1@c Copyright (C) 1988-2020 Free Software Foundation, Inc. 2@c This is part of the GCC manual. 3@c For copying conditions, see the file gcc.texi. 4 5@node Target Macros 6@chapter Target Description Macros and Functions 7@cindex machine description macros 8@cindex target description macros 9@cindex macros, target description 10@cindex @file{tm.h} macros 11 12In addition to the file @file{@var{machine}.md}, a machine description 13includes a C header file conventionally given the name 14@file{@var{machine}.h} and a C source file named @file{@var{machine}.c}. 15The header file defines numerous macros that convey the information 16about the target machine that does not fit into the scheme of the 17@file{.md} file. The file @file{tm.h} should be a link to 18@file{@var{machine}.h}. The header file @file{config.h} includes 19@file{tm.h} and most compiler source files include @file{config.h}. The 20source file defines a variable @code{targetm}, which is a structure 21containing pointers to functions and data relating to the target 22machine. @file{@var{machine}.c} should also contain their definitions, 23if they are not defined elsewhere in GCC, and other functions called 24through the macros defined in the @file{.h} file. 25 26@menu 27* Target Structure:: The @code{targetm} variable. 28* Driver:: Controlling how the driver runs the compilation passes. 29* Run-time Target:: Defining @samp{-m} options like @option{-m68000} and @option{-m68020}. 30* Per-Function Data:: Defining data structures for per-function information. 31* Storage Layout:: Defining sizes and alignments of data. 32* Type Layout:: Defining sizes and properties of basic user data types. 33* Registers:: Naming and describing the hardware registers. 34* Register Classes:: Defining the classes of hardware registers. 35* Stack and Calling:: Defining which way the stack grows and by how much. 36* Varargs:: Defining the varargs macros. 37* Trampolines:: Code set up at run time to enter a nested function. 38* Library Calls:: Controlling how library routines are implicitly called. 39* Addressing Modes:: Defining addressing modes valid for memory operands. 40* Anchored Addresses:: Defining how @option{-fsection-anchors} should work. 41* Condition Code:: Defining how insns update the condition code. 42* Costs:: Defining relative costs of different operations. 43* Scheduling:: Adjusting the behavior of the instruction scheduler. 44* Sections:: Dividing storage into text, data, and other sections. 45* PIC:: Macros for position independent code. 46* Assembler Format:: Defining how to write insns and pseudo-ops to output. 47* Debugging Info:: Defining the format of debugging output. 48* Floating Point:: Handling floating point for cross-compilers. 49* Mode Switching:: Insertion of mode-switching instructions. 50* Target Attributes:: Defining target-specific uses of @code{__attribute__}. 51* Emulated TLS:: Emulated TLS support. 52* MIPS Coprocessors:: MIPS coprocessor support and how to customize it. 53* PCH Target:: Validity checking for precompiled headers. 54* C++ ABI:: Controlling C++ ABI changes. 55* D Language and ABI:: Controlling D ABI changes. 56* Named Address Spaces:: Adding support for named address spaces 57* Misc:: Everything else. 58@end menu 59 60@node Target Structure 61@section The Global @code{targetm} Variable 62@cindex target hooks 63@cindex target functions 64 65@deftypevar {struct gcc_target} targetm 66The target @file{.c} file must define the global @code{targetm} variable 67which contains pointers to functions and data relating to the target 68machine. The variable is declared in @file{target.h}; 69@file{target-def.h} defines the macro @code{TARGET_INITIALIZER} which is 70used to initialize the variable, and macros for the default initializers 71for elements of the structure. The @file{.c} file should override those 72macros for which the default definition is inappropriate. For example: 73@smallexample 74#include "target.h" 75#include "target-def.h" 76 77/* @r{Initialize the GCC target structure.} */ 78 79#undef TARGET_COMP_TYPE_ATTRIBUTES 80#define TARGET_COMP_TYPE_ATTRIBUTES @var{machine}_comp_type_attributes 81 82struct gcc_target targetm = TARGET_INITIALIZER; 83@end smallexample 84@end deftypevar 85 86Where a macro should be defined in the @file{.c} file in this manner to 87form part of the @code{targetm} structure, it is documented below as a 88``Target Hook'' with a prototype. Many macros will change in future 89from being defined in the @file{.h} file to being part of the 90@code{targetm} structure. 91 92Similarly, there is a @code{targetcm} variable for hooks that are 93specific to front ends for C-family languages, documented as ``C 94Target Hook''. This is declared in @file{c-family/c-target.h}, the 95initializer @code{TARGETCM_INITIALIZER} in 96@file{c-family/c-target-def.h}. If targets initialize @code{targetcm} 97themselves, they should set @code{target_has_targetcm=yes} in 98@file{config.gcc}; otherwise a default definition is used. 99 100Similarly, there is a @code{targetm_common} variable for hooks that 101are shared between the compiler driver and the compilers proper, 102documented as ``Common Target Hook''. This is declared in 103@file{common/common-target.h}, the initializer 104@code{TARGETM_COMMON_INITIALIZER} in 105@file{common/common-target-def.h}. If targets initialize 106@code{targetm_common} themselves, they should set 107@code{target_has_targetm_common=yes} in @file{config.gcc}; otherwise a 108default definition is used. 109 110Similarly, there is a @code{targetdm} variable for hooks that are 111specific to the D language front end, documented as ``D Target Hook''. 112This is declared in @file{d/d-target.h}, the initializer 113@code{TARGETDM_INITIALIZER} in @file{d/d-target-def.h}. If targets 114initialize @code{targetdm} themselves, they should set 115@code{target_has_targetdm=yes} in @file{config.gcc}; otherwise a default 116definition is used. 117 118@node Driver 119@section Controlling the Compilation Driver, @file{gcc} 120@cindex driver 121@cindex controlling the compilation driver 122 123@c prevent bad page break with this line 124You can control the compilation driver. 125 126@defmac DRIVER_SELF_SPECS 127A list of specs for the driver itself. It should be a suitable 128initializer for an array of strings, with no surrounding braces. 129 130The driver applies these specs to its own command line between loading 131default @file{specs} files (but not command-line specified ones) and 132choosing the multilib directory or running any subcommands. It 133applies them in the order given, so each spec can depend on the 134options added by earlier ones. It is also possible to remove options 135using @samp{%<@var{option}} in the usual way. 136 137This macro can be useful when a port has several interdependent target 138options. It provides a way of standardizing the command line so 139that the other specs are easier to write. 140 141Do not define this macro if it does not need to do anything. 142@end defmac 143 144@defmac OPTION_DEFAULT_SPECS 145A list of specs used to support configure-time default options (i.e.@: 146@option{--with} options) in the driver. It should be a suitable initializer 147for an array of structures, each containing two strings, without the 148outermost pair of surrounding braces. 149 150The first item in the pair is the name of the default. This must match 151the code in @file{config.gcc} for the target. The second item is a spec 152to apply if a default with this name was specified. The string 153@samp{%(VALUE)} in the spec will be replaced by the value of the default 154everywhere it occurs. 155 156The driver will apply these specs to its own command line between loading 157default @file{specs} files and processing @code{DRIVER_SELF_SPECS}, using 158the same mechanism as @code{DRIVER_SELF_SPECS}. 159 160Do not define this macro if it does not need to do anything. 161@end defmac 162 163@defmac CPP_SPEC 164A C string constant that tells the GCC driver program options to 165pass to CPP@. It can also specify how to translate options you 166give to GCC into options for GCC to pass to the CPP@. 167 168Do not define this macro if it does not need to do anything. 169@end defmac 170 171@defmac CPLUSPLUS_CPP_SPEC 172This macro is just like @code{CPP_SPEC}, but is used for C++, rather 173than C@. If you do not define this macro, then the value of 174@code{CPP_SPEC} (if any) will be used instead. 175@end defmac 176 177@defmac CC1_SPEC 178A C string constant that tells the GCC driver program options to 179pass to @code{cc1}, @code{cc1plus}, @code{f771}, and the other language 180front ends. 181It can also specify how to translate options you give to GCC into options 182for GCC to pass to front ends. 183 184Do not define this macro if it does not need to do anything. 185@end defmac 186 187@defmac CC1PLUS_SPEC 188A C string constant that tells the GCC driver program options to 189pass to @code{cc1plus}. It can also specify how to translate options you 190give to GCC into options for GCC to pass to the @code{cc1plus}. 191 192Do not define this macro if it does not need to do anything. 193Note that everything defined in CC1_SPEC is already passed to 194@code{cc1plus} so there is no need to duplicate the contents of 195CC1_SPEC in CC1PLUS_SPEC@. 196@end defmac 197 198@defmac ASM_SPEC 199A C string constant that tells the GCC driver program options to 200pass to the assembler. It can also specify how to translate options 201you give to GCC into options for GCC to pass to the assembler. 202See the file @file{sun3.h} for an example of this. 203 204Do not define this macro if it does not need to do anything. 205@end defmac 206 207@defmac ASM_FINAL_SPEC 208A C string constant that tells the GCC driver program how to 209run any programs which cleanup after the normal assembler. 210Normally, this is not needed. See the file @file{mips.h} for 211an example of this. 212 213Do not define this macro if it does not need to do anything. 214@end defmac 215 216@defmac AS_NEEDS_DASH_FOR_PIPED_INPUT 217Define this macro, with no value, if the driver should give the assembler 218an argument consisting of a single dash, @option{-}, to instruct it to 219read from its standard input (which will be a pipe connected to the 220output of the compiler proper). This argument is given after any 221@option{-o} option specifying the name of the output file. 222 223If you do not define this macro, the assembler is assumed to read its 224standard input if given no non-option arguments. If your assembler 225cannot read standard input at all, use a @samp{%@{pipe:%e@}} construct; 226see @file{mips.h} for instance. 227@end defmac 228 229@defmac LINK_SPEC 230A C string constant that tells the GCC driver program options to 231pass to the linker. It can also specify how to translate options you 232give to GCC into options for GCC to pass to the linker. 233 234Do not define this macro if it does not need to do anything. 235@end defmac 236 237@defmac LIB_SPEC 238Another C string constant used much like @code{LINK_SPEC}. The difference 239between the two is that @code{LIB_SPEC} is used at the end of the 240command given to the linker. 241 242If this macro is not defined, a default is provided that 243loads the standard C library from the usual place. See @file{gcc.c}. 244@end defmac 245 246@defmac LIBGCC_SPEC 247Another C string constant that tells the GCC driver program 248how and when to place a reference to @file{libgcc.a} into the 249linker command line. This constant is placed both before and after 250the value of @code{LIB_SPEC}. 251 252If this macro is not defined, the GCC driver provides a default that 253passes the string @option{-lgcc} to the linker. 254@end defmac 255 256@defmac REAL_LIBGCC_SPEC 257By default, if @code{ENABLE_SHARED_LIBGCC} is defined, the 258@code{LIBGCC_SPEC} is not directly used by the driver program but is 259instead modified to refer to different versions of @file{libgcc.a} 260depending on the values of the command line flags @option{-static}, 261@option{-shared}, @option{-static-libgcc}, and @option{-shared-libgcc}. On 262targets where these modifications are inappropriate, define 263@code{REAL_LIBGCC_SPEC} instead. @code{REAL_LIBGCC_SPEC} tells the 264driver how to place a reference to @file{libgcc} on the link command 265line, but, unlike @code{LIBGCC_SPEC}, it is used unmodified. 266@end defmac 267 268@defmac USE_LD_AS_NEEDED 269A macro that controls the modifications to @code{LIBGCC_SPEC} 270mentioned in @code{REAL_LIBGCC_SPEC}. If nonzero, a spec will be 271generated that uses @option{--as-needed} or equivalent options and the 272shared @file{libgcc} in place of the 273static exception handler library, when linking without any of 274@code{-static}, @code{-static-libgcc}, or @code{-shared-libgcc}. 275@end defmac 276 277@defmac LINK_EH_SPEC 278If defined, this C string constant is added to @code{LINK_SPEC}. 279When @code{USE_LD_AS_NEEDED} is zero or undefined, it also affects 280the modifications to @code{LIBGCC_SPEC} mentioned in 281@code{REAL_LIBGCC_SPEC}. 282@end defmac 283 284@defmac STARTFILE_SPEC 285Another C string constant used much like @code{LINK_SPEC}. The 286difference between the two is that @code{STARTFILE_SPEC} is used at 287the very beginning of the command given to the linker. 288 289If this macro is not defined, a default is provided that loads the 290standard C startup file from the usual place. See @file{gcc.c}. 291@end defmac 292 293@defmac ENDFILE_SPEC 294Another C string constant used much like @code{LINK_SPEC}. The 295difference between the two is that @code{ENDFILE_SPEC} is used at 296the very end of the command given to the linker. 297 298Do not define this macro if it does not need to do anything. 299@end defmac 300 301@defmac THREAD_MODEL_SPEC 302GCC @code{-v} will print the thread model GCC was configured to use. 303However, this doesn't work on platforms that are multilibbed on thread 304models, such as AIX 4.3. On such platforms, define 305@code{THREAD_MODEL_SPEC} such that it evaluates to a string without 306blanks that names one of the recognized thread models. @code{%*}, the 307default value of this macro, will expand to the value of 308@code{thread_file} set in @file{config.gcc}. 309@end defmac 310 311@defmac SYSROOT_SUFFIX_SPEC 312Define this macro to add a suffix to the target sysroot when GCC is 313configured with a sysroot. This will cause GCC to search for usr/lib, 314et al, within sysroot+suffix. 315@end defmac 316 317@defmac SYSROOT_HEADERS_SUFFIX_SPEC 318Define this macro to add a headers_suffix to the target sysroot when 319GCC is configured with a sysroot. This will cause GCC to pass the 320updated sysroot+headers_suffix to CPP, causing it to search for 321usr/include, et al, within sysroot+headers_suffix. 322@end defmac 323 324@defmac EXTRA_SPECS 325Define this macro to provide additional specifications to put in the 326@file{specs} file that can be used in various specifications like 327@code{CC1_SPEC}. 328 329The definition should be an initializer for an array of structures, 330containing a string constant, that defines the specification name, and a 331string constant that provides the specification. 332 333Do not define this macro if it does not need to do anything. 334 335@code{EXTRA_SPECS} is useful when an architecture contains several 336related targets, which have various @code{@dots{}_SPECS} which are similar 337to each other, and the maintainer would like one central place to keep 338these definitions. 339 340For example, the PowerPC System V.4 targets use @code{EXTRA_SPECS} to 341define either @code{_CALL_SYSV} when the System V calling sequence is 342used or @code{_CALL_AIX} when the older AIX-based calling sequence is 343used. 344 345The @file{config/rs6000/rs6000.h} target file defines: 346 347@smallexample 348#define EXTRA_SPECS \ 349 @{ "cpp_sysv_default", CPP_SYSV_DEFAULT @}, 350 351#define CPP_SYS_DEFAULT "" 352@end smallexample 353 354The @file{config/rs6000/sysv.h} target file defines: 355@smallexample 356#undef CPP_SPEC 357#define CPP_SPEC \ 358"%@{posix: -D_POSIX_SOURCE @} \ 359%@{mcall-sysv: -D_CALL_SYSV @} \ 360%@{!mcall-sysv: %(cpp_sysv_default) @} \ 361%@{msoft-float: -D_SOFT_FLOAT@} %@{mcpu=403: -D_SOFT_FLOAT@}" 362 363#undef CPP_SYSV_DEFAULT 364#define CPP_SYSV_DEFAULT "-D_CALL_SYSV" 365@end smallexample 366 367while the @file{config/rs6000/eabiaix.h} target file defines 368@code{CPP_SYSV_DEFAULT} as: 369 370@smallexample 371#undef CPP_SYSV_DEFAULT 372#define CPP_SYSV_DEFAULT "-D_CALL_AIX" 373@end smallexample 374@end defmac 375 376@defmac LINK_LIBGCC_SPECIAL_1 377Define this macro if the driver program should find the library 378@file{libgcc.a}. If you do not define this macro, the driver program will pass 379the argument @option{-lgcc} to tell the linker to do the search. 380@end defmac 381 382@defmac LINK_GCC_C_SEQUENCE_SPEC 383The sequence in which libgcc and libc are specified to the linker. 384By default this is @code{%G %L %G}. 385@end defmac 386 387@defmac POST_LINK_SPEC 388Define this macro to add additional steps to be executed after linker. 389The default value of this macro is empty string. 390@end defmac 391 392@defmac LINK_COMMAND_SPEC 393A C string constant giving the complete command line need to execute the 394linker. When you do this, you will need to update your port each time a 395change is made to the link command line within @file{gcc.c}. Therefore, 396define this macro only if you need to completely redefine the command 397line for invoking the linker and there is no other way to accomplish 398the effect you need. Overriding this macro may be avoidable by overriding 399@code{LINK_GCC_C_SEQUENCE_SPEC} instead. 400@end defmac 401 402@deftypevr {Common Target Hook} bool TARGET_ALWAYS_STRIP_DOTDOT 403True if @file{..} components should always be removed from directory names computed relative to GCC's internal directories, false (default) if such components should be preserved and directory names containing them passed to other tools such as the linker. 404@end deftypevr 405 406@defmac MULTILIB_DEFAULTS 407Define this macro as a C expression for the initializer of an array of 408string to tell the driver program which options are defaults for this 409target and thus do not need to be handled specially when using 410@code{MULTILIB_OPTIONS}. 411 412Do not define this macro if @code{MULTILIB_OPTIONS} is not defined in 413the target makefile fragment or if none of the options listed in 414@code{MULTILIB_OPTIONS} are set by default. 415@xref{Target Fragment}. 416@end defmac 417 418@defmac RELATIVE_PREFIX_NOT_LINKDIR 419Define this macro to tell @command{gcc} that it should only translate 420a @option{-B} prefix into a @option{-L} linker option if the prefix 421indicates an absolute file name. 422@end defmac 423 424@defmac MD_EXEC_PREFIX 425If defined, this macro is an additional prefix to try after 426@code{STANDARD_EXEC_PREFIX}. @code{MD_EXEC_PREFIX} is not searched 427when the compiler is built as a cross 428compiler. If you define @code{MD_EXEC_PREFIX}, then be sure to add it 429to the list of directories used to find the assembler in @file{configure.ac}. 430@end defmac 431 432@defmac STANDARD_STARTFILE_PREFIX 433Define this macro as a C string constant if you wish to override the 434standard choice of @code{libdir} as the default prefix to 435try when searching for startup files such as @file{crt0.o}. 436@code{STANDARD_STARTFILE_PREFIX} is not searched when the compiler 437is built as a cross compiler. 438@end defmac 439 440@defmac STANDARD_STARTFILE_PREFIX_1 441Define this macro as a C string constant if you wish to override the 442standard choice of @code{/lib} as a prefix to try after the default prefix 443when searching for startup files such as @file{crt0.o}. 444@code{STANDARD_STARTFILE_PREFIX_1} is not searched when the compiler 445is built as a cross compiler. 446@end defmac 447 448@defmac STANDARD_STARTFILE_PREFIX_2 449Define this macro as a C string constant if you wish to override the 450standard choice of @code{/lib} as yet another prefix to try after the 451default prefix when searching for startup files such as @file{crt0.o}. 452@code{STANDARD_STARTFILE_PREFIX_2} is not searched when the compiler 453is built as a cross compiler. 454@end defmac 455 456@defmac MD_STARTFILE_PREFIX 457If defined, this macro supplies an additional prefix to try after the 458standard prefixes. @code{MD_EXEC_PREFIX} is not searched when the 459compiler is built as a cross compiler. 460@end defmac 461 462@defmac MD_STARTFILE_PREFIX_1 463If defined, this macro supplies yet another prefix to try after the 464standard prefixes. It is not searched when the compiler is built as a 465cross compiler. 466@end defmac 467 468@defmac INIT_ENVIRONMENT 469Define this macro as a C string constant if you wish to set environment 470variables for programs called by the driver, such as the assembler and 471loader. The driver passes the value of this macro to @code{putenv} to 472initialize the necessary environment variables. 473@end defmac 474 475@defmac LOCAL_INCLUDE_DIR 476Define this macro as a C string constant if you wish to override the 477standard choice of @file{/usr/local/include} as the default prefix to 478try when searching for local header files. @code{LOCAL_INCLUDE_DIR} 479comes before @code{NATIVE_SYSTEM_HEADER_DIR} (set in 480@file{config.gcc}, normally @file{/usr/include}) in the search order. 481 482Cross compilers do not search either @file{/usr/local/include} or its 483replacement. 484@end defmac 485 486@defmac NATIVE_SYSTEM_HEADER_COMPONENT 487The ``component'' corresponding to @code{NATIVE_SYSTEM_HEADER_DIR}. 488See @code{INCLUDE_DEFAULTS}, below, for the description of components. 489If you do not define this macro, no component is used. 490@end defmac 491 492@defmac INCLUDE_DEFAULTS 493Define this macro if you wish to override the entire default search path 494for include files. For a native compiler, the default search path 495usually consists of @code{GCC_INCLUDE_DIR}, @code{LOCAL_INCLUDE_DIR}, 496@code{GPLUSPLUS_INCLUDE_DIR}, and 497@code{NATIVE_SYSTEM_HEADER_DIR}. In addition, @code{GPLUSPLUS_INCLUDE_DIR} 498and @code{GCC_INCLUDE_DIR} are defined automatically by @file{Makefile}, 499and specify private search areas for GCC@. The directory 500@code{GPLUSPLUS_INCLUDE_DIR} is used only for C++ programs. 501 502The definition should be an initializer for an array of structures. 503Each array element should have four elements: the directory name (a 504string constant), the component name (also a string constant), a flag 505for C++-only directories, 506and a flag showing that the includes in the directory don't need to be 507wrapped in @code{extern @samp{C}} when compiling C++. Mark the end of 508the array with a null element. 509 510The component name denotes what GNU package the include file is part of, 511if any, in all uppercase letters. For example, it might be @samp{GCC} 512or @samp{BINUTILS}. If the package is part of a vendor-supplied 513operating system, code the component name as @samp{0}. 514 515For example, here is the definition used for VAX/VMS: 516 517@smallexample 518#define INCLUDE_DEFAULTS \ 519@{ \ 520 @{ "GNU_GXX_INCLUDE:", "G++", 1, 1@}, \ 521 @{ "GNU_CC_INCLUDE:", "GCC", 0, 0@}, \ 522 @{ "SYS$SYSROOT:[SYSLIB.]", 0, 0, 0@}, \ 523 @{ ".", 0, 0, 0@}, \ 524 @{ 0, 0, 0, 0@} \ 525@} 526@end smallexample 527@end defmac 528 529Here is the order of prefixes tried for exec files: 530 531@enumerate 532@item 533Any prefixes specified by the user with @option{-B}. 534 535@item 536The environment variable @code{GCC_EXEC_PREFIX} or, if @code{GCC_EXEC_PREFIX} 537is not set and the compiler has not been installed in the configure-time 538@var{prefix}, the location in which the compiler has actually been installed. 539 540@item 541The directories specified by the environment variable @code{COMPILER_PATH}. 542 543@item 544The macro @code{STANDARD_EXEC_PREFIX}, if the compiler has been installed 545in the configured-time @var{prefix}. 546 547@item 548The location @file{/usr/libexec/gcc/}, but only if this is a native compiler. 549 550@item 551The location @file{/usr/lib/gcc/}, but only if this is a native compiler. 552 553@item 554The macro @code{MD_EXEC_PREFIX}, if defined, but only if this is a native 555compiler. 556@end enumerate 557 558Here is the order of prefixes tried for startfiles: 559 560@enumerate 561@item 562Any prefixes specified by the user with @option{-B}. 563 564@item 565The environment variable @code{GCC_EXEC_PREFIX} or its automatically determined 566value based on the installed toolchain location. 567 568@item 569The directories specified by the environment variable @code{LIBRARY_PATH} 570(or port-specific name; native only, cross compilers do not use this). 571 572@item 573The macro @code{STANDARD_EXEC_PREFIX}, but only if the toolchain is installed 574in the configured @var{prefix} or this is a native compiler. 575 576@item 577The location @file{/usr/lib/gcc/}, but only if this is a native compiler. 578 579@item 580The macro @code{MD_EXEC_PREFIX}, if defined, but only if this is a native 581compiler. 582 583@item 584The macro @code{MD_STARTFILE_PREFIX}, if defined, but only if this is a 585native compiler, or we have a target system root. 586 587@item 588The macro @code{MD_STARTFILE_PREFIX_1}, if defined, but only if this is a 589native compiler, or we have a target system root. 590 591@item 592The macro @code{STANDARD_STARTFILE_PREFIX}, with any sysroot modifications. 593If this path is relative it will be prefixed by @code{GCC_EXEC_PREFIX} and 594the machine suffix or @code{STANDARD_EXEC_PREFIX} and the machine suffix. 595 596@item 597The macro @code{STANDARD_STARTFILE_PREFIX_1}, but only if this is a native 598compiler, or we have a target system root. The default for this macro is 599@file{/lib/}. 600 601@item 602The macro @code{STANDARD_STARTFILE_PREFIX_2}, but only if this is a native 603compiler, or we have a target system root. The default for this macro is 604@file{/usr/lib/}. 605@end enumerate 606 607@node Run-time Target 608@section Run-time Target Specification 609@cindex run-time target specification 610@cindex predefined macros 611@cindex target specifications 612 613@c prevent bad page break with this line 614Here are run-time target specifications. 615 616@defmac TARGET_CPU_CPP_BUILTINS () 617This function-like macro expands to a block of code that defines 618built-in preprocessor macros and assertions for the target CPU, using 619the functions @code{builtin_define}, @code{builtin_define_std} and 620@code{builtin_assert}. When the front end 621calls this macro it provides a trailing semicolon, and since it has 622finished command line option processing your code can use those 623results freely. 624 625@code{builtin_assert} takes a string in the form you pass to the 626command-line option @option{-A}, such as @code{cpu=mips}, and creates 627the assertion. @code{builtin_define} takes a string in the form 628accepted by option @option{-D} and unconditionally defines the macro. 629 630@code{builtin_define_std} takes a string representing the name of an 631object-like macro. If it doesn't lie in the user's namespace, 632@code{builtin_define_std} defines it unconditionally. Otherwise, it 633defines a version with two leading underscores, and another version 634with two leading and trailing underscores, and defines the original 635only if an ISO standard was not requested on the command line. For 636example, passing @code{unix} defines @code{__unix}, @code{__unix__} 637and possibly @code{unix}; passing @code{_mips} defines @code{__mips}, 638@code{__mips__} and possibly @code{_mips}, and passing @code{_ABI64} 639defines only @code{_ABI64}. 640 641You can also test for the C dialect being compiled. The variable 642@code{c_language} is set to one of @code{clk_c}, @code{clk_cplusplus} 643or @code{clk_objective_c}. Note that if we are preprocessing 644assembler, this variable will be @code{clk_c} but the function-like 645macro @code{preprocessing_asm_p()} will return true, so you might want 646to check for that first. If you need to check for strict ANSI, the 647variable @code{flag_iso} can be used. The function-like macro 648@code{preprocessing_trad_p()} can be used to check for traditional 649preprocessing. 650@end defmac 651 652@defmac TARGET_OS_CPP_BUILTINS () 653Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional 654and is used for the target operating system instead. 655@end defmac 656 657@defmac TARGET_OBJFMT_CPP_BUILTINS () 658Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional 659and is used for the target object format. @file{elfos.h} uses this 660macro to define @code{__ELF__}, so you probably do not need to define 661it yourself. 662@end defmac 663 664@deftypevar {extern int} target_flags 665This variable is declared in @file{options.h}, which is included before 666any target-specific headers. 667@end deftypevar 668 669@deftypevr {Common Target Hook} int TARGET_DEFAULT_TARGET_FLAGS 670This variable specifies the initial value of @code{target_flags}. 671Its default setting is 0. 672@end deftypevr 673 674@cindex optional hardware or system features 675@cindex features, optional, in system conventions 676 677@deftypefn {Common Target Hook} bool TARGET_HANDLE_OPTION (struct gcc_options *@var{opts}, struct gcc_options *@var{opts_set}, const struct cl_decoded_option *@var{decoded}, location_t @var{loc}) 678This hook is called whenever the user specifies one of the 679target-specific options described by the @file{.opt} definition files 680(@pxref{Options}). It has the opportunity to do some option-specific 681processing and should return true if the option is valid. The default 682definition does nothing but return true. 683 684@var{decoded} specifies the option and its arguments. @var{opts} and 685@var{opts_set} are the @code{gcc_options} structures to be used for 686storing option state, and @var{loc} is the location at which the 687option was passed (@code{UNKNOWN_LOCATION} except for options passed 688via attributes). 689@end deftypefn 690 691@deftypefn {C Target Hook} bool TARGET_HANDLE_C_OPTION (size_t @var{code}, const char *@var{arg}, int @var{value}) 692This target hook is called whenever the user specifies one of the 693target-specific C language family options described by the @file{.opt} 694definition files(@pxref{Options}). It has the opportunity to do some 695option-specific processing and should return true if the option is 696valid. The arguments are like for @code{TARGET_HANDLE_OPTION}. The 697default definition does nothing but return false. 698 699In general, you should use @code{TARGET_HANDLE_OPTION} to handle 700options. However, if processing an option requires routines that are 701only available in the C (and related language) front ends, then you 702should use @code{TARGET_HANDLE_C_OPTION} instead. 703@end deftypefn 704 705@deftypefn {C Target Hook} tree TARGET_OBJC_CONSTRUCT_STRING_OBJECT (tree @var{string}) 706Targets may provide a string object type that can be used within and between C, C++ and their respective Objective-C dialects. A string object might, for example, embed encoding and length information. These objects are considered opaque to the compiler and handled as references. An ideal implementation makes the composition of the string object match that of the Objective-C @code{NSString} (@code{NXString} for GNUStep), allowing efficient interworking between C-only and Objective-C code. If a target implements string objects then this hook should return a reference to such an object constructed from the normal `C' string representation provided in @var{string}. At present, the hook is used by Objective-C only, to obtain a common-format string object when the target provides one. 707@end deftypefn 708 709@deftypefn {C Target Hook} void TARGET_OBJC_DECLARE_UNRESOLVED_CLASS_REFERENCE (const char *@var{classname}) 710Declare that Objective C class @var{classname} is referenced by the current TU. 711@end deftypefn 712 713@deftypefn {C Target Hook} void TARGET_OBJC_DECLARE_CLASS_DEFINITION (const char *@var{classname}) 714Declare that Objective C class @var{classname} is defined by the current TU. 715@end deftypefn 716 717@deftypefn {C Target Hook} bool TARGET_STRING_OBJECT_REF_TYPE_P (const_tree @var{stringref}) 718If a target implements string objects then this hook should return @code{true} if @var{stringref} is a valid reference to such an object. 719@end deftypefn 720 721@deftypefn {C Target Hook} void TARGET_CHECK_STRING_OBJECT_FORMAT_ARG (tree @var{format_arg}, tree @var{args_list}) 722If a target implements string objects then this hook should should provide a facility to check the function arguments in @var{args_list} against the format specifiers in @var{format_arg} where the type of @var{format_arg} is one recognized as a valid string reference type. 723@end deftypefn 724 725@deftypefn {Target Hook} void TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE (void) 726This target function is similar to the hook @code{TARGET_OPTION_OVERRIDE} 727but is called when the optimize level is changed via an attribute or 728pragma or when it is reset at the end of the code affected by the 729attribute or pragma. It is not called at the beginning of compilation 730when @code{TARGET_OPTION_OVERRIDE} is called so if you want to perform these 731actions then, you should have @code{TARGET_OPTION_OVERRIDE} call 732@code{TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE}. 733@end deftypefn 734 735@defmac C_COMMON_OVERRIDE_OPTIONS 736This is similar to the @code{TARGET_OPTION_OVERRIDE} hook 737but is only used in the C 738language frontends (C, Objective-C, C++, Objective-C++) and so can be 739used to alter option flag variables which only exist in those 740frontends. 741@end defmac 742 743@deftypevr {Common Target Hook} {const struct default_options *} TARGET_OPTION_OPTIMIZATION_TABLE 744Some machines may desire to change what optimizations are performed for 745various optimization levels. This variable, if defined, describes 746options to enable at particular sets of optimization levels. These 747options are processed once 748just after the optimization level is determined and before the remainder 749of the command options have been parsed, so may be overridden by other 750options passed explicitly. 751 752This processing is run once at program startup and when the optimization 753options are changed via @code{#pragma GCC optimize} or by using the 754@code{optimize} attribute. 755@end deftypevr 756 757@deftypefn {Common Target Hook} void TARGET_OPTION_INIT_STRUCT (struct gcc_options *@var{opts}) 758Set target-dependent initial values of fields in @var{opts}. 759@end deftypefn 760 761@defmac SWITCHABLE_TARGET 762Some targets need to switch between substantially different subtargets 763during compilation. For example, the MIPS target has one subtarget for 764the traditional MIPS architecture and another for MIPS16. Source code 765can switch between these two subarchitectures using the @code{mips16} 766and @code{nomips16} attributes. 767 768Such subtargets can differ in things like the set of available 769registers, the set of available instructions, the costs of various 770operations, and so on. GCC caches a lot of this type of information 771in global variables, and recomputing them for each subtarget takes a 772significant amount of time. The compiler therefore provides a facility 773for maintaining several versions of the global variables and quickly 774switching between them; see @file{target-globals.h} for details. 775 776Define this macro to 1 if your target needs this facility. The default 777is 0. 778@end defmac 779 780@deftypefn {Target Hook} bool TARGET_FLOAT_EXCEPTIONS_ROUNDING_SUPPORTED_P (void) 781Returns true if the target supports IEEE 754 floating-point exceptions and rounding modes, false otherwise. This is intended to relate to the @code{float} and @code{double} types, but not necessarily @code{long double}. By default, returns true if the @code{adddf3} instruction pattern is available and false otherwise, on the assumption that hardware floating point supports exceptions and rounding modes but software floating point does not. 782@end deftypefn 783 784@node Per-Function Data 785@section Defining data structures for per-function information. 786@cindex per-function data 787@cindex data structures 788 789If the target needs to store information on a per-function basis, GCC 790provides a macro and a couple of variables to allow this. Note, just 791using statics to store the information is a bad idea, since GCC supports 792nested functions, so you can be halfway through encoding one function 793when another one comes along. 794 795GCC defines a data structure called @code{struct function} which 796contains all of the data specific to an individual function. This 797structure contains a field called @code{machine} whose type is 798@code{struct machine_function *}, which can be used by targets to point 799to their own specific data. 800 801If a target needs per-function specific data it should define the type 802@code{struct machine_function} and also the macro @code{INIT_EXPANDERS}. 803This macro should be used to initialize the function pointer 804@code{init_machine_status}. This pointer is explained below. 805 806One typical use of per-function, target specific data is to create an 807RTX to hold the register containing the function's return address. This 808RTX can then be used to implement the @code{__builtin_return_address} 809function, for level 0. 810 811Note---earlier implementations of GCC used a single data area to hold 812all of the per-function information. Thus when processing of a nested 813function began the old per-function data had to be pushed onto a 814stack, and when the processing was finished, it had to be popped off the 815stack. GCC used to provide function pointers called 816@code{save_machine_status} and @code{restore_machine_status} to handle 817the saving and restoring of the target specific information. Since the 818single data area approach is no longer used, these pointers are no 819longer supported. 820 821@defmac INIT_EXPANDERS 822Macro called to initialize any target specific information. This macro 823is called once per function, before generation of any RTL has begun. 824The intention of this macro is to allow the initialization of the 825function pointer @code{init_machine_status}. 826@end defmac 827 828@deftypevar {void (*)(struct function *)} init_machine_status 829If this function pointer is non-@code{NULL} it will be called once per 830function, before function compilation starts, in order to allow the 831target to perform any target specific initialization of the 832@code{struct function} structure. It is intended that this would be 833used to initialize the @code{machine} of that structure. 834 835@code{struct machine_function} structures are expected to be freed by GC@. 836Generally, any memory that they reference must be allocated by using 837GC allocation, including the structure itself. 838@end deftypevar 839 840@node Storage Layout 841@section Storage Layout 842@cindex storage layout 843 844Note that the definitions of the macros in this table which are sizes or 845alignments measured in bits do not need to be constant. They can be C 846expressions that refer to static variables, such as the @code{target_flags}. 847@xref{Run-time Target}. 848 849@defmac BITS_BIG_ENDIAN 850Define this macro to have the value 1 if the most significant bit in a 851byte has the lowest number; otherwise define it to have the value zero. 852This means that bit-field instructions count from the most significant 853bit. If the machine has no bit-field instructions, then this must still 854be defined, but it doesn't matter which value it is defined to. This 855macro need not be a constant. 856 857This macro does not affect the way structure fields are packed into 858bytes or words; that is controlled by @code{BYTES_BIG_ENDIAN}. 859@end defmac 860 861@defmac BYTES_BIG_ENDIAN 862Define this macro to have the value 1 if the most significant byte in a 863word has the lowest number. This macro need not be a constant. 864@end defmac 865 866@defmac WORDS_BIG_ENDIAN 867Define this macro to have the value 1 if, in a multiword object, the 868most significant word has the lowest number. This applies to both 869memory locations and registers; see @code{REG_WORDS_BIG_ENDIAN} if the 870order of words in memory is not the same as the order in registers. This 871macro need not be a constant. 872@end defmac 873 874@defmac REG_WORDS_BIG_ENDIAN 875On some machines, the order of words in a multiword object differs between 876registers in memory. In such a situation, define this macro to describe 877the order of words in a register. The macro @code{WORDS_BIG_ENDIAN} controls 878the order of words in memory. 879@end defmac 880 881@defmac FLOAT_WORDS_BIG_ENDIAN 882Define this macro to have the value 1 if @code{DFmode}, @code{XFmode} or 883@code{TFmode} floating point numbers are stored in memory with the word 884containing the sign bit at the lowest address; otherwise define it to 885have the value 0. This macro need not be a constant. 886 887You need not define this macro if the ordering is the same as for 888multi-word integers. 889@end defmac 890 891@defmac BITS_PER_WORD 892Number of bits in a word. If you do not define this macro, the default 893is @code{BITS_PER_UNIT * UNITS_PER_WORD}. 894@end defmac 895 896@defmac MAX_BITS_PER_WORD 897Maximum number of bits in a word. If this is undefined, the default is 898@code{BITS_PER_WORD}. Otherwise, it is the constant value that is the 899largest value that @code{BITS_PER_WORD} can have at run-time. 900@end defmac 901 902@defmac UNITS_PER_WORD 903Number of storage units in a word; normally the size of a general-purpose 904register, a power of two from 1 or 8. 905@end defmac 906 907@defmac MIN_UNITS_PER_WORD 908Minimum number of units in a word. If this is undefined, the default is 909@code{UNITS_PER_WORD}. Otherwise, it is the constant value that is the 910smallest value that @code{UNITS_PER_WORD} can have at run-time. 911@end defmac 912 913@defmac POINTER_SIZE 914Width of a pointer, in bits. You must specify a value no wider than the 915width of @code{Pmode}. If it is not equal to the width of @code{Pmode}, 916you must define @code{POINTERS_EXTEND_UNSIGNED}. If you do not specify 917a value the default is @code{BITS_PER_WORD}. 918@end defmac 919 920@defmac POINTERS_EXTEND_UNSIGNED 921A C expression that determines how pointers should be extended from 922@code{ptr_mode} to either @code{Pmode} or @code{word_mode}. It is 923greater than zero if pointers should be zero-extended, zero if they 924should be sign-extended, and negative if some other sort of conversion 925is needed. In the last case, the extension is done by the target's 926@code{ptr_extend} instruction. 927 928You need not define this macro if the @code{ptr_mode}, @code{Pmode} 929and @code{word_mode} are all the same width. 930@end defmac 931 932@defmac PROMOTE_MODE (@var{m}, @var{unsignedp}, @var{type}) 933A macro to update @var{m} and @var{unsignedp} when an object whose type 934is @var{type} and which has the specified mode and signedness is to be 935stored in a register. This macro is only called when @var{type} is a 936scalar type. 937 938On most RISC machines, which only have operations that operate on a full 939register, define this macro to set @var{m} to @code{word_mode} if 940@var{m} is an integer mode narrower than @code{BITS_PER_WORD}. In most 941cases, only integer modes should be widened because wider-precision 942floating-point operations are usually more expensive than their narrower 943counterparts. 944 945For most machines, the macro definition does not change @var{unsignedp}. 946However, some machines, have instructions that preferentially handle 947either signed or unsigned quantities of certain modes. For example, on 948the DEC Alpha, 32-bit loads from memory and 32-bit add instructions 949sign-extend the result to 64 bits. On such machines, set 950@var{unsignedp} according to which kind of extension is more efficient. 951 952Do not define this macro if it would never modify @var{m}. 953@end defmac 954 955@deftypefn {Target Hook} {enum flt_eval_method} TARGET_C_EXCESS_PRECISION (enum excess_precision_type @var{type}) 956Return a value, with the same meaning as the C99 macro @code{FLT_EVAL_METHOD} that describes which excess precision should be applied. @var{type} is either @code{EXCESS_PRECISION_TYPE_IMPLICIT}, @code{EXCESS_PRECISION_TYPE_FAST}, or @code{EXCESS_PRECISION_TYPE_STANDARD}. For @code{EXCESS_PRECISION_TYPE_IMPLICIT}, the target should return which precision and range operations will be implictly evaluated in regardless of the excess precision explicitly added. For @code{EXCESS_PRECISION_TYPE_STANDARD} and @code{EXCESS_PRECISION_TYPE_FAST}, the target should return the explicit excess precision that should be added depending on the value set for @option{-fexcess-precision=@r{[}standard@r{|}fast@r{]}}. Note that unpredictable explicit excess precision does not make sense, so a target should never return @code{FLT_EVAL_METHOD_UNPREDICTABLE} when @var{type} is @code{EXCESS_PRECISION_TYPE_STANDARD} or @code{EXCESS_PRECISION_TYPE_FAST}. 957@end deftypefn 958 959@deftypefn {Target Hook} machine_mode TARGET_PROMOTE_FUNCTION_MODE (const_tree @var{type}, machine_mode @var{mode}, int *@var{punsignedp}, const_tree @var{funtype}, int @var{for_return}) 960Like @code{PROMOTE_MODE}, but it is applied to outgoing function arguments or 961function return values. The target hook should return the new mode 962and possibly change @code{*@var{punsignedp}} if the promotion should 963change signedness. This function is called only for scalar @emph{or 964pointer} types. 965 966@var{for_return} allows to distinguish the promotion of arguments and 967return values. If it is @code{1}, a return value is being promoted and 968@code{TARGET_FUNCTION_VALUE} must perform the same promotions done here. 969If it is @code{2}, the returned mode should be that of the register in 970which an incoming parameter is copied, or the outgoing result is computed; 971then the hook should return the same mode as @code{promote_mode}, though 972the signedness may be different. 973 974@var{type} can be NULL when promoting function arguments of libcalls. 975 976The default is to not promote arguments and return values. You can 977also define the hook to @code{default_promote_function_mode_always_promote} 978if you would like to apply the same rules given by @code{PROMOTE_MODE}. 979@end deftypefn 980 981@defmac PARM_BOUNDARY 982Normal alignment required for function parameters on the stack, in 983bits. All stack parameters receive at least this much alignment 984regardless of data type. On most machines, this is the same as the 985size of an integer. 986@end defmac 987 988@defmac STACK_BOUNDARY 989Define this macro to the minimum alignment enforced by hardware for the 990stack pointer on this machine. The definition is a C expression for the 991desired alignment (measured in bits). This value is used as a default 992if @code{PREFERRED_STACK_BOUNDARY} is not defined. On most machines, 993this should be the same as @code{PARM_BOUNDARY}. 994@end defmac 995 996@defmac PREFERRED_STACK_BOUNDARY 997Define this macro if you wish to preserve a certain alignment for the 998stack pointer, greater than what the hardware enforces. The definition 999is a C expression for the desired alignment (measured in bits). This 1000macro must evaluate to a value equal to or larger than 1001@code{STACK_BOUNDARY}. 1002@end defmac 1003 1004@defmac INCOMING_STACK_BOUNDARY 1005Define this macro if the incoming stack boundary may be different 1006from @code{PREFERRED_STACK_BOUNDARY}. This macro must evaluate 1007to a value equal to or larger than @code{STACK_BOUNDARY}. 1008@end defmac 1009 1010@defmac FUNCTION_BOUNDARY 1011Alignment required for a function entry point, in bits. 1012@end defmac 1013 1014@defmac BIGGEST_ALIGNMENT 1015Biggest alignment that any data type can require on this machine, in 1016bits. Note that this is not the biggest alignment that is supported, 1017just the biggest alignment that, when violated, may cause a fault. 1018@end defmac 1019 1020@deftypevr {Target Hook} HOST_WIDE_INT TARGET_ABSOLUTE_BIGGEST_ALIGNMENT 1021If defined, this target hook specifies the absolute biggest alignment 1022that a type or variable can have on this machine, otherwise, 1023@code{BIGGEST_ALIGNMENT} is used. 1024@end deftypevr 1025 1026@defmac MALLOC_ABI_ALIGNMENT 1027Alignment, in bits, a C conformant malloc implementation has to 1028provide. If not defined, the default value is @code{BITS_PER_WORD}. 1029@end defmac 1030 1031@defmac ATTRIBUTE_ALIGNED_VALUE 1032Alignment used by the @code{__attribute__ ((aligned))} construct. If 1033not defined, the default value is @code{BIGGEST_ALIGNMENT}. 1034@end defmac 1035 1036@defmac MINIMUM_ATOMIC_ALIGNMENT 1037If defined, the smallest alignment, in bits, that can be given to an 1038object that can be referenced in one operation, without disturbing any 1039nearby object. Normally, this is @code{BITS_PER_UNIT}, but may be larger 1040on machines that don't have byte or half-word store operations. 1041@end defmac 1042 1043@defmac BIGGEST_FIELD_ALIGNMENT 1044Biggest alignment that any structure or union field can require on this 1045machine, in bits. If defined, this overrides @code{BIGGEST_ALIGNMENT} for 1046structure and union fields only, unless the field alignment has been set 1047by the @code{__attribute__ ((aligned (@var{n})))} construct. 1048@end defmac 1049 1050@defmac ADJUST_FIELD_ALIGN (@var{field}, @var{type}, @var{computed}) 1051An expression for the alignment of a structure field @var{field} of 1052type @var{type} if the alignment computed in the usual way (including 1053applying of @code{BIGGEST_ALIGNMENT} and @code{BIGGEST_FIELD_ALIGNMENT} to the 1054alignment) is @var{computed}. It overrides alignment only if the 1055field alignment has not been set by the 1056@code{__attribute__ ((aligned (@var{n})))} construct. Note that @var{field} 1057may be @code{NULL_TREE} in case we just query for the minimum alignment 1058of a field of type @var{type} in structure context. 1059@end defmac 1060 1061@defmac MAX_STACK_ALIGNMENT 1062Biggest stack alignment guaranteed by the backend. Use this macro 1063to specify the maximum alignment of a variable on stack. 1064 1065If not defined, the default value is @code{STACK_BOUNDARY}. 1066 1067@c FIXME: The default should be @code{PREFERRED_STACK_BOUNDARY}. 1068@c But the fix for PR 32893 indicates that we can only guarantee 1069@c maximum stack alignment on stack up to @code{STACK_BOUNDARY}, not 1070@c @code{PREFERRED_STACK_BOUNDARY}, if stack alignment isn't supported. 1071@end defmac 1072 1073@defmac MAX_OFILE_ALIGNMENT 1074Biggest alignment supported by the object file format of this machine. 1075Use this macro to limit the alignment which can be specified using the 1076@code{__attribute__ ((aligned (@var{n})))} construct for functions and 1077objects with static storage duration. The alignment of automatic 1078objects may exceed the object file format maximum up to the maximum 1079supported by GCC. If not defined, the default value is 1080@code{BIGGEST_ALIGNMENT}. 1081 1082On systems that use ELF, the default (in @file{config/elfos.h}) is 1083the largest supported 32-bit ELF section alignment representable on 1084a 32-bit host e.g.@: @samp{(((uint64_t) 1 << 28) * 8)}. 1085On 32-bit ELF the largest supported section alignment in bits is 1086@samp{(0x80000000 * 8)}, but this is not representable on 32-bit hosts. 1087@end defmac 1088 1089@deftypefn {Target Hook} HOST_WIDE_INT TARGET_STATIC_RTX_ALIGNMENT (machine_mode @var{mode}) 1090This hook returns the preferred alignment in bits for a 1091statically-allocated rtx, such as a constant pool entry. @var{mode} 1092is the mode of the rtx. The default implementation returns 1093@samp{GET_MODE_ALIGNMENT (@var{mode})}. 1094@end deftypefn 1095 1096@defmac DATA_ALIGNMENT (@var{type}, @var{basic-align}) 1097If defined, a C expression to compute the alignment for a variable in 1098the static store. @var{type} is the data type, and @var{basic-align} is 1099the alignment that the object would ordinarily have. The value of this 1100macro is used instead of that alignment to align the object. 1101 1102If this macro is not defined, then @var{basic-align} is used. 1103 1104@findex strcpy 1105One use of this macro is to increase alignment of medium-size data to 1106make it all fit in fewer cache lines. Another is to cause character 1107arrays to be word-aligned so that @code{strcpy} calls that copy 1108constants to character arrays can be done inline. 1109@end defmac 1110 1111@defmac DATA_ABI_ALIGNMENT (@var{type}, @var{basic-align}) 1112Similar to @code{DATA_ALIGNMENT}, but for the cases where the ABI mandates 1113some alignment increase, instead of optimization only purposes. E.g.@ 1114AMD x86-64 psABI says that variables with array type larger than 15 bytes 1115must be aligned to 16 byte boundaries. 1116 1117If this macro is not defined, then @var{basic-align} is used. 1118@end defmac 1119 1120@deftypefn {Target Hook} HOST_WIDE_INT TARGET_CONSTANT_ALIGNMENT (const_tree @var{constant}, HOST_WIDE_INT @var{basic_align}) 1121This hook returns the alignment in bits of a constant that is being 1122placed in memory. @var{constant} is the constant and @var{basic_align} 1123is the alignment that the object would ordinarily have. 1124 1125The default definition just returns @var{basic_align}. 1126 1127The typical use of this hook is to increase alignment for string 1128constants to be word aligned so that @code{strcpy} calls that copy 1129constants can be done inline. The function 1130@code{constant_alignment_word_strings} provides such a definition. 1131@end deftypefn 1132 1133@defmac LOCAL_ALIGNMENT (@var{type}, @var{basic-align}) 1134If defined, a C expression to compute the alignment for a variable in 1135the local store. @var{type} is the data type, and @var{basic-align} is 1136the alignment that the object would ordinarily have. The value of this 1137macro is used instead of that alignment to align the object. 1138 1139If this macro is not defined, then @var{basic-align} is used. 1140 1141One use of this macro is to increase alignment of medium-size data to 1142make it all fit in fewer cache lines. 1143 1144If the value of this macro has a type, it should be an unsigned type. 1145@end defmac 1146 1147@deftypefn {Target Hook} HOST_WIDE_INT TARGET_VECTOR_ALIGNMENT (const_tree @var{type}) 1148This hook can be used to define the alignment for a vector of type 1149@var{type}, in order to comply with a platform ABI. The default is to 1150require natural alignment for vector types. The alignment returned by 1151this hook must be a power-of-two multiple of the default alignment of 1152the vector element type. 1153@end deftypefn 1154 1155@defmac STACK_SLOT_ALIGNMENT (@var{type}, @var{mode}, @var{basic-align}) 1156If defined, a C expression to compute the alignment for stack slot. 1157@var{type} is the data type, @var{mode} is the widest mode available, 1158and @var{basic-align} is the alignment that the slot would ordinarily 1159have. The value of this macro is used instead of that alignment to 1160align the slot. 1161 1162If this macro is not defined, then @var{basic-align} is used when 1163@var{type} is @code{NULL}. Otherwise, @code{LOCAL_ALIGNMENT} will 1164be used. 1165 1166This macro is to set alignment of stack slot to the maximum alignment 1167of all possible modes which the slot may have. 1168 1169If the value of this macro has a type, it should be an unsigned type. 1170@end defmac 1171 1172@defmac LOCAL_DECL_ALIGNMENT (@var{decl}) 1173If defined, a C expression to compute the alignment for a local 1174variable @var{decl}. 1175 1176If this macro is not defined, then 1177@code{LOCAL_ALIGNMENT (TREE_TYPE (@var{decl}), DECL_ALIGN (@var{decl}))} 1178is used. 1179 1180One use of this macro is to increase alignment of medium-size data to 1181make it all fit in fewer cache lines. 1182 1183If the value of this macro has a type, it should be an unsigned type. 1184@end defmac 1185 1186@defmac MINIMUM_ALIGNMENT (@var{exp}, @var{mode}, @var{align}) 1187If defined, a C expression to compute the minimum required alignment 1188for dynamic stack realignment purposes for @var{exp} (a type or decl), 1189@var{mode}, assuming normal alignment @var{align}. 1190 1191If this macro is not defined, then @var{align} will be used. 1192@end defmac 1193 1194@defmac EMPTY_FIELD_BOUNDARY 1195Alignment in bits to be given to a structure bit-field that follows an 1196empty field such as @code{int : 0;}. 1197 1198If @code{PCC_BITFIELD_TYPE_MATTERS} is true, it overrides this macro. 1199@end defmac 1200 1201@defmac STRUCTURE_SIZE_BOUNDARY 1202Number of bits which any structure or union's size must be a multiple of. 1203Each structure or union's size is rounded up to a multiple of this. 1204 1205If you do not define this macro, the default is the same as 1206@code{BITS_PER_UNIT}. 1207@end defmac 1208 1209@defmac STRICT_ALIGNMENT 1210Define this macro to be the value 1 if instructions will fail to work 1211if given data not on the nominal alignment. If instructions will merely 1212go slower in that case, define this macro as 0. 1213@end defmac 1214 1215@defmac PCC_BITFIELD_TYPE_MATTERS 1216Define this if you wish to imitate the way many other C compilers handle 1217alignment of bit-fields and the structures that contain them. 1218 1219The behavior is that the type written for a named bit-field (@code{int}, 1220@code{short}, or other integer type) imposes an alignment for the entire 1221structure, as if the structure really did contain an ordinary field of 1222that type. In addition, the bit-field is placed within the structure so 1223that it would fit within such a field, not crossing a boundary for it. 1224 1225Thus, on most machines, a named bit-field whose type is written as 1226@code{int} would not cross a four-byte boundary, and would force 1227four-byte alignment for the whole structure. (The alignment used may 1228not be four bytes; it is controlled by the other alignment parameters.) 1229 1230An unnamed bit-field will not affect the alignment of the containing 1231structure. 1232 1233If the macro is defined, its definition should be a C expression; 1234a nonzero value for the expression enables this behavior. 1235 1236Note that if this macro is not defined, or its value is zero, some 1237bit-fields may cross more than one alignment boundary. The compiler can 1238support such references if there are @samp{insv}, @samp{extv}, and 1239@samp{extzv} insns that can directly reference memory. 1240 1241The other known way of making bit-fields work is to define 1242@code{STRUCTURE_SIZE_BOUNDARY} as large as @code{BIGGEST_ALIGNMENT}. 1243Then every structure can be accessed with fullwords. 1244 1245Unless the machine has bit-field instructions or you define 1246@code{STRUCTURE_SIZE_BOUNDARY} that way, you must define 1247@code{PCC_BITFIELD_TYPE_MATTERS} to have a nonzero value. 1248 1249If your aim is to make GCC use the same conventions for laying out 1250bit-fields as are used by another compiler, here is how to investigate 1251what the other compiler does. Compile and run this program: 1252 1253@smallexample 1254struct foo1 1255@{ 1256 char x; 1257 char :0; 1258 char y; 1259@}; 1260 1261struct foo2 1262@{ 1263 char x; 1264 int :0; 1265 char y; 1266@}; 1267 1268main () 1269@{ 1270 printf ("Size of foo1 is %d\n", 1271 sizeof (struct foo1)); 1272 printf ("Size of foo2 is %d\n", 1273 sizeof (struct foo2)); 1274 exit (0); 1275@} 1276@end smallexample 1277 1278If this prints 2 and 5, then the compiler's behavior is what you would 1279get from @code{PCC_BITFIELD_TYPE_MATTERS}. 1280@end defmac 1281 1282@defmac BITFIELD_NBYTES_LIMITED 1283Like @code{PCC_BITFIELD_TYPE_MATTERS} except that its effect is limited 1284to aligning a bit-field within the structure. 1285@end defmac 1286 1287@deftypefn {Target Hook} bool TARGET_ALIGN_ANON_BITFIELD (void) 1288When @code{PCC_BITFIELD_TYPE_MATTERS} is true this hook will determine 1289whether unnamed bitfields affect the alignment of the containing 1290structure. The hook should return true if the structure should inherit 1291the alignment requirements of an unnamed bitfield's type. 1292@end deftypefn 1293 1294@deftypefn {Target Hook} bool TARGET_NARROW_VOLATILE_BITFIELD (void) 1295This target hook should return @code{true} if accesses to volatile bitfields 1296should use the narrowest mode possible. It should return @code{false} if 1297these accesses should use the bitfield container type. 1298 1299The default is @code{false}. 1300@end deftypefn 1301 1302@deftypefn {Target Hook} bool TARGET_MEMBER_TYPE_FORCES_BLK (const_tree @var{field}, machine_mode @var{mode}) 1303Return true if a structure, union or array containing @var{field} should 1304be accessed using @code{BLKMODE}. 1305 1306If @var{field} is the only field in the structure, @var{mode} is its 1307mode, otherwise @var{mode} is VOIDmode. @var{mode} is provided in the 1308case where structures of one field would require the structure's mode to 1309retain the field's mode. 1310 1311Normally, this is not needed. 1312@end deftypefn 1313 1314@defmac ROUND_TYPE_ALIGN (@var{type}, @var{computed}, @var{specified}) 1315Define this macro as an expression for the alignment of a type (given 1316by @var{type} as a tree node) if the alignment computed in the usual 1317way is @var{computed} and the alignment explicitly specified was 1318@var{specified}. 1319 1320The default is to use @var{specified} if it is larger; otherwise, use 1321the smaller of @var{computed} and @code{BIGGEST_ALIGNMENT} 1322@end defmac 1323 1324@defmac MAX_FIXED_MODE_SIZE 1325An integer expression for the size in bits of the largest integer 1326machine mode that should actually be used. All integer machine modes of 1327this size or smaller can be used for structures and unions with the 1328appropriate sizes. If this macro is undefined, @code{GET_MODE_BITSIZE 1329(DImode)} is assumed. 1330@end defmac 1331 1332@defmac STACK_SAVEAREA_MODE (@var{save_level}) 1333If defined, an expression of type @code{machine_mode} that 1334specifies the mode of the save area operand of a 1335@code{save_stack_@var{level}} named pattern (@pxref{Standard Names}). 1336@var{save_level} is one of @code{SAVE_BLOCK}, @code{SAVE_FUNCTION}, or 1337@code{SAVE_NONLOCAL} and selects which of the three named patterns is 1338having its mode specified. 1339 1340You need not define this macro if it always returns @code{Pmode}. You 1341would most commonly define this macro if the 1342@code{save_stack_@var{level}} patterns need to support both a 32- and a 134364-bit mode. 1344@end defmac 1345 1346@defmac STACK_SIZE_MODE 1347If defined, an expression of type @code{machine_mode} that 1348specifies the mode of the size increment operand of an 1349@code{allocate_stack} named pattern (@pxref{Standard Names}). 1350 1351You need not define this macro if it always returns @code{word_mode}. 1352You would most commonly define this macro if the @code{allocate_stack} 1353pattern needs to support both a 32- and a 64-bit mode. 1354@end defmac 1355 1356@deftypefn {Target Hook} scalar_int_mode TARGET_LIBGCC_CMP_RETURN_MODE (void) 1357This target hook should return the mode to be used for the return value 1358of compare instructions expanded to libgcc calls. If not defined 1359@code{word_mode} is returned which is the right choice for a majority of 1360targets. 1361@end deftypefn 1362 1363@deftypefn {Target Hook} scalar_int_mode TARGET_LIBGCC_SHIFT_COUNT_MODE (void) 1364This target hook should return the mode to be used for the shift count operand 1365of shift instructions expanded to libgcc calls. If not defined 1366@code{word_mode} is returned which is the right choice for a majority of 1367targets. 1368@end deftypefn 1369 1370@deftypefn {Target Hook} scalar_int_mode TARGET_UNWIND_WORD_MODE (void) 1371Return machine mode to be used for @code{_Unwind_Word} type. 1372The default is to use @code{word_mode}. 1373@end deftypefn 1374 1375@deftypefn {Target Hook} bool TARGET_MS_BITFIELD_LAYOUT_P (const_tree @var{record_type}) 1376This target hook returns @code{true} if bit-fields in the given 1377@var{record_type} are to be laid out following the rules of Microsoft 1378Visual C/C++, namely: (i) a bit-field won't share the same storage 1379unit with the previous bit-field if their underlying types have 1380different sizes, and the bit-field will be aligned to the highest 1381alignment of the underlying types of itself and of the previous 1382bit-field; (ii) a zero-sized bit-field will affect the alignment of 1383the whole enclosing structure, even if it is unnamed; except that 1384(iii) a zero-sized bit-field will be disregarded unless it follows 1385another bit-field of nonzero size. If this hook returns @code{true}, 1386other macros that control bit-field layout are ignored. 1387 1388When a bit-field is inserted into a packed record, the whole size 1389of the underlying type is used by one or more same-size adjacent 1390bit-fields (that is, if its long:3, 32 bits is used in the record, 1391and any additional adjacent long bit-fields are packed into the same 1392chunk of 32 bits. However, if the size changes, a new field of that 1393size is allocated). In an unpacked record, this is the same as using 1394alignment, but not equivalent when packing. 1395 1396If both MS bit-fields and @samp{__attribute__((packed))} are used, 1397the latter will take precedence. If @samp{__attribute__((packed))} is 1398used on a single field when MS bit-fields are in use, it will take 1399precedence for that field, but the alignment of the rest of the structure 1400may affect its placement. 1401@end deftypefn 1402 1403@deftypefn {Target Hook} bool TARGET_DECIMAL_FLOAT_SUPPORTED_P (void) 1404Returns true if the target supports decimal floating point. 1405@end deftypefn 1406 1407@deftypefn {Target Hook} bool TARGET_FIXED_POINT_SUPPORTED_P (void) 1408Returns true if the target supports fixed-point arithmetic. 1409@end deftypefn 1410 1411@deftypefn {Target Hook} void TARGET_EXPAND_TO_RTL_HOOK (void) 1412This hook is called just before expansion into rtl, allowing the target 1413to perform additional initializations or analysis before the expansion. 1414For example, the rs6000 port uses it to allocate a scratch stack slot 1415for use in copying SDmode values between memory and floating point 1416registers whenever the function being expanded has any SDmode 1417usage. 1418@end deftypefn 1419 1420@deftypefn {Target Hook} void TARGET_INSTANTIATE_DECLS (void) 1421This hook allows the backend to perform additional instantiations on rtl 1422that are not actually in any insns yet, but will be later. 1423@end deftypefn 1424 1425@deftypefn {Target Hook} {const char *} TARGET_MANGLE_TYPE (const_tree @var{type}) 1426If your target defines any fundamental types, or any types your target 1427uses should be mangled differently from the default, define this hook 1428to return the appropriate encoding for these types as part of a C++ 1429mangled name. The @var{type} argument is the tree structure representing 1430the type to be mangled. The hook may be applied to trees which are 1431not target-specific fundamental types; it should return @code{NULL} 1432for all such types, as well as arguments it does not recognize. If the 1433return value is not @code{NULL}, it must point to a statically-allocated 1434string constant. 1435 1436Target-specific fundamental types might be new fundamental types or 1437qualified versions of ordinary fundamental types. Encode new 1438fundamental types as @samp{@w{u @var{n} @var{name}}}, where @var{name} 1439is the name used for the type in source code, and @var{n} is the 1440length of @var{name} in decimal. Encode qualified versions of 1441ordinary types as @samp{@w{U @var{n} @var{name} @var{code}}}, where 1442@var{name} is the name used for the type qualifier in source code, 1443@var{n} is the length of @var{name} as above, and @var{code} is the 1444code used to represent the unqualified version of this type. (See 1445@code{write_builtin_type} in @file{cp/mangle.c} for the list of 1446codes.) In both cases the spaces are for clarity; do not include any 1447spaces in your string. 1448 1449This hook is applied to types prior to typedef resolution. If the mangled 1450name for a particular type depends only on that type's main variant, you 1451can perform typedef resolution yourself using @code{TYPE_MAIN_VARIANT} 1452before mangling. 1453 1454The default version of this hook always returns @code{NULL}, which is 1455appropriate for a target that does not define any new fundamental 1456types. 1457@end deftypefn 1458 1459@node Type Layout 1460@section Layout of Source Language Data Types 1461 1462These macros define the sizes and other characteristics of the standard 1463basic data types used in programs being compiled. Unlike the macros in 1464the previous section, these apply to specific features of C and related 1465languages, rather than to fundamental aspects of storage layout. 1466 1467@defmac INT_TYPE_SIZE 1468A C expression for the size in bits of the type @code{int} on the 1469target machine. If you don't define this, the default is one word. 1470@end defmac 1471 1472@defmac SHORT_TYPE_SIZE 1473A C expression for the size in bits of the type @code{short} on the 1474target machine. If you don't define this, the default is half a word. 1475(If this would be less than one storage unit, it is rounded up to one 1476unit.) 1477@end defmac 1478 1479@defmac LONG_TYPE_SIZE 1480A C expression for the size in bits of the type @code{long} on the 1481target machine. If you don't define this, the default is one word. 1482@end defmac 1483 1484@defmac ADA_LONG_TYPE_SIZE 1485On some machines, the size used for the Ada equivalent of the type 1486@code{long} by a native Ada compiler differs from that used by C@. In 1487that situation, define this macro to be a C expression to be used for 1488the size of that type. If you don't define this, the default is the 1489value of @code{LONG_TYPE_SIZE}. 1490@end defmac 1491 1492@defmac LONG_LONG_TYPE_SIZE 1493A C expression for the size in bits of the type @code{long long} on the 1494target machine. If you don't define this, the default is two 1495words. If you want to support GNU Ada on your machine, the value of this 1496macro must be at least 64. 1497@end defmac 1498 1499@defmac CHAR_TYPE_SIZE 1500A C expression for the size in bits of the type @code{char} on the 1501target machine. If you don't define this, the default is 1502@code{BITS_PER_UNIT}. 1503@end defmac 1504 1505@defmac BOOL_TYPE_SIZE 1506A C expression for the size in bits of the C++ type @code{bool} and 1507C99 type @code{_Bool} on the target machine. If you don't define 1508this, and you probably shouldn't, the default is @code{CHAR_TYPE_SIZE}. 1509@end defmac 1510 1511@defmac FLOAT_TYPE_SIZE 1512A C expression for the size in bits of the type @code{float} on the 1513target machine. If you don't define this, the default is one word. 1514@end defmac 1515 1516@defmac DOUBLE_TYPE_SIZE 1517A C expression for the size in bits of the type @code{double} on the 1518target machine. If you don't define this, the default is two 1519words. 1520@end defmac 1521 1522@defmac LONG_DOUBLE_TYPE_SIZE 1523A C expression for the size in bits of the type @code{long double} on 1524the target machine. If you don't define this, the default is two 1525words. 1526@end defmac 1527 1528@defmac SHORT_FRACT_TYPE_SIZE 1529A C expression for the size in bits of the type @code{short _Fract} on 1530the target machine. If you don't define this, the default is 1531@code{BITS_PER_UNIT}. 1532@end defmac 1533 1534@defmac FRACT_TYPE_SIZE 1535A C expression for the size in bits of the type @code{_Fract} on 1536the target machine. If you don't define this, the default is 1537@code{BITS_PER_UNIT * 2}. 1538@end defmac 1539 1540@defmac LONG_FRACT_TYPE_SIZE 1541A C expression for the size in bits of the type @code{long _Fract} on 1542the target machine. If you don't define this, the default is 1543@code{BITS_PER_UNIT * 4}. 1544@end defmac 1545 1546@defmac LONG_LONG_FRACT_TYPE_SIZE 1547A C expression for the size in bits of the type @code{long long _Fract} on 1548the target machine. If you don't define this, the default is 1549@code{BITS_PER_UNIT * 8}. 1550@end defmac 1551 1552@defmac SHORT_ACCUM_TYPE_SIZE 1553A C expression for the size in bits of the type @code{short _Accum} on 1554the target machine. If you don't define this, the default is 1555@code{BITS_PER_UNIT * 2}. 1556@end defmac 1557 1558@defmac ACCUM_TYPE_SIZE 1559A C expression for the size in bits of the type @code{_Accum} on 1560the target machine. If you don't define this, the default is 1561@code{BITS_PER_UNIT * 4}. 1562@end defmac 1563 1564@defmac LONG_ACCUM_TYPE_SIZE 1565A C expression for the size in bits of the type @code{long _Accum} on 1566the target machine. If you don't define this, the default is 1567@code{BITS_PER_UNIT * 8}. 1568@end defmac 1569 1570@defmac LONG_LONG_ACCUM_TYPE_SIZE 1571A C expression for the size in bits of the type @code{long long _Accum} on 1572the target machine. If you don't define this, the default is 1573@code{BITS_PER_UNIT * 16}. 1574@end defmac 1575 1576@defmac LIBGCC2_GNU_PREFIX 1577This macro corresponds to the @code{TARGET_LIBFUNC_GNU_PREFIX} target 1578hook and should be defined if that hook is overriden to be true. It 1579causes function names in libgcc to be changed to use a @code{__gnu_} 1580prefix for their name rather than the default @code{__}. A port which 1581uses this macro should also arrange to use @file{t-gnu-prefix} in 1582the libgcc @file{config.host}. 1583@end defmac 1584 1585@defmac WIDEST_HARDWARE_FP_SIZE 1586A C expression for the size in bits of the widest floating-point format 1587supported by the hardware. If you define this macro, you must specify a 1588value less than or equal to the value of @code{LONG_DOUBLE_TYPE_SIZE}. 1589If you do not define this macro, the value of @code{LONG_DOUBLE_TYPE_SIZE} 1590is the default. 1591@end defmac 1592 1593@defmac DEFAULT_SIGNED_CHAR 1594An expression whose value is 1 or 0, according to whether the type 1595@code{char} should be signed or unsigned by default. The user can 1596always override this default with the options @option{-fsigned-char} 1597and @option{-funsigned-char}. 1598@end defmac 1599 1600@deftypefn {Target Hook} bool TARGET_DEFAULT_SHORT_ENUMS (void) 1601This target hook should return true if the compiler should give an 1602@code{enum} type only as many bytes as it takes to represent the range 1603of possible values of that type. It should return false if all 1604@code{enum} types should be allocated like @code{int}. 1605 1606The default is to return false. 1607@end deftypefn 1608 1609@defmac SIZE_TYPE 1610A C expression for a string describing the name of the data type to use 1611for size values. The typedef name @code{size_t} is defined using the 1612contents of the string. 1613 1614The string can contain more than one keyword. If so, separate them with 1615spaces, and write first any length keyword, then @code{unsigned} if 1616appropriate, and finally @code{int}. The string must exactly match one 1617of the data type names defined in the function 1618@code{c_common_nodes_and_builtins} in the file @file{c-family/c-common.c}. 1619You may not omit @code{int} or change the order---that would cause the 1620compiler to crash on startup. 1621 1622If you don't define this macro, the default is @code{"long unsigned 1623int"}. 1624@end defmac 1625 1626@defmac SIZETYPE 1627GCC defines internal types (@code{sizetype}, @code{ssizetype}, 1628@code{bitsizetype} and @code{sbitsizetype}) for expressions 1629dealing with size. This macro is a C expression for a string describing 1630the name of the data type from which the precision of @code{sizetype} 1631is extracted. 1632 1633The string has the same restrictions as @code{SIZE_TYPE} string. 1634 1635If you don't define this macro, the default is @code{SIZE_TYPE}. 1636@end defmac 1637 1638@defmac PTRDIFF_TYPE 1639A C expression for a string describing the name of the data type to use 1640for the result of subtracting two pointers. The typedef name 1641@code{ptrdiff_t} is defined using the contents of the string. See 1642@code{SIZE_TYPE} above for more information. 1643 1644If you don't define this macro, the default is @code{"long int"}. 1645@end defmac 1646 1647@defmac WCHAR_TYPE 1648A C expression for a string describing the name of the data type to use 1649for wide characters. The typedef name @code{wchar_t} is defined using 1650the contents of the string. See @code{SIZE_TYPE} above for more 1651information. 1652 1653If you don't define this macro, the default is @code{"int"}. 1654@end defmac 1655 1656@defmac WCHAR_TYPE_SIZE 1657A C expression for the size in bits of the data type for wide 1658characters. This is used in @code{cpp}, which cannot make use of 1659@code{WCHAR_TYPE}. 1660@end defmac 1661 1662@defmac WINT_TYPE 1663A C expression for a string describing the name of the data type to 1664use for wide characters passed to @code{printf} and returned from 1665@code{getwc}. The typedef name @code{wint_t} is defined using the 1666contents of the string. See @code{SIZE_TYPE} above for more 1667information. 1668 1669If you don't define this macro, the default is @code{"unsigned int"}. 1670@end defmac 1671 1672@defmac INTMAX_TYPE 1673A C expression for a string describing the name of the data type that 1674can represent any value of any standard or extended signed integer type. 1675The typedef name @code{intmax_t} is defined using the contents of the 1676string. See @code{SIZE_TYPE} above for more information. 1677 1678If you don't define this macro, the default is the first of 1679@code{"int"}, @code{"long int"}, or @code{"long long int"} that has as 1680much precision as @code{long long int}. 1681@end defmac 1682 1683@defmac UINTMAX_TYPE 1684A C expression for a string describing the name of the data type that 1685can represent any value of any standard or extended unsigned integer 1686type. The typedef name @code{uintmax_t} is defined using the contents 1687of the string. See @code{SIZE_TYPE} above for more information. 1688 1689If you don't define this macro, the default is the first of 1690@code{"unsigned int"}, @code{"long unsigned int"}, or @code{"long long 1691unsigned int"} that has as much precision as @code{long long unsigned 1692int}. 1693@end defmac 1694 1695@defmac SIG_ATOMIC_TYPE 1696@defmacx INT8_TYPE 1697@defmacx INT16_TYPE 1698@defmacx INT32_TYPE 1699@defmacx INT64_TYPE 1700@defmacx UINT8_TYPE 1701@defmacx UINT16_TYPE 1702@defmacx UINT32_TYPE 1703@defmacx UINT64_TYPE 1704@defmacx INT_LEAST8_TYPE 1705@defmacx INT_LEAST16_TYPE 1706@defmacx INT_LEAST32_TYPE 1707@defmacx INT_LEAST64_TYPE 1708@defmacx UINT_LEAST8_TYPE 1709@defmacx UINT_LEAST16_TYPE 1710@defmacx UINT_LEAST32_TYPE 1711@defmacx UINT_LEAST64_TYPE 1712@defmacx INT_FAST8_TYPE 1713@defmacx INT_FAST16_TYPE 1714@defmacx INT_FAST32_TYPE 1715@defmacx INT_FAST64_TYPE 1716@defmacx UINT_FAST8_TYPE 1717@defmacx UINT_FAST16_TYPE 1718@defmacx UINT_FAST32_TYPE 1719@defmacx UINT_FAST64_TYPE 1720@defmacx INTPTR_TYPE 1721@defmacx UINTPTR_TYPE 1722C expressions for the standard types @code{sig_atomic_t}, 1723@code{int8_t}, @code{int16_t}, @code{int32_t}, @code{int64_t}, 1724@code{uint8_t}, @code{uint16_t}, @code{uint32_t}, @code{uint64_t}, 1725@code{int_least8_t}, @code{int_least16_t}, @code{int_least32_t}, 1726@code{int_least64_t}, @code{uint_least8_t}, @code{uint_least16_t}, 1727@code{uint_least32_t}, @code{uint_least64_t}, @code{int_fast8_t}, 1728@code{int_fast16_t}, @code{int_fast32_t}, @code{int_fast64_t}, 1729@code{uint_fast8_t}, @code{uint_fast16_t}, @code{uint_fast32_t}, 1730@code{uint_fast64_t}, @code{intptr_t}, and @code{uintptr_t}. See 1731@code{SIZE_TYPE} above for more information. 1732 1733If any of these macros evaluates to a null pointer, the corresponding 1734type is not supported; if GCC is configured to provide 1735@code{<stdint.h>} in such a case, the header provided may not conform 1736to C99, depending on the type in question. The defaults for all of 1737these macros are null pointers. 1738@end defmac 1739 1740@defmac TARGET_PTRMEMFUNC_VBIT_LOCATION 1741The C++ compiler represents a pointer-to-member-function with a struct 1742that looks like: 1743 1744@smallexample 1745 struct @{ 1746 union @{ 1747 void (*fn)(); 1748 ptrdiff_t vtable_index; 1749 @}; 1750 ptrdiff_t delta; 1751 @}; 1752@end smallexample 1753 1754@noindent 1755The C++ compiler must use one bit to indicate whether the function that 1756will be called through a pointer-to-member-function is virtual. 1757Normally, we assume that the low-order bit of a function pointer must 1758always be zero. Then, by ensuring that the vtable_index is odd, we can 1759distinguish which variant of the union is in use. But, on some 1760platforms function pointers can be odd, and so this doesn't work. In 1761that case, we use the low-order bit of the @code{delta} field, and shift 1762the remainder of the @code{delta} field to the left. 1763 1764GCC will automatically make the right selection about where to store 1765this bit using the @code{FUNCTION_BOUNDARY} setting for your platform. 1766However, some platforms such as ARM/Thumb have @code{FUNCTION_BOUNDARY} 1767set such that functions always start at even addresses, but the lowest 1768bit of pointers to functions indicate whether the function at that 1769address is in ARM or Thumb mode. If this is the case of your 1770architecture, you should define this macro to 1771@code{ptrmemfunc_vbit_in_delta}. 1772 1773In general, you should not have to define this macro. On architectures 1774in which function addresses are always even, according to 1775@code{FUNCTION_BOUNDARY}, GCC will automatically define this macro to 1776@code{ptrmemfunc_vbit_in_pfn}. 1777@end defmac 1778 1779@defmac TARGET_VTABLE_USES_DESCRIPTORS 1780Normally, the C++ compiler uses function pointers in vtables. This 1781macro allows the target to change to use ``function descriptors'' 1782instead. Function descriptors are found on targets for whom a 1783function pointer is actually a small data structure. Normally the 1784data structure consists of the actual code address plus a data 1785pointer to which the function's data is relative. 1786 1787If vtables are used, the value of this macro should be the number 1788of words that the function descriptor occupies. 1789@end defmac 1790 1791@defmac TARGET_VTABLE_ENTRY_ALIGN 1792By default, the vtable entries are void pointers, the so the alignment 1793is the same as pointer alignment. The value of this macro specifies 1794the alignment of the vtable entry in bits. It should be defined only 1795when special alignment is necessary. */ 1796@end defmac 1797 1798@defmac TARGET_VTABLE_DATA_ENTRY_DISTANCE 1799There are a few non-descriptor entries in the vtable at offsets below 1800zero. If these entries must be padded (say, to preserve the alignment 1801specified by @code{TARGET_VTABLE_ENTRY_ALIGN}), set this to the number 1802of words in each data entry. 1803@end defmac 1804 1805@node Registers 1806@section Register Usage 1807@cindex register usage 1808 1809This section explains how to describe what registers the target machine 1810has, and how (in general) they can be used. 1811 1812The description of which registers a specific instruction can use is 1813done with register classes; see @ref{Register Classes}. For information 1814on using registers to access a stack frame, see @ref{Frame Registers}. 1815For passing values in registers, see @ref{Register Arguments}. 1816For returning values in registers, see @ref{Scalar Return}. 1817 1818@menu 1819* Register Basics:: Number and kinds of registers. 1820* Allocation Order:: Order in which registers are allocated. 1821* Values in Registers:: What kinds of values each reg can hold. 1822* Leaf Functions:: Renumbering registers for leaf functions. 1823* Stack Registers:: Handling a register stack such as 80387. 1824@end menu 1825 1826@node Register Basics 1827@subsection Basic Characteristics of Registers 1828 1829@c prevent bad page break with this line 1830Registers have various characteristics. 1831 1832@defmac FIRST_PSEUDO_REGISTER 1833Number of hardware registers known to the compiler. They receive 1834numbers 0 through @code{FIRST_PSEUDO_REGISTER-1}; thus, the first 1835pseudo register's number really is assigned the number 1836@code{FIRST_PSEUDO_REGISTER}. 1837@end defmac 1838 1839@defmac FIXED_REGISTERS 1840@cindex fixed register 1841An initializer that says which registers are used for fixed purposes 1842all throughout the compiled code and are therefore not available for 1843general allocation. These would include the stack pointer, the frame 1844pointer (except on machines where that can be used as a general 1845register when no frame pointer is needed), the program counter on 1846machines where that is considered one of the addressable registers, 1847and any other numbered register with a standard use. 1848 1849This information is expressed as a sequence of numbers, separated by 1850commas and surrounded by braces. The @var{n}th number is 1 if 1851register @var{n} is fixed, 0 otherwise. 1852 1853The table initialized from this macro, and the table initialized by 1854the following one, may be overridden at run time either automatically, 1855by the actions of the macro @code{CONDITIONAL_REGISTER_USAGE}, or by 1856the user with the command options @option{-ffixed-@var{reg}}, 1857@option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}}. 1858@end defmac 1859 1860@defmac CALL_USED_REGISTERS 1861@cindex call-used register 1862@cindex call-clobbered register 1863@cindex call-saved register 1864Like @code{FIXED_REGISTERS} but has 1 for each register that is 1865clobbered (in general) by function calls as well as for fixed 1866registers. This macro therefore identifies the registers that are not 1867available for general allocation of values that must live across 1868function calls. 1869 1870If a register has 0 in @code{CALL_USED_REGISTERS}, the compiler 1871automatically saves it on function entry and restores it on function 1872exit, if the register is used within the function. 1873 1874Exactly one of @code{CALL_USED_REGISTERS} and @code{CALL_REALLY_USED_REGISTERS} 1875must be defined. Modern ports should define @code{CALL_REALLY_USED_REGISTERS}. 1876@end defmac 1877 1878@defmac CALL_REALLY_USED_REGISTERS 1879@cindex call-used register 1880@cindex call-clobbered register 1881@cindex call-saved register 1882Like @code{CALL_USED_REGISTERS} except this macro doesn't require 1883that the entire set of @code{FIXED_REGISTERS} be included. 1884(@code{CALL_USED_REGISTERS} must be a superset of @code{FIXED_REGISTERS}). 1885 1886Exactly one of @code{CALL_USED_REGISTERS} and @code{CALL_REALLY_USED_REGISTERS} 1887must be defined. Modern ports should define @code{CALL_REALLY_USED_REGISTERS}. 1888@end defmac 1889 1890@cindex call-used register 1891@cindex call-clobbered register 1892@cindex call-saved register 1893@deftypefn {Target Hook} {const predefined_function_abi &} TARGET_FNTYPE_ABI (const_tree @var{type}) 1894Return the ABI used by a function with type @var{type}; see the 1895definition of @code{predefined_function_abi} for details of the ABI 1896descriptor. Targets only need to define this hook if they support 1897interoperability between several ABIs in the same translation unit. 1898@end deftypefn 1899 1900@deftypefn {Target Hook} {const predefined_function_abi &} TARGET_INSN_CALLEE_ABI (const rtx_insn *@var{insn}) 1901This hook returns a description of the ABI used by the target of 1902call instruction @var{insn}; see the definition of 1903@code{predefined_function_abi} for details of the ABI descriptor. 1904Only the global function @code{insn_callee_abi} should call this hook 1905directly. 1906 1907Targets only need to define this hook if they support 1908interoperability between several ABIs in the same translation unit. 1909@end deftypefn 1910 1911@cindex call-used register 1912@cindex call-clobbered register 1913@cindex call-saved register 1914@deftypefn {Target Hook} bool TARGET_HARD_REGNO_CALL_PART_CLOBBERED (unsigned int @var{abi_id}, unsigned int @var{regno}, machine_mode @var{mode}) 1915ABIs usually specify that calls must preserve the full contents 1916of a particular register, or that calls can alter any part of a 1917particular register. This information is captured by the target macro 1918@code{CALL_REALLY_USED_REGISTERS}. However, some ABIs specify that calls 1919must preserve certain bits of a particular register but can alter others. 1920This hook should return true if this applies to at least one of the 1921registers in @samp{(reg:@var{mode} @var{regno})}, and if as a result the 1922call would alter part of the @var{mode} value. For example, if a call 1923preserves the low 32 bits of a 64-bit hard register @var{regno} but can 1924clobber the upper 32 bits, this hook should return true for a 64-bit mode 1925but false for a 32-bit mode. 1926 1927The value of @var{abi_id} comes from the @code{predefined_function_abi} 1928structure that describes the ABI of the call; see the definition of the 1929structure for more details. If (as is usual) the target uses the same ABI 1930for all functions in a translation unit, @var{abi_id} is always 0. 1931 1932The default implementation returns false, which is correct 1933for targets that don't have partly call-clobbered registers. 1934@end deftypefn 1935 1936@deftypefn {Target Hook} {const char *} TARGET_GET_MULTILIB_ABI_NAME (void) 1937This hook returns name of multilib ABI name. 1938@end deftypefn 1939 1940@findex fixed_regs 1941@findex call_used_regs 1942@findex global_regs 1943@findex reg_names 1944@findex reg_class_contents 1945@deftypefn {Target Hook} void TARGET_CONDITIONAL_REGISTER_USAGE (void) 1946This hook may conditionally modify five variables 1947@code{fixed_regs}, @code{call_used_regs}, @code{global_regs}, 1948@code{reg_names}, and @code{reg_class_contents}, to take into account 1949any dependence of these register sets on target flags. The first three 1950of these are of type @code{char []} (interpreted as boolean vectors). 1951@code{global_regs} is a @code{const char *[]}, and 1952@code{reg_class_contents} is a @code{HARD_REG_SET}. Before the macro is 1953called, @code{fixed_regs}, @code{call_used_regs}, 1954@code{reg_class_contents}, and @code{reg_names} have been initialized 1955from @code{FIXED_REGISTERS}, @code{CALL_USED_REGISTERS}, 1956@code{REG_CLASS_CONTENTS}, and @code{REGISTER_NAMES}, respectively. 1957@code{global_regs} has been cleared, and any @option{-ffixed-@var{reg}}, 1958@option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}} 1959command options have been applied. 1960 1961@cindex disabling certain registers 1962@cindex controlling register usage 1963If the usage of an entire class of registers depends on the target 1964flags, you may indicate this to GCC by using this macro to modify 1965@code{fixed_regs} and @code{call_used_regs} to 1 for each of the 1966registers in the classes which should not be used by GCC@. Also make 1967@code{define_register_constraint}s return @code{NO_REGS} for constraints 1968that shouldn't be used. 1969 1970(However, if this class is not included in @code{GENERAL_REGS} and all 1971of the insn patterns whose constraints permit this class are 1972controlled by target switches, then GCC will automatically avoid using 1973these registers when the target switches are opposed to them.) 1974@end deftypefn 1975 1976@defmac INCOMING_REGNO (@var{out}) 1977Define this macro if the target machine has register windows. This C 1978expression returns the register number as seen by the called function 1979corresponding to the register number @var{out} as seen by the calling 1980function. Return @var{out} if register number @var{out} is not an 1981outbound register. 1982@end defmac 1983 1984@defmac OUTGOING_REGNO (@var{in}) 1985Define this macro if the target machine has register windows. This C 1986expression returns the register number as seen by the calling function 1987corresponding to the register number @var{in} as seen by the called 1988function. Return @var{in} if register number @var{in} is not an inbound 1989register. 1990@end defmac 1991 1992@defmac LOCAL_REGNO (@var{regno}) 1993Define this macro if the target machine has register windows. This C 1994expression returns true if the register is call-saved but is in the 1995register window. Unlike most call-saved registers, such registers 1996need not be explicitly restored on function exit or during non-local 1997gotos. 1998@end defmac 1999 2000@defmac PC_REGNUM 2001If the program counter has a register number, define this as that 2002register number. Otherwise, do not define it. 2003@end defmac 2004 2005@node Allocation Order 2006@subsection Order of Allocation of Registers 2007@cindex order of register allocation 2008@cindex register allocation order 2009 2010@c prevent bad page break with this line 2011Registers are allocated in order. 2012 2013@defmac REG_ALLOC_ORDER 2014If defined, an initializer for a vector of integers, containing the 2015numbers of hard registers in the order in which GCC should prefer 2016to use them (from most preferred to least). 2017 2018If this macro is not defined, registers are used lowest numbered first 2019(all else being equal). 2020 2021One use of this macro is on machines where the highest numbered 2022registers must always be saved and the save-multiple-registers 2023instruction supports only sequences of consecutive registers. On such 2024machines, define @code{REG_ALLOC_ORDER} to be an initializer that lists 2025the highest numbered allocable register first. 2026@end defmac 2027 2028@defmac ADJUST_REG_ALLOC_ORDER 2029A C statement (sans semicolon) to choose the order in which to allocate 2030hard registers for pseudo-registers local to a basic block. 2031 2032Store the desired register order in the array @code{reg_alloc_order}. 2033Element 0 should be the register to allocate first; element 1, the next 2034register; and so on. 2035 2036The macro body should not assume anything about the contents of 2037@code{reg_alloc_order} before execution of the macro. 2038 2039On most machines, it is not necessary to define this macro. 2040@end defmac 2041 2042@defmac HONOR_REG_ALLOC_ORDER 2043Normally, IRA tries to estimate the costs for saving a register in the 2044prologue and restoring it in the epilogue. This discourages it from 2045using call-saved registers. If a machine wants to ensure that IRA 2046allocates registers in the order given by REG_ALLOC_ORDER even if some 2047call-saved registers appear earlier than call-used ones, then define this 2048macro as a C expression to nonzero. Default is 0. 2049@end defmac 2050 2051@defmac IRA_HARD_REGNO_ADD_COST_MULTIPLIER (@var{regno}) 2052In some case register allocation order is not enough for the 2053Integrated Register Allocator (@acronym{IRA}) to generate a good code. 2054If this macro is defined, it should return a floating point value 2055based on @var{regno}. The cost of using @var{regno} for a pseudo will 2056be increased by approximately the pseudo's usage frequency times the 2057value returned by this macro. Not defining this macro is equivalent 2058to having it always return @code{0.0}. 2059 2060On most machines, it is not necessary to define this macro. 2061@end defmac 2062 2063@node Values in Registers 2064@subsection How Values Fit in Registers 2065 2066This section discusses the macros that describe which kinds of values 2067(specifically, which machine modes) each register can hold, and how many 2068consecutive registers are needed for a given mode. 2069 2070@deftypefn {Target Hook} {unsigned int} TARGET_HARD_REGNO_NREGS (unsigned int @var{regno}, machine_mode @var{mode}) 2071This hook returns the number of consecutive hard registers, starting 2072at register number @var{regno}, required to hold a value of mode 2073@var{mode}. This hook must never return zero, even if a register 2074cannot hold the requested mode - indicate that with 2075@code{TARGET_HARD_REGNO_MODE_OK} and/or 2076@code{TARGET_CAN_CHANGE_MODE_CLASS} instead. 2077 2078The default definition returns the number of words in @var{mode}. 2079@end deftypefn 2080 2081@defmac HARD_REGNO_NREGS_HAS_PADDING (@var{regno}, @var{mode}) 2082A C expression that is nonzero if a value of mode @var{mode}, stored 2083in memory, ends with padding that causes it to take up more space than 2084in registers starting at register number @var{regno} (as determined by 2085multiplying GCC's notion of the size of the register when containing 2086this mode by the number of registers returned by 2087@code{TARGET_HARD_REGNO_NREGS}). By default this is zero. 2088 2089For example, if a floating-point value is stored in three 32-bit 2090registers but takes up 128 bits in memory, then this would be 2091nonzero. 2092 2093This macros only needs to be defined if there are cases where 2094@code{subreg_get_info} 2095would otherwise wrongly determine that a @code{subreg} can be 2096represented by an offset to the register number, when in fact such a 2097@code{subreg} would contain some of the padding not stored in 2098registers and so not be representable. 2099@end defmac 2100 2101@defmac HARD_REGNO_NREGS_WITH_PADDING (@var{regno}, @var{mode}) 2102For values of @var{regno} and @var{mode} for which 2103@code{HARD_REGNO_NREGS_HAS_PADDING} returns nonzero, a C expression 2104returning the greater number of registers required to hold the value 2105including any padding. In the example above, the value would be four. 2106@end defmac 2107 2108@defmac REGMODE_NATURAL_SIZE (@var{mode}) 2109Define this macro if the natural size of registers that hold values 2110of mode @var{mode} is not the word size. It is a C expression that 2111should give the natural size in bytes for the specified mode. It is 2112used by the register allocator to try to optimize its results. This 2113happens for example on SPARC 64-bit where the natural size of 2114floating-point registers is still 32-bit. 2115@end defmac 2116 2117@deftypefn {Target Hook} bool TARGET_HARD_REGNO_MODE_OK (unsigned int @var{regno}, machine_mode @var{mode}) 2118This hook returns true if it is permissible to store a value 2119of mode @var{mode} in hard register number @var{regno} (or in several 2120registers starting with that one). The default definition returns true 2121unconditionally. 2122 2123You need not include code to check for the numbers of fixed registers, 2124because the allocation mechanism considers them to be always occupied. 2125 2126@cindex register pairs 2127On some machines, double-precision values must be kept in even/odd 2128register pairs. You can implement that by defining this hook to reject 2129odd register numbers for such modes. 2130 2131The minimum requirement for a mode to be OK in a register is that the 2132@samp{mov@var{mode}} instruction pattern support moves between the 2133register and other hard register in the same class and that moving a 2134value into the register and back out not alter it. 2135 2136Since the same instruction used to move @code{word_mode} will work for 2137all narrower integer modes, it is not necessary on any machine for 2138this hook to distinguish between these modes, provided you define 2139patterns @samp{movhi}, etc., to take advantage of this. This is 2140useful because of the interaction between @code{TARGET_HARD_REGNO_MODE_OK} 2141and @code{TARGET_MODES_TIEABLE_P}; it is very desirable for all integer 2142modes to be tieable. 2143 2144Many machines have special registers for floating point arithmetic. 2145Often people assume that floating point machine modes are allowed only 2146in floating point registers. This is not true. Any registers that 2147can hold integers can safely @emph{hold} a floating point machine 2148mode, whether or not floating arithmetic can be done on it in those 2149registers. Integer move instructions can be used to move the values. 2150 2151On some machines, though, the converse is true: fixed-point machine 2152modes may not go in floating registers. This is true if the floating 2153registers normalize any value stored in them, because storing a 2154non-floating value there would garble it. In this case, 2155@code{TARGET_HARD_REGNO_MODE_OK} should reject fixed-point machine modes in 2156floating registers. But if the floating registers do not automatically 2157normalize, if you can store any bit pattern in one and retrieve it 2158unchanged without a trap, then any machine mode may go in a floating 2159register, so you can define this hook to say so. 2160 2161The primary significance of special floating registers is rather that 2162they are the registers acceptable in floating point arithmetic 2163instructions. However, this is of no concern to 2164@code{TARGET_HARD_REGNO_MODE_OK}. You handle it by writing the proper 2165constraints for those instructions. 2166 2167On some machines, the floating registers are especially slow to access, 2168so that it is better to store a value in a stack frame than in such a 2169register if floating point arithmetic is not being done. As long as the 2170floating registers are not in class @code{GENERAL_REGS}, they will not 2171be used unless some pattern's constraint asks for one. 2172@end deftypefn 2173 2174@defmac HARD_REGNO_RENAME_OK (@var{from}, @var{to}) 2175A C expression that is nonzero if it is OK to rename a hard register 2176@var{from} to another hard register @var{to}. 2177 2178One common use of this macro is to prevent renaming of a register to 2179another register that is not saved by a prologue in an interrupt 2180handler. 2181 2182The default is always nonzero. 2183@end defmac 2184 2185@deftypefn {Target Hook} bool TARGET_MODES_TIEABLE_P (machine_mode @var{mode1}, machine_mode @var{mode2}) 2186This hook returns true if a value of mode @var{mode1} is accessible 2187in mode @var{mode2} without copying. 2188 2189If @code{TARGET_HARD_REGNO_MODE_OK (@var{r}, @var{mode1})} and 2190@code{TARGET_HARD_REGNO_MODE_OK (@var{r}, @var{mode2})} are always 2191the same for any @var{r}, then 2192@code{TARGET_MODES_TIEABLE_P (@var{mode1}, @var{mode2})} 2193should be true. If they differ for any @var{r}, you should define 2194this hook to return false unless some other mechanism ensures the 2195accessibility of the value in a narrower mode. 2196 2197You should define this hook to return true in as many cases as 2198possible since doing so will allow GCC to perform better register 2199allocation. The default definition returns true unconditionally. 2200@end deftypefn 2201 2202@deftypefn {Target Hook} bool TARGET_HARD_REGNO_SCRATCH_OK (unsigned int @var{regno}) 2203This target hook should return @code{true} if it is OK to use a hard register 2204@var{regno} as scratch reg in peephole2. 2205 2206One common use of this macro is to prevent using of a register that 2207is not saved by a prologue in an interrupt handler. 2208 2209The default version of this hook always returns @code{true}. 2210@end deftypefn 2211 2212@defmac AVOID_CCMODE_COPIES 2213Define this macro if the compiler should avoid copies to/from @code{CCmode} 2214registers. You should only define this macro if support for copying to/from 2215@code{CCmode} is incomplete. 2216@end defmac 2217 2218@node Leaf Functions 2219@subsection Handling Leaf Functions 2220 2221@cindex leaf functions 2222@cindex functions, leaf 2223On some machines, a leaf function (i.e., one which makes no calls) can run 2224more efficiently if it does not make its own register window. Often this 2225means it is required to receive its arguments in the registers where they 2226are passed by the caller, instead of the registers where they would 2227normally arrive. 2228 2229The special treatment for leaf functions generally applies only when 2230other conditions are met; for example, often they may use only those 2231registers for its own variables and temporaries. We use the term ``leaf 2232function'' to mean a function that is suitable for this special 2233handling, so that functions with no calls are not necessarily ``leaf 2234functions''. 2235 2236GCC assigns register numbers before it knows whether the function is 2237suitable for leaf function treatment. So it needs to renumber the 2238registers in order to output a leaf function. The following macros 2239accomplish this. 2240 2241@defmac LEAF_REGISTERS 2242Name of a char vector, indexed by hard register number, which 2243contains 1 for a register that is allowable in a candidate for leaf 2244function treatment. 2245 2246If leaf function treatment involves renumbering the registers, then the 2247registers marked here should be the ones before renumbering---those that 2248GCC would ordinarily allocate. The registers which will actually be 2249used in the assembler code, after renumbering, should not be marked with 1 2250in this vector. 2251 2252Define this macro only if the target machine offers a way to optimize 2253the treatment of leaf functions. 2254@end defmac 2255 2256@defmac LEAF_REG_REMAP (@var{regno}) 2257A C expression whose value is the register number to which @var{regno} 2258should be renumbered, when a function is treated as a leaf function. 2259 2260If @var{regno} is a register number which should not appear in a leaf 2261function before renumbering, then the expression should yield @minus{}1, which 2262will cause the compiler to abort. 2263 2264Define this macro only if the target machine offers a way to optimize the 2265treatment of leaf functions, and registers need to be renumbered to do 2266this. 2267@end defmac 2268 2269@findex current_function_is_leaf 2270@findex current_function_uses_only_leaf_regs 2271@code{TARGET_ASM_FUNCTION_PROLOGUE} and 2272@code{TARGET_ASM_FUNCTION_EPILOGUE} must usually treat leaf functions 2273specially. They can test the C variable @code{current_function_is_leaf} 2274which is nonzero for leaf functions. @code{current_function_is_leaf} is 2275set prior to local register allocation and is valid for the remaining 2276compiler passes. They can also test the C variable 2277@code{current_function_uses_only_leaf_regs} which is nonzero for leaf 2278functions which only use leaf registers. 2279@code{current_function_uses_only_leaf_regs} is valid after all passes 2280that modify the instructions have been run and is only useful if 2281@code{LEAF_REGISTERS} is defined. 2282@c changed this to fix overfull. ALSO: why the "it" at the beginning 2283@c of the next paragraph?! --mew 2feb93 2284 2285@node Stack Registers 2286@subsection Registers That Form a Stack 2287 2288There are special features to handle computers where some of the 2289``registers'' form a stack. Stack registers are normally written by 2290pushing onto the stack, and are numbered relative to the top of the 2291stack. 2292 2293Currently, GCC can only handle one group of stack-like registers, and 2294they must be consecutively numbered. Furthermore, the existing 2295support for stack-like registers is specific to the 80387 floating 2296point coprocessor. If you have a new architecture that uses 2297stack-like registers, you will need to do substantial work on 2298@file{reg-stack.c} and write your machine description to cooperate 2299with it, as well as defining these macros. 2300 2301@defmac STACK_REGS 2302Define this if the machine has any stack-like registers. 2303@end defmac 2304 2305@defmac STACK_REG_COVER_CLASS 2306This is a cover class containing the stack registers. Define this if 2307the machine has any stack-like registers. 2308@end defmac 2309 2310@defmac FIRST_STACK_REG 2311The number of the first stack-like register. This one is the top 2312of the stack. 2313@end defmac 2314 2315@defmac LAST_STACK_REG 2316The number of the last stack-like register. This one is the bottom of 2317the stack. 2318@end defmac 2319 2320@node Register Classes 2321@section Register Classes 2322@cindex register class definitions 2323@cindex class definitions, register 2324 2325On many machines, the numbered registers are not all equivalent. 2326For example, certain registers may not be allowed for indexed addressing; 2327certain registers may not be allowed in some instructions. These machine 2328restrictions are described to the compiler using @dfn{register classes}. 2329 2330You define a number of register classes, giving each one a name and saying 2331which of the registers belong to it. Then you can specify register classes 2332that are allowed as operands to particular instruction patterns. 2333 2334@findex ALL_REGS 2335@findex NO_REGS 2336In general, each register will belong to several classes. In fact, one 2337class must be named @code{ALL_REGS} and contain all the registers. Another 2338class must be named @code{NO_REGS} and contain no registers. Often the 2339union of two classes will be another class; however, this is not required. 2340 2341@findex GENERAL_REGS 2342One of the classes must be named @code{GENERAL_REGS}. There is nothing 2343terribly special about the name, but the operand constraint letters 2344@samp{r} and @samp{g} specify this class. If @code{GENERAL_REGS} is 2345the same as @code{ALL_REGS}, just define it as a macro which expands 2346to @code{ALL_REGS}. 2347 2348Order the classes so that if class @var{x} is contained in class @var{y} 2349then @var{x} has a lower class number than @var{y}. 2350 2351The way classes other than @code{GENERAL_REGS} are specified in operand 2352constraints is through machine-dependent operand constraint letters. 2353You can define such letters to correspond to various classes, then use 2354them in operand constraints. 2355 2356You must define the narrowest register classes for allocatable 2357registers, so that each class either has no subclasses, or that for 2358some mode, the move cost between registers within the class is 2359cheaper than moving a register in the class to or from memory 2360(@pxref{Costs}). 2361 2362You should define a class for the union of two classes whenever some 2363instruction allows both classes. For example, if an instruction allows 2364either a floating point (coprocessor) register or a general register for a 2365certain operand, you should define a class @code{FLOAT_OR_GENERAL_REGS} 2366which includes both of them. Otherwise you will get suboptimal code, 2367or even internal compiler errors when reload cannot find a register in the 2368class computed via @code{reg_class_subunion}. 2369 2370You must also specify certain redundant information about the register 2371classes: for each class, which classes contain it and which ones are 2372contained in it; for each pair of classes, the largest class contained 2373in their union. 2374 2375When a value occupying several consecutive registers is expected in a 2376certain class, all the registers used must belong to that class. 2377Therefore, register classes cannot be used to enforce a requirement for 2378a register pair to start with an even-numbered register. The way to 2379specify this requirement is with @code{TARGET_HARD_REGNO_MODE_OK}. 2380 2381Register classes used for input-operands of bitwise-and or shift 2382instructions have a special requirement: each such class must have, for 2383each fixed-point machine mode, a subclass whose registers can transfer that 2384mode to or from memory. For example, on some machines, the operations for 2385single-byte values (@code{QImode}) are limited to certain registers. When 2386this is so, each register class that is used in a bitwise-and or shift 2387instruction must have a subclass consisting of registers from which 2388single-byte values can be loaded or stored. This is so that 2389@code{PREFERRED_RELOAD_CLASS} can always have a possible value to return. 2390 2391@deftp {Data type} {enum reg_class} 2392An enumerated type that must be defined with all the register class names 2393as enumerated values. @code{NO_REGS} must be first. @code{ALL_REGS} 2394must be the last register class, followed by one more enumerated value, 2395@code{LIM_REG_CLASSES}, which is not a register class but rather 2396tells how many classes there are. 2397 2398Each register class has a number, which is the value of casting 2399the class name to type @code{int}. The number serves as an index 2400in many of the tables described below. 2401@end deftp 2402 2403@defmac N_REG_CLASSES 2404The number of distinct register classes, defined as follows: 2405 2406@smallexample 2407#define N_REG_CLASSES (int) LIM_REG_CLASSES 2408@end smallexample 2409@end defmac 2410 2411@defmac REG_CLASS_NAMES 2412An initializer containing the names of the register classes as C string 2413constants. These names are used in writing some of the debugging dumps. 2414@end defmac 2415 2416@defmac REG_CLASS_CONTENTS 2417An initializer containing the contents of the register classes, as integers 2418which are bit masks. The @var{n}th integer specifies the contents of class 2419@var{n}. The way the integer @var{mask} is interpreted is that 2420register @var{r} is in the class if @code{@var{mask} & (1 << @var{r})} is 1. 2421 2422When the machine has more than 32 registers, an integer does not suffice. 2423Then the integers are replaced by sub-initializers, braced groupings containing 2424several integers. Each sub-initializer must be suitable as an initializer 2425for the type @code{HARD_REG_SET} which is defined in @file{hard-reg-set.h}. 2426In this situation, the first integer in each sub-initializer corresponds to 2427registers 0 through 31, the second integer to registers 32 through 63, and 2428so on. 2429@end defmac 2430 2431@defmac REGNO_REG_CLASS (@var{regno}) 2432A C expression whose value is a register class containing hard register 2433@var{regno}. In general there is more than one such class; choose a class 2434which is @dfn{minimal}, meaning that no smaller class also contains the 2435register. 2436@end defmac 2437 2438@defmac BASE_REG_CLASS 2439A macro whose definition is the name of the class to which a valid 2440base register must belong. A base register is one used in an address 2441which is the register value plus a displacement. 2442@end defmac 2443 2444@defmac MODE_BASE_REG_CLASS (@var{mode}) 2445This is a variation of the @code{BASE_REG_CLASS} macro which allows 2446the selection of a base register in a mode dependent manner. If 2447@var{mode} is VOIDmode then it should return the same value as 2448@code{BASE_REG_CLASS}. 2449@end defmac 2450 2451@defmac MODE_BASE_REG_REG_CLASS (@var{mode}) 2452A C expression whose value is the register class to which a valid 2453base register must belong in order to be used in a base plus index 2454register address. You should define this macro if base plus index 2455addresses have different requirements than other base register uses. 2456@end defmac 2457 2458@defmac MODE_CODE_BASE_REG_CLASS (@var{mode}, @var{address_space}, @var{outer_code}, @var{index_code}) 2459A C expression whose value is the register class to which a valid 2460base register for a memory reference in mode @var{mode} to address 2461space @var{address_space} must belong. @var{outer_code} and @var{index_code} 2462define the context in which the base register occurs. @var{outer_code} is 2463the code of the immediately enclosing expression (@code{MEM} for the top level 2464of an address, @code{ADDRESS} for something that occurs in an 2465@code{address_operand}). @var{index_code} is the code of the corresponding 2466index expression if @var{outer_code} is @code{PLUS}; @code{SCRATCH} otherwise. 2467@end defmac 2468 2469@defmac INDEX_REG_CLASS 2470A macro whose definition is the name of the class to which a valid 2471index register must belong. An index register is one used in an 2472address where its value is either multiplied by a scale factor or 2473added to another register (as well as added to a displacement). 2474@end defmac 2475 2476@defmac REGNO_OK_FOR_BASE_P (@var{num}) 2477A C expression which is nonzero if register number @var{num} is 2478suitable for use as a base register in operand addresses. 2479@end defmac 2480 2481@defmac REGNO_MODE_OK_FOR_BASE_P (@var{num}, @var{mode}) 2482A C expression that is just like @code{REGNO_OK_FOR_BASE_P}, except that 2483that expression may examine the mode of the memory reference in 2484@var{mode}. You should define this macro if the mode of the memory 2485reference affects whether a register may be used as a base register. If 2486you define this macro, the compiler will use it instead of 2487@code{REGNO_OK_FOR_BASE_P}. The mode may be @code{VOIDmode} for 2488addresses that appear outside a @code{MEM}, i.e., as an 2489@code{address_operand}. 2490@end defmac 2491 2492@defmac REGNO_MODE_OK_FOR_REG_BASE_P (@var{num}, @var{mode}) 2493A C expression which is nonzero if register number @var{num} is suitable for 2494use as a base register in base plus index operand addresses, accessing 2495memory in mode @var{mode}. It may be either a suitable hard register or a 2496pseudo register that has been allocated such a hard register. You should 2497define this macro if base plus index addresses have different requirements 2498than other base register uses. 2499 2500Use of this macro is deprecated; please use the more general 2501@code{REGNO_MODE_CODE_OK_FOR_BASE_P}. 2502@end defmac 2503 2504@defmac REGNO_MODE_CODE_OK_FOR_BASE_P (@var{num}, @var{mode}, @var{address_space}, @var{outer_code}, @var{index_code}) 2505A C expression which is nonzero if register number @var{num} is 2506suitable for use as a base register in operand addresses, accessing 2507memory in mode @var{mode} in address space @var{address_space}. 2508This is similar to @code{REGNO_MODE_OK_FOR_BASE_P}, except 2509that that expression may examine the context in which the register 2510appears in the memory reference. @var{outer_code} is the code of the 2511immediately enclosing expression (@code{MEM} if at the top level of the 2512address, @code{ADDRESS} for something that occurs in an 2513@code{address_operand}). @var{index_code} is the code of the 2514corresponding index expression if @var{outer_code} is @code{PLUS}; 2515@code{SCRATCH} otherwise. The mode may be @code{VOIDmode} for addresses 2516that appear outside a @code{MEM}, i.e., as an @code{address_operand}. 2517@end defmac 2518 2519@defmac REGNO_OK_FOR_INDEX_P (@var{num}) 2520A C expression which is nonzero if register number @var{num} is 2521suitable for use as an index register in operand addresses. It may be 2522either a suitable hard register or a pseudo register that has been 2523allocated such a hard register. 2524 2525The difference between an index register and a base register is that 2526the index register may be scaled. If an address involves the sum of 2527two registers, neither one of them scaled, then either one may be 2528labeled the ``base'' and the other the ``index''; but whichever 2529labeling is used must fit the machine's constraints of which registers 2530may serve in each capacity. The compiler will try both labelings, 2531looking for one that is valid, and will reload one or both registers 2532only if neither labeling works. 2533@end defmac 2534 2535@deftypefn {Target Hook} reg_class_t TARGET_PREFERRED_RENAME_CLASS (reg_class_t @var{rclass}) 2536A target hook that places additional preference on the register class to use when it is necessary to rename a register in class @var{rclass} to another class, or perhaps @var{NO_REGS}, if no preferred register class is found or hook @code{preferred_rename_class} is not implemented. Sometimes returning a more restrictive class makes better code. For example, on ARM, thumb-2 instructions using @code{LO_REGS} may be smaller than instructions using @code{GENERIC_REGS}. By returning @code{LO_REGS} from @code{preferred_rename_class}, code size can be reduced. 2537@end deftypefn 2538 2539@deftypefn {Target Hook} reg_class_t TARGET_PREFERRED_RELOAD_CLASS (rtx @var{x}, reg_class_t @var{rclass}) 2540A target hook that places additional restrictions on the register class 2541to use when it is necessary to copy value @var{x} into a register in class 2542@var{rclass}. The value is a register class; perhaps @var{rclass}, or perhaps 2543another, smaller class. 2544 2545The default version of this hook always returns value of @code{rclass} argument. 2546 2547Sometimes returning a more restrictive class makes better code. For 2548example, on the 68000, when @var{x} is an integer constant that is in range 2549for a @samp{moveq} instruction, the value of this macro is always 2550@code{DATA_REGS} as long as @var{rclass} includes the data registers. 2551Requiring a data register guarantees that a @samp{moveq} will be used. 2552 2553One case where @code{TARGET_PREFERRED_RELOAD_CLASS} must not return 2554@var{rclass} is if @var{x} is a legitimate constant which cannot be 2555loaded into some register class. By returning @code{NO_REGS} you can 2556force @var{x} into a memory location. For example, rs6000 can load 2557immediate values into general-purpose registers, but does not have an 2558instruction for loading an immediate value into a floating-point 2559register, so @code{TARGET_PREFERRED_RELOAD_CLASS} returns @code{NO_REGS} when 2560@var{x} is a floating-point constant. If the constant can't be loaded 2561into any kind of register, code generation will be better if 2562@code{TARGET_LEGITIMATE_CONSTANT_P} makes the constant illegitimate instead 2563of using @code{TARGET_PREFERRED_RELOAD_CLASS}. 2564 2565If an insn has pseudos in it after register allocation, reload will go 2566through the alternatives and call repeatedly @code{TARGET_PREFERRED_RELOAD_CLASS} 2567to find the best one. Returning @code{NO_REGS}, in this case, makes 2568reload add a @code{!} in front of the constraint: the x86 back-end uses 2569this feature to discourage usage of 387 registers when math is done in 2570the SSE registers (and vice versa). 2571@end deftypefn 2572 2573@defmac PREFERRED_RELOAD_CLASS (@var{x}, @var{class}) 2574A C expression that places additional restrictions on the register class 2575to use when it is necessary to copy value @var{x} into a register in class 2576@var{class}. The value is a register class; perhaps @var{class}, or perhaps 2577another, smaller class. On many machines, the following definition is 2578safe: 2579 2580@smallexample 2581#define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS 2582@end smallexample 2583 2584Sometimes returning a more restrictive class makes better code. For 2585example, on the 68000, when @var{x} is an integer constant that is in range 2586for a @samp{moveq} instruction, the value of this macro is always 2587@code{DATA_REGS} as long as @var{class} includes the data registers. 2588Requiring a data register guarantees that a @samp{moveq} will be used. 2589 2590One case where @code{PREFERRED_RELOAD_CLASS} must not return 2591@var{class} is if @var{x} is a legitimate constant which cannot be 2592loaded into some register class. By returning @code{NO_REGS} you can 2593force @var{x} into a memory location. For example, rs6000 can load 2594immediate values into general-purpose registers, but does not have an 2595instruction for loading an immediate value into a floating-point 2596register, so @code{PREFERRED_RELOAD_CLASS} returns @code{NO_REGS} when 2597@var{x} is a floating-point constant. If the constant cannot be loaded 2598into any kind of register, code generation will be better if 2599@code{TARGET_LEGITIMATE_CONSTANT_P} makes the constant illegitimate instead 2600of using @code{TARGET_PREFERRED_RELOAD_CLASS}. 2601 2602If an insn has pseudos in it after register allocation, reload will go 2603through the alternatives and call repeatedly @code{PREFERRED_RELOAD_CLASS} 2604to find the best one. Returning @code{NO_REGS}, in this case, makes 2605reload add a @code{!} in front of the constraint: the x86 back-end uses 2606this feature to discourage usage of 387 registers when math is done in 2607the SSE registers (and vice versa). 2608@end defmac 2609 2610@deftypefn {Target Hook} reg_class_t TARGET_PREFERRED_OUTPUT_RELOAD_CLASS (rtx @var{x}, reg_class_t @var{rclass}) 2611Like @code{TARGET_PREFERRED_RELOAD_CLASS}, but for output reloads instead of 2612input reloads. 2613 2614The default version of this hook always returns value of @code{rclass} 2615argument. 2616 2617You can also use @code{TARGET_PREFERRED_OUTPUT_RELOAD_CLASS} to discourage 2618reload from using some alternatives, like @code{TARGET_PREFERRED_RELOAD_CLASS}. 2619@end deftypefn 2620 2621@defmac LIMIT_RELOAD_CLASS (@var{mode}, @var{class}) 2622A C expression that places additional restrictions on the register class 2623to use when it is necessary to be able to hold a value of mode 2624@var{mode} in a reload register for which class @var{class} would 2625ordinarily be used. 2626 2627Unlike @code{PREFERRED_RELOAD_CLASS}, this macro should be used when 2628there are certain modes that simply cannot go in certain reload classes. 2629 2630The value is a register class; perhaps @var{class}, or perhaps another, 2631smaller class. 2632 2633Don't define this macro unless the target machine has limitations which 2634require the macro to do something nontrivial. 2635@end defmac 2636 2637@deftypefn {Target Hook} reg_class_t TARGET_SECONDARY_RELOAD (bool @var{in_p}, rtx @var{x}, reg_class_t @var{reload_class}, machine_mode @var{reload_mode}, secondary_reload_info *@var{sri}) 2638Many machines have some registers that cannot be copied directly to or 2639from memory or even from other types of registers. An example is the 2640@samp{MQ} register, which on most machines, can only be copied to or 2641from general registers, but not memory. Below, we shall be using the 2642term 'intermediate register' when a move operation cannot be performed 2643directly, but has to be done by copying the source into the intermediate 2644register first, and then copying the intermediate register to the 2645destination. An intermediate register always has the same mode as 2646source and destination. Since it holds the actual value being copied, 2647reload might apply optimizations to re-use an intermediate register 2648and eliding the copy from the source when it can determine that the 2649intermediate register still holds the required value. 2650 2651Another kind of secondary reload is required on some machines which 2652allow copying all registers to and from memory, but require a scratch 2653register for stores to some memory locations (e.g., those with symbolic 2654address on the RT, and those with certain symbolic address on the SPARC 2655when compiling PIC)@. Scratch registers need not have the same mode 2656as the value being copied, and usually hold a different value than 2657that being copied. Special patterns in the md file are needed to 2658describe how the copy is performed with the help of the scratch register; 2659these patterns also describe the number, register class(es) and mode(s) 2660of the scratch register(s). 2661 2662In some cases, both an intermediate and a scratch register are required. 2663 2664For input reloads, this target hook is called with nonzero @var{in_p}, 2665and @var{x} is an rtx that needs to be copied to a register of class 2666@var{reload_class} in @var{reload_mode}. For output reloads, this target 2667hook is called with zero @var{in_p}, and a register of class @var{reload_class} 2668needs to be copied to rtx @var{x} in @var{reload_mode}. 2669 2670If copying a register of @var{reload_class} from/to @var{x} requires 2671an intermediate register, the hook @code{secondary_reload} should 2672return the register class required for this intermediate register. 2673If no intermediate register is required, it should return NO_REGS. 2674If more than one intermediate register is required, describe the one 2675that is closest in the copy chain to the reload register. 2676 2677If scratch registers are needed, you also have to describe how to 2678perform the copy from/to the reload register to/from this 2679closest intermediate register. Or if no intermediate register is 2680required, but still a scratch register is needed, describe the 2681copy from/to the reload register to/from the reload operand @var{x}. 2682 2683You do this by setting @code{sri->icode} to the instruction code of a pattern 2684in the md file which performs the move. Operands 0 and 1 are the output 2685and input of this copy, respectively. Operands from operand 2 onward are 2686for scratch operands. These scratch operands must have a mode, and a 2687single-register-class 2688@c [later: or memory] 2689output constraint. 2690 2691When an intermediate register is used, the @code{secondary_reload} 2692hook will be called again to determine how to copy the intermediate 2693register to/from the reload operand @var{x}, so your hook must also 2694have code to handle the register class of the intermediate operand. 2695 2696@c [For later: maybe we'll allow multi-alternative reload patterns - 2697@c the port maintainer could name a mov<mode> pattern that has clobbers - 2698@c and match the constraints of input and output to determine the required 2699@c alternative. A restriction would be that constraints used to match 2700@c against reloads registers would have to be written as register class 2701@c constraints, or we need a new target macro / hook that tells us if an 2702@c arbitrary constraint can match an unknown register of a given class. 2703@c Such a macro / hook would also be useful in other places.] 2704 2705 2706@var{x} might be a pseudo-register or a @code{subreg} of a 2707pseudo-register, which could either be in a hard register or in memory. 2708Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is 2709in memory and the hard register number if it is in a register. 2710 2711Scratch operands in memory (constraint @code{"=m"} / @code{"=&m"}) are 2712currently not supported. For the time being, you will have to continue 2713to use @code{TARGET_SECONDARY_MEMORY_NEEDED} for that purpose. 2714 2715@code{copy_cost} also uses this target hook to find out how values are 2716copied. If you want it to include some extra cost for the need to allocate 2717(a) scratch register(s), set @code{sri->extra_cost} to the additional cost. 2718Or if two dependent moves are supposed to have a lower cost than the sum 2719of the individual moves due to expected fortuitous scheduling and/or special 2720forwarding logic, you can set @code{sri->extra_cost} to a negative amount. 2721@end deftypefn 2722 2723@defmac SECONDARY_RELOAD_CLASS (@var{class}, @var{mode}, @var{x}) 2724@defmacx SECONDARY_INPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x}) 2725@defmacx SECONDARY_OUTPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x}) 2726These macros are obsolete, new ports should use the target hook 2727@code{TARGET_SECONDARY_RELOAD} instead. 2728 2729These are obsolete macros, replaced by the @code{TARGET_SECONDARY_RELOAD} 2730target hook. Older ports still define these macros to indicate to the 2731reload phase that it may 2732need to allocate at least one register for a reload in addition to the 2733register to contain the data. Specifically, if copying @var{x} to a 2734register @var{class} in @var{mode} requires an intermediate register, 2735you were supposed to define @code{SECONDARY_INPUT_RELOAD_CLASS} to return the 2736largest register class all of whose registers can be used as 2737intermediate registers or scratch registers. 2738 2739If copying a register @var{class} in @var{mode} to @var{x} requires an 2740intermediate or scratch register, @code{SECONDARY_OUTPUT_RELOAD_CLASS} 2741was supposed to be defined be defined to return the largest register 2742class required. If the 2743requirements for input and output reloads were the same, the macro 2744@code{SECONDARY_RELOAD_CLASS} should have been used instead of defining both 2745macros identically. 2746 2747The values returned by these macros are often @code{GENERAL_REGS}. 2748Return @code{NO_REGS} if no spare register is needed; i.e., if @var{x} 2749can be directly copied to or from a register of @var{class} in 2750@var{mode} without requiring a scratch register. Do not define this 2751macro if it would always return @code{NO_REGS}. 2752 2753If a scratch register is required (either with or without an 2754intermediate register), you were supposed to define patterns for 2755@samp{reload_in@var{m}} or @samp{reload_out@var{m}}, as required 2756(@pxref{Standard Names}. These patterns, which were normally 2757implemented with a @code{define_expand}, should be similar to the 2758@samp{mov@var{m}} patterns, except that operand 2 is the scratch 2759register. 2760 2761These patterns need constraints for the reload register and scratch 2762register that 2763contain a single register class. If the original reload register (whose 2764class is @var{class}) can meet the constraint given in the pattern, the 2765value returned by these macros is used for the class of the scratch 2766register. Otherwise, two additional reload registers are required. 2767Their classes are obtained from the constraints in the insn pattern. 2768 2769@var{x} might be a pseudo-register or a @code{subreg} of a 2770pseudo-register, which could either be in a hard register or in memory. 2771Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is 2772in memory and the hard register number if it is in a register. 2773 2774These macros should not be used in the case where a particular class of 2775registers can only be copied to memory and not to another class of 2776registers. In that case, secondary reload registers are not needed and 2777would not be helpful. Instead, a stack location must be used to perform 2778the copy and the @code{mov@var{m}} pattern should use memory as an 2779intermediate storage. This case often occurs between floating-point and 2780general registers. 2781@end defmac 2782 2783@deftypefn {Target Hook} bool TARGET_SECONDARY_MEMORY_NEEDED (machine_mode @var{mode}, reg_class_t @var{class1}, reg_class_t @var{class2}) 2784Certain machines have the property that some registers cannot be copied 2785to some other registers without using memory. Define this hook on 2786those machines to return true if objects of mode @var{m} in registers 2787of @var{class1} can only be copied to registers of class @var{class2} by 2788 storing a register of @var{class1} into memory and loading that memory 2789location into a register of @var{class2}. The default definition returns 2790false for all inputs. 2791@end deftypefn 2792 2793@defmac SECONDARY_MEMORY_NEEDED_RTX (@var{mode}) 2794Normally when @code{TARGET_SECONDARY_MEMORY_NEEDED} is defined, the compiler 2795allocates a stack slot for a memory location needed for register copies. 2796If this macro is defined, the compiler instead uses the memory location 2797defined by this macro. 2798 2799Do not define this macro if you do not define 2800@code{TARGET_SECONDARY_MEMORY_NEEDED}. 2801@end defmac 2802 2803@deftypefn {Target Hook} machine_mode TARGET_SECONDARY_MEMORY_NEEDED_MODE (machine_mode @var{mode}) 2804If @code{TARGET_SECONDARY_MEMORY_NEEDED} tells the compiler to use memory 2805when moving between two particular registers of mode @var{mode}, 2806this hook specifies the mode that the memory should have. 2807 2808The default depends on @code{TARGET_LRA_P}. Without LRA, the default 2809is to use a word-sized mode for integral modes that are smaller than a 2810a word. This is right thing to do on most machines because it ensures 2811that all bits of the register are copied and prevents accesses to the 2812registers in a narrower mode, which some machines prohibit for 2813floating-point registers. 2814 2815However, this default behavior is not correct on some machines, such as 2816the DEC Alpha, that store short integers in floating-point registers 2817differently than in integer registers. On those machines, the default 2818widening will not work correctly and you must define this hook to 2819suppress that widening in some cases. See the file @file{alpha.c} for 2820details. 2821 2822With LRA, the default is to use @var{mode} unmodified. 2823@end deftypefn 2824 2825@deftypefn {Target Hook} void TARGET_SELECT_EARLY_REMAT_MODES (sbitmap @var{modes}) 2826On some targets, certain modes cannot be held in registers around a 2827standard ABI call and are relatively expensive to spill to the stack. 2828The early rematerialization pass can help in such cases by aggressively 2829recomputing values after calls, so that they don't need to be spilled. 2830 2831This hook returns the set of such modes by setting the associated bits 2832in @var{modes}. The default implementation selects no modes, which has 2833the effect of disabling the early rematerialization pass. 2834@end deftypefn 2835 2836@deftypefn {Target Hook} bool TARGET_CLASS_LIKELY_SPILLED_P (reg_class_t @var{rclass}) 2837A target hook which returns @code{true} if pseudos that have been assigned 2838to registers of class @var{rclass} would likely be spilled because 2839registers of @var{rclass} are needed for spill registers. 2840 2841The default version of this target hook returns @code{true} if @var{rclass} 2842has exactly one register and @code{false} otherwise. On most machines, this 2843default should be used. For generally register-starved machines, such as 2844i386, or machines with right register constraints, such as SH, this hook 2845can be used to avoid excessive spilling. 2846 2847This hook is also used by some of the global intra-procedural code 2848transformations to throtle code motion, to avoid increasing register 2849pressure. 2850@end deftypefn 2851 2852@deftypefn {Target Hook} {unsigned char} TARGET_CLASS_MAX_NREGS (reg_class_t @var{rclass}, machine_mode @var{mode}) 2853A target hook returns the maximum number of consecutive registers 2854of class @var{rclass} needed to hold a value of mode @var{mode}. 2855 2856This is closely related to the macro @code{TARGET_HARD_REGNO_NREGS}. 2857In fact, the value returned by @code{TARGET_CLASS_MAX_NREGS (@var{rclass}, 2858@var{mode})} target hook should be the maximum value of 2859@code{TARGET_HARD_REGNO_NREGS (@var{regno}, @var{mode})} for all @var{regno} 2860values in the class @var{rclass}. 2861 2862This target hook helps control the handling of multiple-word values 2863in the reload pass. 2864 2865The default version of this target hook returns the size of @var{mode} 2866in words. 2867@end deftypefn 2868 2869@defmac CLASS_MAX_NREGS (@var{class}, @var{mode}) 2870A C expression for the maximum number of consecutive registers 2871of class @var{class} needed to hold a value of mode @var{mode}. 2872 2873This is closely related to the macro @code{TARGET_HARD_REGNO_NREGS}. In fact, 2874the value of the macro @code{CLASS_MAX_NREGS (@var{class}, @var{mode})} 2875should be the maximum value of @code{TARGET_HARD_REGNO_NREGS (@var{regno}, 2876@var{mode})} for all @var{regno} values in the class @var{class}. 2877 2878This macro helps control the handling of multiple-word values 2879in the reload pass. 2880@end defmac 2881 2882@deftypefn {Target Hook} bool TARGET_CAN_CHANGE_MODE_CLASS (machine_mode @var{from}, machine_mode @var{to}, reg_class_t @var{rclass}) 2883This hook returns true if it is possible to bitcast values held in 2884registers of class @var{rclass} from mode @var{from} to mode @var{to} 2885and if doing so preserves the low-order bits that are common to both modes. 2886The result is only meaningful if @var{rclass} has registers that can hold 2887both @code{from} and @code{to}. The default implementation returns true. 2888 2889As an example of when such bitcasting is invalid, loading 32-bit integer or 2890floating-point objects into floating-point registers on Alpha extends them 2891to 64 bits. Therefore loading a 64-bit object and then storing it as a 289232-bit object does not store the low-order 32 bits, as would be the case 2893for a normal register. Therefore, @file{alpha.h} defines 2894@code{TARGET_CAN_CHANGE_MODE_CLASS} to return: 2895 2896@smallexample 2897(GET_MODE_SIZE (from) == GET_MODE_SIZE (to) 2898 || !reg_classes_intersect_p (FLOAT_REGS, rclass)) 2899@end smallexample 2900 2901Even if storing from a register in mode @var{to} would be valid, 2902if both @var{from} and @code{raw_reg_mode} for @var{rclass} are wider 2903than @code{word_mode}, then we must prevent @var{to} narrowing the 2904mode. This happens when the middle-end assumes that it can load 2905or store pieces of an @var{N}-word pseudo, and that the pseudo will 2906eventually be allocated to @var{N} @code{word_mode} hard registers. 2907Failure to prevent this kind of mode change will result in the 2908entire @code{raw_reg_mode} being modified instead of the partial 2909value that the middle-end intended. 2910@end deftypefn 2911 2912@deftypefn {Target Hook} reg_class_t TARGET_IRA_CHANGE_PSEUDO_ALLOCNO_CLASS (int, @var{reg_class_t}, @var{reg_class_t}) 2913A target hook which can change allocno class for given pseudo from 2914 allocno and best class calculated by IRA. 2915 2916 The default version of this target hook always returns given class. 2917@end deftypefn 2918 2919@deftypefn {Target Hook} bool TARGET_LRA_P (void) 2920A target hook which returns true if we use LRA instead of reload pass. The default version of this target hook returns true. New ports should use LRA, and existing ports are encouraged to convert. 2921@end deftypefn 2922 2923@deftypefn {Target Hook} int TARGET_REGISTER_PRIORITY (int) 2924A target hook which returns the register priority number to which the register @var{hard_regno} belongs to. The bigger the number, the more preferable the hard register usage (when all other conditions are the same). This hook can be used to prefer some hard register over others in LRA. For example, some x86-64 register usage needs additional prefix which makes instructions longer. The hook can return lower priority number for such registers make them less favorable and as result making the generated code smaller. The default version of this target hook returns always zero. 2925@end deftypefn 2926 2927@deftypefn {Target Hook} bool TARGET_REGISTER_USAGE_LEVELING_P (void) 2928A target hook which returns true if we need register usage leveling. That means if a few hard registers are equally good for the assignment, we choose the least used hard register. The register usage leveling may be profitable for some targets. Don't use the usage leveling for targets with conditional execution or targets with big register files as it hurts if-conversion and cross-jumping optimizations. The default version of this target hook returns always false. 2929@end deftypefn 2930 2931@deftypefn {Target Hook} bool TARGET_DIFFERENT_ADDR_DISPLACEMENT_P (void) 2932A target hook which returns true if an address with the same structure can have different maximal legitimate displacement. For example, the displacement can depend on memory mode or on operand combinations in the insn. The default version of this target hook returns always false. 2933@end deftypefn 2934 2935@deftypefn {Target Hook} bool TARGET_CANNOT_SUBSTITUTE_MEM_EQUIV_P (rtx @var{subst}) 2936A target hook which returns @code{true} if @var{subst} can't 2937substitute safely pseudos with equivalent memory values during 2938register allocation. 2939The default version of this target hook returns @code{false}. 2940On most machines, this default should be used. For generally 2941machines with non orthogonal register usage for addressing, such 2942as SH, this hook can be used to avoid excessive spilling. 2943@end deftypefn 2944 2945@deftypefn {Target Hook} bool TARGET_LEGITIMIZE_ADDRESS_DISPLACEMENT (rtx *@var{offset1}, rtx *@var{offset2}, poly_int64 @var{orig_offset}, machine_mode @var{mode}) 2946This hook tries to split address offset @var{orig_offset} into 2947two parts: one that should be added to the base address to create 2948a local anchor point, and an additional offset that can be applied 2949to the anchor to address a value of mode @var{mode}. The idea is that 2950the local anchor could be shared by other accesses to nearby locations. 2951 2952The hook returns true if it succeeds, storing the offset of the 2953anchor from the base in @var{offset1} and the offset of the final address 2954from the anchor in @var{offset2}. The default implementation returns false. 2955@end deftypefn 2956 2957@deftypefn {Target Hook} reg_class_t TARGET_SPILL_CLASS (reg_class_t, @var{machine_mode}) 2958This hook defines a class of registers which could be used for spilling pseudos of the given mode and class, or @code{NO_REGS} if only memory should be used. Not defining this hook is equivalent to returning @code{NO_REGS} for all inputs. 2959@end deftypefn 2960 2961@deftypefn {Target Hook} bool TARGET_ADDITIONAL_ALLOCNO_CLASS_P (reg_class_t) 2962This hook should return @code{true} if given class of registers should be an allocno class in any way. Usually RA uses only one register class from all classes containing the same register set. In some complicated cases, you need to have two or more such classes as allocno ones for RA correct work. Not defining this hook is equivalent to returning @code{false} for all inputs. 2963@end deftypefn 2964 2965@deftypefn {Target Hook} scalar_int_mode TARGET_CSTORE_MODE (enum insn_code @var{icode}) 2966This hook defines the machine mode to use for the boolean result of conditional store patterns. The ICODE argument is the instruction code for the cstore being performed. Not definiting this hook is the same as accepting the mode encoded into operand 0 of the cstore expander patterns. 2967@end deftypefn 2968 2969@deftypefn {Target Hook} int TARGET_COMPUTE_PRESSURE_CLASSES (enum reg_class *@var{pressure_classes}) 2970A target hook which lets a backend compute the set of pressure classes to be used by those optimization passes which take register pressure into account, as opposed to letting IRA compute them. It returns the number of register classes stored in the array @var{pressure_classes}. 2971@end deftypefn 2972 2973@node Stack and Calling 2974@section Stack Layout and Calling Conventions 2975@cindex calling conventions 2976 2977@c prevent bad page break with this line 2978This describes the stack layout and calling conventions. 2979 2980@menu 2981* Frame Layout:: 2982* Exception Handling:: 2983* Stack Checking:: 2984* Frame Registers:: 2985* Elimination:: 2986* Stack Arguments:: 2987* Register Arguments:: 2988* Scalar Return:: 2989* Aggregate Return:: 2990* Caller Saves:: 2991* Function Entry:: 2992* Profiling:: 2993* Tail Calls:: 2994* Shrink-wrapping separate components:: 2995* Stack Smashing Protection:: 2996* Miscellaneous Register Hooks:: 2997@end menu 2998 2999@node Frame Layout 3000@subsection Basic Stack Layout 3001@cindex stack frame layout 3002@cindex frame layout 3003 3004@c prevent bad page break with this line 3005Here is the basic stack layout. 3006 3007@defmac STACK_GROWS_DOWNWARD 3008Define this macro to be true if pushing a word onto the stack moves the stack 3009pointer to a smaller address, and false otherwise. 3010@end defmac 3011 3012@defmac STACK_PUSH_CODE 3013This macro defines the operation used when something is pushed 3014on the stack. In RTL, a push operation will be 3015@code{(set (mem (STACK_PUSH_CODE (reg sp))) @dots{})} 3016 3017The choices are @code{PRE_DEC}, @code{POST_DEC}, @code{PRE_INC}, 3018and @code{POST_INC}. Which of these is correct depends on 3019the stack direction and on whether the stack pointer points 3020to the last item on the stack or whether it points to the 3021space for the next item on the stack. 3022 3023The default is @code{PRE_DEC} when @code{STACK_GROWS_DOWNWARD} is 3024true, which is almost always right, and @code{PRE_INC} otherwise, 3025which is often wrong. 3026@end defmac 3027 3028@defmac FRAME_GROWS_DOWNWARD 3029Define this macro to nonzero value if the addresses of local variable slots 3030are at negative offsets from the frame pointer. 3031@end defmac 3032 3033@defmac ARGS_GROW_DOWNWARD 3034Define this macro if successive arguments to a function occupy decreasing 3035addresses on the stack. 3036@end defmac 3037 3038@deftypefn {Target Hook} HOST_WIDE_INT TARGET_STARTING_FRAME_OFFSET (void) 3039This hook returns the offset from the frame pointer to the first local 3040variable slot to be allocated. If @code{FRAME_GROWS_DOWNWARD}, it is the 3041offset to @emph{end} of the first slot allocated, otherwise it is the 3042offset to @emph{beginning} of the first slot allocated. The default 3043implementation returns 0. 3044@end deftypefn 3045 3046@defmac STACK_ALIGNMENT_NEEDED 3047Define to zero to disable final alignment of the stack during reload. 3048The nonzero default for this macro is suitable for most ports. 3049 3050On ports where @code{TARGET_STARTING_FRAME_OFFSET} is nonzero or where there 3051is a register save block following the local block that doesn't require 3052alignment to @code{STACK_BOUNDARY}, it may be beneficial to disable 3053stack alignment and do it in the backend. 3054@end defmac 3055 3056@defmac STACK_POINTER_OFFSET 3057Offset from the stack pointer register to the first location at which 3058outgoing arguments are placed. If not specified, the default value of 3059zero is used. This is the proper value for most machines. 3060 3061If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above 3062the first location at which outgoing arguments are placed. 3063@end defmac 3064 3065@defmac FIRST_PARM_OFFSET (@var{fundecl}) 3066Offset from the argument pointer register to the first argument's 3067address. On some machines it may depend on the data type of the 3068function. 3069 3070If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above 3071the first argument's address. 3072@end defmac 3073 3074@defmac STACK_DYNAMIC_OFFSET (@var{fundecl}) 3075Offset from the stack pointer register to an item dynamically allocated 3076on the stack, e.g., by @code{alloca}. 3077 3078The default value for this macro is @code{STACK_POINTER_OFFSET} plus the 3079length of the outgoing arguments. The default is correct for most 3080machines. See @file{function.c} for details. 3081@end defmac 3082 3083@defmac INITIAL_FRAME_ADDRESS_RTX 3084A C expression whose value is RTL representing the address of the initial 3085stack frame. This address is passed to @code{RETURN_ADDR_RTX} and 3086@code{DYNAMIC_CHAIN_ADDRESS}. If you don't define this macro, a reasonable 3087default value will be used. Define this macro in order to make frame pointer 3088elimination work in the presence of @code{__builtin_frame_address (count)} and 3089@code{__builtin_return_address (count)} for @code{count} not equal to zero. 3090@end defmac 3091 3092@defmac DYNAMIC_CHAIN_ADDRESS (@var{frameaddr}) 3093A C expression whose value is RTL representing the address in a stack 3094frame where the pointer to the caller's frame is stored. Assume that 3095@var{frameaddr} is an RTL expression for the address of the stack frame 3096itself. 3097 3098If you don't define this macro, the default is to return the value 3099of @var{frameaddr}---that is, the stack frame address is also the 3100address of the stack word that points to the previous frame. 3101@end defmac 3102 3103@defmac SETUP_FRAME_ADDRESSES 3104A C expression that produces the machine-specific code to 3105setup the stack so that arbitrary frames can be accessed. For example, 3106on the SPARC, we must flush all of the register windows to the stack 3107before we can access arbitrary stack frames. You will seldom need to 3108define this macro. The default is to do nothing. 3109@end defmac 3110 3111@deftypefn {Target Hook} rtx TARGET_BUILTIN_SETJMP_FRAME_VALUE (void) 3112This target hook should return an rtx that is used to store 3113the address of the current frame into the built in @code{setjmp} buffer. 3114The default value, @code{virtual_stack_vars_rtx}, is correct for most 3115machines. One reason you may need to define this target hook is if 3116@code{hard_frame_pointer_rtx} is the appropriate value on your machine. 3117@end deftypefn 3118 3119@defmac FRAME_ADDR_RTX (@var{frameaddr}) 3120A C expression whose value is RTL representing the value of the frame 3121address for the current frame. @var{frameaddr} is the frame pointer 3122of the current frame. This is used for __builtin_frame_address. 3123You need only define this macro if the frame address is not the same 3124as the frame pointer. Most machines do not need to define it. 3125@end defmac 3126 3127@defmac RETURN_ADDR_RTX (@var{count}, @var{frameaddr}) 3128A C expression whose value is RTL representing the value of the return 3129address for the frame @var{count} steps up from the current frame, after 3130the prologue. @var{frameaddr} is the frame pointer of the @var{count} 3131frame, or the frame pointer of the @var{count} @minus{} 1 frame if 3132@code{RETURN_ADDR_IN_PREVIOUS_FRAME} is nonzero. 3133 3134The value of the expression must always be the correct address when 3135@var{count} is zero, but may be @code{NULL_RTX} if there is no way to 3136determine the return address of other frames. 3137@end defmac 3138 3139@defmac RETURN_ADDR_IN_PREVIOUS_FRAME 3140Define this macro to nonzero value if the return address of a particular 3141stack frame is accessed from the frame pointer of the previous stack 3142frame. The zero default for this macro is suitable for most ports. 3143@end defmac 3144 3145@defmac INCOMING_RETURN_ADDR_RTX 3146A C expression whose value is RTL representing the location of the 3147incoming return address at the beginning of any function, before the 3148prologue. This RTL is either a @code{REG}, indicating that the return 3149value is saved in @samp{REG}, or a @code{MEM} representing a location in 3150the stack. 3151 3152You only need to define this macro if you want to support call frame 3153debugging information like that provided by DWARF 2. 3154 3155If this RTL is a @code{REG}, you should also define 3156@code{DWARF_FRAME_RETURN_COLUMN} to @code{DWARF_FRAME_REGNUM (REGNO)}. 3157@end defmac 3158 3159@defmac DWARF_ALT_FRAME_RETURN_COLUMN 3160A C expression whose value is an integer giving a DWARF 2 column 3161number that may be used as an alternative return column. The column 3162must not correspond to any gcc hard register (that is, it must not 3163be in the range of @code{DWARF_FRAME_REGNUM}). 3164 3165This macro can be useful if @code{DWARF_FRAME_RETURN_COLUMN} is set to a 3166general register, but an alternative column needs to be used for signal 3167frames. Some targets have also used different frame return columns 3168over time. 3169@end defmac 3170 3171@defmac DWARF_ZERO_REG 3172A C expression whose value is an integer giving a DWARF 2 register 3173number that is considered to always have the value zero. This should 3174only be defined if the target has an architected zero register, and 3175someone decided it was a good idea to use that register number to 3176terminate the stack backtrace. New ports should avoid this. 3177@end defmac 3178 3179@deftypefn {Target Hook} void TARGET_DWARF_HANDLE_FRAME_UNSPEC (const char *@var{label}, rtx @var{pattern}, int @var{index}) 3180This target hook allows the backend to emit frame-related insns that 3181contain UNSPECs or UNSPEC_VOLATILEs. The DWARF 2 call frame debugging 3182info engine will invoke it on insns of the form 3183@smallexample 3184(set (reg) (unspec [@dots{}] UNSPEC_INDEX)) 3185@end smallexample 3186and 3187@smallexample 3188(set (reg) (unspec_volatile [@dots{}] UNSPECV_INDEX)). 3189@end smallexample 3190to let the backend emit the call frame instructions. @var{label} is 3191the CFI label attached to the insn, @var{pattern} is the pattern of 3192the insn and @var{index} is @code{UNSPEC_INDEX} or @code{UNSPECV_INDEX}. 3193@end deftypefn 3194 3195@deftypefn {Target Hook} {unsigned int} TARGET_DWARF_POLY_INDETERMINATE_VALUE (unsigned int @var{i}, unsigned int *@var{factor}, int *@var{offset}) 3196Express the value of @code{poly_int} indeterminate @var{i} as a DWARF 3197expression, with @var{i} counting from 1. Return the number of a DWARF 3198register @var{R} and set @samp{*@var{factor}} and @samp{*@var{offset}} such 3199that the value of the indeterminate is: 3200@smallexample 3201value_of(@var{R}) / @var{factor} - @var{offset} 3202@end smallexample 3203 3204A target only needs to define this hook if it sets 3205@samp{NUM_POLY_INT_COEFFS} to a value greater than 1. 3206@end deftypefn 3207 3208@defmac INCOMING_FRAME_SP_OFFSET 3209A C expression whose value is an integer giving the offset, in bytes, 3210from the value of the stack pointer register to the top of the stack 3211frame at the beginning of any function, before the prologue. The top of 3212the frame is defined to be the value of the stack pointer in the 3213previous frame, just before the call instruction. 3214 3215You only need to define this macro if you want to support call frame 3216debugging information like that provided by DWARF 2. 3217@end defmac 3218 3219@defmac DEFAULT_INCOMING_FRAME_SP_OFFSET 3220Like @code{INCOMING_FRAME_SP_OFFSET}, but must be the same for all 3221functions of the same ABI, and when using GAS @code{.cfi_*} directives 3222must also agree with the default CFI GAS emits. Define this macro 3223only if @code{INCOMING_FRAME_SP_OFFSET} can have different values 3224between different functions of the same ABI or when 3225@code{INCOMING_FRAME_SP_OFFSET} does not agree with GAS default CFI. 3226@end defmac 3227 3228@defmac ARG_POINTER_CFA_OFFSET (@var{fundecl}) 3229A C expression whose value is an integer giving the offset, in bytes, 3230from the argument pointer to the canonical frame address (cfa). The 3231final value should coincide with that calculated by 3232@code{INCOMING_FRAME_SP_OFFSET}. Which is unfortunately not usable 3233during virtual register instantiation. 3234 3235The default value for this macro is 3236@code{FIRST_PARM_OFFSET (fundecl) + crtl->args.pretend_args_size}, 3237which is correct for most machines; in general, the arguments are found 3238immediately before the stack frame. Note that this is not the case on 3239some targets that save registers into the caller's frame, such as SPARC 3240and rs6000, and so such targets need to define this macro. 3241 3242You only need to define this macro if the default is incorrect, and you 3243want to support call frame debugging information like that provided by 3244DWARF 2. 3245@end defmac 3246 3247@defmac FRAME_POINTER_CFA_OFFSET (@var{fundecl}) 3248If defined, a C expression whose value is an integer giving the offset 3249in bytes from the frame pointer to the canonical frame address (cfa). 3250The final value should coincide with that calculated by 3251@code{INCOMING_FRAME_SP_OFFSET}. 3252 3253Normally the CFA is calculated as an offset from the argument pointer, 3254via @code{ARG_POINTER_CFA_OFFSET}, but if the argument pointer is 3255variable due to the ABI, this may not be possible. If this macro is 3256defined, it implies that the virtual register instantiation should be 3257based on the frame pointer instead of the argument pointer. Only one 3258of @code{FRAME_POINTER_CFA_OFFSET} and @code{ARG_POINTER_CFA_OFFSET} 3259should be defined. 3260@end defmac 3261 3262@defmac CFA_FRAME_BASE_OFFSET (@var{fundecl}) 3263If defined, a C expression whose value is an integer giving the offset 3264in bytes from the canonical frame address (cfa) to the frame base used 3265in DWARF 2 debug information. The default is zero. A different value 3266may reduce the size of debug information on some ports. 3267@end defmac 3268 3269@node Exception Handling 3270@subsection Exception Handling Support 3271@cindex exception handling 3272 3273@defmac EH_RETURN_DATA_REGNO (@var{N}) 3274A C expression whose value is the @var{N}th register number used for 3275data by exception handlers, or @code{INVALID_REGNUM} if fewer than 3276@var{N} registers are usable. 3277 3278The exception handling library routines communicate with the exception 3279handlers via a set of agreed upon registers. Ideally these registers 3280should be call-clobbered; it is possible to use call-saved registers, 3281but may negatively impact code size. The target must support at least 32822 data registers, but should define 4 if there are enough free registers. 3283 3284You must define this macro if you want to support call frame exception 3285handling like that provided by DWARF 2. 3286@end defmac 3287 3288@defmac EH_RETURN_STACKADJ_RTX 3289A C expression whose value is RTL representing a location in which 3290to store a stack adjustment to be applied before function return. 3291This is used to unwind the stack to an exception handler's call frame. 3292It will be assigned zero on code paths that return normally. 3293 3294Typically this is a call-clobbered hard register that is otherwise 3295untouched by the epilogue, but could also be a stack slot. 3296 3297Do not define this macro if the stack pointer is saved and restored 3298by the regular prolog and epilog code in the call frame itself; in 3299this case, the exception handling library routines will update the 3300stack location to be restored in place. Otherwise, you must define 3301this macro if you want to support call frame exception handling like 3302that provided by DWARF 2. 3303@end defmac 3304 3305@defmac EH_RETURN_HANDLER_RTX 3306A C expression whose value is RTL representing a location in which 3307to store the address of an exception handler to which we should 3308return. It will not be assigned on code paths that return normally. 3309 3310Typically this is the location in the call frame at which the normal 3311return address is stored. For targets that return by popping an 3312address off the stack, this might be a memory address just below 3313the @emph{target} call frame rather than inside the current call 3314frame. If defined, @code{EH_RETURN_STACKADJ_RTX} will have already 3315been assigned, so it may be used to calculate the location of the 3316target call frame. 3317 3318Some targets have more complex requirements than storing to an 3319address calculable during initial code generation. In that case 3320the @code{eh_return} instruction pattern should be used instead. 3321 3322If you want to support call frame exception handling, you must 3323define either this macro or the @code{eh_return} instruction pattern. 3324@end defmac 3325 3326@defmac RETURN_ADDR_OFFSET 3327If defined, an integer-valued C expression for which rtl will be generated 3328to add it to the exception handler address before it is searched in the 3329exception handling tables, and to subtract it again from the address before 3330using it to return to the exception handler. 3331@end defmac 3332 3333@defmac ASM_PREFERRED_EH_DATA_FORMAT (@var{code}, @var{global}) 3334This macro chooses the encoding of pointers embedded in the exception 3335handling sections. If at all possible, this should be defined such 3336that the exception handling section will not require dynamic relocations, 3337and so may be read-only. 3338 3339@var{code} is 0 for data, 1 for code labels, 2 for function pointers. 3340@var{global} is true if the symbol may be affected by dynamic relocations. 3341The macro should return a combination of the @code{DW_EH_PE_*} defines 3342as found in @file{dwarf2.h}. 3343 3344If this macro is not defined, pointers will not be encoded but 3345represented directly. 3346@end defmac 3347 3348@defmac ASM_MAYBE_OUTPUT_ENCODED_ADDR_RTX (@var{file}, @var{encoding}, @var{size}, @var{addr}, @var{done}) 3349This macro allows the target to emit whatever special magic is required 3350to represent the encoding chosen by @code{ASM_PREFERRED_EH_DATA_FORMAT}. 3351Generic code takes care of pc-relative and indirect encodings; this must 3352be defined if the target uses text-relative or data-relative encodings. 3353 3354This is a C statement that branches to @var{done} if the format was 3355handled. @var{encoding} is the format chosen, @var{size} is the number 3356of bytes that the format occupies, @var{addr} is the @code{SYMBOL_REF} 3357to be emitted. 3358@end defmac 3359 3360@defmac MD_FALLBACK_FRAME_STATE_FOR (@var{context}, @var{fs}) 3361This macro allows the target to add CPU and operating system specific 3362code to the call-frame unwinder for use when there is no unwind data 3363available. The most common reason to implement this macro is to unwind 3364through signal frames. 3365 3366This macro is called from @code{uw_frame_state_for} in 3367@file{unwind-dw2.c}, @file{unwind-dw2-xtensa.c} and 3368@file{unwind-ia64.c}. @var{context} is an @code{_Unwind_Context}; 3369@var{fs} is an @code{_Unwind_FrameState}. Examine @code{context->ra} 3370for the address of the code being executed and @code{context->cfa} for 3371the stack pointer value. If the frame can be decoded, the register 3372save addresses should be updated in @var{fs} and the macro should 3373evaluate to @code{_URC_NO_REASON}. If the frame cannot be decoded, 3374the macro should evaluate to @code{_URC_END_OF_STACK}. 3375 3376For proper signal handling in Java this macro is accompanied by 3377@code{MAKE_THROW_FRAME}, defined in @file{libjava/include/*-signal.h} headers. 3378@end defmac 3379 3380@defmac MD_HANDLE_UNWABI (@var{context}, @var{fs}) 3381This macro allows the target to add operating system specific code to the 3382call-frame unwinder to handle the IA-64 @code{.unwabi} unwinding directive, 3383usually used for signal or interrupt frames. 3384 3385This macro is called from @code{uw_update_context} in libgcc's 3386@file{unwind-ia64.c}. @var{context} is an @code{_Unwind_Context}; 3387@var{fs} is an @code{_Unwind_FrameState}. Examine @code{fs->unwabi} 3388for the abi and context in the @code{.unwabi} directive. If the 3389@code{.unwabi} directive can be handled, the register save addresses should 3390be updated in @var{fs}. 3391@end defmac 3392 3393@defmac TARGET_USES_WEAK_UNWIND_INFO 3394A C expression that evaluates to true if the target requires unwind 3395info to be given comdat linkage. Define it to be @code{1} if comdat 3396linkage is necessary. The default is @code{0}. 3397@end defmac 3398 3399@node Stack Checking 3400@subsection Specifying How Stack Checking is Done 3401 3402GCC will check that stack references are within the boundaries of the 3403stack, if the option @option{-fstack-check} is specified, in one of 3404three ways: 3405 3406@enumerate 3407@item 3408If the value of the @code{STACK_CHECK_BUILTIN} macro is nonzero, GCC 3409will assume that you have arranged for full stack checking to be done 3410at appropriate places in the configuration files. GCC will not do 3411other special processing. 3412 3413@item 3414If @code{STACK_CHECK_BUILTIN} is zero and the value of the 3415@code{STACK_CHECK_STATIC_BUILTIN} macro is nonzero, GCC will assume 3416that you have arranged for static stack checking (checking of the 3417static stack frame of functions) to be done at appropriate places 3418in the configuration files. GCC will only emit code to do dynamic 3419stack checking (checking on dynamic stack allocations) using the third 3420approach below. 3421 3422@item 3423If neither of the above are true, GCC will generate code to periodically 3424``probe'' the stack pointer using the values of the macros defined below. 3425@end enumerate 3426 3427If neither STACK_CHECK_BUILTIN nor STACK_CHECK_STATIC_BUILTIN is defined, 3428GCC will change its allocation strategy for large objects if the option 3429@option{-fstack-check} is specified: they will always be allocated 3430dynamically if their size exceeds @code{STACK_CHECK_MAX_VAR_SIZE} bytes. 3431 3432@defmac STACK_CHECK_BUILTIN 3433A nonzero value if stack checking is done by the configuration files in a 3434machine-dependent manner. You should define this macro if stack checking 3435is required by the ABI of your machine or if you would like to do stack 3436checking in some more efficient way than the generic approach. The default 3437value of this macro is zero. 3438@end defmac 3439 3440@defmac STACK_CHECK_STATIC_BUILTIN 3441A nonzero value if static stack checking is done by the configuration files 3442in a machine-dependent manner. You should define this macro if you would 3443like to do static stack checking in some more efficient way than the generic 3444approach. The default value of this macro is zero. 3445@end defmac 3446 3447@defmac STACK_CHECK_PROBE_INTERVAL_EXP 3448An integer specifying the interval at which GCC must generate stack probe 3449instructions, defined as 2 raised to this integer. You will normally 3450define this macro so that the interval be no larger than the size of 3451the ``guard pages'' at the end of a stack area. The default value 3452of 12 (4096-byte interval) is suitable for most systems. 3453@end defmac 3454 3455@defmac STACK_CHECK_MOVING_SP 3456An integer which is nonzero if GCC should move the stack pointer page by page 3457when doing probes. This can be necessary on systems where the stack pointer 3458contains the bottom address of the memory area accessible to the executing 3459thread at any point in time. In this situation an alternate signal stack 3460is required in order to be able to recover from a stack overflow. The 3461default value of this macro is zero. 3462@end defmac 3463 3464@defmac STACK_CHECK_PROTECT 3465The number of bytes of stack needed to recover from a stack overflow, for 3466languages where such a recovery is supported. The default value of 4KB/8KB 3467with the @code{setjmp}/@code{longjmp}-based exception handling mechanism and 34688KB/12KB with other exception handling mechanisms should be adequate for most 3469architectures and operating systems. 3470@end defmac 3471 3472The following macros are relevant only if neither STACK_CHECK_BUILTIN 3473nor STACK_CHECK_STATIC_BUILTIN is defined; you can omit them altogether 3474in the opposite case. 3475 3476@defmac STACK_CHECK_MAX_FRAME_SIZE 3477The maximum size of a stack frame, in bytes. GCC will generate probe 3478instructions in non-leaf functions to ensure at least this many bytes of 3479stack are available. If a stack frame is larger than this size, stack 3480checking will not be reliable and GCC will issue a warning. The 3481default is chosen so that GCC only generates one instruction on most 3482systems. You should normally not change the default value of this macro. 3483@end defmac 3484 3485@defmac STACK_CHECK_FIXED_FRAME_SIZE 3486GCC uses this value to generate the above warning message. It 3487represents the amount of fixed frame used by a function, not including 3488space for any callee-saved registers, temporaries and user variables. 3489You need only specify an upper bound for this amount and will normally 3490use the default of four words. 3491@end defmac 3492 3493@defmac STACK_CHECK_MAX_VAR_SIZE 3494The maximum size, in bytes, of an object that GCC will place in the 3495fixed area of the stack frame when the user specifies 3496@option{-fstack-check}. 3497GCC computed the default from the values of the above macros and you will 3498normally not need to override that default. 3499@end defmac 3500 3501@deftypefn {Target Hook} HOST_WIDE_INT TARGET_STACK_CLASH_PROTECTION_ALLOCA_PROBE_RANGE (void) 3502Some targets have an ABI defined interval for which no probing needs to be done. 3503When a probe does need to be done this same interval is used as the probe distance up when doing stack clash protection for alloca. 3504On such targets this value can be set to override the default probing up interval. 3505Define this variable to return nonzero if such a probe range is required or zero otherwise. Defining this hook also requires your functions which make use of alloca to have at least 8 byesof outgoing arguments. If this is not the case the stack will be corrupted. 3506You need not define this macro if it would always have the value zero. 3507@end deftypefn 3508 3509@need 2000 3510@node Frame Registers 3511@subsection Registers That Address the Stack Frame 3512 3513@c prevent bad page break with this line 3514This discusses registers that address the stack frame. 3515 3516@defmac STACK_POINTER_REGNUM 3517The register number of the stack pointer register, which must also be a 3518fixed register according to @code{FIXED_REGISTERS}. On most machines, 3519the hardware determines which register this is. 3520@end defmac 3521 3522@defmac FRAME_POINTER_REGNUM 3523The register number of the frame pointer register, which is used to 3524access automatic variables in the stack frame. On some machines, the 3525hardware determines which register this is. On other machines, you can 3526choose any register you wish for this purpose. 3527@end defmac 3528 3529@defmac HARD_FRAME_POINTER_REGNUM 3530On some machines the offset between the frame pointer and starting 3531offset of the automatic variables is not known until after register 3532allocation has been done (for example, because the saved registers are 3533between these two locations). On those machines, define 3534@code{FRAME_POINTER_REGNUM} the number of a special, fixed register to 3535be used internally until the offset is known, and define 3536@code{HARD_FRAME_POINTER_REGNUM} to be the actual hard register number 3537used for the frame pointer. 3538 3539You should define this macro only in the very rare circumstances when it 3540is not possible to calculate the offset between the frame pointer and 3541the automatic variables until after register allocation has been 3542completed. When this macro is defined, you must also indicate in your 3543definition of @code{ELIMINABLE_REGS} how to eliminate 3544@code{FRAME_POINTER_REGNUM} into either @code{HARD_FRAME_POINTER_REGNUM} 3545or @code{STACK_POINTER_REGNUM}. 3546 3547Do not define this macro if it would be the same as 3548@code{FRAME_POINTER_REGNUM}. 3549@end defmac 3550 3551@defmac ARG_POINTER_REGNUM 3552The register number of the arg pointer register, which is used to access 3553the function's argument list. On some machines, this is the same as the 3554frame pointer register. On some machines, the hardware determines which 3555register this is. On other machines, you can choose any register you 3556wish for this purpose. If this is not the same register as the frame 3557pointer register, then you must mark it as a fixed register according to 3558@code{FIXED_REGISTERS}, or arrange to be able to eliminate it 3559(@pxref{Elimination}). 3560@end defmac 3561 3562@defmac HARD_FRAME_POINTER_IS_FRAME_POINTER 3563Define this to a preprocessor constant that is nonzero if 3564@code{hard_frame_pointer_rtx} and @code{frame_pointer_rtx} should be 3565the same. The default definition is @samp{(HARD_FRAME_POINTER_REGNUM 3566== FRAME_POINTER_REGNUM)}; you only need to define this macro if that 3567definition is not suitable for use in preprocessor conditionals. 3568@end defmac 3569 3570@defmac HARD_FRAME_POINTER_IS_ARG_POINTER 3571Define this to a preprocessor constant that is nonzero if 3572@code{hard_frame_pointer_rtx} and @code{arg_pointer_rtx} should be the 3573same. The default definition is @samp{(HARD_FRAME_POINTER_REGNUM == 3574ARG_POINTER_REGNUM)}; you only need to define this macro if that 3575definition is not suitable for use in preprocessor conditionals. 3576@end defmac 3577 3578@defmac RETURN_ADDRESS_POINTER_REGNUM 3579The register number of the return address pointer register, which is used to 3580access the current function's return address from the stack. On some 3581machines, the return address is not at a fixed offset from the frame 3582pointer or stack pointer or argument pointer. This register can be defined 3583to point to the return address on the stack, and then be converted by 3584@code{ELIMINABLE_REGS} into either the frame pointer or stack pointer. 3585 3586Do not define this macro unless there is no other way to get the return 3587address from the stack. 3588@end defmac 3589 3590@defmac STATIC_CHAIN_REGNUM 3591@defmacx STATIC_CHAIN_INCOMING_REGNUM 3592Register numbers used for passing a function's static chain pointer. If 3593register windows are used, the register number as seen by the called 3594function is @code{STATIC_CHAIN_INCOMING_REGNUM}, while the register 3595number as seen by the calling function is @code{STATIC_CHAIN_REGNUM}. If 3596these registers are the same, @code{STATIC_CHAIN_INCOMING_REGNUM} need 3597not be defined. 3598 3599The static chain register need not be a fixed register. 3600 3601If the static chain is passed in memory, these macros should not be 3602defined; instead, the @code{TARGET_STATIC_CHAIN} hook should be used. 3603@end defmac 3604 3605@deftypefn {Target Hook} rtx TARGET_STATIC_CHAIN (const_tree @var{fndecl_or_type}, bool @var{incoming_p}) 3606This hook replaces the use of @code{STATIC_CHAIN_REGNUM} et al for 3607targets that may use different static chain locations for different 3608nested functions. This may be required if the target has function 3609attributes that affect the calling conventions of the function and 3610those calling conventions use different static chain locations. 3611 3612The default version of this hook uses @code{STATIC_CHAIN_REGNUM} et al. 3613 3614If the static chain is passed in memory, this hook should be used to 3615provide rtx giving @code{mem} expressions that denote where they are stored. 3616Often the @code{mem} expression as seen by the caller will be at an offset 3617from the stack pointer and the @code{mem} expression as seen by the callee 3618will be at an offset from the frame pointer. 3619@findex stack_pointer_rtx 3620@findex frame_pointer_rtx 3621@findex arg_pointer_rtx 3622The variables @code{stack_pointer_rtx}, @code{frame_pointer_rtx}, and 3623@code{arg_pointer_rtx} will have been initialized and should be used 3624to refer to those items. 3625@end deftypefn 3626 3627@defmac DWARF_FRAME_REGISTERS 3628This macro specifies the maximum number of hard registers that can be 3629saved in a call frame. This is used to size data structures used in 3630DWARF2 exception handling. 3631 3632Prior to GCC 3.0, this macro was needed in order to establish a stable 3633exception handling ABI in the face of adding new hard registers for ISA 3634extensions. In GCC 3.0 and later, the EH ABI is insulated from changes 3635in the number of hard registers. Nevertheless, this macro can still be 3636used to reduce the runtime memory requirements of the exception handling 3637routines, which can be substantial if the ISA contains a lot of 3638registers that are not call-saved. 3639 3640If this macro is not defined, it defaults to 3641@code{FIRST_PSEUDO_REGISTER}. 3642@end defmac 3643 3644@defmac PRE_GCC3_DWARF_FRAME_REGISTERS 3645 3646This macro is similar to @code{DWARF_FRAME_REGISTERS}, but is provided 3647for backward compatibility in pre GCC 3.0 compiled code. 3648 3649If this macro is not defined, it defaults to 3650@code{DWARF_FRAME_REGISTERS}. 3651@end defmac 3652 3653@defmac DWARF_REG_TO_UNWIND_COLUMN (@var{regno}) 3654 3655Define this macro if the target's representation for dwarf registers 3656is different than the internal representation for unwind column. 3657Given a dwarf register, this macro should return the internal unwind 3658column number to use instead. 3659@end defmac 3660 3661@defmac DWARF_FRAME_REGNUM (@var{regno}) 3662 3663Define this macro if the target's representation for dwarf registers 3664used in .eh_frame or .debug_frame is different from that used in other 3665debug info sections. Given a GCC hard register number, this macro 3666should return the .eh_frame register number. The default is 3667@code{DBX_REGISTER_NUMBER (@var{regno})}. 3668 3669@end defmac 3670 3671@defmac DWARF2_FRAME_REG_OUT (@var{regno}, @var{for_eh}) 3672 3673Define this macro to map register numbers held in the call frame info 3674that GCC has collected using @code{DWARF_FRAME_REGNUM} to those that 3675should be output in .debug_frame (@code{@var{for_eh}} is zero) and 3676.eh_frame (@code{@var{for_eh}} is nonzero). The default is to 3677return @code{@var{regno}}. 3678 3679@end defmac 3680 3681@defmac REG_VALUE_IN_UNWIND_CONTEXT 3682 3683Define this macro if the target stores register values as 3684@code{_Unwind_Word} type in unwind context. It should be defined if 3685target register size is larger than the size of @code{void *}. The 3686default is to store register values as @code{void *} type. 3687 3688@end defmac 3689 3690@defmac ASSUME_EXTENDED_UNWIND_CONTEXT 3691 3692Define this macro to be 1 if the target always uses extended unwind 3693context with version, args_size and by_value fields. If it is undefined, 3694it will be defined to 1 when @code{REG_VALUE_IN_UNWIND_CONTEXT} is 3695defined and 0 otherwise. 3696 3697@end defmac 3698 3699@defmac DWARF_LAZY_REGISTER_VALUE (@var{regno}, @var{value}) 3700Define this macro if the target has pseudo DWARF registers whose 3701values need to be computed lazily on demand by the unwinder (such as when 3702referenced in a CFA expression). The macro returns true if @var{regno} 3703is such a register and stores its value in @samp{*@var{value}} if so. 3704@end defmac 3705 3706@node Elimination 3707@subsection Eliminating Frame Pointer and Arg Pointer 3708 3709@c prevent bad page break with this line 3710This is about eliminating the frame pointer and arg pointer. 3711 3712@deftypefn {Target Hook} bool TARGET_FRAME_POINTER_REQUIRED (void) 3713This target hook should return @code{true} if a function must have and use 3714a frame pointer. This target hook is called in the reload pass. If its return 3715value is @code{true} the function will have a frame pointer. 3716 3717This target hook can in principle examine the current function and decide 3718according to the facts, but on most machines the constant @code{false} or the 3719constant @code{true} suffices. Use @code{false} when the machine allows code 3720to be generated with no frame pointer, and doing so saves some time or space. 3721Use @code{true} when there is no possible advantage to avoiding a frame 3722pointer. 3723 3724In certain cases, the compiler does not know how to produce valid code 3725without a frame pointer. The compiler recognizes those cases and 3726automatically gives the function a frame pointer regardless of what 3727@code{targetm.frame_pointer_required} returns. You don't need to worry about 3728them. 3729 3730In a function that does not require a frame pointer, the frame pointer 3731register can be allocated for ordinary usage, unless you mark it as a 3732fixed register. See @code{FIXED_REGISTERS} for more information. 3733 3734Default return value is @code{false}. 3735@end deftypefn 3736 3737@defmac ELIMINABLE_REGS 3738This macro specifies a table of register pairs used to eliminate 3739unneeded registers that point into the stack frame. 3740 3741The definition of this macro is a list of structure initializations, each 3742of which specifies an original and replacement register. 3743 3744On some machines, the position of the argument pointer is not known until 3745the compilation is completed. In such a case, a separate hard register 3746must be used for the argument pointer. This register can be eliminated by 3747replacing it with either the frame pointer or the argument pointer, 3748depending on whether or not the frame pointer has been eliminated. 3749 3750In this case, you might specify: 3751@smallexample 3752#define ELIMINABLE_REGS \ 3753@{@{ARG_POINTER_REGNUM, STACK_POINTER_REGNUM@}, \ 3754 @{ARG_POINTER_REGNUM, FRAME_POINTER_REGNUM@}, \ 3755 @{FRAME_POINTER_REGNUM, STACK_POINTER_REGNUM@}@} 3756@end smallexample 3757 3758Note that the elimination of the argument pointer with the stack pointer is 3759specified first since that is the preferred elimination. 3760@end defmac 3761 3762@deftypefn {Target Hook} bool TARGET_CAN_ELIMINATE (const int @var{from_reg}, const int @var{to_reg}) 3763This target hook should return @code{true} if the compiler is allowed to 3764try to replace register number @var{from_reg} with register number 3765@var{to_reg}. This target hook will usually be @code{true}, since most of the 3766cases preventing register elimination are things that the compiler already 3767knows about. 3768 3769Default return value is @code{true}. 3770@end deftypefn 3771 3772@defmac INITIAL_ELIMINATION_OFFSET (@var{from-reg}, @var{to-reg}, @var{offset-var}) 3773This macro returns the initial difference between the specified pair 3774of registers. The value would be computed from information 3775such as the result of @code{get_frame_size ()} and the tables of 3776registers @code{df_regs_ever_live_p} and @code{call_used_regs}. 3777@end defmac 3778 3779@deftypefn {Target Hook} void TARGET_COMPUTE_FRAME_LAYOUT (void) 3780This target hook is called once each time the frame layout needs to be 3781recalculated. The calculations can be cached by the target and can then 3782be used by @code{INITIAL_ELIMINATION_OFFSET} instead of re-computing the 3783layout on every invocation of that hook. This is particularly useful 3784for targets that have an expensive frame layout function. Implementing 3785this callback is optional. 3786@end deftypefn 3787 3788@node Stack Arguments 3789@subsection Passing Function Arguments on the Stack 3790@cindex arguments on stack 3791@cindex stack arguments 3792 3793The macros in this section control how arguments are passed 3794on the stack. See the following section for other macros that 3795control passing certain arguments in registers. 3796 3797@deftypefn {Target Hook} bool TARGET_PROMOTE_PROTOTYPES (const_tree @var{fntype}) 3798This target hook returns @code{true} if an argument declared in a 3799prototype as an integral type smaller than @code{int} should actually be 3800passed as an @code{int}. In addition to avoiding errors in certain 3801cases of mismatch, it also makes for better code on certain machines. 3802The default is to not promote prototypes. 3803@end deftypefn 3804 3805@defmac PUSH_ARGS 3806A C expression. If nonzero, push insns will be used to pass 3807outgoing arguments. 3808If the target machine does not have a push instruction, set it to zero. 3809That directs GCC to use an alternate strategy: to 3810allocate the entire argument block and then store the arguments into 3811it. When @code{PUSH_ARGS} is nonzero, @code{PUSH_ROUNDING} must be defined too. 3812@end defmac 3813 3814@defmac PUSH_ARGS_REVERSED 3815A C expression. If nonzero, function arguments will be evaluated from 3816last to first, rather than from first to last. If this macro is not 3817defined, it defaults to @code{PUSH_ARGS} on targets where the stack 3818and args grow in opposite directions, and 0 otherwise. 3819@end defmac 3820 3821@defmac PUSH_ROUNDING (@var{npushed}) 3822A C expression that is the number of bytes actually pushed onto the 3823stack when an instruction attempts to push @var{npushed} bytes. 3824 3825On some machines, the definition 3826 3827@smallexample 3828#define PUSH_ROUNDING(BYTES) (BYTES) 3829@end smallexample 3830 3831@noindent 3832will suffice. But on other machines, instructions that appear 3833to push one byte actually push two bytes in an attempt to maintain 3834alignment. Then the definition should be 3835 3836@smallexample 3837#define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1) 3838@end smallexample 3839 3840If the value of this macro has a type, it should be an unsigned type. 3841@end defmac 3842 3843@findex outgoing_args_size 3844@findex crtl->outgoing_args_size 3845@defmac ACCUMULATE_OUTGOING_ARGS 3846A C expression. If nonzero, the maximum amount of space required for outgoing arguments 3847will be computed and placed into 3848@code{crtl->outgoing_args_size}. No space will be pushed 3849onto the stack for each call; instead, the function prologue should 3850increase the stack frame size by this amount. 3851 3852Setting both @code{PUSH_ARGS} and @code{ACCUMULATE_OUTGOING_ARGS} 3853is not proper. 3854@end defmac 3855 3856@defmac REG_PARM_STACK_SPACE (@var{fndecl}) 3857Define this macro if functions should assume that stack space has been 3858allocated for arguments even when their values are passed in 3859registers. 3860 3861The value of this macro is the size, in bytes, of the area reserved for 3862arguments passed in registers for the function represented by @var{fndecl}, 3863which can be zero if GCC is calling a library function. 3864The argument @var{fndecl} can be the FUNCTION_DECL, or the type itself 3865of the function. 3866 3867This space can be allocated by the caller, or be a part of the 3868machine-dependent stack frame: @code{OUTGOING_REG_PARM_STACK_SPACE} says 3869which. 3870@end defmac 3871@c above is overfull. not sure what to do. --mew 5feb93 did 3872@c something, not sure if it looks good. --mew 10feb93 3873 3874@defmac INCOMING_REG_PARM_STACK_SPACE (@var{fndecl}) 3875Like @code{REG_PARM_STACK_SPACE}, but for incoming register arguments. 3876Define this macro if space guaranteed when compiling a function body 3877is different to space required when making a call, a situation that 3878can arise with K&R style function definitions. 3879@end defmac 3880 3881@defmac OUTGOING_REG_PARM_STACK_SPACE (@var{fntype}) 3882Define this to a nonzero value if it is the responsibility of the 3883caller to allocate the area reserved for arguments passed in registers 3884when calling a function of @var{fntype}. @var{fntype} may be NULL 3885if the function called is a library function. 3886 3887If @code{ACCUMULATE_OUTGOING_ARGS} is defined, this macro controls 3888whether the space for these arguments counts in the value of 3889@code{crtl->outgoing_args_size}. 3890@end defmac 3891 3892@defmac STACK_PARMS_IN_REG_PARM_AREA 3893Define this macro if @code{REG_PARM_STACK_SPACE} is defined, but the 3894stack parameters don't skip the area specified by it. 3895@c i changed this, makes more sens and it should have taken care of the 3896@c overfull.. not as specific, tho. --mew 5feb93 3897 3898Normally, when a parameter is not passed in registers, it is placed on the 3899stack beyond the @code{REG_PARM_STACK_SPACE} area. Defining this macro 3900suppresses this behavior and causes the parameter to be passed on the 3901stack in its natural location. 3902@end defmac 3903 3904@deftypefn {Target Hook} poly_int64 TARGET_RETURN_POPS_ARGS (tree @var{fundecl}, tree @var{funtype}, poly_int64 @var{size}) 3905This target hook returns the number of bytes of its own arguments that 3906a function pops on returning, or 0 if the function pops no arguments 3907and the caller must therefore pop them all after the function returns. 3908 3909@var{fundecl} is a C variable whose value is a tree node that describes 3910the function in question. Normally it is a node of type 3911@code{FUNCTION_DECL} that describes the declaration of the function. 3912From this you can obtain the @code{DECL_ATTRIBUTES} of the function. 3913 3914@var{funtype} is a C variable whose value is a tree node that 3915describes the function in question. Normally it is a node of type 3916@code{FUNCTION_TYPE} that describes the data type of the function. 3917From this it is possible to obtain the data types of the value and 3918arguments (if known). 3919 3920When a call to a library function is being considered, @var{fundecl} 3921will contain an identifier node for the library function. Thus, if 3922you need to distinguish among various library functions, you can do so 3923by their names. Note that ``library function'' in this context means 3924a function used to perform arithmetic, whose name is known specially 3925in the compiler and was not mentioned in the C code being compiled. 3926 3927@var{size} is the number of bytes of arguments passed on the 3928stack. If a variable number of bytes is passed, it is zero, and 3929argument popping will always be the responsibility of the calling function. 3930 3931On the VAX, all functions always pop their arguments, so the definition 3932of this macro is @var{size}. On the 68000, using the standard 3933calling convention, no functions pop their arguments, so the value of 3934the macro is always 0 in this case. But an alternative calling 3935convention is available in which functions that take a fixed number of 3936arguments pop them but other functions (such as @code{printf}) pop 3937nothing (the caller pops all). When this convention is in use, 3938@var{funtype} is examined to determine whether a function takes a fixed 3939number of arguments. 3940@end deftypefn 3941 3942@defmac CALL_POPS_ARGS (@var{cum}) 3943A C expression that should indicate the number of bytes a call sequence 3944pops off the stack. It is added to the value of @code{RETURN_POPS_ARGS} 3945when compiling a function call. 3946 3947@var{cum} is the variable in which all arguments to the called function 3948have been accumulated. 3949 3950On certain architectures, such as the SH5, a call trampoline is used 3951that pops certain registers off the stack, depending on the arguments 3952that have been passed to the function. Since this is a property of the 3953call site, not of the called function, @code{RETURN_POPS_ARGS} is not 3954appropriate. 3955@end defmac 3956 3957@node Register Arguments 3958@subsection Passing Arguments in Registers 3959@cindex arguments in registers 3960@cindex registers arguments 3961 3962This section describes the macros which let you control how various 3963types of arguments are passed in registers or how they are arranged in 3964the stack. 3965 3966@deftypefn {Target Hook} rtx TARGET_FUNCTION_ARG (cumulative_args_t @var{ca}, const function_arg_info @var{&arg}) 3967Return an RTX indicating whether function argument @var{arg} is passed 3968in a register and if so, which register. Argument @var{ca} summarizes all 3969the previous arguments. 3970 3971The return value is usually either a @code{reg} RTX for the hard 3972register in which to pass the argument, or zero to pass the argument 3973on the stack. 3974 3975The return value can be a @code{const_int} which means argument is 3976passed in a target specific slot with specified number. Target hooks 3977should be used to store or load argument in such case. See 3978@code{TARGET_STORE_BOUNDS_FOR_ARG} and @code{TARGET_LOAD_BOUNDS_FOR_ARG} 3979for more information. 3980 3981The value of the expression can also be a @code{parallel} RTX@. This is 3982used when an argument is passed in multiple locations. The mode of the 3983@code{parallel} should be the mode of the entire argument. The 3984@code{parallel} holds any number of @code{expr_list} pairs; each one 3985describes where part of the argument is passed. In each 3986@code{expr_list} the first operand must be a @code{reg} RTX for the hard 3987register in which to pass this part of the argument, and the mode of the 3988register RTX indicates how large this part of the argument is. The 3989second operand of the @code{expr_list} is a @code{const_int} which gives 3990the offset in bytes into the entire argument of where this part starts. 3991As a special exception the first @code{expr_list} in the @code{parallel} 3992RTX may have a first operand of zero. This indicates that the entire 3993argument is also stored on the stack. 3994 3995The last time this hook is called, it is called with @code{MODE == 3996VOIDmode}, and its result is passed to the @code{call} or @code{call_value} 3997pattern as operands 2 and 3 respectively. 3998 3999@cindex @file{stdarg.h} and register arguments 4000The usual way to make the ISO library @file{stdarg.h} work on a 4001machine where some arguments are usually passed in registers, is to 4002cause nameless arguments to be passed on the stack instead. This is 4003done by making @code{TARGET_FUNCTION_ARG} return 0 whenever 4004@var{named} is @code{false}. 4005 4006@cindex @code{TARGET_MUST_PASS_IN_STACK}, and @code{TARGET_FUNCTION_ARG} 4007@cindex @code{REG_PARM_STACK_SPACE}, and @code{TARGET_FUNCTION_ARG} 4008You may use the hook @code{targetm.calls.must_pass_in_stack} 4009in the definition of this macro to determine if this argument is of a 4010type that must be passed in the stack. If @code{REG_PARM_STACK_SPACE} 4011is not defined and @code{TARGET_FUNCTION_ARG} returns nonzero for such an 4012argument, the compiler will abort. If @code{REG_PARM_STACK_SPACE} is 4013defined, the argument will be computed in the stack and then loaded into 4014a register. 4015@end deftypefn 4016 4017@deftypefn {Target Hook} bool TARGET_MUST_PASS_IN_STACK (const function_arg_info @var{&arg}) 4018This target hook should return @code{true} if we should not pass @var{arg} 4019solely in registers. The file @file{expr.h} defines a 4020definition that is usually appropriate, refer to @file{expr.h} for additional 4021documentation. 4022@end deftypefn 4023 4024@deftypefn {Target Hook} rtx TARGET_FUNCTION_INCOMING_ARG (cumulative_args_t @var{ca}, const function_arg_info @var{&arg}) 4025Define this hook if the caller and callee on the target have different 4026views of where arguments are passed. Also define this hook if there are 4027functions that are never directly called, but are invoked by the hardware 4028and which have nonstandard calling conventions. 4029 4030In this case @code{TARGET_FUNCTION_ARG} computes the register in 4031which the caller passes the value, and 4032@code{TARGET_FUNCTION_INCOMING_ARG} should be defined in a similar 4033fashion to tell the function being called where the arguments will 4034arrive. 4035 4036@code{TARGET_FUNCTION_INCOMING_ARG} can also return arbitrary address 4037computation using hard register, which can be forced into a register, 4038so that it can be used to pass special arguments. 4039 4040If @code{TARGET_FUNCTION_INCOMING_ARG} is not defined, 4041@code{TARGET_FUNCTION_ARG} serves both purposes. 4042@end deftypefn 4043 4044@deftypefn {Target Hook} bool TARGET_USE_PSEUDO_PIC_REG (void) 4045This hook should return 1 in case pseudo register should be created 4046for pic_offset_table_rtx during function expand. 4047@end deftypefn 4048 4049@deftypefn {Target Hook} void TARGET_INIT_PIC_REG (void) 4050Perform a target dependent initialization of pic_offset_table_rtx. 4051This hook is called at the start of register allocation. 4052@end deftypefn 4053 4054@deftypefn {Target Hook} int TARGET_ARG_PARTIAL_BYTES (cumulative_args_t @var{cum}, const function_arg_info @var{&arg}) 4055This target hook returns the number of bytes at the beginning of an 4056argument that must be put in registers. The value must be zero for 4057arguments that are passed entirely in registers or that are entirely 4058pushed on the stack. 4059 4060On some machines, certain arguments must be passed partially in 4061registers and partially in memory. On these machines, typically the 4062first few words of arguments are passed in registers, and the rest 4063on the stack. If a multi-word argument (a @code{double} or a 4064structure) crosses that boundary, its first few words must be passed 4065in registers and the rest must be pushed. This macro tells the 4066compiler when this occurs, and how many bytes should go in registers. 4067 4068@code{TARGET_FUNCTION_ARG} for these arguments should return the first 4069register to be used by the caller for this argument; likewise 4070@code{TARGET_FUNCTION_INCOMING_ARG}, for the called function. 4071@end deftypefn 4072 4073@deftypefn {Target Hook} bool TARGET_PASS_BY_REFERENCE (cumulative_args_t @var{cum}, const function_arg_info @var{&arg}) 4074This target hook should return @code{true} if argument @var{arg} at the 4075position indicated by @var{cum} should be passed by reference. This 4076predicate is queried after target independent reasons for being 4077passed by reference, such as @code{TREE_ADDRESSABLE (@var{arg}.type)}. 4078 4079If the hook returns true, a copy of that argument is made in memory and a 4080pointer to the argument is passed instead of the argument itself. 4081The pointer is passed in whatever way is appropriate for passing a pointer 4082to that type. 4083@end deftypefn 4084 4085@deftypefn {Target Hook} bool TARGET_CALLEE_COPIES (cumulative_args_t @var{cum}, const function_arg_info @var{&arg}) 4086The function argument described by the parameters to this hook is 4087known to be passed by reference. The hook should return true if the 4088function argument should be copied by the callee instead of copied 4089by the caller. 4090 4091For any argument for which the hook returns true, if it can be 4092determined that the argument is not modified, then a copy need 4093not be generated. 4094 4095The default version of this hook always returns false. 4096@end deftypefn 4097 4098@defmac CUMULATIVE_ARGS 4099A C type for declaring a variable that is used as the first argument 4100of @code{TARGET_FUNCTION_ARG} and other related values. For some 4101target machines, the type @code{int} suffices and can hold the number 4102of bytes of argument so far. 4103 4104There is no need to record in @code{CUMULATIVE_ARGS} anything about the 4105arguments that have been passed on the stack. The compiler has other 4106variables to keep track of that. For target machines on which all 4107arguments are passed on the stack, there is no need to store anything in 4108@code{CUMULATIVE_ARGS}; however, the data structure must exist and 4109should not be empty, so use @code{int}. 4110@end defmac 4111 4112@defmac OVERRIDE_ABI_FORMAT (@var{fndecl}) 4113If defined, this macro is called before generating any code for a 4114function, but after the @var{cfun} descriptor for the function has been 4115created. The back end may use this macro to update @var{cfun} to 4116reflect an ABI other than that which would normally be used by default. 4117If the compiler is generating code for a compiler-generated function, 4118@var{fndecl} may be @code{NULL}. 4119@end defmac 4120 4121@defmac INIT_CUMULATIVE_ARGS (@var{cum}, @var{fntype}, @var{libname}, @var{fndecl}, @var{n_named_args}) 4122A C statement (sans semicolon) for initializing the variable 4123@var{cum} for the state at the beginning of the argument list. The 4124variable has type @code{CUMULATIVE_ARGS}. The value of @var{fntype} 4125is the tree node for the data type of the function which will receive 4126the args, or 0 if the args are to a compiler support library function. 4127For direct calls that are not libcalls, @var{fndecl} contain the 4128declaration node of the function. @var{fndecl} is also set when 4129@code{INIT_CUMULATIVE_ARGS} is used to find arguments for the function 4130being compiled. @var{n_named_args} is set to the number of named 4131arguments, including a structure return address if it is passed as a 4132parameter, when making a call. When processing incoming arguments, 4133@var{n_named_args} is set to @minus{}1. 4134 4135When processing a call to a compiler support library function, 4136@var{libname} identifies which one. It is a @code{symbol_ref} rtx which 4137contains the name of the function, as a string. @var{libname} is 0 when 4138an ordinary C function call is being processed. Thus, each time this 4139macro is called, either @var{libname} or @var{fntype} is nonzero, but 4140never both of them at once. 4141@end defmac 4142 4143@defmac INIT_CUMULATIVE_LIBCALL_ARGS (@var{cum}, @var{mode}, @var{libname}) 4144Like @code{INIT_CUMULATIVE_ARGS} but only used for outgoing libcalls, 4145it gets a @code{MODE} argument instead of @var{fntype}, that would be 4146@code{NULL}. @var{indirect} would always be zero, too. If this macro 4147is not defined, @code{INIT_CUMULATIVE_ARGS (cum, NULL_RTX, libname, 41480)} is used instead. 4149@end defmac 4150 4151@defmac INIT_CUMULATIVE_INCOMING_ARGS (@var{cum}, @var{fntype}, @var{libname}) 4152Like @code{INIT_CUMULATIVE_ARGS} but overrides it for the purposes of 4153finding the arguments for the function being compiled. If this macro is 4154undefined, @code{INIT_CUMULATIVE_ARGS} is used instead. 4155 4156The value passed for @var{libname} is always 0, since library routines 4157with special calling conventions are never compiled with GCC@. The 4158argument @var{libname} exists for symmetry with 4159@code{INIT_CUMULATIVE_ARGS}. 4160@c could use "this macro" in place of @code{INIT_CUMULATIVE_ARGS}, maybe. 4161@c --mew 5feb93 i switched the order of the sentences. --mew 10feb93 4162@end defmac 4163 4164@deftypefn {Target Hook} void TARGET_FUNCTION_ARG_ADVANCE (cumulative_args_t @var{ca}, const function_arg_info @var{&arg}) 4165This hook updates the summarizer variable pointed to by @var{ca} to 4166advance past argument @var{arg} in the argument list. Once this is done, 4167the variable @var{cum} is suitable for analyzing the @emph{following} 4168argument with @code{TARGET_FUNCTION_ARG}, etc. 4169 4170This hook need not do anything if the argument in question was passed 4171on the stack. The compiler knows how to track the amount of stack space 4172used for arguments without any special help. 4173@end deftypefn 4174 4175@deftypefn {Target Hook} HOST_WIDE_INT TARGET_FUNCTION_ARG_OFFSET (machine_mode @var{mode}, const_tree @var{type}) 4176This hook returns the number of bytes to add to the offset of an 4177argument of type @var{type} and mode @var{mode} when passed in memory. 4178This is needed for the SPU, which passes @code{char} and @code{short} 4179arguments in the preferred slot that is in the middle of the quad word 4180instead of starting at the top. The default implementation returns 0. 4181@end deftypefn 4182 4183@deftypefn {Target Hook} pad_direction TARGET_FUNCTION_ARG_PADDING (machine_mode @var{mode}, const_tree @var{type}) 4184This hook determines whether, and in which direction, to pad out 4185an argument of mode @var{mode} and type @var{type}. It returns 4186@code{PAD_UPWARD} to insert padding above the argument, @code{PAD_DOWNWARD} 4187to insert padding below the argument, or @code{PAD_NONE} to inhibit padding. 4188 4189The @emph{amount} of padding is not controlled by this hook, but by 4190@code{TARGET_FUNCTION_ARG_ROUND_BOUNDARY}. It is always just enough 4191to reach the next multiple of that boundary. 4192 4193This hook has a default definition that is right for most systems. 4194For little-endian machines, the default is to pad upward. For 4195big-endian machines, the default is to pad downward for an argument of 4196constant size shorter than an @code{int}, and upward otherwise. 4197@end deftypefn 4198 4199@defmac PAD_VARARGS_DOWN 4200If defined, a C expression which determines whether the default 4201implementation of va_arg will attempt to pad down before reading the 4202next argument, if that argument is smaller than its aligned space as 4203controlled by @code{PARM_BOUNDARY}. If this macro is not defined, all such 4204arguments are padded down if @code{BYTES_BIG_ENDIAN} is true. 4205@end defmac 4206 4207@defmac BLOCK_REG_PADDING (@var{mode}, @var{type}, @var{first}) 4208Specify padding for the last element of a block move between registers and 4209memory. @var{first} is nonzero if this is the only element. Defining this 4210macro allows better control of register function parameters on big-endian 4211machines, without using @code{PARALLEL} rtl. In particular, 4212@code{MUST_PASS_IN_STACK} need not test padding and mode of types in 4213registers, as there is no longer a "wrong" part of a register; For example, 4214a three byte aggregate may be passed in the high part of a register if so 4215required. 4216@end defmac 4217 4218@deftypefn {Target Hook} {unsigned int} TARGET_FUNCTION_ARG_BOUNDARY (machine_mode @var{mode}, const_tree @var{type}) 4219This hook returns the alignment boundary, in bits, of an argument 4220with the specified mode and type. The default hook returns 4221@code{PARM_BOUNDARY} for all arguments. 4222@end deftypefn 4223 4224@deftypefn {Target Hook} {unsigned int} TARGET_FUNCTION_ARG_ROUND_BOUNDARY (machine_mode @var{mode}, const_tree @var{type}) 4225Normally, the size of an argument is rounded up to @code{PARM_BOUNDARY}, 4226which is the default value for this hook. You can define this hook to 4227return a different value if an argument size must be rounded to a larger 4228value. 4229@end deftypefn 4230 4231@defmac FUNCTION_ARG_REGNO_P (@var{regno}) 4232A C expression that is nonzero if @var{regno} is the number of a hard 4233register in which function arguments are sometimes passed. This does 4234@emph{not} include implicit arguments such as the static chain and 4235the structure-value address. On many machines, no registers can be 4236used for this purpose since all function arguments are pushed on the 4237stack. 4238@end defmac 4239 4240@deftypefn {Target Hook} bool TARGET_SPLIT_COMPLEX_ARG (const_tree @var{type}) 4241This hook should return true if parameter of type @var{type} are passed 4242as two scalar parameters. By default, GCC will attempt to pack complex 4243arguments into the target's word size. Some ABIs require complex arguments 4244to be split and treated as their individual components. For example, on 4245AIX64, complex floats should be passed in a pair of floating point 4246registers, even though a complex float would fit in one 64-bit floating 4247point register. 4248 4249The default value of this hook is @code{NULL}, which is treated as always 4250false. 4251@end deftypefn 4252 4253@deftypefn {Target Hook} tree TARGET_BUILD_BUILTIN_VA_LIST (void) 4254This hook returns a type node for @code{va_list} for the target. 4255The default version of the hook returns @code{void*}. 4256@end deftypefn 4257 4258@deftypefn {Target Hook} int TARGET_ENUM_VA_LIST_P (int @var{idx}, const char **@var{pname}, tree *@var{ptree}) 4259This target hook is used in function @code{c_common_nodes_and_builtins} 4260to iterate through the target specific builtin types for va_list. The 4261variable @var{idx} is used as iterator. @var{pname} has to be a pointer 4262to a @code{const char *} and @var{ptree} a pointer to a @code{tree} typed 4263variable. 4264The arguments @var{pname} and @var{ptree} are used to store the result of 4265this macro and are set to the name of the va_list builtin type and its 4266internal type. 4267If the return value of this macro is zero, then there is no more element. 4268Otherwise the @var{IDX} should be increased for the next call of this 4269macro to iterate through all types. 4270@end deftypefn 4271 4272@deftypefn {Target Hook} tree TARGET_FN_ABI_VA_LIST (tree @var{fndecl}) 4273This hook returns the va_list type of the calling convention specified by 4274@var{fndecl}. 4275The default version of this hook returns @code{va_list_type_node}. 4276@end deftypefn 4277 4278@deftypefn {Target Hook} tree TARGET_CANONICAL_VA_LIST_TYPE (tree @var{type}) 4279This hook returns the va_list type of the calling convention specified by the 4280type of @var{type}. If @var{type} is not a valid va_list type, it returns 4281@code{NULL_TREE}. 4282@end deftypefn 4283 4284@deftypefn {Target Hook} tree TARGET_GIMPLIFY_VA_ARG_EXPR (tree @var{valist}, tree @var{type}, gimple_seq *@var{pre_p}, gimple_seq *@var{post_p}) 4285This hook performs target-specific gimplification of 4286@code{VA_ARG_EXPR}. The first two parameters correspond to the 4287arguments to @code{va_arg}; the latter two are as in 4288@code{gimplify.c:gimplify_expr}. 4289@end deftypefn 4290 4291@deftypefn {Target Hook} bool TARGET_VALID_POINTER_MODE (scalar_int_mode @var{mode}) 4292Define this to return nonzero if the port can handle pointers 4293with machine mode @var{mode}. The default version of this 4294hook returns true for both @code{ptr_mode} and @code{Pmode}. 4295@end deftypefn 4296 4297@deftypefn {Target Hook} bool TARGET_REF_MAY_ALIAS_ERRNO (ao_ref *@var{ref}) 4298Define this to return nonzero if the memory reference @var{ref} may alias with the system C library errno location. The default version of this hook assumes the system C library errno location is either a declaration of type int or accessed by dereferencing a pointer to int. 4299@end deftypefn 4300 4301@deftypefn {Target Hook} machine_mode TARGET_TRANSLATE_MODE_ATTRIBUTE (machine_mode @var{mode}) 4302Define this hook if during mode attribute processing, the port should 4303translate machine_mode @var{mode} to another mode. For example, rs6000's 4304@code{KFmode}, when it is the same as @code{TFmode}. 4305 4306The default version of the hook returns that mode that was passed in. 4307@end deftypefn 4308 4309@deftypefn {Target Hook} bool TARGET_SCALAR_MODE_SUPPORTED_P (scalar_mode @var{mode}) 4310Define this to return nonzero if the port is prepared to handle 4311insns involving scalar mode @var{mode}. For a scalar mode to be 4312considered supported, all the basic arithmetic and comparisons 4313must work. 4314 4315The default version of this hook returns true for any mode 4316required to handle the basic C types (as defined by the port). 4317Included here are the double-word arithmetic supported by the 4318code in @file{optabs.c}. 4319@end deftypefn 4320 4321@deftypefn {Target Hook} bool TARGET_VECTOR_MODE_SUPPORTED_P (machine_mode @var{mode}) 4322Define this to return nonzero if the port is prepared to handle 4323insns involving vector mode @var{mode}. At the very least, it 4324must have move patterns for this mode. 4325@end deftypefn 4326 4327@deftypefn {Target Hook} bool TARGET_COMPATIBLE_VECTOR_TYPES_P (const_tree @var{type1}, const_tree @var{type2}) 4328Return true if there is no target-specific reason for treating 4329vector types @var{type1} and @var{type2} as distinct types. The caller 4330has already checked for target-independent reasons, meaning that the 4331types are known to have the same mode, to have the same number of elements, 4332and to have what the caller considers to be compatible element types. 4333 4334The main reason for defining this hook is to reject pairs of types 4335that are handled differently by the target's calling convention. 4336For example, when a new @var{N}-bit vector architecture is added 4337to a target, the target may want to handle normal @var{N}-bit 4338@code{VECTOR_TYPE} arguments and return values in the same way as 4339before, to maintain backwards compatibility. However, it may also 4340provide new, architecture-specific @code{VECTOR_TYPE}s that are passed 4341and returned in a more efficient way. It is then important to maintain 4342a distinction between the ``normal'' @code{VECTOR_TYPE}s and the new 4343architecture-specific ones. 4344 4345The default implementation returns true, which is correct for most targets. 4346@end deftypefn 4347 4348@deftypefn {Target Hook} opt_machine_mode TARGET_ARRAY_MODE (machine_mode @var{mode}, unsigned HOST_WIDE_INT @var{nelems}) 4349Return the mode that GCC should use for an array that has 4350@var{nelems} elements, with each element having mode @var{mode}. 4351Return no mode if the target has no special requirements. In the 4352latter case, GCC looks for an integer mode of the appropriate size 4353if available and uses BLKmode otherwise. Usually the search for the 4354integer mode is limited to @code{MAX_FIXED_MODE_SIZE}, but the 4355@code{TARGET_ARRAY_MODE_SUPPORTED_P} hook allows a larger mode to be 4356used in specific cases. 4357 4358The main use of this hook is to specify that an array of vectors should 4359also have a vector mode. The default implementation returns no mode. 4360@end deftypefn 4361 4362@deftypefn {Target Hook} bool TARGET_ARRAY_MODE_SUPPORTED_P (machine_mode @var{mode}, unsigned HOST_WIDE_INT @var{nelems}) 4363Return true if GCC should try to use a scalar mode to store an array 4364of @var{nelems} elements, given that each element has mode @var{mode}. 4365Returning true here overrides the usual @code{MAX_FIXED_MODE} limit 4366and allows GCC to use any defined integer mode. 4367 4368One use of this hook is to support vector load and store operations 4369that operate on several homogeneous vectors. For example, ARM NEON 4370has operations like: 4371 4372@smallexample 4373int8x8x3_t vld3_s8 (const int8_t *) 4374@end smallexample 4375 4376where the return type is defined as: 4377 4378@smallexample 4379typedef struct int8x8x3_t 4380@{ 4381 int8x8_t val[3]; 4382@} int8x8x3_t; 4383@end smallexample 4384 4385If this hook allows @code{val} to have a scalar mode, then 4386@code{int8x8x3_t} can have the same mode. GCC can then store 4387@code{int8x8x3_t}s in registers rather than forcing them onto the stack. 4388@end deftypefn 4389 4390@deftypefn {Target Hook} bool TARGET_LIBGCC_FLOATING_MODE_SUPPORTED_P (scalar_float_mode @var{mode}) 4391Define this to return nonzero if libgcc provides support for the 4392floating-point mode @var{mode}, which is known to pass 4393@code{TARGET_SCALAR_MODE_SUPPORTED_P}. The default version of this 4394hook returns true for all of @code{SFmode}, @code{DFmode}, 4395@code{XFmode} and @code{TFmode}, if such modes exist. 4396@end deftypefn 4397 4398@deftypefn {Target Hook} opt_scalar_float_mode TARGET_FLOATN_MODE (int @var{n}, bool @var{extended}) 4399Define this to return the machine mode to use for the type 4400@code{_Float@var{n}}, if @var{extended} is false, or the type 4401@code{_Float@var{n}x}, if @var{extended} is true. If such a type is not 4402supported, return @code{opt_scalar_float_mode ()}. The default version of 4403this hook returns @code{SFmode} for @code{_Float32}, @code{DFmode} for 4404@code{_Float64} and @code{_Float32x} and @code{TFmode} for 4405@code{_Float128}, if those modes exist and satisfy the requirements for 4406those types and pass @code{TARGET_SCALAR_MODE_SUPPORTED_P} and 4407@code{TARGET_LIBGCC_FLOATING_MODE_SUPPORTED_P}; for @code{_Float64x}, it 4408returns the first of @code{XFmode} and @code{TFmode} that exists and 4409satisfies the same requirements; for other types, it returns 4410@code{opt_scalar_float_mode ()}. The hook is only called for values 4411of @var{n} and @var{extended} that are valid according to 4412ISO/IEC TS 18661-3:2015; that is, @var{n} is one of 32, 64, 128, or, 4413if @var{extended} is false, 16 or greater than 128 and a multiple of 32. 4414@end deftypefn 4415 4416@deftypefn {Target Hook} bool TARGET_FLOATN_BUILTIN_P (int @var{func}) 4417Define this to return true if the @code{_Float@var{n}} and 4418@code{_Float@var{n}x} built-in functions should implicitly enable the 4419built-in function without the @code{__builtin_} prefix in addition to the 4420normal built-in function with the @code{__builtin_} prefix. The default is 4421to only enable built-in functions without the @code{__builtin_} prefix for 4422the GNU C langauge. In strict ANSI/ISO mode, the built-in function without 4423the @code{__builtin_} prefix is not enabled. The argument @code{FUNC} is the 4424@code{enum built_in_function} id of the function to be enabled. 4425@end deftypefn 4426 4427@deftypefn {Target Hook} bool TARGET_SMALL_REGISTER_CLASSES_FOR_MODE_P (machine_mode @var{mode}) 4428Define this to return nonzero for machine modes for which the port has 4429small register classes. If this target hook returns nonzero for a given 4430@var{mode}, the compiler will try to minimize the lifetime of registers 4431in @var{mode}. The hook may be called with @code{VOIDmode} as argument. 4432In this case, the hook is expected to return nonzero if it returns nonzero 4433for any mode. 4434 4435On some machines, it is risky to let hard registers live across arbitrary 4436insns. Typically, these machines have instructions that require values 4437to be in specific registers (like an accumulator), and reload will fail 4438if the required hard register is used for another purpose across such an 4439insn. 4440 4441Passes before reload do not know which hard registers will be used 4442in an instruction, but the machine modes of the registers set or used in 4443the instruction are already known. And for some machines, register 4444classes are small for, say, integer registers but not for floating point 4445registers. For example, the AMD x86-64 architecture requires specific 4446registers for the legacy x86 integer instructions, but there are many 4447SSE registers for floating point operations. On such targets, a good 4448strategy may be to return nonzero from this hook for @code{INTEGRAL_MODE_P} 4449machine modes but zero for the SSE register classes. 4450 4451The default version of this hook returns false for any mode. It is always 4452safe to redefine this hook to return with a nonzero value. But if you 4453unnecessarily define it, you will reduce the amount of optimizations 4454that can be performed in some cases. If you do not define this hook 4455to return a nonzero value when it is required, the compiler will run out 4456of spill registers and print a fatal error message. 4457@end deftypefn 4458 4459@node Scalar Return 4460@subsection How Scalar Function Values Are Returned 4461@cindex return values in registers 4462@cindex values, returned by functions 4463@cindex scalars, returned as values 4464 4465This section discusses the macros that control returning scalars as 4466values---values that can fit in registers. 4467 4468@deftypefn {Target Hook} rtx TARGET_FUNCTION_VALUE (const_tree @var{ret_type}, const_tree @var{fn_decl_or_type}, bool @var{outgoing}) 4469 4470Define this to return an RTX representing the place where a function 4471returns or receives a value of data type @var{ret_type}, a tree node 4472representing a data type. @var{fn_decl_or_type} is a tree node 4473representing @code{FUNCTION_DECL} or @code{FUNCTION_TYPE} of a 4474function being called. If @var{outgoing} is false, the hook should 4475compute the register in which the caller will see the return value. 4476Otherwise, the hook should return an RTX representing the place where 4477a function returns a value. 4478 4479On many machines, only @code{TYPE_MODE (@var{ret_type})} is relevant. 4480(Actually, on most machines, scalar values are returned in the same 4481place regardless of mode.) The value of the expression is usually a 4482@code{reg} RTX for the hard register where the return value is stored. 4483The value can also be a @code{parallel} RTX, if the return value is in 4484multiple places. See @code{TARGET_FUNCTION_ARG} for an explanation of the 4485@code{parallel} form. Note that the callee will populate every 4486location specified in the @code{parallel}, but if the first element of 4487the @code{parallel} contains the whole return value, callers will use 4488that element as the canonical location and ignore the others. The m68k 4489port uses this type of @code{parallel} to return pointers in both 4490@samp{%a0} (the canonical location) and @samp{%d0}. 4491 4492If @code{TARGET_PROMOTE_FUNCTION_RETURN} returns true, you must apply 4493the same promotion rules specified in @code{PROMOTE_MODE} if 4494@var{valtype} is a scalar type. 4495 4496If the precise function being called is known, @var{func} is a tree 4497node (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null 4498pointer. This makes it possible to use a different value-returning 4499convention for specific functions when all their calls are 4500known. 4501 4502Some target machines have ``register windows'' so that the register in 4503which a function returns its value is not the same as the one in which 4504the caller sees the value. For such machines, you should return 4505different RTX depending on @var{outgoing}. 4506 4507@code{TARGET_FUNCTION_VALUE} is not used for return values with 4508aggregate data types, because these are returned in another way. See 4509@code{TARGET_STRUCT_VALUE_RTX} and related macros, below. 4510@end deftypefn 4511 4512@defmac FUNCTION_VALUE (@var{valtype}, @var{func}) 4513This macro has been deprecated. Use @code{TARGET_FUNCTION_VALUE} for 4514a new target instead. 4515@end defmac 4516 4517@defmac LIBCALL_VALUE (@var{mode}) 4518A C expression to create an RTX representing the place where a library 4519function returns a value of mode @var{mode}. 4520 4521Note that ``library function'' in this context means a compiler 4522support routine, used to perform arithmetic, whose name is known 4523specially by the compiler and was not mentioned in the C code being 4524compiled. 4525@end defmac 4526 4527@deftypefn {Target Hook} rtx TARGET_LIBCALL_VALUE (machine_mode @var{mode}, const_rtx @var{fun}) 4528Define this hook if the back-end needs to know the name of the libcall 4529function in order to determine where the result should be returned. 4530 4531The mode of the result is given by @var{mode} and the name of the called 4532library function is given by @var{fun}. The hook should return an RTX 4533representing the place where the library function result will be returned. 4534 4535If this hook is not defined, then LIBCALL_VALUE will be used. 4536@end deftypefn 4537 4538@defmac FUNCTION_VALUE_REGNO_P (@var{regno}) 4539A C expression that is nonzero if @var{regno} is the number of a hard 4540register in which the values of called function may come back. 4541 4542A register whose use for returning values is limited to serving as the 4543second of a pair (for a value of type @code{double}, say) need not be 4544recognized by this macro. So for most machines, this definition 4545suffices: 4546 4547@smallexample 4548#define FUNCTION_VALUE_REGNO_P(N) ((N) == 0) 4549@end smallexample 4550 4551If the machine has register windows, so that the caller and the called 4552function use different registers for the return value, this macro 4553should recognize only the caller's register numbers. 4554 4555This macro has been deprecated. Use @code{TARGET_FUNCTION_VALUE_REGNO_P} 4556for a new target instead. 4557@end defmac 4558 4559@deftypefn {Target Hook} bool TARGET_FUNCTION_VALUE_REGNO_P (const unsigned int @var{regno}) 4560A target hook that return @code{true} if @var{regno} is the number of a hard 4561register in which the values of called function may come back. 4562 4563A register whose use for returning values is limited to serving as the 4564second of a pair (for a value of type @code{double}, say) need not be 4565recognized by this target hook. 4566 4567If the machine has register windows, so that the caller and the called 4568function use different registers for the return value, this target hook 4569should recognize only the caller's register numbers. 4570 4571If this hook is not defined, then FUNCTION_VALUE_REGNO_P will be used. 4572@end deftypefn 4573 4574@defmac APPLY_RESULT_SIZE 4575Define this macro if @samp{untyped_call} and @samp{untyped_return} 4576need more space than is implied by @code{FUNCTION_VALUE_REGNO_P} for 4577saving and restoring an arbitrary return value. 4578@end defmac 4579 4580@deftypevr {Target Hook} bool TARGET_OMIT_STRUCT_RETURN_REG 4581Normally, when a function returns a structure by memory, the address 4582is passed as an invisible pointer argument, but the compiler also 4583arranges to return the address from the function like it would a normal 4584pointer return value. Define this to true if that behavior is 4585undesirable on your target. 4586@end deftypevr 4587 4588@deftypefn {Target Hook} bool TARGET_RETURN_IN_MSB (const_tree @var{type}) 4589This hook should return true if values of type @var{type} are returned 4590at the most significant end of a register (in other words, if they are 4591padded at the least significant end). You can assume that @var{type} 4592is returned in a register; the caller is required to check this. 4593 4594Note that the register provided by @code{TARGET_FUNCTION_VALUE} must 4595be able to hold the complete return value. For example, if a 1-, 2- 4596or 3-byte structure is returned at the most significant end of a 45974-byte register, @code{TARGET_FUNCTION_VALUE} should provide an 4598@code{SImode} rtx. 4599@end deftypefn 4600 4601@node Aggregate Return 4602@subsection How Large Values Are Returned 4603@cindex aggregates as return values 4604@cindex large return values 4605@cindex returning aggregate values 4606@cindex structure value address 4607 4608When a function value's mode is @code{BLKmode} (and in some other 4609cases), the value is not returned according to 4610@code{TARGET_FUNCTION_VALUE} (@pxref{Scalar Return}). Instead, the 4611caller passes the address of a block of memory in which the value 4612should be stored. This address is called the @dfn{structure value 4613address}. 4614 4615This section describes how to control returning structure values in 4616memory. 4617 4618@deftypefn {Target Hook} bool TARGET_RETURN_IN_MEMORY (const_tree @var{type}, const_tree @var{fntype}) 4619This target hook should return a nonzero value to say to return the 4620function value in memory, just as large structures are always returned. 4621Here @var{type} will be the data type of the value, and @var{fntype} 4622will be the type of the function doing the returning, or @code{NULL} for 4623libcalls. 4624 4625Note that values of mode @code{BLKmode} must be explicitly handled 4626by this function. Also, the option @option{-fpcc-struct-return} 4627takes effect regardless of this macro. On most systems, it is 4628possible to leave the hook undefined; this causes a default 4629definition to be used, whose value is the constant 1 for @code{BLKmode} 4630values, and 0 otherwise. 4631 4632Do not use this hook to indicate that structures and unions should always 4633be returned in memory. You should instead use @code{DEFAULT_PCC_STRUCT_RETURN} 4634to indicate this. 4635@end deftypefn 4636 4637@defmac DEFAULT_PCC_STRUCT_RETURN 4638Define this macro to be 1 if all structure and union return values must be 4639in memory. Since this results in slower code, this should be defined 4640only if needed for compatibility with other compilers or with an ABI@. 4641If you define this macro to be 0, then the conventions used for structure 4642and union return values are decided by the @code{TARGET_RETURN_IN_MEMORY} 4643target hook. 4644 4645If not defined, this defaults to the value 1. 4646@end defmac 4647 4648@deftypefn {Target Hook} rtx TARGET_STRUCT_VALUE_RTX (tree @var{fndecl}, int @var{incoming}) 4649This target hook should return the location of the structure value 4650address (normally a @code{mem} or @code{reg}), or 0 if the address is 4651passed as an ``invisible'' first argument. Note that @var{fndecl} may 4652be @code{NULL}, for libcalls. You do not need to define this target 4653hook if the address is always passed as an ``invisible'' first 4654argument. 4655 4656On some architectures the place where the structure value address 4657is found by the called function is not the same place that the 4658caller put it. This can be due to register windows, or it could 4659be because the function prologue moves it to a different place. 4660@var{incoming} is @code{1} or @code{2} when the location is needed in 4661the context of the called function, and @code{0} in the context of 4662the caller. 4663 4664If @var{incoming} is nonzero and the address is to be found on the 4665stack, return a @code{mem} which refers to the frame pointer. If 4666@var{incoming} is @code{2}, the result is being used to fetch the 4667structure value address at the beginning of a function. If you need 4668to emit adjusting code, you should do it at this point. 4669@end deftypefn 4670 4671@defmac PCC_STATIC_STRUCT_RETURN 4672Define this macro if the usual system convention on the target machine 4673for returning structures and unions is for the called function to return 4674the address of a static variable containing the value. 4675 4676Do not define this if the usual system convention is for the caller to 4677pass an address to the subroutine. 4678 4679This macro has effect in @option{-fpcc-struct-return} mode, but it does 4680nothing when you use @option{-freg-struct-return} mode. 4681@end defmac 4682 4683@deftypefn {Target Hook} fixed_size_mode TARGET_GET_RAW_RESULT_MODE (int @var{regno}) 4684This target hook returns the mode to be used when accessing raw return registers in @code{__builtin_return}. Define this macro if the value in @var{reg_raw_mode} is not correct. 4685@end deftypefn 4686 4687@deftypefn {Target Hook} fixed_size_mode TARGET_GET_RAW_ARG_MODE (int @var{regno}) 4688This target hook returns the mode to be used when accessing raw argument registers in @code{__builtin_apply_args}. Define this macro if the value in @var{reg_raw_mode} is not correct. 4689@end deftypefn 4690 4691@deftypefn {Target Hook} bool TARGET_EMPTY_RECORD_P (const_tree @var{type}) 4692This target hook returns true if the type is an empty record. The default 4693is to return @code{false}. 4694@end deftypefn 4695 4696@deftypefn {Target Hook} void TARGET_WARN_PARAMETER_PASSING_ABI (cumulative_args_t @var{ca}, tree @var{type}) 4697This target hook warns about the change in empty class parameter passing 4698ABI. 4699@end deftypefn 4700 4701@node Caller Saves 4702@subsection Caller-Saves Register Allocation 4703 4704If you enable it, GCC can save registers around function calls. This 4705makes it possible to use call-clobbered registers to hold variables that 4706must live across calls. 4707 4708@defmac HARD_REGNO_CALLER_SAVE_MODE (@var{regno}, @var{nregs}) 4709A C expression specifying which mode is required for saving @var{nregs} 4710of a pseudo-register in call-clobbered hard register @var{regno}. If 4711@var{regno} is unsuitable for caller save, @code{VOIDmode} should be 4712returned. For most machines this macro need not be defined since GCC 4713will select the smallest suitable mode. 4714@end defmac 4715 4716@node Function Entry 4717@subsection Function Entry and Exit 4718@cindex function entry and exit 4719@cindex prologue 4720@cindex epilogue 4721 4722This section describes the macros that output function entry 4723(@dfn{prologue}) and exit (@dfn{epilogue}) code. 4724 4725@deftypefn {Target Hook} void TARGET_ASM_PRINT_PATCHABLE_FUNCTION_ENTRY (FILE *@var{file}, unsigned HOST_WIDE_INT @var{patch_area_size}, bool @var{record_p}) 4726Generate a patchable area at the function start, consisting of 4727@var{patch_area_size} NOP instructions. If the target supports named 4728sections and if @var{record_p} is true, insert a pointer to the current 4729location in the table of patchable functions. The default implementation 4730of the hook places the table of pointers in the special section named 4731@code{__patchable_function_entries}. 4732@end deftypefn 4733 4734@deftypefn {Target Hook} void TARGET_ASM_FUNCTION_PROLOGUE (FILE *@var{file}) 4735If defined, a function that outputs the assembler code for entry to a 4736function. The prologue is responsible for setting up the stack frame, 4737initializing the frame pointer register, saving registers that must be 4738saved, and allocating @var{size} additional bytes of storage for the 4739local variables. @var{file} is a stdio stream to which the assembler 4740code should be output. 4741 4742The label for the beginning of the function need not be output by this 4743macro. That has already been done when the macro is run. 4744 4745@findex regs_ever_live 4746To determine which registers to save, the macro can refer to the array 4747@code{regs_ever_live}: element @var{r} is nonzero if hard register 4748@var{r} is used anywhere within the function. This implies the function 4749prologue should save register @var{r}, provided it is not one of the 4750call-used registers. (@code{TARGET_ASM_FUNCTION_EPILOGUE} must likewise use 4751@code{regs_ever_live}.) 4752 4753On machines that have ``register windows'', the function entry code does 4754not save on the stack the registers that are in the windows, even if 4755they are supposed to be preserved by function calls; instead it takes 4756appropriate steps to ``push'' the register stack, if any non-call-used 4757registers are used in the function. 4758 4759@findex frame_pointer_needed 4760On machines where functions may or may not have frame-pointers, the 4761function entry code must vary accordingly; it must set up the frame 4762pointer if one is wanted, and not otherwise. To determine whether a 4763frame pointer is in wanted, the macro can refer to the variable 4764@code{frame_pointer_needed}. The variable's value will be 1 at run 4765time in a function that needs a frame pointer. @xref{Elimination}. 4766 4767The function entry code is responsible for allocating any stack space 4768required for the function. This stack space consists of the regions 4769listed below. In most cases, these regions are allocated in the 4770order listed, with the last listed region closest to the top of the 4771stack (the lowest address if @code{STACK_GROWS_DOWNWARD} is defined, and 4772the highest address if it is not defined). You can use a different order 4773for a machine if doing so is more convenient or required for 4774compatibility reasons. Except in cases where required by standard 4775or by a debugger, there is no reason why the stack layout used by GCC 4776need agree with that used by other compilers for a machine. 4777@end deftypefn 4778 4779@deftypefn {Target Hook} void TARGET_ASM_FUNCTION_END_PROLOGUE (FILE *@var{file}) 4780If defined, a function that outputs assembler code at the end of a 4781prologue. This should be used when the function prologue is being 4782emitted as RTL, and you have some extra assembler that needs to be 4783emitted. @xref{prologue instruction pattern}. 4784@end deftypefn 4785 4786@deftypefn {Target Hook} void TARGET_ASM_FUNCTION_BEGIN_EPILOGUE (FILE *@var{file}) 4787If defined, a function that outputs assembler code at the start of an 4788epilogue. This should be used when the function epilogue is being 4789emitted as RTL, and you have some extra assembler that needs to be 4790emitted. @xref{epilogue instruction pattern}. 4791@end deftypefn 4792 4793@deftypefn {Target Hook} void TARGET_ASM_FUNCTION_EPILOGUE (FILE *@var{file}) 4794If defined, a function that outputs the assembler code for exit from a 4795function. The epilogue is responsible for restoring the saved 4796registers and stack pointer to their values when the function was 4797called, and returning control to the caller. This macro takes the 4798same argument as the macro @code{TARGET_ASM_FUNCTION_PROLOGUE}, and the 4799registers to restore are determined from @code{regs_ever_live} and 4800@code{CALL_USED_REGISTERS} in the same way. 4801 4802On some machines, there is a single instruction that does all the work 4803of returning from the function. On these machines, give that 4804instruction the name @samp{return} and do not define the macro 4805@code{TARGET_ASM_FUNCTION_EPILOGUE} at all. 4806 4807Do not define a pattern named @samp{return} if you want the 4808@code{TARGET_ASM_FUNCTION_EPILOGUE} to be used. If you want the target 4809switches to control whether return instructions or epilogues are used, 4810define a @samp{return} pattern with a validity condition that tests the 4811target switches appropriately. If the @samp{return} pattern's validity 4812condition is false, epilogues will be used. 4813 4814On machines where functions may or may not have frame-pointers, the 4815function exit code must vary accordingly. Sometimes the code for these 4816two cases is completely different. To determine whether a frame pointer 4817is wanted, the macro can refer to the variable 4818@code{frame_pointer_needed}. The variable's value will be 1 when compiling 4819a function that needs a frame pointer. 4820 4821Normally, @code{TARGET_ASM_FUNCTION_PROLOGUE} and 4822@code{TARGET_ASM_FUNCTION_EPILOGUE} must treat leaf functions specially. 4823The C variable @code{current_function_is_leaf} is nonzero for such a 4824function. @xref{Leaf Functions}. 4825 4826On some machines, some functions pop their arguments on exit while 4827others leave that for the caller to do. For example, the 68020 when 4828given @option{-mrtd} pops arguments in functions that take a fixed 4829number of arguments. 4830 4831@findex pops_args 4832@findex crtl->args.pops_args 4833Your definition of the macro @code{RETURN_POPS_ARGS} decides which 4834functions pop their own arguments. @code{TARGET_ASM_FUNCTION_EPILOGUE} 4835needs to know what was decided. The number of bytes of the current 4836function's arguments that this function should pop is available in 4837@code{crtl->args.pops_args}. @xref{Scalar Return}. 4838@end deftypefn 4839 4840@itemize @bullet 4841@item 4842@findex pretend_args_size 4843@findex crtl->args.pretend_args_size 4844A region of @code{crtl->args.pretend_args_size} bytes of 4845uninitialized space just underneath the first argument arriving on the 4846stack. (This may not be at the very start of the allocated stack region 4847if the calling sequence has pushed anything else since pushing the stack 4848arguments. But usually, on such machines, nothing else has been pushed 4849yet, because the function prologue itself does all the pushing.) This 4850region is used on machines where an argument may be passed partly in 4851registers and partly in memory, and, in some cases to support the 4852features in @code{<stdarg.h>}. 4853 4854@item 4855An area of memory used to save certain registers used by the function. 4856The size of this area, which may also include space for such things as 4857the return address and pointers to previous stack frames, is 4858machine-specific and usually depends on which registers have been used 4859in the function. Machines with register windows often do not require 4860a save area. 4861 4862@item 4863A region of at least @var{size} bytes, possibly rounded up to an allocation 4864boundary, to contain the local variables of the function. On some machines, 4865this region and the save area may occur in the opposite order, with the 4866save area closer to the top of the stack. 4867 4868@item 4869@cindex @code{ACCUMULATE_OUTGOING_ARGS} and stack frames 4870Optionally, when @code{ACCUMULATE_OUTGOING_ARGS} is defined, a region of 4871@code{crtl->outgoing_args_size} bytes to be used for outgoing 4872argument lists of the function. @xref{Stack Arguments}. 4873@end itemize 4874 4875@defmac EXIT_IGNORE_STACK 4876Define this macro as a C expression that is nonzero if the return 4877instruction or the function epilogue ignores the value of the stack 4878pointer; in other words, if it is safe to delete an instruction to 4879adjust the stack pointer before a return from the function. The 4880default is 0. 4881 4882Note that this macro's value is relevant only for functions for which 4883frame pointers are maintained. It is never safe to delete a final 4884stack adjustment in a function that has no frame pointer, and the 4885compiler knows this regardless of @code{EXIT_IGNORE_STACK}. 4886@end defmac 4887 4888@defmac EPILOGUE_USES (@var{regno}) 4889Define this macro as a C expression that is nonzero for registers that are 4890used by the epilogue or the @samp{return} pattern. The stack and frame 4891pointer registers are already assumed to be used as needed. 4892@end defmac 4893 4894@defmac EH_USES (@var{regno}) 4895Define this macro as a C expression that is nonzero for registers that are 4896used by the exception handling mechanism, and so should be considered live 4897on entry to an exception edge. 4898@end defmac 4899 4900@deftypefn {Target Hook} void TARGET_ASM_OUTPUT_MI_THUNK (FILE *@var{file}, tree @var{thunk_fndecl}, HOST_WIDE_INT @var{delta}, HOST_WIDE_INT @var{vcall_offset}, tree @var{function}) 4901A function that outputs the assembler code for a thunk 4902function, used to implement C++ virtual function calls with multiple 4903inheritance. The thunk acts as a wrapper around a virtual function, 4904adjusting the implicit object parameter before handing control off to 4905the real function. 4906 4907First, emit code to add the integer @var{delta} to the location that 4908contains the incoming first argument. Assume that this argument 4909contains a pointer, and is the one used to pass the @code{this} pointer 4910in C++. This is the incoming argument @emph{before} the function prologue, 4911e.g.@: @samp{%o0} on a sparc. The addition must preserve the values of 4912all other incoming arguments. 4913 4914Then, if @var{vcall_offset} is nonzero, an additional adjustment should be 4915made after adding @code{delta}. In particular, if @var{p} is the 4916adjusted pointer, the following adjustment should be made: 4917 4918@smallexample 4919p += (*((ptrdiff_t **)p))[vcall_offset/sizeof(ptrdiff_t)] 4920@end smallexample 4921 4922After the additions, emit code to jump to @var{function}, which is a 4923@code{FUNCTION_DECL}. This is a direct pure jump, not a call, and does 4924not touch the return address. Hence returning from @var{FUNCTION} will 4925return to whoever called the current @samp{thunk}. 4926 4927The effect must be as if @var{function} had been called directly with 4928the adjusted first argument. This macro is responsible for emitting all 4929of the code for a thunk function; @code{TARGET_ASM_FUNCTION_PROLOGUE} 4930and @code{TARGET_ASM_FUNCTION_EPILOGUE} are not invoked. 4931 4932The @var{thunk_fndecl} is redundant. (@var{delta} and @var{function} 4933have already been extracted from it.) It might possibly be useful on 4934some targets, but probably not. 4935 4936If you do not define this macro, the target-independent code in the C++ 4937front end will generate a less efficient heavyweight thunk that calls 4938@var{function} instead of jumping to it. The generic approach does 4939not support varargs. 4940@end deftypefn 4941 4942@deftypefn {Target Hook} bool TARGET_ASM_CAN_OUTPUT_MI_THUNK (const_tree @var{thunk_fndecl}, HOST_WIDE_INT @var{delta}, HOST_WIDE_INT @var{vcall_offset}, const_tree @var{function}) 4943A function that returns true if TARGET_ASM_OUTPUT_MI_THUNK would be able 4944to output the assembler code for the thunk function specified by the 4945arguments it is passed, and false otherwise. In the latter case, the 4946generic approach will be used by the C++ front end, with the limitations 4947previously exposed. 4948@end deftypefn 4949 4950@node Profiling 4951@subsection Generating Code for Profiling 4952@cindex profiling, code generation 4953 4954These macros will help you generate code for profiling. 4955 4956@defmac FUNCTION_PROFILER (@var{file}, @var{labelno}) 4957A C statement or compound statement to output to @var{file} some 4958assembler code to call the profiling subroutine @code{mcount}. 4959 4960@findex mcount 4961The details of how @code{mcount} expects to be called are determined by 4962your operating system environment, not by GCC@. To figure them out, 4963compile a small program for profiling using the system's installed C 4964compiler and look at the assembler code that results. 4965 4966Older implementations of @code{mcount} expect the address of a counter 4967variable to be loaded into some register. The name of this variable is 4968@samp{LP} followed by the number @var{labelno}, so you would generate 4969the name using @samp{LP%d} in a @code{fprintf}. 4970@end defmac 4971 4972@defmac PROFILE_HOOK 4973A C statement or compound statement to output to @var{file} some assembly 4974code to call the profiling subroutine @code{mcount} even the target does 4975not support profiling. 4976@end defmac 4977 4978@defmac NO_PROFILE_COUNTERS 4979Define this macro to be an expression with a nonzero value if the 4980@code{mcount} subroutine on your system does not need a counter variable 4981allocated for each function. This is true for almost all modern 4982implementations. If you define this macro, you must not use the 4983@var{labelno} argument to @code{FUNCTION_PROFILER}. 4984@end defmac 4985 4986@defmac PROFILE_BEFORE_PROLOGUE 4987Define this macro if the code for function profiling should come before 4988the function prologue. Normally, the profiling code comes after. 4989@end defmac 4990 4991@deftypefn {Target Hook} bool TARGET_KEEP_LEAF_WHEN_PROFILED (void) 4992This target hook returns true if the target wants the leaf flag for the current function to stay true even if it calls mcount. This might make sense for targets using the leaf flag only to determine whether a stack frame needs to be generated or not and for which the call to mcount is generated before the function prologue. 4993@end deftypefn 4994 4995@node Tail Calls 4996@subsection Permitting tail calls 4997@cindex tail calls 4998 4999@deftypefn {Target Hook} bool TARGET_FUNCTION_OK_FOR_SIBCALL (tree @var{decl}, tree @var{exp}) 5000True if it is OK to do sibling call optimization for the specified 5001call expression @var{exp}. @var{decl} will be the called function, 5002or @code{NULL} if this is an indirect call. 5003 5004It is not uncommon for limitations of calling conventions to prevent 5005tail calls to functions outside the current unit of translation, or 5006during PIC compilation. The hook is used to enforce these restrictions, 5007as the @code{sibcall} md pattern cannot fail, or fall over to a 5008``normal'' call. The criteria for successful sibling call optimization 5009may vary greatly between different architectures. 5010@end deftypefn 5011 5012@deftypefn {Target Hook} void TARGET_EXTRA_LIVE_ON_ENTRY (bitmap @var{regs}) 5013Add any hard registers to @var{regs} that are live on entry to the 5014function. This hook only needs to be defined to provide registers that 5015cannot be found by examination of FUNCTION_ARG_REGNO_P, the callee saved 5016registers, STATIC_CHAIN_INCOMING_REGNUM, STATIC_CHAIN_REGNUM, 5017TARGET_STRUCT_VALUE_RTX, FRAME_POINTER_REGNUM, EH_USES, 5018FRAME_POINTER_REGNUM, ARG_POINTER_REGNUM, and the PIC_OFFSET_TABLE_REGNUM. 5019@end deftypefn 5020 5021@deftypefn {Target Hook} void TARGET_SET_UP_BY_PROLOGUE (struct hard_reg_set_container *@var{}) 5022This hook should add additional registers that are computed by the prologue to the hard regset for shrink-wrapping optimization purposes. 5023@end deftypefn 5024 5025@deftypefn {Target Hook} bool TARGET_WARN_FUNC_RETURN (tree) 5026True if a function's return statements should be checked for matching the function's return type. This includes checking for falling off the end of a non-void function. Return false if no such check should be made. 5027@end deftypefn 5028 5029@node Shrink-wrapping separate components 5030@subsection Shrink-wrapping separate components 5031@cindex shrink-wrapping separate components 5032 5033The prologue may perform a variety of target dependent tasks such as 5034saving callee-saved registers, saving the return address, aligning the 5035stack, creating a stack frame, initializing the PIC register, setting 5036up the static chain, etc. 5037 5038On some targets some of these tasks may be independent of others and 5039thus may be shrink-wrapped separately. These independent tasks are 5040referred to as components and are handled generically by the target 5041independent parts of GCC. 5042 5043Using the following hooks those prologue or epilogue components can be 5044shrink-wrapped separately, so that the initialization (and possibly 5045teardown) those components do is not done as frequently on execution 5046paths where this would unnecessary. 5047 5048What exactly those components are is up to the target code; the generic 5049code treats them abstractly, as a bit in an @code{sbitmap}. These 5050@code{sbitmap}s are allocated by the @code{shrink_wrap.get_separate_components} 5051and @code{shrink_wrap.components_for_bb} hooks, and deallocated by the 5052generic code. 5053 5054@deftypefn {Target Hook} sbitmap TARGET_SHRINK_WRAP_GET_SEPARATE_COMPONENTS (void) 5055This hook should return an @code{sbitmap} with the bits set for those 5056components that can be separately shrink-wrapped in the current function. 5057Return @code{NULL} if the current function should not get any separate 5058shrink-wrapping. 5059Don't define this hook if it would always return @code{NULL}. 5060If it is defined, the other hooks in this group have to be defined as well. 5061@end deftypefn 5062 5063@deftypefn {Target Hook} sbitmap TARGET_SHRINK_WRAP_COMPONENTS_FOR_BB (basic_block) 5064This hook should return an @code{sbitmap} with the bits set for those 5065components where either the prologue component has to be executed before 5066the @code{basic_block}, or the epilogue component after it, or both. 5067@end deftypefn 5068 5069@deftypefn {Target Hook} void TARGET_SHRINK_WRAP_DISQUALIFY_COMPONENTS (sbitmap @var{components}, edge @var{e}, sbitmap @var{edge_components}, bool @var{is_prologue}) 5070This hook should clear the bits in the @var{components} bitmap for those 5071components in @var{edge_components} that the target cannot handle on edge 5072@var{e}, where @var{is_prologue} says if this is for a prologue or an 5073epilogue instead. 5074@end deftypefn 5075 5076@deftypefn {Target Hook} void TARGET_SHRINK_WRAP_EMIT_PROLOGUE_COMPONENTS (sbitmap) 5077Emit prologue insns for the components indicated by the parameter. 5078@end deftypefn 5079 5080@deftypefn {Target Hook} void TARGET_SHRINK_WRAP_EMIT_EPILOGUE_COMPONENTS (sbitmap) 5081Emit epilogue insns for the components indicated by the parameter. 5082@end deftypefn 5083 5084@deftypefn {Target Hook} void TARGET_SHRINK_WRAP_SET_HANDLED_COMPONENTS (sbitmap) 5085Mark the components in the parameter as handled, so that the 5086@code{prologue} and @code{epilogue} named patterns know to ignore those 5087components. The target code should not hang on to the @code{sbitmap}, it 5088will be deleted after this call. 5089@end deftypefn 5090 5091@node Stack Smashing Protection 5092@subsection Stack smashing protection 5093@cindex stack smashing protection 5094 5095@deftypefn {Target Hook} tree TARGET_STACK_PROTECT_GUARD (void) 5096This hook returns a @code{DECL} node for the external variable to use 5097for the stack protection guard. This variable is initialized by the 5098runtime to some random value and is used to initialize the guard value 5099that is placed at the top of the local stack frame. The type of this 5100variable must be @code{ptr_type_node}. 5101 5102The default version of this hook creates a variable called 5103@samp{__stack_chk_guard}, which is normally defined in @file{libgcc2.c}. 5104@end deftypefn 5105 5106@deftypefn {Target Hook} tree TARGET_STACK_PROTECT_FAIL (void) 5107This hook returns a @code{CALL_EXPR} that alerts the runtime that the 5108stack protect guard variable has been modified. This expression should 5109involve a call to a @code{noreturn} function. 5110 5111The default version of this hook invokes a function called 5112@samp{__stack_chk_fail}, taking no arguments. This function is 5113normally defined in @file{libgcc2.c}. 5114@end deftypefn 5115 5116@deftypefn {Target Hook} bool TARGET_STACK_PROTECT_RUNTIME_ENABLED_P (void) 5117Returns true if the target wants GCC's default stack protect runtime support, otherwise return false. The default implementation always returns true. 5118@end deftypefn 5119 5120@deftypefn {Common Target Hook} bool TARGET_SUPPORTS_SPLIT_STACK (bool @var{report}, struct gcc_options *@var{opts}) 5121Whether this target supports splitting the stack when the options described in @var{opts} have been passed. This is called after options have been parsed, so the target may reject splitting the stack in some configurations. The default version of this hook returns false. If @var{report} is true, this function may issue a warning or error; if @var{report} is false, it must simply return a value 5122@end deftypefn 5123 5124@deftypefn {Common Target Hook} {vec<const char *>} TARGET_GET_VALID_OPTION_VALUES (int @var{option_code}, const char *@var{prefix}) 5125The hook is used for options that have a non-trivial list of possible option values. OPTION_CODE is option code of opt_code enum type. PREFIX is used for bash completion and allows an implementation to return more specific completion based on the prefix. All string values should be allocated from heap memory and consumers should release them. The result will be pruned to cases with PREFIX if not NULL. 5126@end deftypefn 5127 5128@node Miscellaneous Register Hooks 5129@subsection Miscellaneous register hooks 5130@cindex miscellaneous register hooks 5131 5132@deftypevr {Target Hook} bool TARGET_CALL_FUSAGE_CONTAINS_NON_CALLEE_CLOBBERS 5133Set to true if each call that binds to a local definition explicitly 5134clobbers or sets all non-fixed registers modified by performing the call. 5135That is, by the call pattern itself, or by code that might be inserted by the 5136linker (e.g.@: stubs, veneers, branch islands), but not including those 5137modifiable by the callee. The affected registers may be mentioned explicitly 5138in the call pattern, or included as clobbers in CALL_INSN_FUNCTION_USAGE. 5139The default version of this hook is set to false. The purpose of this hook 5140is to enable the fipa-ra optimization. 5141@end deftypevr 5142 5143@node Varargs 5144@section Implementing the Varargs Macros 5145@cindex varargs implementation 5146 5147GCC comes with an implementation of @code{<varargs.h>} and 5148@code{<stdarg.h>} that work without change on machines that pass arguments 5149on the stack. Other machines require their own implementations of 5150varargs, and the two machine independent header files must have 5151conditionals to include it. 5152 5153ISO @code{<stdarg.h>} differs from traditional @code{<varargs.h>} mainly in 5154the calling convention for @code{va_start}. The traditional 5155implementation takes just one argument, which is the variable in which 5156to store the argument pointer. The ISO implementation of 5157@code{va_start} takes an additional second argument. The user is 5158supposed to write the last named argument of the function here. 5159 5160However, @code{va_start} should not use this argument. The way to find 5161the end of the named arguments is with the built-in functions described 5162below. 5163 5164@defmac __builtin_saveregs () 5165Use this built-in function to save the argument registers in memory so 5166that the varargs mechanism can access them. Both ISO and traditional 5167versions of @code{va_start} must use @code{__builtin_saveregs}, unless 5168you use @code{TARGET_SETUP_INCOMING_VARARGS} (see below) instead. 5169 5170On some machines, @code{__builtin_saveregs} is open-coded under the 5171control of the target hook @code{TARGET_EXPAND_BUILTIN_SAVEREGS}. On 5172other machines, it calls a routine written in assembler language, 5173found in @file{libgcc2.c}. 5174 5175Code generated for the call to @code{__builtin_saveregs} appears at the 5176beginning of the function, as opposed to where the call to 5177@code{__builtin_saveregs} is written, regardless of what the code is. 5178This is because the registers must be saved before the function starts 5179to use them for its own purposes. 5180@c i rewrote the first sentence above to fix an overfull hbox. --mew 5181@c 10feb93 5182@end defmac 5183 5184@defmac __builtin_next_arg (@var{lastarg}) 5185This builtin returns the address of the first anonymous stack 5186argument, as type @code{void *}. If @code{ARGS_GROW_DOWNWARD}, it 5187returns the address of the location above the first anonymous stack 5188argument. Use it in @code{va_start} to initialize the pointer for 5189fetching arguments from the stack. Also use it in @code{va_start} to 5190verify that the second parameter @var{lastarg} is the last named argument 5191of the current function. 5192@end defmac 5193 5194@defmac __builtin_classify_type (@var{object}) 5195Since each machine has its own conventions for which data types are 5196passed in which kind of register, your implementation of @code{va_arg} 5197has to embody these conventions. The easiest way to categorize the 5198specified data type is to use @code{__builtin_classify_type} together 5199with @code{sizeof} and @code{__alignof__}. 5200 5201@code{__builtin_classify_type} ignores the value of @var{object}, 5202considering only its data type. It returns an integer describing what 5203kind of type that is---integer, floating, pointer, structure, and so on. 5204 5205The file @file{typeclass.h} defines an enumeration that you can use to 5206interpret the values of @code{__builtin_classify_type}. 5207@end defmac 5208 5209These machine description macros help implement varargs: 5210 5211@deftypefn {Target Hook} rtx TARGET_EXPAND_BUILTIN_SAVEREGS (void) 5212If defined, this hook produces the machine-specific code for a call to 5213@code{__builtin_saveregs}. This code will be moved to the very 5214beginning of the function, before any parameter access are made. The 5215return value of this function should be an RTX that contains the value 5216to use as the return of @code{__builtin_saveregs}. 5217@end deftypefn 5218 5219@deftypefn {Target Hook} void TARGET_SETUP_INCOMING_VARARGS (cumulative_args_t @var{args_so_far}, const function_arg_info @var{&arg}, int *@var{pretend_args_size}, int @var{second_time}) 5220This target hook offers an alternative to using 5221@code{__builtin_saveregs} and defining the hook 5222@code{TARGET_EXPAND_BUILTIN_SAVEREGS}. Use it to store the anonymous 5223register arguments into the stack so that all the arguments appear to 5224have been passed consecutively on the stack. Once this is done, you can 5225use the standard implementation of varargs that works for machines that 5226pass all their arguments on the stack. 5227 5228The argument @var{args_so_far} points to the @code{CUMULATIVE_ARGS} data 5229structure, containing the values that are obtained after processing the 5230named arguments. The argument @var{arg} describes the last of these named 5231arguments. 5232 5233The target hook should do two things: first, push onto the stack all the 5234argument registers @emph{not} used for the named arguments, and second, 5235store the size of the data thus pushed into the @code{int}-valued 5236variable pointed to by @var{pretend_args_size}. The value that you 5237store here will serve as additional offset for setting up the stack 5238frame. 5239 5240Because you must generate code to push the anonymous arguments at 5241compile time without knowing their data types, 5242@code{TARGET_SETUP_INCOMING_VARARGS} is only useful on machines that 5243have just a single category of argument register and use it uniformly 5244for all data types. 5245 5246If the argument @var{second_time} is nonzero, it means that the 5247arguments of the function are being analyzed for the second time. This 5248happens for an inline function, which is not actually compiled until the 5249end of the source file. The hook @code{TARGET_SETUP_INCOMING_VARARGS} should 5250not generate any instructions in this case. 5251@end deftypefn 5252 5253@deftypefn {Target Hook} bool TARGET_STRICT_ARGUMENT_NAMING (cumulative_args_t @var{ca}) 5254Define this hook to return @code{true} if the location where a function 5255argument is passed depends on whether or not it is a named argument. 5256 5257This hook controls how the @var{named} argument to @code{TARGET_FUNCTION_ARG} 5258is set for varargs and stdarg functions. If this hook returns 5259@code{true}, the @var{named} argument is always true for named 5260arguments, and false for unnamed arguments. If it returns @code{false}, 5261but @code{TARGET_PRETEND_OUTGOING_VARARGS_NAMED} returns @code{true}, 5262then all arguments are treated as named. Otherwise, all named arguments 5263except the last are treated as named. 5264 5265You need not define this hook if it always returns @code{false}. 5266@end deftypefn 5267 5268@deftypefn {Target Hook} void TARGET_CALL_ARGS (rtx, @var{tree}) 5269While generating RTL for a function call, this target hook is invoked once 5270for each argument passed to the function, either a register returned by 5271@code{TARGET_FUNCTION_ARG} or a memory location. It is called just 5272before the point where argument registers are stored. The type of the 5273function to be called is also passed as the second argument; it is 5274@code{NULL_TREE} for libcalls. The @code{TARGET_END_CALL_ARGS} hook is 5275invoked just after the code to copy the return reg has been emitted. 5276This functionality can be used to perform special setup of call argument 5277registers if a target needs it. 5278For functions without arguments, the hook is called once with @code{pc_rtx} 5279passed instead of an argument register. 5280Most ports do not need to implement anything for this hook. 5281@end deftypefn 5282 5283@deftypefn {Target Hook} void TARGET_END_CALL_ARGS (void) 5284This target hook is invoked while generating RTL for a function call, 5285just after the point where the return reg is copied into a pseudo. It 5286signals that all the call argument and return registers for the just 5287emitted call are now no longer in use. 5288Most ports do not need to implement anything for this hook. 5289@end deftypefn 5290 5291@deftypefn {Target Hook} bool TARGET_PRETEND_OUTGOING_VARARGS_NAMED (cumulative_args_t @var{ca}) 5292If you need to conditionally change ABIs so that one works with 5293@code{TARGET_SETUP_INCOMING_VARARGS}, but the other works like neither 5294@code{TARGET_SETUP_INCOMING_VARARGS} nor @code{TARGET_STRICT_ARGUMENT_NAMING} was 5295defined, then define this hook to return @code{true} if 5296@code{TARGET_SETUP_INCOMING_VARARGS} is used, @code{false} otherwise. 5297Otherwise, you should not define this hook. 5298@end deftypefn 5299 5300@deftypefn {Target Hook} rtx TARGET_LOAD_BOUNDS_FOR_ARG (rtx @var{slot}, rtx @var{arg}, rtx @var{slot_no}) 5301This hook is used by expand pass to emit insn to load bounds of 5302@var{arg} passed in @var{slot}. Expand pass uses this hook in case 5303bounds of @var{arg} are not passed in register. If @var{slot} is a 5304memory, then bounds are loaded as for regular pointer loaded from 5305memory. If @var{slot} is not a memory then @var{slot_no} is an integer 5306constant holding number of the target dependent special slot which 5307should be used to obtain bounds. Hook returns RTX holding loaded bounds. 5308@end deftypefn 5309 5310@deftypefn {Target Hook} void TARGET_STORE_BOUNDS_FOR_ARG (rtx @var{arg}, rtx @var{slot}, rtx @var{bounds}, rtx @var{slot_no}) 5311This hook is used by expand pass to emit insns to store @var{bounds} of 5312@var{arg} passed in @var{slot}. Expand pass uses this hook in case 5313@var{bounds} of @var{arg} are not passed in register. If @var{slot} is a 5314memory, then @var{bounds} are stored as for regular pointer stored in 5315memory. If @var{slot} is not a memory then @var{slot_no} is an integer 5316constant holding number of the target dependent special slot which 5317should be used to store @var{bounds}. 5318@end deftypefn 5319 5320@deftypefn {Target Hook} rtx TARGET_LOAD_RETURNED_BOUNDS (rtx @var{slot}) 5321This hook is used by expand pass to emit insn to load bounds 5322returned by function call in @var{slot}. Hook returns RTX holding 5323loaded bounds. 5324@end deftypefn 5325 5326@deftypefn {Target Hook} void TARGET_STORE_RETURNED_BOUNDS (rtx @var{slot}, rtx @var{bounds}) 5327This hook is used by expand pass to emit insn to store @var{bounds} 5328returned by function call into @var{slot}. 5329@end deftypefn 5330 5331@node Trampolines 5332@section Support for Nested Functions 5333@cindex support for nested functions 5334@cindex trampolines for nested functions 5335@cindex descriptors for nested functions 5336@cindex nested functions, support for 5337 5338Taking the address of a nested function requires special compiler 5339handling to ensure that the static chain register is loaded when 5340the function is invoked via an indirect call. 5341 5342GCC has traditionally supported nested functions by creating an 5343executable @dfn{trampoline} at run time when the address of a nested 5344function is taken. This is a small piece of code which normally 5345resides on the stack, in the stack frame of the containing function. 5346The trampoline loads the static chain register and then jumps to the 5347real address of the nested function. 5348 5349The use of trampolines requires an executable stack, which is a 5350security risk. To avoid this problem, GCC also supports another 5351strategy: using descriptors for nested functions. Under this model, 5352taking the address of a nested function results in a pointer to a 5353non-executable function descriptor object. Initializing the static chain 5354from the descriptor is handled at indirect call sites. 5355 5356On some targets, including HPPA and IA-64, function descriptors may be 5357mandated by the ABI or be otherwise handled in a target-specific way 5358by the back end in its code generation strategy for indirect calls. 5359GCC also provides its own generic descriptor implementation to support the 5360@option{-fno-trampolines} option. In this case runtime detection of 5361function descriptors at indirect call sites relies on descriptor 5362pointers being tagged with a bit that is never set in bare function 5363addresses. Since GCC's generic function descriptors are 5364not ABI-compliant, this option is typically used only on a 5365per-language basis (notably by Ada) or when it can otherwise be 5366applied to the whole program. 5367 5368Define the following hook if your backend either implements ABI-specified 5369descriptor support, or can use GCC's generic descriptor implementation 5370for nested functions. 5371 5372@deftypevr {Target Hook} int TARGET_CUSTOM_FUNCTION_DESCRIPTORS 5373If the target can use GCC's generic descriptor mechanism for nested 5374functions, define this hook to a power of 2 representing an unused bit 5375in function pointers which can be used to differentiate descriptors at 5376run time. This value gives the number of bytes by which descriptor 5377pointers are misaligned compared to function pointers. For example, on 5378targets that require functions to be aligned to a 4-byte boundary, a 5379value of either 1 or 2 is appropriate unless the architecture already 5380reserves the bit for another purpose, such as on ARM. 5381 5382Define this hook to 0 if the target implements ABI support for 5383function descriptors in its standard calling sequence, like for example 5384HPPA or IA-64. 5385 5386Using descriptors for nested functions 5387eliminates the need for trampolines that reside on the stack and require 5388it to be made executable. 5389@end deftypevr 5390 5391The following macros tell GCC how to generate code to allocate and 5392initialize an executable trampoline. You can also use this interface 5393if your back end needs to create ABI-specified non-executable descriptors; in 5394this case the "trampoline" created is the descriptor containing data only. 5395 5396The instructions in an executable trampoline must do two things: load 5397a constant address into the static chain register, and jump to the real 5398address of the nested function. On CISC machines such as the m68k, 5399this requires two instructions, a move immediate and a jump. Then the 5400two addresses exist in the trampoline as word-long immediate operands. 5401On RISC machines, it is often necessary to load each address into a 5402register in two parts. Then pieces of each address form separate 5403immediate operands. 5404 5405The code generated to initialize the trampoline must store the variable 5406parts---the static chain value and the function address---into the 5407immediate operands of the instructions. On a CISC machine, this is 5408simply a matter of copying each address to a memory reference at the 5409proper offset from the start of the trampoline. On a RISC machine, it 5410may be necessary to take out pieces of the address and store them 5411separately. 5412 5413@deftypefn {Target Hook} void TARGET_ASM_TRAMPOLINE_TEMPLATE (FILE *@var{f}) 5414This hook is called by @code{assemble_trampoline_template} to output, 5415on the stream @var{f}, assembler code for a block of data that contains 5416the constant parts of a trampoline. This code should not include a 5417label---the label is taken care of automatically. 5418 5419If you do not define this hook, it means no template is needed 5420for the target. Do not define this hook on systems where the block move 5421code to copy the trampoline into place would be larger than the code 5422to generate it on the spot. 5423@end deftypefn 5424 5425@defmac TRAMPOLINE_SECTION 5426Return the section into which the trampoline template is to be placed 5427(@pxref{Sections}). The default value is @code{readonly_data_section}. 5428@end defmac 5429 5430@defmac TRAMPOLINE_SIZE 5431A C expression for the size in bytes of the trampoline, as an integer. 5432@end defmac 5433 5434@defmac TRAMPOLINE_ALIGNMENT 5435Alignment required for trampolines, in bits. 5436 5437If you don't define this macro, the value of @code{FUNCTION_ALIGNMENT} 5438is used for aligning trampolines. 5439@end defmac 5440 5441@deftypefn {Target Hook} void TARGET_TRAMPOLINE_INIT (rtx @var{m_tramp}, tree @var{fndecl}, rtx @var{static_chain}) 5442This hook is called to initialize a trampoline. 5443@var{m_tramp} is an RTX for the memory block for the trampoline; @var{fndecl} 5444is the @code{FUNCTION_DECL} for the nested function; @var{static_chain} is an 5445RTX for the static chain value that should be passed to the function 5446when it is called. 5447 5448If the target defines @code{TARGET_ASM_TRAMPOLINE_TEMPLATE}, then the 5449first thing this hook should do is emit a block move into @var{m_tramp} 5450from the memory block returned by @code{assemble_trampoline_template}. 5451Note that the block move need only cover the constant parts of the 5452trampoline. If the target isolates the variable parts of the trampoline 5453to the end, not all @code{TRAMPOLINE_SIZE} bytes need be copied. 5454 5455If the target requires any other actions, such as flushing caches or 5456enabling stack execution, these actions should be performed after 5457initializing the trampoline proper. 5458@end deftypefn 5459 5460@deftypefn {Target Hook} rtx TARGET_TRAMPOLINE_ADJUST_ADDRESS (rtx @var{addr}) 5461This hook should perform any machine-specific adjustment in 5462the address of the trampoline. Its argument contains the address of the 5463memory block that was passed to @code{TARGET_TRAMPOLINE_INIT}. In case 5464the address to be used for a function call should be different from the 5465address at which the template was stored, the different address should 5466be returned; otherwise @var{addr} should be returned unchanged. 5467If this hook is not defined, @var{addr} will be used for function calls. 5468@end deftypefn 5469 5470Implementing trampolines is difficult on many machines because they have 5471separate instruction and data caches. Writing into a stack location 5472fails to clear the memory in the instruction cache, so when the program 5473jumps to that location, it executes the old contents. 5474 5475Here are two possible solutions. One is to clear the relevant parts of 5476the instruction cache whenever a trampoline is set up. The other is to 5477make all trampolines identical, by having them jump to a standard 5478subroutine. The former technique makes trampoline execution faster; the 5479latter makes initialization faster. 5480 5481To clear the instruction cache when a trampoline is initialized, define 5482the following macro. 5483 5484@defmac CLEAR_INSN_CACHE (@var{beg}, @var{end}) 5485If defined, expands to a C expression clearing the @emph{instruction 5486cache} in the specified interval. The definition of this macro would 5487typically be a series of @code{asm} statements. Both @var{beg} and 5488@var{end} are both pointer expressions. 5489@end defmac 5490 5491To use a standard subroutine, define the following macro. In addition, 5492you must make sure that the instructions in a trampoline fill an entire 5493cache line with identical instructions, or else ensure that the 5494beginning of the trampoline code is always aligned at the same point in 5495its cache line. Look in @file{m68k.h} as a guide. 5496 5497@defmac TRANSFER_FROM_TRAMPOLINE 5498Define this macro if trampolines need a special subroutine to do their 5499work. The macro should expand to a series of @code{asm} statements 5500which will be compiled with GCC@. They go in a library function named 5501@code{__transfer_from_trampoline}. 5502 5503If you need to avoid executing the ordinary prologue code of a compiled 5504C function when you jump to the subroutine, you can do so by placing a 5505special label of your own in the assembler code. Use one @code{asm} 5506statement to generate an assembler label, and another to make the label 5507global. Then trampolines can use that label to jump directly to your 5508special assembler code. 5509@end defmac 5510 5511@node Library Calls 5512@section Implicit Calls to Library Routines 5513@cindex library subroutine names 5514@cindex @file{libgcc.a} 5515 5516@c prevent bad page break with this line 5517Here is an explanation of implicit calls to library routines. 5518 5519@defmac DECLARE_LIBRARY_RENAMES 5520This macro, if defined, should expand to a piece of C code that will get 5521expanded when compiling functions for libgcc.a. It can be used to 5522provide alternate names for GCC's internal library functions if there 5523are ABI-mandated names that the compiler should provide. 5524@end defmac 5525 5526@findex set_optab_libfunc 5527@findex init_one_libfunc 5528@deftypefn {Target Hook} void TARGET_INIT_LIBFUNCS (void) 5529This hook should declare additional library routines or rename 5530existing ones, using the functions @code{set_optab_libfunc} and 5531@code{init_one_libfunc} defined in @file{optabs.c}. 5532@code{init_optabs} calls this macro after initializing all the normal 5533library routines. 5534 5535The default is to do nothing. Most ports don't need to define this hook. 5536@end deftypefn 5537 5538@deftypevr {Target Hook} bool TARGET_LIBFUNC_GNU_PREFIX 5539If false (the default), internal library routines start with two 5540underscores. If set to true, these routines start with @code{__gnu_} 5541instead. E.g., @code{__muldi3} changes to @code{__gnu_muldi3}. This 5542currently only affects functions defined in @file{libgcc2.c}. If this 5543is set to true, the @file{tm.h} file must also 5544@code{#define LIBGCC2_GNU_PREFIX}. 5545@end deftypevr 5546 5547@defmac FLOAT_LIB_COMPARE_RETURNS_BOOL (@var{mode}, @var{comparison}) 5548This macro should return @code{true} if the library routine that 5549implements the floating point comparison operator @var{comparison} in 5550mode @var{mode} will return a boolean, and @var{false} if it will 5551return a tristate. 5552 5553GCC's own floating point libraries return tristates from the 5554comparison operators, so the default returns false always. Most ports 5555don't need to define this macro. 5556@end defmac 5557 5558@defmac TARGET_LIB_INT_CMP_BIASED 5559This macro should evaluate to @code{true} if the integer comparison 5560functions (like @code{__cmpdi2}) return 0 to indicate that the first 5561operand is smaller than the second, 1 to indicate that they are equal, 5562and 2 to indicate that the first operand is greater than the second. 5563If this macro evaluates to @code{false} the comparison functions return 5564@minus{}1, 0, and 1 instead of 0, 1, and 2. If the target uses the routines 5565in @file{libgcc.a}, you do not need to define this macro. 5566@end defmac 5567 5568@defmac TARGET_HAS_NO_HW_DIVIDE 5569This macro should be defined if the target has no hardware divide 5570instructions. If this macro is defined, GCC will use an algorithm which 5571make use of simple logical and arithmetic operations for 64-bit 5572division. If the macro is not defined, GCC will use an algorithm which 5573make use of a 64-bit by 32-bit divide primitive. 5574@end defmac 5575 5576@cindex @code{EDOM}, implicit usage 5577@findex matherr 5578@defmac TARGET_EDOM 5579The value of @code{EDOM} on the target machine, as a C integer constant 5580expression. If you don't define this macro, GCC does not attempt to 5581deposit the value of @code{EDOM} into @code{errno} directly. Look in 5582@file{/usr/include/errno.h} to find the value of @code{EDOM} on your 5583system. 5584 5585If you do not define @code{TARGET_EDOM}, then compiled code reports 5586domain errors by calling the library function and letting it report the 5587error. If mathematical functions on your system use @code{matherr} when 5588there is an error, then you should leave @code{TARGET_EDOM} undefined so 5589that @code{matherr} is used normally. 5590@end defmac 5591 5592@cindex @code{errno}, implicit usage 5593@defmac GEN_ERRNO_RTX 5594Define this macro as a C expression to create an rtl expression that 5595refers to the global ``variable'' @code{errno}. (On certain systems, 5596@code{errno} may not actually be a variable.) If you don't define this 5597macro, a reasonable default is used. 5598@end defmac 5599 5600@deftypefn {Target Hook} bool TARGET_LIBC_HAS_FUNCTION (enum function_class @var{fn_class}) 5601This hook determines whether a function from a class of functions 5602@var{fn_class} is present in the target C library. 5603@end deftypefn 5604 5605@deftypefn {Target Hook} bool TARGET_LIBC_HAS_FAST_FUNCTION (int @var{fcode}) 5606This hook determines whether a function from a class of functions 5607@code{(enum function_class)}@var{fcode} has a fast implementation. 5608@end deftypefn 5609 5610@defmac NEXT_OBJC_RUNTIME 5611Set this macro to 1 to use the "NeXT" Objective-C message sending conventions 5612by default. This calling convention involves passing the object, the selector 5613and the method arguments all at once to the method-lookup library function. 5614This is the usual setting when targeting Darwin/Mac OS X systems, which have 5615the NeXT runtime installed. 5616 5617If the macro is set to 0, the "GNU" Objective-C message sending convention 5618will be used by default. This convention passes just the object and the 5619selector to the method-lookup function, which returns a pointer to the method. 5620 5621In either case, it remains possible to select code-generation for the alternate 5622scheme, by means of compiler command line switches. 5623@end defmac 5624 5625@node Addressing Modes 5626@section Addressing Modes 5627@cindex addressing modes 5628 5629@c prevent bad page break with this line 5630This is about addressing modes. 5631 5632@defmac HAVE_PRE_INCREMENT 5633@defmacx HAVE_PRE_DECREMENT 5634@defmacx HAVE_POST_INCREMENT 5635@defmacx HAVE_POST_DECREMENT 5636A C expression that is nonzero if the machine supports pre-increment, 5637pre-decrement, post-increment, or post-decrement addressing respectively. 5638@end defmac 5639 5640@defmac HAVE_PRE_MODIFY_DISP 5641@defmacx HAVE_POST_MODIFY_DISP 5642A C expression that is nonzero if the machine supports pre- or 5643post-address side-effect generation involving constants other than 5644the size of the memory operand. 5645@end defmac 5646 5647@defmac HAVE_PRE_MODIFY_REG 5648@defmacx HAVE_POST_MODIFY_REG 5649A C expression that is nonzero if the machine supports pre- or 5650post-address side-effect generation involving a register displacement. 5651@end defmac 5652 5653@defmac CONSTANT_ADDRESS_P (@var{x}) 5654A C expression that is 1 if the RTX @var{x} is a constant which 5655is a valid address. On most machines the default definition of 5656@code{(CONSTANT_P (@var{x}) && GET_CODE (@var{x}) != CONST_DOUBLE)} 5657is acceptable, but a few machines are more restrictive as to which 5658constant addresses are supported. 5659@end defmac 5660 5661@defmac CONSTANT_P (@var{x}) 5662@code{CONSTANT_P}, which is defined by target-independent code, 5663accepts integer-values expressions whose values are not explicitly 5664known, such as @code{symbol_ref}, @code{label_ref}, and @code{high} 5665expressions and @code{const} arithmetic expressions, in addition to 5666@code{const_int} and @code{const_double} expressions. 5667@end defmac 5668 5669@defmac MAX_REGS_PER_ADDRESS 5670A number, the maximum number of registers that can appear in a valid 5671memory address. Note that it is up to you to specify a value equal to 5672the maximum number that @code{TARGET_LEGITIMATE_ADDRESS_P} would ever 5673accept. 5674@end defmac 5675 5676@deftypefn {Target Hook} bool TARGET_LEGITIMATE_ADDRESS_P (machine_mode @var{mode}, rtx @var{x}, bool @var{strict}) 5677A function that returns whether @var{x} (an RTX) is a legitimate memory 5678address on the target machine for a memory operand of mode @var{mode}. 5679 5680Legitimate addresses are defined in two variants: a strict variant and a 5681non-strict one. The @var{strict} parameter chooses which variant is 5682desired by the caller. 5683 5684The strict variant is used in the reload pass. It must be defined so 5685that any pseudo-register that has not been allocated a hard register is 5686considered a memory reference. This is because in contexts where some 5687kind of register is required, a pseudo-register with no hard register 5688must be rejected. For non-hard registers, the strict variant should look 5689up the @code{reg_renumber} array; it should then proceed using the hard 5690register number in the array, or treat the pseudo as a memory reference 5691if the array holds @code{-1}. 5692 5693The non-strict variant is used in other passes. It must be defined to 5694accept all pseudo-registers in every context where some kind of 5695register is required. 5696 5697Normally, constant addresses which are the sum of a @code{symbol_ref} 5698and an integer are stored inside a @code{const} RTX to mark them as 5699constant. Therefore, there is no need to recognize such sums 5700specifically as legitimate addresses. Normally you would simply 5701recognize any @code{const} as legitimate. 5702 5703Usually @code{PRINT_OPERAND_ADDRESS} is not prepared to handle constant 5704sums that are not marked with @code{const}. It assumes that a naked 5705@code{plus} indicates indexing. If so, then you @emph{must} reject such 5706naked constant sums as illegitimate addresses, so that none of them will 5707be given to @code{PRINT_OPERAND_ADDRESS}. 5708 5709@cindex @code{TARGET_ENCODE_SECTION_INFO} and address validation 5710On some machines, whether a symbolic address is legitimate depends on 5711the section that the address refers to. On these machines, define the 5712target hook @code{TARGET_ENCODE_SECTION_INFO} to store the information 5713into the @code{symbol_ref}, and then check for it here. When you see a 5714@code{const}, you will have to look inside it to find the 5715@code{symbol_ref} in order to determine the section. @xref{Assembler 5716Format}. 5717 5718@cindex @code{GO_IF_LEGITIMATE_ADDRESS} 5719Some ports are still using a deprecated legacy substitute for 5720this hook, the @code{GO_IF_LEGITIMATE_ADDRESS} macro. This macro 5721has this syntax: 5722 5723@example 5724#define GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{label}) 5725@end example 5726 5727@noindent 5728and should @code{goto @var{label}} if the address @var{x} is a valid 5729address on the target machine for a memory operand of mode @var{mode}. 5730 5731@findex REG_OK_STRICT 5732Compiler source files that want to use the strict variant of this 5733macro define the macro @code{REG_OK_STRICT}. You should use an 5734@code{#ifdef REG_OK_STRICT} conditional to define the strict variant in 5735that case and the non-strict variant otherwise. 5736 5737Using the hook is usually simpler because it limits the number of 5738files that are recompiled when changes are made. 5739@end deftypefn 5740 5741@defmac TARGET_MEM_CONSTRAINT 5742A single character to be used instead of the default @code{'m'} 5743character for general memory addresses. This defines the constraint 5744letter which matches the memory addresses accepted by 5745@code{TARGET_LEGITIMATE_ADDRESS_P}. Define this macro if you want to 5746support new address formats in your back end without changing the 5747semantics of the @code{'m'} constraint. This is necessary in order to 5748preserve functionality of inline assembly constructs using the 5749@code{'m'} constraint. 5750@end defmac 5751 5752@defmac FIND_BASE_TERM (@var{x}) 5753A C expression to determine the base term of address @var{x}, 5754or to provide a simplified version of @var{x} from which @file{alias.c} 5755can easily find the base term. This macro is used in only two places: 5756@code{find_base_value} and @code{find_base_term} in @file{alias.c}. 5757 5758It is always safe for this macro to not be defined. It exists so 5759that alias analysis can understand machine-dependent addresses. 5760 5761The typical use of this macro is to handle addresses containing 5762a label_ref or symbol_ref within an UNSPEC@. 5763@end defmac 5764 5765@deftypefn {Target Hook} rtx TARGET_LEGITIMIZE_ADDRESS (rtx @var{x}, rtx @var{oldx}, machine_mode @var{mode}) 5766This hook is given an invalid memory address @var{x} for an 5767operand of mode @var{mode} and should try to return a valid memory 5768address. 5769 5770@findex break_out_memory_refs 5771@var{x} will always be the result of a call to @code{break_out_memory_refs}, 5772and @var{oldx} will be the operand that was given to that function to produce 5773@var{x}. 5774 5775The code of the hook should not alter the substructure of 5776@var{x}. If it transforms @var{x} into a more legitimate form, it 5777should return the new @var{x}. 5778 5779It is not necessary for this hook to come up with a legitimate address, 5780with the exception of native TLS addresses (@pxref{Emulated TLS}). 5781The compiler has standard ways of doing so in all cases. In fact, if 5782the target supports only emulated TLS, it 5783is safe to omit this hook or make it return @var{x} if it cannot find 5784a valid way to legitimize the address. But often a machine-dependent 5785strategy can generate better code. 5786@end deftypefn 5787 5788@defmac LEGITIMIZE_RELOAD_ADDRESS (@var{x}, @var{mode}, @var{opnum}, @var{type}, @var{ind_levels}, @var{win}) 5789A C compound statement that attempts to replace @var{x}, which is an address 5790that needs reloading, with a valid memory address for an operand of mode 5791@var{mode}. @var{win} will be a C statement label elsewhere in the code. 5792It is not necessary to define this macro, but it might be useful for 5793performance reasons. 5794 5795For example, on the i386, it is sometimes possible to use a single 5796reload register instead of two by reloading a sum of two pseudo 5797registers into a register. On the other hand, for number of RISC 5798processors offsets are limited so that often an intermediate address 5799needs to be generated in order to address a stack slot. By defining 5800@code{LEGITIMIZE_RELOAD_ADDRESS} appropriately, the intermediate addresses 5801generated for adjacent some stack slots can be made identical, and thus 5802be shared. 5803 5804@emph{Note}: This macro should be used with caution. It is necessary 5805to know something of how reload works in order to effectively use this, 5806and it is quite easy to produce macros that build in too much knowledge 5807of reload internals. 5808 5809@emph{Note}: This macro must be able to reload an address created by a 5810previous invocation of this macro. If it fails to handle such addresses 5811then the compiler may generate incorrect code or abort. 5812 5813@findex push_reload 5814The macro definition should use @code{push_reload} to indicate parts that 5815need reloading; @var{opnum}, @var{type} and @var{ind_levels} are usually 5816suitable to be passed unaltered to @code{push_reload}. 5817 5818The code generated by this macro must not alter the substructure of 5819@var{x}. If it transforms @var{x} into a more legitimate form, it 5820should assign @var{x} (which will always be a C variable) a new value. 5821This also applies to parts that you change indirectly by calling 5822@code{push_reload}. 5823 5824@findex strict_memory_address_p 5825The macro definition may use @code{strict_memory_address_p} to test if 5826the address has become legitimate. 5827 5828@findex copy_rtx 5829If you want to change only a part of @var{x}, one standard way of doing 5830this is to use @code{copy_rtx}. Note, however, that it unshares only a 5831single level of rtl. Thus, if the part to be changed is not at the 5832top level, you'll need to replace first the top level. 5833It is not necessary for this macro to come up with a legitimate 5834address; but often a machine-dependent strategy can generate better code. 5835@end defmac 5836 5837@deftypefn {Target Hook} bool TARGET_MODE_DEPENDENT_ADDRESS_P (const_rtx @var{addr}, addr_space_t @var{addrspace}) 5838This hook returns @code{true} if memory address @var{addr} in address 5839space @var{addrspace} can have 5840different meanings depending on the machine mode of the memory 5841reference it is used for or if the address is valid for some modes 5842but not others. 5843 5844Autoincrement and autodecrement addresses typically have mode-dependent 5845effects because the amount of the increment or decrement is the size 5846of the operand being addressed. Some machines have other mode-dependent 5847addresses. Many RISC machines have no mode-dependent addresses. 5848 5849You may assume that @var{addr} is a valid address for the machine. 5850 5851The default version of this hook returns @code{false}. 5852@end deftypefn 5853 5854@deftypefn {Target Hook} bool TARGET_LEGITIMATE_CONSTANT_P (machine_mode @var{mode}, rtx @var{x}) 5855This hook returns true if @var{x} is a legitimate constant for a 5856@var{mode}-mode immediate operand on the target machine. You can assume that 5857@var{x} satisfies @code{CONSTANT_P}, so you need not check this. 5858 5859The default definition returns true. 5860@end deftypefn 5861 5862@deftypefn {Target Hook} rtx TARGET_DELEGITIMIZE_ADDRESS (rtx @var{x}) 5863This hook is used to undo the possibly obfuscating effects of the 5864@code{LEGITIMIZE_ADDRESS} and @code{LEGITIMIZE_RELOAD_ADDRESS} target 5865macros. Some backend implementations of these macros wrap symbol 5866references inside an @code{UNSPEC} rtx to represent PIC or similar 5867addressing modes. This target hook allows GCC's optimizers to understand 5868the semantics of these opaque @code{UNSPEC}s by converting them back 5869into their original form. 5870@end deftypefn 5871 5872@deftypefn {Target Hook} bool TARGET_CONST_NOT_OK_FOR_DEBUG_P (rtx @var{x}) 5873This hook should return true if @var{x} should not be emitted into 5874debug sections. 5875@end deftypefn 5876 5877@deftypefn {Target Hook} bool TARGET_CANNOT_FORCE_CONST_MEM (machine_mode @var{mode}, rtx @var{x}) 5878This hook should return true if @var{x} is of a form that cannot (or 5879should not) be spilled to the constant pool. @var{mode} is the mode 5880of @var{x}. 5881 5882The default version of this hook returns false. 5883 5884The primary reason to define this hook is to prevent reload from 5885deciding that a non-legitimate constant would be better reloaded 5886from the constant pool instead of spilling and reloading a register 5887holding the constant. This restriction is often true of addresses 5888of TLS symbols for various targets. 5889@end deftypefn 5890 5891@deftypefn {Target Hook} bool TARGET_USE_BLOCKS_FOR_CONSTANT_P (machine_mode @var{mode}, const_rtx @var{x}) 5892This hook should return true if pool entries for constant @var{x} can 5893be placed in an @code{object_block} structure. @var{mode} is the mode 5894of @var{x}. 5895 5896The default version returns false for all constants. 5897@end deftypefn 5898 5899@deftypefn {Target Hook} bool TARGET_USE_BLOCKS_FOR_DECL_P (const_tree @var{decl}) 5900This hook should return true if pool entries for @var{decl} should 5901be placed in an @code{object_block} structure. 5902 5903The default version returns true for all decls. 5904@end deftypefn 5905 5906@deftypefn {Target Hook} tree TARGET_BUILTIN_RECIPROCAL (tree @var{fndecl}) 5907This hook should return the DECL of a function that implements the 5908reciprocal of the machine-specific builtin function @var{fndecl}, or 5909@code{NULL_TREE} if such a function is not available. 5910@end deftypefn 5911 5912@deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_MASK_FOR_LOAD (void) 5913This hook should return the DECL of a function @var{f} that given an 5914address @var{addr} as an argument returns a mask @var{m} that can be 5915used to extract from two vectors the relevant data that resides in 5916@var{addr} in case @var{addr} is not properly aligned. 5917 5918The autovectorizer, when vectorizing a load operation from an address 5919@var{addr} that may be unaligned, will generate two vector loads from 5920the two aligned addresses around @var{addr}. It then generates a 5921@code{REALIGN_LOAD} operation to extract the relevant data from the 5922two loaded vectors. The first two arguments to @code{REALIGN_LOAD}, 5923@var{v1} and @var{v2}, are the two vectors, each of size @var{VS}, and 5924the third argument, @var{OFF}, defines how the data will be extracted 5925from these two vectors: if @var{OFF} is 0, then the returned vector is 5926@var{v2}; otherwise, the returned vector is composed from the last 5927@var{VS}-@var{OFF} elements of @var{v1} concatenated to the first 5928@var{OFF} elements of @var{v2}. 5929 5930If this hook is defined, the autovectorizer will generate a call 5931to @var{f} (using the DECL tree that this hook returns) and will 5932use the return value of @var{f} as the argument @var{OFF} to 5933@code{REALIGN_LOAD}. Therefore, the mask @var{m} returned by @var{f} 5934should comply with the semantics expected by @code{REALIGN_LOAD} 5935described above. 5936If this hook is not defined, then @var{addr} will be used as 5937the argument @var{OFF} to @code{REALIGN_LOAD}, in which case the low 5938log2(@var{VS}) @minus{} 1 bits of @var{addr} will be considered. 5939@end deftypefn 5940 5941@deftypefn {Target Hook} int TARGET_VECTORIZE_BUILTIN_VECTORIZATION_COST (enum vect_cost_for_stmt @var{type_of_cost}, tree @var{vectype}, int @var{misalign}) 5942Returns cost of different scalar or vector statements for vectorization cost model. 5943For vector memory operations the cost may depend on type (@var{vectype}) and 5944misalignment value (@var{misalign}). 5945@end deftypefn 5946 5947@deftypefn {Target Hook} poly_uint64 TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT (const_tree @var{type}) 5948This hook returns the preferred alignment in bits for accesses to 5949vectors of type @var{type} in vectorized code. This might be less than 5950or greater than the ABI-defined value returned by 5951@code{TARGET_VECTOR_ALIGNMENT}. It can be equal to the alignment of 5952a single element, in which case the vectorizer will not try to optimize 5953for alignment. 5954 5955The default hook returns @code{TYPE_ALIGN (@var{type})}, which is 5956correct for most targets. 5957@end deftypefn 5958 5959@deftypefn {Target Hook} bool TARGET_VECTORIZE_VECTOR_ALIGNMENT_REACHABLE (const_tree @var{type}, bool @var{is_packed}) 5960Return true if vector alignment is reachable (by peeling N iterations) for the given scalar type @var{type}. @var{is_packed} is false if the scalar access using @var{type} is known to be naturally aligned. 5961@end deftypefn 5962 5963@deftypefn {Target Hook} bool TARGET_VECTORIZE_VEC_PERM_CONST (machine_mode @var{mode}, rtx @var{output}, rtx @var{in0}, rtx @var{in1}, const vec_perm_indices @var{&sel}) 5964This hook is used to test whether the target can permute up to two 5965vectors of mode @var{mode} using the permutation vector @code{sel}, and 5966also to emit such a permutation. In the former case @var{in0}, @var{in1} 5967and @var{out} are all null. In the latter case @var{in0} and @var{in1} are 5968the source vectors and @var{out} is the destination vector; all three are 5969registers of mode @var{mode}. @var{in1} is the same as @var{in0} if 5970@var{sel} describes a permutation on one vector instead of two. 5971 5972Return true if the operation is possible, emitting instructions for it 5973if rtxes are provided. 5974 5975@cindex @code{vec_perm@var{m}} instruction pattern 5976If the hook returns false for a mode with multibyte elements, GCC will 5977try the equivalent byte operation. If that also fails, it will try forcing 5978the selector into a register and using the @var{vec_perm@var{mode}} 5979instruction pattern. There is no need for the hook to handle these two 5980implementation approaches itself. 5981@end deftypefn 5982 5983@deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_VECTORIZED_FUNCTION (unsigned @var{code}, tree @var{vec_type_out}, tree @var{vec_type_in}) 5984This hook should return the decl of a function that implements the 5985vectorized variant of the function with the @code{combined_fn} code 5986@var{code} or @code{NULL_TREE} if such a function is not available. 5987The return type of the vectorized function shall be of vector type 5988@var{vec_type_out} and the argument types should be @var{vec_type_in}. 5989@end deftypefn 5990 5991@deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_MD_VECTORIZED_FUNCTION (tree @var{fndecl}, tree @var{vec_type_out}, tree @var{vec_type_in}) 5992This hook should return the decl of a function that implements the 5993vectorized variant of target built-in function @code{fndecl}. The 5994return type of the vectorized function shall be of vector type 5995@var{vec_type_out} and the argument types should be @var{vec_type_in}. 5996@end deftypefn 5997 5998@deftypefn {Target Hook} bool TARGET_VECTORIZE_SUPPORT_VECTOR_MISALIGNMENT (machine_mode @var{mode}, const_tree @var{type}, int @var{misalignment}, bool @var{is_packed}) 5999This hook should return true if the target supports misaligned vector 6000store/load of a specific factor denoted in the @var{misalignment} 6001parameter. The vector store/load should be of machine mode @var{mode} and 6002the elements in the vectors should be of type @var{type}. @var{is_packed} 6003parameter is true if the memory access is defined in a packed struct. 6004@end deftypefn 6005 6006@deftypefn {Target Hook} machine_mode TARGET_VECTORIZE_PREFERRED_SIMD_MODE (scalar_mode @var{mode}) 6007This hook should return the preferred mode for vectorizing scalar 6008mode @var{mode}. The default is 6009equal to @code{word_mode}, because the vectorizer can do some 6010transformations even in absence of specialized @acronym{SIMD} hardware. 6011@end deftypefn 6012 6013@deftypefn {Target Hook} machine_mode TARGET_VECTORIZE_SPLIT_REDUCTION (machine_mode) 6014This hook should return the preferred mode to split the final reduction 6015step on @var{mode} to. The reduction is then carried out reducing upper 6016against lower halves of vectors recursively until the specified mode is 6017reached. The default is @var{mode} which means no splitting. 6018@end deftypefn 6019 6020@deftypefn {Target Hook} {unsigned int} TARGET_VECTORIZE_AUTOVECTORIZE_VECTOR_MODES (vector_modes *@var{modes}, bool @var{all}) 6021If using the mode returned by @code{TARGET_VECTORIZE_PREFERRED_SIMD_MODE} 6022is not the only approach worth considering, this hook should add one mode to 6023@var{modes} for each useful alternative approach. These modes are then 6024passed to @code{TARGET_VECTORIZE_RELATED_MODE} to obtain the vector mode 6025for a given element mode. 6026 6027The modes returned in @var{modes} should use the smallest element mode 6028possible for the vectorization approach that they represent, preferring 6029integer modes over floating-poing modes in the event of a tie. The first 6030mode should be the @code{TARGET_VECTORIZE_PREFERRED_SIMD_MODE} for its 6031element mode. 6032 6033If @var{all} is true, add suitable vector modes even when they are generally 6034not expected to be worthwhile. 6035 6036The hook returns a bitmask of flags that control how the modes in 6037@var{modes} are used. The flags are: 6038@table @code 6039@item VECT_COMPARE_COSTS 6040Tells the loop vectorizer to try all the provided modes and pick the one 6041with the lowest cost. By default the vectorizer will choose the first 6042mode that works. 6043@end table 6044 6045The hook does not need to do anything if the vector returned by 6046@code{TARGET_VECTORIZE_PREFERRED_SIMD_MODE} is the only one relevant 6047for autovectorization. The default implementation adds no modes and 6048returns 0. 6049@end deftypefn 6050 6051@deftypefn {Target Hook} opt_machine_mode TARGET_VECTORIZE_RELATED_MODE (machine_mode @var{vector_mode}, scalar_mode @var{element_mode}, poly_uint64 @var{nunits}) 6052If a piece of code is using vector mode @var{vector_mode} and also wants 6053to operate on elements of mode @var{element_mode}, return the vector mode 6054it should use for those elements. If @var{nunits} is nonzero, ensure that 6055the mode has exactly @var{nunits} elements, otherwise pick whichever vector 6056size pairs the most naturally with @var{vector_mode}. Return an empty 6057@code{opt_machine_mode} if there is no supported vector mode with the 6058required properties. 6059 6060There is no prescribed way of handling the case in which @var{nunits} 6061is zero. One common choice is to pick a vector mode with the same size 6062as @var{vector_mode}; this is the natural choice if the target has a 6063fixed vector size. Another option is to choose a vector mode with the 6064same number of elements as @var{vector_mode}; this is the natural choice 6065if the target has a fixed number of elements. Alternatively, the hook 6066might choose a middle ground, such as trying to keep the number of 6067elements as similar as possible while applying maximum and minimum 6068vector sizes. 6069 6070The default implementation uses @code{mode_for_vector} to find the 6071requested mode, returning a mode with the same size as @var{vector_mode} 6072when @var{nunits} is zero. This is the correct behavior for most targets. 6073@end deftypefn 6074 6075@deftypefn {Target Hook} opt_machine_mode TARGET_VECTORIZE_GET_MASK_MODE (machine_mode @var{mode}) 6076Return the mode to use for a vector mask that holds one boolean 6077result for each element of vector mode @var{mode}. The returned mask mode 6078can be a vector of integers (class @code{MODE_VECTOR_INT}), a vector of 6079booleans (class @code{MODE_VECTOR_BOOL}) or a scalar integer (class 6080@code{MODE_INT}). Return an empty @code{opt_machine_mode} if no such 6081mask mode exists. 6082 6083The default implementation returns a @code{MODE_VECTOR_INT} with the 6084same size and number of elements as @var{mode}, if such a mode exists. 6085@end deftypefn 6086 6087@deftypefn {Target Hook} bool TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE (unsigned @var{ifn}) 6088This hook returns true if masked internal function @var{ifn} (really of 6089type @code{internal_fn}) should be considered expensive when the mask is 6090all zeros. GCC can then try to branch around the instruction instead. 6091@end deftypefn 6092 6093@deftypefn {Target Hook} {void *} TARGET_VECTORIZE_INIT_COST (class loop *@var{loop_info}) 6094This hook should initialize target-specific data structures in preparation for modeling the costs of vectorizing a loop or basic block. The default allocates three unsigned integers for accumulating costs for the prologue, body, and epilogue of the loop or basic block. If @var{loop_info} is non-NULL, it identifies the loop being vectorized; otherwise a single block is being vectorized. 6095@end deftypefn 6096 6097@deftypefn {Target Hook} unsigned TARGET_VECTORIZE_ADD_STMT_COST (void *@var{data}, int @var{count}, enum vect_cost_for_stmt @var{kind}, class _stmt_vec_info *@var{stmt_info}, int @var{misalign}, enum vect_cost_model_location @var{where}) 6098This hook should update the target-specific @var{data} in response to adding @var{count} copies of the given @var{kind} of statement to a loop or basic block. The default adds the builtin vectorizer cost for the copies of the statement to the accumulator specified by @var{where}, (the prologue, body, or epilogue) and returns the amount added. The return value should be viewed as a tentative cost that may later be revised. 6099@end deftypefn 6100 6101@deftypefn {Target Hook} void TARGET_VECTORIZE_FINISH_COST (void *@var{data}, unsigned *@var{prologue_cost}, unsigned *@var{body_cost}, unsigned *@var{epilogue_cost}) 6102This hook should complete calculations of the cost of vectorizing a loop or basic block based on @var{data}, and return the prologue, body, and epilogue costs as unsigned integers. The default returns the value of the three accumulators. 6103@end deftypefn 6104 6105@deftypefn {Target Hook} void TARGET_VECTORIZE_DESTROY_COST_DATA (void *@var{data}) 6106This hook should release @var{data} and any related data structures allocated by TARGET_VECTORIZE_INIT_COST. The default releases the accumulator. 6107@end deftypefn 6108 6109@deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_GATHER (const_tree @var{mem_vectype}, const_tree @var{index_type}, int @var{scale}) 6110Target builtin that implements vector gather operation. @var{mem_vectype} 6111is the vector type of the load and @var{index_type} is scalar type of 6112the index, scaled by @var{scale}. 6113The default is @code{NULL_TREE} which means to not vectorize gather 6114loads. 6115@end deftypefn 6116 6117@deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_SCATTER (const_tree @var{vectype}, const_tree @var{index_type}, int @var{scale}) 6118Target builtin that implements vector scatter operation. @var{vectype} 6119is the vector type of the store and @var{index_type} is scalar type of 6120the index, scaled by @var{scale}. 6121The default is @code{NULL_TREE} which means to not vectorize scatter 6122stores. 6123@end deftypefn 6124 6125@deftypefn {Target Hook} int TARGET_SIMD_CLONE_COMPUTE_VECSIZE_AND_SIMDLEN (struct cgraph_node *@var{}, struct cgraph_simd_clone *@var{}, @var{tree}, @var{int}) 6126This hook should set @var{vecsize_mangle}, @var{vecsize_int}, @var{vecsize_float} 6127fields in @var{simd_clone} structure pointed by @var{clone_info} argument and also 6128@var{simdlen} field if it was previously 0. 6129The hook should return 0 if SIMD clones shouldn't be emitted, 6130or number of @var{vecsize_mangle} variants that should be emitted. 6131@end deftypefn 6132 6133@deftypefn {Target Hook} void TARGET_SIMD_CLONE_ADJUST (struct cgraph_node *@var{}) 6134This hook should add implicit @code{attribute(target("..."))} attribute 6135to SIMD clone @var{node} if needed. 6136@end deftypefn 6137 6138@deftypefn {Target Hook} int TARGET_SIMD_CLONE_USABLE (struct cgraph_node *@var{}) 6139This hook should return -1 if SIMD clone @var{node} shouldn't be used 6140in vectorized loops in current function, or non-negative number if it is 6141usable. In that case, the smaller the number is, the more desirable it is 6142to use it. 6143@end deftypefn 6144 6145@deftypefn {Target Hook} int TARGET_SIMT_VF (void) 6146Return number of threads in SIMT thread group on the target. 6147@end deftypefn 6148 6149@deftypefn {Target Hook} int TARGET_OMP_DEVICE_KIND_ARCH_ISA (enum omp_device_kind_arch_isa @var{trait}, const char *@var{name}) 6150Return 1 if @var{trait} @var{name} is present in the OpenMP context's 6151device trait set, return 0 if not present in any OpenMP context in the 6152whole translation unit, or -1 if not present in the current OpenMP context 6153but might be present in another OpenMP context in the same TU. 6154@end deftypefn 6155 6156@deftypefn {Target Hook} bool TARGET_GOACC_VALIDATE_DIMS (tree @var{decl}, int *@var{dims}, int @var{fn_level}, unsigned @var{used}) 6157This hook should check the launch dimensions provided for an OpenACC 6158compute region, or routine. Defaulted values are represented as -1 6159and non-constant values as 0. The @var{fn_level} is negative for the 6160function corresponding to the compute region. For a routine it is the 6161outermost level at which partitioned execution may be spawned. The hook 6162should verify non-default values. If DECL is NULL, global defaults 6163are being validated and unspecified defaults should be filled in. 6164Diagnostics should be issued as appropriate. Return 6165true, if changes have been made. You must override this hook to 6166provide dimensions larger than 1. 6167@end deftypefn 6168 6169@deftypefn {Target Hook} int TARGET_GOACC_DIM_LIMIT (int @var{axis}) 6170This hook should return the maximum size of a particular dimension, 6171or zero if unbounded. 6172@end deftypefn 6173 6174@deftypefn {Target Hook} bool TARGET_GOACC_FORK_JOIN (gcall *@var{call}, const int *@var{dims}, bool @var{is_fork}) 6175This hook can be used to convert IFN_GOACC_FORK and IFN_GOACC_JOIN 6176function calls to target-specific gimple, or indicate whether they 6177should be retained. It is executed during the oacc_device_lower pass. 6178It should return true, if the call should be retained. It should 6179return false, if it is to be deleted (either because target-specific 6180gimple has been inserted before it, or there is no need for it). 6181The default hook returns false, if there are no RTL expanders for them. 6182@end deftypefn 6183 6184@deftypefn {Target Hook} void TARGET_GOACC_REDUCTION (gcall *@var{call}) 6185This hook is used by the oacc_transform pass to expand calls to the 6186@var{GOACC_REDUCTION} internal function, into a sequence of gimple 6187instructions. @var{call} is gimple statement containing the call to 6188the function. This hook removes statement @var{call} after the 6189expanded sequence has been inserted. This hook is also responsible 6190for allocating any storage for reductions when necessary. 6191@end deftypefn 6192 6193@deftypefn {Target Hook} tree TARGET_PREFERRED_ELSE_VALUE (unsigned @var{ifn}, tree @var{type}, unsigned @var{nops}, tree *@var{ops}) 6194This hook returns the target's preferred final argument for a call 6195to conditional internal function @var{ifn} (really of type 6196@code{internal_fn}). @var{type} specifies the return type of the 6197function and @var{ops} are the operands to the conditional operation, 6198of which there are @var{nops}. 6199 6200For example, if @var{ifn} is @code{IFN_COND_ADD}, the hook returns 6201a value of type @var{type} that should be used when @samp{@var{ops}[0]} 6202and @samp{@var{ops}[1]} are conditionally added together. 6203 6204This hook is only relevant if the target supports conditional patterns 6205like @code{cond_add@var{m}}. The default implementation returns a zero 6206constant of type @var{type}. 6207@end deftypefn 6208 6209@node Anchored Addresses 6210@section Anchored Addresses 6211@cindex anchored addresses 6212@cindex @option{-fsection-anchors} 6213 6214GCC usually addresses every static object as a separate entity. 6215For example, if we have: 6216 6217@smallexample 6218static int a, b, c; 6219int foo (void) @{ return a + b + c; @} 6220@end smallexample 6221 6222the code for @code{foo} will usually calculate three separate symbolic 6223addresses: those of @code{a}, @code{b} and @code{c}. On some targets, 6224it would be better to calculate just one symbolic address and access 6225the three variables relative to it. The equivalent pseudocode would 6226be something like: 6227 6228@smallexample 6229int foo (void) 6230@{ 6231 register int *xr = &x; 6232 return xr[&a - &x] + xr[&b - &x] + xr[&c - &x]; 6233@} 6234@end smallexample 6235 6236(which isn't valid C). We refer to shared addresses like @code{x} as 6237``section anchors''. Their use is controlled by @option{-fsection-anchors}. 6238 6239The hooks below describe the target properties that GCC needs to know 6240in order to make effective use of section anchors. It won't use 6241section anchors at all unless either @code{TARGET_MIN_ANCHOR_OFFSET} 6242or @code{TARGET_MAX_ANCHOR_OFFSET} is set to a nonzero value. 6243 6244@deftypevr {Target Hook} HOST_WIDE_INT TARGET_MIN_ANCHOR_OFFSET 6245The minimum offset that should be applied to a section anchor. 6246On most targets, it should be the smallest offset that can be 6247applied to a base register while still giving a legitimate address 6248for every mode. The default value is 0. 6249@end deftypevr 6250 6251@deftypevr {Target Hook} HOST_WIDE_INT TARGET_MAX_ANCHOR_OFFSET 6252Like @code{TARGET_MIN_ANCHOR_OFFSET}, but the maximum (inclusive) 6253offset that should be applied to section anchors. The default 6254value is 0. 6255@end deftypevr 6256 6257@deftypefn {Target Hook} void TARGET_ASM_OUTPUT_ANCHOR (rtx @var{x}) 6258Write the assembly code to define section anchor @var{x}, which is a 6259@code{SYMBOL_REF} for which @samp{SYMBOL_REF_ANCHOR_P (@var{x})} is true. 6260The hook is called with the assembly output position set to the beginning 6261of @code{SYMBOL_REF_BLOCK (@var{x})}. 6262 6263If @code{ASM_OUTPUT_DEF} is available, the hook's default definition uses 6264it to define the symbol as @samp{. + SYMBOL_REF_BLOCK_OFFSET (@var{x})}. 6265If @code{ASM_OUTPUT_DEF} is not available, the hook's default definition 6266is @code{NULL}, which disables the use of section anchors altogether. 6267@end deftypefn 6268 6269@deftypefn {Target Hook} bool TARGET_USE_ANCHORS_FOR_SYMBOL_P (const_rtx @var{x}) 6270Return true if GCC should attempt to use anchors to access @code{SYMBOL_REF} 6271@var{x}. You can assume @samp{SYMBOL_REF_HAS_BLOCK_INFO_P (@var{x})} and 6272@samp{!SYMBOL_REF_ANCHOR_P (@var{x})}. 6273 6274The default version is correct for most targets, but you might need to 6275intercept this hook to handle things like target-specific attributes 6276or target-specific sections. 6277@end deftypefn 6278 6279@node Condition Code 6280@section Condition Code Status 6281@cindex condition code status 6282 6283The macros in this section can be split in two families, according to the 6284two ways of representing condition codes in GCC. 6285 6286The first representation is the so called @code{(cc0)} representation 6287(@pxref{Jump Patterns}), where all instructions can have an implicit 6288clobber of the condition codes. The second is the condition code 6289register representation, which provides better schedulability for 6290architectures that do have a condition code register, but on which 6291most instructions do not affect it. The latter category includes 6292most RISC machines. 6293 6294The implicit clobbering poses a strong restriction on the placement of 6295the definition and use of the condition code. In the past the definition 6296and use were always adjacent. However, recent changes to support trapping 6297arithmatic may result in the definition and user being in different blocks. 6298Thus, there may be a @code{NOTE_INSN_BASIC_BLOCK} between them. Additionally, 6299the definition may be the source of exception handling edges. 6300 6301These restrictions can prevent important 6302optimizations on some machines. For example, on the IBM RS/6000, there 6303is a delay for taken branches unless the condition code register is set 6304three instructions earlier than the conditional branch. The instruction 6305scheduler cannot perform this optimization if it is not permitted to 6306separate the definition and use of the condition code register. 6307 6308For this reason, it is possible and suggested to use a register to 6309represent the condition code for new ports. If there is a specific 6310condition code register in the machine, use a hard register. If the 6311condition code or comparison result can be placed in any general register, 6312or if there are multiple condition registers, use a pseudo register. 6313Registers used to store the condition code value will usually have a mode 6314that is in class @code{MODE_CC}. 6315 6316Alternatively, you can use @code{BImode} if the comparison operator is 6317specified already in the compare instruction. In this case, you are not 6318interested in most macros in this section. 6319 6320@menu 6321* CC0 Condition Codes:: Old style representation of condition codes. 6322* MODE_CC Condition Codes:: Modern representation of condition codes. 6323@end menu 6324 6325@node CC0 Condition Codes 6326@subsection Representation of condition codes using @code{(cc0)} 6327@findex cc0 6328 6329@findex cc_status 6330The file @file{conditions.h} defines a variable @code{cc_status} to 6331describe how the condition code was computed (in case the interpretation of 6332the condition code depends on the instruction that it was set by). This 6333variable contains the RTL expressions on which the condition code is 6334currently based, and several standard flags. 6335 6336Sometimes additional machine-specific flags must be defined in the machine 6337description header file. It can also add additional machine-specific 6338information by defining @code{CC_STATUS_MDEP}. 6339 6340@defmac CC_STATUS_MDEP 6341C code for a data type which is used for declaring the @code{mdep} 6342component of @code{cc_status}. It defaults to @code{int}. 6343 6344This macro is not used on machines that do not use @code{cc0}. 6345@end defmac 6346 6347@defmac CC_STATUS_MDEP_INIT 6348A C expression to initialize the @code{mdep} field to ``empty''. 6349The default definition does nothing, since most machines don't use 6350the field anyway. If you want to use the field, you should probably 6351define this macro to initialize it. 6352 6353This macro is not used on machines that do not use @code{cc0}. 6354@end defmac 6355 6356@defmac NOTICE_UPDATE_CC (@var{exp}, @var{insn}) 6357A C compound statement to set the components of @code{cc_status} 6358appropriately for an insn @var{insn} whose body is @var{exp}. It is 6359this macro's responsibility to recognize insns that set the condition 6360code as a byproduct of other activity as well as those that explicitly 6361set @code{(cc0)}. 6362 6363This macro is not used on machines that do not use @code{cc0}. 6364 6365If there are insns that do not set the condition code but do alter 6366other machine registers, this macro must check to see whether they 6367invalidate the expressions that the condition code is recorded as 6368reflecting. For example, on the 68000, insns that store in address 6369registers do not set the condition code, which means that usually 6370@code{NOTICE_UPDATE_CC} can leave @code{cc_status} unaltered for such 6371insns. But suppose that the previous insn set the condition code 6372based on location @samp{a4@@(102)} and the current insn stores a new 6373value in @samp{a4}. Although the condition code is not changed by 6374this, it will no longer be true that it reflects the contents of 6375@samp{a4@@(102)}. Therefore, @code{NOTICE_UPDATE_CC} must alter 6376@code{cc_status} in this case to say that nothing is known about the 6377condition code value. 6378 6379The definition of @code{NOTICE_UPDATE_CC} must be prepared to deal 6380with the results of peephole optimization: insns whose patterns are 6381@code{parallel} RTXs containing various @code{reg}, @code{mem} or 6382constants which are just the operands. The RTL structure of these 6383insns is not sufficient to indicate what the insns actually do. What 6384@code{NOTICE_UPDATE_CC} should do when it sees one is just to run 6385@code{CC_STATUS_INIT}. 6386 6387A possible definition of @code{NOTICE_UPDATE_CC} is to call a function 6388that looks at an attribute (@pxref{Insn Attributes}) named, for example, 6389@samp{cc}. This avoids having detailed information about patterns in 6390two places, the @file{md} file and in @code{NOTICE_UPDATE_CC}. 6391@end defmac 6392 6393@node MODE_CC Condition Codes 6394@subsection Representation of condition codes using registers 6395@findex CCmode 6396@findex MODE_CC 6397 6398@defmac SELECT_CC_MODE (@var{op}, @var{x}, @var{y}) 6399On many machines, the condition code may be produced by other instructions 6400than compares, for example the branch can use directly the condition 6401code set by a subtract instruction. However, on some machines 6402when the condition code is set this way some bits (such as the overflow 6403bit) are not set in the same way as a test instruction, so that a different 6404branch instruction must be used for some conditional branches. When 6405this happens, use the machine mode of the condition code register to 6406record different formats of the condition code register. Modes can 6407also be used to record which compare instruction (e.g.@: a signed or an 6408unsigned comparison) produced the condition codes. 6409 6410If other modes than @code{CCmode} are required, add them to 6411@file{@var{machine}-modes.def} and define @code{SELECT_CC_MODE} to choose 6412a mode given an operand of a compare. This is needed because the modes 6413have to be chosen not only during RTL generation but also, for example, 6414by instruction combination. The result of @code{SELECT_CC_MODE} should 6415be consistent with the mode used in the patterns; for example to support 6416the case of the add on the SPARC discussed above, we have the pattern 6417 6418@smallexample 6419(define_insn "" 6420 [(set (reg:CCNZ 0) 6421 (compare:CCNZ 6422 (plus:SI (match_operand:SI 0 "register_operand" "%r") 6423 (match_operand:SI 1 "arith_operand" "rI")) 6424 (const_int 0)))] 6425 "" 6426 "@dots{}") 6427@end smallexample 6428 6429@noindent 6430together with a @code{SELECT_CC_MODE} that returns @code{CCNZmode} 6431for comparisons whose argument is a @code{plus}: 6432 6433@smallexample 6434#define SELECT_CC_MODE(OP,X,Y) \ 6435 (GET_MODE_CLASS (GET_MODE (X)) == MODE_FLOAT \ 6436 ? ((OP == LT || OP == LE || OP == GT || OP == GE) \ 6437 ? CCFPEmode : CCFPmode) \ 6438 : ((GET_CODE (X) == PLUS || GET_CODE (X) == MINUS \ 6439 || GET_CODE (X) == NEG || GET_CODE (x) == ASHIFT) \ 6440 ? CCNZmode : CCmode)) 6441@end smallexample 6442 6443Another reason to use modes is to retain information on which operands 6444were used by the comparison; see @code{REVERSIBLE_CC_MODE} later in 6445this section. 6446 6447You should define this macro if and only if you define extra CC modes 6448in @file{@var{machine}-modes.def}. 6449@end defmac 6450 6451@deftypefn {Target Hook} void TARGET_CANONICALIZE_COMPARISON (int *@var{code}, rtx *@var{op0}, rtx *@var{op1}, bool @var{op0_preserve_value}) 6452On some machines not all possible comparisons are defined, but you can 6453convert an invalid comparison into a valid one. For example, the Alpha 6454does not have a @code{GT} comparison, but you can use an @code{LT} 6455comparison instead and swap the order of the operands. 6456 6457On such machines, implement this hook to do any required conversions. 6458@var{code} is the initial comparison code and @var{op0} and @var{op1} 6459are the left and right operands of the comparison, respectively. If 6460@var{op0_preserve_value} is @code{true} the implementation is not 6461allowed to change the value of @var{op0} since the value might be used 6462in RTXs which aren't comparisons. E.g. the implementation is not 6463allowed to swap operands in that case. 6464 6465GCC will not assume that the comparison resulting from this macro is 6466valid but will see if the resulting insn matches a pattern in the 6467@file{md} file. 6468 6469You need not to implement this hook if it would never change the 6470comparison code or operands. 6471@end deftypefn 6472 6473@defmac REVERSIBLE_CC_MODE (@var{mode}) 6474A C expression whose value is one if it is always safe to reverse a 6475comparison whose mode is @var{mode}. If @code{SELECT_CC_MODE} 6476can ever return @var{mode} for a floating-point inequality comparison, 6477then @code{REVERSIBLE_CC_MODE (@var{mode})} must be zero. 6478 6479You need not define this macro if it would always returns zero or if the 6480floating-point format is anything other than @code{IEEE_FLOAT_FORMAT}. 6481For example, here is the definition used on the SPARC, where floating-point 6482inequality comparisons are given either @code{CCFPEmode} or @code{CCFPmode}: 6483 6484@smallexample 6485#define REVERSIBLE_CC_MODE(MODE) \ 6486 ((MODE) != CCFPEmode && (MODE) != CCFPmode) 6487@end smallexample 6488@end defmac 6489 6490@defmac REVERSE_CONDITION (@var{code}, @var{mode}) 6491A C expression whose value is reversed condition code of the @var{code} for 6492comparison done in CC_MODE @var{mode}. The macro is used only in case 6493@code{REVERSIBLE_CC_MODE (@var{mode})} is nonzero. Define this macro in case 6494machine has some non-standard way how to reverse certain conditionals. For 6495instance in case all floating point conditions are non-trapping, compiler may 6496freely convert unordered compares to ordered ones. Then definition may look 6497like: 6498 6499@smallexample 6500#define REVERSE_CONDITION(CODE, MODE) \ 6501 ((MODE) != CCFPmode ? reverse_condition (CODE) \ 6502 : reverse_condition_maybe_unordered (CODE)) 6503@end smallexample 6504@end defmac 6505 6506@deftypefn {Target Hook} bool TARGET_FIXED_CONDITION_CODE_REGS (unsigned int *@var{p1}, unsigned int *@var{p2}) 6507On targets which do not use @code{(cc0)}, and which use a hard 6508register rather than a pseudo-register to hold condition codes, the 6509regular CSE passes are often not able to identify cases in which the 6510hard register is set to a common value. Use this hook to enable a 6511small pass which optimizes such cases. This hook should return true 6512to enable this pass, and it should set the integers to which its 6513arguments point to the hard register numbers used for condition codes. 6514When there is only one such register, as is true on most systems, the 6515integer pointed to by @var{p2} should be set to 6516@code{INVALID_REGNUM}. 6517 6518The default version of this hook returns false. 6519@end deftypefn 6520 6521@deftypefn {Target Hook} machine_mode TARGET_CC_MODES_COMPATIBLE (machine_mode @var{m1}, machine_mode @var{m2}) 6522On targets which use multiple condition code modes in class 6523@code{MODE_CC}, it is sometimes the case that a comparison can be 6524validly done in more than one mode. On such a system, define this 6525target hook to take two mode arguments and to return a mode in which 6526both comparisons may be validly done. If there is no such mode, 6527return @code{VOIDmode}. 6528 6529The default version of this hook checks whether the modes are the 6530same. If they are, it returns that mode. If they are different, it 6531returns @code{VOIDmode}. 6532@end deftypefn 6533 6534@deftypevr {Target Hook} {unsigned int} TARGET_FLAGS_REGNUM 6535If the target has a dedicated flags register, and it needs to use the 6536post-reload comparison elimination pass, or the delay slot filler pass, 6537then this value should be set appropriately. 6538@end deftypevr 6539 6540@node Costs 6541@section Describing Relative Costs of Operations 6542@cindex costs of instructions 6543@cindex relative costs 6544@cindex speed of instructions 6545 6546These macros let you describe the relative speed of various operations 6547on the target machine. 6548 6549@defmac REGISTER_MOVE_COST (@var{mode}, @var{from}, @var{to}) 6550A C expression for the cost of moving data of mode @var{mode} from a 6551register in class @var{from} to one in class @var{to}. The classes are 6552expressed using the enumeration values such as @code{GENERAL_REGS}. A 6553value of 2 is the default; other values are interpreted relative to 6554that. 6555 6556It is not required that the cost always equal 2 when @var{from} is the 6557same as @var{to}; on some machines it is expensive to move between 6558registers if they are not general registers. 6559 6560If reload sees an insn consisting of a single @code{set} between two 6561hard registers, and if @code{REGISTER_MOVE_COST} applied to their 6562classes returns a value of 2, reload does not check to ensure that the 6563constraints of the insn are met. Setting a cost of other than 2 will 6564allow reload to verify that the constraints are met. You should do this 6565if the @samp{mov@var{m}} pattern's constraints do not allow such copying. 6566 6567These macros are obsolete, new ports should use the target hook 6568@code{TARGET_REGISTER_MOVE_COST} instead. 6569@end defmac 6570 6571@deftypefn {Target Hook} int TARGET_REGISTER_MOVE_COST (machine_mode @var{mode}, reg_class_t @var{from}, reg_class_t @var{to}) 6572This target hook should return the cost of moving data of mode @var{mode} 6573from a register in class @var{from} to one in class @var{to}. The classes 6574are expressed using the enumeration values such as @code{GENERAL_REGS}. 6575A value of 2 is the default; other values are interpreted relative to 6576that. 6577 6578It is not required that the cost always equal 2 when @var{from} is the 6579same as @var{to}; on some machines it is expensive to move between 6580registers if they are not general registers. 6581 6582If reload sees an insn consisting of a single @code{set} between two 6583hard registers, and if @code{TARGET_REGISTER_MOVE_COST} applied to their 6584classes returns a value of 2, reload does not check to ensure that the 6585constraints of the insn are met. Setting a cost of other than 2 will 6586allow reload to verify that the constraints are met. You should do this 6587if the @samp{mov@var{m}} pattern's constraints do not allow such copying. 6588 6589The default version of this function returns 2. 6590@end deftypefn 6591 6592@defmac MEMORY_MOVE_COST (@var{mode}, @var{class}, @var{in}) 6593A C expression for the cost of moving data of mode @var{mode} between a 6594register of class @var{class} and memory; @var{in} is zero if the value 6595is to be written to memory, nonzero if it is to be read in. This cost 6596is relative to those in @code{REGISTER_MOVE_COST}. If moving between 6597registers and memory is more expensive than between two registers, you 6598should define this macro to express the relative cost. 6599 6600If you do not define this macro, GCC uses a default cost of 4 plus 6601the cost of copying via a secondary reload register, if one is 6602needed. If your machine requires a secondary reload register to copy 6603between memory and a register of @var{class} but the reload mechanism is 6604more complex than copying via an intermediate, define this macro to 6605reflect the actual cost of the move. 6606 6607GCC defines the function @code{memory_move_secondary_cost} if 6608secondary reloads are needed. It computes the costs due to copying via 6609a secondary register. If your machine copies from memory using a 6610secondary register in the conventional way but the default base value of 66114 is not correct for your machine, define this macro to add some other 6612value to the result of that function. The arguments to that function 6613are the same as to this macro. 6614 6615These macros are obsolete, new ports should use the target hook 6616@code{TARGET_MEMORY_MOVE_COST} instead. 6617@end defmac 6618 6619@deftypefn {Target Hook} int TARGET_MEMORY_MOVE_COST (machine_mode @var{mode}, reg_class_t @var{rclass}, bool @var{in}) 6620This target hook should return the cost of moving data of mode @var{mode} 6621between a register of class @var{rclass} and memory; @var{in} is @code{false} 6622if the value is to be written to memory, @code{true} if it is to be read in. 6623This cost is relative to those in @code{TARGET_REGISTER_MOVE_COST}. 6624If moving between registers and memory is more expensive than between two 6625registers, you should add this target hook to express the relative cost. 6626 6627If you do not add this target hook, GCC uses a default cost of 4 plus 6628the cost of copying via a secondary reload register, if one is 6629needed. If your machine requires a secondary reload register to copy 6630between memory and a register of @var{rclass} but the reload mechanism is 6631more complex than copying via an intermediate, use this target hook to 6632reflect the actual cost of the move. 6633 6634GCC defines the function @code{memory_move_secondary_cost} if 6635secondary reloads are needed. It computes the costs due to copying via 6636a secondary register. If your machine copies from memory using a 6637secondary register in the conventional way but the default base value of 66384 is not correct for your machine, use this target hook to add some other 6639value to the result of that function. The arguments to that function 6640are the same as to this target hook. 6641@end deftypefn 6642 6643@defmac BRANCH_COST (@var{speed_p}, @var{predictable_p}) 6644A C expression for the cost of a branch instruction. A value of 1 is 6645the default; other values are interpreted relative to that. Parameter 6646@var{speed_p} is true when the branch in question should be optimized 6647for speed. When it is false, @code{BRANCH_COST} should return a value 6648optimal for code size rather than performance. @var{predictable_p} is 6649true for well-predicted branches. On many architectures the 6650@code{BRANCH_COST} can be reduced then. 6651@end defmac 6652 6653Here are additional macros which do not specify precise relative costs, 6654but only that certain actions are more expensive than GCC would 6655ordinarily expect. 6656 6657@defmac SLOW_BYTE_ACCESS 6658Define this macro as a C expression which is nonzero if accessing less 6659than a word of memory (i.e.@: a @code{char} or a @code{short}) is no 6660faster than accessing a word of memory, i.e., if such access 6661require more than one instruction or if there is no difference in cost 6662between byte and (aligned) word loads. 6663 6664When this macro is not defined, the compiler will access a field by 6665finding the smallest containing object; when it is defined, a fullword 6666load will be used if alignment permits. Unless bytes accesses are 6667faster than word accesses, using word accesses is preferable since it 6668may eliminate subsequent memory access if subsequent accesses occur to 6669other fields in the same word of the structure, but to different bytes. 6670@end defmac 6671 6672@deftypefn {Target Hook} bool TARGET_SLOW_UNALIGNED_ACCESS (machine_mode @var{mode}, unsigned int @var{align}) 6673This hook returns true if memory accesses described by the 6674@var{mode} and @var{alignment} parameters have a cost many times greater 6675than aligned accesses, for example if they are emulated in a trap handler. 6676This hook is invoked only for unaligned accesses, i.e.@: when 6677@code{@var{alignment} < GET_MODE_ALIGNMENT (@var{mode})}. 6678 6679When this hook returns true, the compiler will act as if 6680@code{STRICT_ALIGNMENT} were true when generating code for block 6681moves. This can cause significantly more instructions to be produced. 6682Therefore, do not make this hook return true if unaligned accesses only 6683add a cycle or two to the time for a memory access. 6684 6685The hook must return true whenever @code{STRICT_ALIGNMENT} is true. 6686The default implementation returns @code{STRICT_ALIGNMENT}. 6687@end deftypefn 6688 6689@defmac MOVE_RATIO (@var{speed}) 6690The threshold of number of scalar memory-to-memory move insns, @emph{below} 6691which a sequence of insns should be generated instead of a 6692string move insn or a library call. Increasing the value will always 6693make code faster, but eventually incurs high cost in increased code size. 6694 6695Note that on machines where the corresponding move insn is a 6696@code{define_expand} that emits a sequence of insns, this macro counts 6697the number of such sequences. 6698 6699The parameter @var{speed} is true if the code is currently being 6700optimized for speed rather than size. 6701 6702If you don't define this, a reasonable default is used. 6703@end defmac 6704 6705@deftypefn {Target Hook} bool TARGET_USE_BY_PIECES_INFRASTRUCTURE_P (unsigned HOST_WIDE_INT @var{size}, unsigned int @var{alignment}, enum by_pieces_operation @var{op}, bool @var{speed_p}) 6706GCC will attempt several strategies when asked to copy between 6707two areas of memory, or to set, clear or store to memory, for example 6708when copying a @code{struct}. The @code{by_pieces} infrastructure 6709implements such memory operations as a sequence of load, store or move 6710insns. Alternate strategies are to expand the 6711@code{cpymem} or @code{setmem} optabs, to emit a library call, or to emit 6712unit-by-unit, loop-based operations. 6713 6714This target hook should return true if, for a memory operation with a 6715given @var{size} and @var{alignment}, using the @code{by_pieces} 6716infrastructure is expected to result in better code generation. 6717Both @var{size} and @var{alignment} are measured in terms of storage 6718units. 6719 6720The parameter @var{op} is one of: @code{CLEAR_BY_PIECES}, 6721@code{MOVE_BY_PIECES}, @code{SET_BY_PIECES}, @code{STORE_BY_PIECES} or 6722@code{COMPARE_BY_PIECES}. These describe the type of memory operation 6723under consideration. 6724 6725The parameter @var{speed_p} is true if the code is currently being 6726optimized for speed rather than size. 6727 6728Returning true for higher values of @var{size} can improve code generation 6729for speed if the target does not provide an implementation of the 6730@code{cpymem} or @code{setmem} standard names, if the @code{cpymem} or 6731@code{setmem} implementation would be more expensive than a sequence of 6732insns, or if the overhead of a library call would dominate that of 6733the body of the memory operation. 6734 6735Returning true for higher values of @code{size} may also cause an increase 6736in code size, for example where the number of insns emitted to perform a 6737move would be greater than that of a library call. 6738@end deftypefn 6739 6740@deftypefn {Target Hook} int TARGET_COMPARE_BY_PIECES_BRANCH_RATIO (machine_mode @var{mode}) 6741When expanding a block comparison in MODE, gcc can try to reduce the 6742number of branches at the expense of more memory operations. This hook 6743allows the target to override the default choice. It should return the 6744factor by which branches should be reduced over the plain expansion with 6745one comparison per @var{mode}-sized piece. A port can also prevent a 6746particular mode from being used for block comparisons by returning a 6747negative number from this hook. 6748@end deftypefn 6749 6750@defmac MOVE_MAX_PIECES 6751A C expression used by @code{move_by_pieces} to determine the largest unit 6752a load or store used to copy memory is. Defaults to @code{MOVE_MAX}. 6753@end defmac 6754 6755@defmac STORE_MAX_PIECES 6756A C expression used by @code{store_by_pieces} to determine the largest unit 6757a store used to memory is. Defaults to @code{MOVE_MAX_PIECES}, or two times 6758the size of @code{HOST_WIDE_INT}, whichever is smaller. 6759@end defmac 6760 6761@defmac COMPARE_MAX_PIECES 6762A C expression used by @code{compare_by_pieces} to determine the largest unit 6763a load or store used to compare memory is. Defaults to 6764@code{MOVE_MAX_PIECES}. 6765@end defmac 6766 6767@defmac CLEAR_RATIO (@var{speed}) 6768The threshold of number of scalar move insns, @emph{below} which a sequence 6769of insns should be generated to clear memory instead of a string clear insn 6770or a library call. Increasing the value will always make code faster, but 6771eventually incurs high cost in increased code size. 6772 6773The parameter @var{speed} is true if the code is currently being 6774optimized for speed rather than size. 6775 6776If you don't define this, a reasonable default is used. 6777@end defmac 6778 6779@defmac SET_RATIO (@var{speed}) 6780The threshold of number of scalar move insns, @emph{below} which a sequence 6781of insns should be generated to set memory to a constant value, instead of 6782a block set insn or a library call. 6783Increasing the value will always make code faster, but 6784eventually incurs high cost in increased code size. 6785 6786The parameter @var{speed} is true if the code is currently being 6787optimized for speed rather than size. 6788 6789If you don't define this, it defaults to the value of @code{MOVE_RATIO}. 6790@end defmac 6791 6792@defmac USE_LOAD_POST_INCREMENT (@var{mode}) 6793A C expression used to determine whether a load postincrement is a good 6794thing to use for a given mode. Defaults to the value of 6795@code{HAVE_POST_INCREMENT}. 6796@end defmac 6797 6798@defmac USE_LOAD_POST_DECREMENT (@var{mode}) 6799A C expression used to determine whether a load postdecrement is a good 6800thing to use for a given mode. Defaults to the value of 6801@code{HAVE_POST_DECREMENT}. 6802@end defmac 6803 6804@defmac USE_LOAD_PRE_INCREMENT (@var{mode}) 6805A C expression used to determine whether a load preincrement is a good 6806thing to use for a given mode. Defaults to the value of 6807@code{HAVE_PRE_INCREMENT}. 6808@end defmac 6809 6810@defmac USE_LOAD_PRE_DECREMENT (@var{mode}) 6811A C expression used to determine whether a load predecrement is a good 6812thing to use for a given mode. Defaults to the value of 6813@code{HAVE_PRE_DECREMENT}. 6814@end defmac 6815 6816@defmac USE_STORE_POST_INCREMENT (@var{mode}) 6817A C expression used to determine whether a store postincrement is a good 6818thing to use for a given mode. Defaults to the value of 6819@code{HAVE_POST_INCREMENT}. 6820@end defmac 6821 6822@defmac USE_STORE_POST_DECREMENT (@var{mode}) 6823A C expression used to determine whether a store postdecrement is a good 6824thing to use for a given mode. Defaults to the value of 6825@code{HAVE_POST_DECREMENT}. 6826@end defmac 6827 6828@defmac USE_STORE_PRE_INCREMENT (@var{mode}) 6829This macro is used to determine whether a store preincrement is a good 6830thing to use for a given mode. Defaults to the value of 6831@code{HAVE_PRE_INCREMENT}. 6832@end defmac 6833 6834@defmac USE_STORE_PRE_DECREMENT (@var{mode}) 6835This macro is used to determine whether a store predecrement is a good 6836thing to use for a given mode. Defaults to the value of 6837@code{HAVE_PRE_DECREMENT}. 6838@end defmac 6839 6840@defmac NO_FUNCTION_CSE 6841Define this macro to be true if it is as good or better to call a constant 6842function address than to call an address kept in a register. 6843@end defmac 6844 6845@defmac LOGICAL_OP_NON_SHORT_CIRCUIT 6846Define this macro if a non-short-circuit operation produced by 6847@samp{fold_range_test ()} is optimal. This macro defaults to true if 6848@code{BRANCH_COST} is greater than or equal to the value 2. 6849@end defmac 6850 6851@deftypefn {Target Hook} bool TARGET_OPTAB_SUPPORTED_P (int @var{op}, machine_mode @var{mode1}, machine_mode @var{mode2}, optimization_type @var{opt_type}) 6852Return true if the optimizers should use optab @var{op} with 6853modes @var{mode1} and @var{mode2} for optimization type @var{opt_type}. 6854The optab is known to have an associated @file{.md} instruction 6855whose C condition is true. @var{mode2} is only meaningful for conversion 6856optabs; for direct optabs it is a copy of @var{mode1}. 6857 6858For example, when called with @var{op} equal to @code{rint_optab} and 6859@var{mode1} equal to @code{DFmode}, the hook should say whether the 6860optimizers should use optab @code{rintdf2}. 6861 6862The default hook returns true for all inputs. 6863@end deftypefn 6864 6865@deftypefn {Target Hook} bool TARGET_RTX_COSTS (rtx @var{x}, machine_mode @var{mode}, int @var{outer_code}, int @var{opno}, int *@var{total}, bool @var{speed}) 6866This target hook describes the relative costs of RTL expressions. 6867 6868The cost may depend on the precise form of the expression, which is 6869available for examination in @var{x}, and the fact that @var{x} appears 6870as operand @var{opno} of an expression with rtx code @var{outer_code}. 6871That is, the hook can assume that there is some rtx @var{y} such 6872that @samp{GET_CODE (@var{y}) == @var{outer_code}} and such that 6873either (a) @samp{XEXP (@var{y}, @var{opno}) == @var{x}} or 6874(b) @samp{XVEC (@var{y}, @var{opno})} contains @var{x}. 6875 6876@var{mode} is @var{x}'s machine mode, or for cases like @code{const_int} that 6877do not have a mode, the mode in which @var{x} is used. 6878 6879In implementing this hook, you can use the construct 6880@code{COSTS_N_INSNS (@var{n})} to specify a cost equal to @var{n} fast 6881instructions. 6882 6883On entry to the hook, @code{*@var{total}} contains a default estimate 6884for the cost of the expression. The hook should modify this value as 6885necessary. Traditionally, the default costs are @code{COSTS_N_INSNS (5)} 6886for multiplications, @code{COSTS_N_INSNS (7)} for division and modulus 6887operations, and @code{COSTS_N_INSNS (1)} for all other operations. 6888 6889When optimizing for code size, i.e.@: when @code{speed} is 6890false, this target hook should be used to estimate the relative 6891size cost of an expression, again relative to @code{COSTS_N_INSNS}. 6892 6893The hook returns true when all subexpressions of @var{x} have been 6894processed, and false when @code{rtx_cost} should recurse. 6895@end deftypefn 6896 6897@deftypefn {Target Hook} int TARGET_ADDRESS_COST (rtx @var{address}, machine_mode @var{mode}, addr_space_t @var{as}, bool @var{speed}) 6898This hook computes the cost of an addressing mode that contains 6899@var{address}. If not defined, the cost is computed from 6900the @var{address} expression and the @code{TARGET_RTX_COST} hook. 6901 6902For most CISC machines, the default cost is a good approximation of the 6903true cost of the addressing mode. However, on RISC machines, all 6904instructions normally have the same length and execution time. Hence 6905all addresses will have equal costs. 6906 6907In cases where more than one form of an address is known, the form with 6908the lowest cost will be used. If multiple forms have the same, lowest, 6909cost, the one that is the most complex will be used. 6910 6911For example, suppose an address that is equal to the sum of a register 6912and a constant is used twice in the same basic block. When this macro 6913is not defined, the address will be computed in a register and memory 6914references will be indirect through that register. On machines where 6915the cost of the addressing mode containing the sum is no higher than 6916that of a simple indirect reference, this will produce an additional 6917instruction and possibly require an additional register. Proper 6918specification of this macro eliminates this overhead for such machines. 6919 6920This hook is never called with an invalid address. 6921 6922On machines where an address involving more than one register is as 6923cheap as an address computation involving only one register, defining 6924@code{TARGET_ADDRESS_COST} to reflect this can cause two registers to 6925be live over a region of code where only one would have been if 6926@code{TARGET_ADDRESS_COST} were not defined in that manner. This effect 6927should be considered in the definition of this macro. Equivalent costs 6928should probably only be given to addresses with different numbers of 6929registers on machines with lots of registers. 6930@end deftypefn 6931 6932@deftypefn {Target Hook} int TARGET_INSN_COST (rtx_insn *@var{insn}, bool @var{speed}) 6933This target hook describes the relative costs of RTL instructions. 6934 6935In implementing this hook, you can use the construct 6936@code{COSTS_N_INSNS (@var{n})} to specify a cost equal to @var{n} fast 6937instructions. 6938 6939When optimizing for code size, i.e.@: when @code{speed} is 6940false, this target hook should be used to estimate the relative 6941size cost of an expression, again relative to @code{COSTS_N_INSNS}. 6942@end deftypefn 6943 6944@deftypefn {Target Hook} {unsigned int} TARGET_MAX_NOCE_IFCVT_SEQ_COST (edge @var{e}) 6945This hook returns a value in the same units as @code{TARGET_RTX_COSTS}, 6946giving the maximum acceptable cost for a sequence generated by the RTL 6947if-conversion pass when conditional execution is not available. 6948The RTL if-conversion pass attempts to convert conditional operations 6949that would require a branch to a series of unconditional operations and 6950@code{mov@var{mode}cc} insns. This hook returns the maximum cost of the 6951unconditional instructions and the @code{mov@var{mode}cc} insns. 6952RTL if-conversion is cancelled if the cost of the converted sequence 6953is greater than the value returned by this hook. 6954 6955@code{e} is the edge between the basic block containing the conditional 6956branch to the basic block which would be executed if the condition 6957were true. 6958 6959The default implementation of this hook uses the 6960@code{max-rtl-if-conversion-[un]predictable} parameters if they are set, 6961and uses a multiple of @code{BRANCH_COST} otherwise. 6962@end deftypefn 6963 6964@deftypefn {Target Hook} bool TARGET_NOCE_CONVERSION_PROFITABLE_P (rtx_insn *@var{seq}, struct noce_if_info *@var{if_info}) 6965This hook returns true if the instruction sequence @code{seq} is a good 6966candidate as a replacement for the if-convertible sequence described in 6967@code{if_info}. 6968@end deftypefn 6969 6970@deftypefn {Target Hook} bool TARGET_NO_SPECULATION_IN_DELAY_SLOTS_P (void) 6971This predicate controls the use of the eager delay slot filler to disallow 6972speculatively executed instructions being placed in delay slots. Targets 6973such as certain MIPS architectures possess both branches with and without 6974delay slots. As the eager delay slot filler can decrease performance, 6975disabling it is beneficial when ordinary branches are available. Use of 6976delay slot branches filled using the basic filler is often still desirable 6977as the delay slot can hide a pipeline bubble. 6978@end deftypefn 6979 6980@deftypefn {Target Hook} HOST_WIDE_INT TARGET_ESTIMATED_POLY_VALUE (poly_int64 @var{val}) 6981Return an estimate of the runtime value of @var{val}, for use in 6982things like cost calculations or profiling frequencies. The default 6983implementation returns the lowest possible value of @var{val}. 6984@end deftypefn 6985 6986@node Scheduling 6987@section Adjusting the Instruction Scheduler 6988 6989The instruction scheduler may need a fair amount of machine-specific 6990adjustment in order to produce good code. GCC provides several target 6991hooks for this purpose. It is usually enough to define just a few of 6992them: try the first ones in this list first. 6993 6994@deftypefn {Target Hook} int TARGET_SCHED_ISSUE_RATE (void) 6995This hook returns the maximum number of instructions that can ever 6996issue at the same time on the target machine. The default is one. 6997Although the insn scheduler can define itself the possibility of issue 6998an insn on the same cycle, the value can serve as an additional 6999constraint to issue insns on the same simulated processor cycle (see 7000hooks @samp{TARGET_SCHED_REORDER} and @samp{TARGET_SCHED_REORDER2}). 7001This value must be constant over the entire compilation. If you need 7002it to vary depending on what the instructions are, you must use 7003@samp{TARGET_SCHED_VARIABLE_ISSUE}. 7004@end deftypefn 7005 7006@deftypefn {Target Hook} int TARGET_SCHED_VARIABLE_ISSUE (FILE *@var{file}, int @var{verbose}, rtx_insn *@var{insn}, int @var{more}) 7007This hook is executed by the scheduler after it has scheduled an insn 7008from the ready list. It should return the number of insns which can 7009still be issued in the current cycle. The default is 7010@samp{@w{@var{more} - 1}} for insns other than @code{CLOBBER} and 7011@code{USE}, which normally are not counted against the issue rate. 7012You should define this hook if some insns take more machine resources 7013than others, so that fewer insns can follow them in the same cycle. 7014@var{file} is either a null pointer, or a stdio stream to write any 7015debug output to. @var{verbose} is the verbose level provided by 7016@option{-fsched-verbose-@var{n}}. @var{insn} is the instruction that 7017was scheduled. 7018@end deftypefn 7019 7020@deftypefn {Target Hook} int TARGET_SCHED_ADJUST_COST (rtx_insn *@var{insn}, int @var{dep_type1}, rtx_insn *@var{dep_insn}, int @var{cost}, unsigned int @var{dw}) 7021This function corrects the value of @var{cost} based on the 7022relationship between @var{insn} and @var{dep_insn} through a 7023dependence of type dep_type, and strength @var{dw}. It should return the new 7024value. The default is to make no adjustment to @var{cost}. This can be 7025used for example to specify to the scheduler using the traditional pipeline 7026description that an output- or anti-dependence does not incur the same cost 7027as a data-dependence. If the scheduler using the automaton based pipeline 7028description, the cost of anti-dependence is zero and the cost of 7029output-dependence is maximum of one and the difference of latency 7030times of the first and the second insns. If these values are not 7031acceptable, you could use the hook to modify them too. See also 7032@pxref{Processor pipeline description}. 7033@end deftypefn 7034 7035@deftypefn {Target Hook} int TARGET_SCHED_ADJUST_PRIORITY (rtx_insn *@var{insn}, int @var{priority}) 7036This hook adjusts the integer scheduling priority @var{priority} of 7037@var{insn}. It should return the new priority. Increase the priority to 7038execute @var{insn} earlier, reduce the priority to execute @var{insn} 7039later. Do not define this hook if you do not need to adjust the 7040scheduling priorities of insns. 7041@end deftypefn 7042 7043@deftypefn {Target Hook} int TARGET_SCHED_REORDER (FILE *@var{file}, int @var{verbose}, rtx_insn **@var{ready}, int *@var{n_readyp}, int @var{clock}) 7044This hook is executed by the scheduler after it has scheduled the ready 7045list, to allow the machine description to reorder it (for example to 7046combine two small instructions together on @samp{VLIW} machines). 7047@var{file} is either a null pointer, or a stdio stream to write any 7048debug output to. @var{verbose} is the verbose level provided by 7049@option{-fsched-verbose-@var{n}}. @var{ready} is a pointer to the ready 7050list of instructions that are ready to be scheduled. @var{n_readyp} is 7051a pointer to the number of elements in the ready list. The scheduler 7052reads the ready list in reverse order, starting with 7053@var{ready}[@var{*n_readyp} @minus{} 1] and going to @var{ready}[0]. @var{clock} 7054is the timer tick of the scheduler. You may modify the ready list and 7055the number of ready insns. The return value is the number of insns that 7056can issue this cycle; normally this is just @code{issue_rate}. See also 7057@samp{TARGET_SCHED_REORDER2}. 7058@end deftypefn 7059 7060@deftypefn {Target Hook} int TARGET_SCHED_REORDER2 (FILE *@var{file}, int @var{verbose}, rtx_insn **@var{ready}, int *@var{n_readyp}, int @var{clock}) 7061Like @samp{TARGET_SCHED_REORDER}, but called at a different time. That 7062function is called whenever the scheduler starts a new cycle. This one 7063is called once per iteration over a cycle, immediately after 7064@samp{TARGET_SCHED_VARIABLE_ISSUE}; it can reorder the ready list and 7065return the number of insns to be scheduled in the same cycle. Defining 7066this hook can be useful if there are frequent situations where 7067scheduling one insn causes other insns to become ready in the same 7068cycle. These other insns can then be taken into account properly. 7069@end deftypefn 7070 7071@deftypefn {Target Hook} bool TARGET_SCHED_MACRO_FUSION_P (void) 7072This hook is used to check whether target platform supports macro fusion. 7073@end deftypefn 7074 7075@deftypefn {Target Hook} bool TARGET_SCHED_MACRO_FUSION_PAIR_P (rtx_insn *@var{prev}, rtx_insn *@var{curr}) 7076This hook is used to check whether two insns should be macro fused for 7077a target microarchitecture. If this hook returns true for the given insn pair 7078(@var{prev} and @var{curr}), the scheduler will put them into a sched 7079group, and they will not be scheduled apart. The two insns will be either 7080two SET insns or a compare and a conditional jump and this hook should 7081validate any dependencies needed to fuse the two insns together. 7082@end deftypefn 7083 7084@deftypefn {Target Hook} void TARGET_SCHED_DEPENDENCIES_EVALUATION_HOOK (rtx_insn *@var{head}, rtx_insn *@var{tail}) 7085This hook is called after evaluation forward dependencies of insns in 7086chain given by two parameter values (@var{head} and @var{tail} 7087correspondingly) but before insns scheduling of the insn chain. For 7088example, it can be used for better insn classification if it requires 7089analysis of dependencies. This hook can use backward and forward 7090dependencies of the insn scheduler because they are already 7091calculated. 7092@end deftypefn 7093 7094@deftypefn {Target Hook} void TARGET_SCHED_INIT (FILE *@var{file}, int @var{verbose}, int @var{max_ready}) 7095This hook is executed by the scheduler at the beginning of each block of 7096instructions that are to be scheduled. @var{file} is either a null 7097pointer, or a stdio stream to write any debug output to. @var{verbose} 7098is the verbose level provided by @option{-fsched-verbose-@var{n}}. 7099@var{max_ready} is the maximum number of insns in the current scheduling 7100region that can be live at the same time. This can be used to allocate 7101scratch space if it is needed, e.g.@: by @samp{TARGET_SCHED_REORDER}. 7102@end deftypefn 7103 7104@deftypefn {Target Hook} void TARGET_SCHED_FINISH (FILE *@var{file}, int @var{verbose}) 7105This hook is executed by the scheduler at the end of each block of 7106instructions that are to be scheduled. It can be used to perform 7107cleanup of any actions done by the other scheduling hooks. @var{file} 7108is either a null pointer, or a stdio stream to write any debug output 7109to. @var{verbose} is the verbose level provided by 7110@option{-fsched-verbose-@var{n}}. 7111@end deftypefn 7112 7113@deftypefn {Target Hook} void TARGET_SCHED_INIT_GLOBAL (FILE *@var{file}, int @var{verbose}, int @var{old_max_uid}) 7114This hook is executed by the scheduler after function level initializations. 7115@var{file} is either a null pointer, or a stdio stream to write any debug output to. 7116@var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}. 7117@var{old_max_uid} is the maximum insn uid when scheduling begins. 7118@end deftypefn 7119 7120@deftypefn {Target Hook} void TARGET_SCHED_FINISH_GLOBAL (FILE *@var{file}, int @var{verbose}) 7121This is the cleanup hook corresponding to @code{TARGET_SCHED_INIT_GLOBAL}. 7122@var{file} is either a null pointer, or a stdio stream to write any debug output to. 7123@var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}. 7124@end deftypefn 7125 7126@deftypefn {Target Hook} rtx TARGET_SCHED_DFA_PRE_CYCLE_INSN (void) 7127The hook returns an RTL insn. The automaton state used in the 7128pipeline hazard recognizer is changed as if the insn were scheduled 7129when the new simulated processor cycle starts. Usage of the hook may 7130simplify the automaton pipeline description for some @acronym{VLIW} 7131processors. If the hook is defined, it is used only for the automaton 7132based pipeline description. The default is not to change the state 7133when the new simulated processor cycle starts. 7134@end deftypefn 7135 7136@deftypefn {Target Hook} void TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN (void) 7137The hook can be used to initialize data used by the previous hook. 7138@end deftypefn 7139 7140@deftypefn {Target Hook} {rtx_insn *} TARGET_SCHED_DFA_POST_CYCLE_INSN (void) 7141The hook is analogous to @samp{TARGET_SCHED_DFA_PRE_CYCLE_INSN} but used 7142to changed the state as if the insn were scheduled when the new 7143simulated processor cycle finishes. 7144@end deftypefn 7145 7146@deftypefn {Target Hook} void TARGET_SCHED_INIT_DFA_POST_CYCLE_INSN (void) 7147The hook is analogous to @samp{TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN} but 7148used to initialize data used by the previous hook. 7149@end deftypefn 7150 7151@deftypefn {Target Hook} void TARGET_SCHED_DFA_PRE_ADVANCE_CYCLE (void) 7152The hook to notify target that the current simulated cycle is about to finish. 7153The hook is analogous to @samp{TARGET_SCHED_DFA_PRE_CYCLE_INSN} but used 7154to change the state in more complicated situations - e.g., when advancing 7155state on a single insn is not enough. 7156@end deftypefn 7157 7158@deftypefn {Target Hook} void TARGET_SCHED_DFA_POST_ADVANCE_CYCLE (void) 7159The hook to notify target that new simulated cycle has just started. 7160The hook is analogous to @samp{TARGET_SCHED_DFA_POST_CYCLE_INSN} but used 7161to change the state in more complicated situations - e.g., when advancing 7162state on a single insn is not enough. 7163@end deftypefn 7164 7165@deftypefn {Target Hook} int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD (void) 7166This hook controls better choosing an insn from the ready insn queue 7167for the @acronym{DFA}-based insn scheduler. Usually the scheduler 7168chooses the first insn from the queue. If the hook returns a positive 7169value, an additional scheduler code tries all permutations of 7170@samp{TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD ()} 7171subsequent ready insns to choose an insn whose issue will result in 7172maximal number of issued insns on the same cycle. For the 7173@acronym{VLIW} processor, the code could actually solve the problem of 7174packing simple insns into the @acronym{VLIW} insn. Of course, if the 7175rules of @acronym{VLIW} packing are described in the automaton. 7176 7177This code also could be used for superscalar @acronym{RISC} 7178processors. Let us consider a superscalar @acronym{RISC} processor 7179with 3 pipelines. Some insns can be executed in pipelines @var{A} or 7180@var{B}, some insns can be executed only in pipelines @var{B} or 7181@var{C}, and one insn can be executed in pipeline @var{B}. The 7182processor may issue the 1st insn into @var{A} and the 2nd one into 7183@var{B}. In this case, the 3rd insn will wait for freeing @var{B} 7184until the next cycle. If the scheduler issues the 3rd insn the first, 7185the processor could issue all 3 insns per cycle. 7186 7187Actually this code demonstrates advantages of the automaton based 7188pipeline hazard recognizer. We try quickly and easy many insn 7189schedules to choose the best one. 7190 7191The default is no multipass scheduling. 7192@end deftypefn 7193 7194@deftypefn {Target Hook} int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD (rtx_insn *@var{insn}, int @var{ready_index}) 7195 7196This hook controls what insns from the ready insn queue will be 7197considered for the multipass insn scheduling. If the hook returns 7198zero for @var{insn}, the insn will be considered in multipass scheduling. 7199Positive return values will remove @var{insn} from consideration on 7200the current round of multipass scheduling. 7201Negative return values will remove @var{insn} from consideration for given 7202number of cycles. 7203Backends should be careful about returning non-zero for highest priority 7204instruction at position 0 in the ready list. @var{ready_index} is passed 7205to allow backends make correct judgements. 7206 7207The default is that any ready insns can be chosen to be issued. 7208@end deftypefn 7209 7210@deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_BEGIN (void *@var{data}, signed char *@var{ready_try}, int @var{n_ready}, bool @var{first_cycle_insn_p}) 7211This hook prepares the target backend for a new round of multipass 7212scheduling. 7213@end deftypefn 7214 7215@deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_ISSUE (void *@var{data}, signed char *@var{ready_try}, int @var{n_ready}, rtx_insn *@var{insn}, const void *@var{prev_data}) 7216This hook is called when multipass scheduling evaluates instruction INSN. 7217@end deftypefn 7218 7219@deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_BACKTRACK (const void *@var{data}, signed char *@var{ready_try}, int @var{n_ready}) 7220This is called when multipass scheduling backtracks from evaluation of 7221an instruction. 7222@end deftypefn 7223 7224@deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_END (const void *@var{data}) 7225This hook notifies the target about the result of the concluded current 7226round of multipass scheduling. 7227@end deftypefn 7228 7229@deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_INIT (void *@var{data}) 7230This hook initializes target-specific data used in multipass scheduling. 7231@end deftypefn 7232 7233@deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_FINI (void *@var{data}) 7234This hook finalizes target-specific data used in multipass scheduling. 7235@end deftypefn 7236 7237@deftypefn {Target Hook} int TARGET_SCHED_DFA_NEW_CYCLE (FILE *@var{dump}, int @var{verbose}, rtx_insn *@var{insn}, int @var{last_clock}, int @var{clock}, int *@var{sort_p}) 7238This hook is called by the insn scheduler before issuing @var{insn} 7239on cycle @var{clock}. If the hook returns nonzero, 7240@var{insn} is not issued on this processor cycle. Instead, 7241the processor cycle is advanced. If *@var{sort_p} 7242is zero, the insn ready queue is not sorted on the new cycle 7243start as usually. @var{dump} and @var{verbose} specify the file and 7244verbosity level to use for debugging output. 7245@var{last_clock} and @var{clock} are, respectively, the 7246processor cycle on which the previous insn has been issued, 7247and the current processor cycle. 7248@end deftypefn 7249 7250@deftypefn {Target Hook} bool TARGET_SCHED_IS_COSTLY_DEPENDENCE (struct _dep *@var{_dep}, int @var{cost}, int @var{distance}) 7251This hook is used to define which dependences are considered costly by 7252the target, so costly that it is not advisable to schedule the insns that 7253are involved in the dependence too close to one another. The parameters 7254to this hook are as follows: The first parameter @var{_dep} is the dependence 7255being evaluated. The second parameter @var{cost} is the cost of the 7256dependence as estimated by the scheduler, and the third 7257parameter @var{distance} is the distance in cycles between the two insns. 7258The hook returns @code{true} if considering the distance between the two 7259insns the dependence between them is considered costly by the target, 7260and @code{false} otherwise. 7261 7262Defining this hook can be useful in multiple-issue out-of-order machines, 7263where (a) it's practically hopeless to predict the actual data/resource 7264delays, however: (b) there's a better chance to predict the actual grouping 7265that will be formed, and (c) correctly emulating the grouping can be very 7266important. In such targets one may want to allow issuing dependent insns 7267closer to one another---i.e., closer than the dependence distance; however, 7268not in cases of ``costly dependences'', which this hooks allows to define. 7269@end deftypefn 7270 7271@deftypefn {Target Hook} void TARGET_SCHED_H_I_D_EXTENDED (void) 7272This hook is called by the insn scheduler after emitting a new instruction to 7273the instruction stream. The hook notifies a target backend to extend its 7274per instruction data structures. 7275@end deftypefn 7276 7277@deftypefn {Target Hook} {void *} TARGET_SCHED_ALLOC_SCHED_CONTEXT (void) 7278Return a pointer to a store large enough to hold target scheduling context. 7279@end deftypefn 7280 7281@deftypefn {Target Hook} void TARGET_SCHED_INIT_SCHED_CONTEXT (void *@var{tc}, bool @var{clean_p}) 7282Initialize store pointed to by @var{tc} to hold target scheduling context. 7283It @var{clean_p} is true then initialize @var{tc} as if scheduler is at the 7284beginning of the block. Otherwise, copy the current context into @var{tc}. 7285@end deftypefn 7286 7287@deftypefn {Target Hook} void TARGET_SCHED_SET_SCHED_CONTEXT (void *@var{tc}) 7288Copy target scheduling context pointed to by @var{tc} to the current context. 7289@end deftypefn 7290 7291@deftypefn {Target Hook} void TARGET_SCHED_CLEAR_SCHED_CONTEXT (void *@var{tc}) 7292Deallocate internal data in target scheduling context pointed to by @var{tc}. 7293@end deftypefn 7294 7295@deftypefn {Target Hook} void TARGET_SCHED_FREE_SCHED_CONTEXT (void *@var{tc}) 7296Deallocate a store for target scheduling context pointed to by @var{tc}. 7297@end deftypefn 7298 7299@deftypefn {Target Hook} int TARGET_SCHED_SPECULATE_INSN (rtx_insn *@var{insn}, unsigned int @var{dep_status}, rtx *@var{new_pat}) 7300This hook is called by the insn scheduler when @var{insn} has only 7301speculative dependencies and therefore can be scheduled speculatively. 7302The hook is used to check if the pattern of @var{insn} has a speculative 7303version and, in case of successful check, to generate that speculative 7304pattern. The hook should return 1, if the instruction has a speculative form, 7305or @minus{}1, if it doesn't. @var{request} describes the type of requested 7306speculation. If the return value equals 1 then @var{new_pat} is assigned 7307the generated speculative pattern. 7308@end deftypefn 7309 7310@deftypefn {Target Hook} bool TARGET_SCHED_NEEDS_BLOCK_P (unsigned int @var{dep_status}) 7311This hook is called by the insn scheduler during generation of recovery code 7312for @var{insn}. It should return @code{true}, if the corresponding check 7313instruction should branch to recovery code, or @code{false} otherwise. 7314@end deftypefn 7315 7316@deftypefn {Target Hook} rtx TARGET_SCHED_GEN_SPEC_CHECK (rtx_insn *@var{insn}, rtx_insn *@var{label}, unsigned int @var{ds}) 7317This hook is called by the insn scheduler to generate a pattern for recovery 7318check instruction. If @var{mutate_p} is zero, then @var{insn} is a 7319speculative instruction for which the check should be generated. 7320@var{label} is either a label of a basic block, where recovery code should 7321be emitted, or a null pointer, when requested check doesn't branch to 7322recovery code (a simple check). If @var{mutate_p} is nonzero, then 7323a pattern for a branchy check corresponding to a simple check denoted by 7324@var{insn} should be generated. In this case @var{label} can't be null. 7325@end deftypefn 7326 7327@deftypefn {Target Hook} void TARGET_SCHED_SET_SCHED_FLAGS (struct spec_info_def *@var{spec_info}) 7328This hook is used by the insn scheduler to find out what features should be 7329enabled/used. 7330The structure *@var{spec_info} should be filled in by the target. 7331The structure describes speculation types that can be used in the scheduler. 7332@end deftypefn 7333 7334@deftypefn {Target Hook} bool TARGET_SCHED_CAN_SPECULATE_INSN (rtx_insn *@var{insn}) 7335Some instructions should never be speculated by the schedulers, usually 7336 because the instruction is too expensive to get this wrong. Often such 7337 instructions have long latency, and often they are not fully modeled in the 7338 pipeline descriptions. This hook should return @code{false} if @var{insn} 7339 should not be speculated. 7340@end deftypefn 7341 7342@deftypefn {Target Hook} int TARGET_SCHED_SMS_RES_MII (struct ddg *@var{g}) 7343This hook is called by the swing modulo scheduler to calculate a 7344resource-based lower bound which is based on the resources available in 7345the machine and the resources required by each instruction. The target 7346backend can use @var{g} to calculate such bound. A very simple lower 7347bound will be used in case this hook is not implemented: the total number 7348of instructions divided by the issue rate. 7349@end deftypefn 7350 7351@deftypefn {Target Hook} bool TARGET_SCHED_DISPATCH (rtx_insn *@var{insn}, int @var{x}) 7352This hook is called by Haifa Scheduler. It returns true if dispatch scheduling 7353is supported in hardware and the condition specified in the parameter is true. 7354@end deftypefn 7355 7356@deftypefn {Target Hook} void TARGET_SCHED_DISPATCH_DO (rtx_insn *@var{insn}, int @var{x}) 7357This hook is called by Haifa Scheduler. It performs the operation specified 7358in its second parameter. 7359@end deftypefn 7360 7361@deftypevr {Target Hook} bool TARGET_SCHED_EXPOSED_PIPELINE 7362True if the processor has an exposed pipeline, which means that not just 7363the order of instructions is important for correctness when scheduling, but 7364also the latencies of operations. 7365@end deftypevr 7366 7367@deftypefn {Target Hook} int TARGET_SCHED_REASSOCIATION_WIDTH (unsigned int @var{opc}, machine_mode @var{mode}) 7368This hook is called by tree reassociator to determine a level of 7369parallelism required in output calculations chain. 7370@end deftypefn 7371 7372@deftypefn {Target Hook} void TARGET_SCHED_FUSION_PRIORITY (rtx_insn *@var{insn}, int @var{max_pri}, int *@var{fusion_pri}, int *@var{pri}) 7373This hook is called by scheduling fusion pass. It calculates fusion 7374priorities for each instruction passed in by parameter. The priorities 7375are returned via pointer parameters. 7376 7377@var{insn} is the instruction whose priorities need to be calculated. 7378@var{max_pri} is the maximum priority can be returned in any cases. 7379@var{fusion_pri} is the pointer parameter through which @var{insn}'s 7380fusion priority should be calculated and returned. 7381@var{pri} is the pointer parameter through which @var{insn}'s priority 7382should be calculated and returned. 7383 7384Same @var{fusion_pri} should be returned for instructions which should 7385be scheduled together. Different @var{pri} should be returned for 7386instructions with same @var{fusion_pri}. @var{fusion_pri} is the major 7387sort key, @var{pri} is the minor sort key. All instructions will be 7388scheduled according to the two priorities. All priorities calculated 7389should be between 0 (exclusive) and @var{max_pri} (inclusive). To avoid 7390false dependencies, @var{fusion_pri} of instructions which need to be 7391scheduled together should be smaller than @var{fusion_pri} of irrelevant 7392instructions. 7393 7394Given below example: 7395 7396@smallexample 7397 ldr r10, [r1, 4] 7398 add r4, r4, r10 7399 ldr r15, [r2, 8] 7400 sub r5, r5, r15 7401 ldr r11, [r1, 0] 7402 add r4, r4, r11 7403 ldr r16, [r2, 12] 7404 sub r5, r5, r16 7405@end smallexample 7406 7407On targets like ARM/AArch64, the two pairs of consecutive loads should be 7408merged. Since peephole2 pass can't help in this case unless consecutive 7409loads are actually next to each other in instruction flow. That's where 7410this scheduling fusion pass works. This hook calculates priority for each 7411instruction based on its fustion type, like: 7412 7413@smallexample 7414 ldr r10, [r1, 4] ; fusion_pri=99, pri=96 7415 add r4, r4, r10 ; fusion_pri=100, pri=100 7416 ldr r15, [r2, 8] ; fusion_pri=98, pri=92 7417 sub r5, r5, r15 ; fusion_pri=100, pri=100 7418 ldr r11, [r1, 0] ; fusion_pri=99, pri=100 7419 add r4, r4, r11 ; fusion_pri=100, pri=100 7420 ldr r16, [r2, 12] ; fusion_pri=98, pri=88 7421 sub r5, r5, r16 ; fusion_pri=100, pri=100 7422@end smallexample 7423 7424Scheduling fusion pass then sorts all ready to issue instructions according 7425to the priorities. As a result, instructions of same fusion type will be 7426pushed together in instruction flow, like: 7427 7428@smallexample 7429 ldr r11, [r1, 0] 7430 ldr r10, [r1, 4] 7431 ldr r15, [r2, 8] 7432 ldr r16, [r2, 12] 7433 add r4, r4, r10 7434 sub r5, r5, r15 7435 add r4, r4, r11 7436 sub r5, r5, r16 7437@end smallexample 7438 7439Now peephole2 pass can simply merge the two pairs of loads. 7440 7441Since scheduling fusion pass relies on peephole2 to do real fusion 7442work, it is only enabled by default when peephole2 is in effect. 7443 7444This is firstly introduced on ARM/AArch64 targets, please refer to 7445the hook implementation for how different fusion types are supported. 7446@end deftypefn 7447 7448@deftypefn {Target Hook} void TARGET_EXPAND_DIVMOD_LIBFUNC (rtx @var{libfunc}, machine_mode @var{mode}, rtx @var{op0}, rtx @var{op1}, rtx *@var{quot}, rtx *@var{rem}) 7449Define this hook for enabling divmod transform if the port does not have 7450hardware divmod insn but defines target-specific divmod libfuncs. 7451@end deftypefn 7452 7453@node Sections 7454@section Dividing the Output into Sections (Texts, Data, @dots{}) 7455@c the above section title is WAY too long. maybe cut the part between 7456@c the (...)? --mew 10feb93 7457 7458An object file is divided into sections containing different types of 7459data. In the most common case, there are three sections: the @dfn{text 7460section}, which holds instructions and read-only data; the @dfn{data 7461section}, which holds initialized writable data; and the @dfn{bss 7462section}, which holds uninitialized data. Some systems have other kinds 7463of sections. 7464 7465@file{varasm.c} provides several well-known sections, such as 7466@code{text_section}, @code{data_section} and @code{bss_section}. 7467The normal way of controlling a @code{@var{foo}_section} variable 7468is to define the associated @code{@var{FOO}_SECTION_ASM_OP} macro, 7469as described below. The macros are only read once, when @file{varasm.c} 7470initializes itself, so their values must be run-time constants. 7471They may however depend on command-line flags. 7472 7473@emph{Note:} Some run-time files, such @file{crtstuff.c}, also make 7474use of the @code{@var{FOO}_SECTION_ASM_OP} macros, and expect them 7475to be string literals. 7476 7477Some assemblers require a different string to be written every time a 7478section is selected. If your assembler falls into this category, you 7479should define the @code{TARGET_ASM_INIT_SECTIONS} hook and use 7480@code{get_unnamed_section} to set up the sections. 7481 7482You must always create a @code{text_section}, either by defining 7483@code{TEXT_SECTION_ASM_OP} or by initializing @code{text_section} 7484in @code{TARGET_ASM_INIT_SECTIONS}. The same is true of 7485@code{data_section} and @code{DATA_SECTION_ASM_OP}. If you do not 7486create a distinct @code{readonly_data_section}, the default is to 7487reuse @code{text_section}. 7488 7489All the other @file{varasm.c} sections are optional, and are null 7490if the target does not provide them. 7491 7492@defmac TEXT_SECTION_ASM_OP 7493A C expression whose value is a string, including spacing, containing the 7494assembler operation that should precede instructions and read-only data. 7495Normally @code{"\t.text"} is right. 7496@end defmac 7497 7498@defmac HOT_TEXT_SECTION_NAME 7499If defined, a C string constant for the name of the section containing most 7500frequently executed functions of the program. If not defined, GCC will provide 7501a default definition if the target supports named sections. 7502@end defmac 7503 7504@defmac UNLIKELY_EXECUTED_TEXT_SECTION_NAME 7505If defined, a C string constant for the name of the section containing unlikely 7506executed functions in the program. 7507@end defmac 7508 7509@defmac DATA_SECTION_ASM_OP 7510A C expression whose value is a string, including spacing, containing the 7511assembler operation to identify the following data as writable initialized 7512data. Normally @code{"\t.data"} is right. 7513@end defmac 7514 7515@defmac SDATA_SECTION_ASM_OP 7516If defined, a C expression whose value is a string, including spacing, 7517containing the assembler operation to identify the following data as 7518initialized, writable small data. 7519@end defmac 7520 7521@defmac READONLY_DATA_SECTION_ASM_OP 7522A C expression whose value is a string, including spacing, containing the 7523assembler operation to identify the following data as read-only initialized 7524data. 7525@end defmac 7526 7527@defmac BSS_SECTION_ASM_OP 7528If defined, a C expression whose value is a string, including spacing, 7529containing the assembler operation to identify the following data as 7530uninitialized global data. If not defined, and 7531@code{ASM_OUTPUT_ALIGNED_BSS} not defined, 7532uninitialized global data will be output in the data section if 7533@option{-fno-common} is passed, otherwise @code{ASM_OUTPUT_COMMON} will be 7534used. 7535@end defmac 7536 7537@defmac SBSS_SECTION_ASM_OP 7538If defined, a C expression whose value is a string, including spacing, 7539containing the assembler operation to identify the following data as 7540uninitialized, writable small data. 7541@end defmac 7542 7543@defmac TLS_COMMON_ASM_OP 7544If defined, a C expression whose value is a string containing the 7545assembler operation to identify the following data as thread-local 7546common data. The default is @code{".tls_common"}. 7547@end defmac 7548 7549@defmac TLS_SECTION_ASM_FLAG 7550If defined, a C expression whose value is a character constant 7551containing the flag used to mark a section as a TLS section. The 7552default is @code{'T'}. 7553@end defmac 7554 7555@defmac INIT_SECTION_ASM_OP 7556If defined, a C expression whose value is a string, including spacing, 7557containing the assembler operation to identify the following data as 7558initialization code. If not defined, GCC will assume such a section does 7559not exist. This section has no corresponding @code{init_section} 7560variable; it is used entirely in runtime code. 7561@end defmac 7562 7563@defmac FINI_SECTION_ASM_OP 7564If defined, a C expression whose value is a string, including spacing, 7565containing the assembler operation to identify the following data as 7566finalization code. If not defined, GCC will assume such a section does 7567not exist. This section has no corresponding @code{fini_section} 7568variable; it is used entirely in runtime code. 7569@end defmac 7570 7571@defmac INIT_ARRAY_SECTION_ASM_OP 7572If defined, a C expression whose value is a string, including spacing, 7573containing the assembler operation to identify the following data as 7574part of the @code{.init_array} (or equivalent) section. If not 7575defined, GCC will assume such a section does not exist. Do not define 7576both this macro and @code{INIT_SECTION_ASM_OP}. 7577@end defmac 7578 7579@defmac FINI_ARRAY_SECTION_ASM_OP 7580If defined, a C expression whose value is a string, including spacing, 7581containing the assembler operation to identify the following data as 7582part of the @code{.fini_array} (or equivalent) section. If not 7583defined, GCC will assume such a section does not exist. Do not define 7584both this macro and @code{FINI_SECTION_ASM_OP}. 7585@end defmac 7586 7587@defmac MACH_DEP_SECTION_ASM_FLAG 7588If defined, a C expression whose value is a character constant 7589containing the flag used to mark a machine-dependent section. This 7590corresponds to the @code{SECTION_MACH_DEP} section flag. 7591@end defmac 7592 7593@defmac CRT_CALL_STATIC_FUNCTION (@var{section_op}, @var{function}) 7594If defined, an ASM statement that switches to a different section 7595via @var{section_op}, calls @var{function}, and switches back to 7596the text section. This is used in @file{crtstuff.c} if 7597@code{INIT_SECTION_ASM_OP} or @code{FINI_SECTION_ASM_OP} to calls 7598to initialization and finalization functions from the init and fini 7599sections. By default, this macro uses a simple function call. Some 7600ports need hand-crafted assembly code to avoid dependencies on 7601registers initialized in the function prologue or to ensure that 7602constant pools don't end up too far way in the text section. 7603@end defmac 7604 7605@defmac TARGET_LIBGCC_SDATA_SECTION 7606If defined, a string which names the section into which small 7607variables defined in crtstuff and libgcc should go. This is useful 7608when the target has options for optimizing access to small data, and 7609you want the crtstuff and libgcc routines to be conservative in what 7610they expect of your application yet liberal in what your application 7611expects. For example, for targets with a @code{.sdata} section (like 7612MIPS), you could compile crtstuff with @code{-G 0} so that it doesn't 7613require small data support from your application, but use this macro 7614to put small data into @code{.sdata} so that your application can 7615access these variables whether it uses small data or not. 7616@end defmac 7617 7618@defmac FORCE_CODE_SECTION_ALIGN 7619If defined, an ASM statement that aligns a code section to some 7620arbitrary boundary. This is used to force all fragments of the 7621@code{.init} and @code{.fini} sections to have to same alignment 7622and thus prevent the linker from having to add any padding. 7623@end defmac 7624 7625@defmac JUMP_TABLES_IN_TEXT_SECTION 7626Define this macro to be an expression with a nonzero value if jump 7627tables (for @code{tablejump} insns) should be output in the text 7628section, along with the assembler instructions. Otherwise, the 7629readonly data section is used. 7630 7631This macro is irrelevant if there is no separate readonly data section. 7632@end defmac 7633 7634@deftypefn {Target Hook} void TARGET_ASM_INIT_SECTIONS (void) 7635Define this hook if you need to do something special to set up the 7636@file{varasm.c} sections, or if your target has some special sections 7637of its own that you need to create. 7638 7639GCC calls this hook after processing the command line, but before writing 7640any assembly code, and before calling any of the section-returning hooks 7641described below. 7642@end deftypefn 7643 7644@deftypefn {Target Hook} int TARGET_ASM_RELOC_RW_MASK (void) 7645Return a mask describing how relocations should be treated when 7646selecting sections. Bit 1 should be set if global relocations 7647should be placed in a read-write section; bit 0 should be set if 7648local relocations should be placed in a read-write section. 7649 7650The default version of this function returns 3 when @option{-fpic} 7651is in effect, and 0 otherwise. The hook is typically redefined 7652when the target cannot support (some kinds of) dynamic relocations 7653in read-only sections even in executables. 7654@end deftypefn 7655 7656@deftypefn {Target Hook} bool TARGET_ASM_GENERATE_PIC_ADDR_DIFF_VEC (void) 7657Return true to generate ADDR_DIF_VEC table 7658or false to generate ADDR_VEC table for jumps in case of -fPIC. 7659 7660The default version of this function returns true if flag_pic 7661equals true and false otherwise 7662@end deftypefn 7663 7664@deftypefn {Target Hook} {section *} TARGET_ASM_SELECT_SECTION (tree @var{exp}, int @var{reloc}, unsigned HOST_WIDE_INT @var{align}) 7665Return the section into which @var{exp} should be placed. You can 7666assume that @var{exp} is either a @code{VAR_DECL} node or a constant of 7667some sort. @var{reloc} indicates whether the initial value of @var{exp} 7668requires link-time relocations. Bit 0 is set when variable contains 7669local relocations only, while bit 1 is set for global relocations. 7670@var{align} is the constant alignment in bits. 7671 7672The default version of this function takes care of putting read-only 7673variables in @code{readonly_data_section}. 7674 7675See also @var{USE_SELECT_SECTION_FOR_FUNCTIONS}. 7676@end deftypefn 7677 7678@defmac USE_SELECT_SECTION_FOR_FUNCTIONS 7679Define this macro if you wish TARGET_ASM_SELECT_SECTION to be called 7680for @code{FUNCTION_DECL}s as well as for variables and constants. 7681 7682In the case of a @code{FUNCTION_DECL}, @var{reloc} will be zero if the 7683function has been determined to be likely to be called, and nonzero if 7684it is unlikely to be called. 7685@end defmac 7686 7687@deftypefn {Target Hook} void TARGET_ASM_UNIQUE_SECTION (tree @var{decl}, int @var{reloc}) 7688Build up a unique section name, expressed as a @code{STRING_CST} node, 7689and assign it to @samp{DECL_SECTION_NAME (@var{decl})}. 7690As with @code{TARGET_ASM_SELECT_SECTION}, @var{reloc} indicates whether 7691the initial value of @var{exp} requires link-time relocations. 7692 7693The default version of this function appends the symbol name to the 7694ELF section name that would normally be used for the symbol. For 7695example, the function @code{foo} would be placed in @code{.text.foo}. 7696Whatever the actual target object format, this is often good enough. 7697@end deftypefn 7698 7699@deftypefn {Target Hook} {section *} TARGET_ASM_FUNCTION_RODATA_SECTION (tree @var{decl}) 7700Return the readonly data section associated with 7701@samp{DECL_SECTION_NAME (@var{decl})}. 7702The default version of this function selects @code{.gnu.linkonce.r.name} if 7703the function's section is @code{.gnu.linkonce.t.name}, @code{.rodata.name} 7704if function is in @code{.text.name}, and the normal readonly-data section 7705otherwise. 7706@end deftypefn 7707 7708@deftypevr {Target Hook} {const char *} TARGET_ASM_MERGEABLE_RODATA_PREFIX 7709Usually, the compiler uses the prefix @code{".rodata"} to construct 7710section names for mergeable constant data. Define this macro to override 7711the string if a different section name should be used. 7712@end deftypevr 7713 7714@deftypefn {Target Hook} {section *} TARGET_ASM_TM_CLONE_TABLE_SECTION (void) 7715Return the section that should be used for transactional memory clone tables. 7716@end deftypefn 7717 7718@deftypefn {Target Hook} {section *} TARGET_ASM_SELECT_RTX_SECTION (machine_mode @var{mode}, rtx @var{x}, unsigned HOST_WIDE_INT @var{align}) 7719Return the section into which a constant @var{x}, of mode @var{mode}, 7720should be placed. You can assume that @var{x} is some kind of 7721constant in RTL@. The argument @var{mode} is redundant except in the 7722case of a @code{const_int} rtx. @var{align} is the constant alignment 7723in bits. 7724 7725The default version of this function takes care of putting symbolic 7726constants in @code{flag_pic} mode in @code{data_section} and everything 7727else in @code{readonly_data_section}. 7728@end deftypefn 7729 7730@deftypefn {Target Hook} tree TARGET_MANGLE_DECL_ASSEMBLER_NAME (tree @var{decl}, tree @var{id}) 7731Define this hook if you need to postprocess the assembler name generated 7732by target-independent code. The @var{id} provided to this hook will be 7733the computed name (e.g., the macro @code{DECL_NAME} of the @var{decl} in C, 7734or the mangled name of the @var{decl} in C++). The return value of the 7735hook is an @code{IDENTIFIER_NODE} for the appropriate mangled name on 7736your target system. The default implementation of this hook just 7737returns the @var{id} provided. 7738@end deftypefn 7739 7740@deftypefn {Target Hook} void TARGET_ENCODE_SECTION_INFO (tree @var{decl}, rtx @var{rtl}, int @var{new_decl_p}) 7741Define this hook if references to a symbol or a constant must be 7742treated differently depending on something about the variable or 7743function named by the symbol (such as what section it is in). 7744 7745The hook is executed immediately after rtl has been created for 7746@var{decl}, which may be a variable or function declaration or 7747an entry in the constant pool. In either case, @var{rtl} is the 7748rtl in question. Do @emph{not} use @code{DECL_RTL (@var{decl})} 7749in this hook; that field may not have been initialized yet. 7750 7751In the case of a constant, it is safe to assume that the rtl is 7752a @code{mem} whose address is a @code{symbol_ref}. Most decls 7753will also have this form, but that is not guaranteed. Global 7754register variables, for instance, will have a @code{reg} for their 7755rtl. (Normally the right thing to do with such unusual rtl is 7756leave it alone.) 7757 7758The @var{new_decl_p} argument will be true if this is the first time 7759that @code{TARGET_ENCODE_SECTION_INFO} has been invoked on this decl. It will 7760be false for subsequent invocations, which will happen for duplicate 7761declarations. Whether or not anything must be done for the duplicate 7762declaration depends on whether the hook examines @code{DECL_ATTRIBUTES}. 7763@var{new_decl_p} is always true when the hook is called for a constant. 7764 7765@cindex @code{SYMBOL_REF_FLAG}, in @code{TARGET_ENCODE_SECTION_INFO} 7766The usual thing for this hook to do is to record flags in the 7767@code{symbol_ref}, using @code{SYMBOL_REF_FLAG} or @code{SYMBOL_REF_FLAGS}. 7768Historically, the name string was modified if it was necessary to 7769encode more than one bit of information, but this practice is now 7770discouraged; use @code{SYMBOL_REF_FLAGS}. 7771 7772The default definition of this hook, @code{default_encode_section_info} 7773in @file{varasm.c}, sets a number of commonly-useful bits in 7774@code{SYMBOL_REF_FLAGS}. Check whether the default does what you need 7775before overriding it. 7776@end deftypefn 7777 7778@deftypefn {Target Hook} {const char *} TARGET_STRIP_NAME_ENCODING (const char *@var{name}) 7779Decode @var{name} and return the real name part, sans 7780the characters that @code{TARGET_ENCODE_SECTION_INFO} 7781may have added. 7782@end deftypefn 7783 7784@deftypefn {Target Hook} bool TARGET_IN_SMALL_DATA_P (const_tree @var{exp}) 7785Returns true if @var{exp} should be placed into a ``small data'' section. 7786The default version of this hook always returns false. 7787@end deftypefn 7788 7789@deftypevr {Target Hook} bool TARGET_HAVE_SRODATA_SECTION 7790Contains the value true if the target places read-only 7791``small data'' into a separate section. The default value is false. 7792@end deftypevr 7793 7794@deftypefn {Target Hook} bool TARGET_PROFILE_BEFORE_PROLOGUE (void) 7795It returns true if target wants profile code emitted before prologue. 7796 7797The default version of this hook use the target macro 7798@code{PROFILE_BEFORE_PROLOGUE}. 7799@end deftypefn 7800 7801@deftypefn {Target Hook} bool TARGET_BINDS_LOCAL_P (const_tree @var{exp}) 7802Returns true if @var{exp} names an object for which name resolution 7803rules must resolve to the current ``module'' (dynamic shared library 7804or executable image). 7805 7806The default version of this hook implements the name resolution rules 7807for ELF, which has a looser model of global name binding than other 7808currently supported object file formats. 7809@end deftypefn 7810 7811@deftypevr {Target Hook} bool TARGET_HAVE_TLS 7812Contains the value true if the target supports thread-local storage. 7813The default value is false. 7814@end deftypevr 7815 7816 7817@node PIC 7818@section Position Independent Code 7819@cindex position independent code 7820@cindex PIC 7821 7822This section describes macros that help implement generation of position 7823independent code. Simply defining these macros is not enough to 7824generate valid PIC; you must also add support to the hook 7825@code{TARGET_LEGITIMATE_ADDRESS_P} and to the macro 7826@code{PRINT_OPERAND_ADDRESS}, as well as @code{LEGITIMIZE_ADDRESS}. You 7827must modify the definition of @samp{movsi} to do something appropriate 7828when the source operand contains a symbolic address. You may also 7829need to alter the handling of switch statements so that they use 7830relative addresses. 7831@c i rearranged the order of the macros above to try to force one of 7832@c them to the next line, to eliminate an overfull hbox. --mew 10feb93 7833 7834@defmac PIC_OFFSET_TABLE_REGNUM 7835The register number of the register used to address a table of static 7836data addresses in memory. In some cases this register is defined by a 7837processor's ``application binary interface'' (ABI)@. When this macro 7838is defined, RTL is generated for this register once, as with the stack 7839pointer and frame pointer registers. If this macro is not defined, it 7840is up to the machine-dependent files to allocate such a register (if 7841necessary). Note that this register must be fixed when in use (e.g.@: 7842when @code{flag_pic} is true). 7843@end defmac 7844 7845@defmac PIC_OFFSET_TABLE_REG_CALL_CLOBBERED 7846A C expression that is nonzero if the register defined by 7847@code{PIC_OFFSET_TABLE_REGNUM} is clobbered by calls. If not defined, 7848the default is zero. Do not define 7849this macro if @code{PIC_OFFSET_TABLE_REGNUM} is not defined. 7850@end defmac 7851 7852@defmac LEGITIMATE_PIC_OPERAND_P (@var{x}) 7853A C expression that is nonzero if @var{x} is a legitimate immediate 7854operand on the target machine when generating position independent code. 7855You can assume that @var{x} satisfies @code{CONSTANT_P}, so you need not 7856check this. You can also assume @var{flag_pic} is true, so you need not 7857check it either. You need not define this macro if all constants 7858(including @code{SYMBOL_REF}) can be immediate operands when generating 7859position independent code. 7860@end defmac 7861 7862@node Assembler Format 7863@section Defining the Output Assembler Language 7864 7865This section describes macros whose principal purpose is to describe how 7866to write instructions in assembler language---rather than what the 7867instructions do. 7868 7869@menu 7870* File Framework:: Structural information for the assembler file. 7871* Data Output:: Output of constants (numbers, strings, addresses). 7872* Uninitialized Data:: Output of uninitialized variables. 7873* Label Output:: Output and generation of labels. 7874* Initialization:: General principles of initialization 7875 and termination routines. 7876* Macros for Initialization:: 7877 Specific macros that control the handling of 7878 initialization and termination routines. 7879* Instruction Output:: Output of actual instructions. 7880* Dispatch Tables:: Output of jump tables. 7881* Exception Region Output:: Output of exception region code. 7882* Alignment Output:: Pseudo ops for alignment and skipping data. 7883@end menu 7884 7885@node File Framework 7886@subsection The Overall Framework of an Assembler File 7887@cindex assembler format 7888@cindex output of assembler code 7889 7890@c prevent bad page break with this line 7891This describes the overall framework of an assembly file. 7892 7893@findex default_file_start 7894@deftypefn {Target Hook} void TARGET_ASM_FILE_START (void) 7895Output to @code{asm_out_file} any text which the assembler expects to 7896find at the beginning of a file. The default behavior is controlled 7897by two flags, documented below. Unless your target's assembler is 7898quite unusual, if you override the default, you should call 7899@code{default_file_start} at some point in your target hook. This 7900lets other target files rely on these variables. 7901@end deftypefn 7902 7903@deftypevr {Target Hook} bool TARGET_ASM_FILE_START_APP_OFF 7904If this flag is true, the text of the macro @code{ASM_APP_OFF} will be 7905printed as the very first line in the assembly file, unless 7906@option{-fverbose-asm} is in effect. (If that macro has been defined 7907to the empty string, this variable has no effect.) With the normal 7908definition of @code{ASM_APP_OFF}, the effect is to notify the GNU 7909assembler that it need not bother stripping comments or extra 7910whitespace from its input. This allows it to work a bit faster. 7911 7912The default is false. You should not set it to true unless you have 7913verified that your port does not generate any extra whitespace or 7914comments that will cause GAS to issue errors in NO_APP mode. 7915@end deftypevr 7916 7917@deftypevr {Target Hook} bool TARGET_ASM_FILE_START_FILE_DIRECTIVE 7918If this flag is true, @code{output_file_directive} will be called 7919for the primary source file, immediately after printing 7920@code{ASM_APP_OFF} (if that is enabled). Most ELF assemblers expect 7921this to be done. The default is false. 7922@end deftypevr 7923 7924@deftypefn {Target Hook} void TARGET_ASM_FILE_END (void) 7925Output to @code{asm_out_file} any text which the assembler expects 7926to find at the end of a file. The default is to output nothing. 7927@end deftypefn 7928 7929@deftypefun void file_end_indicate_exec_stack () 7930Some systems use a common convention, the @samp{.note.GNU-stack} 7931special section, to indicate whether or not an object file relies on 7932the stack being executable. If your system uses this convention, you 7933should define @code{TARGET_ASM_FILE_END} to this function. If you 7934need to do other things in that hook, have your hook function call 7935this function. 7936@end deftypefun 7937 7938@deftypefn {Target Hook} void TARGET_ASM_LTO_START (void) 7939Output to @code{asm_out_file} any text which the assembler expects 7940to find at the start of an LTO section. The default is to output 7941nothing. 7942@end deftypefn 7943 7944@deftypefn {Target Hook} void TARGET_ASM_LTO_END (void) 7945Output to @code{asm_out_file} any text which the assembler expects 7946to find at the end of an LTO section. The default is to output 7947nothing. 7948@end deftypefn 7949 7950@deftypefn {Target Hook} void TARGET_ASM_CODE_END (void) 7951Output to @code{asm_out_file} any text which is needed before emitting 7952unwind info and debug info at the end of a file. Some targets emit 7953here PIC setup thunks that cannot be emitted at the end of file, 7954because they couldn't have unwind info then. The default is to output 7955nothing. 7956@end deftypefn 7957 7958@defmac ASM_COMMENT_START 7959A C string constant describing how to begin a comment in the target 7960assembler language. The compiler assumes that the comment will end at 7961the end of the line. 7962@end defmac 7963 7964@defmac ASM_APP_ON 7965A C string constant for text to be output before each @code{asm} 7966statement or group of consecutive ones. Normally this is 7967@code{"#APP"}, which is a comment that has no effect on most 7968assemblers but tells the GNU assembler that it must check the lines 7969that follow for all valid assembler constructs. 7970@end defmac 7971 7972@defmac ASM_APP_OFF 7973A C string constant for text to be output after each @code{asm} 7974statement or group of consecutive ones. Normally this is 7975@code{"#NO_APP"}, which tells the GNU assembler to resume making the 7976time-saving assumptions that are valid for ordinary compiler output. 7977@end defmac 7978 7979@defmac ASM_OUTPUT_SOURCE_FILENAME (@var{stream}, @var{name}) 7980A C statement to output COFF information or DWARF debugging information 7981which indicates that filename @var{name} is the current source file to 7982the stdio stream @var{stream}. 7983 7984This macro need not be defined if the standard form of output 7985for the file format in use is appropriate. 7986@end defmac 7987 7988@deftypefn {Target Hook} void TARGET_ASM_OUTPUT_SOURCE_FILENAME (FILE *@var{file}, const char *@var{name}) 7989Output DWARF debugging information which indicates that filename @var{name} is the current source file to the stdio stream @var{file}. 7990 7991 This target hook need not be defined if the standard form of output for the file format in use is appropriate. 7992@end deftypefn 7993 7994@deftypefn {Target Hook} void TARGET_ASM_OUTPUT_IDENT (const char *@var{name}) 7995Output a string based on @var{name}, suitable for the @samp{#ident} directive, or the equivalent directive or pragma in non-C-family languages. If this hook is not defined, nothing is output for the @samp{#ident} directive. 7996@end deftypefn 7997 7998@defmac OUTPUT_QUOTED_STRING (@var{stream}, @var{string}) 7999A C statement to output the string @var{string} to the stdio stream 8000@var{stream}. If you do not call the function @code{output_quoted_string} 8001in your config files, GCC will only call it to output filenames to 8002the assembler source. So you can use it to canonicalize the format 8003of the filename using this macro. 8004@end defmac 8005 8006@deftypefn {Target Hook} void TARGET_ASM_NAMED_SECTION (const char *@var{name}, unsigned int @var{flags}, tree @var{decl}) 8007Output assembly directives to switch to section @var{name}. The section 8008should have attributes as specified by @var{flags}, which is a bit mask 8009of the @code{SECTION_*} flags defined in @file{output.h}. If @var{decl} 8010is non-NULL, it is the @code{VAR_DECL} or @code{FUNCTION_DECL} with which 8011this section is associated. 8012@end deftypefn 8013 8014@deftypefn {Target Hook} bool TARGET_ASM_ELF_FLAGS_NUMERIC (unsigned int @var{flags}, unsigned int *@var{num}) 8015This hook can be used to encode ELF section flags for which no letter 8016code has been defined in the assembler. It is called by 8017@code{default_asm_named_section} whenever the section flags need to be 8018emitted in the assembler output. If the hook returns true, then the 8019numerical value for ELF section flags should be calculated from 8020@var{flags} and saved in @var{*num}; the value is printed out instead of the 8021normal sequence of letter codes. If the hook is not defined, or if it 8022returns false, then @var{num} is ignored and the traditional letter sequence 8023is emitted. 8024@end deftypefn 8025 8026@deftypefn {Target Hook} {section *} TARGET_ASM_FUNCTION_SECTION (tree @var{decl}, enum node_frequency @var{freq}, bool @var{startup}, bool @var{exit}) 8027Return preferred text (sub)section for function @var{decl}. 8028Main purpose of this function is to separate cold, normal and hot 8029functions. @var{startup} is true when function is known to be used only 8030at startup (from static constructors or it is @code{main()}). 8031@var{exit} is true when function is known to be used only at exit 8032(from static destructors). 8033Return NULL if function should go to default text section. 8034@end deftypefn 8035 8036@deftypefn {Target Hook} void TARGET_ASM_FUNCTION_SWITCHED_TEXT_SECTIONS (FILE *@var{file}, tree @var{decl}, bool @var{new_is_cold}) 8037Used by the target to emit any assembler directives or additional labels needed when a function is partitioned between different sections. Output should be written to @var{file}. The function decl is available as @var{decl} and the new section is `cold' if @var{new_is_cold} is @code{true}. 8038@end deftypefn 8039 8040@deftypevr {Common Target Hook} bool TARGET_HAVE_NAMED_SECTIONS 8041This flag is true if the target supports @code{TARGET_ASM_NAMED_SECTION}. 8042It must not be modified by command-line option processing. 8043@end deftypevr 8044 8045@anchor{TARGET_HAVE_SWITCHABLE_BSS_SECTIONS} 8046@deftypevr {Target Hook} bool TARGET_HAVE_SWITCHABLE_BSS_SECTIONS 8047This flag is true if we can create zeroed data by switching to a BSS 8048section and then using @code{ASM_OUTPUT_SKIP} to allocate the space. 8049This is true on most ELF targets. 8050@end deftypevr 8051 8052@deftypefn {Target Hook} {unsigned int} TARGET_SECTION_TYPE_FLAGS (tree @var{decl}, const char *@var{name}, int @var{reloc}) 8053Choose a set of section attributes for use by @code{TARGET_ASM_NAMED_SECTION} 8054based on a variable or function decl, a section name, and whether or not the 8055declaration's initializer may contain runtime relocations. @var{decl} may be 8056null, in which case read-write data should be assumed. 8057 8058The default version of this function handles choosing code vs data, 8059read-only vs read-write data, and @code{flag_pic}. You should only 8060need to override this if your target has special flags that might be 8061set via @code{__attribute__}. 8062@end deftypefn 8063 8064@deftypefn {Target Hook} int TARGET_ASM_RECORD_GCC_SWITCHES (print_switch_type @var{type}, const char *@var{text}) 8065Provides the target with the ability to record the gcc command line 8066switches that have been passed to the compiler, and options that are 8067enabled. The @var{type} argument specifies what is being recorded. 8068It can take the following values: 8069 8070@table @gcctabopt 8071@item SWITCH_TYPE_PASSED 8072@var{text} is a command line switch that has been set by the user. 8073 8074@item SWITCH_TYPE_ENABLED 8075@var{text} is an option which has been enabled. This might be as a 8076direct result of a command line switch, or because it is enabled by 8077default or because it has been enabled as a side effect of a different 8078command line switch. For example, the @option{-O2} switch enables 8079various different individual optimization passes. 8080 8081@item SWITCH_TYPE_DESCRIPTIVE 8082@var{text} is either NULL or some descriptive text which should be 8083ignored. If @var{text} is NULL then it is being used to warn the 8084target hook that either recording is starting or ending. The first 8085time @var{type} is SWITCH_TYPE_DESCRIPTIVE and @var{text} is NULL, the 8086warning is for start up and the second time the warning is for 8087wind down. This feature is to allow the target hook to make any 8088necessary preparations before it starts to record switches and to 8089perform any necessary tidying up after it has finished recording 8090switches. 8091 8092@item SWITCH_TYPE_LINE_START 8093This option can be ignored by this target hook. 8094 8095@item SWITCH_TYPE_LINE_END 8096This option can be ignored by this target hook. 8097@end table 8098 8099The hook's return value must be zero. Other return values may be 8100supported in the future. 8101 8102By default this hook is set to NULL, but an example implementation is 8103provided for ELF based targets. Called @var{elf_record_gcc_switches}, 8104it records the switches as ASCII text inside a new, string mergeable 8105section in the assembler output file. The name of the new section is 8106provided by the @code{TARGET_ASM_RECORD_GCC_SWITCHES_SECTION} target 8107hook. 8108@end deftypefn 8109 8110@deftypevr {Target Hook} {const char *} TARGET_ASM_RECORD_GCC_SWITCHES_SECTION 8111This is the name of the section that will be created by the example 8112ELF implementation of the @code{TARGET_ASM_RECORD_GCC_SWITCHES} target 8113hook. 8114@end deftypevr 8115 8116@need 2000 8117@node Data Output 8118@subsection Output of Data 8119 8120 8121@deftypevr {Target Hook} {const char *} TARGET_ASM_BYTE_OP 8122@deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_HI_OP 8123@deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_PSI_OP 8124@deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_SI_OP 8125@deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_PDI_OP 8126@deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_DI_OP 8127@deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_PTI_OP 8128@deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_TI_OP 8129@deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_HI_OP 8130@deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_PSI_OP 8131@deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_SI_OP 8132@deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_PDI_OP 8133@deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_DI_OP 8134@deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_PTI_OP 8135@deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_TI_OP 8136These hooks specify assembly directives for creating certain kinds 8137of integer object. The @code{TARGET_ASM_BYTE_OP} directive creates a 8138byte-sized object, the @code{TARGET_ASM_ALIGNED_HI_OP} one creates an 8139aligned two-byte object, and so on. Any of the hooks may be 8140@code{NULL}, indicating that no suitable directive is available. 8141 8142The compiler will print these strings at the start of a new line, 8143followed immediately by the object's initial value. In most cases, 8144the string should contain a tab, a pseudo-op, and then another tab. 8145@end deftypevr 8146 8147@deftypefn {Target Hook} bool TARGET_ASM_INTEGER (rtx @var{x}, unsigned int @var{size}, int @var{aligned_p}) 8148The @code{assemble_integer} function uses this hook to output an 8149integer object. @var{x} is the object's value, @var{size} is its size 8150in bytes and @var{aligned_p} indicates whether it is aligned. The 8151function should return @code{true} if it was able to output the 8152object. If it returns false, @code{assemble_integer} will try to 8153split the object into smaller parts. 8154 8155The default implementation of this hook will use the 8156@code{TARGET_ASM_BYTE_OP} family of strings, returning @code{false} 8157when the relevant string is @code{NULL}. 8158@end deftypefn 8159 8160@deftypefn {Target Hook} void TARGET_ASM_DECL_END (void) 8161Define this hook if the target assembler requires a special marker to 8162terminate an initialized variable declaration. 8163@end deftypefn 8164 8165@deftypefn {Target Hook} bool TARGET_ASM_OUTPUT_ADDR_CONST_EXTRA (FILE *@var{file}, rtx @var{x}) 8166A target hook to recognize @var{rtx} patterns that @code{output_addr_const} 8167can't deal with, and output assembly code to @var{file} corresponding to 8168the pattern @var{x}. This may be used to allow machine-dependent 8169@code{UNSPEC}s to appear within constants. 8170 8171If target hook fails to recognize a pattern, it must return @code{false}, 8172so that a standard error message is printed. If it prints an error message 8173itself, by calling, for example, @code{output_operand_lossage}, it may just 8174return @code{true}. 8175@end deftypefn 8176 8177@defmac ASM_OUTPUT_ASCII (@var{stream}, @var{ptr}, @var{len}) 8178A C statement to output to the stdio stream @var{stream} an assembler 8179instruction to assemble a string constant containing the @var{len} 8180bytes at @var{ptr}. @var{ptr} will be a C expression of type 8181@code{char *} and @var{len} a C expression of type @code{int}. 8182 8183If the assembler has a @code{.ascii} pseudo-op as found in the 8184Berkeley Unix assembler, do not define the macro 8185@code{ASM_OUTPUT_ASCII}. 8186@end defmac 8187 8188@defmac ASM_OUTPUT_FDESC (@var{stream}, @var{decl}, @var{n}) 8189A C statement to output word @var{n} of a function descriptor for 8190@var{decl}. This must be defined if @code{TARGET_VTABLE_USES_DESCRIPTORS} 8191is defined, and is otherwise unused. 8192@end defmac 8193 8194@defmac CONSTANT_POOL_BEFORE_FUNCTION 8195You may define this macro as a C expression. You should define the 8196expression to have a nonzero value if GCC should output the constant 8197pool for a function before the code for the function, or a zero value if 8198GCC should output the constant pool after the function. If you do 8199not define this macro, the usual case, GCC will output the constant 8200pool before the function. 8201@end defmac 8202 8203@defmac ASM_OUTPUT_POOL_PROLOGUE (@var{file}, @var{funname}, @var{fundecl}, @var{size}) 8204A C statement to output assembler commands to define the start of the 8205constant pool for a function. @var{funname} is a string giving 8206the name of the function. Should the return type of the function 8207be required, it can be obtained via @var{fundecl}. @var{size} 8208is the size, in bytes, of the constant pool that will be written 8209immediately after this call. 8210 8211If no constant-pool prefix is required, the usual case, this macro need 8212not be defined. 8213@end defmac 8214 8215@defmac ASM_OUTPUT_SPECIAL_POOL_ENTRY (@var{file}, @var{x}, @var{mode}, @var{align}, @var{labelno}, @var{jumpto}) 8216A C statement (with or without semicolon) to output a constant in the 8217constant pool, if it needs special treatment. (This macro need not do 8218anything for RTL expressions that can be output normally.) 8219 8220The argument @var{file} is the standard I/O stream to output the 8221assembler code on. @var{x} is the RTL expression for the constant to 8222output, and @var{mode} is the machine mode (in case @var{x} is a 8223@samp{const_int}). @var{align} is the required alignment for the value 8224@var{x}; you should output an assembler directive to force this much 8225alignment. 8226 8227The argument @var{labelno} is a number to use in an internal label for 8228the address of this pool entry. The definition of this macro is 8229responsible for outputting the label definition at the proper place. 8230Here is how to do this: 8231 8232@smallexample 8233@code{(*targetm.asm_out.internal_label)} (@var{file}, "LC", @var{labelno}); 8234@end smallexample 8235 8236When you output a pool entry specially, you should end with a 8237@code{goto} to the label @var{jumpto}. This will prevent the same pool 8238entry from being output a second time in the usual manner. 8239 8240You need not define this macro if it would do nothing. 8241@end defmac 8242 8243@defmac ASM_OUTPUT_POOL_EPILOGUE (@var{file} @var{funname} @var{fundecl} @var{size}) 8244A C statement to output assembler commands to at the end of the constant 8245pool for a function. @var{funname} is a string giving the name of the 8246function. Should the return type of the function be required, you can 8247obtain it via @var{fundecl}. @var{size} is the size, in bytes, of the 8248constant pool that GCC wrote immediately before this call. 8249 8250If no constant-pool epilogue is required, the usual case, you need not 8251define this macro. 8252@end defmac 8253 8254@defmac IS_ASM_LOGICAL_LINE_SEPARATOR (@var{C}, @var{STR}) 8255Define this macro as a C expression which is nonzero if @var{C} is 8256used as a logical line separator by the assembler. @var{STR} points 8257to the position in the string where @var{C} was found; this can be used if 8258a line separator uses multiple characters. 8259 8260If you do not define this macro, the default is that only 8261the character @samp{;} is treated as a logical line separator. 8262@end defmac 8263 8264@deftypevr {Target Hook} {const char *} TARGET_ASM_OPEN_PAREN 8265@deftypevrx {Target Hook} {const char *} TARGET_ASM_CLOSE_PAREN 8266These target hooks are C string constants, describing the syntax in the 8267assembler for grouping arithmetic expressions. If not overridden, they 8268default to normal parentheses, which is correct for most assemblers. 8269@end deftypevr 8270 8271These macros are provided by @file{real.h} for writing the definitions 8272of @code{ASM_OUTPUT_DOUBLE} and the like: 8273 8274@defmac REAL_VALUE_TO_TARGET_SINGLE (@var{x}, @var{l}) 8275@defmacx REAL_VALUE_TO_TARGET_DOUBLE (@var{x}, @var{l}) 8276@defmacx REAL_VALUE_TO_TARGET_LONG_DOUBLE (@var{x}, @var{l}) 8277@defmacx REAL_VALUE_TO_TARGET_DECIMAL32 (@var{x}, @var{l}) 8278@defmacx REAL_VALUE_TO_TARGET_DECIMAL64 (@var{x}, @var{l}) 8279@defmacx REAL_VALUE_TO_TARGET_DECIMAL128 (@var{x}, @var{l}) 8280These translate @var{x}, of type @code{REAL_VALUE_TYPE}, to the 8281target's floating point representation, and store its bit pattern in 8282the variable @var{l}. For @code{REAL_VALUE_TO_TARGET_SINGLE} and 8283@code{REAL_VALUE_TO_TARGET_DECIMAL32}, this variable should be a 8284simple @code{long int}. For the others, it should be an array of 8285@code{long int}. The number of elements in this array is determined 8286by the size of the desired target floating point data type: 32 bits of 8287it go in each @code{long int} array element. Each array element holds 828832 bits of the result, even if @code{long int} is wider than 32 bits 8289on the host machine. 8290 8291The array element values are designed so that you can print them out 8292using @code{fprintf} in the order they should appear in the target 8293machine's memory. 8294@end defmac 8295 8296@node Uninitialized Data 8297@subsection Output of Uninitialized Variables 8298 8299Each of the macros in this section is used to do the whole job of 8300outputting a single uninitialized variable. 8301 8302@defmac ASM_OUTPUT_COMMON (@var{stream}, @var{name}, @var{size}, @var{rounded}) 8303A C statement (sans semicolon) to output to the stdio stream 8304@var{stream} the assembler definition of a common-label named 8305@var{name} whose size is @var{size} bytes. The variable @var{rounded} 8306is the size rounded up to whatever alignment the caller wants. It is 8307possible that @var{size} may be zero, for instance if a struct with no 8308other member than a zero-length array is defined. In this case, the 8309backend must output a symbol definition that allocates at least one 8310byte, both so that the address of the resulting object does not compare 8311equal to any other, and because some object formats cannot even express 8312the concept of a zero-sized common symbol, as that is how they represent 8313an ordinary undefined external. 8314 8315Use the expression @code{assemble_name (@var{stream}, @var{name})} to 8316output the name itself; before and after that, output the additional 8317assembler syntax for defining the name, and a newline. 8318 8319This macro controls how the assembler definitions of uninitialized 8320common global variables are output. 8321@end defmac 8322 8323@defmac ASM_OUTPUT_ALIGNED_COMMON (@var{stream}, @var{name}, @var{size}, @var{alignment}) 8324Like @code{ASM_OUTPUT_COMMON} except takes the required alignment as a 8325separate, explicit argument. If you define this macro, it is used in 8326place of @code{ASM_OUTPUT_COMMON}, and gives you more flexibility in 8327handling the required alignment of the variable. The alignment is specified 8328as the number of bits. 8329@end defmac 8330 8331@defmac ASM_OUTPUT_ALIGNED_DECL_COMMON (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment}) 8332Like @code{ASM_OUTPUT_ALIGNED_COMMON} except that @var{decl} of the 8333variable to be output, if there is one, or @code{NULL_TREE} if there 8334is no corresponding variable. If you define this macro, GCC will use it 8335in place of both @code{ASM_OUTPUT_COMMON} and 8336@code{ASM_OUTPUT_ALIGNED_COMMON}. Define this macro when you need to see 8337the variable's decl in order to chose what to output. 8338@end defmac 8339 8340@defmac ASM_OUTPUT_ALIGNED_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment}) 8341A C statement (sans semicolon) to output to the stdio stream 8342@var{stream} the assembler definition of uninitialized global @var{decl} named 8343@var{name} whose size is @var{size} bytes. The variable @var{alignment} 8344is the alignment specified as the number of bits. 8345 8346Try to use function @code{asm_output_aligned_bss} defined in file 8347@file{varasm.c} when defining this macro. If unable, use the expression 8348@code{assemble_name (@var{stream}, @var{name})} to output the name itself; 8349before and after that, output the additional assembler syntax for defining 8350the name, and a newline. 8351 8352There are two ways of handling global BSS@. One is to define this macro. 8353The other is to have @code{TARGET_ASM_SELECT_SECTION} return a 8354switchable BSS section (@pxref{TARGET_HAVE_SWITCHABLE_BSS_SECTIONS}). 8355You do not need to do both. 8356 8357Some languages do not have @code{common} data, and require a 8358non-common form of global BSS in order to handle uninitialized globals 8359efficiently. C++ is one example of this. However, if the target does 8360not support global BSS, the front end may choose to make globals 8361common in order to save space in the object file. 8362@end defmac 8363 8364@defmac ASM_OUTPUT_LOCAL (@var{stream}, @var{name}, @var{size}, @var{rounded}) 8365A C statement (sans semicolon) to output to the stdio stream 8366@var{stream} the assembler definition of a local-common-label named 8367@var{name} whose size is @var{size} bytes. The variable @var{rounded} 8368is the size rounded up to whatever alignment the caller wants. 8369 8370Use the expression @code{assemble_name (@var{stream}, @var{name})} to 8371output the name itself; before and after that, output the additional 8372assembler syntax for defining the name, and a newline. 8373 8374This macro controls how the assembler definitions of uninitialized 8375static variables are output. 8376@end defmac 8377 8378@defmac ASM_OUTPUT_ALIGNED_LOCAL (@var{stream}, @var{name}, @var{size}, @var{alignment}) 8379Like @code{ASM_OUTPUT_LOCAL} except takes the required alignment as a 8380separate, explicit argument. If you define this macro, it is used in 8381place of @code{ASM_OUTPUT_LOCAL}, and gives you more flexibility in 8382handling the required alignment of the variable. The alignment is specified 8383as the number of bits. 8384@end defmac 8385 8386@defmac ASM_OUTPUT_ALIGNED_DECL_LOCAL (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment}) 8387Like @code{ASM_OUTPUT_ALIGNED_LOCAL} except that @var{decl} of the 8388variable to be output, if there is one, or @code{NULL_TREE} if there 8389is no corresponding variable. If you define this macro, GCC will use it 8390in place of both @code{ASM_OUTPUT_LOCAL} and 8391@code{ASM_OUTPUT_ALIGNED_LOCAL}. Define this macro when you need to see 8392the variable's decl in order to chose what to output. 8393@end defmac 8394 8395@node Label Output 8396@subsection Output and Generation of Labels 8397 8398@c prevent bad page break with this line 8399This is about outputting labels. 8400 8401@findex assemble_name 8402@defmac ASM_OUTPUT_LABEL (@var{stream}, @var{name}) 8403A C statement (sans semicolon) to output to the stdio stream 8404@var{stream} the assembler definition of a label named @var{name}. 8405Use the expression @code{assemble_name (@var{stream}, @var{name})} to 8406output the name itself; before and after that, output the additional 8407assembler syntax for defining the name, and a newline. A default 8408definition of this macro is provided which is correct for most systems. 8409@end defmac 8410 8411@defmac ASM_OUTPUT_FUNCTION_LABEL (@var{stream}, @var{name}, @var{decl}) 8412A C statement (sans semicolon) to output to the stdio stream 8413@var{stream} the assembler definition of a label named @var{name} of 8414a function. 8415Use the expression @code{assemble_name (@var{stream}, @var{name})} to 8416output the name itself; before and after that, output the additional 8417assembler syntax for defining the name, and a newline. A default 8418definition of this macro is provided which is correct for most systems. 8419 8420If this macro is not defined, then the function name is defined in the 8421usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}). 8422@end defmac 8423 8424@findex assemble_name_raw 8425@defmac ASM_OUTPUT_INTERNAL_LABEL (@var{stream}, @var{name}) 8426Identical to @code{ASM_OUTPUT_LABEL}, except that @var{name} is known 8427to refer to a compiler-generated label. The default definition uses 8428@code{assemble_name_raw}, which is like @code{assemble_name} except 8429that it is more efficient. 8430@end defmac 8431 8432@defmac SIZE_ASM_OP 8433A C string containing the appropriate assembler directive to specify the 8434size of a symbol, without any arguments. On systems that use ELF, the 8435default (in @file{config/elfos.h}) is @samp{"\t.size\t"}; on other 8436systems, the default is not to define this macro. 8437 8438Define this macro only if it is correct to use the default definitions 8439of @code{ASM_OUTPUT_SIZE_DIRECTIVE} and @code{ASM_OUTPUT_MEASURED_SIZE} 8440for your system. If you need your own custom definitions of those 8441macros, or if you do not need explicit symbol sizes at all, do not 8442define this macro. 8443@end defmac 8444 8445@defmac ASM_OUTPUT_SIZE_DIRECTIVE (@var{stream}, @var{name}, @var{size}) 8446A C statement (sans semicolon) to output to the stdio stream 8447@var{stream} a directive telling the assembler that the size of the 8448symbol @var{name} is @var{size}. @var{size} is a @code{HOST_WIDE_INT}. 8449If you define @code{SIZE_ASM_OP}, a default definition of this macro is 8450provided. 8451@end defmac 8452 8453@defmac ASM_OUTPUT_MEASURED_SIZE (@var{stream}, @var{name}) 8454A C statement (sans semicolon) to output to the stdio stream 8455@var{stream} a directive telling the assembler to calculate the size of 8456the symbol @var{name} by subtracting its address from the current 8457address. 8458 8459If you define @code{SIZE_ASM_OP}, a default definition of this macro is 8460provided. The default assumes that the assembler recognizes a special 8461@samp{.} symbol as referring to the current address, and can calculate 8462the difference between this and another symbol. If your assembler does 8463not recognize @samp{.} or cannot do calculations with it, you will need 8464to redefine @code{ASM_OUTPUT_MEASURED_SIZE} to use some other technique. 8465@end defmac 8466 8467@defmac NO_DOLLAR_IN_LABEL 8468Define this macro if the assembler does not accept the character 8469@samp{$} in label names. By default constructors and destructors in 8470G++ have @samp{$} in the identifiers. If this macro is defined, 8471@samp{.} is used instead. 8472@end defmac 8473 8474@defmac NO_DOT_IN_LABEL 8475Define this macro if the assembler does not accept the character 8476@samp{.} in label names. By default constructors and destructors in G++ 8477have names that use @samp{.}. If this macro is defined, these names 8478are rewritten to avoid @samp{.}. 8479@end defmac 8480 8481@defmac TYPE_ASM_OP 8482A C string containing the appropriate assembler directive to specify the 8483type of a symbol, without any arguments. On systems that use ELF, the 8484default (in @file{config/elfos.h}) is @samp{"\t.type\t"}; on other 8485systems, the default is not to define this macro. 8486 8487Define this macro only if it is correct to use the default definition of 8488@code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own 8489custom definition of this macro, or if you do not need explicit symbol 8490types at all, do not define this macro. 8491@end defmac 8492 8493@defmac TYPE_OPERAND_FMT 8494A C string which specifies (using @code{printf} syntax) the format of 8495the second operand to @code{TYPE_ASM_OP}. On systems that use ELF, the 8496default (in @file{config/elfos.h}) is @samp{"@@%s"}; on other systems, 8497the default is not to define this macro. 8498 8499Define this macro only if it is correct to use the default definition of 8500@code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own 8501custom definition of this macro, or if you do not need explicit symbol 8502types at all, do not define this macro. 8503@end defmac 8504 8505@defmac ASM_OUTPUT_TYPE_DIRECTIVE (@var{stream}, @var{type}) 8506A C statement (sans semicolon) to output to the stdio stream 8507@var{stream} a directive telling the assembler that the type of the 8508symbol @var{name} is @var{type}. @var{type} is a C string; currently, 8509that string is always either @samp{"function"} or @samp{"object"}, but 8510you should not count on this. 8511 8512If you define @code{TYPE_ASM_OP} and @code{TYPE_OPERAND_FMT}, a default 8513definition of this macro is provided. 8514@end defmac 8515 8516@defmac ASM_DECLARE_FUNCTION_NAME (@var{stream}, @var{name}, @var{decl}) 8517A C statement (sans semicolon) to output to the stdio stream 8518@var{stream} any text necessary for declaring the name @var{name} of a 8519function which is being defined. This macro is responsible for 8520outputting the label definition (perhaps using 8521@code{ASM_OUTPUT_FUNCTION_LABEL}). The argument @var{decl} is the 8522@code{FUNCTION_DECL} tree node representing the function. 8523 8524If this macro is not defined, then the function name is defined in the 8525usual manner as a label (by means of @code{ASM_OUTPUT_FUNCTION_LABEL}). 8526 8527You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in the definition 8528of this macro. 8529@end defmac 8530 8531@defmac ASM_DECLARE_FUNCTION_SIZE (@var{stream}, @var{name}, @var{decl}) 8532A C statement (sans semicolon) to output to the stdio stream 8533@var{stream} any text necessary for declaring the size of a function 8534which is being defined. The argument @var{name} is the name of the 8535function. The argument @var{decl} is the @code{FUNCTION_DECL} tree node 8536representing the function. 8537 8538If this macro is not defined, then the function size is not defined. 8539 8540You may wish to use @code{ASM_OUTPUT_MEASURED_SIZE} in the definition 8541of this macro. 8542@end defmac 8543 8544@defmac ASM_DECLARE_COLD_FUNCTION_NAME (@var{stream}, @var{name}, @var{decl}) 8545A C statement (sans semicolon) to output to the stdio stream 8546@var{stream} any text necessary for declaring the name @var{name} of a 8547cold function partition which is being defined. This macro is responsible 8548for outputting the label definition (perhaps using 8549@code{ASM_OUTPUT_FUNCTION_LABEL}). The argument @var{decl} is the 8550@code{FUNCTION_DECL} tree node representing the function. 8551 8552If this macro is not defined, then the cold partition name is defined in the 8553usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}). 8554 8555You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in the definition 8556of this macro. 8557@end defmac 8558 8559@defmac ASM_DECLARE_COLD_FUNCTION_SIZE (@var{stream}, @var{name}, @var{decl}) 8560A C statement (sans semicolon) to output to the stdio stream 8561@var{stream} any text necessary for declaring the size of a cold function 8562partition which is being defined. The argument @var{name} is the name of the 8563cold partition of the function. The argument @var{decl} is the 8564@code{FUNCTION_DECL} tree node representing the function. 8565 8566If this macro is not defined, then the partition size is not defined. 8567 8568You may wish to use @code{ASM_OUTPUT_MEASURED_SIZE} in the definition 8569of this macro. 8570@end defmac 8571 8572@defmac ASM_DECLARE_OBJECT_NAME (@var{stream}, @var{name}, @var{decl}) 8573A C statement (sans semicolon) to output to the stdio stream 8574@var{stream} any text necessary for declaring the name @var{name} of an 8575initialized variable which is being defined. This macro must output the 8576label definition (perhaps using @code{ASM_OUTPUT_LABEL}). The argument 8577@var{decl} is the @code{VAR_DECL} tree node representing the variable. 8578 8579If this macro is not defined, then the variable name is defined in the 8580usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}). 8581 8582You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} and/or 8583@code{ASM_OUTPUT_SIZE_DIRECTIVE} in the definition of this macro. 8584@end defmac 8585 8586@deftypefn {Target Hook} void TARGET_ASM_DECLARE_CONSTANT_NAME (FILE *@var{file}, const char *@var{name}, const_tree @var{expr}, HOST_WIDE_INT @var{size}) 8587A target hook to output to the stdio stream @var{file} any text necessary 8588for declaring the name @var{name} of a constant which is being defined. This 8589target hook is responsible for outputting the label definition (perhaps using 8590@code{assemble_label}). The argument @var{exp} is the value of the constant, 8591and @var{size} is the size of the constant in bytes. The @var{name} 8592will be an internal label. 8593 8594The default version of this target hook, define the @var{name} in the 8595usual manner as a label (by means of @code{assemble_label}). 8596 8597You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in this target hook. 8598@end deftypefn 8599 8600@defmac ASM_DECLARE_REGISTER_GLOBAL (@var{stream}, @var{decl}, @var{regno}, @var{name}) 8601A C statement (sans semicolon) to output to the stdio stream 8602@var{stream} any text necessary for claiming a register @var{regno} 8603for a global variable @var{decl} with name @var{name}. 8604 8605If you don't define this macro, that is equivalent to defining it to do 8606nothing. 8607@end defmac 8608 8609@defmac ASM_FINISH_DECLARE_OBJECT (@var{stream}, @var{decl}, @var{toplevel}, @var{atend}) 8610A C statement (sans semicolon) to finish up declaring a variable name 8611once the compiler has processed its initializer fully and thus has had a 8612chance to determine the size of an array when controlled by an 8613initializer. This is used on systems where it's necessary to declare 8614something about the size of the object. 8615 8616If you don't define this macro, that is equivalent to defining it to do 8617nothing. 8618 8619You may wish to use @code{ASM_OUTPUT_SIZE_DIRECTIVE} and/or 8620@code{ASM_OUTPUT_MEASURED_SIZE} in the definition of this macro. 8621@end defmac 8622 8623@deftypefn {Target Hook} void TARGET_ASM_GLOBALIZE_LABEL (FILE *@var{stream}, const char *@var{name}) 8624This target hook is a function to output to the stdio stream 8625@var{stream} some commands that will make the label @var{name} global; 8626that is, available for reference from other files. 8627 8628The default implementation relies on a proper definition of 8629@code{GLOBAL_ASM_OP}. 8630@end deftypefn 8631 8632@deftypefn {Target Hook} void TARGET_ASM_GLOBALIZE_DECL_NAME (FILE *@var{stream}, tree @var{decl}) 8633This target hook is a function to output to the stdio stream 8634@var{stream} some commands that will make the name associated with @var{decl} 8635global; that is, available for reference from other files. 8636 8637The default implementation uses the TARGET_ASM_GLOBALIZE_LABEL target hook. 8638@end deftypefn 8639 8640@deftypefn {Target Hook} void TARGET_ASM_ASSEMBLE_UNDEFINED_DECL (FILE *@var{stream}, const char *@var{name}, const_tree @var{decl}) 8641This target hook is a function to output to the stdio stream 8642@var{stream} some commands that will declare the name associated with 8643@var{decl} which is not defined in the current translation unit. Most 8644assemblers do not require anything to be output in this case. 8645@end deftypefn 8646 8647@defmac ASM_WEAKEN_LABEL (@var{stream}, @var{name}) 8648A C statement (sans semicolon) to output to the stdio stream 8649@var{stream} some commands that will make the label @var{name} weak; 8650that is, available for reference from other files but only used if 8651no other definition is available. Use the expression 8652@code{assemble_name (@var{stream}, @var{name})} to output the name 8653itself; before and after that, output the additional assembler syntax 8654for making that name weak, and a newline. 8655 8656If you don't define this macro or @code{ASM_WEAKEN_DECL}, GCC will not 8657support weak symbols and you should not define the @code{SUPPORTS_WEAK} 8658macro. 8659@end defmac 8660 8661@defmac ASM_WEAKEN_DECL (@var{stream}, @var{decl}, @var{name}, @var{value}) 8662Combines (and replaces) the function of @code{ASM_WEAKEN_LABEL} and 8663@code{ASM_OUTPUT_WEAK_ALIAS}, allowing access to the associated function 8664or variable decl. If @var{value} is not @code{NULL}, this C statement 8665should output to the stdio stream @var{stream} assembler code which 8666defines (equates) the weak symbol @var{name} to have the value 8667@var{value}. If @var{value} is @code{NULL}, it should output commands 8668to make @var{name} weak. 8669@end defmac 8670 8671@defmac ASM_OUTPUT_WEAKREF (@var{stream}, @var{decl}, @var{name}, @var{value}) 8672Outputs a directive that enables @var{name} to be used to refer to 8673symbol @var{value} with weak-symbol semantics. @code{decl} is the 8674declaration of @code{name}. 8675@end defmac 8676 8677@defmac SUPPORTS_WEAK 8678A preprocessor constant expression which evaluates to true if the target 8679supports weak symbols. 8680 8681If you don't define this macro, @file{defaults.h} provides a default 8682definition. If either @code{ASM_WEAKEN_LABEL} or @code{ASM_WEAKEN_DECL} 8683is defined, the default definition is @samp{1}; otherwise, it is @samp{0}. 8684@end defmac 8685 8686@defmac TARGET_SUPPORTS_WEAK 8687A C expression which evaluates to true if the target supports weak symbols. 8688 8689If you don't define this macro, @file{defaults.h} provides a default 8690definition. The default definition is @samp{(SUPPORTS_WEAK)}. Define 8691this macro if you want to control weak symbol support with a compiler 8692flag such as @option{-melf}. 8693@end defmac 8694 8695@defmac MAKE_DECL_ONE_ONLY (@var{decl}) 8696A C statement (sans semicolon) to mark @var{decl} to be emitted as a 8697public symbol such that extra copies in multiple translation units will 8698be discarded by the linker. Define this macro if your object file 8699format provides support for this concept, such as the @samp{COMDAT} 8700section flags in the Microsoft Windows PE/COFF format, and this support 8701requires changes to @var{decl}, such as putting it in a separate section. 8702@end defmac 8703 8704@defmac SUPPORTS_ONE_ONLY 8705A C expression which evaluates to true if the target supports one-only 8706semantics. 8707 8708If you don't define this macro, @file{varasm.c} provides a default 8709definition. If @code{MAKE_DECL_ONE_ONLY} is defined, the default 8710definition is @samp{1}; otherwise, it is @samp{0}. Define this macro if 8711you want to control one-only symbol support with a compiler flag, or if 8712setting the @code{DECL_ONE_ONLY} flag is enough to mark a declaration to 8713be emitted as one-only. 8714@end defmac 8715 8716@deftypefn {Target Hook} void TARGET_ASM_ASSEMBLE_VISIBILITY (tree @var{decl}, int @var{visibility}) 8717This target hook is a function to output to @var{asm_out_file} some 8718commands that will make the symbol(s) associated with @var{decl} have 8719hidden, protected or internal visibility as specified by @var{visibility}. 8720@end deftypefn 8721 8722@defmac TARGET_WEAK_NOT_IN_ARCHIVE_TOC 8723A C expression that evaluates to true if the target's linker expects 8724that weak symbols do not appear in a static archive's table of contents. 8725The default is @code{0}. 8726 8727Leaving weak symbols out of an archive's table of contents means that, 8728if a symbol will only have a definition in one translation unit and 8729will have undefined references from other translation units, that 8730symbol should not be weak. Defining this macro to be nonzero will 8731thus have the effect that certain symbols that would normally be weak 8732(explicit template instantiations, and vtables for polymorphic classes 8733with noninline key methods) will instead be nonweak. 8734 8735The C++ ABI requires this macro to be zero. Define this macro for 8736targets where full C++ ABI compliance is impossible and where linker 8737restrictions require weak symbols to be left out of a static archive's 8738table of contents. 8739@end defmac 8740 8741@defmac ASM_OUTPUT_EXTERNAL (@var{stream}, @var{decl}, @var{name}) 8742A C statement (sans semicolon) to output to the stdio stream 8743@var{stream} any text necessary for declaring the name of an external 8744symbol named @var{name} which is referenced in this compilation but 8745not defined. The value of @var{decl} is the tree node for the 8746declaration. 8747 8748This macro need not be defined if it does not need to output anything. 8749The GNU assembler and most Unix assemblers don't require anything. 8750@end defmac 8751 8752@deftypefn {Target Hook} void TARGET_ASM_EXTERNAL_LIBCALL (rtx @var{symref}) 8753This target hook is a function to output to @var{asm_out_file} an assembler 8754pseudo-op to declare a library function name external. The name of the 8755library function is given by @var{symref}, which is a @code{symbol_ref}. 8756@end deftypefn 8757 8758@deftypefn {Target Hook} void TARGET_ASM_MARK_DECL_PRESERVED (const char *@var{symbol}) 8759This target hook is a function to output to @var{asm_out_file} an assembler 8760directive to annotate @var{symbol} as used. The Darwin target uses the 8761.no_dead_code_strip directive. 8762@end deftypefn 8763 8764@defmac ASM_OUTPUT_LABELREF (@var{stream}, @var{name}) 8765A C statement (sans semicolon) to output to the stdio stream 8766@var{stream} a reference in assembler syntax to a label named 8767@var{name}. This should add @samp{_} to the front of the name, if that 8768is customary on your operating system, as it is in most Berkeley Unix 8769systems. This macro is used in @code{assemble_name}. 8770@end defmac 8771 8772@deftypefn {Target Hook} tree TARGET_MANGLE_ASSEMBLER_NAME (const char *@var{name}) 8773Given a symbol @var{name}, perform same mangling as @code{varasm.c}'s @code{assemble_name}, but in memory rather than to a file stream, returning result as an @code{IDENTIFIER_NODE}. Required for correct LTO symtabs. The default implementation calls the @code{TARGET_STRIP_NAME_ENCODING} hook and then prepends the @code{USER_LABEL_PREFIX}, if any. 8774@end deftypefn 8775 8776@defmac ASM_OUTPUT_SYMBOL_REF (@var{stream}, @var{sym}) 8777A C statement (sans semicolon) to output a reference to 8778@code{SYMBOL_REF} @var{sym}. If not defined, @code{assemble_name} 8779will be used to output the name of the symbol. This macro may be used 8780to modify the way a symbol is referenced depending on information 8781encoded by @code{TARGET_ENCODE_SECTION_INFO}. 8782@end defmac 8783 8784@defmac ASM_OUTPUT_LABEL_REF (@var{stream}, @var{buf}) 8785A C statement (sans semicolon) to output a reference to @var{buf}, the 8786result of @code{ASM_GENERATE_INTERNAL_LABEL}. If not defined, 8787@code{assemble_name} will be used to output the name of the symbol. 8788This macro is not used by @code{output_asm_label}, or the @code{%l} 8789specifier that calls it; the intention is that this macro should be set 8790when it is necessary to output a label differently when its address is 8791being taken. 8792@end defmac 8793 8794@deftypefn {Target Hook} void TARGET_ASM_INTERNAL_LABEL (FILE *@var{stream}, const char *@var{prefix}, unsigned long @var{labelno}) 8795A function to output to the stdio stream @var{stream} a label whose 8796name is made from the string @var{prefix} and the number @var{labelno}. 8797 8798It is absolutely essential that these labels be distinct from the labels 8799used for user-level functions and variables. Otherwise, certain programs 8800will have name conflicts with internal labels. 8801 8802It is desirable to exclude internal labels from the symbol table of the 8803object file. Most assemblers have a naming convention for labels that 8804should be excluded; on many systems, the letter @samp{L} at the 8805beginning of a label has this effect. You should find out what 8806convention your system uses, and follow it. 8807 8808The default version of this function utilizes @code{ASM_GENERATE_INTERNAL_LABEL}. 8809@end deftypefn 8810 8811@defmac ASM_OUTPUT_DEBUG_LABEL (@var{stream}, @var{prefix}, @var{num}) 8812A C statement to output to the stdio stream @var{stream} a debug info 8813label whose name is made from the string @var{prefix} and the number 8814@var{num}. This is useful for VLIW targets, where debug info labels 8815may need to be treated differently than branch target labels. On some 8816systems, branch target labels must be at the beginning of instruction 8817bundles, but debug info labels can occur in the middle of instruction 8818bundles. 8819 8820If this macro is not defined, then @code{(*targetm.asm_out.internal_label)} will be 8821used. 8822@end defmac 8823 8824@defmac ASM_GENERATE_INTERNAL_LABEL (@var{string}, @var{prefix}, @var{num}) 8825A C statement to store into the string @var{string} a label whose name 8826is made from the string @var{prefix} and the number @var{num}. 8827 8828This string, when output subsequently by @code{assemble_name}, should 8829produce the output that @code{(*targetm.asm_out.internal_label)} would produce 8830with the same @var{prefix} and @var{num}. 8831 8832If the string begins with @samp{*}, then @code{assemble_name} will 8833output the rest of the string unchanged. It is often convenient for 8834@code{ASM_GENERATE_INTERNAL_LABEL} to use @samp{*} in this way. If the 8835string doesn't start with @samp{*}, then @code{ASM_OUTPUT_LABELREF} gets 8836to output the string, and may change it. (Of course, 8837@code{ASM_OUTPUT_LABELREF} is also part of your machine description, so 8838you should know what it does on your machine.) 8839@end defmac 8840 8841@defmac ASM_FORMAT_PRIVATE_NAME (@var{outvar}, @var{name}, @var{number}) 8842A C expression to assign to @var{outvar} (which is a variable of type 8843@code{char *}) a newly allocated string made from the string 8844@var{name} and the number @var{number}, with some suitable punctuation 8845added. Use @code{alloca} to get space for the string. 8846 8847The string will be used as an argument to @code{ASM_OUTPUT_LABELREF} to 8848produce an assembler label for an internal static variable whose name is 8849@var{name}. Therefore, the string must be such as to result in valid 8850assembler code. The argument @var{number} is different each time this 8851macro is executed; it prevents conflicts between similarly-named 8852internal static variables in different scopes. 8853 8854Ideally this string should not be a valid C identifier, to prevent any 8855conflict with the user's own symbols. Most assemblers allow periods 8856or percent signs in assembler symbols; putting at least one of these 8857between the name and the number will suffice. 8858 8859If this macro is not defined, a default definition will be provided 8860which is correct for most systems. 8861@end defmac 8862 8863@defmac ASM_OUTPUT_DEF (@var{stream}, @var{name}, @var{value}) 8864A C statement to output to the stdio stream @var{stream} assembler code 8865which defines (equates) the symbol @var{name} to have the value @var{value}. 8866 8867@findex SET_ASM_OP 8868If @code{SET_ASM_OP} is defined, a default definition is provided which is 8869correct for most systems. 8870@end defmac 8871 8872@defmac ASM_OUTPUT_DEF_FROM_DECLS (@var{stream}, @var{decl_of_name}, @var{decl_of_value}) 8873A C statement to output to the stdio stream @var{stream} assembler code 8874which defines (equates) the symbol whose tree node is @var{decl_of_name} 8875to have the value of the tree node @var{decl_of_value}. This macro will 8876be used in preference to @samp{ASM_OUTPUT_DEF} if it is defined and if 8877the tree nodes are available. 8878 8879@findex SET_ASM_OP 8880If @code{SET_ASM_OP} is defined, a default definition is provided which is 8881correct for most systems. 8882@end defmac 8883 8884@defmac TARGET_DEFERRED_OUTPUT_DEFS (@var{decl_of_name}, @var{decl_of_value}) 8885A C statement that evaluates to true if the assembler code which defines 8886(equates) the symbol whose tree node is @var{decl_of_name} to have the value 8887of the tree node @var{decl_of_value} should be emitted near the end of the 8888current compilation unit. The default is to not defer output of defines. 8889This macro affects defines output by @samp{ASM_OUTPUT_DEF} and 8890@samp{ASM_OUTPUT_DEF_FROM_DECLS}. 8891@end defmac 8892 8893@defmac ASM_OUTPUT_WEAK_ALIAS (@var{stream}, @var{name}, @var{value}) 8894A C statement to output to the stdio stream @var{stream} assembler code 8895which defines (equates) the weak symbol @var{name} to have the value 8896@var{value}. If @var{value} is @code{NULL}, it defines @var{name} as 8897an undefined weak symbol. 8898 8899Define this macro if the target only supports weak aliases; define 8900@code{ASM_OUTPUT_DEF} instead if possible. 8901@end defmac 8902 8903@defmac OBJC_GEN_METHOD_LABEL (@var{buf}, @var{is_inst}, @var{class_name}, @var{cat_name}, @var{sel_name}) 8904Define this macro to override the default assembler names used for 8905Objective-C methods. 8906 8907The default name is a unique method number followed by the name of the 8908class (e.g.@: @samp{_1_Foo}). For methods in categories, the name of 8909the category is also included in the assembler name (e.g.@: 8910@samp{_1_Foo_Bar}). 8911 8912These names are safe on most systems, but make debugging difficult since 8913the method's selector is not present in the name. Therefore, particular 8914systems define other ways of computing names. 8915 8916@var{buf} is an expression of type @code{char *} which gives you a 8917buffer in which to store the name; its length is as long as 8918@var{class_name}, @var{cat_name} and @var{sel_name} put together, plus 891950 characters extra. 8920 8921The argument @var{is_inst} specifies whether the method is an instance 8922method or a class method; @var{class_name} is the name of the class; 8923@var{cat_name} is the name of the category (or @code{NULL} if the method is not 8924in a category); and @var{sel_name} is the name of the selector. 8925 8926On systems where the assembler can handle quoted names, you can use this 8927macro to provide more human-readable names. 8928@end defmac 8929 8930@node Initialization 8931@subsection How Initialization Functions Are Handled 8932@cindex initialization routines 8933@cindex termination routines 8934@cindex constructors, output of 8935@cindex destructors, output of 8936 8937The compiled code for certain languages includes @dfn{constructors} 8938(also called @dfn{initialization routines})---functions to initialize 8939data in the program when the program is started. These functions need 8940to be called before the program is ``started''---that is to say, before 8941@code{main} is called. 8942 8943Compiling some languages generates @dfn{destructors} (also called 8944@dfn{termination routines}) that should be called when the program 8945terminates. 8946 8947To make the initialization and termination functions work, the compiler 8948must output something in the assembler code to cause those functions to 8949be called at the appropriate time. When you port the compiler to a new 8950system, you need to specify how to do this. 8951 8952There are two major ways that GCC currently supports the execution of 8953initialization and termination functions. Each way has two variants. 8954Much of the structure is common to all four variations. 8955 8956@findex __CTOR_LIST__ 8957@findex __DTOR_LIST__ 8958The linker must build two lists of these functions---a list of 8959initialization functions, called @code{__CTOR_LIST__}, and a list of 8960termination functions, called @code{__DTOR_LIST__}. 8961 8962Each list always begins with an ignored function pointer (which may hold 89630, @minus{}1, or a count of the function pointers after it, depending on 8964the environment). This is followed by a series of zero or more function 8965pointers to constructors (or destructors), followed by a function 8966pointer containing zero. 8967 8968Depending on the operating system and its executable file format, either 8969@file{crtstuff.c} or @file{libgcc2.c} traverses these lists at startup 8970time and exit time. Constructors are called in reverse order of the 8971list; destructors in forward order. 8972 8973The best way to handle static constructors works only for object file 8974formats which provide arbitrarily-named sections. A section is set 8975aside for a list of constructors, and another for a list of destructors. 8976Traditionally these are called @samp{.ctors} and @samp{.dtors}. Each 8977object file that defines an initialization function also puts a word in 8978the constructor section to point to that function. The linker 8979accumulates all these words into one contiguous @samp{.ctors} section. 8980Termination functions are handled similarly. 8981 8982This method will be chosen as the default by @file{target-def.h} if 8983@code{TARGET_ASM_NAMED_SECTION} is defined. A target that does not 8984support arbitrary sections, but does support special designated 8985constructor and destructor sections may define @code{CTORS_SECTION_ASM_OP} 8986and @code{DTORS_SECTION_ASM_OP} to achieve the same effect. 8987 8988When arbitrary sections are available, there are two variants, depending 8989upon how the code in @file{crtstuff.c} is called. On systems that 8990support a @dfn{.init} section which is executed at program startup, 8991parts of @file{crtstuff.c} are compiled into that section. The 8992program is linked by the @command{gcc} driver like this: 8993 8994@smallexample 8995ld -o @var{output_file} crti.o crtbegin.o @dots{} -lgcc crtend.o crtn.o 8996@end smallexample 8997 8998The prologue of a function (@code{__init}) appears in the @code{.init} 8999section of @file{crti.o}; the epilogue appears in @file{crtn.o}. Likewise 9000for the function @code{__fini} in the @dfn{.fini} section. Normally these 9001files are provided by the operating system or by the GNU C library, but 9002are provided by GCC for a few targets. 9003 9004The objects @file{crtbegin.o} and @file{crtend.o} are (for most targets) 9005compiled from @file{crtstuff.c}. They contain, among other things, code 9006fragments within the @code{.init} and @code{.fini} sections that branch 9007to routines in the @code{.text} section. The linker will pull all parts 9008of a section together, which results in a complete @code{__init} function 9009that invokes the routines we need at startup. 9010 9011To use this variant, you must define the @code{INIT_SECTION_ASM_OP} 9012macro properly. 9013 9014If no init section is available, when GCC compiles any function called 9015@code{main} (or more accurately, any function designated as a program 9016entry point by the language front end calling @code{expand_main_function}), 9017it inserts a procedure call to @code{__main} as the first executable code 9018after the function prologue. The @code{__main} function is defined 9019in @file{libgcc2.c} and runs the global constructors. 9020 9021In file formats that don't support arbitrary sections, there are again 9022two variants. In the simplest variant, the GNU linker (GNU @code{ld}) 9023and an `a.out' format must be used. In this case, 9024@code{TARGET_ASM_CONSTRUCTOR} is defined to produce a @code{.stabs} 9025entry of type @samp{N_SETT}, referencing the name @code{__CTOR_LIST__}, 9026and with the address of the void function containing the initialization 9027code as its value. The GNU linker recognizes this as a request to add 9028the value to a @dfn{set}; the values are accumulated, and are eventually 9029placed in the executable as a vector in the format described above, with 9030a leading (ignored) count and a trailing zero element. 9031@code{TARGET_ASM_DESTRUCTOR} is handled similarly. Since no init 9032section is available, the absence of @code{INIT_SECTION_ASM_OP} causes 9033the compilation of @code{main} to call @code{__main} as above, starting 9034the initialization process. 9035 9036The last variant uses neither arbitrary sections nor the GNU linker. 9037This is preferable when you want to do dynamic linking and when using 9038file formats which the GNU linker does not support, such as `ECOFF'@. In 9039this case, @code{TARGET_HAVE_CTORS_DTORS} is false, initialization and 9040termination functions are recognized simply by their names. This requires 9041an extra program in the linkage step, called @command{collect2}. This program 9042pretends to be the linker, for use with GCC; it does its job by running 9043the ordinary linker, but also arranges to include the vectors of 9044initialization and termination functions. These functions are called 9045via @code{__main} as described above. In order to use this method, 9046@code{use_collect2} must be defined in the target in @file{config.gcc}. 9047 9048@ifinfo 9049The following section describes the specific macros that control and 9050customize the handling of initialization and termination functions. 9051@end ifinfo 9052 9053@node Macros for Initialization 9054@subsection Macros Controlling Initialization Routines 9055 9056Here are the macros that control how the compiler handles initialization 9057and termination functions: 9058 9059@defmac INIT_SECTION_ASM_OP 9060If defined, a C string constant, including spacing, for the assembler 9061operation to identify the following data as initialization code. If not 9062defined, GCC will assume such a section does not exist. When you are 9063using special sections for initialization and termination functions, this 9064macro also controls how @file{crtstuff.c} and @file{libgcc2.c} arrange to 9065run the initialization functions. 9066@end defmac 9067 9068@defmac HAS_INIT_SECTION 9069If defined, @code{main} will not call @code{__main} as described above. 9070This macro should be defined for systems that control start-up code 9071on a symbol-by-symbol basis, such as OSF/1, and should not 9072be defined explicitly for systems that support @code{INIT_SECTION_ASM_OP}. 9073@end defmac 9074 9075@defmac LD_INIT_SWITCH 9076If defined, a C string constant for a switch that tells the linker that 9077the following symbol is an initialization routine. 9078@end defmac 9079 9080@defmac LD_FINI_SWITCH 9081If defined, a C string constant for a switch that tells the linker that 9082the following symbol is a finalization routine. 9083@end defmac 9084 9085@defmac COLLECT_SHARED_INIT_FUNC (@var{stream}, @var{func}) 9086If defined, a C statement that will write a function that can be 9087automatically called when a shared library is loaded. The function 9088should call @var{func}, which takes no arguments. If not defined, and 9089the object format requires an explicit initialization function, then a 9090function called @code{_GLOBAL__DI} will be generated. 9091 9092This function and the following one are used by collect2 when linking a 9093shared library that needs constructors or destructors, or has DWARF2 9094exception tables embedded in the code. 9095@end defmac 9096 9097@defmac COLLECT_SHARED_FINI_FUNC (@var{stream}, @var{func}) 9098If defined, a C statement that will write a function that can be 9099automatically called when a shared library is unloaded. The function 9100should call @var{func}, which takes no arguments. If not defined, and 9101the object format requires an explicit finalization function, then a 9102function called @code{_GLOBAL__DD} will be generated. 9103@end defmac 9104 9105@defmac INVOKE__main 9106If defined, @code{main} will call @code{__main} despite the presence of 9107@code{INIT_SECTION_ASM_OP}. This macro should be defined for systems 9108where the init section is not actually run automatically, but is still 9109useful for collecting the lists of constructors and destructors. 9110@end defmac 9111 9112@defmac SUPPORTS_INIT_PRIORITY 9113If nonzero, the C++ @code{init_priority} attribute is supported and the 9114compiler should emit instructions to control the order of initialization 9115of objects. If zero, the compiler will issue an error message upon 9116encountering an @code{init_priority} attribute. 9117@end defmac 9118 9119@deftypevr {Target Hook} bool TARGET_HAVE_CTORS_DTORS 9120This value is true if the target supports some ``native'' method of 9121collecting constructors and destructors to be run at startup and exit. 9122It is false if we must use @command{collect2}. 9123@end deftypevr 9124 9125@deftypevr {Target Hook} bool TARGET_DTORS_FROM_CXA_ATEXIT 9126This value is true if the target wants destructors to be queued to be 9127run from __cxa_atexit. If this is the case then, for each priority level, 9128a new constructor will be entered that registers the destructors for that 9129level with __cxa_atexit (and there will be no destructors emitted). 9130It is false the method implied by @code{have_ctors_dtors} is used. 9131@end deftypevr 9132 9133@deftypefn {Target Hook} void TARGET_ASM_CONSTRUCTOR (rtx @var{symbol}, int @var{priority}) 9134If defined, a function that outputs assembler code to arrange to call 9135the function referenced by @var{symbol} at initialization time. 9136 9137Assume that @var{symbol} is a @code{SYMBOL_REF} for a function taking 9138no arguments and with no return value. If the target supports initialization 9139priorities, @var{priority} is a value between 0 and @code{MAX_INIT_PRIORITY}; 9140otherwise it must be @code{DEFAULT_INIT_PRIORITY}. 9141 9142If this macro is not defined by the target, a suitable default will 9143be chosen if (1) the target supports arbitrary section names, (2) the 9144target defines @code{CTORS_SECTION_ASM_OP}, or (3) @code{USE_COLLECT2} 9145is not defined. 9146@end deftypefn 9147 9148@deftypefn {Target Hook} void TARGET_ASM_DESTRUCTOR (rtx @var{symbol}, int @var{priority}) 9149This is like @code{TARGET_ASM_CONSTRUCTOR} but used for termination 9150functions rather than initialization functions. 9151@end deftypefn 9152 9153If @code{TARGET_HAVE_CTORS_DTORS} is true, the initialization routine 9154generated for the generated object file will have static linkage. 9155 9156If your system uses @command{collect2} as the means of processing 9157constructors, then that program normally uses @command{nm} to scan 9158an object file for constructor functions to be called. 9159 9160On certain kinds of systems, you can define this macro to make 9161@command{collect2} work faster (and, in some cases, make it work at all): 9162 9163@defmac OBJECT_FORMAT_COFF 9164Define this macro if the system uses COFF (Common Object File Format) 9165object files, so that @command{collect2} can assume this format and scan 9166object files directly for dynamic constructor/destructor functions. 9167 9168This macro is effective only in a native compiler; @command{collect2} as 9169part of a cross compiler always uses @command{nm} for the target machine. 9170@end defmac 9171 9172@defmac REAL_NM_FILE_NAME 9173Define this macro as a C string constant containing the file name to use 9174to execute @command{nm}. The default is to search the path normally for 9175@command{nm}. 9176@end defmac 9177 9178@defmac NM_FLAGS 9179@command{collect2} calls @command{nm} to scan object files for static 9180constructors and destructors and LTO info. By default, @option{-n} is 9181passed. Define @code{NM_FLAGS} to a C string constant if other options 9182are needed to get the same output format as GNU @command{nm -n} 9183produces. 9184@end defmac 9185 9186If your system supports shared libraries and has a program to list the 9187dynamic dependencies of a given library or executable, you can define 9188these macros to enable support for running initialization and 9189termination functions in shared libraries: 9190 9191@defmac LDD_SUFFIX 9192Define this macro to a C string constant containing the name of the program 9193which lists dynamic dependencies, like @command{ldd} under SunOS 4. 9194@end defmac 9195 9196@defmac PARSE_LDD_OUTPUT (@var{ptr}) 9197Define this macro to be C code that extracts filenames from the output 9198of the program denoted by @code{LDD_SUFFIX}. @var{ptr} is a variable 9199of type @code{char *} that points to the beginning of a line of output 9200from @code{LDD_SUFFIX}. If the line lists a dynamic dependency, the 9201code must advance @var{ptr} to the beginning of the filename on that 9202line. Otherwise, it must set @var{ptr} to @code{NULL}. 9203@end defmac 9204 9205@defmac SHLIB_SUFFIX 9206Define this macro to a C string constant containing the default shared 9207library extension of the target (e.g., @samp{".so"}). @command{collect2} 9208strips version information after this suffix when generating global 9209constructor and destructor names. This define is only needed on targets 9210that use @command{collect2} to process constructors and destructors. 9211@end defmac 9212 9213@node Instruction Output 9214@subsection Output of Assembler Instructions 9215 9216@c prevent bad page break with this line 9217This describes assembler instruction output. 9218 9219@defmac REGISTER_NAMES 9220A C initializer containing the assembler's names for the machine 9221registers, each one as a C string constant. This is what translates 9222register numbers in the compiler into assembler language. 9223@end defmac 9224 9225@defmac ADDITIONAL_REGISTER_NAMES 9226If defined, a C initializer for an array of structures containing a name 9227and a register number. This macro defines additional names for hard 9228registers, thus allowing the @code{asm} option in declarations to refer 9229to registers using alternate names. 9230@end defmac 9231 9232@defmac OVERLAPPING_REGISTER_NAMES 9233If defined, a C initializer for an array of structures containing a 9234name, a register number and a count of the number of consecutive 9235machine registers the name overlaps. This macro defines additional 9236names for hard registers, thus allowing the @code{asm} option in 9237declarations to refer to registers using alternate names. Unlike 9238@code{ADDITIONAL_REGISTER_NAMES}, this macro should be used when the 9239register name implies multiple underlying registers. 9240 9241This macro should be used when it is important that a clobber in an 9242@code{asm} statement clobbers all the underlying values implied by the 9243register name. For example, on ARM, clobbering the double-precision 9244VFP register ``d0'' implies clobbering both single-precision registers 9245``s0'' and ``s1''. 9246@end defmac 9247 9248@defmac ASM_OUTPUT_OPCODE (@var{stream}, @var{ptr}) 9249Define this macro if you are using an unusual assembler that 9250requires different names for the machine instructions. 9251 9252The definition is a C statement or statements which output an 9253assembler instruction opcode to the stdio stream @var{stream}. The 9254macro-operand @var{ptr} is a variable of type @code{char *} which 9255points to the opcode name in its ``internal'' form---the form that is 9256written in the machine description. The definition should output the 9257opcode name to @var{stream}, performing any translation you desire, and 9258increment the variable @var{ptr} to point at the end of the opcode 9259so that it will not be output twice. 9260 9261In fact, your macro definition may process less than the entire opcode 9262name, or more than the opcode name; but if you want to process text 9263that includes @samp{%}-sequences to substitute operands, you must take 9264care of the substitution yourself. Just be sure to increment 9265@var{ptr} over whatever text should not be output normally. 9266 9267@findex recog_data.operand 9268If you need to look at the operand values, they can be found as the 9269elements of @code{recog_data.operand}. 9270 9271If the macro definition does nothing, the instruction is output 9272in the usual way. 9273@end defmac 9274 9275@defmac FINAL_PRESCAN_INSN (@var{insn}, @var{opvec}, @var{noperands}) 9276If defined, a C statement to be executed just prior to the output of 9277assembler code for @var{insn}, to modify the extracted operands so 9278they will be output differently. 9279 9280Here the argument @var{opvec} is the vector containing the operands 9281extracted from @var{insn}, and @var{noperands} is the number of 9282elements of the vector which contain meaningful data for this insn. 9283The contents of this vector are what will be used to convert the insn 9284template into assembler code, so you can change the assembler output 9285by changing the contents of the vector. 9286 9287This macro is useful when various assembler syntaxes share a single 9288file of instruction patterns; by defining this macro differently, you 9289can cause a large class of instructions to be output differently (such 9290as with rearranged operands). Naturally, variations in assembler 9291syntax affecting individual insn patterns ought to be handled by 9292writing conditional output routines in those patterns. 9293 9294If this macro is not defined, it is equivalent to a null statement. 9295@end defmac 9296 9297@deftypefn {Target Hook} void TARGET_ASM_FINAL_POSTSCAN_INSN (FILE *@var{file}, rtx_insn *@var{insn}, rtx *@var{opvec}, int @var{noperands}) 9298If defined, this target hook is a function which is executed just after the 9299output of assembler code for @var{insn}, to change the mode of the assembler 9300if necessary. 9301 9302Here the argument @var{opvec} is the vector containing the operands 9303extracted from @var{insn}, and @var{noperands} is the number of 9304elements of the vector which contain meaningful data for this insn. 9305The contents of this vector are what was used to convert the insn 9306template into assembler code, so you can change the assembler mode 9307by checking the contents of the vector. 9308@end deftypefn 9309 9310@defmac PRINT_OPERAND (@var{stream}, @var{x}, @var{code}) 9311A C compound statement to output to stdio stream @var{stream} the 9312assembler syntax for an instruction operand @var{x}. @var{x} is an 9313RTL expression. 9314 9315@var{code} is a value that can be used to specify one of several ways 9316of printing the operand. It is used when identical operands must be 9317printed differently depending on the context. @var{code} comes from 9318the @samp{%} specification that was used to request printing of the 9319operand. If the specification was just @samp{%@var{digit}} then 9320@var{code} is 0; if the specification was @samp{%@var{ltr} 9321@var{digit}} then @var{code} is the ASCII code for @var{ltr}. 9322 9323@findex reg_names 9324If @var{x} is a register, this macro should print the register's name. 9325The names can be found in an array @code{reg_names} whose type is 9326@code{char *[]}. @code{reg_names} is initialized from 9327@code{REGISTER_NAMES}. 9328 9329When the machine description has a specification @samp{%@var{punct}} 9330(a @samp{%} followed by a punctuation character), this macro is called 9331with a null pointer for @var{x} and the punctuation character for 9332@var{code}. 9333@end defmac 9334 9335@defmac PRINT_OPERAND_PUNCT_VALID_P (@var{code}) 9336A C expression which evaluates to true if @var{code} is a valid 9337punctuation character for use in the @code{PRINT_OPERAND} macro. If 9338@code{PRINT_OPERAND_PUNCT_VALID_P} is not defined, it means that no 9339punctuation characters (except for the standard one, @samp{%}) are used 9340in this way. 9341@end defmac 9342 9343@defmac PRINT_OPERAND_ADDRESS (@var{stream}, @var{x}) 9344A C compound statement to output to stdio stream @var{stream} the 9345assembler syntax for an instruction operand that is a memory reference 9346whose address is @var{x}. @var{x} is an RTL expression. 9347 9348@cindex @code{TARGET_ENCODE_SECTION_INFO} usage 9349On some machines, the syntax for a symbolic address depends on the 9350section that the address refers to. On these machines, define the hook 9351@code{TARGET_ENCODE_SECTION_INFO} to store the information into the 9352@code{symbol_ref}, and then check for it here. @xref{Assembler 9353Format}. 9354@end defmac 9355 9356@findex dbr_sequence_length 9357@defmac DBR_OUTPUT_SEQEND (@var{file}) 9358A C statement, to be executed after all slot-filler instructions have 9359been output. If necessary, call @code{dbr_sequence_length} to 9360determine the number of slots filled in a sequence (zero if not 9361currently outputting a sequence), to decide how many no-ops to output, 9362or whatever. 9363 9364Don't define this macro if it has nothing to do, but it is helpful in 9365reading assembly output if the extent of the delay sequence is made 9366explicit (e.g.@: with white space). 9367@end defmac 9368 9369@findex final_sequence 9370Note that output routines for instructions with delay slots must be 9371prepared to deal with not being output as part of a sequence 9372(i.e.@: when the scheduling pass is not run, or when no slot fillers could be 9373found.) The variable @code{final_sequence} is null when not 9374processing a sequence, otherwise it contains the @code{sequence} rtx 9375being output. 9376 9377@findex asm_fprintf 9378@defmac REGISTER_PREFIX 9379@defmacx LOCAL_LABEL_PREFIX 9380@defmacx USER_LABEL_PREFIX 9381@defmacx IMMEDIATE_PREFIX 9382If defined, C string expressions to be used for the @samp{%R}, @samp{%L}, 9383@samp{%U}, and @samp{%I} options of @code{asm_fprintf} (see 9384@file{final.c}). These are useful when a single @file{md} file must 9385support multiple assembler formats. In that case, the various @file{tm.h} 9386files can define these macros differently. 9387@end defmac 9388 9389@defmac ASM_FPRINTF_EXTENSIONS (@var{file}, @var{argptr}, @var{format}) 9390If defined this macro should expand to a series of @code{case} 9391statements which will be parsed inside the @code{switch} statement of 9392the @code{asm_fprintf} function. This allows targets to define extra 9393printf formats which may useful when generating their assembler 9394statements. Note that uppercase letters are reserved for future 9395generic extensions to asm_fprintf, and so are not available to target 9396specific code. The output file is given by the parameter @var{file}. 9397The varargs input pointer is @var{argptr} and the rest of the format 9398string, starting the character after the one that is being switched 9399upon, is pointed to by @var{format}. 9400@end defmac 9401 9402@defmac ASSEMBLER_DIALECT 9403If your target supports multiple dialects of assembler language (such as 9404different opcodes), define this macro as a C expression that gives the 9405numeric index of the assembler language dialect to use, with zero as the 9406first variant. 9407 9408If this macro is defined, you may use constructs of the form 9409@smallexample 9410@samp{@{option0|option1|option2@dots{}@}} 9411@end smallexample 9412@noindent 9413in the output templates of patterns (@pxref{Output Template}) or in the 9414first argument of @code{asm_fprintf}. This construct outputs 9415@samp{option0}, @samp{option1}, @samp{option2}, etc., if the value of 9416@code{ASSEMBLER_DIALECT} is zero, one, two, etc. Any special characters 9417within these strings retain their usual meaning. If there are fewer 9418alternatives within the braces than the value of 9419@code{ASSEMBLER_DIALECT}, the construct outputs nothing. If it's needed 9420to print curly braces or @samp{|} character in assembler output directly, 9421@samp{%@{}, @samp{%@}} and @samp{%|} can be used. 9422 9423If you do not define this macro, the characters @samp{@{}, @samp{|} and 9424@samp{@}} do not have any special meaning when used in templates or 9425operands to @code{asm_fprintf}. 9426 9427Define the macros @code{REGISTER_PREFIX}, @code{LOCAL_LABEL_PREFIX}, 9428@code{USER_LABEL_PREFIX} and @code{IMMEDIATE_PREFIX} if you can express 9429the variations in assembler language syntax with that mechanism. Define 9430@code{ASSEMBLER_DIALECT} and use the @samp{@{option0|option1@}} syntax 9431if the syntax variant are larger and involve such things as different 9432opcodes or operand order. 9433@end defmac 9434 9435@defmac ASM_OUTPUT_REG_PUSH (@var{stream}, @var{regno}) 9436A C expression to output to @var{stream} some assembler code 9437which will push hard register number @var{regno} onto the stack. 9438The code need not be optimal, since this macro is used only when 9439profiling. 9440@end defmac 9441 9442@defmac ASM_OUTPUT_REG_POP (@var{stream}, @var{regno}) 9443A C expression to output to @var{stream} some assembler code 9444which will pop hard register number @var{regno} off of the stack. 9445The code need not be optimal, since this macro is used only when 9446profiling. 9447@end defmac 9448 9449@node Dispatch Tables 9450@subsection Output of Dispatch Tables 9451 9452@c prevent bad page break with this line 9453This concerns dispatch tables. 9454 9455@cindex dispatch table 9456@defmac ASM_OUTPUT_ADDR_DIFF_ELT (@var{stream}, @var{body}, @var{value}, @var{rel}) 9457A C statement to output to the stdio stream @var{stream} an assembler 9458pseudo-instruction to generate a difference between two labels. 9459@var{value} and @var{rel} are the numbers of two internal labels. The 9460definitions of these labels are output using 9461@code{(*targetm.asm_out.internal_label)}, and they must be printed in the same 9462way here. For example, 9463 9464@smallexample 9465fprintf (@var{stream}, "\t.word L%d-L%d\n", 9466 @var{value}, @var{rel}) 9467@end smallexample 9468 9469You must provide this macro on machines where the addresses in a 9470dispatch table are relative to the table's own address. If defined, GCC 9471will also use this macro on all machines when producing PIC@. 9472@var{body} is the body of the @code{ADDR_DIFF_VEC}; it is provided so that the 9473mode and flags can be read. 9474@end defmac 9475 9476@defmac ASM_OUTPUT_ADDR_VEC_ELT (@var{stream}, @var{value}) 9477This macro should be provided on machines where the addresses 9478in a dispatch table are absolute. 9479 9480The definition should be a C statement to output to the stdio stream 9481@var{stream} an assembler pseudo-instruction to generate a reference to 9482a label. @var{value} is the number of an internal label whose 9483definition is output using @code{(*targetm.asm_out.internal_label)}. 9484For example, 9485 9486@smallexample 9487fprintf (@var{stream}, "\t.word L%d\n", @var{value}) 9488@end smallexample 9489@end defmac 9490 9491@defmac ASM_OUTPUT_CASE_LABEL (@var{stream}, @var{prefix}, @var{num}, @var{table}) 9492Define this if the label before a jump-table needs to be output 9493specially. The first three arguments are the same as for 9494@code{(*targetm.asm_out.internal_label)}; the fourth argument is the 9495jump-table which follows (a @code{jump_table_data} containing an 9496@code{addr_vec} or @code{addr_diff_vec}). 9497 9498This feature is used on system V to output a @code{swbeg} statement 9499for the table. 9500 9501If this macro is not defined, these labels are output with 9502@code{(*targetm.asm_out.internal_label)}. 9503@end defmac 9504 9505@defmac ASM_OUTPUT_CASE_END (@var{stream}, @var{num}, @var{table}) 9506Define this if something special must be output at the end of a 9507jump-table. The definition should be a C statement to be executed 9508after the assembler code for the table is written. It should write 9509the appropriate code to stdio stream @var{stream}. The argument 9510@var{table} is the jump-table insn, and @var{num} is the label-number 9511of the preceding label. 9512 9513If this macro is not defined, nothing special is output at the end of 9514the jump-table. 9515@end defmac 9516 9517@deftypefn {Target Hook} void TARGET_ASM_POST_CFI_STARTPROC (FILE *@var{}, @var{tree}) 9518This target hook is used to emit assembly strings required by the target 9519after the .cfi_startproc directive. The first argument is the file stream to 9520write the strings to and the second argument is the function's declaration. The 9521expected use is to add more .cfi_* directives. 9522 9523The default is to not output any assembly strings. 9524@end deftypefn 9525 9526@deftypefn {Target Hook} void TARGET_ASM_EMIT_UNWIND_LABEL (FILE *@var{stream}, tree @var{decl}, int @var{for_eh}, int @var{empty}) 9527This target hook emits a label at the beginning of each FDE@. It 9528should be defined on targets where FDEs need special labels, and it 9529should write the appropriate label, for the FDE associated with the 9530function declaration @var{decl}, to the stdio stream @var{stream}. 9531The third argument, @var{for_eh}, is a boolean: true if this is for an 9532exception table. The fourth argument, @var{empty}, is a boolean: 9533true if this is a placeholder label for an omitted FDE@. 9534 9535The default is that FDEs are not given nonlocal labels. 9536@end deftypefn 9537 9538@deftypefn {Target Hook} void TARGET_ASM_EMIT_EXCEPT_TABLE_LABEL (FILE *@var{stream}) 9539This target hook emits a label at the beginning of the exception table. 9540It should be defined on targets where it is desirable for the table 9541to be broken up according to function. 9542 9543The default is that no label is emitted. 9544@end deftypefn 9545 9546@deftypefn {Target Hook} void TARGET_ASM_EMIT_EXCEPT_PERSONALITY (rtx @var{personality}) 9547If the target implements @code{TARGET_ASM_UNWIND_EMIT}, this hook may be used to emit a directive to install a personality hook into the unwind info. This hook should not be used if dwarf2 unwind info is used. 9548@end deftypefn 9549 9550@deftypefn {Target Hook} void TARGET_ASM_UNWIND_EMIT (FILE *@var{stream}, rtx_insn *@var{insn}) 9551This target hook emits assembly directives required to unwind the 9552given instruction. This is only used when @code{TARGET_EXCEPT_UNWIND_INFO} 9553returns @code{UI_TARGET}. 9554@end deftypefn 9555 9556@deftypevr {Target Hook} bool TARGET_ASM_UNWIND_EMIT_BEFORE_INSN 9557True if the @code{TARGET_ASM_UNWIND_EMIT} hook should be called before the assembly for @var{insn} has been emitted, false if the hook should be called afterward. 9558@end deftypevr 9559 9560@deftypefn {Target Hook} bool TARGET_ASM_SHOULD_RESTORE_CFA_STATE (void) 9561For DWARF-based unwind frames, two CFI instructions provide for save and restore of register state. GCC maintains the current frame address (CFA) separately from the register bank but the unwinder in libgcc preserves this state along with the registers (and this is expected by the code that writes the unwind frames). This hook allows the target to specify that the CFA data is not saved/restored along with the registers by the target unwinder so that suitable additional instructions should be emitted to restore it. 9562@end deftypefn 9563 9564@node Exception Region Output 9565@subsection Assembler Commands for Exception Regions 9566 9567@c prevent bad page break with this line 9568 9569This describes commands marking the start and the end of an exception 9570region. 9571 9572@defmac EH_FRAME_SECTION_NAME 9573If defined, a C string constant for the name of the section containing 9574exception handling frame unwind information. If not defined, GCC will 9575provide a default definition if the target supports named sections. 9576@file{crtstuff.c} uses this macro to switch to the appropriate section. 9577 9578You should define this symbol if your target supports DWARF 2 frame 9579unwind information and the default definition does not work. 9580@end defmac 9581 9582@defmac EH_FRAME_THROUGH_COLLECT2 9583If defined, DWARF 2 frame unwind information will identified by 9584specially named labels. The collect2 process will locate these 9585labels and generate code to register the frames. 9586 9587This might be necessary, for instance, if the system linker will not 9588place the eh_frames in-between the sentinals from @file{crtstuff.c}, 9589or if the system linker does garbage collection and sections cannot 9590be marked as not to be collected. 9591@end defmac 9592 9593@defmac EH_TABLES_CAN_BE_READ_ONLY 9594Define this macro to 1 if your target is such that no frame unwind 9595information encoding used with non-PIC code will ever require a 9596runtime relocation, but the linker may not support merging read-only 9597and read-write sections into a single read-write section. 9598@end defmac 9599 9600@defmac MASK_RETURN_ADDR 9601An rtx used to mask the return address found via @code{RETURN_ADDR_RTX}, so 9602that it does not contain any extraneous set bits in it. 9603@end defmac 9604 9605@defmac DWARF2_UNWIND_INFO 9606Define this macro to 0 if your target supports DWARF 2 frame unwind 9607information, but it does not yet work with exception handling. 9608Otherwise, if your target supports this information (if it defines 9609@code{INCOMING_RETURN_ADDR_RTX} and @code{OBJECT_FORMAT_ELF}), 9610GCC will provide a default definition of 1. 9611@end defmac 9612 9613@deftypefn {Common Target Hook} {enum unwind_info_type} TARGET_EXCEPT_UNWIND_INFO (struct gcc_options *@var{opts}) 9614This hook defines the mechanism that will be used for exception handling 9615by the target. If the target has ABI specified unwind tables, the hook 9616should return @code{UI_TARGET}. If the target is to use the 9617@code{setjmp}/@code{longjmp}-based exception handling scheme, the hook 9618should return @code{UI_SJLJ}. If the target supports DWARF 2 frame unwind 9619information, the hook should return @code{UI_DWARF2}. 9620 9621A target may, if exceptions are disabled, choose to return @code{UI_NONE}. 9622This may end up simplifying other parts of target-specific code. The 9623default implementation of this hook never returns @code{UI_NONE}. 9624 9625Note that the value returned by this hook should be constant. It should 9626not depend on anything except the command-line switches described by 9627@var{opts}. In particular, the 9628setting @code{UI_SJLJ} must be fixed at compiler start-up as C pre-processor 9629macros and builtin functions related to exception handling are set up 9630depending on this setting. 9631 9632The default implementation of the hook first honors the 9633@option{--enable-sjlj-exceptions} configure option, then 9634@code{DWARF2_UNWIND_INFO}, and finally defaults to @code{UI_SJLJ}. If 9635@code{DWARF2_UNWIND_INFO} depends on command-line options, the target 9636must define this hook so that @var{opts} is used correctly. 9637@end deftypefn 9638 9639@deftypevr {Common Target Hook} bool TARGET_UNWIND_TABLES_DEFAULT 9640This variable should be set to @code{true} if the target ABI requires unwinding 9641tables even when exceptions are not used. It must not be modified by 9642command-line option processing. 9643@end deftypevr 9644 9645@defmac DONT_USE_BUILTIN_SETJMP 9646Define this macro to 1 if the @code{setjmp}/@code{longjmp}-based scheme 9647should use the @code{setjmp}/@code{longjmp} functions from the C library 9648instead of the @code{__builtin_setjmp}/@code{__builtin_longjmp} machinery. 9649@end defmac 9650 9651@defmac JMP_BUF_SIZE 9652This macro has no effect unless @code{DONT_USE_BUILTIN_SETJMP} is also 9653defined. Define this macro if the default size of @code{jmp_buf} buffer 9654for the @code{setjmp}/@code{longjmp}-based exception handling mechanism 9655is not large enough, or if it is much too large. 9656The default size is @code{FIRST_PSEUDO_REGISTER * sizeof(void *)}. 9657@end defmac 9658 9659@defmac DWARF_CIE_DATA_ALIGNMENT 9660This macro need only be defined if the target might save registers in the 9661function prologue at an offset to the stack pointer that is not aligned to 9662@code{UNITS_PER_WORD}. The definition should be the negative minimum 9663alignment if @code{STACK_GROWS_DOWNWARD} is true, and the positive 9664minimum alignment otherwise. @xref{DWARF}. Only applicable if 9665the target supports DWARF 2 frame unwind information. 9666@end defmac 9667 9668@deftypevr {Target Hook} bool TARGET_TERMINATE_DW2_EH_FRAME_INFO 9669Contains the value true if the target should add a zero word onto the 9670end of a Dwarf-2 frame info section when used for exception handling. 9671Default value is false if @code{EH_FRAME_SECTION_NAME} is defined, and 9672true otherwise. 9673@end deftypevr 9674 9675@deftypefn {Target Hook} rtx TARGET_DWARF_REGISTER_SPAN (rtx @var{reg}) 9676Given a register, this hook should return a parallel of registers to 9677represent where to find the register pieces. Define this hook if the 9678register and its mode are represented in Dwarf in non-contiguous 9679locations, or if the register should be represented in more than one 9680register in Dwarf. Otherwise, this hook should return @code{NULL_RTX}. 9681If not defined, the default is to return @code{NULL_RTX}. 9682@end deftypefn 9683 9684@deftypefn {Target Hook} machine_mode TARGET_DWARF_FRAME_REG_MODE (int @var{regno}) 9685Given a register, this hook should return the mode which the 9686corresponding Dwarf frame register should have. This is normally 9687used to return a smaller mode than the raw mode to prevent call 9688clobbered parts of a register altering the frame register size 9689@end deftypefn 9690 9691@deftypefn {Target Hook} void TARGET_INIT_DWARF_REG_SIZES_EXTRA (tree @var{address}) 9692If some registers are represented in Dwarf-2 unwind information in 9693multiple pieces, define this hook to fill in information about the 9694sizes of those pieces in the table used by the unwinder at runtime. 9695It will be called by @code{expand_builtin_init_dwarf_reg_sizes} after 9696filling in a single size corresponding to each hard register; 9697@var{address} is the address of the table. 9698@end deftypefn 9699 9700@deftypefn {Target Hook} bool TARGET_ASM_TTYPE (rtx @var{sym}) 9701This hook is used to output a reference from a frame unwinding table to 9702the type_info object identified by @var{sym}. It should return @code{true} 9703if the reference was output. Returning @code{false} will cause the 9704reference to be output using the normal Dwarf2 routines. 9705@end deftypefn 9706 9707@deftypevr {Target Hook} bool TARGET_ARM_EABI_UNWINDER 9708This flag should be set to @code{true} on targets that use an ARM EABI 9709based unwinding library, and @code{false} on other targets. This effects 9710the format of unwinding tables, and how the unwinder in entered after 9711running a cleanup. The default is @code{false}. 9712@end deftypevr 9713 9714@node Alignment Output 9715@subsection Assembler Commands for Alignment 9716 9717@c prevent bad page break with this line 9718This describes commands for alignment. 9719 9720@defmac JUMP_ALIGN (@var{label}) 9721The alignment (log base 2) to put in front of @var{label}, which is 9722a common destination of jumps and has no fallthru incoming edge. 9723 9724This macro need not be defined if you don't want any special alignment 9725to be done at such a time. Most machine descriptions do not currently 9726define the macro. 9727 9728Unless it's necessary to inspect the @var{label} parameter, it is better 9729to set the variable @var{align_jumps} in the target's 9730@code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's 9731selection in @var{align_jumps} in a @code{JUMP_ALIGN} implementation. 9732@end defmac 9733 9734@defmac LABEL_ALIGN_AFTER_BARRIER (@var{label}) 9735The alignment (log base 2) to put in front of @var{label}, which follows 9736a @code{BARRIER}. 9737 9738This macro need not be defined if you don't want any special alignment 9739to be done at such a time. Most machine descriptions do not currently 9740define the macro. 9741@end defmac 9742 9743@defmac LOOP_ALIGN (@var{label}) 9744The alignment (log base 2) to put in front of @var{label} that heads 9745a frequently executed basic block (usually the header of a loop). 9746 9747This macro need not be defined if you don't want any special alignment 9748to be done at such a time. Most machine descriptions do not currently 9749define the macro. 9750 9751Unless it's necessary to inspect the @var{label} parameter, it is better 9752to set the variable @code{align_loops} in the target's 9753@code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's 9754selection in @code{align_loops} in a @code{LOOP_ALIGN} implementation. 9755@end defmac 9756 9757@defmac LABEL_ALIGN (@var{label}) 9758The alignment (log base 2) to put in front of @var{label}. 9759If @code{LABEL_ALIGN_AFTER_BARRIER} / @code{LOOP_ALIGN} specify a different alignment, 9760the maximum of the specified values is used. 9761 9762Unless it's necessary to inspect the @var{label} parameter, it is better 9763to set the variable @code{align_labels} in the target's 9764@code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's 9765selection in @code{align_labels} in a @code{LABEL_ALIGN} implementation. 9766@end defmac 9767 9768@defmac ASM_OUTPUT_SKIP (@var{stream}, @var{nbytes}) 9769A C statement to output to the stdio stream @var{stream} an assembler 9770instruction to advance the location counter by @var{nbytes} bytes. 9771Those bytes should be zero when loaded. @var{nbytes} will be a C 9772expression of type @code{unsigned HOST_WIDE_INT}. 9773@end defmac 9774 9775@defmac ASM_NO_SKIP_IN_TEXT 9776Define this macro if @code{ASM_OUTPUT_SKIP} should not be used in the 9777text section because it fails to put zeros in the bytes that are skipped. 9778This is true on many Unix systems, where the pseudo--op to skip bytes 9779produces no-op instructions rather than zeros when used in the text 9780section. 9781@end defmac 9782 9783@defmac ASM_OUTPUT_ALIGN (@var{stream}, @var{power}) 9784A C statement to output to the stdio stream @var{stream} an assembler 9785command to advance the location counter to a multiple of 2 to the 9786@var{power} bytes. @var{power} will be a C expression of type @code{int}. 9787@end defmac 9788 9789@defmac ASM_OUTPUT_ALIGN_WITH_NOP (@var{stream}, @var{power}) 9790Like @code{ASM_OUTPUT_ALIGN}, except that the ``nop'' instruction is used 9791for padding, if necessary. 9792@end defmac 9793 9794@defmac ASM_OUTPUT_MAX_SKIP_ALIGN (@var{stream}, @var{power}, @var{max_skip}) 9795A C statement to output to the stdio stream @var{stream} an assembler 9796command to advance the location counter to a multiple of 2 to the 9797@var{power} bytes, but only if @var{max_skip} or fewer bytes are needed to 9798satisfy the alignment request. @var{power} and @var{max_skip} will be 9799a C expression of type @code{int}. 9800@end defmac 9801 9802@need 3000 9803@node Debugging Info 9804@section Controlling Debugging Information Format 9805 9806@c prevent bad page break with this line 9807This describes how to specify debugging information. 9808 9809@menu 9810* All Debuggers:: Macros that affect all debugging formats uniformly. 9811* DBX Options:: Macros enabling specific options in DBX format. 9812* DBX Hooks:: Hook macros for varying DBX format. 9813* File Names and DBX:: Macros controlling output of file names in DBX format. 9814* DWARF:: Macros for DWARF format. 9815* VMS Debug:: Macros for VMS debug format. 9816@end menu 9817 9818@node All Debuggers 9819@subsection Macros Affecting All Debugging Formats 9820 9821@c prevent bad page break with this line 9822These macros affect all debugging formats. 9823 9824@defmac DBX_REGISTER_NUMBER (@var{regno}) 9825A C expression that returns the DBX register number for the compiler 9826register number @var{regno}. In the default macro provided, the value 9827of this expression will be @var{regno} itself. But sometimes there are 9828some registers that the compiler knows about and DBX does not, or vice 9829versa. In such cases, some register may need to have one number in the 9830compiler and another for DBX@. 9831 9832If two registers have consecutive numbers inside GCC, and they can be 9833used as a pair to hold a multiword value, then they @emph{must} have 9834consecutive numbers after renumbering with @code{DBX_REGISTER_NUMBER}. 9835Otherwise, debuggers will be unable to access such a pair, because they 9836expect register pairs to be consecutive in their own numbering scheme. 9837 9838If you find yourself defining @code{DBX_REGISTER_NUMBER} in way that 9839does not preserve register pairs, then what you must do instead is 9840redefine the actual register numbering scheme. 9841@end defmac 9842 9843@defmac DEBUGGER_AUTO_OFFSET (@var{x}) 9844A C expression that returns the integer offset value for an automatic 9845variable having address @var{x} (an RTL expression). The default 9846computation assumes that @var{x} is based on the frame-pointer and 9847gives the offset from the frame-pointer. This is required for targets 9848that produce debugging output for DBX and allow the frame-pointer to be 9849eliminated when the @option{-g} option is used. 9850@end defmac 9851 9852@defmac DEBUGGER_ARG_OFFSET (@var{offset}, @var{x}) 9853A C expression that returns the integer offset value for an argument 9854having address @var{x} (an RTL expression). The nominal offset is 9855@var{offset}. 9856@end defmac 9857 9858@defmac PREFERRED_DEBUGGING_TYPE 9859A C expression that returns the type of debugging output GCC should 9860produce when the user specifies just @option{-g}. Define 9861this if you have arranged for GCC to support more than one format of 9862debugging output. Currently, the allowable values are @code{DBX_DEBUG}, 9863@code{DWARF2_DEBUG}, @code{XCOFF_DEBUG}, @code{VMS_DEBUG}, 9864and @code{VMS_AND_DWARF2_DEBUG}. 9865 9866When the user specifies @option{-ggdb}, GCC normally also uses the 9867value of this macro to select the debugging output format, but with two 9868exceptions. If @code{DWARF2_DEBUGGING_INFO} is defined, GCC uses the 9869value @code{DWARF2_DEBUG}. Otherwise, if @code{DBX_DEBUGGING_INFO} is 9870defined, GCC uses @code{DBX_DEBUG}. 9871 9872The value of this macro only affects the default debugging output; the 9873user can always get a specific type of output by using @option{-gstabs}, 9874@option{-gdwarf-2}, @option{-gxcoff}, or @option{-gvms}. 9875@end defmac 9876 9877@node DBX Options 9878@subsection Specific Options for DBX Output 9879 9880@c prevent bad page break with this line 9881These are specific options for DBX output. 9882 9883@defmac DBX_DEBUGGING_INFO 9884Define this macro if GCC should produce debugging output for DBX 9885in response to the @option{-g} option. 9886@end defmac 9887 9888@defmac XCOFF_DEBUGGING_INFO 9889Define this macro if GCC should produce XCOFF format debugging output 9890in response to the @option{-g} option. This is a variant of DBX format. 9891@end defmac 9892 9893@defmac DEFAULT_GDB_EXTENSIONS 9894Define this macro to control whether GCC should by default generate 9895GDB's extended version of DBX debugging information (assuming DBX-format 9896debugging information is enabled at all). If you don't define the 9897macro, the default is 1: always generate the extended information 9898if there is any occasion to. 9899@end defmac 9900 9901@defmac DEBUG_SYMS_TEXT 9902Define this macro if all @code{.stabs} commands should be output while 9903in the text section. 9904@end defmac 9905 9906@defmac ASM_STABS_OP 9907A C string constant, including spacing, naming the assembler pseudo op to 9908use instead of @code{"\t.stabs\t"} to define an ordinary debugging symbol. 9909If you don't define this macro, @code{"\t.stabs\t"} is used. This macro 9910applies only to DBX debugging information format. 9911@end defmac 9912 9913@defmac ASM_STABD_OP 9914A C string constant, including spacing, naming the assembler pseudo op to 9915use instead of @code{"\t.stabd\t"} to define a debugging symbol whose 9916value is the current location. If you don't define this macro, 9917@code{"\t.stabd\t"} is used. This macro applies only to DBX debugging 9918information format. 9919@end defmac 9920 9921@defmac ASM_STABN_OP 9922A C string constant, including spacing, naming the assembler pseudo op to 9923use instead of @code{"\t.stabn\t"} to define a debugging symbol with no 9924name. If you don't define this macro, @code{"\t.stabn\t"} is used. This 9925macro applies only to DBX debugging information format. 9926@end defmac 9927 9928@defmac DBX_NO_XREFS 9929Define this macro if DBX on your system does not support the construct 9930@samp{xs@var{tagname}}. On some systems, this construct is used to 9931describe a forward reference to a structure named @var{tagname}. 9932On other systems, this construct is not supported at all. 9933@end defmac 9934 9935@defmac DBX_CONTIN_LENGTH 9936A symbol name in DBX-format debugging information is normally 9937continued (split into two separate @code{.stabs} directives) when it 9938exceeds a certain length (by default, 80 characters). On some 9939operating systems, DBX requires this splitting; on others, splitting 9940must not be done. You can inhibit splitting by defining this macro 9941with the value zero. You can override the default splitting-length by 9942defining this macro as an expression for the length you desire. 9943@end defmac 9944 9945@defmac DBX_CONTIN_CHAR 9946Normally continuation is indicated by adding a @samp{\} character to 9947the end of a @code{.stabs} string when a continuation follows. To use 9948a different character instead, define this macro as a character 9949constant for the character you want to use. Do not define this macro 9950if backslash is correct for your system. 9951@end defmac 9952 9953@defmac DBX_STATIC_STAB_DATA_SECTION 9954Define this macro if it is necessary to go to the data section before 9955outputting the @samp{.stabs} pseudo-op for a non-global static 9956variable. 9957@end defmac 9958 9959@defmac DBX_TYPE_DECL_STABS_CODE 9960The value to use in the ``code'' field of the @code{.stabs} directive 9961for a typedef. The default is @code{N_LSYM}. 9962@end defmac 9963 9964@defmac DBX_STATIC_CONST_VAR_CODE 9965The value to use in the ``code'' field of the @code{.stabs} directive 9966for a static variable located in the text section. DBX format does not 9967provide any ``right'' way to do this. The default is @code{N_FUN}. 9968@end defmac 9969 9970@defmac DBX_REGPARM_STABS_CODE 9971The value to use in the ``code'' field of the @code{.stabs} directive 9972for a parameter passed in registers. DBX format does not provide any 9973``right'' way to do this. The default is @code{N_RSYM}. 9974@end defmac 9975 9976@defmac DBX_REGPARM_STABS_LETTER 9977The letter to use in DBX symbol data to identify a symbol as a parameter 9978passed in registers. DBX format does not customarily provide any way to 9979do this. The default is @code{'P'}. 9980@end defmac 9981 9982@defmac DBX_FUNCTION_FIRST 9983Define this macro if the DBX information for a function and its 9984arguments should precede the assembler code for the function. Normally, 9985in DBX format, the debugging information entirely follows the assembler 9986code. 9987@end defmac 9988 9989@defmac DBX_BLOCKS_FUNCTION_RELATIVE 9990Define this macro, with value 1, if the value of a symbol describing 9991the scope of a block (@code{N_LBRAC} or @code{N_RBRAC}) should be 9992relative to the start of the enclosing function. Normally, GCC uses 9993an absolute address. 9994@end defmac 9995 9996@defmac DBX_LINES_FUNCTION_RELATIVE 9997Define this macro, with value 1, if the value of a symbol indicating 9998the current line number (@code{N_SLINE}) should be relative to the 9999start of the enclosing function. Normally, GCC uses an absolute address. 10000@end defmac 10001 10002@defmac DBX_USE_BINCL 10003Define this macro if GCC should generate @code{N_BINCL} and 10004@code{N_EINCL} stabs for included header files, as on Sun systems. This 10005macro also directs GCC to output a type number as a pair of a file 10006number and a type number within the file. Normally, GCC does not 10007generate @code{N_BINCL} or @code{N_EINCL} stabs, and it outputs a single 10008number for a type number. 10009@end defmac 10010 10011@node DBX Hooks 10012@subsection Open-Ended Hooks for DBX Format 10013 10014@c prevent bad page break with this line 10015These are hooks for DBX format. 10016 10017@defmac DBX_OUTPUT_SOURCE_LINE (@var{stream}, @var{line}, @var{counter}) 10018A C statement to output DBX debugging information before code for line 10019number @var{line} of the current source file to the stdio stream 10020@var{stream}. @var{counter} is the number of time the macro was 10021invoked, including the current invocation; it is intended to generate 10022unique labels in the assembly output. 10023 10024This macro should not be defined if the default output is correct, or 10025if it can be made correct by defining @code{DBX_LINES_FUNCTION_RELATIVE}. 10026@end defmac 10027 10028@defmac NO_DBX_FUNCTION_END 10029Some stabs encapsulation formats (in particular ECOFF), cannot handle the 10030@code{.stabs "",N_FUN,,0,0,Lscope-function-1} gdb dbx extension construct. 10031On those machines, define this macro to turn this feature off without 10032disturbing the rest of the gdb extensions. 10033@end defmac 10034 10035@defmac NO_DBX_BNSYM_ENSYM 10036Some assemblers cannot handle the @code{.stabd BNSYM/ENSYM,0,0} gdb dbx 10037extension construct. On those machines, define this macro to turn this 10038feature off without disturbing the rest of the gdb extensions. 10039@end defmac 10040 10041@node File Names and DBX 10042@subsection File Names in DBX Format 10043 10044@c prevent bad page break with this line 10045This describes file names in DBX format. 10046 10047@defmac DBX_OUTPUT_MAIN_SOURCE_FILENAME (@var{stream}, @var{name}) 10048A C statement to output DBX debugging information to the stdio stream 10049@var{stream}, which indicates that file @var{name} is the main source 10050file---the file specified as the input file for compilation. 10051This macro is called only once, at the beginning of compilation. 10052 10053This macro need not be defined if the standard form of output 10054for DBX debugging information is appropriate. 10055 10056It may be necessary to refer to a label equal to the beginning of the 10057text section. You can use @samp{assemble_name (stream, ltext_label_name)} 10058to do so. If you do this, you must also set the variable 10059@var{used_ltext_label_name} to @code{true}. 10060@end defmac 10061 10062@defmac NO_DBX_MAIN_SOURCE_DIRECTORY 10063Define this macro, with value 1, if GCC should not emit an indication 10064of the current directory for compilation and current source language at 10065the beginning of the file. 10066@end defmac 10067 10068@defmac NO_DBX_GCC_MARKER 10069Define this macro, with value 1, if GCC should not emit an indication 10070that this object file was compiled by GCC@. The default is to emit 10071an @code{N_OPT} stab at the beginning of every source file, with 10072@samp{gcc2_compiled.} for the string and value 0. 10073@end defmac 10074 10075@defmac DBX_OUTPUT_MAIN_SOURCE_FILE_END (@var{stream}, @var{name}) 10076A C statement to output DBX debugging information at the end of 10077compilation of the main source file @var{name}. Output should be 10078written to the stdio stream @var{stream}. 10079 10080If you don't define this macro, nothing special is output at the end 10081of compilation, which is correct for most machines. 10082@end defmac 10083 10084@defmac DBX_OUTPUT_NULL_N_SO_AT_MAIN_SOURCE_FILE_END 10085Define this macro @emph{instead of} defining 10086@code{DBX_OUTPUT_MAIN_SOURCE_FILE_END}, if what needs to be output at 10087the end of compilation is an @code{N_SO} stab with an empty string, 10088whose value is the highest absolute text address in the file. 10089@end defmac 10090 10091@need 2000 10092@node DWARF 10093@subsection Macros for DWARF Output 10094 10095@c prevent bad page break with this line 10096Here are macros for DWARF output. 10097 10098@defmac DWARF2_DEBUGGING_INFO 10099Define this macro if GCC should produce dwarf version 2 format 10100debugging output in response to the @option{-g} option. 10101 10102@deftypefn {Target Hook} int TARGET_DWARF_CALLING_CONVENTION (const_tree @var{function}) 10103Define this to enable the dwarf attribute @code{DW_AT_calling_convention} to 10104be emitted for each function. Instead of an integer return the enum 10105value for the @code{DW_CC_} tag. 10106@end deftypefn 10107 10108To support optional call frame debugging information, you must also 10109define @code{INCOMING_RETURN_ADDR_RTX} and either set 10110@code{RTX_FRAME_RELATED_P} on the prologue insns if you use RTL for the 10111prologue, or call @code{dwarf2out_def_cfa} and @code{dwarf2out_reg_save} 10112as appropriate from @code{TARGET_ASM_FUNCTION_PROLOGUE} if you don't. 10113@end defmac 10114 10115@defmac DWARF2_FRAME_INFO 10116Define this macro to a nonzero value if GCC should always output 10117Dwarf 2 frame information. If @code{TARGET_EXCEPT_UNWIND_INFO} 10118(@pxref{Exception Region Output}) returns @code{UI_DWARF2}, and 10119exceptions are enabled, GCC will output this information not matter 10120how you define @code{DWARF2_FRAME_INFO}. 10121@end defmac 10122 10123@deftypefn {Target Hook} {enum unwind_info_type} TARGET_DEBUG_UNWIND_INFO (void) 10124This hook defines the mechanism that will be used for describing frame 10125unwind information to the debugger. Normally the hook will return 10126@code{UI_DWARF2} if DWARF 2 debug information is enabled, and 10127return @code{UI_NONE} otherwise. 10128 10129A target may return @code{UI_DWARF2} even when DWARF 2 debug information 10130is disabled in order to always output DWARF 2 frame information. 10131 10132A target may return @code{UI_TARGET} if it has ABI specified unwind tables. 10133This will suppress generation of the normal debug frame unwind information. 10134@end deftypefn 10135 10136@defmac DWARF2_ASM_LINE_DEBUG_INFO 10137Define this macro to be a nonzero value if the assembler can generate Dwarf 2 10138line debug info sections. This will result in much more compact line number 10139tables, and hence is desirable if it works. 10140@end defmac 10141 10142@defmac DWARF2_ASM_VIEW_DEBUG_INFO 10143Define this macro to be a nonzero value if the assembler supports view 10144assignment and verification in @code{.loc}. If it does not, but the 10145user enables location views, the compiler may have to fallback to 10146internal line number tables. 10147@end defmac 10148 10149@deftypefn {Target Hook} int TARGET_RESET_LOCATION_VIEW (rtx_insn *@var{}) 10150This hook, if defined, enables -ginternal-reset-location-views, and 10151uses its result to override cases in which the estimated min insn 10152length might be nonzero even when a PC advance (i.e., a view reset) 10153cannot be taken for granted. 10154 10155If the hook is defined, it must return a positive value to indicate 10156the insn definitely advances the PC, and so the view number can be 10157safely assumed to be reset; a negative value to mean the insn 10158definitely does not advance the PC, and os the view number must not 10159be reset; or zero to decide based on the estimated insn length. 10160 10161If insn length is to be regarded as reliable, set the hook to 10162@code{hook_int_rtx_insn_0}. 10163@end deftypefn 10164 10165@deftypevr {Target Hook} bool TARGET_WANT_DEBUG_PUB_SECTIONS 10166True if the @code{.debug_pubtypes} and @code{.debug_pubnames} sections should be emitted. These sections are not used on most platforms, and in particular GDB does not use them. 10167@end deftypevr 10168 10169@deftypevr {Target Hook} bool TARGET_DELAY_SCHED2 10170True if sched2 is not to be run at its normal place. 10171This usually means it will be run as part of machine-specific reorg. 10172@end deftypevr 10173 10174@deftypevr {Target Hook} bool TARGET_DELAY_VARTRACK 10175True if vartrack is not to be run at its normal place. 10176This usually means it will be run as part of machine-specific reorg. 10177@end deftypevr 10178 10179@deftypevr {Target Hook} bool TARGET_NO_REGISTER_ALLOCATION 10180True if register allocation and the passes 10181following it should not be run. Usually true only for virtual assembler 10182targets. 10183@end deftypevr 10184 10185@defmac ASM_OUTPUT_DWARF_DELTA (@var{stream}, @var{size}, @var{label1}, @var{label2}) 10186A C statement to issue assembly directives that create a difference 10187@var{lab1} minus @var{lab2}, using an integer of the given @var{size}. 10188@end defmac 10189 10190@defmac ASM_OUTPUT_DWARF_VMS_DELTA (@var{stream}, @var{size}, @var{label1}, @var{label2}) 10191A C statement to issue assembly directives that create a difference 10192between the two given labels in system defined units, e.g.@: instruction 10193slots on IA64 VMS, using an integer of the given size. 10194@end defmac 10195 10196@defmac ASM_OUTPUT_DWARF_OFFSET (@var{stream}, @var{size}, @var{label}, @var{offset}, @var{section}) 10197A C statement to issue assembly directives that create a 10198section-relative reference to the given @var{label} plus @var{offset}, using 10199an integer of the given @var{size}. The label is known to be defined in the 10200given @var{section}. 10201@end defmac 10202 10203@defmac ASM_OUTPUT_DWARF_PCREL (@var{stream}, @var{size}, @var{label}) 10204A C statement to issue assembly directives that create a self-relative 10205reference to the given @var{label}, using an integer of the given @var{size}. 10206@end defmac 10207 10208@defmac ASM_OUTPUT_DWARF_DATAREL (@var{stream}, @var{size}, @var{label}) 10209A C statement to issue assembly directives that create a reference to the 10210given @var{label} relative to the dbase, using an integer of the given @var{size}. 10211@end defmac 10212 10213@defmac ASM_OUTPUT_DWARF_TABLE_REF (@var{label}) 10214A C statement to issue assembly directives that create a reference to 10215the DWARF table identifier @var{label} from the current section. This 10216is used on some systems to avoid garbage collecting a DWARF table which 10217is referenced by a function. 10218@end defmac 10219 10220@deftypefn {Target Hook} void TARGET_ASM_OUTPUT_DWARF_DTPREL (FILE *@var{file}, int @var{size}, rtx @var{x}) 10221If defined, this target hook is a function which outputs a DTP-relative 10222reference to the given TLS symbol of the specified size. 10223@end deftypefn 10224 10225@need 2000 10226@node VMS Debug 10227@subsection Macros for VMS Debug Format 10228 10229@c prevent bad page break with this line 10230Here are macros for VMS debug format. 10231 10232@defmac VMS_DEBUGGING_INFO 10233Define this macro if GCC should produce debugging output for VMS 10234in response to the @option{-g} option. The default behavior for VMS 10235is to generate minimal debug info for a traceback in the absence of 10236@option{-g} unless explicitly overridden with @option{-g0}. This 10237behavior is controlled by @code{TARGET_OPTION_OPTIMIZATION} and 10238@code{TARGET_OPTION_OVERRIDE}. 10239@end defmac 10240 10241@node Floating Point 10242@section Cross Compilation and Floating Point 10243@cindex cross compilation and floating point 10244@cindex floating point and cross compilation 10245 10246While all modern machines use twos-complement representation for integers, 10247there are a variety of representations for floating point numbers. This 10248means that in a cross-compiler the representation of floating point numbers 10249in the compiled program may be different from that used in the machine 10250doing the compilation. 10251 10252Because different representation systems may offer different amounts of 10253range and precision, all floating point constants must be represented in 10254the target machine's format. Therefore, the cross compiler cannot 10255safely use the host machine's floating point arithmetic; it must emulate 10256the target's arithmetic. To ensure consistency, GCC always uses 10257emulation to work with floating point values, even when the host and 10258target floating point formats are identical. 10259 10260The following macros are provided by @file{real.h} for the compiler to 10261use. All parts of the compiler which generate or optimize 10262floating-point calculations must use these macros. They may evaluate 10263their operands more than once, so operands must not have side effects. 10264 10265@defmac REAL_VALUE_TYPE 10266The C data type to be used to hold a floating point value in the target 10267machine's format. Typically this is a @code{struct} containing an 10268array of @code{HOST_WIDE_INT}, but all code should treat it as an opaque 10269quantity. 10270@end defmac 10271 10272@deftypefn Macro HOST_WIDE_INT REAL_VALUE_FIX (REAL_VALUE_TYPE @var{x}) 10273Truncates @var{x} to a signed integer, rounding toward zero. 10274@end deftypefn 10275 10276@deftypefn Macro {unsigned HOST_WIDE_INT} REAL_VALUE_UNSIGNED_FIX (REAL_VALUE_TYPE @var{x}) 10277Truncates @var{x} to an unsigned integer, rounding toward zero. If 10278@var{x} is negative, returns zero. 10279@end deftypefn 10280 10281@deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ATOF (const char *@var{string}, machine_mode @var{mode}) 10282Converts @var{string} into a floating point number in the target machine's 10283representation for mode @var{mode}. This routine can handle both 10284decimal and hexadecimal floating point constants, using the syntax 10285defined by the C language for both. 10286@end deftypefn 10287 10288@deftypefn Macro int REAL_VALUE_NEGATIVE (REAL_VALUE_TYPE @var{x}) 10289Returns 1 if @var{x} is negative (including negative zero), 0 otherwise. 10290@end deftypefn 10291 10292@deftypefn Macro int REAL_VALUE_ISINF (REAL_VALUE_TYPE @var{x}) 10293Determines whether @var{x} represents infinity (positive or negative). 10294@end deftypefn 10295 10296@deftypefn Macro int REAL_VALUE_ISNAN (REAL_VALUE_TYPE @var{x}) 10297Determines whether @var{x} represents a ``NaN'' (not-a-number). 10298@end deftypefn 10299 10300@deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_NEGATE (REAL_VALUE_TYPE @var{x}) 10301Returns the negative of the floating point value @var{x}. 10302@end deftypefn 10303 10304@deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ABS (REAL_VALUE_TYPE @var{x}) 10305Returns the absolute value of @var{x}. 10306@end deftypefn 10307 10308@node Mode Switching 10309@section Mode Switching Instructions 10310@cindex mode switching 10311The following macros control mode switching optimizations: 10312 10313@defmac OPTIMIZE_MODE_SWITCHING (@var{entity}) 10314Define this macro if the port needs extra instructions inserted for mode 10315switching in an optimizing compilation. 10316 10317For an example, the SH4 can perform both single and double precision 10318floating point operations, but to perform a single precision operation, 10319the FPSCR PR bit has to be cleared, while for a double precision 10320operation, this bit has to be set. Changing the PR bit requires a general 10321purpose register as a scratch register, hence these FPSCR sets have to 10322be inserted before reload, i.e.@: you cannot put this into instruction emitting 10323or @code{TARGET_MACHINE_DEPENDENT_REORG}. 10324 10325You can have multiple entities that are mode-switched, and select at run time 10326which entities actually need it. @code{OPTIMIZE_MODE_SWITCHING} should 10327return nonzero for any @var{entity} that needs mode-switching. 10328If you define this macro, you also have to define 10329@code{NUM_MODES_FOR_MODE_SWITCHING}, @code{TARGET_MODE_NEEDED}, 10330@code{TARGET_MODE_PRIORITY} and @code{TARGET_MODE_EMIT}. 10331@code{TARGET_MODE_AFTER}, @code{TARGET_MODE_ENTRY}, and @code{TARGET_MODE_EXIT} 10332are optional. 10333@end defmac 10334 10335@defmac NUM_MODES_FOR_MODE_SWITCHING 10336If you define @code{OPTIMIZE_MODE_SWITCHING}, you have to define this as 10337initializer for an array of integers. Each initializer element 10338N refers to an entity that needs mode switching, and specifies the number 10339of different modes that might need to be set for this entity. 10340The position of the initializer in the initializer---starting counting at 10341zero---determines the integer that is used to refer to the mode-switched 10342entity in question. 10343In macros that take mode arguments / yield a mode result, modes are 10344represented as numbers 0 @dots{} N @minus{} 1. N is used to specify that no mode 10345switch is needed / supplied. 10346@end defmac 10347 10348@deftypefn {Target Hook} void TARGET_MODE_EMIT (int @var{entity}, int @var{mode}, int @var{prev_mode}, HARD_REG_SET @var{regs_live}) 10349Generate one or more insns to set @var{entity} to @var{mode}. @var{hard_reg_live} is the set of hard registers live at the point where the insn(s) are to be inserted. @var{prev_moxde} indicates the mode to switch from. Sets of a lower numbered entity will be emitted before sets of a higher numbered entity to a mode of the same or lower priority. 10350@end deftypefn 10351 10352@deftypefn {Target Hook} int TARGET_MODE_NEEDED (int @var{entity}, rtx_insn *@var{insn}) 10353@var{entity} is an integer specifying a mode-switched entity. If @code{OPTIMIZE_MODE_SWITCHING} is defined, you must define this macro to return an integer value not larger than the corresponding element in @code{NUM_MODES_FOR_MODE_SWITCHING}, to denote the mode that @var{entity} must be switched into prior to the execution of @var{insn}. 10354@end deftypefn 10355 10356@deftypefn {Target Hook} int TARGET_MODE_AFTER (int @var{entity}, int @var{mode}, rtx_insn *@var{insn}) 10357@var{entity} is an integer specifying a mode-switched entity. If this macro is defined, it is evaluated for every @var{insn} during mode switching. It determines the mode that an insn results in (if different from the incoming mode). 10358@end deftypefn 10359 10360@deftypefn {Target Hook} int TARGET_MODE_ENTRY (int @var{entity}) 10361If this macro is defined, it is evaluated for every @var{entity} that needs mode switching. It should evaluate to an integer, which is a mode that @var{entity} is assumed to be switched to at function entry. If @code{TARGET_MODE_ENTRY} is defined then @code{TARGET_MODE_EXIT} must be defined. 10362@end deftypefn 10363 10364@deftypefn {Target Hook} int TARGET_MODE_EXIT (int @var{entity}) 10365If this macro is defined, it is evaluated for every @var{entity} that needs mode switching. It should evaluate to an integer, which is a mode that @var{entity} is assumed to be switched to at function exit. If @code{TARGET_MODE_EXIT} is defined then @code{TARGET_MODE_ENTRY} must be defined. 10366@end deftypefn 10367 10368@deftypefn {Target Hook} int TARGET_MODE_PRIORITY (int @var{entity}, int @var{n}) 10369This macro specifies the order in which modes for @var{entity} are processed. 0 is the highest priority, @code{NUM_MODES_FOR_MODE_SWITCHING[@var{entity}] - 1} the lowest. The value of the macro should be an integer designating a mode for @var{entity}. For any fixed @var{entity}, @code{mode_priority} (@var{entity}, @var{n}) shall be a bijection in 0 @dots{} @code{num_modes_for_mode_switching[@var{entity}] - 1}. 10370@end deftypefn 10371 10372@node Target Attributes 10373@section Defining target-specific uses of @code{__attribute__} 10374@cindex target attributes 10375@cindex machine attributes 10376@cindex attributes, target-specific 10377 10378Target-specific attributes may be defined for functions, data and types. 10379These are described using the following target hooks; they also need to 10380be documented in @file{extend.texi}. 10381 10382@deftypevr {Target Hook} {const struct attribute_spec *} TARGET_ATTRIBUTE_TABLE 10383If defined, this target hook points to an array of @samp{struct 10384attribute_spec} (defined in @file{tree-core.h}) specifying the machine 10385specific attributes for this target and some of the restrictions on the 10386entities to which these attributes are applied and the arguments they 10387take. 10388@end deftypevr 10389 10390@deftypefn {Target Hook} bool TARGET_ATTRIBUTE_TAKES_IDENTIFIER_P (const_tree @var{name}) 10391If defined, this target hook is a function which returns true if the 10392machine-specific attribute named @var{name} expects an identifier 10393given as its first argument to be passed on as a plain identifier, not 10394subjected to name lookup. If this is not defined, the default is 10395false for all machine-specific attributes. 10396@end deftypefn 10397 10398@deftypefn {Target Hook} int TARGET_COMP_TYPE_ATTRIBUTES (const_tree @var{type1}, const_tree @var{type2}) 10399If defined, this target hook is a function which returns zero if the attributes on 10400@var{type1} and @var{type2} are incompatible, one if they are compatible, 10401and two if they are nearly compatible (which causes a warning to be 10402generated). If this is not defined, machine-specific attributes are 10403supposed always to be compatible. 10404@end deftypefn 10405 10406@deftypefn {Target Hook} void TARGET_SET_DEFAULT_TYPE_ATTRIBUTES (tree @var{type}) 10407If defined, this target hook is a function which assigns default attributes to 10408the newly defined @var{type}. 10409@end deftypefn 10410 10411@deftypefn {Target Hook} tree TARGET_MERGE_TYPE_ATTRIBUTES (tree @var{type1}, tree @var{type2}) 10412Define this target hook if the merging of type attributes needs special 10413handling. If defined, the result is a list of the combined 10414@code{TYPE_ATTRIBUTES} of @var{type1} and @var{type2}. It is assumed 10415that @code{comptypes} has already been called and returned 1. This 10416function may call @code{merge_attributes} to handle machine-independent 10417merging. 10418@end deftypefn 10419 10420@deftypefn {Target Hook} tree TARGET_MERGE_DECL_ATTRIBUTES (tree @var{olddecl}, tree @var{newdecl}) 10421Define this target hook if the merging of decl attributes needs special 10422handling. If defined, the result is a list of the combined 10423@code{DECL_ATTRIBUTES} of @var{olddecl} and @var{newdecl}. 10424@var{newdecl} is a duplicate declaration of @var{olddecl}. Examples of 10425when this is needed are when one attribute overrides another, or when an 10426attribute is nullified by a subsequent definition. This function may 10427call @code{merge_attributes} to handle machine-independent merging. 10428 10429@findex TARGET_DLLIMPORT_DECL_ATTRIBUTES 10430If the only target-specific handling you require is @samp{dllimport} 10431for Microsoft Windows targets, you should define the macro 10432@code{TARGET_DLLIMPORT_DECL_ATTRIBUTES} to @code{1}. The compiler 10433will then define a function called 10434@code{merge_dllimport_decl_attributes} which can then be defined as 10435the expansion of @code{TARGET_MERGE_DECL_ATTRIBUTES}. You can also 10436add @code{handle_dll_attribute} in the attribute table for your port 10437to perform initial processing of the @samp{dllimport} and 10438@samp{dllexport} attributes. This is done in @file{i386/cygwin.h} and 10439@file{i386/i386.c}, for example. 10440@end deftypefn 10441 10442@deftypefn {Target Hook} bool TARGET_VALID_DLLIMPORT_ATTRIBUTE_P (const_tree @var{decl}) 10443@var{decl} is a variable or function with @code{__attribute__((dllimport))} specified. Use this hook if the target needs to add extra validation checks to @code{handle_dll_attribute}. 10444@end deftypefn 10445 10446@defmac TARGET_DECLSPEC 10447Define this macro to a nonzero value if you want to treat 10448@code{__declspec(X)} as equivalent to @code{__attribute((X))}. By 10449default, this behavior is enabled only for targets that define 10450@code{TARGET_DLLIMPORT_DECL_ATTRIBUTES}. The current implementation 10451of @code{__declspec} is via a built-in macro, but you should not rely 10452on this implementation detail. 10453@end defmac 10454 10455@deftypefn {Target Hook} void TARGET_INSERT_ATTRIBUTES (tree @var{node}, tree *@var{attr_ptr}) 10456Define this target hook if you want to be able to add attributes to a decl 10457when it is being created. This is normally useful for back ends which 10458wish to implement a pragma by using the attributes which correspond to 10459the pragma's effect. The @var{node} argument is the decl which is being 10460created. The @var{attr_ptr} argument is a pointer to the attribute list 10461for this decl. The list itself should not be modified, since it may be 10462shared with other decls, but attributes may be chained on the head of 10463the list and @code{*@var{attr_ptr}} modified to point to the new 10464attributes, or a copy of the list may be made if further changes are 10465needed. 10466@end deftypefn 10467 10468@deftypefn {Target Hook} tree TARGET_HANDLE_GENERIC_ATTRIBUTE (tree *@var{node}, tree @var{name}, tree @var{args}, int @var{flags}, bool *@var{no_add_attrs}) 10469Define this target hook if you want to be able to perform additional 10470target-specific processing of an attribute which is handled generically 10471by a front end. The arguments are the same as those which are passed to 10472attribute handlers. So far this only affects the @var{noinit} and 10473@var{section} attribute. 10474@end deftypefn 10475 10476@deftypefn {Target Hook} bool TARGET_FUNCTION_ATTRIBUTE_INLINABLE_P (const_tree @var{fndecl}) 10477@cindex inlining 10478This target hook returns @code{true} if it is OK to inline @var{fndecl} 10479into the current function, despite its having target-specific 10480attributes, @code{false} otherwise. By default, if a function has a 10481target specific attribute attached to it, it will not be inlined. 10482@end deftypefn 10483 10484@deftypefn {Target Hook} bool TARGET_OPTION_VALID_ATTRIBUTE_P (tree @var{fndecl}, tree @var{name}, tree @var{args}, int @var{flags}) 10485This hook is called to parse @code{attribute(target("..."))}, which 10486allows setting target-specific options on individual functions. 10487These function-specific options may differ 10488from the options specified on the command line. The hook should return 10489@code{true} if the options are valid. 10490 10491The hook should set the @code{DECL_FUNCTION_SPECIFIC_TARGET} field in 10492the function declaration to hold a pointer to a target-specific 10493@code{struct cl_target_option} structure. 10494@end deftypefn 10495 10496@deftypefn {Target Hook} void TARGET_OPTION_SAVE (struct cl_target_option *@var{ptr}, struct gcc_options *@var{opts}) 10497This hook is called to save any additional target-specific information 10498in the @code{struct cl_target_option} structure for function-specific 10499options from the @code{struct gcc_options} structure. 10500@xref{Option file format}. 10501@end deftypefn 10502 10503@deftypefn {Target Hook} void TARGET_OPTION_RESTORE (struct gcc_options *@var{opts}, struct cl_target_option *@var{ptr}) 10504This hook is called to restore any additional target-specific 10505information in the @code{struct cl_target_option} structure for 10506function-specific options to the @code{struct gcc_options} structure. 10507@end deftypefn 10508 10509@deftypefn {Target Hook} void TARGET_OPTION_POST_STREAM_IN (struct cl_target_option *@var{ptr}) 10510This hook is called to update target-specific information in the 10511@code{struct cl_target_option} structure after it is streamed in from 10512LTO bytecode. 10513@end deftypefn 10514 10515@deftypefn {Target Hook} void TARGET_OPTION_PRINT (FILE *@var{file}, int @var{indent}, struct cl_target_option *@var{ptr}) 10516This hook is called to print any additional target-specific 10517information in the @code{struct cl_target_option} structure for 10518function-specific options. 10519@end deftypefn 10520 10521@deftypefn {Target Hook} bool TARGET_OPTION_PRAGMA_PARSE (tree @var{args}, tree @var{pop_target}) 10522This target hook parses the options for @code{#pragma GCC target}, which 10523sets the target-specific options for functions that occur later in the 10524input stream. The options accepted should be the same as those handled by the 10525@code{TARGET_OPTION_VALID_ATTRIBUTE_P} hook. 10526@end deftypefn 10527 10528@deftypefn {Target Hook} void TARGET_OPTION_OVERRIDE (void) 10529Sometimes certain combinations of command options do not make sense on 10530a particular target machine. You can override the hook 10531@code{TARGET_OPTION_OVERRIDE} to take account of this. This hooks is called 10532once just after all the command options have been parsed. 10533 10534Don't use this hook to turn on various extra optimizations for 10535@option{-O}. That is what @code{TARGET_OPTION_OPTIMIZATION} is for. 10536 10537If you need to do something whenever the optimization level is 10538changed via the optimize attribute or pragma, see 10539@code{TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE} 10540@end deftypefn 10541 10542@deftypefn {Target Hook} bool TARGET_OPTION_FUNCTION_VERSIONS (tree @var{decl1}, tree @var{decl2}) 10543This target hook returns @code{true} if @var{DECL1} and @var{DECL2} are 10544versions of the same function. @var{DECL1} and @var{DECL2} are function 10545versions if and only if they have the same function signature and 10546different target specific attributes, that is, they are compiled for 10547different target machines. 10548@end deftypefn 10549 10550@deftypefn {Target Hook} bool TARGET_CAN_INLINE_P (tree @var{caller}, tree @var{callee}) 10551This target hook returns @code{false} if the @var{caller} function 10552cannot inline @var{callee}, based on target specific information. By 10553default, inlining is not allowed if the callee function has function 10554specific target options and the caller does not use the same options. 10555@end deftypefn 10556 10557@deftypefn {Target Hook} void TARGET_RELAYOUT_FUNCTION (tree @var{fndecl}) 10558This target hook fixes function @var{fndecl} after attributes are processed. Default does nothing. On ARM, the default function's alignment is updated with the attribute target. 10559@end deftypefn 10560 10561@node Emulated TLS 10562@section Emulating TLS 10563@cindex Emulated TLS 10564 10565For targets whose psABI does not provide Thread Local Storage via 10566specific relocations and instruction sequences, an emulation layer is 10567used. A set of target hooks allows this emulation layer to be 10568configured for the requirements of a particular target. For instance 10569the psABI may in fact specify TLS support in terms of an emulation 10570layer. 10571 10572The emulation layer works by creating a control object for every TLS 10573object. To access the TLS object, a lookup function is provided 10574which, when given the address of the control object, will return the 10575address of the current thread's instance of the TLS object. 10576 10577@deftypevr {Target Hook} {const char *} TARGET_EMUTLS_GET_ADDRESS 10578Contains the name of the helper function that uses a TLS control 10579object to locate a TLS instance. The default causes libgcc's 10580emulated TLS helper function to be used. 10581@end deftypevr 10582 10583@deftypevr {Target Hook} {const char *} TARGET_EMUTLS_REGISTER_COMMON 10584Contains the name of the helper function that should be used at 10585program startup to register TLS objects that are implicitly 10586initialized to zero. If this is @code{NULL}, all TLS objects will 10587have explicit initializers. The default causes libgcc's emulated TLS 10588registration function to be used. 10589@end deftypevr 10590 10591@deftypevr {Target Hook} {const char *} TARGET_EMUTLS_VAR_SECTION 10592Contains the name of the section in which TLS control variables should 10593be placed. The default of @code{NULL} allows these to be placed in 10594any section. 10595@end deftypevr 10596 10597@deftypevr {Target Hook} {const char *} TARGET_EMUTLS_TMPL_SECTION 10598Contains the name of the section in which TLS initializers should be 10599placed. The default of @code{NULL} allows these to be placed in any 10600section. 10601@end deftypevr 10602 10603@deftypevr {Target Hook} {const char *} TARGET_EMUTLS_VAR_PREFIX 10604Contains the prefix to be prepended to TLS control variable names. 10605The default of @code{NULL} uses a target-specific prefix. 10606@end deftypevr 10607 10608@deftypevr {Target Hook} {const char *} TARGET_EMUTLS_TMPL_PREFIX 10609Contains the prefix to be prepended to TLS initializer objects. The 10610default of @code{NULL} uses a target-specific prefix. 10611@end deftypevr 10612 10613@deftypefn {Target Hook} tree TARGET_EMUTLS_VAR_FIELDS (tree @var{type}, tree *@var{name}) 10614Specifies a function that generates the FIELD_DECLs for a TLS control 10615object type. @var{type} is the RECORD_TYPE the fields are for and 10616@var{name} should be filled with the structure tag, if the default of 10617@code{__emutls_object} is unsuitable. The default creates a type suitable 10618for libgcc's emulated TLS function. 10619@end deftypefn 10620 10621@deftypefn {Target Hook} tree TARGET_EMUTLS_VAR_INIT (tree @var{var}, tree @var{decl}, tree @var{tmpl_addr}) 10622Specifies a function that generates the CONSTRUCTOR to initialize a 10623TLS control object. @var{var} is the TLS control object, @var{decl} 10624is the TLS object and @var{tmpl_addr} is the address of the 10625initializer. The default initializes libgcc's emulated TLS control object. 10626@end deftypefn 10627 10628@deftypevr {Target Hook} bool TARGET_EMUTLS_VAR_ALIGN_FIXED 10629Specifies whether the alignment of TLS control variable objects is 10630fixed and should not be increased as some backends may do to optimize 10631single objects. The default is false. 10632@end deftypevr 10633 10634@deftypevr {Target Hook} bool TARGET_EMUTLS_DEBUG_FORM_TLS_ADDRESS 10635Specifies whether a DWARF @code{DW_OP_form_tls_address} location descriptor 10636may be used to describe emulated TLS control objects. 10637@end deftypevr 10638 10639@node MIPS Coprocessors 10640@section Defining coprocessor specifics for MIPS targets. 10641@cindex MIPS coprocessor-definition macros 10642 10643The MIPS specification allows MIPS implementations to have as many as 4 10644coprocessors, each with as many as 32 private registers. GCC supports 10645accessing these registers and transferring values between the registers 10646and memory using asm-ized variables. For example: 10647 10648@smallexample 10649 register unsigned int cp0count asm ("c0r1"); 10650 unsigned int d; 10651 10652 d = cp0count + 3; 10653@end smallexample 10654 10655(``c0r1'' is the default name of register 1 in coprocessor 0; alternate 10656names may be added as described below, or the default names may be 10657overridden entirely in @code{SUBTARGET_CONDITIONAL_REGISTER_USAGE}.) 10658 10659Coprocessor registers are assumed to be epilogue-used; sets to them will 10660be preserved even if it does not appear that the register is used again 10661later in the function. 10662 10663Another note: according to the MIPS spec, coprocessor 1 (if present) is 10664the FPU@. One accesses COP1 registers through standard mips 10665floating-point support; they are not included in this mechanism. 10666 10667@node PCH Target 10668@section Parameters for Precompiled Header Validity Checking 10669@cindex parameters, precompiled headers 10670 10671@deftypefn {Target Hook} {void *} TARGET_GET_PCH_VALIDITY (size_t *@var{sz}) 10672This hook returns a pointer to the data needed by 10673@code{TARGET_PCH_VALID_P} and sets 10674@samp{*@var{sz}} to the size of the data in bytes. 10675@end deftypefn 10676 10677@deftypefn {Target Hook} {const char *} TARGET_PCH_VALID_P (const void *@var{data}, size_t @var{sz}) 10678This hook checks whether the options used to create a PCH file are 10679compatible with the current settings. It returns @code{NULL} 10680if so and a suitable error message if not. Error messages will 10681be presented to the user and must be localized using @samp{_(@var{msg})}. 10682 10683@var{data} is the data that was returned by @code{TARGET_GET_PCH_VALIDITY} 10684when the PCH file was created and @var{sz} is the size of that data in bytes. 10685It's safe to assume that the data was created by the same version of the 10686compiler, so no format checking is needed. 10687 10688The default definition of @code{default_pch_valid_p} should be 10689suitable for most targets. 10690@end deftypefn 10691 10692@deftypefn {Target Hook} {const char *} TARGET_CHECK_PCH_TARGET_FLAGS (int @var{pch_flags}) 10693If this hook is nonnull, the default implementation of 10694@code{TARGET_PCH_VALID_P} will use it to check for compatible values 10695of @code{target_flags}. @var{pch_flags} specifies the value that 10696@code{target_flags} had when the PCH file was created. The return 10697value is the same as for @code{TARGET_PCH_VALID_P}. 10698@end deftypefn 10699 10700@deftypefn {Target Hook} void TARGET_PREPARE_PCH_SAVE (void) 10701Called before writing out a PCH file. If the target has some 10702garbage-collected data that needs to be in a particular state on PCH loads, 10703it can use this hook to enforce that state. Very few targets need 10704to do anything here. 10705@end deftypefn 10706 10707@node C++ ABI 10708@section C++ ABI parameters 10709@cindex parameters, c++ abi 10710 10711@deftypefn {Target Hook} tree TARGET_CXX_GUARD_TYPE (void) 10712Define this hook to override the integer type used for guard variables. 10713These are used to implement one-time construction of static objects. The 10714default is long_long_integer_type_node. 10715@end deftypefn 10716 10717@deftypefn {Target Hook} bool TARGET_CXX_GUARD_MASK_BIT (void) 10718This hook determines how guard variables are used. It should return 10719@code{false} (the default) if the first byte should be used. A return value of 10720@code{true} indicates that only the least significant bit should be used. 10721@end deftypefn 10722 10723@deftypefn {Target Hook} tree TARGET_CXX_GET_COOKIE_SIZE (tree @var{type}) 10724This hook returns the size of the cookie to use when allocating an array 10725whose elements have the indicated @var{type}. Assumes that it is already 10726known that a cookie is needed. The default is 10727@code{max(sizeof (size_t), alignof(type))}, as defined in section 2.7 of the 10728IA64/Generic C++ ABI@. 10729@end deftypefn 10730 10731@deftypefn {Target Hook} bool TARGET_CXX_COOKIE_HAS_SIZE (void) 10732This hook should return @code{true} if the element size should be stored in 10733array cookies. The default is to return @code{false}. 10734@end deftypefn 10735 10736@deftypefn {Target Hook} int TARGET_CXX_IMPORT_EXPORT_CLASS (tree @var{type}, int @var{import_export}) 10737If defined by a backend this hook allows the decision made to export 10738class @var{type} to be overruled. Upon entry @var{import_export} 10739will contain 1 if the class is going to be exported, @minus{}1 if it is going 10740to be imported and 0 otherwise. This function should return the 10741modified value and perform any other actions necessary to support the 10742backend's targeted operating system. 10743@end deftypefn 10744 10745@deftypefn {Target Hook} bool TARGET_CXX_CDTOR_RETURNS_THIS (void) 10746This hook should return @code{true} if constructors and destructors return 10747the address of the object created/destroyed. The default is to return 10748@code{false}. 10749@end deftypefn 10750 10751@deftypefn {Target Hook} bool TARGET_CXX_KEY_METHOD_MAY_BE_INLINE (void) 10752This hook returns true if the key method for a class (i.e., the method 10753which, if defined in the current translation unit, causes the virtual 10754table to be emitted) may be an inline function. Under the standard 10755Itanium C++ ABI the key method may be an inline function so long as 10756the function is not declared inline in the class definition. Under 10757some variants of the ABI, an inline function can never be the key 10758method. The default is to return @code{true}. 10759@end deftypefn 10760 10761@deftypefn {Target Hook} void TARGET_CXX_DETERMINE_CLASS_DATA_VISIBILITY (tree @var{decl}) 10762@var{decl} is a virtual table, virtual table table, typeinfo object, or other similar implicit class data object that will be emitted with external linkage in this translation unit. No ELF visibility has been explicitly specified. If the target needs to specify a visibility other than that of the containing class, use this hook to set @code{DECL_VISIBILITY} and @code{DECL_VISIBILITY_SPECIFIED}. 10763@end deftypefn 10764 10765@deftypefn {Target Hook} bool TARGET_CXX_CLASS_DATA_ALWAYS_COMDAT (void) 10766This hook returns true (the default) if virtual tables and other 10767similar implicit class data objects are always COMDAT if they have 10768external linkage. If this hook returns false, then class data for 10769classes whose virtual table will be emitted in only one translation 10770unit will not be COMDAT. 10771@end deftypefn 10772 10773@deftypefn {Target Hook} bool TARGET_CXX_LIBRARY_RTTI_COMDAT (void) 10774This hook returns true (the default) if the RTTI information for 10775the basic types which is defined in the C++ runtime should always 10776be COMDAT, false if it should not be COMDAT. 10777@end deftypefn 10778 10779@deftypefn {Target Hook} bool TARGET_CXX_USE_AEABI_ATEXIT (void) 10780This hook returns true if @code{__aeabi_atexit} (as defined by the ARM EABI) 10781should be used to register static destructors when @option{-fuse-cxa-atexit} 10782is in effect. The default is to return false to use @code{__cxa_atexit}. 10783@end deftypefn 10784 10785@deftypefn {Target Hook} bool TARGET_CXX_USE_ATEXIT_FOR_CXA_ATEXIT (void) 10786This hook returns true if the target @code{atexit} function can be used 10787in the same manner as @code{__cxa_atexit} to register C++ static 10788destructors. This requires that @code{atexit}-registered functions in 10789shared libraries are run in the correct order when the libraries are 10790unloaded. The default is to return false. 10791@end deftypefn 10792 10793@deftypefn {Target Hook} void TARGET_CXX_ADJUST_CLASS_AT_DEFINITION (tree @var{type}) 10794@var{type} is a C++ class (i.e., RECORD_TYPE or UNION_TYPE) that has just been defined. Use this hook to make adjustments to the class (eg, tweak visibility or perform any other required target modifications). 10795@end deftypefn 10796 10797@deftypefn {Target Hook} tree TARGET_CXX_DECL_MANGLING_CONTEXT (const_tree @var{decl}) 10798Return target-specific mangling context of @var{decl} or @code{NULL_TREE}. 10799@end deftypefn 10800 10801@node D Language and ABI 10802@section D ABI parameters 10803@cindex parameters, d abi 10804 10805@deftypefn {D Target Hook} void TARGET_D_CPU_VERSIONS (void) 10806Declare all environmental version identifiers relating to the target CPU 10807using the function @code{builtin_version}, which takes a string representing 10808the name of the version. Version identifiers predefined by this hook apply 10809to all modules that are being compiled and imported. 10810@end deftypefn 10811 10812@deftypefn {D Target Hook} void TARGET_D_OS_VERSIONS (void) 10813Similarly to @code{TARGET_D_CPU_VERSIONS}, but is used for versions 10814relating to the target operating system. 10815@end deftypefn 10816 10817@deftypefn {D Target Hook} unsigned TARGET_D_CRITSEC_SIZE (void) 10818Returns the size of the data structure used by the target operating system 10819for critical sections and monitors. For example, on Microsoft Windows this 10820would return the @code{sizeof(CRITICAL_SECTION)}, while other platforms that 10821implement pthreads would return @code{sizeof(pthread_mutex_t)}. 10822@end deftypefn 10823 10824@node Named Address Spaces 10825@section Adding support for named address spaces 10826@cindex named address spaces 10827 10828The draft technical report of the ISO/IEC JTC1 S22 WG14 N1275 10829standards committee, @cite{Programming Languages - C - Extensions to 10830support embedded processors}, specifies a syntax for embedded 10831processors to specify alternate address spaces. You can configure a 10832GCC port to support section 5.1 of the draft report to add support for 10833address spaces other than the default address space. These address 10834spaces are new keywords that are similar to the @code{volatile} and 10835@code{const} type attributes. 10836 10837Pointers to named address spaces can have a different size than 10838pointers to the generic address space. 10839 10840For example, the SPU port uses the @code{__ea} address space to refer 10841to memory in the host processor, rather than memory local to the SPU 10842processor. Access to memory in the @code{__ea} address space involves 10843issuing DMA operations to move data between the host processor and the 10844local processor memory address space. Pointers in the @code{__ea} 10845address space are either 32 bits or 64 bits based on the 10846@option{-mea32} or @option{-mea64} switches (native SPU pointers are 10847always 32 bits). 10848 10849Internally, address spaces are represented as a small integer in the 10850range 0 to 15 with address space 0 being reserved for the generic 10851address space. 10852 10853To register a named address space qualifier keyword with the C front end, 10854the target may call the @code{c_register_addr_space} routine. For example, 10855the SPU port uses the following to declare @code{__ea} as the keyword for 10856named address space #1: 10857@smallexample 10858#define ADDR_SPACE_EA 1 10859c_register_addr_space ("__ea", ADDR_SPACE_EA); 10860@end smallexample 10861 10862@deftypefn {Target Hook} scalar_int_mode TARGET_ADDR_SPACE_POINTER_MODE (addr_space_t @var{address_space}) 10863Define this to return the machine mode to use for pointers to 10864@var{address_space} if the target supports named address spaces. 10865The default version of this hook returns @code{ptr_mode}. 10866@end deftypefn 10867 10868@deftypefn {Target Hook} scalar_int_mode TARGET_ADDR_SPACE_ADDRESS_MODE (addr_space_t @var{address_space}) 10869Define this to return the machine mode to use for addresses in 10870@var{address_space} if the target supports named address spaces. 10871The default version of this hook returns @code{Pmode}. 10872@end deftypefn 10873 10874@deftypefn {Target Hook} bool TARGET_ADDR_SPACE_VALID_POINTER_MODE (scalar_int_mode @var{mode}, addr_space_t @var{as}) 10875Define this to return nonzero if the port can handle pointers 10876with machine mode @var{mode} to address space @var{as}. This target 10877hook is the same as the @code{TARGET_VALID_POINTER_MODE} target hook, 10878except that it includes explicit named address space support. The default 10879version of this hook returns true for the modes returned by either the 10880@code{TARGET_ADDR_SPACE_POINTER_MODE} or @code{TARGET_ADDR_SPACE_ADDRESS_MODE} 10881target hooks for the given address space. 10882@end deftypefn 10883 10884@deftypefn {Target Hook} bool TARGET_ADDR_SPACE_LEGITIMATE_ADDRESS_P (machine_mode @var{mode}, rtx @var{exp}, bool @var{strict}, addr_space_t @var{as}) 10885Define this to return true if @var{exp} is a valid address for mode 10886@var{mode} in the named address space @var{as}. The @var{strict} 10887parameter says whether strict addressing is in effect after reload has 10888finished. This target hook is the same as the 10889@code{TARGET_LEGITIMATE_ADDRESS_P} target hook, except that it includes 10890explicit named address space support. 10891@end deftypefn 10892 10893@deftypefn {Target Hook} rtx TARGET_ADDR_SPACE_LEGITIMIZE_ADDRESS (rtx @var{x}, rtx @var{oldx}, machine_mode @var{mode}, addr_space_t @var{as}) 10894Define this to modify an invalid address @var{x} to be a valid address 10895with mode @var{mode} in the named address space @var{as}. This target 10896hook is the same as the @code{TARGET_LEGITIMIZE_ADDRESS} target hook, 10897except that it includes explicit named address space support. 10898@end deftypefn 10899 10900@deftypefn {Target Hook} bool TARGET_ADDR_SPACE_SUBSET_P (addr_space_t @var{subset}, addr_space_t @var{superset}) 10901Define this to return whether the @var{subset} named address space is 10902contained within the @var{superset} named address space. Pointers to 10903a named address space that is a subset of another named address space 10904will be converted automatically without a cast if used together in 10905arithmetic operations. Pointers to a superset address space can be 10906converted to pointers to a subset address space via explicit casts. 10907@end deftypefn 10908 10909@deftypefn {Target Hook} bool TARGET_ADDR_SPACE_ZERO_ADDRESS_VALID (addr_space_t @var{as}) 10910Define this to modify the default handling of address 0 for the 10911address space. Return true if 0 should be considered a valid address. 10912@end deftypefn 10913 10914@deftypefn {Target Hook} rtx TARGET_ADDR_SPACE_CONVERT (rtx @var{op}, tree @var{from_type}, tree @var{to_type}) 10915Define this to convert the pointer expression represented by the RTL 10916@var{op} with type @var{from_type} that points to a named address 10917space to a new pointer expression with type @var{to_type} that points 10918to a different named address space. When this hook it called, it is 10919guaranteed that one of the two address spaces is a subset of the other, 10920as determined by the @code{TARGET_ADDR_SPACE_SUBSET_P} target hook. 10921@end deftypefn 10922 10923@deftypefn {Target Hook} int TARGET_ADDR_SPACE_DEBUG (addr_space_t @var{as}) 10924Define this to define how the address space is encoded in dwarf. 10925The result is the value to be used with @code{DW_AT_address_class}. 10926@end deftypefn 10927 10928@deftypefn {Target Hook} void TARGET_ADDR_SPACE_DIAGNOSE_USAGE (addr_space_t @var{as}, location_t @var{loc}) 10929Define this hook if the availability of an address space depends on 10930command line options and some diagnostics should be printed when the 10931address space is used. This hook is called during parsing and allows 10932to emit a better diagnostic compared to the case where the address space 10933was not registered with @code{c_register_addr_space}. @var{as} is 10934the address space as registered with @code{c_register_addr_space}. 10935@var{loc} is the location of the address space qualifier token. 10936The default implementation does nothing. 10937@end deftypefn 10938 10939@node Misc 10940@section Miscellaneous Parameters 10941@cindex parameters, miscellaneous 10942 10943@c prevent bad page break with this line 10944Here are several miscellaneous parameters. 10945 10946@defmac HAS_LONG_COND_BRANCH 10947Define this boolean macro to indicate whether or not your architecture 10948has conditional branches that can span all of memory. It is used in 10949conjunction with an optimization that partitions hot and cold basic 10950blocks into separate sections of the executable. If this macro is 10951set to false, gcc will convert any conditional branches that attempt 10952to cross between sections into unconditional branches or indirect jumps. 10953@end defmac 10954 10955@defmac HAS_LONG_UNCOND_BRANCH 10956Define this boolean macro to indicate whether or not your architecture 10957has unconditional branches that can span all of memory. It is used in 10958conjunction with an optimization that partitions hot and cold basic 10959blocks into separate sections of the executable. If this macro is 10960set to false, gcc will convert any unconditional branches that attempt 10961to cross between sections into indirect jumps. 10962@end defmac 10963 10964@defmac CASE_VECTOR_MODE 10965An alias for a machine mode name. This is the machine mode that 10966elements of a jump-table should have. 10967@end defmac 10968 10969@defmac CASE_VECTOR_SHORTEN_MODE (@var{min_offset}, @var{max_offset}, @var{body}) 10970Optional: return the preferred mode for an @code{addr_diff_vec} 10971when the minimum and maximum offset are known. If you define this, 10972it enables extra code in branch shortening to deal with @code{addr_diff_vec}. 10973To make this work, you also have to define @code{INSN_ALIGN} and 10974make the alignment for @code{addr_diff_vec} explicit. 10975The @var{body} argument is provided so that the offset_unsigned and scale 10976flags can be updated. 10977@end defmac 10978 10979@defmac CASE_VECTOR_PC_RELATIVE 10980Define this macro to be a C expression to indicate when jump-tables 10981should contain relative addresses. You need not define this macro if 10982jump-tables never contain relative addresses, or jump-tables should 10983contain relative addresses only when @option{-fPIC} or @option{-fPIC} 10984is in effect. 10985@end defmac 10986 10987@deftypefn {Target Hook} {unsigned int} TARGET_CASE_VALUES_THRESHOLD (void) 10988This function return the smallest number of different values for which it 10989is best to use a jump-table instead of a tree of conditional branches. 10990The default is four for machines with a @code{casesi} instruction and 10991five otherwise. This is best for most machines. 10992@end deftypefn 10993 10994@defmac WORD_REGISTER_OPERATIONS 10995Define this macro to 1 if operations between registers with integral mode 10996smaller than a word are always performed on the entire register. To be 10997more explicit, if you start with a pair of @code{word_mode} registers with 10998known values and you do a subword, for example @code{QImode}, addition on 10999the low part of the registers, then the compiler may consider that the 11000result has a known value in @code{word_mode} too if the macro is defined 11001to 1. Most RISC machines have this property and most CISC machines do not. 11002@end defmac 11003 11004@deftypefn {Target Hook} {unsigned int} TARGET_MIN_ARITHMETIC_PRECISION (void) 11005On some RISC architectures with 64-bit registers, the processor also 11006maintains 32-bit condition codes that make it possible to do real 32-bit 11007arithmetic, although the operations are performed on the full registers. 11008 11009On such architectures, defining this hook to 32 tells the compiler to try 11010using 32-bit arithmetical operations setting the condition codes instead 11011of doing full 64-bit arithmetic. 11012 11013More generally, define this hook on RISC architectures if you want the 11014compiler to try using arithmetical operations setting the condition codes 11015with a precision lower than the word precision. 11016 11017You need not define this hook if @code{WORD_REGISTER_OPERATIONS} is not 11018defined to 1. 11019@end deftypefn 11020 11021@defmac LOAD_EXTEND_OP (@var{mem_mode}) 11022Define this macro to be a C expression indicating when insns that read 11023memory in @var{mem_mode}, an integral mode narrower than a word, set the 11024bits outside of @var{mem_mode} to be either the sign-extension or the 11025zero-extension of the data read. Return @code{SIGN_EXTEND} for values 11026of @var{mem_mode} for which the 11027insn sign-extends, @code{ZERO_EXTEND} for which it zero-extends, and 11028@code{UNKNOWN} for other modes. 11029 11030This macro is not called with @var{mem_mode} non-integral or with a width 11031greater than or equal to @code{BITS_PER_WORD}, so you may return any 11032value in this case. Do not define this macro if it would always return 11033@code{UNKNOWN}. On machines where this macro is defined, you will normally 11034define it as the constant @code{SIGN_EXTEND} or @code{ZERO_EXTEND}. 11035 11036You may return a non-@code{UNKNOWN} value even if for some hard registers 11037the sign extension is not performed, if for the @code{REGNO_REG_CLASS} 11038of these hard registers @code{TARGET_CAN_CHANGE_MODE_CLASS} returns false 11039when the @var{from} mode is @var{mem_mode} and the @var{to} mode is any 11040integral mode larger than this but not larger than @code{word_mode}. 11041 11042You must return @code{UNKNOWN} if for some hard registers that allow this 11043mode, @code{TARGET_CAN_CHANGE_MODE_CLASS} says that they cannot change to 11044@code{word_mode}, but that they can change to another integral mode that 11045is larger then @var{mem_mode} but still smaller than @code{word_mode}. 11046@end defmac 11047 11048@defmac SHORT_IMMEDIATES_SIGN_EXTEND 11049Define this macro to 1 if loading short immediate values into registers sign 11050extends. 11051@end defmac 11052 11053@deftypefn {Target Hook} {unsigned int} TARGET_MIN_DIVISIONS_FOR_RECIP_MUL (machine_mode @var{mode}) 11054When @option{-ffast-math} is in effect, GCC tries to optimize 11055divisions by the same divisor, by turning them into multiplications by 11056the reciprocal. This target hook specifies the minimum number of divisions 11057that should be there for GCC to perform the optimization for a variable 11058of mode @var{mode}. The default implementation returns 3 if the machine 11059has an instruction for the division, and 2 if it does not. 11060@end deftypefn 11061 11062@defmac MOVE_MAX 11063The maximum number of bytes that a single instruction can move quickly 11064between memory and registers or between two memory locations. 11065@end defmac 11066 11067@defmac MAX_MOVE_MAX 11068The maximum number of bytes that a single instruction can move quickly 11069between memory and registers or between two memory locations. If this 11070is undefined, the default is @code{MOVE_MAX}. Otherwise, it is the 11071constant value that is the largest value that @code{MOVE_MAX} can have 11072at run-time. 11073@end defmac 11074 11075@defmac SHIFT_COUNT_TRUNCATED 11076A C expression that is nonzero if on this machine the number of bits 11077actually used for the count of a shift operation is equal to the number 11078of bits needed to represent the size of the object being shifted. When 11079this macro is nonzero, the compiler will assume that it is safe to omit 11080a sign-extend, zero-extend, and certain bitwise `and' instructions that 11081truncates the count of a shift operation. On machines that have 11082instructions that act on bit-fields at variable positions, which may 11083include `bit test' instructions, a nonzero @code{SHIFT_COUNT_TRUNCATED} 11084also enables deletion of truncations of the values that serve as 11085arguments to bit-field instructions. 11086 11087If both types of instructions truncate the count (for shifts) and 11088position (for bit-field operations), or if no variable-position bit-field 11089instructions exist, you should define this macro. 11090 11091However, on some machines, such as the 80386 and the 680x0, truncation 11092only applies to shift operations and not the (real or pretended) 11093bit-field operations. Define @code{SHIFT_COUNT_TRUNCATED} to be zero on 11094such machines. Instead, add patterns to the @file{md} file that include 11095the implied truncation of the shift instructions. 11096 11097You need not define this macro if it would always have the value of zero. 11098@end defmac 11099 11100@anchor{TARGET_SHIFT_TRUNCATION_MASK} 11101@deftypefn {Target Hook} {unsigned HOST_WIDE_INT} TARGET_SHIFT_TRUNCATION_MASK (machine_mode @var{mode}) 11102This function describes how the standard shift patterns for @var{mode} 11103deal with shifts by negative amounts or by more than the width of the mode. 11104@xref{shift patterns}. 11105 11106On many machines, the shift patterns will apply a mask @var{m} to the 11107shift count, meaning that a fixed-width shift of @var{x} by @var{y} is 11108equivalent to an arbitrary-width shift of @var{x} by @var{y & m}. If 11109this is true for mode @var{mode}, the function should return @var{m}, 11110otherwise it should return 0. A return value of 0 indicates that no 11111particular behavior is guaranteed. 11112 11113Note that, unlike @code{SHIFT_COUNT_TRUNCATED}, this function does 11114@emph{not} apply to general shift rtxes; it applies only to instructions 11115that are generated by the named shift patterns. 11116 11117The default implementation of this function returns 11118@code{GET_MODE_BITSIZE (@var{mode}) - 1} if @code{SHIFT_COUNT_TRUNCATED} 11119and 0 otherwise. This definition is always safe, but if 11120@code{SHIFT_COUNT_TRUNCATED} is false, and some shift patterns 11121nevertheless truncate the shift count, you may get better code 11122by overriding it. 11123@end deftypefn 11124 11125@deftypefn {Target Hook} bool TARGET_TRULY_NOOP_TRUNCATION (poly_uint64 @var{outprec}, poly_uint64 @var{inprec}) 11126This hook returns true if it is safe to ``convert'' a value of 11127@var{inprec} bits to one of @var{outprec} bits (where @var{outprec} is 11128smaller than @var{inprec}) by merely operating on it as if it had only 11129@var{outprec} bits. The default returns true unconditionally, which 11130is correct for most machines. 11131 11132If @code{TARGET_MODES_TIEABLE_P} returns false for a pair of modes, 11133suboptimal code can result if this hook returns true for the corresponding 11134mode sizes. Making this hook return false in such cases may improve things. 11135@end deftypefn 11136 11137@deftypefn {Target Hook} int TARGET_MODE_REP_EXTENDED (scalar_int_mode @var{mode}, scalar_int_mode @var{rep_mode}) 11138The representation of an integral mode can be such that the values 11139are always extended to a wider integral mode. Return 11140@code{SIGN_EXTEND} if values of @var{mode} are represented in 11141sign-extended form to @var{rep_mode}. Return @code{UNKNOWN} 11142otherwise. (Currently, none of the targets use zero-extended 11143representation this way so unlike @code{LOAD_EXTEND_OP}, 11144@code{TARGET_MODE_REP_EXTENDED} is expected to return either 11145@code{SIGN_EXTEND} or @code{UNKNOWN}. Also no target extends 11146@var{mode} to @var{rep_mode} so that @var{rep_mode} is not the next 11147widest integral mode and currently we take advantage of this fact.) 11148 11149Similarly to @code{LOAD_EXTEND_OP} you may return a non-@code{UNKNOWN} 11150value even if the extension is not performed on certain hard registers 11151as long as for the @code{REGNO_REG_CLASS} of these hard registers 11152@code{TARGET_CAN_CHANGE_MODE_CLASS} returns false. 11153 11154Note that @code{TARGET_MODE_REP_EXTENDED} and @code{LOAD_EXTEND_OP} 11155describe two related properties. If you define 11156@code{TARGET_MODE_REP_EXTENDED (mode, word_mode)} you probably also want 11157to define @code{LOAD_EXTEND_OP (mode)} to return the same type of 11158extension. 11159 11160In order to enforce the representation of @code{mode}, 11161@code{TARGET_TRULY_NOOP_TRUNCATION} should return false when truncating to 11162@code{mode}. 11163@end deftypefn 11164 11165@deftypefn {Target Hook} bool TARGET_SETJMP_PRESERVES_NONVOLATILE_REGS_P (void) 11166On some targets, it is assumed that the compiler will spill all pseudos 11167 that are live across a call to @code{setjmp}, while other targets treat 11168 @code{setjmp} calls as normal function calls. 11169 11170 This hook returns false if @code{setjmp} calls do not preserve all 11171 non-volatile registers so that gcc that must spill all pseudos that are 11172 live across @code{setjmp} calls. Define this to return true if the 11173 target does not need to spill all pseudos live across @code{setjmp} calls. 11174 The default implementation conservatively assumes all pseudos must be 11175 spilled across @code{setjmp} calls. 11176@end deftypefn 11177 11178@defmac STORE_FLAG_VALUE 11179A C expression describing the value returned by a comparison operator 11180with an integral mode and stored by a store-flag instruction 11181(@samp{cstore@var{mode}4}) when the condition is true. This description must 11182apply to @emph{all} the @samp{cstore@var{mode}4} patterns and all the 11183comparison operators whose results have a @code{MODE_INT} mode. 11184 11185A value of 1 or @minus{}1 means that the instruction implementing the 11186comparison operator returns exactly 1 or @minus{}1 when the comparison is true 11187and 0 when the comparison is false. Otherwise, the value indicates 11188which bits of the result are guaranteed to be 1 when the comparison is 11189true. This value is interpreted in the mode of the comparison 11190operation, which is given by the mode of the first operand in the 11191@samp{cstore@var{mode}4} pattern. Either the low bit or the sign bit of 11192@code{STORE_FLAG_VALUE} be on. Presently, only those bits are used by 11193the compiler. 11194 11195If @code{STORE_FLAG_VALUE} is neither 1 or @minus{}1, the compiler will 11196generate code that depends only on the specified bits. It can also 11197replace comparison operators with equivalent operations if they cause 11198the required bits to be set, even if the remaining bits are undefined. 11199For example, on a machine whose comparison operators return an 11200@code{SImode} value and where @code{STORE_FLAG_VALUE} is defined as 11201@samp{0x80000000}, saying that just the sign bit is relevant, the 11202expression 11203 11204@smallexample 11205(ne:SI (and:SI @var{x} (const_int @var{power-of-2})) (const_int 0)) 11206@end smallexample 11207 11208@noindent 11209can be converted to 11210 11211@smallexample 11212(ashift:SI @var{x} (const_int @var{n})) 11213@end smallexample 11214 11215@noindent 11216where @var{n} is the appropriate shift count to move the bit being 11217tested into the sign bit. 11218 11219There is no way to describe a machine that always sets the low-order bit 11220for a true value, but does not guarantee the value of any other bits, 11221but we do not know of any machine that has such an instruction. If you 11222are trying to port GCC to such a machine, include an instruction to 11223perform a logical-and of the result with 1 in the pattern for the 11224comparison operators and let us know at @email{gcc@@gcc.gnu.org}. 11225 11226Often, a machine will have multiple instructions that obtain a value 11227from a comparison (or the condition codes). Here are rules to guide the 11228choice of value for @code{STORE_FLAG_VALUE}, and hence the instructions 11229to be used: 11230 11231@itemize @bullet 11232@item 11233Use the shortest sequence that yields a valid definition for 11234@code{STORE_FLAG_VALUE}. It is more efficient for the compiler to 11235``normalize'' the value (convert it to, e.g., 1 or 0) than for the 11236comparison operators to do so because there may be opportunities to 11237combine the normalization with other operations. 11238 11239@item 11240For equal-length sequences, use a value of 1 or @minus{}1, with @minus{}1 being 11241slightly preferred on machines with expensive jumps and 1 preferred on 11242other machines. 11243 11244@item 11245As a second choice, choose a value of @samp{0x80000001} if instructions 11246exist that set both the sign and low-order bits but do not define the 11247others. 11248 11249@item 11250Otherwise, use a value of @samp{0x80000000}. 11251@end itemize 11252 11253Many machines can produce both the value chosen for 11254@code{STORE_FLAG_VALUE} and its negation in the same number of 11255instructions. On those machines, you should also define a pattern for 11256those cases, e.g., one matching 11257 11258@smallexample 11259(set @var{A} (neg:@var{m} (ne:@var{m} @var{B} @var{C}))) 11260@end smallexample 11261 11262Some machines can also perform @code{and} or @code{plus} operations on 11263condition code values with less instructions than the corresponding 11264@samp{cstore@var{mode}4} insn followed by @code{and} or @code{plus}. On those 11265machines, define the appropriate patterns. Use the names @code{incscc} 11266and @code{decscc}, respectively, for the patterns which perform 11267@code{plus} or @code{minus} operations on condition code values. See 11268@file{rs6000.md} for some examples. The GNU Superoptimizer can be used to 11269find such instruction sequences on other machines. 11270 11271If this macro is not defined, the default value, 1, is used. You need 11272not define @code{STORE_FLAG_VALUE} if the machine has no store-flag 11273instructions, or if the value generated by these instructions is 1. 11274@end defmac 11275 11276@defmac FLOAT_STORE_FLAG_VALUE (@var{mode}) 11277A C expression that gives a nonzero @code{REAL_VALUE_TYPE} value that is 11278returned when comparison operators with floating-point results are true. 11279Define this macro on machines that have comparison operations that return 11280floating-point values. If there are no such operations, do not define 11281this macro. 11282@end defmac 11283 11284@defmac VECTOR_STORE_FLAG_VALUE (@var{mode}) 11285A C expression that gives a rtx representing the nonzero true element 11286for vector comparisons. The returned rtx should be valid for the inner 11287mode of @var{mode} which is guaranteed to be a vector mode. Define 11288this macro on machines that have vector comparison operations that 11289return a vector result. If there are no such operations, do not define 11290this macro. Typically, this macro is defined as @code{const1_rtx} or 11291@code{constm1_rtx}. This macro may return @code{NULL_RTX} to prevent 11292the compiler optimizing such vector comparison operations for the 11293given mode. 11294@end defmac 11295 11296@defmac CLZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value}) 11297@defmacx CTZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value}) 11298A C expression that indicates whether the architecture defines a value 11299for @code{clz} or @code{ctz} with a zero operand. 11300A result of @code{0} indicates the value is undefined. 11301If the value is defined for only the RTL expression, the macro should 11302evaluate to @code{1}; if the value applies also to the corresponding optab 11303entry (which is normally the case if it expands directly into 11304the corresponding RTL), then the macro should evaluate to @code{2}. 11305In the cases where the value is defined, @var{value} should be set to 11306this value. 11307 11308If this macro is not defined, the value of @code{clz} or 11309@code{ctz} at zero is assumed to be undefined. 11310 11311This macro must be defined if the target's expansion for @code{ffs} 11312relies on a particular value to get correct results. Otherwise it 11313is not necessary, though it may be used to optimize some corner cases, and 11314to provide a default expansion for the @code{ffs} optab. 11315 11316Note that regardless of this macro the ``definedness'' of @code{clz} 11317and @code{ctz} at zero do @emph{not} extend to the builtin functions 11318visible to the user. Thus one may be free to adjust the value at will 11319to match the target expansion of these operations without fear of 11320breaking the API@. 11321@end defmac 11322 11323@defmac Pmode 11324An alias for the machine mode for pointers. On most machines, define 11325this to be the integer mode corresponding to the width of a hardware 11326pointer; @code{SImode} on 32-bit machine or @code{DImode} on 64-bit machines. 11327On some machines you must define this to be one of the partial integer 11328modes, such as @code{PSImode}. 11329 11330The width of @code{Pmode} must be at least as large as the value of 11331@code{POINTER_SIZE}. If it is not equal, you must define the macro 11332@code{POINTERS_EXTEND_UNSIGNED} to specify how pointers are extended 11333to @code{Pmode}. 11334@end defmac 11335 11336@defmac FUNCTION_MODE 11337An alias for the machine mode used for memory references to functions 11338being called, in @code{call} RTL expressions. On most CISC machines, 11339where an instruction can begin at any byte address, this should be 11340@code{QImode}. On most RISC machines, where all instructions have fixed 11341size and alignment, this should be a mode with the same size and alignment 11342as the machine instruction words - typically @code{SImode} or @code{HImode}. 11343@end defmac 11344 11345@defmac STDC_0_IN_SYSTEM_HEADERS 11346In normal operation, the preprocessor expands @code{__STDC__} to the 11347constant 1, to signify that GCC conforms to ISO Standard C@. On some 11348hosts, like Solaris, the system compiler uses a different convention, 11349where @code{__STDC__} is normally 0, but is 1 if the user specifies 11350strict conformance to the C Standard. 11351 11352Defining @code{STDC_0_IN_SYSTEM_HEADERS} makes GNU CPP follows the host 11353convention when processing system header files, but when processing user 11354files @code{__STDC__} will always expand to 1. 11355@end defmac 11356 11357@deftypefn {C Target Hook} {const char *} TARGET_C_PREINCLUDE (void) 11358Define this hook to return the name of a header file to be included at the start of all compilations, as if it had been included with @code{#include <@var{file}>}. If this hook returns @code{NULL}, or is not defined, or the header is not found, or if the user specifies @option{-ffreestanding} or @option{-nostdinc}, no header is included. 11359 11360 This hook can be used together with a header provided by the system C library to implement ISO C requirements for certain macros to be predefined that describe properties of the whole implementation rather than just the compiler. 11361@end deftypefn 11362 11363@deftypefn {C Target Hook} bool TARGET_CXX_IMPLICIT_EXTERN_C (const char*@var{}) 11364Define this hook to add target-specific C++ implicit extern C functions. If this function returns true for the name of a file-scope function, that function implicitly gets extern "C" linkage rather than whatever language linkage the declaration would normally have. An example of such function is WinMain on Win32 targets. 11365@end deftypefn 11366 11367@defmac SYSTEM_IMPLICIT_EXTERN_C 11368Define this macro if the system header files do not support C++@. 11369This macro handles system header files by pretending that system 11370header files are enclosed in @samp{extern "C" @{@dots{}@}}. 11371@end defmac 11372 11373@findex #pragma 11374@findex pragma 11375@defmac REGISTER_TARGET_PRAGMAS () 11376Define this macro if you want to implement any target-specific pragmas. 11377If defined, it is a C expression which makes a series of calls to 11378@code{c_register_pragma} or @code{c_register_pragma_with_expansion} 11379for each pragma. The macro may also do any 11380setup required for the pragmas. 11381 11382The primary reason to define this macro is to provide compatibility with 11383other compilers for the same target. In general, we discourage 11384definition of target-specific pragmas for GCC@. 11385 11386If the pragma can be implemented by attributes then you should consider 11387defining the target hook @samp{TARGET_INSERT_ATTRIBUTES} as well. 11388 11389Preprocessor macros that appear on pragma lines are not expanded. All 11390@samp{#pragma} directives that do not match any registered pragma are 11391silently ignored, unless the user specifies @option{-Wunknown-pragmas}. 11392@end defmac 11393 11394@deftypefun void c_register_pragma (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *)) 11395@deftypefunx void c_register_pragma_with_expansion (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *)) 11396 11397Each call to @code{c_register_pragma} or 11398@code{c_register_pragma_with_expansion} establishes one pragma. The 11399@var{callback} routine will be called when the preprocessor encounters a 11400pragma of the form 11401 11402@smallexample 11403#pragma [@var{space}] @var{name} @dots{} 11404@end smallexample 11405 11406@var{space} is the case-sensitive namespace of the pragma, or 11407@code{NULL} to put the pragma in the global namespace. The callback 11408routine receives @var{pfile} as its first argument, which can be passed 11409on to cpplib's functions if necessary. You can lex tokens after the 11410@var{name} by calling @code{pragma_lex}. Tokens that are not read by the 11411callback will be silently ignored. The end of the line is indicated by 11412a token of type @code{CPP_EOF}. Macro expansion occurs on the 11413arguments of pragmas registered with 11414@code{c_register_pragma_with_expansion} but not on the arguments of 11415pragmas registered with @code{c_register_pragma}. 11416 11417Note that the use of @code{pragma_lex} is specific to the C and C++ 11418compilers. It will not work in the Java or Fortran compilers, or any 11419other language compilers for that matter. Thus if @code{pragma_lex} is going 11420to be called from target-specific code, it must only be done so when 11421building the C and C++ compilers. This can be done by defining the 11422variables @code{c_target_objs} and @code{cxx_target_objs} in the 11423target entry in the @file{config.gcc} file. These variables should name 11424the target-specific, language-specific object file which contains the 11425code that uses @code{pragma_lex}. Note it will also be necessary to add a 11426rule to the makefile fragment pointed to by @code{tmake_file} that shows 11427how to build this object file. 11428@end deftypefun 11429 11430@defmac HANDLE_PRAGMA_PACK_WITH_EXPANSION 11431Define this macro if macros should be expanded in the 11432arguments of @samp{#pragma pack}. 11433@end defmac 11434 11435@defmac TARGET_DEFAULT_PACK_STRUCT 11436If your target requires a structure packing default other than 0 (meaning 11437the machine default), define this macro to the necessary value (in bytes). 11438This must be a value that would also be valid to use with 11439@samp{#pragma pack()} (that is, a small power of two). 11440@end defmac 11441 11442@defmac DOLLARS_IN_IDENTIFIERS 11443Define this macro to control use of the character @samp{$} in 11444identifier names for the C family of languages. 0 means @samp{$} is 11445not allowed by default; 1 means it is allowed. 1 is the default; 11446there is no need to define this macro in that case. 11447@end defmac 11448 11449@defmac INSN_SETS_ARE_DELAYED (@var{insn}) 11450Define this macro as a C expression that is nonzero if it is safe for the 11451delay slot scheduler to place instructions in the delay slot of @var{insn}, 11452even if they appear to use a resource set or clobbered in @var{insn}. 11453@var{insn} is always a @code{jump_insn} or an @code{insn}; GCC knows that 11454every @code{call_insn} has this behavior. On machines where some @code{insn} 11455or @code{jump_insn} is really a function call and hence has this behavior, 11456you should define this macro. 11457 11458You need not define this macro if it would always return zero. 11459@end defmac 11460 11461@defmac INSN_REFERENCES_ARE_DELAYED (@var{insn}) 11462Define this macro as a C expression that is nonzero if it is safe for the 11463delay slot scheduler to place instructions in the delay slot of @var{insn}, 11464even if they appear to set or clobber a resource referenced in @var{insn}. 11465@var{insn} is always a @code{jump_insn} or an @code{insn}. On machines where 11466some @code{insn} or @code{jump_insn} is really a function call and its operands 11467are registers whose use is actually in the subroutine it calls, you should 11468define this macro. Doing so allows the delay slot scheduler to move 11469instructions which copy arguments into the argument registers into the delay 11470slot of @var{insn}. 11471 11472You need not define this macro if it would always return zero. 11473@end defmac 11474 11475@defmac MULTIPLE_SYMBOL_SPACES 11476Define this macro as a C expression that is nonzero if, in some cases, 11477global symbols from one translation unit may not be bound to undefined 11478symbols in another translation unit without user intervention. For 11479instance, under Microsoft Windows symbols must be explicitly imported 11480from shared libraries (DLLs). 11481 11482You need not define this macro if it would always evaluate to zero. 11483@end defmac 11484 11485@deftypefn {Target Hook} {rtx_insn *} TARGET_MD_ASM_ADJUST (vec<rtx>& @var{outputs}, vec<rtx>& @var{inputs}, vec<const char *>& @var{constraints}, vec<rtx>& @var{clobbers}, HARD_REG_SET& @var{clobbered_regs}) 11486This target hook may add @dfn{clobbers} to @var{clobbers} and 11487@var{clobbered_regs} for any hard regs the port wishes to automatically 11488clobber for an asm. The @var{outputs} and @var{inputs} may be inspected 11489to avoid clobbering a register that is already used by the asm. 11490 11491It may modify the @var{outputs}, @var{inputs}, and @var{constraints} 11492as necessary for other pre-processing. In this case the return value is 11493a sequence of insns to emit after the asm. 11494@end deftypefn 11495 11496@defmac MATH_LIBRARY 11497Define this macro as a C string constant for the linker argument to link 11498in the system math library, minus the initial @samp{"-l"}, or 11499@samp{""} if the target does not have a 11500separate math library. 11501 11502You need only define this macro if the default of @samp{"m"} is wrong. 11503@end defmac 11504 11505@defmac LIBRARY_PATH_ENV 11506Define this macro as a C string constant for the environment variable that 11507specifies where the linker should look for libraries. 11508 11509You need only define this macro if the default of @samp{"LIBRARY_PATH"} 11510is wrong. 11511@end defmac 11512 11513@defmac TARGET_POSIX_IO 11514Define this macro if the target supports the following POSIX@ file 11515functions, access, mkdir and file locking with fcntl / F_SETLKW@. 11516Defining @code{TARGET_POSIX_IO} will enable the test coverage code 11517to use file locking when exiting a program, which avoids race conditions 11518if the program has forked. It will also create directories at run-time 11519for cross-profiling. 11520@end defmac 11521 11522@defmac MAX_CONDITIONAL_EXECUTE 11523 11524A C expression for the maximum number of instructions to execute via 11525conditional execution instructions instead of a branch. A value of 11526@code{BRANCH_COST}+1 is the default if the machine does not use cc0, and 115271 if it does use cc0. 11528@end defmac 11529 11530@defmac IFCVT_MODIFY_TESTS (@var{ce_info}, @var{true_expr}, @var{false_expr}) 11531Used if the target needs to perform machine-dependent modifications on the 11532conditionals used for turning basic blocks into conditionally executed code. 11533@var{ce_info} points to a data structure, @code{struct ce_if_block}, which 11534contains information about the currently processed blocks. @var{true_expr} 11535and @var{false_expr} are the tests that are used for converting the 11536then-block and the else-block, respectively. Set either @var{true_expr} or 11537@var{false_expr} to a null pointer if the tests cannot be converted. 11538@end defmac 11539 11540@defmac IFCVT_MODIFY_MULTIPLE_TESTS (@var{ce_info}, @var{bb}, @var{true_expr}, @var{false_expr}) 11541Like @code{IFCVT_MODIFY_TESTS}, but used when converting more complicated 11542if-statements into conditions combined by @code{and} and @code{or} operations. 11543@var{bb} contains the basic block that contains the test that is currently 11544being processed and about to be turned into a condition. 11545@end defmac 11546 11547@defmac IFCVT_MODIFY_INSN (@var{ce_info}, @var{pattern}, @var{insn}) 11548A C expression to modify the @var{PATTERN} of an @var{INSN} that is to 11549be converted to conditional execution format. @var{ce_info} points to 11550a data structure, @code{struct ce_if_block}, which contains information 11551about the currently processed blocks. 11552@end defmac 11553 11554@defmac IFCVT_MODIFY_FINAL (@var{ce_info}) 11555A C expression to perform any final machine dependent modifications in 11556converting code to conditional execution. The involved basic blocks 11557can be found in the @code{struct ce_if_block} structure that is pointed 11558to by @var{ce_info}. 11559@end defmac 11560 11561@defmac IFCVT_MODIFY_CANCEL (@var{ce_info}) 11562A C expression to cancel any machine dependent modifications in 11563converting code to conditional execution. The involved basic blocks 11564can be found in the @code{struct ce_if_block} structure that is pointed 11565to by @var{ce_info}. 11566@end defmac 11567 11568@defmac IFCVT_MACHDEP_INIT (@var{ce_info}) 11569A C expression to initialize any machine specific data for if-conversion 11570of the if-block in the @code{struct ce_if_block} structure that is pointed 11571to by @var{ce_info}. 11572@end defmac 11573 11574@deftypefn {Target Hook} void TARGET_MACHINE_DEPENDENT_REORG (void) 11575If non-null, this hook performs a target-specific pass over the 11576instruction stream. The compiler will run it at all optimization levels, 11577just before the point at which it normally does delayed-branch scheduling. 11578 11579The exact purpose of the hook varies from target to target. Some use 11580it to do transformations that are necessary for correctness, such as 11581laying out in-function constant pools or avoiding hardware hazards. 11582Others use it as an opportunity to do some machine-dependent optimizations. 11583 11584You need not implement the hook if it has nothing to do. The default 11585definition is null. 11586@end deftypefn 11587 11588@deftypefn {Target Hook} void TARGET_INIT_BUILTINS (void) 11589Define this hook if you have any machine-specific built-in functions 11590that need to be defined. It should be a function that performs the 11591necessary setup. 11592 11593Machine specific built-in functions can be useful to expand special machine 11594instructions that would otherwise not normally be generated because 11595they have no equivalent in the source language (for example, SIMD vector 11596instructions or prefetch instructions). 11597 11598To create a built-in function, call the function 11599@code{lang_hooks.builtin_function} 11600which is defined by the language front end. You can use any type nodes set 11601up by @code{build_common_tree_nodes}; 11602only language front ends that use those two functions will call 11603@samp{TARGET_INIT_BUILTINS}. 11604@end deftypefn 11605 11606@deftypefn {Target Hook} tree TARGET_BUILTIN_DECL (unsigned @var{code}, bool @var{initialize_p}) 11607Define this hook if you have any machine-specific built-in functions 11608that need to be defined. It should be a function that returns the 11609builtin function declaration for the builtin function code @var{code}. 11610If there is no such builtin and it cannot be initialized at this time 11611if @var{initialize_p} is true the function should return @code{NULL_TREE}. 11612If @var{code} is out of range the function should return 11613@code{error_mark_node}. 11614@end deftypefn 11615 11616@deftypefn {Target Hook} rtx TARGET_EXPAND_BUILTIN (tree @var{exp}, rtx @var{target}, rtx @var{subtarget}, machine_mode @var{mode}, int @var{ignore}) 11617 11618Expand a call to a machine specific built-in function that was set up by 11619@samp{TARGET_INIT_BUILTINS}. @var{exp} is the expression for the 11620function call; the result should go to @var{target} if that is 11621convenient, and have mode @var{mode} if that is convenient. 11622@var{subtarget} may be used as the target for computing one of 11623@var{exp}'s operands. @var{ignore} is nonzero if the value is to be 11624ignored. This function should return the result of the call to the 11625built-in function. 11626@end deftypefn 11627 11628@deftypefn {Target Hook} tree TARGET_RESOLVE_OVERLOADED_BUILTIN (unsigned int @var{loc}, tree @var{fndecl}, void *@var{arglist}) 11629Select a replacement for a machine specific built-in function that 11630was set up by @samp{TARGET_INIT_BUILTINS}. This is done 11631@emph{before} regular type checking, and so allows the target to 11632implement a crude form of function overloading. @var{fndecl} is the 11633declaration of the built-in function. @var{arglist} is the list of 11634arguments passed to the built-in function. The result is a 11635complete expression that implements the operation, usually 11636another @code{CALL_EXPR}. 11637@var{arglist} really has type @samp{VEC(tree,gc)*} 11638@end deftypefn 11639 11640@deftypefn {Target Hook} bool TARGET_CHECK_BUILTIN_CALL (location_t @var{loc}, vec<location_t> @var{arg_loc}, tree @var{fndecl}, tree @var{orig_fndecl}, unsigned int @var{nargs}, tree *@var{args}) 11641Perform semantic checking on a call to a machine-specific built-in 11642function after its arguments have been constrained to the function 11643signature. Return true if the call is valid, otherwise report an error 11644and return false. 11645 11646This hook is called after @code{TARGET_RESOLVE_OVERLOADED_BUILTIN}. 11647The call was originally to built-in function @var{orig_fndecl}, 11648but after the optional @code{TARGET_RESOLVE_OVERLOADED_BUILTIN} 11649step is now to built-in function @var{fndecl}. @var{loc} is the 11650location of the call and @var{args} is an array of function arguments, 11651of which there are @var{nargs}. @var{arg_loc} specifies the location 11652of each argument. 11653@end deftypefn 11654 11655@deftypefn {Target Hook} tree TARGET_FOLD_BUILTIN (tree @var{fndecl}, int @var{n_args}, tree *@var{argp}, bool @var{ignore}) 11656Fold a call to a machine specific built-in function that was set up by 11657@samp{TARGET_INIT_BUILTINS}. @var{fndecl} is the declaration of the 11658built-in function. @var{n_args} is the number of arguments passed to 11659the function; the arguments themselves are pointed to by @var{argp}. 11660The result is another tree, valid for both GIMPLE and GENERIC, 11661containing a simplified expression for the call's result. If 11662@var{ignore} is true the value will be ignored. 11663@end deftypefn 11664 11665@deftypefn {Target Hook} bool TARGET_GIMPLE_FOLD_BUILTIN (gimple_stmt_iterator *@var{gsi}) 11666Fold a call to a machine specific built-in function that was set up 11667by @samp{TARGET_INIT_BUILTINS}. @var{gsi} points to the gimple 11668statement holding the function call. Returns true if any change 11669was made to the GIMPLE stream. 11670@end deftypefn 11671 11672@deftypefn {Target Hook} int TARGET_COMPARE_VERSION_PRIORITY (tree @var{decl1}, tree @var{decl2}) 11673This hook is used to compare the target attributes in two functions to 11674determine which function's features get higher priority. This is used 11675during function multi-versioning to figure out the order in which two 11676versions must be dispatched. A function version with a higher priority 11677is checked for dispatching earlier. @var{decl1} and @var{decl2} are 11678 the two function decls that will be compared. 11679@end deftypefn 11680 11681@deftypefn {Target Hook} tree TARGET_GET_FUNCTION_VERSIONS_DISPATCHER (void *@var{decl}) 11682This hook is used to get the dispatcher function for a set of function 11683versions. The dispatcher function is called to invoke the right function 11684version at run-time. @var{decl} is one version from a set of semantically 11685identical versions. 11686@end deftypefn 11687 11688@deftypefn {Target Hook} tree TARGET_GENERATE_VERSION_DISPATCHER_BODY (void *@var{arg}) 11689This hook is used to generate the dispatcher logic to invoke the right 11690function version at run-time for a given set of function versions. 11691@var{arg} points to the callgraph node of the dispatcher function whose 11692body must be generated. 11693@end deftypefn 11694 11695@deftypefn {Target Hook} bool TARGET_PREDICT_DOLOOP_P (class loop *@var{loop}) 11696Return true if we can predict it is possible to use a low-overhead loop 11697for a particular loop. The parameter @var{loop} is a pointer to the loop. 11698This target hook is required only when the target supports low-overhead 11699loops, and will help ivopts to make some decisions. 11700The default version of this hook returns false. 11701@end deftypefn 11702 11703@deftypevr {Target Hook} bool TARGET_HAVE_COUNT_REG_DECR_P 11704Return true if the target supports hardware count register for decrement 11705and branch. 11706The default value is false. 11707@end deftypevr 11708 11709@deftypevr {Target Hook} int64_t TARGET_DOLOOP_COST_FOR_GENERIC 11710One IV candidate dedicated for doloop is introduced in IVOPTs, we can 11711calculate the computation cost of adopting it to any generic IV use by 11712function get_computation_cost as before. But for targets which have 11713hardware count register support for decrement and branch, it may have to 11714move IV value from hardware count register to general purpose register 11715while doloop IV candidate is used for generic IV uses. It probably takes 11716expensive penalty. This hook allows target owners to define the cost for 11717this especially for generic IV uses. 11718The default value is zero. 11719@end deftypevr 11720 11721@deftypevr {Target Hook} int64_t TARGET_DOLOOP_COST_FOR_ADDRESS 11722One IV candidate dedicated for doloop is introduced in IVOPTs, we can 11723calculate the computation cost of adopting it to any address IV use by 11724function get_computation_cost as before. But for targets which have 11725hardware count register support for decrement and branch, it may have to 11726move IV value from hardware count register to general purpose register 11727while doloop IV candidate is used for address IV uses. It probably takes 11728expensive penalty. This hook allows target owners to define the cost for 11729this escpecially for address IV uses. 11730The default value is zero. 11731@end deftypevr 11732 11733@deftypefn {Target Hook} bool TARGET_CAN_USE_DOLOOP_P (const widest_int @var{&iterations}, const widest_int @var{&iterations_max}, unsigned int @var{loop_depth}, bool @var{entered_at_top}) 11734Return true if it is possible to use low-overhead loops (@code{doloop_end} 11735and @code{doloop_begin}) for a particular loop. @var{iterations} gives the 11736exact number of iterations, or 0 if not known. @var{iterations_max} gives 11737the maximum number of iterations, or 0 if not known. @var{loop_depth} is 11738the nesting depth of the loop, with 1 for innermost loops, 2 for loops that 11739contain innermost loops, and so on. @var{entered_at_top} is true if the 11740loop is only entered from the top. 11741 11742This hook is only used if @code{doloop_end} is available. The default 11743implementation returns true. You can use @code{can_use_doloop_if_innermost} 11744if the loop must be the innermost, and if there are no other restrictions. 11745@end deftypefn 11746 11747@deftypefn {Target Hook} {const char *} TARGET_INVALID_WITHIN_DOLOOP (const rtx_insn *@var{insn}) 11748 11749Take an instruction in @var{insn} and return NULL if it is valid within a 11750low-overhead loop, otherwise return a string explaining why doloop 11751could not be applied. 11752 11753Many targets use special registers for low-overhead looping. For any 11754instruction that clobbers these this function should return a string indicating 11755the reason why the doloop could not be applied. 11756By default, the RTL loop optimizer does not use a present doloop pattern for 11757loops containing function calls or branch on table instructions. 11758@end deftypefn 11759 11760@deftypefn {Target Hook} bool TARGET_LEGITIMATE_COMBINED_INSN (rtx_insn *@var{insn}) 11761Take an instruction in @var{insn} and return @code{false} if the instruction is not appropriate as a combination of two or more instructions. The default is to accept all instructions. 11762@end deftypefn 11763 11764@deftypefn {Target Hook} bool TARGET_CAN_FOLLOW_JUMP (const rtx_insn *@var{follower}, const rtx_insn *@var{followee}) 11765FOLLOWER and FOLLOWEE are JUMP_INSN instructions; return true if FOLLOWER may be modified to follow FOLLOWEE; false, if it can't. For example, on some targets, certain kinds of branches can't be made to follow through a hot/cold partitioning. 11766@end deftypefn 11767 11768@deftypefn {Target Hook} bool TARGET_COMMUTATIVE_P (const_rtx @var{x}, int @var{outer_code}) 11769This target hook returns @code{true} if @var{x} is considered to be commutative. 11770Usually, this is just COMMUTATIVE_P (@var{x}), but the HP PA doesn't consider 11771PLUS to be commutative inside a MEM@. @var{outer_code} is the rtx code 11772of the enclosing rtl, if known, otherwise it is UNKNOWN. 11773@end deftypefn 11774 11775@deftypefn {Target Hook} rtx TARGET_ALLOCATE_INITIAL_VALUE (rtx @var{hard_reg}) 11776 11777When the initial value of a hard register has been copied in a pseudo 11778register, it is often not necessary to actually allocate another register 11779to this pseudo register, because the original hard register or a stack slot 11780it has been saved into can be used. @code{TARGET_ALLOCATE_INITIAL_VALUE} 11781is called at the start of register allocation once for each hard register 11782that had its initial value copied by using 11783@code{get_func_hard_reg_initial_val} or @code{get_hard_reg_initial_val}. 11784Possible values are @code{NULL_RTX}, if you don't want 11785to do any special allocation, a @code{REG} rtx---that would typically be 11786the hard register itself, if it is known not to be clobbered---or a 11787@code{MEM}. 11788If you are returning a @code{MEM}, this is only a hint for the allocator; 11789it might decide to use another register anyways. 11790You may use @code{current_function_is_leaf} or 11791@code{REG_N_SETS} in the hook to determine if the hard 11792register in question will not be clobbered. 11793The default value of this hook is @code{NULL}, which disables any special 11794allocation. 11795@end deftypefn 11796 11797@deftypefn {Target Hook} int TARGET_UNSPEC_MAY_TRAP_P (const_rtx @var{x}, unsigned @var{flags}) 11798This target hook returns nonzero if @var{x}, an @code{unspec} or 11799@code{unspec_volatile} operation, might cause a trap. Targets can use 11800this hook to enhance precision of analysis for @code{unspec} and 11801@code{unspec_volatile} operations. You may call @code{may_trap_p_1} 11802to analyze inner elements of @var{x} in which case @var{flags} should be 11803passed along. 11804@end deftypefn 11805 11806@deftypefn {Target Hook} int TARGET_BITFIELD_MAY_TRAP_P (const_rtx @var{x}, unsigned @var{flags}) 11807This target hook returns nonzero if @var{x}, an @code{sign_extract} or 11808@code{zero_extract} operation, might cause a trap. Targets can use 11809this hook to enhance precision of analysis for @code{sign_extract} and 11810@code{zero_extract} operations. You may call @code{may_trap_p_1} 11811to analyze inner elements of @var{x} in which case @var{flags} should be 11812passed along. 11813@end deftypefn 11814 11815@deftypefn {Target Hook} void TARGET_SET_CURRENT_FUNCTION (tree @var{decl}) 11816The compiler invokes this hook whenever it changes its current function 11817context (@code{cfun}). You can define this function if 11818the back end needs to perform any initialization or reset actions on a 11819per-function basis. For example, it may be used to implement function 11820attributes that affect register usage or code generation patterns. 11821The argument @var{decl} is the declaration for the new function context, 11822and may be null to indicate that the compiler has left a function context 11823and is returning to processing at the top level. 11824The default hook function does nothing. 11825 11826GCC sets @code{cfun} to a dummy function context during initialization of 11827some parts of the back end. The hook function is not invoked in this 11828situation; you need not worry about the hook being invoked recursively, 11829or when the back end is in a partially-initialized state. 11830@code{cfun} might be @code{NULL} to indicate processing at top level, 11831outside of any function scope. 11832@end deftypefn 11833 11834@defmac TARGET_OBJECT_SUFFIX 11835Define this macro to be a C string representing the suffix for object 11836files on your target machine. If you do not define this macro, GCC will 11837use @samp{.o} as the suffix for object files. 11838@end defmac 11839 11840@defmac TARGET_EXECUTABLE_SUFFIX 11841Define this macro to be a C string representing the suffix to be 11842automatically added to executable files on your target machine. If you 11843do not define this macro, GCC will use the null string as the suffix for 11844executable files. 11845@end defmac 11846 11847@defmac COLLECT_EXPORT_LIST 11848If defined, @code{collect2} will scan the individual object files 11849specified on its command line and create an export list for the linker. 11850Define this macro for systems like AIX, where the linker discards 11851object files that are not referenced from @code{main} and uses export 11852lists. 11853@end defmac 11854 11855@deftypefn {Target Hook} bool TARGET_CANNOT_MODIFY_JUMPS_P (void) 11856This target hook returns @code{true} past the point in which new jump 11857instructions could be created. On machines that require a register for 11858every jump such as the SHmedia ISA of SH5, this point would typically be 11859reload, so this target hook should be defined to a function such as: 11860 11861@smallexample 11862static bool 11863cannot_modify_jumps_past_reload_p () 11864@{ 11865 return (reload_completed || reload_in_progress); 11866@} 11867@end smallexample 11868@end deftypefn 11869 11870@deftypefn {Target Hook} bool TARGET_HAVE_CONDITIONAL_EXECUTION (void) 11871This target hook returns true if the target supports conditional execution. 11872This target hook is required only when the target has several different 11873modes and they have different conditional execution capability, such as ARM. 11874@end deftypefn 11875 11876@deftypefn {Target Hook} rtx TARGET_GEN_CCMP_FIRST (rtx_insn **@var{prep_seq}, rtx_insn **@var{gen_seq}, int @var{code}, tree @var{op0}, tree @var{op1}) 11877This function prepares to emit a comparison insn for the first compare in a 11878 sequence of conditional comparisions. It returns an appropriate comparison 11879 with @code{CC} for passing to @code{gen_ccmp_next} or @code{cbranch_optab}. 11880 The insns to prepare the compare are saved in @var{prep_seq} and the compare 11881 insns are saved in @var{gen_seq}. They will be emitted when all the 11882 compares in the conditional comparision are generated without error. 11883 @var{code} is the @code{rtx_code} of the compare for @var{op0} and @var{op1}. 11884@end deftypefn 11885 11886@deftypefn {Target Hook} rtx TARGET_GEN_CCMP_NEXT (rtx_insn **@var{prep_seq}, rtx_insn **@var{gen_seq}, rtx @var{prev}, int @var{cmp_code}, tree @var{op0}, tree @var{op1}, int @var{bit_code}) 11887This function prepares to emit a conditional comparison within a sequence 11888 of conditional comparisons. It returns an appropriate comparison with 11889 @code{CC} for passing to @code{gen_ccmp_next} or @code{cbranch_optab}. 11890 The insns to prepare the compare are saved in @var{prep_seq} and the compare 11891 insns are saved in @var{gen_seq}. They will be emitted when all the 11892 compares in the conditional comparision are generated without error. The 11893 @var{prev} expression is the result of a prior call to @code{gen_ccmp_first} 11894 or @code{gen_ccmp_next}. It may return @code{NULL} if the combination of 11895 @var{prev} and this comparison is not supported, otherwise the result must 11896 be appropriate for passing to @code{gen_ccmp_next} or @code{cbranch_optab}. 11897 @var{code} is the @code{rtx_code} of the compare for @var{op0} and @var{op1}. 11898 @var{bit_code} is @code{AND} or @code{IOR}, which is the op on the compares. 11899@end deftypefn 11900 11901@deftypefn {Target Hook} unsigned TARGET_LOOP_UNROLL_ADJUST (unsigned @var{nunroll}, class loop *@var{loop}) 11902This target hook returns a new value for the number of times @var{loop} 11903should be unrolled. The parameter @var{nunroll} is the number of times 11904the loop is to be unrolled. The parameter @var{loop} is a pointer to 11905the loop, which is going to be checked for unrolling. This target hook 11906is required only when the target has special constraints like maximum 11907number of memory accesses. 11908@end deftypefn 11909 11910@defmac POWI_MAX_MULTS 11911If defined, this macro is interpreted as a signed integer C expression 11912that specifies the maximum number of floating point multiplications 11913that should be emitted when expanding exponentiation by an integer 11914constant inline. When this value is defined, exponentiation requiring 11915more than this number of multiplications is implemented by calling the 11916system library's @code{pow}, @code{powf} or @code{powl} routines. 11917The default value places no upper bound on the multiplication count. 11918@end defmac 11919 11920@deftypefn Macro void TARGET_EXTRA_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc}) 11921This target hook should register any extra include files for the 11922target. The parameter @var{stdinc} indicates if normal include files 11923are present. The parameter @var{sysroot} is the system root directory. 11924The parameter @var{iprefix} is the prefix for the gcc directory. 11925@end deftypefn 11926 11927@deftypefn Macro void TARGET_EXTRA_PRE_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc}) 11928This target hook should register any extra include files for the 11929target before any standard headers. The parameter @var{stdinc} 11930indicates if normal include files are present. The parameter 11931@var{sysroot} is the system root directory. The parameter 11932@var{iprefix} is the prefix for the gcc directory. 11933@end deftypefn 11934 11935@deftypefn Macro void TARGET_OPTF (char *@var{path}) 11936This target hook should register special include paths for the target. 11937The parameter @var{path} is the include to register. On Darwin 11938systems, this is used for Framework includes, which have semantics 11939that are different from @option{-I}. 11940@end deftypefn 11941 11942@defmac bool TARGET_USE_LOCAL_THUNK_ALIAS_P (tree @var{fndecl}) 11943This target macro returns @code{true} if it is safe to use a local alias 11944for a virtual function @var{fndecl} when constructing thunks, 11945@code{false} otherwise. By default, the macro returns @code{true} for all 11946functions, if a target supports aliases (i.e.@: defines 11947@code{ASM_OUTPUT_DEF}), @code{false} otherwise, 11948@end defmac 11949 11950@defmac TARGET_FORMAT_TYPES 11951If defined, this macro is the name of a global variable containing 11952target-specific format checking information for the @option{-Wformat} 11953option. The default is to have no target-specific format checks. 11954@end defmac 11955 11956@defmac TARGET_N_FORMAT_TYPES 11957If defined, this macro is the number of entries in 11958@code{TARGET_FORMAT_TYPES}. 11959@end defmac 11960 11961@defmac TARGET_OVERRIDES_FORMAT_ATTRIBUTES 11962If defined, this macro is the name of a global variable containing 11963target-specific format overrides for the @option{-Wformat} option. The 11964default is to have no target-specific format overrides. If defined, 11965@code{TARGET_FORMAT_TYPES} must be defined, too. 11966@end defmac 11967 11968@defmac TARGET_OVERRIDES_FORMAT_ATTRIBUTES_COUNT 11969If defined, this macro specifies the number of entries in 11970@code{TARGET_OVERRIDES_FORMAT_ATTRIBUTES}. 11971@end defmac 11972 11973@defmac TARGET_OVERRIDES_FORMAT_INIT 11974If defined, this macro specifies the optional initialization 11975routine for target specific customizations of the system printf 11976and scanf formatter settings. 11977@end defmac 11978 11979@deftypefn {Target Hook} {const char *} TARGET_INVALID_ARG_FOR_UNPROTOTYPED_FN (const_tree @var{typelist}, const_tree @var{funcdecl}, const_tree @var{val}) 11980If defined, this macro returns the diagnostic message when it is 11981illegal to pass argument @var{val} to function @var{funcdecl} 11982with prototype @var{typelist}. 11983@end deftypefn 11984 11985@deftypefn {Target Hook} {const char *} TARGET_INVALID_CONVERSION (const_tree @var{fromtype}, const_tree @var{totype}) 11986If defined, this macro returns the diagnostic message when it is 11987invalid to convert from @var{fromtype} to @var{totype}, or @code{NULL} 11988if validity should be determined by the front end. 11989@end deftypefn 11990 11991@deftypefn {Target Hook} {const char *} TARGET_INVALID_UNARY_OP (int @var{op}, const_tree @var{type}) 11992If defined, this macro returns the diagnostic message when it is 11993invalid to apply operation @var{op} (where unary plus is denoted by 11994@code{CONVERT_EXPR}) to an operand of type @var{type}, or @code{NULL} 11995if validity should be determined by the front end. 11996@end deftypefn 11997 11998@deftypefn {Target Hook} {const char *} TARGET_INVALID_BINARY_OP (int @var{op}, const_tree @var{type1}, const_tree @var{type2}) 11999If defined, this macro returns the diagnostic message when it is 12000invalid to apply operation @var{op} to operands of types @var{type1} 12001and @var{type2}, or @code{NULL} if validity should be determined by 12002the front end. 12003@end deftypefn 12004 12005@deftypefn {Target Hook} tree TARGET_PROMOTED_TYPE (const_tree @var{type}) 12006If defined, this target hook returns the type to which values of 12007@var{type} should be promoted when they appear in expressions, 12008analogous to the integer promotions, or @code{NULL_TREE} to use the 12009front end's normal promotion rules. This hook is useful when there are 12010target-specific types with special promotion rules. 12011This is currently used only by the C and C++ front ends. 12012@end deftypefn 12013 12014@deftypefn {Target Hook} tree TARGET_CONVERT_TO_TYPE (tree @var{type}, tree @var{expr}) 12015If defined, this hook returns the result of converting @var{expr} to 12016@var{type}. It should return the converted expression, 12017or @code{NULL_TREE} to apply the front end's normal conversion rules. 12018This hook is useful when there are target-specific types with special 12019conversion rules. 12020This is currently used only by the C and C++ front ends. 12021@end deftypefn 12022 12023@deftypefn {Target Hook} bool TARGET_VERIFY_TYPE_CONTEXT (location_t @var{loc}, type_context_kind @var{context}, const_tree @var{type}, bool @var{silent_p}) 12024If defined, this hook returns false if there is a target-specific reason 12025why type @var{type} cannot be used in the source language context described 12026by @var{context}. When @var{silent_p} is false, the hook also reports an 12027error against @var{loc} for invalid uses of @var{type}. 12028 12029Calls to this hook should be made through the global function 12030@code{verify_type_context}, which makes the @var{silent_p} parameter 12031default to false and also handles @code{error_mark_node}. 12032 12033The default implementation always returns true. 12034@end deftypefn 12035 12036@defmac OBJC_JBLEN 12037This macro determines the size of the objective C jump buffer for the 12038NeXT runtime. By default, OBJC_JBLEN is defined to an innocuous value. 12039@end defmac 12040 12041@defmac LIBGCC2_UNWIND_ATTRIBUTE 12042Define this macro if any target-specific attributes need to be attached 12043to the functions in @file{libgcc} that provide low-level support for 12044call stack unwinding. It is used in declarations in @file{unwind-generic.h} 12045and the associated definitions of those functions. 12046@end defmac 12047 12048@deftypefn {Target Hook} void TARGET_UPDATE_STACK_BOUNDARY (void) 12049Define this macro to update the current function stack boundary if 12050necessary. 12051@end deftypefn 12052 12053@deftypefn {Target Hook} rtx TARGET_GET_DRAP_RTX (void) 12054This hook should return an rtx for Dynamic Realign Argument Pointer (DRAP) if a 12055different argument pointer register is needed to access the function's 12056argument list due to stack realignment. Return @code{NULL} if no DRAP 12057is needed. 12058@end deftypefn 12059 12060@deftypefn {Target Hook} bool TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS (void) 12061When optimization is disabled, this hook indicates whether or not 12062arguments should be allocated to stack slots. Normally, GCC allocates 12063stacks slots for arguments when not optimizing in order to make 12064debugging easier. However, when a function is declared with 12065@code{__attribute__((naked))}, there is no stack frame, and the compiler 12066cannot safely move arguments from the registers in which they are passed 12067to the stack. Therefore, this hook should return true in general, but 12068false for naked functions. The default implementation always returns true. 12069@end deftypefn 12070 12071@deftypevr {Target Hook} {unsigned HOST_WIDE_INT} TARGET_CONST_ANCHOR 12072On some architectures it can take multiple instructions to synthesize 12073a constant. If there is another constant already in a register that 12074is close enough in value then it is preferable that the new constant 12075is computed from this register using immediate addition or 12076subtraction. We accomplish this through CSE. Besides the value of 12077the constant we also add a lower and an upper constant anchor to the 12078available expressions. These are then queried when encountering new 12079constants. The anchors are computed by rounding the constant up and 12080down to a multiple of the value of @code{TARGET_CONST_ANCHOR}. 12081@code{TARGET_CONST_ANCHOR} should be the maximum positive value 12082accepted by immediate-add plus one. We currently assume that the 12083value of @code{TARGET_CONST_ANCHOR} is a power of 2. For example, on 12084MIPS, where add-immediate takes a 16-bit signed value, 12085@code{TARGET_CONST_ANCHOR} is set to @samp{0x8000}. The default value 12086is zero, which disables this optimization. 12087@end deftypevr 12088 12089@deftypefn {Target Hook} {unsigned HOST_WIDE_INT} TARGET_ASAN_SHADOW_OFFSET (void) 12090Return the offset bitwise ored into shifted address to get corresponding 12091Address Sanitizer shadow memory address. NULL if Address Sanitizer is not 12092supported by the target. 12093@end deftypefn 12094 12095@deftypefn {Target Hook} {unsigned HOST_WIDE_INT} TARGET_MEMMODEL_CHECK (unsigned HOST_WIDE_INT @var{val}) 12096Validate target specific memory model mask bits. When NULL no target specific 12097memory model bits are allowed. 12098@end deftypefn 12099 12100@deftypevr {Target Hook} {unsigned char} TARGET_ATOMIC_TEST_AND_SET_TRUEVAL 12101This value should be set if the result written by @code{atomic_test_and_set} is not exactly 1, i.e.@: the @code{bool} @code{true}. 12102@end deftypevr 12103 12104@deftypefn {Target Hook} bool TARGET_HAS_IFUNC_P (void) 12105It returns true if the target supports GNU indirect functions. 12106The support includes the assembler, linker and dynamic linker. 12107The default value of this hook is based on target's libc. 12108@end deftypefn 12109 12110@deftypefn {Target Hook} {unsigned int} TARGET_ATOMIC_ALIGN_FOR_MODE (machine_mode @var{mode}) 12111If defined, this function returns an appropriate alignment in bits for an atomic object of machine_mode @var{mode}. If 0 is returned then the default alignment for the specified mode is used. 12112@end deftypefn 12113 12114@deftypefn {Target Hook} void TARGET_ATOMIC_ASSIGN_EXPAND_FENV (tree *@var{hold}, tree *@var{clear}, tree *@var{update}) 12115ISO C11 requires atomic compound assignments that may raise floating-point exceptions to raise exceptions corresponding to the arithmetic operation whose result was successfully stored in a compare-and-exchange sequence. This requires code equivalent to calls to @code{feholdexcept}, @code{feclearexcept} and @code{feupdateenv} to be generated at appropriate points in the compare-and-exchange sequence. This hook should set @code{*@var{hold}} to an expression equivalent to the call to @code{feholdexcept}, @code{*@var{clear}} to an expression equivalent to the call to @code{feclearexcept} and @code{*@var{update}} to an expression equivalent to the call to @code{feupdateenv}. The three expressions are @code{NULL_TREE} on entry to the hook and may be left as @code{NULL_TREE} if no code is required in a particular place. The default implementation leaves all three expressions as @code{NULL_TREE}. The @code{__atomic_feraiseexcept} function from @code{libatomic} may be of use as part of the code generated in @code{*@var{update}}. 12116@end deftypefn 12117 12118@deftypefn {Target Hook} void TARGET_RECORD_OFFLOAD_SYMBOL (tree) 12119Used when offloaded functions are seen in the compilation unit and no named 12120sections are available. It is called once for each symbol that must be 12121recorded in the offload function and variable table. 12122@end deftypefn 12123 12124@deftypefn {Target Hook} {char *} TARGET_OFFLOAD_OPTIONS (void) 12125Used when writing out the list of options into an LTO file. It should 12126translate any relevant target-specific options (such as the ABI in use) 12127into one of the @option{-foffload} options that exist as a common interface 12128to express such options. It should return a string containing these options, 12129separated by spaces, which the caller will free. 12130 12131@end deftypefn 12132 12133@defmac TARGET_SUPPORTS_WIDE_INT 12134 12135On older ports, large integers are stored in @code{CONST_DOUBLE} rtl 12136objects. Newer ports define @code{TARGET_SUPPORTS_WIDE_INT} to be nonzero 12137to indicate that large integers are stored in 12138@code{CONST_WIDE_INT} rtl objects. The @code{CONST_WIDE_INT} allows 12139very large integer constants to be represented. @code{CONST_DOUBLE} 12140is limited to twice the size of the host's @code{HOST_WIDE_INT} 12141representation. 12142 12143Converting a port mostly requires looking for the places where 12144@code{CONST_DOUBLE}s are used with @code{VOIDmode} and replacing that 12145code with code that accesses @code{CONST_WIDE_INT}s. @samp{"grep -i 12146const_double"} at the port level gets you to 95% of the changes that 12147need to be made. There are a few places that require a deeper look. 12148 12149@itemize @bullet 12150@item 12151There is no equivalent to @code{hval} and @code{lval} for 12152@code{CONST_WIDE_INT}s. This would be difficult to express in the md 12153language since there are a variable number of elements. 12154 12155Most ports only check that @code{hval} is either 0 or -1 to see if the 12156value is small. As mentioned above, this will no longer be necessary 12157since small constants are always @code{CONST_INT}. Of course there 12158are still a few exceptions, the alpha's constraint used by the zap 12159instruction certainly requires careful examination by C code. 12160However, all the current code does is pass the hval and lval to C 12161code, so evolving the c code to look at the @code{CONST_WIDE_INT} is 12162not really a large change. 12163 12164@item 12165Because there is no standard template that ports use to materialize 12166constants, there is likely to be some futzing that is unique to each 12167port in this code. 12168 12169@item 12170The rtx costs may have to be adjusted to properly account for larger 12171constants that are represented as @code{CONST_WIDE_INT}. 12172@end itemize 12173 12174All and all it does not take long to convert ports that the 12175maintainer is familiar with. 12176 12177@end defmac 12178 12179@deftypefn {Target Hook} bool TARGET_HAVE_SPECULATION_SAFE_VALUE (bool @var{active}) 12180This hook is used to determine the level of target support for 12181 @code{__builtin_speculation_safe_value}. If called with an argument 12182 of false, it returns true if the target has been modified to support 12183 this builtin. If called with an argument of true, it returns true 12184 if the target requires active mitigation execution might be speculative. 12185 12186 The default implementation returns false if the target does not define 12187 a pattern named @code{speculation_barrier}. Else it returns true 12188 for the first case and whether the pattern is enabled for the current 12189 compilation for the second case. 12190 12191 For targets that have no processors that can execute instructions 12192 speculatively an alternative implemenation of this hook is available: 12193 simply redefine this hook to @code{speculation_safe_value_not_needed} 12194 along with your other target hooks. 12195@end deftypefn 12196 12197@deftypefn {Target Hook} rtx TARGET_SPECULATION_SAFE_VALUE (machine_mode @var{mode}, rtx @var{result}, rtx @var{val}, rtx @var{failval}) 12198This target hook can be used to generate a target-specific code 12199 sequence that implements the @code{__builtin_speculation_safe_value} 12200 built-in function. The function must always return @var{val} in 12201 @var{result} in mode @var{mode} when the cpu is not executing 12202 speculatively, but must never return that when speculating until it 12203 is known that the speculation will not be unwound. The hook supports 12204 two primary mechanisms for implementing the requirements. The first 12205 is to emit a speculation barrier which forces the processor to wait 12206 until all prior speculative operations have been resolved; the second 12207 is to use a target-specific mechanism that can track the speculation 12208 state and to return @var{failval} if it can determine that 12209 speculation must be unwound at a later time. 12210 12211 The default implementation simply copies @var{val} to @var{result} and 12212 emits a @code{speculation_barrier} instruction if that is defined. 12213@end deftypefn 12214 12215@deftypefn {Target Hook} void TARGET_RUN_TARGET_SELFTESTS (void) 12216If selftests are enabled, run any selftests for this target. 12217@end deftypefn 12218