagnostic block as specified in the 53415 Principles of Operation manual chapter 5-91. 53416 53417 -- Built-in Function: int __builtin_tbegin_nofloat (void*) 53418 Same as '__builtin_tbegin' but without FPR saves and restores. 53419 Using this variant in code making use of FPRs will leave the FPRs 53420 in undefined state when entering the transaction abort handler 53421 code. 53422 53423 -- Built-in Function: int __builtin_tbegin_retry (void*, int) 53424 In addition to '__builtin_tbegin' a loop for transient failures is 53425 generated. If tbegin returns a condition code of 2 the transaction 53426 will be retried as often as specified in the second argument. The 53427 perform processor assist instruction is used to tell the CPU about 53428 the number of fails so far. 53429 53430 -- Built-in Function: int __builtin_tbegin_retry_nofloat (void*, int) 53431 Same as '__builtin_tbegin_retry' but without FPR saves and 53432 restores. Using this variant in code making use of FPRs will leave 53433 the FPRs in undefined state when entering the transaction abort 53434 handler code. 53435 53436 -- Built-in Function: void __builtin_tbeginc (void) 53437 Generates the 'tbeginc' machine instruction starting a constrained 53438 hardware transaction. The second operand is set to '0xff08'. 53439 53440 -- Built-in Function: int __builtin_tend (void) 53441 Generates the 'tend' machine instruction finishing a transaction 53442 and making the changes visible to other threads. The condition 53443 code generated by tend is returned as integer value. 53444 53445 -- Built-in Function: void __builtin_tabort (int) 53446 Generates the 'tabort' machine instruction with the specified abort 53447 code. Abort codes from 0 through 255 are reserved and will result 53448 in an error message. 53449 53450 -- Built-in Function: void __builtin_tx_assist (int) 53451 Generates the 'ppa rX,rY,1' machine instruction. Where the integer 53452 parameter is loaded into rX and a value of zero is loaded into rY. 53453 The integer parameter specifies the number of times the transaction 53454 repeatedly aborted. 53455 53456 -- Built-in Function: int __builtin_tx_nesting_depth (void) 53457 Generates the 'etnd' machine instruction. The current nesting 53458 depth is returned as integer value. For a nesting depth of 0 the 53459 code is not executed as part of an transaction. 53460 53461 -- Built-in Function: void __builtin_non_tx_store (uint64_t *, 53462 uint64_t) 53463 53464 Generates the 'ntstg' machine instruction. The second argument is 53465 written to the first arguments location. The store operation will 53466 not be rolled-back in case of an transaction abort. 53467 53468 53469File: gcc.info, Node: SH Built-in Functions, Next: SPARC VIS Built-in Functions, Prev: S/390 System z Built-in Functions, Up: Target Builtins 53470 534716.60.31 SH Built-in Functions 53472----------------------------- 53473 53474The following built-in functions are supported on the SH1, SH2, SH3 and 53475SH4 families of processors: 53476 53477 -- Built-in Function: void __builtin_set_thread_pointer (void *PTR) 53478 Sets the 'GBR' register to the specified value PTR. This is 53479 usually used by system code that manages threads and execution 53480 contexts. The compiler normally does not generate code that 53481 modifies the contents of 'GBR' and thus the value is preserved 53482 across function calls. Changing the 'GBR' value in user code must 53483 be done with caution, since the compiler might use 'GBR' in order 53484 to access thread local variables. 53485 53486 -- Built-in Function: void * __builtin_thread_pointer (void) 53487 Returns the value that is currently set in the 'GBR' register. 53488 Memory loads and stores that use the thread pointer as a base 53489 address are turned into 'GBR' based displacement loads and stores, 53490 if possible. For example: 53491 struct my_tcb 53492 { 53493 int a, b, c, d, e; 53494 }; 53495 53496 int get_tcb_value (void) 53497 { 53498 // Generate 'mov.l @(8,gbr),r0' instruction 53499 return ((my_tcb*)__builtin_thread_pointer ())->c; 53500 } 53501 53502 53503 -- Built-in Function: unsigned int __builtin_sh_get_fpscr (void) 53504 Returns the value that is currently set in the 'FPSCR' register. 53505 53506 -- Built-in Function: void __builtin_sh_set_fpscr (unsigned int VAL) 53507 Sets the 'FPSCR' register to the specified value VAL, while 53508 preserving the current values of the FR, SZ and PR bits. 53509 53510 53511File: gcc.info, Node: SPARC VIS Built-in Functions, Next: TI C6X Built-in Functions, Prev: SH Built-in Functions, Up: Target Builtins 53512 535136.60.32 SPARC VIS Built-in Functions 53514------------------------------------ 53515 53516GCC supports SIMD operations on the SPARC using both the generic vector 53517extensions (*note Vector Extensions::) as well as built-in functions for 53518the SPARC Visual Instruction Set (VIS). When you use the '-mvis' switch, 53519the VIS extension is exposed as the following built-in functions: 53520 53521 typedef int v1si __attribute__ ((vector_size (4))); 53522 typedef int v2si __attribute__ ((vector_size (8))); 53523 typedef short v4hi __attribute__ ((vector_size (8))); 53524 typedef short v2hi __attribute__ ((vector_size (4))); 53525 typedef unsigned char v8qi __attribute__ ((vector_size (8))); 53526 typedef unsigned char v4qi __attribute__ ((vector_size (4))); 53527 53528 void __builtin_vis_write_gsr (int64_t); 53529 int64_t __builtin_vis_read_gsr (void); 53530 53531 void * __builtin_vis_alignaddr (void *, long); 53532 void * __builtin_vis_alignaddrl (void *, long); 53533 int64_t __builtin_vis_faligndatadi (int64_t, int64_t); 53534 v2si __builtin_vis_faligndatav2si (v2si, v2si); 53535 v4hi __builtin_vis_faligndatav4hi (v4si, v4si); 53536 v8qi __builtin_vis_faligndatav8qi (v8qi, v8qi); 53537 53538 v4hi __builtin_vis_fexpand (v4qi); 53539 53540 v4hi __builtin_vis_fmul8x16 (v4qi, v4hi); 53541 v4hi __builtin_vis_fmul8x16au (v4qi, v2hi); 53542 v4hi __builtin_vis_fmul8x16al (v4qi, v2hi); 53543 v4hi __builtin_vis_fmul8sux16 (v8qi, v4hi); 53544 v4hi __builtin_vis_fmul8ulx16 (v8qi, v4hi); 53545 v2si __builtin_vis_fmuld8sux16 (v4qi, v2hi); 53546 v2si __builtin_vis_fmuld8ulx16 (v4qi, v2hi); 53547 53548 v4qi __builtin_vis_fpack16 (v4hi); 53549 v8qi __builtin_vis_fpack32 (v2si, v8qi); 53550 v2hi __builtin_vis_fpackfix (v2si); 53551 v8qi __builtin_vis_fpmerge (v4qi, v4qi); 53552 53553 int64_t __builtin_vis_pdist (v8qi, v8qi, int64_t); 53554 53555 long __builtin_vis_edge8 (void *, void *); 53556 long __builtin_vis_edge8l (void *, void *); 53557 long __builtin_vis_edge16 (void *, void *); 53558 long __builtin_vis_edge16l (void *, void *); 53559 long __builtin_vis_edge32 (void *, void *); 53560 long __builtin_vis_edge32l (void *, void *); 53561 53562 long __builtin_vis_fcmple16 (v4hi, v4hi); 53563 long __builtin_vis_fcmple32 (v2si, v2si); 53564 long __builtin_vis_fcmpne16 (v4hi, v4hi); 53565 long __builtin_vis_fcmpne32 (v2si, v2si); 53566 long __builtin_vis_fcmpgt16 (v4hi, v4hi); 53567 long __builtin_vis_fcmpgt32 (v2si, v2si); 53568 long __builtin_vis_fcmpeq16 (v4hi, v4hi); 53569 long __builtin_vis_fcmpeq32 (v2si, v2si); 53570 53571 v4hi __builtin_vis_fpadd16 (v4hi, v4hi); 53572 v2hi __builtin_vis_fpadd16s (v2hi, v2hi); 53573 v2si __builtin_vis_fpadd32 (v2si, v2si); 53574 v1si __builtin_vis_fpadd32s (v1si, v1si); 53575 v4hi __builtin_vis_fpsub16 (v4hi, v4hi); 53576 v2hi __builtin_vis_fpsub16s (v2hi, v2hi); 53577 v2si __builtin_vis_fpsub32 (v2si, v2si); 53578 v1si __builtin_vis_fpsub32s (v1si, v1si); 53579 53580 long __builtin_vis_array8 (long, long); 53581 long __builtin_vis_array16 (long, long); 53582 long __builtin_vis_array32 (long, long); 53583 53584 When you use the '-mvis2' switch, the VIS version 2.0 built-in 53585functions also become available: 53586 53587 long __builtin_vis_bmask (long, long); 53588 int64_t __builtin_vis_bshuffledi (int64_t, int64_t); 53589 v2si __builtin_vis_bshufflev2si (v2si, v2si); 53590 v4hi __builtin_vis_bshufflev2si (v4hi, v4hi); 53591 v8qi __builtin_vis_bshufflev2si (v8qi, v8qi); 53592 53593 long __builtin_vis_edge8n (void *, void *); 53594 long __builtin_vis_edge8ln (void *, void *); 53595 long __builtin_vis_edge16n (void *, void *); 53596 long __builtin_vis_edge16ln (void *, void *); 53597 long __builtin_vis_edge32n (void *, void *); 53598 long __builtin_vis_edge32ln (void *, void *); 53599 53600 When you use the '-mvis3' switch, the VIS version 3.0 built-in 53601functions also become available: 53602 53603 void __builtin_vis_cmask8 (long); 53604 void __builtin_vis_cmask16 (long); 53605 void __builtin_vis_cmask32 (long); 53606 53607 v4hi __builtin_vis_fchksm16 (v4hi, v4hi); 53608 53609 v4hi __builtin_vis_fsll16 (v4hi, v4hi); 53610 v4hi __builtin_vis_fslas16 (v4hi, v4hi); 53611 v4hi __builtin_vis_fsrl16 (v4hi, v4hi); 53612 v4hi __builtin_vis_fsra16 (v4hi, v4hi); 53613 v2si __builtin_vis_fsll16 (v2si, v2si); 53614 v2si __builtin_vis_fslas16 (v2si, v2si); 53615 v2si __builtin_vis_fsrl16 (v2si, v2si); 53616 v2si __builtin_vis_fsra16 (v2si, v2si); 53617 53618 long __builtin_vis_pdistn (v8qi, v8qi); 53619 53620 v4hi __builtin_vis_fmean16 (v4hi, v4hi); 53621 53622 int64_t __builtin_vis_fpadd64 (int64_t, int64_t); 53623 int64_t __builtin_vis_fpsub64 (int64_t, int64_t); 53624 53625 v4hi __builtin_vis_fpadds16 (v4hi, v4hi); 53626 v2hi __builtin_vis_fpadds16s (v2hi, v2hi); 53627 v4hi __builtin_vis_fpsubs16 (v4hi, v4hi); 53628 v2hi __builtin_vis_fpsubs16s (v2hi, v2hi); 53629 v2si __builtin_vis_fpadds32 (v2si, v2si); 53630 v1si __builtin_vis_fpadds32s (v1si, v1si); 53631 v2si __builtin_vis_fpsubs32 (v2si, v2si); 53632 v1si __builtin_vis_fpsubs32s (v1si, v1si); 53633 53634 long __builtin_vis_fucmple8 (v8qi, v8qi); 53635 long __builtin_vis_fucmpne8 (v8qi, v8qi); 53636 long __builtin_vis_fucmpgt8 (v8qi, v8qi); 53637 long __builtin_vis_fucmpeq8 (v8qi, v8qi); 53638 53639 float __builtin_vis_fhadds (float, float); 53640 double __builtin_vis_fhaddd (double, double); 53641 float __builtin_vis_fhsubs (float, float); 53642 double __builtin_vis_fhsubd (double, double); 53643 float __builtin_vis_fnhadds (float, float); 53644 double __builtin_vis_fnhaddd (double, double); 53645 53646 int64_t __builtin_vis_umulxhi (int64_t, int64_t); 53647 int64_t __builtin_vis_xmulx (int64_t, int64_t); 53648 int64_t __builtin_vis_xmulxhi (int64_t, int64_t); 53649 53650 When you use the '-mvis4' switch, the VIS version 4.0 built-in 53651functions also become available: 53652 53653 v8qi __builtin_vis_fpadd8 (v8qi, v8qi); 53654 v8qi __builtin_vis_fpadds8 (v8qi, v8qi); 53655 v8qi __builtin_vis_fpaddus8 (v8qi, v8qi); 53656 v4hi __builtin_vis_fpaddus16 (v4hi, v4hi); 53657 53658 v8qi __builtin_vis_fpsub8 (v8qi, v8qi); 53659 v8qi __builtin_vis_fpsubs8 (v8qi, v8qi); 53660 v8qi __builtin_vis_fpsubus8 (v8qi, v8qi); 53661 v4hi __builtin_vis_fpsubus16 (v4hi, v4hi); 53662 53663 long __builtin_vis_fpcmple8 (v8qi, v8qi); 53664 long __builtin_vis_fpcmpgt8 (v8qi, v8qi); 53665 long __builtin_vis_fpcmpule16 (v4hi, v4hi); 53666 long __builtin_vis_fpcmpugt16 (v4hi, v4hi); 53667 long __builtin_vis_fpcmpule32 (v2si, v2si); 53668 long __builtin_vis_fpcmpugt32 (v2si, v2si); 53669 53670 v8qi __builtin_vis_fpmax8 (v8qi, v8qi); 53671 v4hi __builtin_vis_fpmax16 (v4hi, v4hi); 53672 v2si __builtin_vis_fpmax32 (v2si, v2si); 53673 53674 v8qi __builtin_vis_fpmaxu8 (v8qi, v8qi); 53675 v4hi __builtin_vis_fpmaxu16 (v4hi, v4hi); 53676 v2si __builtin_vis_fpmaxu32 (v2si, v2si); 53677 53678 v8qi __builtin_vis_fpmin8 (v8qi, v8qi); 53679 v4hi __builtin_vis_fpmin16 (v4hi, v4hi); 53680 v2si __builtin_vis_fpmin32 (v2si, v2si); 53681 53682 v8qi __builtin_vis_fpminu8 (v8qi, v8qi); 53683 v4hi __builtin_vis_fpminu16 (v4hi, v4hi); 53684 v2si __builtin_vis_fpminu32 (v2si, v2si); 53685 53686 When you use the '-mvis4b' switch, the VIS version 4.0B built-in 53687functions also become available: 53688 53689 v8qi __builtin_vis_dictunpack8 (double, int); 53690 v4hi __builtin_vis_dictunpack16 (double, int); 53691 v2si __builtin_vis_dictunpack32 (double, int); 53692 53693 long __builtin_vis_fpcmple8shl (v8qi, v8qi, int); 53694 long __builtin_vis_fpcmpgt8shl (v8qi, v8qi, int); 53695 long __builtin_vis_fpcmpeq8shl (v8qi, v8qi, int); 53696 long __builtin_vis_fpcmpne8shl (v8qi, v8qi, int); 53697 53698 long __builtin_vis_fpcmple16shl (v4hi, v4hi, int); 53699 long __builtin_vis_fpcmpgt16shl (v4hi, v4hi, int); 53700 long __builtin_vis_fpcmpeq16shl (v4hi, v4hi, int); 53701 long __builtin_vis_fpcmpne16shl (v4hi, v4hi, int); 53702 53703 long __builtin_vis_fpcmple32shl (v2si, v2si, int); 53704 long __builtin_vis_fpcmpgt32shl (v2si, v2si, int); 53705 long __builtin_vis_fpcmpeq32shl (v2si, v2si, int); 53706 long __builtin_vis_fpcmpne32shl (v2si, v2si, int); 53707 53708 long __builtin_vis_fpcmpule8shl (v8qi, v8qi, int); 53709 long __builtin_vis_fpcmpugt8shl (v8qi, v8qi, int); 53710 long __builtin_vis_fpcmpule16shl (v4hi, v4hi, int); 53711 long __builtin_vis_fpcmpugt16shl (v4hi, v4hi, int); 53712 long __builtin_vis_fpcmpule32shl (v2si, v2si, int); 53713 long __builtin_vis_fpcmpugt32shl (v2si, v2si, int); 53714 53715 long __builtin_vis_fpcmpde8shl (v8qi, v8qi, int); 53716 long __builtin_vis_fpcmpde16shl (v4hi, v4hi, int); 53717 long __builtin_vis_fpcmpde32shl (v2si, v2si, int); 53718 53719 long __builtin_vis_fpcmpur8shl (v8qi, v8qi, int); 53720 long __builtin_vis_fpcmpur16shl (v4hi, v4hi, int); 53721 long __builtin_vis_fpcmpur32shl (v2si, v2si, int); 53722 53723 53724File: gcc.info, Node: TI C6X Built-in Functions, Next: TILE-Gx Built-in Functions, Prev: SPARC VIS Built-in Functions, Up: Target Builtins 53725 537266.60.33 TI C6X Built-in Functions 53727--------------------------------- 53728 53729GCC provides intrinsics to access certain instructions of the TI C6X 53730processors. These intrinsics, listed below, are available after 53731inclusion of the 'c6x_intrinsics.h' header file. They map directly to 53732C6X instructions. 53733 53734 int _sadd (int, int); 53735 int _ssub (int, int); 53736 int _sadd2 (int, int); 53737 int _ssub2 (int, int); 53738 long long _mpy2 (int, int); 53739 long long _smpy2 (int, int); 53740 int _add4 (int, int); 53741 int _sub4 (int, int); 53742 int _saddu4 (int, int); 53743 53744 int _smpy (int, int); 53745 int _smpyh (int, int); 53746 int _smpyhl (int, int); 53747 int _smpylh (int, int); 53748 53749 int _sshl (int, int); 53750 int _subc (int, int); 53751 53752 int _avg2 (int, int); 53753 int _avgu4 (int, int); 53754 53755 int _clrr (int, int); 53756 int _extr (int, int); 53757 int _extru (int, int); 53758 int _abs (int); 53759 int _abs2 (int); 53760 53761 53762File: gcc.info, Node: TILE-Gx Built-in Functions, Next: TILEPro Built-in Functions, Prev: TI C6X Built-in Functions, Up: Target Builtins 53763 537646.60.34 TILE-Gx Built-in Functions 53765---------------------------------- 53766 53767GCC provides intrinsics to access every instruction of the TILE-Gx 53768processor. The intrinsics are of the form: 53769 53770 53771 unsigned long long __insn_OP (...) 53772 53773 53774 Where OP is the name of the instruction. Refer to the ISA manual for 53775the complete list of instructions. 53776 53777 GCC also provides intrinsics to directly access the network registers. 53778The intrinsics are: 53779 53780 unsigned long long __tile_idn0_receive (void); 53781 unsigned long long __tile_idn1_receive (void); 53782 unsigned long long __tile_udn0_receive (void); 53783 unsigned long long __tile_udn1_receive (void); 53784 unsigned long long __tile_udn2_receive (void); 53785 unsigned long long __tile_udn3_receive (void); 53786 void __tile_idn_send (unsigned long long); 53787 void __tile_udn_send (unsigned long long); 53788 53789 The intrinsic 'void __tile_network_barrier (void)' is used to guarantee 53790that no network operations before it are reordered with those after it. 53791 53792 53793File: gcc.info, Node: TILEPro Built-in Functions, Next: x86 Built-in Functions, Prev: TILE-Gx Built-in Functions, Up: Target Builtins 53794 537956.60.35 TILEPro Built-in Functions 53796---------------------------------- 53797 53798GCC provides intrinsics to access every instruction of the TILEPro 53799processor. The intrinsics are of the form: 53800 53801 53802 unsigned __insn_OP (...) 53803 53804 53805where OP is the name of the instruction. Refer to the ISA manual for 53806the complete list of instructions. 53807 53808 GCC also provides intrinsics to directly access the network registers. 53809The intrinsics are: 53810 53811 unsigned __tile_idn0_receive (void); 53812 unsigned __tile_idn1_receive (void); 53813 unsigned __tile_sn_receive (void); 53814 unsigned __tile_udn0_receive (void); 53815 unsigned __tile_udn1_receive (void); 53816 unsigned __tile_udn2_receive (void); 53817 unsigned __tile_udn3_receive (void); 53818 void __tile_idn_send (unsigned); 53819 void __tile_sn_send (unsigned); 53820 void __tile_udn_send (unsigned); 53821 53822 The intrinsic 'void __tile_network_barrier (void)' is used to guarantee 53823that no network operations before it are reordered with those after it. 53824 53825 53826File: gcc.info, Node: x86 Built-in Functions, Next: x86 transactional memory intrinsics, Prev: TILEPro Built-in Functions, Up: Target Builtins 53827 538286.60.36 x86 Built-in Functions 53829------------------------------ 53830 53831These built-in functions are available for the x86-32 and x86-64 family 53832of computers, depending on the command-line switches used. 53833 53834 If you specify command-line switches such as '-msse', the compiler 53835could use the extended instruction sets even if the built-ins are not 53836used explicitly in the program. For this reason, applications that 53837perform run-time CPU detection must compile separate files for each 53838supported architecture, using the appropriate flags. In particular, the 53839file containing the CPU detection code should be compiled without these 53840options. 53841 53842 The following machine modes are available for use with MMX built-in 53843functions (*note Vector Extensions::): 'V2SI' for a vector of two 32-bit 53844integers, 'V4HI' for a vector of four 16-bit integers, and 'V8QI' for a 53845vector of eight 8-bit integers. Some of the built-in functions operate 53846on MMX registers as a whole 64-bit entity, these use 'V1DI' as their 53847mode. 53848 53849 If 3DNow! extensions are enabled, 'V2SF' is used as a mode for a vector 53850of two 32-bit floating-point values. 53851 53852 If SSE extensions are enabled, 'V4SF' is used for a vector of four 5385332-bit floating-point values. Some instructions use a vector of four 5385432-bit integers, these use 'V4SI'. Finally, some instructions operate 53855on an entire vector register, interpreting it as a 128-bit integer, 53856these use mode 'TI'. 53857 53858 The x86-32 and x86-64 family of processors use additional built-in 53859functions for efficient use of 'TF' ('__float128') 128-bit floating 53860point and 'TC' 128-bit complex floating-point values. 53861 53862 The following floating-point built-in functions are always available. 53863All of them implement the function that is part of the name. 53864 53865 __float128 __builtin_fabsq (__float128) 53866 __float128 __builtin_copysignq (__float128, __float128) 53867 53868 The following built-in functions are always available. 53869 53870'__float128 __builtin_infq (void)' 53871 Similar to '__builtin_inf', except the return type is '__float128'. 53872 53873'__float128 __builtin_huge_valq (void)' 53874 Similar to '__builtin_huge_val', except the return type is 53875 '__float128'. 53876 53877'__float128 __builtin_nanq (void)' 53878 Similar to '__builtin_nan', except the return type is '__float128'. 53879 53880'__float128 __builtin_nansq (void)' 53881 Similar to '__builtin_nans', except the return type is 53882 '__float128'. 53883 53884 The following built-in function is always available. 53885 53886'void __builtin_ia32_pause (void)' 53887 Generates the 'pause' machine instruction with a compiler memory 53888 barrier. 53889 53890 The following built-in functions are always available and can be used 53891to check the target platform type. 53892 53893 -- Built-in Function: void __builtin_cpu_init (void) 53894 This function runs the CPU detection code to check the type of CPU 53895 and the features supported. This built-in function needs to be 53896 invoked along with the built-in functions to check CPU type and 53897 features, '__builtin_cpu_is' and '__builtin_cpu_supports', only 53898 when used in a function that is executed before any constructors 53899 are called. The CPU detection code is automatically executed in a 53900 very high priority constructor. 53901 53902 For example, this function has to be used in 'ifunc' resolvers that 53903 check for CPU type using the built-in functions '__builtin_cpu_is' 53904 and '__builtin_cpu_supports', or in constructors on targets that 53905 don't support constructor priority. 53906 53907 static void (*resolve_memcpy (void)) (void) 53908 { 53909 // ifunc resolvers fire before constructors, explicitly call the init 53910 // function. 53911 __builtin_cpu_init (); 53912 if (__builtin_cpu_supports ("ssse3")) 53913 return ssse3_memcpy; // super fast memcpy with ssse3 instructions. 53914 else 53915 return default_memcpy; 53916 } 53917 53918 void *memcpy (void *, const void *, size_t) 53919 __attribute__ ((ifunc ("resolve_memcpy"))); 53920 53921 -- Built-in Function: int __builtin_cpu_is (const char *CPUNAME) 53922 This function returns a positive integer if the run-time CPU is of 53923 type CPUNAME and returns '0' otherwise. The following CPU names 53924 can be detected: 53925 53926 'amd' 53927 AMD CPU. 53928 53929 'intel' 53930 Intel CPU. 53931 53932 'atom' 53933 Intel Atom CPU. 53934 53935 'slm' 53936 Intel Silvermont CPU. 53937 53938 'core2' 53939 Intel Core 2 CPU. 53940 53941 'corei7' 53942 Intel Core i7 CPU. 53943 53944 'nehalem' 53945 Intel Core i7 Nehalem CPU. 53946 53947 'westmere' 53948 Intel Core i7 Westmere CPU. 53949 53950 'sandybridge' 53951 Intel Core i7 Sandy Bridge CPU. 53952 53953 'ivybridge' 53954 Intel Core i7 Ivy Bridge CPU. 53955 53956 'haswell' 53957 Intel Core i7 Haswell CPU. 53958 53959 'broadwell' 53960 Intel Core i7 Broadwell CPU. 53961 53962 'skylake' 53963 Intel Core i7 Skylake CPU. 53964 53965 'skylake-avx512' 53966 Intel Core i7 Skylake AVX512 CPU. 53967 53968 'cannonlake' 53969 Intel Core i7 Cannon Lake CPU. 53970 53971 'icelake-client' 53972 Intel Core i7 Ice Lake Client CPU. 53973 53974 'icelake-server' 53975 Intel Core i7 Ice Lake Server CPU. 53976 53977 'cascadelake' 53978 Intel Core i7 Cascadelake CPU. 53979 53980 'tigerlake' 53981 Intel Core i7 Tigerlake CPU. 53982 53983 'cooperlake' 53984 Intel Core i7 Cooperlake CPU. 53985 53986 'sapphirerapids' 53987 Intel Core i7 sapphirerapids CPU. 53988 53989 'alderlake' 53990 Intel Core i7 Alderlake CPU. 53991 53992 'rocketlake' 53993 Intel Core i7 Rocketlake CPU. 53994 53995 'bonnell' 53996 Intel Atom Bonnell CPU. 53997 53998 'silvermont' 53999 Intel Atom Silvermont CPU. 54000 54001 'goldmont' 54002 Intel Atom Goldmont CPU. 54003 54004 'goldmont-plus' 54005 Intel Atom Goldmont Plus CPU. 54006 54007 'tremont' 54008 Intel Atom Tremont CPU. 54009 54010 'knl' 54011 Intel Knights Landing CPU. 54012 54013 'knm' 54014 Intel Knights Mill CPU. 54015 54016 'amdfam10h' 54017 AMD Family 10h CPU. 54018 54019 'barcelona' 54020 AMD Family 10h Barcelona CPU. 54021 54022 'shanghai' 54023 AMD Family 10h Shanghai CPU. 54024 54025 'istanbul' 54026 AMD Family 10h Istanbul CPU. 54027 54028 'btver1' 54029 AMD Family 14h CPU. 54030 54031 'amdfam15h' 54032 AMD Family 15h CPU. 54033 54034 'bdver1' 54035 AMD Family 15h Bulldozer version 1. 54036 54037 'bdver2' 54038 AMD Family 15h Bulldozer version 2. 54039 54040 'bdver3' 54041 AMD Family 15h Bulldozer version 3. 54042 54043 'bdver4' 54044 AMD Family 15h Bulldozer version 4. 54045 54046 'btver2' 54047 AMD Family 16h CPU. 54048 54049 'amdfam17h' 54050 AMD Family 17h CPU. 54051 54052 'znver1' 54053 AMD Family 17h Zen version 1. 54054 54055 'znver2' 54056 AMD Family 17h Zen version 2. 54057 54058 'amdfam19h' 54059 AMD Family 19h CPU. 54060 54061 'znver3' 54062 AMD Family 19h Zen version 3. 54063 54064 'znver4' 54065 AMD Family 19h Zen version 4. 54066 54067 Here is an example: 54068 if (__builtin_cpu_is ("corei7")) 54069 { 54070 do_corei7 (); // Core i7 specific implementation. 54071 } 54072 else 54073 { 54074 do_generic (); // Generic implementation. 54075 } 54076 54077 -- Built-in Function: int __builtin_cpu_supports (const char *FEATURE) 54078 This function returns a positive integer if the run-time CPU 54079 supports FEATURE and returns '0' otherwise. The following features 54080 can be detected: 54081 54082 'cmov' 54083 CMOV instruction. 54084 'mmx' 54085 MMX instructions. 54086 'popcnt' 54087 POPCNT instruction. 54088 'sse' 54089 SSE instructions. 54090 'sse2' 54091 SSE2 instructions. 54092 'sse3' 54093 SSE3 instructions. 54094 'ssse3' 54095 SSSE3 instructions. 54096 'sse4.1' 54097 SSE4.1 instructions. 54098 'sse4.2' 54099 SSE4.2 instructions. 54100 'avx' 54101 AVX instructions. 54102 'avx2' 54103 AVX2 instructions. 54104 'sse4a' 54105 SSE4A instructions. 54106 'fma4' 54107 FMA4 instructions. 54108 'xop' 54109 XOP instructions. 54110 'fma' 54111 FMA instructions. 54112 'avx512f' 54113 AVX512F instructions. 54114 'bmi' 54115 BMI instructions. 54116 'bmi2' 54117 BMI2 instructions. 54118 'aes' 54119 AES instructions. 54120 'pclmul' 54121 PCLMUL instructions. 54122 'avx512vl' 54123 AVX512VL instructions. 54124 'avx512bw' 54125 AVX512BW instructions. 54126 'avx512dq' 54127 AVX512DQ instructions. 54128 'avx512cd' 54129 AVX512CD instructions. 54130 'avx512er' 54131 AVX512ER instructions. 54132 'avx512pf' 54133 AVX512PF instructions. 54134 'avx512vbmi' 54135 AVX512VBMI instructions. 54136 'avx512ifma' 54137 AVX512IFMA instructions. 54138 'avx5124vnniw' 54139 AVX5124VNNIW instructions. 54140 'avx5124fmaps' 54141 AVX5124FMAPS instructions. 54142 'avx512vpopcntdq' 54143 AVX512VPOPCNTDQ instructions. 54144 'avx512vbmi2' 54145 AVX512VBMI2 instructions. 54146 'gfni' 54147 GFNI instructions. 54148 'vpclmulqdq' 54149 VPCLMULQDQ instructions. 54150 'avx512vnni' 54151 AVX512VNNI instructions. 54152 'avx512bitalg' 54153 AVX512BITALG instructions. 54154 'x86-64' 54155 Baseline x86-64 microarchitecture level (as defined in x86-64 54156 psABI). 54157 'x86-64-v2' 54158 x86-64-v2 microarchitecture level. 54159 'x86-64-v3' 54160 x86-64-v3 microarchitecture level. 54161 'x86-64-v4' 54162 x86-64-v4 microarchitecture level. 54163 54164 Here is an example: 54165 if (__builtin_cpu_supports ("popcnt")) 54166 { 54167 asm("popcnt %1,%0" : "=r"(count) : "rm"(n) : "cc"); 54168 } 54169 else 54170 { 54171 count = generic_countbits (n); //generic implementation. 54172 } 54173 54174 The following built-in functions are made available by '-mmmx'. All of 54175them generate the machine instruction that is part of the name. 54176 54177 v8qi __builtin_ia32_paddb (v8qi, v8qi); 54178 v4hi __builtin_ia32_paddw (v4hi, v4hi); 54179 v2si __builtin_ia32_paddd (v2si, v2si); 54180 v8qi __builtin_ia32_psubb (v8qi, v8qi); 54181 v4hi __builtin_ia32_psubw (v4hi, v4hi); 54182 v2si __builtin_ia32_psubd (v2si, v2si); 54183 v8qi __builtin_ia32_paddsb (v8qi, v8qi); 54184 v4hi __builtin_ia32_paddsw (v4hi, v4hi); 54185 v8qi __builtin_ia32_psubsb (v8qi, v8qi); 54186 v4hi __builtin_ia32_psubsw (v4hi, v4hi); 54187 v8qi __builtin_ia32_paddusb (v8qi, v8qi); 54188 v4hi __builtin_ia32_paddusw (v4hi, v4hi); 54189 v8qi __builtin_ia32_psubusb (v8qi, v8qi); 54190 v4hi __builtin_ia32_psubusw (v4hi, v4hi); 54191 v4hi __builtin_ia32_pmullw (v4hi, v4hi); 54192 v4hi __builtin_ia32_pmulhw (v4hi, v4hi); 54193 di __builtin_ia32_pand (di, di); 54194 di __builtin_ia32_pandn (di,di); 54195 di __builtin_ia32_por (di, di); 54196 di __builtin_ia32_pxor (di, di); 54197 v8qi __builtin_ia32_pcmpeqb (v8qi, v8qi); 54198 v4hi __builtin_ia32_pcmpeqw (v4hi, v4hi); 54199 v2si __builtin_ia32_pcmpeqd (v2si, v2si); 54200 v8qi __builtin_ia32_pcmpgtb (v8qi, v8qi); 54201 v4hi __builtin_ia32_pcmpgtw (v4hi, v4hi); 54202 v2si __builtin_ia32_pcmpgtd (v2si, v2si); 54203 v8qi __builtin_ia32_punpckhbw (v8qi, v8qi); 54204 v4hi __builtin_ia32_punpckhwd (v4hi, v4hi); 54205 v2si __builtin_ia32_punpckhdq (v2si, v2si); 54206 v8qi __builtin_ia32_punpcklbw (v8qi, v8qi); 54207 v4hi __builtin_ia32_punpcklwd (v4hi, v4hi); 54208 v2si __builtin_ia32_punpckldq (v2si, v2si); 54209 v8qi __builtin_ia32_packsswb (v4hi, v4hi); 54210 v4hi __builtin_ia32_packssdw (v2si, v2si); 54211 v8qi __builtin_ia32_packuswb (v4hi, v4hi); 54212 54213 v4hi __builtin_ia32_psllw (v4hi, v4hi); 54214 v2si __builtin_ia32_pslld (v2si, v2si); 54215 v1di __builtin_ia32_psllq (v1di, v1di); 54216 v4hi __builtin_ia32_psrlw (v4hi, v4hi); 54217 v2si __builtin_ia32_psrld (v2si, v2si); 54218 v1di __builtin_ia32_psrlq (v1di, v1di); 54219 v4hi __builtin_ia32_psraw (v4hi, v4hi); 54220 v2si __builtin_ia32_psrad (v2si, v2si); 54221 v4hi __builtin_ia32_psllwi (v4hi, int); 54222 v2si __builtin_ia32_pslldi (v2si, int); 54223 v1di __builtin_ia32_psllqi (v1di, int); 54224 v4hi __builtin_ia32_psrlwi (v4hi, int); 54225 v2si __builtin_ia32_psrldi (v2si, int); 54226 v1di __builtin_ia32_psrlqi (v1di, int); 54227 v4hi __builtin_ia32_psrawi (v4hi, int); 54228 v2si __builtin_ia32_psradi (v2si, int); 54229 54230 The following built-in functions are made available either with 54231'-msse', or with '-m3dnowa'. All of them generate the machine 54232instruction that is part of the name. 54233 54234 v4hi __builtin_ia32_pmulhuw (v4hi, v4hi); 54235 v8qi __builtin_ia32_pavgb (v8qi, v8qi); 54236 v4hi __builtin_ia32_pavgw (v4hi, v4hi); 54237 v1di __builtin_ia32_psadbw (v8qi, v8qi); 54238 v8qi __builtin_ia32_pmaxub (v8qi, v8qi); 54239 v4hi __builtin_ia32_pmaxsw (v4hi, v4hi); 54240 v8qi __builtin_ia32_pminub (v8qi, v8qi); 54241 v4hi __builtin_ia32_pminsw (v4hi, v4hi); 54242 int __builtin_ia32_pmovmskb (v8qi); 54243 void __builtin_ia32_maskmovq (v8qi, v8qi, char *); 54244 void __builtin_ia32_movntq (di *, di); 54245 void __builtin_ia32_sfence (void); 54246 54247 The following built-in functions are available when '-msse' is used. 54248All of them generate the machine instruction that is part of the name. 54249 54250 int __builtin_ia32_comieq (v4sf, v4sf); 54251 int __builtin_ia32_comineq (v4sf, v4sf); 54252 int __builtin_ia32_comilt (v4sf, v4sf); 54253 int __builtin_ia32_comile (v4sf, v4sf); 54254 int __builtin_ia32_comigt (v4sf, v4sf); 54255 int __builtin_ia32_comige (v4sf, v4sf); 54256 int __builtin_ia32_ucomieq (v4sf, v4sf); 54257 int __builtin_ia32_ucomineq (v4sf, v4sf); 54258 int __builtin_ia32_ucomilt (v4sf, v4sf); 54259 int __builtin_ia32_ucomile (v4sf, v4sf); 54260 int __builtin_ia32_ucomigt (v4sf, v4sf); 54261 int __builtin_ia32_ucomige (v4sf, v4sf); 54262 v4sf __builtin_ia32_addps (v4sf, v4sf); 54263 v4sf __builtin_ia32_subps (v4sf, v4sf); 54264 v4sf __builtin_ia32_mulps (v4sf, v4sf); 54265 v4sf __builtin_ia32_divps (v4sf, v4sf); 54266 v4sf __builtin_ia32_addss (v4sf, v4sf); 54267 v4sf __builtin_ia32_subss (v4sf, v4sf); 54268 v4sf __builtin_ia32_mulss (v4sf, v4sf); 54269 v4sf __builtin_ia32_divss (v4sf, v4sf); 54270 v4sf __builtin_ia32_cmpeqps (v4sf, v4sf); 54271 v4sf __builtin_ia32_cmpltps (v4sf, v4sf); 54272 v4sf __builtin_ia32_cmpleps (v4sf, v4sf); 54273 v4sf __builtin_ia32_cmpgtps (v4sf, v4sf); 54274 v4sf __builtin_ia32_cmpgeps (v4sf, v4sf); 54275 v4sf __builtin_ia32_cmpunordps (v4sf, v4sf); 54276 v4sf __builtin_ia32_cmpneqps (v4sf, v4sf); 54277 v4sf __builtin_ia32_cmpnltps (v4sf, v4sf); 54278 v4sf __builtin_ia32_cmpnleps (v4sf, v4sf); 54279 v4sf __builtin_ia32_cmpngtps (v4sf, v4sf); 54280 v4sf __builtin_ia32_cmpngeps (v4sf, v4sf); 54281 v4sf __builtin_ia32_cmpordps (v4sf, v4sf); 54282 v4sf __builtin_ia32_cmpeqss (v4sf, v4sf); 54283 v4sf __builtin_ia32_cmpltss (v4sf, v4sf); 54284 v4sf __builtin_ia32_cmpless (v4sf, v4sf); 54285 v4sf __builtin_ia32_cmpunordss (v4sf, v4sf); 54286 v4sf __builtin_ia32_cmpneqss (v4sf, v4sf); 54287 v4sf __builtin_ia32_cmpnltss (v4sf, v4sf); 54288 v4sf __builtin_ia32_cmpnless (v4sf, v4sf); 54289 v4sf __builtin_ia32_cmpordss (v4sf, v4sf); 54290 v4sf __builtin_ia32_maxps (v4sf, v4sf); 54291 v4sf __builtin_ia32_maxss (v4sf, v4sf); 54292 v4sf __builtin_ia32_minps (v4sf, v4sf); 54293 v4sf __builtin_ia32_minss (v4sf, v4sf); 54294 v4sf __builtin_ia32_andps (v4sf, v4sf); 54295 v4sf __builtin_ia32_andnps (v4sf, v4sf); 54296 v4sf __builtin_ia32_orps (v4sf, v4sf); 54297 v4sf __builtin_ia32_xorps (v4sf, v4sf); 54298 v4sf __builtin_ia32_movss (v4sf, v4sf); 54299 v4sf __builtin_ia32_movhlps (v4sf, v4sf); 54300 v4sf __builtin_ia32_movlhps (v4sf, v4sf); 54301 v4sf __builtin_ia32_unpckhps (v4sf, v4sf); 54302 v4sf __builtin_ia32_unpcklps (v4sf, v4sf); 54303 v4sf __builtin_ia32_cvtpi2ps (v4sf, v2si); 54304 v4sf __builtin_ia32_cvtsi2ss (v4sf, int); 54305 v2si __builtin_ia32_cvtps2pi (v4sf); 54306 int __builtin_ia32_cvtss2si (v4sf); 54307 v2si __builtin_ia32_cvttps2pi (v4sf); 54308 int __builtin_ia32_cvttss2si (v4sf); 54309 v4sf __builtin_ia32_rcpps (v4sf); 54310 v4sf __builtin_ia32_rsqrtps (v4sf); 54311 v4sf __builtin_ia32_sqrtps (v4sf); 54312 v4sf __builtin_ia32_rcpss (v4sf); 54313 v4sf __builtin_ia32_rsqrtss (v4sf); 54314 v4sf __builtin_ia32_sqrtss (v4sf); 54315 v4sf __builtin_ia32_shufps (v4sf, v4sf, int); 54316 void __builtin_ia32_movntps (float *, v4sf); 54317 int __builtin_ia32_movmskps (v4sf); 54318 54319 The following built-in functions are available when '-msse' is used. 54320 54321'v4sf __builtin_ia32_loadups (float *)' 54322 Generates the 'movups' machine instruction as a load from memory. 54323'void __builtin_ia32_storeups (float *, v4sf)' 54324 Generates the 'movups' machine instruction as a store to memory. 54325'v4sf __builtin_ia32_loadss (float *)' 54326 Generates the 'movss' machine instruction as a load from memory. 54327'v4sf __builtin_ia32_loadhps (v4sf, const v2sf *)' 54328 Generates the 'movhps' machine instruction as a load from memory. 54329'v4sf __builtin_ia32_loadlps (v4sf, const v2sf *)' 54330 Generates the 'movlps' machine instruction as a load from memory 54331'void __builtin_ia32_storehps (v2sf *, v4sf)' 54332 Generates the 'movhps' machine instruction as a store to memory. 54333'void __builtin_ia32_storelps (v2sf *, v4sf)' 54334 Generates the 'movlps' machine instruction as a store to memory. 54335 54336 The following built-in functions are available when '-msse2' is used. 54337All of them generate the machine instruction that is part of the name. 54338 54339 int __builtin_ia32_comisdeq (v2df, v2df); 54340 int __builtin_ia32_comisdlt (v2df, v2df); 54341 int __builtin_ia32_comisdle (v2df, v2df); 54342 int __builtin_ia32_comisdgt (v2df, v2df); 54343 int __builtin_ia32_comisdge (v2df, v2df); 54344 int __builtin_ia32_comisdneq (v2df, v2df); 54345 int __builtin_ia32_ucomisdeq (v2df, v2df); 54346 int __builtin_ia32_ucomisdlt (v2df, v2df); 54347 int __builtin_ia32_ucomisdle (v2df, v2df); 54348 int __builtin_ia32_ucomisdgt (v2df, v2df); 54349 int __builtin_ia32_ucomisdge (v2df, v2df); 54350 int __builtin_ia32_ucomisdneq (v2df, v2df); 54351 v2df __builtin_ia32_cmpeqpd (v2df, v2df); 54352 v2df __builtin_ia32_cmpltpd (v2df, v2df); 54353 v2df __builtin_ia32_cmplepd (v2df, v2df); 54354 v2df __builtin_ia32_cmpgtpd (v2df, v2df); 54355 v2df __builtin_ia32_cmpgepd (v2df, v2df); 54356 v2df __builtin_ia32_cmpunordpd (v2df, v2df); 54357 v2df __builtin_ia32_cmpneqpd (v2df, v2df); 54358 v2df __builtin_ia32_cmpnltpd (v2df, v2df); 54359 v2df __builtin_ia32_cmpnlepd (v2df, v2df); 54360 v2df __builtin_ia32_cmpngtpd (v2df, v2df); 54361 v2df __builtin_ia32_cmpngepd (v2df, v2df); 54362 v2df __builtin_ia32_cmpordpd (v2df, v2df); 54363 v2df __builtin_ia32_cmpeqsd (v2df, v2df); 54364 v2df __builtin_ia32_cmpltsd (v2df, v2df); 54365 v2df __builtin_ia32_cmplesd (v2df, v2df); 54366 v2df __builtin_ia32_cmpunordsd (v2df, v2df); 54367 v2df __builtin_ia32_cmpneqsd (v2df, v2df); 54368 v2df __builtin_ia32_cmpnltsd (v2df, v2df); 54369 v2df __builtin_ia32_cmpnlesd (v2df, v2df); 54370 v2df __builtin_ia32_cmpordsd (v2df, v2df); 54371 v2di __builtin_ia32_paddq (v2di, v2di); 54372 v2di __builtin_ia32_psubq (v2di, v2di); 54373 v2df __builtin_ia32_addpd (v2df, v2df); 54374 v2df __builtin_ia32_subpd (v2df, v2df); 54375 v2df __builtin_ia32_mulpd (v2df, v2df); 54376 v2df __builtin_ia32_divpd (v2df, v2df); 54377 v2df __builtin_ia32_addsd (v2df, v2df); 54378 v2df __builtin_ia32_subsd (v2df, v2df); 54379 v2df __builtin_ia32_mulsd (v2df, v2df); 54380 v2df __builtin_ia32_divsd (v2df, v2df); 54381 v2df __builtin_ia32_minpd (v2df, v2df); 54382 v2df __builtin_ia32_maxpd (v2df, v2df); 54383 v2df __builtin_ia32_minsd (v2df, v2df); 54384 v2df __builtin_ia32_maxsd (v2df, v2df); 54385 v2df __builtin_ia32_andpd (v2df, v2df); 54386 v2df __builtin_ia32_andnpd (v2df, v2df); 54387 v2df __builtin_ia32_orpd (v2df, v2df); 54388 v2df __builtin_ia32_xorpd (v2df, v2df); 54389 v2df __builtin_ia32_movsd (v2df, v2df); 54390 v2df __builtin_ia32_unpckhpd (v2df, v2df); 54391 v2df __builtin_ia32_unpcklpd (v2df, v2df); 54392 v16qi __builtin_ia32_paddb128 (v16qi, v16qi); 54393 v8hi __builtin_ia32_paddw128 (v8hi, v8hi); 54394 v4si __builtin_ia32_paddd128 (v4si, v4si); 54395 v2di __builtin_ia32_paddq128 (v2di, v2di); 54396 v16qi __builtin_ia32_psubb128 (v16qi, v16qi); 54397 v8hi __builtin_ia32_psubw128 (v8hi, v8hi); 54398 v4si __builtin_ia32_psubd128 (v4si, v4si); 54399 v2di __builtin_ia32_psubq128 (v2di, v2di); 54400 v8hi __builtin_ia32_pmullw128 (v8hi, v8hi); 54401 v8hi __builtin_ia32_pmulhw128 (v8hi, v8hi); 54402 v2di __builtin_ia32_pand128 (v2di, v2di); 54403 v2di __builtin_ia32_pandn128 (v2di, v2di); 54404 v2di __builtin_ia32_por128 (v2di, v2di); 54405 v2di __builtin_ia32_pxor128 (v2di, v2di); 54406 v16qi __builtin_ia32_pavgb128 (v16qi, v16qi); 54407 v8hi __builtin_ia32_pavgw128 (v8hi, v8hi); 54408 v16qi __builtin_ia32_pcmpeqb128 (v16qi, v16qi); 54409 v8hi __builtin_ia32_pcmpeqw128 (v8hi, v8hi); 54410 v4si __builtin_ia32_pcmpeqd128 (v4si, v4si); 54411 v16qi __builtin_ia32_pcmpgtb128 (v16qi, v16qi); 54412 v8hi __builtin_ia32_pcmpgtw128 (v8hi, v8hi); 54413 v4si __builtin_ia32_pcmpgtd128 (v4si, v4si); 54414 v16qi __builtin_ia32_pmaxub128 (v16qi, v16qi); 54415 v8hi __builtin_ia32_pmaxsw128 (v8hi, v8hi); 54416 v16qi __builtin_ia32_pminub128 (v16qi, v16qi); 54417 v8hi __builtin_ia32_pminsw128 (v8hi, v8hi); 54418 v16qi __builtin_ia32_punpckhbw128 (v16qi, v16qi); 54419 v8hi __builtin_ia32_punpckhwd128 (v8hi, v8hi); 54420 v4si __builtin_ia32_punpckhdq128 (v4si, v4si); 54421 v2di __builtin_ia32_punpckhqdq128 (v2di, v2di); 54422 v16qi __builtin_ia32_punpcklbw128 (v16qi, v16qi); 54423 v8hi __builtin_ia32_punpcklwd128 (v8hi, v8hi); 54424 v4si __builtin_ia32_punpckldq128 (v4si, v4si); 54425 v2di __builtin_ia32_punpcklqdq128 (v2di, v2di); 54426 v16qi __builtin_ia32_packsswb128 (v8hi, v8hi); 54427 v8hi __builtin_ia32_packssdw128 (v4si, v4si); 54428 v16qi __builtin_ia32_packuswb128 (v8hi, v8hi); 54429 v8hi __builtin_ia32_pmulhuw128 (v8hi, v8hi); 54430 void __builtin_ia32_maskmovdqu (v16qi, v16qi); 54431 v2df __builtin_ia32_loadupd (double *); 54432 void __builtin_ia32_storeupd (double *, v2df); 54433 v2df __builtin_ia32_loadhpd (v2df, double const *); 54434 v2df __builtin_ia32_loadlpd (v2df, double const *); 54435 int __builtin_ia32_movmskpd (v2df); 54436 int __builtin_ia32_pmovmskb128 (v16qi); 54437 void __builtin_ia32_movnti (int *, int); 54438 void __builtin_ia32_movnti64 (long long int *, long long int); 54439 void __builtin_ia32_movntpd (double *, v2df); 54440 void __builtin_ia32_movntdq (v2df *, v2df); 54441 v4si __builtin_ia32_pshufd (v4si, int); 54442 v8hi __builtin_ia32_pshuflw (v8hi, int); 54443 v8hi __builtin_ia32_pshufhw (v8hi, int); 54444 v2di __builtin_ia32_psadbw128 (v16qi, v16qi); 54445 v2df __builtin_ia32_sqrtpd (v2df); 54446 v2df __builtin_ia32_sqrtsd (v2df); 54447 v2df __builtin_ia32_shufpd (v2df, v2df, int); 54448 v2df __builtin_ia32_cvtdq2pd (v4si); 54449 v4sf __builtin_ia32_cvtdq2ps (v4si); 54450 v4si __builtin_ia32_cvtpd2dq (v2df); 54451 v2si __builtin_ia32_cvtpd2pi (v2df); 54452 v4sf __builtin_ia32_cvtpd2ps (v2df); 54453 v4si __builtin_ia32_cvttpd2dq (v2df); 54454 v2si __builtin_ia32_cvttpd2pi (v2df); 54455 v2df __builtin_ia32_cvtpi2pd (v2si); 54456 int __builtin_ia32_cvtsd2si (v2df); 54457 int __builtin_ia32_cvttsd2si (v2df); 54458 long long __builtin_ia32_cvtsd2si64 (v2df); 54459 long long __builtin_ia32_cvttsd2si64 (v2df); 54460 v4si __builtin_ia32_cvtps2dq (v4sf); 54461 v2df __builtin_ia32_cvtps2pd (v4sf); 54462 v4si __builtin_ia32_cvttps2dq (v4sf); 54463 v2df __builtin_ia32_cvtsi2sd (v2df, int); 54464 v2df __builtin_ia32_cvtsi642sd (v2df, long long); 54465 v4sf __builtin_ia32_cvtsd2ss (v4sf, v2df); 54466 v2df __builtin_ia32_cvtss2sd (v2df, v4sf); 54467 void __builtin_ia32_clflush (const void *); 54468 void __builtin_ia32_lfence (void); 54469 void __builtin_ia32_mfence (void); 54470 v16qi __builtin_ia32_loaddqu (const char *); 54471 void __builtin_ia32_storedqu (char *, v16qi); 54472 v1di __builtin_ia32_pmuludq (v2si, v2si); 54473 v2di __builtin_ia32_pmuludq128 (v4si, v4si); 54474 v8hi __builtin_ia32_psllw128 (v8hi, v8hi); 54475 v4si __builtin_ia32_pslld128 (v4si, v4si); 54476 v2di __builtin_ia32_psllq128 (v2di, v2di); 54477 v8hi __builtin_ia32_psrlw128 (v8hi, v8hi); 54478 v4si __builtin_ia32_psrld128 (v4si, v4si); 54479 v2di __builtin_ia32_psrlq128 (v2di, v2di); 54480 v8hi __builtin_ia32_psraw128 (v8hi, v8hi); 54481 v4si __builtin_ia32_psrad128 (v4si, v4si); 54482 v2di __builtin_ia32_pslldqi128 (v2di, int); 54483 v8hi __builtin_ia32_psllwi128 (v8hi, int); 54484 v4si __builtin_ia32_pslldi128 (v4si, int); 54485 v2di __builtin_ia32_psllqi128 (v2di, int); 54486 v2di __builtin_ia32_psrldqi128 (v2di, int); 54487 v8hi __builtin_ia32_psrlwi128 (v8hi, int); 54488 v4si __builtin_ia32_psrldi128 (v4si, int); 54489 v2di __builtin_ia32_psrlqi128 (v2di, int); 54490 v8hi __builtin_ia32_psrawi128 (v8hi, int); 54491 v4si __builtin_ia32_psradi128 (v4si, int); 54492 v4si __builtin_ia32_pmaddwd128 (v8hi, v8hi); 54493 v2di __builtin_ia32_movq128 (v2di); 54494 54495 The following built-in functions are available when '-msse3' is used. 54496All of them generate the machine instruction that is part of the name. 54497 54498 v2df __builtin_ia32_addsubpd (v2df, v2df); 54499 v4sf __builtin_ia32_addsubps (v4sf, v4sf); 54500 v2df __builtin_ia32_haddpd (v2df, v2df); 54501 v4sf __builtin_ia32_haddps (v4sf, v4sf); 54502 v2df __builtin_ia32_hsubpd (v2df, v2df); 54503 v4sf __builtin_ia32_hsubps (v4sf, v4sf); 54504 v16qi __builtin_ia32_lddqu (char const *); 54505 void __builtin_ia32_monitor (void *, unsigned int, unsigned int); 54506 v4sf __builtin_ia32_movshdup (v4sf); 54507 v4sf __builtin_ia32_movsldup (v4sf); 54508 void __builtin_ia32_mwait (unsigned int, unsigned int); 54509 54510 The following built-in functions are available when '-mssse3' is used. 54511All of them generate the machine instruction that is part of the name. 54512 54513 v2si __builtin_ia32_phaddd (v2si, v2si); 54514 v4hi __builtin_ia32_phaddw (v4hi, v4hi); 54515 v4hi __builtin_ia32_phaddsw (v4hi, v4hi); 54516 v2si __builtin_ia32_phsubd (v2si, v2si); 54517 v4hi __builtin_ia32_phsubw (v4hi, v4hi); 54518 v4hi __builtin_ia32_phsubsw (v4hi, v4hi); 54519 v4hi __builtin_ia32_pmaddubsw (v8qi, v8qi); 54520 v4hi __builtin_ia32_pmulhrsw (v4hi, v4hi); 54521 v8qi __builtin_ia32_pshufb (v8qi, v8qi); 54522 v8qi __builtin_ia32_psignb (v8qi, v8qi); 54523 v2si __builtin_ia32_psignd (v2si, v2si); 54524 v4hi __builtin_ia32_psignw (v4hi, v4hi); 54525 v1di __builtin_ia32_palignr (v1di, v1di, int); 54526 v8qi __builtin_ia32_pabsb (v8qi); 54527 v2si __builtin_ia32_pabsd (v2si); 54528 v4hi __builtin_ia32_pabsw (v4hi); 54529 54530 The following built-in functions are available when '-mssse3' is used. 54531All of them generate the machine instruction that is part of the name. 54532 54533 v4si __builtin_ia32_phaddd128 (v4si, v4si); 54534 v8hi __builtin_ia32_phaddw128 (v8hi, v8hi); 54535 v8hi __builtin_ia32_phaddsw128 (v8hi, v8hi); 54536 v4si __builtin_ia32_phsubd128 (v4si, v4si); 54537 v8hi __builtin_ia32_phsubw128 (v8hi, v8hi); 54538 v8hi __builtin_ia32_phsubsw128 (v8hi, v8hi); 54539 v8hi __builtin_ia32_pmaddubsw128 (v16qi, v16qi); 54540 v8hi __builtin_ia32_pmulhrsw128 (v8hi, v8hi); 54541 v16qi __builtin_ia32_pshufb128 (v16qi, v16qi); 54542 v16qi __builtin_ia32_psignb128 (v16qi, v16qi); 54543 v4si __builtin_ia32_psignd128 (v4si, v4si); 54544 v8hi __builtin_ia32_psignw128 (v8hi, v8hi); 54545 v2di __builtin_ia32_palignr128 (v2di, v2di, int); 54546 v16qi __builtin_ia32_pabsb128 (v16qi); 54547 v4si __builtin_ia32_pabsd128 (v4si); 54548 v8hi __builtin_ia32_pabsw128 (v8hi); 54549 54550 The following built-in functions are available when '-msse4.1' is used. 54551All of them generate the machine instruction that is part of the name. 54552 54553 v2df __builtin_ia32_blendpd (v2df, v2df, const int); 54554 v4sf __builtin_ia32_blendps (v4sf, v4sf, const int); 54555 v2df __builtin_ia32_blendvpd (v2df, v2df, v2df); 54556 v4sf __builtin_ia32_blendvps (v4sf, v4sf, v4sf); 54557 v2df __builtin_ia32_dppd (v2df, v2df, const int); 54558 v4sf __builtin_ia32_dpps (v4sf, v4sf, const int); 54559 v4sf __builtin_ia32_insertps128 (v4sf, v4sf, const int); 54560 v2di __builtin_ia32_movntdqa (v2di *); 54561 v16qi __builtin_ia32_mpsadbw128 (v16qi, v16qi, const int); 54562 v8hi __builtin_ia32_packusdw128 (v4si, v4si); 54563 v16qi __builtin_ia32_pblendvb128 (v16qi, v16qi, v16qi); 54564 v8hi __builtin_ia32_pblendw128 (v8hi, v8hi, const int); 54565 v2di __builtin_ia32_pcmpeqq (v2di, v2di); 54566 v8hi __builtin_ia32_phminposuw128 (v8hi); 54567 v16qi __builtin_ia32_pmaxsb128 (v16qi, v16qi); 54568 v4si __builtin_ia32_pmaxsd128 (v4si, v4si); 54569 v4si __builtin_ia32_pmaxud128 (v4si, v4si); 54570 v8hi __builtin_ia32_pmaxuw128 (v8hi, v8hi); 54571 v16qi __builtin_ia32_pminsb128 (v16qi, v16qi); 54572 v4si __builtin_ia32_pminsd128 (v4si, v4si); 54573 v4si __builtin_ia32_pminud128 (v4si, v4si); 54574 v8hi __builtin_ia32_pminuw128 (v8hi, v8hi); 54575 v4si __builtin_ia32_pmovsxbd128 (v16qi); 54576 v2di __builtin_ia32_pmovsxbq128 (v16qi); 54577 v8hi __builtin_ia32_pmovsxbw128 (v16qi); 54578 v2di __builtin_ia32_pmovsxdq128 (v4si); 54579 v4si __builtin_ia32_pmovsxwd128 (v8hi); 54580 v2di __builtin_ia32_pmovsxwq128 (v8hi); 54581 v4si __builtin_ia32_pmovzxbd128 (v16qi); 54582 v2di __builtin_ia32_pmovzxbq128 (v16qi); 54583 v8hi __builtin_ia32_pmovzxbw128 (v16qi); 54584 v2di __builtin_ia32_pmovzxdq128 (v4si); 54585 v4si __builtin_ia32_pmovzxwd128 (v8hi); 54586 v2di __builtin_ia32_pmovzxwq128 (v8hi); 54587 v2di __builtin_ia32_pmuldq128 (v4si, v4si); 54588 v4si __builtin_ia32_pmulld128 (v4si, v4si); 54589 int __builtin_ia32_ptestc128 (v2di, v2di); 54590 int __builtin_ia32_ptestnzc128 (v2di, v2di); 54591 int __builtin_ia32_ptestz128 (v2di, v2di); 54592 v2df __builtin_ia32_roundpd (v2df, const int); 54593 v4sf __builtin_ia32_roundps (v4sf, const int); 54594 v2df __builtin_ia32_roundsd (v2df, v2df, const int); 54595 v4sf __builtin_ia32_roundss (v4sf, v4sf, const int); 54596 54597 The following built-in functions are available when '-msse4.1' is used. 54598 54599'v4sf __builtin_ia32_vec_set_v4sf (v4sf, float, const int)' 54600 Generates the 'insertps' machine instruction. 54601'int __builtin_ia32_vec_ext_v16qi (v16qi, const int)' 54602 Generates the 'pextrb' machine instruction. 54603'v16qi __builtin_ia32_vec_set_v16qi (v16qi, int, const int)' 54604 Generates the 'pinsrb' machine instruction. 54605'v4si __builtin_ia32_vec_set_v4si (v4si, int, const int)' 54606 Generates the 'pinsrd' machine instruction. 54607'v2di __builtin_ia32_vec_set_v2di (v2di, long long, const int)' 54608 Generates the 'pinsrq' machine instruction in 64bit mode. 54609 54610 The following built-in functions are changed to generate new SSE4.1 54611instructions when '-msse4.1' is used. 54612 54613'float __builtin_ia32_vec_ext_v4sf (v4sf, const int)' 54614 Generates the 'extractps' machine instruction. 54615'int __builtin_ia32_vec_ext_v4si (v4si, const int)' 54616 Generates the 'pextrd' machine instruction. 54617'long long __builtin_ia32_vec_ext_v2di (v2di, const int)' 54618 Generates the 'pextrq' machine instruction in 64bit mode. 54619 54620 The following built-in functions are available when '-msse4.2' is used. 54621All of them generate the machine instruction that is part of the name. 54622 54623 v16qi __builtin_ia32_pcmpestrm128 (v16qi, int, v16qi, int, const int); 54624 int __builtin_ia32_pcmpestri128 (v16qi, int, v16qi, int, const int); 54625 int __builtin_ia32_pcmpestria128 (v16qi, int, v16qi, int, const int); 54626 int __builtin_ia32_pcmpestric128 (v16qi, int, v16qi, int, const int); 54627 int __builtin_ia32_pcmpestrio128 (v16qi, int, v16qi, int, const int); 54628 int __builtin_ia32_pcmpestris128 (v16qi, int, v16qi, int, const int); 54629 int __builtin_ia32_pcmpestriz128 (v16qi, int, v16qi, int, const int); 54630 v16qi __builtin_ia32_pcmpistrm128 (v16qi, v16qi, const int); 54631 int __builtin_ia32_pcmpistri128 (v16qi, v16qi, const int); 54632 int __builtin_ia32_pcmpistria128 (v16qi, v16qi, const int); 54633 int __builtin_ia32_pcmpistric128 (v16qi, v16qi, const int); 54634 int __builtin_ia32_pcmpistrio128 (v16qi, v16qi, const int); 54635 int __builtin_ia32_pcmpistris128 (v16qi, v16qi, const int); 54636 int __builtin_ia32_pcmpistriz128 (v16qi, v16qi, const int); 54637 v2di __builtin_ia32_pcmpgtq (v2di, v2di); 54638 54639 The following built-in functions are available when '-msse4.2' is used. 54640 54641'unsigned int __builtin_ia32_crc32qi (unsigned int, unsigned char)' 54642 Generates the 'crc32b' machine instruction. 54643'unsigned int __builtin_ia32_crc32hi (unsigned int, unsigned short)' 54644 Generates the 'crc32w' machine instruction. 54645'unsigned int __builtin_ia32_crc32si (unsigned int, unsigned int)' 54646 Generates the 'crc32l' machine instruction. 54647'unsigned long long __builtin_ia32_crc32di (unsigned long long, unsigned long long)' 54648 Generates the 'crc32q' machine instruction. 54649 54650 The following built-in functions are changed to generate new SSE4.2 54651instructions when '-msse4.2' is used. 54652 54653'int __builtin_popcount (unsigned int)' 54654 Generates the 'popcntl' machine instruction. 54655'int __builtin_popcountl (unsigned long)' 54656 Generates the 'popcntl' or 'popcntq' machine instruction, depending 54657 on the size of 'unsigned long'. 54658'int __builtin_popcountll (unsigned long long)' 54659 Generates the 'popcntq' machine instruction. 54660 54661 The following built-in functions are available when '-mavx' is used. 54662All of them generate the machine instruction that is part of the name. 54663 54664 v4df __builtin_ia32_addpd256 (v4df,v4df); 54665 v8sf __builtin_ia32_addps256 (v8sf,v8sf); 54666 v4df __builtin_ia32_addsubpd256 (v4df,v4df); 54667 v8sf __builtin_ia32_addsubps256 (v8sf,v8sf); 54668 v4df __builtin_ia32_andnpd256 (v4df,v4df); 54669 v8sf __builtin_ia32_andnps256 (v8sf,v8sf); 54670 v4df __builtin_ia32_andpd256 (v4df,v4df); 54671 v8sf __builtin_ia32_andps256 (v8sf,v8sf); 54672 v4df __builtin_ia32_blendpd256 (v4df,v4df,int); 54673 v8sf __builtin_ia32_blendps256 (v8sf,v8sf,int); 54674 v4df __builtin_ia32_blendvpd256 (v4df,v4df,v4df); 54675 v8sf __builtin_ia32_blendvps256 (v8sf,v8sf,v8sf); 54676 v2df __builtin_ia32_cmppd (v2df,v2df,int); 54677 v4df __builtin_ia32_cmppd256 (v4df,v4df,int); 54678 v4sf __builtin_ia32_cmpps (v4sf,v4sf,int); 54679 v8sf __builtin_ia32_cmpps256 (v8sf,v8sf,int); 54680 v2df __builtin_ia32_cmpsd (v2df,v2df,int); 54681 v4sf __builtin_ia32_cmpss (v4sf,v4sf,int); 54682 v4df __builtin_ia32_cvtdq2pd256 (v4si); 54683 v8sf __builtin_ia32_cvtdq2ps256 (v8si); 54684 v4si __builtin_ia32_cvtpd2dq256 (v4df); 54685 v4sf __builtin_ia32_cvtpd2ps256 (v4df); 54686 v8si __builtin_ia32_cvtps2dq256 (v8sf); 54687 v4df __builtin_ia32_cvtps2pd256 (v4sf); 54688 v4si __builtin_ia32_cvttpd2dq256 (v4df); 54689 v8si __builtin_ia32_cvttps2dq256 (v8sf); 54690 v4df __builtin_ia32_divpd256 (v4df,v4df); 54691 v8sf __builtin_ia32_divps256 (v8sf,v8sf); 54692 v8sf __builtin_ia32_dpps256 (v8sf,v8sf,int); 54693 v4df __builtin_ia32_haddpd256 (v4df,v4df); 54694 v8sf __builtin_ia32_haddps256 (v8sf,v8sf); 54695 v4df __builtin_ia32_hsubpd256 (v4df,v4df); 54696 v8sf __builtin_ia32_hsubps256 (v8sf,v8sf); 54697 v32qi __builtin_ia32_lddqu256 (pcchar); 54698 v32qi __builtin_ia32_loaddqu256 (pcchar); 54699 v4df __builtin_ia32_loadupd256 (pcdouble); 54700 v8sf __builtin_ia32_loadups256 (pcfloat); 54701 v2df __builtin_ia32_maskloadpd (pcv2df,v2df); 54702 v4df __builtin_ia32_maskloadpd256 (pcv4df,v4df); 54703 v4sf __builtin_ia32_maskloadps (pcv4sf,v4sf); 54704 v8sf __builtin_ia32_maskloadps256 (pcv8sf,v8sf); 54705 void __builtin_ia32_maskstorepd (pv2df,v2df,v2df); 54706 void __builtin_ia32_maskstorepd256 (pv4df,v4df,v4df); 54707 void __builtin_ia32_maskstoreps (pv4sf,v4sf,v4sf); 54708 void __builtin_ia32_maskstoreps256 (pv8sf,v8sf,v8sf); 54709 v4df __builtin_ia32_maxpd256 (v4df,v4df); 54710 v8sf __builtin_ia32_maxps256 (v8sf,v8sf); 54711 v4df __builtin_ia32_minpd256 (v4df,v4df); 54712 v8sf __builtin_ia32_minps256 (v8sf,v8sf); 54713 v4df __builtin_ia32_movddup256 (v4df); 54714 int __builtin_ia32_movmskpd256 (v4df); 54715 int __builtin_ia32_movmskps256 (v8sf); 54716 v8sf __builtin_ia32_movshdup256 (v8sf); 54717 v8sf __builtin_ia32_movsldup256 (v8sf); 54718 v4df __builtin_ia32_mulpd256 (v4df,v4df); 54719 v8sf __builtin_ia32_mulps256 (v8sf,v8sf); 54720 v4df __builtin_ia32_orpd256 (v4df,v4df); 54721 v8sf __builtin_ia32_orps256 (v8sf,v8sf); 54722 v2df __builtin_ia32_pd_pd256 (v4df); 54723 v4df __builtin_ia32_pd256_pd (v2df); 54724 v4sf __builtin_ia32_ps_ps256 (v8sf); 54725 v8sf __builtin_ia32_ps256_ps (v4sf); 54726 int __builtin_ia32_ptestc256 (v4di,v4di,ptest); 54727 int __builtin_ia32_ptestnzc256 (v4di,v4di,ptest); 54728 int __builtin_ia32_ptestz256 (v4di,v4di,ptest); 54729 v8sf __builtin_ia32_rcpps256 (v8sf); 54730 v4df __builtin_ia32_roundpd256 (v4df,int); 54731 v8sf __builtin_ia32_roundps256 (v8sf,int); 54732 v8sf __builtin_ia32_rsqrtps_nr256 (v8sf); 54733 v8sf __builtin_ia32_rsqrtps256 (v8sf); 54734 v4df __builtin_ia32_shufpd256 (v4df,v4df,int); 54735 v8sf __builtin_ia32_shufps256 (v8sf,v8sf,int); 54736 v4si __builtin_ia32_si_si256 (v8si); 54737 v8si __builtin_ia32_si256_si (v4si); 54738 v4df __builtin_ia32_sqrtpd256 (v4df); 54739 v8sf __builtin_ia32_sqrtps_nr256 (v8sf); 54740 v8sf __builtin_ia32_sqrtps256 (v8sf); 54741 void __builtin_ia32_storedqu256 (pchar,v32qi); 54742 void __builtin_ia32_storeupd256 (pdouble,v4df); 54743 void __builtin_ia32_storeups256 (pfloat,v8sf); 54744 v4df __builtin_ia32_subpd256 (v4df,v4df); 54745 v8sf __builtin_ia32_subps256 (v8sf,v8sf); 54746 v4df __builtin_ia32_unpckhpd256 (v4df,v4df); 54747 v8sf __builtin_ia32_unpckhps256 (v8sf,v8sf); 54748 v4df __builtin_ia32_unpcklpd256 (v4df,v4df); 54749 v8sf __builtin_ia32_unpcklps256 (v8sf,v8sf); 54750 v4df __builtin_ia32_vbroadcastf128_pd256 (pcv2df); 54751 v8sf __builtin_ia32_vbroadcastf128_ps256 (pcv4sf); 54752 v4df __builtin_ia32_vbroadcastsd256 (pcdouble); 54753 v4sf __builtin_ia32_vbroadcastss (pcfloat); 54754 v8sf __builtin_ia32_vbroadcastss256 (pcfloat); 54755 v2df __builtin_ia32_vextractf128_pd256 (v4df,int); 54756 v4sf __builtin_ia32_vextractf128_ps256 (v8sf,int); 54757 v4si __builtin_ia32_vextractf128_si256 (v8si,int); 54758 v4df __builtin_ia32_vinsertf128_pd256 (v4df,v2df,int); 54759 v8sf __builtin_ia32_vinsertf128_ps256 (v8sf,v4sf,int); 54760 v8si __builtin_ia32_vinsertf128_si256 (v8si,v4si,int); 54761 v4df __builtin_ia32_vperm2f128_pd256 (v4df,v4df,int); 54762 v8sf __builtin_ia32_vperm2f128_ps256 (v8sf,v8sf,int); 54763 v8si __builtin_ia32_vperm2f128_si256 (v8si,v8si,int); 54764 v2df __builtin_ia32_vpermil2pd (v2df,v2df,v2di,int); 54765 v4df __builtin_ia32_vpermil2pd256 (v4df,v4df,v4di,int); 54766 v4sf __builtin_ia32_vpermil2ps (v4sf,v4sf,v4si,int); 54767 v8sf __builtin_ia32_vpermil2ps256 (v8sf,v8sf,v8si,int); 54768 v2df __builtin_ia32_vpermilpd (v2df,int); 54769 v4df __builtin_ia32_vpermilpd256 (v4df,int); 54770 v4sf __builtin_ia32_vpermilps (v4sf,int); 54771 v8sf __builtin_ia32_vpermilps256 (v8sf,int); 54772 v2df __builtin_ia32_vpermilvarpd (v2df,v2di); 54773 v4df __builtin_ia32_vpermilvarpd256 (v4df,v4di); 54774 v4sf __builtin_ia32_vpermilvarps (v4sf,v4si); 54775 v8sf __builtin_ia32_vpermilvarps256 (v8sf,v8si); 54776 int __builtin_ia32_vtestcpd (v2df,v2df,ptest); 54777 int __builtin_ia32_vtestcpd256 (v4df,v4df,ptest); 54778 int __builtin_ia32_vtestcps (v4sf,v4sf,ptest); 54779 int __builtin_ia32_vtestcps256 (v8sf,v8sf,ptest); 54780 int __builtin_ia32_vtestnzcpd (v2df,v2df,ptest); 54781 int __builtin_ia32_vtestnzcpd256 (v4df,v4df,ptest); 54782 int __builtin_ia32_vtestnzcps (v4sf,v4sf,ptest); 54783 int __builtin_ia32_vtestnzcps256 (v8sf,v8sf,ptest); 54784 int __builtin_ia32_vtestzpd (v2df,v2df,ptest); 54785 int __builtin_ia32_vtestzpd256 (v4df,v4df,ptest); 54786 int __builtin_ia32_vtestzps (v4sf,v4sf,ptest); 54787 int __builtin_ia32_vtestzps256 (v8sf,v8sf,ptest); 54788 void __builtin_ia32_vzeroall (void); 54789 void __builtin_ia32_vzeroupper (void); 54790 v4df __builtin_ia32_xorpd256 (v4df,v4df); 54791 v8sf __builtin_ia32_xorps256 (v8sf,v8sf); 54792 54793 The following built-in functions are available when '-mavx2' is used. 54794All of them generate the machine instruction that is part of the name. 54795 54796 v32qi __builtin_ia32_mpsadbw256 (v32qi,v32qi,int); 54797 v32qi __builtin_ia32_pabsb256 (v32qi); 54798 v16hi __builtin_ia32_pabsw256 (v16hi); 54799 v8si __builtin_ia32_pabsd256 (v8si); 54800 v16hi __builtin_ia32_packssdw256 (v8si,v8si); 54801 v32qi __builtin_ia32_packsswb256 (v16hi,v16hi); 54802 v16hi __builtin_ia32_packusdw256 (v8si,v8si); 54803 v32qi __builtin_ia32_packuswb256 (v16hi,v16hi); 54804 v32qi __builtin_ia32_paddb256 (v32qi,v32qi); 54805 v16hi __builtin_ia32_paddw256 (v16hi,v16hi); 54806 v8si __builtin_ia32_paddd256 (v8si,v8si); 54807 v4di __builtin_ia32_paddq256 (v4di,v4di); 54808 v32qi __builtin_ia32_paddsb256 (v32qi,v32qi); 54809 v16hi __builtin_ia32_paddsw256 (v16hi,v16hi); 54810 v32qi __builtin_ia32_paddusb256 (v32qi,v32qi); 54811 v16hi __builtin_ia32_paddusw256 (v16hi,v16hi); 54812 v4di __builtin_ia32_palignr256 (v4di,v4di,int); 54813 v4di __builtin_ia32_andsi256 (v4di,v4di); 54814 v4di __builtin_ia32_andnotsi256 (v4di,v4di); 54815 v32qi __builtin_ia32_pavgb256 (v32qi,v32qi); 54816 v16hi __builtin_ia32_pavgw256 (v16hi,v16hi); 54817 v32qi __builtin_ia32_pblendvb256 (v32qi,v32qi,v32qi); 54818 v16hi __builtin_ia32_pblendw256 (v16hi,v16hi,int); 54819 v32qi __builtin_ia32_pcmpeqb256 (v32qi,v32qi); 54820 v16hi __builtin_ia32_pcmpeqw256 (v16hi,v16hi); 54821 v8si __builtin_ia32_pcmpeqd256 (c8si,v8si); 54822 v4di __builtin_ia32_pcmpeqq256 (v4di,v4di); 54823 v32qi __builtin_ia32_pcmpgtb256 (v32qi,v32qi); 54824 v16hi __builtin_ia32_pcmpgtw256 (16hi,v16hi); 54825 v8si __builtin_ia32_pcmpgtd256 (v8si,v8si); 54826 v4di __builtin_ia32_pcmpgtq256 (v4di,v4di); 54827 v16hi __builtin_ia32_phaddw256 (v16hi,v16hi); 54828 v8si __builtin_ia32_phaddd256 (v8si,v8si); 54829 v16hi __builtin_ia32_phaddsw256 (v16hi,v16hi); 54830 v16hi __builtin_ia32_phsubw256 (v16hi,v16hi); 54831 v8si __builtin_ia32_phsubd256 (v8si,v8si); 54832 v16hi __builtin_ia32_phsubsw256 (v16hi,v16hi); 54833 v32qi __builtin_ia32_pmaddubsw256 (v32qi,v32qi); 54834 v16hi __builtin_ia32_pmaddwd256 (v16hi,v16hi); 54835 v32qi __builtin_ia32_pmaxsb256 (v32qi,v32qi); 54836 v16hi __builtin_ia32_pmaxsw256 (v16hi,v16hi); 54837 v8si __builtin_ia32_pmaxsd256 (v8si,v8si); 54838 v32qi __builtin_ia32_pmaxub256 (v32qi,v32qi); 54839 v16hi __builtin_ia32_pmaxuw256 (v16hi,v16hi); 54840 v8si __builtin_ia32_pmaxud256 (v8si,v8si); 54841 v32qi __builtin_ia32_pminsb256 (v32qi,v32qi); 54842 v16hi __builtin_ia32_pminsw256 (v16hi,v16hi); 54843 v8si __builtin_ia32_pminsd256 (v8si,v8si); 54844 v32qi __builtin_ia32_pminub256 (v32qi,v32qi); 54845 v16hi __builtin_ia32_pminuw256 (v16hi,v16hi); 54846 v8si __builtin_ia32_pminud256 (v8si,v8si); 54847 int __builtin_ia32_pmovmskb256 (v32qi); 54848 v16hi __builtin_ia32_pmovsxbw256 (v16qi); 54849 v8si __builtin_ia32_pmovsxbd256 (v16qi); 54850 v4di __builtin_ia32_pmovsxbq256 (v16qi); 54851 v8si __builtin_ia32_pmovsxwd256 (v8hi); 54852 v4di __builtin_ia32_pmovsxwq256 (v8hi); 54853 v4di __builtin_ia32_pmovsxdq256 (v4si); 54854 v16hi __builtin_ia32_pmovzxbw256 (v16qi); 54855 v8si __builtin_ia32_pmovzxbd256 (v16qi); 54856 v4di __builtin_ia32_pmovzxbq256 (v16qi); 54857 v8si __builtin_ia32_pmovzxwd256 (v8hi); 54858 v4di __builtin_ia32_pmovzxwq256 (v8hi); 54859 v4di __builtin_ia32_pmovzxdq256 (v4si); 54860 v4di __builtin_ia32_pmuldq256 (v8si,v8si); 54861 v16hi __builtin_ia32_pmulhrsw256 (v16hi, v16hi); 54862 v16hi __builtin_ia32_pmulhuw256 (v16hi,v16hi); 54863 v16hi __builtin_ia32_pmulhw256 (v16hi,v16hi); 54864 v16hi __builtin_ia32_pmullw256 (v16hi,v16hi); 54865 v8si __builtin_ia32_pmulld256 (v8si,v8si); 54866 v4di __builtin_ia32_pmuludq256 (v8si,v8si); 54867 v4di __builtin_ia32_por256 (v4di,v4di); 54868 v16hi __builtin_ia32_psadbw256 (v32qi,v32qi); 54869 v32qi __builtin_ia32_pshufb256 (v32qi,v32qi); 54870 v8si __builtin_ia32_pshufd256 (v8si,int); 54871 v16hi __builtin_ia32_pshufhw256 (v16hi,int); 54872 v16hi __builtin_ia32_pshuflw256 (v16hi,int); 54873 v32qi __builtin_ia32_psignb256 (v32qi,v32qi); 54874 v16hi __builtin_ia32_psignw256 (v16hi,v16hi); 54875 v8si __builtin_ia32_psignd256 (v8si,v8si); 54876 v4di __builtin_ia32_pslldqi256 (v4di,int); 54877 v16hi __builtin_ia32_psllwi256 (16hi,int); 54878 v16hi __builtin_ia32_psllw256(v16hi,v8hi); 54879 v8si __builtin_ia32_pslldi256 (v8si,int); 54880 v8si __builtin_ia32_pslld256(v8si,v4si); 54881 v4di __builtin_ia32_psllqi256 (v4di,int); 54882 v4di __builtin_ia32_psllq256(v4di,v2di); 54883 v16hi __builtin_ia32_psrawi256 (v16hi,int); 54884 v16hi __builtin_ia32_psraw256 (v16hi,v8hi); 54885 v8si __builtin_ia32_psradi256 (v8si,int); 54886 v8si __builtin_ia32_psrad256 (v8si,v4si); 54887 v4di __builtin_ia32_psrldqi256 (v4di, int); 54888 v16hi __builtin_ia32_psrlwi256 (v16hi,int); 54889 v16hi __builtin_ia32_psrlw256 (v16hi,v8hi); 54890 v8si __builtin_ia32_psrldi256 (v8si,int); 54891 v8si __builtin_ia32_psrld256 (v8si,v4si); 54892 v4di __builtin_ia32_psrlqi256 (v4di,int); 54893 v4di __builtin_ia32_psrlq256(v4di,v2di); 54894 v32qi __builtin_ia32_psubb256 (v32qi,v32qi); 54895 v32hi __builtin_ia32_psubw256 (v16hi,v16hi); 54896 v8si __builtin_ia32_psubd256 (v8si,v8si); 54897 v4di __builtin_ia32_psubq256 (v4di,v4di); 54898 v32qi __builtin_ia32_psubsb256 (v32qi,v32qi); 54899 v16hi __builtin_ia32_psubsw256 (v16hi,v16hi); 54900 v32qi __builtin_ia32_psubusb256 (v32qi,v32qi); 54901 v16hi __builtin_ia32_psubusw256 (v16hi,v16hi); 54902 v32qi __builtin_ia32_punpckhbw256 (v32qi,v32qi); 54903 v16hi __builtin_ia32_punpckhwd256 (v16hi,v16hi); 54904 v8si __builtin_ia32_punpckhdq256 (v8si,v8si); 54905 v4di __builtin_ia32_punpckhqdq256 (v4di,v4di); 54906 v32qi __builtin_ia32_punpcklbw256 (v32qi,v32qi); 54907 v16hi __builtin_ia32_punpcklwd256 (v16hi,v16hi); 54908 v8si __builtin_ia32_punpckldq256 (v8si,v8si); 54909 v4di __builtin_ia32_punpcklqdq256 (v4di,v4di); 54910 v4di __builtin_ia32_pxor256 (v4di,v4di); 54911 v4di __builtin_ia32_movntdqa256 (pv4di); 54912 v4sf __builtin_ia32_vbroadcastss_ps (v4sf); 54913 v8sf __builtin_ia32_vbroadcastss_ps256 (v4sf); 54914 v4df __builtin_ia32_vbroadcastsd_pd256 (v2df); 54915 v4di __builtin_ia32_vbroadcastsi256 (v2di); 54916 v4si __builtin_ia32_pblendd128 (v4si,v4si); 54917 v8si __builtin_ia32_pblendd256 (v8si,v8si); 54918 v32qi __builtin_ia32_pbroadcastb256 (v16qi); 54919 v16hi __builtin_ia32_pbroadcastw256 (v8hi); 54920 v8si __builtin_ia32_pbroadcastd256 (v4si); 54921 v4di __builtin_ia32_pbroadcastq256 (v2di); 54922 v16qi __builtin_ia32_pbroadcastb128 (v16qi); 54923 v8hi __builtin_ia32_pbroadcastw128 (v8hi); 54924 v4si __builtin_ia32_pbroadcastd128 (v4si); 54925 v2di __builtin_ia32_pbroadcastq128 (v2di); 54926 v8si __builtin_ia32_permvarsi256 (v8si,v8si); 54927 v4df __builtin_ia32_permdf256 (v4df,int); 54928 v8sf __builtin_ia32_permvarsf256 (v8sf,v8sf); 54929 v4di __builtin_ia32_permdi256 (v4di,int); 54930 v4di __builtin_ia32_permti256 (v4di,v4di,int); 54931 v4di __builtin_ia32_extract128i256 (v4di,int); 54932 v4di __builtin_ia32_insert128i256 (v4di,v2di,int); 54933 v8si __builtin_ia32_maskloadd256 (pcv8si,v8si); 54934 v4di __builtin_ia32_maskloadq256 (pcv4di,v4di); 54935 v4si __builtin_ia32_maskloadd (pcv4si,v4si); 54936 v2di __builtin_ia32_maskloadq (pcv2di,v2di); 54937 void __builtin_ia32_maskstored256 (pv8si,v8si,v8si); 54938 void __builtin_ia32_maskstoreq256 (pv4di,v4di,v4di); 54939 void __builtin_ia32_maskstored (pv4si,v4si,v4si); 54940 void __builtin_ia32_maskstoreq (pv2di,v2di,v2di); 54941 v8si __builtin_ia32_psllv8si (v8si,v8si); 54942 v4si __builtin_ia32_psllv4si (v4si,v4si); 54943 v4di __builtin_ia32_psllv4di (v4di,v4di); 54944 v2di __builtin_ia32_psllv2di (v2di,v2di); 54945 v8si __builtin_ia32_psrav8si (v8si,v8si); 54946 v4si __builtin_ia32_psrav4si (v4si,v4si); 54947 v8si __builtin_ia32_psrlv8si (v8si,v8si); 54948 v4si __builtin_ia32_psrlv4si (v4si,v4si); 54949 v4di __builtin_ia32_psrlv4di (v4di,v4di); 54950 v2di __builtin_ia32_psrlv2di (v2di,v2di); 54951 v2df __builtin_ia32_gathersiv2df (v2df, pcdouble,v4si,v2df,int); 54952 v4df __builtin_ia32_gathersiv4df (v4df, pcdouble,v4si,v4df,int); 54953 v2df __builtin_ia32_gatherdiv2df (v2df, pcdouble,v2di,v2df,int); 54954 v4df __builtin_ia32_gatherdiv4df (v4df, pcdouble,v4di,v4df,int); 54955 v4sf __builtin_ia32_gathersiv4sf (v4sf, pcfloat,v4si,v4sf,int); 54956 v8sf __builtin_ia32_gathersiv8sf (v8sf, pcfloat,v8si,v8sf,int); 54957 v4sf __builtin_ia32_gatherdiv4sf (v4sf, pcfloat,v2di,v4sf,int); 54958 v4sf __builtin_ia32_gatherdiv4sf256 (v4sf, pcfloat,v4di,v4sf,int); 54959 v2di __builtin_ia32_gathersiv2di (v2di, pcint64,v4si,v2di,int); 54960 v4di __builtin_ia32_gathersiv4di (v4di, pcint64,v4si,v4di,int); 54961 v2di __builtin_ia32_gatherdiv2di (v2di, pcint64,v2di,v2di,int); 54962 v4di __builtin_ia32_gatherdiv4di (v4di, pcint64,v4di,v4di,int); 54963 v4si __builtin_ia32_gathersiv4si (v4si, pcint,v4si,v4si,int); 54964 v8si __builtin_ia32_gathersiv8si (v8si, pcint,v8si,v8si,int); 54965 v4si __builtin_ia32_gatherdiv4si (v4si, pcint,v2di,v4si,int); 54966 v4si __builtin_ia32_gatherdiv4si256 (v4si, pcint,v4di,v4si,int); 54967 54968 The following built-in functions are available when '-maes' is used. 54969All of them generate the machine instruction that is part of the name. 54970 54971 v2di __builtin_ia32_aesenc128 (v2di, v2di); 54972 v2di __builtin_ia32_aesenclast128 (v2di, v2di); 54973 v2di __builtin_ia32_aesdec128 (v2di, v2di); 54974 v2di __builtin_ia32_aesdeclast128 (v2di, v2di); 54975 v2di __builtin_ia32_aeskeygenassist128 (v2di, const int); 54976 v2di __builtin_ia32_aesimc128 (v2di); 54977 54978 The following built-in function is available when '-mpclmul' is used. 54979 54980'v2di __builtin_ia32_pclmulqdq128 (v2di, v2di, const int)' 54981 Generates the 'pclmulqdq' machine instruction. 54982 54983 The following built-in function is available when '-mfsgsbase' is used. 54984All of them generate the machine instruction that is part of the name. 54985 54986 unsigned int __builtin_ia32_rdfsbase32 (void); 54987 unsigned long long __builtin_ia32_rdfsbase64 (void); 54988 unsigned int __builtin_ia32_rdgsbase32 (void); 54989 unsigned long long __builtin_ia32_rdgsbase64 (void); 54990 void _writefsbase_u32 (unsigned int); 54991 void _writefsbase_u64 (unsigned long long); 54992 void _writegsbase_u32 (unsigned int); 54993 void _writegsbase_u64 (unsigned long long); 54994 54995 The following built-in function is available when '-mrdrnd' is used. 54996All of them generate the machine instruction that is part of the name. 54997 54998 unsigned int __builtin_ia32_rdrand16_step (unsigned short *); 54999 unsigned int __builtin_ia32_rdrand32_step (unsigned int *); 55000 unsigned int __builtin_ia32_rdrand64_step (unsigned long long *); 55001 55002 The following built-in function is available when '-mptwrite' is used. 55003All of them generate the machine instruction that is part of the name. 55004 55005 void __builtin_ia32_ptwrite32 (unsigned); 55006 void __builtin_ia32_ptwrite64 (unsigned long long); 55007 55008 The following built-in functions are available when '-msse4a' is used. 55009All of them generate the machine instruction that is part of the name. 55010 55011 void __builtin_ia32_movntsd (double *, v2df); 55012 void __builtin_ia32_movntss (float *, v4sf); 55013 v2di __builtin_ia32_extrq (v2di, v16qi); 55014 v2di __builtin_ia32_extrqi (v2di, const unsigned int, const unsigned int); 55015 v2di __builtin_ia32_insertq (v2di, v2di); 55016 v2di __builtin_ia32_insertqi (v2di, v2di, const unsigned int, const unsigned int); 55017 55018 The following built-in functions are available when '-mxop' is used. 55019 v2df __builtin_ia32_vfrczpd (v2df); 55020 v4sf __builtin_ia32_vfrczps (v4sf); 55021 v2df __builtin_ia32_vfrczsd (v2df); 55022 v4sf __builtin_ia32_vfrczss (v4sf); 55023 v4df __builtin_ia32_vfrczpd256 (v4df); 55024 v8sf __builtin_ia32_vfrczps256 (v8sf); 55025 v2di __builtin_ia32_vpcmov (v2di, v2di, v2di); 55026 v2di __builtin_ia32_vpcmov_v2di (v2di, v2di, v2di); 55027 v4si __builtin_ia32_vpcmov_v4si (v4si, v4si, v4si); 55028 v8hi __builtin_ia32_vpcmov_v8hi (v8hi, v8hi, v8hi); 55029 v16qi __builtin_ia32_vpcmov_v16qi (v16qi, v16qi, v16qi); 55030 v2df __builtin_ia32_vpcmov_v2df (v2df, v2df, v2df); 55031 v4sf __builtin_ia32_vpcmov_v4sf (v4sf, v4sf, v4sf); 55032 v4di __builtin_ia32_vpcmov_v4di256 (v4di, v4di, v4di); 55033 v8si __builtin_ia32_vpcmov_v8si256 (v8si, v8si, v8si); 55034 v16hi __builtin_ia32_vpcmov_v16hi256 (v16hi, v16hi, v16hi); 55035 v32qi __builtin_ia32_vpcmov_v32qi256 (v32qi, v32qi, v32qi); 55036 v4df __builtin_ia32_vpcmov_v4df256 (v4df, v4df, v4df); 55037 v8sf __builtin_ia32_vpcmov_v8sf256 (v8sf, v8sf, v8sf); 55038 v16qi __builtin_ia32_vpcomeqb (v16qi, v16qi); 55039 v8hi __builtin_ia32_vpcomeqw (v8hi, v8hi); 55040 v4si __builtin_ia32_vpcomeqd (v4si, v4si); 55041 v2di __builtin_ia32_vpcomeqq (v2di, v2di); 55042 v16qi __builtin_ia32_vpcomequb (v16qi, v16qi); 55043 v4si __builtin_ia32_vpcomequd (v4si, v4si); 55044 v2di __builtin_ia32_vpcomequq (v2di, v2di); 55045 v8hi __builtin_ia32_vpcomequw (v8hi, v8hi); 55046 v8hi __builtin_ia32_vpcomeqw (v8hi, v8hi); 55047 v16qi __builtin_ia32_vpcomfalseb (v16qi, v16qi); 55048 v4si __builtin_ia32_vpcomfalsed (v4si, v4si); 55049 v2di __builtin_ia32_vpcomfalseq (v2di, v2di); 55050 v16qi __builtin_ia32_vpcomfalseub (v16qi, v16qi); 55051 v4si __builtin_ia32_vpcomfalseud (v4si, v4si); 55052 v2di __builtin_ia32_vpcomfalseuq (v2di, v2di); 55053 v8hi __builtin_ia32_vpcomfalseuw (v8hi, v8hi); 55054 v8hi __builtin_ia32_vpcomfalsew (v8hi, v8hi); 55055 v16qi __builtin_ia32_vpcomgeb (v16qi, v16qi); 55056 v4si __builtin_ia32_vpcomged (v4si, v4si); 55057 v2di __builtin_ia32_vpcomgeq (v2di, v2di); 55058 v16qi __builtin_ia32_vpcomgeub (v16qi, v16qi); 55059 v4si __builtin_ia32_vpcomgeud (v4si, v4si); 55060 v2di __builtin_ia32_vpcomgeuq (v2di, v2di); 55061 v8hi __builtin_ia32_vpcomgeuw (v8hi, v8hi); 55062 v8hi __builtin_ia32_vpcomgew (v8hi, v8hi); 55063 v16qi __builtin_ia32_vpcomgtb (v16qi, v16qi); 55064 v4si __builtin_ia32_vpcomgtd (v4si, v4si); 55065 v2di __builtin_ia32_vpcomgtq (v2di, v2di); 55066 v16qi __builtin_ia32_vpcomgtub (v16qi, v16qi); 55067 v4si __builtin_ia32_vpcomgtud (v4si, v4si); 55068 v2di __builtin_ia32_vpcomgtuq (v2di, v2di); 55069 v8hi __builtin_ia32_vpcomgtuw (v8hi, v8hi); 55070 v8hi __builtin_ia32_vpcomgtw (v8hi, v8hi); 55071 v16qi __builtin_ia32_vpcomleb (v16qi, v16qi); 55072 v4si __builtin_ia32_vpcomled (v4si, v4si); 55073 v2di __builtin_ia32_vpcomleq (v2di, v2di); 55074 v16qi __builtin_ia32_vpcomleub (v16qi, v16qi); 55075 v4si __builtin_ia32_vpcomleud (v4si, v4si); 55076 v2di __builtin_ia32_vpcomleuq (v2di, v2di); 55077 v8hi __builtin_ia32_vpcomleuw (v8hi, v8hi); 55078 v8hi __builtin_ia32_vpcomlew (v8hi, v8hi); 55079 v16qi __builtin_ia32_vpcomltb (v16qi, v16qi); 55080 v4si __builtin_ia32_vpcomltd (v4si, v4si); 55081 v2di __builtin_ia32_vpcomltq (v2di, v2di); 55082 v16qi __builtin_ia32_vpcomltub (v16qi, v16qi); 55083 v4si __builtin_ia32_vpcomltud (v4si, v4si); 55084 v2di __builtin_ia32_vpcomltuq (v2di, v2di); 55085 v8hi __builtin_ia32_vpcomltuw (v8hi, v8hi); 55086 v8hi __builtin_ia32_vpcomltw (v8hi, v8hi); 55087 v16qi __builtin_ia32_vpcomneb (v16qi, v16qi); 55088 v4si __builtin_ia32_vpcomned (v4si, v4si); 55089 v2di __builtin_ia32_vpcomneq (v2di, v2di); 55090 v16qi __builtin_ia32_vpcomneub (v16qi, v16qi); 55091 v4si __builtin_ia32_vpcomneud (v4si, v4si); 55092 v2di __builtin_ia32_vpcomneuq (v2di, v2di); 55093 v8hi __builtin_ia32_vpcomneuw (v8hi, v8hi); 55094 v8hi __builtin_ia32_vpcomnew (v8hi, v8hi); 55095 v16qi __builtin_ia32_vpcomtrueb (v16qi, v16qi); 55096 v4si __builtin_ia32_vpcomtrued (v4si, v4si); 55097 v2di __builtin_ia32_vpcomtrueq (v2di, v2di); 55098 v16qi __builtin_ia32_vpcomtrueub (v16qi, v16qi); 55099 v4si __builtin_ia32_vpcomtrueud (v4si, v4si); 55100 v2di __builtin_ia32_vpcomtrueuq (v2di, v2di); 55101 v8hi __builtin_ia32_vpcomtrueuw (v8hi, v8hi); 55102 v8hi __builtin_ia32_vpcomtruew (v8hi, v8hi); 55103 v4si __builtin_ia32_vphaddbd (v16qi); 55104 v2di __builtin_ia32_vphaddbq (v16qi); 55105 v8hi __builtin_ia32_vphaddbw (v16qi); 55106 v2di __builtin_ia32_vphadddq (v4si); 55107 v4si __builtin_ia32_vphaddubd (v16qi); 55108 v2di __builtin_ia32_vphaddubq (v16qi); 55109 v8hi __builtin_ia32_vphaddubw (v16qi); 55110 v2di __builtin_ia32_vphaddudq (v4si); 55111 v4si __builtin_ia32_vphadduwd (v8hi); 55112 v2di __builtin_ia32_vphadduwq (v8hi); 55113 v4si __builtin_ia32_vphaddwd (v8hi); 55114 v2di __builtin_ia32_vphaddwq (v8hi); 55115 v8hi __builtin_ia32_vphsubbw (v16qi); 55116 v2di __builtin_ia32_vphsubdq (v4si); 55117 v4si __builtin_ia32_vphsubwd (v8hi); 55118 v4si __builtin_ia32_vpmacsdd (v4si, v4si, v4si); 55119 v2di __builtin_ia32_vpmacsdqh (v4si, v4si, v2di); 55120 v2di __builtin_ia32_vpmacsdql (v4si, v4si, v2di); 55121 v4si __builtin_ia32_vpmacssdd (v4si, v4si, v4si); 55122 v2di __builtin_ia32_vpmacssdqh (v4si, v4si, v2di); 55123 v2di __builtin_ia32_vpmacssdql (v4si, v4si, v2di); 55124 v4si __builtin_ia32_vpmacsswd (v8hi, v8hi, v4si); 55125 v8hi __builtin_ia32_vpmacssww (v8hi, v8hi, v8hi); 55126 v4si __builtin_ia32_vpmacswd (v8hi, v8hi, v4si); 55127 v8hi __builtin_ia32_vpmacsww (v8hi, v8hi, v8hi); 55128 v4si __builtin_ia32_vpmadcsswd (v8hi, v8hi, v4si); 55129 v4si __builtin_ia32_vpmadcswd (v8hi, v8hi, v4si); 55130 v16qi __builtin_ia32_vpperm (v16qi, v16qi, v16qi); 55131 v16qi __builtin_ia32_vprotb (v16qi, v16qi); 55132 v4si __builtin_ia32_vprotd (v4si, v4si); 55133 v2di __builtin_ia32_vprotq (v2di, v2di); 55134 v8hi __builtin_ia32_vprotw (v8hi, v8hi); 55135 v16qi __builtin_ia32_vpshab (v16qi, v16qi); 55136 v4si __builtin_ia32_vpshad (v4si, v4si); 55137 v2di __builtin_ia32_vpshaq (v2di, v2di); 55138 v8hi __builtin_ia32_vpshaw (v8hi, v8hi); 55139 v16qi __builtin_ia32_vpshlb (v16qi, v16qi); 55140 v4si __builtin_ia32_vpshld (v4si, v4si); 55141 v2di __builtin_ia32_vpshlq (v2di, v2di); 55142 v8hi __builtin_ia32_vpshlw (v8hi, v8hi); 55143 55144 The following built-in functions are available when '-mfma4' is used. 55145All of them generate the machine instruction that is part of the name. 55146 55147 v2df __builtin_ia32_vfmaddpd (v2df, v2df, v2df); 55148 v4sf __builtin_ia32_vfmaddps (v4sf, v4sf, v4sf); 55149 v2df __builtin_ia32_vfmaddsd (v2df, v2df, v2df); 55150 v4sf __builtin_ia32_vfmaddss (v4sf, v4sf, v4sf); 55151 v2df __builtin_ia32_vfmsubpd (v2df, v2df, v2df); 55152 v4sf __builtin_ia32_vfmsubps (v4sf, v4sf, v4sf); 55153 v2df __builtin_ia32_vfmsubsd (v2df, v2df, v2df); 55154 v4sf __builtin_ia32_vfmsubss (v4sf, v4sf, v4sf); 55155 v2df __builtin_ia32_vfnmaddpd (v2df, v2df, v2df); 55156 v4sf __builtin_ia32_vfnmaddps (v4sf, v4sf, v4sf); 55157 v2df __builtin_ia32_vfnmaddsd (v2df, v2df, v2df); 55158 v4sf __builtin_ia32_vfnmaddss (v4sf, v4sf, v4sf); 55159 v2df __builtin_ia32_vfnmsubpd (v2df, v2df, v2df); 55160 v4sf __builtin_ia32_vfnmsubps (v4sf, v4sf, v4sf); 55161 v2df __builtin_ia32_vfnmsubsd (v2df, v2df, v2df); 55162 v4sf __builtin_ia32_vfnmsubss (v4sf, v4sf, v4sf); 55163 v2df __builtin_ia32_vfmaddsubpd (v2df, v2df, v2df); 55164 v4sf __builtin_ia32_vfmaddsubps (v4sf, v4sf, v4sf); 55165 v2df __builtin_ia32_vfmsubaddpd (v2df, v2df, v2df); 55166 v4sf __builtin_ia32_vfmsubaddps (v4sf, v4sf, v4sf); 55167 v4df __builtin_ia32_vfmaddpd256 (v4df, v4df, v4df); 55168 v8sf __builtin_ia32_vfmaddps256 (v8sf, v8sf, v8sf); 55169 v4df __builtin_ia32_vfmsubpd256 (v4df, v4df, v4df); 55170 v8sf __builtin_ia32_vfmsubps256 (v8sf, v8sf, v8sf); 55171 v4df __builtin_ia32_vfnmaddpd256 (v4df, v4df, v4df); 55172 v8sf __builtin_ia32_vfnmaddps256 (v8sf, v8sf, v8sf); 55173 v4df __builtin_ia32_vfnmsubpd256 (v4df, v4df, v4df); 55174 v8sf __builtin_ia32_vfnmsubps256 (v8sf, v8sf, v8sf); 55175 v4df __builtin_ia32_vfmaddsubpd256 (v4df, v4df, v4df); 55176 v8sf __builtin_ia32_vfmaddsubps256 (v8sf, v8sf, v8sf); 55177 v4df __builtin_ia32_vfmsubaddpd256 (v4df, v4df, v4df); 55178 v8sf __builtin_ia32_vfmsubaddps256 (v8sf, v8sf, v8sf); 55179 55180 55181 The following built-in functions are available when '-mlwp' is used. 55182 55183 void __builtin_ia32_llwpcb16 (void *); 55184 void __builtin_ia32_llwpcb32 (void *); 55185 void __builtin_ia32_llwpcb64 (void *); 55186 void * __builtin_ia32_llwpcb16 (void); 55187 void * __builtin_ia32_llwpcb32 (void); 55188 void * __builtin_ia32_llwpcb64 (void); 55189 void __builtin_ia32_lwpval16 (unsigned short, unsigned int, unsigned short); 55190 void __builtin_ia32_lwpval32 (unsigned int, unsigned int, unsigned int); 55191 void __builtin_ia32_lwpval64 (unsigned __int64, unsigned int, unsigned int); 55192 unsigned char __builtin_ia32_lwpins16 (unsigned short, unsigned int, unsigned short); 55193 unsigned char __builtin_ia32_lwpins32 (unsigned int, unsigned int, unsigned int); 55194 unsigned char __builtin_ia32_lwpins64 (unsigned __int64, unsigned int, unsigned int); 55195 55196 The following built-in functions are available when '-mbmi' is used. 55197All of them generate the machine instruction that is part of the name. 55198 unsigned int __builtin_ia32_bextr_u32(unsigned int, unsigned int); 55199 unsigned long long __builtin_ia32_bextr_u64 (unsigned long long, unsigned long long); 55200 55201 The following built-in functions are available when '-mbmi2' is used. 55202All of them generate the machine instruction that is part of the name. 55203 unsigned int _bzhi_u32 (unsigned int, unsigned int); 55204 unsigned int _pdep_u32 (unsigned int, unsigned int); 55205 unsigned int _pext_u32 (unsigned int, unsigned int); 55206 unsigned long long _bzhi_u64 (unsigned long long, unsigned long long); 55207 unsigned long long _pdep_u64 (unsigned long long, unsigned long long); 55208 unsigned long long _pext_u64 (unsigned long long, unsigned long long); 55209 55210 The following built-in functions are available when '-mlzcnt' is used. 55211All of them generate the machine instruction that is part of the name. 55212 unsigned short __builtin_ia32_lzcnt_u16(unsigned short); 55213 unsigned int __builtin_ia32_lzcnt_u32(unsigned int); 55214 unsigned long long __builtin_ia32_lzcnt_u64 (unsigned long long); 55215 55216 The following built-in functions are available when '-mfxsr' is used. 55217All of them generate the machine instruction that is part of the name. 55218 void __builtin_ia32_fxsave (void *); 55219 void __builtin_ia32_fxrstor (void *); 55220 void __builtin_ia32_fxsave64 (void *); 55221 void __builtin_ia32_fxrstor64 (void *); 55222 55223 The following built-in functions are available when '-mxsave' is used. 55224All of them generate the machine instruction that is part of the name. 55225 void __builtin_ia32_xsave (void *, long long); 55226 void __builtin_ia32_xrstor (void *, long long); 55227 void __builtin_ia32_xsave64 (void *, long long); 55228 void __builtin_ia32_xrstor64 (void *, long long); 55229 55230 The following built-in functions are available when '-mxsaveopt' is 55231used. All of them generate the machine instruction that is part of the 55232name. 55233 void __builtin_ia32_xsaveopt (void *, long long); 55234 void __builtin_ia32_xsaveopt64 (void *, long long); 55235 55236 The following built-in functions are available when '-mtbm' is used. 55237Both of them generate the immediate form of the bextr machine 55238instruction. 55239 unsigned int __builtin_ia32_bextri_u32 (unsigned int, 55240 const unsigned int); 55241 unsigned long long __builtin_ia32_bextri_u64 (unsigned long long, 55242 const unsigned long long); 55243 55244 The following built-in functions are available when '-m3dnow' is used. 55245All of them generate the machine instruction that is part of the name. 55246 55247 void __builtin_ia32_femms (void); 55248 v8qi __builtin_ia32_pavgusb (v8qi, v8qi); 55249 v2si __builtin_ia32_pf2id (v2sf); 55250 v2sf __builtin_ia32_pfacc (v2sf, v2sf); 55251 v2sf __builtin_ia32_pfadd (v2sf, v2sf); 55252 v2si __builtin_ia32_pfcmpeq (v2sf, v2sf); 55253 v2si __builtin_ia32_pfcmpge (v2sf, v2sf); 55254 v2si __builtin_ia32_pfcmpgt (v2sf, v2sf); 55255 v2sf __builtin_ia32_pfmax (v2sf, v2sf); 55256 v2sf __builtin_ia32_pfmin (v2sf, v2sf); 55257 v2sf __builtin_ia32_pfmul (v2sf, v2sf); 55258 v2sf __builtin_ia32_pfrcp (v2sf); 55259 v2sf __builtin_ia32_pfrcpit1 (v2sf, v2sf); 55260 v2sf __builtin_ia32_pfrcpit2 (v2sf, v2sf); 55261 v2sf __builtin_ia32_pfrsqrt (v2sf); 55262 v2sf __builtin_ia32_pfsub (v2sf, v2sf); 55263 v2sf __builtin_ia32_pfsubr (v2sf, v2sf); 55264 v2sf __builtin_ia32_pi2fd (v2si); 55265 v4hi __builtin_ia32_pmulhrw (v4hi, v4hi); 55266 55267 The following built-in functions are available when '-m3dnowa' is used. 55268All of them generate the machine instruction that is part of the name. 55269 55270 v2si __builtin_ia32_pf2iw (v2sf); 55271 v2sf __builtin_ia32_pfnacc (v2sf, v2sf); 55272 v2sf __builtin_ia32_pfpnacc (v2sf, v2sf); 55273 v2sf __builtin_ia32_pi2fw (v2si); 55274 v2sf __builtin_ia32_pswapdsf (v2sf); 55275 v2si __builtin_ia32_pswapdsi (v2si); 55276 55277 The following built-in functions are available when '-mrtm' is used 55278They are used for restricted transactional memory. These are the 55279internal low level functions. Normally the functions in *note x86 55280transactional memory intrinsics:: should be used instead. 55281 55282 int __builtin_ia32_xbegin (); 55283 void __builtin_ia32_xend (); 55284 void __builtin_ia32_xabort (status); 55285 int __builtin_ia32_xtest (); 55286 55287 The following built-in functions are available when '-mmwaitx' is used. 55288All of them generate the machine instruction that is part of the name. 55289 void __builtin_ia32_monitorx (void *, unsigned int, unsigned int); 55290 void __builtin_ia32_mwaitx (unsigned int, unsigned int, unsigned int); 55291 55292 The following built-in functions are available when '-mclzero' is used. 55293All of them generate the machine instruction that is part of the name. 55294 void __builtin_i32_clzero (void *); 55295 55296 The following built-in functions are available when '-mpku' is used. 55297They generate reads and writes to PKRU. 55298 void __builtin_ia32_wrpkru (unsigned int); 55299 unsigned int __builtin_ia32_rdpkru (); 55300 55301 The following built-in functions are available when '-mshstk' option is 55302used. They support shadow stack machine instructions from Intel 55303Control-flow Enforcement Technology (CET). Each built-in function 55304generates the machine instruction that is part of the function's name. 55305These are the internal low-level functions. Normally the functions in 55306*note x86 control-flow protection intrinsics:: should be used instead. 55307 55308 unsigned int __builtin_ia32_rdsspd (void); 55309 unsigned long long __builtin_ia32_rdsspq (void); 55310 void __builtin_ia32_incsspd (unsigned int); 55311 void __builtin_ia32_incsspq (unsigned long long); 55312 void __builtin_ia32_saveprevssp(void); 55313 void __builtin_ia32_rstorssp(void *); 55314 void __builtin_ia32_wrssd(unsigned int, void *); 55315 void __builtin_ia32_wrssq(unsigned long long, void *); 55316 void __builtin_ia32_wrussd(unsigned int, void *); 55317 void __builtin_ia32_wrussq(unsigned long long, void *); 55318 void __builtin_ia32_setssbsy(void); 55319 void __builtin_ia32_clrssbsy(void *); 55320 55321 55322File: gcc.info, Node: x86 transactional memory intrinsics, Next: x86 control-flow protection intrinsics, Prev: x86 Built-in Functions, Up: Target Builtins 55323 553246.60.37 x86 Transactional Memory Intrinsics 55325------------------------------------------- 55326 55327These hardware transactional memory intrinsics for x86 allow you to use 55328memory transactions with RTM (Restricted Transactional Memory). This 55329support is enabled with the '-mrtm' option. For using HLE (Hardware 55330Lock Elision) see *note x86 specific memory model extensions for 55331transactional memory:: instead. 55332 55333 A memory transaction commits all changes to memory in an atomic way, as 55334visible to other threads. If the transaction fails it is rolled back 55335and all side effects discarded. 55336 55337 Generally there is no guarantee that a memory transaction ever succeeds 55338and suitable fallback code always needs to be supplied. 55339 55340 -- RTM Function: unsigned _xbegin () 55341 Start a RTM (Restricted Transactional Memory) transaction. Returns 55342 '_XBEGIN_STARTED' when the transaction started successfully (note 55343 this is not 0, so the constant has to be explicitly tested). 55344 55345 If the transaction aborts, all side effects are undone and an abort 55346 code encoded as a bit mask is returned. The following macros are 55347 defined: 55348 55349 '_XABORT_EXPLICIT' 55350 Transaction was explicitly aborted with '_xabort'. The 55351 parameter passed to '_xabort' is available with 55352 '_XABORT_CODE(status)'. 55353 '_XABORT_RETRY' 55354 Transaction retry is possible. 55355 '_XABORT_CONFLICT' 55356 Transaction abort due to a memory conflict with another 55357 thread. 55358 '_XABORT_CAPACITY' 55359 Transaction abort due to the transaction using too much 55360 memory. 55361 '_XABORT_DEBUG' 55362 Transaction abort due to a debug trap. 55363 '_XABORT_NESTED' 55364 Transaction abort in an inner nested transaction. 55365 55366 There is no guarantee any transaction ever succeeds, so there 55367 always needs to be a valid fallback path. 55368 55369 -- RTM Function: void _xend () 55370 Commit the current transaction. When no transaction is active this 55371 faults. All memory side effects of the transaction become visible 55372 to other threads in an atomic manner. 55373 55374 -- RTM Function: int _xtest () 55375 Return a nonzero value if a transaction is currently active, 55376 otherwise 0. 55377 55378 -- RTM Function: void _xabort (status) 55379 Abort the current transaction. When no transaction is active this 55380 is a no-op. The STATUS is an 8-bit constant; its value is encoded 55381 in the return value from '_xbegin'. 55382 55383 Here is an example showing handling for '_XABORT_RETRY' and a fallback 55384path for other failures: 55385 55386 #include <immintrin.h> 55387 55388 int n_tries, max_tries; 55389 unsigned status = _XABORT_EXPLICIT; 55390 ... 55391 55392 for (n_tries = 0; n_tries < max_tries; n_tries++) 55393 { 55394 status = _xbegin (); 55395 if (status == _XBEGIN_STARTED || !(status & _XABORT_RETRY)) 55396 break; 55397 } 55398 if (status == _XBEGIN_STARTED) 55399 { 55400 ... transaction code... 55401 _xend (); 55402 } 55403 else 55404 { 55405 ... non-transactional fallback path... 55406 } 55407 55408Note that, in most cases, the transactional and non-transactional code 55409must synchronize together to ensure consistency. 55410 55411 55412File: gcc.info, Node: x86 control-flow protection intrinsics, Prev: x86 transactional memory intrinsics, Up: Target Builtins 55413 554146.60.38 x86 Control-Flow Protection Intrinsics 55415---------------------------------------------- 55416 55417 -- CET Function: ret_type _get_ssp (void) 55418 Get the current value of shadow stack pointer if shadow stack 55419 support from Intel CET is enabled in the hardware or '0' otherwise. 55420 The 'ret_type' is 'unsigned long long' for 64-bit targets and 55421 'unsigned int' for 32-bit targets. 55422 55423 -- CET Function: void _inc_ssp (unsigned int) 55424 Increment the current shadow stack pointer by the size specified by 55425 the function argument. The argument is masked to a byte value for 55426 security reasons, so to increment by more than 255 bytes you must 55427 call the function multiple times. 55428 55429 The shadow stack unwind code looks like: 55430 55431 #include <immintrin.h> 55432 55433 /* Unwind the shadow stack for EH. */ 55434 #define _Unwind_Frames_Extra(x) \ 55435 do \ 55436 { \ 55437 _Unwind_Word ssp = _get_ssp (); \ 55438 if (ssp != 0) \ 55439 { \ 55440 _Unwind_Word tmp = (x); \ 55441 while (tmp > 255) \ 55442 { \ 55443 _inc_ssp (tmp); \ 55444 tmp -= 255; \ 55445 } \ 55446 _inc_ssp (tmp); \ 55447 } \ 55448 } \ 55449 while (0) 55450 55451This code runs unconditionally on all 64-bit processors. For 32-bit 55452processors the code runs on those that support multi-byte NOP 55453instructions. 55454 55455 55456File: gcc.info, Node: Target Format Checks, Next: Pragmas, Prev: Target Builtins, Up: C Extensions 55457 554586.61 Format Checks Specific to Particular Target Machines 55459========================================================= 55460 55461For some target machines, GCC supports additional options to the format 55462attribute (*note Declaring Attributes of Functions: Function 55463Attributes.). 55464 55465* Menu: 55466 55467* Solaris Format Checks:: 55468* Darwin Format Checks:: 55469 55470 55471File: gcc.info, Node: Solaris Format Checks, Next: Darwin Format Checks, Up: Target Format Checks 55472 554736.61.1 Solaris Format Checks 55474---------------------------- 55475 55476Solaris targets support the 'cmn_err' (or '__cmn_err__') format check. 55477'cmn_err' accepts a subset of the standard 'printf' conversions, and the 55478two-argument '%b' conversion for displaying bit-fields. See the Solaris 55479man page for 'cmn_err' for more information. 55480 55481 55482File: gcc.info, Node: Darwin Format Checks, Prev: Solaris Format Checks, Up: Target Format Checks 55483 554846.61.2 Darwin Format Checks 55485--------------------------- 55486 55487In addition to the full set of format archetypes (attribute format style 55488arguments such as 'printf', 'scanf', 'strftime', and 'strfmon'), Darwin 55489targets also support the 'CFString' (or '__CFString__') archetype in the 55490'format' attribute. Declarations with this archetype are parsed for 55491correct syntax and argument types. However, parsing of the format 55492string itself and validating arguments against it in calls to such 55493functions is currently not performed. 55494 55495 Additionally, 'CFStringRefs' (defined by the 'CoreFoundation' headers) 55496may also be used as format arguments. Note that the relevant headers 55497are only likely to be available on Darwin (OSX) installations. On such 55498installations, the XCode and system documentation provide descriptions 55499of 'CFString', 'CFStringRefs' and associated functions. 55500 55501 55502File: gcc.info, Node: Pragmas, Next: Unnamed Fields, Prev: Target Format Checks, Up: C Extensions 55503 555046.62 Pragmas Accepted by GCC 55505============================ 55506 55507GCC supports several types of pragmas, primarily in order to compile 55508code originally written for other compilers. Note that in general we do 55509not recommend the use of pragmas; *Note Function Attributes::, for 55510further explanation. 55511 55512 The GNU C preprocessor recognizes several pragmas in addition to the 55513compiler pragmas documented here. Refer to the CPP manual for more 55514information. 55515 55516* Menu: 55517 55518* AArch64 Pragmas:: 55519* ARM Pragmas:: 55520* M32C Pragmas:: 55521* MeP Pragmas:: 55522* PRU Pragmas:: 55523* RS/6000 and PowerPC Pragmas:: 55524* S/390 Pragmas:: 55525* Darwin Pragmas:: 55526* Solaris Pragmas:: 55527* Symbol-Renaming Pragmas:: 55528* Structure-Layout Pragmas:: 55529* Weak Pragmas:: 55530* Diagnostic Pragmas:: 55531* Visibility Pragmas:: 55532* Push/Pop Macro Pragmas:: 55533* Function Specific Option Pragmas:: 55534* Loop-Specific Pragmas:: 55535 55536 55537File: gcc.info, Node: AArch64 Pragmas, Next: ARM Pragmas, Up: Pragmas 55538 555396.62.1 AArch64 Pragmas 55540---------------------- 55541 55542The pragmas defined by the AArch64 target correspond to the AArch64 55543target function attributes. They can be specified as below: 55544 #pragma GCC target("string") 55545 55546 where 'STRING' can be any string accepted as an AArch64 target 55547attribute. *Note AArch64 Function Attributes::, for more details on the 55548permissible values of 'string'. 55549 55550 55551File: gcc.info, Node: ARM Pragmas, Next: M32C Pragmas, Prev: AArch64 Pragmas, Up: Pragmas 55552 555536.62.2 ARM Pragmas 55554------------------ 55555 55556The ARM target defines pragmas for controlling the default addition of 55557'long_call' and 'short_call' attributes to functions. *Note Function 55558Attributes::, for information about the effects of these attributes. 55559 55560'long_calls' 55561 Set all subsequent functions to have the 'long_call' attribute. 55562 55563'no_long_calls' 55564 Set all subsequent functions to have the 'short_call' attribute. 55565 55566'long_calls_off' 55567 Do not affect the 'long_call' or 'short_call' attributes of 55568 subsequent functions. 55569 55570 55571File: gcc.info, Node: M32C Pragmas, Next: MeP Pragmas, Prev: ARM Pragmas, Up: Pragmas 55572 555736.62.3 M32C Pragmas 55574------------------- 55575 55576'GCC memregs NUMBER' 55577 Overrides the command-line option '-memregs=' for the current file. 55578 Use with care! This pragma must be before any function in the 55579 file, and mixing different memregs values in different objects may 55580 make them incompatible. This pragma is useful when a 55581 performance-critical function uses a memreg for temporary values, 55582 as it may allow you to reduce the number of memregs used. 55583 55584'ADDRESS NAME ADDRESS' 55585 For any declared symbols matching NAME, this does three things to 55586 that symbol: it forces the symbol to be located at the given 55587 address (a number), it forces the symbol to be volatile, and it 55588 changes the symbol's scope to be static. This pragma exists for 55589 compatibility with other compilers, but note that the common 55590 '1234H' numeric syntax is not supported (use '0x1234' instead). 55591 Example: 55592 55593 #pragma ADDRESS port3 0x103 55594 char port3; 55595 55596 55597File: gcc.info, Node: MeP Pragmas, Next: PRU Pragmas, Prev: M32C Pragmas, Up: Pragmas 55598 555996.62.4 MeP Pragmas 55600------------------ 55601 55602'custom io_volatile (on|off)' 55603 Overrides the command-line option '-mio-volatile' for the current 55604 file. Note that for compatibility with future GCC releases, this 55605 option should only be used once before any 'io' variables in each 55606 file. 55607 55608'GCC coprocessor available REGISTERS' 55609 Specifies which coprocessor registers are available to the register 55610 allocator. REGISTERS may be a single register, register range 55611 separated by ellipses, or comma-separated list of those. Example: 55612 55613 #pragma GCC coprocessor available $c0...$c10, $c28 55614 55615'GCC coprocessor call_saved REGISTERS' 55616 Specifies which coprocessor registers are to be saved and restored 55617 by any function using them. REGISTERS may be a single register, 55618 register range separated by ellipses, or comma-separated list of 55619 those. Example: 55620 55621 #pragma GCC coprocessor call_saved $c4...$c6, $c31 55622 55623'GCC coprocessor subclass '(A|B|C|D)' = REGISTERS' 55624 Creates and defines a register class. These register classes can 55625 be used by inline 'asm' constructs. REGISTERS may be a single 55626 register, register range separated by ellipses, or comma-separated 55627 list of those. Example: 55628 55629 #pragma GCC coprocessor subclass 'B' = $c2, $c4, $c6 55630 55631 asm ("cpfoo %0" : "=B" (x)); 55632 55633'GCC disinterrupt NAME , NAME ...' 55634 For the named functions, the compiler adds code to disable 55635 interrupts for the duration of those functions. If any functions 55636 so named are not encountered in the source, a warning is emitted 55637 that the pragma is not used. Examples: 55638 55639 #pragma disinterrupt foo 55640 #pragma disinterrupt bar, grill 55641 int foo () { ... } 55642 55643'GCC call NAME , NAME ...' 55644 For the named functions, the compiler always uses a 55645 register-indirect call model when calling the named functions. 55646 Examples: 55647 55648 extern int foo (); 55649 #pragma call foo 55650 55651 55652File: gcc.info, Node: PRU Pragmas, Next: RS/6000 and PowerPC Pragmas, Prev: MeP Pragmas, Up: Pragmas 55653 556546.62.5 PRU Pragmas 55655------------------ 55656 55657'ctable_entry INDEX CONSTANT_ADDRESS' 55658 Specifies that the PRU CTABLE entry given by INDEX has the value 55659 CONSTANT_ADDRESS. This enables GCC to emit LBCO/SBCO instructions 55660 when the load/store address is known and can be addressed with some 55661 CTABLE entry. For example: 55662 55663 /* will compile to "sbco Rx, 2, 0x10, 4" */ 55664 #pragma ctable_entry 2 0x4802a000 55665 *(unsigned int *)0x4802a010 = val; 55666 55667 55668File: gcc.info, Node: RS/6000 and PowerPC Pragmas, Next: S/390 Pragmas, Prev: PRU Pragmas, Up: Pragmas 55669 556706.62.6 RS/6000 and PowerPC Pragmas 55671---------------------------------- 55672 55673The RS/6000 and PowerPC targets define one pragma for controlling 55674whether or not the 'longcall' attribute is added to function 55675declarations by default. This pragma overrides the '-mlongcall' option, 55676but not the 'longcall' and 'shortcall' attributes. *Note RS/6000 and 55677PowerPC Options::, for more information about when long calls are and 55678are not necessary. 55679 55680'longcall (1)' 55681 Apply the 'longcall' attribute to all subsequent function 55682 declarations. 55683 55684'longcall (0)' 55685 Do not apply the 'longcall' attribute to subsequent function 55686 declarations. 55687 55688 55689File: gcc.info, Node: S/390 Pragmas, Next: Darwin Pragmas, Prev: RS/6000 and PowerPC Pragmas, Up: Pragmas 55690 556916.62.7 S/390 Pragmas 55692-------------------- 55693 55694The pragmas defined by the S/390 target correspond to the S/390 target 55695function attributes and some the additional options: 55696 55697'zvector' 55698'no-zvector' 55699 55700 Note that options of the pragma, unlike options of the target 55701attribute, do change the value of preprocessor macros like '__VEC__'. 55702They can be specified as below: 55703 55704 #pragma GCC target("string[,string]...") 55705 #pragma GCC target("string"[,"string"]...) 55706 55707 55708File: gcc.info, Node: Darwin Pragmas, Next: Solaris Pragmas, Prev: S/390 Pragmas, Up: Pragmas 55709 557106.62.8 Darwin Pragmas 55711--------------------- 55712 55713The following pragmas are available for all architectures running the 55714Darwin operating system. These are useful for compatibility with other 55715Mac OS compilers. 55716 55717'mark TOKENS...' 55718 This pragma is accepted, but has no effect. 55719 55720'options align=ALIGNMENT' 55721 This pragma sets the alignment of fields in structures. The values 55722 of ALIGNMENT may be 'mac68k', to emulate m68k alignment, or 55723 'power', to emulate PowerPC alignment. Uses of this pragma nest 55724 properly; to restore the previous setting, use 'reset' for the 55725 ALIGNMENT. 55726 55727'segment TOKENS...' 55728 This pragma is accepted, but has no effect. 55729 55730'unused (VAR [, VAR]...)' 55731 This pragma declares variables to be possibly unused. GCC does not 55732 produce warnings for the listed variables. The effect is similar 55733 to that of the 'unused' attribute, except that this pragma may 55734 appear anywhere within the variables' scopes. 55735 55736 55737File: gcc.info, Node: Solaris Pragmas, Next: Symbol-Renaming Pragmas, Prev: Darwin Pragmas, Up: Pragmas 55738 557396.62.9 Solaris Pragmas 55740---------------------- 55741 55742The Solaris target supports '#pragma redefine_extname' (*note 55743Symbol-Renaming Pragmas::). It also supports additional '#pragma' 55744directives for compatibility with the system compiler. 55745 55746'align ALIGNMENT (VARIABLE [, VARIABLE]...)' 55747 55748 Increase the minimum alignment of each VARIABLE to ALIGNMENT. This 55749 is the same as GCC's 'aligned' attribute *note Variable 55750 Attributes::). Macro expansion occurs on the arguments to this 55751 pragma when compiling C and Objective-C. It does not currently 55752 occur when compiling C++, but this is a bug which may be fixed in a 55753 future release. 55754 55755'fini (FUNCTION [, FUNCTION]...)' 55756 55757 This pragma causes each listed FUNCTION to be called after main, or 55758 during shared module unloading, by adding a call to the '.fini' 55759 section. 55760 55761'init (FUNCTION [, FUNCTION]...)' 55762 55763 This pragma causes each listed FUNCTION to be called during 55764 initialization (before 'main') or during shared module loading, by 55765 adding a call to the '.init' section. 55766 55767 55768File: gcc.info, Node: Symbol-Renaming Pragmas, Next: Structure-Layout Pragmas, Prev: Solaris Pragmas, Up: Pragmas 55769 557706.62.10 Symbol-Renaming Pragmas 55771------------------------------- 55772 55773GCC supports a '#pragma' directive that changes the name used in 55774assembly for a given declaration. While this pragma is supported on all 55775platforms, it is intended primarily to provide compatibility with the 55776Solaris system headers. This effect can also be achieved using the asm 55777labels extension (*note Asm Labels::). 55778 55779'redefine_extname OLDNAME NEWNAME' 55780 55781 This pragma gives the C function OLDNAME the assembly symbol 55782 NEWNAME. The preprocessor macro '__PRAGMA_REDEFINE_EXTNAME' is 55783 defined if this pragma is available (currently on all platforms). 55784 55785 This pragma and the 'asm' labels extension interact in a complicated 55786manner. Here are some corner cases you may want to be aware of: 55787 55788 1. This pragma silently applies only to declarations with external 55789 linkage. The 'asm' label feature does not have this restriction. 55790 55791 2. In C++, this pragma silently applies only to declarations with "C" 55792 linkage. Again, 'asm' labels do not have this restriction. 55793 55794 3. If either of the ways of changing the assembly name of a 55795 declaration are applied to a declaration whose assembly name has 55796 already been determined (either by a previous use of one of these 55797 features, or because the compiler needed the assembly name in order 55798 to generate code), and the new name is different, a warning issues 55799 and the name does not change. 55800 55801 4. The OLDNAME used by '#pragma redefine_extname' is always the 55802 C-language name. 55803 55804 55805File: gcc.info, Node: Structure-Layout Pragmas, Next: Weak Pragmas, Prev: Symbol-Renaming Pragmas, Up: Pragmas 55806 558076.62.11 Structure-Layout Pragmas 55808-------------------------------- 55809 55810For compatibility with Microsoft Windows compilers, GCC supports a set 55811of '#pragma' directives that change the maximum alignment of members of 55812structures (other than zero-width bit-fields), unions, and classes 55813subsequently defined. The N value below always is required to be a 55814small power of two and specifies the new alignment in bytes. 55815 55816 1. '#pragma pack(N)' simply sets the new alignment. 55817 2. '#pragma pack()' sets the alignment to the one that was in effect 55818 when compilation started (see also command-line option 55819 '-fpack-struct[=N]' *note Code Gen Options::). 55820 3. '#pragma pack(push[,N])' pushes the current alignment setting on an 55821 internal stack and then optionally sets the new alignment. 55822 4. '#pragma pack(pop)' restores the alignment setting to the one saved 55823 at the top of the internal stack (and removes that stack entry). 55824 Note that '#pragma pack([N])' does not influence this internal 55825 stack; thus it is possible to have '#pragma pack(push)' followed by 55826 multiple '#pragma pack(N)' instances and finalized by a single 55827 '#pragma pack(pop)'. 55828 55829 Some targets, e.g. x86 and PowerPC, support the '#pragma ms_struct' 55830directive which lays out structures and unions subsequently defined as 55831the documented '__attribute__ ((ms_struct))'. 55832 55833 1. '#pragma ms_struct on' turns on the Microsoft layout. 55834 2. '#pragma ms_struct off' turns off the Microsoft layout. 55835 3. '#pragma ms_struct reset' goes back to the default layout. 55836 55837 Most targets also support the '#pragma scalar_storage_order' directive 55838which lays out structures and unions subsequently defined as the 55839documented '__attribute__ ((scalar_storage_order))'. 55840 55841 1. '#pragma scalar_storage_order big-endian' sets the storage order of 55842 the scalar fields to big-endian. 55843 2. '#pragma scalar_storage_order little-endian' sets the storage order 55844 of the scalar fields to little-endian. 55845 3. '#pragma scalar_storage_order default' goes back to the endianness 55846 that was in effect when compilation started (see also command-line 55847 option '-fsso-struct=ENDIANNESS' *note C Dialect Options::). 55848 55849 55850File: gcc.info, Node: Weak Pragmas, Next: Diagnostic Pragmas, Prev: Structure-Layout Pragmas, Up: Pragmas 55851 558526.62.12 Weak Pragmas 55853-------------------- 55854 55855For compatibility with SVR4, GCC supports a set of '#pragma' directives 55856for declaring symbols to be weak, and defining weak aliases. 55857 55858'#pragma weak SYMBOL' 55859 This pragma declares SYMBOL to be weak, as if the declaration had 55860 the attribute of the same name. The pragma may appear before or 55861 after the declaration of SYMBOL. It is not an error for SYMBOL to 55862 never be defined at all. 55863 55864'#pragma weak SYMBOL1 = SYMBOL2' 55865 This pragma declares SYMBOL1 to be a weak alias of SYMBOL2. It is 55866 an error if SYMBOL2 is not defined in the current translation unit. 55867 55868 55869File: gcc.info, Node: Diagnostic Pragmas, Next: Visibility Pragmas, Prev: Weak Pragmas, Up: Pragmas 55870 558716.62.13 Diagnostic Pragmas 55872-------------------------- 55873 55874GCC allows the user to selectively enable or disable certain types of 55875diagnostics, and change the kind of the diagnostic. For example, a 55876project's policy might require that all sources compile with '-Werror' 55877but certain files might have exceptions allowing specific types of 55878warnings. Or, a project might selectively enable diagnostics and treat 55879them as errors depending on which preprocessor macros are defined. 55880 55881'#pragma GCC diagnostic KIND OPTION' 55882 55883 Modifies the disposition of a diagnostic. Note that not all 55884 diagnostics are modifiable; at the moment only warnings (normally 55885 controlled by '-W...') can be controlled, and not all of them. Use 55886 '-fdiagnostics-show-option' to determine which diagnostics are 55887 controllable and which option controls them. 55888 55889 KIND is 'error' to treat this diagnostic as an error, 'warning' to 55890 treat it like a warning (even if '-Werror' is in effect), or 55891 'ignored' if the diagnostic is to be ignored. OPTION is a double 55892 quoted string that matches the command-line option. 55893 55894 #pragma GCC diagnostic warning "-Wformat" 55895 #pragma GCC diagnostic error "-Wformat" 55896 #pragma GCC diagnostic ignored "-Wformat" 55897 55898 Note that these pragmas override any command-line options. GCC 55899 keeps track of the location of each pragma, and issues diagnostics 55900 according to the state as of that point in the source file. Thus, 55901 pragmas occurring after a line do not affect diagnostics caused by 55902 that line. 55903 55904'#pragma GCC diagnostic push' 55905'#pragma GCC diagnostic pop' 55906 55907 Causes GCC to remember the state of the diagnostics as of each 55908 'push', and restore to that point at each 'pop'. If a 'pop' has no 55909 matching 'push', the command-line options are restored. 55910 55911 #pragma GCC diagnostic error "-Wuninitialized" 55912 foo(a); /* error is given for this one */ 55913 #pragma GCC diagnostic push 55914 #pragma GCC diagnostic ignored "-Wuninitialized" 55915 foo(b); /* no diagnostic for this one */ 55916 #pragma GCC diagnostic pop 55917 foo(c); /* error is given for this one */ 55918 #pragma GCC diagnostic pop 55919 foo(d); /* depends on command-line options */ 55920 55921'#pragma GCC diagnostic ignored_attributes' 55922 55923 Similarly to '-Wno-attributes=', this pragma allows users to 55924 suppress warnings about unknown scoped attributes (in C++11 and 55925 C2X). For example, '#pragma GCC diagnostic ignored_attributes 55926 "vendor::attr"' disables warning about the following declaration: 55927 55928 [[vendor::attr]] void f(); 55929 55930 whereas '#pragma GCC diagnostic ignored_attributes "vendor::"' 55931 prevents warning about both of these declarations: 55932 55933 [[vendor::safe]] void f(); 55934 [[vendor::unsafe]] void f2(); 55935 55936 GCC also offers a simple mechanism for printing messages during 55937compilation. 55938 55939'#pragma message STRING' 55940 55941 Prints STRING as a compiler message on compilation. The message is 55942 informational only, and is neither a compilation warning nor an 55943 error. Newlines can be included in the string by using the '\n' 55944 escape sequence. 55945 55946 #pragma message "Compiling " __FILE__ "..." 55947 55948 STRING may be parenthesized, and is printed with location 55949 information. For example, 55950 55951 #define DO_PRAGMA(x) _Pragma (#x) 55952 #define TODO(x) DO_PRAGMA(message ("TODO - " #x)) 55953 55954 TODO(Remember to fix this) 55955 55956 prints '/tmp/file.c:4: note: #pragma message: TODO - Remember to 55957 fix this'. 55958 55959'#pragma GCC error MESSAGE' 55960 Generates an error message. This pragma _is_ considered to 55961 indicate an error in the compilation, and it will be treated as 55962 such. 55963 55964 Newlines can be included in the string by using the '\n' escape 55965 sequence. They will be displayed as newlines even if the 55966 '-fmessage-length' option is set to zero. 55967 55968 The error is only generated if the pragma is present in the code 55969 after pre-processing has been completed. It does not matter 55970 however if the code containing the pragma is unreachable: 55971 55972 #if 0 55973 #pragma GCC error "this error is not seen" 55974 #endif 55975 void foo (void) 55976 { 55977 return; 55978 #pragma GCC error "this error is seen" 55979 } 55980 55981'#pragma GCC warning MESSAGE' 55982 This is just like 'pragma GCC error' except that a warning message 55983 is issued instead of an error message. Unless '-Werror' is in 55984 effect, in which case this pragma will generate an error as well. 55985 55986 55987File: gcc.info, Node: Visibility Pragmas, Next: Push/Pop Macro Pragmas, Prev: Diagnostic Pragmas, Up: Pragmas 55988 559896.62.14 Visibility Pragmas 55990-------------------------- 55991 55992'#pragma GCC visibility push(VISIBILITY)' 55993'#pragma GCC visibility pop' 55994 55995 This pragma allows the user to set the visibility for multiple 55996 declarations without having to give each a visibility attribute 55997 (*note Function Attributes::). 55998 55999 In C++, '#pragma GCC visibility' affects only namespace-scope 56000 declarations. Class members and template specializations are not 56001 affected; if you want to override the visibility for a particular 56002 member or instantiation, you must use an attribute. 56003 56004 56005File: gcc.info, Node: Push/Pop Macro Pragmas, Next: Function Specific Option Pragmas, Prev: Visibility Pragmas, Up: Pragmas 56006 560076.62.15 Push/Pop Macro Pragmas 56008------------------------------ 56009 56010For compatibility with Microsoft Windows compilers, GCC supports 56011'#pragma push_macro("MACRO_NAME")' and '#pragma 56012pop_macro("MACRO_NAME")'. 56013 56014'#pragma push_macro("MACRO_NAME")' 56015 This pragma saves the value of the macro named as MACRO_NAME to the 56016 top of the stack for this macro. 56017 56018'#pragma pop_macro("MACRO_NAME")' 56019 This pragma sets the value of the macro named as MACRO_NAME to the 56020 value on top of the stack for this macro. If the stack for 56021 MACRO_NAME is empty, the value of the macro remains unchanged. 56022 56023 For example: 56024 56025 #define X 1 56026 #pragma push_macro("X") 56027 #undef X 56028 #define X -1 56029 #pragma pop_macro("X") 56030 int x [X]; 56031 56032In this example, the definition of X as 1 is saved by '#pragma 56033push_macro' and restored by '#pragma pop_macro'. 56034 56035 56036File: gcc.info, Node: Function Specific Option Pragmas, Next: Loop-Specific Pragmas, Prev: Push/Pop Macro Pragmas, Up: Pragmas 56037 560386.62.16 Function Specific Option Pragmas 56039---------------------------------------- 56040 56041'#pragma GCC target (STRING, ...)' 56042 56043 This pragma allows you to set target-specific options for functions 56044 defined later in the source file. One or more strings can be 56045 specified. Each function that is defined after this point is 56046 treated as if it had been declared with one 'target('STRING')' 56047 attribute for each STRING argument. The parentheses around the 56048 strings in the pragma are optional. *Note Function Attributes::, 56049 for more information about the 'target' attribute and the attribute 56050 syntax. 56051 56052 The '#pragma GCC target' pragma is presently implemented for x86, 56053 ARM, AArch64, PowerPC, S/390, and Nios II targets only. 56054 56055'#pragma GCC optimize (STRING, ...)' 56056 56057 This pragma allows you to set global optimization options for 56058 functions defined later in the source file. One or more strings 56059 can be specified. Each function that is defined after this point 56060 is treated as if it had been declared with one 'optimize('STRING')' 56061 attribute for each STRING argument. The parentheses around the 56062 strings in the pragma are optional. *Note Function Attributes::, 56063 for more information about the 'optimize' attribute and the 56064 attribute syntax. 56065 56066'#pragma GCC push_options' 56067'#pragma GCC pop_options' 56068 56069 These pragmas maintain a stack of the current target and 56070 optimization options. It is intended for include files where you 56071 temporarily want to switch to using a different '#pragma GCC 56072 target' or '#pragma GCC optimize' and then to pop back to the 56073 previous options. 56074 56075'#pragma GCC reset_options' 56076 56077 This pragma clears the current '#pragma GCC target' and '#pragma 56078 GCC optimize' to use the default switches as specified on the 56079 command line. 56080 56081 56082File: gcc.info, Node: Loop-Specific Pragmas, Prev: Function Specific Option Pragmas, Up: Pragmas 56083 560846.62.17 Loop-Specific Pragmas 56085----------------------------- 56086 56087'#pragma GCC ivdep' 56088 56089 With this pragma, the programmer asserts that there are no 56090 loop-carried dependencies which would prevent consecutive 56091 iterations of the following loop from executing concurrently with 56092 SIMD (single instruction multiple data) instructions. 56093 56094 For example, the compiler can only unconditionally vectorize the 56095 following loop with the pragma: 56096 56097 void foo (int n, int *a, int *b, int *c) 56098 { 56099 int i, j; 56100 #pragma GCC ivdep 56101 for (i = 0; i < n; ++i) 56102 a[i] = b[i] + c[i]; 56103 } 56104 56105 In this example, using the 'restrict' qualifier had the same 56106 effect. In the following example, that would not be possible. 56107 Assume k < -m or k >= m. Only with the pragma, the compiler knows 56108 that it can unconditionally vectorize the following loop: 56109 56110 void ignore_vec_dep (int *a, int k, int c, int m) 56111 { 56112 #pragma GCC ivdep 56113 for (int i = 0; i < m; i++) 56114 a[i] = a[i + k] * c; 56115 } 56116 56117'#pragma GCC unroll N' 56118 56119 You can use this pragma to control how many times a loop should be 56120 unrolled. It must be placed immediately before a 'for', 'while' or 56121 'do' loop or a '#pragma GCC ivdep', and applies only to the loop 56122 that follows. N is an integer constant expression specifying the 56123 unrolling factor. The values of 0 and 1 block any unrolling of the 56124 loop. 56125 56126 56127File: gcc.info, Node: Unnamed Fields, Next: Thread-Local, Prev: Pragmas, Up: C Extensions 56128 561296.63 Unnamed Structure and Union Fields 56130======================================= 56131 56132As permitted by ISO C11 and for compatibility with other compilers, GCC 56133allows you to define a structure or union that contains, as fields, 56134structures and unions without names. For example: 56135 56136 struct { 56137 int a; 56138 union { 56139 int b; 56140 float c; 56141 }; 56142 int d; 56143 } foo; 56144 56145In this example, you are able to access members of the unnamed union 56146with code like 'foo.b'. Note that only unnamed structs and unions are 56147allowed, you may not have, for example, an unnamed 'int'. 56148 56149 You must never create such structures that cause ambiguous field 56150definitions. For example, in this structure: 56151 56152 struct { 56153 int a; 56154 struct { 56155 int a; 56156 }; 56157 } foo; 56158 56159it is ambiguous which 'a' is being referred to with 'foo.a'. The 56160compiler gives errors for such constructs. 56161 56162 Unless '-fms-extensions' is used, the unnamed field must be a structure 56163or union definition without a tag (for example, 'struct { int a; };'). 56164If '-fms-extensions' is used, the field may also be a definition with a 56165tag such as 'struct foo { int a; };', a reference to a previously 56166defined structure or union such as 'struct foo;', or a reference to a 56167'typedef' name for a previously defined structure or union type. 56168 56169 The option '-fplan9-extensions' enables '-fms-extensions' as well as 56170two other extensions. First, a pointer to a structure is automatically 56171converted to a pointer to an anonymous field for assignments and 56172function calls. For example: 56173 56174 struct s1 { int a; }; 56175 struct s2 { struct s1; }; 56176 extern void f1 (struct s1 *); 56177 void f2 (struct s2 *p) { f1 (p); } 56178 56179In the call to 'f1' inside 'f2', the pointer 'p' is converted into a 56180pointer to the anonymous field. 56181 56182 Second, when the type of an anonymous field is a 'typedef' for a 56183'struct' or 'union', code may refer to the field using the name of the 56184'typedef'. 56185 56186 typedef struct { int a; } s1; 56187 struct s2 { s1; }; 56188 s1 f1 (struct s2 *p) { return p->s1; } 56189 56190 These usages are only permitted when they are not ambiguous. 56191 56192 56193File: gcc.info, Node: Thread-Local, Next: Binary constants, Prev: Unnamed Fields, Up: C Extensions 56194 561956.64 Thread-Local Storage 56196========================= 56197 56198Thread-local storage (TLS) is a mechanism by which variables are 56199allocated such that there is one instance of the variable per extant 56200thread. The runtime model GCC uses to implement this originates in the 56201IA-64 processor-specific ABI, but has since been migrated to other 56202processors as well. It requires significant support from the linker 56203('ld'), dynamic linker ('ld.so'), and system libraries ('libc.so' and 56204'libpthread.so'), so it is not available everywhere. 56205 56206 At the user level, the extension is visible with a new storage class 56207keyword: '__thread'. For example: 56208 56209 __thread int i; 56210 extern __thread struct state s; 56211 static __thread char *p; 56212 56213 The '__thread' specifier may be used alone, with the 'extern' or 56214'static' specifiers, but with no other storage class specifier. When 56215used with 'extern' or 'static', '__thread' must appear immediately after 56216the other storage class specifier. 56217 56218 The '__thread' specifier may be applied to any global, file-scoped 56219static, function-scoped static, or static data member of a class. It 56220may not be applied to block-scoped automatic or non-static data member. 56221 56222 When the address-of operator is applied to a thread-local variable, it 56223is evaluated at run time and returns the address of the current thread's 56224instance of that variable. An address so obtained may be used by any 56225thread. When a thread terminates, any pointers to thread-local 56226variables in that thread become invalid. 56227 56228 No static initialization may refer to the address of a thread-local 56229variable. 56230 56231 In C++, if an initializer is present for a thread-local variable, it 56232must be a CONSTANT-EXPRESSION, as defined in 5.19.2 of the ANSI/ISO C++ 56233standard. 56234 56235 See ELF Handling For Thread-Local Storage 56236(https://www.akkadia.org/drepper/tls.pdf) for a detailed explanation of 56237the four thread-local storage addressing models, and how the runtime is 56238expected to function. 56239 56240* Menu: 56241 56242* C99 Thread-Local Edits:: 56243* C++98 Thread-Local Edits:: 56244 56245 56246File: gcc.info, Node: C99 Thread-Local Edits, Next: C++98 Thread-Local Edits, Up: Thread-Local 56247 562486.64.1 ISO/IEC 9899:1999 Edits for Thread-Local Storage 56249------------------------------------------------------- 56250 56251The following are a set of changes to ISO/IEC 9899:1999 (aka C99) that 56252document the exact semantics of the language extension. 56253 56254 * '5.1.2 Execution environments' 56255 56256 Add new text after paragraph 1 56257 56258 Within either execution environment, a "thread" is a flow of 56259 control within a program. It is implementation defined 56260 whether or not there may be more than one thread associated 56261 with a program. It is implementation defined how threads 56262 beyond the first are created, the name and type of the 56263 function called at thread startup, and how threads may be 56264 terminated. However, objects with thread storage duration 56265 shall be initialized before thread startup. 56266 56267 * '6.2.4 Storage durations of objects' 56268 56269 Add new text before paragraph 3 56270 56271 An object whose identifier is declared with the storage-class 56272 specifier '__thread' has "thread storage duration". Its 56273 lifetime is the entire execution of the thread, and its stored 56274 value is initialized only once, prior to thread startup. 56275 56276 * '6.4.1 Keywords' 56277 56278 Add '__thread'. 56279 56280 * '6.7.1 Storage-class specifiers' 56281 56282 Add '__thread' to the list of storage class specifiers in paragraph 56283 1. 56284 56285 Change paragraph 2 to 56286 56287 With the exception of '__thread', at most one storage-class 56288 specifier may be given [...]. The '__thread' specifier may be 56289 used alone, or immediately following 'extern' or 'static'. 56290 56291 Add new text after paragraph 6 56292 56293 The declaration of an identifier for a variable that has block 56294 scope that specifies '__thread' shall also specify either 56295 'extern' or 'static'. 56296 56297 The '__thread' specifier shall be used only with variables. 56298 56299 56300File: gcc.info, Node: C++98 Thread-Local Edits, Prev: C99 Thread-Local Edits, Up: Thread-Local 56301 563026.64.2 ISO/IEC 14882:1998 Edits for Thread-Local Storage 56303-------------------------------------------------------- 56304 56305The following are a set of changes to ISO/IEC 14882:1998 (aka C++98) 56306that document the exact semantics of the language extension. 56307 56308 * [intro.execution] 56309 56310 New text after paragraph 4 56311 56312 A "thread" is a flow of control within the abstract machine. 56313 It is implementation defined whether or not there may be more 56314 than one thread. 56315 56316 New text after paragraph 7 56317 56318 It is unspecified whether additional action must be taken to 56319 ensure when and whether side effects are visible to other 56320 threads. 56321 56322 * [lex.key] 56323 56324 Add '__thread'. 56325 56326 * [basic.start.main] 56327 56328 Add after paragraph 5 56329 56330 The thread that begins execution at the 'main' function is 56331 called the "main thread". It is implementation defined how 56332 functions beginning threads other than the main thread are 56333 designated or typed. A function so designated, as well as the 56334 'main' function, is called a "thread startup function". It is 56335 implementation defined what happens if a thread startup 56336 function returns. It is implementation defined what happens 56337 to other threads when any thread calls 'exit'. 56338 56339 * [basic.start.init] 56340 56341 Add after paragraph 4 56342 56343 The storage for an object of thread storage duration shall be 56344 statically initialized before the first statement of the 56345 thread startup function. An object of thread storage duration 56346 shall not require dynamic initialization. 56347 56348 * [basic.start.term] 56349 56350 Add after paragraph 3 56351 56352 The type of an object with thread storage duration shall not 56353 have a non-trivial destructor, nor shall it be an array type 56354 whose elements (directly or indirectly) have non-trivial 56355 destructors. 56356 56357 * [basic.stc] 56358 56359 Add "thread storage duration" to the list in paragraph 1. 56360 56361 Change paragraph 2 56362 56363 Thread, static, and automatic storage durations are associated 56364 with objects introduced by declarations [...]. 56365 56366 Add '__thread' to the list of specifiers in paragraph 3. 56367 56368 * [basic.stc.thread] 56369 56370 New section before [basic.stc.static] 56371 56372 The keyword '__thread' applied to a non-local object gives the 56373 object thread storage duration. 56374 56375 A local variable or class data member declared both 'static' 56376 and '__thread' gives the variable or member thread storage 56377 duration. 56378 56379 * [basic.stc.static] 56380 56381 Change paragraph 1 56382 56383 All objects that have neither thread storage duration, dynamic 56384 storage duration nor are local [...]. 56385 56386 * [dcl.stc] 56387 56388 Add '__thread' to the list in paragraph 1. 56389 56390 Change paragraph 1 56391 56392 With the exception of '__thread', at most one 56393 STORAGE-CLASS-SPECIFIER shall appear in a given 56394 DECL-SPECIFIER-SEQ. The '__thread' specifier may be used 56395 alone, or immediately following the 'extern' or 'static' 56396 specifiers. [...] 56397 56398 Add after paragraph 5 56399 56400 The '__thread' specifier can be applied only to the names of 56401 objects and to anonymous unions. 56402 56403 * [class.mem] 56404 56405 Add after paragraph 6 56406 56407 Non-'static' members shall not be '__thread'. 56408 56409 56410File: gcc.info, Node: Binary constants, Prev: Thread-Local, Up: C Extensions 56411 564126.65 Binary Constants using the '0b' Prefix 56413=========================================== 56414 56415Integer constants can be written as binary constants, consisting of a 56416sequence of '0' and '1' digits, prefixed by '0b' or '0B'. This is 56417particularly useful in environments that operate a lot on the bit level 56418(like microcontrollers). 56419 56420 The following statements are identical: 56421 56422 i = 42; 56423 i = 0x2a; 56424 i = 052; 56425 i = 0b101010; 56426 56427 The type of these constants follows the same rules as for octal or 56428hexadecimal integer constants, so suffixes like 'L' or 'UL' can be 56429applied. 56430 56431 56432File: gcc.info, Node: C++ Extensions, Next: Objective-C, Prev: C Extensions, Up: Top 56433 564347 Extensions to the C++ Language 56435******************************** 56436 56437The GNU compiler provides these extensions to the C++ language (and you 56438can also use most of the C language extensions in your C++ programs). 56439If you want to write code that checks whether these features are 56440available, you can test for the GNU compiler the same way as for C 56441programs: check for a predefined macro '__GNUC__'. You can also use 56442'__GNUG__' to test specifically for GNU C++ (*note Predefined Macros: 56443(cpp)Common Predefined Macros.). 56444 56445* Menu: 56446 56447* C++ Volatiles:: What constitutes an access to a volatile object. 56448* Restricted Pointers:: C99 restricted pointers and references. 56449* Vague Linkage:: Where G++ puts inlines, vtables and such. 56450* C++ Interface:: You can use a single C++ header file for both 56451 declarations and definitions. 56452* Template Instantiation:: Methods for ensuring that exactly one copy of 56453 each needed template instantiation is emitted. 56454* Bound member functions:: You can extract a function pointer to the 56455 method denoted by a '->*' or '.*' expression. 56456* C++ Attributes:: Variable, function, and type attributes for C++ only. 56457* Function Multiversioning:: Declaring multiple function versions. 56458* Type Traits:: Compiler support for type traits. 56459* C++ Concepts:: Improved support for generic programming. 56460* Deprecated Features:: Things will disappear from G++. 56461* Backwards Compatibility:: Compatibilities with earlier definitions of C++. 56462 56463 56464File: gcc.info, Node: C++ Volatiles, Next: Restricted Pointers, Up: C++ Extensions 56465 564667.1 When is a Volatile C++ Object Accessed? 56467=========================================== 56468 56469The C++ standard differs from the C standard in its treatment of 56470volatile objects. It fails to specify what constitutes a volatile 56471access, except to say that C++ should behave in a similar manner to C 56472with respect to volatiles, where possible. However, the different 56473lvalueness of expressions between C and C++ complicate the behavior. 56474G++ behaves the same as GCC for volatile access, *Note Volatiles: C 56475Extensions, for a description of GCC's behavior. 56476 56477 The C and C++ language specifications differ when an object is accessed 56478in a void context: 56479 56480 volatile int *src = SOMEVALUE; 56481 *src; 56482 56483 The C++ standard specifies that such expressions do not undergo lvalue 56484to rvalue conversion, and that the type of the dereferenced object may 56485be incomplete. The C++ standard does not specify explicitly that it is 56486lvalue to rvalue conversion that is responsible for causing an access. 56487There is reason to believe that it is, because otherwise certain simple 56488expressions become undefined. However, because it would surprise most 56489programmers, G++ treats dereferencing a pointer to volatile object of 56490complete type as GCC would do for an equivalent type in C. When the 56491object has incomplete type, G++ issues a warning; if you wish to force 56492an error, you must force a conversion to rvalue with, for instance, a 56493static cast. 56494 56495 When using a reference to volatile, G++ does not treat equivalent 56496expressions as accesses to volatiles, but instead issues a warning that 56497no volatile is accessed. The rationale for this is that otherwise it 56498becomes difficult to determine where volatile access occur, and not 56499possible to ignore the return value from functions returning volatile 56500references. Again, if you wish to force a read, cast the reference to 56501an rvalue. 56502 56503 G++ implements the same behavior as GCC does when assigning to a 56504volatile object--there is no reread of the assigned-to object, the 56505assigned rvalue is reused. Note that in C++ assignment expressions are 56506lvalues, and if used as an lvalue, the volatile object is referred to. 56507For instance, VREF refers to VOBJ, as expected, in the following 56508example: 56509 56510 volatile int vobj; 56511 volatile int &vref = vobj = SOMETHING; 56512 56513 56514File: gcc.info, Node: Restricted Pointers, Next: Vague Linkage, Prev: C++ Volatiles, Up: C++ Extensions 56515 565167.2 Restricting Pointer Aliasing 56517================================ 56518 56519As with the C front end, G++ understands the C99 feature of restricted 56520pointers, specified with the '__restrict__', or '__restrict' type 56521qualifier. Because you cannot compile C++ by specifying the '-std=c99' 56522language flag, 'restrict' is not a keyword in C++. 56523 56524 In addition to allowing restricted pointers, you can specify restricted 56525references, which indicate that the reference is not aliased in the 56526local context. 56527 56528 void fn (int *__restrict__ rptr, int &__restrict__ rref) 56529 { 56530 /* ... */ 56531 } 56532 56533In the body of 'fn', RPTR points to an unaliased integer and RREF refers 56534to a (different) unaliased integer. 56535 56536 You may also specify whether a member function's THIS pointer is 56537unaliased by using '__restrict__' as a member function qualifier. 56538 56539 void T::fn () __restrict__ 56540 { 56541 /* ... */ 56542 } 56543 56544Within the body of 'T::fn', THIS has the effective definition 'T 56545*__restrict__ const this'. Notice that the interpretation of a 56546'__restrict__' member function qualifier is different to that of 'const' 56547or 'volatile' qualifier, in that it is applied to the pointer rather 56548than the object. This is consistent with other compilers that implement 56549restricted pointers. 56550 56551 As with all outermost parameter qualifiers, '__restrict__' is ignored 56552in function definition matching. This means you only need to specify 56553'__restrict__' in a function definition, rather than in a function 56554prototype as well. 56555 56556 56557File: gcc.info, Node: Vague Linkage, Next: C++ Interface, Prev: Restricted Pointers, Up: C++ Extensions 56558 565597.3 Vague Linkage 56560================= 56561 56562There are several constructs in C++ that require space in the object 56563file but are not clearly tied to a single translation unit. We say that 56564these constructs have "vague linkage". Typically such constructs are 56565emitted wherever they are needed, though sometimes we can be more 56566clever. 56567 56568Inline Functions 56569 Inline functions are typically defined in a header file which can 56570 be included in many different compilations. Hopefully they can 56571 usually be inlined, but sometimes an out-of-line copy is necessary, 56572 if the address of the function is taken or if inlining fails. In 56573 general, we emit an out-of-line copy in all translation units where 56574 one is needed. As an exception, we only emit inline virtual 56575 functions with the vtable, since it always requires a copy. 56576 56577 Local static variables and string constants used in an inline 56578 function are also considered to have vague linkage, since they must 56579 be shared between all inlined and out-of-line instances of the 56580 function. 56581 56582VTables 56583 C++ virtual functions are implemented in most compilers using a 56584 lookup table, known as a vtable. The vtable contains pointers to 56585 the virtual functions provided by a class, and each object of the 56586 class contains a pointer to its vtable (or vtables, in some 56587 multiple-inheritance situations). If the class declares any 56588 non-inline, non-pure virtual functions, the first one is chosen as 56589 the "key method" for the class, and the vtable is only emitted in 56590 the translation unit where the key method is defined. 56591 56592 _Note:_ If the chosen key method is later defined as inline, the 56593 vtable is still emitted in every translation unit that defines it. 56594 Make sure that any inline virtuals are declared inline in the class 56595 body, even if they are not defined there. 56596 56597'type_info' objects 56598 C++ requires information about types to be written out in order to 56599 implement 'dynamic_cast', 'typeid' and exception handling. For 56600 polymorphic classes (classes with virtual functions), the 56601 'type_info' object is written out along with the vtable so that 56602 'dynamic_cast' can determine the dynamic type of a class object at 56603 run time. For all other types, we write out the 'type_info' object 56604 when it is used: when applying 'typeid' to an expression, throwing 56605 an object, or referring to a type in a catch clause or exception 56606 specification. 56607 56608Template Instantiations 56609 Most everything in this section also applies to template 56610 instantiations, but there are other options as well. *Note Where's 56611 the Template?: Template Instantiation. 56612 56613 When used with GNU ld version 2.8 or later on an ELF system such as 56614GNU/Linux or Solaris 2, or on Microsoft Windows, duplicate copies of 56615these constructs will be discarded at link time. This is known as 56616COMDAT support. 56617 56618 On targets that don't support COMDAT, but do support weak symbols, GCC 56619uses them. This way one copy overrides all the others, but the unused 56620copies still take up space in the executable. 56621 56622 For targets that do not support either COMDAT or weak symbols, most 56623entities with vague linkage are emitted as local symbols to avoid 56624duplicate definition errors from the linker. This does not happen for 56625local statics in inlines, however, as having multiple copies almost 56626certainly breaks things. 56627 56628 *Note Declarations and Definitions in One Header: C++ Interface, for 56629another way to control placement of these constructs. 56630 56631 56632File: gcc.info, Node: C++ Interface, Next: Template Instantiation, Prev: Vague Linkage, Up: C++ Extensions 56633 566347.4 C++ Interface and Implementation Pragmas 56635============================================ 56636 56637'#pragma interface' and '#pragma implementation' provide the user with a 56638way of explicitly directing the compiler to emit entities with vague 56639linkage (and debugging information) in a particular translation unit. 56640 56641 _Note:_ These '#pragma's have been superceded as of GCC 2.7.2 by COMDAT 56642support and the "key method" heuristic mentioned in *note Vague 56643Linkage::. Using them can actually cause your program to grow due to 56644unnecessary out-of-line copies of inline functions. 56645 56646'#pragma interface' 56647'#pragma interface "SUBDIR/OBJECTS.h"' 56648 Use this directive in _header files_ that define object classes, to 56649 save space in most of the object files that use those classes. 56650 Normally, local copies of certain information (backup copies of 56651 inline member functions, debugging information, and the internal 56652 tables that implement virtual functions) must be kept in each 56653 object file that includes class definitions. You can use this 56654 pragma to avoid such duplication. When a header file containing 56655 '#pragma interface' is included in a compilation, this auxiliary 56656 information is not generated (unless the main input source file 56657 itself uses '#pragma implementation'). Instead, the object files 56658 contain references to be resolved at link time. 56659 56660 The second form of this directive is useful for the case where you 56661 have multiple headers with the same name in different directories. 56662 If you use this form, you must specify the same string to '#pragma 56663 implementation'. 56664 56665'#pragma implementation' 56666'#pragma implementation "OBJECTS.h"' 56667 Use this pragma in a _main input file_, when you want full output 56668 from included header files to be generated (and made globally 56669 visible). The included header file, in turn, should use '#pragma 56670 interface'. Backup copies of inline member functions, debugging 56671 information, and the internal tables used to implement virtual 56672 functions are all generated in implementation files. 56673 56674 If you use '#pragma implementation' with no argument, it applies to 56675 an include file with the same basename(1) as your source file. For 56676 example, in 'allclass.cc', giving just '#pragma implementation' by 56677 itself is equivalent to '#pragma implementation "allclass.h"'. 56678 56679 Use the string argument if you want a single implementation file to 56680 include code from multiple header files. (You must also use 56681 '#include' to include the header file; '#pragma implementation' 56682 only specifies how to use the file--it doesn't actually include 56683 it.) 56684 56685 There is no way to split up the contents of a single header file 56686 into multiple implementation files. 56687 56688 '#pragma implementation' and '#pragma interface' also have an effect on 56689function inlining. 56690 56691 If you define a class in a header file marked with '#pragma interface', 56692the effect on an inline function defined in that class is similar to an 56693explicit 'extern' declaration--the compiler emits no code at all to 56694define an independent version of the function. Its definition is used 56695only for inlining with its callers. 56696 56697 Conversely, when you include the same header file in a main source file 56698that declares it as '#pragma implementation', the compiler emits code 56699for the function itself; this defines a version of the function that can 56700be found via pointers (or by callers compiled without inlining). If all 56701calls to the function can be inlined, you can avoid emitting the 56702function by compiling with '-fno-implement-inlines'. If any calls are 56703not inlined, you will get linker errors. 56704 56705 ---------- Footnotes ---------- 56706 56707 (1) A file's "basename" is the name stripped of all leading path 56708information and of trailing suffixes, such as '.h' or '.C' or '.cc'. 56709 56710 56711File: gcc.info, Node: Template Instantiation, Next: Bound member functions, Prev: C++ Interface, Up: C++ Extensions 56712 567137.5 Where's the Template? 56714========================= 56715 56716C++ templates were the first language feature to require more 56717intelligence from the environment than was traditionally found on a UNIX 56718system. Somehow the compiler and linker have to make sure that each 56719template instance occurs exactly once in the executable if it is needed, 56720and not at all otherwise. There are two basic approaches to this 56721problem, which are referred to as the Borland model and the Cfront 56722model. 56723 56724Borland model 56725 Borland C++ solved the template instantiation problem by adding the 56726 code equivalent of common blocks to their linker; the compiler 56727 emits template instances in each translation unit that uses them, 56728 and the linker collapses them together. The advantage of this 56729 model is that the linker only has to consider the object files 56730 themselves; there is no external complexity to worry about. The 56731 disadvantage is that compilation time is increased because the 56732 template code is being compiled repeatedly. Code written for this 56733 model tends to include definitions of all templates in the header 56734 file, since they must be seen to be instantiated. 56735 56736Cfront model 56737 The AT&T C++ translator, Cfront, solved the template instantiation 56738 problem by creating the notion of a template repository, an 56739 automatically maintained place where template instances are stored. 56740 A more modern version of the repository works as follows: As 56741 individual object files are built, the compiler places any template 56742 definitions and instantiations encountered in the repository. At 56743 link time, the link wrapper adds in the objects in the repository 56744 and compiles any needed instances that were not previously emitted. 56745 The advantages of this model are more optimal compilation speed and 56746 the ability to use the system linker; to implement the Borland 56747 model a compiler vendor also needs to replace the linker. The 56748 disadvantages are vastly increased complexity, and thus potential 56749 for error; for some code this can be just as transparent, but in 56750 practice it can been very difficult to build multiple programs in 56751 one directory and one program in multiple directories. Code 56752 written for this model tends to separate definitions of non-inline 56753 member templates into a separate file, which should be compiled 56754 separately. 56755 56756 G++ implements the Borland model on targets where the linker supports 56757it, including ELF targets (such as GNU/Linux), Mac OS X and Microsoft 56758Windows. Otherwise G++ implements neither automatic model. 56759 56760 You have the following options for dealing with template 56761instantiations: 56762 56763 1. Do nothing. Code written for the Borland model works fine, but 56764 each translation unit contains instances of each of the templates 56765 it uses. The duplicate instances will be discarded by the linker, 56766 but in a large program, this can lead to an unacceptable amount of 56767 code duplication in object files or shared libraries. 56768 56769 Duplicate instances of a template can be avoided by defining an 56770 explicit instantiation in one object file, and preventing the 56771 compiler from doing implicit instantiations in any other object 56772 files by using an explicit instantiation declaration, using the 56773 'extern template' syntax: 56774 56775 extern template int max (int, int); 56776 56777 This syntax is defined in the C++ 2011 standard, but has been 56778 supported by G++ and other compilers since well before 2011. 56779 56780 Explicit instantiations can be used for the largest or most 56781 frequently duplicated instances, without having to know exactly 56782 which other instances are used in the rest of the program. You can 56783 scatter the explicit instantiations throughout your program, 56784 perhaps putting them in the translation units where the instances 56785 are used or the translation units that define the templates 56786 themselves; you can put all of the explicit instantiations you need 56787 into one big file; or you can create small files like 56788 56789 #include "Foo.h" 56790 #include "Foo.cc" 56791 56792 template class Foo<int>; 56793 template ostream& operator << 56794 (ostream&, const Foo<int>&); 56795 56796 for each of the instances you need, and create a template 56797 instantiation library from those. 56798 56799 This is the simplest option, but also offers flexibility and 56800 fine-grained control when necessary. It is also the most portable 56801 alternative and programs using this approach will work with most 56802 modern compilers. 56803 56804 2. Compile your code with '-fno-implicit-templates' to disable the 56805 implicit generation of template instances, and explicitly 56806 instantiate all the ones you use. This approach requires more 56807 knowledge of exactly which instances you need than do the others, 56808 but it's less mysterious and allows greater control if you want to 56809 ensure that only the intended instances are used. 56810 56811 If you are using Cfront-model code, you can probably get away with 56812 not using '-fno-implicit-templates' when compiling files that don't 56813 '#include' the member template definitions. 56814 56815 If you use one big file to do the instantiations, you may want to 56816 compile it without '-fno-implicit-templates' so you get all of the 56817 instances required by your explicit instantiations (but not by any 56818 other files) without having to specify them as well. 56819 56820 In addition to forward declaration of explicit instantiations (with 56821 'extern'), G++ has extended the template instantiation syntax to 56822 support instantiation of the compiler support data for a template 56823 class (i.e. the vtable) without instantiating any of its members 56824 (with 'inline'), and instantiation of only the static data members 56825 of a template class, without the support data or member functions 56826 (with 'static'): 56827 56828 inline template class Foo<int>; 56829 static template class Foo<int>; 56830 56831 56832File: gcc.info, Node: Bound member functions, Next: C++ Attributes, Prev: Template Instantiation, Up: C++ Extensions 56833 568347.6 Extracting the Function Pointer from a Bound Pointer to Member Function 56835=========================================================================== 56836 56837In C++, pointer to member functions (PMFs) are implemented using a wide 56838pointer of sorts to handle all the possible call mechanisms; the PMF 56839needs to store information about how to adjust the 'this' pointer, and 56840if the function pointed to is virtual, where to find the vtable, and 56841where in the vtable to look for the member function. If you are using 56842PMFs in an inner loop, you should really reconsider that decision. If 56843that is not an option, you can extract the pointer to the function that 56844would be called for a given object/PMF pair and call it directly inside 56845the inner loop, to save a bit of time. 56846 56847 Note that you still pay the penalty for the call through a function 56848pointer; on most modern architectures, such a call defeats the branch 56849prediction features of the CPU. This is also true of normal virtual 56850function calls. 56851 56852 The syntax for this extension is 56853 56854 extern A a; 56855 extern int (A::*fp)(); 56856 typedef int (*fptr)(A *); 56857 56858 fptr p = (fptr)(a.*fp); 56859 56860 For PMF constants (i.e. expressions of the form '&Klasse::Member'), no 56861object is needed to obtain the address of the function. They can be 56862converted to function pointers directly: 56863 56864 fptr p1 = (fptr)(&A::foo); 56865 56866 You must specify '-Wno-pmf-conversions' to use this extension. 56867 56868 56869File: gcc.info, Node: C++ Attributes, Next: Function Multiversioning, Prev: Bound member functions, Up: C++ Extensions 56870 568717.7 C++-Specific Variable, Function, and Type Attributes 56872======================================================== 56873 56874Some attributes only make sense for C++ programs. 56875 56876'abi_tag ("TAG", ...)' 56877 The 'abi_tag' attribute can be applied to a function, variable, or 56878 class declaration. It modifies the mangled name of the entity to 56879 incorporate the tag name, in order to distinguish the function or 56880 class from an earlier version with a different ABI; perhaps the 56881 class has changed size, or the function has a different return type 56882 that is not encoded in the mangled name. 56883 56884 The attribute can also be applied to an inline namespace, but does 56885 not affect the mangled name of the namespace; in this case it is 56886 only used for '-Wabi-tag' warnings and automatic tagging of 56887 functions and variables. Tagging inline namespaces is generally 56888 preferable to tagging individual declarations, but the latter is 56889 sometimes necessary, such as when only certain members of a class 56890 need to be tagged. 56891 56892 The argument can be a list of strings of arbitrary length. The 56893 strings are sorted on output, so the order of the list is 56894 unimportant. 56895 56896 A redeclaration of an entity must not add new ABI tags, since doing 56897 so would change the mangled name. 56898 56899 The ABI tags apply to a name, so all instantiations and 56900 specializations of a template have the same tags. The attribute 56901 will be ignored if applied to an explicit specialization or 56902 instantiation. 56903 56904 The '-Wabi-tag' flag enables a warning about a class which does not 56905 have all the ABI tags used by its subobjects and virtual functions; 56906 for users with code that needs to coexist with an earlier ABI, 56907 using this option can help to find all affected types that need to 56908 be tagged. 56909 56910 When a type involving an ABI tag is used as the type of a variable 56911 or return type of a function where that tag is not already present 56912 in the signature of the function, the tag is automatically applied 56913 to the variable or function. '-Wabi-tag' also warns about this 56914 situation; this warning can be avoided by explicitly tagging the 56915 variable or function or moving it into a tagged inline namespace. 56916 56917'init_priority (PRIORITY)' 56918 56919 In Standard C++, objects defined at namespace scope are guaranteed 56920 to be initialized in an order in strict accordance with that of 56921 their definitions _in a given translation unit_. No guarantee is 56922 made for initializations across translation units. However, GNU 56923 C++ allows users to control the order of initialization of objects 56924 defined at namespace scope with the 'init_priority' attribute by 56925 specifying a relative PRIORITY, a constant integral expression 56926 currently bounded between 101 and 65535 inclusive. Lower numbers 56927 indicate a higher priority. 56928 56929 In the following example, 'A' would normally be created before 'B', 56930 but the 'init_priority' attribute reverses that order: 56931 56932 Some_Class A __attribute__ ((init_priority (2000))); 56933 Some_Class B __attribute__ ((init_priority (543))); 56934 56935 Note that the particular values of PRIORITY do not matter; only 56936 their relative ordering. 56937 56938'warn_unused' 56939 56940 For C++ types with non-trivial constructors and/or destructors it 56941 is impossible for the compiler to determine whether a variable of 56942 this type is truly unused if it is not referenced. This type 56943 attribute informs the compiler that variables of this type should 56944 be warned about if they appear to be unused, just like variables of 56945 fundamental types. 56946 56947 This attribute is appropriate for types which just represent a 56948 value, such as 'std::string'; it is not appropriate for types which 56949 control a resource, such as 'std::lock_guard'. 56950 56951 This attribute is also accepted in C, but it is unnecessary because 56952 C does not have constructors or destructors. 56953 56954 56955File: gcc.info, Node: Function Multiversioning, Next: Type Traits, Prev: C++ Attributes, Up: C++ Extensions 56956 569577.8 Function Multiversioning 56958============================ 56959 56960With the GNU C++ front end, for x86 targets, you may specify multiple 56961versions of a function, where each function is specialized for a 56962specific target feature. At runtime, the appropriate version of the 56963function is automatically executed depending on the characteristics of 56964the execution platform. Here is an example. 56965 56966 __attribute__ ((target ("default"))) 56967 int foo () 56968 { 56969 // The default version of foo. 56970 return 0; 56971 } 56972 56973 __attribute__ ((target ("sse4.2"))) 56974 int foo () 56975 { 56976 // foo version for SSE4.2 56977 return 1; 56978 } 56979 56980 __attribute__ ((target ("arch=atom"))) 56981 int foo () 56982 { 56983 // foo version for the Intel ATOM processor 56984 return 2; 56985 } 56986 56987 __attribute__ ((target ("arch=amdfam10"))) 56988 int foo () 56989 { 56990 // foo version for the AMD Family 0x10 processors. 56991 return 3; 56992 } 56993 56994 int main () 56995 { 56996 int (*p)() = &foo; 56997 assert ((*p) () == foo ()); 56998 return 0; 56999 } 57000 57001 In the above example, four versions of function foo are created. The 57002first version of foo with the target attribute "default" is the default 57003version. This version gets executed when no other target specific 57004version qualifies for execution on a particular platform. A new version 57005of foo is created by using the same function signature but with a 57006different target string. Function foo is called or a pointer to it is 57007taken just like a regular function. GCC takes care of doing the 57008dispatching to call the right version at runtime. Refer to the GCC wiki 57009on Function Multiversioning 57010(https://gcc.gnu.org/wiki/FunctionMultiVersioning) for more details. 57011 57012 57013File: gcc.info, Node: Type Traits, Next: C++ Concepts, Prev: Function Multiversioning, Up: C++ Extensions 57014 570157.9 Type Traits 57016=============== 57017 57018The C++ front end implements syntactic extensions that allow 57019compile-time determination of various characteristics of a type (or of a 57020pair of types). 57021 57022'__has_nothrow_assign (type)' 57023 If 'type' is 'const'-qualified or is a reference type then the 57024 trait is 'false'. Otherwise if '__has_trivial_assign (type)' is 57025 'true' then the trait is 'true', else if 'type' is a cv-qualified 57026 class or union type with copy assignment operators that are known 57027 not to throw an exception then the trait is 'true', else it is 57028 'false'. Requires: 'type' shall be a complete type, (possibly 57029 cv-qualified) 'void', or an array of unknown bound. 57030 57031'__has_nothrow_copy (type)' 57032 If '__has_trivial_copy (type)' is 'true' then the trait is 'true', 57033 else if 'type' is a cv-qualified class or union type with copy 57034 constructors that are known not to throw an exception then the 57035 trait is 'true', else it is 'false'. Requires: 'type' shall be a 57036 complete type, (possibly cv-qualified) 'void', or an array of 57037 unknown bound. 57038 57039'__has_nothrow_constructor (type)' 57040 If '__has_trivial_constructor (type)' is 'true' then the trait is 57041 'true', else if 'type' is a cv class or union type (or array 57042 thereof) with a default constructor that is known not to throw an 57043 exception then the trait is 'true', else it is 'false'. Requires: 57044 'type' shall be a complete type, (possibly cv-qualified) 'void', or 57045 an array of unknown bound. 57046 57047'__has_trivial_assign (type)' 57048 If 'type' is 'const'- qualified or is a reference type then the 57049 trait is 'false'. Otherwise if '__is_trivial (type)' is 'true' 57050 then the trait is 'true', else if 'type' is a cv-qualified class or 57051 union type with a trivial copy assignment ([class.copy]) then the 57052 trait is 'true', else it is 'false'. Requires: 'type' shall be a 57053 complete type, (possibly cv-qualified) 'void', or an array of 57054 unknown bound. 57055 57056'__has_trivial_copy (type)' 57057 If '__is_trivial (type)' is 'true' or 'type' is a reference type 57058 then the trait is 'true', else if 'type' is a cv class or union 57059 type with a trivial copy constructor ([class.copy]) then the trait 57060 is 'true', else it is 'false'. Requires: 'type' shall be a 57061 complete type, (possibly cv-qualified) 'void', or an array of 57062 unknown bound. 57063 57064'__has_trivial_constructor (type)' 57065 If '__is_trivial (type)' is 'true' then the trait is 'true', else 57066 if 'type' is a cv-qualified class or union type (or array thereof) 57067 with a trivial default constructor ([class.ctor]) then the trait is 57068 'true', else it is 'false'. Requires: 'type' shall be a complete 57069 type, (possibly cv-qualified) 'void', or an array of unknown bound. 57070 57071'__has_trivial_destructor (type)' 57072 If '__is_trivial (type)' is 'true' or 'type' is a reference type 57073 then the trait is 'true', else if 'type' is a cv class or union 57074 type (or array thereof) with a trivial destructor ([class.dtor]) 57075 then the trait is 'true', else it is 'false'. Requires: 'type' 57076 shall be a complete type, (possibly cv-qualified) 'void', or an 57077 array of unknown bound. 57078 57079'__has_virtual_destructor (type)' 57080 If 'type' is a class type with a virtual destructor ([class.dtor]) 57081 then the trait is 'true', else it is 'false'. Requires: 'type' 57082 shall be a complete type, (possibly cv-qualified) 'void', or an 57083 array of unknown bound. 57084 57085'__is_abstract (type)' 57086 If 'type' is an abstract class ([class.abstract]) then the trait is 57087 'true', else it is 'false'. Requires: 'type' shall be a complete 57088 type, (possibly cv-qualified) 'void', or an array of unknown bound. 57089 57090'__is_base_of (base_type, derived_type)' 57091 If 'base_type' is a base class of 'derived_type' ([class.derived]) 57092 then the trait is 'true', otherwise it is 'false'. Top-level 57093 cv-qualifications of 'base_type' and 'derived_type' are ignored. 57094 For the purposes of this trait, a class type is considered is own 57095 base. Requires: if '__is_class (base_type)' and '__is_class 57096 (derived_type)' are 'true' and 'base_type' and 'derived_type' are 57097 not the same type (disregarding cv-qualifiers), 'derived_type' 57098 shall be a complete type. A diagnostic is produced if this 57099 requirement is not met. 57100 57101'__is_class (type)' 57102 If 'type' is a cv-qualified class type, and not a union type 57103 ([basic.compound]) the trait is 'true', else it is 'false'. 57104 57105'__is_empty (type)' 57106 If '__is_class (type)' is 'false' then the trait is 'false'. 57107 Otherwise 'type' is considered empty if and only if: 'type' has no 57108 non-static data members, or all non-static data members, if any, 57109 are bit-fields of length 0, and 'type' has no virtual members, and 57110 'type' has no virtual base classes, and 'type' has no base classes 57111 'base_type' for which '__is_empty (base_type)' is 'false'. 57112 Requires: 'type' shall be a complete type, (possibly cv-qualified) 57113 'void', or an array of unknown bound. 57114 57115'__is_enum (type)' 57116 If 'type' is a cv enumeration type ([basic.compound]) the trait is 57117 'true', else it is 'false'. 57118 57119'__is_literal_type (type)' 57120 If 'type' is a literal type ([basic.types]) the trait is 'true', 57121 else it is 'false'. Requires: 'type' shall be a complete type, 57122 (possibly cv-qualified) 'void', or an array of unknown bound. 57123 57124'__is_pod (type)' 57125 If 'type' is a cv POD type ([basic.types]) then the trait is 57126 'true', else it is 'false'. Requires: 'type' shall be a complete 57127 type, (possibly cv-qualified) 'void', or an array of unknown bound. 57128 57129'__is_polymorphic (type)' 57130 If 'type' is a polymorphic class ([class.virtual]) then the trait 57131 is 'true', else it is 'false'. Requires: 'type' shall be a 57132 complete type, (possibly cv-qualified) 'void', or an array of 57133 unknown bound. 57134 57135'__is_standard_layout (type)' 57136 If 'type' is a standard-layout type ([basic.types]) the trait is 57137 'true', else it is 'false'. Requires: 'type' shall be a complete 57138 type, (possibly cv-qualified) 'void', or an array of unknown bound. 57139 57140'__is_trivial (type)' 57141 If 'type' is a trivial type ([basic.types]) the trait is 'true', 57142 else it is 'false'. Requires: 'type' shall be a complete type, 57143 (possibly cv-qualified) 'void', or an array of unknown bound. 57144 57145'__is_union (type)' 57146 If 'type' is a cv union type ([basic.compound]) the trait is 57147 'true', else it is 'false'. 57148 57149'__underlying_type (type)' 57150 The underlying type of 'type'. Requires: 'type' shall be an 57151 enumeration type ([dcl.enum]). 57152 57153'__integer_pack (length)' 57154 When used as the pattern of a pack expansion within a template 57155 definition, expands to a template argument pack containing integers 57156 from '0' to 'length-1'. This is provided for efficient 57157 implementation of 'std::make_integer_sequence'. 57158 57159 57160File: gcc.info, Node: C++ Concepts, Next: Deprecated Features, Prev: Type Traits, Up: C++ Extensions 57161 571627.10 C++ Concepts 57163================= 57164 57165C++ concepts provide much-improved support for generic programming. In 57166particular, they allow the specification of constraints on template 57167arguments. The constraints are used to extend the usual overloading and 57168partial specialization capabilities of the language, allowing generic 57169data structures and algorithms to be "refined" based on their properties 57170rather than their type names. 57171 57172 The following keywords are reserved for concepts. 57173 57174'assumes' 57175 States an expression as an assumption, and if possible, verifies 57176 that the assumption is valid. For example, 'assume(n > 0)'. 57177 57178'axiom' 57179 Introduces an axiom definition. Axioms introduce requirements on 57180 values. 57181 57182'forall' 57183 Introduces a universally quantified object in an axiom. For 57184 example, 'forall (int n) n + 0 == n'). 57185 57186'concept' 57187 Introduces a concept definition. Concepts are sets of syntactic 57188 and semantic requirements on types and their values. 57189 57190'requires' 57191 Introduces constraints on template arguments or requirements for a 57192 member function of a class template. 57193 57194 The front end also exposes a number of internal mechanism that can be 57195used to simplify the writing of type traits. Note that some of these 57196traits are likely to be removed in the future. 57197 57198'__is_same (type1, type2)' 57199 A binary type trait: 'true' whenever the type arguments are the 57200 same. 57201 57202 57203File: gcc.info, Node: Deprecated Features, Next: Backwards Compatibility, Prev: C++ Concepts, Up: C++ Extensions 57204 572057.11 Deprecated Features 57206======================== 57207 57208In the past, the GNU C++ compiler was extended to experiment with new 57209features, at a time when the C++ language was still evolving. Now that 57210the C++ standard is complete, some of those features are superseded by 57211superior alternatives. Using the old features might cause a warning in 57212some cases that the feature will be dropped in the future. In other 57213cases, the feature might be gone already. 57214 57215 G++ allows a virtual function returning 'void *' to be overridden by 57216one returning a different pointer type. This extension to the covariant 57217return type rules is now deprecated and will be removed from a future 57218version. 57219 57220 The use of default arguments in function pointers, function typedefs 57221and other places where they are not permitted by the standard is 57222deprecated and will be removed from a future version of G++. 57223 57224 G++ allows floating-point literals to appear in integral constant 57225expressions, e.g. ' enum E { e = int(2.2 * 3.7) } ' This extension is 57226deprecated and will be removed from a future version. 57227 57228 G++ allows static data members of const floating-point type to be 57229declared with an initializer in a class definition. The standard only 57230allows initializers for static members of const integral types and const 57231enumeration types so this extension has been deprecated and will be 57232removed from a future version. 57233 57234 G++ allows attributes to follow a parenthesized direct initializer, 57235e.g. ' int f (0) __attribute__ ((something)); ' This extension has been 57236ignored since G++ 3.3 and is deprecated. 57237 57238 G++ allows anonymous structs and unions to have members that are not 57239public non-static data members (i.e. fields). These extensions are 57240deprecated. 57241 57242 57243File: gcc.info, Node: Backwards Compatibility, Prev: Deprecated Features, Up: C++ Extensions 57244 572457.12 Backwards Compatibility 57246============================ 57247 57248Now that there is a definitive ISO standard C++, G++ has a specification 57249to adhere to. The C++ language evolved over time, and features that 57250used to be acceptable in previous drafts of the standard, such as the 57251ARM [Annotated C++ Reference Manual], are no longer accepted. In order 57252to allow compilation of C++ written to such drafts, G++ contains some 57253backwards compatibilities. _All such backwards compatibility features 57254are liable to disappear in future versions of G++._ They should be 57255considered deprecated. *Note Deprecated Features::. 57256 57257'Implicit C language' 57258 Old C system header files did not contain an 'extern "C" {...}' 57259 scope to set the language. On such systems, all system header 57260 files are implicitly scoped inside a C language scope. Such 57261 headers must correctly prototype function argument types, there is 57262 no leeway for '()' to indicate an unspecified set of arguments. 57263 57264 57265File: gcc.info, Node: Objective-C, Next: Compatibility, Prev: C++ Extensions, Up: Top 57266 572678 GNU Objective-C Features 57268************************** 57269 57270This document is meant to describe some of the GNU Objective-C features. 57271It is not intended to teach you Objective-C. There are several resources 57272on the Internet that present the language. 57273 57274* Menu: 57275 57276* GNU Objective-C runtime API:: 57277* Executing code before main:: 57278* Type encoding:: 57279* Garbage Collection:: 57280* Constant string objects:: 57281* compatibility_alias:: 57282* Exceptions:: 57283* Synchronization:: 57284* Fast enumeration:: 57285* Messaging with the GNU Objective-C runtime:: 57286 57287 57288File: gcc.info, Node: GNU Objective-C runtime API, Next: Executing code before main, Up: Objective-C 57289 572908.1 GNU Objective-C Runtime API 57291=============================== 57292 57293This section is specific for the GNU Objective-C runtime. If you are 57294using a different runtime, you can skip it. 57295 57296 The GNU Objective-C runtime provides an API that allows you to interact 57297with the Objective-C runtime system, querying the live runtime 57298structures and even manipulating them. This allows you for example to 57299inspect and navigate classes, methods and protocols; to define new 57300classes or new methods, and even to modify existing classes or 57301protocols. 57302 57303 If you are using a "Foundation" library such as GNUstep-Base, this 57304library will provide you with a rich set of functionality to do most of 57305the inspection tasks, and you probably will only need direct access to 57306the GNU Objective-C runtime API to define new classes or methods. 57307 57308* Menu: 57309 57310* Modern GNU Objective-C runtime API:: 57311* Traditional GNU Objective-C runtime API:: 57312 57313 57314File: gcc.info, Node: Modern GNU Objective-C runtime API, Next: Traditional GNU Objective-C runtime API, Up: GNU Objective-C runtime API 57315 573168.1.1 Modern GNU Objective-C Runtime API 57317---------------------------------------- 57318 57319The GNU Objective-C runtime provides an API which is similar to the one 57320provided by the "Objective-C 2.0" Apple/NeXT Objective-C runtime. The 57321API is documented in the public header files of the GNU Objective-C 57322runtime: 57323 57324 * 'objc/objc.h': this is the basic Objective-C header file, defining 57325 the basic Objective-C types such as 'id', 'Class' and 'BOOL'. You 57326 have to include this header to do almost anything with Objective-C. 57327 57328 * 'objc/runtime.h': this header declares most of the public runtime 57329 API functions allowing you to inspect and manipulate the 57330 Objective-C runtime data structures. These functions are fairly 57331 standardized across Objective-C runtimes and are almost identical 57332 to the Apple/NeXT Objective-C runtime ones. It does not declare 57333 functions in some specialized areas (constructing and forwarding 57334 message invocations, threading) which are in the other headers 57335 below. You have to include 'objc/objc.h' and 'objc/runtime.h' to 57336 use any of the functions, such as 'class_getName()', declared in 57337 'objc/runtime.h'. 57338 57339 * 'objc/message.h': this header declares public functions used to 57340 construct, deconstruct and forward message invocations. Because 57341 messaging is done in quite a different way on different runtimes, 57342 functions in this header are specific to the GNU Objective-C 57343 runtime implementation. 57344 57345 * 'objc/objc-exception.h': this header declares some public functions 57346 related to Objective-C exceptions. For example functions in this 57347 header allow you to throw an Objective-C exception from plain C/C++ 57348 code. 57349 57350 * 'objc/objc-sync.h': this header declares some public functions 57351 related to the Objective-C '@synchronized()' syntax, allowing you 57352 to emulate an Objective-C '@synchronized()' block in plain C/C++ 57353 code. 57354 57355 * 'objc/thr.h': this header declares a public runtime API threading 57356 layer that is only provided by the GNU Objective-C runtime. It 57357 declares functions such as 'objc_mutex_lock()', which provide a 57358 platform-independent set of threading functions. 57359 57360 The header files contain detailed documentation for each function in 57361the GNU Objective-C runtime API. 57362 57363 57364File: gcc.info, Node: Traditional GNU Objective-C runtime API, Prev: Modern GNU Objective-C runtime API, Up: GNU Objective-C runtime API 57365 573668.1.2 Traditional GNU Objective-C Runtime API 57367--------------------------------------------- 57368 57369The GNU Objective-C runtime used to provide a different API, which we 57370call the "traditional" GNU Objective-C runtime API. Functions belonging 57371to this API are easy to recognize because they use a different naming 57372convention, such as 'class_get_super_class()' (traditional API) instead 57373of 'class_getSuperclass()' (modern API). Software using this API 57374includes the file 'objc/objc-api.h' where it is declared. 57375 57376 Starting with GCC 4.7.0, the traditional GNU runtime API is no longer 57377available. 57378 57379 57380File: gcc.info, Node: Executing code before main, Next: Type encoding, Prev: GNU Objective-C runtime API, Up: Objective-C 57381 573828.2 '+load': Executing Code before 'main' 57383========================================= 57384 57385This section is specific for the GNU Objective-C runtime. If you are 57386using a different runtime, you can skip it. 57387 57388 The GNU Objective-C runtime provides a way that allows you to execute 57389code before the execution of the program enters the 'main' function. 57390The code is executed on a per-class and a per-category basis, through a 57391special class method '+load'. 57392 57393 This facility is very useful if you want to initialize global variables 57394which can be accessed by the program directly, without sending a message 57395to the class first. The usual way to initialize global variables, in 57396the '+initialize' method, might not be useful because '+initialize' is 57397only called when the first message is sent to a class object, which in 57398some cases could be too late. 57399 57400 Suppose for example you have a 'FileStream' class that declares 57401'Stdin', 'Stdout' and 'Stderr' as global variables, like below: 57402 57403 57404 FileStream *Stdin = nil; 57405 FileStream *Stdout = nil; 57406 FileStream *Stderr = nil; 57407 57408 @implementation FileStream 57409 57410 + (void)initialize 57411 { 57412 Stdin = [[FileStream new] initWithFd:0]; 57413 Stdout = [[FileStream new] initWithFd:1]; 57414 Stderr = [[FileStream new] initWithFd:2]; 57415 } 57416 57417 /* Other methods here */ 57418 @end 57419 57420 57421 In this example, the initialization of 'Stdin', 'Stdout' and 'Stderr' 57422in '+initialize' occurs too late. The programmer can send a message to 57423one of these objects before the variables are actually initialized, thus 57424sending messages to the 'nil' object. The '+initialize' method which 57425actually initializes the global variables is not invoked until the first 57426message is sent to the class object. The solution would require these 57427variables to be initialized just before entering 'main'. 57428 57429 The correct solution of the above problem is to use the '+load' method 57430instead of '+initialize': 57431 57432 57433 @implementation FileStream 57434 57435 + (void)load 57436 { 57437 Stdin = [[FileStream new] initWithFd:0]; 57438 Stdout = [[FileStream new] initWithFd:1]; 57439 Stderr = [[FileStream new] initWithFd:2]; 57440 } 57441 57442 /* Other methods here */ 57443 @end 57444 57445 57446 The '+load' is a method that is not overridden by categories. If a 57447class and a category of it both implement '+load', both methods are 57448invoked. This allows some additional initializations to be performed in 57449a category. 57450 57451 This mechanism is not intended to be a replacement for '+initialize'. 57452You should be aware of its limitations when you decide to use it instead 57453of '+initialize'. 57454 57455* Menu: 57456 57457* What you can and what you cannot do in +load:: 57458 57459 57460File: gcc.info, Node: What you can and what you cannot do in +load, Up: Executing code before main 57461 574628.2.1 What You Can and Cannot Do in '+load' 57463------------------------------------------- 57464 57465'+load' is to be used only as a last resort. Because it is executed 57466very early, most of the Objective-C runtime machinery will not be ready 57467when '+load' is executed; hence '+load' works best for executing C code 57468that is independent on the Objective-C runtime. 57469 57470 The '+load' implementation in the GNU runtime guarantees you the 57471following things: 57472 57473 * you can write whatever C code you like; 57474 57475 * you can allocate and send messages to objects whose class is 57476 implemented in the same file; 57477 57478 * the '+load' implementation of all super classes of a class are 57479 executed before the '+load' of that class is executed; 57480 57481 * the '+load' implementation of a class is executed before the 57482 '+load' implementation of any category. 57483 57484 In particular, the following things, even if they can work in a 57485particular case, are not guaranteed: 57486 57487 * allocation of or sending messages to arbitrary objects; 57488 57489 * allocation of or sending messages to objects whose classes have a 57490 category implemented in the same file; 57491 57492 * sending messages to Objective-C constant strings ('@"this is a 57493 constant string"'); 57494 57495 You should make no assumptions about receiving '+load' in sibling 57496classes when you write '+load' of a class. The order in which sibling 57497classes receive '+load' is not guaranteed. 57498 57499 The order in which '+load' and '+initialize' are called could be 57500problematic if this matters. If you don't allocate objects inside 57501'+load', it is guaranteed that '+load' is called before '+initialize'. 57502If you create an object inside '+load' the '+initialize' method of 57503object's class is invoked even if '+load' was not invoked. Note if you 57504explicitly call '+load' on a class, '+initialize' will be called first. 57505To avoid possible problems try to implement only one of these methods. 57506 57507 The '+load' method is also invoked when a bundle is dynamically loaded 57508into your running program. This happens automatically without any 57509intervening operation from you. When you write bundles and you need to 57510write '+load' you can safely create and send messages to objects whose 57511classes already exist in the running program. The same restrictions as 57512above apply to classes defined in bundle. 57513 57514 57515File: gcc.info, Node: Type encoding, Next: Garbage Collection, Prev: Executing code before main, Up: Objective-C 57516 575178.3 Type Encoding 57518================= 57519 57520This is an advanced section. Type encodings are used extensively by the 57521compiler and by the runtime, but you generally do not need to know about 57522them to use Objective-C. 57523 57524 The Objective-C compiler generates type encodings for all the types. 57525These type encodings are used at runtime to find out information about 57526selectors and methods and about objects and classes. 57527 57528 The types are encoded in the following way: 57529 57530'_Bool' 'B' 57531'char' 'c' 57532'unsigned char' 'C' 57533'short' 's' 57534'unsigned short' 'S' 57535'int' 'i' 57536'unsigned int' 'I' 57537'long' 'l' 57538'unsigned long' 'L' 57539'long long' 'q' 57540'unsigned long 'Q' 57541long' 57542'float' 'f' 57543'double' 'd' 57544'long double' 'D' 57545'void' 'v' 57546'id' '@' 57547'Class' '#' 57548'SEL' ':' 57549'char*' '*' 57550'enum' an 'enum' is encoded exactly as the integer type 57551 that the compiler uses for it, which depends on the 57552 enumeration values. Often the compiler users 57553 'unsigned int', which is then encoded as 'I'. 57554unknown type '?' 57555Complex types 'j' followed by the inner type. For example 57556 '_Complex double' is encoded as "jd". 57557bit-fields 'b' followed by the starting position of the 57558 bit-field, the type of the bit-field and the size of 57559 the bit-field (the bit-fields encoding was changed 57560 from the NeXT's compiler encoding, see below) 57561 57562 The encoding of bit-fields has changed to allow bit-fields to be 57563properly handled by the runtime functions that compute sizes and 57564alignments of types that contain bit-fields. The previous encoding 57565contained only the size of the bit-field. Using only this information 57566it is not possible to reliably compute the size occupied by the 57567bit-field. This is very important in the presence of the Boehm's 57568garbage collector because the objects are allocated using the typed 57569memory facility available in this collector. The typed memory 57570allocation requires information about where the pointers are located 57571inside the object. 57572 57573 The position in the bit-field is the position, counting in bits, of the 57574bit closest to the beginning of the structure. 57575 57576 The non-atomic types are encoded as follows: 57577 57578pointers '^' followed by the pointed type. 57579arrays '[' followed by the number of elements in the array 57580 followed by the type of the elements followed by ']' 57581structures '{' followed by the name of the structure (or '?' if the 57582 structure is unnamed), the '=' sign, the type of the 57583 members and by '}' 57584unions '(' followed by the name of the structure (or '?' if the 57585 union is unnamed), the '=' sign, the type of the members 57586 followed by ')' 57587vectors '![' followed by the vector_size (the number of bytes 57588 composing the vector) followed by a comma, followed by 57589 the alignment (in bytes) of the vector, followed by the 57590 type of the elements followed by ']' 57591 57592 Here are some types and their encodings, as they are generated by the 57593compiler on an i386 machine: 57594 57595 57596Objective-C type Compiler encoding 57597-------------------------------------------------------------------------- 57598 int a[10]; '[10i]' 57599 struct { '{?=i[3f]b128i3b131i2c}' 57600 int i; 57601 float f[3]; 57602 int a:3; 57603 int b:2; 57604 char c; 57605 } 57606 int a __attribute__ ((vector_size (16)));'![16,16i]' (alignment 57607 depends on the machine) 57608 57609 57610 In addition to the types the compiler also encodes the type specifiers. 57611The table below describes the encoding of the current Objective-C type 57612specifiers: 57613 57614 57615Specifier Encoding 57616-------------------------------------------------------------------------- 57617'const' 'r' 57618'in' 'n' 57619'inout' 'N' 57620'out' 'o' 57621'bycopy' 'O' 57622'byref' 'R' 57623'oneway' 'V' 57624 57625 57626 The type specifiers are encoded just before the type. Unlike types 57627however, the type specifiers are only encoded when they appear in method 57628argument types. 57629 57630 Note how 'const' interacts with pointers: 57631 57632 57633Objective-C type Compiler encoding 57634-------------------------------------------------------------------------- 57635 const int 'ri' 57636 const int* '^ri' 57637 int *const 'r^i' 57638 57639 57640 'const int*' is a pointer to a 'const int', and so is encoded as '^ri'. 57641'int* const', instead, is a 'const' pointer to an 'int', and so is 57642encoded as 'r^i'. 57643 57644 Finally, there is a complication when encoding 'const char *' versus 57645'char * const'. Because 'char *' is encoded as '*' and not as '^c', 57646there is no way to express the fact that 'r' applies to the pointer or 57647to the pointee. 57648 57649 Hence, it is assumed as a convention that 'r*' means 'const char *' 57650(since it is what is most often meant), and there is no way to encode 57651'char *const'. 'char *const' would simply be encoded as '*', and the 57652'const' is lost. 57653 57654* Menu: 57655 57656* Legacy type encoding:: 57657* @encode:: 57658* Method signatures:: 57659 57660 57661File: gcc.info, Node: Legacy type encoding, Next: @encode, Up: Type encoding 57662 576638.3.1 Legacy Type Encoding 57664-------------------------- 57665 57666Unfortunately, historically GCC used to have a number of bugs in its 57667encoding code. The NeXT runtime expects GCC to emit type encodings in 57668this historical format (compatible with GCC-3.3), so when using the NeXT 57669runtime, GCC will introduce on purpose a number of incorrect encodings: 57670 57671 * the read-only qualifier of the pointee gets emitted before the '^'. 57672 The read-only qualifier of the pointer itself gets ignored, unless 57673 it is a typedef. Also, the 'r' is only emitted for the outermost 57674 type. 57675 57676 * 32-bit longs are encoded as 'l' or 'L', but not always. For 57677 typedefs, the compiler uses 'i' or 'I' instead if encoding a struct 57678 field or a pointer. 57679 57680 * 'enum's are always encoded as 'i' (int) even if they are actually 57681 unsigned or long. 57682 57683 In addition to that, the NeXT runtime uses a different encoding for 57684bitfields. It encodes them as 'b' followed by the size, without a bit 57685offset or the underlying field type. 57686 57687 57688File: gcc.info, Node: @encode, Next: Method signatures, Prev: Legacy type encoding, Up: Type encoding 57689 576908.3.2 '@encode' 57691--------------- 57692 57693GNU Objective-C supports the '@encode' syntax that allows you to create 57694a type encoding from a C/Objective-C type. For example, '@encode(int)' 57695is compiled by the compiler into '"i"'. 57696 57697 '@encode' does not support type qualifiers other than 'const'. For 57698example, '@encode(const char*)' is valid and is compiled into '"r*"', 57699while '@encode(bycopy char *)' is invalid and will cause a compilation 57700error. 57701 57702 57703File: gcc.info, Node: Method signatures, Prev: @encode, Up: Type encoding 57704 577058.3.3 Method Signatures 57706----------------------- 57707 57708This section documents the encoding of method types, which is rarely 57709needed to use Objective-C. You should skip it at a first reading; the 57710runtime provides functions that will work on methods and can walk 57711through the list of parameters and interpret them for you. These 57712functions are part of the public "API" and are the preferred way to 57713interact with method signatures from user code. 57714 57715 But if you need to debug a problem with method signatures and need to 57716know how they are implemented (i.e., the "ABI"), read on. 57717 57718 Methods have their "signature" encoded and made available to the 57719runtime. The "signature" encodes all the information required to 57720dynamically build invocations of the method at runtime: return type and 57721arguments. 57722 57723 The "signature" is a null-terminated string, composed of the following: 57724 57725 * The return type, including type qualifiers. For example, a method 57726 returning 'int' would have 'i' here. 57727 57728 * The total size (in bytes) required to pass all the parameters. 57729 This includes the two hidden parameters (the object 'self' and the 57730 method selector '_cmd'). 57731 57732 * Each argument, with the type encoding, followed by the offset (in 57733 bytes) of the argument in the list of parameters. 57734 57735 For example, a method with no arguments and returning 'int' would have 57736the signature 'i8@0:4' if the size of a pointer is 4. The signature is 57737interpreted as follows: the 'i' is the return type (an 'int'), the '8' 57738is the total size of the parameters in bytes (two pointers each of size 577394), the '@0' is the first parameter (an object at byte offset '0') and 57740':4' is the second parameter (a 'SEL' at byte offset '4'). 57741 57742 You can easily find more examples by running the "strings" program on 57743an Objective-C object file compiled by GCC. You'll see a lot of strings 57744that look very much like 'i8@0:4'. They are signatures of Objective-C 57745methods. 57746 57747 57748File: gcc.info, Node: Garbage Collection, Next: Constant string objects, Prev: Type encoding, Up: Objective-C 57749 577508.4 Garbage Collection 57751====================== 57752 57753This section is specific for the GNU Objective-C runtime. If you are 57754using a different runtime, you can skip it. 57755 57756 Support for garbage collection with the GNU runtime has been added by 57757using a powerful conservative garbage collector, known as the 57758Boehm-Demers-Weiser conservative garbage collector. 57759 57760 To enable the support for it you have to configure the compiler using 57761an additional argument, '--enable-objc-gc'. This will build the 57762boehm-gc library, and build an additional runtime library which has 57763several enhancements to support the garbage collector. The new library 57764has a new name, 'libobjc_gc.a' to not conflict with the 57765non-garbage-collected library. 57766 57767 When the garbage collector is used, the objects are allocated using the 57768so-called typed memory allocation mechanism available in the 57769Boehm-Demers-Weiser collector. This mode requires precise information 57770on where pointers are located inside objects. This information is 57771computed once per class, immediately after the class has been 57772initialized. 57773 57774 There is a new runtime function 'class_ivar_set_gcinvisible()' which 57775can be used to declare a so-called "weak pointer" reference. Such a 57776pointer is basically hidden for the garbage collector; this can be 57777useful in certain situations, especially when you want to keep track of 57778the allocated objects, yet allow them to be collected. This kind of 57779pointers can only be members of objects, you cannot declare a global 57780pointer as a weak reference. Every type which is a pointer type can be 57781declared a weak pointer, including 'id', 'Class' and 'SEL'. 57782 57783 Here is an example of how to use this feature. Suppose you want to 57784implement a class whose instances hold a weak pointer reference; the 57785following class does this: 57786 57787 57788 @interface WeakPointer : Object 57789 { 57790 const void* weakPointer; 57791 } 57792 57793 - initWithPointer:(const void*)p; 57794 - (const void*)weakPointer; 57795 @end 57796 57797 57798 @implementation WeakPointer 57799 57800 + (void)initialize 57801 { 57802 if (self == objc_lookUpClass ("WeakPointer")) 57803 class_ivar_set_gcinvisible (self, "weakPointer", YES); 57804 } 57805 57806 - initWithPointer:(const void*)p 57807 { 57808 weakPointer = p; 57809 return self; 57810 } 57811 57812 - (const void*)weakPointer 57813 { 57814 return weakPointer; 57815 } 57816 57817 @end 57818 57819 57820 Weak pointers are supported through a new type character specifier 57821represented by the '!' character. The 'class_ivar_set_gcinvisible()' 57822function adds or removes this specifier to the string type description 57823of the instance variable named as argument. 57824 57825 57826File: gcc.info, Node: Constant string objects, Next: compatibility_alias, Prev: Garbage Collection, Up: Objective-C 57827 578288.5 Constant String Objects 57829=========================== 57830 57831GNU Objective-C provides constant string objects that are generated 57832directly by the compiler. You declare a constant string object by 57833prefixing a C constant string with the character '@': 57834 57835 id myString = @"this is a constant string object"; 57836 57837 The constant string objects are by default instances of the 57838'NXConstantString' class which is provided by the GNU Objective-C 57839runtime. To get the definition of this class you must include the 57840'objc/NXConstStr.h' header file. 57841 57842 User defined libraries may want to implement their own constant string 57843class. To be able to support them, the GNU Objective-C compiler 57844provides a new command line options 57845'-fconstant-string-class=CLASS-NAME'. The provided class should adhere 57846to a strict structure, the same as 'NXConstantString''s structure: 57847 57848 57849 @interface MyConstantStringClass 57850 { 57851 Class isa; 57852 char *c_string; 57853 unsigned int len; 57854 } 57855 @end 57856 57857 57858 'NXConstantString' inherits from 'Object'; user class libraries may 57859choose to inherit the customized constant string class from a different 57860class than 'Object'. There is no requirement in the methods the 57861constant string class has to implement, but the final ivar layout of the 57862class must be the compatible with the given structure. 57863 57864 When the compiler creates the statically allocated constant string 57865object, the 'c_string' field will be filled by the compiler with the 57866string; the 'length' field will be filled by the compiler with the 57867string length; the 'isa' pointer will be filled with 'NULL' by the 57868compiler, and it will later be fixed up automatically at runtime by the 57869GNU Objective-C runtime library to point to the class which was set by 57870the '-fconstant-string-class' option when the object file is loaded (if 57871you wonder how it works behind the scenes, the name of the class to use, 57872and the list of static objects to fixup, are stored by the compiler in 57873the object file in a place where the GNU runtime library will find them 57874at runtime). 57875 57876 As a result, when a file is compiled with the '-fconstant-string-class' 57877option, all the constant string objects will be instances of the class 57878specified as argument to this option. It is possible to have multiple 57879compilation units referring to different constant string classes, 57880neither the compiler nor the linker impose any restrictions in doing 57881this. 57882 57883 57884File: gcc.info, Node: compatibility_alias, Next: Exceptions, Prev: Constant string objects, Up: Objective-C 57885 578868.6 'compatibility_alias' 57887========================= 57888 57889The keyword '@compatibility_alias' allows you to define a class name as 57890equivalent to another class name. For example: 57891 57892 @compatibility_alias WOApplication GSWApplication; 57893 57894 tells the compiler that each time it encounters 'WOApplication' as a 57895class name, it should replace it with 'GSWApplication' (that is, 57896'WOApplication' is just an alias for 'GSWApplication'). 57897 57898 There are some constraints on how this can be used-- 57899 57900 * 'WOApplication' (the alias) must not be an existing class; 57901 57902 * 'GSWApplication' (the real class) must be an existing class. 57903 57904 57905File: gcc.info, Node: Exceptions, Next: Synchronization, Prev: compatibility_alias, Up: Objective-C 57906 579078.7 Exceptions 57908============== 57909 57910GNU Objective-C provides exception support built into the language, as 57911in the following example: 57912 57913 @try { 57914 ... 57915 @throw expr; 57916 ... 57917 } 57918 @catch (AnObjCClass *exc) { 57919 ... 57920 @throw expr; 57921 ... 57922 @throw; 57923 ... 57924 } 57925 @catch (AnotherClass *exc) { 57926 ... 57927 } 57928 @catch (id allOthers) { 57929 ... 57930 } 57931 @finally { 57932 ... 57933 @throw expr; 57934 ... 57935 } 57936 57937 The '@throw' statement may appear anywhere in an Objective-C or 57938Objective-C++ program; when used inside of a '@catch' block, the 57939'@throw' may appear without an argument (as shown above), in which case 57940the object caught by the '@catch' will be rethrown. 57941 57942 Note that only (pointers to) Objective-C objects may be thrown and 57943caught using this scheme. When an object is thrown, it will be caught 57944by the nearest '@catch' clause capable of handling objects of that type, 57945analogously to how 'catch' blocks work in C++ and Java. A '@catch(id 57946...)' clause (as shown above) may also be provided to catch any and all 57947Objective-C exceptions not caught by previous '@catch' clauses (if any). 57948 57949 The '@finally' clause, if present, will be executed upon exit from the 57950immediately preceding '@try ... @catch' section. This will happen 57951regardless of whether any exceptions are thrown, caught or rethrown 57952inside the '@try ... @catch' section, analogously to the behavior of the 57953'finally' clause in Java. 57954 57955 There are several caveats to using the new exception mechanism: 57956 57957 * The '-fobjc-exceptions' command line option must be used when 57958 compiling Objective-C files that use exceptions. 57959 57960 * With the GNU runtime, exceptions are always implemented as "native" 57961 exceptions and it is recommended that the '-fexceptions' and 57962 '-shared-libgcc' options are used when linking. 57963 57964 * With the NeXT runtime, although currently designed to be binary 57965 compatible with 'NS_HANDLER'-style idioms provided by the 57966 'NSException' class, the new exceptions can only be used on Mac OS 57967 X 10.3 (Panther) and later systems, due to additional functionality 57968 needed in the NeXT Objective-C runtime. 57969 57970 * As mentioned above, the new exceptions do not support handling 57971 types other than Objective-C objects. Furthermore, when used from 57972 Objective-C++, the Objective-C exception model does not 57973 interoperate with C++ exceptions at this time. This means you 57974 cannot '@throw' an exception from Objective-C and 'catch' it in 57975 C++, or vice versa (i.e., 'throw ... @catch'). 57976 57977 57978File: gcc.info, Node: Synchronization, Next: Fast enumeration, Prev: Exceptions, Up: Objective-C 57979 579808.8 Synchronization 57981=================== 57982 57983GNU Objective-C provides support for synchronized blocks: 57984 57985 @synchronized (ObjCClass *guard) { 57986 ... 57987 } 57988 57989 Upon entering the '@synchronized' block, a thread of execution shall 57990first check whether a lock has been placed on the corresponding 'guard' 57991object by another thread. If it has, the current thread shall wait 57992until the other thread relinquishes its lock. Once 'guard' becomes 57993available, the current thread will place its own lock on it, execute the 57994code contained in the '@synchronized' block, and finally relinquish the 57995lock (thereby making 'guard' available to other threads). 57996 57997 Unlike Java, Objective-C does not allow for entire methods to be marked 57998'@synchronized'. Note that throwing exceptions out of '@synchronized' 57999blocks is allowed, and will cause the guarding object to be unlocked 58000properly. 58001 58002 Because of the interactions between synchronization and exception 58003handling, you can only use '@synchronized' when compiling with 58004exceptions enabled, that is with the command line option 58005'-fobjc-exceptions'. 58006 58007 58008File: gcc.info, Node: Fast enumeration, Next: Messaging with the GNU Objective-C runtime, Prev: Synchronization, Up: Objective-C 58009 580108.9 Fast Enumeration 58011==================== 58012 58013* Menu: 58014 58015* Using fast enumeration:: 58016* c99-like fast enumeration syntax:: 58017* Fast enumeration details:: 58018* Fast enumeration protocol:: 58019 58020 58021File: gcc.info, Node: Using fast enumeration, Next: c99-like fast enumeration syntax, Up: Fast enumeration 58022 580238.9.1 Using Fast Enumeration 58024---------------------------- 58025 58026GNU Objective-C provides support for the fast enumeration syntax: 58027 58028 id array = ...; 58029 id object; 58030 58031 for (object in array) 58032 { 58033 /* Do something with 'object' */ 58034 } 58035 58036 'array' needs to be an Objective-C object (usually a collection object, 58037for example an array, a dictionary or a set) which implements the "Fast 58038Enumeration Protocol" (see below). If you are using a Foundation 58039library such as GNUstep Base or Apple Cocoa Foundation, all collection 58040objects in the library implement this protocol and can be used in this 58041way. 58042 58043 The code above would iterate over all objects in 'array'. For each of 58044them, it assigns it to 'object', then executes the 'Do something with 58045'object'' statements. 58046 58047 Here is a fully worked-out example using a Foundation library (which 58048provides the implementation of 'NSArray', 'NSString' and 'NSLog'): 58049 58050 NSArray *array = [NSArray arrayWithObjects: @"1", @"2", @"3", nil]; 58051 NSString *object; 58052 58053 for (object in array) 58054 NSLog (@"Iterating over %@", object); 58055 58056 58057File: gcc.info, Node: c99-like fast enumeration syntax, Next: Fast enumeration details, Prev: Using fast enumeration, Up: Fast enumeration 58058 580598.9.2 C99-Like Fast Enumeration Syntax 58060-------------------------------------- 58061 58062A c99-like declaration syntax is also allowed: 58063 58064 id array = ...; 58065 58066 for (id object in array) 58067 { 58068 /* Do something with 'object' */ 58069 } 58070 58071 this is completely equivalent to: 58072 58073 id array = ...; 58074 58075 { 58076 id object; 58077 for (object in array) 58078 { 58079 /* Do something with 'object' */ 58080 } 58081 } 58082 58083 but can save some typing. 58084 58085 Note that the option '-std=c99' is not required to allow this syntax in 58086Objective-C. 58087 58088 58089File: gcc.info, Node: Fast enumeration details, Next: Fast enumeration protocol, Prev: c99-like fast enumeration syntax, Up: Fast enumeration 58090 580918.9.3 Fast Enumeration Details 58092------------------------------ 58093 58094Here is a more technical description with the gory details. Consider 58095the code 58096 58097 for (OBJECT EXPRESSION in COLLECTION EXPRESSION) 58098 { 58099 STATEMENTS 58100 } 58101 58102 here is what happens when you run it: 58103 58104 * 'COLLECTION EXPRESSION' is evaluated exactly once and the result is 58105 used as the collection object to iterate over. This means it is 58106 safe to write code such as 'for (object in [NSDictionary 58107 keyEnumerator]) ...'. 58108 58109 * the iteration is implemented by the compiler by repeatedly getting 58110 batches of objects from the collection object using the fast 58111 enumeration protocol (see below), then iterating over all objects 58112 in the batch. This is faster than a normal enumeration where 58113 objects are retrieved one by one (hence the name "fast 58114 enumeration"). 58115 58116 * if there are no objects in the collection, then 'OBJECT EXPRESSION' 58117 is set to 'nil' and the loop immediately terminates. 58118 58119 * if there are objects in the collection, then for each object in the 58120 collection (in the order they are returned) 'OBJECT EXPRESSION' is 58121 set to the object, then 'STATEMENTS' are executed. 58122 58123 * 'STATEMENTS' can contain 'break' and 'continue' commands, which 58124 will abort the iteration or skip to the next loop iteration as 58125 expected. 58126 58127 * when the iteration ends because there are no more objects to 58128 iterate over, 'OBJECT EXPRESSION' is set to 'nil'. This allows you 58129 to determine whether the iteration finished because a 'break' 58130 command was used (in which case 'OBJECT EXPRESSION' will remain set 58131 to the last object that was iterated over) or because it iterated 58132 over all the objects (in which case 'OBJECT EXPRESSION' will be set 58133 to 'nil'). 58134 58135 * 'STATEMENTS' must not make any changes to the collection object; if 58136 they do, it is a hard error and the fast enumeration terminates by 58137 invoking 'objc_enumerationMutation', a runtime function that 58138 normally aborts the program but which can be customized by 58139 Foundation libraries via 'objc_set_mutation_handler' to do 58140 something different, such as raising an exception. 58141 58142 58143File: gcc.info, Node: Fast enumeration protocol, Prev: Fast enumeration details, Up: Fast enumeration 58144 581458.9.4 Fast Enumeration Protocol 58146------------------------------- 58147 58148If you want your own collection object to be usable with fast 58149enumeration, you need to have it implement the method 58150 58151 - (unsigned long) countByEnumeratingWithState: (NSFastEnumerationState *)state 58152 objects: (id *)objects 58153 count: (unsigned long)len; 58154 58155 where 'NSFastEnumerationState' must be defined in your code as follows: 58156 58157 typedef struct 58158 { 58159 unsigned long state; 58160 id *itemsPtr; 58161 unsigned long *mutationsPtr; 58162 unsigned long extra[5]; 58163 } NSFastEnumerationState; 58164 58165 If no 'NSFastEnumerationState' is defined in your code, the compiler 58166will automatically replace 'NSFastEnumerationState *' with 'struct 58167__objcFastEnumerationState *', where that type is silently defined by 58168the compiler in an identical way. This can be confusing and we 58169recommend that you define 'NSFastEnumerationState' (as shown above) 58170instead. 58171 58172 The method is called repeatedly during a fast enumeration to retrieve 58173batches of objects. Each invocation of the method should retrieve the 58174next batch of objects. 58175 58176 The return value of the method is the number of objects in the current 58177batch; this should not exceed 'len', which is the maximum size of a 58178batch as requested by the caller. The batch itself is returned in the 58179'itemsPtr' field of the 'NSFastEnumerationState' struct. 58180 58181 To help with returning the objects, the 'objects' array is a C array 58182preallocated by the caller (on the stack) of size 'len'. In many cases 58183you can put the objects you want to return in that 'objects' array, then 58184do 'itemsPtr = objects'. But you don't have to; if your collection 58185already has the objects to return in some form of C array, it could 58186return them from there instead. 58187 58188 The 'state' and 'extra' fields of the 'NSFastEnumerationState' 58189structure allows your collection object to keep track of the state of 58190the enumeration. In a simple array implementation, 'state' may keep 58191track of the index of the last object that was returned, and 'extra' may 58192be unused. 58193 58194 The 'mutationsPtr' field of the 'NSFastEnumerationState' is used to 58195keep track of mutations. It should point to a number; before working on 58196each object, the fast enumeration loop will check that this number has 58197not changed. If it has, a mutation has happened and the fast 58198enumeration will abort. So, 'mutationsPtr' could be set to point to 58199some sort of version number of your collection, which is increased by 58200one every time there is a change (for example when an object is added or 58201removed). Or, if you are content with less strict mutation checks, it 58202could point to the number of objects in your collection or some other 58203value that can be checked to perform an approximate check that the 58204collection has not been mutated. 58205 58206 Finally, note how we declared the 'len' argument and the return value 58207to be of type 'unsigned long'. They could also be declared to be of 58208type 'unsigned int' and everything would still work. 58209 58210 58211File: gcc.info, Node: Messaging with the GNU Objective-C runtime, Prev: Fast enumeration, Up: Objective-C 58212 582138.10 Messaging with the GNU Objective-C Runtime 58214=============================================== 58215 58216This section is specific for the GNU Objective-C runtime. If you are 58217using a different runtime, you can skip it. 58218 58219 The implementation of messaging in the GNU Objective-C runtime is 58220designed to be portable, and so is based on standard C. 58221 58222 Sending a message in the GNU Objective-C runtime is composed of two 58223separate steps. First, there is a call to the lookup function, 58224'objc_msg_lookup ()' (or, in the case of messages to super, 58225'objc_msg_lookup_super ()'). This runtime function takes as argument 58226the receiver and the selector of the method to be called; it returns the 58227'IMP', that is a pointer to the function implementing the method. The 58228second step of method invocation consists of casting this pointer 58229function to the appropriate function pointer type, and calling the 58230function pointed to it with the right arguments. 58231 58232 For example, when the compiler encounters a method invocation such as 58233'[object init]', it compiles it into a call to 'objc_msg_lookup (object, 58234@selector(init))' followed by a cast of the returned value to the 58235appropriate function pointer type, and then it calls it. 58236 58237* Menu: 58238 58239* Dynamically registering methods:: 58240* Forwarding hook:: 58241 58242 58243File: gcc.info, Node: Dynamically registering methods, Next: Forwarding hook, Up: Messaging with the GNU Objective-C runtime 58244 582458.10.1 Dynamically Registering Methods 58246-------------------------------------- 58247 58248If 'objc_msg_lookup()' does not find a suitable method implementation, 58249because the receiver does not implement the required method, it tries to 58250see if the class can dynamically register the method. 58251 58252 To do so, the runtime checks if the class of the receiver implements 58253the method 58254 58255 + (BOOL) resolveInstanceMethod: (SEL)selector; 58256 58257 in the case of an instance method, or 58258 58259 + (BOOL) resolveClassMethod: (SEL)selector; 58260 58261 in the case of a class method. If the class implements it, the runtime 58262invokes it, passing as argument the selector of the original method, and 58263if it returns 'YES', the runtime tries the lookup again, which could now 58264succeed if a matching method was added dynamically by 58265'+resolveInstanceMethod:' or '+resolveClassMethod:'. 58266 58267 This allows classes to dynamically register methods (by adding them to 58268the class using 'class_addMethod') when they are first called. To do 58269so, a class should implement '+resolveInstanceMethod:' (or, depending on 58270the case, '+resolveClassMethod:') and have it recognize the selectors of 58271methods that can be registered dynamically at runtime, register them, 58272and return 'YES'. It should return 'NO' for methods that it does not 58273dynamically registered at runtime. 58274 58275 If '+resolveInstanceMethod:' (or '+resolveClassMethod:') is not 58276implemented or returns 'NO', the runtime then tries the forwarding hook. 58277 58278 Support for '+resolveInstanceMethod:' and 'resolveClassMethod:' was 58279added to the GNU Objective-C runtime in GCC version 4.6. 58280 58281 58282File: gcc.info, Node: Forwarding hook, Prev: Dynamically registering methods, Up: Messaging with the GNU Objective-C runtime 58283 582848.10.2 Forwarding Hook 58285---------------------- 58286 58287The GNU Objective-C runtime provides a hook, called 58288'__objc_msg_forward2', which is called by 'objc_msg_lookup()' when it 58289cannot find a method implementation in the runtime tables and after 58290calling '+resolveInstanceMethod:' and '+resolveClassMethod:' has been 58291attempted and did not succeed in dynamically registering the method. 58292 58293 To configure the hook, you set the global variable 58294'__objc_msg_forward2' to a function with the same argument and return 58295types of 'objc_msg_lookup()'. When 'objc_msg_lookup()' cannot find a 58296method implementation, it invokes the hook function you provided to get 58297a method implementation to return. So, in practice 58298'__objc_msg_forward2' allows you to extend 'objc_msg_lookup()' by adding 58299some custom code that is called to do a further lookup when no standard 58300method implementation can be found using the normal lookup. 58301 58302 This hook is generally reserved for "Foundation" libraries such as 58303GNUstep Base, which use it to implement their high-level method 58304forwarding API, typically based around the 'forwardInvocation:' method. 58305So, unless you are implementing your own "Foundation" library, you 58306should not set this hook. 58307 58308 In a typical forwarding implementation, the '__objc_msg_forward2' hook 58309function determines the argument and return type of the method that is 58310being looked up, and then creates a function that takes these arguments 58311and has that return type, and returns it to the caller. Creating this 58312function is non-trivial and is typically performed using a dedicated 58313library such as 'libffi'. 58314 58315 The forwarding method implementation thus created is returned by 58316'objc_msg_lookup()' and is executed as if it was a normal method 58317implementation. When the forwarding method implementation is called, it 58318is usually expected to pack all arguments into some sort of object 58319(typically, an 'NSInvocation' in a "Foundation" library), and hand it 58320over to the programmer ('forwardInvocation:') who is then allowed to 58321manipulate the method invocation using a high-level API provided by the 58322"Foundation" library. For example, the programmer may want to examine 58323the method invocation arguments and name and potentially change them 58324before forwarding the method invocation to one or more local objects 58325('performInvocation:') or even to remote objects (by using Distributed 58326Objects or some other mechanism). When all this completes, the return 58327value is passed back and must be returned correctly to the original 58328caller. 58329 58330 Note that the GNU Objective-C runtime currently provides no support for 58331method forwarding or method invocations other than the 58332'__objc_msg_forward2' hook. 58333 58334 If the forwarding hook does not exist or returns 'NULL', the runtime 58335currently attempts forwarding using an older, deprecated API, and if 58336that fails, it aborts the program. In future versions of the GNU 58337Objective-C runtime, the runtime will immediately abort. 58338 58339 58340File: gcc.info, Node: Compatibility, Next: Gcov, Prev: Objective-C, Up: Top 58341 583429 Binary Compatibility 58343********************** 58344 58345Binary compatibility encompasses several related concepts: 58346 58347"application binary interface (ABI)" 58348 The set of runtime conventions followed by all of the tools that 58349 deal with binary representations of a program, including compilers, 58350 assemblers, linkers, and language runtime support. Some ABIs are 58351 formal with a written specification, possibly designed by multiple 58352 interested parties. Others are simply the way things are actually 58353 done by a particular set of tools. 58354 58355"ABI conformance" 58356 A compiler conforms to an ABI if it generates code that follows all 58357 of the specifications enumerated by that ABI. A library conforms 58358 to an ABI if it is implemented according to that ABI. An 58359 application conforms to an ABI if it is built using tools that 58360 conform to that ABI and does not contain source code that 58361 specifically changes behavior specified by the ABI. 58362 58363"calling conventions" 58364 Calling conventions are a subset of an ABI that specify of how 58365 arguments are passed and function results are returned. 58366 58367"interoperability" 58368 Different sets of tools are interoperable if they generate files 58369 that can be used in the same program. The set of tools includes 58370 compilers, assemblers, linkers, libraries, header files, startup 58371 files, and debuggers. Binaries produced by different sets of tools 58372 are not interoperable unless they implement the same ABI. This 58373 applies to different versions of the same tools as well as tools 58374 from different vendors. 58375 58376"intercallability" 58377 Whether a function in a binary built by one set of tools can call a 58378 function in a binary built by a different set of tools is a subset 58379 of interoperability. 58380 58381"implementation-defined features" 58382 Language standards include lists of implementation-defined features 58383 whose behavior can vary from one implementation to another. Some 58384 of these features are normally covered by a platform's ABI and 58385 others are not. The features that are not covered by an ABI 58386 generally affect how a program behaves, but not intercallability. 58387 58388"compatibility" 58389 Conformance to the same ABI and the same behavior of 58390 implementation-defined features are both relevant for 58391 compatibility. 58392 58393 The application binary interface implemented by a C or C++ compiler 58394affects code generation and runtime support for: 58395 58396 * size and alignment of data types 58397 * layout of structured types 58398 * calling conventions 58399 * register usage conventions 58400 * interfaces for runtime arithmetic support 58401 * object file formats 58402 58403 In addition, the application binary interface implemented by a C++ 58404compiler affects code generation and runtime support for: 58405 * name mangling 58406 * exception handling 58407 * invoking constructors and destructors 58408 * layout, alignment, and padding of classes 58409 * layout and alignment of virtual tables 58410 58411 Some GCC compilation options cause the compiler to generate code that 58412does not conform to the platform's default ABI. Other options cause 58413different program behavior for implementation-defined features that are 58414not covered by an ABI. These options are provided for consistency with 58415other compilers that do not follow the platform's default ABI or the 58416usual behavior of implementation-defined features for the platform. Be 58417very careful about using such options. 58418 58419 Most platforms have a well-defined ABI that covers C code, but ABIs 58420that cover C++ functionality are not yet common. 58421 58422 Starting with GCC 3.2, GCC binary conventions for C++ are based on a 58423written, vendor-neutral C++ ABI that was designed to be specific to 5842464-bit Itanium but also includes generic specifications that apply to 58425any platform. This C++ ABI is also implemented by other compiler 58426vendors on some platforms, notably GNU/Linux and BSD systems. We have 58427tried hard to provide a stable ABI that will be compatible with future 58428GCC releases, but it is possible that we will encounter problems that 58429make this difficult. Such problems could include different 58430interpretations of the C++ ABI by different vendors, bugs in the ABI, or 58431bugs in the implementation of the ABI in different compilers. GCC's 58432'-Wabi' switch warns when G++ generates code that is probably not 58433compatible with the C++ ABI. 58434 58435 The C++ library used with a C++ compiler includes the Standard C++ 58436Library, with functionality defined in the C++ Standard, plus language 58437runtime support. The runtime support is included in a C++ ABI, but 58438there is no formal ABI for the Standard C++ Library. Two 58439implementations of that library are interoperable if one follows the 58440de-facto ABI of the other and if they are both built with the same 58441compiler, or with compilers that conform to the same ABI for C++ 58442compiler and runtime support. 58443 58444 When G++ and another C++ compiler conform to the same C++ ABI, but the 58445implementations of the Standard C++ Library that they normally use do 58446not follow the same ABI for the Standard C++ Library, object files built 58447with those compilers can be used in the same program only if they use 58448the same C++ library. This requires specifying the location of the C++ 58449library header files when invoking the compiler whose usual library is 58450not being used. The location of GCC's C++ header files depends on how 58451the GCC build was configured, but can be seen by using the G++ '-v' 58452option. With default configuration options for G++ 3.3 the compile line 58453for a different C++ compiler needs to include 58454 58455 -IGCC_INSTALL_DIRECTORY/include/c++/3.3 58456 58457 Similarly, compiling code with G++ that must use a C++ library other 58458than the GNU C++ library requires specifying the location of the header 58459files for that other library. 58460 58461 The most straightforward way to link a program to use a particular C++ 58462library is to use a C++ driver that specifies that C++ library by 58463default. The 'g++' driver, for example, tells the linker where to find 58464GCC's C++ library ('libstdc++') plus the other libraries and startup 58465files it needs, in the proper order. 58466 58467 If a program must use a different C++ library and it's not possible to 58468do the final link using a C++ driver that uses that library by default, 58469it is necessary to tell 'g++' the location and name of that library. It 58470might also be necessary to specify different startup files and other 58471runtime support libraries, and to suppress the use of GCC's support 58472libraries with one or more of the options '-nostdlib', '-nostartfiles', 58473and '-nodefaultlibs'. 58474 58475 58476File: gcc.info, Node: Gcov, Next: Gcov-tool, Prev: Compatibility, Up: Top 58477 5847810 'gcov'--a Test Coverage Program 58479********************************** 58480 58481'gcov' is a tool you can use in conjunction with GCC to test code 58482coverage in your programs. 58483 58484* Menu: 58485 58486* Gcov Intro:: Introduction to gcov. 58487* Invoking Gcov:: How to use gcov. 58488* Gcov and Optimization:: Using gcov with GCC optimization. 58489* Gcov Data Files:: The files used by gcov. 58490* Cross-profiling:: Data file relocation. 58491 58492 58493File: gcc.info, Node: Gcov Intro, Next: Invoking Gcov, Up: Gcov 58494 5849510.1 Introduction to 'gcov' 58496=========================== 58497 58498'gcov' is a test coverage program. Use it in concert with GCC to 58499analyze your programs to help create more efficient, faster running code 58500and to discover untested parts of your program. You can use 'gcov' as a 58501profiling tool to help discover where your optimization efforts will 58502best affect your code. You can also use 'gcov' along with the other 58503profiling tool, 'gprof', to assess which parts of your code use the 58504greatest amount of computing time. 58505 58506 Profiling tools help you analyze your code's performance. Using a 58507profiler such as 'gcov' or 'gprof', you can find out some basic 58508performance statistics, such as: 58509 58510 * how often each line of code executes 58511 58512 * what lines of code are actually executed 58513 58514 * how much computing time each section of code uses 58515 58516 Once you know these things about how your code works when compiled, you 58517can look at each module to see which modules should be optimized. 58518'gcov' helps you determine where to work on optimization. 58519 58520 Software developers also use coverage testing in concert with 58521testsuites, to make sure software is actually good enough for a release. 58522Testsuites can verify that a program works as expected; a coverage 58523program tests to see how much of the program is exercised by the 58524testsuite. Developers can then determine what kinds of test cases need 58525to be added to the testsuites to create both better testing and a better 58526final product. 58527 58528 You should compile your code without optimization if you plan to use 58529'gcov' because the optimization, by combining some lines of code into 58530one function, may not give you as much information as you need to look 58531for 'hot spots' where the code is using a great deal of computer time. 58532Likewise, because 'gcov' accumulates statistics by line (at the lowest 58533resolution), it works best with a programming style that places only one 58534statement on each line. If you use complicated macros that expand to 58535loops or to other control structures, the statistics are less 58536helpful--they only report on the line where the macro call appears. If 58537your complex macros behave like functions, you can replace them with 58538inline functions to solve this problem. 58539 58540 'gcov' creates a logfile called 'SOURCEFILE.gcov' which indicates how 58541many times each line of a source file 'SOURCEFILE.c' has executed. You 58542can use these logfiles along with 'gprof' to aid in fine-tuning the 58543performance of your programs. 'gprof' gives timing information you can 58544use along with the information you get from 'gcov'. 58545 58546 'gcov' works only on code compiled with GCC. It is not compatible with 58547any other profiling or test coverage mechanism. 58548 58549 58550File: gcc.info, Node: Invoking Gcov, Next: Gcov and Optimization, Prev: Gcov Intro, Up: Gcov 58551 5855210.2 Invoking 'gcov' 58553==================== 58554 58555 gcov [OPTIONS] FILES 58556 58557 'gcov' accepts the following options: 58558 58559'-a' 58560'--all-blocks' 58561 Write individual execution counts for every basic block. Normally 58562 gcov outputs execution counts only for the main blocks of a line. 58563 With this option you can determine if blocks within a single line 58564 are not being executed. 58565 58566'-b' 58567'--branch-probabilities' 58568 Write branch frequencies to the output file, and write branch 58569 summary info to the standard output. This option allows you to see 58570 how often each branch in your program was taken. Unconditional 58571 branches will not be shown, unless the '-u' option is given. 58572 58573'-c' 58574'--branch-counts' 58575 Write branch frequencies as the number of branches taken, rather 58576 than the percentage of branches taken. 58577 58578'-d' 58579'--display-progress' 58580 Display the progress on the standard output. 58581 58582'-f' 58583'--function-summaries' 58584 Output summaries for each function in addition to the file level 58585 summary. 58586 58587'-h' 58588'--help' 58589 Display help about using 'gcov' (on the standard output), and exit 58590 without doing any further processing. 58591 58592'-j' 58593'--json-format' 58594 Output gcov file in an easy-to-parse JSON intermediate format which 58595 does not require source code for generation. The JSON file is 58596 compressed with gzip compression algorithm and the files have 58597 '.gcov.json.gz' extension. 58598 58599 Structure of the JSON is following: 58600 58601 { 58602 "current_working_directory": "foo/bar", 58603 "data_file": "a.out", 58604 "format_version": "1", 58605 "gcc_version": "11.1.1 20210510" 58606 "files": ["$file"] 58607 } 58608 58609 Fields of the root element have following semantics: 58610 58611 * CURRENT_WORKING_DIRECTORY: working directory where a 58612 compilation unit was compiled 58613 58614 * DATA_FILE: name of the data file (GCDA) 58615 58616 * FORMAT_VERSION: semantic version of the format 58617 58618 * GCC_VERSION: version of the GCC compiler 58619 58620 Each FILE has the following form: 58621 58622 { 58623 "file": "a.c", 58624 "functions": ["$function"], 58625 "lines": ["$line"] 58626 } 58627 58628 Fields of the FILE element have following semantics: 58629 58630 * FILE_NAME: name of the source file 58631 58632 Each FUNCTION has the following form: 58633 58634 { 58635 "blocks": 2, 58636 "blocks_executed": 2, 58637 "demangled_name": "foo", 58638 "end_column": 1, 58639 "end_line": 4, 58640 "execution_count": 1, 58641 "name": "foo", 58642 "start_column": 5, 58643 "start_line": 1 58644 } 58645 58646 Fields of the FUNCTION element have following semantics: 58647 58648 * BLOCKS: number of blocks that are in the function 58649 58650 * BLOCKS_EXECUTED: number of executed blocks of the function 58651 58652 * DEMANGLED_NAME: demangled name of the function 58653 58654 * END_COLUMN: column in the source file where the function ends 58655 58656 * END_LINE: line in the source file where the function ends 58657 58658 * EXECUTION_COUNT: number of executions of the function 58659 58660 * NAME: name of the function 58661 58662 * START_COLUMN: column in the source file where the function 58663 begins 58664 58665 * START_LINE: line in the source file where the function begins 58666 58667 Note that line numbers and column numbers number from 1. In the 58668 current implementation, START_LINE and START_COLUMN do not include 58669 any template parameters and the leading return type but that this 58670 is likely to be fixed in the future. 58671 58672 Each LINE has the following form: 58673 58674 { 58675 "branches": ["$branch"], 58676 "count": 2, 58677 "line_number": 15, 58678 "unexecuted_block": false, 58679 "function_name": "foo", 58680 } 58681 58682 Branches are present only with -B option. Fields of the LINE 58683 element have following semantics: 58684 58685 * COUNT: number of executions of the line 58686 58687 * LINE_NUMBER: line number 58688 58689 * UNEXECUTED_BLOCK: flag whether the line contains an unexecuted 58690 block (not all statements on the line are executed) 58691 58692 * FUNCTION_NAME: a name of a function this LINE belongs to (for 58693 a line with an inlined statements can be not set) 58694 58695 Each BRANCH has the following form: 58696 58697 { 58698 "count": 11, 58699 "fallthrough": true, 58700 "throw": false 58701 } 58702 58703 Fields of the BRANCH element have following semantics: 58704 58705 * COUNT: number of executions of the branch 58706 58707 * FALLTHROUGH: true when the branch is a fall through branch 58708 58709 * THROW: true when the branch is an exceptional branch 58710 58711'-H' 58712'--human-readable' 58713 Write counts in human readable format (like 24.6k). 58714 58715'-k' 58716'--use-colors' 58717 58718 Use colors for lines of code that have zero coverage. We use red 58719 color for non-exceptional lines and cyan for exceptional. Same 58720 colors are used for basic blocks with '-a' option. 58721 58722'-l' 58723'--long-file-names' 58724 Create long file names for included source files. For example, if 58725 the header file 'x.h' contains code, and was included in the file 58726 'a.c', then running 'gcov' on the file 'a.c' will produce an output 58727 file called 'a.c##x.h.gcov' instead of 'x.h.gcov'. This can be 58728 useful if 'x.h' is included in multiple source files and you want 58729 to see the individual contributions. If you use the '-p' option, 58730 both the including and included file names will be complete path 58731 names. 58732 58733'-m' 58734'--demangled-names' 58735 Display demangled function names in output. The default is to show 58736 mangled function names. 58737 58738'-n' 58739'--no-output' 58740 Do not create the 'gcov' output file. 58741 58742'-o DIRECTORY|FILE' 58743'--object-directory DIRECTORY' 58744'--object-file FILE' 58745 Specify either the directory containing the gcov data files, or the 58746 object path name. The '.gcno', and '.gcda' data files are searched 58747 for using this option. If a directory is specified, the data files 58748 are in that directory and named after the input file name, without 58749 its extension. If a file is specified here, the data files are 58750 named after that file, without its extension. 58751 58752'-p' 58753'--preserve-paths' 58754 Preserve complete path information in the names of generated 58755 '.gcov' files. Without this option, just the filename component is 58756 used. With this option, all directories are used, with '/' 58757 characters translated to '#' characters, '.' directory components 58758 removed and unremoveable '..' components renamed to '^'. This is 58759 useful if sourcefiles are in several different directories. 58760 58761'-q' 58762'--use-hotness-colors' 58763 58764 Emit perf-like colored output for hot lines. Legend of the color 58765 scale is printed at the very beginning of the output file. 58766 58767'-r' 58768'--relative-only' 58769 Only output information about source files with a relative pathname 58770 (after source prefix elision). Absolute paths are usually system 58771 header files and coverage of any inline functions therein is 58772 normally uninteresting. 58773 58774'-s DIRECTORY' 58775'--source-prefix DIRECTORY' 58776 A prefix for source file names to remove when generating the output 58777 coverage files. This option is useful when building in a separate 58778 directory, and the pathname to the source directory is not wanted 58779 when determining the output file names. Note that this prefix 58780 detection is applied before determining whether the source file is 58781 absolute. 58782 58783'-t' 58784'--stdout' 58785 Output to standard output instead of output files. 58786 58787'-u' 58788'--unconditional-branches' 58789 When branch probabilities are given, include those of unconditional 58790 branches. Unconditional branches are normally not interesting. 58791 58792'-v' 58793'--version' 58794 Display the 'gcov' version number (on the standard output), and 58795 exit without doing any further processing. 58796 58797'-w' 58798'--verbose' 58799 Print verbose informations related to basic blocks and arcs. 58800 58801'-x' 58802'--hash-filenames' 58803 When using -PRESERVE-PATHS, gcov uses the full pathname of the 58804 source files to create an output filename. This can lead to long 58805 filenames that can overflow filesystem limits. This option creates 58806 names of the form 'SOURCE-FILE##MD5.gcov', where the SOURCE-FILE 58807 component is the final filename part and the MD5 component is 58808 calculated from the full mangled name that would have been used 58809 otherwise. The option is an alternative to the -PRESERVE-PATHS on 58810 systems which have a filesystem limit. 58811 58812 'gcov' should be run with the current directory the same as that when 58813you invoked the compiler. Otherwise it will not be able to locate the 58814source files. 'gcov' produces files called 'MANGLEDNAME.gcov' in the 58815current directory. These contain the coverage information of the source 58816file they correspond to. One '.gcov' file is produced for each source 58817(or header) file containing code, which was compiled to produce the data 58818files. The MANGLEDNAME part of the output file name is usually simply 58819the source file name, but can be something more complicated if the '-l' 58820or '-p' options are given. Refer to those options for details. 58821 58822 If you invoke 'gcov' with multiple input files, the contributions from 58823each input file are summed. Typically you would invoke it with the same 58824list of files as the final link of your executable. 58825 58826 The '.gcov' files contain the ':' separated fields along with program 58827source code. The format is 58828 58829 EXECUTION_COUNT:LINE_NUMBER:SOURCE LINE TEXT 58830 58831 Additional block information may succeed each line, when requested by 58832command line option. The EXECUTION_COUNT is '-' for lines containing no 58833code. Unexecuted lines are marked '#####' or '=====', depending on 58834whether they are reachable by non-exceptional paths or only exceptional 58835paths such as C++ exception handlers, respectively. Given the '-a' 58836option, unexecuted blocks are marked '$$$$$' or '%%%%%', depending on 58837whether a basic block is reachable via non-exceptional or exceptional 58838paths. Executed basic blocks having a statement with zero 58839EXECUTION_COUNT end with '*' character and are colored with magenta 58840color with the '-k' option. This functionality is not supported in Ada. 58841 58842 Note that GCC can completely remove the bodies of functions that are 58843not needed - for instance if they are inlined everywhere. Such 58844functions are marked with '-', which can be confusing. Use the 58845'-fkeep-inline-functions' and '-fkeep-static-functions' options to 58846retain these functions and allow gcov to properly show their 58847EXECUTION_COUNT. 58848 58849 Some lines of information at the start have LINE_NUMBER of zero. These 58850preamble lines are of the form 58851 58852 -:0:TAG:VALUE 58853 58854 The ordering and number of these preamble lines will be augmented as 58855'gcov' development progresses -- do not rely on them remaining 58856unchanged. Use TAG to locate a particular preamble line. 58857 58858 The additional block information is of the form 58859 58860 TAG INFORMATION 58861 58862 The INFORMATION is human readable, but designed to be simple enough for 58863machine parsing too. 58864 58865 When printing percentages, 0% and 100% are only printed when the values 58866are _exactly_ 0% and 100% respectively. Other values which would 58867conventionally be rounded to 0% or 100% are instead printed as the 58868nearest non-boundary value. 58869 58870 When using 'gcov', you must first compile your program with a special 58871GCC option '--coverage'. This tells the compiler to generate additional 58872information needed by gcov (basically a flow graph of the program) and 58873also includes additional code in the object files for generating the 58874extra profiling information needed by gcov. These additional files are 58875placed in the directory where the object file is located. 58876 58877 Running the program will cause profile output to be generated. For 58878each source file compiled with '-fprofile-arcs', an accompanying '.gcda' 58879file will be placed in the object file directory. 58880 58881 Running 'gcov' with your program's source file names as arguments will 58882now produce a listing of the code along with frequency of execution for 58883each line. For example, if your program is called 'tmp.cpp', this is 58884what you see when you use the basic 'gcov' facility: 58885 58886 $ g++ --coverage tmp.cpp -c 58887 $ g++ --coverage tmp.o 58888 $ a.out 58889 $ gcov tmp.cpp -m 58890 File 'tmp.cpp' 58891 Lines executed:92.86% of 14 58892 Creating 'tmp.cpp.gcov' 58893 58894 The file 'tmp.cpp.gcov' contains output from 'gcov'. Here is a sample: 58895 58896 -: 0:Source:tmp.cpp 58897 -: 0:Working directory:/home/gcc/testcase 58898 -: 0:Graph:tmp.gcno 58899 -: 0:Data:tmp.gcda 58900 -: 0:Runs:1 58901 -: 0:Programs:1 58902 -: 1:#include <stdio.h> 58903 -: 2: 58904 -: 3:template<class T> 58905 -: 4:class Foo 58906 -: 5:{ 58907 -: 6: public: 58908 1*: 7: Foo(): b (1000) {} 58909 ------------------ 58910 Foo<char>::Foo(): 58911 #####: 7: Foo(): b (1000) {} 58912 ------------------ 58913 Foo<int>::Foo(): 58914 1: 7: Foo(): b (1000) {} 58915 ------------------ 58916 2*: 8: void inc () { b++; } 58917 ------------------ 58918 Foo<char>::inc(): 58919 #####: 8: void inc () { b++; } 58920 ------------------ 58921 Foo<int>::inc(): 58922 2: 8: void inc () { b++; } 58923 ------------------ 58924 -: 9: 58925 -: 10: private: 58926 -: 11: int b; 58927 -: 12:}; 58928 -: 13: 58929 -: 14:template class Foo<int>; 58930 -: 15:template class Foo<char>; 58931 -: 16: 58932 -: 17:int 58933 1: 18:main (void) 58934 -: 19:{ 58935 -: 20: int i, total; 58936 1: 21: Foo<int> counter; 58937 -: 22: 58938 1: 23: counter.inc(); 58939 1: 24: counter.inc(); 58940 1: 25: total = 0; 58941 -: 26: 58942 11: 27: for (i = 0; i < 10; i++) 58943 10: 28: total += i; 58944 -: 29: 58945 1*: 30: int v = total > 100 ? 1 : 2; 58946 -: 31: 58947 1: 32: if (total != 45) 58948 #####: 33: printf ("Failure\n"); 58949 -: 34: else 58950 1: 35: printf ("Success\n"); 58951 1: 36: return 0; 58952 -: 37:} 58953 58954 Note that line 7 is shown in the report multiple times. First 58955occurrence presents total number of execution of the line and the next 58956two belong to instances of class Foo constructors. As you can also see, 58957line 30 contains some unexecuted basic blocks and thus execution count 58958has asterisk symbol. 58959 58960 When you use the '-a' option, you will get individual block counts, and 58961the output looks like this: 58962 58963 -: 0:Source:tmp.cpp 58964 -: 0:Working directory:/home/gcc/testcase 58965 -: 0:Graph:tmp.gcno 58966 -: 0:Data:tmp.gcda 58967 -: 0:Runs:1 58968 -: 0:Programs:1 58969 -: 1:#include <stdio.h> 58970 -: 2: 58971 -: 3:template<class T> 58972 -: 4:class Foo 58973 -: 5:{ 58974 -: 6: public: 58975 1*: 7: Foo(): b (1000) {} 58976 ------------------ 58977 Foo<char>::Foo(): 58978 #####: 7: Foo(): b (1000) {} 58979 ------------------ 58980 Foo<int>::Foo(): 58981 1: 7: Foo(): b (1000) {} 58982 ------------------ 58983 2*: 8: void inc () { b++; } 58984 ------------------ 58985 Foo<char>::inc(): 58986 #####: 8: void inc () { b++; } 58987 ------------------ 58988 Foo<int>::inc(): 58989 2: 8: void inc () { b++; } 58990 ------------------ 58991 -: 9: 58992 -: 10: private: 58993 -: 11: int b; 58994 -: 12:}; 58995 -: 13: 58996 -: 14:template class Foo<int>; 58997 -: 15:template class Foo<char>; 58998 -: 16: 58999 -: 17:int 59000 1: 18:main (void) 59001 -: 19:{ 59002 -: 20: int i, total; 59003 1: 21: Foo<int> counter; 59004 1: 21-block 0 59005 -: 22: 59006 1: 23: counter.inc(); 59007 1: 23-block 0 59008 1: 24: counter.inc(); 59009 1: 24-block 0 59010 1: 25: total = 0; 59011 -: 26: 59012 11: 27: for (i = 0; i < 10; i++) 59013 1: 27-block 0 59014 11: 27-block 1 59015 10: 28: total += i; 59016 10: 28-block 0 59017 -: 29: 59018 1*: 30: int v = total > 100 ? 1 : 2; 59019 1: 30-block 0 59020 %%%%%: 30-block 1 59021 1: 30-block 2 59022 -: 31: 59023 1: 32: if (total != 45) 59024 1: 32-block 0 59025 #####: 33: printf ("Failure\n"); 59026 %%%%%: 33-block 0 59027 -: 34: else 59028 1: 35: printf ("Success\n"); 59029 1: 35-block 0 59030 1: 36: return 0; 59031 1: 36-block 0 59032 -: 37:} 59033 59034 In this mode, each basic block is only shown on one line - the last 59035line of the block. A multi-line block will only contribute to the 59036execution count of that last line, and other lines will not be shown to 59037contain code, unless previous blocks end on those lines. The total 59038execution count of a line is shown and subsequent lines show the 59039execution counts for individual blocks that end on that line. After 59040each block, the branch and call counts of the block will be shown, if 59041the '-b' option is given. 59042 59043 Because of the way GCC instruments calls, a call count can be shown 59044after a line with no individual blocks. As you can see, line 33 59045contains a basic block that was not executed. 59046 59047 When you use the '-b' option, your output looks like this: 59048 59049 -: 0:Source:tmp.cpp 59050 -: 0:Working directory:/home/gcc/testcase 59051 -: 0:Graph:tmp.gcno 59052 -: 0:Data:tmp.gcda 59053 -: 0:Runs:1 59054 -: 0:Programs:1 59055 -: 1:#include <stdio.h> 59056 -: 2: 59057 -: 3:template<class T> 59058 -: 4:class Foo 59059 -: 5:{ 59060 -: 6: public: 59061 1*: 7: Foo(): b (1000) {} 59062 ------------------ 59063 Foo<char>::Foo(): 59064 function Foo<char>::Foo() called 0 returned 0% blocks executed 0% 59065 #####: 7: Foo(): b (1000) {} 59066 ------------------ 59067 Foo<int>::Foo(): 59068 function Foo<int>::Foo() called 1 returned 100% blocks executed 100% 59069 1: 7: Foo(): b (1000) {} 59070 ------------------ 59071 2*: 8: void inc () { b++; } 59072 ------------------ 59073 Foo<char>::inc(): 59074 function Foo<char>::inc() called 0 returned 0% blocks executed 0% 59075 #####: 8: void inc () { b++; } 59076 ------------------ 59077 Foo<int>::inc(): 59078 function Foo<int>::inc() called 2 returned 100% blocks executed 100% 59079 2: 8: void inc () { b++; } 59080 ------------------ 59081 -: 9: 59082 -: 10: private: 59083 -: 11: int b; 59084 -: 12:}; 59085 -: 13: 59086 -: 14:template class Foo<int>; 59087 -: 15:template class Foo<char>; 59088 -: 16: 59089 -: 17:int 59090 function main called 1 returned 100% blocks executed 81% 59091 1: 18:main (void) 59092 -: 19:{ 59093 -: 20: int i, total; 59094 1: 21: Foo<int> counter; 59095 call 0 returned 100% 59096 branch 1 taken 100% (fallthrough) 59097 branch 2 taken 0% (throw) 59098 -: 22: 59099 1: 23: counter.inc(); 59100 call 0 returned 100% 59101 branch 1 taken 100% (fallthrough) 59102 branch 2 taken 0% (throw) 59103 1: 24: counter.inc(); 59104 call 0 returned 100% 59105 branch 1 taken 100% (fallthrough) 59106 branch 2 taken 0% (throw) 59107 1: 25: total = 0; 59108 -: 26: 59109 11: 27: for (i = 0; i < 10; i++) 59110 branch 0 taken 91% (fallthrough) 59111 branch 1 taken 9% 59112 10: 28: total += i; 59113 -: 29: 59114 1*: 30: int v = total > 100 ? 1 : 2; 59115 branch 0 taken 0% (fallthrough) 59116 branch 1 taken 100% 59117 -: 31: 59118 1: 32: if (total != 45) 59119 branch 0 taken 0% (fallthrough) 59120 branch 1 taken 100% 59121 #####: 33: printf ("Failure\n"); 59122 call 0 never executed 59123 branch 1 never executed 59124 branch 2 never executed 59125 -: 34: else 59126 1: 35: printf ("Success\n"); 59127 call 0 returned 100% 59128 branch 1 taken 100% (fallthrough) 59129 branch 2 taken 0% (throw) 59130 1: 36: return 0; 59131 -: 37:} 59132 59133 For each function, a line is printed showing how many times the 59134function is called, how many times it returns and what percentage of the 59135function's blocks were executed. 59136 59137 For each basic block, a line is printed after the last line of the 59138basic block describing the branch or call that ends the basic block. 59139There can be multiple branches and calls listed for a single source line 59140if there are multiple basic blocks that end on that line. In this case, 59141the branches and calls are each given a number. There is no simple way 59142to map these branches and calls back to source constructs. In general, 59143though, the lowest numbered branch or call will correspond to the 59144leftmost construct on the source line. 59145 59146 For a branch, if it was executed at least once, then a percentage 59147indicating the number of times the branch was taken divided by the 59148number of times the branch was executed will be printed. Otherwise, the 59149message "never executed" is printed. 59150 59151 For a call, if it was executed at least once, then a percentage 59152indicating the number of times the call returned divided by the number 59153of times the call was executed will be printed. This will usually be 59154100%, but may be less for functions that call 'exit' or 'longjmp', and 59155thus may not return every time they are called. 59156 59157 The execution counts are cumulative. If the example program were 59158executed again without removing the '.gcda' file, the count for the 59159number of times each line in the source was executed would be added to 59160the results of the previous run(s). This is potentially useful in 59161several ways. For example, it could be used to accumulate data over a 59162number of program runs as part of a test verification suite, or to 59163provide more accurate long-term information over a large number of 59164program runs. 59165 59166 The data in the '.gcda' files is saved immediately before the program 59167exits. For each source file compiled with '-fprofile-arcs', the 59168profiling code first attempts to read in an existing '.gcda' file; if 59169the file doesn't match the executable (differing number of basic block 59170counts) it will ignore the contents of the file. It then adds in the 59171new execution counts and finally writes the data to the file. 59172 59173 59174File: gcc.info, Node: Gcov and Optimization, Next: Gcov Data Files, Prev: Invoking Gcov, Up: Gcov 59175 5917610.3 Using 'gcov' with GCC Optimization 59177======================================= 59178 59179If you plan to use 'gcov' to help optimize your code, you must first 59180compile your program with a special GCC option '--coverage'. Aside from 59181that, you can use any other GCC options; but if you want to prove that 59182every single line in your program was executed, you should not compile 59183with optimization at the same time. On some machines the optimizer can 59184eliminate some simple code lines by combining them with other lines. 59185For example, code like this: 59186 59187 if (a != b) 59188 c = 1; 59189 else 59190 c = 0; 59191 59192can be compiled into one instruction on some machines. In this case, 59193there is no way for 'gcov' to calculate separate execution counts for 59194each line because there isn't separate code for each line. Hence the 59195'gcov' output looks like this if you compiled the program with 59196optimization: 59197 59198 100: 12:if (a != b) 59199 100: 13: c = 1; 59200 100: 14:else 59201 100: 15: c = 0; 59202 59203 The output shows that this block of code, combined by optimization, 59204executed 100 times. In one sense this result is correct, because there 59205was only one instruction representing all four of these lines. However, 59206the output does not indicate how many times the result was 0 and how 59207many times the result was 1. 59208 59209 Inlineable functions can create unexpected line counts. Line counts 59210are shown for the source code of the inlineable function, but what is 59211shown depends on where the function is inlined, or if it is not inlined 59212at all. 59213 59214 If the function is not inlined, the compiler must emit an out of line 59215copy of the function, in any object file that needs it. If 'fileA.o' 59216and 'fileB.o' both contain out of line bodies of a particular inlineable 59217function, they will also both contain coverage counts for that function. 59218When 'fileA.o' and 'fileB.o' are linked together, the linker will, on 59219many systems, select one of those out of line bodies for all calls to 59220that function, and remove or ignore the other. Unfortunately, it will 59221not remove the coverage counters for the unused function body. Hence 59222when instrumented, all but one use of that function will show zero 59223counts. 59224 59225 If the function is inlined in several places, the block structure in 59226each location might not be the same. For instance, a condition might 59227now be calculable at compile time in some instances. Because the 59228coverage of all the uses of the inline function will be shown for the 59229same source lines, the line counts themselves might seem inconsistent. 59230 59231 Long-running applications can use the '__gcov_reset' and '__gcov_dump' 59232facilities to restrict profile collection to the program region of 59233interest. Calling '__gcov_reset(void)' will clear all run-time profile 59234counters to zero, and calling '__gcov_dump(void)' will cause the profile 59235information collected at that point to be dumped to '.gcda' output 59236files. Instrumented applications use a static destructor with priority 5923799 to invoke the '__gcov_dump' function. Thus '__gcov_dump' is executed 59238after all user defined static destructors, as well as handlers 59239registered with 'atexit'. 59240 59241 If an executable loads a dynamic shared object via dlopen 59242functionality, '-Wl,--dynamic-list-data' is needed to dump all profile 59243data. 59244 59245 Profiling run-time library reports various errors related to profile 59246manipulation and profile saving. Errors are printed into standard error 59247output or 'GCOV_ERROR_FILE' file, if environment variable is used. In 59248order to terminate immediately after an errors occurs set 59249'GCOV_EXIT_AT_ERROR' environment variable. That can help users to find 59250profile clashing which leads to a misleading profile. 59251 59252 59253File: gcc.info, Node: Gcov Data Files, Next: Cross-profiling, Prev: Gcov and Optimization, Up: Gcov 59254 5925510.4 Brief Description of 'gcov' Data Files 59256=========================================== 59257 59258'gcov' uses two files for profiling. The names of these files are 59259derived from the original _object_ file by substituting the file suffix 59260with either '.gcno', or '.gcda'. The files contain coverage and profile 59261data stored in a platform-independent format. The '.gcno' files are 59262placed in the same directory as the object file. By default, the 59263'.gcda' files are also stored in the same directory as the object file, 59264but the GCC '-fprofile-dir' option may be used to store the '.gcda' 59265files in a separate directory. 59266 59267 The '.gcno' notes file is generated when the source file is compiled 59268with the GCC '-ftest-coverage' option. It contains information to 59269reconstruct the basic block graphs and assign source line numbers to 59270blocks. 59271 59272 The '.gcda' count data file is generated when a program containing 59273object files built with the GCC '-fprofile-arcs' option is executed. A 59274separate '.gcda' file is created for each object file compiled with this 59275option. It contains arc transition counts, value profile counts, and 59276some summary information. 59277 59278 It is not recommended to access the coverage files directly. Consumers 59279should use the intermediate format that is provided by 'gcov' tool via 59280'--json-format' option. 59281 59282 59283File: gcc.info, Node: Cross-profiling, Prev: Gcov Data Files, Up: Gcov 59284 5928510.5 Data File Relocation to Support Cross-Profiling 59286==================================================== 59287 59288Running the program will cause profile output to be generated. For each 59289source file compiled with '-fprofile-arcs', an accompanying '.gcda' file 59290will be placed in the object file directory. That implicitly requires 59291running the program on the same system as it was built or having the 59292same absolute directory structure on the target system. The program 59293will try to create the needed directory structure, if it is not already 59294present. 59295 59296 To support cross-profiling, a program compiled with '-fprofile-arcs' 59297can relocate the data files based on two environment variables: 59298 59299 * GCOV_PREFIX contains the prefix to add to the absolute paths in the 59300 object file. Prefix can be absolute, or relative. The default is 59301 no prefix. 59302 59303 * GCOV_PREFIX_STRIP indicates the how many initial directory names to 59304 strip off the hardwired absolute paths. Default value is 0. 59305 59306 _Note:_ If GCOV_PREFIX_STRIP is set without GCOV_PREFIX is 59307 undefined, then a relative path is made out of the hardwired 59308 absolute paths. 59309 59310 For example, if the object file '/user/build/foo.o' was built with 59311'-fprofile-arcs', the final executable will try to create the data file 59312'/user/build/foo.gcda' when running on the target system. This will 59313fail if the corresponding directory does not exist and it is unable to 59314create it. This can be overcome by, for example, setting the 59315environment as 'GCOV_PREFIX=/target/run' and 'GCOV_PREFIX_STRIP=1'. 59316Such a setting will name the data file '/target/run/build/foo.gcda'. 59317 59318 You must move the data files to the expected directory tree in order to 59319use them for profile directed optimizations ('-fprofile-use'), or to use 59320the 'gcov' tool. 59321 59322 59323File: gcc.info, Node: Gcov-tool, Next: Gcov-dump, Prev: Gcov, Up: Top 59324 5932511 'gcov-tool'--an Offline Gcda Profile Processing Tool 59326******************************************************* 59327 59328'gcov-tool' is a tool you can use in conjunction with GCC to manipulate 59329or process gcda profile files offline. 59330 59331* Menu: 59332 59333* Gcov-tool Intro:: Introduction to gcov-tool. 59334* Invoking Gcov-tool:: How to use gcov-tool. 59335 59336 59337File: gcc.info, Node: Gcov-tool Intro, Next: Invoking Gcov-tool, Up: Gcov-tool 59338 5933911.1 Introduction to 'gcov-tool' 59340================================ 59341 59342'gcov-tool' is an offline tool to process gcc's gcda profile files. 59343 59344 Current gcov-tool supports the following functionalities: 59345 59346 * merge two sets of profiles with weights. 59347 59348 * read one set of profile and rewrite profile contents. One can 59349 scale or normalize the count values. 59350 59351 Examples of the use cases for this tool are: 59352 * Collect the profiles for different set of inputs, and use this tool 59353 to merge them. One can specify the weight to factor in the 59354 relative importance of each input. 59355 59356 * Rewrite the profile after removing a subset of the gcda files, 59357 while maintaining the consistency of the summary and the histogram. 59358 59359 * It can also be used to debug or libgcov code as the tools shares 59360 the majority code as the runtime library. 59361 59362 Note that for the merging operation, this profile generated offline may 59363contain slight different values from the online merged profile. Here 59364are a list of typical differences: 59365 59366 * histogram difference: This offline tool recomputes the histogram 59367 after merging the counters. The resulting histogram, therefore, is 59368 precise. The online merging does not have this capability - the 59369 histogram is merged from two histograms and the result is an 59370 approximation. 59371 59372 * summary checksum difference: Summary checksum uses a CRC32 59373 operation. The value depends on the link list order of gcov-info 59374 objects. This order is different in gcov-tool from that in the 59375 online merge. It's expected to have different summary checksums. 59376 It does not really matter as the compiler does not use this 59377 checksum anywhere. 59378 59379 * value profile counter values difference: Some counter values for 59380 value profile are runtime dependent, like heap addresses. It's 59381 normal to see some difference in these kind of counters. 59382 59383 59384File: gcc.info, Node: Invoking Gcov-tool, Prev: Gcov-tool Intro, Up: Gcov-tool 59385 5938611.2 Invoking 'gcov-tool' 59387========================= 59388 59389 gcov-tool [GLOBAL-OPTIONS] SUB_COMMAND [SUB_COMMAND-OPTIONS] PROFILE_DIR 59390 59391 'gcov-tool' accepts the following options: 59392 59393'-h' 59394'--help' 59395 Display help about using 'gcov-tool' (on the standard output), and 59396 exit without doing any further processing. 59397 59398'-v' 59399'--version' 59400 Display the 'gcov-tool' version number (on the standard output), 59401 and exit without doing any further processing. 59402 59403'merge' 59404 Merge two profile directories. 59405 59406 '-o DIRECTORY' 59407 '--output DIRECTORY' 59408 Set the output profile directory. Default output directory 59409 name is MERGED_PROFILE. 59410 59411 '-v' 59412 '--verbose' 59413 Set the verbose mode. 59414 59415 '-w W1,W2' 59416 '--weight W1,W2' 59417 Set the merge weights of the DIRECTORY1 and DIRECTORY2, 59418 respectively. The default weights are 1 for both. 59419 59420'rewrite' 59421 Read the specified profile directory and rewrite to a new 59422 directory. 59423 59424 '-n LONG_LONG_VALUE' 59425 '--normalize <long_long_value>' 59426 Normalize the profile. The specified value is the max counter 59427 value in the new profile. 59428 59429 '-o DIRECTORY' 59430 '--output DIRECTORY' 59431 Set the output profile directory. Default output name is 59432 REWRITE_PROFILE. 59433 59434 '-s FLOAT_OR_SIMPLE-FRAC_VALUE' 59435 '--scale FLOAT_OR_SIMPLE-FRAC_VALUE' 59436 Scale the profile counters. The specified value can be in 59437 floating point value, or simple fraction value form, such 1, 59438 2, 2/3, and 5/3. 59439 59440 '-v' 59441 '--verbose' 59442 Set the verbose mode. 59443 59444'overlap' 59445 Compute the overlap score between the two specified profile 59446 directories. The overlap score is computed based on the arc 59447 profiles. It is defined as the sum of min (p1_counter[i] / 59448 p1_sum_all, p2_counter[i] / p2_sum_all), for all arc counter i, 59449 where p1_counter[i] and p2_counter[i] are two matched counters and 59450 p1_sum_all and p2_sum_all are the sum of counter values in profile 59451 1 and profile 2, respectively. 59452 59453 '-f' 59454 '--function' 59455 Print function level overlap score. 59456 59457 '-F' 59458 '--fullname' 59459 Print full gcda filename. 59460 59461 '-h' 59462 '--hotonly' 59463 Only print info for hot objects/functions. 59464 59465 '-o' 59466 '--object' 59467 Print object level overlap score. 59468 59469 '-t FLOAT' 59470 '--hot_threshold <float>' 59471 Set the threshold for hot counter value. 59472 59473 '-v' 59474 '--verbose' 59475 Set the verbose mode. 59476 59477 59478File: gcc.info, Node: Gcov-dump, Next: lto-dump, Prev: Gcov-tool, Up: Top 59479 5948012 'gcov-dump'--an Offline Gcda and Gcno Profile Dump Tool 59481********************************************************** 59482 59483* Menu: 59484 59485* Gcov-dump Intro:: Introduction to gcov-dump. 59486* Invoking Gcov-dump:: How to use gcov-dump. 59487 59488 59489File: gcc.info, Node: Gcov-dump Intro, Next: Invoking Gcov-dump, Up: Gcov-dump 59490 5949112.1 Introduction to 'gcov-dump' 59492================================ 59493 59494'gcov-dump' is a tool you can use in conjunction with GCC to dump 59495content of gcda and gcno profile files offline. 59496 59497 59498File: gcc.info, Node: Invoking Gcov-dump, Prev: Gcov-dump Intro, Up: Gcov-dump 59499 5950012.2 Invoking 'gcov-dump' 59501========================= 59502 59503 Usage: gcov-dump [OPTION] ... GCOVFILES 59504 59505 'gcov-dump' accepts the following options: 59506 59507'-h' 59508'--help' 59509 Display help about using 'gcov-dump' (on the standard output), and 59510 exit without doing any further processing. 59511 59512'-l' 59513'--long' 59514 Dump content of records. 59515 59516'-p' 59517'--positions' 59518 Dump positions of records. 59519 59520'-r' 59521'--raw' 59522 Print content records in raw format. 59523 59524'-v' 59525'--version' 59526 Display the 'gcov-dump' version number (on the standard output), 59527 and exit without doing any further processing. 59528 59529 59530File: gcc.info, Node: lto-dump, Next: Trouble, Prev: Gcov-dump, Up: Top 59531 5953213 'lto-dump'--Tool for dumping LTO object files. 59533************************************************* 59534 59535* Menu: 59536 59537* lto-dump Intro:: Introduction to lto-dump. 59538* Invoking lto-dump:: How to use lto-dump. 59539 59540 59541File: gcc.info, Node: lto-dump Intro, Next: Invoking lto-dump, Up: lto-dump 59542 5954313.1 Introduction to 'lto-dump' 59544=============================== 59545 59546'lto-dump' is a tool you can use in conjunction with GCC to dump link 59547time optimization object files. 59548 59549 59550File: gcc.info, Node: Invoking lto-dump, Prev: lto-dump Intro, Up: lto-dump 59551 5955213.2 Invoking 'lto-dump' 59553======================== 59554 59555 Usage: lto-dump [OPTION] ... OBJFILES 59556 59557 'lto-dump' accepts the following options: 59558 59559'-list' 59560 Dumps list of details of functions and variables. 59561 59562'-demangle' 59563 Dump the demangled output. 59564 59565'-defined-only' 59566 Dump only the defined symbols. 59567 59568'-print-value' 59569 Dump initial values of the variables. 59570 59571'-name-sort' 59572 Sort the symbols alphabetically. 59573 59574'-size-sort' 59575 Sort the symbols according to size. 59576 59577'-reverse-sort' 59578 Dump the symbols in reverse order. 59579 59580'-no-sort' 59581 Dump the symbols in order of occurrence. 59582 59583'-symbol=' 59584 Dump the details of specific symbol. 59585 59586'-objects' 59587 Dump the details of LTO objects. 59588 59589'-type-stats' 59590 Dump the statistics of tree types. 59591 59592'-tree-stats' 59593 Dump the statistics of trees. 59594 59595'-gimple-stats' 59596 Dump the statistics of gimple statements. 59597 59598'-dump-level=' 59599 For deciding the optimization level of body. 59600 59601'-dump-body=' 59602 Dump the specific gimple body. 59603 59604'-help' 59605 Display the dump tool help. 59606 59607 59608File: gcc.info, Node: Trouble, Next: Bugs, Prev: lto-dump, Up: Top 59609 5961014 Known Causes of Trouble with GCC 59611*********************************** 59612 59613This section describes known problems that affect users of GCC. Most of 59614these are not GCC bugs per se--if they were, we would fix them. But the 59615result for a user may be like the result of a bug. 59616 59617 Some of these problems are due to bugs in other software, some are 59618missing features that are too much work to add, and some are places 59619where people's opinions differ as to what is best. 59620 59621* Menu: 59622 59623* Actual Bugs:: Bugs we will fix later. 59624* Interoperation:: Problems using GCC with other compilers, 59625 and with certain linkers, assemblers and debuggers. 59626* Incompatibilities:: GCC is incompatible with traditional C. 59627* Fixed Headers:: GCC uses corrected versions of system header files. 59628 This is necessary, but doesn't always work smoothly. 59629* Standard Libraries:: GCC uses the system C library, which might not be 59630 compliant with the ISO C standard. 59631* Disappointments:: Regrettable things we cannot change, but not quite bugs. 59632* C++ Misunderstandings:: Common misunderstandings with GNU C++. 59633* Non-bugs:: Things we think are right, but some others disagree. 59634* Warnings and Errors:: Which problems in your code get warnings, 59635 and which get errors. 59636 59637 59638File: gcc.info, Node: Actual Bugs, Next: Interoperation, Up: Trouble 59639 5964014.1 Actual Bugs We Haven't Fixed Yet 59641===================================== 59642 59643 * The 'fixincludes' script interacts badly with automounters; if the 59644 directory of system header files is automounted, it tends to be 59645 unmounted while 'fixincludes' is running. This would seem to be a 59646 bug in the automounter. We don't know any good way to work around 59647 it. 59648 59649 59650File: gcc.info, Node: Interoperation, Next: Incompatibilities, Prev: Actual Bugs, Up: Trouble 59651 5965214.2 Interoperation 59653=================== 59654 59655This section lists various difficulties encountered in using GCC 59656together with other compilers or with the assemblers, linkers, libraries 59657and debuggers on certain systems. 59658 59659 * On many platforms, GCC supports a different ABI for C++ than do 59660 other compilers, so the object files compiled by GCC cannot be used 59661 with object files generated by another C++ compiler. 59662 59663 An area where the difference is most apparent is name mangling. 59664 The use of different name mangling is intentional, to protect you 59665 from more subtle problems. Compilers differ as to many internal 59666 details of C++ implementation, including: how class instances are 59667 laid out, how multiple inheritance is implemented, and how virtual 59668 function calls are handled. If the name encoding were made the 59669 same, your programs would link against libraries provided from 59670 other compilers--but the programs would then crash when run. 59671 Incompatible libraries are then detected at link time, rather than 59672 at run time. 59673 59674 * On some BSD systems, including some versions of Ultrix, use of 59675 profiling causes static variable destructors (currently used only 59676 in C++) not to be run. 59677 59678 * On a SPARC, GCC aligns all values of type 'double' on an 8-byte 59679 boundary, and it expects every 'double' to be so aligned. The Sun 59680 compiler usually gives 'double' values 8-byte alignment, with one 59681 exception: function arguments of type 'double' may not be aligned. 59682 59683 As a result, if a function compiled with Sun CC takes the address 59684 of an argument of type 'double' and passes this pointer of type 59685 'double *' to a function compiled with GCC, dereferencing the 59686 pointer may cause a fatal signal. 59687 59688 One way to solve this problem is to compile your entire program 59689 with GCC. Another solution is to modify the function that is 59690 compiled with Sun CC to copy the argument into a local variable; 59691 local variables are always properly aligned. A third solution is 59692 to modify the function that uses the pointer to dereference it via 59693 the following function 'access_double' instead of directly with 59694 '*': 59695 59696 inline double 59697 access_double (double *unaligned_ptr) 59698 { 59699 union d2i { double d; int i[2]; }; 59700 59701 union d2i *p = (union d2i *) unaligned_ptr; 59702 union d2i u; 59703 59704 u.i[0] = p->i[0]; 59705 u.i[1] = p->i[1]; 59706 59707 return u.d; 59708 } 59709 59710 Storing into the pointer can be done likewise with the same union. 59711 59712 * On Solaris, the 'malloc' function in the 'libmalloc.a' library may 59713 allocate memory that is only 4 byte aligned. Since GCC on the 59714 SPARC assumes that doubles are 8 byte aligned, this may result in a 59715 fatal signal if doubles are stored in memory allocated by the 59716 'libmalloc.a' library. 59717 59718 The solution is to not use the 'libmalloc.a' library. Use instead 59719 'malloc' and related functions from 'libc.a'; they do not have this 59720 problem. 59721 59722 * On the HP PA machine, ADB sometimes fails to work on functions 59723 compiled with GCC. Specifically, it fails to work on functions 59724 that use 'alloca' or variable-size arrays. This is because GCC 59725 doesn't generate HP-UX unwind descriptors for such functions. It 59726 may even be impossible to generate them. 59727 59728 * Debugging ('-g') is not supported on the HP PA machine, unless you 59729 use the preliminary GNU tools. 59730 59731 * Taking the address of a label may generate errors from the HP-UX PA 59732 assembler. GAS for the PA does not have this problem. 59733 59734 * Using floating point parameters for indirect calls to static 59735 functions will not work when using the HP assembler. There simply 59736 is no way for GCC to specify what registers hold arguments for 59737 static functions when using the HP assembler. GAS for the PA does 59738 not have this problem. 59739 59740 * In extremely rare cases involving some very large functions you may 59741 receive errors from the HP linker complaining about an out of 59742 bounds unconditional branch offset. This used to occur more often 59743 in previous versions of GCC, but is now exceptionally rare. If you 59744 should run into it, you can work around by making your function 59745 smaller. 59746 59747 * GCC compiled code sometimes emits warnings from the HP-UX assembler 59748 of the form: 59749 59750 (warning) Use of GR3 when 59751 frame >= 8192 may cause conflict. 59752 59753 These warnings are harmless and can be safely ignored. 59754 59755 * In extremely rare cases involving some very large functions you may 59756 receive errors from the AIX Assembler complaining about a 59757 displacement that is too large. If you should run into it, you can 59758 work around by making your function smaller. 59759 59760 * The 'libstdc++.a' library in GCC relies on the SVR4 dynamic linker 59761 semantics which merges global symbols between libraries and 59762 applications, especially necessary for C++ streams functionality. 59763 This is not the default behavior of AIX shared libraries and 59764 dynamic linking. 'libstdc++.a' is built on AIX with 59765 "runtime-linking" enabled so that symbol merging can occur. To 59766 utilize this feature, the application linked with 'libstdc++.a' 59767 must include the '-Wl,-brtl' flag on the link line. G++ cannot 59768 impose this because this option may interfere with the semantics of 59769 the user program and users may not always use 'g++' to link his or 59770 her application. Applications are not required to use the 59771 '-Wl,-brtl' flag on the link line--the rest of the 'libstdc++.a' 59772 library which is not dependent on the symbol merging semantics will 59773 continue to function correctly. 59774 59775 * An application can interpose its own definition of functions for 59776 functions invoked by 'libstdc++.a' with "runtime-linking" enabled 59777 on AIX. To accomplish this the application must be linked with 59778 "runtime-linking" option and the functions explicitly must be 59779 exported by the application ('-Wl,-brtl,-bE:exportfile'). 59780 59781 * AIX on the RS/6000 provides support (NLS) for environments outside 59782 of the United States. Compilers and assemblers use NLS to support 59783 locale-specific representations of various objects including 59784 floating-point numbers ('.' vs ',' for separating decimal 59785 fractions). There have been problems reported where the library 59786 linked with GCC does not produce the same floating-point formats 59787 that the assembler accepts. If you have this problem, set the 59788 'LANG' environment variable to 'C' or 'En_US'. 59789 59790 * Even if you specify '-fdollars-in-identifiers', you cannot 59791 successfully use '$' in identifiers on the RS/6000 due to a 59792 restriction in the IBM assembler. GAS supports these identifiers. 59793 59794 59795File: gcc.info, Node: Incompatibilities, Next: Fixed Headers, Prev: Interoperation, Up: Trouble 59796 5979714.3 Incompatibilities of GCC 59798============================= 59799 59800There are several noteworthy incompatibilities between GNU C and K&R 59801(non-ISO) versions of C. 59802 59803 * GCC normally makes string constants read-only. If several 59804 identical-looking string constants are used, GCC stores only one 59805 copy of the string. 59806 59807 One consequence is that you cannot call 'mktemp' with a string 59808 constant argument. The function 'mktemp' always alters the string 59809 its argument points to. 59810 59811 Another consequence is that 'sscanf' does not work on some very old 59812 systems when passed a string constant as its format control string 59813 or input. This is because 'sscanf' incorrectly tries to write into 59814 the string constant. Likewise 'fscanf' and 'scanf'. 59815 59816 The solution to these problems is to change the program to use 59817 'char'-array variables with initialization strings for these 59818 purposes instead of string constants. 59819 59820 * '-2147483648' is positive. 59821 59822 This is because 2147483648 cannot fit in the type 'int', so 59823 (following the ISO C rules) its data type is 'unsigned long int'. 59824 Negating this value yields 2147483648 again. 59825 59826 * GCC does not substitute macro arguments when they appear inside of 59827 string constants. For example, the following macro in GCC 59828 59829 #define foo(a) "a" 59830 59831 will produce output '"a"' regardless of what the argument A is. 59832 59833 * When you use 'setjmp' and 'longjmp', the only automatic variables 59834 guaranteed to remain valid are those declared 'volatile'. This is 59835 a consequence of automatic register allocation. Consider this 59836 function: 59837 59838 jmp_buf j; 59839 59840 foo () 59841 { 59842 int a, b; 59843 59844 a = fun1 (); 59845 if (setjmp (j)) 59846 return a; 59847 59848 a = fun2 (); 59849 /* 'longjmp (j)' may occur in 'fun3'. */ 59850 return a + fun3 (); 59851 } 59852 59853 Here 'a' may or may not be restored to its first value when the 59854 'longjmp' occurs. If 'a' is allocated in a register, then its 59855 first value is restored; otherwise, it keeps the last value stored 59856 in it. 59857 59858 If you use the '-W' option with the '-O' option, you will get a 59859 warning when GCC thinks such a problem might be possible. 59860 59861 * Programs that use preprocessing directives in the middle of macro 59862 arguments do not work with GCC. For example, a program like this 59863 will not work: 59864 59865 foobar ( 59866 #define luser 59867 hack) 59868 59869 ISO C does not permit such a construct. 59870 59871 * K&R compilers allow comments to cross over an inclusion boundary 59872 (i.e. started in an include file and ended in the including file). 59873 59874 * Declarations of external variables and functions within a block 59875 apply only to the block containing the declaration. In other 59876 words, they have the same scope as any other declaration in the 59877 same place. 59878 59879 In some other C compilers, an 'extern' declaration affects all the 59880 rest of the file even if it happens within a block. 59881 59882 * In traditional C, you can combine 'long', etc., with a typedef 59883 name, as shown here: 59884 59885 typedef int foo; 59886 typedef long foo bar; 59887 59888 In ISO C, this is not allowed: 'long' and other type modifiers 59889 require an explicit 'int'. 59890 59891 * PCC allows typedef names to be used as function parameters. 59892 59893 * Traditional C allows the following erroneous pair of declarations 59894 to appear together in a given scope: 59895 59896 typedef int foo; 59897 typedef foo foo; 59898 59899 * GCC treats all characters of identifiers as significant. According 59900 to K&R-1 (2.2), "No more than the first eight characters are 59901 significant, although more may be used.". Also according to K&R-1 59902 (2.2), "An identifier is a sequence of letters and digits; the 59903 first character must be a letter. The underscore _ counts as a 59904 letter.", but GCC also allows dollar signs in identifiers. 59905 59906 * PCC allows whitespace in the middle of compound assignment 59907 operators such as '+='. GCC, following the ISO standard, does not 59908 allow this. 59909 59910 * GCC complains about unterminated character constants inside of 59911 preprocessing conditionals that fail. Some programs have English 59912 comments enclosed in conditionals that are guaranteed to fail; if 59913 these comments contain apostrophes, GCC will probably report an 59914 error. For example, this code would produce an error: 59915 59916 #if 0 59917 You can't expect this to work. 59918 #endif 59919 59920 The best solution to such a problem is to put the text into an 59921 actual C comment delimited by '/*...*/'. 59922 59923 * Many user programs contain the declaration 'long time ();'. In the 59924 past, the system header files on many systems did not actually 59925 declare 'time', so it did not matter what type your program 59926 declared it to return. But in systems with ISO C headers, 'time' 59927 is declared to return 'time_t', and if that is not the same as 59928 'long', then 'long time ();' is erroneous. 59929 59930 The solution is to change your program to use appropriate system 59931 headers ('<time.h>' on systems with ISO C headers) and not to 59932 declare 'time' if the system header files declare it, or failing 59933 that to use 'time_t' as the return type of 'time'. 59934 59935 * When compiling functions that return 'float', PCC converts it to a 59936 double. GCC actually returns a 'float'. If you are concerned with 59937 PCC compatibility, you should declare your functions to return 59938 'double'; you might as well say what you mean. 59939 59940 * When compiling functions that return structures or unions, GCC 59941 output code normally uses a method different from that used on most 59942 versions of Unix. As a result, code compiled with GCC cannot call 59943 a structure-returning function compiled with PCC, and vice versa. 59944 59945 The method used by GCC is as follows: a structure or union which is 59946 1, 2, 4 or 8 bytes long is returned like a scalar. A structure or 59947 union with any other size is stored into an address supplied by the 59948 caller (usually in a special, fixed register, but on some machines 59949 it is passed on the stack). The target hook 59950 'TARGET_STRUCT_VALUE_RTX' tells GCC where to pass this address. 59951 59952 By contrast, PCC on most target machines returns structures and 59953 unions of any size by copying the data into an area of static 59954 storage, and then returning the address of that storage as if it 59955 were a pointer value. The caller must copy the data from that 59956 memory area to the place where the value is wanted. GCC does not 59957 use this method because it is slower and nonreentrant. 59958 59959 On some newer machines, PCC uses a reentrant convention for all 59960 structure and union returning. GCC on most of these machines uses 59961 a compatible convention when returning structures and unions in 59962 memory, but still returns small structures and unions in registers. 59963 59964 You can tell GCC to use a compatible convention for all structure 59965 and union returning with the option '-fpcc-struct-return'. 59966 59967 * GCC complains about program fragments such as '0x74ae-0x4000' which 59968 appear to be two hexadecimal constants separated by the minus 59969 operator. Actually, this string is a single "preprocessing token". 59970 Each such token must correspond to one token in C. Since this does 59971 not, GCC prints an error message. Although it may appear obvious 59972 that what is meant is an operator and two values, the ISO C 59973 standard specifically requires that this be treated as erroneous. 59974 59975 A "preprocessing token" is a "preprocessing number" if it begins 59976 with a digit and is followed by letters, underscores, digits, 59977 periods and 'e+', 'e-', 'E+', 'E-', 'p+', 'p-', 'P+', or 'P-' 59978 character sequences. (In strict C90 mode, the sequences 'p+', 59979 'p-', 'P+' and 'P-' cannot appear in preprocessing numbers.) 59980 59981 To make the above program fragment valid, place whitespace in front 59982 of the minus sign. This whitespace will end the preprocessing 59983 number. 59984 59985 59986File: gcc.info, Node: Fixed Headers, Next: Standard Libraries, Prev: Incompatibilities, Up: Trouble 59987 5998814.4 Fixed Header Files 59989======================= 59990 59991GCC needs to install corrected versions of some system header files. 59992This is because most target systems have some header files that won't 59993work with GCC unless they are changed. Some have bugs, some are 59994incompatible with ISO C, and some depend on special features of other 59995compilers. 59996 59997 Installing GCC automatically creates and installs the fixed header 59998files, by running a program called 'fixincludes'. Normally, you don't 59999need to pay attention to this. But there are cases where it doesn't do 60000the right thing automatically. 60001 60002 * If you update the system's header files, such as by installing a 60003 new system version, the fixed header files of GCC are not 60004 automatically updated. They can be updated using the 'mkheaders' 60005 script installed in 'LIBEXECDIR/gcc/TARGET/VERSION/install-tools/'. 60006 60007 * On some systems, header file directories contain machine-specific 60008 symbolic links in certain places. This makes it possible to share 60009 most of the header files among hosts running the same version of 60010 the system on different machine models. 60011 60012 The programs that fix the header files do not understand this 60013 special way of using symbolic links; therefore, the directory of 60014 fixed header files is good only for the machine model used to build 60015 it. 60016 60017 It is possible to make separate sets of fixed header files for the 60018 different machine models, and arrange a structure of symbolic links 60019 so as to use the proper set, but you'll have to do this by hand. 60020 60021 60022File: gcc.info, Node: Standard Libraries, Next: Disappointments, Prev: Fixed Headers, Up: Trouble 60023 6002414.5 Standard Libraries 60025======================= 60026 60027GCC by itself attempts to be a conforming freestanding implementation. 60028*Note Language Standards Supported by GCC: Standards, for details of 60029what this means. Beyond the library facilities required of such an 60030implementation, the rest of the C library is supplied by the vendor of 60031the operating system. If that C library doesn't conform to the C 60032standards, then your programs might get warnings (especially when using 60033'-Wall') that you don't expect. 60034 60035 For example, the 'sprintf' function on SunOS 4.1.3 returns 'char *' 60036while the C standard says that 'sprintf' returns an 'int'. The 60037'fixincludes' program could make the prototype for this function match 60038the Standard, but that would be wrong, since the function will still 60039return 'char *'. 60040 60041 If you need a Standard compliant library, then you need to find one, as 60042GCC does not provide one. The GNU C library (called 'glibc') provides 60043ISO C, POSIX, BSD, SystemV and X/Open compatibility for GNU/Linux and 60044HURD-based GNU systems; no recent version of it supports other systems, 60045though some very old versions did. Version 2.2 of the GNU C library 60046includes nearly complete C99 support. You could also ask your operating 60047system vendor if newer libraries are available. 60048 60049 60050File: gcc.info, Node: Disappointments, Next: C++ Misunderstandings, Prev: Standard Libraries, Up: Trouble 60051 6005214.6 Disappointments and Misunderstandings 60053========================================== 60054 60055These problems are perhaps regrettable, but we don't know any practical 60056way around them. 60057 60058 * Certain local variables aren't recognized by debuggers when you 60059 compile with optimization. 60060 60061 This occurs because sometimes GCC optimizes the variable out of 60062 existence. There is no way to tell the debugger how to compute the 60063 value such a variable "would have had", and it is not clear that 60064 would be desirable anyway. So GCC simply does not mention the 60065 eliminated variable when it writes debugging information. 60066 60067 You have to expect a certain amount of disagreement between the 60068 executable and your source code, when you use optimization. 60069 60070 * Users often think it is a bug when GCC reports an error for code 60071 like this: 60072 60073 int foo (struct mumble *); 60074 60075 struct mumble { ... }; 60076 60077 int foo (struct mumble *x) 60078 { ... } 60079 60080 This code really is erroneous, because the scope of 'struct mumble' 60081 in the prototype is limited to the argument list containing it. It 60082 does not refer to the 'struct mumble' defined with file scope 60083 immediately below--they are two unrelated types with similar names 60084 in different scopes. 60085 60086 But in the definition of 'foo', the file-scope type is used because 60087 that is available to be inherited. Thus, the definition and the 60088 prototype do not match, and you get an error. 60089 60090 This behavior may seem silly, but it's what the ISO standard 60091 specifies. It is easy enough for you to make your code work by 60092 moving the definition of 'struct mumble' above the prototype. It's 60093 not worth being incompatible with ISO C just to avoid an error for 60094 the example shown above. 60095 60096 * Accesses to bit-fields even in volatile objects works by accessing 60097 larger objects, such as a byte or a word. You cannot rely on what 60098 size of object is accessed in order to read or write the bit-field; 60099 it may even vary for a given bit-field according to the precise 60100 usage. 60101 60102 If you care about controlling the amount of memory that is 60103 accessed, use volatile but do not use bit-fields. 60104 60105 * GCC comes with shell scripts to fix certain known problems in 60106 system header files. They install corrected copies of various 60107 header files in a special directory where only GCC will normally 60108 look for them. The scripts adapt to various systems by searching 60109 all the system header files for the problem cases that we know 60110 about. 60111 60112 If new system header files are installed, nothing automatically 60113 arranges to update the corrected header files. They can be updated 60114 using the 'mkheaders' script installed in 60115 'LIBEXECDIR/gcc/TARGET/VERSION/install-tools/'. 60116 60117 * On 68000 and x86 systems, for instance, you can get paradoxical 60118 results if you test the precise values of floating point numbers. 60119 For example, you can find that a floating point value which is not 60120 a NaN is not equal to itself. This results from the fact that the 60121 floating point registers hold a few more bits of precision than fit 60122 in a 'double' in memory. Compiled code moves values between memory 60123 and floating point registers at its convenience, and moving them 60124 into memory truncates them. 60125 60126 You can partially avoid this problem by using the '-ffloat-store' 60127 option (*note Optimize Options::). 60128 60129 * On AIX and other platforms without weak symbol support, templates 60130 need to be instantiated explicitly and symbols for static members 60131 of templates will not be generated. 60132 60133 * On AIX, GCC scans object files and library archives for static 60134 constructors and destructors when linking an application before the 60135 linker prunes unreferenced symbols. This is necessary to prevent 60136 the AIX linker from mistakenly assuming that static constructor or 60137 destructor are unused and removing them before the scanning can 60138 occur. All static constructors and destructors found will be 60139 referenced even though the modules in which they occur may not be 60140 used by the program. This may lead to both increased executable 60141 size and unexpected symbol references. 60142 60143 60144File: gcc.info, Node: C++ Misunderstandings, Next: Non-bugs, Prev: Disappointments, Up: Trouble 60145 6014614.7 Common Misunderstandings with GNU C++ 60147========================================== 60148 60149C++ is a complex language and an evolving one, and its standard 60150definition (the ISO C++ standard) was only recently completed. As a 60151result, your C++ compiler may occasionally surprise you, even when its 60152behavior is correct. This section discusses some areas that frequently 60153give rise to questions of this sort. 60154 60155* Menu: 60156 60157* Static Definitions:: Static member declarations are not definitions 60158* Name lookup:: Name lookup, templates, and accessing members of base classes 60159* Temporaries:: Temporaries may vanish before you expect 60160* Copy Assignment:: Copy Assignment operators copy virtual bases twice 60161 60162 60163File: gcc.info, Node: Static Definitions, Next: Name lookup, Up: C++ Misunderstandings 60164 6016514.7.1 Declare _and_ Define Static Members 60166------------------------------------------ 60167 60168When a class has static data members, it is not enough to _declare_ the 60169static member; you must also _define_ it. For example: 60170 60171 class Foo 60172 { 60173 ... 60174 void method(); 60175 static int bar; 60176 }; 60177 60178 This declaration only establishes that the class 'Foo' has an 'int' 60179named 'Foo::bar', and a member function named 'Foo::method'. But you 60180still need to define _both_ 'method' and 'bar' elsewhere. According to 60181the ISO standard, you must supply an initializer in one (and only one) 60182source file, such as: 60183 60184 int Foo::bar = 0; 60185 60186 Other C++ compilers may not correctly implement the standard behavior. 60187As a result, when you switch to 'g++' from one of these compilers, you 60188may discover that a program that appeared to work correctly in fact does 60189not conform to the standard: 'g++' reports as undefined symbols any 60190static data members that lack definitions. 60191 60192 60193File: gcc.info, Node: Name lookup, Next: Temporaries, Prev: Static Definitions, Up: C++ Misunderstandings 60194 6019514.7.2 Name Lookup, Templates, and Accessing Members of Base Classes 60196-------------------------------------------------------------------- 60197 60198The C++ standard prescribes that all names that are not dependent on 60199template parameters are bound to their present definitions when parsing 60200a template function or class.(1) Only names that are dependent are 60201looked up at the point of instantiation. For example, consider 60202 60203 void foo(double); 60204 60205 struct A { 60206 template <typename T> 60207 void f () { 60208 foo (1); // 1 60209 int i = N; // 2 60210 T t; 60211 t.bar(); // 3 60212 foo (t); // 4 60213 } 60214 60215 static const int N; 60216 }; 60217 60218 Here, the names 'foo' and 'N' appear in a context that does not depend 60219on the type of 'T'. The compiler will thus require that they are 60220defined in the context of use in the template, not only before the point 60221of instantiation, and will here use '::foo(double)' and 'A::N', 60222respectively. In particular, it will convert the integer value to a 60223'double' when passing it to '::foo(double)'. 60224 60225 Conversely, 'bar' and the call to 'foo' in the fourth marked line are 60226used in contexts that do depend on the type of 'T', so they are only 60227looked up at the point of instantiation, and you can provide 60228declarations for them after declaring the template, but before 60229instantiating it. In particular, if you instantiate 'A::f<int>', the 60230last line will call an overloaded '::foo(int)' if one was provided, even 60231if after the declaration of 'struct A'. 60232 60233 This distinction between lookup of dependent and non-dependent names is 60234called two-stage (or dependent) name lookup. G++ implements it since 60235version 3.4. 60236 60237 Two-stage name lookup sometimes leads to situations with behavior 60238different from non-template codes. The most common is probably this: 60239 60240 template <typename T> struct Base { 60241 int i; 60242 }; 60243 60244 template <typename T> struct Derived : public Base<T> { 60245 int get_i() { return i; } 60246 }; 60247 60248 In 'get_i()', 'i' is not used in a dependent context, so the compiler 60249will look for a name declared at the enclosing namespace scope (which is 60250the global scope here). It will not look into the base class, since 60251that is dependent and you may declare specializations of 'Base' even 60252after declaring 'Derived', so the compiler cannot really know what 'i' 60253would refer to. If there is no global variable 'i', then you will get 60254an error message. 60255 60256 In order to make it clear that you want the member of the base class, 60257you need to defer lookup until instantiation time, at which the base 60258class is known. For this, you need to access 'i' in a dependent 60259context, by either using 'this->i' (remember that 'this' is of type 60260'Derived<T>*', so is obviously dependent), or using 'Base<T>::i'. 60261Alternatively, 'Base<T>::i' might be brought into scope by a 60262'using'-declaration. 60263 60264 Another, similar example involves calling member functions of a base 60265class: 60266 60267 template <typename T> struct Base { 60268 int f(); 60269 }; 60270 60271 template <typename T> struct Derived : Base<T> { 60272 int g() { return f(); }; 60273 }; 60274 60275 Again, the call to 'f()' is not dependent on template arguments (there 60276are no arguments that depend on the type 'T', and it is also not 60277otherwise specified that the call should be in a dependent context). 60278Thus a global declaration of such a function must be available, since 60279the one in the base class is not visible until instantiation time. The 60280compiler will consequently produce the following error message: 60281 60282 x.cc: In member function `int Derived<T>::g()': 60283 x.cc:6: error: there are no arguments to `f' that depend on a template 60284 parameter, so a declaration of `f' must be available 60285 x.cc:6: error: (if you use `-fpermissive', G++ will accept your code, but 60286 allowing the use of an undeclared name is deprecated) 60287 60288 To make the code valid either use 'this->f()', or 'Base<T>::f()'. 60289Using the '-fpermissive' flag will also let the compiler accept the 60290code, by marking all function calls for which no declaration is visible 60291at the time of definition of the template for later lookup at 60292instantiation time, as if it were a dependent call. We do not recommend 60293using '-fpermissive' to work around invalid code, and it will also only 60294catch cases where functions in base classes are called, not where 60295variables in base classes are used (as in the example above). 60296 60297 Note that some compilers (including G++ versions prior to 3.4) get 60298these examples wrong and accept above code without an error. Those 60299compilers do not implement two-stage name lookup correctly. 60300 60301 ---------- Footnotes ---------- 60302 60303 (1) The C++ standard just uses the term "dependent" for names that 60304depend on the type or value of template parameters. This shorter term 60305will also be used in the rest of this section. 60306 60307 60308File: gcc.info, Node: Temporaries, Next: Copy Assignment, Prev: Name lookup, Up: C++ Misunderstandings 60309 6031014.7.3 Temporaries May Vanish Before You Expect 60311----------------------------------------------- 60312 60313It is dangerous to use pointers or references to _portions_ of a 60314temporary object. The compiler may very well delete the object before 60315you expect it to, leaving a pointer to garbage. The most common place 60316where this problem crops up is in classes like string classes, 60317especially ones that define a conversion function to type 'char *' or 60318'const char *'--which is one reason why the standard 'string' class 60319requires you to call the 'c_str' member function. However, any class 60320that returns a pointer to some internal structure is potentially subject 60321to this problem. 60322 60323 For example, a program may use a function 'strfunc' that returns 60324'string' objects, and another function 'charfunc' that operates on 60325pointers to 'char': 60326 60327 string strfunc (); 60328 void charfunc (const char *); 60329 60330 void 60331 f () 60332 { 60333 const char *p = strfunc().c_str(); 60334 ... 60335 charfunc (p); 60336 ... 60337 charfunc (p); 60338 } 60339 60340In this situation, it may seem reasonable to save a pointer to the C 60341string returned by the 'c_str' member function and use that rather than 60342call 'c_str' repeatedly. However, the temporary string created by the 60343call to 'strfunc' is destroyed after 'p' is initialized, at which point 60344'p' is left pointing to freed memory. 60345 60346 Code like this may run successfully under some other compilers, 60347particularly obsolete cfront-based compilers that delete temporaries 60348along with normal local variables. However, the GNU C++ behavior is 60349standard-conforming, so if your program depends on late destruction of 60350temporaries it is not portable. 60351 60352 The safe way to write such code is to give the temporary a name, which 60353forces it to remain until the end of the scope of the name. For 60354example: 60355 60356 const string& tmp = strfunc (); 60357 charfunc (tmp.c_str ()); 60358 60359 60360File: gcc.info, Node: Copy Assignment, Prev: Temporaries, Up: C++ Misunderstandings 60361 6036214.7.4 Implicit Copy-Assignment for Virtual Bases 60363------------------------------------------------- 60364 60365When a base class is virtual, only one subobject of the base class 60366belongs to each full object. Also, the constructors and destructors are 60367invoked only once, and called from the most-derived class. However, 60368such objects behave unspecified when being assigned. For example: 60369 60370 struct Base{ 60371 char *name; 60372 Base(const char *n) : name(strdup(n)){} 60373 Base& operator= (const Base& other){ 60374 free (name); 60375 name = strdup (other.name); 60376 return *this; 60377 } 60378 }; 60379 60380 struct A:virtual Base{ 60381 int val; 60382 A():Base("A"){} 60383 }; 60384 60385 struct B:virtual Base{ 60386 int bval; 60387 B():Base("B"){} 60388 }; 60389 60390 struct Derived:public A, public B{ 60391 Derived():Base("Derived"){} 60392 }; 60393 60394 void func(Derived &d1, Derived &d2) 60395 { 60396 d1 = d2; 60397 } 60398 60399 The C++ standard specifies that 'Base::Base' is only called once when 60400constructing or copy-constructing a Derived object. It is unspecified 60401whether 'Base::operator=' is called more than once when the implicit 60402copy-assignment for Derived objects is invoked (as it is inside 'func' 60403in the example). 60404 60405 G++ implements the "intuitive" algorithm for copy-assignment: assign 60406all direct bases, then assign all members. In that algorithm, the 60407virtual base subobject can be encountered more than once. In the 60408example, copying proceeds in the following order: 'name' (via 'strdup'), 60409'val', 'name' again, and 'bval'. 60410 60411 If application code relies on copy-assignment, a user-defined 60412copy-assignment operator removes any uncertainties. With such an 60413operator, the application can define whether and how the virtual base 60414subobject is assigned. 60415 60416 60417File: gcc.info, Node: Non-bugs, Next: Warnings and Errors, Prev: C++ Misunderstandings, Up: Trouble 60418 6041914.8 Certain Changes We Don't Want to Make 60420========================================== 60421 60422This section lists changes that people frequently request, but which we 60423do not make because we think GCC is better without them. 60424 60425 * Checking the number and type of arguments to a function which has 60426 an old-fashioned definition and no prototype. 60427 60428 Such a feature would work only occasionally--only for calls that 60429 appear in the same file as the called function, following the 60430 definition. The only way to check all calls reliably is to add a 60431 prototype for the function. But adding a prototype eliminates the 60432 motivation for this feature. So the feature is not worthwhile. 60433 60434 * Warning about using an expression whose type is signed as a shift 60435 count. 60436 60437 Shift count operands are probably signed more often than unsigned. 60438 Warning about this would cause far more annoyance than good. 60439 60440 * Warning about assigning a signed value to an unsigned variable. 60441 60442 Such assignments must be very common; warning about them would 60443 cause more annoyance than good. 60444 60445 * Warning when a non-void function value is ignored. 60446 60447 C contains many standard functions that return a value that most 60448 programs choose to ignore. One obvious example is 'printf'. 60449 Warning about this practice only leads the defensive programmer to 60450 clutter programs with dozens of casts to 'void'. Such casts are 60451 required so frequently that they become visual noise. Writing 60452 those casts becomes so automatic that they no longer convey useful 60453 information about the intentions of the programmer. For functions 60454 where the return value should never be ignored, use the 60455 'warn_unused_result' function attribute (*note Function 60456 Attributes::). 60457 60458 * Making '-fshort-enums' the default. 60459 60460 This would cause storage layout to be incompatible with most other 60461 C compilers. And it doesn't seem very important, given that you 60462 can get the same result in other ways. The case where it matters 60463 most is when the enumeration-valued object is inside a structure, 60464 and in that case you can specify a field width explicitly. 60465 60466 * Making bit-fields unsigned by default on particular machines where 60467 "the ABI standard" says to do so. 60468 60469 The ISO C standard leaves it up to the implementation whether a 60470 bit-field declared plain 'int' is signed or not. This in effect 60471 creates two alternative dialects of C. 60472 60473 The GNU C compiler supports both dialects; you can specify the 60474 signed dialect with '-fsigned-bitfields' and the unsigned dialect 60475 with '-funsigned-bitfields'. However, this leaves open the 60476 question of which dialect to use by default. 60477 60478 Currently, the preferred dialect makes plain bit-fields signed, 60479 because this is simplest. Since 'int' is the same as 'signed int' 60480 in every other context, it is cleanest for them to be the same in 60481 bit-fields as well. 60482 60483 Some computer manufacturers have published Application Binary 60484 Interface standards which specify that plain bit-fields should be 60485 unsigned. It is a mistake, however, to say anything about this 60486 issue in an ABI. This is because the handling of plain bit-fields 60487 distinguishes two dialects of C. Both dialects are meaningful on 60488 every type of machine. Whether a particular object file was 60489 compiled using signed bit-fields or unsigned is of no concern to 60490 other object files, even if they access the same bit-fields in the 60491 same data structures. 60492 60493 A given program is written in one or the other of these two 60494 dialects. The program stands a chance to work on most any machine 60495 if it is compiled with the proper dialect. It is unlikely to work 60496 at all if compiled with the wrong dialect. 60497 60498 Many users appreciate the GNU C compiler because it provides an 60499 environment that is uniform across machines. These users would be 60500 inconvenienced if the compiler treated plain bit-fields differently 60501 on certain machines. 60502 60503 Occasionally users write programs intended only for a particular 60504 machine type. On these occasions, the users would benefit if the 60505 GNU C compiler were to support by default the same dialect as the 60506 other compilers on that machine. But such applications are rare. 60507 And users writing a program to run on more than one type of machine 60508 cannot possibly benefit from this kind of compatibility. 60509 60510 This is why GCC does and will treat plain bit-fields in the same 60511 fashion on all types of machines (by default). 60512 60513 There are some arguments for making bit-fields unsigned by default 60514 on all machines. If, for example, this becomes a universal de 60515 facto standard, it would make sense for GCC to go along with it. 60516 This is something to be considered in the future. 60517 60518 (Of course, users strongly concerned about portability should 60519 indicate explicitly in each bit-field whether it is signed or not. 60520 In this way, they write programs which have the same meaning in 60521 both C dialects.) 60522 60523 * Undefining '__STDC__' when '-ansi' is not used. 60524 60525 Currently, GCC defines '__STDC__' unconditionally. This provides 60526 good results in practice. 60527 60528 Programmers normally use conditionals on '__STDC__' to ask whether 60529 it is safe to use certain features of ISO C, such as function 60530 prototypes or ISO token concatenation. Since plain 'gcc' supports 60531 all the features of ISO C, the correct answer to these questions is 60532 "yes". 60533 60534 Some users try to use '__STDC__' to check for the availability of 60535 certain library facilities. This is actually incorrect usage in an 60536 ISO C program, because the ISO C standard says that a conforming 60537 freestanding implementation should define '__STDC__' even though it 60538 does not have the library facilities. 'gcc -ansi -pedantic' is a 60539 conforming freestanding implementation, and it is therefore 60540 required to define '__STDC__', even though it does not come with an 60541 ISO C library. 60542 60543 Sometimes people say that defining '__STDC__' in a compiler that 60544 does not completely conform to the ISO C standard somehow violates 60545 the standard. This is illogical. The standard is a standard for 60546 compilers that claim to support ISO C, such as 'gcc -ansi'--not for 60547 other compilers such as plain 'gcc'. Whatever the ISO C standard 60548 says is relevant to the design of plain 'gcc' without '-ansi' only 60549 for pragmatic reasons, not as a requirement. 60550 60551 GCC normally defines '__STDC__' to be 1, and in addition defines 60552 '__STRICT_ANSI__' if you specify the '-ansi' option, or a '-std' 60553 option for strict conformance to some version of ISO C. On some 60554 hosts, system include files use a different convention, where 60555 '__STDC__' is normally 0, but is 1 if the user specifies strict 60556 conformance to the C Standard. GCC follows the host convention 60557 when processing system include files, but when processing user 60558 files it follows the usual GNU C convention. 60559 60560 * Undefining '__STDC__' in C++. 60561 60562 Programs written to compile with C++-to-C translators get the value 60563 of '__STDC__' that goes with the C compiler that is subsequently 60564 used. These programs must test '__STDC__' to determine what kind 60565 of C preprocessor that compiler uses: whether they should 60566 concatenate tokens in the ISO C fashion or in the traditional 60567 fashion. 60568 60569 These programs work properly with GNU C++ if '__STDC__' is defined. 60570 They would not work otherwise. 60571 60572 In addition, many header files are written to provide prototypes in 60573 ISO C but not in traditional C. Many of these header files can 60574 work without change in C++ provided '__STDC__' is defined. If 60575 '__STDC__' is not defined, they will all fail, and will all need to 60576 be changed to test explicitly for C++ as well. 60577 60578 * Deleting "empty" loops. 60579 60580 Historically, GCC has not deleted "empty" loops under the 60581 assumption that the most likely reason you would put one in a 60582 program is to have a delay, so deleting them will not make real 60583 programs run any faster. 60584 60585 However, the rationale here is that optimization of a nonempty loop 60586 cannot produce an empty one. This held for carefully written C 60587 compiled with less powerful optimizers but is not always the case 60588 for carefully written C++ or with more powerful optimizers. Thus 60589 GCC will remove operations from loops whenever it can determine 60590 those operations are not externally visible (apart from the time 60591 taken to execute them, of course). In case the loop can be proved 60592 to be finite, GCC will also remove the loop itself. 60593 60594 Be aware of this when performing timing tests, for instance the 60595 following loop can be completely removed, provided 60596 'some_expression' can provably not change any global state. 60597 60598 { 60599 int sum = 0; 60600 int ix; 60601 60602 for (ix = 0; ix != 10000; ix++) 60603 sum += some_expression; 60604 } 60605 60606 Even though 'sum' is accumulated in the loop, no use is made of 60607 that summation, so the accumulation can be removed. 60608 60609 * Making side effects happen in the same order as in some other 60610 compiler. 60611 60612 It is never safe to depend on the order of evaluation of side 60613 effects. For example, a function call like this may very well 60614 behave differently from one compiler to another: 60615 60616 void func (int, int); 60617 60618 int i = 2; 60619 func (i++, i++); 60620 60621 There is no guarantee (in either the C or the C++ standard language 60622 definitions) that the increments will be evaluated in any 60623 particular order. Either increment might happen first. 'func' 60624 might get the arguments '2, 3', or it might get '3, 2', or even '2, 60625 2'. 60626 60627 * Making certain warnings into errors by default. 60628 60629 Some ISO C testsuites report failure when the compiler does not 60630 produce an error message for a certain program. 60631 60632 ISO C requires a "diagnostic" message for certain kinds of invalid 60633 programs, but a warning is defined by GCC to count as a diagnostic. 60634 If GCC produces a warning but not an error, that is correct ISO C 60635 support. If testsuites call this "failure", they should be run 60636 with the GCC option '-pedantic-errors', which will turn these 60637 warnings into errors. 60638 60639 60640File: gcc.info, Node: Warnings and Errors, Prev: Non-bugs, Up: Trouble 60641 6064214.9 Warning Messages and Error Messages 60643======================================== 60644 60645The GNU compiler can produce two kinds of diagnostics: errors and 60646warnings. Each kind has a different purpose: 60647 60648 "Errors" report problems that make it impossible to compile your 60649 program. GCC reports errors with the source file name and line 60650 number where the problem is apparent. 60651 60652 "Warnings" report other unusual conditions in your code that _may_ 60653 indicate a problem, although compilation can (and does) proceed. 60654 Warning messages also report the source file name and line number, 60655 but include the text 'warning:' to distinguish them from error 60656 messages. 60657 60658 Warnings may indicate danger points where you should check to make sure 60659that your program really does what you intend; or the use of obsolete 60660features; or the use of nonstandard features of GNU C or C++. Many 60661warnings are issued only if you ask for them, with one of the '-W' 60662options (for instance, '-Wall' requests a variety of useful warnings). 60663 60664 GCC always tries to compile your program if possible; it never 60665gratuitously rejects a program whose meaning is clear merely because 60666(for instance) it fails to conform to a standard. In some cases, 60667however, the C and C++ standards specify that certain extensions are 60668forbidden, and a diagnostic _must_ be issued by a conforming compiler. 60669The '-pedantic' option tells GCC to issue warnings in such cases; 60670'-pedantic-errors' says to make them errors instead. This does not mean 60671that _all_ non-ISO constructs get warnings or errors. 60672 60673 *Note Options to Request or Suppress Warnings: Warning Options, for 60674more detail on these and related command-line options. 60675 60676 60677File: gcc.info, Node: Bugs, Next: Service, Prev: Trouble, Up: Top 60678 6067915 Reporting Bugs 60680***************** 60681 60682Your bug reports play an essential role in making GCC reliable. 60683 60684 When you encounter a problem, the first thing to do is to see if it is 60685already known. *Note Trouble::. If it isn't known, then you should 60686report the problem. 60687 60688* Menu: 60689 60690* Criteria: Bug Criteria. Have you really found a bug? 60691* Reporting: Bug Reporting. How to report a bug effectively. 60692 60693 60694File: gcc.info, Node: Bug Criteria, Next: Bug Reporting, Up: Bugs 60695 6069615.1 Have You Found a Bug? 60697========================== 60698 60699If you are not sure whether you have found a bug, here are some 60700guidelines: 60701 60702 * If the compiler gets a fatal signal, for any input whatever, that 60703 is a compiler bug. Reliable compilers never crash. 60704 60705 * If the compiler produces invalid assembly code, for any input 60706 whatever (except an 'asm' statement), that is a compiler bug, 60707 unless the compiler reports errors (not just warnings) which would 60708 ordinarily prevent the assembler from being run. 60709 60710 * If the compiler produces valid assembly code that does not 60711 correctly execute the input source code, that is a compiler bug. 60712 60713 However, you must double-check to make sure, because you may have a 60714 program whose behavior is undefined, which happened by chance to 60715 give the desired results with another C or C++ compiler. 60716 60717 For example, in many nonoptimizing compilers, you can write 'x;' at 60718 the end of a function instead of 'return x;', with the same 60719 results. But the value of the function is undefined if 'return' is 60720 omitted; it is not a bug when GCC produces different results. 60721 60722 Problems often result from expressions with two increment 60723 operators, as in 'f (*p++, *p++)'. Your previous compiler might 60724 have interpreted that expression the way you intended; GCC might 60725 interpret it another way. Neither compiler is wrong. The bug is 60726 in your code. 60727 60728 After you have localized the error to a single source line, it 60729 should be easy to check for these things. If your program is 60730 correct and well defined, you have found a compiler bug. 60731 60732 * If the compiler produces an error message for valid input, that is 60733 a compiler bug. 60734 60735 * If the compiler does not produce an error message for invalid 60736 input, that is a compiler bug. However, you should note that your 60737 idea of "invalid input" might be someone else's idea of "an 60738 extension" or "support for traditional practice". 60739 60740 * If you are an experienced user of one of the languages GCC 60741 supports, your suggestions for improvement of GCC are welcome in 60742 any case. 60743 60744 60745File: gcc.info, Node: Bug Reporting, Prev: Bug Criteria, Up: Bugs 60746 6074715.2 How and Where to Report Bugs 60748================================= 60749 60750Bugs should be reported to the bug database at 60751<https://gcc.gnu.org/bugs/>. 60752 60753 60754File: gcc.info, Node: Service, Next: Contributing, Prev: Bugs, Up: Top 60755 6075616 How To Get Help with GCC 60757*************************** 60758 60759If you need help installing, using or changing GCC, there are two ways 60760to find it: 60761 60762 * Send a message to a suitable network mailing list. First try 60763 <gcc-help@gcc.gnu.org> (for help installing or using GCC), and if 60764 that brings no response, try <gcc@gcc.gnu.org>. For help changing 60765 GCC, ask <gcc@gcc.gnu.org>. If you think you have found a bug in 60766 GCC, please report it following the instructions at *note Bug 60767 Reporting::. 60768 60769 * Look in the service directory for someone who might help you for a 60770 fee. The service directory is found at 60771 <https://www.fsf.org/resources/service>. 60772 60773 For further information, see <https://gcc.gnu.org/faq.html#support>. 60774 60775 60776File: gcc.info, Node: Contributing, Next: Funding, Prev: Service, Up: Top 60777 6077817 Contributing to GCC Development 60779********************************** 60780 60781If you would like to help pretest GCC releases to assure they work well, 60782current development sources are available via Git (see 60783<https://gcc.gnu.org/git.html>). Source and binary snapshots are also 60784available for FTP; see <https://gcc.gnu.org/snapshots.html>. 60785 60786 If you would like to work on improvements to GCC, please read the 60787advice at these URLs: 60788 60789 <https://gcc.gnu.org/contribute.html> 60790 <https://gcc.gnu.org/contributewhy.html> 60791 60792for information on how to make useful contributions and avoid 60793duplication of effort. Suggested projects are listed at 60794<https://gcc.gnu.org/projects/>. 60795 60796 60797File: gcc.info, Node: Funding, Next: GNU Project, Prev: Contributing, Up: Top 60798 60799Funding Free Software 60800********************* 60801 60802If you want to have more free software a few years from now, it makes 60803sense for you to help encourage people to contribute funds for its 60804development. The most effective approach known is to encourage 60805commercial redistributors to donate. 60806 60807 Users of free software systems can boost the pace of development by 60808encouraging for-a-fee distributors to donate part of their selling price 60809to free software developers--the Free Software Foundation, and others. 60810 60811 The way to convince distributors to do this is to demand it and expect 60812it from them. So when you compare distributors, judge them partly by 60813how much they give to free software development. Show distributors they 60814must compete to be the one who gives the most. 60815 60816 To make this approach work, you must insist on numbers that you can 60817compare, such as, "We will donate ten dollars to the Frobnitz project 60818for each disk sold." Don't be satisfied with a vague promise, such as 60819"A portion of the profits are donated," since it doesn't give a basis 60820for comparison. 60821 60822 Even a precise fraction "of the profits from this disk" is not very 60823meaningful, since creative accounting and unrelated business decisions 60824can greatly alter what fraction of the sales price counts as profit. If 60825the price you pay is $50, ten percent of the profit is probably less 60826than a dollar; it might be a few cents, or nothing at all. 60827 60828 Some redistributors do development work themselves. This is useful 60829too; but to keep everyone honest, you need to inquire how much they do, 60830and what kind. Some kinds of development make much more long-term 60831difference than others. For example, maintaining a separate version of 60832a program contributes very little; maintaining the standard version of a 60833program for the whole community contributes much. Easy new ports 60834contribute little, since someone else would surely do them; difficult 60835ports such as adding a new CPU to the GNU Compiler Collection contribute 60836more; major new features or packages contribute the most. 60837 60838 By establishing the idea that supporting further development is "the 60839proper thing to do" when distributing free software for a fee, we can 60840assure a steady flow of resources into making more free software. 60841 60842 Copyright (C) 1994 Free Software Foundation, Inc. 60843 Verbatim copying and redistribution of this section is permitted 60844 without royalty; alteration is not permitted. 60845 60846 60847File: gcc.info, Node: GNU Project, Next: Copying, Prev: Funding, Up: Top 60848 60849The GNU Project and GNU/Linux 60850***************************** 60851 60852The GNU Project was launched in 1984 to develop a complete Unix-like 60853operating system which is free software: the GNU system. (GNU is a 60854recursive acronym for "GNU's Not Unix"; it is pronounced "guh-NEW".) 60855Variants of the GNU operating system, which use the kernel Linux, are 60856now widely used; though these systems are often referred to as "Linux", 60857they are more accurately called GNU/Linux systems. 60858 60859 For more information, see: 60860 <https://www.gnu.org/> 60861 <https://www.gnu.org/gnu/linux-and-gnu.html> 60862 60863 60864File: gcc.info, Node: Copying, Next: GNU Free Documentation License, Prev: GNU Project, Up: Top 60865 60866GNU General Public License 60867************************** 60868 60869 Version 3, 29 June 2007 60870 60871 Copyright (C) 2007 Free Software Foundation, Inc. <https://fsf.org/> 60872 60873 Everyone is permitted to copy and distribute verbatim copies of this 60874 license document, but changing it is not allowed. 60875 60876Preamble 60877======== 60878 60879The GNU General Public License is a free, copyleft license for software 60880and other kinds of works. 60881 60882 The licenses for most software and other practical works are designed 60883to take away your freedom to share and change the works. By contrast, 60884the GNU General Public License is intended to guarantee your freedom to 60885share and change all versions of a program-to make sure it remains free 60886software for all its users. We, the Free Software Foundation, use the 60887GNU General Public License for most of our software; it applies also to 60888any other work released this way by its authors. You can apply it to 60889your programs, too. 60890 60891 When we speak of free software, we are referring to freedom, not price. 60892Our General Public Licenses are designed to make sure that you have the 60893freedom to distribute copies of free software (and charge for them if 60894you wish), that you receive source code or can get it if you want it, 60895that you can change the software or use pieces of it in new free 60896programs, and that you know you can do these things. 60897 60898 To protect your rights, we need to prevent others from denying you 60899these rights or asking you to surrender the rights. Therefore, you have 60900certain responsibilities if you distribute copies of the software, or if 60901you modify it: responsibilities to respect the freedom of others. 60902 60903 For example, if you distribute copies of such a program, whether gratis 60904or for a fee, you must pass on to the recipients the same freedoms that 60905you received. You must make sure that they, too, receive or can get the 60906source code. And you must show them these terms so they know their 60907rights. 60908 60909 Developers that use the GNU GPL protect your rights with two steps: (1) 60910assert copyright on the software, and (2) offer you this License giving 60911you legal permission to copy, distribute and/or modify it. 60912 60913 For the developers' and authors' protection, the GPL clearly explains 60914that there is no warranty for this free software. For both users' and 60915authors' sake, the GPL requires that modified versions be marked as 60916changed, so that their problems will not be attributed erroneously to 60917authors of previous versions. 60918 60919 Some devices are designed to deny users access to install or run 60920modified versions of the software inside them, although the manufacturer 60921can do so. This is fundamentally incompatible with the aim of 60922protecting users' freedom to change the software. The systematic 60923pattern of such abuse occurs in the area of products for individuals to 60924use, which is precisely where it is most unacceptable. Therefore, we 60925have designed this version of the GPL to prohibit the practice for those 60926products. If such problems arise substantially in other domains, we 60927stand ready to extend this provision to those domains in future versions 60928of the GPL, as needed to protect the freedom of users. 60929 60930 Finally, every program is threatened constantly by software patents. 60931States should not allow patents to restrict development and use of 60932software on general-purpose computers, but in those that do, we wish to 60933avoid the special danger that patents applied to a free program could 60934make it effectively proprietary. To prevent this, the GPL assures that 60935patents cannot be used to render the program non-free. 60936 60937 The precise terms and conditions for copying, distribution and 60938modification follow. 60939 60940TERMS AND CONDITIONS 60941==================== 60942 60943 0. Definitions. 60944 60945 "This License" refers to version 3 of the GNU General Public 60946 License. 60947 60948 "Copyright" also means copyright-like laws that apply to other 60949 kinds of works, such as semiconductor masks. 60950 60951 "The Program" refers to any copyrightable work licensed under this 60952 License. Each licensee is addressed as "you". "Licensees" and 60953 "recipients" may be individuals or organizations. 60954 60955 To "modify" a work means to copy from or adapt all or part of the 60956 work in a fashion requiring copyright permission, other than the 60957 making of an exact copy. The resulting work is called a "modified 60958 version" of the earlier work or a work "based on" the earlier work. 60959 60960 A "covered work" means either the unmodified Program or a work 60961 based on the Program. 60962 60963 To "propagate" a work means to do anything with it that, without 60964 permission, would make you directly or secondarily liable for 60965 infringement under applicable copyright law, except executing it on 60966 a computer or modifying a private copy. Propagation includes 60967 copying, distribution (with or without modification), making 60968 available to the public, and in some countries other activities as 60969 well. 60970 60971 To "convey" a work means any kind of propagation that enables other 60972 parties to make or receive copies. Mere interaction with a user 60973 through a computer network, with no transfer of a copy, is not 60974 conveying. 60975 60976 An interactive user interface displays "Appropriate Legal Notices" 60977 to the extent that it includes a convenient and prominently visible 60978 feature that (1) displays an appropriate copyright notice, and (2) 60979 tells the user that there is no warranty for the work (except to 60980 the extent that warranties are provided), that licensees may convey 60981 the work under this License, and how to view a copy of this 60982 License. If the interface presents a list of user commands or 60983 options, such as a menu, a prominent item in the list meets this 60984 criterion. 60985 60986 1. Source Code. 60987 60988 The "source code" for a work means the preferred form of the work 60989 for making modifications to it. "Object code" means any non-source 60990 form of a work. 60991 60992 A "Standard Interface" means an interface that either is an 60993 official standard defined by a recognized standards body, or, in 60994 the case of interfaces specified for a particular programming 60995 language, one that is widely used among developers working in that 60996 language. 60997 60998 The "System Libraries" of an executable work include anything, 60999 other than the work as a whole, that (a) is included in the normal 61000 form of packaging a Major Component, but which is not part of that 61001 Major Component, and (b) serves only to enable use of the work with 61002 that Major Component, or to implement a Standard Interface for 61003 which an implementation is available to the public in source code 61004 form. A "Major Component", in this context, means a major 61005 essential component (kernel, window system, and so on) of the 61006 specific operating system (if any) on which the executable work 61007 runs, or a compiler used to produce the work, or an object code 61008 interpreter used to run it. 61009 61010 The "Corresponding Source" for a work in object code form means all 61011 the source code needed to generate, install, and (for an executable 61012 work) run the object code and to modify the work, including scripts 61013 to control those activities. However, it does not include the 61014 work's System Libraries, or general-purpose tools or generally 61015 available free programs which are used unmodified in performing 61016 those activities but which are not part of the work. For example, 61017 Corresponding Source includes interface definition files associated 61018 with source files for the work, and the source code for shared 61019 libraries and dynamically linked subprograms that the work is 61020 specifically designed to require, such as by intimate data 61021 communication or control flow between those subprograms and other 61022 parts of the work. 61023 61024 The Corresponding Source need not include anything that users can 61025 regenerate automatically from other parts of the Corresponding 61026 Source. 61027 61028 The Corresponding Source for a work in source code form is that 61029 same work. 61030 61031 2. Basic Permissions. 61032 61033 All rights granted under this License are granted for the term of 61034 copyright on the Program, and are irrevocable provided the stated 61035 conditions are met. This License explicitly affirms your unlimited 61036 permission to run the unmodified Program. The output from running 61037 a covered work is covered by this License only if the output, given 61038 its content, constitutes a covered work. This License acknowledges 61039 your rights of fair use or other equivalent, as provided by 61040 copyright law. 61041 61042 You may make, run and propagate covered works that you do not 61043 convey, without conditions so long as your license otherwise 61044 remains in force. You may convey covered works to others for the 61045 sole purpose of having them make modifications exclusively for you, 61046 or provide you with facilities for running those works, provided 61047 that you comply with the terms of this License in conveying all 61048 material for which you do not control copyright. Those thus making 61049 or running the covered works for you must do so exclusively on your 61050 behalf, under your direction and control, on terms that prohibit 61051 them from making any copies of your copyrighted material outside 61052 their relationship with you. 61053 61054 Conveying under any other circumstances is permitted solely under 61055 the conditions stated below. Sublicensing is not allowed; section 61056 10 makes it unnecessary. 61057 61058 3. Protecting Users' Legal Rights From Anti-Circumvention Law. 61059 61060 No covered work shall be deemed part of an effective technological 61061 measure under any applicable law fulfilling obligations under 61062 article 11 of the WIPO copyright treaty adopted on 20 December 61063 1996, or similar laws prohibiting or restricting circumvention of 61064 such measures. 61065 61066 When you convey a covered work, you waive any legal power to forbid 61067 circumvention of technological measures to the extent such 61068 circumvention is effected by exercising rights under this License 61069 with respect to the covered work, and you disclaim any intention to 61070 limit operation or modification of the work as a means of 61071 enforcing, against the work's users, your or third parties' legal 61072 rights to forbid circumvention of technological measures. 61073 61074 4. Conveying Verbatim Copies. 61075 61076 You may convey verbatim copies of the Program's source code as you 61077 receive it, in any medium, provided that you conspicuously and 61078 appropriately publish on each copy an appropriate copyright notice; 61079 keep intact all notices stating that this License and any 61080 non-permissive terms added in accord with section 7 apply to the 61081 code; keep intact all notices of the absence of any warranty; and 61082 give all recipients a copy of this License along with the Program. 61083 61084 You may charge any price or no price for each copy that you convey, 61085 and you may offer support or warranty protection for a fee. 61086 61087 5. Conveying Modified Source Versions. 61088 61089 You may convey a work based on the Program, or the modifications to 61090 produce it from the Program, in the form of source code under the 61091 terms of section 4, provided that you also meet all of these 61092 conditions: 61093 61094 a. The work must carry prominent notices stating that you 61095 modified it, and giving a relevant date. 61096 61097 b. The work must carry prominent notices stating that it is 61098 released under this License and any conditions added under 61099 section 7. This requirement modifies the requirement in 61100 section 4 to "keep intact all notices". 61101 61102 c. You must license the entire work, as a whole, under this 61103 License to anyone who comes into possession of a copy. This 61104 License will therefore apply, along with any applicable 61105 section 7 additional terms, to the whole of the work, and all 61106 its parts, regardless of how they are packaged. This License 61107 gives no permission to license the work in any other way, but 61108 it does not invalidate such permission if you have separately 61109 received it. 61110 61111 d. If the work has interactive user interfaces, each must display 61112 Appropriate Legal Notices; however, if the Program has 61113 interactive interfaces that do not display Appropriate Legal 61114 Notices, your work need not make them do so. 61115 61116 A compilation of a covered work with other separate and independent 61117 works, which are not by their nature extensions of the covered 61118 work, and which are not combined with it such as to form a larger 61119 program, in or on a volume of a storage or distribution medium, is 61120 called an "aggregate" if the compilation and its resulting 61121 copyright are not used to limit the access or legal rights of the 61122 compilation's users beyond what the individual works permit. 61123 Inclusion of a covered work in an aggregate does not cause this 61124 License to apply to the other parts of the aggregate. 61125 61126 6. Conveying Non-Source Forms. 61127 61128 You may convey a covered work in object code form under the terms 61129 of sections 4 and 5, provided that you also convey the 61130 machine-readable Corresponding Source under the terms of this 61131 License, in one of these ways: 61132 61133 a. Convey the object code in, or embodied in, a physical product 61134 (including a physical distribution medium), accompanied by the 61135 Corresponding Source fixed on a durable physical medium 61136 customarily used for software interchange. 61137 61138 b. Convey the object code in, or embodied in, a physical product 61139 (including a physical distribution medium), accompanied by a 61140 written offer, valid for at least three years and valid for as 61141 long as you offer spare parts or customer support for that 61142 product model, to give anyone who possesses the object code 61143 either (1) a copy of the Corresponding Source for all the 61144 software in the product that is covered by this License, on a 61145 durable physical medium customarily used for software 61146 interchange, for a price no more than your reasonable cost of 61147 physically performing this conveying of source, or (2) access 61148 to copy the Corresponding Source from a network server at no 61149 charge. 61150 61151 c. Convey individual copies of the object code with a copy of the 61152 written offer to provide the Corresponding Source. This 61153 alternative is allowed only occasionally and noncommercially, 61154 and only if you received the object code with such an offer, 61155 in accord with subsection 6b. 61156 61157 d. Convey the object code by offering access from a designated 61158 place (gratis or for a charge), and offer equivalent access to 61159 the Corresponding Source in the same way through the same 61160 place at no further charge. You need not require recipients 61161 to copy the Corresponding Source along with the object code. 61162 If the place to copy the object code is a network server, the 61163 Corresponding Source may be on a different server (operated by 61164 you or a third party) that supports equivalent copying 61165 facilities, provided you maintain clear directions next to the 61166 object code saying where to find the Corresponding Source. 61167 Regardless of what server hosts the Corresponding Source, you 61168 remain obligated to ensure that it is available for as long as 61169 needed to satisfy these requirements. 61170 61171 e. Convey the object code using peer-to-peer transmission, 61172 provided you inform other peers where the object code and 61173 Corresponding Source of the work are being offered to the 61174 general public at no charge under subsection 6d. 61175 61176 A separable portion of the object code, whose source code is 61177 excluded from the Corresponding Source as a System Library, need 61178 not be included in conveying the object code work. 61179 61180 A "User Product" is either (1) a "consumer product", which means 61181 any tangible personal property which is normally used for personal, 61182 family, or household purposes, or (2) anything designed or sold for 61183 incorporation into a dwelling. In determining whether a product is 61184 a consumer product, doubtful cases shall be resolved in favor of 61185 coverage. For a particular product received by a particular user, 61186 "normally used" refers to a typical or common use of that class of 61187 product, regardless of the status of the particular user or of the 61188 way in which the particular user actually uses, or expects or is 61189 expected to use, the product. A product is a consumer product 61190 regardless of whether the product has substantial commercial, 61191 industrial or non-consumer uses, unless such uses represent the 61192 only significant mode of use of the product. 61193 61194 "Installation Information" for a User Product means any methods, 61195 procedures, authorization keys, or other information required to 61196 install and execute modified versions of a covered work in that 61197 User Product from a modified version of its Corresponding Source. 61198 The information must suffice to ensure that the continued 61199 functioning of the modified object code is in no case prevented or 61200 interfered with solely because modification has been made. 61201 61202 If you convey an object code work under this section in, or with, 61203 or specifically for use in, a User Product, and the conveying 61204 occurs as part of a transaction in which the right of possession 61205 and use of the User Product is transferred to the recipient in 61206 perpetuity or for a fixed term (regardless of how the transaction 61207 is characterized), the Corresponding Source conveyed under this 61208 section must be accompanied by the Installation Information. But 61209 this requirement does not apply if neither you nor any third party 61210 retains the ability to install modified object code on the User 61211 Product (for example, the work has been installed in ROM). 61212 61213 The requirement to provide Installation Information does not 61214 include a requirement to continue to provide support service, 61215 warranty, or updates for a work that has been modified or installed 61216 by the recipient, or for the User Product in which it has been 61217 modified or installed. Access to a network may be denied when the 61218 modification itself materially and adversely affects the operation 61219 of the network or violates the rules and protocols for 61220 communication across the network. 61221 61222 Corresponding Source conveyed, and Installation Information 61223 provided, in accord with this section must be in a format that is 61224 publicly documented (and with an implementation available to the 61225 public in source code form), and must require no special password 61226 or key for unpacking, reading or copying. 61227 61228 7. Additional Terms. 61229 61230 "Additional permissions" are terms that supplement the terms of 61231 this License by making exceptions from one or more of its 61232 conditions. Additional permissions that are applicable to the 61233 entire Program shall be treated as though they were included in 61234 this License, to the extent that they are valid under applicable 61235 law. If additional permissions apply only to part of the Program, 61236 that part may be used separately under those permissions, but the 61237 entire Program remains governed by this License without regard to 61238 the additional permissions. 61239 61240 When you convey a copy of a covered work, you may at your option 61241 remove any additional permissions from that copy, or from any part 61242 of it. (Additional permissions may be written to require their own 61243 removal in certain cases when you modify the work.) You may place 61244 additional permissions on material, added by you to a covered work, 61245 for which you have or can give appropriate copyright permission. 61246 61247 Notwithstanding any other provision of this License, for material 61248 you add to a covered work, you may (if authorized by the copyright 61249 holders of that material) supplement the terms of this License with 61250 terms: 61251 61252 a. Disclaiming warranty or limiting liability differently from 61253 the terms of sections 15 and 16 of this License; or 61254 61255 b. Requiring preservation of specified reasonable legal notices 61256 or author attributions in that material or in the Appropriate 61257 Legal Notices displayed by works containing it; or 61258 61259 c. Prohibiting misrepresentation of the origin of that material, 61260 or requiring that modified versions of such material be marked 61261 in reasonable ways as different from the original version; or 61262 61263 d. Limiting the use for publicity purposes of names of licensors 61264 or authors of the material; or 61265 61266 e. Declining to grant rights under trademark law for use of some 61267 trade names, trademarks, or service marks; or 61268 61269 f. Requiring indemnification of licensors and authors of that 61270 material by anyone who conveys the material (or modified 61271 versions of it) with contractual assumptions of liability to 61272 the recipient, for any liability that these contractual 61273 assumptions directly impose on those licensors and authors. 61274 61275 All other non-permissive additional terms are considered "further 61276 restrictions" within the meaning of section 10. If the Program as 61277 you received it, or any part of it, contains a notice stating that 61278 it is governed by this License along with a term that is a further 61279 restriction, you may remove that term. If a license document 61280 contains a further restriction but permits relicensing or conveying 61281 under this License, you may add to a covered work material governed 61282 by the terms of that license document, provided that the further 61283 restriction does not survive such relicensing or conveying. 61284 61285 If you add terms to a covered work in accord with this section, you 61286 must place, in the relevant source files, a statement of the 61287 additional terms that apply to those files, or a notice indicating 61288 where to find the applicable terms. 61289 61290 Additional terms, permissive or non-permissive, may be stated in 61291 the form of a separately written license, or stated as exceptions; 61292 the above requirements apply either way. 61293 61294 8. Termination. 61295 61296 You may not propagate or modify a covered work except as expressly 61297 provided under this License. Any attempt otherwise to propagate or 61298 modify it is void, and will automatically terminate your rights 61299 under this License (including any patent licenses granted under the 61300 third paragraph of section 11). 61301 61302 However, if you cease all violation of this License, then your 61303 license from a particular copyright holder is reinstated (a) 61304 provisionally, unless and until the copyright holder explicitly and 61305 finally terminates your license, and (b) permanently, if the 61306 copyright holder fails to notify you of the violation by some 61307 reasonable means prior to 60 days after the cessation. 61308 61309 Moreover, your license from a particular copyright holder is 61310 reinstated permanently if the copyright holder notifies you of the 61311 violation by some reasonable means, this is the first time you have 61312 received notice of violation of this License (for any work) from 61313 that copyright holder, and you cure the violation prior to 30 days 61314 after your receipt of the notice. 61315 61316 Termination of your rights under this section does not terminate 61317 the licenses of parties who have received copies or rights from you 61318 under this License. If your rights have been terminated and not 61319 permanently reinstated, you do not qualify to receive new licenses 61320 for the same material under section 10. 61321 61322 9. Acceptance Not Required for Having Copies. 61323 61324 You are not required to accept this License in order to receive or 61325 run a copy of the Program. Ancillary propagation of a covered work 61326 occurring solely as a consequence of using peer-to-peer 61327 transmission to receive a copy likewise does not require 61328 acceptance. However, nothing other than this License grants you 61329 permission to propagate or modify any covered work. These actions 61330 infringe copyright if you do not accept this License. Therefore, 61331 by modifying or propagating a covered work, you indicate your 61332 acceptance of this License to do so. 61333 61334 10. Automatic Licensing of Downstream Recipients. 61335 61336 Each time you convey a covered work, the recipient automatically 61337 receives a license from the original licensors, to run, modify and 61338 propagate that work, subject to this License. You are not 61339 responsible for enforcing compliance by third parties with this 61340 License. 61341 61342 An "entity transaction" is a transaction transferring control of an 61343 organization, or substantially all assets of one, or subdividing an 61344 organization, or merging organizations. If propagation of a 61345 covered work results from an entity transaction, each party to that 61346 transaction who receives a copy of the work also receives whatever 61347 licenses to the work the party's predecessor in interest had or 61348 could give under the previous paragraph, plus a right to possession 61349 of the Corresponding Source of the work from the predecessor in 61350 interest, if the predecessor has it or can get it with reasonable 61351 efforts. 61352 61353 You may not impose any further restrictions on the exercise of the 61354 rights granted or affirmed under this License. For example, you 61355 may not impose a license fee, royalty, or other charge for exercise 61356 of rights granted under this License, and you may not initiate 61357 litigation (including a cross-claim or counterclaim in a lawsuit) 61358 alleging that any patent claim is infringed by making, using, 61359 selling, offering for sale, or importing the Program or any portion 61360 of it. 61361 61362 11. Patents. 61363 61364 A "contributor" is a copyright holder who authorizes use under this 61365 License of the Program or a work on which the Program is based. 61366 The work thus licensed is called the contributor's "contributor 61367 version". 61368 61369 A contributor's "essential patent claims" are all patent claims 61370 owned or controlled by the contributor, whether already acquired or 61371 hereafter acquired, that would be infringed by some manner, 61372 permitted by this License, of making, using, or selling its 61373 contributor version, but do not include claims that would be 61374 infringed only as a consequence of further modification of the 61375 contributor version. For purposes of this definition, "control" 61376 includes the right to grant patent sublicenses in a manner 61377 consistent with the requirements of this License. 61378 61379 Each contributor grants you a non-exclusive, worldwide, 61380 royalty-free patent license under the contributor's essential 61381 patent claims, to make, use, sell, offer for sale, import and 61382 otherwise run, modify and propagate the contents of its contributor 61383 version. 61384 61385 In the following three paragraphs, a "patent license" is any 61386 express agreement or commitment, however denominated, not to 61387 enforce a patent (such as an express permission to practice a 61388 patent or covenant not to sue for patent infringement). To "grant" 61389 such a patent license to a party means to make such an agreement or 61390 commitment not to enforce a patent against the party. 61391 61392 If you convey a covered work, knowingly relying on a patent 61393 license, and the Corresponding Source of the work is not available 61394 for anyone to copy, free of charge and under the terms of this 61395 License, through a publicly available network server or other 61396 readily accessible means, then you must either (1) cause the 61397 Corresponding Source to be so available, or (2) arrange to deprive 61398 yourself of the benefit of the patent license for this particular 61399 work, or (3) arrange, in a manner consistent with the requirements 61400 of this License, to extend the patent license to downstream 61401 recipients. "Knowingly relying" means you have actual knowledge 61402 that, but for the patent license, your conveying the covered work 61403 in a country, or your recipient's use of the covered work in a 61404 country, would infringe one or more identifiable patents in that 61405 country that you have reason to believe are valid. 61406 61407 If, pursuant to or in connection with a single transaction or 61408 arrangement, you convey, or propagate by procuring conveyance of, a 61409 covered work, and grant a patent license to some of the parties 61410 receiving the covered work authorizing them to use, propagate, 61411 modify or convey a specific copy of the covered work, then the 61412 patent license you grant is automatically extended to all 61413 recipients of the covered work and works based on it. 61414 61415 A patent license is "discriminatory" if it does not include within 61416 the scope of its coverage, prohibits the exercise of, or is 61417 conditioned on the non-exercise of one or more of the rights that 61418 are specifically granted under this License. You may not convey a 61419 covered work if you are a party to an arrangement with a third 61420 party that is in the business of distributing software, under which 61421 you make payment to the third party based on the extent of your 61422 activity of conveying the work, and under which the third party 61423 grants, to any of the parties who would receive the covered work 61424 from you, a discriminatory patent license (a) in connection with 61425 copies of the covered work conveyed by you (or copies made from 61426 those copies), or (b) primarily for and in connection with specific 61427 products or compilations that contain the covered work, unless you 61428 entered into that arrangement, or that patent license was granted, 61429 prior to 28 March 2007. 61430 61431 Nothing in this License shall be construed as excluding or limiting 61432 any implied license or other defenses to infringement that may 61433 otherwise be available to you under applicable patent law. 61434 61435 12. No Surrender of Others' Freedom. 61436 61437 If conditions are imposed on you (whether by court order, agreement 61438 or otherwise) that contradict the conditions of this License, they 61439 do not excuse you from the conditions of this License. If you 61440 cannot convey a covered work so as to satisfy simultaneously your 61441 obligations under this License and any other pertinent obligations, 61442 then as a consequence you may not convey it at all. For example, 61443 if you agree to terms that obligate you to collect a royalty for 61444 further conveying from those to whom you convey the Program, the 61445 only way you could satisfy both those terms and this License would 61446 be to refrain entirely from conveying the Program. 61447 61448 13. Use with the GNU Affero General Public License. 61449 61450 Notwithstanding any other provision of this License, you have 61451 permission to link or combine any covered work with a work licensed 61452 under version 3 of the GNU Affero General Public License into a 61453 single combined work, and to convey the resulting work. The terms 61454 of this License will continue to apply to the part which is the 61455 covered work, but the special requirements of the GNU Affero 61456 General Public License, section 13, concerning interaction through 61457 a network will apply to the combination as such. 61458 61459 14. Revised Versions of this License. 61460 61461 The Free Software Foundation may publish revised and/or new 61462 versions of the GNU General Public License from time to time. Such 61463 new versions will be similar in spirit to the present version, but 61464 may differ in detail to address new problems or concerns. 61465 61466 Each version is given a distinguishing version number. If the 61467 Program specifies that a certain numbered version of the GNU 61468 General Public License "or any later version" applies to it, you 61469 have the option of following the terms and conditions either of 61470 that numbered version or of any later version published by the Free 61471 Software Foundation. If the Program does not specify a version 61472 number of the GNU General Public License, you may choose any 61473 version ever published by the Free Software Foundation. 61474 61475 If the Program specifies that a proxy can decide which future 61476 versions of the GNU General Public License can be used, that 61477 proxy's public statement of acceptance of a version permanently 61478 authorizes you to choose that version for the Program. 61479 61480 Later license versions may give you additional or different 61481 permissions. However, no additional obligations are imposed on any 61482 author or copyright holder as a result of your choosing to follow a 61483 later version. 61484 61485 15. Disclaimer of Warranty. 61486 61487 THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY 61488 APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE 61489 COPYRIGHT HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" 61490 WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESSED OR IMPLIED, 61491 INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF 61492 MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. THE ENTIRE 61493 RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM IS WITH YOU. 61494 SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF ALL 61495 NECESSARY SERVICING, REPAIR OR CORRECTION. 61496 61497 16. Limitation of Liability. 61498 61499 IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN 61500 WRITING WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES 61501 AND/OR CONVEYS THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR 61502 DAMAGES, INCLUDING ANY GENERAL, SPECIAL, INCIDENTAL OR 61503 CONSEQUENTIAL DAMAGES ARISING OUT OF THE USE OR INABILITY TO USE 61504 THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF DATA OR DATA 61505 BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD 61506 PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER 61507 PROGRAMS), EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF 61508 THE POSSIBILITY OF SUCH DAMAGES. 61509 61510 17. Interpretation of Sections 15 and 16. 61511 61512 If the disclaimer of warranty and limitation of liability provided 61513 above cannot be given local legal effect according to their terms, 61514 reviewing courts shall apply local law that most closely 61515 approximates an absolute waiver of all civil liability in 61516 connection with the Program, unless a warranty or assumption of 61517 liability accompanies a copy of the Program in return for a fee. 61518 61519END OF TERMS AND CONDITIONS 61520=========================== 61521 61522How to Apply These Terms to Your New Programs 61523============================================= 61524 61525If you develop a new program, and you want it to be of the greatest 61526possible use to the public, the best way to achieve this is to make it 61527free software which everyone can redistribute and change under these 61528terms. 61529 61530 To do so, attach the following notices to the program. It is safest to 61531attach them to the start of each source file to most effectively state 61532the exclusion of warranty; and each file should have at least the 61533"copyright" line and a pointer to where the full notice is found. 61534 61535 ONE LINE TO GIVE THE PROGRAM'S NAME AND A BRIEF IDEA OF WHAT IT DOES. 61536 Copyright (C) YEAR NAME OF AUTHOR 61537 61538 This program is free software: you can redistribute it and/or modify 61539 it under the terms of the GNU General Public License as published by 61540 the Free Software Foundation, either version 3 of the License, or (at 61541 your option) any later version. 61542 61543 This program is distributed in the hope that it will be useful, but 61544 WITHOUT ANY WARRANTY; without even the implied warranty of 61545 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 61546 General Public License for more details. 61547 61548 You should have received a copy of the GNU General Public License 61549 along with this program. If not, see <https://www.gnu.org/licenses/>. 61550 61551 Also add information on how to contact you by electronic and paper 61552mail. 61553 61554 If the program does terminal interaction, make it output a short notice 61555like this when it starts in an interactive mode: 61556 61557 PROGRAM Copyright (C) YEAR NAME OF AUTHOR 61558 This program comes with ABSOLUTELY NO WARRANTY; for details type 'show w'. 61559 This is free software, and you are welcome to redistribute it 61560 under certain conditions; type 'show c' for details. 61561 61562 The hypothetical commands 'show w' and 'show c' should show the 61563appropriate parts of the General Public License. Of course, your 61564program's commands might be different; for a GUI interface, you would 61565use an "about box". 61566 61567 You should also get your employer (if you work as a programmer) or 61568school, if any, to sign a "copyright disclaimer" for the program, if 61569necessary. For more information on this, and how to apply and follow 61570the GNU GPL, see <https://www.gnu.org/licenses/>. 61571 61572 The GNU General Public License does not permit incorporating your 61573program into proprietary programs. If your program is a subroutine 61574library, you may consider it more useful to permit linking proprietary 61575applications with the library. If this is what you want to do, use the 61576GNU Lesser General Public License instead of this License. But first, 61577please read <https://www.gnu.org/licenses/why-not-lgpl.html>. 61578 61579 61580File: gcc.info, Node: GNU Free Documentation License, Next: Contributors, Prev: Copying, Up: Top 61581 61582GNU Free Documentation License 61583****************************** 61584 61585 Version 1.3, 3 November 2008 61586 61587 Copyright (C) 2000, 2001, 2002, 2007, 2008 Free Software Foundation, Inc. 61588 <https://fsf.org/> 61589 61590 Everyone is permitted to copy and distribute verbatim copies 61591 of this license document, but changing it is not allowed. 61592 61593 0. PREAMBLE 61594 61595 The purpose of this License is to make a manual, textbook, or other 61596 functional and useful document "free" in the sense of freedom: to 61597 assure everyone the effective freedom to copy and redistribute it, 61598 with or without modifying it, either commercially or 61599 noncommercially. Secondarily, this License preserves for the 61600 author and publisher a way to get credit for their work, while not 61601 being considered responsible for modifications made by others. 61602 61603 This License is a kind of "copyleft", which means that derivative 61604 works of the document must themselves be free in the same sense. 61605 It complements the GNU General Public License, which is a copyleft 61606 license designed for free software. 61607 61608 We have designed this License in order to use it for manuals for 61609 free software, because free software needs free documentation: a 61610 free program should come with manuals providing the same freedoms 61611 that the software does. But this License is not limited to 61612 software manuals; it can be used for any textual work, regardless 61613 of subject matter or whether it is published as a printed book. We 61614 recommend this License principally for works whose purpose is 61615 instruction or reference. 61616 61617 1. APPLICABILITY AND DEFINITIONS 61618 61619 This License applies to any manual or other work, in any medium, 61620 that contains a notice placed by the copyright holder saying it can 61621 be distributed under the terms of this License. Such a notice 61622 grants a world-wide, royalty-free license, unlimited in duration, 61623 to use that work under the conditions stated herein. The 61624 "Document", below, refers to any such manual or work. Any member 61625 of the public is a licensee, and is addressed as "you". You accept 61626 the license if you copy, modify or distribute the work in a way 61627 requiring permission under copyright law. 61628 61629 A "Modified Version" of the Document means any work containing the 61630 Document or a portion of it, either copied verbatim, or with 61631 modifications and/or translated into another language. 61632 61633 A "Secondary Section" is a named appendix or a front-matter section 61634 of the Document that deals exclusively with the relationship of the 61635 publishers or authors of the Document to the Document's overall 61636 subject (or to related matters) and contains nothing that could 61637 fall directly within that overall subject. (Thus, if the Document 61638 is in part a textbook of mathematics, a Secondary Section may not 61639 explain any mathematics.) The relationship could be a matter of 61640 historical connection with the subject or with related matters, or 61641 of legal, commercial, philosophical, ethical or political position 61642 regarding them. 61643 61644 The "Invariant Sections" are certain Secondary Sections whose 61645 titles are designated, as being those of Invariant Sections, in the 61646 notice that says that the Document is released under this License. 61647 If a section does not fit the above definition of Secondary then it 61648 is not allowed to be designated as Invariant. The Document may 61649 contain zero Invariant Sections. If the Document does not identify 61650 any Invariant Sections then there are none. 61651 61652 The "Cover Texts" are certain short passages of text that are 61653 listed, as Front-Cover Texts or Back-Cover Texts, in the notice 61654 that says that the Document is released under this License. A 61655 Front-Cover Text may be at most 5 words, and a Back-Cover Text may 61656 be at most 25 words. 61657 61658 A "Transparent" copy of the Document means a machine-readable copy, 61659 represented in a format whose specification is available to the 61660 general public, that is suitable for revising the document 61661 straightforwardly with generic text editors or (for images composed 61662 of pixels) generic paint programs or (for drawings) some widely 61663 available drawing editor, and that is suitable for input to text 61664 formatters or for automatic translation to a variety of formats 61665 suitable for input to text formatters. A copy made in an otherwise 61666 Transparent file format whose markup, or absence of markup, has 61667 been arranged to thwart or discourage subsequent modification by 61668 readers is not Transparent. An image format is not Transparent if 61669 used for any substantial amount of text. A copy that is not 61670 "Transparent" is called "Opaque". 61671 61672 Examples of suitable formats for Transparent copies include plain 61673 ASCII without markup, Texinfo input format, LaTeX input format, 61674 SGML or XML using a publicly available DTD, and standard-conforming 61675 simple HTML, PostScript or PDF designed for human modification. 61676 Examples of transparent image formats include PNG, XCF and JPG. 61677 Opaque formats include proprietary formats that can be read and 61678 edited only by proprietary word processors, SGML or XML for which 61679 the DTD and/or processing tools are not generally available, and 61680 the machine-generated HTML, PostScript or PDF produced by some word 61681 processors for output purposes only. 61682 61683 The "Title Page" means, for a printed book, the title page itself, 61684 plus such following pages as are needed to hold, legibly, the 61685 material this License requires to appear in the title page. For 61686 works in formats which do not have any title page as such, "Title 61687 Page" means the text near the most prominent appearance of the 61688 work's title, preceding the beginning of the body of the text. 61689 61690 The "publisher" means any person or entity that distributes copies 61691 of the Document to the public. 61692 61693 A section "Entitled XYZ" means a named subunit of the Document 61694 whose title either is precisely XYZ or contains XYZ in parentheses 61695 following text that translates XYZ in another language. (Here XYZ 61696 stands for a specific section name mentioned below, such as 61697 "Acknowledgements", "Dedications", "Endorsements", or "History".) 61698 To "Preserve the Title" of such a section when you modify the 61699 Document means that it remains a section "Entitled XYZ" according 61700 to this definition. 61701 61702 The Document may include Warranty Disclaimers next to the notice 61703 which states that this License applies to the Document. These 61704 Warranty Disclaimers are considered to be included by reference in 61705 this License, but only as regards disclaiming warranties: any other 61706 implication that these Warranty Disclaimers may have is void and 61707 has no effect on the meaning of this License. 61708 61709 2. VERBATIM COPYING 61710 61711 You may copy and distribute the Document in any medium, either 61712 commercially or noncommercially, provided that this License, the 61713 copyright notices, and the license notice saying this License 61714 applies to the Document are reproduced in all copies, and that you 61715 add no other conditions whatsoever to those of this License. You 61716 may not use technical measures to obstruct or control the reading 61717 or further copying of the copies you make or distribute. However, 61718 you may accept compensation in exchange for copies. If you 61719 distribute a large enough number of copies you must also follow the 61720 conditions in section 3. 61721 61722 You may also lend copies, under the same conditions stated above, 61723 and you may publicly display copies. 61724 61725 3. COPYING IN QUANTITY 61726 61727 If you publish printed copies (or copies in media that commonly 61728 have printed covers) of the Document, numbering more than 100, and 61729 the Document's license notice requires Cover Texts, you must 61730 enclose the copies in covers that carry, clearly and legibly, all 61731 these Cover Texts: Front-Cover Texts on the front cover, and 61732 Back-Cover Texts on the back cover. Both covers must also clearly 61733 and legibly identify you as the publisher of these copies. The 61734 front cover must present the full title with all words of the title 61735 equally prominent and visible. You may add other material on the 61736 covers in addition. Copying with changes limited to the covers, as 61737 long as they preserve the title of the Document and satisfy these 61738 conditions, can be treated as verbatim copying in other respects. 61739 61740 If the required texts for either cover are too voluminous to fit 61741 legibly, you should put the first ones listed (as many as fit 61742 reasonably) on the actual cover, and continue the rest onto 61743 adjacent pages. 61744 61745 If you publish or distribute Opaque copies of the Document 61746 numbering more than 100, you must either include a machine-readable 61747 Transparent copy along with each Opaque copy, or state in or with 61748 each Opaque copy a computer-network location from which the general 61749 network-using public has access to download using public-standard 61750 network protocols a complete Transparent copy of the Document, free 61751 of added material. If you use the latter option, you must take 61752 reasonably prudent steps, when you begin distribution of Opaque 61753 copies in quantity, to ensure that this Transparent copy will 61754 remain thus accessible at the stated location until at least one 61755 year after the last time you distribute an Opaque copy (directly or 61756 through your agents or retailers) of that edition to the public. 61757 61758 It is requested, but not required, that you contact the authors of 61759 the Document well before redistributing any large number of copies, 61760 to give them a chance to provide you with an updated version of the 61761 Document. 61762 61763 4. MODIFICATIONS 61764 61765 You may copy and distribute a Modified Version of the Document 61766 under the conditions of sections 2 and 3 above, provided that you 61767 release the Modified Version under precisely this License, with the 61768 Modified Version filling the role of the Document, thus licensing 61769 distribution and modification of the Modified Version to whoever 61770 possesses a copy of it. In addition, you must do these things in 61771 the Modified Version: 61772 61773 A. Use in the Title Page (and on the covers, if any) a title 61774 distinct from that of the Document, and from those of previous 61775 versions (which should, if there were any, be listed in the 61776 History section of the Document). You may use the same title 61777 as a previous version if the original publisher of that 61778 version gives permission. 61779 61780 B. List on the Title Page, as authors, one or more persons or 61781 entities responsible for authorship of the modifications in 61782 the Modified Version, together with at least five of the 61783 principal authors of the Document (all of its principal 61784 authors, if it has fewer than five), unless they release you 61785 from this requirement. 61786 61787 C. State on the Title page the name of the publisher of the 61788 Modified Version, as the publisher. 61789 61790 D. Preserve all the copyright notices of the Document. 61791 61792 E. Add an appropriate copyright notice for your modifications 61793 adjacent to the other copyright notices. 61794 61795 F. Include, immediately after the copyright notices, a license 61796 notice giving the public permission to use the Modified 61797 Version under the terms of this License, in the form shown in 61798 the Addendum below. 61799 61800 G. Preserve in that license notice the full lists of Invariant 61801 Sections and required Cover Texts given in the Document's 61802 license notice. 61803 61804 H. Include an unaltered copy of this License. 61805 61806 I. Preserve the section Entitled "History", Preserve its Title, 61807 and add to it an item stating at least the title, year, new 61808 authors, and publisher of the Modified Version as given on the 61809 Title Page. If there is no section Entitled "History" in the 61810 Document, create one stating the title, year, authors, and 61811 publisher of the Document as given on its Title Page, then add 61812 an item describing the Modified Version as stated in the 61813 previous sentence. 61814 61815 J. Preserve the network location, if any, given in the Document 61816 for public access to a Transparent copy of the Document, and 61817 likewise the network locations given in the Document for 61818 previous versions it was based on. These may be placed in the 61819 "History" section. You may omit a network location for a work 61820 that was published at least four years before the Document 61821 itself, or if the original publisher of the version it refers 61822 to gives permission. 61823 61824 K. For any section Entitled "Acknowledgements" or "Dedications", 61825 Preserve the Title of the section, and preserve in the section 61826 all the substance and tone of each of the contributor 61827 acknowledgements and/or dedications given therein. 61828 61829 L. Preserve all the Invariant Sections of the Document, unaltered 61830 in their text and in their titles. Section numbers or the 61831 equivalent are not considered part of the section titles. 61832 61833 M. Delete any section Entitled "Endorsements". Such a section 61834 may not be included in the Modified Version. 61835 61836 N. Do not retitle any existing section to be Entitled 61837 "Endorsements" or to conflict in title with any Invariant 61838 Section. 61839 61840 O. Preserve any Warranty Disclaimers. 61841 61842 If the Modified Version includes new front-matter sections or 61843 appendices that qualify as Secondary Sections and contain no 61844 material copied from the Document, you may at your option designate 61845 some or all of these sections as invariant. To do this, add their 61846 titles to the list of Invariant Sections in the Modified Version's 61847 license notice. These titles must be distinct from any other 61848 section titles. 61849 61850 You may add a section Entitled "Endorsements", provided it contains 61851 nothing but endorsements of your Modified Version by various 61852 parties--for example, statements of peer review or that the text 61853 has been approved by an organization as the authoritative 61854 definition of a standard. 61855 61856 You may add a passage of up to five words as a Front-Cover Text, 61857 and a passage of up to 25 words as a Back-Cover Text, to the end of 61858 the list of Cover Texts in the Modified Version. Only one passage 61859 of Front-Cover Text and one of Back-Cover Text may be added by (or 61860 through arrangements made by) any one entity. If the Document 61861 already includes a cover text for the same cover, previously added 61862 by you or by arrangement made by the same entity you are acting on 61863 behalf of, you may not add another; but you may replace the old 61864 one, on explicit permission from the previous publisher that added 61865 the old one. 61866 61867 The author(s) and publisher(s) of the Document do not by this 61868 License give permission to use their names for publicity for or to 61869 assert or imply endorsement of any Modified Version. 61870 61871 5. COMBINING DOCUMENTS 61872 61873 You may combine the Document with other documents released under 61874 this License, under the terms defined in section 4 above for 61875 modified versions, provided that you include in the combination all 61876 of the Invariant Sections of all of the original documents, 61877 unmodified, and list them all as Invariant Sections of your 61878 combined work in its license notice, and that you preserve all 61879 their Warranty Disclaimers. 61880 61881 The combined work need only contain one copy of this License, and 61882 multiple identical Invariant Sections may be replaced with a single 61883 copy. If there are multiple Invariant Sections with the same name 61884 but different contents, make the title of each such section unique 61885 by adding at the end of it, in parentheses, the name of the 61886 original author or publisher of that section if known, or else a 61887 unique number. Make the same adjustment to the section titles in 61888 the list of Invariant Sections in the license notice of the 61889 combined work. 61890 61891 In the combination, you must combine any sections Entitled 61892 "History" in the various original documents, forming one section 61893 Entitled "History"; likewise combine any sections Entitled 61894 "Acknowledgements", and any sections Entitled "Dedications". You 61895 must delete all sections Entitled "Endorsements." 61896 61897 6. COLLECTIONS OF DOCUMENTS 61898 61899 You may make a collection consisting of the Document and other 61900 documents released under this License, and replace the individual 61901 copies of this License in the various documents with a single copy 61902 that is included in the collection, provided that you follow the 61903 rules of this License for verbatim copying of each of the documents 61904 in all other respects. 61905 61906 You may extract a single document from such a collection, and 61907 distribute it individually under this License, provided you insert 61908 a copy of this License into the extracted document, and follow this 61909 License in all other respects regarding verbatim copying of that 61910 document. 61911 61912 7. AGGREGATION WITH INDEPENDENT WORKS 61913 61914 A compilation of the Document or its derivatives with other 61915 separate and independent documents or works, in or on a volume of a 61916 storage or distribution medium, is called an "aggregate" if the 61917 copyright resulting from the compilation is not used to limit the 61918 legal rights of the compilation's users beyond what the individual 61919 works permit. When the Document is included in an aggregate, this 61920 License does not apply to the other works in the aggregate which 61921 are not themselves derivative works of the Document. 61922 61923 If the Cover Text requirement of section 3 is applicable to these 61924 copies of the Document, then if the Document is less than one half 61925 of the entire aggregate, the Document's Cover Texts may be placed 61926 on covers that bracket the Document within the aggregate, or the 61927 electronic equivalent of covers if the Document is in electronic 61928 form. Otherwise they must appear on printed covers that bracket 61929 the whole aggregate. 61930 61931 8. TRANSLATION 61932 61933 Translation is considered a kind of modification, so you may 61934 distribute translations of the Document under the terms of section 61935 4. Replacing Invariant Sections with translations requires special 61936 permission from their copyright holders, but you may include 61937 translations of some or all Invariant Sections in addition to the 61938 original versions of these Invariant Sections. You may include a 61939 translation of this License, and all the license notices in the 61940 Document, and any Warranty Disclaimers, provided that you also 61941 include the original English version of this License and the 61942 original versions of those notices and disclaimers. In case of a 61943 disagreement between the translation and the original version of 61944 this License or a notice or disclaimer, the original version will 61945 prevail. 61946 61947 If a section in the Document is Entitled "Acknowledgements", 61948 "Dedications", or "History", the requirement (section 4) to 61949 Preserve its Title (section 1) will typically require changing the 61950 actual title. 61951 61952 9. TERMINATION 61953 61954 You may not copy, modify, sublicense, or distribute the Document 61955 except as expressly provided under this License. Any attempt 61956 otherwise to copy, modify, sublicense, or distribute it is void, 61957 and will automatically terminate your rights under this License. 61958 61959 However, if you cease all violation of this License, then your 61960 license from a particular copyright holder is reinstated (a) 61961 provisionally, unless and until the copyright holder explicitly and 61962 finally terminates your license, and (b) permanently, if the 61963 copyright holder fails to notify you of the violation by some 61964 reasonable means prior to 60 days after the cessation. 61965 61966 Moreover, your license from a particular copyright holder is 61967 reinstated permanently if the copyright holder notifies you of the 61968 violation by some reasonable means, this is the first time you have 61969 received notice of violation of this License (for any work) from 61970 that copyright holder, and you cure the violation prior to 30 days 61971 after your receipt of the notice. 61972 61973 Termination of your rights under this section does not terminate 61974 the licenses of parties who have received copies or rights from you 61975 under this License. If your rights have been terminated and not 61976 permanently reinstated, receipt of a copy of some or all of the 61977 same material does not give you any rights to use it. 61978 61979 10. FUTURE REVISIONS OF THIS LICENSE 61980 61981 The Free Software Foundation may publish new, revised versions of 61982 the GNU Free Documentation License from time to time. Such new 61983 versions will be similar in spirit to the present version, but may 61984 differ in detail to address new problems or concerns. See 61985 <https://www.gnu.org/copyleft/>. 61986 61987 Each version of the License is given a distinguishing version 61988 number. If the Document specifies that a particular numbered 61989 version of this License "or any later version" applies to it, you 61990 have the option of following the terms and conditions either of 61991 that specified version or of any later version that has been 61992 published (not as a draft) by the Free Software Foundation. If the 61993 Document does not specify a version number of this License, you may 61994 choose any version ever published (not as a draft) by the Free 61995 Software Foundation. If the Document specifies that a proxy can 61996 decide which future versions of this License can be used, that 61997 proxy's public statement of acceptance of a version permanently 61998 authorizes you to choose that version for the Document. 61999 62000 11. RELICENSING 62001 62002 "Massive Multiauthor Collaboration Site" (or "MMC Site") means any 62003 World Wide Web server that publishes copyrightable works and also 62004 provides prominent facilities for anybody to edit those works. A 62005 public wiki that anybody can edit is an example of such a server. 62006 A "Massive Multiauthor Collaboration" (or "MMC") contained in the 62007 site means any set of copyrightable works thus published on the MMC 62008 site. 62009 62010 "CC-BY-SA" means the Creative Commons Attribution-Share Alike 3.0 62011 license published by Creative Commons Corporation, a not-for-profit 62012 corporation with a principal place of business in San Francisco, 62013 California, as well as future copyleft versions of that license 62014 published by that same organization. 62015 62016 "Incorporate" means to publish or republish a Document, in whole or 62017 in part, as part of another Document. 62018 62019 An MMC is "eligible for relicensing" if it is licensed under this 62020 License, and if all works that were first published under this 62021 License somewhere other than this MMC, and subsequently 62022 incorporated in whole or in part into the MMC, (1) had no cover 62023 texts or invariant sections, and (2) were thus incorporated prior 62024 to November 1, 2008. 62025 62026 The operator of an MMC Site may republish an MMC contained in the 62027 site under CC-BY-SA on the same site at any time before August 1, 62028 2009, provided the MMC is eligible for relicensing. 62029 62030ADDENDUM: How to use this License for your documents 62031==================================================== 62032 62033To use this License in a document you have written, include a copy of 62034the License in the document and put the following copyright and license 62035notices just after the title page: 62036 62037 Copyright (C) YEAR YOUR NAME. 62038 Permission is granted to copy, distribute and/or modify this document 62039 under the terms of the GNU Free Documentation License, Version 1.3 62040 or any later version published by the Free Software Foundation; 62041 with no Invariant Sections, no Front-Cover Texts, and no Back-Cover 62042 Texts. A copy of the license is included in the section entitled ``GNU 62043 Free Documentation License''. 62044 62045 If you have Invariant Sections, Front-Cover Texts and Back-Cover Texts, 62046replace the "with...Texts." line with this: 62047 62048 with the Invariant Sections being LIST THEIR TITLES, with 62049 the Front-Cover Texts being LIST, and with the Back-Cover Texts 62050 being LIST. 62051 62052 If you have Invariant Sections without Cover Texts, or some other 62053combination of the three, merge those two alternatives to suit the 62054situation. 62055 62056 If your document contains nontrivial examples of program code, we 62057recommend releasing these examples in parallel under your choice of free 62058software license, such as the GNU General Public License, to permit 62059their use in free software. 62060 62061 62062File: gcc.info, Node: Contributors, Next: Option Index, Prev: GNU Free Documentation License, Up: Top 62063 62064Contributors to GCC 62065******************* 62066 62067The GCC project would like to thank its many contributors. Without them 62068the project would not have been nearly as successful as it has been. 62069Any omissions in this list are accidental. Feel free to contact 62070<law@redhat.com> or <gerald@pfeifer.com> if you have been left out or 62071some of your contributions are not listed. Please keep this list in 62072alphabetical order. 62073 62074 * Analog Devices helped implement the support for complex data types 62075 and iterators. 62076 62077 * John David Anglin for threading-related fixes and improvements to 62078 libstdc++-v3, and the HP-UX port. 62079 62080 * James van Artsdalen wrote the code that makes efficient use of the 62081 Intel 80387 register stack. 62082 62083 * Abramo and Roberto Bagnara for the SysV68 Motorola 3300 Delta 62084 Series port. 62085 62086 * Alasdair Baird for various bug fixes. 62087 62088 * Giovanni Bajo for analyzing lots of complicated C++ problem 62089 reports. 62090 62091 * Peter Barada for his work to improve code generation for new 62092 ColdFire cores. 62093 62094 * Gerald Baumgartner added the signature extension to the C++ front 62095 end. 62096 62097 * Godmar Back for his Java improvements and encouragement. 62098 62099 * Scott Bambrough for help porting the Java compiler. 62100 62101 * Wolfgang Bangerth for processing tons of bug reports. 62102 62103 * Jon Beniston for his Microsoft Windows port of Java and port to 62104 Lattice Mico32. 62105 62106 * Daniel Berlin for better DWARF 2 support, faster/better 62107 optimizations, improved alias analysis, plus migrating GCC to 62108 Bugzilla. 62109 62110 * Geoff Berry for his Java object serialization work and various 62111 patches. 62112 62113 * David Binderman tests weekly snapshots of GCC trunk against Fedora 62114 Rawhide for several architectures. 62115 62116 * Laurynas Biveinis for memory management work and DJGPP port fixes. 62117 62118 * Uros Bizjak for the implementation of x87 math built-in functions 62119 and for various middle end and i386 back end improvements and bug 62120 fixes. 62121 62122 * Eric Blake for helping to make GCJ and libgcj conform to the 62123 specifications. 62124 62125 * Janne Blomqvist for contributions to GNU Fortran. 62126 62127 * Hans-J. Boehm for his garbage collector, IA-64 libffi port, and 62128 other Java work. 62129 62130 * Segher Boessenkool for helping maintain the PowerPC port and the 62131 instruction combiner plus various contributions to the middle end. 62132 62133 * Neil Booth for work on cpplib, lang hooks, debug hooks and other 62134 miscellaneous clean-ups. 62135 62136 * Steven Bosscher for integrating the GNU Fortran front end into GCC 62137 and for contributing to the tree-ssa branch. 62138 62139 * Eric Botcazou for fixing middle- and backend bugs left and right. 62140 62141 * Per Bothner for his direction via the steering committee and 62142 various improvements to the infrastructure for supporting new 62143 languages. Chill front end implementation. Initial 62144 implementations of cpplib, fix-header, config.guess, libio, and 62145 past C++ library (libg++) maintainer. Dreaming up, designing and 62146 implementing much of GCJ. 62147 62148 * Devon Bowen helped port GCC to the Tahoe. 62149 62150 * Don Bowman for mips-vxworks contributions. 62151 62152 * James Bowman for the FT32 port. 62153 62154 * Dave Brolley for work on cpplib and Chill. 62155 62156 * Paul Brook for work on the ARM architecture and maintaining GNU 62157 Fortran. 62158 62159 * Robert Brown implemented the support for Encore 32000 systems. 62160 62161 * Christian Bruel for improvements to local store elimination. 62162 62163 * Herman A.J. ten Brugge for various fixes. 62164 62165 * Joerg Brunsmann for Java compiler hacking and help with the GCJ 62166 FAQ. 62167 62168 * Joe Buck for his direction via the steering committee from its 62169 creation to 2013. 62170 62171 * Iain Buclaw for the D frontend. 62172 62173 * Craig Burley for leadership of the G77 Fortran effort. 62174 62175 * Tobias Burnus for contributions to GNU Fortran. 62176 62177 * Stephan Buys for contributing Doxygen notes for libstdc++. 62178 62179 * Paolo Carlini for libstdc++ work: lots of efficiency improvements 62180 to the C++ strings, streambufs and formatted I/O, hard detective 62181 work on the frustrating localization issues, and keeping up with 62182 the problem reports. 62183 62184 * John Carr for his alias work, SPARC hacking, infrastructure 62185 improvements, previous contributions to the steering committee, 62186 loop optimizations, etc. 62187 62188 * Stephane Carrez for 68HC11 and 68HC12 ports. 62189 62190 * Steve Chamberlain for support for the Renesas SH and H8 processors 62191 and the PicoJava processor, and for GCJ config fixes. 62192 62193 * Glenn Chambers for help with the GCJ FAQ. 62194 62195 * John-Marc Chandonia for various libgcj patches. 62196 62197 * Denis Chertykov for contributing and maintaining the AVR port, the 62198 first GCC port for an 8-bit architecture. 62199 62200 * Kito Cheng for his work on the RISC-V port, including bringing up 62201 the test suite and maintenance. 62202 62203 * Scott Christley for his Objective-C contributions. 62204 62205 * Eric Christopher for his Java porting help and clean-ups. 62206 62207 * Branko Cibej for more warning contributions. 62208 62209 * The GNU Classpath project for all of their merged runtime code. 62210 62211 * Nick Clifton for arm, mcore, fr30, v850, m32r, msp430 rx work, 62212 '--help', and other random hacking. 62213 62214 * Michael Cook for libstdc++ cleanup patches to reduce warnings. 62215 62216 * R. Kelley Cook for making GCC buildable from a read-only directory 62217 as well as other miscellaneous build process and documentation 62218 clean-ups. 62219 62220 * Ralf Corsepius for SH testing and minor bug fixing. 62221 62222 * Franc,ois-Xavier Coudert for contributions to GNU Fortran. 62223 62224 * Stan Cox for care and feeding of the x86 port and lots of behind 62225 the scenes hacking. 62226 62227 * Alex Crain provided changes for the 3b1. 62228 62229 * Ian Dall for major improvements to the NS32k port. 62230 62231 * Paul Dale for his work to add uClinux platform support to the m68k 62232 backend. 62233 62234 * Palmer Dabbelt for his work maintaining the RISC-V port. 62235 62236 * Dario Dariol contributed the four varieties of sample programs that 62237 print a copy of their source. 62238 62239 * Russell Davidson for fstream and stringstream fixes in libstdc++. 62240 62241 * Bud Davis for work on the G77 and GNU Fortran compilers. 62242 62243 * Mo DeJong for GCJ and libgcj bug fixes. 62244 62245 * Jerry DeLisle for contributions to GNU Fortran. 62246 62247 * DJ Delorie for the DJGPP port, build and libiberty maintenance, 62248 various bug fixes, and the M32C, MeP, MSP430, and RL78 ports. 62249 62250 * Arnaud Desitter for helping to debug GNU Fortran. 62251 62252 * Gabriel Dos Reis for contributions to G++, contributions and 62253 maintenance of GCC diagnostics infrastructure, libstdc++-v3, 62254 including 'valarray<>', 'complex<>', maintaining the numerics 62255 library (including that pesky '<limits>' :-) and keeping up-to-date 62256 anything to do with numbers. 62257 62258 * Ulrich Drepper for his work on glibc, testing of GCC using glibc, 62259 ISO C99 support, CFG dumping support, etc., plus support of the C++ 62260 runtime libraries including for all kinds of C interface issues, 62261 contributing and maintaining 'complex<>', sanity checking and 62262 disbursement, configuration architecture, libio maintenance, and 62263 early math work. 62264 62265 * Franc,ois Dumont for his work on libstdc++-v3, especially 62266 maintaining and improving 'debug-mode' and associative and 62267 unordered containers. 62268 62269 * Zdenek Dvorak for a new loop unroller and various fixes. 62270 62271 * Michael Eager for his work on the Xilinx MicroBlaze port. 62272 62273 * Richard Earnshaw for his ongoing work with the ARM. 62274 62275 * David Edelsohn for his direction via the steering committee, 62276 ongoing work with the RS6000/PowerPC port, help cleaning up Haifa 62277 loop changes, doing the entire AIX port of libstdc++ with his bare 62278 hands, and for ensuring GCC properly keeps working on AIX. 62279 62280 * Kevin Ediger for the floating point formatting of num_put::do_put 62281 in libstdc++. 62282 62283 * Phil Edwards for libstdc++ work including configuration hackery, 62284 documentation maintainer, chief breaker of the web pages, the 62285 occasional iostream bug fix, and work on shared library symbol 62286 versioning. 62287 62288 * Paul Eggert for random hacking all over GCC. 62289 62290 * Mark Elbrecht for various DJGPP improvements, and for libstdc++ 62291 configuration support for locales and fstream-related fixes. 62292 62293 * Vadim Egorov for libstdc++ fixes in strings, streambufs, and 62294 iostreams. 62295 62296 * Christian Ehrhardt for dealing with bug reports. 62297 62298 * Ben Elliston for his work to move the Objective-C runtime into its 62299 own subdirectory and for his work on autoconf. 62300 62301 * Revital Eres for work on the PowerPC 750CL port. 62302 62303 * Marc Espie for OpenBSD support. 62304 62305 * Doug Evans for much of the global optimization framework, arc, 62306 m32r, and SPARC work. 62307 62308 * Christopher Faylor for his work on the Cygwin port and for caring 62309 and feeding the gcc.gnu.org box and saving its users tons of spam. 62310 62311 * Fred Fish for BeOS support and Ada fixes. 62312 62313 * Ivan Fontes Garcia for the Portuguese translation of the GCJ FAQ. 62314 62315 * Peter Gerwinski for various bug fixes and the Pascal front end. 62316 62317 * Kaveh R. Ghazi for his direction via the steering committee, 62318 amazing work to make '-W -Wall -W* -Werror' useful, and testing GCC 62319 on a plethora of platforms. Kaveh extends his gratitude to the 62320 CAIP Center at Rutgers University for providing him with computing 62321 resources to work on Free Software from the late 1980s to 2010. 62322 62323 * John Gilmore for a donation to the FSF earmarked improving GNU 62324 Java. 62325 62326 * Judy Goldberg for c++ contributions. 62327 62328 * Torbjorn Granlund for various fixes and the c-torture testsuite, 62329 multiply- and divide-by-constant optimization, improved long long 62330 support, improved leaf function register allocation, and his 62331 direction via the steering committee. 62332 62333 * Jonny Grant for improvements to 'collect2's' '--help' 62334 documentation. 62335 62336 * Anthony Green for his '-Os' contributions, the moxie port, and Java 62337 front end work. 62338 62339 * Stu Grossman for gdb hacking, allowing GCJ developers to debug Java 62340 code. 62341 62342 * Michael K. Gschwind contributed the port to the PDP-11. 62343 62344 * Richard Biener for his ongoing middle-end contributions and bug 62345 fixes and for release management. 62346 62347 * Ron Guilmette implemented the 'protoize' and 'unprotoize' tools, 62348 the support for DWARF 1 symbolic debugging information, and much of 62349 the support for System V Release 4. He has also worked heavily on 62350 the Intel 386 and 860 support. 62351 62352 * Sumanth Gundapaneni for contributing the CR16 port. 62353 62354 * Mostafa Hagog for Swing Modulo Scheduling (SMS) and post reload 62355 GCSE. 62356 62357 * Bruno Haible for improvements in the runtime overhead for EH, new 62358 warnings and assorted bug fixes. 62359 62360 * Andrew Haley for his amazing Java compiler and library efforts. 62361 62362 * Chris Hanson assisted in making GCC work on HP-UX for the 9000 62363 series 300. 62364 62365 * Michael Hayes for various thankless work he's done trying to get 62366 the c30/c40 ports functional. Lots of loop and unroll improvements 62367 and fixes. 62368 62369 * Dara Hazeghi for wading through myriads of target-specific bug 62370 reports. 62371 62372 * Kate Hedstrom for staking the G77 folks with an initial testsuite. 62373 62374 * Richard Henderson for his ongoing SPARC, alpha, ia32, and ia64 62375 work, loop opts, and generally fixing lots of old problems we've 62376 ignored for years, flow rewrite and lots of further stuff, 62377 including reviewing tons of patches. 62378 62379 * Aldy Hernandez for working on the PowerPC port, SIMD support, and 62380 various fixes. 62381 62382 * Nobuyuki Hikichi of Software Research Associates, Tokyo, 62383 contributed the support for the Sony NEWS machine. 62384 62385 * Kazu Hirata for caring and feeding the Renesas H8/300 port and 62386 various fixes. 62387 62388 * Katherine Holcomb for work on GNU Fortran. 62389 62390 * Manfred Hollstein for his ongoing work to keep the m88k alive, lots 62391 of testing and bug fixing, particularly of GCC configury code. 62392 62393 * Steve Holmgren for MachTen patches. 62394 62395 * Mat Hostetter for work on the TILE-Gx and TILEPro ports. 62396 62397 * Jan Hubicka for his x86 port improvements. 62398 62399 * Falk Hueffner for working on C and optimization bug reports. 62400 62401 * Bernardo Innocenti for his m68k work, including merging of ColdFire 62402 improvements and uClinux support. 62403 62404 * Christian Iseli for various bug fixes. 62405 62406 * Kamil Iskra for general m68k hacking. 62407 62408 * Lee Iverson for random fixes and MIPS testing. 62409 62410 * Balaji V. Iyer for Cilk+ development and merging. 62411 62412 * Andreas Jaeger for testing and benchmarking of GCC and various bug 62413 fixes. 62414 62415 * Martin Jambor for his work on inter-procedural optimizations, the 62416 switch conversion pass, and scalar replacement of aggregates. 62417 62418 * Jakub Jelinek for his SPARC work and sibling call optimizations as 62419 well as lots of bug fixes and test cases, and for improving the 62420 Java build system. 62421 62422 * Janis Johnson for ia64 testing and fixes, her quality improvement 62423 sidetracks, and web page maintenance. 62424 62425 * Kean Johnston for SCO OpenServer support and various fixes. 62426 62427 * Tim Josling for the sample language treelang based originally on 62428 Richard Kenner's "toy" language. 62429 62430 * Nicolai Josuttis for additional libstdc++ documentation. 62431 62432 * Klaus Kaempf for his ongoing work to make alpha-vms a viable 62433 target. 62434 62435 * Steven G. Kargl for work on GNU Fortran. 62436 62437 * David Kashtan of SRI adapted GCC to VMS. 62438 62439 * Ryszard Kabatek for many, many libstdc++ bug fixes and 62440 optimizations of strings, especially member functions, and for 62441 auto_ptr fixes. 62442 62443 * Geoffrey Keating for his ongoing work to make the PPC work for 62444 GNU/Linux and his automatic regression tester. 62445 62446 * Brendan Kehoe for his ongoing work with G++ and for a lot of early 62447 work in just about every part of libstdc++. 62448 62449 * Oliver M. Kellogg of Deutsche Aerospace contributed the port to the 62450 MIL-STD-1750A. 62451 62452 * Richard Kenner of the New York University Ultracomputer Research 62453 Laboratory wrote the machine descriptions for the AMD 29000, the 62454 DEC Alpha, the IBM RT PC, and the IBM RS/6000 as well as the 62455 support for instruction attributes. He also made changes to better 62456 support RISC processors including changes to common subexpression 62457 elimination, strength reduction, function calling sequence 62458 handling, and condition code support, in addition to generalizing 62459 the code for frame pointer elimination and delay slot scheduling. 62460 Richard Kenner was also the head maintainer of GCC for several 62461 years. 62462 62463 * Mumit Khan for various contributions to the Cygwin and Mingw32 62464 ports and maintaining binary releases for Microsoft Windows hosts, 62465 and for massive libstdc++ porting work to Cygwin/Mingw32. 62466 62467 * Robin Kirkham for cpu32 support. 62468 62469 * Mark Klein for PA improvements. 62470 62471 * Thomas Koenig for various bug fixes. 62472 62473 * Bruce Korb for the new and improved fixincludes code. 62474 62475 * Benjamin Kosnik for his G++ work and for leading the libstdc++-v3 62476 effort. 62477 62478 * Maxim Kuvyrkov for contributions to the instruction scheduler, the 62479 Android and m68k/Coldfire ports, and optimizations. 62480 62481 * Charles LaBrec contributed the support for the Integrated Solutions 62482 68020 system. 62483 62484 * Asher Langton and Mike Kumbera for contributing Cray pointer 62485 support to GNU Fortran, and for other GNU Fortran improvements. 62486 62487 * Jeff Law for his direction via the steering committee, coordinating 62488 the entire egcs project and GCC 2.95, rolling out snapshots and 62489 releases, handling merges from GCC2, reviewing tons of patches that 62490 might have fallen through the cracks else, and random but extensive 62491 hacking. 62492 62493 * Walter Lee for work on the TILE-Gx and TILEPro ports. 62494 62495 * Marc Lehmann for his direction via the steering committee and 62496 helping with analysis and improvements of x86 performance. 62497 62498 * Victor Leikehman for work on GNU Fortran. 62499 62500 * Ted Lemon wrote parts of the RTL reader and printer. 62501 62502 * Kriang Lerdsuwanakij for C++ improvements including template as 62503 template parameter support, and many C++ fixes. 62504 62505 * Warren Levy for tremendous work on libgcj (Java Runtime Library) 62506 and random work on the Java front end. 62507 62508 * Alain Lichnewsky ported GCC to the MIPS CPU. 62509 62510 * Oskar Liljeblad for hacking on AWT and his many Java bug reports 62511 and patches. 62512 62513 * Robert Lipe for OpenServer support, new testsuites, testing, etc. 62514 62515 * Chen Liqin for various S+core related fixes/improvement, and for 62516 maintaining the S+core port. 62517 62518 * Martin Liska for his work on identical code folding, the 62519 sanitizers, HSA, general bug fixing and for running automated 62520 regression testing of GCC and reporting numerous bugs. 62521 62522 * Weiwen Liu for testing and various bug fixes. 62523 62524 * Manuel Lo'pez-Iba'n~ez for improving '-Wconversion' and many other 62525 diagnostics fixes and improvements. 62526 62527 * Dave Love for his ongoing work with the Fortran front end and 62528 runtime libraries. 62529 62530 * Martin von Lo"wis for internal consistency checking infrastructure, 62531 various C++ improvements including namespace support, and tons of 62532 assistance with libstdc++/compiler merges. 62533 62534 * H.J. Lu for his previous contributions to the steering committee, 62535 many x86 bug reports, prototype patches, and keeping the GNU/Linux 62536 ports working. 62537 62538 * Greg McGary for random fixes and (someday) bounded pointers. 62539 62540 * Andrew MacLeod for his ongoing work in building a real EH system, 62541 various code generation improvements, work on the global optimizer, 62542 etc. 62543 62544 * Vladimir Makarov for hacking some ugly i960 problems, PowerPC 62545 hacking improvements to compile-time performance, overall knowledge 62546 and direction in the area of instruction scheduling, design and 62547 implementation of the automaton based instruction scheduler and 62548 design and implementation of the integrated and local register 62549 allocators. 62550 62551 * David Malcolm for his work on improving GCC diagnostics, JIT, 62552 self-tests and unit testing. 62553 62554 * Bob Manson for his behind the scenes work on dejagnu. 62555 62556 * John Marino for contributing the DragonFly BSD port. 62557 62558 * Philip Martin for lots of libstdc++ string and vector iterator 62559 fixes and improvements, and string clean up and testsuites. 62560 62561 * Michael Matz for his work on dominance tree discovery, the x86-64 62562 port, link-time optimization framework and general optimization 62563 improvements. 62564 62565 * All of the Mauve project contributors for Java test code. 62566 62567 * Bryce McKinlay for numerous GCJ and libgcj fixes and improvements. 62568 62569 * Adam Megacz for his work on the Microsoft Windows port of GCJ. 62570 62571 * Michael Meissner for LRS framework, ia32, m32r, v850, m88k, MIPS, 62572 powerpc, haifa, ECOFF debug support, and other assorted hacking. 62573 62574 * Jason Merrill for his direction via the steering committee and 62575 leading the G++ effort. 62576 62577 * Martin Michlmayr for testing GCC on several architectures using the 62578 entire Debian archive. 62579 62580 * David Miller for his direction via the steering committee, lots of 62581 SPARC work, improvements in jump.cc and interfacing with the Linux 62582 kernel developers. 62583 62584 * Gary Miller ported GCC to Charles River Data Systems machines. 62585 62586 * Alfred Minarik for libstdc++ string and ios bug fixes, and turning 62587 the entire libstdc++ testsuite namespace-compatible. 62588 62589 * Mark Mitchell for his direction via the steering committee, 62590 mountains of C++ work, load/store hoisting out of loops, alias 62591 analysis improvements, ISO C 'restrict' support, and serving as 62592 release manager from 2000 to 2011. 62593 62594 * Alan Modra for various GNU/Linux bits and testing. 62595 62596 * Toon Moene for his direction via the steering committee, Fortran 62597 maintenance, and his ongoing work to make us make Fortran run fast. 62598 62599 * Jason Molenda for major help in the care and feeding of all the 62600 services on the gcc.gnu.org (formerly egcs.cygnus.com) 62601 machine--mail, web services, ftp services, etc etc. Doing all this 62602 work on scrap paper and the backs of envelopes would have been... 62603 difficult. 62604 62605 * Catherine Moore for fixing various ugly problems we have sent her 62606 way, including the haifa bug which was killing the Alpha & PowerPC 62607 Linux kernels. 62608 62609 * Mike Moreton for his various Java patches. 62610 62611 * David Mosberger-Tang for various Alpha improvements, and for the 62612 initial IA-64 port. 62613 62614 * Stephen Moshier contributed the floating point emulator that 62615 assists in cross-compilation and permits support for floating point 62616 numbers wider than 64 bits and for ISO C99 support. 62617 62618 * Bill Moyer for his behind the scenes work on various issues. 62619 62620 * Philippe De Muyter for his work on the m68k port. 62621 62622 * Joseph S. Myers for his work on the PDP-11 port, format checking 62623 and ISO C99 support, and continuous emphasis on (and contributions 62624 to) documentation. 62625 62626 * Nathan Myers for his work on libstdc++-v3: architecture and 62627 authorship through the first three snapshots, including 62628 implementation of locale infrastructure, string, shadow C headers, 62629 and the initial project documentation (DESIGN, CHECKLIST, and so 62630 forth). Later, more work on MT-safe string and shadow headers. 62631 62632 * Felix Natter for documentation on porting libstdc++. 62633 62634 * Nathanael Nerode for cleaning up the configuration/build process. 62635 62636 * NeXT, Inc. donated the front end that supports the Objective-C 62637 language. 62638 62639 * Hans-Peter Nilsson for the CRIS and MMIX ports, improvements to the 62640 search engine setup, various documentation fixes and other small 62641 fixes. 62642 62643 * Geoff Noer for his work on getting cygwin native builds working. 62644 62645 * Vegard Nossum for running automated regression testing of GCC and 62646 reporting numerous bugs. 62647 62648 * Diego Novillo for his work on Tree SSA, OpenMP, SPEC performance 62649 tracking web pages, GIMPLE tuples, and assorted fixes. 62650 62651 * David O'Brien for the FreeBSD/alpha, FreeBSD/AMD x86-64, 62652 FreeBSD/ARM, FreeBSD/PowerPC, and FreeBSD/SPARC64 ports and related 62653 infrastructure improvements. 62654 62655 * Alexandre Oliva for various build infrastructure improvements, 62656 scripts and amazing testing work, including keeping libtool issues 62657 sane and happy. 62658 62659 * Stefan Olsson for work on mt_alloc. 62660 62661 * Melissa O'Neill for various NeXT fixes. 62662 62663 * Rainer Orth for random MIPS work, including improvements to GCC's 62664 o32 ABI support, improvements to dejagnu's MIPS support, Java 62665 configuration clean-ups and porting work, and maintaining the IRIX, 62666 Solaris 2, and Tru64 UNIX ports. 62667 62668 * Steven Pemberton for his contribution of 'enquire' which allowed 62669 GCC to determine various properties of the floating point unit and 62670 generate 'float.h' in older versions of GCC. 62671 62672 * Hartmut Penner for work on the s390 port. 62673 62674 * Paul Petersen wrote the machine description for the Alliant FX/8. 62675 62676 * Alexandre Petit-Bianco for implementing much of the Java compiler 62677 and continued Java maintainership. 62678 62679 * Matthias Pfaller for major improvements to the NS32k port. 62680 62681 * Gerald Pfeifer for his direction via the steering committee, 62682 pointing out lots of problems we need to solve, maintenance of the 62683 web pages, and taking care of documentation maintenance in general. 62684 62685 * Marek Polacek for his work on the C front end, the sanitizers and 62686 general bug fixing. 62687 62688 * Andrew Pinski for processing bug reports by the dozen. 62689 62690 * Ovidiu Predescu for his work on the Objective-C front end and 62691 runtime libraries. 62692 62693 * Jerry Quinn for major performance improvements in C++ formatted 62694 I/O. 62695 62696 * Ken Raeburn for various improvements to checker, MIPS ports and 62697 various cleanups in the compiler. 62698 62699 * Rolf W. Rasmussen for hacking on AWT. 62700 62701 * David Reese of Sun Microsystems contributed to the Solaris on 62702 PowerPC port. 62703 62704 * John Regehr for running automated regression testing of GCC and 62705 reporting numerous bugs. 62706 62707 * Volker Reichelt for running automated regression testing of GCC and 62708 reporting numerous bugs and for keeping up with the problem 62709 reports. 62710 62711 * Joern Rennecke for maintaining the sh port, loop, regmove & reload 62712 hacking and developing and maintaining the Epiphany port. 62713 62714 * Loren J. Rittle for improvements to libstdc++-v3 including the 62715 FreeBSD port, threading fixes, thread-related configury changes, 62716 critical threading documentation, and solutions to really tricky 62717 I/O problems, as well as keeping GCC properly working on FreeBSD 62718 and continuous testing. 62719 62720 * Craig Rodrigues for processing tons of bug reports. 62721 62722 * Ola Ro"nnerup for work on mt_alloc. 62723 62724 * Gavin Romig-Koch for lots of behind the scenes MIPS work. 62725 62726 * David Ronis inspired and encouraged Craig to rewrite the G77 62727 documentation in texinfo format by contributing a first pass at a 62728 translation of the old 'g77-0.5.16/f/DOC' file. 62729 62730 * Ken Rose for fixes to GCC's delay slot filling code. 62731 62732 * Ira Rosen for her contributions to the auto-vectorizer. 62733 62734 * Paul Rubin wrote most of the preprocessor. 62735 62736 * Pe'tur Runo'lfsson for major performance improvements in C++ 62737 formatted I/O and large file support in C++ filebuf. 62738 62739 * Chip Salzenberg for libstdc++ patches and improvements to locales, 62740 traits, Makefiles, libio, libtool hackery, and "long long" support. 62741 62742 * Juha Sarlin for improvements to the H8 code generator. 62743 62744 * Greg Satz assisted in making GCC work on HP-UX for the 9000 series 62745 300. 62746 62747 * Roger Sayle for improvements to constant folding and GCC's RTL 62748 optimizers as well as for fixing numerous bugs. 62749 62750 * Bradley Schatz for his work on the GCJ FAQ. 62751 62752 * Peter Schauer wrote the code to allow debugging to work on the 62753 Alpha. 62754 62755 * William Schelter did most of the work on the Intel 80386 support. 62756 62757 * Tobias Schlu"ter for work on GNU Fortran. 62758 62759 * Bernd Schmidt for various code generation improvements and major 62760 work in the reload pass, serving as release manager for GCC 2.95.3, 62761 and work on the Blackfin and C6X ports. 62762 62763 * Peter Schmid for constant testing of libstdc++--especially 62764 application testing, going above and beyond what was requested for 62765 the release criteria--and libstdc++ header file tweaks. 62766 62767 * Jason Schroeder for jcf-dump patches. 62768 62769 * Andreas Schwab for his work on the m68k port. 62770 62771 * Lars Segerlund for work on GNU Fortran. 62772 62773 * Dodji Seketeli for numerous C++ bug fixes and debug info 62774 improvements. 62775 62776 * Tim Shen for major work on '<regex>'. 62777 62778 * Joel Sherrill for his direction via the steering committee, RTEMS 62779 contributions and RTEMS testing. 62780 62781 * Nathan Sidwell for many C++ fixes/improvements. 62782 62783 * Jeffrey Siegal for helping RMS with the original design of GCC, 62784 some code which handles the parse tree and RTL data structures, 62785 constant folding and help with the original VAX & m68k ports. 62786 62787 * Kenny Simpson for prompting libstdc++ fixes due to defect reports 62788 from the LWG (thereby keeping GCC in line with updates from the 62789 ISO). 62790 62791 * Franz Sirl for his ongoing work with making the PPC port stable for 62792 GNU/Linux. 62793 62794 * Andrey Slepuhin for assorted AIX hacking. 62795 62796 * Trevor Smigiel for contributing the SPU port. 62797 62798 * Christopher Smith did the port for Convex machines. 62799 62800 * Danny Smith for his major efforts on the Mingw (and Cygwin) ports. 62801 Retired from GCC maintainership August 2010, having mentored two 62802 new maintainers into the role. 62803 62804 * Randy Smith finished the Sun FPA support. 62805 62806 * Ed Smith-Rowland for his continuous work on libstdc++-v3, special 62807 functions, '<random>', and various improvements to C++11 features. 62808 62809 * Scott Snyder for queue, iterator, istream, and string fixes and 62810 libstdc++ testsuite entries. Also for providing the patch to G77 62811 to add rudimentary support for 'INTEGER*1', 'INTEGER*2', and 62812 'LOGICAL*1'. 62813 62814 * Zdenek Sojka for running automated regression testing of GCC and 62815 reporting numerous bugs. 62816 62817 * Arseny Solokha for running automated regression testing of GCC and 62818 reporting numerous bugs. 62819 62820 * Jayant Sonar for contributing the CR16 port. 62821 62822 * Brad Spencer for contributions to the GLIBCPP_FORCE_NEW technique. 62823 62824 * Richard Stallman, for writing the original GCC and launching the 62825 GNU project. 62826 62827 * Jan Stein of the Chalmers Computer Society provided support for 62828 Genix, as well as part of the 32000 machine description. 62829 62830 * Gerhard Steinmetz for running automated regression testing of GCC 62831 and reporting numerous bugs. 62832 62833 * Nigel Stephens for various mips16 related fixes/improvements. 62834 62835 * Jonathan Stone wrote the machine description for the Pyramid 62836 computer. 62837 62838 * Graham Stott for various infrastructure improvements. 62839 62840 * John Stracke for his Java HTTP protocol fixes. 62841 62842 * Mike Stump for his Elxsi port, G++ contributions over the years and 62843 more recently his vxworks contributions 62844 62845 * Jeff Sturm for Java porting help, bug fixes, and encouragement. 62846 62847 * Zhendong Su for running automated regression testing of GCC and 62848 reporting numerous bugs. 62849 62850 * Chengnian Sun for running automated regression testing of GCC and 62851 reporting numerous bugs. 62852 62853 * Shigeya Suzuki for this fixes for the bsdi platforms. 62854 62855 * Ian Lance Taylor for the Go frontend, the initial mips16 and mips64 62856 support, general configury hacking, fixincludes, etc. 62857 62858 * Holger Teutsch provided the support for the Clipper CPU. 62859 62860 * Gary Thomas for his ongoing work to make the PPC work for 62861 GNU/Linux. 62862 62863 * Paul Thomas for contributions to GNU Fortran. 62864 62865 * Philipp Thomas for random bug fixes throughout the compiler 62866 62867 * Jason Thorpe for thread support in libstdc++ on NetBSD. 62868 62869 * Kresten Krab Thorup wrote the run time support for the Objective-C 62870 language and the fantastic Java bytecode interpreter. 62871 62872 * Michael Tiemann for random bug fixes, the first instruction 62873 scheduler, initial C++ support, function integration, NS32k, SPARC 62874 and M88k machine description work, delay slot scheduling. 62875 62876 * Andreas Tobler for his work porting libgcj to Darwin. 62877 62878 * Teemu Torma for thread safe exception handling support. 62879 62880 * Leonard Tower wrote parts of the parser, RTL generator, and RTL 62881 definitions, and of the VAX machine description. 62882 62883 * Daniel Towner and Hariharan Sandanagobalane contributed and 62884 maintain the picoChip port. 62885 62886 * Tom Tromey for internationalization support and for his many Java 62887 contributions and libgcj maintainership. 62888 62889 * Lassi Tuura for improvements to config.guess to determine HP 62890 processor types. 62891 62892 * Petter Urkedal for libstdc++ CXXFLAGS, math, and algorithms fixes. 62893 62894 * Andy Vaught for the design and initial implementation of the GNU 62895 Fortran front end. 62896 62897 * Brent Verner for work with the libstdc++ cshadow files and their 62898 associated configure steps. 62899 62900 * Todd Vierling for contributions for NetBSD ports. 62901 62902 * Andrew Waterman for contributing the RISC-V port, as well as 62903 maintaining it. 62904 62905 * Jonathan Wakely for contributing libstdc++ Doxygen notes and XHTML 62906 guidance and maintaining libstdc++. 62907 62908 * Dean Wakerley for converting the install documentation from HTML to 62909 texinfo in time for GCC 3.0. 62910 62911 * Krister Walfridsson for random bug fixes. 62912 62913 * Feng Wang for contributions to GNU Fortran. 62914 62915 * Stephen M. Webb for time and effort on making libstdc++ shadow 62916 files work with the tricky Solaris 8+ headers, and for pushing the 62917 build-time header tree. Also, for starting and driving the 62918 '<regex>' effort. 62919 62920 * John Wehle for various improvements for the x86 code generator, 62921 related infrastructure improvements to help x86 code generation, 62922 value range propagation and other work, WE32k port. 62923 62924 * Ulrich Weigand for work on the s390 port. 62925 62926 * Janus Weil for contributions to GNU Fortran. 62927 62928 * Zack Weinberg for major work on cpplib and various other bug fixes. 62929 62930 * Matt Welsh for help with Linux Threads support in GCJ. 62931 62932 * Urban Widmark for help fixing java.io. 62933 62934 * Mark Wielaard for new Java library code and his work integrating 62935 with Classpath. 62936 62937 * Dale Wiles helped port GCC to the Tahoe. 62938 62939 * Bob Wilson from Tensilica, Inc. for the Xtensa port. 62940 62941 * Jim Wilson for his direction via the steering committee, tackling 62942 hard problems in various places that nobody else wanted to work on, 62943 strength reduction and other loop optimizations. 62944 62945 * Paul Woegerer and Tal Agmon for the CRX port. 62946 62947 * Carlo Wood for various fixes. 62948 62949 * Tom Wood for work on the m88k port. 62950 62951 * Chung-Ju Wu for his work on the Andes NDS32 port. 62952 62953 * Canqun Yang for work on GNU Fortran. 62954 62955 * Masanobu Yuhara of Fujitsu Laboratories implemented the machine 62956 description for the Tron architecture (specifically, the Gmicro). 62957 62958 * Kevin Zachmann helped port GCC to the Tahoe. 62959 62960 * Ayal Zaks for Swing Modulo Scheduling (SMS). 62961 62962 * Qirun Zhang for running automated regression testing of GCC and 62963 reporting numerous bugs. 62964 62965 * Xiaoqiang Zhang for work on GNU Fortran. 62966 62967 * Gilles Zunino for help porting Java to Irix. 62968 62969 The following people are recognized for their contributions to GNAT, 62970the Ada front end of GCC: 62971 * Bernard Banner 62972 62973 * Romain Berrendonner 62974 62975 * Geert Bosch 62976 62977 * Emmanuel Briot 62978 62979 * Joel Brobecker 62980 62981 * Ben Brosgol 62982 62983 * Vincent Celier 62984 62985 * Arnaud Charlet 62986 62987 * Chien Chieng 62988 62989 * Cyrille Comar 62990 62991 * Cyrille Crozes 62992 62993 * Robert Dewar 62994 62995 * Gary Dismukes 62996 62997 * Robert Duff 62998 62999 * Ed Falis 63000 63001 * Ramon Fernandez 63002 63003 * Sam Figueroa 63004 63005 * Vasiliy Fofanov 63006 63007 * Michael Friess 63008 63009 * Franco Gasperoni 63010 63011 * Ted Giering 63012 63013 * Matthew Gingell 63014 63015 * Laurent Guerby 63016 63017 * Jerome Guitton 63018 63019 * Olivier Hainque 63020 63021 * Jerome Hugues 63022 63023 * Hristian Kirtchev 63024 63025 * Jerome Lambourg 63026 63027 * Bruno Leclerc 63028 63029 * Albert Lee 63030 63031 * Sean McNeil 63032 63033 * Javier Miranda 63034 63035 * Laurent Nana 63036 63037 * Pascal Obry 63038 63039 * Dong-Ik Oh 63040 63041 * Laurent Pautet 63042 63043 * Brett Porter 63044 63045 * Thomas Quinot 63046 63047 * Nicolas Roche 63048 63049 * Pat Rogers 63050 63051 * Jose Ruiz 63052 63053 * Douglas Rupp 63054 63055 * Sergey Rybin 63056 63057 * Gail Schenker 63058 63059 * Ed Schonberg 63060 63061 * Nicolas Setton 63062 63063 * Samuel Tardieu 63064 63065 The following people are recognized for their contributions of new 63066features, bug reports, testing and integration of classpath/libgcj for 63067GCC version 4.1: 63068 * Lillian Angel for 'JTree' implementation and lots Free Swing 63069 additions and bug fixes. 63070 63071 * Wolfgang Baer for 'GapContent' bug fixes. 63072 63073 * Anthony Balkissoon for 'JList', Free Swing 1.5 updates and mouse 63074 event fixes, lots of Free Swing work including 'JTable' editing. 63075 63076 * Stuart Ballard for RMI constant fixes. 63077 63078 * Goffredo Baroncelli for 'HTTPURLConnection' fixes. 63079 63080 * Gary Benson for 'MessageFormat' fixes. 63081 63082 * Daniel Bonniot for 'Serialization' fixes. 63083 63084 * Chris Burdess for lots of gnu.xml and http protocol fixes, 'StAX' 63085 and 'DOM xml:id' support. 63086 63087 * Ka-Hing Cheung for 'TreePath' and 'TreeSelection' fixes. 63088 63089 * Archie Cobbs for build fixes, VM interface updates, 63090 'URLClassLoader' updates. 63091 63092 * Kelley Cook for build fixes. 63093 63094 * Martin Cordova for Suggestions for better 'SocketTimeoutException'. 63095 63096 * David Daney for 'BitSet' bug fixes, 'HttpURLConnection' rewrite and 63097 improvements. 63098 63099 * Thomas Fitzsimmons for lots of upgrades to the gtk+ AWT and Cairo 63100 2D support. Lots of imageio framework additions, lots of AWT and 63101 Free Swing bug fixes. 63102 63103 * Jeroen Frijters for 'ClassLoader' and nio cleanups, serialization 63104 fixes, better 'Proxy' support, bug fixes and IKVM integration. 63105 63106 * Santiago Gala for 'AccessControlContext' fixes. 63107 63108 * Nicolas Geoffray for 'VMClassLoader' and 'AccessController' 63109 improvements. 63110 63111 * David Gilbert for 'basic' and 'metal' icon and plaf support and 63112 lots of documenting, Lots of Free Swing and metal theme additions. 63113 'MetalIconFactory' implementation. 63114 63115 * Anthony Green for 'MIDI' framework, 'ALSA' and 'DSSI' providers. 63116 63117 * Andrew Haley for 'Serialization' and 'URLClassLoader' fixes, gcj 63118 build speedups. 63119 63120 * Kim Ho for 'JFileChooser' implementation. 63121 63122 * Andrew John Hughes for 'Locale' and net fixes, URI RFC2986 updates, 63123 'Serialization' fixes, 'Properties' XML support and generic branch 63124 work, VMIntegration guide update. 63125 63126 * Bastiaan Huisman for 'TimeZone' bug fixing. 63127 63128 * Andreas Jaeger for mprec updates. 63129 63130 * Paul Jenner for better '-Werror' support. 63131 63132 * Ito Kazumitsu for 'NetworkInterface' implementation and updates. 63133 63134 * Roman Kennke for 'BoxLayout', 'GrayFilter' and 'SplitPane', plus 63135 bug fixes all over. Lots of Free Swing work including styled text. 63136 63137 * Simon Kitching for 'String' cleanups and optimization suggestions. 63138 63139 * Michael Koch for configuration fixes, 'Locale' updates, bug and 63140 build fixes. 63141 63142 * Guilhem Lavaux for configuration, thread and channel fixes and 63143 Kaffe integration. JCL native 'Pointer' updates. Logger bug 63144 fixes. 63145 63146 * David Lichteblau for JCL support library global/local reference 63147 cleanups. 63148 63149 * Aaron Luchko for JDWP updates and documentation fixes. 63150 63151 * Ziga Mahkovec for 'Graphics2D' upgraded to Cairo 0.5 and new regex 63152 features. 63153 63154 * Sven de Marothy for BMP imageio support, CSS and 'TextLayout' 63155 fixes. 'GtkImage' rewrite, 2D, awt, free swing and date/time fixes 63156 and implementing the Qt4 peers. 63157 63158 * Casey Marshall for crypto algorithm fixes, 'FileChannel' lock, 63159 'SystemLogger' and 'FileHandler' rotate implementations, NIO 63160 'FileChannel.map' support, security and policy updates. 63161 63162 * Bryce McKinlay for RMI work. 63163 63164 * Audrius Meskauskas for lots of Free Corba, RMI and HTML work plus 63165 testing and documenting. 63166 63167 * Kalle Olavi Niemitalo for build fixes. 63168 63169 * Rainer Orth for build fixes. 63170 63171 * Andrew Overholt for 'File' locking fixes. 63172 63173 * Ingo Proetel for 'Image', 'Logger' and 'URLClassLoader' updates. 63174 63175 * Olga Rodimina for 'MenuSelectionManager' implementation. 63176 63177 * Jan Roehrich for 'BasicTreeUI' and 'JTree' fixes. 63178 63179 * Julian Scheid for documentation updates and gjdoc support. 63180 63181 * Christian Schlichtherle for zip fixes and cleanups. 63182 63183 * Robert Schuster for documentation updates and beans fixes, 63184 'TreeNode' enumerations and 'ActionCommand' and various fixes, XML 63185 and URL, AWT and Free Swing bug fixes. 63186 63187 * Keith Seitz for lots of JDWP work. 63188 63189 * Christian Thalinger for 64-bit cleanups, Configuration and VM 63190 interface fixes and 'CACAO' integration, 'fdlibm' updates. 63191 63192 * Gael Thomas for 'VMClassLoader' boot packages support suggestions. 63193 63194 * Andreas Tobler for Darwin and Solaris testing and fixing, 'Qt4' 63195 support for Darwin/OS X, 'Graphics2D' support, 'gtk+' updates. 63196 63197 * Dalibor Topic for better 'DEBUG' support, build cleanups and Kaffe 63198 integration. 'Qt4' build infrastructure, 'SHA1PRNG' and 63199 'GdkPixbugDecoder' updates. 63200 63201 * Tom Tromey for Eclipse integration, generics work, lots of bug 63202 fixes and gcj integration including coordinating The Big Merge. 63203 63204 * Mark Wielaard for bug fixes, packaging and release management, 63205 'Clipboard' implementation, system call interrupts and network 63206 timeouts and 'GdkPixpufDecoder' fixes. 63207 63208 In addition to the above, all of which also contributed time and energy 63209in testing GCC, we would like to thank the following for their 63210contributions to testing: 63211 63212 * Michael Abd-El-Malek 63213 63214 * Thomas Arend 63215 63216 * Bonzo Armstrong 63217 63218 * Steven Ashe 63219 63220 * Chris Baldwin 63221 63222 * David Billinghurst 63223 63224 * Jim Blandy 63225 63226 * Stephane Bortzmeyer 63227 63228 * Horst von Brand 63229 63230 * Frank Braun 63231 63232 * Rodney Brown 63233 63234 * Sidney Cadot 63235 63236 * Bradford Castalia 63237 63238 * Robert Clark 63239 63240 * Jonathan Corbet 63241 63242 * Ralph Doncaster 63243 63244 * Richard Emberson 63245 63246 * Levente Farkas 63247 63248 * Graham Fawcett 63249 63250 * Mark Fernyhough 63251 63252 * Robert A. French 63253 63254 * Jo"rgen Freyh 63255 63256 * Mark K. Gardner 63257 63258 * Charles-Antoine Gauthier 63259 63260 * Yung Shing Gene 63261 63262 * David Gilbert 63263 63264 * Simon Gornall 63265 63266 * Fred Gray 63267 63268 * John Griffin 63269 63270 * Patrik Hagglund 63271 63272 * Phil Hargett 63273 63274 * Amancio Hasty 63275 63276 * Takafumi Hayashi 63277 63278 * Bryan W. Headley 63279 63280 * Kevin B. Hendricks 63281 63282 * Joep Jansen 63283 63284 * Christian Joensson 63285 63286 * Michel Kern 63287 63288 * David Kidd 63289 63290 * Tobias Kuipers 63291 63292 * Anand Krishnaswamy 63293 63294 * A. O. V. Le Blanc 63295 63296 * llewelly 63297 63298 * Damon Love 63299 63300 * Brad Lucier 63301 63302 * Matthias Klose 63303 63304 * Martin Knoblauch 63305 63306 * Rick Lutowski 63307 63308 * Jesse Macnish 63309 63310 * Stefan Morrell 63311 63312 * Anon A. Mous 63313 63314 * Matthias Mueller 63315 63316 * Pekka Nikander 63317 63318 * Rick Niles 63319 63320 * Jon Olson 63321 63322 * Magnus Persson 63323 63324 * Chris Pollard 63325 63326 * Richard Polton 63327 63328 * Derk Reefman 63329 63330 * David Rees 63331 63332 * Paul Reilly 63333 63334 * Tom Reilly 63335 63336 * Torsten Rueger 63337 63338 * Danny Sadinoff 63339 63340 * Marc Schifer 63341 63342 * Erik Schnetter 63343 63344 * Wayne K. Schroll 63345 63346 * David Schuler 63347 63348 * Vin Shelton 63349 63350 * Tim Souder 63351 63352 * Adam Sulmicki 63353 63354 * Bill Thorson 63355 63356 * George Talbot 63357 63358 * Pedro A. M. Vazquez 63359 63360 * Gregory Warnes 63361 63362 * Ian Watson 63363 63364 * David E. Young 63365 63366 * And many others 63367 63368 And finally we'd like to thank everyone who uses the compiler, provides 63369feedback and generally reminds us why we're doing this work in the first 63370place. 63371 63372 63373File: gcc.info, Node: Option Index, Next: Keyword Index, Prev: Contributors, Up: Top 63374 63375Option Index 63376************ 63377 63378GCC's command line options are indexed here without any initial '-' or 63379'--'. Where an option has both positive and negative forms (such as 63380'-fOPTION' and '-fno-OPTION'), relevant entries in the manual are 63381indexed under the most appropriate form; it may sometimes be useful to 63382look up both forms. 63383 63384[index] 63385* Menu: 63386 63387* ###: Overall Options. (line 472) 63388* -march: LoongArch Options. (line 9) 63389* -mbranch-cost: LoongArch Options. (line 80) 63390* -mcheck-zero-division: LoongArch Options. (line 84) 63391* -mcond-move-float: LoongArch Options. (line 95) 63392* -mcond-move-int: LoongArch Options. (line 90) 63393* -mdouble-float: LoongArch Options. (line 75) 63394* -mmax-inline-memcpy-size: LoongArch Options. (line 118) 63395* -mmemcpy: LoongArch Options. (line 100) 63396* -msingle-float: LoongArch Options. (line 70) 63397* -msmall-data-limit: LoongArch Options. (line 114) 63398* -mstrict-align: LoongArch Options. (line 109) 63399* 80387: x86 Options. (line 603) 63400* A: Preprocessor Options. 63401 (line 337) 63402* allowable_client: Darwin Options. (line 196) 63403* all_load: Darwin Options. (line 110) 63404* analyzer: Static Analyzer Options. 63405 (line 7) 63406* ansi: Standards. (line 13) 63407* ansi <1>: C Dialect Options. (line 11) 63408* ansi <2>: Other Builtins. (line 31) 63409* ansi <3>: Non-bugs. (line 107) 63410* arch_errors_fatal: Darwin Options. (line 114) 63411* aux-info: C Dialect Options. (line 202) 63412* B: Directory Options. (line 122) 63413* Bdynamic: VxWorks Options. (line 22) 63414* bind_at_load: Darwin Options. (line 118) 63415* block-ops-unaligned-vsx: RS/6000 and PowerPC Options. 63416 (line 959) 63417* Bstatic: VxWorks Options. (line 22) 63418* bundle: Darwin Options. (line 123) 63419* bundle_loader: Darwin Options. (line 127) 63420* c: Overall Options. (line 165) 63421* C: Preprocessor Options. 63422 (line 346) 63423* c <1>: Link Options. (line 20) 63424* CC: Preprocessor Options. 63425 (line 358) 63426* client_name: Darwin Options. (line 196) 63427* compatibility_version: Darwin Options. (line 196) 63428* coverage: Instrumentation Options. 63429 (line 56) 63430* current_version: Darwin Options. (line 196) 63431* D: Preprocessor Options. 63432 (line 19) 63433* d: Preprocessor Options. 63434 (line 412) 63435* d <1>: Developer Options. (line 52) 63436* da: Developer Options. (line 246) 63437* dA: Developer Options. (line 249) 63438* dD: Preprocessor Options. 63439 (line 436) 63440* dD <1>: Developer Options. (line 253) 63441* dead_strip: Darwin Options. (line 196) 63442* dependency-file: Darwin Options. (line 196) 63443* dH: Developer Options. (line 257) 63444* dI: Preprocessor Options. 63445 (line 446) 63446* dM: Preprocessor Options. 63447 (line 421) 63448* dN: Preprocessor Options. 63449 (line 442) 63450* dp: Developer Options. (line 260) 63451* dP: Developer Options. (line 265) 63452* dU: Preprocessor Options. 63453 (line 450) 63454* dump-analyzer-exploded-nodes: Static Analyzer Options. 63455 (line 448) 63456* dump-analyzer-exploded-nodes-2: Static Analyzer Options. 63457 (line 452) 63458* dump-analyzer-exploded-nodes-3: Static Analyzer Options. 63459 (line 456) 63460* dump-analyzer-feasibility: Static Analyzer Options. 63461 (line 465) 63462* dumpbase: Overall Options. (line 252) 63463* dumpbase-ext: Overall Options. (line 325) 63464* dumpdir: Overall Options. (line 353) 63465* dumpfullversion: Developer Options. (line 1014) 63466* dumpmachine: Developer Options. (line 1002) 63467* dumpspecs: Developer Options. (line 1019) 63468* dumpversion: Developer Options. (line 1006) 63469* dx: Developer Options. (line 269) 63470* dylib_file: Darwin Options. (line 196) 63471* dylinker_install_name: Darwin Options. (line 196) 63472* dynamic: Darwin Options. (line 196) 63473* dynamiclib: Darwin Options. (line 131) 63474* E: Overall Options. (line 186) 63475* E <1>: Link Options. (line 20) 63476* e: Link Options. (line 172) 63477* EB: ARC Options. (line 590) 63478* EB <1>: C-SKY Options. (line 29) 63479* EB <2>: MIPS Options. (line 7) 63480* EL: ARC Options. (line 599) 63481* EL <1>: C-SKY Options. (line 31) 63482* EL <2>: MIPS Options. (line 10) 63483* entry: Link Options. (line 172) 63484* exported_symbols_list: Darwin Options. (line 196) 63485* F: Darwin Options. (line 31) 63486* fabi-compat-version: C++ Dialect Options. 63487 (line 92) 63488* fabi-version: C++ Dialect Options. 63489 (line 24) 63490* faccess-control: C++ Dialect Options. 63491 (line 108) 63492* fada-spec-parent: Overall Options. (line 655) 63493* faggressive-loop-optimizations: Optimize Options. (line 554) 63494* falign-functions: Optimize Options. (line 1734) 63495* falign-jumps: Optimize Options. (line 1816) 63496* falign-labels: Optimize Options. (line 1775) 63497* falign-loops: Optimize Options. (line 1795) 63498* faligned-new: C++ Dialect Options. 63499 (line 112) 63500* fallow-parameterless-variadic-functions: C Dialect Options. 63501 (line 218) 63502* fallow-store-data-races: Optimize Options. (line 1836) 63503* fanalyzer: Static Analyzer Options. 63504 (line 7) 63505* fanalyzer-call-summaries: Static Analyzer Options. 63506 (line 303) 63507* fanalyzer-checker: Static Analyzer Options. 63508 (line 312) 63509* fanalyzer-feasibility: Static Analyzer Options. 63510 (line 339) 63511* fanalyzer-fine-grained: Static Analyzer Options. 63512 (line 349) 63513* fanalyzer-show-duplicate-count: Static Analyzer Options. 63514 (line 359) 63515* fanalyzer-state-merge: Static Analyzer Options. 63516 (line 366) 63517* fanalyzer-state-purge: Static Analyzer Options. 63518 (line 374) 63519* fanalyzer-transitivity: Static Analyzer Options. 63520 (line 385) 63521* fasan-shadow-offset: Instrumentation Options. 63522 (line 619) 63523* fasm: C Dialect Options. (line 225) 63524* fassociative-math: Optimize Options. (line 2373) 63525* fasynchronous-unwind-tables: Code Gen Options. (line 159) 63526* fauto-inc-dec: Optimize Options. (line 576) 63527* fauto-profile: Optimize Options. (line 2248) 63528* fbit-tests: Code Gen Options. (line 425) 63529* fbranch-count-reg: Optimize Options. (line 428) 63530* fbranch-probabilities: Optimize Options. (line 2519) 63531* fbuiltin: C Dialect Options. (line 239) 63532* fcall-saved: Code Gen Options. (line 455) 63533* fcall-used: Code Gen Options. (line 441) 63534* fcaller-saves: Optimize Options. (line 934) 63535* fcallgraph-info: Developer Options. (line 34) 63536* fcf-protection: Instrumentation Options. 63537 (line 692) 63538* fchar8_t: C++ Dialect Options. 63539 (line 122) 63540* fcheck-new: C++ Dialect Options. 63541 (line 165) 63542* fchecking: Developer Options. (line 716) 63543* fcode-hoisting: Optimize Options. (line 975) 63544* fcombine-stack-adjustments: Optimize Options. (line 946) 63545* fcommon: Code Gen Options. (line 234) 63546* fcommon <1>: Common Variable Attributes. 63547 (line 176) 63548* fcompare-debug: Developer Options. (line 788) 63549* fcompare-debug-second: Developer Options. (line 814) 63550* fcompare-elim: Optimize Options. (line 2178) 63551* fconcepts: C++ Dialect Options. 63552 (line 176) 63553* fconcepts-ts: C++ Dialect Options. 63554 (line 176) 63555* fcond-mismatch: C Dialect Options. (line 273) 63556* fconserve-stack: Optimize Options. (line 965) 63557* fconstant-string-class: Objective-C and Objective-C++ Dialect Options. 63558 (line 30) 63559* fconstexpr-cache-depth: C++ Dialect Options. 63560 (line 192) 63561* fconstexpr-depth: C++ Dialect Options. 63562 (line 186) 63563* fconstexpr-fp-except: C++ Dialect Options. 63564 (line 202) 63565* fconstexpr-loop-limit: C++ Dialect Options. 63566 (line 213) 63567* fconstexpr-ops-limit: C++ Dialect Options. 63568 (line 218) 63569* fcoroutines: C++ Dialect Options. 63570 (line 227) 63571* fcprop-registers: Optimize Options. (line 2190) 63572* fcrossjumping: Optimize Options. (line 569) 63573* fcse-follow-jumps: Optimize Options. (line 488) 63574* fcse-skip-blocks: Optimize Options. (line 497) 63575* fcx-fortran-rules: Optimize Options. (line 2506) 63576* fcx-limited-range: Optimize Options. (line 2494) 63577* fdata-sections: Optimize Options. (line 2666) 63578* fdbg-cnt: Developer Options. (line 924) 63579* fdbg-cnt-list: Developer Options. (line 921) 63580* fdce: Optimize Options. (line 582) 63581* fdebug-cpp: Preprocessor Options. 63582 (line 457) 63583* fdebug-prefix-map: Debugging Options. (line 173) 63584* fdebug-types-section: Debugging Options. (line 235) 63585* fdeclone-ctor-dtor: Optimize Options. (line 605) 63586* fdefer-pop: Optimize Options. (line 229) 63587* fdelayed-branch: Optimize Options. (line 758) 63588* fdelete-dead-exceptions: Code Gen Options. (line 142) 63589* fdelete-null-pointer-checks: Optimize Options. (line 616) 63590* fdevirtualize: Optimize Options. (line 637) 63591* fdevirtualize-at-ltrans: Optimize Options. (line 654) 63592* fdevirtualize-speculatively: Optimize Options. (line 644) 63593* fdiagnostics-color: Diagnostic Message Formatting Options. 63594 (line 55) 63595* fdiagnostics-column-origin: Diagnostic Message Formatting Options. 63596 (line 432) 63597* fdiagnostics-column-unit: Diagnostic Message Formatting Options. 63598 (line 413) 63599* fdiagnostics-escape-format: Diagnostic Message Formatting Options. 63600 (line 439) 63601* fdiagnostics-format: Diagnostic Message Formatting Options. 63602 (line 469) 63603* fdiagnostics-generate-patch: Diagnostic Message Formatting Options. 63604 (line 255) 63605* fdiagnostics-minimum-margin-width: Diagnostic Message Formatting Options. 63606 (line 224) 63607* fdiagnostics-parseable-fixits: Diagnostic Message Formatting Options. 63608 (line 228) 63609* fdiagnostics-path-format: Diagnostic Message Formatting Options. 63610 (line 305) 63611* fdiagnostics-show-caret: Diagnostic Message Formatting Options. 63612 (line 188) 63613* fdiagnostics-show-cwe: Diagnostic Message Formatting Options. 63614 (line 210) 63615* fdiagnostics-show-labels: Diagnostic Message Formatting Options. 63616 (line 197) 63617* fdiagnostics-show-line-numbers: Diagnostic Message Formatting Options. 63618 (line 219) 63619* fdiagnostics-show-location: Diagnostic Message Formatting Options. 63620 (line 40) 63621* fdiagnostics-show-option: Diagnostic Message Formatting Options. 63622 (line 182) 63623* fdiagnostics-show-path-depths: Diagnostic Message Formatting Options. 63624 (line 396) 63625* fdiagnostics-show-template-tree: Diagnostic Message Formatting Options. 63626 (line 273) 63627* fdiagnostics-urls: Diagnostic Message Formatting Options. 63628 (line 144) 63629* fdirectives-only: Preprocessor Options. 63630 (line 205) 63631* fdisable-: Developer Options. (line 647) 63632* fdollars-in-identifiers: Preprocessor Options. 63633 (line 226) 63634* fdollars-in-identifiers <1>: Interoperation. (line 141) 63635* fdpic: SH Options. (line 388) 63636* fdse: Optimize Options. (line 586) 63637* fdump-ada-spec: Overall Options. (line 650) 63638* fdump-analyzer: Static Analyzer Options. 63639 (line 430) 63640* fdump-analyzer-callgraph: Static Analyzer Options. 63641 (line 439) 63642* fdump-analyzer-exploded-graph: Static Analyzer Options. 63643 (line 443) 63644* fdump-analyzer-exploded-paths: Static Analyzer Options. 63645 (line 461) 63646* fdump-analyzer-json: Static Analyzer Options. 63647 (line 471) 63648* fdump-analyzer-state-purge: Static Analyzer Options. 63649 (line 475) 63650* fdump-analyzer-stderr: Static Analyzer Options. 63651 (line 435) 63652* fdump-analyzer-supergraph: Static Analyzer Options. 63653 (line 481) 63654* fdump-analyzer-untracked: Static Analyzer Options. 63655 (line 489) 63656* fdump-debug: Developer Options. (line 273) 63657* fdump-earlydebug: Developer Options. (line 277) 63658* fdump-final-insns: Developer Options. (line 782) 63659* fdump-go-spec: Overall Options. (line 659) 63660* fdump-ipa: Developer Options. (line 303) 63661* fdump-lang: Developer Options. (line 332) 63662* fdump-lang <1>: Developer Options. (line 339) 63663* fdump-lang-all: Developer Options. (line 339) 63664* fdump-noaddr: Developer Options. (line 281) 63665* fdump-passes: Developer Options. (line 364) 63666* fdump-rtl-alignments: Developer Options. (line 65) 63667* fdump-rtl-all: Developer Options. (line 246) 63668* fdump-rtl-asmcons: Developer Options. (line 68) 63669* fdump-rtl-auto_inc_dec: Developer Options. (line 72) 63670* fdump-rtl-barriers: Developer Options. (line 76) 63671* fdump-rtl-bbpart: Developer Options. (line 79) 63672* fdump-rtl-bbro: Developer Options. (line 82) 63673* fdump-rtl-btl2: Developer Options. (line 86) 63674* fdump-rtl-btl2 <1>: Developer Options. (line 86) 63675* fdump-rtl-bypass: Developer Options. (line 90) 63676* fdump-rtl-ce1: Developer Options. (line 101) 63677* fdump-rtl-ce2: Developer Options. (line 101) 63678* fdump-rtl-ce3: Developer Options. (line 101) 63679* fdump-rtl-combine: Developer Options. (line 93) 63680* fdump-rtl-compgotos: Developer Options. (line 96) 63681* fdump-rtl-cprop_hardreg: Developer Options. (line 105) 63682* fdump-rtl-csa: Developer Options. (line 108) 63683* fdump-rtl-cse1: Developer Options. (line 112) 63684* fdump-rtl-cse2: Developer Options. (line 112) 63685* fdump-rtl-dbr: Developer Options. (line 119) 63686* fdump-rtl-dce: Developer Options. (line 116) 63687* fdump-rtl-dce1: Developer Options. (line 123) 63688* fdump-rtl-dce2: Developer Options. (line 123) 63689* fdump-rtl-dfinish: Developer Options. (line 242) 63690* fdump-rtl-dfinit: Developer Options. (line 242) 63691* fdump-rtl-eh: Developer Options. (line 127) 63692* fdump-rtl-eh_ranges: Developer Options. (line 130) 63693* fdump-rtl-expand: Developer Options. (line 133) 63694* fdump-rtl-fwprop1: Developer Options. (line 137) 63695* fdump-rtl-fwprop2: Developer Options. (line 137) 63696* fdump-rtl-gcse1: Developer Options. (line 142) 63697* fdump-rtl-gcse2: Developer Options. (line 142) 63698* fdump-rtl-init-regs: Developer Options. (line 146) 63699* fdump-rtl-initvals: Developer Options. (line 149) 63700* fdump-rtl-into_cfglayout: Developer Options. (line 152) 63701* fdump-rtl-ira: Developer Options. (line 155) 63702* fdump-rtl-jump: Developer Options. (line 158) 63703* fdump-rtl-loop2: Developer Options. (line 161) 63704* fdump-rtl-mach: Developer Options. (line 165) 63705* fdump-rtl-mode_sw: Developer Options. (line 169) 63706* fdump-rtl-outof_cfglayout: Developer Options. (line 175) 63707* fdump-rtl-PASS: Developer Options. (line 52) 63708* fdump-rtl-peephole2: Developer Options. (line 178) 63709* fdump-rtl-postreload: Developer Options. (line 181) 63710* fdump-rtl-pro_and_epilogue: Developer Options. (line 184) 63711* fdump-rtl-ree: Developer Options. (line 192) 63712* fdump-rtl-regclass: Developer Options. (line 242) 63713* fdump-rtl-rnreg: Developer Options. (line 172) 63714* fdump-rtl-sched1: Developer Options. (line 188) 63715* fdump-rtl-sched2: Developer Options. (line 188) 63716* fdump-rtl-seqabstr: Developer Options. (line 195) 63717* fdump-rtl-shorten: Developer Options. (line 198) 63718* fdump-rtl-sibling: Developer Options. (line 201) 63719* fdump-rtl-sms: Developer Options. (line 212) 63720* fdump-rtl-split1: Developer Options. (line 208) 63721* fdump-rtl-split2: Developer Options. (line 208) 63722* fdump-rtl-split3: Developer Options. (line 208) 63723* fdump-rtl-split4: Developer Options. (line 208) 63724* fdump-rtl-split5: Developer Options. (line 208) 63725* fdump-rtl-stack: Developer Options. (line 216) 63726* fdump-rtl-subreg1: Developer Options. (line 222) 63727* fdump-rtl-subreg2: Developer Options. (line 222) 63728* fdump-rtl-subregs_of_mode_finish: Developer Options. (line 242) 63729* fdump-rtl-subregs_of_mode_init: Developer Options. (line 242) 63730* fdump-rtl-unshare: Developer Options. (line 226) 63731* fdump-rtl-vartrack: Developer Options. (line 229) 63732* fdump-rtl-vregs: Developer Options. (line 232) 63733* fdump-rtl-web: Developer Options. (line 235) 63734* fdump-statistics: Developer Options. (line 368) 63735* fdump-tree: Developer Options. (line 381) 63736* fdump-tree-all: Developer Options. (line 381) 63737* fdump-unnumbered: Developer Options. (line 291) 63738* fdump-unnumbered-links: Developer Options. (line 297) 63739* fdwarf2-cfi-asm: Debugging Options. (line 440) 63740* fearly-inlining: Optimize Options. (line 328) 63741* felide-constructors: C++ Dialect Options. 63742 (line 230) 63743* felide-type: Diagnostic Message Formatting Options. 63744 (line 293) 63745* feliminate-unused-debug-symbols: Debugging Options. (line 153) 63746* feliminate-unused-debug-types: Debugging Options. (line 444) 63747* femit-class-debug-always: Debugging Options. (line 158) 63748* femit-struct-debug-baseonly: Debugging Options. (line 371) 63749* femit-struct-debug-detailed: Debugging Options. (line 398) 63750* femit-struct-debug-reduced: Debugging Options. (line 384) 63751* fenable-: Developer Options. (line 647) 63752* fenforce-eh-specs: C++ Dialect Options. 63753 (line 241) 63754* fexceptions: Code Gen Options. (line 119) 63755* fexcess-precision: Optimize Options. (line 2299) 63756* fexec-charset: Preprocessor Options. 63757 (line 273) 63758* fexpensive-optimizations: Optimize Options. (line 661) 63759* fext-numeric-literals: C++ Dialect Options. 63760 (line 546) 63761* fextended-identifiers: Preprocessor Options. 63762 (line 229) 63763* fextern-tls-init: C++ Dialect Options. 63764 (line 251) 63765* ffast-math: Optimize Options. (line 2323) 63766* ffat-lto-objects: Optimize Options. (line 2155) 63767* ffile-prefix-map: Overall Options. (line 630) 63768* ffinite-loops: Optimize Options. (line 1222) 63769* ffinite-math-only: Optimize Options. (line 2400) 63770* ffix-and-continue: Darwin Options. (line 104) 63771* ffixed: Code Gen Options. (line 429) 63772* ffloat-store: Optimize Options. (line 2285) 63773* ffloat-store <1>: Disappointments. (line 77) 63774* ffold-simple-inlines: C++ Dialect Options. 63775 (line 272) 63776* fforward-propagate: Optimize Options. (line 236) 63777* ffp-contract: Optimize Options. (line 245) 63778* ffp-int-builtin-inexact: Optimize Options. (line 2472) 63779* ffreestanding: Standards. (line 99) 63780* ffreestanding <1>: C Dialect Options. (line 278) 63781* ffreestanding <2>: Warning Options. (line 432) 63782* ffreestanding <3>: Common Function Attributes. 63783 (line 451) 63784* ffunction-cse: Optimize Options. (line 442) 63785* ffunction-sections: Optimize Options. (line 2666) 63786* fgcse: Optimize Options. (line 511) 63787* fgcse-after-reload: Optimize Options. (line 547) 63788* fgcse-las: Optimize Options. (line 540) 63789* fgcse-lm: Optimize Options. (line 522) 63790* fgcse-sm: Optimize Options. (line 531) 63791* fgimple: C Dialect Options. (line 289) 63792* fgnu-keywords: C++ Dialect Options. 63793 (line 279) 63794* fgnu-runtime: Objective-C and Objective-C++ Dialect Options. 63795 (line 39) 63796* fgnu-tm: C Dialect Options. (line 295) 63797* fgnu-unique: Code Gen Options. (line 165) 63798* fgnu89-inline: C Dialect Options. (line 310) 63799* fgraphite-identity: Optimize Options. (line 1264) 63800* fguess-branch-probability: Optimize Options. (line 1605) 63801* fharden-compares: Instrumentation Options. 63802 (line 725) 63803* fharden-conditional-branches: Instrumentation Options. 63804 (line 734) 63805* fhoist-adjacent-loads: Optimize Options. (line 1005) 63806* fhosted: C Dialect Options. (line 327) 63807* fident: Code Gen Options. (line 255) 63808* fif-conversion: Optimize Options. (line 590) 63809* fif-conversion2: Optimize Options. (line 599) 63810* fiji: AMD GCN Options. (line 13) 63811* filelist: Darwin Options. (line 196) 63812* fimplement-inlines: C++ Dialect Options. 63813 (line 305) 63814* fimplicit-constexpr: C++ Dialect Options. 63815 (line 285) 63816* fimplicit-inline-templates: C++ Dialect Options. 63817 (line 299) 63818* fimplicit-templates: C++ Dialect Options. 63819 (line 291) 63820* findirect-data: Darwin Options. (line 104) 63821* findirect-inlining: Optimize Options. (line 300) 63822* finhibit-size-directive: Code Gen Options. (line 258) 63823* finline: Optimize Options. (line 283) 63824* finline-functions: Optimize Options. (line 308) 63825* finline-functions-called-once: Optimize Options. (line 320) 63826* finline-limit: Optimize Options. (line 344) 63827* finline-small-functions: Optimize Options. (line 291) 63828* finput-charset: Preprocessor Options. 63829 (line 288) 63830* finstrument-functions: Instrumentation Options. 63831 (line 911) 63832* finstrument-functions <1>: Common Function Attributes. 63833 (line 814) 63834* finstrument-functions-exclude-file-list: Instrumentation Options. 63835 (line 947) 63836* finstrument-functions-exclude-function-list: Instrumentation Options. 63837 (line 968) 63838* fipa-bit-cp: Optimize Options. (line 1072) 63839* fipa-cp: Optimize Options. (line 1053) 63840* fipa-cp-clone: Optimize Options. (line 1062) 63841* fipa-icf: Optimize Options. (line 1082) 63842* fipa-modref: Optimize Options. (line 1046) 63843* fipa-profile: Optimize Options. (line 1038) 63844* fipa-pta: Optimize Options. (line 1032) 63845* fipa-pure-const: Optimize Options. (line 1016) 63846* fipa-ra: Optimize Options. (line 952) 63847* fipa-reference: Optimize Options. (line 1020) 63848* fipa-reference-addressable: Optimize Options. (line 1024) 63849* fipa-sra: Optimize Options. (line 337) 63850* fipa-stack-alignment: Optimize Options. (line 1028) 63851* fipa-strict-aliasing: Optimize Options. (line 1721) 63852* fipa-vrp: Optimize Options. (line 1077) 63853* fira-algorithm: Optimize Options. (line 695) 63854* fira-hoist-pressure: Optimize Options. (line 724) 63855* fira-loop-pressure: Optimize Options. (line 731) 63856* fira-region: Optimize Options. (line 703) 63857* fira-share-save-slots: Optimize Options. (line 739) 63858* fira-share-spill-slots: Optimize Options. (line 745) 63859* fira-verbose: Developer Options. (line 851) 63860* fisolate-erroneous-paths-attribute: Optimize Options. (line 1164) 63861* fisolate-erroneous-paths-dereference: Optimize Options. (line 1156) 63862* fivar-visibility: Objective-C and Objective-C++ Dialect Options. 63863 (line 161) 63864* fivopts: Optimize Options. (line 1389) 63865* fjump-tables: Code Gen Options. (line 417) 63866* fkeep-inline-dllexport: Optimize Options. (line 369) 63867* fkeep-inline-functions: Optimize Options. (line 375) 63868* fkeep-inline-functions <1>: Inline. (line 51) 63869* fkeep-static-consts: Optimize Options. (line 386) 63870* fkeep-static-functions: Optimize Options. (line 382) 63871* flang-info-include-translate: C++ Dialect Options. 63872 (line 562) 63873* flang-info-include-translate-not: C++ Dialect Options. 63874 (line 562) 63875* flang-info-module-cmi: C++ Dialect Options. 63876 (line 571) 63877* flarge-source-files: Preprocessor Options. 63878 (line 497) 63879* flat_namespace: Darwin Options. (line 196) 63880* flax-vector-conversions: C Dialect Options. (line 335) 63881* fleading-underscore: Code Gen Options. (line 485) 63882* flifetime-dse: Optimize Options. (line 675) 63883* flinker-output: Link Options. (line 25) 63884* flive-patching: Optimize Options. (line 1096) 63885* flive-range-shrinkage: Optimize Options. (line 690) 63886* flocal-ivars: Objective-C and Objective-C++ Dialect Options. 63887 (line 152) 63888* floop-block: Optimize Options. (line 1258) 63889* floop-interchange: Optimize Options. (line 1342) 63890* floop-nest-optimize: Optimize Options. (line 1272) 63891* floop-parallelize-all: Optimize Options. (line 1278) 63892* floop-strip-mine: Optimize Options. (line 1258) 63893* floop-unroll-and-jam: Optimize Options. (line 1359) 63894* flra-remat: Optimize Options. (line 751) 63895* flto: Optimize Options. (line 1894) 63896* flto-compression-level: Optimize Options. (line 2126) 63897* flto-partition: Optimize Options. (line 2112) 63898* flto-report: Developer Options. (line 857) 63899* flto-report-wpa: Developer Options. (line 865) 63900* fmacro-prefix-map: Preprocessor Options. 63901 (line 264) 63902* fmath-errno: Optimize Options. (line 2337) 63903* fmax-errors: Warning Options. (line 18) 63904* fmax-include-depth: Preprocessor Options. 63905 (line 238) 63906* fmem-report: Developer Options. (line 869) 63907* fmem-report-wpa: Developer Options. (line 873) 63908* fmerge-all-constants: Optimize Options. (line 405) 63909* fmerge-constants: Optimize Options. (line 395) 63910* fmerge-debug-strings: Debugging Options. (line 166) 63911* fmessage-length: Diagnostic Message Formatting Options. 63912 (line 14) 63913* fmodule-header: C++ Dialect Options. 63914 (line 319) 63915* fmodule-implicit-inline: C++ Dialect Options. 63916 (line 322) 63917* fmodule-lazy: C++ Dialect Options. 63918 (line 332) 63919* fmodule-mapper: C++ Dialect Options. 63920 (line 340) 63921* fmodule-only: C++ Dialect Options. 63922 (line 345) 63923* fmodules-ts: C++ Dialect Options. 63924 (line 311) 63925* fmodulo-sched: Optimize Options. (line 416) 63926* fmodulo-sched-allow-regmoves: Optimize Options. (line 421) 63927* fmove-loop-invariants: Optimize Options. (line 2618) 63928* fmove-loop-stores: Optimize Options. (line 2622) 63929* fms-extensions: C Dialect Options. (line 340) 63930* fms-extensions <1>: C++ Dialect Options. 63931 (line 349) 63932* fms-extensions <2>: Unnamed Fields. (line 36) 63933* fnew-inheriting-ctors: C++ Dialect Options. 63934 (line 354) 63935* fnew-ttp-matching: C++ Dialect Options. 63936 (line 360) 63937* fnext-runtime: Objective-C and Objective-C++ Dialect Options. 63938 (line 43) 63939* fnil-receivers: Objective-C and Objective-C++ Dialect Options. 63940 (line 49) 63941* fno-access-control: C++ Dialect Options. 63942 (line 108) 63943* fno-allocation-dce: Optimize Options. (line 1833) 63944* fno-analyzer: Static Analyzer Options. 63945 (line 7) 63946* fno-analyzer-call-summaries: Static Analyzer Options. 63947 (line 303) 63948* fno-analyzer-feasibility: Static Analyzer Options. 63949 (line 339) 63950* fno-analyzer-fine-grained: Static Analyzer Options. 63951 (line 349) 63952* fno-analyzer-show-duplicate-count: Static Analyzer Options. 63953 (line 359) 63954* fno-analyzer-state-merge: Static Analyzer Options. 63955 (line 366) 63956* fno-analyzer-state-purge: Static Analyzer Options. 63957 (line 374) 63958* fno-analyzer-transitivity: Static Analyzer Options. 63959 (line 385) 63960* fno-asm: C Dialect Options. (line 225) 63961* fno-bit-tests: Code Gen Options. (line 425) 63962* fno-branch-count-reg: Optimize Options. (line 428) 63963* fno-builtin: C Dialect Options. (line 239) 63964* fno-builtin <1>: Warning Options. (line 432) 63965* fno-builtin <2>: Common Function Attributes. 63966 (line 451) 63967* fno-builtin <3>: Other Builtins. (line 21) 63968* fno-canonical-system-headers: Preprocessor Options. 63969 (line 234) 63970* fno-char8_t: C++ Dialect Options. 63971 (line 122) 63972* fno-checking: Developer Options. (line 716) 63973* fno-common: Code Gen Options. (line 234) 63974* fno-common <1>: Common Variable Attributes. 63975 (line 176) 63976* fno-compare-debug: Developer Options. (line 788) 63977* fno-debug-types-section: Debugging Options. (line 235) 63978* fno-default-inline: Inline. (line 68) 63979* fno-defer-pop: Optimize Options. (line 229) 63980* fno-diagnostics-show-caret: Diagnostic Message Formatting Options. 63981 (line 188) 63982* fno-diagnostics-show-cwe: Diagnostic Message Formatting Options. 63983 (line 210) 63984* fno-diagnostics-show-labels: Diagnostic Message Formatting Options. 63985 (line 197) 63986* fno-diagnostics-show-line-numbers: Diagnostic Message Formatting Options. 63987 (line 219) 63988* fno-diagnostics-show-option: Diagnostic Message Formatting Options. 63989 (line 182) 63990* fno-dwarf2-cfi-asm: Debugging Options. (line 440) 63991* fno-elide-constructors: C++ Dialect Options. 63992 (line 230) 63993* fno-elide-type: Diagnostic Message Formatting Options. 63994 (line 293) 63995* fno-eliminate-unused-debug-symbols: Debugging Options. (line 153) 63996* fno-eliminate-unused-debug-types: Debugging Options. (line 444) 63997* fno-enforce-eh-specs: C++ Dialect Options. 63998 (line 241) 63999* fno-ext-numeric-literals: C++ Dialect Options. 64000 (line 546) 64001* fno-extern-tls-init: C++ Dialect Options. 64002 (line 251) 64003* fno-finite-loops: Optimize Options. (line 1222) 64004* fno-fold-simple-inlines: C++ Dialect Options. 64005 (line 272) 64006* fno-fp-int-builtin-inexact: Optimize Options. (line 2472) 64007* fno-function-cse: Optimize Options. (line 442) 64008* fno-gnu-keywords: C++ Dialect Options. 64009 (line 279) 64010* fno-gnu-unique: Code Gen Options. (line 165) 64011* fno-guess-branch-probability: Optimize Options. (line 1605) 64012* fno-ident: Code Gen Options. (line 255) 64013* fno-implement-inlines: C++ Dialect Options. 64014 (line 305) 64015* fno-implement-inlines <1>: C++ Interface. (line 66) 64016* fno-implicit-inline-templates: C++ Dialect Options. 64017 (line 299) 64018* fno-implicit-templates: C++ Dialect Options. 64019 (line 291) 64020* fno-implicit-templates <1>: Template Instantiation. 64021 (line 94) 64022* fno-inline: Optimize Options. (line 283) 64023* fno-ira-share-save-slots: Optimize Options. (line 739) 64024* fno-ira-share-spill-slots: Optimize Options. (line 745) 64025* fno-jump-tables: Code Gen Options. (line 417) 64026* fno-keep-inline-dllexport: Optimize Options. (line 369) 64027* fno-lifetime-dse: Optimize Options. (line 675) 64028* fno-local-ivars: Objective-C and Objective-C++ Dialect Options. 64029 (line 152) 64030* fno-math-errno: Optimize Options. (line 2337) 64031* fno-merge-debug-strings: Debugging Options. (line 166) 64032* fno-module-lazy: C++ Dialect Options. 64033 (line 332) 64034* fno-modules-ts: C++ Dialect Options. 64035 (line 311) 64036* fno-nil-receivers: Objective-C and Objective-C++ Dialect Options. 64037 (line 49) 64038* fno-nonansi-builtins: C++ Dialect Options. 64039 (line 367) 64040* fno-operator-names: C++ Dialect Options. 64041 (line 383) 64042* fno-optional-diags: C++ Dialect Options. 64043 (line 387) 64044* fno-peephole: Optimize Options. (line 1596) 64045* fno-peephole2: Optimize Options. (line 1596) 64046* fno-plt: Code Gen Options. (line 399) 64047* fno-pretty-templates: C++ Dialect Options. 64048 (line 397) 64049* fno-printf-return-value: Optimize Options. (line 1573) 64050* fno-rtti: C++ Dialect Options. 64051 (line 409) 64052* fno-sanitize-recover: Instrumentation Options. 64053 (line 628) 64054* fno-sanitize=all: Instrumentation Options. 64055 (line 613) 64056* fno-sched-interblock: Optimize Options. (line 784) 64057* fno-sched-spec: Optimize Options. (line 789) 64058* fno-set-stack-executable: x86 Windows Options. 64059 (line 46) 64060* fno-show-column: Diagnostic Message Formatting Options. 64061 (line 408) 64062* fno-signed-bitfields: C Dialect Options. (line 451) 64063* fno-signed-zeros: Optimize Options. (line 2412) 64064* fno-stack-limit: Instrumentation Options. 64065 (line 823) 64066* fno-threadsafe-statics: C++ Dialect Options. 64067 (line 464) 64068* fno-toplevel-reorder: Optimize Options. (line 1859) 64069* fno-trapping-math: Optimize Options. (line 2422) 64070* fno-unsigned-bitfields: C Dialect Options. (line 451) 64071* fno-use-cxa-get-exception-ptr: C++ Dialect Options. 64072 (line 477) 64073* fno-var-tracking-assignments: Debugging Options. (line 193) 64074* fno-var-tracking-assignments-toggle: Developer Options. (line 835) 64075* fno-weak: C++ Dialect Options. 64076 (line 539) 64077* fno-working-directory: Preprocessor Options. 64078 (line 323) 64079* fno-writable-relocated-rdata: x86 Windows Options. 64080 (line 53) 64081* fno-zero-initialized-in-bss: Optimize Options. (line 453) 64082* fnon-call-exceptions: Code Gen Options. (line 133) 64083* fnonansi-builtins: C++ Dialect Options. 64084 (line 367) 64085* fnothrow-opt: C++ Dialect Options. 64086 (line 372) 64087* fobjc-abi-version: Objective-C and Objective-C++ Dialect Options. 64088 (line 56) 64089* fobjc-call-cxx-cdtors: Objective-C and Objective-C++ Dialect Options. 64090 (line 67) 64091* fobjc-direct-dispatch: Objective-C and Objective-C++ Dialect Options. 64092 (line 92) 64093* fobjc-exceptions: Objective-C and Objective-C++ Dialect Options. 64094 (line 96) 64095* fobjc-gc: Objective-C and Objective-C++ Dialect Options. 64096 (line 104) 64097* fobjc-nilcheck: Objective-C and Objective-C++ Dialect Options. 64098 (line 110) 64099* fobjc-std: Objective-C and Objective-C++ Dialect Options. 64100 (line 119) 64101* foffload: C Dialect Options. (line 360) 64102* foffload-options: C Dialect Options. (line 373) 64103* fomit-frame-pointer: Optimize Options. (line 256) 64104* fopenacc: C Dialect Options. (line 388) 64105* fopenacc-dim: C Dialect Options. (line 397) 64106* fopenmp: C Dialect Options. (line 403) 64107* fopenmp-simd: C Dialect Options. (line 412) 64108* foperator-names: C++ Dialect Options. 64109 (line 383) 64110* fopt-info: Developer Options. (line 487) 64111* foptimize-sibling-calls: Optimize Options. (line 271) 64112* foptimize-strlen: Optimize Options. (line 276) 64113* foptional-diags: C++ Dialect Options. 64114 (line 387) 64115* force_cpusubtype_ALL: Darwin Options. (line 135) 64116* force_flat_namespace: Darwin Options. (line 196) 64117* fpack-struct: Code Gen Options. (line 472) 64118* fpartial-inlining: Optimize Options. (line 1548) 64119* fpatchable-function-entry: Instrumentation Options. 64120 (line 980) 64121* fpcc-struct-return: Code Gen Options. (line 178) 64122* fpcc-struct-return <1>: Incompatibilities. (line 170) 64123* fpch-deps: Preprocessor Options. 64124 (line 298) 64125* fpch-preprocess: Preprocessor Options. 64126 (line 306) 64127* fpeel-loops: Optimize Options. (line 2610) 64128* fpeephole: Optimize Options. (line 1596) 64129* fpeephole2: Optimize Options. (line 1596) 64130* fpermissive: C++ Dialect Options. 64131 (line 392) 64132* fpermitted-flt-eval-methods: C Dialect Options. (line 416) 64133* fpermitted-flt-eval-methods=c11: C Dialect Options. (line 416) 64134* fpermitted-flt-eval-methods=ts-18661-3: C Dialect Options. (line 416) 64135* fpic: Code Gen Options. (line 356) 64136* fPIC: Code Gen Options. (line 377) 64137* fpie: Code Gen Options. (line 390) 64138* fPIE: Code Gen Options. (line 390) 64139* fplan9-extensions: C Dialect Options. (line 438) 64140* fplan9-extensions <1>: Unnamed Fields. (line 43) 64141* fplt: Code Gen Options. (line 399) 64142* fplugin: Overall Options. (line 639) 64143* fplugin-arg: Overall Options. (line 646) 64144* fpost-ipa-mem-report: Developer Options. (line 878) 64145* fpre-ipa-mem-report: Developer Options. (line 877) 64146* fpredictive-commoning: Optimize Options. (line 1555) 64147* fprefetch-loop-arrays: Optimize Options. (line 1563) 64148* fpreprocessed: Preprocessor Options. 64149 (line 192) 64150* fpretty-templates: C++ Dialect Options. 64151 (line 397) 64152* fprintf-return-value: Optimize Options. (line 1573) 64153* fprofile-abs-path: Instrumentation Options. 64154 (line 116) 64155* fprofile-arcs: Instrumentation Options. 64156 (line 30) 64157* fprofile-arcs <1>: Other Builtins. (line 596) 64158* fprofile-correction: Optimize Options. (line 2197) 64159* fprofile-dir: Instrumentation Options. 64160 (line 122) 64161* fprofile-exclude-files: Instrumentation Options. 64162 (line 299) 64163* fprofile-filter-files: Instrumentation Options. 64164 (line 291) 64165* fprofile-generate: Instrumentation Options. 64166 (line 148) 64167* fprofile-info-section: Instrumentation Options. 64168 (line 166) 64169* fprofile-note: Instrumentation Options. 64170 (line 248) 64171* fprofile-partial-training: Optimize Options. (line 2206) 64172* fprofile-prefix-map: Instrumentation Options. 64173 (line 269) 64174* fprofile-prefix-path: Instrumentation Options. 64175 (line 254) 64176* fprofile-reorder-functions: Optimize Options. (line 2550) 64177* fprofile-report: Developer Options. (line 882) 64178* fprofile-reproducible: Instrumentation Options. 64179 (line 308) 64180* fprofile-update: Instrumentation Options. 64181 (line 274) 64182* fprofile-use: Optimize Options. (line 2220) 64183* fprofile-values: Optimize Options. (line 2540) 64184* fpu: RX Options. (line 17) 64185* frandom-seed: Developer Options. (line 721) 64186* freciprocal-math: Optimize Options. (line 2390) 64187* frecord-gcc-switches: Code Gen Options. (line 344) 64188* free: Optimize Options. (line 667) 64189* freg-struct-return: Code Gen Options. (line 196) 64190* frename-registers: Optimize Options. (line 2569) 64191* freorder-blocks: Optimize Options. (line 1626) 64192* freorder-blocks-algorithm: Optimize Options. (line 1632) 64193* freorder-blocks-and-partition: Optimize Options. (line 1643) 64194* freorder-functions: Optimize Options. (line 1660) 64195* freplace-objc-classes: Objective-C and Objective-C++ Dialect Options. 64196 (line 130) 64197* freport-bug: Developer Options. (line 287) 64198* frerun-cse-after-loop: Optimize Options. (line 505) 64199* freschedule-modulo-scheduled-loops: Optimize Options. (line 883) 64200* frounding-math: Optimize Options. (line 2437) 64201* frtti: C++ Dialect Options. 64202 (line 409) 64203* fsanitize-address-use-after-scope: Instrumentation Options. 64204 (line 664) 64205* fsanitize-coverage=trace-cmp: Instrumentation Options. 64206 (line 679) 64207* fsanitize-coverage=trace-pc: Instrumentation Options. 64208 (line 675) 64209* fsanitize-recover: Instrumentation Options. 64210 (line 628) 64211* fsanitize-sections: Instrumentation Options. 64212 (line 624) 64213* fsanitize-undefined-trap-on-error: Instrumentation Options. 64214 (line 668) 64215* fsanitize=address: Instrumentation Options. 64216 (line 336) 64217* fsanitize=alignment: Instrumentation Options. 64218 (line 539) 64219* fsanitize=bool: Instrumentation Options. 64220 (line 577) 64221* fsanitize=bounds: Instrumentation Options. 64222 (line 526) 64223* fsanitize=bounds-strict: Instrumentation Options. 64224 (line 532) 64225* fsanitize=builtin: Instrumentation Options. 64226 (line 601) 64227* fsanitize=enum: Instrumentation Options. 64228 (line 582) 64229* fsanitize=float-cast-overflow: Instrumentation Options. 64230 (line 557) 64231* fsanitize=float-divide-by-zero: Instrumentation Options. 64232 (line 551) 64233* fsanitize=hwaddress: Instrumentation Options. 64234 (line 356) 64235* fsanitize=integer-divide-by-zero: Instrumentation Options. 64236 (line 489) 64237* fsanitize=kernel-address: Instrumentation Options. 64238 (line 352) 64239* fsanitize=kernel-hwaddress: Instrumentation Options. 64240 (line 370) 64241* fsanitize=leak: Instrumentation Options. 64242 (line 451) 64243* fsanitize=nonnull-attribute: Instrumentation Options. 64244 (line 565) 64245* fsanitize=null: Instrumentation Options. 64246 (line 502) 64247* fsanitize=object-size: Instrumentation Options. 64248 (line 546) 64249* fsanitize=pointer-compare: Instrumentation Options. 64250 (line 386) 64251* fsanitize=pointer-overflow: Instrumentation Options. 64252 (line 595) 64253* fsanitize=pointer-subtract: Instrumentation Options. 64254 (line 396) 64255* fsanitize=return: Instrumentation Options. 64256 (line 510) 64257* fsanitize=returns-nonnull-attribute: Instrumentation Options. 64258 (line 571) 64259* fsanitize=shadow-call-stack: Instrumentation Options. 64260 (line 406) 64261* fsanitize=shift: Instrumentation Options. 64262 (line 469) 64263* fsanitize=shift-base: Instrumentation Options. 64264 (line 482) 64265* fsanitize=shift-exponent: Instrumentation Options. 64266 (line 477) 64267* fsanitize=signed-integer-overflow: Instrumentation Options. 64268 (line 516) 64269* fsanitize=thread: Instrumentation Options. 64270 (line 437) 64271* fsanitize=undefined: Instrumentation Options. 64272 (line 461) 64273* fsanitize=unreachable: Instrumentation Options. 64274 (line 492) 64275* fsanitize=vla-bound: Instrumentation Options. 64276 (line 498) 64277* fsanitize=vptr: Instrumentation Options. 64278 (line 588) 64279* fsave-optimization-record: Developer Options. (line 593) 64280* fsched-critical-path-heuristic: Optimize Options. (line 849) 64281* fsched-dep-count-heuristic: Optimize Options. (line 876) 64282* fsched-group-heuristic: Optimize Options. (line 843) 64283* fsched-interblock: Optimize Options. (line 784) 64284* fsched-last-insn-heuristic: Optimize Options. (line 869) 64285* fsched-pressure: Optimize Options. (line 794) 64286* fsched-rank-heuristic: Optimize Options. (line 862) 64287* fsched-spec: Optimize Options. (line 789) 64288* fsched-spec-insn-heuristic: Optimize Options. (line 855) 64289* fsched-spec-load: Optimize Options. (line 803) 64290* fsched-spec-load-dangerous: Optimize Options. (line 808) 64291* fsched-stalled-insns: Optimize Options. (line 814) 64292* fsched-stalled-insns-dep: Optimize Options. (line 824) 64293* fsched-verbose: Developer Options. (line 633) 64294* fsched2-use-superblocks: Optimize Options. (line 833) 64295* fschedule-fusion: Optimize Options. (line 2579) 64296* fschedule-insns: Optimize Options. (line 765) 64297* fschedule-insns2: Optimize Options. (line 775) 64298* fsection-anchors: Optimize Options. (line 2699) 64299* fsel-sched-pipelining: Optimize Options. (line 896) 64300* fsel-sched-pipelining-outer-loops: Optimize Options. (line 901) 64301* fselective-scheduling: Optimize Options. (line 888) 64302* fselective-scheduling2: Optimize Options. (line 892) 64303* fsemantic-interposition: Optimize Options. (line 906) 64304* fset-stack-executable: x86 Windows Options. 64305 (line 46) 64306* fshort-enums: Code Gen Options. (line 214) 64307* fshort-enums <1>: Structures unions enumerations and bit-fields implementation. 64308 (line 48) 64309* fshort-enums <2>: Common Type Attributes. 64310 (line 299) 64311* fshort-enums <3>: Non-bugs. (line 42) 64312* fshort-wchar: Code Gen Options. (line 224) 64313* fshow-column: Diagnostic Message Formatting Options. 64314 (line 408) 64315* fshrink-wrap: Optimize Options. (line 923) 64316* fshrink-wrap-separate: Optimize Options. (line 928) 64317* fsignaling-nans: Optimize Options. (line 2457) 64318* fsigned-bitfields: C Dialect Options. (line 451) 64319* fsigned-bitfields <1>: Non-bugs. (line 57) 64320* fsigned-char: C Dialect Options. (line 457) 64321* fsigned-char <1>: Characters implementation. 64322 (line 31) 64323* fsigned-zeros: Optimize Options. (line 2412) 64324* fsimd-cost-model: Optimize Options. (line 1505) 64325* fsingle-precision-constant: Optimize Options. (line 2490) 64326* fsized-deallocation: C++ Dialect Options. 64327 (line 424) 64328* fsplit-ivs-in-unroller: Optimize Options. (line 1526) 64329* fsplit-loops: Optimize Options. (line 2630) 64330* fsplit-paths: Optimize Options. (line 1521) 64331* fsplit-stack: Instrumentation Options. 64332 (line 840) 64333* fsplit-stack <1>: Common Function Attributes. 64334 (line 869) 64335* fsplit-wide-types: Optimize Options. (line 474) 64336* fsplit-wide-types-early: Optimize Options. (line 482) 64337* fssa-backprop: Optimize Options. (line 1188) 64338* fssa-phiopt: Optimize Options. (line 1194) 64339* fsso-struct: C Dialect Options. (line 482) 64340* fstack-check: Instrumentation Options. 64341 (line 766) 64342* fstack-clash-protection: Instrumentation Options. 64343 (line 808) 64344* fstack-limit-register: Instrumentation Options. 64345 (line 823) 64346* fstack-limit-symbol: Instrumentation Options. 64347 (line 823) 64348* fstack-protector: Instrumentation Options. 64349 (line 741) 64350* fstack-protector-all: Instrumentation Options. 64351 (line 752) 64352* fstack-protector-explicit: Instrumentation Options. 64353 (line 762) 64354* fstack-protector-strong: Instrumentation Options. 64355 (line 755) 64356* fstack-usage: Developer Options. (line 886) 64357* fstack_reuse: Code Gen Options. (line 15) 64358* fstats: Developer Options. (line 915) 64359* fstdarg-opt: Optimize Options. (line 2695) 64360* fstore-merging: Optimize Options. (line 1413) 64361* fstrict-aliasing: Optimize Options. (line 1675) 64362* fstrict-enums: C++ Dialect Options. 64363 (line 434) 64364* fstrict-overflow: Code Gen Options. (line 115) 64365* fstrict-volatile-bitfields: Code Gen Options. (line 601) 64366* fstrong-eval-order: C++ Dialect Options. 64367 (line 443) 64368* fsync-libcalls: Code Gen Options. (line 633) 64369* fsyntax-only: Warning Options. (line 14) 64370* ftabstop: Preprocessor Options. 64371 (line 241) 64372* ftemplate-backtrace-limit: C++ Dialect Options. 64373 (line 451) 64374* ftemplate-depth: C++ Dialect Options. 64375 (line 455) 64376* ftest-coverage: Instrumentation Options. 64377 (line 107) 64378* fthread-jumps: Optimize Options. (line 465) 64379* fthreadsafe-statics: C++ Dialect Options. 64380 (line 464) 64381* ftime-report: Developer Options. (line 843) 64382* ftime-report-details: Developer Options. (line 847) 64383* ftls-model: Code Gen Options. (line 496) 64384* ftoplevel-reorder: Optimize Options. (line 1859) 64385* ftracer: Optimize Options. (line 2587) 64386* ftrack-macro-expansion: Preprocessor Options. 64387 (line 247) 64388* ftrampolines: Code Gen Options. (line 507) 64389* ftrapping-math: Optimize Options. (line 2422) 64390* ftrapv: Code Gen Options. (line 91) 64391* ftree-bit-ccp: Optimize Options. (line 1176) 64392* ftree-builtin-call-dce: Optimize Options. (line 1216) 64393* ftree-ccp: Optimize Options. (line 1183) 64394* ftree-ch: Optimize Options. (line 1245) 64395* ftree-coalesce-vars: Optimize Options. (line 1284) 64396* ftree-copy-prop: Optimize Options. (line 1011) 64397* ftree-dce: Optimize Options. (line 1212) 64398* ftree-dominator-opts: Optimize Options. (line 1231) 64399* ftree-dse: Optimize Options. (line 1238) 64400* ftree-forwprop: Optimize Options. (line 990) 64401* ftree-fre: Optimize Options. (line 994) 64402* ftree-loop-distribute-patterns: Optimize Options. (line 1320) 64403* ftree-loop-distribution: Optimize Options. (line 1301) 64404* ftree-loop-if-convert: Optimize Options. (line 1294) 64405* ftree-loop-im: Optimize Options. (line 1365) 64406* ftree-loop-ivcanon: Optimize Options. (line 1374) 64407* ftree-loop-linear: Optimize Options. (line 1258) 64408* ftree-loop-optimize: Optimize Options. (line 1252) 64409* ftree-loop-vectorize: Optimize Options. (line 1439) 64410* ftree-parallelize-loops: Optimize Options. (line 1394) 64411* ftree-partial-pre: Optimize Options. (line 986) 64412* ftree-phiprop: Optimize Options. (line 1001) 64413* ftree-pre: Optimize Options. (line 982) 64414* ftree-pta: Optimize Options. (line 1403) 64415* ftree-reassoc: Optimize Options. (line 971) 64416* ftree-scev-cprop: Optimize Options. (line 1380) 64417* ftree-sink: Optimize Options. (line 1172) 64418* ftree-slp-vectorize: Optimize Options. (line 1444) 64419* ftree-slsr: Optimize Options. (line 1428) 64420* ftree-sra: Optimize Options. (line 1407) 64421* ftree-switch-conversion: Optimize Options. (line 1199) 64422* ftree-tail-merge: Optimize Options. (line 1204) 64423* ftree-ter: Optimize Options. (line 1420) 64424* ftree-vectorize: Optimize Options. (line 1434) 64425* ftree-vrp: Optimize Options. (line 1512) 64426* ftrivial-auto-var-init: Optimize Options. (line 1449) 64427* funconstrained-commons: Optimize Options. (line 563) 64428* funit-at-a-time: Optimize Options. (line 1852) 64429* funroll-all-loops: Optimize Options. (line 2604) 64430* funroll-loops: Optimize Options. (line 2594) 64431* funsafe-math-optimizations: Optimize Options. (line 2355) 64432* funsigned-bitfields: C Dialect Options. (line 451) 64433* funsigned-bitfields <1>: Structures unions enumerations and bit-fields implementation. 64434 (line 17) 64435* funsigned-bitfields <2>: Non-bugs. (line 57) 64436* funsigned-char: C Dialect Options. (line 464) 64437* funsigned-char <1>: Characters implementation. 64438 (line 31) 64439* funswitch-loops: Optimize Options. (line 2636) 64440* funwind-tables: Code Gen Options. (line 152) 64441* fuse-cxa-atexit: C++ Dialect Options. 64442 (line 470) 64443* fuse-cxa-get-exception-ptr: C++ Dialect Options. 64444 (line 477) 64445* fuse-ld=bfd: Link Options. (line 74) 64446* fuse-ld=gold: Link Options. (line 77) 64447* fuse-ld=lld: Link Options. (line 80) 64448* fuse-ld=mold: Link Options. (line 83) 64449* fuse-linker-plugin: Optimize Options. (line 2137) 64450* fvar-tracking: Debugging Options. (line 183) 64451* fvar-tracking-assignments: Debugging Options. (line 193) 64452* fvar-tracking-assignments-toggle: Developer Options. (line 835) 64453* fvariable-expansion-in-unroller: Optimize Options. (line 1540) 64454* fvect-cost-model: Optimize Options. (line 1484) 64455* fverbose-asm: Code Gen Options. (line 265) 64456* fversion-loops-for-strides: Optimize Options. (line 2643) 64457* fvisibility: Code Gen Options. (line 536) 64458* fvisibility-inlines-hidden: C++ Dialect Options. 64459 (line 482) 64460* fvisibility-ms-compat: C++ Dialect Options. 64461 (line 510) 64462* fvpt: Optimize Options. (line 2557) 64463* fvtable-verify: Instrumentation Options. 64464 (line 858) 64465* fvtv-counts: Instrumentation Options. 64466 (line 894) 64467* fvtv-debug: Instrumentation Options. 64468 (line 881) 64469* fweak: C++ Dialect Options. 64470 (line 539) 64471* fweb: Optimize Options. (line 1872) 64472* fwhole-program: Optimize Options. (line 1883) 64473* fwide-exec-charset: Preprocessor Options. 64474 (line 278) 64475* fworking-directory: Preprocessor Options. 64476 (line 323) 64477* fwrapv: Code Gen Options. (line 99) 64478* fwrapv-pointer: Code Gen Options. (line 109) 64479* fwritable-relocated-rdata: x86 Windows Options. 64480 (line 53) 64481* fzero-call-used-regs: Optimize Options. (line 2723) 64482* fzero-initialized-in-bss: Optimize Options. (line 453) 64483* fzero-link: Objective-C and Objective-C++ Dialect Options. 64484 (line 140) 64485* g: Debugging Options. (line 28) 64486* G: ARC Options. (line 416) 64487* G <1>: M32R/D Options. (line 57) 64488* G <2>: MIPS Options. (line 460) 64489* G <3>: Nios II Options. (line 9) 64490* G <4>: RS/6000 and PowerPC Options. 64491 (line 714) 64492* G <5>: System V Options. (line 10) 64493* gas-loc-support: Debugging Options. (line 264) 64494* gas-locview-support: Debugging Options. (line 280) 64495* gbtf: Debugging Options. (line 67) 64496* gcolumn-info: Debugging Options. (line 292) 64497* gctf: Debugging Options. (line 75) 64498* gdescribe-dies: Debugging Options. (line 222) 64499* gdwarf: Debugging Options. (line 48) 64500* gdwarf32: Debugging Options. (line 212) 64501* gdwarf64: Debugging Options. (line 212) 64502* gen-decls: Objective-C and Objective-C++ Dialect Options. 64503 (line 166) 64504* gfull: Darwin Options. (line 69) 64505* ggdb: Debugging Options. (line 41) 64506* ggnu-pubnames: Debugging Options. (line 230) 64507* ginline-points: Debugging Options. (line 351) 64508* ginternal-reset-location-views: Debugging Options. (line 340) 64509* gno-as-loc-support: Debugging Options. (line 276) 64510* gno-column-info: Debugging Options. (line 292) 64511* gno-inline-points: Debugging Options. (line 351) 64512* gno-internal-reset-location-views: Debugging Options. (line 340) 64513* gno-record-gcc-switches: Debugging Options. (line 245) 64514* gno-statement-frontiers: Debugging Options. (line 297) 64515* gno-strict-dwarf: Debugging Options. (line 260) 64516* gno-variable-location-views: Debugging Options. (line 308) 64517* gpubnames: Debugging Options. (line 227) 64518* grecord-gcc-switches: Debugging Options. (line 245) 64519* gsplit-dwarf: Debugging Options. (line 204) 64520* gstabs: Debugging Options. (line 95) 64521* gstabs+: Debugging Options. (line 103) 64522* gstatement-frontiers: Debugging Options. (line 297) 64523* gstrict-dwarf: Debugging Options. (line 254) 64524* gtoggle: Developer Options. (line 827) 64525* gused: Darwin Options. (line 64) 64526* gvariable-location-views: Debugging Options. (line 308) 64527* gvariable-location-views=incompat5: Debugging Options. (line 308) 64528* gvms: Debugging Options. (line 122) 64529* gxcoff: Debugging Options. (line 109) 64530* gxcoff+: Debugging Options. (line 114) 64531* gz: Debugging Options. (line 360) 64532* H: Preprocessor Options. 64533 (line 405) 64534* headerpad_max_install_names: Darwin Options. (line 196) 64535* help: Overall Options. (line 478) 64536* I: Directory Options. (line 13) 64537* I-: Directory Options. (line 65) 64538* idirafter: Directory Options. (line 13) 64539* iframework: Darwin Options. (line 57) 64540* imacros: Preprocessor Options. 64541 (line 57) 64542* image_base: Darwin Options. (line 196) 64543* imultilib: Directory Options. (line 98) 64544* include: Preprocessor Options. 64545 (line 46) 64546* init: Darwin Options. (line 196) 64547* install_name: Darwin Options. (line 196) 64548* iplugindir=: Directory Options. (line 113) 64549* iprefix: Directory Options. (line 80) 64550* iquote: Directory Options. (line 13) 64551* isysroot: Directory Options. (line 92) 64552* isystem: Directory Options. (line 13) 64553* iwithprefix: Directory Options. (line 86) 64554* iwithprefixbefore: Directory Options. (line 86) 64555* keep_private_externs: Darwin Options. (line 196) 64556* l: Link Options. (line 87) 64557* L: Directory Options. (line 118) 64558* lobjc: Link Options. (line 113) 64559* M: Preprocessor Options. 64560 (line 77) 64561* m: RS/6000 and PowerPC Options. 64562 (line 521) 64563* m1: SH Options. (line 9) 64564* m10: PDP-11 Options. (line 29) 64565* m128bit-long-double: x86 Options. (line 655) 64566* m16: x86 Options. (line 1542) 64567* m16-bit: CRIS Options. (line 63) 64568* m16-bit <1>: NDS32 Options. (line 51) 64569* m1reg-: Adapteva Epiphany Options. 64570 (line 131) 64571* m2: SH Options. (line 12) 64572* m210: MCore Options. (line 43) 64573* m2a: SH Options. (line 30) 64574* m2a-nofpu: SH Options. (line 18) 64575* m2a-single: SH Options. (line 26) 64576* m2a-single-only: SH Options. (line 22) 64577* m3: SH Options. (line 34) 64578* m31: S/390 and zSeries Options. 64579 (line 86) 64580* m32: RS/6000 and PowerPC Options. 64581 (line 245) 64582* m32 <1>: SPARC Options. (line 315) 64583* m32 <2>: TILE-Gx Options. (line 23) 64584* m32 <3>: TILEPro Options. (line 13) 64585* m32 <4>: x86 Options. (line 1542) 64586* m32-bit: CRIS Options. (line 63) 64587* m32bit-doubles: RL78 Options. (line 73) 64588* m32bit-doubles <1>: RX Options. (line 10) 64589* m32r: M32R/D Options. (line 15) 64590* m32r2: M32R/D Options. (line 9) 64591* m32rx: M32R/D Options. (line 12) 64592* m340: MCore Options. (line 43) 64593* m3dnow: x86 Options. (line 878) 64594* m3dnowa: x86 Options. (line 879) 64595* m3e: SH Options. (line 37) 64596* m4: SH Options. (line 51) 64597* m4-100: SH Options. (line 54) 64598* m4-100-nofpu: SH Options. (line 57) 64599* m4-100-single: SH Options. (line 61) 64600* m4-100-single-only: SH Options. (line 65) 64601* m4-200: SH Options. (line 69) 64602* m4-200-nofpu: SH Options. (line 72) 64603* m4-200-single: SH Options. (line 76) 64604* m4-200-single-only: SH Options. (line 80) 64605* m4-300: SH Options. (line 84) 64606* m4-300-nofpu: SH Options. (line 87) 64607* m4-300-single: SH Options. (line 91) 64608* m4-300-single-only: SH Options. (line 95) 64609* m4-340: SH Options. (line 99) 64610* m4-500: SH Options. (line 102) 64611* m4-nofpu: SH Options. (line 40) 64612* m4-single: SH Options. (line 47) 64613* m4-single-only: SH Options. (line 43) 64614* m40: PDP-11 Options. (line 23) 64615* m45: PDP-11 Options. (line 26) 64616* m4a: SH Options. (line 118) 64617* m4a-nofpu: SH Options. (line 106) 64618* m4a-single: SH Options. (line 114) 64619* m4a-single-only: SH Options. (line 110) 64620* m4al: SH Options. (line 121) 64621* m4byte-functions: MCore Options. (line 27) 64622* m5200: M680x0 Options. (line 145) 64623* m5206e: M680x0 Options. (line 154) 64624* m528x: M680x0 Options. (line 158) 64625* m5307: M680x0 Options. (line 162) 64626* m5407: M680x0 Options. (line 166) 64627* m64: Nvidia PTX Options. (line 9) 64628* m64 <1>: RS/6000 and PowerPC Options. 64629 (line 245) 64630* m64 <2>: S/390 and zSeries Options. 64631 (line 86) 64632* m64 <3>: SPARC Options. (line 315) 64633* m64 <4>: TILE-Gx Options. (line 23) 64634* m64 <5>: x86 Options. (line 1542) 64635* m64bit-doubles: RL78 Options. (line 73) 64636* m64bit-doubles <1>: RX Options. (line 10) 64637* m68000: M680x0 Options. (line 94) 64638* m68010: M680x0 Options. (line 102) 64639* m68020: M680x0 Options. (line 108) 64640* m68020-40: M680x0 Options. (line 176) 64641* m68020-60: M680x0 Options. (line 185) 64642* m68030: M680x0 Options. (line 113) 64643* m68040: M680x0 Options. (line 118) 64644* m68060: M680x0 Options. (line 127) 64645* m68881: M680x0 Options. (line 195) 64646* m8-bit: CRIS Options. (line 63) 64647* m8bit-idiv: x86 Options. (line 1444) 64648* m8byte-align: V850 Options. (line 170) 64649* m96bit-long-double: x86 Options. (line 655) 64650* mA6: ARC Options. (line 23) 64651* mA7: ARC Options. (line 30) 64652* mabi: AArch64 Options. (line 9) 64653* mabi <1>: ARM Options. (line 9) 64654* mabi <2>: LoongArch Options. (line 36) 64655* mabi <3>: PRU Options. (line 28) 64656* mabi <4>: RISC-V Options. (line 17) 64657* mabi <5>: RS/6000 and PowerPC Options. 64658 (line 552) 64659* mabi <6>: x86 Options. (line 1137) 64660* mabi <7>: Xtensa Options. (line 103) 64661* mabi=32: MIPS Options. (line 156) 64662* mabi=64: MIPS Options. (line 156) 64663* mabi=call0: Xtensa Options. (line 108) 64664* mabi=eabi: MIPS Options. (line 156) 64665* mabi=elfv1: RS/6000 and PowerPC Options. 64666 (line 574) 64667* mabi=elfv2: RS/6000 and PowerPC Options. 64668 (line 580) 64669* mabi=gnu: MMIX Options. (line 20) 64670* mabi=ibmlongdouble: RS/6000 and PowerPC Options. 64671 (line 558) 64672* mabi=ieeelongdouble: RS/6000 and PowerPC Options. 64673 (line 566) 64674* mabi=mmixware: MMIX Options. (line 20) 64675* mabi=n32: MIPS Options. (line 156) 64676* mabi=o64: MIPS Options. (line 156) 64677* mabi=windowed: Xtensa Options. (line 115) 64678* mabicalls: MIPS Options. (line 192) 64679* mabm: x86 Options. (line 881) 64680* mabort-on-noreturn: ARM Options. (line 752) 64681* mabs=2008: MIPS Options. (line 300) 64682* mabs=legacy: MIPS Options. (line 300) 64683* mabsdata: AVR Options. (line 163) 64684* mabsdiff: MeP Options. (line 7) 64685* mac0: PDP-11 Options. (line 16) 64686* macc-4: FRV Options. (line 139) 64687* macc-8: FRV Options. (line 143) 64688* maccumulate-args: AVR Options. (line 170) 64689* maccumulate-outgoing-args: SH Options. (line 314) 64690* maccumulate-outgoing-args <1>: x86 Options. (line 1181) 64691* maddress-mode=long: x86 Options. (line 1592) 64692* maddress-mode=short: x86 Options. (line 1597) 64693* mads: RS/6000 and PowerPC Options. 64694 (line 614) 64695* madx: x86 Options. (line 882) 64696* maes: x86 Options. (line 859) 64697* maix-struct-return: RS/6000 and PowerPC Options. 64698 (line 545) 64699* maix32: RS/6000 and PowerPC Options. 64700 (line 283) 64701* maix64: RS/6000 and PowerPC Options. 64702 (line 283) 64703* malign-300: H8/300 Options. (line 41) 64704* malign-call: ARC Options. (line 435) 64705* malign-data: RISC-V Options. (line 168) 64706* malign-data <1>: x86 Options. (line 695) 64707* malign-double: x86 Options. (line 640) 64708* malign-int: M680x0 Options. (line 262) 64709* malign-labels: FRV Options. (line 128) 64710* malign-loops: M32R/D Options. (line 73) 64711* malign-natural: RS/6000 and PowerPC Options. 64712 (line 321) 64713* malign-power: RS/6000 and PowerPC Options. 64714 (line 321) 64715* malign-stringops: x86 Options. (line 1317) 64716* mall-opts: MeP Options. (line 11) 64717* malloc-cc: FRV Options. (line 31) 64718* mallow-string-insns: RX Options. (line 150) 64719* mallregs: RL78 Options. (line 66) 64720* maltivec: RS/6000 and PowerPC Options. 64721 (line 137) 64722* malu32: eBPF Options. (line 35) 64723* mam33: MN10300 Options. (line 17) 64724* mam33-2: MN10300 Options. (line 24) 64725* mam34: MN10300 Options. (line 27) 64726* mamx-bf16: x86 Options. (line 920) 64727* mamx-int8: x86 Options. (line 919) 64728* mamx-tile: x86 Options. (line 918) 64729* manchor: C-SKY Options. (line 126) 64730* mandroid: GNU/Linux Options. (line 26) 64731* mannotate-align: ARC Options. (line 382) 64732* mapcs: ARM Options. (line 21) 64733* mapcs-frame: ARM Options. (line 13) 64734* mapp-regs: SPARC Options. (line 10) 64735* mapp-regs <1>: V850 Options. (line 181) 64736* mARC600: ARC Options. (line 23) 64737* mARC601: ARC Options. (line 27) 64738* mARC700: ARC Options. (line 30) 64739* march: AArch64 Options. (line 150) 64740* march <1>: AMD GCN Options. (line 9) 64741* march <2>: ARM Options. (line 80) 64742* march <3>: C6X Options. (line 7) 64743* march <4>: CRIS Options. (line 10) 64744* march <5>: HPPA Options. (line 9) 64745* march <6>: HPPA Options. (line 162) 64746* march <7>: M680x0 Options. (line 12) 64747* march <8>: MIPS Options. (line 14) 64748* march <9>: NDS32 Options. (line 64) 64749* march <10>: Nios II Options. (line 94) 64750* march <11>: Nvidia PTX Options. (line 13) 64751* march <12>: Nvidia PTX Options. (line 25) 64752* march <13>: RISC-V Options. (line 67) 64753* march <14>: S/390 and zSeries Options. 64754 (line 148) 64755* march <15>: x86 Options. (line 9) 64756* march=: C-SKY Options. (line 9) 64757* marclinux: ARC Options. (line 388) 64758* marclinux_prof: ARC Options. (line 395) 64759* margonaut: ARC Options. (line 586) 64760* marm: ARM Options. (line 824) 64761* mas100-syntax: RX Options. (line 76) 64762* masm-hex: MSP430 Options. (line 9) 64763* masm-syntax-unified: ARM Options. (line 932) 64764* masm=DIALECT: x86 Options. (line 589) 64765* matomic: ARC Options. (line 155) 64766* matomic-model=MODEL: SH Options. (line 193) 64767* mauto-litpools: Xtensa Options. (line 60) 64768* mauto-modify-reg: ARC Options. (line 438) 64769* mauto-pic: IA-64 Options. (line 50) 64770* maverage: MeP Options. (line 16) 64771* mavoid-indexed-addresses: RS/6000 and PowerPC Options. 64772 (line 360) 64773* mavx: x86 Options. (line 847) 64774* mavx2: x86 Options. (line 848) 64775* mavx256-split-unaligned-load: x86 Options. (line 1452) 64776* mavx256-split-unaligned-store: x86 Options. (line 1452) 64777* mavx5124fmaps: x86 Options. (line 912) 64778* mavx5124vnniw: x86 Options. (line 915) 64779* mavx512bf16: x86 Options. (line 898) 64780* mavx512bitalg: x86 Options. (line 904) 64781* mavx512bw: x86 Options. (line 854) 64782* mavx512cd: x86 Options. (line 852) 64783* mavx512dq: x86 Options. (line 855) 64784* mavx512er: x86 Options. (line 851) 64785* mavx512f: x86 Options. (line 849) 64786* mavx512fp16: x86 Options. (line 899) 64787* mavx512ifma: x86 Options. (line 856) 64788* mavx512pf: x86 Options. (line 850) 64789* mavx512vbmi: x86 Options. (line 857) 64790* mavx512vbmi2: x86 Options. (line 897) 64791* mavx512vl: x86 Options. (line 853) 64792* mavx512vnni: x86 Options. (line 913) 64793* mavx512vp2intersect: x86 Options. (line 911) 64794* mavx512vpopcntdq: x86 Options. (line 910) 64795* mavxvnni: x86 Options. (line 914) 64796* max-vect-align: Adapteva Epiphany Options. 64797 (line 119) 64798* mb: SH Options. (line 126) 64799* mbackchain: S/390 and zSeries Options. 64800 (line 35) 64801* mbarrel-shift-enabled: LM32 Options. (line 9) 64802* mbarrel-shifter: ARC Options. (line 10) 64803* mbarrel_shifter: ARC Options. (line 606) 64804* mbase-addresses: MMIX Options. (line 53) 64805* mbased=: MeP Options. (line 20) 64806* mbbit-peephole: ARC Options. (line 441) 64807* mbe8: ARM Options. (line 72) 64808* mbig: RS/6000 and PowerPC Options. 64809 (line 440) 64810* mbig-endian: AArch64 Options. (line 20) 64811* mbig-endian <1>: ARC Options. (line 589) 64812* mbig-endian <2>: ARM Options. (line 67) 64813* mbig-endian <3>: C6X Options. (line 13) 64814* mbig-endian <4>: C-SKY Options. (line 28) 64815* mbig-endian <5>: eBPF Options. (line 22) 64816* mbig-endian <6>: IA-64 Options. (line 9) 64817* mbig-endian <7>: MCore Options. (line 39) 64818* mbig-endian <8>: MicroBlaze Options. (line 56) 64819* mbig-endian <9>: NDS32 Options. (line 9) 64820* mbig-endian <10>: RISC-V Options. (line 174) 64821* mbig-endian <11>: RS/6000 and PowerPC Options. 64822 (line 440) 64823* mbig-endian <12>: TILE-Gx Options. (line 29) 64824* mbig-endian-data: RX Options. (line 42) 64825* mbig-switch: V850 Options. (line 176) 64826* mbigtable: SH Options. (line 141) 64827* mbionic: GNU/Linux Options. (line 22) 64828* mbit-align: RS/6000 and PowerPC Options. 64829 (line 392) 64830* mbit-ops: CR16 Options. (line 25) 64831* mbitfield: M680x0 Options. (line 232) 64832* mbitops: MeP Options. (line 26) 64833* mbitops <1>: SH Options. (line 145) 64834* mblock-compare-inline-limit: RS/6000 and PowerPC Options. 64835 (line 694) 64836* mblock-compare-inline-loop-limit: RS/6000 and PowerPC Options. 64837 (line 700) 64838* mblock-move-inline-limit: RS/6000 and PowerPC Options. 64839 (line 688) 64840* mbmi: x86 Options. (line 883) 64841* mbmi2: x86 Options. (line 884) 64842* mboard: OpenRISC Options. (line 9) 64843* mbranch-cost: Adapteva Epiphany Options. 64844 (line 18) 64845* mbranch-cost <1>: AVR Options. (line 185) 64846* mbranch-cost <2>: MIPS Options. (line 791) 64847* mbranch-cost <3>: RISC-V Options. (line 9) 64848* mbranch-cost=: C-SKY Options. (line 159) 64849* mbranch-cost=NUM: SH Options. (line 334) 64850* mbranch-cost=NUMBER: M32R/D Options. (line 82) 64851* mbranch-index: ARC Options. (line 329) 64852* mbranch-likely: MIPS Options. (line 798) 64853* mbranch-predict: MMIX Options. (line 48) 64854* mbranch-protection: AArch64 Options. (line 288) 64855* mbss-plt: RS/6000 and PowerPC Options. 64856 (line 161) 64857* mbuild-constants: DEC Alpha Options. (line 141) 64858* mbwx: DEC Alpha Options. (line 163) 64859* mbypass-cache: Nios II Options. (line 103) 64860* mc68000: M680x0 Options. (line 94) 64861* mc68020: M680x0 Options. (line 108) 64862* mc=: MeP Options. (line 31) 64863* mcache: C-SKY Options. (line 93) 64864* mcache-block-size: NDS32 Options. (line 60) 64865* mcache-volatile: Nios II Options. (line 109) 64866* mcall-eabi: RS/6000 and PowerPC Options. 64867 (line 515) 64868* mcall-freebsd: RS/6000 and PowerPC Options. 64869 (line 529) 64870* mcall-linux: RS/6000 and PowerPC Options. 64871 (line 525) 64872* mcall-ms2sysv-xlogues: x86 Options. (line 1157) 64873* mcall-netbsd: RS/6000 and PowerPC Options. 64874 (line 533) 64875* mcall-netbsd <1>: RS/6000 and PowerPC Options. 64876 (line 537) 64877* mcall-prologues: AVR Options. (line 190) 64878* mcall-sysv: RS/6000 and PowerPC Options. 64879 (line 507) 64880* mcall-sysv-eabi: RS/6000 and PowerPC Options. 64881 (line 515) 64882* mcall-sysv-noeabi: RS/6000 and PowerPC Options. 64883 (line 518) 64884* mcallee-super-interworking: ARM Options. (line 853) 64885* mcaller-copies: HPPA Options. (line 23) 64886* mcaller-super-interworking: ARM Options. (line 860) 64887* mcallgraph-data: MCore Options. (line 31) 64888* mcase-vector-pcrel: ARC Options. (line 450) 64889* mcbcond: SPARC Options. (line 260) 64890* mcbranch-force-delay-slot: SH Options. (line 349) 64891* mcc-init: CRIS Options. (line 41) 64892* mccrt: C-SKY Options. (line 155) 64893* mcfv4e: M680x0 Options. (line 170) 64894* mcheck-zero-division: MIPS Options. (line 570) 64895* mcix: DEC Alpha Options. (line 163) 64896* mcld: x86 Options. (line 975) 64897* mcldemote: x86 Options. (line 916) 64898* mclear-hwcap: Solaris 2 Options. (line 9) 64899* mclflushopt: x86 Options. (line 861) 64900* mclip: MeP Options. (line 35) 64901* mclwb: x86 Options. (line 862) 64902* mclzero: x86 Options. (line 895) 64903* mcmodel: NDS32 Options. (line 67) 64904* mcmodel <1>: SPARC Options. (line 320) 64905* mcmodel=kernel: x86 Options. (line 1576) 64906* mcmodel=large: AArch64 Options. (line 45) 64907* mcmodel=large <1>: OpenRISC Options. (line 77) 64908* mcmodel=large <2>: RS/6000 and PowerPC Options. 64909 (line 131) 64910* mcmodel=large <3>: TILE-Gx Options. (line 14) 64911* mcmodel=large <4>: x86 Options. (line 1588) 64912* mcmodel=medany: RISC-V Options. (line 142) 64913* mcmodel=medium: RS/6000 and PowerPC Options. 64914 (line 126) 64915* mcmodel=medium <1>: x86 Options. (line 1581) 64916* mcmodel=medlow: RISC-V Options. (line 135) 64917* mcmodel=small: AArch64 Options. (line 39) 64918* mcmodel=small <1>: OpenRISC Options. (line 73) 64919* mcmodel=small <2>: RS/6000 and PowerPC Options. 64920 (line 122) 64921* mcmodel=small <3>: TILE-Gx Options. (line 9) 64922* mcmodel=small <4>: x86 Options. (line 1570) 64923* mcmodel=tiny: AArch64 Options. (line 34) 64924* mcmov: NDS32 Options. (line 21) 64925* mcmov <1>: OpenRISC Options. (line 45) 64926* mcmove: Adapteva Epiphany Options. 64927 (line 23) 64928* mcmpb: RS/6000 and PowerPC Options. 64929 (line 25) 64930* mcmse: ARM Options. (line 961) 64931* mco-re: eBPF Options. (line 56) 64932* mcode-density: ARC Options. (line 163) 64933* mcode-density-frame: ARC Options. (line 511) 64934* mcode-readable: MIPS Options. (line 530) 64935* mcode-region: MSP430 Options. (line 150) 64936* mcompact-branches=always: MIPS Options. (line 810) 64937* mcompact-branches=never: MIPS Options. (line 810) 64938* mcompact-branches=optimal: MIPS Options. (line 810) 64939* mcompact-casesi: ARC Options. (line 454) 64940* mcompat-align-parm: RS/6000 and PowerPC Options. 64941 (line 899) 64942* mcompress: FT32 Options. (line 26) 64943* mcond-exec: FRV Options. (line 187) 64944* mcond-move: FRV Options. (line 159) 64945* mconfig=: MeP Options. (line 39) 64946* mconsole: x86 Windows Options. 64947 (line 9) 64948* mconst-align: CRIS Options. (line 54) 64949* mconst16: Xtensa Options. (line 10) 64950* mconstant-gp: IA-64 Options. (line 46) 64951* mconstpool: C-SKY Options. (line 143) 64952* mcop: MeP Options. (line 48) 64953* mcop32: MeP Options. (line 53) 64954* mcop64: MeP Options. (line 56) 64955* mcorea: Blackfin Options. (line 154) 64956* mcoreb: Blackfin Options. (line 161) 64957* mcp: C-SKY Options. (line 90) 64958* mcpu: AArch64 Options. (line 235) 64959* mcpu <1>: ARC Options. (line 18) 64960* mcpu <2>: ARM Options. (line 622) 64961* mcpu <3>: CRIS Options. (line 10) 64962* mcpu <4>: DEC Alpha Options. (line 215) 64963* mcpu <5>: eBPF Options. (line 38) 64964* mcpu <6>: FRV Options. (line 258) 64965* mcpu <7>: M680x0 Options. (line 28) 64966* mcpu <8>: picoChip Options. (line 9) 64967* mcpu <9>: RISC-V Options. (line 78) 64968* mcpu <10>: RL78 Options. (line 32) 64969* mcpu <11>: RS/6000 and PowerPC Options. 64970 (line 62) 64971* mcpu <12>: RX Options. (line 30) 64972* mcpu <13>: SPARC Options. (line 115) 64973* mcpu <14>: TILE-Gx Options. (line 18) 64974* mcpu <15>: TILEPro Options. (line 9) 64975* mcpu <16>: Visium Options. (line 33) 64976* mcpu <17>: x86 Options. (line 535) 64977* mcpu32: M680x0 Options. (line 136) 64978* mcpu=: Blackfin Options. (line 7) 64979* mcpu= <1>: C-SKY Options. (line 14) 64980* mcpu= <2>: M32C Options. (line 7) 64981* mcpu= <3>: MicroBlaze Options. (line 20) 64982* mcpu= <4>: MSP430 Options. (line 72) 64983* mcr16c: CR16 Options. (line 14) 64984* mcr16cplus: CR16 Options. (line 14) 64985* mcrc: MIPS Options. (line 416) 64986* mcrc32: x86 Options. (line 1053) 64987* mcrypto: RS/6000 and PowerPC Options. 64988 (line 178) 64989* mcsync-anomaly: Blackfin Options. (line 57) 64990* mcsync-anomaly <1>: Blackfin Options. (line 63) 64991* mctor-dtor: NDS32 Options. (line 81) 64992* mcustom-fpu-cfg: Nios II Options. (line 259) 64993* mcustom-INSN: Nios II Options. (line 139) 64994* mcx16: x86 Options. (line 1026) 64995* MD: Preprocessor Options. 64996 (line 172) 64997* mdalign: SH Options. (line 132) 64998* mdata-align: CRIS Options. (line 54) 64999* mdata-model: CR16 Options. (line 28) 65000* mdata-region: MSP430 Options. (line 150) 65001* mdc: MeP Options. (line 62) 65002* mdebug: M32R/D Options. (line 69) 65003* mdebug <1>: S/390 and zSeries Options. 65004 (line 144) 65005* mdebug <2>: Visium Options. (line 7) 65006* mdebug-main=PREFIX: VMS Options. (line 13) 65007* mdec-asm: PDP-11 Options. (line 46) 65008* mdirect-extern-access: x86 Options. (line 1608) 65009* mdisable-callt: V850 Options. (line 92) 65010* mdisable-fpregs: HPPA Options. (line 34) 65011* mdisable-indexing: HPPA Options. (line 40) 65012* mdiv: C-SKY Options. (line 109) 65013* mdiv <1>: M680x0 Options. (line 207) 65014* mdiv <2>: MCore Options. (line 15) 65015* mdiv <3>: MeP Options. (line 65) 65016* mdiv <4>: RISC-V Options. (line 49) 65017* mdiv-rem: ARC Options. (line 160) 65018* mdiv=STRATEGY: SH Options. (line 284) 65019* mdivide-breaks: MIPS Options. (line 576) 65020* mdivide-enabled: LM32 Options. (line 12) 65021* mdivide-traps: MIPS Options. (line 576) 65022* mdivsi3_libfunc=NAME: SH Options. (line 320) 65023* mdll: x86 Windows Options. 65024 (line 16) 65025* mdlmzb: RS/6000 and PowerPC Options. 65026 (line 385) 65027* mdmx: MIPS Options. (line 376) 65028* mdouble: AVR Options. (line 195) 65029* mdouble <1>: FRV Options. (line 48) 65030* mdouble-float: C-SKY Options. (line 58) 65031* mdouble-float <1>: MIPS Options. (line 288) 65032* mdouble-float <2>: OpenRISC Options. (line 33) 65033* mdpfp: ARC Options. (line 99) 65034* mdpfp-compact: ARC Options. (line 100) 65035* mdpfp-fast: ARC Options. (line 104) 65036* mdpfp_compact: ARC Options. (line 609) 65037* mdpfp_fast: ARC Options. (line 612) 65038* mdsp: C-SKY Options. (line 102) 65039* mdsp <1>: MIPS Options. (line 353) 65040* mdsp-packa: ARC Options. (line 335) 65041* mdspr2: MIPS Options. (line 359) 65042* mdsp_packa: ARC Options. (line 615) 65043* mdump-tune-features: x86 Options. (line 957) 65044* mdvbf: ARC Options. (line 340) 65045* mdwarf2-asm: IA-64 Options. (line 94) 65046* mdword: FRV Options. (line 40) 65047* mdword <1>: FRV Options. (line 44) 65048* mdynamic-no-pic: RS/6000 and PowerPC Options. 65049 (line 445) 65050* mea: ARC Options. (line 112) 65051* mEA: ARC Options. (line 618) 65052* meabi: RS/6000 and PowerPC Options. 65053 (line 633) 65054* mearly-cbranchsi: ARC Options. (line 476) 65055* mearly-stop-bits: IA-64 Options. (line 100) 65056* meb: MeP Options. (line 68) 65057* meb <1>: Moxie Options. (line 7) 65058* meb <2>: Nios II Options. (line 90) 65059* meb <3>: Score Options. (line 9) 65060* medsp: C-SKY Options. (line 103) 65061* mel: MeP Options. (line 71) 65062* mel <1>: Moxie Options. (line 11) 65063* mel <2>: Nios II Options. (line 90) 65064* mel <3>: Score Options. (line 12) 65065* melf: CRIS Options. (line 79) 65066* melf <1>: MMIX Options. (line 43) 65067* melrw: C-SKY Options. (line 76) 65068* memb: RS/6000 and PowerPC Options. 65069 (line 628) 65070* membedded-data: MIPS Options. (line 517) 65071* memregs=: M32C Options. (line 21) 65072* menqcmd: x86 Options. (line 907) 65073* mep: V850 Options. (line 16) 65074* mepsilon: MMIX Options. (line 15) 65075* mesa: S/390 and zSeries Options. 65076 (line 94) 65077* metrax100: CRIS Options. (line 26) 65078* metrax4: CRIS Options. (line 26) 65079* meva: MIPS Options. (line 403) 65080* mexpand-adddi: ARC Options. (line 479) 65081* mexplicit-relocs: DEC Alpha Options. (line 176) 65082* mexplicit-relocs <1>: MIPS Options. (line 561) 65083* mexr: H8/300 Options. (line 28) 65084* mexr <1>: H8/300 Options. (line 33) 65085* mext-perf: NDS32 Options. (line 27) 65086* mext-perf2: NDS32 Options. (line 33) 65087* mext-string: NDS32 Options. (line 39) 65088* mextern-sdata: MIPS Options. (line 480) 65089* MF: Preprocessor Options. 65090 (line 111) 65091* mf16c: x86 Options. (line 866) 65092* mfancy-math-387: x86 Options. (line 630) 65093* mfast-fp: Blackfin Options. (line 130) 65094* mfast-indirect-calls: HPPA Options. (line 52) 65095* mfast-sw-div: Nios II Options. (line 115) 65096* mfaster-structs: SPARC Options. (line 91) 65097* mfdiv: RISC-V Options. (line 42) 65098* mfdivdu: C-SKY Options. (line 64) 65099* mfdpic: ARM Options. (line 984) 65100* mfdpic <1>: FRV Options. (line 72) 65101* mfentry: x86 Options. (line 1391) 65102* mfentry-name: x86 Options. (line 1422) 65103* mfentry-section: x86 Options. (line 1426) 65104* mfix: DEC Alpha Options. (line 163) 65105* mfix-24k: MIPS Options. (line 647) 65106* mfix-and-continue: Darwin Options. (line 104) 65107* mfix-at697f: SPARC Options. (line 294) 65108* mfix-cmse-cve-2021-35465: ARM Options. (line 967) 65109* mfix-cortex-a53-835769: AArch64 Options. (line 90) 65110* mfix-cortex-a53-843419: AArch64 Options. (line 97) 65111* mfix-cortex-m3-ldrd: ARM Options. (line 894) 65112* mfix-gr712rc: SPARC Options. (line 307) 65113* mfix-r10000: MIPS Options. (line 669) 65114* mfix-r4000: MIPS Options. (line 653) 65115* mfix-r4400: MIPS Options. (line 663) 65116* mfix-r5900: MIPS Options. (line 680) 65117* mfix-rm7000: MIPS Options. (line 690) 65118* mfix-sb1: MIPS Options. (line 715) 65119* mfix-ut699: SPARC Options. (line 299) 65120* mfix-ut700: SPARC Options. (line 303) 65121* mfix-vr4120: MIPS Options. (line 695) 65122* mfix-vr4130: MIPS Options. (line 708) 65123* mfixed-cc: FRV Options. (line 35) 65124* mfixed-range: HPPA Options. (line 59) 65125* mfixed-range <1>: IA-64 Options. (line 105) 65126* mfixed-range <2>: SH Options. (line 327) 65127* mflat: SPARC Options. (line 22) 65128* mflip-mips16: MIPS Options. (line 128) 65129* mflip-thumb: ARM Options. (line 836) 65130* mfloat-abi: ARM Options. (line 41) 65131* mfloat-abi <1>: C-SKY Options. (line 35) 65132* mfloat-ieee: DEC Alpha Options. (line 171) 65133* mfloat-vax: DEC Alpha Options. (line 171) 65134* mfloat128: RS/6000 and PowerPC Options. 65135 (line 215) 65136* mfloat128-hardware: RS/6000 and PowerPC Options. 65137 (line 236) 65138* mflush-func: MIPS Options. (line 782) 65139* mflush-func=NAME: M32R/D Options. (line 93) 65140* mflush-trap=NUMBER: M32R/D Options. (line 86) 65141* mfma: x86 Options. (line 867) 65142* mfma4: x86 Options. (line 870) 65143* mfmaf: SPARC Options. (line 267) 65144* mfmovd: SH Options. (line 148) 65145* mforce-indirect-call: x86 Options. (line 1146) 65146* mforce-no-pic: Xtensa Options. (line 41) 65147* mfp-exceptions: MIPS Options. (line 830) 65148* mfp-mode: Adapteva Epiphany Options. 65149 (line 71) 65150* mfp-reg: DEC Alpha Options. (line 25) 65151* mfp-ret-in-387: x86 Options. (line 620) 65152* mfp-rounding-mode: DEC Alpha Options. (line 85) 65153* mfp-trap-mode: DEC Alpha Options. (line 63) 65154* mfp16-format: ARM Options. (line 730) 65155* mfp32: MIPS Options. (line 258) 65156* mfp64: MIPS Options. (line 261) 65157* mfpmath: Optimize Options. (line 2316) 65158* mfpmath <1>: x86 Options. (line 538) 65159* mfpr-32: FRV Options. (line 15) 65160* mfpr-64: FRV Options. (line 19) 65161* mfprnd: RS/6000 and PowerPC Options. 65162 (line 25) 65163* mfpu: ARC Options. (line 231) 65164* mfpu <1>: ARM Options. (line 702) 65165* mfpu <2>: LoongArch Options. (line 54) 65166* mfpu <3>: PDP-11 Options. (line 9) 65167* mfpu <4>: SPARC Options. (line 34) 65168* mfpu <5>: Visium Options. (line 19) 65169* mfpu=: C-SKY Options. (line 69) 65170* mfpxx: MIPS Options. (line 264) 65171* mfract-convert-truncate: AVR Options. (line 286) 65172* mframe-header-opt: MIPS Options. (line 891) 65173* mfriz: RS/6000 and PowerPC Options. 65174 (line 870) 65175* mfsca: SH Options. (line 365) 65176* mfsgsbase: x86 Options. (line 863) 65177* mfsmuld: SPARC Options. (line 274) 65178* mfsrra: SH Options. (line 374) 65179* mft32b: FT32 Options. (line 23) 65180* mfull-regs: NDS32 Options. (line 18) 65181* mfull-toc: RS/6000 and PowerPC Options. 65182 (line 256) 65183* mfunction-return: x86 Options. (line 1500) 65184* mfused-madd: IA-64 Options. (line 88) 65185* mfused-madd <1>: MIPS Options. (line 630) 65186* mfused-madd <2>: RS/6000 and PowerPC Options. 65187 (line 369) 65188* mfused-madd <3>: S/390 and zSeries Options. 65189 (line 183) 65190* mfused-madd <4>: SH Options. (line 356) 65191* mfused-madd <5>: Xtensa Options. (line 19) 65192* mfxsr: x86 Options. (line 886) 65193* MG: Preprocessor Options. 65194 (line 122) 65195* mg: VAX Options. (line 17) 65196* mg10: RL78 Options. (line 62) 65197* mg13: RL78 Options. (line 62) 65198* mg14: RL78 Options. (line 62) 65199* mgas: HPPA Options. (line 75) 65200* mgas-isr-prologues: AVR Options. (line 203) 65201* mgcc-abi: V850 Options. (line 148) 65202* mgeneral-regs-only: AArch64 Options. (line 24) 65203* mgeneral-regs-only <1>: ARM Options. (line 57) 65204* mgeneral-regs-only <2>: x86 Options. (line 1470) 65205* mgfni: x86 Options. (line 900) 65206* mghs: V850 Options. (line 127) 65207* mginv: MIPS Options. (line 421) 65208* mglibc: GNU/Linux Options. (line 9) 65209* mgnu: VAX Options. (line 13) 65210* mgnu-as: IA-64 Options. (line 18) 65211* mgnu-asm: PDP-11 Options. (line 49) 65212* mgnu-attribute: RS/6000 and PowerPC Options. 65213 (line 587) 65214* mgnu-ld: HPPA Options. (line 111) 65215* mgnu-ld <1>: IA-64 Options. (line 23) 65216* mgomp: Nvidia PTX Options. (line 76) 65217* mgp32: MIPS Options. (line 252) 65218* mgp64: MIPS Options. (line 255) 65219* mgpopt: MIPS Options. (line 502) 65220* mgpopt <1>: Nios II Options. (line 16) 65221* mgpr-32: FRV Options. (line 7) 65222* mgpr-64: FRV Options. (line 11) 65223* mgprel-ro: FRV Options. (line 99) 65224* mgprel-sec: Nios II Options. (line 65) 65225* mh: H8/300 Options. (line 14) 65226* mhal: Nios II Options. (line 332) 65227* mhalf-reg-file: Adapteva Epiphany Options. 65228 (line 9) 65229* mhard-dfp: RS/6000 and PowerPC Options. 65230 (line 25) 65231* mhard-dfp <1>: S/390 and zSeries Options. 65232 (line 20) 65233* mhard-div: OpenRISC Options. (line 19) 65234* mhard-float: C-SKY Options. (line 51) 65235* mhard-float <1>: FRV Options. (line 23) 65236* mhard-float <2>: M680x0 Options. (line 195) 65237* mhard-float <3>: MicroBlaze Options. (line 10) 65238* mhard-float <4>: MIPS Options. (line 267) 65239* mhard-float <5>: OpenRISC Options. (line 29) 65240* mhard-float <6>: RS/6000 and PowerPC Options. 65241 (line 333) 65242* mhard-float <7>: S/390 and zSeries Options. 65243 (line 11) 65244* mhard-float <8>: SPARC Options. (line 34) 65245* mhard-float <9>: V850 Options. (line 113) 65246* mhard-float <10>: Visium Options. (line 19) 65247* mhard-float <11>: x86 Options. (line 603) 65248* mhard-mul: OpenRISC Options. (line 24) 65249* mhard-quad-float: SPARC Options. (line 55) 65250* mharden-sls: AArch64 Options. (line 301) 65251* mharden-sls <1>: x86 Options. (line 1522) 65252* mhardlit: MCore Options. (line 10) 65253* mhigh-registers: C-SKY Options. (line 120) 65254* mhle: x86 Options. (line 892) 65255* mhotpatch: S/390 and zSeries Options. 65256 (line 218) 65257* mhp-ld: HPPA Options. (line 123) 65258* mhreset: x86 Options. (line 921) 65259* mhtm: RS/6000 and PowerPC Options. 65260 (line 184) 65261* mhtm <1>: S/390 and zSeries Options. 65262 (line 104) 65263* mhw-div: Nios II Options. (line 124) 65264* mhw-mul: Nios II Options. (line 124) 65265* mhw-mulx: Nios II Options. (line 124) 65266* mhwmult=: MSP430 Options. (line 93) 65267* miamcu: x86 Options. (line 1542) 65268* micplb: Blackfin Options. (line 175) 65269* mid-shared-library: Blackfin Options. (line 78) 65270* mid-shared-library <1>: Blackfin Options. (line 85) 65271* mieee: DEC Alpha Options. (line 39) 65272* mieee <1>: SH Options. (line 165) 65273* mieee-conformant: DEC Alpha Options. (line 134) 65274* mieee-fp: x86 Options. (line 597) 65275* mieee-with-inexact: DEC Alpha Options. (line 52) 65276* milp32: IA-64 Options. (line 121) 65277* mimadd: MIPS Options. (line 623) 65278* mimpure-text: Solaris 2 Options. (line 15) 65279* mincoming-stack-boundary: x86 Options. (line 813) 65280* mindexed-loads: ARC Options. (line 483) 65281* mindirect-branch: x86 Options. (line 1481) 65282* mindirect-branch-cs-prefix: x86 Options. (line 1529) 65283* mindirect-branch-register: x86 Options. (line 1519) 65284* minline-all-stringops: x86 Options. (line 1322) 65285* minline-float-divide-max-throughput: IA-64 Options. (line 58) 65286* minline-float-divide-min-latency: IA-64 Options. (line 54) 65287* minline-ic_invalidate: SH Options. (line 174) 65288* minline-int-divide: IA-64 Options. (line 73) 65289* minline-int-divide-max-throughput: IA-64 Options. (line 69) 65290* minline-int-divide-min-latency: IA-64 Options. (line 65) 65291* minline-plt: Blackfin Options. (line 135) 65292* minline-plt <1>: FRV Options. (line 81) 65293* minline-sqrt-max-throughput: IA-64 Options. (line 80) 65294* minline-sqrt-min-latency: IA-64 Options. (line 76) 65295* minline-stringops-dynamically: x86 Options. (line 1330) 65296* minrt: MSP430 Options. (line 115) 65297* minrt <1>: PRU Options. (line 9) 65298* minsert-sched-nops: RS/6000 and PowerPC Options. 65299 (line 485) 65300* minstrument-return: x86 Options. (line 1410) 65301* mint-register: RX Options. (line 100) 65302* mint16: PDP-11 Options. (line 33) 65303* mint32: CR16 Options. (line 22) 65304* mint32 <1>: H8/300 Options. (line 38) 65305* mint32 <2>: PDP-11 Options. (line 37) 65306* mint8: AVR Options. (line 213) 65307* minterlink-compressed: MIPS Options. (line 135) 65308* minterlink-mips16: MIPS Options. (line 147) 65309* mio-volatile: MeP Options. (line 74) 65310* mips1: MIPS Options. (line 80) 65311* mips16: MIPS Options. (line 120) 65312* mips2: MIPS Options. (line 83) 65313* mips3: MIPS Options. (line 86) 65314* mips32: MIPS Options. (line 92) 65315* mips32r3: MIPS Options. (line 95) 65316* mips32r5: MIPS Options. (line 98) 65317* mips32r6: MIPS Options. (line 101) 65318* mips3d: MIPS Options. (line 382) 65319* mips4: MIPS Options. (line 89) 65320* mips64: MIPS Options. (line 104) 65321* mips64r2: MIPS Options. (line 107) 65322* mips64r3: MIPS Options. (line 110) 65323* mips64r5: MIPS Options. (line 113) 65324* mips64r6: MIPS Options. (line 116) 65325* mirq-ctrl-saved: ARC Options. (line 296) 65326* misa: Nvidia PTX Options. (line 22) 65327* misa-spec: RISC-V Options. (line 54) 65328* misel: RS/6000 and PowerPC Options. 65329 (line 167) 65330* misize: ARC Options. (line 379) 65331* misize <1>: SH Options. (line 186) 65332* misr-vector-size: NDS32 Options. (line 57) 65333* missue-rate=NUMBER: M32R/D Options. (line 79) 65334* mistack: C-SKY Options. (line 81) 65335* mivc2: MeP Options. (line 59) 65336* mjli-always: ARC Options. (line 14) 65337* mjmp32: eBPF Options. (line 32) 65338* mjmpext: eBPF Options. (line 28) 65339* mjsr: RX Options. (line 169) 65340* mjump-in-delay: HPPA Options. (line 30) 65341* mkernel: Darwin Options. (line 82) 65342* mkernel <1>: eBPF Options. (line 13) 65343* mkl: x86 Options. (line 922) 65344* mknuthdiv: MMIX Options. (line 32) 65345* ml: MeP Options. (line 78) 65346* ml <1>: SH Options. (line 129) 65347* mlarge: MSP430 Options. (line 82) 65348* mlarge-data: DEC Alpha Options. (line 187) 65349* mlarge-data-threshold: x86 Options. (line 702) 65350* mlarge-text: DEC Alpha Options. (line 205) 65351* mleadz: MeP Options. (line 81) 65352* mleaf-id-shared-library: Blackfin Options. (line 89) 65353* mleaf-id-shared-library <1>: Blackfin Options. (line 95) 65354* mlibfuncs: MMIX Options. (line 10) 65355* mlibrary-pic: FRV Options. (line 135) 65356* mlinked-fp: FRV Options. (line 116) 65357* mlinker-opt: HPPA Options. (line 85) 65358* mlittle: RS/6000 and PowerPC Options. 65359 (line 434) 65360* mlittle-endian: AArch64 Options. (line 30) 65361* mlittle-endian <1>: ARC Options. (line 598) 65362* mlittle-endian <2>: ARM Options. (line 63) 65363* mlittle-endian <3>: C6X Options. (line 16) 65364* mlittle-endian <4>: C-SKY Options. (line 30) 65365* mlittle-endian <5>: eBPF Options. (line 25) 65366* mlittle-endian <6>: IA-64 Options. (line 13) 65367* mlittle-endian <7>: MCore Options. (line 39) 65368* mlittle-endian <8>: MicroBlaze Options. (line 59) 65369* mlittle-endian <9>: NDS32 Options. (line 12) 65370* mlittle-endian <10>: RISC-V Options. (line 178) 65371* mlittle-endian <11>: RS/6000 and PowerPC Options. 65372 (line 434) 65373* mlittle-endian <12>: TILE-Gx Options. (line 29) 65374* mlittle-endian-data: RX Options. (line 42) 65375* mliw: MN10300 Options. (line 54) 65376* mll64: ARC Options. (line 167) 65377* mllsc: MIPS Options. (line 339) 65378* mload-store-pairs: MIPS Options. (line 590) 65379* mlocal-sdata: MIPS Options. (line 468) 65380* mlock: ARC Options. (line 345) 65381* mlong-calls: Adapteva Epiphany Options. 65382 (line 55) 65383* mlong-calls <1>: ARC Options. (line 404) 65384* mlong-calls <2>: ARM Options. (line 757) 65385* mlong-calls <3>: Blackfin Options. (line 118) 65386* mlong-calls <4>: FRV Options. (line 122) 65387* mlong-calls <5>: HPPA Options. (line 136) 65388* mlong-calls <6>: MIPS Options. (line 609) 65389* mlong-calls <7>: V850 Options. (line 10) 65390* mlong-double: AVR Options. (line 195) 65391* mlong-double-128: S/390 and zSeries Options. 65392 (line 29) 65393* mlong-double-128 <1>: x86 Options. (line 681) 65394* mlong-double-64: S/390 and zSeries Options. 65395 (line 29) 65396* mlong-double-64 <1>: x86 Options. (line 681) 65397* mlong-double-80: x86 Options. (line 681) 65398* mlong-jump-table-offsets: M680x0 Options. (line 340) 65399* mlong-jumps: V850 Options. (line 108) 65400* mlong-load-store: HPPA Options. (line 66) 65401* mlong32: MIPS Options. (line 443) 65402* mlong64: MIPS Options. (line 438) 65403* mlongcall: RS/6000 and PowerPC Options. 65404 (line 728) 65405* mlongcalls: Xtensa Options. (line 87) 65406* mloongson-ext: MIPS Options. (line 430) 65407* mloongson-ext2: MIPS Options. (line 434) 65408* mloongson-mmi: MIPS Options. (line 425) 65409* mloop: PRU Options. (line 25) 65410* mloop <1>: V850 Options. (line 121) 65411* mlow-precision-div: AArch64 Options. (line 120) 65412* mlow-precision-recip-sqrt: AArch64 Options. (line 103) 65413* mlow-precision-sqrt: AArch64 Options. (line 111) 65414* mlow64k: Blackfin Options. (line 67) 65415* mlp64: IA-64 Options. (line 121) 65416* mlpc-width: ARC Options. (line 313) 65417* mlra: ARC Options. (line 488) 65418* mlra <1>: FT32 Options. (line 16) 65419* mlra <2>: PDP-11 Options. (line 52) 65420* mlra <3>: SPARC Options. (line 111) 65421* mlra <4>: VAX Options. (line 22) 65422* mlra-priority-compact: ARC Options. (line 496) 65423* mlra-priority-noncompact: ARC Options. (line 499) 65424* mlra-priority-none: ARC Options. (line 493) 65425* mlwp: x86 Options. (line 877) 65426* mlxc1-sxc1: MIPS Options. (line 901) 65427* mlzcnt: x86 Options. (line 885) 65428* MM: Preprocessor Options. 65429 (line 102) 65430* mm: MeP Options. (line 84) 65431* mmac: CR16 Options. (line 9) 65432* mmac <1>: Score Options. (line 21) 65433* mmac-24: ARC Options. (line 354) 65434* mmac-d16: ARC Options. (line 350) 65435* mmac_24: ARC Options. (line 621) 65436* mmac_d16: ARC Options. (line 624) 65437* mmad: MIPS Options. (line 618) 65438* mmadd4: MIPS Options. (line 906) 65439* mmain-is-OS_task: AVR Options. (line 219) 65440* mmainkernel: Nvidia PTX Options. (line 41) 65441* mmalloc64: VMS Options. (line 17) 65442* mmanual-endbr: x86 Options. (line 1151) 65443* mmax: DEC Alpha Options. (line 163) 65444* mmax-constant-size: RX Options. (line 82) 65445* mmax-inline-shift=: MSP430 Options. (line 134) 65446* mmax-stack-frame: CRIS Options. (line 22) 65447* mmcount-ra-address: MIPS Options. (line 878) 65448* mmcu: AVR Options. (line 9) 65449* mmcu <1>: MIPS Options. (line 399) 65450* mmcu <2>: PRU Options. (line 17) 65451* mmcu=: MSP430 Options. (line 14) 65452* MMD: Preprocessor Options. 65453 (line 188) 65454* mmedia: FRV Options. (line 56) 65455* mmedium-calls: ARC Options. (line 408) 65456* mmemcpy: MicroBlaze Options. (line 13) 65457* mmemcpy <1>: MIPS Options. (line 603) 65458* mmemcpy-strategy=STRATEGY: x86 Options. (line 1352) 65459* mmemory-latency: DEC Alpha Options. (line 268) 65460* mmemory-model: SPARC Options. (line 348) 65461* mmemset-strategy=STRATEGY: x86 Options. (line 1364) 65462* mmfcrf: RS/6000 and PowerPC Options. 65463 (line 25) 65464* mmicromips: MIPS Options. (line 387) 65465* mmillicode: ARC Options. (line 502) 65466* mminimal-toc: RS/6000 and PowerPC Options. 65467 (line 256) 65468* mminmax: MeP Options. (line 87) 65469* mmixed-code: ARC Options. (line 516) 65470* mmma: RS/6000 and PowerPC Options. 65471 (line 944) 65472* mmmx: x86 Options. (line 838) 65473* mmodel=large: M32R/D Options. (line 33) 65474* mmodel=medium: M32R/D Options. (line 27) 65475* mmodel=small: M32R/D Options. (line 18) 65476* mmovbe: x86 Options. (line 1045) 65477* mmovdir64b: x86 Options. (line 906) 65478* mmovdiri: x86 Options. (line 905) 65479* mmove-max: x86 Options. (line 1003) 65480* mmp: C-SKY Options. (line 87) 65481* mmpy: ARC Options. (line 117) 65482* mmpy-option: ARC Options. (line 173) 65483* mms-bitfields: x86 Options. (line 1197) 65484* mmt: MIPS Options. (line 395) 65485* mmul: RL78 Options. (line 15) 65486* mmul-bug-workaround: CRIS Options. (line 31) 65487* mmul.x: Moxie Options. (line 14) 65488* mmul32x16: ARC Options. (line 121) 65489* mmul64: ARC Options. (line 124) 65490* mmuladd: FRV Options. (line 64) 65491* mmulhw: RS/6000 and PowerPC Options. 65492 (line 378) 65493* mmult: MeP Options. (line 90) 65494* mmult-bug: MN10300 Options. (line 9) 65495* mmultcost: ARC Options. (line 576) 65496* mmulti-cond-exec: FRV Options. (line 215) 65497* mmulticore: Blackfin Options. (line 139) 65498* mmultiple: RS/6000 and PowerPC Options. 65499 (line 339) 65500* mmultiple-stld: C-SKY Options. (line 137) 65501* mmusl: GNU/Linux Options. (line 18) 65502* mmvcle: S/390 and zSeries Options. 65503 (line 138) 65504* mmvme: RS/6000 and PowerPC Options. 65505 (line 609) 65506* mmwait: x86 Options. (line 1059) 65507* mmwaitx: x86 Options. (line 894) 65508* mn: H8/300 Options. (line 20) 65509* mn-flash: AVR Options. (line 224) 65510* mnan=2008: MIPS Options. (line 320) 65511* mnan=legacy: MIPS Options. (line 320) 65512* mneeded: x86 Options. (line 1603) 65513* mneon-for-64bits: ARM Options. (line 923) 65514* mnested-cond-exec: FRV Options. (line 230) 65515* mnewlib: OpenRISC Options. (line 13) 65516* mnhwloop: Score Options. (line 15) 65517* mno-16-bit: NDS32 Options. (line 54) 65518* mno-4byte-functions: MCore Options. (line 27) 65519* mno-8byte-align: V850 Options. (line 170) 65520* mno-abicalls: MIPS Options. (line 192) 65521* mno-ac0: PDP-11 Options. (line 20) 65522* mno-align-double: x86 Options. (line 640) 65523* mno-align-int: M680x0 Options. (line 262) 65524* mno-align-loops: M32R/D Options. (line 76) 65525* mno-align-stringops: x86 Options. (line 1317) 65526* mno-allow-string-insns: RX Options. (line 150) 65527* mno-altivec: RS/6000 and PowerPC Options. 65528 (line 137) 65529* mno-am33: MN10300 Options. (line 20) 65530* mno-app-regs: SPARC Options. (line 10) 65531* mno-app-regs <1>: V850 Options. (line 185) 65532* mno-as100-syntax: RX Options. (line 76) 65533* mno-auto-litpools: Xtensa Options. (line 60) 65534* mno-avoid-indexed-addresses: RS/6000 and PowerPC Options. 65535 (line 360) 65536* mno-backchain: S/390 and zSeries Options. 65537 (line 35) 65538* mno-base-addresses: MMIX Options. (line 53) 65539* mno-bit-align: RS/6000 and PowerPC Options. 65540 (line 392) 65541* mno-bitfield: M680x0 Options. (line 228) 65542* mno-branch-likely: MIPS Options. (line 798) 65543* mno-branch-predict: MMIX Options. (line 48) 65544* mno-brcc: ARC Options. (line 444) 65545* mno-bwx: DEC Alpha Options. (line 163) 65546* mno-bypass-cache: Nios II Options. (line 103) 65547* mno-cache-volatile: Nios II Options. (line 109) 65548* mno-call-ms2sysv-xlogues: x86 Options. (line 1157) 65549* mno-callgraph-data: MCore Options. (line 31) 65550* mno-cbcond: SPARC Options. (line 260) 65551* mno-check-zero-division: MIPS Options. (line 570) 65552* mno-cix: DEC Alpha Options. (line 163) 65553* mno-clearbss: MicroBlaze Options. (line 16) 65554* mno-cmov: NDS32 Options. (line 24) 65555* mno-cmpb: RS/6000 and PowerPC Options. 65556 (line 25) 65557* mno-co-re: eBPF Options. (line 60) 65558* mno-cond-exec: ARC Options. (line 458) 65559* mno-cond-exec <1>: FRV Options. (line 194) 65560* mno-cond-move: FRV Options. (line 166) 65561* mno-const-align: CRIS Options. (line 54) 65562* mno-const16: Xtensa Options. (line 10) 65563* mno-crc: MIPS Options. (line 416) 65564* mno-crt0: MN10300 Options. (line 43) 65565* mno-crt0 <1>: Moxie Options. (line 18) 65566* mno-crypto: RS/6000 and PowerPC Options. 65567 (line 178) 65568* mno-csync-anomaly: Blackfin Options. (line 63) 65569* mno-custom-INSN: Nios II Options. (line 139) 65570* mno-data-align: CRIS Options. (line 54) 65571* mno-debug: S/390 and zSeries Options. 65572 (line 144) 65573* mno-default: x86 Options. (line 971) 65574* mno-direct-extern-access: x86 Options. (line 1608) 65575* mno-direct-extern-access <1>: x86 Function Attributes. 65576 (line 712) 65577* mno-disable-callt: V850 Options. (line 92) 65578* mno-div: M680x0 Options. (line 207) 65579* mno-div <1>: MCore Options. (line 15) 65580* mno-dlmzb: RS/6000 and PowerPC Options. 65581 (line 385) 65582* mno-double: FRV Options. (line 52) 65583* mno-dpfp-lrsr: ARC Options. (line 108) 65584* mno-dsp: MIPS Options. (line 353) 65585* mno-dspr2: MIPS Options. (line 359) 65586* mno-dwarf2-asm: IA-64 Options. (line 94) 65587* mno-dword: FRV Options. (line 44) 65588* mno-eabi: RS/6000 and PowerPC Options. 65589 (line 633) 65590* mno-early-stop-bits: IA-64 Options. (line 100) 65591* mno-eflags: FRV Options. (line 155) 65592* mno-embedded-data: MIPS Options. (line 517) 65593* mno-ep: V850 Options. (line 16) 65594* mno-epsilon: MMIX Options. (line 15) 65595* mno-eva: MIPS Options. (line 403) 65596* mno-explicit-relocs: DEC Alpha Options. (line 176) 65597* mno-explicit-relocs <1>: MIPS Options. (line 561) 65598* mno-exr: H8/300 Options. (line 33) 65599* mno-ext-perf: NDS32 Options. (line 30) 65600* mno-ext-perf2: NDS32 Options. (line 36) 65601* mno-ext-string: NDS32 Options. (line 42) 65602* mno-extern-sdata: MIPS Options. (line 480) 65603* mno-fancy-math-387: x86 Options. (line 630) 65604* mno-fast-sw-div: Nios II Options. (line 115) 65605* mno-faster-structs: SPARC Options. (line 91) 65606* mno-fdpic: ARM Options. (line 984) 65607* mno-fix: DEC Alpha Options. (line 163) 65608* mno-fix-24k: MIPS Options. (line 647) 65609* mno-fix-cortex-a53-835769: AArch64 Options. (line 90) 65610* mno-fix-cortex-a53-843419: AArch64 Options. (line 97) 65611* mno-fix-r10000: MIPS Options. (line 669) 65612* mno-fix-r4000: MIPS Options. (line 653) 65613* mno-fix-r4400: MIPS Options. (line 663) 65614* mno-flat: SPARC Options. (line 22) 65615* mno-float: MIPS Options. (line 274) 65616* mno-float128: RS/6000 and PowerPC Options. 65617 (line 215) 65618* mno-float128-hardware: RS/6000 and PowerPC Options. 65619 (line 236) 65620* mno-flush-func: M32R/D Options. (line 98) 65621* mno-flush-trap: M32R/D Options. (line 90) 65622* mno-fmaf: SPARC Options. (line 267) 65623* mno-fp-in-toc: RS/6000 and PowerPC Options. 65624 (line 256) 65625* mno-fp-regs: DEC Alpha Options. (line 25) 65626* mno-fp-ret-in-387: x86 Options. (line 620) 65627* mno-fprnd: RS/6000 and PowerPC Options. 65628 (line 25) 65629* mno-fpu: SPARC Options. (line 39) 65630* mno-fpu <1>: Visium Options. (line 24) 65631* mno-fsca: SH Options. (line 365) 65632* mno-fsmuld: SPARC Options. (line 274) 65633* mno-fsrra: SH Options. (line 374) 65634* mno-fused-madd: IA-64 Options. (line 88) 65635* mno-fused-madd <1>: MIPS Options. (line 630) 65636* mno-fused-madd <2>: RS/6000 and PowerPC Options. 65637 (line 369) 65638* mno-fused-madd <3>: S/390 and zSeries Options. 65639 (line 183) 65640* mno-fused-madd <4>: SH Options. (line 356) 65641* mno-fused-madd <5>: Xtensa Options. (line 19) 65642* mno-ginv: MIPS Options. (line 421) 65643* mno-gnu-as: IA-64 Options. (line 18) 65644* mno-gnu-attribute: RS/6000 and PowerPC Options. 65645 (line 587) 65646* mno-gnu-ld: IA-64 Options. (line 23) 65647* mno-gpopt: MIPS Options. (line 502) 65648* mno-gpopt <1>: Nios II Options. (line 16) 65649* mno-hard-dfp: RS/6000 and PowerPC Options. 65650 (line 25) 65651* mno-hard-dfp <1>: S/390 and zSeries Options. 65652 (line 20) 65653* mno-hardlit: MCore Options. (line 10) 65654* mno-htm: RS/6000 and PowerPC Options. 65655 (line 184) 65656* mno-htm <1>: S/390 and zSeries Options. 65657 (line 104) 65658* mno-hw-div: Nios II Options. (line 124) 65659* mno-hw-mul: Nios II Options. (line 124) 65660* mno-hw-mulx: Nios II Options. (line 124) 65661* mno-id-shared-library: Blackfin Options. (line 85) 65662* mno-ieee: SH Options. (line 165) 65663* mno-ieee-fp: x86 Options. (line 597) 65664* mno-imadd: MIPS Options. (line 623) 65665* mno-inline-float-divide: IA-64 Options. (line 62) 65666* mno-inline-int-divide: IA-64 Options. (line 73) 65667* mno-inline-sqrt: IA-64 Options. (line 84) 65668* mno-int16: PDP-11 Options. (line 37) 65669* mno-int32: PDP-11 Options. (line 33) 65670* mno-interlink-compressed: MIPS Options. (line 135) 65671* mno-interlink-mips16: MIPS Options. (line 147) 65672* mno-interrupts: AVR Options. (line 227) 65673* mno-isel: RS/6000 and PowerPC Options. 65674 (line 167) 65675* mno-jsr: RX Options. (line 169) 65676* mno-knuthdiv: MMIX Options. (line 32) 65677* mno-leaf-id-shared-library: Blackfin Options. (line 95) 65678* mno-libfuncs: MMIX Options. (line 10) 65679* mno-liw: MN10300 Options. (line 59) 65680* mno-llsc: MIPS Options. (line 339) 65681* mno-load-store-pairs: MIPS Options. (line 590) 65682* mno-local-sdata: MIPS Options. (line 468) 65683* mno-long-calls: ARM Options. (line 757) 65684* mno-long-calls <1>: Blackfin Options. (line 118) 65685* mno-long-calls <2>: HPPA Options. (line 136) 65686* mno-long-calls <3>: MIPS Options. (line 609) 65687* mno-long-calls <4>: V850 Options. (line 10) 65688* mno-long-jumps: V850 Options. (line 108) 65689* mno-longcall: RS/6000 and PowerPC Options. 65690 (line 728) 65691* mno-longcalls: Xtensa Options. (line 87) 65692* mno-loongson-ext: MIPS Options. (line 430) 65693* mno-loongson-ext2: MIPS Options. (line 434) 65694* mno-loongson-mmi: MIPS Options. (line 425) 65695* mno-low-precision-div: AArch64 Options. (line 120) 65696* mno-low-precision-recip-sqrt: AArch64 Options. (line 103) 65697* mno-low-precision-sqrt: AArch64 Options. (line 111) 65698* mno-low64k: Blackfin Options. (line 71) 65699* mno-lra: SPARC Options. (line 111) 65700* mno-lra <1>: VAX Options. (line 22) 65701* mno-lsim: FR30 Options. (line 14) 65702* mno-lsim <1>: MCore Options. (line 46) 65703* mno-mad: MIPS Options. (line 618) 65704* mno-max: DEC Alpha Options. (line 163) 65705* mno-mcount-ra-address: MIPS Options. (line 878) 65706* mno-mcu: MIPS Options. (line 399) 65707* mno-mdmx: MIPS Options. (line 376) 65708* mno-media: FRV Options. (line 60) 65709* mno-memcpy: MIPS Options. (line 603) 65710* mno-mfcrf: RS/6000 and PowerPC Options. 65711 (line 25) 65712* mno-mips16: MIPS Options. (line 120) 65713* mno-mips3d: MIPS Options. (line 382) 65714* mno-mma: RS/6000 and PowerPC Options. 65715 (line 944) 65716* mno-mmicromips: MIPS Options. (line 387) 65717* Mno-modules: Preprocessor Options. 65718 (line 132) 65719* mno-mpy: ARC Options. (line 117) 65720* mno-ms-bitfields: x86 Options. (line 1197) 65721* mno-mt: MIPS Options. (line 395) 65722* mno-mul-bug-workaround: CRIS Options. (line 31) 65723* mno-muladd: FRV Options. (line 68) 65724* mno-mulhw: RS/6000 and PowerPC Options. 65725 (line 378) 65726* mno-mult-bug: MN10300 Options. (line 13) 65727* mno-multi-cond-exec: FRV Options. (line 223) 65728* mno-multiple: RS/6000 and PowerPC Options. 65729 (line 339) 65730* mno-mvcle: S/390 and zSeries Options. 65731 (line 138) 65732* mno-nested-cond-exec: FRV Options. (line 237) 65733* mno-odd-spreg: MIPS Options. (line 293) 65734* mno-omit-leaf-frame-pointer: AArch64 Options. (line 58) 65735* mno-optimize-membar: FRV Options. (line 249) 65736* mno-opts: MeP Options. (line 93) 65737* mno-pack: FRV Options. (line 151) 65738* mno-packed-stack: S/390 and zSeries Options. 65739 (line 54) 65740* mno-paired-single: MIPS Options. (line 370) 65741* mno-pc-relative-literal-loads: AArch64 Options. (line 274) 65742* mno-pcrel: RS/6000 and PowerPC Options. 65743 (line 932) 65744* mno-pic: IA-64 Options. (line 26) 65745* mno-pid: RX Options. (line 117) 65746* mno-plt: MIPS Options. (line 219) 65747* mno-pltseq: RS/6000 and PowerPC Options. 65748 (line 765) 65749* mno-popc: SPARC Options. (line 281) 65750* mno-popcntb: RS/6000 and PowerPC Options. 65751 (line 25) 65752* mno-popcntd: RS/6000 and PowerPC Options. 65753 (line 25) 65754* mno-postinc: Adapteva Epiphany Options. 65755 (line 109) 65756* mno-postmodify: Adapteva Epiphany Options. 65757 (line 109) 65758* mno-power8-fusion: RS/6000 and PowerPC Options. 65759 (line 190) 65760* mno-power8-vector: RS/6000 and PowerPC Options. 65761 (line 196) 65762* mno-powerpc-gfxopt: RS/6000 and PowerPC Options. 65763 (line 25) 65764* mno-powerpc-gpopt: RS/6000 and PowerPC Options. 65765 (line 25) 65766* mno-powerpc64: RS/6000 and PowerPC Options. 65767 (line 25) 65768* mno-prefixed: RS/6000 and PowerPC Options. 65769 (line 938) 65770* mno-privileged: RS/6000 and PowerPC Options. 65771 (line 955) 65772* mno-prolog-function: V850 Options. (line 23) 65773* mno-prologue-epilogue: CRIS Options. (line 70) 65774* mno-prototype: RS/6000 and PowerPC Options. 65775 (line 593) 65776* mno-push-args: x86 Options. (line 1174) 65777* mno-quad-memory: RS/6000 and PowerPC Options. 65778 (line 203) 65779* mno-quad-memory-atomic: RS/6000 and PowerPC Options. 65780 (line 209) 65781* mno-readonly-in-sdata: RS/6000 and PowerPC Options. 65782 (line 684) 65783* mno-red-zone: x86 Options. (line 1562) 65784* mno-register-names: IA-64 Options. (line 37) 65785* mno-regnames: RS/6000 and PowerPC Options. 65786 (line 722) 65787* mno-relax: PRU Options. (line 21) 65788* mno-relax <1>: V850 Options. (line 103) 65789* mno-relax-immediate: MCore Options. (line 19) 65790* mno-relocatable: RS/6000 and PowerPC Options. 65791 (line 408) 65792* mno-relocatable-lib: RS/6000 and PowerPC Options. 65793 (line 419) 65794* mno-renesas: SH Options. (line 155) 65795* mno-rop-protect: RS/6000 and PowerPC Options. 65796 (line 949) 65797* mno-round-nearest: Adapteva Epiphany Options. 65798 (line 51) 65799* mno-save-mduc-in-interrupts: RL78 Options. (line 79) 65800* mno-scc: FRV Options. (line 180) 65801* mno-sched-ar-data-spec: IA-64 Options. (line 135) 65802* mno-sched-ar-in-data-spec: IA-64 Options. (line 157) 65803* mno-sched-br-data-spec: IA-64 Options. (line 128) 65804* mno-sched-br-in-data-spec: IA-64 Options. (line 150) 65805* mno-sched-control-spec: IA-64 Options. (line 142) 65806* mno-sched-count-spec-in-critical-path: IA-64 Options. (line 185) 65807* mno-sched-in-control-spec: IA-64 Options. (line 164) 65808* mno-sched-prefer-non-control-spec-insns: IA-64 Options. (line 178) 65809* mno-sched-prefer-non-data-spec-insns: IA-64 Options. (line 171) 65810* mno-sched-prolog: ARM Options. (line 32) 65811* mno-sdata: ARC Options. (line 422) 65812* mno-sdata <1>: IA-64 Options. (line 42) 65813* mno-sdata <2>: RS/6000 and PowerPC Options. 65814 (line 679) 65815* mno-sep-data: Blackfin Options. (line 113) 65816* mno-serialize-volatile: Xtensa Options. (line 35) 65817* mno-setlb: MN10300 Options. (line 69) 65818* mno-short: M680x0 Options. (line 223) 65819* mno-side-effects: CRIS Options. (line 45) 65820* mno-sim: RX Options. (line 71) 65821* mno-single-exit: MMIX Options. (line 65) 65822* mno-slow-bytes: MCore Options. (line 35) 65823* mno-small-exec: S/390 and zSeries Options. 65824 (line 79) 65825* mno-smartmips: MIPS Options. (line 366) 65826* mno-soft-cmpsf: Adapteva Epiphany Options. 65827 (line 29) 65828* mno-soft-float: DEC Alpha Options. (line 10) 65829* mno-space-regs: HPPA Options. (line 45) 65830* mno-specld-anomaly: Blackfin Options. (line 53) 65831* mno-split-addresses: MIPS Options. (line 555) 65832* mno-split-lohi: Adapteva Epiphany Options. 65833 (line 109) 65834* mno-stack-align: CRIS Options. (line 54) 65835* mno-stack-bias: SPARC Options. (line 372) 65836* mno-std-struct-return: SPARC Options. (line 102) 65837* mno-strict-align: AArch64 Options. (line 52) 65838* mno-strict-align <1>: M680x0 Options. (line 281) 65839* mno-strict-align <2>: RS/6000 and PowerPC Options. 65840 (line 403) 65841* mno-subxc: SPARC Options. (line 288) 65842* mno-sum-in-toc: RS/6000 and PowerPC Options. 65843 (line 256) 65844* mno-sym32: MIPS Options. (line 453) 65845* mno-target-align: Xtensa Options. (line 74) 65846* mno-text-section-literals: Xtensa Options. (line 47) 65847* mno-tls-markers: RS/6000 and PowerPC Options. 65848 (line 777) 65849* mno-toc: RS/6000 and PowerPC Options. 65850 (line 428) 65851* mno-toplevel-symbols: MMIX Options. (line 39) 65852* mno-tpf-trace: S/390 and zSeries Options. 65853 (line 169) 65854* mno-tpf-trace-skip: S/390 and zSeries Options. 65855 (line 175) 65856* mno-unaligned-access: ARM Options. (line 910) 65857* mno-unaligned-access <1>: MIPS Options. (line 597) 65858* mno-unaligned-doubles: SPARC Options. (line 73) 65859* mno-uninit-const-in-rodata: MIPS Options. (line 525) 65860* mno-update: RS/6000 and PowerPC Options. 65861 (line 350) 65862* mno-user-mode: SPARC Options. (line 85) 65863* mno-usermode: SH Options. (line 274) 65864* mno-v3push: NDS32 Options. (line 48) 65865* mno-v8plus: SPARC Options. (line 214) 65866* mno-vect-double: Adapteva Epiphany Options. 65867 (line 115) 65868* mno-virt: MIPS Options. (line 407) 65869* mno-vis: SPARC Options. (line 221) 65870* mno-vis2: SPARC Options. (line 227) 65871* mno-vis3: SPARC Options. (line 235) 65872* mno-vis4: SPARC Options. (line 243) 65873* mno-vis4b: SPARC Options. (line 251) 65874* mno-vliw-branch: FRV Options. (line 208) 65875* mno-volatile-asm-stop: IA-64 Options. (line 32) 65876* mno-volatile-cache: ARC Options. (line 431) 65877* mno-vrsave: RS/6000 and PowerPC Options. 65878 (line 153) 65879* mno-vsx: RS/6000 and PowerPC Options. 65880 (line 172) 65881* mno-vx: S/390 and zSeries Options. 65882 (line 112) 65883* mno-warn-devices-csv: MSP430 Options. (line 168) 65884* mno-warn-mcu: MSP430 Options. (line 65) 65885* mno-warn-multiple-fast-interrupts: RX Options. (line 143) 65886* mno-wide-bitfields: MCore Options. (line 23) 65887* mno-xgot: M680x0 Options. (line 313) 65888* mno-xgot <1>: MIPS Options. (line 229) 65889* mno-xl-compat: RS/6000 and PowerPC Options. 65890 (line 291) 65891* mno-xpa: MIPS Options. (line 411) 65892* mno-zdcbranch: SH Options. (line 341) 65893* mno-zero-extend: MMIX Options. (line 26) 65894* mno-zvector: S/390 and zSeries Options. 65895 (line 123) 65896* mnobitfield: M680x0 Options. (line 228) 65897* mnodiv: FT32 Options. (line 20) 65898* mnomacsave: SH Options. (line 160) 65899* mnop-fun-dllimport: x86 Windows Options. 65900 (line 22) 65901* mnop-mcount: x86 Options. (line 1404) 65902* mnopm: FT32 Options. (line 29) 65903* mnops: Adapteva Epiphany Options. 65904 (line 26) 65905* mnorm: ARC Options. (line 128) 65906* modd-spreg: MIPS Options. (line 293) 65907* momit-leaf-frame-pointer: AArch64 Options. (line 58) 65908* momit-leaf-frame-pointer <1>: Blackfin Options. (line 43) 65909* momit-leaf-frame-pointer <2>: x86 Options. (line 1368) 65910* mone-byte-bool: Darwin Options. (line 90) 65911* moptimize: Nvidia PTX Options. (line 45) 65912* moptimize-membar: FRV Options. (line 244) 65913* moptimize-membar <1>: FRV Options. (line 249) 65914* moverride: AArch64 Options. (line 261) 65915* MP: Preprocessor Options. 65916 (line 135) 65917* mpa-risc-1-0: HPPA Options. (line 19) 65918* mpa-risc-1-1: HPPA Options. (line 19) 65919* mpa-risc-2-0: HPPA Options. (line 19) 65920* mpack: FRV Options. (line 147) 65921* mpacked-stack: S/390 and zSeries Options. 65922 (line 54) 65923* mpadstruct: SH Options. (line 189) 65924* mpaired-single: MIPS Options. (line 370) 65925* mpc-relative-literal-loads: AArch64 Options. (line 274) 65926* mpc32: x86 Options. (line 762) 65927* mpc64: x86 Options. (line 762) 65928* mpc80: x86 Options. (line 762) 65929* mpclmul: x86 Options. (line 860) 65930* mpconfig: x86 Options. (line 868) 65931* mpcrel: M680x0 Options. (line 273) 65932* mpcrel <1>: RS/6000 and PowerPC Options. 65933 (line 932) 65934* mpdebug: CRIS Options. (line 35) 65935* mpe: RS/6000 and PowerPC Options. 65936 (line 310) 65937* mpe-aligned-commons: x86 Windows Options. 65938 (line 59) 65939* mpic-data-is-text-relative: ARM Options. (line 794) 65940* mpic-data-is-text-relative <1>: MicroBlaze Options. (line 92) 65941* mpic-register: ARM Options. (line 787) 65942* mpid: RX Options. (line 117) 65943* mpku: x86 Options. (line 896) 65944* mplt: MIPS Options. (line 219) 65945* mpltseq: RS/6000 and PowerPC Options. 65946 (line 765) 65947* mpointer-size=SIZE: VMS Options. (line 20) 65948* mpointers-to-nested-functions: RS/6000 and PowerPC Options. 65949 (line 878) 65950* mpoke-function-name: ARM Options. (line 802) 65951* mpopc: SPARC Options. (line 281) 65952* mpopcnt: x86 Options. (line 880) 65953* mpopcntb: RS/6000 and PowerPC Options. 65954 (line 25) 65955* mpopcntd: RS/6000 and PowerPC Options. 65956 (line 25) 65957* mportable-runtime: HPPA Options. (line 71) 65958* mpostinc: Adapteva Epiphany Options. 65959 (line 109) 65960* mpostmodify: Adapteva Epiphany Options. 65961 (line 109) 65962* mpower8-fusion: RS/6000 and PowerPC Options. 65963 (line 190) 65964* mpower8-vector: RS/6000 and PowerPC Options. 65965 (line 196) 65966* mpowerpc-gfxopt: RS/6000 and PowerPC Options. 65967 (line 25) 65968* mpowerpc-gpopt: RS/6000 and PowerPC Options. 65969 (line 25) 65970* mpowerpc64: RS/6000 and PowerPC Options. 65971 (line 25) 65972* mprefer-avx128: x86 Options. (line 995) 65973* mprefer-short-insn-regs: Adapteva Epiphany Options. 65974 (line 13) 65975* mprefer-vector-width: x86 Options. (line 999) 65976* mprefergot: SH Options. (line 268) 65977* mpreferred-stack-boundary: RISC-V Options. (line 101) 65978* mpreferred-stack-boundary <1>: x86 Options. (line 792) 65979* mprefetchwt1: x86 Options. (line 873) 65980* mprefixed: RS/6000 and PowerPC Options. 65981 (line 938) 65982* mpretend-cmove: SH Options. (line 383) 65983* mprfchw: x86 Options. (line 871) 65984* mprint-tune-info: ARM Options. (line 944) 65985* mprioritize-restricted-insns: RS/6000 and PowerPC Options. 65986 (line 457) 65987* mprivileged: RS/6000 and PowerPC Options. 65988 (line 955) 65989* mprolog-function: V850 Options. (line 23) 65990* mprologue-epilogue: CRIS Options. (line 70) 65991* mprototype: RS/6000 and PowerPC Options. 65992 (line 593) 65993* mptwrite: x86 Options. (line 864) 65994* mptx: Nvidia PTX Options. (line 30) 65995* mpure-code: ARM Options. (line 954) 65996* mpush-args: x86 Options. (line 1174) 65997* mpushpop: C-SKY Options. (line 130) 65998* MQ: Preprocessor Options. 65999 (line 162) 66000* mq-class: ARC Options. (line 519) 66001* mquad-memory: RS/6000 and PowerPC Options. 66002 (line 203) 66003* mquad-memory-atomic: RS/6000 and PowerPC Options. 66004 (line 209) 66005* mr0rel-sec: Nios II Options. (line 76) 66006* mr10k-cache-barrier: MIPS Options. (line 720) 66007* mRcq: ARC Options. (line 523) 66008* mRcw: ARC Options. (line 527) 66009* mrdpid: x86 Options. (line 872) 66010* mrdrnd: x86 Options. (line 865) 66011* mrdseed: x86 Options. (line 874) 66012* mreadonly-in-sdata: RS/6000 and PowerPC Options. 66013 (line 684) 66014* mrecip: RS/6000 and PowerPC Options. 66015 (line 785) 66016* mrecip <1>: x86 Options. (line 1064) 66017* mrecip-precision: RS/6000 and PowerPC Options. 66018 (line 842) 66019* mrecip=opt: RS/6000 and PowerPC Options. 66020 (line 798) 66021* mrecip=opt <1>: x86 Options. (line 1086) 66022* mrecord-mcount: x86 Options. (line 1398) 66023* mrecord-return: x86 Options. (line 1418) 66024* mred-zone: x86 Options. (line 1562) 66025* mreduced-regs: NDS32 Options. (line 15) 66026* mregister-names: IA-64 Options. (line 37) 66027* mregnames: RS/6000 and PowerPC Options. 66028 (line 722) 66029* mregparm: x86 Options. (line 732) 66030* mrelax: AVR Options. (line 231) 66031* mrelax <1>: H8/300 Options. (line 9) 66032* mrelax <2>: MN10300 Options. (line 46) 66033* mrelax <3>: MSP430 Options. (line 88) 66034* mrelax <4>: NDS32 Options. (line 84) 66035* mrelax <5>: RX Options. (line 95) 66036* mrelax <6>: SH Options. (line 137) 66037* mrelax <7>: V850 Options. (line 103) 66038* mrelax-cmpxchg-loop: x86 Options. (line 1475) 66039* mrelax-immediate: MCore Options. (line 19) 66040* mrelax-pic-calls: MIPS Options. (line 865) 66041* mrelocatable: RS/6000 and PowerPC Options. 66042 (line 408) 66043* mrelocatable-lib: RS/6000 and PowerPC Options. 66044 (line 419) 66045* mrenesas: SH Options. (line 152) 66046* mrepeat: MeP Options. (line 96) 66047* mrestrict-it: ARM Options. (line 938) 66048* mreturn-pointer-on-d0: MN10300 Options. (line 36) 66049* mrf16: ARC Options. (line 324) 66050* mrgf-banked-regs: ARC Options. (line 304) 66051* mrh850-abi: V850 Options. (line 127) 66052* mrl78: RL78 Options. (line 62) 66053* mrmw: AVR Options. (line 245) 66054* mrop-protect: RS/6000 and PowerPC Options. 66055 (line 949) 66056* mror: OpenRISC Options. (line 49) 66057* mrori: OpenRISC Options. (line 54) 66058* mround-nearest: Adapteva Epiphany Options. 66059 (line 51) 66060* mrtd: M680x0 Options. (line 237) 66061* mrtd <1>: x86 Options. (line 708) 66062* mrtd <2>: x86 Function Attributes. 66063 (line 9) 66064* mrtm: x86 Options. (line 891) 66065* mrtp: VxWorks Options. (line 11) 66066* mrtsc: ARC Options. (line 358) 66067* ms: H8/300 Options. (line 17) 66068* ms <1>: MeP Options. (line 100) 66069* ms2600: H8/300 Options. (line 24) 66070* msahf: x86 Options. (line 1035) 66071* msatur: MeP Options. (line 105) 66072* msave-acc-in-interrupts: RX Options. (line 109) 66073* msave-mduc-in-interrupts: RL78 Options. (line 79) 66074* msave-restore: RISC-V Options. (line 115) 66075* msave-toc-indirect: RS/6000 and PowerPC Options. 66076 (line 890) 66077* mscc: FRV Options. (line 173) 66078* msched-ar-data-spec: IA-64 Options. (line 135) 66079* msched-ar-in-data-spec: IA-64 Options. (line 157) 66080* msched-br-data-spec: IA-64 Options. (line 128) 66081* msched-br-in-data-spec: IA-64 Options. (line 150) 66082* msched-control-spec: IA-64 Options. (line 142) 66083* msched-costly-dep: RS/6000 and PowerPC Options. 66084 (line 464) 66085* msched-count-spec-in-critical-path: IA-64 Options. (line 185) 66086* msched-fp-mem-deps-zero-cost: IA-64 Options. (line 202) 66087* msched-in-control-spec: IA-64 Options. (line 164) 66088* msched-max-memory-insns: IA-64 Options. (line 211) 66089* msched-max-memory-insns-hard-limit: IA-64 Options. (line 217) 66090* msched-prefer-non-control-spec-insns: IA-64 Options. (line 178) 66091* msched-prefer-non-data-spec-insns: IA-64 Options. (line 171) 66092* msched-prolog: ARM Options. (line 32) 66093* msched-prolog <1>: C-SKY Options. (line 164) 66094* msched-spec-ldc: IA-64 Options. (line 191) 66095* msched-spec-ldc <1>: IA-64 Options. (line 194) 66096* msched-stop-bits-after-every-cycle: IA-64 Options. (line 198) 66097* mschedule: HPPA Options. (line 78) 66098* mscore5: Score Options. (line 25) 66099* mscore5u: Score Options. (line 28) 66100* mscore7: Score Options. (line 31) 66101* mscore7d: Score Options. (line 35) 66102* msda: V850 Options. (line 40) 66103* msdata: ARC Options. (line 422) 66104* msdata <1>: IA-64 Options. (line 42) 66105* msdata <2>: RS/6000 and PowerPC Options. 66106 (line 666) 66107* msdata=all: C6X Options. (line 30) 66108* msdata=data: RS/6000 and PowerPC Options. 66109 (line 671) 66110* msdata=default: C6X Options. (line 22) 66111* msdata=default <1>: RS/6000 and PowerPC Options. 66112 (line 666) 66113* msdata=eabi: RS/6000 and PowerPC Options. 66114 (line 647) 66115* msdata=none: C6X Options. (line 35) 66116* msdata=none <1>: M32R/D Options. (line 40) 66117* msdata=none <2>: RS/6000 and PowerPC Options. 66118 (line 679) 66119* msdata=sdata: M32R/D Options. (line 49) 66120* msdata=sysv: RS/6000 and PowerPC Options. 66121 (line 657) 66122* msdata=use: M32R/D Options. (line 53) 66123* msdram: Blackfin Options. (line 169) 66124* msdram <1>: MeP Options. (line 110) 66125* msecure-plt: RS/6000 and PowerPC Options. 66126 (line 156) 66127* msecurity: C-SKY Options. (line 96) 66128* msel-sched-dont-check-control-spec: IA-64 Options. (line 207) 66129* msep-data: Blackfin Options. (line 107) 66130* msep-data <1>: Blackfin Options. (line 113) 66131* mserialize: x86 Options. (line 917) 66132* mserialize-volatile: Xtensa Options. (line 35) 66133* msetlb: MN10300 Options. (line 64) 66134* msext: OpenRISC Options. (line 59) 66135* msfimm: OpenRISC Options. (line 63) 66136* msgx: x86 Options. (line 875) 66137* msha: x86 Options. (line 858) 66138* mshared-library-id: Blackfin Options. (line 100) 66139* mshftimm: OpenRISC Options. (line 68) 66140* mshort: M680x0 Options. (line 217) 66141* mshort-calls: AVR Options. (line 249) 66142* mshorten-memrefs: RISC-V Options. (line 121) 66143* mshstk: x86 Options. (line 1049) 66144* mside-effects: CRIS Options. (line 45) 66145* msign-extend-enabled: LM32 Options. (line 18) 66146* msign-return-address: AArch64 Options. (line 280) 66147* msilicon-errata: MSP430 Options. (line 159) 66148* msilicon-errata-warn: MSP430 Options. (line 163) 66149* msim: Blackfin Options. (line 36) 66150* msim <1>: C6X Options. (line 19) 66151* msim <2>: CR16 Options. (line 18) 66152* msim <3>: C-SKY Options. (line 170) 66153* msim <4>: FT32 Options. (line 9) 66154* msim <5>: M32C Options. (line 13) 66155* msim <6>: MeP Options. (line 114) 66156* msim <7>: MSP430 Options. (line 77) 66157* msim <8>: RL78 Options. (line 7) 66158* msim <9>: RS/6000 and PowerPC Options. 66159 (line 603) 66160* msim <10>: RX Options. (line 71) 66161* msim <11>: Visium Options. (line 13) 66162* msim <12>: Xstormy16 Options. (line 9) 66163* msimd: ARC Options. (line 141) 66164* msimnovec: MeP Options. (line 117) 66165* msingle-exit: MMIX Options. (line 65) 66166* msingle-float: MIPS Options. (line 284) 66167* msingle-pic-base: ARM Options. (line 781) 66168* msingle-pic-base <1>: RS/6000 and PowerPC Options. 66169 (line 451) 66170* msio: HPPA Options. (line 105) 66171* msize-level: ARC Options. (line 531) 66172* mskip-rax-setup: x86 Options. (line 1431) 66173* mslow-bytes: MCore Options. (line 35) 66174* mslow-flash-data: ARM Options. (line 926) 66175* msmall: MSP430 Options. (line 85) 66176* msmall-data: DEC Alpha Options. (line 187) 66177* msmall-data-limit: RISC-V Options. (line 110) 66178* msmall-data-limit <1>: RX Options. (line 47) 66179* msmall-divides: MicroBlaze Options. (line 38) 66180* msmall-exec: S/390 and zSeries Options. 66181 (line 79) 66182* msmall-model: FR30 Options. (line 9) 66183* msmall-text: DEC Alpha Options. (line 205) 66184* msmall16: Adapteva Epiphany Options. 66185 (line 66) 66186* msmallc: Nios II Options. (line 338) 66187* msmart: C-SKY Options. (line 113) 66188* msmartmips: MIPS Options. (line 366) 66189* msoft-cmpsf: Adapteva Epiphany Options. 66190 (line 29) 66191* msoft-div: OpenRISC Options. (line 19) 66192* msoft-float: ARC Options. (line 145) 66193* msoft-float <1>: C-SKY Options. (line 52) 66194* msoft-float <2>: DEC Alpha Options. (line 10) 66195* msoft-float <3>: FRV Options. (line 27) 66196* msoft-float <4>: HPPA Options. (line 91) 66197* msoft-float <5>: LoongArch Options. (line 66) 66198* msoft-float <6>: M680x0 Options. (line 201) 66199* msoft-float <7>: MicroBlaze Options. (line 7) 66200* msoft-float <8>: MIPS Options. (line 270) 66201* msoft-float <9>: OpenRISC Options. (line 29) 66202* msoft-float <10>: PDP-11 Options. (line 13) 66203* msoft-float <11>: RS/6000 and PowerPC Options. 66204 (line 333) 66205* msoft-float <12>: S/390 and zSeries Options. 66206 (line 11) 66207* msoft-float <13>: SPARC Options. (line 39) 66208* msoft-float <14>: V850 Options. (line 113) 66209* msoft-float <15>: Visium Options. (line 24) 66210* msoft-float <16>: x86 Options. (line 607) 66211* msoft-mul: OpenRISC Options. (line 24) 66212* msoft-quad-float: SPARC Options. (line 59) 66213* msoft-stack: Nvidia PTX Options. (line 49) 66214* msp8: AVR Options. (line 256) 66215* mspace: V850 Options. (line 30) 66216* mspace-regs: HPPA Options. (line 45) 66217* mspecld-anomaly: Blackfin Options. (line 48) 66218* mspecld-anomaly <1>: Blackfin Options. (line 53) 66219* mspfp: ARC Options. (line 132) 66220* mspfp-compact: ARC Options. (line 133) 66221* mspfp-fast: ARC Options. (line 137) 66222* mspfp_compact: ARC Options. (line 627) 66223* mspfp_fast: ARC Options. (line 630) 66224* msplit: PDP-11 Options. (line 40) 66225* msplit-addresses: MIPS Options. (line 555) 66226* msplit-lohi: Adapteva Epiphany Options. 66227 (line 109) 66228* msplit-vecmove-early: Adapteva Epiphany Options. 66229 (line 126) 66230* msram-ecc: AMD GCN Options. (line 25) 66231* msse: x86 Options. (line 839) 66232* msse2: x86 Options. (line 840) 66233* msse2avx: x86 Options. (line 1386) 66234* msse3: x86 Options. (line 841) 66235* msse4: x86 Options. (line 843) 66236* msse4.1: x86 Options. (line 845) 66237* msse4.2: x86 Options. (line 846) 66238* msse4a: x86 Options. (line 844) 66239* msseregparm: x86 Options. (line 743) 66240* mssse3: x86 Options. (line 842) 66241* mstack-align: CRIS Options. (line 54) 66242* mstack-bias: SPARC Options. (line 372) 66243* mstack-check-l1: Blackfin Options. (line 74) 66244* mstack-guard: S/390 and zSeries Options. 66245 (line 202) 66246* mstack-increment: MCore Options. (line 50) 66247* mstack-offset: Adapteva Epiphany Options. 66248 (line 37) 66249* mstack-protector-guard: AArch64 Options. (line 64) 66250* mstack-protector-guard <1>: ARM Options. (line 976) 66251* mstack-protector-guard <2>: RISC-V Options. (line 185) 66252* mstack-protector-guard <3>: RS/6000 and PowerPC Options. 66253 (line 916) 66254* mstack-protector-guard <4>: x86 Options. (line 1457) 66255* mstack-protector-guard-offset: AArch64 Options. (line 64) 66256* mstack-protector-guard-offset <1>: ARM Options. (line 976) 66257* mstack-protector-guard-offset <2>: RISC-V Options. (line 185) 66258* mstack-protector-guard-offset <3>: RS/6000 and PowerPC Options. 66259 (line 916) 66260* mstack-protector-guard-offset <4>: x86 Options. (line 1457) 66261* mstack-protector-guard-reg: AArch64 Options. (line 64) 66262* mstack-protector-guard-reg <1>: RISC-V Options. (line 185) 66263* mstack-protector-guard-reg <2>: RS/6000 and PowerPC Options. 66264 (line 916) 66265* mstack-protector-guard-reg <3>: x86 Options. (line 1457) 66266* mstack-protector-guard-symbol: RS/6000 and PowerPC Options. 66267 (line 916) 66268* mstack-size: AMD GCN Options. (line 31) 66269* mstack-size <1>: C-SKY Options. (line 150) 66270* mstack-size <2>: S/390 and zSeries Options. 66271 (line 202) 66272* mstackrealign: x86 Options. (line 783) 66273* mstd-struct-return: SPARC Options. (line 102) 66274* mstore-max: x86 Options. (line 1008) 66275* mstrict-align: AArch64 Options. (line 52) 66276* mstrict-align <1>: M680x0 Options. (line 281) 66277* mstrict-align <2>: RISC-V Options. (line 130) 66278* mstrict-align <3>: RS/6000 and PowerPC Options. 66279 (line 403) 66280* mstrict-X: AVR Options. (line 269) 66281* mstring-compare-inline-limit: RS/6000 and PowerPC Options. 66282 (line 708) 66283* mstringop-strategy=ALG: x86 Options. (line 1334) 66284* mstructure-size-boundary: ARM Options. (line 736) 66285* msubxc: SPARC Options. (line 288) 66286* msv-mode: Visium Options. (line 52) 66287* msve-vector-bits: AArch64 Options. (line 309) 66288* msvr4-struct-return: RS/6000 and PowerPC Options. 66289 (line 548) 66290* mswap: ARC Options. (line 152) 66291* mswape: ARC Options. (line 363) 66292* msym32: MIPS Options. (line 453) 66293* msynci: MIPS Options. (line 851) 66294* msys-crt0: Nios II Options. (line 342) 66295* msys-lib: Nios II Options. (line 346) 66296* MT: Preprocessor Options. 66297 (line 147) 66298* mtarget-align: Xtensa Options. (line 74) 66299* mtas: SH Options. (line 259) 66300* mtbm: x86 Options. (line 893) 66301* mtda: V850 Options. (line 34) 66302* mtelephony: ARC Options. (line 368) 66303* mtext-section-literals: Xtensa Options. (line 47) 66304* mtf: MeP Options. (line 121) 66305* mthreads: x86 Options. (line 1189) 66306* mthreads <1>: x86 Windows Options. 66307 (line 26) 66308* mthumb: ARM Options. (line 824) 66309* mthumb-interwork: ARM Options. (line 24) 66310* mtiny-printf: MSP430 Options. (line 122) 66311* mtiny-stack: AVR Options. (line 283) 66312* mtiny=: MeP Options. (line 125) 66313* mTLS: FRV Options. (line 90) 66314* mtls: FRV Options. (line 94) 66315* mtls-dialect: ARM Options. (line 876) 66316* mtls-dialect <1>: x86 Options. (line 1167) 66317* mtls-dialect=desc: AArch64 Options. (line 77) 66318* mtls-dialect=traditional: AArch64 Options. (line 81) 66319* mtls-direct-seg-refs: x86 Options. (line 1376) 66320* mtls-markers: RS/6000 and PowerPC Options. 66321 (line 777) 66322* mtls-size: AArch64 Options. (line 85) 66323* mtls-size <1>: IA-64 Options. (line 112) 66324* mtoc: RS/6000 and PowerPC Options. 66325 (line 428) 66326* mtomcat-stats: FRV Options. (line 254) 66327* mtoplevel-symbols: MMIX Options. (line 39) 66328* mtp: ARM Options. (line 868) 66329* mtp-regno: ARC Options. (line 170) 66330* mtpcs-frame: ARM Options. (line 841) 66331* mtpcs-leaf-frame: ARM Options. (line 847) 66332* mtpf-trace: S/390 and zSeries Options. 66333 (line 169) 66334* mtpf-trace-skip: S/390 and zSeries Options. 66335 (line 175) 66336* mtraceback: RS/6000 and PowerPC Options. 66337 (line 541) 66338* mtrap-precision: DEC Alpha Options. (line 109) 66339* mtrust: C-SKY Options. (line 99) 66340* mtsxldtrk: x86 Options. (line 909) 66341* mtune: AArch64 Options. (line 189) 66342* mtune <1>: AMD GCN Options. (line 10) 66343* mtune <2>: ARC Options. (line 552) 66344* mtune <3>: ARC Options. (line 633) 66345* mtune <4>: ARM Options. (line 571) 66346* mtune <5>: CRIS Options. (line 16) 66347* mtune <6>: DEC Alpha Options. (line 259) 66348* mtune <7>: IA-64 Options. (line 116) 66349* mtune <8>: LoongArch Options. (line 32) 66350* mtune <9>: M680x0 Options. (line 69) 66351* mtune <10>: MIPS Options. (line 66) 66352* mtune <11>: MN10300 Options. (line 30) 66353* mtune <12>: RISC-V Options. (line 86) 66354* mtune <13>: RS/6000 and PowerPC Options. 66355 (line 114) 66356* mtune <14>: S/390 and zSeries Options. 66357 (line 162) 66358* mtune <15>: SPARC Options. (line 199) 66359* mtune <16>: Visium Options. (line 47) 66360* mtune <17>: x86 Options. (line 481) 66361* mtune-ctrl=FEATURE-LIST: x86 Options. (line 962) 66362* muclibc: GNU/Linux Options. (line 14) 66363* muintr: x86 Options. (line 908) 66364* muls: Score Options. (line 18) 66365* multcost: ARC Options. (line 638) 66366* multcost=NUMBER: SH Options. (line 281) 66367* multilib-library-pic: FRV Options. (line 110) 66368* multiply-enabled: LM32 Options. (line 15) 66369* multiply_defined: Darwin Options. (line 196) 66370* multiply_defined_unused: Darwin Options. (line 196) 66371* multi_module: Darwin Options. (line 196) 66372* munalign-prob-threshold: ARC Options. (line 580) 66373* munaligned-access: ARM Options. (line 910) 66374* munaligned-access <1>: MIPS Options. (line 597) 66375* munaligned-doubles: SPARC Options. (line 73) 66376* municode: x86 Windows Options. 66377 (line 30) 66378* muniform-simt: Nvidia PTX Options. (line 61) 66379* muninit-const-in-rodata: MIPS Options. (line 525) 66380* munix: VAX Options. (line 9) 66381* munix-asm: PDP-11 Options. (line 43) 66382* munordered-float: OpenRISC Options. (line 39) 66383* mupdate: RS/6000 and PowerPC Options. 66384 (line 350) 66385* muser-enabled: LM32 Options. (line 21) 66386* muser-mode: SPARC Options. (line 85) 66387* muser-mode <1>: Visium Options. (line 57) 66388* musermode: SH Options. (line 274) 66389* mv3push: NDS32 Options. (line 45) 66390* mv850: V850 Options. (line 49) 66391* mv850e: V850 Options. (line 79) 66392* mv850e1: V850 Options. (line 70) 66393* mv850e2: V850 Options. (line 66) 66394* mv850e2v3: V850 Options. (line 61) 66395* mv850e2v4: V850 Options. (line 57) 66396* mv850e3v5: V850 Options. (line 52) 66397* mv850es: V850 Options. (line 75) 66398* mv8plus: SPARC Options. (line 214) 66399* mvaes: x86 Options. (line 901) 66400* mvdsp: C-SKY Options. (line 104) 66401* mveclibabi: RS/6000 and PowerPC Options. 66402 (line 851) 66403* mveclibabi <1>: x86 Options. (line 1115) 66404* mvect-double: Adapteva Epiphany Options. 66405 (line 115) 66406* mvect8-ret-in-mem: x86 Options. (line 753) 66407* mverbose-cost-dump: AArch64 Options. (line 269) 66408* mverbose-cost-dump <1>: ARM Options. (line 950) 66409* mvirt: MIPS Options. (line 407) 66410* mvis: SPARC Options. (line 221) 66411* mvis2: SPARC Options. (line 227) 66412* mvis3: SPARC Options. (line 235) 66413* mvis4: SPARC Options. (line 243) 66414* mvis4b: SPARC Options. (line 251) 66415* mvliw-branch: FRV Options. (line 201) 66416* mvms-return-codes: VMS Options. (line 9) 66417* mvolatile-asm-stop: IA-64 Options. (line 32) 66418* mvolatile-cache: ARC Options. (line 427) 66419* mvolatile-cache <1>: ARC Options. (line 431) 66420* mvpclmulqdq: x86 Options. (line 903) 66421* mvr4130-align: MIPS Options. (line 840) 66422* mvrsave: RS/6000 and PowerPC Options. 66423 (line 153) 66424* mvsx: RS/6000 and PowerPC Options. 66425 (line 172) 66426* mvx: S/390 and zSeries Options. 66427 (line 112) 66428* mvxworks: RS/6000 and PowerPC Options. 66429 (line 624) 66430* mvzeroupper: x86 Options. (line 989) 66431* mwaitpkg: x86 Options. (line 902) 66432* mwarn-devices-csv: MSP430 Options. (line 168) 66433* mwarn-dynamicstack: S/390 and zSeries Options. 66434 (line 196) 66435* mwarn-framesize: S/390 and zSeries Options. 66436 (line 188) 66437* mwarn-mcu: MSP430 Options. (line 65) 66438* mwarn-multiple-fast-interrupts: RX Options. (line 143) 66439* mwbnoinvd: x86 Options. (line 869) 66440* mwide-bitfields: MCore Options. (line 23) 66441* mwidekl: x86 Options. (line 923) 66442* mwin32: x86 Windows Options. 66443 (line 35) 66444* mwindows: x86 Windows Options. 66445 (line 41) 66446* mword-relocations: ARM Options. (line 887) 66447* mx32: x86 Options. (line 1542) 66448* mxgot: M680x0 Options. (line 313) 66449* mxgot <1>: MIPS Options. (line 229) 66450* mxl-barrel-shift: MicroBlaze Options. (line 32) 66451* mxl-compat: RS/6000 and PowerPC Options. 66452 (line 291) 66453* mxl-float-convert: MicroBlaze Options. (line 50) 66454* mxl-float-sqrt: MicroBlaze Options. (line 53) 66455* mxl-gp-opt: MicroBlaze Options. (line 44) 66456* mxl-multiply-high: MicroBlaze Options. (line 47) 66457* mxl-pattern-compare: MicroBlaze Options. (line 35) 66458* mxl-reorder: MicroBlaze Options. (line 62) 66459* mxl-soft-div: MicroBlaze Options. (line 29) 66460* mxl-soft-mul: MicroBlaze Options. (line 26) 66461* mxl-stack-check: MicroBlaze Options. (line 41) 66462* mxnack: AMD GCN Options. (line 38) 66463* mxop: x86 Options. (line 876) 66464* mxpa: MIPS Options. (line 411) 66465* mxsave: x86 Options. (line 887) 66466* mxsavec: x86 Options. (line 889) 66467* mxsaveopt: x86 Options. (line 888) 66468* mxsaves: x86 Options. (line 890) 66469* mxy: ARC Options. (line 373) 66470* myellowknife: RS/6000 and PowerPC Options. 66471 (line 619) 66472* mzarch: S/390 and zSeries Options. 66473 (line 94) 66474* mzda: V850 Options. (line 45) 66475* mzdcbranch: SH Options. (line 341) 66476* mzero-extend: MMIX Options. (line 26) 66477* mzvector: S/390 and zSeries Options. 66478 (line 123) 66479* no-80387: x86 Options. (line 607) 66480* no-block-ops-unaligned-vsx: RS/6000 and PowerPC Options. 66481 (line 959) 66482* no-canonical-prefixes: Directory Options. (line 164) 66483* no-integrated-cpp: Preprocessor Options. 66484 (line 485) 66485* no-pie: Link Options. (line 184) 66486* no-sysroot-suffix: Directory Options. (line 183) 66487* noall_load: Darwin Options. (line 196) 66488* nocpp: MIPS Options. (line 642) 66489* nodefaultlibs: Link Options. (line 122) 66490* nodevicelib: AVR Options. (line 290) 66491* nodevicespecs: AVR Options. (line 293) 66492* nofixprebinding: Darwin Options. (line 196) 66493* nofpu: RX Options. (line 17) 66494* nolibc: Link Options. (line 134) 66495* nolibdld: HPPA Options. (line 188) 66496* nomultidefs: Darwin Options. (line 196) 66497* non-static: VxWorks Options. (line 16) 66498* noprebind: Darwin Options. (line 196) 66499* noseglinkedit: Darwin Options. (line 196) 66500* nostartfiles: Link Options. (line 117) 66501* nostdinc: Directory Options. (line 102) 66502* nostdinc++: C++ Dialect Options. 66503 (line 555) 66504* nostdinc++ <1>: Directory Options. (line 108) 66505* nostdlib: Link Options. (line 146) 66506* no_dead_strip_inits_and_terms: Darwin Options. (line 196) 66507* o: Overall Options. (line 193) 66508* O: Optimize Options. (line 39) 66509* O0: Optimize Options. (line 166) 66510* O1: Optimize Options. (line 39) 66511* O2: Optimize Options. (line 97) 66512* O3: Optimize Options. (line 148) 66513* Ofast: Optimize Options. (line 182) 66514* Og: Optimize Options. (line 190) 66515* Os: Optimize Options. (line 170) 66516* Oz: Optimize Options. (line 209) 66517* p: Instrumentation Options. 66518 (line 20) 66519* P: Preprocessor Options. 66520 (line 370) 66521* p <1>: Common Function Attributes. 66522 (line 814) 66523* pagezero_size: Darwin Options. (line 196) 66524* param: Optimize Options. (line 2736) 66525* pass-exit-codes: Overall Options. (line 597) 66526* pedantic: Standards. (line 13) 66527* pedantic <1>: Warning Options. (line 88) 66528* pedantic <2>: C Extensions. (line 6) 66529* pedantic <3>: Alternate Keywords. (line 30) 66530* pedantic <4>: Warnings and Errors. 66531 (line 25) 66532* pedantic-errors: Standards. (line 13) 66533* pedantic-errors <1>: Warning Options. (line 131) 66534* pedantic-errors <2>: Non-bugs. (line 216) 66535* pedantic-errors <3>: Warnings and Errors. 66536 (line 25) 66537* pg: Instrumentation Options. 66538 (line 20) 66539* pg <1>: Common Function Attributes. 66540 (line 814) 66541* pie: Link Options. (line 178) 66542* pipe: Overall Options. (line 605) 66543* plt: RISC-V Options. (line 13) 66544* prebind: Darwin Options. (line 196) 66545* prebind_all_twolevel_modules: Darwin Options. (line 196) 66546* print-file-name: Developer Options. (line 935) 66547* print-libgcc-file-name: Developer Options. (line 969) 66548* print-multi-directory: Developer Options. (line 941) 66549* print-multi-lib: Developer Options. (line 946) 66550* print-multi-os-directory: Developer Options. (line 953) 66551* print-multiarch: Developer Options. (line 962) 66552* print-objc-runtime-info: Objective-C and Objective-C++ Dialect Options. 66553 (line 227) 66554* print-prog-name: Developer Options. (line 966) 66555* print-search-dirs: Developer Options. (line 977) 66556* print-sysroot: Developer Options. (line 990) 66557* print-sysroot-headers-suffix: Developer Options. (line 997) 66558* private_bundle: Darwin Options. (line 196) 66559* pthread: Preprocessor Options. 66560 (line 70) 66561* pthread <1>: Link Options. (line 195) 66562* pthreads: Solaris 2 Options. (line 30) 66563* Q: Developer Options. (line 839) 66564* Qn: System V Options. (line 18) 66565* Qy: System V Options. (line 14) 66566* r: Link Options. (line 202) 66567* rdynamic: Link Options. (line 206) 66568* read_only_relocs: Darwin Options. (line 196) 66569* remap: Preprocessor Options. 66570 (line 401) 66571* S: Overall Options. (line 176) 66572* S <1>: Link Options. (line 20) 66573* s: Link Options. (line 213) 66574* save-temps: Developer Options. (line 735) 66575* save-temps=cwd: Developer Options. (line 746) 66576* save-temps=obj: Developer Options. (line 749) 66577* sectalign: Darwin Options. (line 196) 66578* sectcreate: Darwin Options. (line 196) 66579* sectobjectsymbols: Darwin Options. (line 196) 66580* sectobjectsymbols <1>: Darwin Options. (line 196) 66581* sectorder: Darwin Options. (line 196) 66582* seg1addr: Darwin Options. (line 196) 66583* segaddr: Darwin Options. (line 196) 66584* seglinkedit: Darwin Options. (line 196) 66585* segprot: Darwin Options. (line 196) 66586* segs_read_only_addr: Darwin Options. (line 196) 66587* segs_read_only_addr <1>: Darwin Options. (line 196) 66588* segs_read_write_addr: Darwin Options. (line 196) 66589* segs_read_write_addr <1>: Darwin Options. (line 196) 66590* seg_addr_table: Darwin Options. (line 196) 66591* seg_addr_table_filename: Darwin Options. (line 196) 66592* shared: Link Options. (line 222) 66593* shared-libgcc: Link Options. (line 230) 66594* short-calls: Adapteva Epiphany Options. 66595 (line 61) 66596* sim: CRIS Options. (line 82) 66597* sim2: CRIS Options. (line 87) 66598* single_module: Darwin Options. (line 196) 66599* specs: Overall Options. (line 611) 66600* static: Link Options. (line 217) 66601* static <1>: Darwin Options. (line 196) 66602* static <2>: HPPA Options. (line 192) 66603* static-libasan: Link Options. (line 264) 66604* static-libgcc: Link Options. (line 230) 66605* static-liblsan: Link Options. (line 280) 66606* static-libstdc++: Link Options. (line 297) 66607* static-libtsan: Link Options. (line 272) 66608* static-libubsan: Link Options. (line 288) 66609* static-pie: Link Options. (line 187) 66610* std: Standards. (line 13) 66611* std <1>: C Dialect Options. (line 46) 66612* std <2>: Other Builtins. (line 31) 66613* std <3>: Non-bugs. (line 107) 66614* stdlib: C++ Dialect Options. 66615 (line 578) 66616* sub_library: Darwin Options. (line 196) 66617* sub_umbrella: Darwin Options. (line 196) 66618* symbolic: Link Options. (line 308) 66619* sysroot: Directory Options. (line 168) 66620* T: Link Options. (line 314) 66621* target-help: Overall Options. (line 487) 66622* threads: HPPA Options. (line 205) 66623* time: Developer Options. (line 755) 66624* tno-android-cc: GNU/Linux Options. (line 36) 66625* tno-android-ld: GNU/Linux Options. (line 40) 66626* traditional: Preprocessor Options. 66627 (line 377) 66628* traditional <1>: Incompatibilities. (line 6) 66629* traditional-cpp: Preprocessor Options. 66630 (line 377) 66631* trigraphs: Preprocessor Options. 66632 (line 387) 66633* twolevel_namespace: Darwin Options. (line 196) 66634* U: Preprocessor Options. 66635 (line 42) 66636* u: Link Options. (line 346) 66637* umbrella: Darwin Options. (line 196) 66638* undef: Preprocessor Options. 66639 (line 66) 66640* undefined: Darwin Options. (line 196) 66641* unexported_symbols_list: Darwin Options. (line 196) 66642* v: Overall Options. (line 466) 66643* version: Overall Options. (line 594) 66644* w: Warning Options. (line 25) 66645* W: Warning Options. (line 222) 66646* W <1>: Warning Options. (line 2948) 66647* W <2>: Warning Options. (line 3105) 66648* W <3>: Incompatibilities. (line 64) 66649* Wa: Assembler Options. (line 9) 66650* Wabi: Warning Options. (line 267) 66651* Wabi-tag: C++ Dialect Options. 66652 (line 590) 66653* Wabsolute-value: Warning Options. (line 2352) 66654* Waddr-space-convert: AVR Options. (line 308) 66655* Waddress: Warning Options. (line 2781) 66656* Waddress-of-packed-member: Warning Options. (line 2815) 66657* Waggregate-return: Warning Options. (line 2843) 66658* Waggressive-loop-optimizations: Warning Options. (line 2848) 66659* Waligned-new: C++ Dialect Options. 66660 (line 1180) 66661* Wall: Warning Options. (line 140) 66662* Wall <1>: Standard Libraries. (line 6) 66663* Walloc-size-larger-than=: Warning Options. (line 1715) 66664* Walloc-zero: Warning Options. (line 1705) 66665* Walloca: Warning Options. (line 1730) 66666* Walloca-larger-than=: Warning Options. (line 1733) 66667* Wanalyzer-double-fclose: Static Analyzer Options. 66668 (line 59) 66669* Wanalyzer-double-free: Static Analyzer Options. 66670 (line 66) 66671* Wanalyzer-exposure-through-output-file: Static Analyzer Options. 66672 (line 74) 66673* Wanalyzer-file-leak: Static Analyzer Options. 66674 (line 82) 66675* Wanalyzer-free-of-non-heap: Static Analyzer Options. 66676 (line 89) 66677* Wanalyzer-malloc-leak: Static Analyzer Options. 66678 (line 97) 66679* Wanalyzer-mismatching-deallocation: Static Analyzer Options. 66680 (line 105) 66681* Wanalyzer-null-argument: Static Analyzer Options. 66682 (line 131) 66683* Wanalyzer-null-dereference: Static Analyzer Options. 66684 (line 139) 66685* Wanalyzer-possible-null-argument: Static Analyzer Options. 66686 (line 116) 66687* Wanalyzer-possible-null-dereference: Static Analyzer Options. 66688 (line 124) 66689* Wanalyzer-shift-count-negative: Static Analyzer Options. 66690 (line 146) 66691* Wanalyzer-shift-count-overflow: Static Analyzer Options. 66692 (line 158) 66693* Wanalyzer-stale-setjmp-buffer: Static Analyzer Options. 66694 (line 171) 66695* Wanalyzer-tainted-allocation-size: Static Analyzer Options. 66696 (line 185) 66697* Wanalyzer-tainted-array-index: Static Analyzer Options. 66698 (line 199) 66699* Wanalyzer-tainted-divisor: Static Analyzer Options. 66700 (line 212) 66701* Wanalyzer-tainted-offset: Static Analyzer Options. 66702 (line 222) 66703* Wanalyzer-tainted-size: Static Analyzer Options. 66704 (line 235) 66705* Wanalyzer-too-complex: Static Analyzer Options. 66706 (line 49) 66707* Wanalyzer-unsafe-call-within-signal-handler: Static Analyzer Options. 66708 (line 245) 66709* Wanalyzer-use-after-free: Static Analyzer Options. 66710 (line 253) 66711* Wanalyzer-use-of-pointer-in-stale-stack-frame: Static Analyzer Options. 66712 (line 261) 66713* Wanalyzer-use-of-uninitialized-value: Static Analyzer Options. 66714 (line 288) 66715* Wanalyzer-write-to-const: Static Analyzer Options. 66716 (line 268) 66717* Wanalyzer-write-to-string-literal: Static Analyzer Options. 66718 (line 278) 66719* Warith-conversion: Warning Options. (line 1798) 66720* Warray-bounds: Warning Options. (line 1811) 66721* Warray-compare: Warning Options. (line 1830) 66722* Wassign-intercept: Objective-C and Objective-C++ Dialect Options. 66723 (line 170) 66724* Wattribute-alias: Warning Options. (line 1886) 66725* Wattribute-warning: Warning Options. (line 3072) 66726* Wattributes: Warning Options. (line 2853) 66727* Wbad-function-cast: Warning Options. (line 2421) 66728* Wbidi-chars: Warning Options. (line 1918) 66729* Wbidi-chars=: Warning Options. (line 1918) 66730* Wbool-compare: Warning Options. (line 1940) 66731* Wbool-operation: Warning Options. (line 1949) 66732* Wbuiltin-declaration-mismatch: Warning Options. (line 2875) 66733* Wbuiltin-macro-redefined: Warning Options. (line 2896) 66734* Wc++-compat: Warning Options. (line 2449) 66735* Wc++11-compat: Warning Options. (line 2454) 66736* Wc++11-extensions: Warning Options. (line 2472) 66737* Wc++14-compat: Warning Options. (line 2460) 66738* Wc++14-extensions: Warning Options. (line 2477) 66739* Wc++17-compat: Warning Options. (line 2464) 66740* Wc++17-extensions: Warning Options. (line 2482) 66741* Wc++20-compat: Warning Options. (line 2468) 66742* Wc++20-extensions: Warning Options. (line 2487) 66743* Wc++23-extensions: Warning Options. (line 2492) 66744* Wc11-c2x-compat: Warning Options. (line 2441) 66745* Wc90-c99-compat: Warning Options. (line 2426) 66746* Wc99-c11-compat: Warning Options. (line 2433) 66747* Wcast-align: Warning Options. (line 2513) 66748* Wcast-align=strict: Warning Options. (line 2519) 66749* Wcast-function-type: Warning Options. (line 2524) 66750* Wcast-qual: Warning Options. (line 2497) 66751* Wcatch-value: C++ Dialect Options. 66752 (line 1227) 66753* Wchar-subscripts: Warning Options. (line 377) 66754* Wclass-conversion: C++ Dialect Options. 66755 (line 1158) 66756* Wclass-memaccess: C++ Dialect Options. 66757 (line 785) 66758* Wclobbered: Warning Options. (line 2550) 66759* Wcomma-subscript: C++ Dialect Options. 66760 (line 595) 66761* Wcomment: Warning Options. (line 2363) 66762* Wcomments: Warning Options. (line 2363) 66763* Wconditionally-supported: C++ Dialect Options. 66764 (line 1235) 66765* Wconversion: Warning Options. (line 2554) 66766* Wconversion-null: C++ Dialect Options. 66767 (line 1369) 66768* Wcoverage-invalid-line-number: Warning Options. (line 396) 66769* Wcoverage-mismatch: Warning Options. (line 382) 66770* Wcpp: Warning Options. (line 406) 66771* Wctad-maybe-unsupported: C++ Dialect Options. 66772 (line 616) 66773* Wctor-dtor-privacy: C++ Dialect Options. 66774 (line 632) 66775* Wdangling-else: Warning Options. (line 2575) 66776* Wdangling-pointer: Warning Options. (line 2610) 66777* Wdate-time: Warning Options. (line 2657) 66778* Wdeclaration-after-statement: Warning Options. (line 2169) 66779* Wdelete-incomplete: C++ Dialect Options. 66780 (line 1238) 66781* Wdelete-non-virtual-dtor: C++ Dialect Options. 66782 (line 639) 66783* Wdeprecated: Warning Options. (line 3079) 66784* Wdeprecated-copy: C++ Dialect Options. 66785 (line 646) 66786* Wdeprecated-declarations: Warning Options. (line 3083) 66787* Wdeprecated-enum-enum-conversion: C++ Dialect Options. 66788 (line 654) 66789* Wdeprecated-enum-float-conversion: C++ Dialect Options. 66790 (line 668) 66791* Wdesignated-init: Warning Options. (line 3410) 66792* Wdisabled-optimization: Warning Options. (line 3361) 66793* Wdiscarded-array-qualifiers: Warning Options. (line 1988) 66794* Wdiscarded-qualifiers: Warning Options. (line 1982) 66795* Wdiv-by-zero: Warning Options. (line 2029) 66796* Wdouble-promotion: Warning Options. (line 409) 66797* Wduplicate-decl-specifier: Warning Options. (line 427) 66798* Wduplicated-branches: Warning Options. (line 1959) 66799* Wduplicated-cond: Warning Options. (line 1970) 66800* weak_reference_mismatches: Darwin Options. (line 196) 66801* Weffc++: C++ Dialect Options. 66802 (line 951) 66803* Wempty-body: Warning Options. (line 2662) 66804* Wendif-labels: Warning Options. (line 2407) 66805* Wendif-labels <1>: Warning Options. (line 2666) 66806* Wenum-compare: Warning Options. (line 2669) 66807* Wenum-conversion: Warning Options. (line 2675) 66808* Werror: Warning Options. (line 28) 66809* Werror=: Warning Options. (line 31) 66810* Wexceptions: C++ Dialect Options. 66811 (line 979) 66812* Wexpansion-to-defined: Warning Options. (line 2382) 66813* Wextra: Warning Options. (line 222) 66814* Wextra <1>: Warning Options. (line 2948) 66815* Wextra <2>: Warning Options. (line 3105) 66816* Wextra-semi: C++ Dialect Options. 66817 (line 1243) 66818* Wfatal-errors: Warning Options. (line 48) 66819* Wfloat-conversion: Warning Options. (line 2703) 66820* Wfloat-equal: Warning Options. (line 2069) 66821* Wformat: Warning Options. (line 432) 66822* Wformat <1>: Warning Options. (line 457) 66823* Wformat <2>: Warning Options. (line 1676) 66824* Wformat <3>: Common Function Attributes. 66825 (line 411) 66826* Wformat-contains-nul: Warning Options. (line 471) 66827* Wformat-extra-args: Warning Options. (line 475) 66828* Wformat-nonliteral: Warning Options. (line 568) 66829* Wformat-nonliteral <1>: Common Function Attributes. 66830 (line 476) 66831* Wformat-overflow: Warning Options. (line 489) 66832* Wformat-overflow <1>: Warning Options. (line 500) 66833* Wformat-security: Warning Options. (line 573) 66834* Wformat-signedness: Warning Options. (line 584) 66835* Wformat-truncation: Warning Options. (line 589) 66836* Wformat-truncation <1>: Warning Options. (line 601) 66837* Wformat-y2k: Warning Options. (line 612) 66838* Wformat-zero-length: Warning Options. (line 564) 66839* Wformat=: Warning Options. (line 432) 66840* Wformat=1: Warning Options. (line 457) 66841* Wformat=2: Warning Options. (line 466) 66842* Wframe-address: Warning Options. (line 1976) 66843* Wframe-larger-than=: Warning Options. (line 2243) 66844* Wfree-nonheap-object: Warning Options. (line 2260) 66845* whatsloaded: Darwin Options. (line 196) 66846* whyload: Darwin Options. (line 196) 66847* Wif-not-aligned: Warning Options. (line 791) 66848* Wignored-attributes: Warning Options. (line 806) 66849* Wignored-qualifiers: Warning Options. (line 795) 66850* Wimplicit: Warning Options. (line 669) 66851* Wimplicit-fallthrough: Warning Options. (line 673) 66852* Wimplicit-fallthrough=: Warning Options. (line 678) 66853* Wimplicit-function-declaration: Warning Options. (line 663) 66854* Wimplicit-int: Warning Options. (line 658) 66855* Winaccessible-base: C++ Dialect Options. 66856 (line 1247) 66857* Wincompatible-pointer-types: Warning Options. (line 1994) 66858* Winfinite-recursion: Warning Options. (line 637) 66859* Winherited-variadic-ctor: C++ Dialect Options. 66860 (line 1258) 66861* Winit-list-lifetime: C++ Dialect Options. 66862 (line 682) 66863* Winit-self: Warning Options. (line 643) 66864* Winline: Warning Options. (line 3198) 66865* Winline <1>: Inline. (line 60) 66866* Wint-conversion: Warning Options. (line 2000) 66867* Wint-in-bool-context: Warning Options. (line 3246) 66868* Wint-to-pointer-cast: Warning Options. (line 3254) 66869* Winterference-size: Warning Options. (line 3211) 66870* Winvalid-imported-macros: C++ Dialect Options. 66871 (line 717) 66872* Winvalid-memory-model: Warning Options. (line 1320) 66873* Winvalid-offsetof: C++ Dialect Options. 66874 (line 1263) 66875* Winvalid-pch: Warning Options. (line 3263) 66876* Wjump-misses-init: Warning Options. (line 2680) 66877* Wl: Link Options. (line 338) 66878* Wlarger-than-BYTE-SIZE: Warning Options. (line 2228) 66879* Wlarger-than=: Warning Options. (line 2228) 66880* Wliteral-suffix: C++ Dialect Options. 66881 (line 723) 66882* Wlogical-not-parentheses: Warning Options. (line 2828) 66883* Wlogical-op: Warning Options. (line 2820) 66884* Wlong-long: Warning Options. (line 3267) 66885* Wlto-type-mismatch: Warning Options. (line 3404) 66886* Wmain: Warning Options. (line 813) 66887* Wmaybe-uninitialized: Warning Options. (line 1337) 66888* Wmemset-elt-size: Warning Options. (line 2762) 66889* Wmemset-transposed-args: Warning Options. (line 2770) 66890* Wmisleading-indentation: Warning Options. (line 820) 66891* Wmismatched-dealloc: Warning Options. (line 930) 66892* Wmismatched-new-delete: C++ Dialect Options. 66893 (line 1042) 66894* Wmismatched-tags: C++ Dialect Options. 66895 (line 1073) 66896* Wmissing-attributes: Warning Options. (line 854) 66897* Wmissing-braces: Warning Options. (line 898) 66898* Wmissing-declarations: Warning Options. (line 2938) 66899* Wmissing-field-initializers: Warning Options. (line 2948) 66900* Wmissing-format-attribute: Warning Options. (line 1676) 66901* Wmissing-include-dirs: Warning Options. (line 908) 66902* Wmissing-noreturn: Warning Options. (line 1662) 66903* Wmissing-parameter-type: Warning Options. (line 2920) 66904* Wmissing-profile: Warning Options. (line 914) 66905* Wmissing-prototypes: Warning Options. (line 2928) 66906* Wmissing-requires: Warning Options. (line 2978) 66907* Wmissing-template-keyword: Warning Options. (line 2995) 66908* Wmisspelled-isr: AVR Options. (line 313) 66909* Wmultichar: Warning Options. (line 3023) 66910* Wmultiple-inheritance: C++ Dialect Options. 66911 (line 1098) 66912* Wmultistatement-macros: Warning Options. (line 961) 66913* Wnamespaces: C++ Dialect Options. 66914 (line 1121) 66915* Wnarrowing: C++ Dialect Options. 66916 (line 749) 66917* Wnested-externs: Warning Options. (line 3195) 66918* Wno-abi: Warning Options. (line 267) 66919* Wno-absolute-value: Warning Options. (line 2352) 66920* Wno-addr-space-convert: AVR Options. (line 308) 66921* Wno-address: Warning Options. (line 2781) 66922* Wno-address-of-packed-member: Warning Options. (line 2815) 66923* Wno-aggregate-return: Warning Options. (line 2843) 66924* Wno-aggressive-loop-optimizations: Warning Options. (line 2848) 66925* Wno-aligned-new: C++ Dialect Options. 66926 (line 1180) 66927* Wno-all: Warning Options. (line 140) 66928* Wno-alloc-size-larger-than: Warning Options. (line 1715) 66929* Wno-alloc-size-larger-than <1>: Warning Options. (line 1726) 66930* Wno-alloc-zero: Warning Options. (line 1705) 66931* Wno-alloca: Warning Options. (line 1730) 66932* Wno-alloca-larger-than: Warning Options. (line 1733) 66933* Wno-alloca-larger-than <1>: Warning Options. (line 1794) 66934* Wno-analyzer-double-fclose: Static Analyzer Options. 66935 (line 59) 66936* Wno-analyzer-double-free: Static Analyzer Options. 66937 (line 66) 66938* Wno-analyzer-exposure-through-output-file: Static Analyzer Options. 66939 (line 74) 66940* Wno-analyzer-file-leak: Static Analyzer Options. 66941 (line 82) 66942* Wno-analyzer-free-of-non-heap: Static Analyzer Options. 66943 (line 89) 66944* Wno-analyzer-malloc-leak: Static Analyzer Options. 66945 (line 97) 66946* Wno-analyzer-mismatching-deallocation: Static Analyzer Options. 66947 (line 105) 66948* Wno-analyzer-null-argument: Static Analyzer Options. 66949 (line 131) 66950* Wno-analyzer-null-dereference: Static Analyzer Options. 66951 (line 139) 66952* Wno-analyzer-possible-null-argument: Static Analyzer Options. 66953 (line 116) 66954* Wno-analyzer-possible-null-dereference: Static Analyzer Options. 66955 (line 124) 66956* Wno-analyzer-shift-count-negative: Static Analyzer Options. 66957 (line 146) 66958* Wno-analyzer-shift-count-overflow: Static Analyzer Options. 66959 (line 158) 66960* Wno-analyzer-stale-setjmp-buffer: Static Analyzer Options. 66961 (line 171) 66962* Wno-analyzer-tainted-allocation-size: Static Analyzer Options. 66963 (line 185) 66964* Wno-analyzer-tainted-array-index: Static Analyzer Options. 66965 (line 199) 66966* Wno-analyzer-tainted-divisor: Static Analyzer Options. 66967 (line 212) 66968* Wno-analyzer-tainted-offset: Static Analyzer Options. 66969 (line 222) 66970* Wno-analyzer-tainted-size: Static Analyzer Options. 66971 (line 235) 66972* Wno-analyzer-too-complex: Static Analyzer Options. 66973 (line 49) 66974* Wno-analyzer-unsafe-call-within-signal-handler: Static Analyzer Options. 66975 (line 245) 66976* Wno-analyzer-use-after-free: Static Analyzer Options. 66977 (line 253) 66978* Wno-analyzer-use-of-pointer-in-stale-stack-frame: Static Analyzer Options. 66979 (line 261) 66980* Wno-analyzer-use-of-uninitialized-value: Static Analyzer Options. 66981 (line 288) 66982* Wno-analyzer-write-to-const: Static Analyzer Options. 66983 (line 268) 66984* Wno-analyzer-write-to-string-literal: Static Analyzer Options. 66985 (line 278) 66986* Wno-arith-conversion: Warning Options. (line 1798) 66987* Wno-array-bounds: Warning Options. (line 1811) 66988* Wno-array-compare: Warning Options. (line 1830) 66989* Wno-array-parameter: Warning Options. (line 1842) 66990* Wno-assign-intercept: Objective-C and Objective-C++ Dialect Options. 66991 (line 170) 66992* Wno-attribute-alias: Warning Options. (line 1886) 66993* Wno-attribute-warning: Warning Options. (line 3072) 66994* Wno-attributes: Warning Options. (line 2853) 66995* Wno-bad-function-cast: Warning Options. (line 2421) 66996* Wno-bidi-chars: Warning Options. (line 1918) 66997* Wno-bool-compare: Warning Options. (line 1940) 66998* Wno-bool-operation: Warning Options. (line 1949) 66999* Wno-builtin-declaration-mismatch: Warning Options. (line 2875) 67000* Wno-builtin-macro-redefined: Warning Options. (line 2896) 67001* Wno-c++-compat: Warning Options. (line 2449) 67002* Wno-c++11-compat: Warning Options. (line 2454) 67003* Wno-c++11-extensions: Warning Options. (line 2472) 67004* Wno-c++14-compat: Warning Options. (line 2460) 67005* Wno-c++14-extensions: Warning Options. (line 2477) 67006* Wno-c++17-compat: Warning Options. (line 2464) 67007* Wno-c++17-extensions: Warning Options. (line 2482) 67008* Wno-c++20-compat: Warning Options. (line 2468) 67009* Wno-c++20-extensions: Warning Options. (line 2487) 67010* Wno-c++23-extensions: Warning Options. (line 2492) 67011* Wno-c11-c2x-compat: Warning Options. (line 2441) 67012* Wno-c90-c99-compat: Warning Options. (line 2426) 67013* Wno-c99-c11-compat: Warning Options. (line 2433) 67014* Wno-cast-align: Warning Options. (line 2513) 67015* Wno-cast-function-type: Warning Options. (line 2524) 67016* Wno-cast-qual: Warning Options. (line 2497) 67017* Wno-catch-value: C++ Dialect Options. 67018 (line 1227) 67019* Wno-char-subscripts: Warning Options. (line 377) 67020* Wno-class-conversion: C++ Dialect Options. 67021 (line 1158) 67022* Wno-class-memaccess: C++ Dialect Options. 67023 (line 785) 67024* Wno-clobbered: Warning Options. (line 2550) 67025* Wno-comma-subscript: C++ Dialect Options. 67026 (line 595) 67027* Wno-conditionally-supported: C++ Dialect Options. 67028 (line 1235) 67029* Wno-conversion: Warning Options. (line 2554) 67030* Wno-conversion-null: C++ Dialect Options. 67031 (line 1369) 67032* Wno-coverage-invalid-line-number: Warning Options. (line 396) 67033* Wno-coverage-mismatch: Warning Options. (line 382) 67034* Wno-cpp: Warning Options. (line 406) 67035* Wno-ctad-maybe-unsupported: C++ Dialect Options. 67036 (line 616) 67037* Wno-ctor-dtor-privacy: C++ Dialect Options. 67038 (line 632) 67039* Wno-dangling-else: Warning Options. (line 2575) 67040* Wno-dangling-pointer: Warning Options. (line 2610) 67041* Wno-date-time: Warning Options. (line 2657) 67042* Wno-declaration-after-statement: Warning Options. (line 2169) 67043* Wno-delete-incomplete: C++ Dialect Options. 67044 (line 1238) 67045* Wno-delete-non-virtual-dtor: C++ Dialect Options. 67046 (line 639) 67047* Wno-deprecated: Warning Options. (line 3079) 67048* Wno-deprecated-copy: C++ Dialect Options. 67049 (line 646) 67050* Wno-deprecated-declarations: Warning Options. (line 3083) 67051* Wno-deprecated-enum-enum-conversion: C++ Dialect Options. 67052 (line 654) 67053* Wno-deprecated-enum-float-conversion: C++ Dialect Options. 67054 (line 668) 67055* Wno-designated-init: Warning Options. (line 3410) 67056* Wno-disabled-optimization: Warning Options. (line 3361) 67057* Wno-discarded-array-qualifiers: Warning Options. (line 1988) 67058* Wno-discarded-qualifiers: Warning Options. (line 1982) 67059* Wno-div-by-zero: Warning Options. (line 2029) 67060* Wno-double-promotion: Warning Options. (line 409) 67061* Wno-duplicate-decl-specifier: Warning Options. (line 427) 67062* Wno-duplicated-branches: Warning Options. (line 1959) 67063* Wno-duplicated-cond: Warning Options. (line 1970) 67064* Wno-effc++: C++ Dialect Options. 67065 (line 951) 67066* Wno-empty-body: Warning Options. (line 2662) 67067* Wno-endif-labels: Warning Options. (line 2407) 67068* Wno-endif-labels <1>: Warning Options. (line 2666) 67069* Wno-enum-compare: Warning Options. (line 2669) 67070* Wno-enum-conversion: Warning Options. (line 2675) 67071* Wno-error: Warning Options. (line 28) 67072* Wno-error=: Warning Options. (line 31) 67073* Wno-exceptions: C++ Dialect Options. 67074 (line 979) 67075* Wno-extra: Warning Options. (line 222) 67076* Wno-extra <1>: Warning Options. (line 2948) 67077* Wno-extra <2>: Warning Options. (line 3105) 67078* Wno-extra-semi: C++ Dialect Options. 67079 (line 1243) 67080* Wno-fatal-errors: Warning Options. (line 48) 67081* Wno-float-conversion: Warning Options. (line 2703) 67082* Wno-float-equal: Warning Options. (line 2069) 67083* Wno-format: Warning Options. (line 432) 67084* Wno-format <1>: Warning Options. (line 1676) 67085* Wno-format-contains-nul: Warning Options. (line 471) 67086* Wno-format-extra-args: Warning Options. (line 475) 67087* Wno-format-nonliteral: Warning Options. (line 568) 67088* Wno-format-overflow: Warning Options. (line 489) 67089* Wno-format-overflow <1>: Warning Options. (line 500) 67090* Wno-format-security: Warning Options. (line 573) 67091* Wno-format-signedness: Warning Options. (line 584) 67092* Wno-format-truncation: Warning Options. (line 589) 67093* Wno-format-truncation <1>: Warning Options. (line 601) 67094* Wno-format-y2k: Warning Options. (line 612) 67095* Wno-format-zero-length: Warning Options. (line 564) 67096* Wno-frame-address: Warning Options. (line 1976) 67097* Wno-frame-larger-than: Warning Options. (line 2243) 67098* Wno-frame-larger-than <1>: Warning Options. (line 2256) 67099* Wno-free-nonheap-object: Warning Options. (line 2260) 67100* Wno-if-not-aligned: Warning Options. (line 791) 67101* Wno-ignored-attributes: Warning Options. (line 806) 67102* Wno-ignored-qualifiers: Warning Options. (line 795) 67103* Wno-implicit: Warning Options. (line 669) 67104* Wno-implicit-fallthrough: Warning Options. (line 673) 67105* Wno-implicit-function-declaration: Warning Options. (line 663) 67106* Wno-implicit-int: Warning Options. (line 658) 67107* Wno-inaccessible-base: C++ Dialect Options. 67108 (line 1247) 67109* Wno-incompatible-pointer-types: Warning Options. (line 1994) 67110* Wno-infinite-recursion: Warning Options. (line 637) 67111* Wno-inherited-variadic-ctor: C++ Dialect Options. 67112 (line 1258) 67113* Wno-init-list-lifetime: C++ Dialect Options. 67114 (line 682) 67115* Wno-init-self: Warning Options. (line 643) 67116* Wno-inline: Warning Options. (line 3198) 67117* Wno-int-conversion: Warning Options. (line 2000) 67118* Wno-int-in-bool-context: Warning Options. (line 3246) 67119* Wno-int-to-pointer-cast: Warning Options. (line 3254) 67120* Wno-invalid-imported-macros: C++ Dialect Options. 67121 (line 717) 67122* Wno-invalid-memory-model: Warning Options. (line 1320) 67123* Wno-invalid-offsetof: C++ Dialect Options. 67124 (line 1263) 67125* Wno-invalid-pch: Warning Options. (line 3263) 67126* Wno-jump-misses-init: Warning Options. (line 2680) 67127* Wno-larger-than: Warning Options. (line 2239) 67128* Wno-literal-suffix: C++ Dialect Options. 67129 (line 723) 67130* Wno-logical-not-parentheses: Warning Options. (line 2828) 67131* Wno-logical-op: Warning Options. (line 2820) 67132* Wno-long-long: Warning Options. (line 3267) 67133* Wno-lto-type-mismatch: Warning Options. (line 3404) 67134* Wno-main: Warning Options. (line 813) 67135* Wno-maybe-uninitialized: Warning Options. (line 1337) 67136* Wno-memset-elt-size: Warning Options. (line 2762) 67137* Wno-memset-transposed-args: Warning Options. (line 2770) 67138* Wno-misleading-indentation: Warning Options. (line 820) 67139* Wno-mismatched-dealloc: Warning Options. (line 930) 67140* Wno-mismatched-new-delete: C++ Dialect Options. 67141 (line 1042) 67142* Wno-mismatched-tags: C++ Dialect Options. 67143 (line 1073) 67144* Wno-missing-attributes: Warning Options. (line 854) 67145* Wno-missing-braces: Warning Options. (line 898) 67146* Wno-missing-declarations: Warning Options. (line 2938) 67147* Wno-missing-field-initializers: Warning Options. (line 2948) 67148* Wno-missing-format-attribute: Warning Options. (line 1676) 67149* Wno-missing-include-dirs: Warning Options. (line 908) 67150* Wno-missing-noreturn: Warning Options. (line 1662) 67151* Wno-missing-parameter-type: Warning Options. (line 2920) 67152* Wno-missing-profile: Warning Options. (line 914) 67153* Wno-missing-prototypes: Warning Options. (line 2928) 67154* Wno-missing-requires: Warning Options. (line 2978) 67155* Wno-missing-template-keyword: Warning Options. (line 2995) 67156* Wno-misspelled-isr: AVR Options. (line 313) 67157* Wno-multichar: Warning Options. (line 3023) 67158* Wno-multiple-inheritance: C++ Dialect Options. 67159 (line 1098) 67160* Wno-multistatement-macros: Warning Options. (line 961) 67161* Wno-namespaces: C++ Dialect Options. 67162 (line 1121) 67163* Wno-narrowing: C++ Dialect Options. 67164 (line 749) 67165* Wno-nested-externs: Warning Options. (line 3195) 67166* Wno-noexcept: C++ Dialect Options. 67167 (line 765) 67168* Wno-noexcept-type: C++ Dialect Options. 67169 (line 771) 67170* Wno-non-template-friend: C++ Dialect Options. 67171 (line 991) 67172* Wno-non-virtual-dtor: C++ Dialect Options. 67173 (line 805) 67174* Wno-nonnull: Warning Options. (line 616) 67175* Wno-nonnull-compare: Warning Options. (line 623) 67176* Wno-normalized: Warning Options. (line 3029) 67177* Wno-null-dereference: Warning Options. (line 630) 67178* Wno-odr: Warning Options. (line 3092) 67179* Wno-old-style-cast: C++ Dialect Options. 67180 (line 1000) 67181* Wno-old-style-declaration: Warning Options. (line 2907) 67182* Wno-old-style-definition: Warning Options. (line 2913) 67183* Wno-openacc-parallelism: Warning Options. (line 3096) 67184* Wno-openmp-simd: Warning Options. (line 3100) 67185* Wno-overflow: Warning Options. (line 3089) 67186* Wno-overlength-strings: Warning Options. (line 3381) 67187* Wno-overloaded-virtual: C++ Dialect Options. 67188 (line 1006) 67189* Wno-override-init: Warning Options. (line 3105) 67190* Wno-override-init-side-effects: Warning Options. (line 3113) 67191* Wno-packed: Warning Options. (line 3118) 67192* Wno-packed-bitfield-compat: Warning Options. (line 3135) 67193* Wno-packed-not-aligned: Warning Options. (line 3152) 67194* Wno-padded: Warning Options. (line 3165) 67195* Wno-parentheses: Warning Options. (line 981) 67196* Wno-pedantic: Warning Options. (line 88) 67197* Wno-pedantic-ms-format: Warning Options. (line 2311) 67198* Wno-pessimizing-move: C++ Dialect Options. 67199 (line 834) 67200* Wno-placement-new: C++ Dialect Options. 67201 (line 1191) 67202* Wno-pmf-conversions: C++ Dialect Options. 67203 (line 1025) 67204* Wno-pmf-conversions <1>: Bound member functions. 67205 (line 35) 67206* Wno-pointer-arith: Warning Options. (line 2317) 67207* Wno-pointer-compare: Warning Options. (line 2324) 67208* Wno-pointer-sign: Warning Options. (line 3370) 67209* Wno-pointer-to-int-cast: Warning Options. (line 3259) 67210* Wno-pragmas: Warning Options. (line 1397) 67211* Wno-prio-ctor-dtor: Warning Options. (line 1402) 67212* Wno-property-assign-default: Objective-C and Objective-C++ Dialect Options. 67213 (line 174) 67214* Wno-protocol: Objective-C and Objective-C++ Dialect Options. 67215 (line 178) 67216* Wno-range-loop-construct: C++ Dialect Options. 67217 (line 895) 67218* Wno-redundant-decls: Warning Options. (line 3172) 67219* Wno-redundant-move: C++ Dialect Options. 67220 (line 856) 67221* Wno-redundant-tags: C++ Dialect Options. 67222 (line 924) 67223* Wno-register: C++ Dialect Options. 67224 (line 813) 67225* Wno-reorder: C++ Dialect Options. 67226 (line 820) 67227* Wno-restrict: Warning Options. (line 3176) 67228* Wno-return-local-addr: Warning Options. (line 1061) 67229* Wno-return-type: Warning Options. (line 1065) 67230* Wno-scalar-storage-order: Warning Options. (line 2709) 67231* Wno-selector: Objective-C and Objective-C++ Dialect Options. 67232 (line 195) 67233* Wno-sequence-point: Warning Options. (line 1008) 67234* Wno-shadow: Warning Options. (line 2175) 67235* Wno-shadow-ivar: Warning Options. (line 2186) 67236* Wno-shift-count-negative: Warning Options. (line 1086) 67237* Wno-shift-count-overflow: Warning Options. (line 1090) 67238* Wno-shift-negative-value: Warning Options. (line 1094) 67239* Wno-shift-overflow: Warning Options. (line 1099) 67240* Wno-sign-compare: Warning Options. (line 2691) 67241* Wno-sign-conversion: Warning Options. (line 2697) 67242* Wno-sign-promo: C++ Dialect Options. 67243 (line 1029) 67244* Wno-sized-deallocation: C++ Dialect Options. 67245 (line 1275) 67246* Wno-sizeof-array-argument: Warning Options. (line 2757) 67247* Wno-sizeof-array-div: Warning Options. (line 2713) 67248* Wno-sizeof-pointer-div: Warning Options. (line 2727) 67249* Wno-sizeof-pointer-memaccess: Warning Options. (line 2735) 67250* Wno-stack-protector: Warning Options. (line 3376) 67251* Wno-stack-usage: Warning Options. (line 2277) 67252* Wno-stack-usage <1>: Warning Options. (line 2301) 67253* Wno-strict-aliasing: Warning Options. (line 1410) 67254* Wno-strict-null-sentinel: C++ Dialect Options. 67255 (line 984) 67256* Wno-strict-overflow: Warning Options. (line 1449) 67257* Wno-strict-prototypes: Warning Options. (line 2901) 67258* Wno-strict-selector-match: Objective-C and Objective-C++ Dialect Options. 67259 (line 207) 67260* Wno-string-compare: Warning Options. (line 1497) 67261* Wno-stringop-overflow: Warning Options. (line 1519) 67262* Wno-stringop-overflow <1>: Warning Options. (line 1558) 67263* Wno-stringop-overread: Warning Options. (line 1596) 67264* Wno-stringop-truncation: Warning Options. (line 1603) 67265* Wno-subobject-linkage: C++ Dialect Options. 67266 (line 938) 67267* Wno-suggest-attribute=: Warning Options. (line 1654) 67268* Wno-suggest-attribute=cold: Warning Options. (line 1697) 67269* Wno-suggest-attribute=const: Warning Options. (line 1662) 67270* Wno-suggest-attribute=format: Warning Options. (line 1676) 67271* Wno-suggest-attribute=malloc: Warning Options. (line 1662) 67272* Wno-suggest-attribute=noreturn: Warning Options. (line 1662) 67273* Wno-suggest-attribute=pure: Warning Options. (line 1662) 67274* Wno-suggest-final-methods: C++ Dialect Options. 67275 (line 1295) 67276* Wno-suggest-final-types: C++ Dialect Options. 67277 (line 1286) 67278* Wno-suggest-override: C++ Dialect Options. 67279 (line 1305) 67280* Wno-switch: Warning Options. (line 1115) 67281* Wno-switch-bool: Warning Options. (line 1135) 67282* Wno-switch-default: Warning Options. (line 1123) 67283* Wno-switch-enum: Warning Options. (line 1126) 67284* Wno-switch-outside-range: Warning Options. (line 1146) 67285* Wno-switch-unreachable: Warning Options. (line 1151) 67286* Wno-sync-nand: Warning Options. (line 1175) 67287* Wno-system-headers: Warning Options. (line 2034) 67288* Wno-tautological-compare: Warning Options. (line 2045) 67289* Wno-templates: C++ Dialect Options. 67290 (line 1035) 67291* Wno-terminate: C++ Dialect Options. 67292 (line 1128) 67293* Wno-traditional: Warning Options. (line 2084) 67294* Wno-traditional-conversion: Warning Options. (line 2161) 67295* Wno-trampolines: Warning Options. (line 2059) 67296* Wno-trivial-auto-var-init: Warning Options. (line 1180) 67297* Wno-tsan: Warning Options. (line 2337) 67298* Wno-type-limits: Warning Options. (line 2345) 67299* Wno-undeclared-selector: Objective-C and Objective-C++ Dialect Options. 67300 (line 215) 67301* Wno-undef: Warning Options. (line 2378) 67302* Wno-uninitialized: Warning Options. (line 1282) 67303* Wno-unknown-pragmas: Warning Options. (line 1390) 67304* Wno-unsafe-loop-optimizations: Warning Options. (line 2305) 67305* Wno-unsuffixed-float-constants: Warning Options. (line 3396) 67306* Wno-unused: Warning Options. (line 1275) 67307* Wno-unused-but-set-parameter: Warning Options. (line 1186) 67308* Wno-unused-but-set-variable: Warning Options. (line 1195) 67309* Wno-unused-const-variable: Warning Options. (line 1242) 67310* Wno-unused-function: Warning Options. (line 1205) 67311* Wno-unused-label: Warning Options. (line 1210) 67312* Wno-unused-local-typedefs: Warning Options. (line 1217) 67313* Wno-unused-parameter: Warning Options. (line 1221) 67314* Wno-unused-result: Warning Options. (line 1228) 67315* Wno-unused-value: Warning Options. (line 1265) 67316* Wno-unused-variable: Warning Options. (line 1233) 67317* Wno-use-after-free: C++ Dialect Options. 67318 (line 1310) 67319* Wno-useless-cast: C++ Dialect Options. 67320 (line 1366) 67321* Wno-varargs: Warning Options. (line 3278) 67322* Wno-variadic-macros: Warning Options. (line 3272) 67323* Wno-vector-operation-performance: Warning Options. (line 3283) 67324* Wno-vexing-parse: C++ Dialect Options. 67325 (line 1132) 67326* Wno-virtual-inheritance: C++ Dialect Options. 67327 (line 1105) 67328* Wno-virtual-move-assign: C++ Dialect Options. 67329 (line 1112) 67330* Wno-vla: Warning Options. (line 3293) 67331* Wno-vla-larger-than: Warning Options. (line 3297) 67332* Wno-vla-larger-than <1>: Warning Options. (line 3314) 67333* Wno-vla-parameter: Warning Options. (line 3318) 67334* Wno-volatile: C++ Dialect Options. 67335 (line 1163) 67336* Wno-volatile-register-var: Warning Options. (line 3355) 67337* Wno-write-strings: Warning Options. (line 2537) 67338* Wno-zero-as-null-pointer-constant: C++ Dialect Options. 67339 (line 1176) 67340* Wnoexcept: C++ Dialect Options. 67341 (line 765) 67342* Wnoexcept-type: C++ Dialect Options. 67343 (line 771) 67344* Wnon-template-friend: C++ Dialect Options. 67345 (line 991) 67346* Wnon-virtual-dtor: C++ Dialect Options. 67347 (line 805) 67348* Wnonnull: Warning Options. (line 616) 67349* Wnonnull-compare: Warning Options. (line 623) 67350* Wnormalized: Warning Options. (line 3029) 67351* Wnormalized=: Warning Options. (line 3029) 67352* Wnull-dereference: Warning Options. (line 630) 67353* Wobjc-root-class: Objective-C and Objective-C++ Dialect Options. 67354 (line 188) 67355* Wodr: Warning Options. (line 3092) 67356* Wold-style-cast: C++ Dialect Options. 67357 (line 1000) 67358* Wold-style-declaration: Warning Options. (line 2907) 67359* Wold-style-definition: Warning Options. (line 2913) 67360* Wopenacc-parallelism: Warning Options. (line 3096) 67361* Wopenmp-simd: Warning Options. (line 3100) 67362* Woverflow: Warning Options. (line 3089) 67363* Woverlength-strings: Warning Options. (line 3381) 67364* Woverloaded-virtual: C++ Dialect Options. 67365 (line 1006) 67366* Woverride-init: Warning Options. (line 3105) 67367* Woverride-init-side-effects: Warning Options. (line 3113) 67368* Wp: Preprocessor Options. 67369 (line 465) 67370* Wpacked: Warning Options. (line 3118) 67371* Wpacked-bitfield-compat: Warning Options. (line 3135) 67372* Wpacked-not-aligned: Warning Options. (line 3152) 67373* Wpadded: Warning Options. (line 3165) 67374* Wparentheses: Warning Options. (line 981) 67375* Wpedantic: Warning Options. (line 88) 67376* Wpedantic-ms-format: Warning Options. (line 2311) 67377* Wpessimizing-move: C++ Dialect Options. 67378 (line 834) 67379* Wplacement-new: C++ Dialect Options. 67380 (line 1191) 67381* Wpmf-conversions: C++ Dialect Options. 67382 (line 1025) 67383* Wpointer-arith: Warning Options. (line 2317) 67384* Wpointer-arith <1>: Pointer Arith. (line 13) 67385* Wpointer-compare: Warning Options. (line 2324) 67386* Wpointer-sign: Warning Options. (line 3370) 67387* Wpointer-to-int-cast: Warning Options. (line 3259) 67388* Wpragmas: Warning Options. (line 1397) 67389* Wprio-ctor-dtor: Warning Options. (line 1402) 67390* Wproperty-assign-default: Objective-C and Objective-C++ Dialect Options. 67391 (line 174) 67392* Wprotocol: Objective-C and Objective-C++ Dialect Options. 67393 (line 178) 67394* Wrange-loop-construct: C++ Dialect Options. 67395 (line 895) 67396* wrapper: Overall Options. (line 620) 67397* Wredundant-decls: Warning Options. (line 3172) 67398* Wredundant-move: C++ Dialect Options. 67399 (line 856) 67400* Wredundant-tags: C++ Dialect Options. 67401 (line 924) 67402* Wregister: C++ Dialect Options. 67403 (line 813) 67404* Wreorder: C++ Dialect Options. 67405 (line 820) 67406* Wrestrict: Warning Options. (line 3176) 67407* Wreturn-local-addr: Warning Options. (line 1061) 67408* Wreturn-type: Warning Options. (line 1065) 67409* Wscalar-storage-order: Warning Options. (line 2709) 67410* Wselector: Objective-C and Objective-C++ Dialect Options. 67411 (line 195) 67412* Wsequence-point: Warning Options. (line 1008) 67413* Wshadow: Warning Options. (line 2175) 67414* Wshadow-ivar: Warning Options. (line 2186) 67415* Wshadow=compatible-local: Warning Options. (line 2197) 67416* Wshadow=global: Warning Options. (line 2190) 67417* Wshadow=local: Warning Options. (line 2193) 67418* Wshift-count-negative: Warning Options. (line 1086) 67419* Wshift-count-overflow: Warning Options. (line 1090) 67420* Wshift-negative-value: Warning Options. (line 1094) 67421* Wshift-overflow: Warning Options. (line 1099) 67422* Wsign-compare: Warning Options. (line 2691) 67423* Wsign-conversion: Warning Options. (line 2697) 67424* Wsign-promo: C++ Dialect Options. 67425 (line 1029) 67426* Wsized-deallocation: C++ Dialect Options. 67427 (line 1275) 67428* Wsizeof-array-argument: Warning Options. (line 2757) 67429* Wsizeof-array-div: Warning Options. (line 2713) 67430* Wsizeof-pointer-div: Warning Options. (line 2727) 67431* Wsizeof-pointer-memaccess: Warning Options. (line 2735) 67432* Wstack-protector: Warning Options. (line 3376) 67433* Wstack-usage: Warning Options. (line 2277) 67434* Wstrict-aliasing: Warning Options. (line 1410) 67435* Wstrict-aliasing=n: Warning Options. (line 1417) 67436* Wstrict-null-sentinel: C++ Dialect Options. 67437 (line 984) 67438* Wstrict-overflow: Warning Options. (line 1449) 67439* Wstrict-prototypes: Warning Options. (line 2901) 67440* Wstrict-selector-match: Objective-C and Objective-C++ Dialect Options. 67441 (line 207) 67442* Wstring-compare: Warning Options. (line 1497) 67443* Wstringop-overflow: Warning Options. (line 1519) 67444* Wstringop-overflow <1>: Warning Options. (line 1558) 67445* Wstringop-overread: Warning Options. (line 1596) 67446* Wstringop-truncation: Warning Options. (line 1603) 67447* Wsubobject-linkage: C++ Dialect Options. 67448 (line 938) 67449* Wsuggest-attribute=: Warning Options. (line 1654) 67450* Wsuggest-attribute=cold: Warning Options. (line 1697) 67451* Wsuggest-attribute=const: Warning Options. (line 1662) 67452* Wsuggest-attribute=format: Warning Options. (line 1676) 67453* Wsuggest-attribute=malloc: Warning Options. (line 1662) 67454* Wsuggest-attribute=noreturn: Warning Options. (line 1662) 67455* Wsuggest-attribute=pure: Warning Options. (line 1662) 67456* Wsuggest-final-methods: C++ Dialect Options. 67457 (line 1295) 67458* Wsuggest-final-types: C++ Dialect Options. 67459 (line 1286) 67460* Wsuggest-override: C++ Dialect Options. 67461 (line 1305) 67462* Wswitch: Warning Options. (line 1115) 67463* Wswitch-bool: Warning Options. (line 1135) 67464* Wswitch-default: Warning Options. (line 1123) 67465* Wswitch-enum: Warning Options. (line 1126) 67466* Wswitch-outside-range: Warning Options. (line 1146) 67467* Wswitch-unreachable: Warning Options. (line 1151) 67468* Wsync-nand: Warning Options. (line 1175) 67469* Wsystem-headers: Warning Options. (line 2034) 67470* Wtautological-compare: Warning Options. (line 2045) 67471* Wtemplates: C++ Dialect Options. 67472 (line 1035) 67473* Wterminate: C++ Dialect Options. 67474 (line 1128) 67475* Wtraditional: Warning Options. (line 2084) 67476* Wtraditional-conversion: Warning Options. (line 2161) 67477* Wtrampolines: Warning Options. (line 2059) 67478* Wtrigraphs: Warning Options. (line 2368) 67479* Wtrivial-auto-var-init: Warning Options. (line 1180) 67480* Wtsan: Warning Options. (line 2337) 67481* Wtype-limits: Warning Options. (line 2345) 67482* Wundeclared-selector: Objective-C and Objective-C++ Dialect Options. 67483 (line 215) 67484* Wundef: Warning Options. (line 2378) 67485* Wuninitialized: Warning Options. (line 1282) 67486* Wunknown-pragmas: Warning Options. (line 1390) 67487* Wunsafe-loop-optimizations: Warning Options. (line 2305) 67488* Wunsuffixed-float-constants: Warning Options. (line 3396) 67489* Wunused: Warning Options. (line 1275) 67490* Wunused-but-set-parameter: Warning Options. (line 1186) 67491* Wunused-but-set-variable: Warning Options. (line 1195) 67492* Wunused-const-variable: Warning Options. (line 1242) 67493* Wunused-function: Warning Options. (line 1205) 67494* Wunused-label: Warning Options. (line 1210) 67495* Wunused-local-typedefs: Warning Options. (line 1217) 67496* Wunused-macros: Warning Options. (line 2388) 67497* Wunused-parameter: Warning Options. (line 1221) 67498* Wunused-result: Warning Options. (line 1228) 67499* Wunused-value: Warning Options. (line 1265) 67500* Wunused-variable: Warning Options. (line 1233) 67501* Wuse-after-free: C++ Dialect Options. 67502 (line 1310) 67503* Wuseless-cast: C++ Dialect Options. 67504 (line 1366) 67505* Wvarargs: Warning Options. (line 3278) 67506* Wvariadic-macros: Warning Options. (line 3272) 67507* Wvector-operation-performance: Warning Options. (line 3283) 67508* Wvexing-parse: C++ Dialect Options. 67509 (line 1132) 67510* Wvirtual-inheritance: C++ Dialect Options. 67511 (line 1105) 67512* Wvirtual-move-assign: C++ Dialect Options. 67513 (line 1112) 67514* Wvla: Warning Options. (line 3293) 67515* Wvla-larger-than=: Warning Options. (line 3297) 67516* Wvolatile: C++ Dialect Options. 67517 (line 1163) 67518* Wvolatile-register-var: Warning Options. (line 3355) 67519* Wwrite-strings: Warning Options. (line 2537) 67520* Wzero-as-null-pointer-constant: C++ Dialect Options. 67521 (line 1176) 67522* Wzero-length-bounds: Warning Options. (line 2006) 67523* Wzero-length-bounds <1>: Warning Options. (line 2006) 67524* x: Overall Options. (line 135) 67525* Xassembler: Assembler Options. (line 13) 67526* Xbind-lazy: VxWorks Options. (line 26) 67527* Xbind-now: VxWorks Options. (line 30) 67528* Xlinker: Link Options. (line 320) 67529* Xpreprocessor: Preprocessor Options. 67530 (line 476) 67531* Ym: System V Options. (line 26) 67532* YP: System V Options. (line 22) 67533* z: Link Options. (line 351) 67534 67535 67536File: gcc.info, Node: Keyword Index, Prev: Option Index, Up: Top 67537 67538Keyword Index 67539************* 67540 67541[index] 67542* Menu: 67543 67544* #pragma: Pragmas. (line 6) 67545* #pragma implementation: C++ Interface. (line 36) 67546* #pragma implementation, implied: C++ Interface. (line 43) 67547* #pragma interface: C++ Interface. (line 17) 67548* $: Dollar Signs. (line 6) 67549* % in constraint: Modifiers. (line 52) 67550* %include: Spec Files. (line 26) 67551* %include_noerr: Spec Files. (line 30) 67552* %rename: Spec Files. (line 34) 67553* & in constraint: Modifiers. (line 25) 67554* ': Incompatibilities. (line 116) 67555* *: BPF Built-in Functions. 67556 (line 23) 67557* *__builtin_alloca: Other Builtins. (line 137) 67558* *__builtin_alloca_with_align: Other Builtins. (line 174) 67559* *__builtin_alloca_with_align_and_max: Other Builtins. (line 219) 67560* + in constraint: Modifiers. (line 12) 67561* -lgcc, use with -nodefaultlibs: Link Options. (line 157) 67562* -lgcc, use with -nostdlib: Link Options. (line 157) 67563* -march feature modifiers: AArch64 Options. (line 331) 67564* -mcpu feature modifiers: AArch64 Options. (line 331) 67565* -nodefaultlibs and unresolved references: Link Options. (line 157) 67566* -nostdlib and unresolved references: Link Options. (line 157) 67567* .sdata/.sdata2 references (PowerPC): RS/6000 and PowerPC Options. 67568 (line 714) 67569* //: C++ Comments. (line 6) 67570* 0 in constraint: Simple Constraints. (line 125) 67571* < in constraint: Simple Constraints. (line 47) 67572* = in constraint: Modifiers. (line 8) 67573* > in constraint: Simple Constraints. (line 59) 67574* ?: extensions: Conditionals. (line 6) 67575* ?: side effect: Conditionals. (line 20) 67576* _ in variables in macros: Typeof. (line 46) 67577* _Accum data type: Fixed-Point. (line 6) 67578* _Complex keyword: Complex. (line 6) 67579* _Decimal128 data type: Decimal Float. (line 6) 67580* _Decimal32 data type: Decimal Float. (line 6) 67581* _Decimal64 data type: Decimal Float. (line 6) 67582* _Exit: Other Builtins. (line 6) 67583* _exit: Other Builtins. (line 6) 67584* _FloatN data types: Floating Types. (line 6) 67585* _FloatNx data types: Floating Types. (line 6) 67586* _Fract data type: Fixed-Point. (line 6) 67587* _get_ssp: x86 control-flow protection intrinsics. 67588 (line 6) 67589* _HTM_FIRST_USER_ABORT_CODE: S/390 System z Built-in Functions. 67590 (line 44) 67591* _inc_ssp: x86 control-flow protection intrinsics. 67592 (line 12) 67593* _Sat data type: Fixed-Point. (line 6) 67594* _xabort: x86 transactional memory intrinsics. 67595 (line 57) 67596* _xbegin: x86 transactional memory intrinsics. 67597 (line 19) 67598* _xend: x86 transactional memory intrinsics. 67599 (line 48) 67600* _xtest: x86 transactional memory intrinsics. 67601 (line 53) 67602* __atomic_add_fetch: __atomic Builtins. (line 177) 67603* __atomic_always_lock_free: __atomic Builtins. (line 265) 67604* __atomic_and_fetch: __atomic Builtins. (line 181) 67605* __atomic_clear: __atomic Builtins. (line 239) 67606* __atomic_compare_exchange: __atomic Builtins. (line 169) 67607* __atomic_compare_exchange_n: __atomic Builtins. (line 145) 67608* __atomic_exchange: __atomic Builtins. (line 139) 67609* __atomic_exchange_n: __atomic Builtins. (line 131) 67610* __atomic_fetch_add: __atomic Builtins. (line 202) 67611* __atomic_fetch_and: __atomic Builtins. (line 206) 67612* __atomic_fetch_nand: __atomic Builtins. (line 212) 67613* __atomic_fetch_or: __atomic Builtins. (line 210) 67614* __atomic_fetch_sub: __atomic Builtins. (line 204) 67615* __atomic_fetch_xor: __atomic Builtins. (line 208) 67616* __atomic_is_lock_free: __atomic Builtins. (line 279) 67617* __atomic_load: __atomic Builtins. (line 113) 67618* __atomic_load_n: __atomic Builtins. (line 106) 67619* __atomic_nand_fetch: __atomic Builtins. (line 187) 67620* __atomic_or_fetch: __atomic Builtins. (line 185) 67621* __atomic_signal_fence: __atomic Builtins. (line 258) 67622* __atomic_store: __atomic Builtins. (line 126) 67623* __atomic_store_n: __atomic Builtins. (line 118) 67624* __atomic_sub_fetch: __atomic Builtins. (line 179) 67625* __atomic_test_and_set: __atomic Builtins. (line 227) 67626* __atomic_thread_fence: __atomic Builtins. (line 251) 67627* __atomic_xor_fetch: __atomic Builtins. (line 183) 67628* __builtin_addf128_round_to_odd: Basic PowerPC Built-in Functions Available on ISA 3.0. 67629 (line 17) 67630* __builtin_add_overflow: Integer Overflow Builtins. 67631 (line 9) 67632* __builtin_add_overflow_p: Integer Overflow Builtins. 67633 (line 86) 67634* __builtin_alloca: Other Builtins. (line 6) 67635* __builtin_alloca_with_align: Other Builtins. (line 6) 67636* __builtin_alloca_with_align_and_max: Other Builtins. (line 6) 67637* __builtin_apply: Constructing Calls. (line 29) 67638* __builtin_apply_args: Constructing Calls. (line 19) 67639* __builtin_arc_aligned: ARC Built-in Functions. 67640 (line 18) 67641* __builtin_arc_brk: ARC Built-in Functions. 67642 (line 28) 67643* __builtin_arc_core_read: ARC Built-in Functions. 67644 (line 32) 67645* __builtin_arc_core_write: ARC Built-in Functions. 67646 (line 39) 67647* __builtin_arc_divaw: ARC Built-in Functions. 67648 (line 46) 67649* __builtin_arc_flag: ARC Built-in Functions. 67650 (line 53) 67651* __builtin_arc_lr: ARC Built-in Functions. 67652 (line 57) 67653* __builtin_arc_mul64: ARC Built-in Functions. 67654 (line 64) 67655* __builtin_arc_mulu64: ARC Built-in Functions. 67656 (line 68) 67657* __builtin_arc_nop: ARC Built-in Functions. 67658 (line 73) 67659* __builtin_arc_norm: ARC Built-in Functions. 67660 (line 77) 67661* __builtin_arc_normw: ARC Built-in Functions. 67662 (line 84) 67663* __builtin_arc_rtie: ARC Built-in Functions. 67664 (line 91) 67665* __builtin_arc_sleep: ARC Built-in Functions. 67666 (line 95) 67667* __builtin_arc_sr: ARC Built-in Functions. 67668 (line 99) 67669* __builtin_arc_swap: ARC Built-in Functions. 67670 (line 106) 67671* __builtin_arc_swi: ARC Built-in Functions. 67672 (line 112) 67673* __builtin_arc_sync: ARC Built-in Functions. 67674 (line 116) 67675* __builtin_arc_trap_s: ARC Built-in Functions. 67676 (line 120) 67677* __builtin_arc_unimp_s: ARC Built-in Functions. 67678 (line 124) 67679* __builtin_assoc_barrier: Other Builtins. (line 695) 67680* __builtin_assume_aligned: Other Builtins. (line 710) 67681* __builtin_bit_cast: Other Builtins. (line 581) 67682* __builtin_bswap128: Other Builtins. (line 1055) 67683* __builtin_bswap16: Other Builtins. (line 1043) 67684* __builtin_bswap32: Other Builtins. (line 1047) 67685* __builtin_bswap64: Other Builtins. (line 1051) 67686* __builtin_call_with_static_chain: Other Builtins. (line 6) 67687* __builtin_call_with_static_chain <1>: Other Builtins. (line 393) 67688* __builtin_cfuged: Basic PowerPC Built-in Functions Available on ISA 3.1. 67689 (line 18) 67690* __builtin_choose_expr: Other Builtins. (line 404) 67691* __builtin_clear_padding: Other Builtins. (line 561) 67692* __builtin_clrsb: Other Builtins. (line 973) 67693* __builtin_clrsbl: Other Builtins. (line 995) 67694* __builtin_clrsbll: Other Builtins. (line 1018) 67695* __builtin_clz: Other Builtins. (line 965) 67696* __builtin_clzl: Other Builtins. (line 987) 67697* __builtin_clzll: Other Builtins. (line 1010) 67698* __builtin_cntlzdm: Basic PowerPC Built-in Functions Available on ISA 3.1. 67699 (line 23) 67700* __builtin_cnttzdm: Basic PowerPC Built-in Functions Available on ISA 3.1. 67701 (line 28) 67702* __builtin_complex: Complex. (line 82) 67703* __builtin_constant_p: Other Builtins. (line 498) 67704* __builtin_convertvector: Vector Extensions. (line 189) 67705* __builtin_cpu_init: Basic PowerPC Built-in Functions Available on all Configurations. 67706 (line 6) 67707* __builtin_cpu_init <1>: x86 Built-in Functions. 67708 (line 68) 67709* __builtin_cpu_is: Basic PowerPC Built-in Functions Available on all Configurations. 67710 (line 10) 67711* __builtin_cpu_is <1>: x86 Built-in Functions. 67712 (line 96) 67713* __builtin_cpu_supports: Basic PowerPC Built-in Functions Available on all Configurations. 67714 (line 70) 67715* __builtin_cpu_supports <1>: x86 Built-in Functions. 67716 (line 252) 67717* __builtin_ctz: Other Builtins. (line 969) 67718* __builtin_ctzl: Other Builtins. (line 991) 67719* __builtin_ctzll: Other Builtins. (line 1014) 67720* __builtin_divf128_round_to_odd: Basic PowerPC Built-in Functions Available on ISA 3.0. 67721 (line 29) 67722* __builtin_dynamic_object_size: Object Size Checking. 67723 (line 6) 67724* __builtin_dynamic_object_size <1>: Object Size Checking. 67725 (line 53) 67726* __builtin_expect: Other Builtins. (line 595) 67727* __builtin_expect_with_probability: Other Builtins. (line 630) 67728* __builtin_extend_pointer: Other Builtins. (line 6) 67729* __builtin_extend_pointer <1>: Other Builtins. (line 1060) 67730* __builtin_extract_return_addr: Return Address. (line 50) 67731* __builtin_ffs: Other Builtins. (line 961) 67732* __builtin_ffsl: Other Builtins. (line 984) 67733* __builtin_ffsll: Other Builtins. (line 1006) 67734* __builtin_FILE: Other Builtins. (line 743) 67735* __builtin_fmaf128_round_to_odd: Basic PowerPC Built-in Functions Available on ISA 3.0. 67736 (line 37) 67737* __builtin_fpclassify: Other Builtins. (line 6) 67738* __builtin_fpclassify <1>: Other Builtins. (line 845) 67739* __builtin_frame_address: Return Address. (line 62) 67740* __builtin_frob_return_addr: Return Address. (line 59) 67741* __builtin_FUNCTION: Other Builtins. (line 735) 67742* __builtin_goacc_parlevel_id: Other Builtins. (line 1067) 67743* __builtin_goacc_parlevel_size: Other Builtins. (line 1071) 67744* __builtin_has_attribute: Other Builtins. (line 6) 67745* __builtin_has_attribute <1>: Other Builtins. (line 228) 67746* __builtin_huge_val: Other Builtins. (line 825) 67747* __builtin_huge_valf: Other Builtins. (line 830) 67748* __builtin_huge_valfN: Other Builtins. (line 837) 67749* __builtin_huge_valfNx: Other Builtins. (line 841) 67750* __builtin_huge_vall: Other Builtins. (line 833) 67751* __builtin_huge_valq: x86 Built-in Functions. 67752 (line 50) 67753* __builtin_inf: Other Builtins. (line 856) 67754* __builtin_infd128: Other Builtins. (line 866) 67755* __builtin_infd32: Other Builtins. (line 860) 67756* __builtin_infd64: Other Builtins. (line 863) 67757* __builtin_inff: Other Builtins. (line 870) 67758* __builtin_inffN: Other Builtins. (line 879) 67759* __builtin_inffNx: Other Builtins. (line 882) 67760* __builtin_infl: Other Builtins. (line 875) 67761* __builtin_infq: x86 Built-in Functions. 67762 (line 47) 67763* __builtin_isfinite: Other Builtins. (line 6) 67764* __builtin_isgreater: Other Builtins. (line 6) 67765* __builtin_isgreaterequal: Other Builtins. (line 6) 67766* __builtin_isinf_sign: Other Builtins. (line 6) 67767* __builtin_isinf_sign <1>: Other Builtins. (line 885) 67768* __builtin_isless: Other Builtins. (line 6) 67769* __builtin_islessequal: Other Builtins. (line 6) 67770* __builtin_islessgreater: Other Builtins. (line 6) 67771* __builtin_isnormal: Other Builtins. (line 6) 67772* __builtin_isunordered: Other Builtins. (line 6) 67773* __builtin_is_constant_evaluated: Other Builtins. (line 543) 67774* __builtin_LINE: Other Builtins. (line 728) 67775* __builtin_longjmp: Nonlocal Gotos. (line 37) 67776* __builtin_mulf128_round_to_odd: Basic PowerPC Built-in Functions Available on ISA 3.0. 67777 (line 25) 67778* __builtin_mul_overflow: Integer Overflow Builtins. 67779 (line 63) 67780* __builtin_mul_overflow_p: Integer Overflow Builtins. 67781 (line 90) 67782* __builtin_nan: Other Builtins. (line 893) 67783* __builtin_nand128: Other Builtins. (line 915) 67784* __builtin_nand32: Other Builtins. (line 909) 67785* __builtin_nand64: Other Builtins. (line 912) 67786* __builtin_nanf: Other Builtins. (line 919) 67787* __builtin_nanfN: Other Builtins. (line 926) 67788* __builtin_nanfNx: Other Builtins. (line 929) 67789* __builtin_nanl: Other Builtins. (line 922) 67790* __builtin_nanq: x86 Built-in Functions. 67791 (line 54) 67792* __builtin_nans: Other Builtins. (line 932) 67793* __builtin_nansd128: Other Builtins. (line 944) 67794* __builtin_nansd32: Other Builtins. (line 936) 67795* __builtin_nansd64: Other Builtins. (line 940) 67796* __builtin_nansf: Other Builtins. (line 948) 67797* __builtin_nansfN: Other Builtins. (line 955) 67798* __builtin_nansfNx: Other Builtins. (line 958) 67799* __builtin_nansl: Other Builtins. (line 951) 67800* __builtin_nansq: x86 Built-in Functions. 67801 (line 57) 67802* __builtin_nds32_isb: NDS32 Built-in Functions. 67803 (line 12) 67804* __builtin_nds32_isync: NDS32 Built-in Functions. 67805 (line 8) 67806* __builtin_nds32_mfsr: NDS32 Built-in Functions. 67807 (line 15) 67808* __builtin_nds32_mfusr: NDS32 Built-in Functions. 67809 (line 18) 67810* __builtin_nds32_mtsr: NDS32 Built-in Functions. 67811 (line 21) 67812* __builtin_nds32_mtusr: NDS32 Built-in Functions. 67813 (line 24) 67814* __builtin_nds32_setgie_dis: NDS32 Built-in Functions. 67815 (line 30) 67816* __builtin_nds32_setgie_en: NDS32 Built-in Functions. 67817 (line 27) 67818* __builtin_non_tx_store: S/390 System z Built-in Functions. 67819 (line 98) 67820* __builtin_object_size: Object Size Checking. 67821 (line 6) 67822* __builtin_object_size <1>: Object Size Checking. 67823 (line 16) 67824* __builtin_object_size <2>: Other Builtins. (line 6) 67825* __builtin_object_size <3>: Other Builtins. (line 820) 67826* __builtin_offsetof: Offsetof. (line 6) 67827* __builtin_parity: Other Builtins. (line 981) 67828* __builtin_parityl: Other Builtins. (line 1002) 67829* __builtin_parityll: Other Builtins. (line 1026) 67830* __builtin_pdepd: Basic PowerPC Built-in Functions Available on ISA 3.1. 67831 (line 33) 67832* __builtin_pextd: Basic PowerPC Built-in Functions Available on ISA 3.1. 67833 (line 38) 67834* __builtin_popcount: Other Builtins. (line 978) 67835* __builtin_popcountl: Other Builtins. (line 998) 67836* __builtin_popcountll: Other Builtins. (line 1022) 67837* __builtin_powi: Other Builtins. (line 6) 67838* __builtin_powi <1>: Other Builtins. (line 1030) 67839* __builtin_powif: Other Builtins. (line 6) 67840* __builtin_powif <1>: Other Builtins. (line 1035) 67841* __builtin_powil: Other Builtins. (line 6) 67842* __builtin_powil <1>: Other Builtins. (line 1039) 67843* __builtin_prefetch: Other Builtins. (line 781) 67844* __builtin_return: Constructing Calls. (line 47) 67845* __builtin_return_address: Return Address. (line 9) 67846* __builtin_rx_brk: RX Built-in Functions. 67847 (line 10) 67848* __builtin_rx_clrpsw: RX Built-in Functions. 67849 (line 13) 67850* __builtin_rx_int: RX Built-in Functions. 67851 (line 17) 67852* __builtin_rx_machi: RX Built-in Functions. 67853 (line 21) 67854* __builtin_rx_maclo: RX Built-in Functions. 67855 (line 26) 67856* __builtin_rx_mulhi: RX Built-in Functions. 67857 (line 31) 67858* __builtin_rx_mullo: RX Built-in Functions. 67859 (line 36) 67860* __builtin_rx_mvfachi: RX Built-in Functions. 67861 (line 41) 67862* __builtin_rx_mvfacmi: RX Built-in Functions. 67863 (line 45) 67864* __builtin_rx_mvfc: RX Built-in Functions. 67865 (line 49) 67866* __builtin_rx_mvtachi: RX Built-in Functions. 67867 (line 53) 67868* __builtin_rx_mvtaclo: RX Built-in Functions. 67869 (line 57) 67870* __builtin_rx_mvtc: RX Built-in Functions. 67871 (line 61) 67872* __builtin_rx_mvtipl: RX Built-in Functions. 67873 (line 65) 67874* __builtin_rx_racw: RX Built-in Functions. 67875 (line 69) 67876* __builtin_rx_revw: RX Built-in Functions. 67877 (line 73) 67878* __builtin_rx_rmpa: RX Built-in Functions. 67879 (line 78) 67880* __builtin_rx_round: RX Built-in Functions. 67881 (line 82) 67882* __builtin_rx_sat: RX Built-in Functions. 67883 (line 87) 67884* __builtin_rx_setpsw: RX Built-in Functions. 67885 (line 91) 67886* __builtin_rx_wait: RX Built-in Functions. 67887 (line 95) 67888* __builtin_saddll_overflow: Integer Overflow Builtins. 67889 (line 15) 67890* __builtin_saddl_overflow: Integer Overflow Builtins. 67891 (line 13) 67892* __builtin_sadd_overflow: Integer Overflow Builtins. 67893 (line 11) 67894* __builtin_setjmp: Nonlocal Gotos. (line 32) 67895* __builtin_set_thread_pointer: SH Built-in Functions. 67896 (line 9) 67897* __builtin_shuffle: Vector Extensions. (line 127) 67898* __builtin_shufflevector: Vector Extensions. (line 165) 67899* __builtin_sh_get_fpscr: SH Built-in Functions. 67900 (line 35) 67901* __builtin_sh_set_fpscr: SH Built-in Functions. 67902 (line 38) 67903* __builtin_smulll_overflow: Integer Overflow Builtins. 67904 (line 69) 67905* __builtin_smull_overflow: Integer Overflow Builtins. 67906 (line 67) 67907* __builtin_smul_overflow: Integer Overflow Builtins. 67908 (line 65) 67909* __builtin_speculation_safe_value: Other Builtins. (line 6) 67910* __builtin_speculation_safe_value <1>: Other Builtins. (line 269) 67911* __builtin_sqrtf128_round_to_odd: Basic PowerPC Built-in Functions Available on ISA 3.0. 67912 (line 33) 67913* __builtin_ssubll_overflow: Integer Overflow Builtins. 67914 (line 49) 67915* __builtin_ssubl_overflow: Integer Overflow Builtins. 67916 (line 47) 67917* __builtin_ssub_overflow: Integer Overflow Builtins. 67918 (line 45) 67919* __builtin_subf128_round_to_odd: Basic PowerPC Built-in Functions Available on ISA 3.0. 67920 (line 21) 67921* __builtin_sub_overflow: Integer Overflow Builtins. 67922 (line 43) 67923* __builtin_sub_overflow_p: Integer Overflow Builtins. 67924 (line 88) 67925* __builtin_tabort: S/390 System z Built-in Functions. 67926 (line 82) 67927* __builtin_tbegin: S/390 System z Built-in Functions. 67928 (line 6) 67929* __builtin_tbeginc: S/390 System z Built-in Functions. 67930 (line 73) 67931* __builtin_tbegin_nofloat: S/390 System z Built-in Functions. 67932 (line 54) 67933* __builtin_tbegin_retry: S/390 System z Built-in Functions. 67934 (line 60) 67935* __builtin_tbegin_retry_nofloat: S/390 System z Built-in Functions. 67936 (line 67) 67937* __builtin_tend: S/390 System z Built-in Functions. 67938 (line 77) 67939* __builtin_tgmath: Other Builtins. (line 444) 67940* __builtin_thread_pointer: RISC-V Built-in Functions. 67941 (line 9) 67942* __builtin_thread_pointer <1>: SH Built-in Functions. 67943 (line 18) 67944* __builtin_trap: Other Builtins. (line 639) 67945* __builtin_truncf128_round_to_odd: Basic PowerPC Built-in Functions Available on ISA 3.0. 67946 (line 41) 67947* __builtin_tx_assist: S/390 System z Built-in Functions. 67948 (line 87) 67949* __builtin_tx_nesting_depth: S/390 System z Built-in Functions. 67950 (line 93) 67951* __builtin_types_compatible_p: Other Builtins. (line 348) 67952* __builtin_uaddll_overflow: Integer Overflow Builtins. 67953 (line 21) 67954* __builtin_uaddl_overflow: Integer Overflow Builtins. 67955 (line 19) 67956* __builtin_uadd_overflow: Integer Overflow Builtins. 67957 (line 17) 67958* __builtin_umulll_overflow: Integer Overflow Builtins. 67959 (line 75) 67960* __builtin_umull_overflow: Integer Overflow Builtins. 67961 (line 73) 67962* __builtin_umul_overflow: Integer Overflow Builtins. 67963 (line 71) 67964* __builtin_unreachable: Other Builtins. (line 646) 67965* __builtin_usubll_overflow: Integer Overflow Builtins. 67966 (line 55) 67967* __builtin_usubl_overflow: Integer Overflow Builtins. 67968 (line 53) 67969* __builtin_usub_overflow: Integer Overflow Builtins. 67970 (line 51) 67971* __builtin_va_arg_pack: Constructing Calls. (line 52) 67972* __builtin_va_arg_pack_len: Constructing Calls. (line 75) 67973* __builtin___clear_cache: Other Builtins. (line 768) 67974* __builtin___fprintf_chk: Object Size Checking. 67975 (line 6) 67976* __builtin___memcpy_chk: Object Size Checking. 67977 (line 6) 67978* __builtin___memmove_chk: Object Size Checking. 67979 (line 6) 67980* __builtin___mempcpy_chk: Object Size Checking. 67981 (line 6) 67982* __builtin___memset_chk: Object Size Checking. 67983 (line 6) 67984* __builtin___printf_chk: Object Size Checking. 67985 (line 6) 67986* __builtin___snprintf_chk: Object Size Checking. 67987 (line 6) 67988* __builtin___sprintf_chk: Object Size Checking. 67989 (line 6) 67990* __builtin___stpcpy_chk: Object Size Checking. 67991 (line 6) 67992* __builtin___strcat_chk: Object Size Checking. 67993 (line 6) 67994* __builtin___strcpy_chk: Object Size Checking. 67995 (line 6) 67996* __builtin___strncat_chk: Object Size Checking. 67997 (line 6) 67998* __builtin___strncpy_chk: Object Size Checking. 67999 (line 6) 68000* __builtin___vfprintf_chk: Object Size Checking. 68001 (line 6) 68002* __builtin___vprintf_chk: Object Size Checking. 68003 (line 6) 68004* __builtin___vsnprintf_chk: Object Size Checking. 68005 (line 6) 68006* __builtin___vsprintf_chk: Object Size Checking. 68007 (line 6) 68008* __complex__ keyword: Complex. (line 6) 68009* __declspec(dllexport): Microsoft Windows Function Attributes. 68010 (line 10) 68011* __declspec(dllimport): Microsoft Windows Function Attributes. 68012 (line 42) 68013* __extension__: Alternate Keywords. (line 30) 68014* __far M32C Named Address Spaces: Named Address Spaces. 68015 (line 153) 68016* __far RL78 Named Address Spaces: Named Address Spaces. 68017 (line 177) 68018* __flash AVR Named Address Spaces: Named Address Spaces. 68019 (line 44) 68020* __flash1 AVR Named Address Spaces: Named Address Spaces. 68021 (line 53) 68022* __flash2 AVR Named Address Spaces: Named Address Spaces. 68023 (line 53) 68024* __flash3 AVR Named Address Spaces: Named Address Spaces. 68025 (line 53) 68026* __flash4 AVR Named Address Spaces: Named Address Spaces. 68027 (line 53) 68028* __flash5 AVR Named Address Spaces: Named Address Spaces. 68029 (line 53) 68030* __float128 data type: Floating Types. (line 6) 68031* __Float16 data type: Half-Precision. (line 6) 68032* __float80 data type: Floating Types. (line 6) 68033* __fp16 data type: Half-Precision. (line 6) 68034* __FUNCTION__ identifier: Function Names. (line 6) 68035* __func__ identifier: Function Names. (line 6) 68036* __ibm128 data type: Floating Types. (line 6) 68037* __imag__ keyword: Complex. (line 43) 68038* __int128 data types: __int128. (line 6) 68039* __memx AVR Named Address Spaces: Named Address Spaces. 68040 (line 59) 68041* __PRETTY_FUNCTION__ identifier: Function Names. (line 6) 68042* __real__ keyword: Complex. (line 43) 68043* __regio_symbol PRU Named Address Spaces: Named Address Spaces. 68044 (line 162) 68045* __seg_fs x86 named address space: Named Address Spaces. 68046 (line 190) 68047* __seg_gs x86 named address space: Named Address Spaces. 68048 (line 190) 68049* __STDC_HOSTED__: Standards. (line 13) 68050* __sync_add_and_fetch: __sync Builtins. (line 72) 68051* __sync_and_and_fetch: __sync Builtins. (line 72) 68052* __sync_bool_compare_and_swap: __sync Builtins. (line 88) 68053* __sync_fetch_and_add: __sync Builtins. (line 50) 68054* __sync_fetch_and_and: __sync Builtins. (line 50) 68055* __sync_fetch_and_nand: __sync Builtins. (line 50) 68056* __sync_fetch_and_or: __sync Builtins. (line 50) 68057* __sync_fetch_and_sub: __sync Builtins. (line 50) 68058* __sync_fetch_and_xor: __sync Builtins. (line 50) 68059* __sync_lock_release: __sync Builtins. (line 118) 68060* __sync_lock_test_and_set: __sync Builtins. (line 100) 68061* __sync_nand_and_fetch: __sync Builtins. (line 72) 68062* __sync_or_and_fetch: __sync Builtins. (line 72) 68063* __sync_sub_and_fetch: __sync Builtins. (line 72) 68064* __sync_synchronize: __sync Builtins. (line 97) 68065* __sync_val_compare_and_swap: __sync Builtins. (line 88) 68066* __sync_xor_and_fetch: __sync Builtins. (line 72) 68067* __thread: Thread-Local. (line 6) 68068* AArch64 Options: AArch64 Options. (line 6) 68069* ABI: Compatibility. (line 6) 68070* abi_tag function attribute: C++ Attributes. (line 9) 68071* abi_tag type attribute: C++ Attributes. (line 9) 68072* abi_tag variable attribute: C++ Attributes. (line 9) 68073* abort: Other Builtins. (line 6) 68074* abs: Other Builtins. (line 6) 68075* absdata variable attribute, AVR: AVR Variable Attributes. 68076 (line 104) 68077* accessing volatiles: Volatiles. (line 6) 68078* accessing volatiles <1>: C++ Volatiles. (line 6) 68079* acos: Other Builtins. (line 6) 68080* acosf: Other Builtins. (line 6) 68081* acosh: Other Builtins. (line 6) 68082* acoshf: Other Builtins. (line 6) 68083* acoshl: Other Builtins. (line 6) 68084* acosl: Other Builtins. (line 6) 68085* Ada: G++ and GCC. (line 6) 68086* Ada <1>: G++ and GCC. (line 29) 68087* additional floating types: Floating Types. (line 6) 68088* address constraints: Simple Constraints. (line 152) 68089* address of a label: Labels as Values. (line 6) 68090* address variable attribute, AVR: AVR Variable Attributes. 68091 (line 97) 68092* address_operand: Simple Constraints. (line 156) 68093* alias function attribute: Common Function Attributes. 68094 (line 93) 68095* alias variable attribute: Common Variable Attributes. 68096 (line 9) 68097* aligned function attribute: Common Function Attributes. 68098 (line 112) 68099* aligned type attribute: Common Type Attributes. 68100 (line 8) 68101* aligned variable attribute: Common Variable Attributes. 68102 (line 31) 68103* alignment: Alignment. (line 6) 68104* alloca: Other Builtins. (line 6) 68105* alloca vs variable-length arrays: Variable Length. (line 35) 68106* alloc_align function attribute: Common Function Attributes. 68107 (line 140) 68108* alloc_size function attribute: Common Function Attributes. 68109 (line 160) 68110* alloc_size type attribute: Common Type Attributes. 68111 (line 136) 68112* alloc_size variable attribute: Common Variable Attributes. 68113 (line 137) 68114* Allow nesting in an interrupt handler on the Blackfin processor: Blackfin Function Attributes. 68115 (line 45) 68116* Altera Nios II options: Nios II Options. (line 6) 68117* alternate keywords: Alternate Keywords. (line 6) 68118* altivec type attribute, PowerPC: PowerPC Type Attributes. 68119 (line 12) 68120* altivec variable attribute, PowerPC: PowerPC Variable Attributes. 68121 (line 12) 68122* always_inline function attribute: Common Function Attributes. 68123 (line 186) 68124* AMD GCN Options: AMD GCN Options. (line 6) 68125* AMD1: Standards. (line 13) 68126* amdgpu_hsa_kernel function attribute, AMD GCN: AMD GCN Function Attributes. 68127 (line 9) 68128* ANSI C: Standards. (line 13) 68129* ANSI C standard: Standards. (line 13) 68130* ANSI C89: Standards. (line 13) 68131* ANSI support: C Dialect Options. (line 10) 68132* ANSI X3.159-1989: Standards. (line 13) 68133* apostrophes: Incompatibilities. (line 116) 68134* application binary interface: Compatibility. (line 6) 68135* ARC options: ARC Options. (line 6) 68136* arch= function attribute, AArch64: AArch64 Function Attributes. 68137 (line 53) 68138* arch= function attribute, ARM: ARM Function Attributes. 68139 (line 98) 68140* ARM options: ARM Options. (line 6) 68141* ARM [Annotated C++ Reference Manual]: Backwards Compatibility. 68142 (line 6) 68143* arrays of length zero: Zero Length. (line 6) 68144* arrays of variable length: Variable Length. (line 6) 68145* arrays, non-lvalue: Subscripting. (line 6) 68146* artificial function attribute: Common Function Attributes. 68147 (line 196) 68148* asin: Other Builtins. (line 6) 68149* asinf: Other Builtins. (line 6) 68150* asinh: Other Builtins. (line 6) 68151* asinhf: Other Builtins. (line 6) 68152* asinhl: Other Builtins. (line 6) 68153* asinl: Other Builtins. (line 6) 68154* asm assembler template: Extended Asm. (line 226) 68155* asm clobbers: Extended Asm. (line 693) 68156* asm constraints: Constraints. (line 6) 68157* asm expressions: Extended Asm. (line 598) 68158* asm flag output operands: Extended Asm. (line 488) 68159* asm goto labels: Extended Asm. (line 880) 68160* asm inline: Size of an asm. (line 25) 68161* asm input operands: Extended Asm. (line 598) 68162* asm keyword: Using Assembly Language with C. 68163 (line 6) 68164* asm output operands: Extended Asm. (line 329) 68165* asm scratch registers: Extended Asm. (line 693) 68166* asm volatile: Extended Asm. (line 116) 68167* assembler names for identifiers: Asm Labels. (line 6) 68168* assembly code, invalid: Bug Criteria. (line 12) 68169* assembly language in C: Using Assembly Language with C. 68170 (line 6) 68171* assembly language in C, basic: Basic Asm. (line 6) 68172* assembly language in C, extended: Extended Asm. (line 6) 68173* assume_aligned function attribute: Common Function Attributes. 68174 (line 204) 68175* atan: Other Builtins. (line 6) 68176* atan2: Other Builtins. (line 6) 68177* atan2f: Other Builtins. (line 6) 68178* atan2l: Other Builtins. (line 6) 68179* atanf: Other Builtins. (line 6) 68180* atanh: Other Builtins. (line 6) 68181* atanhf: Other Builtins. (line 6) 68182* atanhl: Other Builtins. (line 6) 68183* atanl: Other Builtins. (line 6) 68184* attribute of types: Type Attributes. (line 6) 68185* attribute of variables: Variable Attributes. 68186 (line 6) 68187* attribute syntax: Attribute Syntax. (line 6) 68188* autoincrement/decrement addressing: Simple Constraints. (line 30) 68189* automatic inline for C++ member fns: Inline. (line 68) 68190* aux variable attribute, ARC: ARC Variable Attributes. 68191 (line 7) 68192* AVR Options: AVR Options. (line 6) 68193* Backwards Compatibility: Backwards Compatibility. 68194 (line 6) 68195* bank_switch function attribute, M32C: M32C Function Attributes. 68196 (line 9) 68197* base class members: Name lookup. (line 6) 68198* based type attribute, MeP: MeP Type Attributes. 68199 (line 6) 68200* based variable attribute, MeP: MeP Variable Attributes. 68201 (line 16) 68202* basic asm: Basic Asm. (line 6) 68203* bcmp: Other Builtins. (line 6) 68204* below100 variable attribute, Xstormy16: Xstormy16 Variable Attributes. 68205 (line 10) 68206* binary compatibility: Compatibility. (line 6) 68207* Binary constants using the 0b prefix: Binary constants. (line 6) 68208* Blackfin Options: Blackfin Options. (line 6) 68209* bound pointer to member function: Bound member functions. 68210 (line 6) 68211* branch-protection function attribute, AArch64: AArch64 Function Attributes. 68212 (line 76) 68213* break handler functions: MicroBlaze Function Attributes. 68214 (line 17) 68215* break_handler function attribute, MicroBlaze: MicroBlaze Function Attributes. 68216 (line 17) 68217* brk_interrupt function attribute, RL78: RL78 Function Attributes. 68218 (line 10) 68219* bug criteria: Bug Criteria. (line 6) 68220* bugs: Bugs. (line 6) 68221* bugs, known: Trouble. (line 6) 68222* built-in functions: C Dialect Options. (line 239) 68223* built-in functions <1>: Other Builtins. (line 6) 68224* bzero: Other Builtins. (line 6) 68225* C compilation options: Invoking GCC. (line 18) 68226* C intermediate output, nonexistent: G++ and GCC. (line 34) 68227* C language extensions: C Extensions. (line 6) 68228* C language, traditional: Preprocessor Options. 68229 (line 375) 68230* C standard: Standards. (line 13) 68231* C standards: Standards. (line 13) 68232* c++: Invoking G++. (line 14) 68233* C++: G++ and GCC. (line 29) 68234* C++ comments: C++ Comments. (line 6) 68235* C++ Compiled Module Interface: C++ Compiled Module Interface. 68236 (line 6) 68237* C++ interface and implementation headers: C++ Interface. (line 6) 68238* C++ language extensions: C++ Extensions. (line 6) 68239* C++ member fns, automatically inline: Inline. (line 68) 68240* C++ misunderstandings: C++ Misunderstandings. 68241 (line 6) 68242* C++ Module Mapper: C++ Module Mapper. (line 6) 68243* C++ Module Preprocessing: C++ Module Preprocessing. 68244 (line 6) 68245* C++ options, command-line: C++ Dialect Options. 68246 (line 6) 68247* C++ pragmas, effect on inlining: C++ Interface. (line 57) 68248* C++ source file suffixes: Invoking G++. (line 6) 68249* C++ static data, declaring and defining: Static Definitions. 68250 (line 6) 68251* C-SKY Options: C-SKY Options. (line 6) 68252* C11: Standards. (line 13) 68253* C17: Standards. (line 13) 68254* C1X: Standards. (line 13) 68255* C2X: Standards. (line 13) 68256* C6X Options: C6X Options. (line 6) 68257* C89: Standards. (line 13) 68258* C90: Standards. (line 13) 68259* C94: Standards. (line 13) 68260* C95: Standards. (line 13) 68261* C99: Standards. (line 13) 68262* C9X: Standards. (line 13) 68263* cabs: Other Builtins. (line 6) 68264* cabsf: Other Builtins. (line 6) 68265* cabsl: Other Builtins. (line 6) 68266* cacos: Other Builtins. (line 6) 68267* cacosf: Other Builtins. (line 6) 68268* cacosh: Other Builtins. (line 6) 68269* cacoshf: Other Builtins. (line 6) 68270* cacoshl: Other Builtins. (line 6) 68271* cacosl: Other Builtins. (line 6) 68272* callee_pop_aggregate_return function attribute, x86: x86 Function Attributes. 68273 (line 47) 68274* calling functions through the function vector on SH2A: SH Function Attributes. 68275 (line 9) 68276* calloc: Other Builtins. (line 6) 68277* carg: Other Builtins. (line 6) 68278* cargf: Other Builtins. (line 6) 68279* cargl: Other Builtins. (line 6) 68280* case labels in initializers: Designated Inits. (line 6) 68281* case ranges: Case Ranges. (line 6) 68282* casin: Other Builtins. (line 6) 68283* casinf: Other Builtins. (line 6) 68284* casinh: Other Builtins. (line 6) 68285* casinhf: Other Builtins. (line 6) 68286* casinhl: Other Builtins. (line 6) 68287* casinl: Other Builtins. (line 6) 68288* cast to a union: Cast to Union. (line 6) 68289* catan: Other Builtins. (line 6) 68290* catanf: Other Builtins. (line 6) 68291* catanh: Other Builtins. (line 6) 68292* catanhf: Other Builtins. (line 6) 68293* catanhl: Other Builtins. (line 6) 68294* catanl: Other Builtins. (line 6) 68295* cb variable attribute, MeP: MeP Variable Attributes. 68296 (line 46) 68297* cbrt: Other Builtins. (line 6) 68298* cbrtf: Other Builtins. (line 6) 68299* cbrtl: Other Builtins. (line 6) 68300* ccos: Other Builtins. (line 6) 68301* ccosf: Other Builtins. (line 6) 68302* ccosh: Other Builtins. (line 6) 68303* ccoshf: Other Builtins. (line 6) 68304* ccoshl: Other Builtins. (line 6) 68305* ccosl: Other Builtins. (line 6) 68306* cdecl function attribute, x86-32: x86 Function Attributes. 68307 (line 9) 68308* ceil: Other Builtins. (line 6) 68309* ceilf: Other Builtins. (line 6) 68310* ceill: Other Builtins. (line 6) 68311* cexp: Other Builtins. (line 6) 68312* cexpf: Other Builtins. (line 6) 68313* cexpl: Other Builtins. (line 6) 68314* cf_check function attribute, x86: x86 Function Attributes. 68315 (line 689) 68316* character set, execution: Preprocessor Options. 68317 (line 273) 68318* character set, input: Preprocessor Options. 68319 (line 288) 68320* character set, input normalization: Warning Options. (line 3029) 68321* character set, wide execution: Preprocessor Options. 68322 (line 278) 68323* cimag: Other Builtins. (line 6) 68324* cimagf: Other Builtins. (line 6) 68325* cimagl: Other Builtins. (line 6) 68326* cleanup variable attribute: Common Variable Attributes. 68327 (line 161) 68328* clog: Other Builtins. (line 6) 68329* clog10: Other Builtins. (line 6) 68330* clog10f: Other Builtins. (line 6) 68331* clog10l: Other Builtins. (line 6) 68332* clogf: Other Builtins. (line 6) 68333* clogl: Other Builtins. (line 6) 68334* cmodel= function attribute, AArch64: AArch64 Function Attributes. 68335 (line 27) 68336* COBOL: G++ and GCC. (line 23) 68337* code generation conventions: Code Gen Options. (line 6) 68338* code, mixed with declarations: Mixed Labels and Declarations. 68339 (line 6) 68340* cold function attribute: Common Function Attributes. 68341 (line 220) 68342* cold label attribute: Label Attributes. (line 46) 68343* command options: Invoking GCC. (line 6) 68344* comments, C++ style: C++ Comments. (line 6) 68345* common variable attribute: Common Variable Attributes. 68346 (line 176) 68347* comparison of signed and unsigned values, warning: Warning Options. 68348 (line 2691) 68349* compilation statistics: Developer Options. (line 6) 68350* compiler bugs, reporting: Bug Reporting. (line 6) 68351* compiler compared to C++ preprocessor: G++ and GCC. (line 34) 68352* compiler options, C++: C++ Dialect Options. 68353 (line 6) 68354* compiler options, Objective-C and Objective-C++: Objective-C and Objective-C++ Dialect Options. 68355 (line 6) 68356* compiler version, specifying: Invoking GCC. (line 24) 68357* COMPILER_PATH: Environment Variables. 68358 (line 91) 68359* complex conjugation: Complex. (line 64) 68360* complex numbers: Complex. (line 6) 68361* compound literals: Compound Literals. (line 6) 68362* computed gotos: Labels as Values. (line 6) 68363* conditional expressions, extensions: Conditionals. (line 6) 68364* conflicting types: Disappointments. (line 21) 68365* conj: Other Builtins. (line 6) 68366* conjf: Other Builtins. (line 6) 68367* conjl: Other Builtins. (line 6) 68368* const applied to function: Function Attributes. 68369 (line 6) 68370* const function attribute: Common Function Attributes. 68371 (line 236) 68372* const qualifier: Pointers to Arrays. (line 6) 68373* constants in constraints: Simple Constraints. (line 68) 68374* constraint modifier characters: Modifiers. (line 6) 68375* constraint, matching: Simple Constraints. (line 137) 68376* constraints, asm: Constraints. (line 6) 68377* constraints, machine specific: Machine Constraints. 68378 (line 6) 68379* constructing calls: Constructing Calls. (line 6) 68380* constructor expressions: Compound Literals. (line 6) 68381* constructor function attribute: Common Function Attributes. 68382 (line 277) 68383* contributors: Contributors. (line 6) 68384* copy function attribute: Common Function Attributes. 68385 (line 306) 68386* copy type attribute: Common Type Attributes. 68387 (line 161) 68388* copy variable attribute: Common Variable Attributes. 68389 (line 185) 68390* copysign: Other Builtins. (line 6) 68391* copysignf: Other Builtins. (line 6) 68392* copysignl: Other Builtins. (line 6) 68393* core dump: Bug Criteria. (line 9) 68394* cos: Other Builtins. (line 6) 68395* cosf: Other Builtins. (line 6) 68396* cosh: Other Builtins. (line 6) 68397* coshf: Other Builtins. (line 6) 68398* coshl: Other Builtins. (line 6) 68399* cosl: Other Builtins. (line 6) 68400* CPATH: Environment Variables. 68401 (line 144) 68402* CPLUS_INCLUDE_PATH: Environment Variables. 68403 (line 146) 68404* cpow: Other Builtins. (line 6) 68405* cpowf: Other Builtins. (line 6) 68406* cpowl: Other Builtins. (line 6) 68407* cproj: Other Builtins. (line 6) 68408* cprojf: Other Builtins. (line 6) 68409* cprojl: Other Builtins. (line 6) 68410* cpu= function attribute, AArch64: AArch64 Function Attributes. 68411 (line 63) 68412* CR16 Options: CR16 Options. (line 6) 68413* creal: Other Builtins. (line 6) 68414* crealf: Other Builtins. (line 6) 68415* creall: Other Builtins. (line 6) 68416* CRIS Options: CRIS Options. (line 6) 68417* critical function attribute, MSP430: MSP430 Function Attributes. 68418 (line 9) 68419* cross compiling: Invoking GCC. (line 24) 68420* csin: Other Builtins. (line 6) 68421* csinf: Other Builtins. (line 6) 68422* csinh: Other Builtins. (line 6) 68423* csinhf: Other Builtins. (line 6) 68424* csinhl: Other Builtins. (line 6) 68425* csinl: Other Builtins. (line 6) 68426* csqrt: Other Builtins. (line 6) 68427* csqrtf: Other Builtins. (line 6) 68428* csqrtl: Other Builtins. (line 6) 68429* ctan: Other Builtins. (line 6) 68430* ctanf: Other Builtins. (line 6) 68431* ctanh: Other Builtins. (line 6) 68432* ctanhf: Other Builtins. (line 6) 68433* ctanhl: Other Builtins. (line 6) 68434* ctanl: Other Builtins. (line 6) 68435* CXX_MODULE_MAPPER environment variable: C++ Dialect Options. 68436 (line 340) 68437* C_INCLUDE_PATH: Environment Variables. 68438 (line 145) 68439* D: G++ and GCC. (line 6) 68440* Darwin options: Darwin Options. (line 6) 68441* dcgettext: Other Builtins. (line 6) 68442* dd integer suffix: Decimal Float. (line 6) 68443* DD integer suffix: Decimal Float. (line 6) 68444* deallocating variable length arrays: Variable Length. (line 22) 68445* debug dump options: Developer Options. (line 6) 68446* debugging GCC: Developer Options. (line 6) 68447* debugging information options: Debugging Options. (line 6) 68448* decimal floating types: Decimal Float. (line 6) 68449* declaration scope: Incompatibilities. (line 80) 68450* declarations inside expressions: Statement Exprs. (line 6) 68451* declarations, mixed with code: Mixed Labels and Declarations. 68452 (line 6) 68453* declaring attributes of functions: Function Attributes. 68454 (line 6) 68455* declaring static data in C++: Static Definitions. (line 6) 68456* defining static data in C++: Static Definitions. (line 6) 68457* dependencies for make as output: Environment Variables. 68458 (line 173) 68459* dependencies for make as output <1>: Environment Variables. 68460 (line 189) 68461* dependencies, make: Preprocessor Options. 68462 (line 77) 68463* DEPENDENCIES_OUTPUT: Environment Variables. 68464 (line 172) 68465* dependent name lookup: Name lookup. (line 6) 68466* deprecated enumerator attribute: Enumerator Attributes. 68467 (line 28) 68468* deprecated function attribute: Common Function Attributes. 68469 (line 338) 68470* deprecated type attribute: Common Type Attributes. 68471 (line 189) 68472* deprecated variable attribute: Common Variable Attributes. 68473 (line 201) 68474* designated initializers: Designated Inits. (line 6) 68475* designated_init type attribute: Common Type Attributes. 68476 (line 234) 68477* designator lists: Designated Inits. (line 96) 68478* designators: Designated Inits. (line 64) 68479* destructor function attribute: Common Function Attributes. 68480 (line 277) 68481* developer options: Developer Options. (line 6) 68482* df integer suffix: Decimal Float. (line 6) 68483* DF integer suffix: Decimal Float. (line 6) 68484* dgettext: Other Builtins. (line 6) 68485* diagnostic messages: Diagnostic Message Formatting Options. 68486 (line 6) 68487* dialect options: C Dialect Options. (line 6) 68488* diff-delete GCC_COLORS capability: Diagnostic Message Formatting Options. 68489 (line 134) 68490* diff-filename GCC_COLORS capability: Diagnostic Message Formatting Options. 68491 (line 127) 68492* diff-hunk GCC_COLORS capability: Diagnostic Message Formatting Options. 68493 (line 130) 68494* diff-insert GCC_COLORS capability: Diagnostic Message Formatting Options. 68495 (line 137) 68496* digits in constraint: Simple Constraints. (line 125) 68497* directory options: Directory Options. (line 6) 68498* disinterrupt function attribute, Epiphany: Epiphany Function Attributes. 68499 (line 9) 68500* disinterrupt function attribute, MeP: MeP Function Attributes. 68501 (line 9) 68502* dl integer suffix: Decimal Float. (line 6) 68503* DL integer suffix: Decimal Float. (line 6) 68504* dllexport function attribute: Microsoft Windows Function Attributes. 68505 (line 10) 68506* dllexport variable attribute: Microsoft Windows Variable Attributes. 68507 (line 12) 68508* dllimport function attribute: Microsoft Windows Function Attributes. 68509 (line 42) 68510* dllimport variable attribute: Microsoft Windows Variable Attributes. 68511 (line 12) 68512* dollar signs in identifier names: Dollar Signs. (line 6) 68513* double-word arithmetic: Long Long. (line 6) 68514* downward funargs: Nested Functions. (line 6) 68515* drem: Other Builtins. (line 6) 68516* dremf: Other Builtins. (line 6) 68517* dreml: Other Builtins. (line 6) 68518* dump options: Developer Options. (line 6) 68519* E in constraint: Simple Constraints. (line 87) 68520* earlyclobber operand: Modifiers. (line 25) 68521* eBPF Options: eBPF Options. (line 6) 68522* eight-bit data on the H8/300, H8/300H, and H8S: H8/300 Variable Attributes. 68523 (line 9) 68524* eightbit_data variable attribute, H8/300: H8/300 Variable Attributes. 68525 (line 9) 68526* EIND: AVR Options. (line 319) 68527* either function attribute, MSP430: MSP430 Function Attributes. 68528 (line 57) 68529* either variable attribute, MSP430: MSP430 Variable Attributes. 68530 (line 8) 68531* empty structures: Empty Structures. (line 6) 68532* Enumerator Attributes: Enumerator Attributes. 68533 (line 6) 68534* environment variables: Environment Variables. 68535 (line 6) 68536* erf: Other Builtins. (line 6) 68537* erfc: Other Builtins. (line 6) 68538* erfcf: Other Builtins. (line 6) 68539* erfcl: Other Builtins. (line 6) 68540* erff: Other Builtins. (line 6) 68541* erfl: Other Builtins. (line 6) 68542* error function attribute: Common Function Attributes. 68543 (line 373) 68544* error GCC_COLORS capability: Diagnostic Message Formatting Options. 68545 (line 92) 68546* error messages: Warnings and Errors. 68547 (line 6) 68548* escaped newlines: Escaped Newlines. (line 6) 68549* exception function attribute: NDS32 Function Attributes. 68550 (line 9) 68551* exception handler functions, Blackfin: Blackfin Function Attributes. 68552 (line 9) 68553* exception handler functions, NDS32: NDS32 Function Attributes. 68554 (line 9) 68555* exception_handler function attribute: Blackfin Function Attributes. 68556 (line 9) 68557* exit: Other Builtins. (line 6) 68558* exp: Other Builtins. (line 6) 68559* exp10: Other Builtins. (line 6) 68560* exp10f: Other Builtins. (line 6) 68561* exp10l: Other Builtins. (line 6) 68562* exp2: Other Builtins. (line 6) 68563* exp2f: Other Builtins. (line 6) 68564* exp2l: Other Builtins. (line 6) 68565* expf: Other Builtins. (line 6) 68566* expl: Other Builtins. (line 6) 68567* explicit register variables: Explicit Register Variables. 68568 (line 6) 68569* expm1: Other Builtins. (line 6) 68570* expm1f: Other Builtins. (line 6) 68571* expm1l: Other Builtins. (line 6) 68572* expressions containing statements: Statement Exprs. (line 6) 68573* expressions, constructor: Compound Literals. (line 6) 68574* extended asm: Extended Asm. (line 6) 68575* extensible constraints: Simple Constraints. (line 161) 68576* extensions, ?:: Conditionals. (line 6) 68577* extensions, C language: C Extensions. (line 6) 68578* extensions, C++ language: C++ Extensions. (line 6) 68579* external declaration scope: Incompatibilities. (line 80) 68580* externally_visible function attribute: Common Function Attributes. 68581 (line 390) 68582* extra NOP instructions at the function entry point: Common Function Attributes. 68583 (line 1052) 68584* F in constraint: Simple Constraints. (line 92) 68585* fabs: Other Builtins. (line 6) 68586* fabsf: Other Builtins. (line 6) 68587* fabsl: Other Builtins. (line 6) 68588* fallthrough statement attribute: Statement Attributes. 68589 (line 26) 68590* far function attribute, MeP: MeP Function Attributes. 68591 (line 25) 68592* far function attribute, MIPS: MIPS Function Attributes. 68593 (line 63) 68594* far type attribute, MeP: MeP Type Attributes. 68595 (line 6) 68596* far variable attribute, MeP: MeP Variable Attributes. 68597 (line 30) 68598* fastcall function attribute, x86-32: x86 Function Attributes. 68599 (line 15) 68600* fast_interrupt function attribute, M32C: M32C Function Attributes. 68601 (line 14) 68602* fast_interrupt function attribute, MicroBlaze: MicroBlaze Function Attributes. 68603 (line 27) 68604* fast_interrupt function attribute, RX: RX Function Attributes. 68605 (line 9) 68606* fatal signal: Bug Criteria. (line 9) 68607* fdim: Other Builtins. (line 6) 68608* fdimf: Other Builtins. (line 6) 68609* fdiml: Other Builtins. (line 6) 68610* FDL, GNU Free Documentation License: GNU Free Documentation License. 68611 (line 6) 68612* fentry_name function attribute, x86: x86 Function Attributes. 68613 (line 701) 68614* fentry_section function attribute, x86: x86 Function Attributes. 68615 (line 707) 68616* ffs: Other Builtins. (line 6) 68617* file name suffix: Overall Options. (line 14) 68618* file names: Link Options. (line 10) 68619* fix-cortex-a53-835769 function attribute, AArch64: AArch64 Function Attributes. 68620 (line 19) 68621* fixed-point types: Fixed-Point. (line 6) 68622* fixit-delete GCC_COLORS capability: Diagnostic Message Formatting Options. 68623 (line 124) 68624* fixit-insert GCC_COLORS capability: Diagnostic Message Formatting Options. 68625 (line 120) 68626* flatten function attribute: Common Function Attributes. 68627 (line 403) 68628* flexible array members: Zero Length. (line 6) 68629* float as function value type: Incompatibilities. (line 141) 68630* floating point precision: Disappointments. (line 68) 68631* floating-point precision: Optimize Options. (line 2289) 68632* floor: Other Builtins. (line 6) 68633* floorf: Other Builtins. (line 6) 68634* floorl: Other Builtins. (line 6) 68635* fma: Other Builtins. (line 6) 68636* fmaf: Other Builtins. (line 6) 68637* fmal: Other Builtins. (line 6) 68638* fmax: Other Builtins. (line 6) 68639* fmaxf: Other Builtins. (line 6) 68640* fmaxl: Other Builtins. (line 6) 68641* fmin: Other Builtins. (line 6) 68642* fminf: Other Builtins. (line 6) 68643* fminl: Other Builtins. (line 6) 68644* fmod: Other Builtins. (line 6) 68645* fmodf: Other Builtins. (line 6) 68646* fmodl: Other Builtins. (line 6) 68647* force_align_arg_pointer function attribute, x86: x86 Function Attributes. 68648 (line 100) 68649* format function attribute: Common Function Attributes. 68650 (line 411) 68651* format_arg function attribute: Common Function Attributes. 68652 (line 476) 68653* Fortran: G++ and GCC. (line 6) 68654* forwarder_section function attribute, Epiphany: Epiphany Function Attributes. 68655 (line 13) 68656* forwarding calls: Constructing Calls. (line 6) 68657* fprintf: Other Builtins. (line 6) 68658* fprintf_unlocked: Other Builtins. (line 6) 68659* fputs: Other Builtins. (line 6) 68660* fputs_unlocked: Other Builtins. (line 6) 68661* FR30 Options: FR30 Options. (line 6) 68662* free: Other Builtins. (line 6) 68663* freestanding environment: Standards. (line 13) 68664* freestanding implementation: Standards. (line 13) 68665* frexp: Other Builtins. (line 6) 68666* frexpf: Other Builtins. (line 6) 68667* frexpl: Other Builtins. (line 6) 68668* FRV Options: FRV Options. (line 6) 68669* fscanf: Other Builtins. (line 6) 68670* fscanf, and constant strings: Incompatibilities. (line 17) 68671* FT32 Options: FT32 Options. (line 6) 68672* function addressability on the M32R/D: M32R/D Function Attributes. 68673 (line 15) 68674* function attributes: Function Attributes. 68675 (line 6) 68676* function pointers, arithmetic: Pointer Arith. (line 6) 68677* function prototype declarations: Function Prototypes. 68678 (line 6) 68679* function versions: Function Multiversioning. 68680 (line 6) 68681* function, size of pointer to: Pointer Arith. (line 6) 68682* functions in arbitrary sections: Common Function Attributes. 68683 (line 1136) 68684* functions that are dynamically resolved: Common Function Attributes. 68685 (line 577) 68686* functions that are passed arguments in registers on x86-32: x86 Function Attributes. 68687 (line 76) 68688* functions that behave like malloc: Common Function Attributes. 68689 (line 706) 68690* functions that have no side effects: Common Function Attributes. 68691 (line 236) 68692* functions that have no side effects <1>: Common Function Attributes. 68693 (line 1070) 68694* functions that never return: Common Function Attributes. 68695 (line 982) 68696* functions that pop the argument stack on x86-32: x86 Function Attributes. 68697 (line 9) 68698* functions that pop the argument stack on x86-32 <1>: x86 Function Attributes. 68699 (line 15) 68700* functions that pop the argument stack on x86-32 <2>: x86 Function Attributes. 68701 (line 23) 68702* functions that pop the argument stack on x86-32 <3>: x86 Function Attributes. 68703 (line 108) 68704* functions that return more than once: Common Function Attributes. 68705 (line 1127) 68706* functions with non-null pointer arguments: Common Function Attributes. 68707 (line 911) 68708* functions with printf, scanf, strftime or strfmon style arguments: Common Function Attributes. 68709 (line 411) 68710* function_return function attribute, x86: x86 Function Attributes. 68711 (line 628) 68712* function_vector function attribute, H8/300: H8/300 Function Attributes. 68713 (line 9) 68714* function_vector function attribute, M16C/M32C: M32C Function Attributes. 68715 (line 20) 68716* function_vector function attribute, SH: SH Function Attributes. 68717 (line 9) 68718* G in constraint: Simple Constraints. (line 96) 68719* g in constraint: Simple Constraints. (line 118) 68720* g++: Invoking G++. (line 14) 68721* G++: G++ and GCC. (line 29) 68722* gamma: Other Builtins. (line 6) 68723* gammaf: Other Builtins. (line 6) 68724* gammaf_r: Other Builtins. (line 6) 68725* gammal: Other Builtins. (line 6) 68726* gammal_r: Other Builtins. (line 6) 68727* gamma_r: Other Builtins. (line 6) 68728* GCC: G++ and GCC. (line 6) 68729* GCC command options: Invoking GCC. (line 6) 68730* GCC_COLORS environment variable: Diagnostic Message Formatting Options. 68731 (line 55) 68732* GCC_COMPARE_DEBUG: Environment Variables. 68733 (line 52) 68734* GCC_EXEC_PREFIX: Environment Variables. 68735 (line 57) 68736* GCC_EXTRA_DIAGNOSTIC_OUTPUT: Environment Variables. 68737 (line 125) 68738* gcc_struct type attribute, PowerPC: PowerPC Type Attributes. 68739 (line 9) 68740* gcc_struct type attribute, x86: x86 Type Attributes. 68741 (line 11) 68742* gcc_struct variable attribute, PowerPC: PowerPC Variable Attributes. 68743 (line 9) 68744* gcc_struct variable attribute, x86: x86 Variable Attributes. 68745 (line 11) 68746* GCC_URLS environment variable: Diagnostic Message Formatting Options. 68747 (line 144) 68748* gcov: Instrumentation Options. 68749 (line 55) 68750* general-regs-only function attribute, AArch64: AArch64 Function Attributes. 68751 (line 12) 68752* general-regs-only function attribute, ARM: ARM Function Attributes. 68753 (line 9) 68754* gettext: Other Builtins. (line 6) 68755* global offset table: Code Gen Options. (line 356) 68756* global register after longjmp: Global Register Variables. 68757 (line 92) 68758* global register variables: Global Register Variables. 68759 (line 6) 68760* GNAT: G++ and GCC. (line 29) 68761* GNU C Compiler: G++ and GCC. (line 6) 68762* GNU Compiler Collection: G++ and GCC. (line 6) 68763* gnu_inline function attribute: Common Function Attributes. 68764 (line 531) 68765* Go: G++ and GCC. (line 6) 68766* goto with computed label: Labels as Values. (line 6) 68767* gprof: Instrumentation Options. 68768 (line 18) 68769* grouping options: Invoking GCC. (line 31) 68770* H in constraint: Simple Constraints. (line 96) 68771* half-precision floating point: Half-Precision. (line 6) 68772* hardware models and configurations, specifying: Submodel Options. 68773 (line 6) 68774* hex floats: Hex Floats. (line 6) 68775* highlight, color: Diagnostic Message Formatting Options. 68776 (line 55) 68777* hk fixed-suffix: Fixed-Point. (line 6) 68778* HK fixed-suffix: Fixed-Point. (line 6) 68779* hosted environment: Standards. (line 13) 68780* hosted environment <1>: C Dialect Options. (line 278) 68781* hosted environment <2>: C Dialect Options. (line 327) 68782* hosted implementation: Standards. (line 13) 68783* hot function attribute: Common Function Attributes. 68784 (line 567) 68785* hot label attribute: Label Attributes. (line 39) 68786* hotpatch function attribute, S/390: S/390 Function Attributes. 68787 (line 9) 68788* HPPA Options: HPPA Options. (line 6) 68789* hr fixed-suffix: Fixed-Point. (line 6) 68790* HR fixed-suffix: Fixed-Point. (line 6) 68791* hypot: Other Builtins. (line 6) 68792* hypotf: Other Builtins. (line 6) 68793* hypotl: Other Builtins. (line 6) 68794* i in constraint: Simple Constraints. (line 68) 68795* I in constraint: Simple Constraints. (line 79) 68796* IA-64 Options: IA-64 Options. (line 6) 68797* IBM RS/6000 and PowerPC Options: RS/6000 and PowerPC Options. 68798 (line 6) 68799* identifier names, dollar signs in: Dollar Signs. (line 6) 68800* identifiers, names in assembler code: Asm Labels. (line 6) 68801* ifunc function attribute: Common Function Attributes. 68802 (line 577) 68803* ilogb: Other Builtins. (line 6) 68804* ilogbf: Other Builtins. (line 6) 68805* ilogbl: Other Builtins. (line 6) 68806* imaxabs: Other Builtins. (line 6) 68807* implementation-defined behavior, C language: C Implementation. 68808 (line 6) 68809* implementation-defined behavior, C++ language: C++ Implementation. 68810 (line 6) 68811* implied #pragma implementation: C++ Interface. (line 43) 68812* incompatibilities of GCC: Incompatibilities. (line 6) 68813* increment operators: Bug Criteria. (line 17) 68814* index: Other Builtins. (line 6) 68815* indirect calls, ARC: ARC Function Attributes. 68816 (line 27) 68817* indirect calls, ARM: ARM Function Attributes. 68818 (line 38) 68819* indirect calls, Blackfin: Blackfin Function Attributes. 68820 (line 38) 68821* indirect calls, Epiphany: Epiphany Function Attributes. 68822 (line 57) 68823* indirect calls, MIPS: MIPS Function Attributes. 68824 (line 63) 68825* indirect calls, PowerPC: PowerPC Function Attributes. 68826 (line 10) 68827* indirect functions: Common Function Attributes. 68828 (line 577) 68829* indirect_branch function attribute, x86: x86 Function Attributes. 68830 (line 619) 68831* indirect_return function attribute, x86: x86 Function Attributes. 68832 (line 695) 68833* initializations in expressions: Compound Literals. (line 6) 68834* initializers with labeled elements: Designated Inits. (line 6) 68835* initializers, non-constant: Initializers. (line 6) 68836* init_priority variable attribute: C++ Attributes. (line 50) 68837* inline assembly language: Using Assembly Language with C. 68838 (line 6) 68839* inline automatic for C++ member fns: Inline. (line 68) 68840* inline functions: Inline. (line 6) 68841* inline functions, omission of: Inline. (line 51) 68842* inlining and C++ pragmas: C++ Interface. (line 57) 68843* installation trouble: Trouble. (line 6) 68844* instrumentation options: Instrumentation Options. 68845 (line 6) 68846* integrating function code: Inline. (line 6) 68847* interface and implementation headers, C++: C++ Interface. (line 6) 68848* intermediate C version, nonexistent: G++ and GCC. (line 34) 68849* interrupt function attribute, ARC: ARC Function Attributes. 68850 (line 9) 68851* interrupt function attribute, ARM: ARM Function Attributes. 68852 (line 16) 68853* interrupt function attribute, AVR: AVR Function Attributes. 68854 (line 9) 68855* interrupt function attribute, C-SKY: C-SKY Function Attributes. 68856 (line 10) 68857* interrupt function attribute, CR16: CR16 Function Attributes. 68858 (line 9) 68859* interrupt function attribute, Epiphany: Epiphany Function Attributes. 68860 (line 20) 68861* interrupt function attribute, M32C: M32C Function Attributes. 68862 (line 53) 68863* interrupt function attribute, M32R/D: M32R/D Function Attributes. 68864 (line 9) 68865* interrupt function attribute, m68k: m68k Function Attributes. 68866 (line 10) 68867* interrupt function attribute, MeP: MeP Function Attributes. 68868 (line 14) 68869* interrupt function attribute, MIPS: MIPS Function Attributes. 68870 (line 9) 68871* interrupt function attribute, MSP430: MSP430 Function Attributes. 68872 (line 19) 68873* interrupt function attribute, NDS32: NDS32 Function Attributes. 68874 (line 14) 68875* interrupt function attribute, RISC-V: RISC-V Function Attributes. 68876 (line 19) 68877* interrupt function attribute, RL78: RL78 Function Attributes. 68878 (line 10) 68879* interrupt function attribute, RX: RX Function Attributes. 68880 (line 15) 68881* interrupt function attribute, V850: V850 Function Attributes. 68882 (line 10) 68883* interrupt function attribute, Visium: Visium Function Attributes. 68884 (line 9) 68885* interrupt function attribute, x86: x86 Function Attributes. 68886 (line 124) 68887* interrupt function attribute, Xstormy16: Xstormy16 Function Attributes. 68888 (line 9) 68889* interrupt_handler function attribute, Blackfin: Blackfin Function Attributes. 68890 (line 15) 68891* interrupt_handler function attribute, H8/300: H8/300 Function Attributes. 68892 (line 17) 68893* interrupt_handler function attribute, m68k: m68k Function Attributes. 68894 (line 10) 68895* interrupt_handler function attribute, MicroBlaze: MicroBlaze Function Attributes. 68896 (line 27) 68897* interrupt_handler function attribute, SH: SH Function Attributes. 68898 (line 28) 68899* interrupt_handler function attribute, V850: V850 Function Attributes. 68900 (line 10) 68901* interrupt_thread function attribute, fido: m68k Function Attributes. 68902 (line 16) 68903* introduction: Top. (line 6) 68904* invalid assembly code: Bug Criteria. (line 12) 68905* invalid input: Bug Criteria. (line 42) 68906* invoking g++: Invoking G++. (line 22) 68907* io variable attribute, AVR: AVR Variable Attributes. 68908 (line 73) 68909* io variable attribute, MeP: MeP Variable Attributes. 68910 (line 36) 68911* io_low variable attribute, AVR: AVR Variable Attributes. 68912 (line 91) 68913* isalnum: Other Builtins. (line 6) 68914* isalpha: Other Builtins. (line 6) 68915* isascii: Other Builtins. (line 6) 68916* isblank: Other Builtins. (line 6) 68917* iscntrl: Other Builtins. (line 6) 68918* isdigit: Other Builtins. (line 6) 68919* isgraph: Other Builtins. (line 6) 68920* islower: Other Builtins. (line 6) 68921* ISO 9899: Standards. (line 13) 68922* ISO C: Standards. (line 13) 68923* ISO C standard: Standards. (line 13) 68924* ISO C11: Standards. (line 13) 68925* ISO C17: Standards. (line 13) 68926* ISO C1X: Standards. (line 13) 68927* ISO C2X: Standards. (line 13) 68928* ISO C90: Standards. (line 13) 68929* ISO C94: Standards. (line 13) 68930* ISO C95: Standards. (line 13) 68931* ISO C99: Standards. (line 13) 68932* ISO C9X: Standards. (line 13) 68933* ISO support: C Dialect Options. (line 10) 68934* ISO/IEC 9899: Standards. (line 13) 68935* isprint: Other Builtins. (line 6) 68936* ispunct: Other Builtins. (line 6) 68937* isr function attribute, ARM: ARM Function Attributes. 68938 (line 33) 68939* isr function attribute, C-SKY: C-SKY Function Attributes. 68940 (line 10) 68941* isspace: Other Builtins. (line 6) 68942* isupper: Other Builtins. (line 6) 68943* iswalnum: Other Builtins. (line 6) 68944* iswalpha: Other Builtins. (line 6) 68945* iswblank: Other Builtins. (line 6) 68946* iswcntrl: Other Builtins. (line 6) 68947* iswdigit: Other Builtins. (line 6) 68948* iswgraph: Other Builtins. (line 6) 68949* iswlower: Other Builtins. (line 6) 68950* iswprint: Other Builtins. (line 6) 68951* iswpunct: Other Builtins. (line 6) 68952* iswspace: Other Builtins. (line 6) 68953* iswupper: Other Builtins. (line 6) 68954* iswxdigit: Other Builtins. (line 6) 68955* isxdigit: Other Builtins. (line 6) 68956* j0: Other Builtins. (line 6) 68957* j0f: Other Builtins. (line 6) 68958* j0l: Other Builtins. (line 6) 68959* j1: Other Builtins. (line 6) 68960* j1f: Other Builtins. (line 6) 68961* j1l: Other Builtins. (line 6) 68962* jli_always function attribute, ARC: ARC Function Attributes. 68963 (line 44) 68964* jli_fixed function attribute, ARC: ARC Function Attributes. 68965 (line 50) 68966* jn: Other Builtins. (line 6) 68967* jnf: Other Builtins. (line 6) 68968* jnl: Other Builtins. (line 6) 68969* k fixed-suffix: Fixed-Point. (line 6) 68970* K fixed-suffix: Fixed-Point. (line 6) 68971* keep_interrupts_masked function attribute, MIPS: MIPS Function Attributes. 68972 (line 34) 68973* kernel attribute, Nvidia PTX: Nvidia PTX Function Attributes. 68974 (line 9) 68975* kernel helper, function attribute, BPF: BPF Function Attributes. 68976 (line 9) 68977* keywords, alternate: Alternate Keywords. (line 6) 68978* known causes of trouble: Trouble. (line 6) 68979* kspisusp function attribute, Blackfin: Blackfin Function Attributes. 68980 (line 21) 68981* l1_data variable attribute, Blackfin: Blackfin Variable Attributes. 68982 (line 11) 68983* l1_data_A variable attribute, Blackfin: Blackfin Variable Attributes. 68984 (line 11) 68985* l1_data_B variable attribute, Blackfin: Blackfin Variable Attributes. 68986 (line 11) 68987* l1_text function attribute, Blackfin: Blackfin Function Attributes. 68988 (line 26) 68989* l2 function attribute, Blackfin: Blackfin Function Attributes. 68990 (line 32) 68991* l2 variable attribute, Blackfin: Blackfin Variable Attributes. 68992 (line 19) 68993* Label Attributes: Label Attributes. (line 6) 68994* labeled elements in initializers: Designated Inits. (line 6) 68995* labels as values: Labels as Values. (line 6) 68996* labs: Other Builtins. (line 6) 68997* LANG: Environment Variables. 68998 (line 21) 68999* LANG <1>: Environment Variables. 69000 (line 106) 69001* language dialect options: C Dialect Options. (line 6) 69002* LC_ALL: Environment Variables. 69003 (line 21) 69004* LC_CTYPE: Environment Variables. 69005 (line 21) 69006* LC_MESSAGES: Environment Variables. 69007 (line 21) 69008* ldexp: Other Builtins. (line 6) 69009* ldexpf: Other Builtins. (line 6) 69010* ldexpl: Other Builtins. (line 6) 69011* leaf function attribute: Common Function Attributes. 69012 (line 667) 69013* length-zero arrays: Zero Length. (line 6) 69014* lgamma: Other Builtins. (line 6) 69015* lgammaf: Other Builtins. (line 6) 69016* lgammaf_r: Other Builtins. (line 6) 69017* lgammal: Other Builtins. (line 6) 69018* lgammal_r: Other Builtins. (line 6) 69019* lgamma_r: Other Builtins. (line 6) 69020* Libraries: Link Options. (line 85) 69021* LIBRARY_PATH: Environment Variables. 69022 (line 97) 69023* link options: Link Options. (line 6) 69024* linker script: Link Options. (line 314) 69025* lk fixed-suffix: Fixed-Point. (line 6) 69026* LK fixed-suffix: Fixed-Point. (line 6) 69027* LL integer suffix: Long Long. (line 6) 69028* llabs: Other Builtins. (line 6) 69029* llk fixed-suffix: Fixed-Point. (line 6) 69030* LLK fixed-suffix: Fixed-Point. (line 6) 69031* llr fixed-suffix: Fixed-Point. (line 6) 69032* LLR fixed-suffix: Fixed-Point. (line 6) 69033* llrint: Other Builtins. (line 6) 69034* llrintf: Other Builtins. (line 6) 69035* llrintl: Other Builtins. (line 6) 69036* llround: Other Builtins. (line 6) 69037* llroundf: Other Builtins. (line 6) 69038* llroundl: Other Builtins. (line 6) 69039* LM32 options: LM32 Options. (line 6) 69040* load address instruction: Simple Constraints. (line 152) 69041* local labels: Local Labels. (line 6) 69042* local variables in macros: Typeof. (line 46) 69043* local variables, specifying registers: Local Register Variables. 69044 (line 6) 69045* locale: Environment Variables. 69046 (line 21) 69047* locale definition: Environment Variables. 69048 (line 106) 69049* locus GCC_COLORS capability: Diagnostic Message Formatting Options. 69050 (line 113) 69051* log: Other Builtins. (line 6) 69052* log10: Other Builtins. (line 6) 69053* log10f: Other Builtins. (line 6) 69054* log10l: Other Builtins. (line 6) 69055* log1p: Other Builtins. (line 6) 69056* log1pf: Other Builtins. (line 6) 69057* log1pl: Other Builtins. (line 6) 69058* log2: Other Builtins. (line 6) 69059* log2f: Other Builtins. (line 6) 69060* log2l: Other Builtins. (line 6) 69061* logb: Other Builtins. (line 6) 69062* logbf: Other Builtins. (line 6) 69063* logbl: Other Builtins. (line 6) 69064* logf: Other Builtins. (line 6) 69065* logl: Other Builtins. (line 6) 69066* long: BPF Built-in Functions. 69067 (line 8) 69068* long <1>: BPF Built-in Functions. 69069 (line 13) 69070* long <2>: BPF Built-in Functions. 69071 (line 18) 69072* long long data types: Long Long. (line 6) 69073* longcall function attribute, Blackfin: Blackfin Function Attributes. 69074 (line 38) 69075* longcall function attribute, PowerPC: PowerPC Function Attributes. 69076 (line 10) 69077* longjmp: Global Register Variables. 69078 (line 92) 69079* longjmp incompatibilities: Incompatibilities. (line 39) 69080* longjmp warnings: Warning Options. (line 1375) 69081* long_call function attribute, ARC: ARC Function Attributes. 69082 (line 27) 69083* long_call function attribute, ARM: ARM Function Attributes. 69084 (line 38) 69085* long_call function attribute, Epiphany: Epiphany Function Attributes. 69086 (line 57) 69087* long_call function attribute, MIPS: MIPS Function Attributes. 69088 (line 63) 69089* LoongArch Options: LoongArch Options. (line 6) 69090* lower function attribute, MSP430: MSP430 Function Attributes. 69091 (line 57) 69092* lower variable attribute, MSP430: MSP430 Variable Attributes. 69093 (line 12) 69094* lr fixed-suffix: Fixed-Point. (line 6) 69095* LR fixed-suffix: Fixed-Point. (line 6) 69096* lrint: Other Builtins. (line 6) 69097* lrintf: Other Builtins. (line 6) 69098* lrintl: Other Builtins. (line 6) 69099* lround: Other Builtins. (line 6) 69100* lroundf: Other Builtins. (line 6) 69101* lroundl: Other Builtins. (line 6) 69102* m in constraint: Simple Constraints. (line 17) 69103* M32C options: M32C Options. (line 6) 69104* M32R/D options: M32R/D Options. (line 6) 69105* M680x0 options: M680x0 Options. (line 6) 69106* machine specific constraints: Machine Constraints. 69107 (line 6) 69108* machine-dependent options: Submodel Options. (line 6) 69109* macro with variable arguments: Variadic Macros. (line 6) 69110* macros, inline alternative: Inline. (line 6) 69111* macros, local labels: Local Labels. (line 6) 69112* macros, local variables in: Typeof. (line 46) 69113* macros, statements in expressions: Statement Exprs. (line 6) 69114* macros, types of arguments: Typeof. (line 6) 69115* make: Preprocessor Options. 69116 (line 77) 69117* malloc: Other Builtins. (line 6) 69118* malloc function attribute: Common Function Attributes. 69119 (line 706) 69120* matching constraint: Simple Constraints. (line 137) 69121* may_alias type attribute: Common Type Attributes. 69122 (line 245) 69123* MCore options: MCore Options. (line 6) 69124* medium_call function attribute, ARC: ARC Function Attributes. 69125 (line 27) 69126* member fns, automatically inline: Inline. (line 68) 69127* memchr: Other Builtins. (line 6) 69128* memcmp: Other Builtins. (line 6) 69129* memcpy: Other Builtins. (line 6) 69130* memory references in constraints: Simple Constraints. (line 17) 69131* mempcpy: Other Builtins. (line 6) 69132* memset: Other Builtins. (line 6) 69133* MeP options: MeP Options. (line 6) 69134* Mercury: G++ and GCC. (line 23) 69135* message formatting: Diagnostic Message Formatting Options. 69136 (line 6) 69137* messages, warning: Warning Options. (line 6) 69138* messages, warning and error: Warnings and Errors. 69139 (line 6) 69140* MicroBlaze Options: MicroBlaze Options. (line 6) 69141* micromips function attribute: MIPS Function Attributes. 69142 (line 91) 69143* middle-operands, omitted: Conditionals. (line 6) 69144* MIPS options: MIPS Options. (line 6) 69145* mips16 function attribute, MIPS: MIPS Function Attributes. 69146 (line 75) 69147* misunderstandings in C++: C++ Misunderstandings. 69148 (line 6) 69149* mixed declarations and code: Mixed Labels and Declarations. 69150 (line 6) 69151* mixing assembly language and C: Using Assembly Language with C. 69152 (line 6) 69153* mktemp, and constant strings: Incompatibilities. (line 13) 69154* MMIX Options: MMIX Options. (line 6) 69155* MN10300 options: MN10300 Options. (line 6) 69156* mode type attribute: Common Type Attributes. 69157 (line 281) 69158* mode variable attribute: Common Variable Attributes. 69159 (line 240) 69160* model function attribute, M32R/D: M32R/D Function Attributes. 69161 (line 15) 69162* model variable attribute, IA-64: IA-64 Variable Attributes. 69163 (line 9) 69164* model-name variable attribute, M32R/D: M32R/D Variable Attributes. 69165 (line 9) 69166* modf: Other Builtins. (line 6) 69167* modff: Other Builtins. (line 6) 69168* modfl: Other Builtins. (line 6) 69169* modifiers in constraints: Modifiers. (line 6) 69170* Moxie Options: Moxie Options. (line 6) 69171* MSP430 Options: MSP430 Options. (line 6) 69172* ms_abi function attribute, x86: x86 Function Attributes. 69173 (line 34) 69174* ms_hook_prologue function attribute, x86: x86 Function Attributes. 69175 (line 59) 69176* ms_struct type attribute, PowerPC: PowerPC Type Attributes. 69177 (line 9) 69178* ms_struct type attribute, x86: x86 Type Attributes. 69179 (line 11) 69180* ms_struct variable attribute, PowerPC: PowerPC Variable Attributes. 69181 (line 9) 69182* ms_struct variable attribute, x86: x86 Variable Attributes. 69183 (line 11) 69184* multiple alternative constraints: Multi-Alternative. (line 6) 69185* multiprecision arithmetic: Long Long. (line 6) 69186* n in constraint: Simple Constraints. (line 73) 69187* naked function attribute, ARC: ARC Function Attributes. 69188 (line 59) 69189* naked function attribute, ARM: ARM Function Attributes. 69190 (line 48) 69191* naked function attribute, AVR: AVR Function Attributes. 69192 (line 23) 69193* naked function attribute, C-SKY: C-SKY Function Attributes. 69194 (line 20) 69195* naked function attribute, MCORE: MCORE Function Attributes. 69196 (line 9) 69197* naked function attribute, MSP430: MSP430 Function Attributes. 69198 (line 34) 69199* naked function attribute, NDS32: NDS32 Function Attributes. 69200 (line 35) 69201* naked function attribute, RISC-V: RISC-V Function Attributes. 69202 (line 9) 69203* naked function attribute, RL78: RL78 Function Attributes. 69204 (line 20) 69205* naked function attribute, RX: RX Function Attributes. 69206 (line 39) 69207* naked function attribute, x86: x86 Function Attributes. 69208 (line 66) 69209* Named Address Spaces: Named Address Spaces. 69210 (line 6) 69211* names used in assembler code: Asm Labels. (line 6) 69212* naming convention, implementation headers: C++ Interface. (line 43) 69213* NDS32 Options: NDS32 Options. (line 6) 69214* near function attribute, MeP: MeP Function Attributes. 69215 (line 20) 69216* near function attribute, MIPS: MIPS Function Attributes. 69217 (line 63) 69218* near type attribute, MeP: MeP Type Attributes. 69219 (line 6) 69220* near variable attribute, MeP: MeP Variable Attributes. 69221 (line 24) 69222* nearbyint: Other Builtins. (line 6) 69223* nearbyintf: Other Builtins. (line 6) 69224* nearbyintl: Other Builtins. (line 6) 69225* nested function attribute, NDS32: NDS32 Function Attributes. 69226 (line 19) 69227* nested functions: Nested Functions. (line 6) 69228* nested_ready function attribute, NDS32: NDS32 Function Attributes. 69229 (line 23) 69230* nesting function attribute, Blackfin: Blackfin Function Attributes. 69231 (line 45) 69232* newlines (escaped): Escaped Newlines. (line 6) 69233* nextafter: Other Builtins. (line 6) 69234* nextafterf: Other Builtins. (line 6) 69235* nextafterl: Other Builtins. (line 6) 69236* nexttoward: Other Builtins. (line 6) 69237* nexttowardf: Other Builtins. (line 6) 69238* nexttowardl: Other Builtins. (line 6) 69239* NFC: Warning Options. (line 3029) 69240* NFKC: Warning Options. (line 3029) 69241* Nios II options: Nios II Options. (line 6) 69242* nmi function attribute, NDS32: NDS32 Function Attributes. 69243 (line 50) 69244* NMI handler functions on the Blackfin processor: Blackfin Function Attributes. 69245 (line 50) 69246* nmi_handler function attribute, Blackfin: Blackfin Function Attributes. 69247 (line 50) 69248* nocf_check function attribute: x86 Function Attributes. 69249 (line 636) 69250* noclone function attribute: Common Function Attributes. 69251 (line 880) 69252* nocommon variable attribute: Common Variable Attributes. 69253 (line 176) 69254* nocompression function attribute, MIPS: MIPS Function Attributes. 69255 (line 108) 69256* nodirect_extern_access function attribute: x86 Function Attributes. 69257 (line 712) 69258* noinit variable attribute: Common Variable Attributes. 69259 (line 438) 69260* noinline function attribute: Common Function Attributes. 69261 (line 886) 69262* noipa function attribute: Common Function Attributes. 69263 (line 897) 69264* nomicromips function attribute: MIPS Function Attributes. 69265 (line 91) 69266* nomips16 function attribute, MIPS: MIPS Function Attributes. 69267 (line 75) 69268* non-constant initializers: Initializers. (line 6) 69269* non-static inline function: Inline. (line 82) 69270* nonlocal gotos: Nonlocal Gotos. (line 6) 69271* nonnull function attribute: Common Function Attributes. 69272 (line 911) 69273* nonstring variable attribute: Common Variable Attributes. 69274 (line 252) 69275* noplt function attribute: Common Function Attributes. 69276 (line 958) 69277* noreturn function attribute: Common Function Attributes. 69278 (line 982) 69279* nosave_low_regs function attribute, SH: SH Function Attributes. 69280 (line 34) 69281* note GCC_COLORS capability: Diagnostic Message Formatting Options. 69282 (line 98) 69283* nothrow function attribute: Common Function Attributes. 69284 (line 1016) 69285* notshared type attribute, ARM: ARM Type Attributes. 69286 (line 6) 69287* not_nested function attribute, NDS32: NDS32 Function Attributes. 69288 (line 21) 69289* no_caller_saved_registers function attribute, x86: x86 Function Attributes. 69290 (line 113) 69291* no_gccisr function attribute, AVR: AVR Function Attributes. 69292 (line 33) 69293* no_icf function attribute: Common Function Attributes. 69294 (line 810) 69295* no_instrument_function function attribute: Common Function Attributes. 69296 (line 814) 69297* no_profile_instrument_function function attribute: Common Function Attributes. 69298 (line 820) 69299* no_reorder function attribute: Common Function Attributes. 69300 (line 825) 69301* no_sanitize function attribute: Common Function Attributes. 69302 (line 833) 69303* no_sanitize_address function attribute: Common Function Attributes. 69304 (line 845) 69305* no_sanitize_coverage function attribute: Common Function Attributes. 69306 (line 864) 69307* no_sanitize_thread function attribute: Common Function Attributes. 69308 (line 853) 69309* no_sanitize_undefined function attribute: Common Function Attributes. 69310 (line 858) 69311* no_split_stack function attribute: Common Function Attributes. 69312 (line 869) 69313* no_stack_limit function attribute: Common Function Attributes. 69314 (line 875) 69315* no_stack_protector function attribute: Common Function Attributes. 69316 (line 1204) 69317* Nvidia PTX options: Nvidia PTX Options. (line 6) 69318* nvptx options: Nvidia PTX Options. (line 6) 69319* o in constraint: Simple Constraints. (line 23) 69320* OBJC_INCLUDE_PATH: Environment Variables. 69321 (line 147) 69322* objc_nullability variable attribute: Common Variable Attributes. 69323 (line 458) 69324* objc_root_class type attribute: Common Type Attributes. 69325 (line 475) 69326* Objective-C: G++ and GCC. (line 6) 69327* Objective-C <1>: Standards. (line 197) 69328* Objective-C and Objective-C++ options, command-line: Objective-C and Objective-C++ Dialect Options. 69329 (line 6) 69330* Objective-C++: G++ and GCC. (line 6) 69331* Objective-C++ <1>: Standards. (line 197) 69332* Offloading options: C Dialect Options. (line 373) 69333* Offloading targets: C Dialect Options. (line 360) 69334* offsettable address: Simple Constraints. (line 23) 69335* old-style function definitions: Function Prototypes. 69336 (line 6) 69337* omit-leaf-frame-pointer function attribute, AArch64: AArch64 Function Attributes. 69338 (line 41) 69339* omitted middle-operands: Conditionals. (line 6) 69340* open coding: Inline. (line 6) 69341* OpenACC accelerator programming: C Dialect Options. (line 388) 69342* OpenACC accelerator programming <1>: C Dialect Options. (line 397) 69343* OpenACC accelerator programming <2>: Warning Options. (line 3096) 69344* OpenACC offloading options: C Dialect Options. (line 373) 69345* OpenACC offloading targets: C Dialect Options. (line 360) 69346* OpenMP offloading options: C Dialect Options. (line 373) 69347* OpenMP offloading targets: C Dialect Options. (line 360) 69348* OpenMP parallel: C Dialect Options. (line 403) 69349* OpenMP SIMD: C Dialect Options. (line 412) 69350* OpenRISC Options: OpenRISC Options. (line 6) 69351* operand constraints, asm: Constraints. (line 6) 69352* optimize function attribute: Common Function Attributes. 69353 (line 1024) 69354* optimize options: Optimize Options. (line 6) 69355* options to control diagnostics formatting: Diagnostic Message Formatting Options. 69356 (line 6) 69357* options to control warnings: Warning Options. (line 6) 69358* options, C++: C++ Dialect Options. 69359 (line 6) 69360* options, code generation: Code Gen Options. (line 6) 69361* options, debugging: Debugging Options. (line 6) 69362* options, dialect: C Dialect Options. (line 6) 69363* options, directory search: Directory Options. (line 6) 69364* options, GCC command: Invoking GCC. (line 6) 69365* options, grouping: Invoking GCC. (line 31) 69366* options, linking: Link Options. (line 6) 69367* options, Objective-C and Objective-C++: Objective-C and Objective-C++ Dialect Options. 69368 (line 6) 69369* options, optimization: Optimize Options. (line 6) 69370* options, order: Invoking GCC. (line 35) 69371* options, preprocessor: Preprocessor Options. 69372 (line 6) 69373* options, profiling: Instrumentation Options. 69374 (line 6) 69375* options, program instrumentation: Instrumentation Options. 69376 (line 6) 69377* options, run-time error checking: Instrumentation Options. 69378 (line 6) 69379* order of evaluation, side effects: Non-bugs. (line 196) 69380* order of options: Invoking GCC. (line 35) 69381* OS_main function attribute, AVR: AVR Function Attributes. 69382 (line 56) 69383* OS_task function attribute, AVR: AVR Function Attributes. 69384 (line 56) 69385* other register constraints: Simple Constraints. (line 161) 69386* outline-atomics function attribute, AArch64: AArch64 Function Attributes. 69387 (line 82) 69388* output file option: Overall Options. (line 192) 69389* overloaded virtual function, warning: C++ Dialect Options. 69390 (line 1006) 69391* p in constraint: Simple Constraints. (line 152) 69392* packed type attribute: Common Type Attributes. 69393 (line 293) 69394* packed variable attribute: Common Variable Attributes. 69395 (line 285) 69396* parameter forward declaration: Variable Length. (line 66) 69397* partial_save function attribute, NDS32: NDS32 Function Attributes. 69398 (line 31) 69399* patchable_function_entry function attribute: Common Function Attributes. 69400 (line 1052) 69401* path GCC_COLORS capability: Diagnostic Message Formatting Options. 69402 (line 101) 69403* pcs function attribute, ARM: ARM Function Attributes. 69404 (line 58) 69405* PDP-11 Options: PDP-11 Options. (line 6) 69406* persistent variable attribute: Common Variable Attributes. 69407 (line 447) 69408* PIC: Code Gen Options. (line 356) 69409* picoChip options: picoChip Options. (line 6) 69410* pmf: Bound member functions. 69411 (line 6) 69412* pointer arguments: Common Function Attributes. 69413 (line 265) 69414* pointer arguments in variadic functions: Variadic Pointer Args. 69415 (line 6) 69416* pointer to member function: Bound member functions. 69417 (line 6) 69418* pointers to arrays: Pointers to Arrays. (line 6) 69419* portions of temporary objects, pointers to: Temporaries. (line 6) 69420* pow: Other Builtins. (line 6) 69421* pow10: Other Builtins. (line 6) 69422* pow10f: Other Builtins. (line 6) 69423* pow10l: Other Builtins. (line 6) 69424* PowerPC options: PowerPC Options. (line 6) 69425* powf: Other Builtins. (line 6) 69426* powl: Other Builtins. (line 6) 69427* pragma GCC ivdep: Loop-Specific Pragmas. 69428 (line 7) 69429* pragma GCC optimize: Function Specific Option Pragmas. 69430 (line 21) 69431* pragma GCC pop_options: Function Specific Option Pragmas. 69432 (line 33) 69433* pragma GCC push_options: Function Specific Option Pragmas. 69434 (line 33) 69435* pragma GCC reset_options: Function Specific Option Pragmas. 69436 (line 41) 69437* pragma GCC target: Function Specific Option Pragmas. 69438 (line 7) 69439* pragma GCC unroll N: Loop-Specific Pragmas. 69440 (line 37) 69441* pragma, address: M32C Pragmas. (line 15) 69442* pragma, align: Solaris Pragmas. (line 11) 69443* pragma, call: MeP Pragmas. (line 48) 69444* pragma, coprocessor available: MeP Pragmas. (line 13) 69445* pragma, coprocessor call_saved: MeP Pragmas. (line 20) 69446* pragma, coprocessor subclass: MeP Pragmas. (line 28) 69447* pragma, ctable_entry: PRU Pragmas. (line 7) 69448* pragma, custom io_volatile: MeP Pragmas. (line 7) 69449* pragma, diagnostic: Diagnostic Pragmas. (line 14) 69450* pragma, diagnostic <1>: Diagnostic Pragmas. (line 72) 69451* pragma, diagnostic <2>: Diagnostic Pragmas. (line 92) 69452* pragma, diagnostic <3>: Diagnostic Pragmas. (line 114) 69453* pragma, disinterrupt: MeP Pragmas. (line 38) 69454* pragma, fini: Solaris Pragmas. (line 20) 69455* pragma, init: Solaris Pragmas. (line 26) 69456* pragma, longcall: RS/6000 and PowerPC Pragmas. 69457 (line 14) 69458* pragma, long_calls: ARM Pragmas. (line 11) 69459* pragma, long_calls_off: ARM Pragmas. (line 17) 69460* pragma, mark: Darwin Pragmas. (line 11) 69461* pragma, memregs: M32C Pragmas. (line 7) 69462* pragma, no_long_calls: ARM Pragmas. (line 14) 69463* pragma, options align: Darwin Pragmas. (line 14) 69464* pragma, pop_macro: Push/Pop Macro Pragmas. 69465 (line 15) 69466* pragma, push_macro: Push/Pop Macro Pragmas. 69467 (line 11) 69468* pragma, redefine_extname: Symbol-Renaming Pragmas. 69469 (line 13) 69470* pragma, segment: Darwin Pragmas. (line 21) 69471* pragma, unused: Darwin Pragmas. (line 24) 69472* pragma, visibility: Visibility Pragmas. (line 8) 69473* pragma, weak: Weak Pragmas. (line 10) 69474* pragmas: Pragmas. (line 6) 69475* pragmas in C++, effect on inlining: C++ Interface. (line 57) 69476* pragmas, interface and implementation: C++ Interface. (line 6) 69477* pragmas, warning of unknown: Warning Options. (line 1390) 69478* precompiled headers: Precompiled Headers. 69479 (line 6) 69480* prefer-vector-width function attribute, x86: x86 Function Attributes. 69481 (line 592) 69482* preprocessing numbers: Incompatibilities. (line 173) 69483* preprocessing tokens: Incompatibilities. (line 173) 69484* preprocessor options: Preprocessor Options. 69485 (line 6) 69486* preserve_access_index type attribute, BPF: BPF Type Attributes. 69487 (line 6) 69488* printf: Other Builtins. (line 6) 69489* printf_unlocked: Other Builtins. (line 6) 69490* prof: Instrumentation Options. 69491 (line 18) 69492* profiling options: Instrumentation Options. 69493 (line 6) 69494* progmem variable attribute, AVR: AVR Variable Attributes. 69495 (line 7) 69496* program instrumentation options: Instrumentation Options. 69497 (line 6) 69498* promotion of formal parameters: Function Prototypes. 69499 (line 6) 69500* PRU Options: PRU Options. (line 6) 69501* pure function attribute: Common Function Attributes. 69502 (line 1070) 69503* push address instruction: Simple Constraints. (line 152) 69504* putchar: Other Builtins. (line 6) 69505* puts: Other Builtins. (line 6) 69506* q floating point suffix: Floating Types. (line 6) 69507* Q floating point suffix: Floating Types. (line 6) 69508* qsort, and global register variables: Global Register Variables. 69509 (line 75) 69510* quote GCC_COLORS capability: Diagnostic Message Formatting Options. 69511 (line 117) 69512* r fixed-suffix: Fixed-Point. (line 6) 69513* R fixed-suffix: Fixed-Point. (line 6) 69514* r in constraint: Simple Constraints. (line 64) 69515* RAMPD: AVR Options. (line 430) 69516* RAMPX: AVR Options. (line 430) 69517* RAMPY: AVR Options. (line 430) 69518* RAMPZ: AVR Options. (line 430) 69519* range1 GCC_COLORS capability: Diagnostic Message Formatting Options. 69520 (line 107) 69521* range2 GCC_COLORS capability: Diagnostic Message Formatting Options. 69522 (line 110) 69523* ranges in case statements: Case Ranges. (line 6) 69524* read-only strings: Incompatibilities. (line 9) 69525* realloc: Other Builtins. (line 6) 69526* reentrant function attribute, MSP430: MSP430 Function Attributes. 69527 (line 44) 69528* register variable after longjmp: Global Register Variables. 69529 (line 92) 69530* registers for local variables: Local Register Variables. 69531 (line 6) 69532* registers in constraints: Simple Constraints. (line 64) 69533* registers, global allocation: Global Register Variables. 69534 (line 6) 69535* registers, global variables in: Global Register Variables. 69536 (line 6) 69537* regparm function attribute, x86: x86 Function Attributes. 69538 (line 76) 69539* relocation truncated to fit (ColdFire): M680x0 Options. (line 323) 69540* relocation truncated to fit (MIPS): MIPS Options. (line 237) 69541* remainder: Other Builtins. (line 6) 69542* remainderf: Other Builtins. (line 6) 69543* remainderl: Other Builtins. (line 6) 69544* remquo: Other Builtins. (line 6) 69545* remquof: Other Builtins. (line 6) 69546* remquol: Other Builtins. (line 6) 69547* renesas function attribute, SH: SH Function Attributes. 69548 (line 40) 69549* reordering, warning: C++ Dialect Options. 69550 (line 820) 69551* reporting bugs: Bugs. (line 6) 69552* resbank function attribute, SH: SH Function Attributes. 69553 (line 44) 69554* reset function attribute, NDS32: NDS32 Function Attributes. 69555 (line 45) 69556* reset handler functions: NDS32 Function Attributes. 69557 (line 45) 69558* rest argument (in macro): Variadic Macros. (line 6) 69559* restricted pointers: Restricted Pointers. 69560 (line 6) 69561* restricted references: Restricted Pointers. 69562 (line 6) 69563* restricted this pointer: Restricted Pointers. 69564 (line 6) 69565* retain function attribute: Common Function Attributes. 69566 (line 1350) 69567* retain variable attribute: Common Variable Attributes. 69568 (line 375) 69569* returns_nonnull function attribute: Common Function Attributes. 69570 (line 1117) 69571* returns_twice function attribute: Common Function Attributes. 69572 (line 1127) 69573* rindex: Other Builtins. (line 6) 69574* rint: Other Builtins. (line 6) 69575* rintf: Other Builtins. (line 6) 69576* rintl: Other Builtins. (line 6) 69577* RISC-V Options: RISC-V Options. (line 6) 69578* RL78 Options: RL78 Options. (line 6) 69579* round: Other Builtins. (line 6) 69580* roundf: Other Builtins. (line 6) 69581* roundl: Other Builtins. (line 6) 69582* RS/6000 and PowerPC Options: RS/6000 and PowerPC Options. 69583 (line 6) 69584* RTTI: Vague Linkage. (line 42) 69585* run-time error checking options: Instrumentation Options. 69586 (line 6) 69587* run-time options: Code Gen Options. (line 6) 69588* RX Options: RX Options. (line 6) 69589* s in constraint: Simple Constraints. (line 100) 69590* S/390 and zSeries Options: S/390 and zSeries Options. 69591 (line 6) 69592* saddr variable attribute, RL78: RL78 Variable Attributes. 69593 (line 6) 69594* save all registers on the Blackfin: Blackfin Function Attributes. 69595 (line 56) 69596* save all registers on the H8/300, H8/300H, and H8S: H8/300 Function Attributes. 69597 (line 23) 69598* saveall function attribute, Blackfin: Blackfin Function Attributes. 69599 (line 56) 69600* saveall function attribute, H8/300: H8/300 Function Attributes. 69601 (line 23) 69602* save_all function attribute, NDS32: NDS32 Function Attributes. 69603 (line 28) 69604* save_volatiles function attribute, MicroBlaze: MicroBlaze Function Attributes. 69605 (line 9) 69606* scalar_storage_order type attribute: Common Type Attributes. 69607 (line 328) 69608* scalb: Other Builtins. (line 6) 69609* scalbf: Other Builtins. (line 6) 69610* scalbl: Other Builtins. (line 6) 69611* scalbln: Other Builtins. (line 6) 69612* scalblnf: Other Builtins. (line 6) 69613* scalblnf <1>: Other Builtins. (line 6) 69614* scalbn: Other Builtins. (line 6) 69615* scalbnf: Other Builtins. (line 6) 69616* scanf, and constant strings: Incompatibilities. (line 17) 69617* scanfnl: Other Builtins. (line 6) 69618* scope of a variable length array: Variable Length. (line 22) 69619* scope of declaration: Disappointments. (line 21) 69620* scope of external declarations: Incompatibilities. (line 80) 69621* Score Options: Score Options. (line 6) 69622* sda variable attribute, V850: V850 Variable Attributes. 69623 (line 9) 69624* search path: Directory Options. (line 6) 69625* section function attribute: Common Function Attributes. 69626 (line 1136) 69627* section variable attribute: Common Variable Attributes. 69628 (line 307) 69629* secure_call function attribute, ARC: ARC Function Attributes. 69630 (line 54) 69631* selectany variable attribute: Microsoft Windows Variable Attributes. 69632 (line 16) 69633* sentinel function attribute: Common Function Attributes. 69634 (line 1153) 69635* setjmp: Global Register Variables. 69636 (line 92) 69637* setjmp incompatibilities: Incompatibilities. (line 39) 69638* shared attribute, Nvidia PTX: Nvidia PTX Variable Attributes. 69639 (line 9) 69640* shared strings: Incompatibilities. (line 9) 69641* shared variable attribute: Microsoft Windows Variable Attributes. 69642 (line 37) 69643* shortcall function attribute, Blackfin: Blackfin Function Attributes. 69644 (line 38) 69645* shortcall function attribute, PowerPC: PowerPC Function Attributes. 69646 (line 10) 69647* short_call function attribute, ARC: ARC Function Attributes. 69648 (line 27) 69649* short_call function attribute, ARM: ARM Function Attributes. 69650 (line 38) 69651* short_call function attribute, Epiphany: Epiphany Function Attributes. 69652 (line 57) 69653* short_call function attribute, MIPS: MIPS Function Attributes. 69654 (line 63) 69655* side effect in ?:: Conditionals. (line 20) 69656* side effects, macro argument: Statement Exprs. (line 35) 69657* side effects, order of evaluation: Non-bugs. (line 196) 69658* sign-return-address function attribute, AArch64: AArch64 Function Attributes. 69659 (line 69) 69660* signal function attribute, AVR: AVR Function Attributes. 69661 (line 80) 69662* signbit: Other Builtins. (line 6) 69663* signbitd128: Other Builtins. (line 6) 69664* signbitd32: Other Builtins. (line 6) 69665* signbitd64: Other Builtins. (line 6) 69666* signbitf: Other Builtins. (line 6) 69667* signbitl: Other Builtins. (line 6) 69668* signed and unsigned values, comparison warning: Warning Options. 69669 (line 2691) 69670* significand: Other Builtins. (line 6) 69671* significandf: Other Builtins. (line 6) 69672* significandl: Other Builtins. (line 6) 69673* SIMD: C Dialect Options. (line 412) 69674* simd function attribute: Common Function Attributes. 69675 (line 1180) 69676* simple constraints: Simple Constraints. (line 6) 69677* sin: Other Builtins. (line 6) 69678* sincos: Other Builtins. (line 6) 69679* sincosf: Other Builtins. (line 6) 69680* sincosl: Other Builtins. (line 6) 69681* sinf: Other Builtins. (line 6) 69682* sinh: Other Builtins. (line 6) 69683* sinhf: Other Builtins. (line 6) 69684* sinhl: Other Builtins. (line 6) 69685* sinl: Other Builtins. (line 6) 69686* sizeof: Typeof. (line 6) 69687* smaller data references: M32R/D Options. (line 57) 69688* smaller data references <1>: Nios II Options. (line 9) 69689* smaller data references (PowerPC): RS/6000 and PowerPC Options. 69690 (line 714) 69691* snprintf: Other Builtins. (line 6) 69692* Solaris 2 options: Solaris 2 Options. (line 6) 69693* SOURCE_DATE_EPOCH: Environment Variables. 69694 (line 194) 69695* SPARC options: SPARC Options. (line 6) 69696* Spec Files: Spec Files. (line 6) 69697* specified registers: Explicit Register Variables. 69698 (line 6) 69699* specifying compiler version and target machine: Invoking GCC. 69700 (line 24) 69701* specifying hardware config: Submodel Options. (line 6) 69702* specifying machine version: Invoking GCC. (line 24) 69703* specifying registers for local variables: Local Register Variables. 69704 (line 6) 69705* speed of compilation: Precompiled Headers. 69706 (line 6) 69707* speed of compilation <1>: C++ Modules. (line 6) 69708* sprintf: Other Builtins. (line 6) 69709* sp_switch function attribute, SH: SH Function Attributes. 69710 (line 58) 69711* sqrt: Other Builtins. (line 6) 69712* sqrtf: Other Builtins. (line 6) 69713* sqrtl: Other Builtins. (line 6) 69714* sscanf: Other Builtins. (line 6) 69715* sscanf, and constant strings: Incompatibilities. (line 17) 69716* sseregparm function attribute, x86: x86 Function Attributes. 69717 (line 93) 69718* stack_protect function attribute: Common Function Attributes. 69719 (line 1199) 69720* Statement Attributes: Statement Attributes. 69721 (line 6) 69722* statements inside expressions: Statement Exprs. (line 6) 69723* static data in C++, declaring and defining: Static Definitions. 69724 (line 6) 69725* stdcall function attribute, x86-32: x86 Function Attributes. 69726 (line 108) 69727* stpcpy: Other Builtins. (line 6) 69728* stpncpy: Other Builtins. (line 6) 69729* strcasecmp: Other Builtins. (line 6) 69730* strcat: Other Builtins. (line 6) 69731* strchr: Other Builtins. (line 6) 69732* strcmp: Other Builtins. (line 6) 69733* strcpy: Other Builtins. (line 6) 69734* strcspn: Other Builtins. (line 6) 69735* strdup: Other Builtins. (line 6) 69736* strfmon: Other Builtins. (line 6) 69737* strftime: Other Builtins. (line 6) 69738* strict-align function attribute, AArch64: AArch64 Function Attributes. 69739 (line 33) 69740* string constants: Incompatibilities. (line 9) 69741* strlen: Other Builtins. (line 6) 69742* strncasecmp: Other Builtins. (line 6) 69743* strncat: Other Builtins. (line 6) 69744* strncmp: Other Builtins. (line 6) 69745* strncpy: Other Builtins. (line 6) 69746* strndup: Other Builtins. (line 6) 69747* strnlen: Other Builtins. (line 6) 69748* strpbrk: Other Builtins. (line 6) 69749* strrchr: Other Builtins. (line 6) 69750* strspn: Other Builtins. (line 6) 69751* strstr: Other Builtins. (line 6) 69752* struct: Unnamed Fields. (line 6) 69753* struct __htm_tdb: S/390 System z Built-in Functions. 69754 (line 49) 69755* structures: Incompatibilities. (line 146) 69756* structures, constructor expression: Compound Literals. (line 6) 69757* submodel options: Submodel Options. (line 6) 69758* subscripting: Subscripting. (line 6) 69759* subscripting and function values: Subscripting. (line 6) 69760* suffixes for C++ source: Invoking G++. (line 6) 69761* SUNPRO_DEPENDENCIES: Environment Variables. 69762 (line 188) 69763* suppressing warnings: Warning Options. (line 6) 69764* surprises in C++: C++ Misunderstandings. 69765 (line 6) 69766* symver function attribute: Common Function Attributes. 69767 (line 1247) 69768* syntax checking: Warning Options. (line 13) 69769* syscall_linkage function attribute, IA-64: IA-64 Function Attributes. 69770 (line 9) 69771* system headers, warnings from: Warning Options. (line 2034) 69772* sysv_abi function attribute, x86: x86 Function Attributes. 69773 (line 34) 69774* tainted_args function attribute: Common Function Attributes. 69775 (line 1291) 69776* tan: Other Builtins. (line 6) 69777* tanf: Other Builtins. (line 6) 69778* tanh: Other Builtins. (line 6) 69779* tanhf: Other Builtins. (line 6) 69780* tanhl: Other Builtins. (line 6) 69781* tanl: Other Builtins. (line 6) 69782* target function attribute: Common Function Attributes. 69783 (line 1207) 69784* target function attribute <1>: ARM Function Attributes. 69785 (line 77) 69786* target function attribute <2>: Nios II Function Attributes. 69787 (line 9) 69788* target function attribute <3>: PowerPC Function Attributes. 69789 (line 21) 69790* target function attribute <4>: S/390 Function Attributes. 69791 (line 22) 69792* target function attribute <5>: x86 Function Attributes. 69793 (line 180) 69794* target machine, specifying: Invoking GCC. (line 24) 69795* target("3dnow") function attribute, x86: x86 Function Attributes. 69796 (line 186) 69797* target("3dnowa") function attribute, x86: x86 Function Attributes. 69798 (line 190) 69799* target("abm") function attribute, x86: x86 Function Attributes. 69800 (line 195) 69801* target("adx") function attribute, x86: x86 Function Attributes. 69802 (line 200) 69803* target("aes") function attribute, x86: x86 Function Attributes. 69804 (line 204) 69805* target("align-stringops") function attribute, x86: x86 Function Attributes. 69806 (line 566) 69807* target("altivec") function attribute, PowerPC: PowerPC Function Attributes. 69808 (line 28) 69809* target("amx-bf16") function attribute, x86: x86 Function Attributes. 69810 (line 518) 69811* target("amx-int8") function attribute, x86: x86 Function Attributes. 69812 (line 514) 69813* target("amx-tile") function attribute, x86: x86 Function Attributes. 69814 (line 510) 69815* target("arch=ARCH") function attribute, x86: x86 Function Attributes. 69816 (line 578) 69817* target("arm") function attribute, ARM: ARM Function Attributes. 69818 (line 87) 69819* target("avoid-indexed-addresses") function attribute, PowerPC: PowerPC Function Attributes. 69820 (line 142) 69821* target("avx") function attribute, x86: x86 Function Attributes. 69822 (line 208) 69823* target("avx2") function attribute, x86: x86 Function Attributes. 69824 (line 212) 69825* target("avx5124fmaps") function attribute, x86: x86 Function Attributes. 69826 (line 216) 69827* target("avx5124vnniw") function attribute, x86: x86 Function Attributes. 69828 (line 221) 69829* target("avx512bitalg") function attribute, x86: x86 Function Attributes. 69830 (line 226) 69831* target("avx512bw") function attribute, x86: x86 Function Attributes. 69832 (line 231) 69833* target("avx512cd") function attribute, x86: x86 Function Attributes. 69834 (line 235) 69835* target("avx512dq") function attribute, x86: x86 Function Attributes. 69836 (line 239) 69837* target("avx512er") function attribute, x86: x86 Function Attributes. 69838 (line 243) 69839* target("avx512f") function attribute, x86: x86 Function Attributes. 69840 (line 247) 69841* target("avx512ifma") function attribute, x86: x86 Function Attributes. 69842 (line 251) 69843* target("avx512pf") function attribute, x86: x86 Function Attributes. 69844 (line 255) 69845* target("avx512vbmi") function attribute, x86: x86 Function Attributes. 69846 (line 259) 69847* target("avx512vbmi2") function attribute, x86: x86 Function Attributes. 69848 (line 263) 69849* target("avx512vl") function attribute, x86: x86 Function Attributes. 69850 (line 267) 69851* target("avx512vnni") function attribute, x86: x86 Function Attributes. 69852 (line 271) 69853* target("avx512vpopcntdq") function attribute, x86: x86 Function Attributes. 69854 (line 275) 69855* target("avxvnni") function attribute, x86: x86 Function Attributes. 69856 (line 538) 69857* target("bmi") function attribute, x86: x86 Function Attributes. 69858 (line 280) 69859* target("bmi2") function attribute, x86: x86 Function Attributes. 69860 (line 284) 69861* target("cld") function attribute, x86: x86 Function Attributes. 69862 (line 542) 69863* target("cldemote") function attribute, x86: x86 Function Attributes. 69864 (line 288) 69865* target("clflushopt") function attribute, x86: x86 Function Attributes. 69866 (line 292) 69867* target("clwb") function attribute, x86: x86 Function Attributes. 69868 (line 296) 69869* target("clzero") function attribute, x86: x86 Function Attributes. 69870 (line 300) 69871* target("cmpb") function attribute, PowerPC: PowerPC Function Attributes. 69872 (line 34) 69873* target("cpu=CPU") function attribute, PowerPC: PowerPC Function Attributes. 69874 (line 157) 69875* target("crc32") function attribute, x86: x86 Function Attributes. 69876 (line 304) 69877* target("custom-fpu-cfg=NAME") function attribute, Nios II: Nios II Function Attributes. 69878 (line 25) 69879* target("custom-INSN=N") function attribute, Nios II: Nios II Function Attributes. 69880 (line 16) 69881* target("cx16") function attribute, x86: x86 Function Attributes. 69882 (line 308) 69883* target("default") function attribute, x86: x86 Function Attributes. 69884 (line 311) 69885* target("dlmzb") function attribute, PowerPC: PowerPC Function Attributes. 69886 (line 40) 69887* target("f16c") function attribute, x86: x86 Function Attributes. 69888 (line 316) 69889* target("fancy-math-387") function attribute, x86: x86 Function Attributes. 69890 (line 546) 69891* target("fma") function attribute, x86: x86 Function Attributes. 69892 (line 320) 69893* target("fma4") function attribute, x86: x86 Function Attributes. 69894 (line 324) 69895* target("fpmath=FPMATH") function attribute, x86: x86 Function Attributes. 69896 (line 586) 69897* target("fprnd") function attribute, PowerPC: PowerPC Function Attributes. 69898 (line 47) 69899* target("fpu=") function attribute, ARM: ARM Function Attributes. 69900 (line 93) 69901* target("friz") function attribute, PowerPC: PowerPC Function Attributes. 69902 (line 133) 69903* target("fsgsbase") function attribute, x86: x86 Function Attributes. 69904 (line 328) 69905* target("fxsr") function attribute, x86: x86 Function Attributes. 69906 (line 332) 69907* target("general-regs-only") function attribute, x86: x86 Function Attributes. 69908 (line 575) 69909* target("gfni") function attribute, x86: x86 Function Attributes. 69910 (line 336) 69911* target("hard-dfp") function attribute, PowerPC: PowerPC Function Attributes. 69912 (line 53) 69913* target("hle") function attribute, x86: x86 Function Attributes. 69914 (line 340) 69915* target("hreset") function attribute, x86: x86 Function Attributes. 69916 (line 526) 69917* target("ieee-fp") function attribute, x86: x86 Function Attributes. 69918 (line 551) 69919* target("inline-all-stringops") function attribute, x86: x86 Function Attributes. 69920 (line 556) 69921* target("inline-stringops-dynamically") function attribute, x86: x86 Function Attributes. 69922 (line 560) 69923* target("isel") function attribute, PowerPC: PowerPC Function Attributes. 69924 (line 59) 69925* target("kl") function attribute, x86: x86 Function Attributes. 69926 (line 530) 69927* target("longcall") function attribute, PowerPC: PowerPC Function Attributes. 69928 (line 152) 69929* target("lwp") function attribute, x86: x86 Function Attributes. 69930 (line 344) 69931* target("lzcnt") function attribute, x86: x86 Function Attributes. 69932 (line 348) 69933* target("mfcrf") function attribute, PowerPC: PowerPC Function Attributes. 69934 (line 63) 69935* target("mmx") function attribute, x86: x86 Function Attributes. 69936 (line 352) 69937* target("movbe") function attribute, x86: x86 Function Attributes. 69938 (line 356) 69939* target("movdir64b") function attribute, x86: x86 Function Attributes. 69940 (line 360) 69941* target("movdiri") function attribute, x86: x86 Function Attributes. 69942 (line 364) 69943* target("mulhw") function attribute, PowerPC: PowerPC Function Attributes. 69944 (line 70) 69945* target("multiple") function attribute, PowerPC: PowerPC Function Attributes. 69946 (line 77) 69947* target("mwait") function attribute, x86: x86 Function Attributes. 69948 (line 368) 69949* target("mwaitx") function attribute, x86: x86 Function Attributes. 69950 (line 373) 69951* target("no-custom-INSN") function attribute, Nios II: Nios II Function Attributes. 69952 (line 16) 69953* target("paired") function attribute, PowerPC: PowerPC Function Attributes. 69954 (line 147) 69955* target("pclmul") function attribute, x86: x86 Function Attributes. 69956 (line 377) 69957* target("pconfig") function attribute, x86: x86 Function Attributes. 69958 (line 381) 69959* target("pku") function attribute, x86: x86 Function Attributes. 69960 (line 385) 69961* target("popcnt") function attribute, x86: x86 Function Attributes. 69962 (line 389) 69963* target("popcntb") function attribute, PowerPC: PowerPC Function Attributes. 69964 (line 88) 69965* target("popcntd") function attribute, PowerPC: PowerPC Function Attributes. 69966 (line 95) 69967* target("powerpc-gfxopt") function attribute, PowerPC: PowerPC Function Attributes. 69968 (line 101) 69969* target("powerpc-gpopt") function attribute, PowerPC: PowerPC Function Attributes. 69970 (line 107) 69971* target("prefetchwt1") function attribute, x86: x86 Function Attributes. 69972 (line 393) 69973* target("prfchw") function attribute, x86: x86 Function Attributes. 69974 (line 397) 69975* target("ptwrite") function attribute, x86: x86 Function Attributes. 69976 (line 401) 69977* target("rdpid") function attribute, x86: x86 Function Attributes. 69978 (line 405) 69979* target("rdrnd") function attribute, x86: x86 Function Attributes. 69980 (line 409) 69981* target("rdseed") function attribute, x86: x86 Function Attributes. 69982 (line 413) 69983* target("recip") function attribute, x86: x86 Function Attributes. 69984 (line 570) 69985* target("recip-precision") function attribute, PowerPC: PowerPC Function Attributes. 69986 (line 113) 69987* target("rtm") function attribute, x86: x86 Function Attributes. 69988 (line 417) 69989* target("sahf") function attribute, x86: x86 Function Attributes. 69990 (line 421) 69991* target("sgx") function attribute, x86: x86 Function Attributes. 69992 (line 425) 69993* target("sha") function attribute, x86: x86 Function Attributes. 69994 (line 429) 69995* target("shstk") function attribute, x86: x86 Function Attributes. 69996 (line 433) 69997* target("sse") function attribute, x86: x86 Function Attributes. 69998 (line 437) 69999* target("sse2") function attribute, x86: x86 Function Attributes. 70000 (line 441) 70001* target("sse3") function attribute, x86: x86 Function Attributes. 70002 (line 445) 70003* target("sse4") function attribute, x86: x86 Function Attributes. 70004 (line 449) 70005* target("sse4.1") function attribute, x86: x86 Function Attributes. 70006 (line 454) 70007* target("sse4.2") function attribute, x86: x86 Function Attributes. 70008 (line 458) 70009* target("sse4a") function attribute, x86: x86 Function Attributes. 70010 (line 462) 70011* target("ssse3") function attribute, x86: x86 Function Attributes. 70012 (line 466) 70013* target("string") function attribute, PowerPC: PowerPC Function Attributes. 70014 (line 119) 70015* target("tbm") function attribute, x86: x86 Function Attributes. 70016 (line 470) 70017* target("thumb") function attribute, ARM: ARM Function Attributes. 70018 (line 83) 70019* target("tune=TUNE") function attribute, PowerPC: PowerPC Function Attributes. 70020 (line 164) 70021* target("tune=TUNE") function attribute, x86: x86 Function Attributes. 70022 (line 582) 70023* target("uintr") function attribute, x86: x86 Function Attributes. 70024 (line 522) 70025* target("update") function attribute, PowerPC: PowerPC Function Attributes. 70026 (line 82) 70027* target("vaes") function attribute, x86: x86 Function Attributes. 70028 (line 474) 70029* target("vpclmulqdq") function attribute, x86: x86 Function Attributes. 70030 (line 478) 70031* target("vsx") function attribute, PowerPC: PowerPC Function Attributes. 70032 (line 125) 70033* target("waitpkg") function attribute, x86: x86 Function Attributes. 70034 (line 482) 70035* target("wbnoinvd") function attribute, x86: x86 Function Attributes. 70036 (line 486) 70037* target("widekl") function attribute, x86: x86 Function Attributes. 70038 (line 534) 70039* target("xop") function attribute, x86: x86 Function Attributes. 70040 (line 490) 70041* target("xsave") function attribute, x86: x86 Function Attributes. 70042 (line 494) 70043* target("xsavec") function attribute, x86: x86 Function Attributes. 70044 (line 498) 70045* target("xsaveopt") function attribute, x86: x86 Function Attributes. 70046 (line 502) 70047* target("xsaves") function attribute, x86: x86 Function Attributes. 70048 (line 506) 70049* target-dependent options: Submodel Options. (line 6) 70050* target_clones function attribute: Common Function Attributes. 70051 (line 1308) 70052* TC1: Standards. (line 13) 70053* TC2: Standards. (line 13) 70054* TC3: Standards. (line 13) 70055* tda variable attribute, V850: V850 Variable Attributes. 70056 (line 13) 70057* Technical Corrigenda: Standards. (line 13) 70058* Technical Corrigendum 1: Standards. (line 13) 70059* Technical Corrigendum 2: Standards. (line 13) 70060* Technical Corrigendum 3: Standards. (line 13) 70061* template instantiation: Template Instantiation. 70062 (line 6) 70063* temporaries, lifetime of: Temporaries. (line 6) 70064* tentative definitions: Code Gen Options. (line 234) 70065* TERM_URLS environment variable: Diagnostic Message Formatting Options. 70066 (line 144) 70067* tgamma: Other Builtins. (line 6) 70068* tgammaf: Other Builtins. (line 6) 70069* tgammal: Other Builtins. (line 6) 70070* thiscall function attribute, x86-32: x86 Function Attributes. 70071 (line 23) 70072* Thread-Local Storage: Thread-Local. (line 6) 70073* thunks: Nested Functions. (line 6) 70074* TILE-Gx options: TILE-Gx Options. (line 6) 70075* TILEPro options: TILEPro Options. (line 6) 70076* tiny data section on the H8/300H and H8S: H8/300 Variable Attributes. 70077 (line 19) 70078* tiny type attribute, MeP: MeP Type Attributes. 70079 (line 6) 70080* tiny variable attribute, MeP: MeP Variable Attributes. 70081 (line 20) 70082* tiny_data variable attribute, H8/300: H8/300 Variable Attributes. 70083 (line 19) 70084* TLS: Thread-Local. (line 6) 70085* tls-dialect= function attribute, AArch64: AArch64 Function Attributes. 70086 (line 48) 70087* tls_model variable attribute: Common Variable Attributes. 70088 (line 352) 70089* TMPDIR: Environment Variables. 70090 (line 45) 70091* toascii: Other Builtins. (line 6) 70092* tolower: Other Builtins. (line 6) 70093* toupper: Other Builtins. (line 6) 70094* towlower: Other Builtins. (line 6) 70095* towupper: Other Builtins. (line 6) 70096* traditional C language: Preprocessor Options. 70097 (line 375) 70098* transparent_union type attribute: Common Type Attributes. 70099 (line 372) 70100* trapa_handler function attribute, SH: SH Function Attributes. 70101 (line 73) 70102* trap_exit function attribute, SH: SH Function Attributes. 70103 (line 68) 70104* trunc: Other Builtins. (line 6) 70105* truncf: Other Builtins. (line 6) 70106* truncl: Other Builtins. (line 6) 70107* tune= function attribute, AArch64: AArch64 Function Attributes. 70108 (line 58) 70109* two-stage name lookup: Name lookup. (line 6) 70110* type alignment: Alignment. (line 6) 70111* type attributes: Type Attributes. (line 6) 70112* type-diff GCC_COLORS capability: Diagnostic Message Formatting Options. 70113 (line 140) 70114* typedef names as function parameters: Incompatibilities. (line 97) 70115* typeof: Typeof. (line 6) 70116* type_info: Vague Linkage. (line 42) 70117* uhk fixed-suffix: Fixed-Point. (line 6) 70118* UHK fixed-suffix: Fixed-Point. (line 6) 70119* uhr fixed-suffix: Fixed-Point. (line 6) 70120* UHR fixed-suffix: Fixed-Point. (line 6) 70121* uk fixed-suffix: Fixed-Point. (line 6) 70122* UK fixed-suffix: Fixed-Point. (line 6) 70123* ulk fixed-suffix: Fixed-Point. (line 6) 70124* ULK fixed-suffix: Fixed-Point. (line 6) 70125* ULL integer suffix: Long Long. (line 6) 70126* ullk fixed-suffix: Fixed-Point. (line 6) 70127* ULLK fixed-suffix: Fixed-Point. (line 6) 70128* ullr fixed-suffix: Fixed-Point. (line 6) 70129* ULLR fixed-suffix: Fixed-Point. (line 6) 70130* ulr fixed-suffix: Fixed-Point. (line 6) 70131* ULR fixed-suffix: Fixed-Point. (line 6) 70132* unavailable enumerator attribute: Enumerator Attributes. 70133 (line 38) 70134* unavailable function attribute: Common Function Attributes. 70135 (line 362) 70136* unavailable type attribute: Common Type Attributes. 70137 (line 224) 70138* unavailable variable attribute: Common Variable Attributes. 70139 (line 226) 70140* uncached type attribute, ARC: ARC Type Attributes. 70141 (line 6) 70142* undefined behavior: Bug Criteria. (line 17) 70143* undefined function value: Bug Criteria. (line 17) 70144* underscores in variables in macros: Typeof. (line 46) 70145* uninitialized variable attribute: Common Variable Attributes. 70146 (line 385) 70147* union: Unnamed Fields. (line 6) 70148* union, casting to a: Cast to Union. (line 6) 70149* unions: Incompatibilities. (line 146) 70150* unknown pragmas, warning: Warning Options. (line 1390) 70151* unresolved references and -nodefaultlibs: Link Options. (line 157) 70152* unresolved references and -nostdlib: Link Options. (line 157) 70153* unused function attribute: Common Function Attributes. 70154 (line 1335) 70155* unused label attribute: Label Attributes. (line 32) 70156* unused type attribute: Common Type Attributes. 70157 (line 425) 70158* unused variable attribute: Common Variable Attributes. 70159 (line 361) 70160* upper function attribute, MSP430: MSP430 Function Attributes. 70161 (line 57) 70162* upper variable attribute, MSP430: MSP430 Variable Attributes. 70163 (line 8) 70164* ur fixed-suffix: Fixed-Point. (line 6) 70165* UR fixed-suffix: Fixed-Point. (line 6) 70166* urls: Diagnostic Message Formatting Options. 70167 (line 144) 70168* used function attribute: Common Function Attributes. 70169 (line 1340) 70170* used variable attribute: Common Variable Attributes. 70171 (line 366) 70172* User stack pointer in interrupts on the Blackfin: Blackfin Function Attributes. 70173 (line 21) 70174* use_debug_exception_return function attribute, MIPS: MIPS Function Attributes. 70175 (line 39) 70176* use_shadow_register_set function attribute, MIPS: MIPS Function Attributes. 70177 (line 28) 70178* V in constraint: Simple Constraints. (line 43) 70179* V850 Options: V850 Options. (line 6) 70180* vague linkage: Vague Linkage. (line 6) 70181* value after longjmp: Global Register Variables. 70182 (line 92) 70183* variable addressability on the M32R/D: M32R/D Variable Attributes. 70184 (line 9) 70185* variable alignment: Alignment. (line 6) 70186* variable attributes: Variable Attributes. 70187 (line 6) 70188* variable number of arguments: Variadic Macros. (line 6) 70189* variable-length array in a structure: Variable Length. (line 26) 70190* variable-length array scope: Variable Length. (line 22) 70191* variable-length arrays: Variable Length. (line 6) 70192* variables in specified registers: Explicit Register Variables. 70193 (line 6) 70194* variables, local, in macros: Typeof. (line 46) 70195* variadic functions, pointer arguments: Variadic Pointer Args. 70196 (line 6) 70197* variadic macros: Variadic Macros. (line 6) 70198* VAX options: VAX Options. (line 6) 70199* vector function attribute, RX: RX Function Attributes. 70200 (line 49) 70201* vector types, using with x86 intrinsics: Vector Extensions. 70202 (line 212) 70203* vector_size type attribute: Common Type Attributes. 70204 (line 434) 70205* vector_size variable attribute: Common Variable Attributes. 70206 (line 401) 70207* vec_blendv: PowerPC AltiVec Built-in Functions Available on ISA 3.1. 70208 (line 308) 70209* vec_cfuge: PowerPC AltiVec Built-in Functions Available on ISA 3.1. 70210 (line 12) 70211* vec_clrl: PowerPC AltiVec Built-in Functions Available on ISA 3.1. 70212 (line 31) 70213* vec_clrr: PowerPC AltiVec Built-in Functions Available on ISA 3.1. 70214 (line 40) 70215* vec_cntlzm: PowerPC AltiVec Built-in Functions Available on ISA 3.1. 70216 (line 17) 70217* vec_cnttzm: PowerPC AltiVec Built-in Functions Available on ISA 3.1. 70218 (line 22) 70219* vec_extracth: PowerPC AltiVec Built-in Functions Available on ISA 3.1. 70220 (line 101) 70221* vec_extractl: PowerPC AltiVec Built-in Functions Available on ISA 3.1. 70222 (line 72) 70223* vec_genpcvm: PowerPC AltiVec Built-in Functions Available on ISA 3.1. 70224 (line 514) 70225* vec_gnb: PowerPC AltiVec Built-in Functions Available on ISA 3.1. 70226 (line 46) 70227* vec_inserth: PowerPC AltiVec Built-in Functions Available on ISA 3.1. 70228 (line 168) 70229* vec_insertl: PowerPC AltiVec Built-in Functions Available on ISA 3.1. 70230 (line 137) 70231* vec_pdep: PowerPC AltiVec Built-in Functions Available on ISA 3.1. 70232 (line 105) 70233* vec_permx: PowerPC AltiVec Built-in Functions Available on ISA 3.1. 70234 (line 335) 70235* vec_pext: PowerPC AltiVec Built-in Functions Available on ISA 3.1. 70236 (line 340) 70237* vec_replace_element: PowerPC AltiVec Built-in Functions Available on ISA 3.1. 70238 (line 189) 70239* vec_replace_unaligned: PowerPC AltiVec Built-in Functions Available on ISA 3.1. 70240 (line 211) 70241* vec_sldb: PowerPC AltiVec Built-in Functions Available on ISA 3.1. 70242 (line 235) 70243* vec_splati: PowerPC AltiVec Built-in Functions Available on ISA 3.1. 70244 (line 265) 70245* vec_splatid: PowerPC AltiVec Built-in Functions Available on ISA 3.1. 70246 (line 270) 70247* vec_splati_ins: PowerPC AltiVec Built-in Functions Available on ISA 3.1. 70248 (line 281) 70249* vec_srdb: PowerPC AltiVec Built-in Functions Available on ISA 3.1. 70250 (line 258) 70251* vec_stril: PowerPC AltiVec Built-in Functions Available on ISA 3.1. 70252 (line 350) 70253* vec_stril_p: PowerPC AltiVec Built-in Functions Available on ISA 3.1. 70254 (line 361) 70255* vec_strir: PowerPC AltiVec Built-in Functions Available on ISA 3.1. 70256 (line 371) 70257* vec_strir_p: PowerPC AltiVec Built-in Functions Available on ISA 3.1. 70258 (line 382) 70259* vec_ternarylogic: PowerPC AltiVec Built-in Functions Available on ISA 3.1. 70260 (line 401) 70261* vec_xst_trunc: Basic PowerPC Built-in Functions Available on ISA 3.1. 70262 (line 77) 70263* version_id function attribute, IA-64: IA-64 Function Attributes. 70264 (line 16) 70265* vfprintf: Other Builtins. (line 6) 70266* vfscanf: Other Builtins. (line 6) 70267* visibility function attribute: Common Function Attributes. 70268 (line 1361) 70269* visibility type attribute: Common Type Attributes. 70270 (line 461) 70271* visibility variable attribute: Common Variable Attributes. 70272 (line 429) 70273* Visium options: Visium Options. (line 6) 70274* VLAs: Variable Length. (line 6) 70275* vliw function attribute, MeP: MeP Function Attributes. 70276 (line 30) 70277* void pointers, arithmetic: Pointer Arith. (line 6) 70278* void, size of pointer to: Pointer Arith. (line 6) 70279* volatile access: Volatiles. (line 6) 70280* volatile access <1>: C++ Volatiles. (line 6) 70281* volatile applied to function: Function Attributes. 70282 (line 6) 70283* volatile asm: Extended Asm. (line 116) 70284* volatile read: Volatiles. (line 6) 70285* volatile read <1>: C++ Volatiles. (line 6) 70286* volatile write: Volatiles. (line 6) 70287* volatile write <1>: C++ Volatiles. (line 6) 70288* vprintf: Other Builtins. (line 6) 70289* vscanf: Other Builtins. (line 6) 70290* vsnprintf: Other Builtins. (line 6) 70291* vsprintf: Other Builtins. (line 6) 70292* vsscanf: Other Builtins. (line 6) 70293* vsx_xl_sext: Basic PowerPC Built-in Functions Available on ISA 3.1. 70294 (line 57) 70295* vsx_xl_zext: Basic PowerPC Built-in Functions Available on ISA 3.1. 70296 (line 57) 70297* vtable: Vague Linkage. (line 27) 70298* VxWorks Options: VxWorks Options. (line 6) 70299* w floating point suffix: Floating Types. (line 6) 70300* W floating point suffix: Floating Types. (line 6) 70301* wakeup function attribute, MSP430: MSP430 Function Attributes. 70302 (line 49) 70303* warm function attribute, NDS32: NDS32 Function Attributes. 70304 (line 52) 70305* warning for comparison of signed and unsigned values: Warning Options. 70306 (line 2691) 70307* warning for overloaded virtual function: C++ Dialect Options. 70308 (line 1006) 70309* warning for reordering of member initializers: C++ Dialect Options. 70310 (line 820) 70311* warning for unknown pragmas: Warning Options. (line 1390) 70312* warning function attribute: Common Function Attributes. 70313 (line 373) 70314* warning GCC_COLORS capability: Diagnostic Message Formatting Options. 70315 (line 95) 70316* warning messages: Warning Options. (line 6) 70317* warnings from system headers: Warning Options. (line 2034) 70318* warnings vs errors: Warnings and Errors. 70319 (line 6) 70320* warn_if_not_aligned type attribute: Common Type Attributes. 70321 (line 91) 70322* warn_if_not_aligned variable attribute: Common Variable Attributes. 70323 (line 106) 70324* warn_unused type attribute: C++ Attributes. (line 71) 70325* warn_unused_result function attribute: Common Function Attributes. 70326 (line 1461) 70327* weak function attribute: Common Function Attributes. 70328 (line 1478) 70329* weak variable attribute: Common Variable Attributes. 70330 (line 434) 70331* weakref function attribute: Common Function Attributes. 70332 (line 1490) 70333* whitespace: Incompatibilities. (line 112) 70334* Windows Options for x86: x86 Windows Options. 70335 (line 6) 70336* X in constraint: Simple Constraints. (line 122) 70337* X3.159-1989: Standards. (line 13) 70338* x86 named address spaces: Named Address Spaces. 70339 (line 185) 70340* x86 Options: x86 Options. (line 6) 70341* x86 Windows Options: x86 Windows Options. 70342 (line 6) 70343* Xstormy16 Options: Xstormy16 Options. (line 6) 70344* Xtensa Options: Xtensa Options. (line 6) 70345* y0: Other Builtins. (line 6) 70346* y0f: Other Builtins. (line 6) 70347* y0l: Other Builtins. (line 6) 70348* y1: Other Builtins. (line 6) 70349* y1f: Other Builtins. (line 6) 70350* y1l: Other Builtins. (line 6) 70351* yn: Other Builtins. (line 6) 70352* ynf: Other Builtins. (line 6) 70353* ynl: Other Builtins. (line 6) 70354* zda variable attribute, V850: V850 Variable Attributes. 70355 (line 17) 70356* zero-length arrays: Zero Length. (line 6) 70357* zero-size structures: Empty Structures. (line 6) 70358* zero_call_used_regs function attribute: Common Function Attributes. 70359 (line 1532) 70360* zSeries options: zSeries Options. (line 6) 70361 70362 70363 70364Tag Table: 70365Node: Top2135 70366Node: G++ and GCC4093 70367Node: Standards6139 70368Node: Invoking GCC19766 70369Node: Option Summary24993 70370Node: Overall Options83928 70371Node: Invoking G++111454 70372Node: C Dialect Options112977 70373Node: C++ Dialect Options134016 70374Node: Objective-C and Objective-C++ Dialect Options196575 70375Node: Diagnostic Message Formatting Options208322 70376Node: Warning Options238895 70377Ref: Wtrigraphs343793 70378Node: Static Analyzer Options389156 70379Node: Debugging Options409708 70380Node: Optimize Options430945 70381Ref: Type-punning503405 70382Node: Instrumentation Options623696 70383Node: Preprocessor Options671808 70384Ref: dashMF676655 70385Ref: fdollars-in-identifiers681317 70386Node: Assembler Options694642 70387Node: Link Options695333 70388Ref: Link Options-Footnote-1712293 70389Node: Directory Options712629 70390Node: Code Gen Options721032 70391Node: Developer Options749905 70392Node: Submodel Options791290 70393Node: AArch64 Options793122 70394Ref: aarch64-feature-modifiers808695 70395Node: Adapteva Epiphany Options814256 70396Node: AMD GCN Options820208 70397Node: ARC Options821771 70398Node: ARM Options842540 70399Node: AVR Options884565 70400Node: Blackfin Options911102 70401Node: C6X Options918994 70402Node: CRIS Options920537 70403Node: CR16 Options923723 70404Node: C-SKY Options924635 70405Node: Darwin Options930362 70406Node: DEC Alpha Options937803 70407Node: eBPF Options949419 70408Node: FR30 Options951795 70409Node: FT32 Options952355 70410Node: FRV Options953301 70411Node: GNU/Linux Options960065 70412Node: H8/300 Options961446 70413Node: HPPA Options962898 70414Node: IA-64 Options972430 70415Node: LM32 Options980558 70416Node: LoongArch Options981086 70417Node: M32C Options987364 70418Node: M32R/D Options988642 70419Node: M680x0 Options992187 70420Node: MCore Options1006342 70421Node: MeP Options1007844 70422Node: MicroBlaze Options1011804 70423Node: MIPS Options1014869 70424Node: MMIX Options1051666 70425Node: MN10300 Options1054143 70426Node: Moxie Options1056686 70427Node: MSP430 Options1057173 70428Node: NDS32 Options1064934 70429Node: Nios II Options1067104 70430Node: Nvidia PTX Options1080320 70431Node: OpenRISC Options1083816 70432Node: PDP-11 Options1086570 70433Node: picoChip Options1087819 70434Node: PowerPC Options1089957 70435Node: PRU Options1090177 70436Node: RISC-V Options1092386 70437Node: RL78 Options1100787 70438Node: RS/6000 and PowerPC Options1104562 70439Node: RX Options1146078 70440Node: S/390 and zSeries Options1154680 70441Node: Score Options1165480 70442Node: SH Options1166329 70443Node: Solaris 2 Options1181469 70444Node: SPARC Options1182707 70445Node: System V Options1198359 70446Node: TILE-Gx Options1199187 70447Node: TILEPro Options1200205 70448Node: V850 Options1200709 70449Node: VAX Options1207396 70450Node: Visium Options1208088 70451Node: VMS Options1210396 70452Node: VxWorks Options1211212 70453Node: x86 Options1212364 70454Node: x86 Windows Options1281467 70455Node: Xstormy16 Options1284273 70456Node: Xtensa Options1284567 70457Node: zSeries Options1290703 70458Node: Spec Files1290899 70459Node: Environment Variables1316978 70460Node: Precompiled Headers1326411 70461Node: C++ Modules1332620 70462Ref: C++ Modules-Footnote-11338615 70463Node: C++ Module Mapper1338868 70464Node: C++ Module Preprocessing1344021 70465Node: C++ Compiled Module Interface1345850 70466Ref: C++ Compiled Module Interface-Footnote-11349234 70467Ref: C++ Compiled Module Interface-Footnote-21349323 70468Node: C Implementation1349379 70469Node: Translation implementation1351070 70470Node: Environment implementation1351661 70471Node: Identifiers implementation1352215 70472Node: Characters implementation1353301 70473Node: Integers implementation1356951 70474Node: Floating point implementation1359000 70475Node: Arrays and pointers implementation1362063 70476Ref: Arrays and pointers implementation-Footnote-11363523 70477Node: Hints implementation1363649 70478Node: Structures unions enumerations and bit-fields implementation1365144 70479Node: Qualifiers implementation1367368 70480Node: Declarators implementation1369429 70481Node: Statements implementation1369770 70482Node: Preprocessing directives implementation1370096 70483Node: Library functions implementation1372417 70484Node: Architecture implementation1373066 70485Node: Locale-specific behavior implementation1374711 70486Node: C++ Implementation1375016 70487Node: Conditionally-supported behavior1376300 70488Node: Exception handling1376917 70489Node: C Extensions1377384 70490Node: Statement Exprs1382607 70491Node: Local Labels1387979 70492Node: Labels as Values1390952 70493Ref: Labels as Values-Footnote-11393479 70494Node: Nested Functions1393664 70495Node: Nonlocal Gotos1397618 70496Node: Constructing Calls1399884 70497Node: Typeof1404596 70498Node: Conditionals1408525 70499Node: __int1281409414 70500Node: Long Long1409939 70501Node: Complex1411430 70502Node: Floating Types1416175 70503Node: Half-Precision1419696 70504Node: Decimal Float1423164 70505Node: Hex Floats1425018 70506Node: Fixed-Point1426092 70507Node: Named Address Spaces1429350 70508Ref: AVR Named Address Spaces1430036 70509Node: Zero Length1437330 70510Node: Empty Structures1441511 70511Node: Variable Length1441917 70512Node: Variadic Macros1444635 70513Node: Escaped Newlines1447013 70514Node: Subscripting1447874 70515Node: Pointer Arith1448599 70516Node: Variadic Pointer Args1449176 70517Node: Pointers to Arrays1449901 70518Node: Initializers1450654 70519Node: Compound Literals1451155 70520Node: Designated Inits1454722 70521Node: Case Ranges1458646 70522Node: Cast to Union1459327 70523Node: Mixed Labels and Declarations1461061 70524Node: Function Attributes1461698 70525Node: Common Function Attributes1466209 70526Ref: Common Function Attributes-Footnote-11543150 70527Node: AArch64 Function Attributes1543467 70528Node: AMD GCN Function Attributes1549661 70529Node: ARC Function Attributes1552714 70530Node: ARM Function Attributes1555651 70531Node: AVR Function Attributes1560790 70532Node: Blackfin Function Attributes1565325 70533Node: BPF Function Attributes1567822 70534Node: CR16 Function Attributes1568410 70535Node: C-SKY Function Attributes1568929 70536Node: Epiphany Function Attributes1570228 70537Node: H8/300 Function Attributes1572983 70538Node: IA-64 Function Attributes1574181 70539Node: M32C Function Attributes1575223 70540Node: M32R/D Function Attributes1577561 70541Node: m68k Function Attributes1579035 70542Node: MCORE Function Attributes1579979 70543Node: MeP Function Attributes1580790 70544Node: MicroBlaze Function Attributes1582091 70545Node: Microsoft Windows Function Attributes1583598 70546Node: MIPS Function Attributes1588167 70547Node: MSP430 Function Attributes1593785 70548Node: NDS32 Function Attributes1597864 70549Node: Nios II Function Attributes1600288 70550Node: Nvidia PTX Function Attributes1601585 70551Node: PowerPC Function Attributes1602200 70552Node: RISC-V Function Attributes1608974 70553Node: RL78 Function Attributes1610390 70554Node: RX Function Attributes1611629 70555Node: S/390 Function Attributes1614176 70556Node: SH Function Attributes1616004 70557Node: Symbian OS Function Attributes1619432 70558Node: V850 Function Attributes1619768 70559Node: Visium Function Attributes1620313 70560Node: x86 Function Attributes1620841 70561Node: Xstormy16 Function Attributes1644796 70562Node: Variable Attributes1645303 70563Node: Common Variable Attributes1646836 70564Node: ARC Variable Attributes1668743 70565Node: AVR Variable Attributes1669125 70566Node: Blackfin Variable Attributes1674288 70567Node: H8/300 Variable Attributes1675146 70568Node: IA-64 Variable Attributes1676219 70569Node: M32R/D Variable Attributes1676970 70570Node: MeP Variable Attributes1677753 70571Node: Microsoft Windows Variable Attributes1679846 70572Node: MSP430 Variable Attributes1682299 70573Node: Nvidia PTX Variable Attributes1683513 70574Node: PowerPC Variable Attributes1684130 70575Node: RL78 Variable Attributes1684687 70576Node: V850 Variable Attributes1685106 70577Node: x86 Variable Attributes1685739 70578Node: Xstormy16 Variable Attributes1686795 70579Node: Type Attributes1687365 70580Node: Common Type Attributes1689053 70581Node: ARC Type Attributes1711414 70582Node: ARM Type Attributes1711886 70583Node: BPF Type Attributes1712668 70584Node: MeP Type Attributes1713185 70585Node: PowerPC Type Attributes1713587 70586Node: x86 Type Attributes1714576 70587Node: Label Attributes1715568 70588Node: Enumerator Attributes1717595 70589Node: Statement Attributes1719127 70590Node: Attribute Syntax1720610 70591Node: Function Prototypes1731868 70592Node: C++ Comments1733648 70593Node: Dollar Signs1734167 70594Node: Character Escapes1734632 70595Node: Alignment1734916 70596Node: Inline1736569 70597Node: Volatiles1741386 70598Node: Using Assembly Language with C1744285 70599Node: Basic Asm1745522 70600Node: Extended Asm1751145 70601Ref: Volatile1755259 70602Ref: AssemblerTemplate1759372 70603Ref: OutputOperands1763612 70604Ref: FlagOutputOperands1770575 70605Ref: InputOperands1773522 70606Ref: Clobbers and Scratch Registers1777790 70607Ref: GotoLabels1786431 70608Ref: x86Operandmodifiers1790054 70609Ref: x86floatingpointasmoperands1794870 70610Ref: msp430Operandmodifiers1798199 70611Node: Constraints1799857 70612Node: Simple Constraints1800963 70613Node: Multi-Alternative1808277 70614Node: Modifiers1809952 70615Node: Machine Constraints1812751 70616Node: Asm Labels1870522 70617Node: Explicit Register Variables1872142 70618Ref: Explicit Reg Vars1872356 70619Node: Global Register Variables1872965 70620Ref: Global Reg Vars1873173 70621Node: Local Register Variables1877955 70622Ref: Local Reg Vars1878175 70623Node: Size of an asm1881803 70624Node: Alternate Keywords1883281 70625Node: Incomplete Enums1884786 70626Node: Function Names1885543 70627Node: Return Address1887447 70628Node: Vector Extensions1892034 70629Node: Offsetof1903017 70630Node: __sync Builtins1903850 70631Node: __atomic Builtins1910293 70632Node: Integer Overflow Builtins1923806 70633Node: x86 specific memory model extensions for transactional memory1930289 70634Node: Object Size Checking1931555 70635Node: Other Builtins1938571 70636Node: Target Builtins1991175 70637Node: AArch64 Built-in Functions1992950 70638Node: Alpha Built-in Functions1993646 70639Node: Altera Nios II Built-in Functions1996740 70640Node: ARC Built-in Functions2001179 70641Node: ARC SIMD Built-in Functions2006391 70642Node: ARM iWMMXt Built-in Functions2015389 70643Node: ARM C Language Extensions (ACLE)2022525 70644Node: ARM Floating Point Status and Control Intrinsics2023802 70645Node: ARM ARMv8-M Security Extensions2024289 70646Node: AVR Built-in Functions2025581 70647Node: Blackfin Built-in Functions2029347 70648Node: BPF Built-in Functions2029967 70649Node: FR-V Built-in Functions2031426 70650Node: Argument Types2032289 70651Node: Directly-mapped Integer Functions2034120 70652Node: Directly-mapped Media Functions2035280 70653Node: Raw read/write Functions2043562 70654Node: Other Built-in Functions2044470 70655Node: MIPS DSP Built-in Functions2045656 70656Node: MIPS Paired-Single Support2058411 70657Node: MIPS Loongson Built-in Functions2059910 70658Node: Paired-Single Arithmetic2066432 70659Node: Paired-Single Built-in Functions2067455 70660Node: MIPS-3D Built-in Functions2070122 70661Node: MIPS SIMD Architecture (MSA) Support2075516 70662Node: MIPS SIMD Architecture Built-in Functions2078356 70663Node: Other MIPS Built-in Functions2105210 70664Node: MSP430 Built-in Functions2106219 70665Node: NDS32 Built-in Functions2107620 70666Node: picoChip Built-in Functions2108913 70667Node: Basic PowerPC Built-in Functions2110262 70668Node: Basic PowerPC Built-in Functions Available on all Configurations2111120 70669Node: Basic PowerPC Built-in Functions Available on ISA 2.052119630 70670Node: Basic PowerPC Built-in Functions Available on ISA 2.062124465 70671Node: Basic PowerPC Built-in Functions Available on ISA 2.072126545 70672Node: Basic PowerPC Built-in Functions Available on ISA 3.02127399 70673Node: Basic PowerPC Built-in Functions Available on ISA 3.12134377 70674Node: PowerPC AltiVec/VSX Built-in Functions2137542 70675Node: PowerPC AltiVec Built-in Functions on ISA 2.052145611 70676Node: PowerPC AltiVec Built-in Functions Available on ISA 2.062171011 70677Node: PowerPC AltiVec Built-in Functions Available on ISA 2.072178077 70678Node: PowerPC AltiVec Built-in Functions Available on ISA 3.02187887 70679Node: PowerPC AltiVec Built-in Functions Available on ISA 3.12203610 70680Node: PowerPC Hardware Transactional Memory Built-in Functions2236894 70681Node: PowerPC Atomic Memory Operation Functions2245444 70682Node: PowerPC Matrix-Multiply Assist Built-in Functions2248007 70683Node: PRU Built-in Functions2254132 70684Node: RISC-V Built-in Functions2255320 70685Node: RX Built-in Functions2255731 70686Node: S/390 System z Built-in Functions2259733 70687Node: SH Built-in Functions2264963 70688Node: SPARC VIS Built-in Functions2266691 70689Node: TI C6X Built-in Functions2275222 70690Node: TILE-Gx Built-in Functions2276273 70691Node: TILEPro Built-in Functions2277398 70692Node: x86 Built-in Functions2278506 70693Node: x86 transactional memory intrinsics2343856 70694Node: x86 control-flow protection intrinsics2347123 70695Node: Target Format Checks2348894 70696Node: Solaris Format Checks2349326 70697Node: Darwin Format Checks2349752 70698Node: Pragmas2350715 70699Node: AArch64 Pragmas2351656 70700Node: ARM Pragmas2352113 70701Node: M32C Pragmas2352740 70702Node: MeP Pragmas2353812 70703Node: PRU Pragmas2355864 70704Node: RS/6000 and PowerPC Pragmas2356442 70705Node: S/390 Pragmas2357182 70706Node: Darwin Pragmas2357748 70707Node: Solaris Pragmas2358801 70708Node: Symbol-Renaming Pragmas2359965 70709Node: Structure-Layout Pragmas2361602 70710Node: Weak Pragmas2363882 70711Node: Diagnostic Pragmas2364617 70712Node: Visibility Pragmas2369368 70713Node: Push/Pop Macro Pragmas2370053 70714Node: Function Specific Option Pragmas2371026 70715Node: Loop-Specific Pragmas2372992 70716Node: Unnamed Fields2374592 70717Node: Thread-Local2376789 70718Node: C99 Thread-Local Edits2378895 70719Node: C++98 Thread-Local Edits2380893 70720Node: Binary constants2384338 70721Node: C++ Extensions2385009 70722Node: C++ Volatiles2386639 70723Node: Restricted Pointers2388987 70724Node: Vague Linkage2390578 70725Node: C++ Interface2394201 70726Ref: C++ Interface-Footnote-12397998 70727Node: Template Instantiation2398136 70728Node: Bound member functions2404227 70729Node: C++ Attributes2405759 70730Node: Function Multiversioning2409831 70731Node: Type Traits2411639 70732Node: C++ Concepts2418610 70733Node: Deprecated Features2420116 70734Node: Backwards Compatibility2421941 70735Node: Objective-C2423013 70736Node: GNU Objective-C runtime API2423620 70737Node: Modern GNU Objective-C runtime API2424627 70738Node: Traditional GNU Objective-C runtime API2427063 70739Node: Executing code before main2427790 70740Node: What you can and what you cannot do in +load2430534 70741Node: Type encoding2432904 70742Node: Legacy type encoding2438270 70743Node: @encode2439360 70744Node: Method signatures2439905 70745Node: Garbage Collection2441897 70746Node: Constant string objects2444587 70747Node: compatibility_alias2447096 70748Node: Exceptions2447821 70749Node: Synchronization2450531 70750Node: Fast enumeration2451715 70751Node: Using fast enumeration2452027 70752Node: c99-like fast enumeration syntax2453238 70753Node: Fast enumeration details2453941 70754Node: Fast enumeration protocol2456281 70755Node: Messaging with the GNU Objective-C runtime2459433 70756Node: Dynamically registering methods2460805 70757Node: Forwarding hook2462496 70758Node: Compatibility2465536 70759Node: Gcov2472092 70760Node: Gcov Intro2472627 70761Node: Invoking Gcov2475345 70762Node: Gcov and Optimization2498078 70763Node: Gcov Data Files2501831 70764Node: Cross-profiling2503240 70765Node: Gcov-tool2505094 70766Node: Gcov-tool Intro2505519 70767Node: Invoking Gcov-tool2507489 70768Node: Gcov-dump2510067 70769Node: Gcov-dump Intro2510390 70770Node: Invoking Gcov-dump2510657 70771Node: lto-dump2511314 70772Node: lto-dump Intro2511613 70773Node: Invoking lto-dump2511863 70774Node: Trouble2512959 70775Node: Actual Bugs2514376 70776Node: Interoperation2514823 70777Node: Incompatibilities2521714 70778Node: Fixed Headers2529866 70779Node: Standard Libraries2531524 70780Node: Disappointments2532896 70781Node: C++ Misunderstandings2537255 70782Node: Static Definitions2538066 70783Node: Name lookup2539119 70784Ref: Name lookup-Footnote-12543900 70785Node: Temporaries2544089 70786Node: Copy Assignment2546065 70787Node: Non-bugs2547900 70788Node: Warnings and Errors2558406 70789Node: Bugs2560168 70790Node: Bug Criteria2560635 70791Node: Bug Reporting2562845 70792Node: Service2563063 70793Node: Contributing2563884 70794Node: Funding2564630 70795Node: GNU Project2567120 70796Node: Copying2567768 70797Node: GNU Free Documentation License2605279 70798Node: Contributors2630399 70799Node: Option Index2671373 70800Node: Keyword Index2970472 70801 70802End Tag Table 70803