tm.texi revision 146895
11638Srgrimes@c Copyright (C) 1988,1989,1992,1993,1994,1995,1996,1997,1998,1999,2000,2001, 21638Srgrimes@c 2002, 2003, 2004 Free Software Foundation, Inc. 31638Srgrimes@c This is part of the GCC manual. 41638Srgrimes@c For copying conditions, see the file gcc.texi. 51638Srgrimes 61638Srgrimes@node Target Macros 71638Srgrimes@chapter Target Description Macros and Functions 81638Srgrimes@cindex machine description macros 91638Srgrimes@cindex target description macros 101638Srgrimes@cindex macros, target description 1112554Sjkh@cindex @file{tm.h} macros 121638Srgrimes 131638SrgrimesIn addition to the file @file{@var{machine}.md}, a machine description 141638Srgrimesincludes a C header file conventionally given the name 1512554Sjkh@file{@var{machine}.h} and a C source file named @file{@var{machine}.c}. 1612554SjkhThe header file defines numerous macros that convey the information 1712554Sjkhabout the target machine that does not fit into the scheme of the 1812554Sjkh@file{.md} file. The file @file{tm.h} should be a link to 191638Srgrimes@file{@var{machine}.h}. The header file @file{config.h} includes 2012554Sjkh@file{tm.h} and most compiler source files include @file{config.h}. The 2112554Sjkhsource file defines a variable @code{targetm}, which is a structure 221638Srgrimescontaining pointers to functions and data relating to the target 2312554Sjkhmachine. @file{@var{machine}.c} should also contain their definitions, 2412554Sjkhif they are not defined elsewhere in GCC, and other functions called 2512554Sjkhthrough the macros defined in the @file{.h} file. 2612554Sjkh 2712554Sjkh@menu 2812554Sjkh* Target Structure:: The @code{targetm} variable. 2912554Sjkh* Driver:: Controlling how the driver runs the compilation passes. 3012554Sjkh* Run-time Target:: Defining @samp{-m} options like @option{-m68000} and @option{-m68020}. 3112554Sjkh* Per-Function Data:: Defining data structures for per-function information. 3212554Sjkh* Storage Layout:: Defining sizes and alignments of data. 3312554Sjkh* Type Layout:: Defining sizes and properties of basic user data types. 3412554Sjkh* Escape Sequences:: Defining the value of target character escape sequences 3512554Sjkh* Registers:: Naming and describing the hardware registers. 361638Srgrimes* Register Classes:: Defining the classes of hardware registers. 37* Stack and Calling:: Defining which way the stack grows and by how much. 38* Varargs:: Defining the varargs macros. 39* Trampolines:: Code set up at run time to enter a nested function. 40* Library Calls:: Controlling how library routines are implicitly called. 41* Addressing Modes:: Defining addressing modes valid for memory operands. 42* Condition Code:: Defining how insns update the condition code. 43* Costs:: Defining relative costs of different operations. 44* Scheduling:: Adjusting the behavior of the instruction scheduler. 45* Sections:: Dividing storage into text, data, and other sections. 46* PIC:: Macros for position independent code. 47* Assembler Format:: Defining how to write insns and pseudo-ops to output. 48* Debugging Info:: Defining the format of debugging output. 49* Floating Point:: Handling floating point for cross-compilers. 50* Mode Switching:: Insertion of mode-switching instructions. 51* Target Attributes:: Defining target-specific uses of @code{__attribute__}. 52* MIPS Coprocessors:: MIPS coprocessor support and how to customize it. 53* PCH Target:: Validity checking for precompiled headers. 54* Misc:: Everything else. 55@end menu 56 57@node Target Structure 58@section The Global @code{targetm} Variable 59@cindex target hooks 60@cindex target functions 61 62@deftypevar {struct gcc_target} targetm 63The target @file{.c} file must define the global @code{targetm} variable 64which contains pointers to functions and data relating to the target 65machine. The variable is declared in @file{target.h}; 66@file{target-def.h} defines the macro @code{TARGET_INITIALIZER} which is 67used to initialize the variable, and macros for the default initializers 68for elements of the structure. The @file{.c} file should override those 69macros for which the default definition is inappropriate. For example: 70@smallexample 71#include "target.h" 72#include "target-def.h" 73 74/* @r{Initialize the GCC target structure.} */ 75 76#undef TARGET_COMP_TYPE_ATTRIBUTES 77#define TARGET_COMP_TYPE_ATTRIBUTES @var{machine}_comp_type_attributes 78 79struct gcc_target targetm = TARGET_INITIALIZER; 80@end smallexample 81@end deftypevar 82 83Where a macro should be defined in the @file{.c} file in this manner to 84form part of the @code{targetm} structure, it is documented below as a 85``Target Hook'' with a prototype. Many macros will change in future 86from being defined in the @file{.h} file to being part of the 87@code{targetm} structure. 88 89@node Driver 90@section Controlling the Compilation Driver, @file{gcc} 91@cindex driver 92@cindex controlling the compilation driver 93 94@c prevent bad page break with this line 95You can control the compilation driver. 96 97@defmac SWITCH_TAKES_ARG (@var{char}) 98A C expression which determines whether the option @option{-@var{char}} 99takes arguments. The value should be the number of arguments that 100option takes--zero, for many options. 101 102By default, this macro is defined as 103@code{DEFAULT_SWITCH_TAKES_ARG}, which handles the standard options 104properly. You need not define @code{SWITCH_TAKES_ARG} unless you 105wish to add additional options which take arguments. Any redefinition 106should call @code{DEFAULT_SWITCH_TAKES_ARG} and then check for 107additional options. 108@end defmac 109 110@defmac WORD_SWITCH_TAKES_ARG (@var{name}) 111A C expression which determines whether the option @option{-@var{name}} 112takes arguments. The value should be the number of arguments that 113option takes--zero, for many options. This macro rather than 114@code{SWITCH_TAKES_ARG} is used for multi-character option names. 115 116By default, this macro is defined as 117@code{DEFAULT_WORD_SWITCH_TAKES_ARG}, which handles the standard options 118properly. You need not define @code{WORD_SWITCH_TAKES_ARG} unless you 119wish to add additional options which take arguments. Any redefinition 120should call @code{DEFAULT_WORD_SWITCH_TAKES_ARG} and then check for 121additional options. 122@end defmac 123 124@defmac SWITCH_CURTAILS_COMPILATION (@var{char}) 125A C expression which determines whether the option @option{-@var{char}} 126stops compilation before the generation of an executable. The value is 127boolean, nonzero if the option does stop an executable from being 128generated, zero otherwise. 129 130By default, this macro is defined as 131@code{DEFAULT_SWITCH_CURTAILS_COMPILATION}, which handles the standard 132options properly. You need not define 133@code{SWITCH_CURTAILS_COMPILATION} unless you wish to add additional 134options which affect the generation of an executable. Any redefinition 135should call @code{DEFAULT_SWITCH_CURTAILS_COMPILATION} and then check 136for additional options. 137@end defmac 138 139@defmac SWITCHES_NEED_SPACES 140A string-valued C expression which enumerates the options for which 141the linker needs a space between the option and its argument. 142 143If this macro is not defined, the default value is @code{""}. 144@end defmac 145 146@defmac TARGET_OPTION_TRANSLATE_TABLE 147If defined, a list of pairs of strings, the first of which is a 148potential command line target to the @file{gcc} driver program, and the 149second of which is a space-separated (tabs and other whitespace are not 150supported) list of options with which to replace the first option. The 151target defining this list is responsible for assuring that the results 152are valid. Replacement options may not be the @code{--opt} style, they 153must be the @code{-opt} style. It is the intention of this macro to 154provide a mechanism for substitution that affects the multilibs chosen, 155such as one option that enables many options, some of which select 156multilibs. Example nonsensical definition, where @code{-malt-abi}, 157@code{-EB}, and @code{-mspoo} cause different multilibs to be chosen: 158 159@smallexample 160#define TARGET_OPTION_TRANSLATE_TABLE \ 161@{ "-fast", "-march=fast-foo -malt-abi -I/usr/fast-foo" @}, \ 162@{ "-compat", "-EB -malign=4 -mspoo" @} 163@end smallexample 164@end defmac 165 166@defmac DRIVER_SELF_SPECS 167A list of specs for the driver itself. It should be a suitable 168initializer for an array of strings, with no surrounding braces. 169 170The driver applies these specs to its own command line between loading 171default @file{specs} files (but not command-line specified ones) and 172choosing the multilib directory or running any subcommands. It 173applies them in the order given, so each spec can depend on the 174options added by earlier ones. It is also possible to remove options 175using @samp{%<@var{option}} in the usual way. 176 177This macro can be useful when a port has several interdependent target 178options. It provides a way of standardizing the command line so 179that the other specs are easier to write. 180 181Do not define this macro if it does not need to do anything. 182@end defmac 183 184@defmac OPTION_DEFAULT_SPECS 185A list of specs used to support configure-time default options (i.e.@: 186@option{--with} options) in the driver. It should be a suitable initializer 187for an array of structures, each containing two strings, without the 188outermost pair of surrounding braces. 189 190The first item in the pair is the name of the default. This must match 191the code in @file{config.gcc} for the target. The second item is a spec 192to apply if a default with this name was specified. The string 193@samp{%(VALUE)} in the spec will be replaced by the value of the default 194everywhere it occurs. 195 196The driver will apply these specs to its own command line between loading 197default @file{specs} files and processing @code{DRIVER_SELF_SPECS}, using 198the same mechanism as @code{DRIVER_SELF_SPECS}. 199 200Do not define this macro if it does not need to do anything. 201@end defmac 202 203@defmac CPP_SPEC 204A C string constant that tells the GCC driver program options to 205pass to CPP@. It can also specify how to translate options you 206give to GCC into options for GCC to pass to the CPP@. 207 208Do not define this macro if it does not need to do anything. 209@end defmac 210 211@defmac CPLUSPLUS_CPP_SPEC 212This macro is just like @code{CPP_SPEC}, but is used for C++, rather 213than C@. If you do not define this macro, then the value of 214@code{CPP_SPEC} (if any) will be used instead. 215@end defmac 216 217@defmac CC1_SPEC 218A C string constant that tells the GCC driver program options to 219pass to @code{cc1}, @code{cc1plus}, @code{f771}, and the other language 220front ends. 221It can also specify how to translate options you give to GCC into options 222for GCC to pass to front ends. 223 224Do not define this macro if it does not need to do anything. 225@end defmac 226 227@defmac CC1PLUS_SPEC 228A C string constant that tells the GCC driver program options to 229pass to @code{cc1plus}. It can also specify how to translate options you 230give to GCC into options for GCC to pass to the @code{cc1plus}. 231 232Do not define this macro if it does not need to do anything. 233Note that everything defined in CC1_SPEC is already passed to 234@code{cc1plus} so there is no need to duplicate the contents of 235CC1_SPEC in CC1PLUS_SPEC@. 236@end defmac 237 238@defmac ASM_SPEC 239A C string constant that tells the GCC driver program options to 240pass to the assembler. It can also specify how to translate options 241you give to GCC into options for GCC to pass to the assembler. 242See the file @file{sun3.h} for an example of this. 243 244Do not define this macro if it does not need to do anything. 245@end defmac 246 247@defmac ASM_FINAL_SPEC 248A C string constant that tells the GCC driver program how to 249run any programs which cleanup after the normal assembler. 250Normally, this is not needed. See the file @file{mips.h} for 251an example of this. 252 253Do not define this macro if it does not need to do anything. 254@end defmac 255 256@defmac AS_NEEDS_DASH_FOR_PIPED_INPUT 257Define this macro, with no value, if the driver should give the assembler 258an argument consisting of a single dash, @option{-}, to instruct it to 259read from its standard input (which will be a pipe connected to the 260output of the compiler proper). This argument is given after any 261@option{-o} option specifying the name of the output file. 262 263If you do not define this macro, the assembler is assumed to read its 264standard input if given no non-option arguments. If your assembler 265cannot read standard input at all, use a @samp{%@{pipe:%e@}} construct; 266see @file{mips.h} for instance. 267@end defmac 268 269@defmac LINK_SPEC 270A C string constant that tells the GCC driver program options to 271pass to the linker. It can also specify how to translate options you 272give to GCC into options for GCC to pass to the linker. 273 274Do not define this macro if it does not need to do anything. 275@end defmac 276 277@defmac LIB_SPEC 278Another C string constant used much like @code{LINK_SPEC}. The difference 279between the two is that @code{LIB_SPEC} is used at the end of the 280command given to the linker. 281 282If this macro is not defined, a default is provided that 283loads the standard C library from the usual place. See @file{gcc.c}. 284@end defmac 285 286@defmac LIBGCC_SPEC 287Another C string constant that tells the GCC driver program 288how and when to place a reference to @file{libgcc.a} into the 289linker command line. This constant is placed both before and after 290the value of @code{LIB_SPEC}. 291 292If this macro is not defined, the GCC driver provides a default that 293passes the string @option{-lgcc} to the linker. 294 295By default, if @code{ENABLE_SHARED_LIBGCC} is defined, the 296@code{LIBGCC_SPEC} is not directly used by the driver program but is 297instead modified to refer to different versions of @file{libgcc.a} 298depending on the values of the command line flags @option{-static}, 299@option{-shared}, @option{-static-libgcc}, and @option{-shared-libgcc}. 300@end defmac 301 302@defmac USE_LD_AS_NEEDED 303A macro that controls the modifications to @code{LIBGCC_SPEC}. 304If nonzero, a spec will be 305generated that uses --as-needed and the shared libgcc in place of the 306static exception handler library, when linking without any of 307@code{-static}, @code{-static-libgcc}, or @code{-shared-libgcc}. 308@end defmac 309 310@defmac LINK_EH_SPEC 311If defined, this C string constant is added to @code{LINK_SPEC}. 312When @code{USE_LD_AS_NEEDED} is zero or undefined, it also affects 313the modifications to @code{LIBGCC_SPEC}. 314@end defmac 315 316@defmac STARTFILE_SPEC 317Another C string constant used much like @code{LINK_SPEC}. The 318difference between the two is that @code{STARTFILE_SPEC} is used at 319the very beginning of the command given to the linker. 320 321If this macro is not defined, a default is provided that loads the 322standard C startup file from the usual place. See @file{gcc.c}. 323@end defmac 324 325@defmac ENDFILE_SPEC 326Another C string constant used much like @code{LINK_SPEC}. The 327difference between the two is that @code{ENDFILE_SPEC} is used at 328the very end of the command given to the linker. 329 330Do not define this macro if it does not need to do anything. 331@end defmac 332 333@defmac THREAD_MODEL_SPEC 334GCC @code{-v} will print the thread model GCC was configured to use. 335However, this doesn't work on platforms that are multilibbed on thread 336models, such as AIX 4.3. On such platforms, define 337@code{THREAD_MODEL_SPEC} such that it evaluates to a string without 338blanks that names one of the recognized thread models. @code{%*}, the 339default value of this macro, will expand to the value of 340@code{thread_file} set in @file{config.gcc}. 341@end defmac 342 343@defmac SYSROOT_SUFFIX_SPEC 344Define this macro to add a suffix to the target sysroot when GCC is 345configured with a sysroot. This will cause GCC to search for usr/lib, 346et al, within sysroot+suffix. 347@end defmac 348 349@defmac SYSROOT_HEADERS_SUFFIX_SPEC 350Define this macro to add a headers_suffix to the target sysroot when 351GCC is configured with a sysroot. This will cause GCC to pass the 352updated sysroot+headers_suffix to CPP, causing it to search for 353usr/include, et al, within sysroot+headers_suffix. 354@end defmac 355 356@defmac EXTRA_SPECS 357Define this macro to provide additional specifications to put in the 358@file{specs} file that can be used in various specifications like 359@code{CC1_SPEC}. 360 361The definition should be an initializer for an array of structures, 362containing a string constant, that defines the specification name, and a 363string constant that provides the specification. 364 365Do not define this macro if it does not need to do anything. 366 367@code{EXTRA_SPECS} is useful when an architecture contains several 368related targets, which have various @code{@dots{}_SPECS} which are similar 369to each other, and the maintainer would like one central place to keep 370these definitions. 371 372For example, the PowerPC System V.4 targets use @code{EXTRA_SPECS} to 373define either @code{_CALL_SYSV} when the System V calling sequence is 374used or @code{_CALL_AIX} when the older AIX-based calling sequence is 375used. 376 377The @file{config/rs6000/rs6000.h} target file defines: 378 379@smallexample 380#define EXTRA_SPECS \ 381 @{ "cpp_sysv_default", CPP_SYSV_DEFAULT @}, 382 383#define CPP_SYS_DEFAULT "" 384@end smallexample 385 386The @file{config/rs6000/sysv.h} target file defines: 387@smallexample 388#undef CPP_SPEC 389#define CPP_SPEC \ 390"%@{posix: -D_POSIX_SOURCE @} \ 391%@{mcall-sysv: -D_CALL_SYSV @} \ 392%@{!mcall-sysv: %(cpp_sysv_default) @} \ 393%@{msoft-float: -D_SOFT_FLOAT@} %@{mcpu=403: -D_SOFT_FLOAT@}" 394 395#undef CPP_SYSV_DEFAULT 396#define CPP_SYSV_DEFAULT "-D_CALL_SYSV" 397@end smallexample 398 399while the @file{config/rs6000/eabiaix.h} target file defines 400@code{CPP_SYSV_DEFAULT} as: 401 402@smallexample 403#undef CPP_SYSV_DEFAULT 404#define CPP_SYSV_DEFAULT "-D_CALL_AIX" 405@end smallexample 406@end defmac 407 408@defmac LINK_LIBGCC_SPECIAL 409Define this macro if the driver program should find the library 410@file{libgcc.a} itself and should not pass @option{-L} options to the 411linker. If you do not define this macro, the driver program will pass 412the argument @option{-lgcc} to tell the linker to do the search and will 413pass @option{-L} options to it. 414@end defmac 415 416@defmac LINK_LIBGCC_SPECIAL_1 417Define this macro if the driver program should find the library 418@file{libgcc.a}. If you do not define this macro, the driver program will pass 419the argument @option{-lgcc} to tell the linker to do the search. 420This macro is similar to @code{LINK_LIBGCC_SPECIAL}, except that it does 421not affect @option{-L} options. 422@end defmac 423 424@defmac LINK_GCC_C_SEQUENCE_SPEC 425The sequence in which libgcc and libc are specified to the linker. 426By default this is @code{%G %L %G}. 427@end defmac 428 429@defmac LINK_COMMAND_SPEC 430A C string constant giving the complete command line need to execute the 431linker. When you do this, you will need to update your port each time a 432change is made to the link command line within @file{gcc.c}. Therefore, 433define this macro only if you need to completely redefine the command 434line for invoking the linker and there is no other way to accomplish 435the effect you need. Overriding this macro may be avoidable by overriding 436@code{LINK_GCC_C_SEQUENCE_SPEC} instead. 437@end defmac 438 439@defmac LINK_ELIMINATE_DUPLICATE_LDIRECTORIES 440A nonzero value causes @command{collect2} to remove duplicate @option{-L@var{directory}} search 441directories from linking commands. Do not give it a nonzero value if 442removing duplicate search directories changes the linker's semantics. 443@end defmac 444 445@defmac MULTILIB_DEFAULTS 446Define this macro as a C expression for the initializer of an array of 447string to tell the driver program which options are defaults for this 448target and thus do not need to be handled specially when using 449@code{MULTILIB_OPTIONS}. 450 451Do not define this macro if @code{MULTILIB_OPTIONS} is not defined in 452the target makefile fragment or if none of the options listed in 453@code{MULTILIB_OPTIONS} are set by default. 454@xref{Target Fragment}. 455@end defmac 456 457@defmac RELATIVE_PREFIX_NOT_LINKDIR 458Define this macro to tell @command{gcc} that it should only translate 459a @option{-B} prefix into a @option{-L} linker option if the prefix 460indicates an absolute file name. 461@end defmac 462 463@defmac MD_EXEC_PREFIX 464If defined, this macro is an additional prefix to try after 465@code{STANDARD_EXEC_PREFIX}. @code{MD_EXEC_PREFIX} is not searched 466when the @option{-b} option is used, or the compiler is built as a cross 467compiler. If you define @code{MD_EXEC_PREFIX}, then be sure to add it 468to the list of directories used to find the assembler in @file{configure.in}. 469@end defmac 470 471@defmac STANDARD_STARTFILE_PREFIX 472Define this macro as a C string constant if you wish to override the 473standard choice of @code{libdir} as the default prefix to 474try when searching for startup files such as @file{crt0.o}. 475@code{STANDARD_STARTFILE_PREFIX} is not searched when the compiler 476is built as a cross compiler. 477@end defmac 478 479@defmac MD_STARTFILE_PREFIX 480If defined, this macro supplies an additional prefix to try after the 481standard prefixes. @code{MD_EXEC_PREFIX} is not searched when the 482@option{-b} option is used, or when the compiler is built as a cross 483compiler. 484@end defmac 485 486@defmac MD_STARTFILE_PREFIX_1 487If defined, this macro supplies yet another prefix to try after the 488standard prefixes. It is not searched when the @option{-b} option is 489used, or when the compiler is built as a cross compiler. 490@end defmac 491 492@defmac INIT_ENVIRONMENT 493Define this macro as a C string constant if you wish to set environment 494variables for programs called by the driver, such as the assembler and 495loader. The driver passes the value of this macro to @code{putenv} to 496initialize the necessary environment variables. 497@end defmac 498 499@defmac LOCAL_INCLUDE_DIR 500Define this macro as a C string constant if you wish to override the 501standard choice of @file{/usr/local/include} as the default prefix to 502try when searching for local header files. @code{LOCAL_INCLUDE_DIR} 503comes before @code{SYSTEM_INCLUDE_DIR} in the search order. 504 505Cross compilers do not search either @file{/usr/local/include} or its 506replacement. 507@end defmac 508 509@defmac MODIFY_TARGET_NAME 510Define this macro if you wish to define command-line switches that 511modify the default target name. 512 513For each switch, you can include a string to be appended to the first 514part of the configuration name or a string to be deleted from the 515configuration name, if present. The definition should be an initializer 516for an array of structures. Each array element should have three 517elements: the switch name (a string constant, including the initial 518dash), one of the enumeration codes @code{ADD} or @code{DELETE} to 519indicate whether the string should be inserted or deleted, and the string 520to be inserted or deleted (a string constant). 521 522For example, on a machine where @samp{64} at the end of the 523configuration name denotes a 64-bit target and you want the @option{-32} 524and @option{-64} switches to select between 32- and 64-bit targets, you would 525code 526 527@smallexample 528#define MODIFY_TARGET_NAME \ 529 @{ @{ "-32", DELETE, "64"@}, \ 530 @{"-64", ADD, "64"@}@} 531@end smallexample 532@end defmac 533 534@defmac SYSTEM_INCLUDE_DIR 535Define this macro as a C string constant if you wish to specify a 536system-specific directory to search for header files before the standard 537directory. @code{SYSTEM_INCLUDE_DIR} comes before 538@code{STANDARD_INCLUDE_DIR} in the search order. 539 540Cross compilers do not use this macro and do not search the directory 541specified. 542@end defmac 543 544@defmac STANDARD_INCLUDE_DIR 545Define this macro as a C string constant if you wish to override the 546standard choice of @file{/usr/include} as the default prefix to 547try when searching for header files. 548 549Cross compilers ignore this macro and do not search either 550@file{/usr/include} or its replacement. 551@end defmac 552 553@defmac STANDARD_INCLUDE_COMPONENT 554The ``component'' corresponding to @code{STANDARD_INCLUDE_DIR}. 555See @code{INCLUDE_DEFAULTS}, below, for the description of components. 556If you do not define this macro, no component is used. 557@end defmac 558 559@defmac INCLUDE_DEFAULTS 560Define this macro if you wish to override the entire default search path 561for include files. For a native compiler, the default search path 562usually consists of @code{GCC_INCLUDE_DIR}, @code{LOCAL_INCLUDE_DIR}, 563@code{SYSTEM_INCLUDE_DIR}, @code{GPLUSPLUS_INCLUDE_DIR}, and 564@code{STANDARD_INCLUDE_DIR}. In addition, @code{GPLUSPLUS_INCLUDE_DIR} 565and @code{GCC_INCLUDE_DIR} are defined automatically by @file{Makefile}, 566and specify private search areas for GCC@. The directory 567@code{GPLUSPLUS_INCLUDE_DIR} is used only for C++ programs. 568 569The definition should be an initializer for an array of structures. 570Each array element should have four elements: the directory name (a 571string constant), the component name (also a string constant), a flag 572for C++-only directories, 573and a flag showing that the includes in the directory don't need to be 574wrapped in @code{extern @samp{C}} when compiling C++. Mark the end of 575the array with a null element. 576 577The component name denotes what GNU package the include file is part of, 578if any, in all uppercase letters. For example, it might be @samp{GCC} 579or @samp{BINUTILS}. If the package is part of a vendor-supplied 580operating system, code the component name as @samp{0}. 581 582For example, here is the definition used for VAX/VMS: 583 584@smallexample 585#define INCLUDE_DEFAULTS \ 586@{ \ 587 @{ "GNU_GXX_INCLUDE:", "G++", 1, 1@}, \ 588 @{ "GNU_CC_INCLUDE:", "GCC", 0, 0@}, \ 589 @{ "SYS$SYSROOT:[SYSLIB.]", 0, 0, 0@}, \ 590 @{ ".", 0, 0, 0@}, \ 591 @{ 0, 0, 0, 0@} \ 592@} 593@end smallexample 594@end defmac 595 596Here is the order of prefixes tried for exec files: 597 598@enumerate 599@item 600Any prefixes specified by the user with @option{-B}. 601 602@item 603The environment variable @code{GCC_EXEC_PREFIX}, if any. 604 605@item 606The directories specified by the environment variable @code{COMPILER_PATH}. 607 608@item 609The macro @code{STANDARD_EXEC_PREFIX}. 610 611@item 612@file{/usr/lib/gcc/}. 613 614@item 615The macro @code{MD_EXEC_PREFIX}, if any. 616@end enumerate 617 618Here is the order of prefixes tried for startfiles: 619 620@enumerate 621@item 622Any prefixes specified by the user with @option{-B}. 623 624@item 625The environment variable @code{GCC_EXEC_PREFIX}, if any. 626 627@item 628The directories specified by the environment variable @code{LIBRARY_PATH} 629(or port-specific name; native only, cross compilers do not use this). 630 631@item 632The macro @code{STANDARD_EXEC_PREFIX}. 633 634@item 635@file{/usr/lib/gcc/}. 636 637@item 638The macro @code{MD_EXEC_PREFIX}, if any. 639 640@item 641The macro @code{MD_STARTFILE_PREFIX}, if any. 642 643@item 644The macro @code{STANDARD_STARTFILE_PREFIX}. 645 646@item 647@file{/lib/}. 648 649@item 650@file{/usr/lib/}. 651@end enumerate 652 653@node Run-time Target 654@section Run-time Target Specification 655@cindex run-time target specification 656@cindex predefined macros 657@cindex target specifications 658 659@c prevent bad page break with this line 660Here are run-time target specifications. 661 662@defmac TARGET_CPU_CPP_BUILTINS () 663This function-like macro expands to a block of code that defines 664built-in preprocessor macros and assertions for the target cpu, using 665the functions @code{builtin_define}, @code{builtin_define_std} and 666@code{builtin_assert}. When the front end 667calls this macro it provides a trailing semicolon, and since it has 668finished command line option processing your code can use those 669results freely. 670 671@code{builtin_assert} takes a string in the form you pass to the 672command-line option @option{-A}, such as @code{cpu=mips}, and creates 673the assertion. @code{builtin_define} takes a string in the form 674accepted by option @option{-D} and unconditionally defines the macro. 675 676@code{builtin_define_std} takes a string representing the name of an 677object-like macro. If it doesn't lie in the user's namespace, 678@code{builtin_define_std} defines it unconditionally. Otherwise, it 679defines a version with two leading underscores, and another version 680with two leading and trailing underscores, and defines the original 681only if an ISO standard was not requested on the command line. For 682example, passing @code{unix} defines @code{__unix}, @code{__unix__} 683and possibly @code{unix}; passing @code{_mips} defines @code{__mips}, 684@code{__mips__} and possibly @code{_mips}, and passing @code{_ABI64} 685defines only @code{_ABI64}. 686 687You can also test for the C dialect being compiled. The variable 688@code{c_language} is set to one of @code{clk_c}, @code{clk_cplusplus} 689or @code{clk_objective_c}. Note that if we are preprocessing 690assembler, this variable will be @code{clk_c} but the function-like 691macro @code{preprocessing_asm_p()} will return true, so you might want 692to check for that first. If you need to check for strict ANSI, the 693variable @code{flag_iso} can be used. The function-like macro 694@code{preprocessing_trad_p()} can be used to check for traditional 695preprocessing. 696@end defmac 697 698@defmac TARGET_OS_CPP_BUILTINS () 699Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional 700and is used for the target operating system instead. 701@end defmac 702 703@defmac TARGET_OBJFMT_CPP_BUILTINS () 704Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional 705and is used for the target object format. @file{elfos.h} uses this 706macro to define @code{__ELF__}, so you probably do not need to define 707it yourself. 708@end defmac 709 710@deftypevar {extern int} target_flags 711This declaration should be present. 712@end deftypevar 713 714@cindex optional hardware or system features 715@cindex features, optional, in system conventions 716 717@defmac TARGET_@var{featurename} 718This series of macros is to allow compiler command arguments to 719enable or disable the use of optional features of the target machine. 720For example, one machine description serves both the 68000 and 721the 68020; a command argument tells the compiler whether it should 722use 68020-only instructions or not. This command argument works 723by means of a macro @code{TARGET_68020} that tests a bit in 724@code{target_flags}. 725 726Define a macro @code{TARGET_@var{featurename}} for each such option. 727Its definition should test a bit in @code{target_flags}. It is 728recommended that a helper macro @code{MASK_@var{featurename}} 729is defined for each bit-value to test, and used in 730@code{TARGET_@var{featurename}} and @code{TARGET_SWITCHES}. For 731example: 732 733@smallexample 734#define TARGET_MASK_68020 1 735#define TARGET_68020 (target_flags & MASK_68020) 736@end smallexample 737 738One place where these macros are used is in the condition-expressions 739of instruction patterns. Note how @code{TARGET_68020} appears 740frequently in the 68000 machine description file, @file{m68k.md}. 741Another place they are used is in the definitions of the other 742macros in the @file{@var{machine}.h} file. 743@end defmac 744 745@defmac TARGET_SWITCHES 746This macro defines names of command options to set and clear 747bits in @code{target_flags}. Its definition is an initializer 748with a subgrouping for each command option. 749 750Each subgrouping contains a string constant, that defines the option 751name, a number, which contains the bits to set in 752@code{target_flags}, and a second string which is the description 753displayed by @option{--help}. If the number is negative then the bits specified 754by the number are cleared instead of being set. If the description 755string is present but empty, then no help information will be displayed 756for that option, but it will not count as an undocumented option. The 757actual option name is made by appending @samp{-m} to the specified name. 758Non-empty description strings should be marked with @code{N_(@dots{})} for 759@command{xgettext}. Please do not mark empty strings because the empty 760string is reserved by GNU gettext. @code{gettext("")} returns the header entry 761of the message catalog with meta information, not the empty string. 762 763In addition to the description for @option{--help}, 764more detailed documentation for each option should be added to 765@file{invoke.texi}. 766 767One of the subgroupings should have a null string. The number in 768this grouping is the default value for @code{target_flags}. Any 769target options act starting with that value. 770 771Here is an example which defines @option{-m68000} and @option{-m68020} 772with opposite meanings, and picks the latter as the default: 773 774@smallexample 775#define TARGET_SWITCHES \ 776 @{ @{ "68020", MASK_68020, "" @}, \ 777 @{ "68000", -MASK_68020, \ 778 N_("Compile for the 68000") @}, \ 779 @{ "", MASK_68020, "" @}, \ 780 @} 781@end smallexample 782@end defmac 783 784@defmac TARGET_OPTIONS 785This macro is similar to @code{TARGET_SWITCHES} but defines names of command 786options that have values. Its definition is an initializer with a 787subgrouping for each command option. 788 789Each subgrouping contains a string constant, that defines the option 790name, the address of a variable, a description string, and a value. 791Non-empty description strings should be marked with @code{N_(@dots{})} 792for @command{xgettext}. Please do not mark empty strings because the 793empty string is reserved by GNU gettext. @code{gettext("")} returns the 794header entry of the message catalog with meta information, not the empty 795string. 796 797If the value listed in the table is @code{NULL}, then the variable, type 798@code{char *}, is set to the variable part of the given option if the 799fixed part matches. In other words, if the first part of the option 800matches what's in the table, the variable will be set to point to the 801rest of the option. This allows the user to specify a value for that 802option. The actual option name is made by appending @samp{-m} to the 803specified name. Again, each option should also be documented in 804@file{invoke.texi}. 805 806If the value listed in the table is non-@code{NULL}, then the option 807must match the option in the table exactly (with @samp{-m}), and the 808variable is set to point to the value listed in the table. 809 810Here is an example which defines @option{-mshort-data-@var{number}}. If the 811given option is @option{-mshort-data-512}, the variable @code{m88k_short_data} 812will be set to the string @code{"512"}. 813 814@smallexample 815extern char *m88k_short_data; 816#define TARGET_OPTIONS \ 817 @{ @{ "short-data-", &m88k_short_data, \ 818 N_("Specify the size of the short data section"), 0 @} @} 819@end smallexample 820 821Here is a variant of the above that allows the user to also specify 822just @option{-mshort-data} where a default of @code{"64"} is used. 823 824@smallexample 825extern char *m88k_short_data; 826#define TARGET_OPTIONS \ 827 @{ @{ "short-data-", &m88k_short_data, \ 828 N_("Specify the size of the short data section"), 0 @} \ 829 @{ "short-data", &m88k_short_data, "", "64" @}, 830 @} 831@end smallexample 832 833Here is an example which defines @option{-mno-alu}, @option{-malu1}, and 834@option{-malu2} as a three-state switch, along with suitable macros for 835checking the state of the option (documentation is elided for brevity). 836 837@smallexample 838[chip.c] 839char *chip_alu = ""; /* Specify default here. */ 840 841[chip.h] 842extern char *chip_alu; 843#define TARGET_OPTIONS \ 844 @{ @{ "no-alu", &chip_alu, "", "" @}, \ 845 @{ "alu1", &chip_alu, "", "1" @}, \ 846 @{ "alu2", &chip_alu, "", "2" @}, @} 847#define TARGET_ALU (chip_alu[0] != '\0') 848#define TARGET_ALU1 (chip_alu[0] == '1') 849#define TARGET_ALU2 (chip_alu[0] == '2') 850@end smallexample 851@end defmac 852 853@defmac TARGET_VERSION 854This macro is a C statement to print on @code{stderr} a string 855describing the particular machine description choice. Every machine 856description should define @code{TARGET_VERSION}. For example: 857 858@smallexample 859#ifdef MOTOROLA 860#define TARGET_VERSION \ 861 fprintf (stderr, " (68k, Motorola syntax)"); 862#else 863#define TARGET_VERSION \ 864 fprintf (stderr, " (68k, MIT syntax)"); 865#endif 866@end smallexample 867@end defmac 868 869@defmac OVERRIDE_OPTIONS 870Sometimes certain combinations of command options do not make sense on 871a particular target machine. You can define a macro 872@code{OVERRIDE_OPTIONS} to take account of this. This macro, if 873defined, is executed once just after all the command options have been 874parsed. 875 876Don't use this macro to turn on various extra optimizations for 877@option{-O}. That is what @code{OPTIMIZATION_OPTIONS} is for. 878@end defmac 879 880@defmac OPTIMIZATION_OPTIONS (@var{level}, @var{size}) 881Some machines may desire to change what optimizations are performed for 882various optimization levels. This macro, if defined, is executed once 883just after the optimization level is determined and before the remainder 884of the command options have been parsed. Values set in this macro are 885used as the default values for the other command line options. 886 887@var{level} is the optimization level specified; 2 if @option{-O2} is 888specified, 1 if @option{-O} is specified, and 0 if neither is specified. 889 890@var{size} is nonzero if @option{-Os} is specified and zero otherwise. 891 892You should not use this macro to change options that are not 893machine-specific. These should uniformly selected by the same 894optimization level on all supported machines. Use this macro to enable 895machine-specific optimizations. 896 897@strong{Do not examine @code{write_symbols} in 898this macro!} The debugging options are not supposed to alter the 899generated code. 900@end defmac 901 902@defmac CAN_DEBUG_WITHOUT_FP 903Define this macro if debugging can be performed even without a frame 904pointer. If this macro is defined, GCC will turn on the 905@option{-fomit-frame-pointer} option whenever @option{-O} is specified. 906@end defmac 907 908@node Per-Function Data 909@section Defining data structures for per-function information. 910@cindex per-function data 911@cindex data structures 912 913If the target needs to store information on a per-function basis, GCC 914provides a macro and a couple of variables to allow this. Note, just 915using statics to store the information is a bad idea, since GCC supports 916nested functions, so you can be halfway through encoding one function 917when another one comes along. 918 919GCC defines a data structure called @code{struct function} which 920contains all of the data specific to an individual function. This 921structure contains a field called @code{machine} whose type is 922@code{struct machine_function *}, which can be used by targets to point 923to their own specific data. 924 925If a target needs per-function specific data it should define the type 926@code{struct machine_function} and also the macro @code{INIT_EXPANDERS}. 927This macro should be used to initialize the function pointer 928@code{init_machine_status}. This pointer is explained below. 929 930One typical use of per-function, target specific data is to create an 931RTX to hold the register containing the function's return address. This 932RTX can then be used to implement the @code{__builtin_return_address} 933function, for level 0. 934 935Note---earlier implementations of GCC used a single data area to hold 936all of the per-function information. Thus when processing of a nested 937function began the old per-function data had to be pushed onto a 938stack, and when the processing was finished, it had to be popped off the 939stack. GCC used to provide function pointers called 940@code{save_machine_status} and @code{restore_machine_status} to handle 941the saving and restoring of the target specific information. Since the 942single data area approach is no longer used, these pointers are no 943longer supported. 944 945@defmac INIT_EXPANDERS 946Macro called to initialize any target specific information. This macro 947is called once per function, before generation of any RTL has begun. 948The intention of this macro is to allow the initialization of the 949function pointer @code{init_machine_status}. 950@end defmac 951 952@deftypevar {void (*)(struct function *)} init_machine_status 953If this function pointer is non-@code{NULL} it will be called once per 954function, before function compilation starts, in order to allow the 955target to perform any target specific initialization of the 956@code{struct function} structure. It is intended that this would be 957used to initialize the @code{machine} of that structure. 958 959@code{struct machine_function} structures are expected to be freed by GC. 960Generally, any memory that they reference must be allocated by using 961@code{ggc_alloc}, including the structure itself. 962@end deftypevar 963 964@node Storage Layout 965@section Storage Layout 966@cindex storage layout 967 968Note that the definitions of the macros in this table which are sizes or 969alignments measured in bits do not need to be constant. They can be C 970expressions that refer to static variables, such as the @code{target_flags}. 971@xref{Run-time Target}. 972 973@defmac BITS_BIG_ENDIAN 974Define this macro to have the value 1 if the most significant bit in a 975byte has the lowest number; otherwise define it to have the value zero. 976This means that bit-field instructions count from the most significant 977bit. If the machine has no bit-field instructions, then this must still 978be defined, but it doesn't matter which value it is defined to. This 979macro need not be a constant. 980 981This macro does not affect the way structure fields are packed into 982bytes or words; that is controlled by @code{BYTES_BIG_ENDIAN}. 983@end defmac 984 985@defmac BYTES_BIG_ENDIAN 986Define this macro to have the value 1 if the most significant byte in a 987word has the lowest number. This macro need not be a constant. 988@end defmac 989 990@defmac WORDS_BIG_ENDIAN 991Define this macro to have the value 1 if, in a multiword object, the 992most significant word has the lowest number. This applies to both 993memory locations and registers; GCC fundamentally assumes that the 994order of words in memory is the same as the order in registers. This 995macro need not be a constant. 996@end defmac 997 998@defmac LIBGCC2_WORDS_BIG_ENDIAN 999Define this macro if @code{WORDS_BIG_ENDIAN} is not constant. This must be a 1000constant value with the same meaning as @code{WORDS_BIG_ENDIAN}, which will be 1001used only when compiling @file{libgcc2.c}. Typically the value will be set 1002based on preprocessor defines. 1003@end defmac 1004 1005@defmac FLOAT_WORDS_BIG_ENDIAN 1006Define this macro to have the value 1 if @code{DFmode}, @code{XFmode} or 1007@code{TFmode} floating point numbers are stored in memory with the word 1008containing the sign bit at the lowest address; otherwise define it to 1009have the value 0. This macro need not be a constant. 1010 1011You need not define this macro if the ordering is the same as for 1012multi-word integers. 1013@end defmac 1014 1015@defmac BITS_PER_UNIT 1016Define this macro to be the number of bits in an addressable storage 1017unit (byte). If you do not define this macro the default is 8. 1018@end defmac 1019 1020@defmac BITS_PER_WORD 1021Number of bits in a word. If you do not define this macro, the default 1022is @code{BITS_PER_UNIT * UNITS_PER_WORD}. 1023@end defmac 1024 1025@defmac MAX_BITS_PER_WORD 1026Maximum number of bits in a word. If this is undefined, the default is 1027@code{BITS_PER_WORD}. Otherwise, it is the constant value that is the 1028largest value that @code{BITS_PER_WORD} can have at run-time. 1029@end defmac 1030 1031@defmac UNITS_PER_WORD 1032Number of storage units in a word; normally 4. 1033@end defmac 1034 1035@defmac MIN_UNITS_PER_WORD 1036Minimum number of units in a word. If this is undefined, the default is 1037@code{UNITS_PER_WORD}. Otherwise, it is the constant value that is the 1038smallest value that @code{UNITS_PER_WORD} can have at run-time. 1039@end defmac 1040 1041@defmac POINTER_SIZE 1042Width of a pointer, in bits. You must specify a value no wider than the 1043width of @code{Pmode}. If it is not equal to the width of @code{Pmode}, 1044you must define @code{POINTERS_EXTEND_UNSIGNED}. If you do not specify 1045a value the default is @code{BITS_PER_WORD}. 1046@end defmac 1047 1048@defmac POINTERS_EXTEND_UNSIGNED 1049A C expression whose value is greater than zero if pointers that need to be 1050extended from being @code{POINTER_SIZE} bits wide to @code{Pmode} are to 1051be zero-extended and zero if they are to be sign-extended. If the value 1052is less then zero then there must be an "ptr_extend" instruction that 1053extends a pointer from @code{POINTER_SIZE} to @code{Pmode}. 1054 1055You need not define this macro if the @code{POINTER_SIZE} is equal 1056to the width of @code{Pmode}. 1057@end defmac 1058 1059@defmac PROMOTE_MODE (@var{m}, @var{unsignedp}, @var{type}) 1060A macro to update @var{m} and @var{unsignedp} when an object whose type 1061is @var{type} and which has the specified mode and signedness is to be 1062stored in a register. This macro is only called when @var{type} is a 1063scalar type. 1064 1065On most RISC machines, which only have operations that operate on a full 1066register, define this macro to set @var{m} to @code{word_mode} if 1067@var{m} is an integer mode narrower than @code{BITS_PER_WORD}. In most 1068cases, only integer modes should be widened because wider-precision 1069floating-point operations are usually more expensive than their narrower 1070counterparts. 1071 1072For most machines, the macro definition does not change @var{unsignedp}. 1073However, some machines, have instructions that preferentially handle 1074either signed or unsigned quantities of certain modes. For example, on 1075the DEC Alpha, 32-bit loads from memory and 32-bit add instructions 1076sign-extend the result to 64 bits. On such machines, set 1077@var{unsignedp} according to which kind of extension is more efficient. 1078 1079Do not define this macro if it would never modify @var{m}. 1080@end defmac 1081 1082@deftypefn {Target Hook} bool TARGET_PROMOTE_FUNCTION_ARGS (tree @var{fntype}) 1083This target hook should return @code{true} if the promotion described by 1084@code{PROMOTE_MODE} should also be done for outgoing function arguments. 1085@end deftypefn 1086 1087@deftypefn {Target Hook} bool TARGET_PROMOTE_FUNCTION_RETURN (tree @var{fntype}) 1088This target hook should return @code{true} if the promotion described by 1089@code{PROMOTE_MODE} should also be done for the return value of 1090functions. 1091 1092If this target hook returns @code{true}, @code{FUNCTION_VALUE} must 1093perform the same promotions done by @code{PROMOTE_MODE}. 1094@end deftypefn 1095 1096@defmac PROMOTE_FOR_CALL_ONLY 1097Define this macro if the promotion described by @code{PROMOTE_MODE} 1098should @emph{only} be performed for outgoing function arguments or 1099function return values, as specified by @code{TARGET_PROMOTE_FUNCTION_ARGS} 1100and @code{TARGET_PROMOTE_FUNCTION_RETURN}, respectively. 1101@end defmac 1102 1103@defmac PARM_BOUNDARY 1104Normal alignment required for function parameters on the stack, in 1105bits. All stack parameters receive at least this much alignment 1106regardless of data type. On most machines, this is the same as the 1107size of an integer. 1108@end defmac 1109 1110@defmac STACK_BOUNDARY 1111Define this macro to the minimum alignment enforced by hardware for the 1112stack pointer on this machine. The definition is a C expression for the 1113desired alignment (measured in bits). This value is used as a default 1114if @code{PREFERRED_STACK_BOUNDARY} is not defined. On most machines, 1115this should be the same as @code{PARM_BOUNDARY}. 1116@end defmac 1117 1118@defmac PREFERRED_STACK_BOUNDARY 1119Define this macro if you wish to preserve a certain alignment for the 1120stack pointer, greater than what the hardware enforces. The definition 1121is a C expression for the desired alignment (measured in bits). This 1122macro must evaluate to a value equal to or larger than 1123@code{STACK_BOUNDARY}. 1124@end defmac 1125 1126@defmac FORCE_PREFERRED_STACK_BOUNDARY_IN_MAIN 1127A C expression that evaluates true if @code{PREFERRED_STACK_BOUNDARY} is 1128not guaranteed by the runtime and we should emit code to align the stack 1129at the beginning of @code{main}. 1130 1131@cindex @code{PUSH_ROUNDING}, interaction with @code{PREFERRED_STACK_BOUNDARY} 1132If @code{PUSH_ROUNDING} is not defined, the stack will always be aligned 1133to the specified boundary. If @code{PUSH_ROUNDING} is defined and specifies 1134a less strict alignment than @code{PREFERRED_STACK_BOUNDARY}, the stack may 1135be momentarily unaligned while pushing arguments. 1136@end defmac 1137 1138@defmac FUNCTION_BOUNDARY 1139Alignment required for a function entry point, in bits. 1140@end defmac 1141 1142@defmac BIGGEST_ALIGNMENT 1143Biggest alignment that any data type can require on this machine, in bits. 1144@end defmac 1145 1146@defmac MINIMUM_ATOMIC_ALIGNMENT 1147If defined, the smallest alignment, in bits, that can be given to an 1148object that can be referenced in one operation, without disturbing any 1149nearby object. Normally, this is @code{BITS_PER_UNIT}, but may be larger 1150on machines that don't have byte or half-word store operations. 1151@end defmac 1152 1153@defmac BIGGEST_FIELD_ALIGNMENT 1154Biggest alignment that any structure or union field can require on this 1155machine, in bits. If defined, this overrides @code{BIGGEST_ALIGNMENT} for 1156structure and union fields only, unless the field alignment has been set 1157by the @code{__attribute__ ((aligned (@var{n})))} construct. 1158@end defmac 1159 1160@defmac ADJUST_FIELD_ALIGN (@var{field}, @var{computed}) 1161An expression for the alignment of a structure field @var{field} if the 1162alignment computed in the usual way (including applying of 1163@code{BIGGEST_ALIGNMENT} and @code{BIGGEST_FIELD_ALIGNMENT} to the 1164alignment) is @var{computed}. It overrides alignment only if the 1165field alignment has not been set by the 1166@code{__attribute__ ((aligned (@var{n})))} construct. 1167@end defmac 1168 1169@defmac MAX_OFILE_ALIGNMENT 1170Biggest alignment supported by the object file format of this machine. 1171Use this macro to limit the alignment which can be specified using the 1172@code{__attribute__ ((aligned (@var{n})))} construct. If not defined, 1173the default value is @code{BIGGEST_ALIGNMENT}. 1174@end defmac 1175 1176@defmac DATA_ALIGNMENT (@var{type}, @var{basic-align}) 1177If defined, a C expression to compute the alignment for a variable in 1178the static store. @var{type} is the data type, and @var{basic-align} is 1179the alignment that the object would ordinarily have. The value of this 1180macro is used instead of that alignment to align the object. 1181 1182If this macro is not defined, then @var{basic-align} is used. 1183 1184@findex strcpy 1185One use of this macro is to increase alignment of medium-size data to 1186make it all fit in fewer cache lines. Another is to cause character 1187arrays to be word-aligned so that @code{strcpy} calls that copy 1188constants to character arrays can be done inline. 1189@end defmac 1190 1191@defmac CONSTANT_ALIGNMENT (@var{constant}, @var{basic-align}) 1192If defined, a C expression to compute the alignment given to a constant 1193that is being placed in memory. @var{constant} is the constant and 1194@var{basic-align} is the alignment that the object would ordinarily 1195have. The value of this macro is used instead of that alignment to 1196align the object. 1197 1198If this macro is not defined, then @var{basic-align} is used. 1199 1200The typical use of this macro is to increase alignment for string 1201constants to be word aligned so that @code{strcpy} calls that copy 1202constants can be done inline. 1203@end defmac 1204 1205@defmac LOCAL_ALIGNMENT (@var{type}, @var{basic-align}) 1206If defined, a C expression to compute the alignment for a variable in 1207the local store. @var{type} is the data type, and @var{basic-align} is 1208the alignment that the object would ordinarily have. The value of this 1209macro is used instead of that alignment to align the object. 1210 1211If this macro is not defined, then @var{basic-align} is used. 1212 1213One use of this macro is to increase alignment of medium-size data to 1214make it all fit in fewer cache lines. 1215@end defmac 1216 1217@defmac EMPTY_FIELD_BOUNDARY 1218Alignment in bits to be given to a structure bit-field that follows an 1219empty field such as @code{int : 0;}. 1220 1221If @code{PCC_BITFIELD_TYPE_MATTERS} is true, it overrides this macro. 1222@end defmac 1223 1224@defmac STRUCTURE_SIZE_BOUNDARY 1225Number of bits which any structure or union's size must be a multiple of. 1226Each structure or union's size is rounded up to a multiple of this. 1227 1228If you do not define this macro, the default is the same as 1229@code{BITS_PER_UNIT}. 1230@end defmac 1231 1232@defmac STRICT_ALIGNMENT 1233Define this macro to be the value 1 if instructions will fail to work 1234if given data not on the nominal alignment. If instructions will merely 1235go slower in that case, define this macro as 0. 1236@end defmac 1237 1238@defmac PCC_BITFIELD_TYPE_MATTERS 1239Define this if you wish to imitate the way many other C compilers handle 1240alignment of bit-fields and the structures that contain them. 1241 1242The behavior is that the type written for a named bit-field (@code{int}, 1243@code{short}, or other integer type) imposes an alignment for the entire 1244structure, as if the structure really did contain an ordinary field of 1245that type. In addition, the bit-field is placed within the structure so 1246that it would fit within such a field, not crossing a boundary for it. 1247 1248Thus, on most machines, a named bit-field whose type is written as 1249@code{int} would not cross a four-byte boundary, and would force 1250four-byte alignment for the whole structure. (The alignment used may 1251not be four bytes; it is controlled by the other alignment parameters.) 1252 1253An unnamed bit-field will not affect the alignment of the containing 1254structure. 1255 1256If the macro is defined, its definition should be a C expression; 1257a nonzero value for the expression enables this behavior. 1258 1259Note that if this macro is not defined, or its value is zero, some 1260bit-fields may cross more than one alignment boundary. The compiler can 1261support such references if there are @samp{insv}, @samp{extv}, and 1262@samp{extzv} insns that can directly reference memory. 1263 1264The other known way of making bit-fields work is to define 1265@code{STRUCTURE_SIZE_BOUNDARY} as large as @code{BIGGEST_ALIGNMENT}. 1266Then every structure can be accessed with fullwords. 1267 1268Unless the machine has bit-field instructions or you define 1269@code{STRUCTURE_SIZE_BOUNDARY} that way, you must define 1270@code{PCC_BITFIELD_TYPE_MATTERS} to have a nonzero value. 1271 1272If your aim is to make GCC use the same conventions for laying out 1273bit-fields as are used by another compiler, here is how to investigate 1274what the other compiler does. Compile and run this program: 1275 1276@smallexample 1277struct foo1 1278@{ 1279 char x; 1280 char :0; 1281 char y; 1282@}; 1283 1284struct foo2 1285@{ 1286 char x; 1287 int :0; 1288 char y; 1289@}; 1290 1291main () 1292@{ 1293 printf ("Size of foo1 is %d\n", 1294 sizeof (struct foo1)); 1295 printf ("Size of foo2 is %d\n", 1296 sizeof (struct foo2)); 1297 exit (0); 1298@} 1299@end smallexample 1300 1301If this prints 2 and 5, then the compiler's behavior is what you would 1302get from @code{PCC_BITFIELD_TYPE_MATTERS}. 1303@end defmac 1304 1305@defmac BITFIELD_NBYTES_LIMITED 1306Like @code{PCC_BITFIELD_TYPE_MATTERS} except that its effect is limited 1307to aligning a bit-field within the structure. 1308@end defmac 1309 1310@defmac MEMBER_TYPE_FORCES_BLK (@var{field}, @var{mode}) 1311Return 1 if a structure or array containing @var{field} should be accessed using 1312@code{BLKMODE}. 1313 1314If @var{field} is the only field in the structure, @var{mode} is its 1315mode, otherwise @var{mode} is VOIDmode. @var{mode} is provided in the 1316case where structures of one field would require the structure's mode to 1317retain the field's mode. 1318 1319Normally, this is not needed. See the file @file{c4x.h} for an example 1320of how to use this macro to prevent a structure having a floating point 1321field from being accessed in an integer mode. 1322@end defmac 1323 1324@defmac ROUND_TYPE_ALIGN (@var{type}, @var{computed}, @var{specified}) 1325Define this macro as an expression for the alignment of a type (given 1326by @var{type} as a tree node) if the alignment computed in the usual 1327way is @var{computed} and the alignment explicitly specified was 1328@var{specified}. 1329 1330The default is to use @var{specified} if it is larger; otherwise, use 1331the smaller of @var{computed} and @code{BIGGEST_ALIGNMENT} 1332@end defmac 1333 1334@defmac MAX_FIXED_MODE_SIZE 1335An integer expression for the size in bits of the largest integer 1336machine mode that should actually be used. All integer machine modes of 1337this size or smaller can be used for structures and unions with the 1338appropriate sizes. If this macro is undefined, @code{GET_MODE_BITSIZE 1339(DImode)} is assumed. 1340@end defmac 1341 1342@defmac VECTOR_MODE_SUPPORTED_P (@var{mode}) 1343Define this macro to be nonzero if the port is prepared to handle insns 1344involving vector mode @var{mode}. At the very least, it must have move 1345patterns for this mode. 1346@end defmac 1347 1348@defmac STACK_SAVEAREA_MODE (@var{save_level}) 1349If defined, an expression of type @code{enum machine_mode} that 1350specifies the mode of the save area operand of a 1351@code{save_stack_@var{level}} named pattern (@pxref{Standard Names}). 1352@var{save_level} is one of @code{SAVE_BLOCK}, @code{SAVE_FUNCTION}, or 1353@code{SAVE_NONLOCAL} and selects which of the three named patterns is 1354having its mode specified. 1355 1356You need not define this macro if it always returns @code{Pmode}. You 1357would most commonly define this macro if the 1358@code{save_stack_@var{level}} patterns need to support both a 32- and a 135964-bit mode. 1360@end defmac 1361 1362@defmac STACK_SIZE_MODE 1363If defined, an expression of type @code{enum machine_mode} that 1364specifies the mode of the size increment operand of an 1365@code{allocate_stack} named pattern (@pxref{Standard Names}). 1366 1367You need not define this macro if it always returns @code{word_mode}. 1368You would most commonly define this macro if the @code{allocate_stack} 1369pattern needs to support both a 32- and a 64-bit mode. 1370@end defmac 1371 1372@defmac TARGET_FLOAT_FORMAT 1373A code distinguishing the floating point format of the target machine. 1374There are four defined values: 1375 1376@ftable @code 1377@item IEEE_FLOAT_FORMAT 1378This code indicates IEEE floating point. It is the default; there is no 1379need to define @code{TARGET_FLOAT_FORMAT} when the format is IEEE@. 1380 1381@item VAX_FLOAT_FORMAT 1382This code indicates the ``F float'' (for @code{float}) and ``D float'' 1383or ``G float'' formats (for @code{double}) used on the VAX and PDP-11@. 1384 1385@item IBM_FLOAT_FORMAT 1386This code indicates the format used on the IBM System/370. 1387 1388@item C4X_FLOAT_FORMAT 1389This code indicates the format used on the TMS320C3x/C4x. 1390@end ftable 1391 1392If your target uses a floating point format other than these, you must 1393define a new @var{name}_FLOAT_FORMAT code for it, and add support for 1394it to @file{real.c}. 1395 1396The ordering of the component words of floating point values stored in 1397memory is controlled by @code{FLOAT_WORDS_BIG_ENDIAN}. 1398@end defmac 1399 1400@defmac MODE_HAS_NANS (@var{mode}) 1401When defined, this macro should be true if @var{mode} has a NaN 1402representation. The compiler assumes that NaNs are not equal to 1403anything (including themselves) and that addition, subtraction, 1404multiplication and division all return NaNs when one operand is 1405NaN@. 1406 1407By default, this macro is true if @var{mode} is a floating-point 1408mode and the target floating-point format is IEEE@. 1409@end defmac 1410 1411@defmac MODE_HAS_INFINITIES (@var{mode}) 1412This macro should be true if @var{mode} can represent infinity. At 1413present, the compiler uses this macro to decide whether @samp{x - x} 1414is always defined. By default, the macro is true when @var{mode} 1415is a floating-point mode and the target format is IEEE@. 1416@end defmac 1417 1418@defmac MODE_HAS_SIGNED_ZEROS (@var{mode}) 1419True if @var{mode} distinguishes between positive and negative zero. 1420The rules are expected to follow the IEEE standard: 1421 1422@itemize @bullet 1423@item 1424@samp{x + x} has the same sign as @samp{x}. 1425 1426@item 1427If the sum of two values with opposite sign is zero, the result is 1428positive for all rounding modes expect towards @minus{}infinity, for 1429which it is negative. 1430 1431@item 1432The sign of a product or quotient is negative when exactly one 1433of the operands is negative. 1434@end itemize 1435 1436The default definition is true if @var{mode} is a floating-point 1437mode and the target format is IEEE@. 1438@end defmac 1439 1440@defmac MODE_HAS_SIGN_DEPENDENT_ROUNDING (@var{mode}) 1441If defined, this macro should be true for @var{mode} if it has at 1442least one rounding mode in which @samp{x} and @samp{-x} can be 1443rounded to numbers of different magnitude. Two such modes are 1444towards @minus{}infinity and towards +infinity. 1445 1446The default definition of this macro is true if @var{mode} is 1447a floating-point mode and the target format is IEEE@. 1448@end defmac 1449 1450@defmac ROUND_TOWARDS_ZERO 1451If defined, this macro should be true if the prevailing rounding 1452mode is towards zero. A true value has the following effects: 1453 1454@itemize @bullet 1455@item 1456@code{MODE_HAS_SIGN_DEPENDENT_ROUNDING} will be false for all modes. 1457 1458@item 1459@file{libgcc.a}'s floating-point emulator will round towards zero 1460rather than towards nearest. 1461 1462@item 1463The compiler's floating-point emulator will round towards zero after 1464doing arithmetic, and when converting from the internal float format to 1465the target format. 1466@end itemize 1467 1468The macro does not affect the parsing of string literals. When the 1469primary rounding mode is towards zero, library functions like 1470@code{strtod} might still round towards nearest, and the compiler's 1471parser should behave like the target's @code{strtod} where possible. 1472 1473Not defining this macro is equivalent to returning zero. 1474@end defmac 1475 1476@defmac LARGEST_EXPONENT_IS_NORMAL (@var{size}) 1477This macro should return true if floats with @var{size} 1478bits do not have a NaN or infinity representation, but use the largest 1479exponent for normal numbers instead. 1480 1481Defining this macro to true for @var{size} causes @code{MODE_HAS_NANS} 1482and @code{MODE_HAS_INFINITIES} to be false for @var{size}-bit modes. 1483It also affects the way @file{libgcc.a} and @file{real.c} emulate 1484floating-point arithmetic. 1485 1486The default definition of this macro returns false for all sizes. 1487@end defmac 1488 1489@deftypefn {Target Hook} bool TARGET_VECTOR_OPAQUE_P (tree @var{type}) 1490This target hook should return @code{true} a vector is opaque. That 1491is, if no cast is needed when copying a vector value of type 1492@var{type} into another vector lvalue of the same size. Vector opaque 1493types cannot be initialized. The default is that there are no such 1494types. 1495@end deftypefn 1496 1497@deftypefn {Target Hook} bool TARGET_MS_BITFIELD_LAYOUT_P (tree @var{record_type}) 1498This target hook returns @code{true} if bit-fields in the given 1499@var{record_type} are to be laid out following the rules of Microsoft 1500Visual C/C++, namely: (i) a bit-field won't share the same storage 1501unit with the previous bit-field if their underlying types have 1502different sizes, and the bit-field will be aligned to the highest 1503alignment of the underlying types of itself and of the previous 1504bit-field; (ii) a zero-sized bit-field will affect the alignment of 1505the whole enclosing structure, even if it is unnamed; except that 1506(iii) a zero-sized bit-field will be disregarded unless it follows 1507another bit-field of nonzero size. If this hook returns @code{true}, 1508other macros that control bit-field layout are ignored. 1509 1510When a bit-field is inserted into a packed record, the whole size 1511of the underlying type is used by one or more same-size adjacent 1512bit-fields (that is, if its long:3, 32 bits is used in the record, 1513and any additional adjacent long bit-fields are packed into the same 1514chunk of 32 bits. However, if the size changes, a new field of that 1515size is allocated). In an unpacked record, this is the same as using 1516alignment, but not equivalent when packing. 1517 1518If both MS bit-fields and @samp{__attribute__((packed))} are used, 1519the latter will take precedence. If @samp{__attribute__((packed))} is 1520used on a single field when MS bit-fields are in use, it will take 1521precedence for that field, but the alignment of the rest of the structure 1522may affect its placement. 1523@end deftypefn 1524 1525@deftypefn {Target Hook} {const char *} TARGET_MANGLE_FUNDAMENTAL_TYPE (tree @var{type}) 1526If your target defines any fundamental types, define this hook to 1527return the appropriate encoding for these types as part of a C++ 1528mangled name. The @var{type} argument is the tree structure 1529representing the type to be mangled. The hook may be applied to trees 1530which are not target-specific fundamental types; it should return 1531@code{NULL} for all such types, as well as arguments it does not 1532recognize. If the return value is not @code{NULL}, it must point to 1533a statically-allocated string constant. 1534 1535Target-specific fundamental types might be new fundamental types or 1536qualified versions of ordinary fundamental types. Encode new 1537fundamental types as @samp{@w{u @var{n} @var{name}}}, where @var{name} 1538is the name used for the type in source code, and @var{n} is the 1539length of @var{name} in decimal. Encode qualified versions of 1540ordinary types as @samp{@w{U @var{n} @var{name} @var{code}}}, where 1541@var{name} is the name used for the type qualifier in source code, 1542@var{n} is the length of @var{name} as above, and @var{code} is the 1543code used to represent the unqualified version of this type. (See 1544@code{write_builtin_type} in @file{cp/mangle.c} for the list of 1545codes.) In both cases the spaces are for clarity; do not include any 1546spaces in your string. 1547 1548The default version of this hook always returns @code{NULL}, which is 1549appropriate for a target that does not define any new fundamental 1550types. 1551@end deftypefn 1552 1553@node Type Layout 1554@section Layout of Source Language Data Types 1555 1556These macros define the sizes and other characteristics of the standard 1557basic data types used in programs being compiled. Unlike the macros in 1558the previous section, these apply to specific features of C and related 1559languages, rather than to fundamental aspects of storage layout. 1560 1561@defmac INT_TYPE_SIZE 1562A C expression for the size in bits of the type @code{int} on the 1563target machine. If you don't define this, the default is one word. 1564@end defmac 1565 1566@defmac SHORT_TYPE_SIZE 1567A C expression for the size in bits of the type @code{short} on the 1568target machine. If you don't define this, the default is half a word. 1569(If this would be less than one storage unit, it is rounded up to one 1570unit.) 1571@end defmac 1572 1573@defmac LONG_TYPE_SIZE 1574A C expression for the size in bits of the type @code{long} on the 1575target machine. If you don't define this, the default is one word. 1576@end defmac 1577 1578@defmac ADA_LONG_TYPE_SIZE 1579On some machines, the size used for the Ada equivalent of the type 1580@code{long} by a native Ada compiler differs from that used by C. In 1581that situation, define this macro to be a C expression to be used for 1582the size of that type. If you don't define this, the default is the 1583value of @code{LONG_TYPE_SIZE}. 1584@end defmac 1585 1586@defmac MAX_LONG_TYPE_SIZE 1587Maximum number for the size in bits of the type @code{long} on the 1588target machine. If this is undefined, the default is 1589@code{LONG_TYPE_SIZE}. Otherwise, it is the constant value that is the 1590largest value that @code{LONG_TYPE_SIZE} can have at run-time. This is 1591used in @code{cpp}. 1592@end defmac 1593 1594@defmac LONG_LONG_TYPE_SIZE 1595A C expression for the size in bits of the type @code{long long} on the 1596target machine. If you don't define this, the default is two 1597words. If you want to support GNU Ada on your machine, the value of this 1598macro must be at least 64. 1599@end defmac 1600 1601@defmac CHAR_TYPE_SIZE 1602A C expression for the size in bits of the type @code{char} on the 1603target machine. If you don't define this, the default is 1604@code{BITS_PER_UNIT}. 1605@end defmac 1606 1607@defmac BOOL_TYPE_SIZE 1608A C expression for the size in bits of the C++ type @code{bool} and 1609C99 type @code{_Bool} on the target machine. If you don't define 1610this, and you probably shouldn't, the default is @code{CHAR_TYPE_SIZE}. 1611@end defmac 1612 1613@defmac FLOAT_TYPE_SIZE 1614A C expression for the size in bits of the type @code{float} on the 1615target machine. If you don't define this, the default is one word. 1616@end defmac 1617 1618@defmac DOUBLE_TYPE_SIZE 1619A C expression for the size in bits of the type @code{double} on the 1620target machine. If you don't define this, the default is two 1621words. 1622@end defmac 1623 1624@defmac LONG_DOUBLE_TYPE_SIZE 1625A C expression for the size in bits of the type @code{long double} on 1626the target machine. If you don't define this, the default is two 1627words. 1628@end defmac 1629 1630@defmac MAX_LONG_DOUBLE_TYPE_SIZE 1631Maximum number for the size in bits of the type @code{long double} on the 1632target machine. If this is undefined, the default is 1633@code{LONG_DOUBLE_TYPE_SIZE}. Otherwise, it is the constant value that is 1634the largest value that @code{LONG_DOUBLE_TYPE_SIZE} can have at run-time. 1635This is used in @code{cpp}. 1636@end defmac 1637 1638@defmac TARGET_FLT_EVAL_METHOD 1639A C expression for the value for @code{FLT_EVAL_METHOD} in @file{float.h}, 1640assuming, if applicable, that the floating-point control word is in its 1641default state. If you do not define this macro the value of 1642@code{FLT_EVAL_METHOD} will be zero. 1643@end defmac 1644 1645@defmac WIDEST_HARDWARE_FP_SIZE 1646A C expression for the size in bits of the widest floating-point format 1647supported by the hardware. If you define this macro, you must specify a 1648value less than or equal to the value of @code{LONG_DOUBLE_TYPE_SIZE}. 1649If you do not define this macro, the value of @code{LONG_DOUBLE_TYPE_SIZE} 1650is the default. 1651@end defmac 1652 1653@defmac DEFAULT_SIGNED_CHAR 1654An expression whose value is 1 or 0, according to whether the type 1655@code{char} should be signed or unsigned by default. The user can 1656always override this default with the options @option{-fsigned-char} 1657and @option{-funsigned-char}. 1658@end defmac 1659 1660@defmac DEFAULT_SHORT_ENUMS 1661A C expression to determine whether to give an @code{enum} type 1662only as many bytes as it takes to represent the range of possible values 1663of that type. A nonzero value means to do that; a zero value means all 1664@code{enum} types should be allocated like @code{int}. 1665 1666If you don't define the macro, the default is 0. 1667@end defmac 1668 1669@defmac SIZE_TYPE 1670A C expression for a string describing the name of the data type to use 1671for size values. The typedef name @code{size_t} is defined using the 1672contents of the string. 1673 1674The string can contain more than one keyword. If so, separate them with 1675spaces, and write first any length keyword, then @code{unsigned} if 1676appropriate, and finally @code{int}. The string must exactly match one 1677of the data type names defined in the function 1678@code{init_decl_processing} in the file @file{c-decl.c}. You may not 1679omit @code{int} or change the order---that would cause the compiler to 1680crash on startup. 1681 1682If you don't define this macro, the default is @code{"long unsigned 1683int"}. 1684@end defmac 1685 1686@defmac PTRDIFF_TYPE 1687A C expression for a string describing the name of the data type to use 1688for the result of subtracting two pointers. The typedef name 1689@code{ptrdiff_t} is defined using the contents of the string. See 1690@code{SIZE_TYPE} above for more information. 1691 1692If you don't define this macro, the default is @code{"long int"}. 1693@end defmac 1694 1695@defmac WCHAR_TYPE 1696A C expression for a string describing the name of the data type to use 1697for wide characters. The typedef name @code{wchar_t} is defined using 1698the contents of the string. See @code{SIZE_TYPE} above for more 1699information. 1700 1701If you don't define this macro, the default is @code{"int"}. 1702@end defmac 1703 1704@defmac WCHAR_TYPE_SIZE 1705A C expression for the size in bits of the data type for wide 1706characters. This is used in @code{cpp}, which cannot make use of 1707@code{WCHAR_TYPE}. 1708@end defmac 1709 1710@defmac MAX_WCHAR_TYPE_SIZE 1711Maximum number for the size in bits of the data type for wide 1712characters. If this is undefined, the default is 1713@code{WCHAR_TYPE_SIZE}. Otherwise, it is the constant value that is the 1714largest value that @code{WCHAR_TYPE_SIZE} can have at run-time. This is 1715used in @code{cpp}. 1716@end defmac 1717 1718@defmac GCOV_TYPE_SIZE 1719A C expression for the size in bits of the type used for gcov counters on the 1720target machine. If you don't define this, the default is one 1721@code{LONG_TYPE_SIZE} in case it is greater or equal to 64-bit and 1722@code{LONG_LONG_TYPE_SIZE} otherwise. You may want to re-define the type to 1723ensure atomicity for counters in multithreaded programs. 1724@end defmac 1725 1726@defmac WINT_TYPE 1727A C expression for a string describing the name of the data type to 1728use for wide characters passed to @code{printf} and returned from 1729@code{getwc}. The typedef name @code{wint_t} is defined using the 1730contents of the string. See @code{SIZE_TYPE} above for more 1731information. 1732 1733If you don't define this macro, the default is @code{"unsigned int"}. 1734@end defmac 1735 1736@defmac INTMAX_TYPE 1737A C expression for a string describing the name of the data type that 1738can represent any value of any standard or extended signed integer type. 1739The typedef name @code{intmax_t} is defined using the contents of the 1740string. See @code{SIZE_TYPE} above for more information. 1741 1742If you don't define this macro, the default is the first of 1743@code{"int"}, @code{"long int"}, or @code{"long long int"} that has as 1744much precision as @code{long long int}. 1745@end defmac 1746 1747@defmac UINTMAX_TYPE 1748A C expression for a string describing the name of the data type that 1749can represent any value of any standard or extended unsigned integer 1750type. The typedef name @code{uintmax_t} is defined using the contents 1751of the string. See @code{SIZE_TYPE} above for more information. 1752 1753If you don't define this macro, the default is the first of 1754@code{"unsigned int"}, @code{"long unsigned int"}, or @code{"long long 1755unsigned int"} that has as much precision as @code{long long unsigned 1756int}. 1757@end defmac 1758 1759@defmac TARGET_PTRMEMFUNC_VBIT_LOCATION 1760The C++ compiler represents a pointer-to-member-function with a struct 1761that looks like: 1762 1763@smallexample 1764 struct @{ 1765 union @{ 1766 void (*fn)(); 1767 ptrdiff_t vtable_index; 1768 @}; 1769 ptrdiff_t delta; 1770 @}; 1771@end smallexample 1772 1773@noindent 1774The C++ compiler must use one bit to indicate whether the function that 1775will be called through a pointer-to-member-function is virtual. 1776Normally, we assume that the low-order bit of a function pointer must 1777always be zero. Then, by ensuring that the vtable_index is odd, we can 1778distinguish which variant of the union is in use. But, on some 1779platforms function pointers can be odd, and so this doesn't work. In 1780that case, we use the low-order bit of the @code{delta} field, and shift 1781the remainder of the @code{delta} field to the left. 1782 1783GCC will automatically make the right selection about where to store 1784this bit using the @code{FUNCTION_BOUNDARY} setting for your platform. 1785However, some platforms such as ARM/Thumb have @code{FUNCTION_BOUNDARY} 1786set such that functions always start at even addresses, but the lowest 1787bit of pointers to functions indicate whether the function at that 1788address is in ARM or Thumb mode. If this is the case of your 1789architecture, you should define this macro to 1790@code{ptrmemfunc_vbit_in_delta}. 1791 1792In general, you should not have to define this macro. On architectures 1793in which function addresses are always even, according to 1794@code{FUNCTION_BOUNDARY}, GCC will automatically define this macro to 1795@code{ptrmemfunc_vbit_in_pfn}. 1796@end defmac 1797 1798@defmac TARGET_VTABLE_USES_DESCRIPTORS 1799Normally, the C++ compiler uses function pointers in vtables. This 1800macro allows the target to change to use ``function descriptors'' 1801instead. Function descriptors are found on targets for whom a 1802function pointer is actually a small data structure. Normally the 1803data structure consists of the actual code address plus a data 1804pointer to which the function's data is relative. 1805 1806If vtables are used, the value of this macro should be the number 1807of words that the function descriptor occupies. 1808@end defmac 1809 1810@defmac TARGET_VTABLE_ENTRY_ALIGN 1811By default, the vtable entries are void pointers, the so the alignment 1812is the same as pointer alignment. The value of this macro specifies 1813the alignment of the vtable entry in bits. It should be defined only 1814when special alignment is necessary. */ 1815@end defmac 1816 1817@defmac TARGET_VTABLE_DATA_ENTRY_DISTANCE 1818There are a few non-descriptor entries in the vtable at offsets below 1819zero. If these entries must be padded (say, to preserve the alignment 1820specified by @code{TARGET_VTABLE_ENTRY_ALIGN}), set this to the number 1821of words in each data entry. 1822@end defmac 1823 1824@node Escape Sequences 1825@section Target Character Escape Sequences 1826@cindex escape sequences 1827 1828By default, GCC assumes that the C character escape sequences take on 1829their ASCII values for the target. If this is not correct, you must 1830explicitly define all of the macros below. All of them must evaluate 1831to constants; they are used in @code{case} statements. 1832 1833@findex TARGET_BELL 1834@findex TARGET_CR 1835@findex TARGET_ESC 1836@findex TARGET_FF 1837@findex TARGET_NEWLINE 1838@findex TARGET_TAB 1839@findex TARGET_VT 1840@multitable {@code{TARGET_NEWLINE}} {Escape} {ASCII character} 1841@item Macro @tab Escape @tab ASCII character 1842@item @code{TARGET_BELL} @tab @kbd{\a} @tab @code{07}, @code{BEL} 1843@item @code{TARGET_CR} @tab @kbd{\r} @tab @code{0D}, @code{CR} 1844@item @code{TARGET_ESC} @tab @kbd{\e}, @kbd{\E} @tab @code{1B}, @code{ESC} 1845@item @code{TARGET_FF} @tab @kbd{\f} @tab @code{0C}, @code{FF} 1846@item @code{TARGET_NEWLINE} @tab @kbd{\n} @tab @code{0A}, @code{LF} 1847@item @code{TARGET_TAB} @tab @kbd{\t} @tab @code{09}, @code{HT} 1848@item @code{TARGET_VT} @tab @kbd{\v} @tab @code{0B}, @code{VT} 1849@end multitable 1850 1851@noindent 1852Note that the @kbd{\e} and @kbd{\E} escapes are GNU extensions, not 1853part of the C standard. 1854 1855@node Registers 1856@section Register Usage 1857@cindex register usage 1858 1859This section explains how to describe what registers the target machine 1860has, and how (in general) they can be used. 1861 1862The description of which registers a specific instruction can use is 1863done with register classes; see @ref{Register Classes}. For information 1864on using registers to access a stack frame, see @ref{Frame Registers}. 1865For passing values in registers, see @ref{Register Arguments}. 1866For returning values in registers, see @ref{Scalar Return}. 1867 1868@menu 1869* Register Basics:: Number and kinds of registers. 1870* Allocation Order:: Order in which registers are allocated. 1871* Values in Registers:: What kinds of values each reg can hold. 1872* Leaf Functions:: Renumbering registers for leaf functions. 1873* Stack Registers:: Handling a register stack such as 80387. 1874@end menu 1875 1876@node Register Basics 1877@subsection Basic Characteristics of Registers 1878 1879@c prevent bad page break with this line 1880Registers have various characteristics. 1881 1882@defmac FIRST_PSEUDO_REGISTER 1883Number of hardware registers known to the compiler. They receive 1884numbers 0 through @code{FIRST_PSEUDO_REGISTER-1}; thus, the first 1885pseudo register's number really is assigned the number 1886@code{FIRST_PSEUDO_REGISTER}. 1887@end defmac 1888 1889@defmac FIXED_REGISTERS 1890@cindex fixed register 1891An initializer that says which registers are used for fixed purposes 1892all throughout the compiled code and are therefore not available for 1893general allocation. These would include the stack pointer, the frame 1894pointer (except on machines where that can be used as a general 1895register when no frame pointer is needed), the program counter on 1896machines where that is considered one of the addressable registers, 1897and any other numbered register with a standard use. 1898 1899This information is expressed as a sequence of numbers, separated by 1900commas and surrounded by braces. The @var{n}th number is 1 if 1901register @var{n} is fixed, 0 otherwise. 1902 1903The table initialized from this macro, and the table initialized by 1904the following one, may be overridden at run time either automatically, 1905by the actions of the macro @code{CONDITIONAL_REGISTER_USAGE}, or by 1906the user with the command options @option{-ffixed-@var{reg}}, 1907@option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}}. 1908@end defmac 1909 1910@defmac CALL_USED_REGISTERS 1911@cindex call-used register 1912@cindex call-clobbered register 1913@cindex call-saved register 1914Like @code{FIXED_REGISTERS} but has 1 for each register that is 1915clobbered (in general) by function calls as well as for fixed 1916registers. This macro therefore identifies the registers that are not 1917available for general allocation of values that must live across 1918function calls. 1919 1920If a register has 0 in @code{CALL_USED_REGISTERS}, the compiler 1921automatically saves it on function entry and restores it on function 1922exit, if the register is used within the function. 1923@end defmac 1924 1925@defmac CALL_REALLY_USED_REGISTERS 1926@cindex call-used register 1927@cindex call-clobbered register 1928@cindex call-saved register 1929Like @code{CALL_USED_REGISTERS} except this macro doesn't require 1930that the entire set of @code{FIXED_REGISTERS} be included. 1931(@code{CALL_USED_REGISTERS} must be a superset of @code{FIXED_REGISTERS}). 1932This macro is optional. If not specified, it defaults to the value 1933of @code{CALL_USED_REGISTERS}. 1934@end defmac 1935 1936@defmac HARD_REGNO_CALL_PART_CLOBBERED (@var{regno}, @var{mode}) 1937@cindex call-used register 1938@cindex call-clobbered register 1939@cindex call-saved register 1940A C expression that is nonzero if it is not permissible to store a 1941value of mode @var{mode} in hard register number @var{regno} across a 1942call without some part of it being clobbered. For most machines this 1943macro need not be defined. It is only required for machines that do not 1944preserve the entire contents of a register across a call. 1945@end defmac 1946 1947@findex fixed_regs 1948@findex call_used_regs 1949@findex global_regs 1950@findex reg_names 1951@findex reg_class_contents 1952@defmac CONDITIONAL_REGISTER_USAGE 1953Zero or more C statements that may conditionally modify five variables 1954@code{fixed_regs}, @code{call_used_regs}, @code{global_regs}, 1955@code{reg_names}, and @code{reg_class_contents}, to take into account 1956any dependence of these register sets on target flags. The first three 1957of these are of type @code{char []} (interpreted as Boolean vectors). 1958@code{global_regs} is a @code{const char *[]}, and 1959@code{reg_class_contents} is a @code{HARD_REG_SET}. Before the macro is 1960called, @code{fixed_regs}, @code{call_used_regs}, 1961@code{reg_class_contents}, and @code{reg_names} have been initialized 1962from @code{FIXED_REGISTERS}, @code{CALL_USED_REGISTERS}, 1963@code{REG_CLASS_CONTENTS}, and @code{REGISTER_NAMES}, respectively. 1964@code{global_regs} has been cleared, and any @option{-ffixed-@var{reg}}, 1965@option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}} 1966command options have been applied. 1967 1968You need not define this macro if it has no work to do. 1969 1970@cindex disabling certain registers 1971@cindex controlling register usage 1972If the usage of an entire class of registers depends on the target 1973flags, you may indicate this to GCC by using this macro to modify 1974@code{fixed_regs} and @code{call_used_regs} to 1 for each of the 1975registers in the classes which should not be used by GCC@. Also define 1976the macro @code{REG_CLASS_FROM_LETTER} / @code{REG_CLASS_FROM_CONSTRAINT} 1977to return @code{NO_REGS} if it 1978is called with a letter for a class that shouldn't be used. 1979 1980(However, if this class is not included in @code{GENERAL_REGS} and all 1981of the insn patterns whose constraints permit this class are 1982controlled by target switches, then GCC will automatically avoid using 1983these registers when the target switches are opposed to them.) 1984@end defmac 1985 1986@defmac NON_SAVING_SETJMP 1987If this macro is defined and has a nonzero value, it means that 1988@code{setjmp} and related functions fail to save the registers, or that 1989@code{longjmp} fails to restore them. To compensate, the compiler 1990avoids putting variables in registers in functions that use 1991@code{setjmp}. 1992@end defmac 1993 1994@defmac INCOMING_REGNO (@var{out}) 1995Define this macro if the target machine has register windows. This C 1996expression returns the register number as seen by the called function 1997corresponding to the register number @var{out} as seen by the calling 1998function. Return @var{out} if register number @var{out} is not an 1999outbound register. 2000@end defmac 2001 2002@defmac OUTGOING_REGNO (@var{in}) 2003Define this macro if the target machine has register windows. This C 2004expression returns the register number as seen by the calling function 2005corresponding to the register number @var{in} as seen by the called 2006function. Return @var{in} if register number @var{in} is not an inbound 2007register. 2008@end defmac 2009 2010@defmac LOCAL_REGNO (@var{regno}) 2011Define this macro if the target machine has register windows. This C 2012expression returns true if the register is call-saved but is in the 2013register window. Unlike most call-saved registers, such registers 2014need not be explicitly restored on function exit or during non-local 2015gotos. 2016@end defmac 2017 2018@defmac PC_REGNUM 2019If the program counter has a register number, define this as that 2020register number. Otherwise, do not define it. 2021@end defmac 2022 2023@node Allocation Order 2024@subsection Order of Allocation of Registers 2025@cindex order of register allocation 2026@cindex register allocation order 2027 2028@c prevent bad page break with this line 2029Registers are allocated in order. 2030 2031@defmac REG_ALLOC_ORDER 2032If defined, an initializer for a vector of integers, containing the 2033numbers of hard registers in the order in which GCC should prefer 2034to use them (from most preferred to least). 2035 2036If this macro is not defined, registers are used lowest numbered first 2037(all else being equal). 2038 2039One use of this macro is on machines where the highest numbered 2040registers must always be saved and the save-multiple-registers 2041instruction supports only sequences of consecutive registers. On such 2042machines, define @code{REG_ALLOC_ORDER} to be an initializer that lists 2043the highest numbered allocable register first. 2044@end defmac 2045 2046@defmac ORDER_REGS_FOR_LOCAL_ALLOC 2047A C statement (sans semicolon) to choose the order in which to allocate 2048hard registers for pseudo-registers local to a basic block. 2049 2050Store the desired register order in the array @code{reg_alloc_order}. 2051Element 0 should be the register to allocate first; element 1, the next 2052register; and so on. 2053 2054The macro body should not assume anything about the contents of 2055@code{reg_alloc_order} before execution of the macro. 2056 2057On most machines, it is not necessary to define this macro. 2058@end defmac 2059 2060@node Values in Registers 2061@subsection How Values Fit in Registers 2062 2063This section discusses the macros that describe which kinds of values 2064(specifically, which machine modes) each register can hold, and how many 2065consecutive registers are needed for a given mode. 2066 2067@defmac HARD_REGNO_NREGS (@var{regno}, @var{mode}) 2068A C expression for the number of consecutive hard registers, starting 2069at register number @var{regno}, required to hold a value of mode 2070@var{mode}. 2071 2072On a machine where all registers are exactly one word, a suitable 2073definition of this macro is 2074 2075@smallexample 2076#define HARD_REGNO_NREGS(REGNO, MODE) \ 2077 ((GET_MODE_SIZE (MODE) + UNITS_PER_WORD - 1) \ 2078 / UNITS_PER_WORD) 2079@end smallexample 2080@end defmac 2081 2082@defmac REGMODE_NATURAL_SIZE (@var{mode}) 2083Define this macro if the natural size of registers that hold values 2084of mode @var{mode} is not the word size. It is a C expression that 2085should give the natural size in bytes for the specified mode. It is 2086used by the register allocator to try to optimize its results. This 2087happens for example on SPARC 64-bit where the natural size of 2088floating-point registers is still 32-bit. 2089@end defmac 2090 2091@defmac HARD_REGNO_MODE_OK (@var{regno}, @var{mode}) 2092A C expression that is nonzero if it is permissible to store a value 2093of mode @var{mode} in hard register number @var{regno} (or in several 2094registers starting with that one). For a machine where all registers 2095are equivalent, a suitable definition is 2096 2097@smallexample 2098#define HARD_REGNO_MODE_OK(REGNO, MODE) 1 2099@end smallexample 2100 2101You need not include code to check for the numbers of fixed registers, 2102because the allocation mechanism considers them to be always occupied. 2103 2104@cindex register pairs 2105On some machines, double-precision values must be kept in even/odd 2106register pairs. You can implement that by defining this macro to reject 2107odd register numbers for such modes. 2108 2109The minimum requirement for a mode to be OK in a register is that the 2110@samp{mov@var{mode}} instruction pattern support moves between the 2111register and other hard register in the same class and that moving a 2112value into the register and back out not alter it. 2113 2114Since the same instruction used to move @code{word_mode} will work for 2115all narrower integer modes, it is not necessary on any machine for 2116@code{HARD_REGNO_MODE_OK} to distinguish between these modes, provided 2117you define patterns @samp{movhi}, etc., to take advantage of this. This 2118is useful because of the interaction between @code{HARD_REGNO_MODE_OK} 2119and @code{MODES_TIEABLE_P}; it is very desirable for all integer modes 2120to be tieable. 2121 2122Many machines have special registers for floating point arithmetic. 2123Often people assume that floating point machine modes are allowed only 2124in floating point registers. This is not true. Any registers that 2125can hold integers can safely @emph{hold} a floating point machine 2126mode, whether or not floating arithmetic can be done on it in those 2127registers. Integer move instructions can be used to move the values. 2128 2129On some machines, though, the converse is true: fixed-point machine 2130modes may not go in floating registers. This is true if the floating 2131registers normalize any value stored in them, because storing a 2132non-floating value there would garble it. In this case, 2133@code{HARD_REGNO_MODE_OK} should reject fixed-point machine modes in 2134floating registers. But if the floating registers do not automatically 2135normalize, if you can store any bit pattern in one and retrieve it 2136unchanged without a trap, then any machine mode may go in a floating 2137register, so you can define this macro to say so. 2138 2139The primary significance of special floating registers is rather that 2140they are the registers acceptable in floating point arithmetic 2141instructions. However, this is of no concern to 2142@code{HARD_REGNO_MODE_OK}. You handle it by writing the proper 2143constraints for those instructions. 2144 2145On some machines, the floating registers are especially slow to access, 2146so that it is better to store a value in a stack frame than in such a 2147register if floating point arithmetic is not being done. As long as the 2148floating registers are not in class @code{GENERAL_REGS}, they will not 2149be used unless some pattern's constraint asks for one. 2150@end defmac 2151 2152@defmac MODES_TIEABLE_P (@var{mode1}, @var{mode2}) 2153A C expression that is nonzero if a value of mode 2154@var{mode1} is accessible in mode @var{mode2} without copying. 2155 2156If @code{HARD_REGNO_MODE_OK (@var{r}, @var{mode1})} and 2157@code{HARD_REGNO_MODE_OK (@var{r}, @var{mode2})} are always the same for 2158any @var{r}, then @code{MODES_TIEABLE_P (@var{mode1}, @var{mode2})} 2159should be nonzero. If they differ for any @var{r}, you should define 2160this macro to return zero unless some other mechanism ensures the 2161accessibility of the value in a narrower mode. 2162 2163You should define this macro to return nonzero in as many cases as 2164possible since doing so will allow GCC to perform better register 2165allocation. 2166@end defmac 2167 2168@defmac AVOID_CCMODE_COPIES 2169Define this macro if the compiler should avoid copies to/from @code{CCmode} 2170registers. You should only define this macro if support for copying to/from 2171@code{CCmode} is incomplete. 2172@end defmac 2173 2174@node Leaf Functions 2175@subsection Handling Leaf Functions 2176 2177@cindex leaf functions 2178@cindex functions, leaf 2179On some machines, a leaf function (i.e., one which makes no calls) can run 2180more efficiently if it does not make its own register window. Often this 2181means it is required to receive its arguments in the registers where they 2182are passed by the caller, instead of the registers where they would 2183normally arrive. 2184 2185The special treatment for leaf functions generally applies only when 2186other conditions are met; for example, often they may use only those 2187registers for its own variables and temporaries. We use the term ``leaf 2188function'' to mean a function that is suitable for this special 2189handling, so that functions with no calls are not necessarily ``leaf 2190functions''. 2191 2192GCC assigns register numbers before it knows whether the function is 2193suitable for leaf function treatment. So it needs to renumber the 2194registers in order to output a leaf function. The following macros 2195accomplish this. 2196 2197@defmac LEAF_REGISTERS 2198Name of a char vector, indexed by hard register number, which 2199contains 1 for a register that is allowable in a candidate for leaf 2200function treatment. 2201 2202If leaf function treatment involves renumbering the registers, then the 2203registers marked here should be the ones before renumbering---those that 2204GCC would ordinarily allocate. The registers which will actually be 2205used in the assembler code, after renumbering, should not be marked with 1 2206in this vector. 2207 2208Define this macro only if the target machine offers a way to optimize 2209the treatment of leaf functions. 2210@end defmac 2211 2212@defmac LEAF_REG_REMAP (@var{regno}) 2213A C expression whose value is the register number to which @var{regno} 2214should be renumbered, when a function is treated as a leaf function. 2215 2216If @var{regno} is a register number which should not appear in a leaf 2217function before renumbering, then the expression should yield @minus{}1, which 2218will cause the compiler to abort. 2219 2220Define this macro only if the target machine offers a way to optimize the 2221treatment of leaf functions, and registers need to be renumbered to do 2222this. 2223@end defmac 2224 2225@findex current_function_is_leaf 2226@findex current_function_uses_only_leaf_regs 2227@code{TARGET_ASM_FUNCTION_PROLOGUE} and 2228@code{TARGET_ASM_FUNCTION_EPILOGUE} must usually treat leaf functions 2229specially. They can test the C variable @code{current_function_is_leaf} 2230which is nonzero for leaf functions. @code{current_function_is_leaf} is 2231set prior to local register allocation and is valid for the remaining 2232compiler passes. They can also test the C variable 2233@code{current_function_uses_only_leaf_regs} which is nonzero for leaf 2234functions which only use leaf registers. 2235@code{current_function_uses_only_leaf_regs} is valid after reload and is 2236only useful if @code{LEAF_REGISTERS} is defined. 2237@c changed this to fix overfull. ALSO: why the "it" at the beginning 2238@c of the next paragraph?! --mew 2feb93 2239 2240@node Stack Registers 2241@subsection Registers That Form a Stack 2242 2243There are special features to handle computers where some of the 2244``registers'' form a stack. Stack registers are normally written by 2245pushing onto the stack, and are numbered relative to the top of the 2246stack. 2247 2248Currently, GCC can only handle one group of stack-like registers, and 2249they must be consecutively numbered. Furthermore, the existing 2250support for stack-like registers is specific to the 80387 floating 2251point coprocessor. If you have a new architecture that uses 2252stack-like registers, you will need to do substantial work on 2253@file{reg-stack.c} and write your machine description to cooperate 2254with it, as well as defining these macros. 2255 2256@defmac STACK_REGS 2257Define this if the machine has any stack-like registers. 2258@end defmac 2259 2260@defmac FIRST_STACK_REG 2261The number of the first stack-like register. This one is the top 2262of the stack. 2263@end defmac 2264 2265@defmac LAST_STACK_REG 2266The number of the last stack-like register. This one is the bottom of 2267the stack. 2268@end defmac 2269 2270@node Register Classes 2271@section Register Classes 2272@cindex register class definitions 2273@cindex class definitions, register 2274 2275On many machines, the numbered registers are not all equivalent. 2276For example, certain registers may not be allowed for indexed addressing; 2277certain registers may not be allowed in some instructions. These machine 2278restrictions are described to the compiler using @dfn{register classes}. 2279 2280You define a number of register classes, giving each one a name and saying 2281which of the registers belong to it. Then you can specify register classes 2282that are allowed as operands to particular instruction patterns. 2283 2284@findex ALL_REGS 2285@findex NO_REGS 2286In general, each register will belong to several classes. In fact, one 2287class must be named @code{ALL_REGS} and contain all the registers. Another 2288class must be named @code{NO_REGS} and contain no registers. Often the 2289union of two classes will be another class; however, this is not required. 2290 2291@findex GENERAL_REGS 2292One of the classes must be named @code{GENERAL_REGS}. There is nothing 2293terribly special about the name, but the operand constraint letters 2294@samp{r} and @samp{g} specify this class. If @code{GENERAL_REGS} is 2295the same as @code{ALL_REGS}, just define it as a macro which expands 2296to @code{ALL_REGS}. 2297 2298Order the classes so that if class @var{x} is contained in class @var{y} 2299then @var{x} has a lower class number than @var{y}. 2300 2301The way classes other than @code{GENERAL_REGS} are specified in operand 2302constraints is through machine-dependent operand constraint letters. 2303You can define such letters to correspond to various classes, then use 2304them in operand constraints. 2305 2306You should define a class for the union of two classes whenever some 2307instruction allows both classes. For example, if an instruction allows 2308either a floating point (coprocessor) register or a general register for a 2309certain operand, you should define a class @code{FLOAT_OR_GENERAL_REGS} 2310which includes both of them. Otherwise you will get suboptimal code. 2311 2312You must also specify certain redundant information about the register 2313classes: for each class, which classes contain it and which ones are 2314contained in it; for each pair of classes, the largest class contained 2315in their union. 2316 2317When a value occupying several consecutive registers is expected in a 2318certain class, all the registers used must belong to that class. 2319Therefore, register classes cannot be used to enforce a requirement for 2320a register pair to start with an even-numbered register. The way to 2321specify this requirement is with @code{HARD_REGNO_MODE_OK}. 2322 2323Register classes used for input-operands of bitwise-and or shift 2324instructions have a special requirement: each such class must have, for 2325each fixed-point machine mode, a subclass whose registers can transfer that 2326mode to or from memory. For example, on some machines, the operations for 2327single-byte values (@code{QImode}) are limited to certain registers. When 2328this is so, each register class that is used in a bitwise-and or shift 2329instruction must have a subclass consisting of registers from which 2330single-byte values can be loaded or stored. This is so that 2331@code{PREFERRED_RELOAD_CLASS} can always have a possible value to return. 2332 2333@deftp {Data type} {enum reg_class} 2334An enumerated type that must be defined with all the register class names 2335as enumerated values. @code{NO_REGS} must be first. @code{ALL_REGS} 2336must be the last register class, followed by one more enumerated value, 2337@code{LIM_REG_CLASSES}, which is not a register class but rather 2338tells how many classes there are. 2339 2340Each register class has a number, which is the value of casting 2341the class name to type @code{int}. The number serves as an index 2342in many of the tables described below. 2343@end deftp 2344 2345@defmac N_REG_CLASSES 2346The number of distinct register classes, defined as follows: 2347 2348@smallexample 2349#define N_REG_CLASSES (int) LIM_REG_CLASSES 2350@end smallexample 2351@end defmac 2352 2353@defmac REG_CLASS_NAMES 2354An initializer containing the names of the register classes as C string 2355constants. These names are used in writing some of the debugging dumps. 2356@end defmac 2357 2358@defmac REG_CLASS_CONTENTS 2359An initializer containing the contents of the register classes, as integers 2360which are bit masks. The @var{n}th integer specifies the contents of class 2361@var{n}. The way the integer @var{mask} is interpreted is that 2362register @var{r} is in the class if @code{@var{mask} & (1 << @var{r})} is 1. 2363 2364When the machine has more than 32 registers, an integer does not suffice. 2365Then the integers are replaced by sub-initializers, braced groupings containing 2366several integers. Each sub-initializer must be suitable as an initializer 2367for the type @code{HARD_REG_SET} which is defined in @file{hard-reg-set.h}. 2368In this situation, the first integer in each sub-initializer corresponds to 2369registers 0 through 31, the second integer to registers 32 through 63, and 2370so on. 2371@end defmac 2372 2373@defmac REGNO_REG_CLASS (@var{regno}) 2374A C expression whose value is a register class containing hard register 2375@var{regno}. In general there is more than one such class; choose a class 2376which is @dfn{minimal}, meaning that no smaller class also contains the 2377register. 2378@end defmac 2379 2380@defmac BASE_REG_CLASS 2381A macro whose definition is the name of the class to which a valid 2382base register must belong. A base register is one used in an address 2383which is the register value plus a displacement. 2384@end defmac 2385 2386@defmac MODE_BASE_REG_CLASS (@var{mode}) 2387This is a variation of the @code{BASE_REG_CLASS} macro which allows 2388the selection of a base register in a mode dependent manner. If 2389@var{mode} is VOIDmode then it should return the same value as 2390@code{BASE_REG_CLASS}. 2391@end defmac 2392 2393@defmac INDEX_REG_CLASS 2394A macro whose definition is the name of the class to which a valid 2395index register must belong. An index register is one used in an 2396address where its value is either multiplied by a scale factor or 2397added to another register (as well as added to a displacement). 2398@end defmac 2399 2400@defmac CONSTRAINT_LEN (@var{char}, @var{str}) 2401For the constraint at the start of @var{str}, which starts with the letter 2402@var{c}, return the length. This allows you to have register class / 2403constant / extra constraints that are longer than a single letter; 2404you don't need to define this macro if you can do with single-letter 2405constraints only. The definition of this macro should use 2406DEFAULT_CONSTRAINT_LEN for all the characters that you don't want 2407to handle specially. 2408There are some sanity checks in genoutput.c that check the constraint lengths 2409for the md file, so you can also use this macro to help you while you are 2410transitioning from a byzantine single-letter-constraint scheme: when you 2411return a negative length for a constraint you want to re-use, genoutput 2412will complain about every instance where it is used in the md file. 2413@end defmac 2414 2415@defmac REG_CLASS_FROM_LETTER (@var{char}) 2416A C expression which defines the machine-dependent operand constraint 2417letters for register classes. If @var{char} is such a letter, the 2418value should be the register class corresponding to it. Otherwise, 2419the value should be @code{NO_REGS}. The register letter @samp{r}, 2420corresponding to class @code{GENERAL_REGS}, will not be passed 2421to this macro; you do not need to handle it. 2422@end defmac 2423 2424@defmac REG_CLASS_FROM_CONSTRAINT (@var{char}, @var{str}) 2425Like @code{REG_CLASS_FROM_LETTER}, but you also get the constraint string 2426passed in @var{str}, so that you can use suffixes to distinguish between 2427different variants. 2428@end defmac 2429 2430@defmac REGNO_OK_FOR_BASE_P (@var{num}) 2431A C expression which is nonzero if register number @var{num} is 2432suitable for use as a base register in operand addresses. It may be 2433either a suitable hard register or a pseudo register that has been 2434allocated such a hard register. 2435@end defmac 2436 2437@defmac REGNO_MODE_OK_FOR_BASE_P (@var{num}, @var{mode}) 2438A C expression that is just like @code{REGNO_OK_FOR_BASE_P}, except that 2439that expression may examine the mode of the memory reference in 2440@var{mode}. You should define this macro if the mode of the memory 2441reference affects whether a register may be used as a base register. If 2442you define this macro, the compiler will use it instead of 2443@code{REGNO_OK_FOR_BASE_P}. 2444@end defmac 2445 2446@defmac REGNO_OK_FOR_INDEX_P (@var{num}) 2447A C expression which is nonzero if register number @var{num} is 2448suitable for use as an index register in operand addresses. It may be 2449either a suitable hard register or a pseudo register that has been 2450allocated such a hard register. 2451 2452The difference between an index register and a base register is that 2453the index register may be scaled. If an address involves the sum of 2454two registers, neither one of them scaled, then either one may be 2455labeled the ``base'' and the other the ``index''; but whichever 2456labeling is used must fit the machine's constraints of which registers 2457may serve in each capacity. The compiler will try both labelings, 2458looking for one that is valid, and will reload one or both registers 2459only if neither labeling works. 2460@end defmac 2461 2462@defmac PREFERRED_RELOAD_CLASS (@var{x}, @var{class}) 2463A C expression that places additional restrictions on the register class 2464to use when it is necessary to copy value @var{x} into a register in class 2465@var{class}. The value is a register class; perhaps @var{class}, or perhaps 2466another, smaller class. On many machines, the following definition is 2467safe: 2468 2469@smallexample 2470#define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS 2471@end smallexample 2472 2473Sometimes returning a more restrictive class makes better code. For 2474example, on the 68000, when @var{x} is an integer constant that is in range 2475for a @samp{moveq} instruction, the value of this macro is always 2476@code{DATA_REGS} as long as @var{class} includes the data registers. 2477Requiring a data register guarantees that a @samp{moveq} will be used. 2478 2479One case where @code{PREFERRED_RELOAD_CLASS} must not return 2480@var{class} is if @var{x} is a legitimate constant which cannot be 2481loaded into some register class. By returning @code{NO_REGS} you can 2482force @var{x} into a memory location. For example, rs6000 can load 2483immediate values into general-purpose registers, but does not have an 2484instruction for loading an immediate value into a floating-point 2485register, so @code{PREFERRED_RELOAD_CLASS} returns @code{NO_REGS} when 2486@var{x} is a floating-point constant. If the constant can't be loaded 2487into any kind of register, code generation will be better if 2488@code{LEGITIMATE_CONSTANT_P} makes the constant illegitimate instead 2489of using @code{PREFERRED_RELOAD_CLASS}. 2490@end defmac 2491 2492@defmac PREFERRED_OUTPUT_RELOAD_CLASS (@var{x}, @var{class}) 2493Like @code{PREFERRED_RELOAD_CLASS}, but for output reloads instead of 2494input reloads. If you don't define this macro, the default is to use 2495@var{class}, unchanged. 2496@end defmac 2497 2498@defmac LIMIT_RELOAD_CLASS (@var{mode}, @var{class}) 2499A C expression that places additional restrictions on the register class 2500to use when it is necessary to be able to hold a value of mode 2501@var{mode} in a reload register for which class @var{class} would 2502ordinarily be used. 2503 2504Unlike @code{PREFERRED_RELOAD_CLASS}, this macro should be used when 2505there are certain modes that simply can't go in certain reload classes. 2506 2507The value is a register class; perhaps @var{class}, or perhaps another, 2508smaller class. 2509 2510Don't define this macro unless the target machine has limitations which 2511require the macro to do something nontrivial. 2512@end defmac 2513 2514@defmac SECONDARY_RELOAD_CLASS (@var{class}, @var{mode}, @var{x}) 2515@defmacx SECONDARY_INPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x}) 2516@defmacx SECONDARY_OUTPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x}) 2517Many machines have some registers that cannot be copied directly to or 2518from memory or even from other types of registers. An example is the 2519@samp{MQ} register, which on most machines, can only be copied to or 2520from general registers, but not memory. Some machines allow copying all 2521registers to and from memory, but require a scratch register for stores 2522to some memory locations (e.g., those with symbolic address on the RT, 2523and those with certain symbolic address on the SPARC when compiling 2524PIC)@. In some cases, both an intermediate and a scratch register are 2525required. 2526 2527You should define these macros to indicate to the reload phase that it may 2528need to allocate at least one register for a reload in addition to the 2529register to contain the data. Specifically, if copying @var{x} to a 2530register @var{class} in @var{mode} requires an intermediate register, 2531you should define @code{SECONDARY_INPUT_RELOAD_CLASS} to return the 2532largest register class all of whose registers can be used as 2533intermediate registers or scratch registers. 2534 2535If copying a register @var{class} in @var{mode} to @var{x} requires an 2536intermediate or scratch register, @code{SECONDARY_OUTPUT_RELOAD_CLASS} 2537should be defined to return the largest register class required. If the 2538requirements for input and output reloads are the same, the macro 2539@code{SECONDARY_RELOAD_CLASS} should be used instead of defining both 2540macros identically. 2541 2542The values returned by these macros are often @code{GENERAL_REGS}. 2543Return @code{NO_REGS} if no spare register is needed; i.e., if @var{x} 2544can be directly copied to or from a register of @var{class} in 2545@var{mode} without requiring a scratch register. Do not define this 2546macro if it would always return @code{NO_REGS}. 2547 2548If a scratch register is required (either with or without an 2549intermediate register), you should define patterns for 2550@samp{reload_in@var{m}} or @samp{reload_out@var{m}}, as required 2551(@pxref{Standard Names}. These patterns, which will normally be 2552implemented with a @code{define_expand}, should be similar to the 2553@samp{mov@var{m}} patterns, except that operand 2 is the scratch 2554register. 2555 2556Define constraints for the reload register and scratch register that 2557contain a single register class. If the original reload register (whose 2558class is @var{class}) can meet the constraint given in the pattern, the 2559value returned by these macros is used for the class of the scratch 2560register. Otherwise, two additional reload registers are required. 2561Their classes are obtained from the constraints in the insn pattern. 2562 2563@var{x} might be a pseudo-register or a @code{subreg} of a 2564pseudo-register, which could either be in a hard register or in memory. 2565Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is 2566in memory and the hard register number if it is in a register. 2567 2568These macros should not be used in the case where a particular class of 2569registers can only be copied to memory and not to another class of 2570registers. In that case, secondary reload registers are not needed and 2571would not be helpful. Instead, a stack location must be used to perform 2572the copy and the @code{mov@var{m}} pattern should use memory as an 2573intermediate storage. This case often occurs between floating-point and 2574general registers. 2575@end defmac 2576 2577@defmac SECONDARY_MEMORY_NEEDED (@var{class1}, @var{class2}, @var{m}) 2578Certain machines have the property that some registers cannot be copied 2579to some other registers without using memory. Define this macro on 2580those machines to be a C expression that is nonzero if objects of mode 2581@var{m} in registers of @var{class1} can only be copied to registers of 2582class @var{class2} by storing a register of @var{class1} into memory 2583and loading that memory location into a register of @var{class2}. 2584 2585Do not define this macro if its value would always be zero. 2586@end defmac 2587 2588@defmac SECONDARY_MEMORY_NEEDED_RTX (@var{mode}) 2589Normally when @code{SECONDARY_MEMORY_NEEDED} is defined, the compiler 2590allocates a stack slot for a memory location needed for register copies. 2591If this macro is defined, the compiler instead uses the memory location 2592defined by this macro. 2593 2594Do not define this macro if you do not define 2595@code{SECONDARY_MEMORY_NEEDED}. 2596@end defmac 2597 2598@defmac SECONDARY_MEMORY_NEEDED_MODE (@var{mode}) 2599When the compiler needs a secondary memory location to copy between two 2600registers of mode @var{mode}, it normally allocates sufficient memory to 2601hold a quantity of @code{BITS_PER_WORD} bits and performs the store and 2602load operations in a mode that many bits wide and whose class is the 2603same as that of @var{mode}. 2604 2605This is right thing to do on most machines because it ensures that all 2606bits of the register are copied and prevents accesses to the registers 2607in a narrower mode, which some machines prohibit for floating-point 2608registers. 2609 2610However, this default behavior is not correct on some machines, such as 2611the DEC Alpha, that store short integers in floating-point registers 2612differently than in integer registers. On those machines, the default 2613widening will not work correctly and you must define this macro to 2614suppress that widening in some cases. See the file @file{alpha.h} for 2615details. 2616 2617Do not define this macro if you do not define 2618@code{SECONDARY_MEMORY_NEEDED} or if widening @var{mode} to a mode that 2619is @code{BITS_PER_WORD} bits wide is correct for your machine. 2620@end defmac 2621 2622@defmac SMALL_REGISTER_CLASSES 2623On some machines, it is risky to let hard registers live across arbitrary 2624insns. Typically, these machines have instructions that require values 2625to be in specific registers (like an accumulator), and reload will fail 2626if the required hard register is used for another purpose across such an 2627insn. 2628 2629Define @code{SMALL_REGISTER_CLASSES} to be an expression with a nonzero 2630value on these machines. When this macro has a nonzero value, the 2631compiler will try to minimize the lifetime of hard registers. 2632 2633It is always safe to define this macro with a nonzero value, but if you 2634unnecessarily define it, you will reduce the amount of optimizations 2635that can be performed in some cases. If you do not define this macro 2636with a nonzero value when it is required, the compiler will run out of 2637spill registers and print a fatal error message. For most machines, you 2638should not define this macro at all. 2639@end defmac 2640 2641@defmac CLASS_LIKELY_SPILLED_P (@var{class}) 2642A C expression whose value is nonzero if pseudos that have been assigned 2643to registers of class @var{class} would likely be spilled because 2644registers of @var{class} are needed for spill registers. 2645 2646The default value of this macro returns 1 if @var{class} has exactly one 2647register and zero otherwise. On most machines, this default should be 2648used. Only define this macro to some other expression if pseudos 2649allocated by @file{local-alloc.c} end up in memory because their hard 2650registers were needed for spill registers. If this macro returns nonzero 2651for those classes, those pseudos will only be allocated by 2652@file{global.c}, which knows how to reallocate the pseudo to another 2653register. If there would not be another register available for 2654reallocation, you should not change the definition of this macro since 2655the only effect of such a definition would be to slow down register 2656allocation. 2657@end defmac 2658 2659@defmac CLASS_MAX_NREGS (@var{class}, @var{mode}) 2660A C expression for the maximum number of consecutive registers 2661of class @var{class} needed to hold a value of mode @var{mode}. 2662 2663This is closely related to the macro @code{HARD_REGNO_NREGS}. In fact, 2664the value of the macro @code{CLASS_MAX_NREGS (@var{class}, @var{mode})} 2665should be the maximum value of @code{HARD_REGNO_NREGS (@var{regno}, 2666@var{mode})} for all @var{regno} values in the class @var{class}. 2667 2668This macro helps control the handling of multiple-word values 2669in the reload pass. 2670@end defmac 2671 2672@defmac CANNOT_CHANGE_MODE_CLASS (@var{from}, @var{to}, @var{class}) 2673If defined, a C expression that returns nonzero for a @var{class} for which 2674a change from mode @var{from} to mode @var{to} is invalid. 2675 2676For the example, loading 32-bit integer or floating-point objects into 2677floating-point registers on the Alpha extends them to 64 bits. 2678Therefore loading a 64-bit object and then storing it as a 32-bit object 2679does not store the low-order 32 bits, as would be the case for a normal 2680register. Therefore, @file{alpha.h} defines @code{CANNOT_CHANGE_MODE_CLASS} 2681as below: 2682 2683@smallexample 2684#define CANNOT_CHANGE_MODE_CLASS(FROM, TO, CLASS) \ 2685 (GET_MODE_SIZE (FROM) != GET_MODE_SIZE (TO) \ 2686 ? reg_classes_intersect_p (FLOAT_REGS, (CLASS)) : 0) 2687@end smallexample 2688@end defmac 2689 2690Three other special macros describe which operands fit which constraint 2691letters. 2692 2693@defmac CONST_OK_FOR_LETTER_P (@var{value}, @var{c}) 2694A C expression that defines the machine-dependent operand constraint 2695letters (@samp{I}, @samp{J}, @samp{K}, @dots{} @samp{P}) that specify 2696particular ranges of integer values. If @var{c} is one of those 2697letters, the expression should check that @var{value}, an integer, is in 2698the appropriate range and return 1 if so, 0 otherwise. If @var{c} is 2699not one of those letters, the value should be 0 regardless of 2700@var{value}. 2701@end defmac 2702 2703@defmac CONST_OK_FOR_CONSTRAINT_P (@var{value}, @var{c}, @var{str}) 2704Like @code{CONST_OK_FOR_LETTER_P}, but you also get the constraint 2705string passed in @var{str}, so that you can use suffixes to distinguish 2706between different variants. 2707@end defmac 2708 2709@defmac CONST_DOUBLE_OK_FOR_LETTER_P (@var{value}, @var{c}) 2710A C expression that defines the machine-dependent operand constraint 2711letters that specify particular ranges of @code{const_double} values 2712(@samp{G} or @samp{H}). 2713 2714If @var{c} is one of those letters, the expression should check that 2715@var{value}, an RTX of code @code{const_double}, is in the appropriate 2716range and return 1 if so, 0 otherwise. If @var{c} is not one of those 2717letters, the value should be 0 regardless of @var{value}. 2718 2719@code{const_double} is used for all floating-point constants and for 2720@code{DImode} fixed-point constants. A given letter can accept either 2721or both kinds of values. It can use @code{GET_MODE} to distinguish 2722between these kinds. 2723@end defmac 2724 2725@defmac CONST_DOUBLE_OK_FOR_CONSTRAINT_P (@var{value}, @var{c}, @var{str}) 2726Like @code{CONST_DOUBLE_OK_FOR_LETTER_P}, but you also get the constraint 2727string passed in @var{str}, so that you can use suffixes to distinguish 2728between different variants. 2729@end defmac 2730 2731@defmac EXTRA_CONSTRAINT (@var{value}, @var{c}) 2732A C expression that defines the optional machine-dependent constraint 2733letters that can be used to segregate specific types of operands, usually 2734memory references, for the target machine. Any letter that is not 2735elsewhere defined and not matched by @code{REG_CLASS_FROM_LETTER} / 2736@code{REG_CLASS_FROM_CONSTRAINT} 2737may be used. Normally this macro will not be defined. 2738 2739If it is required for a particular target machine, it should return 1 2740if @var{value} corresponds to the operand type represented by the 2741constraint letter @var{c}. If @var{c} is not defined as an extra 2742constraint, the value returned should be 0 regardless of @var{value}. 2743 2744For example, on the ROMP, load instructions cannot have their output 2745in r0 if the memory reference contains a symbolic address. Constraint 2746letter @samp{Q} is defined as representing a memory address that does 2747@emph{not} contain a symbolic address. An alternative is specified with 2748a @samp{Q} constraint on the input and @samp{r} on the output. The next 2749alternative specifies @samp{m} on the input and a register class that 2750does not include r0 on the output. 2751@end defmac 2752 2753@defmac EXTRA_CONSTRAINT_STR (@var{value}, @var{c}, @var{str}) 2754Like @code{EXTRA_CONSTRAINT}, but you also get the constraint string passed 2755in @var{str}, so that you can use suffixes to distinguish between different 2756variants. 2757@end defmac 2758 2759@defmac EXTRA_MEMORY_CONSTRAINT (@var{c}, @var{str}) 2760A C expression that defines the optional machine-dependent constraint 2761letters, amongst those accepted by @code{EXTRA_CONSTRAINT}, that should 2762be treated like memory constraints by the reload pass. 2763 2764It should return 1 if the operand type represented by the constraint 2765at the start of @var{str}, the first letter of which is the letter @var{c}, 2766 comprises a subset of all memory references including 2767all those whose address is simply a base register. This allows the reload 2768pass to reload an operand, if it does not directly correspond to the operand 2769type of @var{c}, by copying its address into a base register. 2770 2771For example, on the S/390, some instructions do not accept arbitrary 2772memory references, but only those that do not make use of an index 2773register. The constraint letter @samp{Q} is defined via 2774@code{EXTRA_CONSTRAINT} as representing a memory address of this type. 2775If the letter @samp{Q} is marked as @code{EXTRA_MEMORY_CONSTRAINT}, 2776a @samp{Q} constraint can handle any memory operand, because the 2777reload pass knows it can be reloaded by copying the memory address 2778into a base register if required. This is analogous to the way 2779a @samp{o} constraint can handle any memory operand. 2780@end defmac 2781 2782@defmac EXTRA_ADDRESS_CONSTRAINT (@var{c}, @var{str}) 2783A C expression that defines the optional machine-dependent constraint 2784letters, amongst those accepted by @code{EXTRA_CONSTRAINT} / 2785@code{EXTRA_CONSTRAINT_STR}, that should 2786be treated like address constraints by the reload pass. 2787 2788It should return 1 if the operand type represented by the constraint 2789at the start of @var{str}, which starts with the letter @var{c}, comprises 2790a subset of all memory addresses including 2791all those that consist of just a base register. This allows the reload 2792pass to reload an operand, if it does not directly correspond to the operand 2793type of @var{str}, by copying it into a base register. 2794 2795Any constraint marked as @code{EXTRA_ADDRESS_CONSTRAINT} can only 2796be used with the @code{address_operand} predicate. It is treated 2797analogously to the @samp{p} constraint. 2798@end defmac 2799 2800@node Stack and Calling 2801@section Stack Layout and Calling Conventions 2802@cindex calling conventions 2803 2804@c prevent bad page break with this line 2805This describes the stack layout and calling conventions. 2806 2807@menu 2808* Frame Layout:: 2809* Exception Handling:: 2810* Stack Checking:: 2811* Frame Registers:: 2812* Elimination:: 2813* Stack Arguments:: 2814* Register Arguments:: 2815* Scalar Return:: 2816* Aggregate Return:: 2817* Caller Saves:: 2818* Function Entry:: 2819* Profiling:: 2820* Tail Calls:: 2821@end menu 2822 2823@node Frame Layout 2824@subsection Basic Stack Layout 2825@cindex stack frame layout 2826@cindex frame layout 2827 2828@c prevent bad page break with this line 2829Here is the basic stack layout. 2830 2831@defmac STACK_GROWS_DOWNWARD 2832Define this macro if pushing a word onto the stack moves the stack 2833pointer to a smaller address. 2834 2835When we say, ``define this macro if @dots{},'' it means that the 2836compiler checks this macro only with @code{#ifdef} so the precise 2837definition used does not matter. 2838@end defmac 2839 2840@defmac STACK_PUSH_CODE 2841This macro defines the operation used when something is pushed 2842on the stack. In RTL, a push operation will be 2843@code{(set (mem (STACK_PUSH_CODE (reg sp))) @dots{})} 2844 2845The choices are @code{PRE_DEC}, @code{POST_DEC}, @code{PRE_INC}, 2846and @code{POST_INC}. Which of these is correct depends on 2847the stack direction and on whether the stack pointer points 2848to the last item on the stack or whether it points to the 2849space for the next item on the stack. 2850 2851The default is @code{PRE_DEC} when @code{STACK_GROWS_DOWNWARD} is 2852defined, which is almost always right, and @code{PRE_INC} otherwise, 2853which is often wrong. 2854@end defmac 2855 2856@defmac FRAME_GROWS_DOWNWARD 2857Define this macro if the addresses of local variable slots are at negative 2858offsets from the frame pointer. 2859@end defmac 2860 2861@defmac ARGS_GROW_DOWNWARD 2862Define this macro if successive arguments to a function occupy decreasing 2863addresses on the stack. 2864@end defmac 2865 2866@defmac STARTING_FRAME_OFFSET 2867Offset from the frame pointer to the first local variable slot to be allocated. 2868 2869If @code{FRAME_GROWS_DOWNWARD}, find the next slot's offset by 2870subtracting the first slot's length from @code{STARTING_FRAME_OFFSET}. 2871Otherwise, it is found by adding the length of the first slot to the 2872value @code{STARTING_FRAME_OFFSET}. 2873@c i'm not sure if the above is still correct.. had to change it to get 2874@c rid of an overfull. --mew 2feb93 2875@end defmac 2876 2877@defmac STACK_ALIGNMENT_NEEDED 2878Define to zero to disable final alignment of the stack during reload. 2879The nonzero default for this macro is suitable for most ports. 2880 2881On ports where @code{STARTING_FRAME_OFFSET} is nonzero or where there 2882is a register save block following the local block that doesn't require 2883alignment to @code{STACK_BOUNDARY}, it may be beneficial to disable 2884stack alignment and do it in the backend. 2885@end defmac 2886 2887@defmac STACK_POINTER_OFFSET 2888Offset from the stack pointer register to the first location at which 2889outgoing arguments are placed. If not specified, the default value of 2890zero is used. This is the proper value for most machines. 2891 2892If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above 2893the first location at which outgoing arguments are placed. 2894@end defmac 2895 2896@defmac FIRST_PARM_OFFSET (@var{fundecl}) 2897Offset from the argument pointer register to the first argument's 2898address. On some machines it may depend on the data type of the 2899function. 2900 2901If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above 2902the first argument's address. 2903@end defmac 2904 2905@defmac STACK_DYNAMIC_OFFSET (@var{fundecl}) 2906Offset from the stack pointer register to an item dynamically allocated 2907on the stack, e.g., by @code{alloca}. 2908 2909The default value for this macro is @code{STACK_POINTER_OFFSET} plus the 2910length of the outgoing arguments. The default is correct for most 2911machines. See @file{function.c} for details. 2912@end defmac 2913 2914@defmac DYNAMIC_CHAIN_ADDRESS (@var{frameaddr}) 2915A C expression whose value is RTL representing the address in a stack 2916frame where the pointer to the caller's frame is stored. Assume that 2917@var{frameaddr} is an RTL expression for the address of the stack frame 2918itself. 2919 2920If you don't define this macro, the default is to return the value 2921of @var{frameaddr}---that is, the stack frame address is also the 2922address of the stack word that points to the previous frame. 2923@end defmac 2924 2925@defmac SETUP_FRAME_ADDRESSES 2926If defined, a C expression that produces the machine-specific code to 2927setup the stack so that arbitrary frames can be accessed. For example, 2928on the SPARC, we must flush all of the register windows to the stack 2929before we can access arbitrary stack frames. You will seldom need to 2930define this macro. 2931@end defmac 2932 2933@defmac BUILTIN_SETJMP_FRAME_VALUE 2934If defined, a C expression that contains an rtx that is used to store 2935the address of the current frame into the built in @code{setjmp} buffer. 2936The default value, @code{virtual_stack_vars_rtx}, is correct for most 2937machines. One reason you may need to define this macro is if 2938@code{hard_frame_pointer_rtx} is the appropriate value on your machine. 2939@end defmac 2940 2941@defmac RETURN_ADDR_RTX (@var{count}, @var{frameaddr}) 2942A C expression whose value is RTL representing the value of the return 2943address for the frame @var{count} steps up from the current frame, after 2944the prologue. @var{frameaddr} is the frame pointer of the @var{count} 2945frame, or the frame pointer of the @var{count} @minus{} 1 frame if 2946@code{RETURN_ADDR_IN_PREVIOUS_FRAME} is defined. 2947 2948The value of the expression must always be the correct address when 2949@var{count} is zero, but may be @code{NULL_RTX} if there is not way to 2950determine the return address of other frames. 2951@end defmac 2952 2953@defmac RETURN_ADDR_IN_PREVIOUS_FRAME 2954Define this if the return address of a particular stack frame is accessed 2955from the frame pointer of the previous stack frame. 2956@end defmac 2957 2958@defmac INCOMING_RETURN_ADDR_RTX 2959A C expression whose value is RTL representing the location of the 2960incoming return address at the beginning of any function, before the 2961prologue. This RTL is either a @code{REG}, indicating that the return 2962value is saved in @samp{REG}, or a @code{MEM} representing a location in 2963the stack. 2964 2965You only need to define this macro if you want to support call frame 2966debugging information like that provided by DWARF 2. 2967 2968If this RTL is a @code{REG}, you should also define 2969@code{DWARF_FRAME_RETURN_COLUMN} to @code{DWARF_FRAME_REGNUM (REGNO)}. 2970@end defmac 2971 2972@defmac DWARF_ALT_FRAME_RETURN_COLUMN 2973A C expression whose value is an integer giving a DWARF 2 column 2974number that may be used as an alternate return column. This should 2975be defined only if @code{DWARF_FRAME_RETURN_COLUMN} is set to a 2976general register, but an alternate column needs to be used for 2977signal frames. 2978@end defmac 2979 2980@defmac DWARF_ZERO_REG 2981A C expression whose value is an integer giving a DWARF 2 register 2982number that is considered to always have the value zero. This should 2983only be defined if the target has an architected zero register, and 2984someone decided it was a good idea to use that register number to 2985terminate the stack backtrace. New ports should avoid this. 2986@end defmac 2987 2988@defmac INCOMING_FRAME_SP_OFFSET 2989A C expression whose value is an integer giving the offset, in bytes, 2990from the value of the stack pointer register to the top of the stack 2991frame at the beginning of any function, before the prologue. The top of 2992the frame is defined to be the value of the stack pointer in the 2993previous frame, just before the call instruction. 2994 2995You only need to define this macro if you want to support call frame 2996debugging information like that provided by DWARF 2. 2997@end defmac 2998 2999@defmac ARG_POINTER_CFA_OFFSET (@var{fundecl}) 3000A C expression whose value is an integer giving the offset, in bytes, 3001from the argument pointer to the canonical frame address (cfa). The 3002final value should coincide with that calculated by 3003@code{INCOMING_FRAME_SP_OFFSET}. Which is unfortunately not usable 3004during virtual register instantiation. 3005 3006The default value for this macro is @code{FIRST_PARM_OFFSET (fundecl)}, 3007which is correct for most machines; in general, the arguments are found 3008immediately before the stack frame. Note that this is not the case on 3009some targets that save registers into the caller's frame, such as SPARC 3010and rs6000, and so such targets need to define this macro. 3011 3012You only need to define this macro if the default is incorrect, and you 3013want to support call frame debugging information like that provided by 3014DWARF 2. 3015@end defmac 3016 3017@node Exception Handling 3018@subsection Exception Handling Support 3019@cindex exception handling 3020 3021@defmac EH_RETURN_DATA_REGNO (@var{N}) 3022A C expression whose value is the @var{N}th register number used for 3023data by exception handlers, or @code{INVALID_REGNUM} if fewer than 3024@var{N} registers are usable. 3025 3026The exception handling library routines communicate with the exception 3027handlers via a set of agreed upon registers. Ideally these registers 3028should be call-clobbered; it is possible to use call-saved registers, 3029but may negatively impact code size. The target must support at least 30302 data registers, but should define 4 if there are enough free registers. 3031 3032You must define this macro if you want to support call frame exception 3033handling like that provided by DWARF 2. 3034@end defmac 3035 3036@defmac EH_RETURN_STACKADJ_RTX 3037A C expression whose value is RTL representing a location in which 3038to store a stack adjustment to be applied before function return. 3039This is used to unwind the stack to an exception handler's call frame. 3040It will be assigned zero on code paths that return normally. 3041 3042Typically this is a call-clobbered hard register that is otherwise 3043untouched by the epilogue, but could also be a stack slot. 3044 3045Do not define this macro if the stack pointer is saved and restored 3046by the regular prolog and epilog code in the call frame itself; in 3047this case, the exception handling library routines will update the 3048stack location to be restored in place. Otherwise, you must define 3049this macro if you want to support call frame exception handling like 3050that provided by DWARF 2. 3051@end defmac 3052 3053@defmac EH_RETURN_HANDLER_RTX 3054A C expression whose value is RTL representing a location in which 3055to store the address of an exception handler to which we should 3056return. It will not be assigned on code paths that return normally. 3057 3058Typically this is the location in the call frame at which the normal 3059return address is stored. For targets that return by popping an 3060address off the stack, this might be a memory address just below 3061the @emph{target} call frame rather than inside the current call 3062frame. If defined, @code{EH_RETURN_STACKADJ_RTX} will have already 3063been assigned, so it may be used to calculate the location of the 3064target call frame. 3065 3066Some targets have more complex requirements than storing to an 3067address calculable during initial code generation. In that case 3068the @code{eh_return} instruction pattern should be used instead. 3069 3070If you want to support call frame exception handling, you must 3071define either this macro or the @code{eh_return} instruction pattern. 3072@end defmac 3073 3074@defmac RETURN_ADDR_OFFSET 3075If defined, an integer-valued C expression for which rtl will be generated 3076to add it to the exception handler address before it is searched in the 3077exception handling tables, and to subtract it again from the address before 3078using it to return to the exception handler. 3079@end defmac 3080 3081@defmac ASM_PREFERRED_EH_DATA_FORMAT (@var{code}, @var{global}) 3082This macro chooses the encoding of pointers embedded in the exception 3083handling sections. If at all possible, this should be defined such 3084that the exception handling section will not require dynamic relocations, 3085and so may be read-only. 3086 3087@var{code} is 0 for data, 1 for code labels, 2 for function pointers. 3088@var{global} is true if the symbol may be affected by dynamic relocations. 3089The macro should return a combination of the @code{DW_EH_PE_*} defines 3090as found in @file{dwarf2.h}. 3091 3092If this macro is not defined, pointers will not be encoded but 3093represented directly. 3094@end defmac 3095 3096@defmac ASM_MAYBE_OUTPUT_ENCODED_ADDR_RTX (@var{file}, @var{encoding}, @var{size}, @var{addr}, @var{done}) 3097This macro allows the target to emit whatever special magic is required 3098to represent the encoding chosen by @code{ASM_PREFERRED_EH_DATA_FORMAT}. 3099Generic code takes care of pc-relative and indirect encodings; this must 3100be defined if the target uses text-relative or data-relative encodings. 3101 3102This is a C statement that branches to @var{done} if the format was 3103handled. @var{encoding} is the format chosen, @var{size} is the number 3104of bytes that the format occupies, @var{addr} is the @code{SYMBOL_REF} 3105to be emitted. 3106@end defmac 3107 3108@defmac MD_FALLBACK_FRAME_STATE_FOR (@var{context}, @var{fs}, @var{success}) 3109This macro allows the target to add cpu and operating system specific 3110code to the call-frame unwinder for use when there is no unwind data 3111available. The most common reason to implement this macro is to unwind 3112through signal frames. 3113 3114This macro is called from @code{uw_frame_state_for} in @file{unwind-dw2.c} 3115and @file{unwind-ia64.c}. @var{context} is an @code{_Unwind_Context}; 3116@var{fs} is an @code{_Unwind_FrameState}. Examine @code{context->ra} 3117for the address of the code being executed and @code{context->cfa} for 3118the stack pointer value. If the frame can be decoded, the register save 3119addresses should be updated in @var{fs} and the macro should branch to 3120@var{success}. If the frame cannot be decoded, the macro should do 3121nothing. 3122 3123For proper signal handling in Java this macro is accompanied by 3124@code{MAKE_THROW_FRAME}, defined in @file{libjava/include/*-signal.h} headers. 3125@end defmac 3126 3127@defmac MD_HANDLE_UNWABI (@var{context}, @var{fs}) 3128This macro allows the target to add operating system specific code to the 3129call-frame unwinder to handle the IA-64 @code{.unwabi} unwinding directive, 3130usually used for signal or interrupt frames. 3131 3132This macro is called from @code{uw_update_context} in @file{unwind-ia64.c}. 3133@var{context} is an @code{_Unwind_Context}; 3134@var{fs} is an @code{_Unwind_FrameState}. Examine @code{fs->unwabi} 3135for the abi and context in the @code{.unwabi} directive. If the 3136@code{.unwabi} directive can be handled, the register save addresses should 3137be updated in @var{fs}. 3138@end defmac 3139 3140@node Stack Checking 3141@subsection Specifying How Stack Checking is Done 3142 3143GCC will check that stack references are within the boundaries of 3144the stack, if the @option{-fstack-check} is specified, in one of three ways: 3145 3146@enumerate 3147@item 3148If the value of the @code{STACK_CHECK_BUILTIN} macro is nonzero, GCC 3149will assume that you have arranged for stack checking to be done at 3150appropriate places in the configuration files, e.g., in 3151@code{TARGET_ASM_FUNCTION_PROLOGUE}. GCC will do not other special 3152processing. 3153 3154@item 3155If @code{STACK_CHECK_BUILTIN} is zero and you defined a named pattern 3156called @code{check_stack} in your @file{md} file, GCC will call that 3157pattern with one argument which is the address to compare the stack 3158value against. You must arrange for this pattern to report an error if 3159the stack pointer is out of range. 3160 3161@item 3162If neither of the above are true, GCC will generate code to periodically 3163``probe'' the stack pointer using the values of the macros defined below. 3164@end enumerate 3165 3166Normally, you will use the default values of these macros, so GCC 3167will use the third approach. 3168 3169@defmac STACK_CHECK_BUILTIN 3170A nonzero value if stack checking is done by the configuration files in a 3171machine-dependent manner. You should define this macro if stack checking 3172is require by the ABI of your machine or if you would like to have to stack 3173checking in some more efficient way than GCC's portable approach. 3174The default value of this macro is zero. 3175@end defmac 3176 3177@defmac STACK_CHECK_PROBE_INTERVAL 3178An integer representing the interval at which GCC must generate stack 3179probe instructions. You will normally define this macro to be no larger 3180than the size of the ``guard pages'' at the end of a stack area. The 3181default value of 4096 is suitable for most systems. 3182@end defmac 3183 3184@defmac STACK_CHECK_PROBE_LOAD 3185A integer which is nonzero if GCC should perform the stack probe 3186as a load instruction and zero if GCC should use a store instruction. 3187The default is zero, which is the most efficient choice on most systems. 3188@end defmac 3189 3190@defmac STACK_CHECK_PROTECT 3191The number of bytes of stack needed to recover from a stack overflow, 3192for languages where such a recovery is supported. The default value of 319375 words should be adequate for most machines. 3194@end defmac 3195 3196@defmac STACK_CHECK_MAX_FRAME_SIZE 3197The maximum size of a stack frame, in bytes. GCC will generate probe 3198instructions in non-leaf functions to ensure at least this many bytes of 3199stack are available. If a stack frame is larger than this size, stack 3200checking will not be reliable and GCC will issue a warning. The 3201default is chosen so that GCC only generates one instruction on most 3202systems. You should normally not change the default value of this macro. 3203@end defmac 3204 3205@defmac STACK_CHECK_FIXED_FRAME_SIZE 3206GCC uses this value to generate the above warning message. It 3207represents the amount of fixed frame used by a function, not including 3208space for any callee-saved registers, temporaries and user variables. 3209You need only specify an upper bound for this amount and will normally 3210use the default of four words. 3211@end defmac 3212 3213@defmac STACK_CHECK_MAX_VAR_SIZE 3214The maximum size, in bytes, of an object that GCC will place in the 3215fixed area of the stack frame when the user specifies 3216@option{-fstack-check}. 3217GCC computed the default from the values of the above macros and you will 3218normally not need to override that default. 3219@end defmac 3220 3221@need 2000 3222@node Frame Registers 3223@subsection Registers That Address the Stack Frame 3224 3225@c prevent bad page break with this line 3226This discusses registers that address the stack frame. 3227 3228@defmac STACK_POINTER_REGNUM 3229The register number of the stack pointer register, which must also be a 3230fixed register according to @code{FIXED_REGISTERS}. On most machines, 3231the hardware determines which register this is. 3232@end defmac 3233 3234@defmac FRAME_POINTER_REGNUM 3235The register number of the frame pointer register, which is used to 3236access automatic variables in the stack frame. On some machines, the 3237hardware determines which register this is. On other machines, you can 3238choose any register you wish for this purpose. 3239@end defmac 3240 3241@defmac HARD_FRAME_POINTER_REGNUM 3242On some machines the offset between the frame pointer and starting 3243offset of the automatic variables is not known until after register 3244allocation has been done (for example, because the saved registers are 3245between these two locations). On those machines, define 3246@code{FRAME_POINTER_REGNUM} the number of a special, fixed register to 3247be used internally until the offset is known, and define 3248@code{HARD_FRAME_POINTER_REGNUM} to be the actual hard register number 3249used for the frame pointer. 3250 3251You should define this macro only in the very rare circumstances when it 3252is not possible to calculate the offset between the frame pointer and 3253the automatic variables until after register allocation has been 3254completed. When this macro is defined, you must also indicate in your 3255definition of @code{ELIMINABLE_REGS} how to eliminate 3256@code{FRAME_POINTER_REGNUM} into either @code{HARD_FRAME_POINTER_REGNUM} 3257or @code{STACK_POINTER_REGNUM}. 3258 3259Do not define this macro if it would be the same as 3260@code{FRAME_POINTER_REGNUM}. 3261@end defmac 3262 3263@defmac ARG_POINTER_REGNUM 3264The register number of the arg pointer register, which is used to access 3265the function's argument list. On some machines, this is the same as the 3266frame pointer register. On some machines, the hardware determines which 3267register this is. On other machines, you can choose any register you 3268wish for this purpose. If this is not the same register as the frame 3269pointer register, then you must mark it as a fixed register according to 3270@code{FIXED_REGISTERS}, or arrange to be able to eliminate it 3271(@pxref{Elimination}). 3272@end defmac 3273 3274@defmac RETURN_ADDRESS_POINTER_REGNUM 3275The register number of the return address pointer register, which is used to 3276access the current function's return address from the stack. On some 3277machines, the return address is not at a fixed offset from the frame 3278pointer or stack pointer or argument pointer. This register can be defined 3279to point to the return address on the stack, and then be converted by 3280@code{ELIMINABLE_REGS} into either the frame pointer or stack pointer. 3281 3282Do not define this macro unless there is no other way to get the return 3283address from the stack. 3284@end defmac 3285 3286@defmac STATIC_CHAIN_REGNUM 3287@defmacx STATIC_CHAIN_INCOMING_REGNUM 3288Register numbers used for passing a function's static chain pointer. If 3289register windows are used, the register number as seen by the called 3290function is @code{STATIC_CHAIN_INCOMING_REGNUM}, while the register 3291number as seen by the calling function is @code{STATIC_CHAIN_REGNUM}. If 3292these registers are the same, @code{STATIC_CHAIN_INCOMING_REGNUM} need 3293not be defined. 3294 3295The static chain register need not be a fixed register. 3296 3297If the static chain is passed in memory, these macros should not be 3298defined; instead, the next two macros should be defined. 3299@end defmac 3300 3301@defmac STATIC_CHAIN 3302@defmacx STATIC_CHAIN_INCOMING 3303If the static chain is passed in memory, these macros provide rtx giving 3304@code{mem} expressions that denote where they are stored. 3305@code{STATIC_CHAIN} and @code{STATIC_CHAIN_INCOMING} give the locations 3306as seen by the calling and called functions, respectively. Often the former 3307will be at an offset from the stack pointer and the latter at an offset from 3308the frame pointer. 3309 3310@findex stack_pointer_rtx 3311@findex frame_pointer_rtx 3312@findex arg_pointer_rtx 3313The variables @code{stack_pointer_rtx}, @code{frame_pointer_rtx}, and 3314@code{arg_pointer_rtx} will have been initialized prior to the use of these 3315macros and should be used to refer to those items. 3316 3317If the static chain is passed in a register, the two previous macros should 3318be defined instead. 3319@end defmac 3320 3321@defmac DWARF_FRAME_REGISTERS 3322This macro specifies the maximum number of hard registers that can be 3323saved in a call frame. This is used to size data structures used in 3324DWARF2 exception handling. 3325 3326Prior to GCC 3.0, this macro was needed in order to establish a stable 3327exception handling ABI in the face of adding new hard registers for ISA 3328extensions. In GCC 3.0 and later, the EH ABI is insulated from changes 3329in the number of hard registers. Nevertheless, this macro can still be 3330used to reduce the runtime memory requirements of the exception handling 3331routines, which can be substantial if the ISA contains a lot of 3332registers that are not call-saved. 3333 3334If this macro is not defined, it defaults to 3335@code{FIRST_PSEUDO_REGISTER}. 3336@end defmac 3337 3338@defmac PRE_GCC3_DWARF_FRAME_REGISTERS 3339 3340This macro is similar to @code{DWARF_FRAME_REGISTERS}, but is provided 3341for backward compatibility in pre GCC 3.0 compiled code. 3342 3343If this macro is not defined, it defaults to 3344@code{DWARF_FRAME_REGISTERS}. 3345@end defmac 3346 3347@defmac DWARF_REG_TO_UNWIND_COLUMN (@var{regno}) 3348 3349Define this macro if the target's representation for dwarf registers 3350is different than the internal representation for unwind column. 3351Given a dwarf register, this macro should return the internal unwind 3352column number to use instead. 3353 3354See the PowerPC's SPE target for an example. 3355@end defmac 3356 3357@defmac DWARF_FRAME_REGNUM (@var{regno}) 3358 3359Define this macro if the target's representation for dwarf registers 3360used in .eh_frame or .debug_frame is different from that used in other 3361debug info sections. Given a GCC hard register number, this macro 3362should return the .eh_frame register number. The default is 3363@code{DBX_REGISTER_NUMBER (@var{regno})}. 3364 3365@end defmac 3366 3367@defmac DWARF2_FRAME_REG_OUT (@var{regno}, @var{for_eh}) 3368 3369Define this macro to map register numbers held in the call frame info 3370that GCC has collected using @code{DWARF_FRAME_REGNUM} to those that 3371should be output in .debug_frame (@code{@var{for_eh}} is zero) and 3372.eh_frame (@code{@var{for_eh}} is nonzero). The default is to 3373return @code{@var{regno}}. 3374 3375@end defmac 3376 3377@node Elimination 3378@subsection Eliminating Frame Pointer and Arg Pointer 3379 3380@c prevent bad page break with this line 3381This is about eliminating the frame pointer and arg pointer. 3382 3383@defmac FRAME_POINTER_REQUIRED 3384A C expression which is nonzero if a function must have and use a frame 3385pointer. This expression is evaluated in the reload pass. If its value is 3386nonzero the function will have a frame pointer. 3387 3388The expression can in principle examine the current function and decide 3389according to the facts, but on most machines the constant 0 or the 3390constant 1 suffices. Use 0 when the machine allows code to be generated 3391with no frame pointer, and doing so saves some time or space. Use 1 3392when there is no possible advantage to avoiding a frame pointer. 3393 3394In certain cases, the compiler does not know how to produce valid code 3395without a frame pointer. The compiler recognizes those cases and 3396automatically gives the function a frame pointer regardless of what 3397@code{FRAME_POINTER_REQUIRED} says. You don't need to worry about 3398them. 3399 3400In a function that does not require a frame pointer, the frame pointer 3401register can be allocated for ordinary usage, unless you mark it as a 3402fixed register. See @code{FIXED_REGISTERS} for more information. 3403@end defmac 3404 3405@findex get_frame_size 3406@defmac INITIAL_FRAME_POINTER_OFFSET (@var{depth-var}) 3407A C statement to store in the variable @var{depth-var} the difference 3408between the frame pointer and the stack pointer values immediately after 3409the function prologue. The value would be computed from information 3410such as the result of @code{get_frame_size ()} and the tables of 3411registers @code{regs_ever_live} and @code{call_used_regs}. 3412 3413If @code{ELIMINABLE_REGS} is defined, this macro will be not be used and 3414need not be defined. Otherwise, it must be defined even if 3415@code{FRAME_POINTER_REQUIRED} is defined to always be true; in that 3416case, you may set @var{depth-var} to anything. 3417@end defmac 3418 3419@defmac ELIMINABLE_REGS 3420If defined, this macro specifies a table of register pairs used to 3421eliminate unneeded registers that point into the stack frame. If it is not 3422defined, the only elimination attempted by the compiler is to replace 3423references to the frame pointer with references to the stack pointer. 3424 3425The definition of this macro is a list of structure initializations, each 3426of which specifies an original and replacement register. 3427 3428On some machines, the position of the argument pointer is not known until 3429the compilation is completed. In such a case, a separate hard register 3430must be used for the argument pointer. This register can be eliminated by 3431replacing it with either the frame pointer or the argument pointer, 3432depending on whether or not the frame pointer has been eliminated. 3433 3434In this case, you might specify: 3435@smallexample 3436#define ELIMINABLE_REGS \ 3437@{@{ARG_POINTER_REGNUM, STACK_POINTER_REGNUM@}, \ 3438 @{ARG_POINTER_REGNUM, FRAME_POINTER_REGNUM@}, \ 3439 @{FRAME_POINTER_REGNUM, STACK_POINTER_REGNUM@}@} 3440@end smallexample 3441 3442Note that the elimination of the argument pointer with the stack pointer is 3443specified first since that is the preferred elimination. 3444@end defmac 3445 3446@defmac CAN_ELIMINATE (@var{from-reg}, @var{to-reg}) 3447A C expression that returns nonzero if the compiler is allowed to try 3448to replace register number @var{from-reg} with register number 3449@var{to-reg}. This macro need only be defined if @code{ELIMINABLE_REGS} 3450is defined, and will usually be the constant 1, since most of the cases 3451preventing register elimination are things that the compiler already 3452knows about. 3453@end defmac 3454 3455@defmac INITIAL_ELIMINATION_OFFSET (@var{from-reg}, @var{to-reg}, @var{offset-var}) 3456This macro is similar to @code{INITIAL_FRAME_POINTER_OFFSET}. It 3457specifies the initial difference between the specified pair of 3458registers. This macro must be defined if @code{ELIMINABLE_REGS} is 3459defined. 3460@end defmac 3461 3462@node Stack Arguments 3463@subsection Passing Function Arguments on the Stack 3464@cindex arguments on stack 3465@cindex stack arguments 3466 3467The macros in this section control how arguments are passed 3468on the stack. See the following section for other macros that 3469control passing certain arguments in registers. 3470 3471@deftypefn {Target Hook} bool TARGET_PROMOTE_PROTOTYPES (tree @var{fntype}) 3472This target hook returns @code{true} if an argument declared in a 3473prototype as an integral type smaller than @code{int} should actually be 3474passed as an @code{int}. In addition to avoiding errors in certain 3475cases of mismatch, it also makes for better code on certain machines. 3476The default is to not promote prototypes. 3477@end deftypefn 3478 3479@defmac PUSH_ARGS 3480A C expression. If nonzero, push insns will be used to pass 3481outgoing arguments. 3482If the target machine does not have a push instruction, set it to zero. 3483That directs GCC to use an alternate strategy: to 3484allocate the entire argument block and then store the arguments into 3485it. When @code{PUSH_ARGS} is nonzero, @code{PUSH_ROUNDING} must be defined too. 3486@end defmac 3487 3488@defmac PUSH_ARGS_REVERSED 3489A C expression. If nonzero, function arguments will be evaluated from 3490last to first, rather than from first to last. If this macro is not 3491defined, it defaults to @code{PUSH_ARGS} on targets where the stack 3492and args grow in opposite directions, and 0 otherwise. 3493@end defmac 3494 3495@defmac PUSH_ROUNDING (@var{npushed}) 3496A C expression that is the number of bytes actually pushed onto the 3497stack when an instruction attempts to push @var{npushed} bytes. 3498 3499On some machines, the definition 3500 3501@smallexample 3502#define PUSH_ROUNDING(BYTES) (BYTES) 3503@end smallexample 3504 3505@noindent 3506will suffice. But on other machines, instructions that appear 3507to push one byte actually push two bytes in an attempt to maintain 3508alignment. Then the definition should be 3509 3510@smallexample 3511#define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1) 3512@end smallexample 3513@end defmac 3514 3515@findex current_function_outgoing_args_size 3516@defmac ACCUMULATE_OUTGOING_ARGS 3517A C expression. If nonzero, the maximum amount of space required for outgoing arguments 3518will be computed and placed into the variable 3519@code{current_function_outgoing_args_size}. No space will be pushed 3520onto the stack for each call; instead, the function prologue should 3521increase the stack frame size by this amount. 3522 3523Setting both @code{PUSH_ARGS} and @code{ACCUMULATE_OUTGOING_ARGS} 3524is not proper. 3525@end defmac 3526 3527@defmac REG_PARM_STACK_SPACE (@var{fndecl}) 3528Define this macro if functions should assume that stack space has been 3529allocated for arguments even when their values are passed in 3530registers. 3531 3532The value of this macro is the size, in bytes, of the area reserved for 3533arguments passed in registers for the function represented by @var{fndecl}, 3534which can be zero if GCC is calling a library function. 3535 3536This space can be allocated by the caller, or be a part of the 3537machine-dependent stack frame: @code{OUTGOING_REG_PARM_STACK_SPACE} says 3538which. 3539@end defmac 3540@c above is overfull. not sure what to do. --mew 5feb93 did 3541@c something, not sure if it looks good. --mew 10feb93 3542 3543@defmac MAYBE_REG_PARM_STACK_SPACE 3544@defmacx FINAL_REG_PARM_STACK_SPACE (@var{const_size}, @var{var_size}) 3545Define these macros in addition to the one above if functions might 3546allocate stack space for arguments even when their values are passed 3547in registers. These should be used when the stack space allocated 3548for arguments in registers is not a simple constant independent of the 3549function declaration. 3550 3551The value of the first macro is the size, in bytes, of the area that 3552we should initially assume would be reserved for arguments passed in registers. 3553 3554The value of the second macro is the actual size, in bytes, of the area 3555that will be reserved for arguments passed in registers. This takes two 3556arguments: an integer representing the number of bytes of fixed sized 3557arguments on the stack, and a tree representing the number of bytes of 3558variable sized arguments on the stack. 3559 3560When these macros are defined, @code{REG_PARM_STACK_SPACE} will only be 3561called for libcall functions, the current function, or for a function 3562being called when it is known that such stack space must be allocated. 3563In each case this value can be easily computed. 3564 3565When deciding whether a called function needs such stack space, and how 3566much space to reserve, GCC uses these two macros instead of 3567@code{REG_PARM_STACK_SPACE}. 3568@end defmac 3569 3570@defmac OUTGOING_REG_PARM_STACK_SPACE 3571Define this if it is the responsibility of the caller to allocate the area 3572reserved for arguments passed in registers. 3573 3574If @code{ACCUMULATE_OUTGOING_ARGS} is defined, this macro controls 3575whether the space for these arguments counts in the value of 3576@code{current_function_outgoing_args_size}. 3577@end defmac 3578 3579@defmac STACK_PARMS_IN_REG_PARM_AREA 3580Define this macro if @code{REG_PARM_STACK_SPACE} is defined, but the 3581stack parameters don't skip the area specified by it. 3582@c i changed this, makes more sens and it should have taken care of the 3583@c overfull.. not as specific, tho. --mew 5feb93 3584 3585Normally, when a parameter is not passed in registers, it is placed on the 3586stack beyond the @code{REG_PARM_STACK_SPACE} area. Defining this macro 3587suppresses this behavior and causes the parameter to be passed on the 3588stack in its natural location. 3589@end defmac 3590 3591@defmac RETURN_POPS_ARGS (@var{fundecl}, @var{funtype}, @var{stack-size}) 3592A C expression that should indicate the number of bytes of its own 3593arguments that a function pops on returning, or 0 if the 3594function pops no arguments and the caller must therefore pop them all 3595after the function returns. 3596 3597@var{fundecl} is a C variable whose value is a tree node that describes 3598the function in question. Normally it is a node of type 3599@code{FUNCTION_DECL} that describes the declaration of the function. 3600From this you can obtain the @code{DECL_ATTRIBUTES} of the function. 3601 3602@var{funtype} is a C variable whose value is a tree node that 3603describes the function in question. Normally it is a node of type 3604@code{FUNCTION_TYPE} that describes the data type of the function. 3605From this it is possible to obtain the data types of the value and 3606arguments (if known). 3607 3608When a call to a library function is being considered, @var{fundecl} 3609will contain an identifier node for the library function. Thus, if 3610you need to distinguish among various library functions, you can do so 3611by their names. Note that ``library function'' in this context means 3612a function used to perform arithmetic, whose name is known specially 3613in the compiler and was not mentioned in the C code being compiled. 3614 3615@var{stack-size} is the number of bytes of arguments passed on the 3616stack. If a variable number of bytes is passed, it is zero, and 3617argument popping will always be the responsibility of the calling function. 3618 3619On the VAX, all functions always pop their arguments, so the definition 3620of this macro is @var{stack-size}. On the 68000, using the standard 3621calling convention, no functions pop their arguments, so the value of 3622the macro is always 0 in this case. But an alternative calling 3623convention is available in which functions that take a fixed number of 3624arguments pop them but other functions (such as @code{printf}) pop 3625nothing (the caller pops all). When this convention is in use, 3626@var{funtype} is examined to determine whether a function takes a fixed 3627number of arguments. 3628@end defmac 3629 3630@defmac CALL_POPS_ARGS (@var{cum}) 3631A C expression that should indicate the number of bytes a call sequence 3632pops off the stack. It is added to the value of @code{RETURN_POPS_ARGS} 3633when compiling a function call. 3634 3635@var{cum} is the variable in which all arguments to the called function 3636have been accumulated. 3637 3638On certain architectures, such as the SH5, a call trampoline is used 3639that pops certain registers off the stack, depending on the arguments 3640that have been passed to the function. Since this is a property of the 3641call site, not of the called function, @code{RETURN_POPS_ARGS} is not 3642appropriate. 3643@end defmac 3644 3645@node Register Arguments 3646@subsection Passing Arguments in Registers 3647@cindex arguments in registers 3648@cindex registers arguments 3649 3650This section describes the macros which let you control how various 3651types of arguments are passed in registers or how they are arranged in 3652the stack. 3653 3654@defmac FUNCTION_ARG (@var{cum}, @var{mode}, @var{type}, @var{named}) 3655A C expression that controls whether a function argument is passed 3656in a register, and which register. 3657 3658The arguments are @var{cum}, which summarizes all the previous 3659arguments; @var{mode}, the machine mode of the argument; @var{type}, 3660the data type of the argument as a tree node or 0 if that is not known 3661(which happens for C support library functions); and @var{named}, 3662which is 1 for an ordinary argument and 0 for nameless arguments that 3663correspond to @samp{@dots{}} in the called function's prototype. 3664@var{type} can be an incomplete type if a syntax error has previously 3665occurred. 3666 3667The value of the expression is usually either a @code{reg} RTX for the 3668hard register in which to pass the argument, or zero to pass the 3669argument on the stack. 3670 3671For machines like the VAX and 68000, where normally all arguments are 3672pushed, zero suffices as a definition. 3673 3674The value of the expression can also be a @code{parallel} RTX@. This is 3675used when an argument is passed in multiple locations. The mode of the 3676@code{parallel} should be the mode of the entire argument. The 3677@code{parallel} holds any number of @code{expr_list} pairs; each one 3678describes where part of the argument is passed. In each 3679@code{expr_list} the first operand must be a @code{reg} RTX for the hard 3680register in which to pass this part of the argument, and the mode of the 3681register RTX indicates how large this part of the argument is. The 3682second operand of the @code{expr_list} is a @code{const_int} which gives 3683the offset in bytes into the entire argument of where this part starts. 3684As a special exception the first @code{expr_list} in the @code{parallel} 3685RTX may have a first operand of zero. This indicates that the entire 3686argument is also stored on the stack. 3687 3688The last time this macro is called, it is called with @code{MODE == 3689VOIDmode}, and its result is passed to the @code{call} or @code{call_value} 3690pattern as operands 2 and 3 respectively. 3691 3692@cindex @file{stdarg.h} and register arguments 3693The usual way to make the ISO library @file{stdarg.h} work on a machine 3694where some arguments are usually passed in registers, is to cause 3695nameless arguments to be passed on the stack instead. This is done 3696by making @code{FUNCTION_ARG} return 0 whenever @var{named} is 0. 3697 3698@cindex @code{MUST_PASS_IN_STACK}, and @code{FUNCTION_ARG} 3699@cindex @code{REG_PARM_STACK_SPACE}, and @code{FUNCTION_ARG} 3700You may use the macro @code{MUST_PASS_IN_STACK (@var{mode}, @var{type})} 3701in the definition of this macro to determine if this argument is of a 3702type that must be passed in the stack. If @code{REG_PARM_STACK_SPACE} 3703is not defined and @code{FUNCTION_ARG} returns nonzero for such an 3704argument, the compiler will abort. If @code{REG_PARM_STACK_SPACE} is 3705defined, the argument will be computed in the stack and then loaded into 3706a register. 3707@end defmac 3708 3709@defmac MUST_PASS_IN_STACK (@var{mode}, @var{type}) 3710Define as a C expression that evaluates to nonzero if we do not know how 3711to pass TYPE solely in registers. The file @file{expr.h} defines a 3712definition that is usually appropriate, refer to @file{expr.h} for additional 3713documentation. 3714@end defmac 3715 3716@defmac FUNCTION_INCOMING_ARG (@var{cum}, @var{mode}, @var{type}, @var{named}) 3717Define this macro if the target machine has ``register windows'', so 3718that the register in which a function sees an arguments is not 3719necessarily the same as the one in which the caller passed the 3720argument. 3721 3722For such machines, @code{FUNCTION_ARG} computes the register in which 3723the caller passes the value, and @code{FUNCTION_INCOMING_ARG} should 3724be defined in a similar fashion to tell the function being called 3725where the arguments will arrive. 3726 3727If @code{FUNCTION_INCOMING_ARG} is not defined, @code{FUNCTION_ARG} 3728serves both purposes. 3729@end defmac 3730 3731@defmac FUNCTION_ARG_PARTIAL_NREGS (@var{cum}, @var{mode}, @var{type}, @var{named}) 3732A C expression for the number of words, at the beginning of an 3733argument, that must be put in registers. The value must be zero for 3734arguments that are passed entirely in registers or that are entirely 3735pushed on the stack. 3736 3737On some machines, certain arguments must be passed partially in 3738registers and partially in memory. On these machines, typically the 3739first @var{n} words of arguments are passed in registers, and the rest 3740on the stack. If a multi-word argument (a @code{double} or a 3741structure) crosses that boundary, its first few words must be passed 3742in registers and the rest must be pushed. This macro tells the 3743compiler when this occurs, and how many of the words should go in 3744registers. 3745 3746@code{FUNCTION_ARG} for these arguments should return the first 3747register to be used by the caller for this argument; likewise 3748@code{FUNCTION_INCOMING_ARG}, for the called function. 3749@end defmac 3750 3751@defmac FUNCTION_ARG_PASS_BY_REFERENCE (@var{cum}, @var{mode}, @var{type}, @var{named}) 3752A C expression that indicates when an argument must be passed by reference. 3753If nonzero for an argument, a copy of that argument is made in memory and a 3754pointer to the argument is passed instead of the argument itself. 3755The pointer is passed in whatever way is appropriate for passing a pointer 3756to that type. 3757 3758On machines where @code{REG_PARM_STACK_SPACE} is not defined, a suitable 3759definition of this macro might be 3760@smallexample 3761#define FUNCTION_ARG_PASS_BY_REFERENCE\ 3762(CUM, MODE, TYPE, NAMED) \ 3763 MUST_PASS_IN_STACK (MODE, TYPE) 3764@end smallexample 3765@c this is *still* too long. --mew 5feb93 3766@end defmac 3767 3768@defmac FUNCTION_ARG_CALLEE_COPIES (@var{cum}, @var{mode}, @var{type}, @var{named}) 3769If defined, a C expression that indicates when it is the called function's 3770responsibility to make a copy of arguments passed by invisible reference. 3771Normally, the caller makes a copy and passes the address of the copy to the 3772routine being called. When @code{FUNCTION_ARG_CALLEE_COPIES} is defined and is 3773nonzero, the caller does not make a copy. Instead, it passes a pointer to the 3774``live'' value. The called function must not modify this value. If it can be 3775determined that the value won't be modified, it need not make a copy; 3776otherwise a copy must be made. 3777@end defmac 3778 3779@defmac CUMULATIVE_ARGS 3780A C type for declaring a variable that is used as the first argument of 3781@code{FUNCTION_ARG} and other related values. For some target machines, 3782the type @code{int} suffices and can hold the number of bytes of 3783argument so far. 3784 3785There is no need to record in @code{CUMULATIVE_ARGS} anything about the 3786arguments that have been passed on the stack. The compiler has other 3787variables to keep track of that. For target machines on which all 3788arguments are passed on the stack, there is no need to store anything in 3789@code{CUMULATIVE_ARGS}; however, the data structure must exist and 3790should not be empty, so use @code{int}. 3791@end defmac 3792 3793@defmac INIT_CUMULATIVE_ARGS (@var{cum}, @var{fntype}, @var{libname}, @var{fndecl}, @var{n_named_args}) 3794A C statement (sans semicolon) for initializing the variable 3795@var{cum} for the state at the beginning of the argument list. The 3796variable has type @code{CUMULATIVE_ARGS}. The value of @var{fntype} 3797is the tree node for the data type of the function which will receive 3798the args, or 0 if the args are to a compiler support library function. 3799For direct calls that are not libcalls, @var{fndecl} contain the 3800declaration node of the function. @var{fndecl} is also set when 3801@code{INIT_CUMULATIVE_ARGS} is used to find arguments for the function 3802being compiled. @var{n_named_args} is set to the number of named 3803arguments, including a structure return address if it is passed as a 3804parameter, when making a call. When processing incoming arguments, 3805@var{n_named_args} is set to -1. 3806 3807When processing a call to a compiler support library function, 3808@var{libname} identifies which one. It is a @code{symbol_ref} rtx which 3809contains the name of the function, as a string. @var{libname} is 0 when 3810an ordinary C function call is being processed. Thus, each time this 3811macro is called, either @var{libname} or @var{fntype} is nonzero, but 3812never both of them at once. 3813@end defmac 3814 3815@defmac INIT_CUMULATIVE_LIBCALL_ARGS (@var{cum}, @var{mode}, @var{libname}) 3816Like @code{INIT_CUMULATIVE_ARGS} but only used for outgoing libcalls, 3817it gets a @code{MODE} argument instead of @var{fntype}, that would be 3818@code{NULL}. @var{indirect} would always be zero, too. If this macro 3819is not defined, @code{INIT_CUMULATIVE_ARGS (cum, NULL_RTX, libname, 38200)} is used instead. 3821@end defmac 3822 3823@defmac INIT_CUMULATIVE_INCOMING_ARGS (@var{cum}, @var{fntype}, @var{libname}) 3824Like @code{INIT_CUMULATIVE_ARGS} but overrides it for the purposes of 3825finding the arguments for the function being compiled. If this macro is 3826undefined, @code{INIT_CUMULATIVE_ARGS} is used instead. 3827 3828The value passed for @var{libname} is always 0, since library routines 3829with special calling conventions are never compiled with GCC@. The 3830argument @var{libname} exists for symmetry with 3831@code{INIT_CUMULATIVE_ARGS}. 3832@c could use "this macro" in place of @code{INIT_CUMULATIVE_ARGS}, maybe. 3833@c --mew 5feb93 i switched the order of the sentences. --mew 10feb93 3834@end defmac 3835 3836@defmac FUNCTION_ARG_ADVANCE (@var{cum}, @var{mode}, @var{type}, @var{named}) 3837A C statement (sans semicolon) to update the summarizer variable 3838@var{cum} to advance past an argument in the argument list. The 3839values @var{mode}, @var{type} and @var{named} describe that argument. 3840Once this is done, the variable @var{cum} is suitable for analyzing 3841the @emph{following} argument with @code{FUNCTION_ARG}, etc. 3842 3843This macro need not do anything if the argument in question was passed 3844on the stack. The compiler knows how to track the amount of stack space 3845used for arguments without any special help. 3846@end defmac 3847 3848@defmac FUNCTION_ARG_PADDING (@var{mode}, @var{type}) 3849If defined, a C expression which determines whether, and in which direction, 3850to pad out an argument with extra space. The value should be of type 3851@code{enum direction}: either @code{upward} to pad above the argument, 3852@code{downward} to pad below, or @code{none} to inhibit padding. 3853 3854The @emph{amount} of padding is always just enough to reach the next 3855multiple of @code{FUNCTION_ARG_BOUNDARY}; this macro does not control 3856it. 3857 3858This macro has a default definition which is right for most systems. 3859For little-endian machines, the default is to pad upward. For 3860big-endian machines, the default is to pad downward for an argument of 3861constant size shorter than an @code{int}, and upward otherwise. 3862@end defmac 3863 3864@defmac PAD_VARARGS_DOWN 3865If defined, a C expression which determines whether the default 3866implementation of va_arg will attempt to pad down before reading the 3867next argument, if that argument is smaller than its aligned space as 3868controlled by @code{PARM_BOUNDARY}. If this macro is not defined, all such 3869arguments are padded down if @code{BYTES_BIG_ENDIAN} is true. 3870@end defmac 3871 3872@defmac BLOCK_REG_PADDING (@var{mode}, @var{type}, @var{first}) 3873Specify padding for the last element of a block move between registers and 3874memory. @var{first} is nonzero if this is the only element. Defining this 3875macro allows better control of register function parameters on big-endian 3876machines, without using @code{PARALLEL} rtl. In particular, 3877@code{MUST_PASS_IN_STACK} need not test padding and mode of types in 3878registers, as there is no longer a "wrong" part of a register; For example, 3879a three byte aggregate may be passed in the high part of a register if so 3880required. 3881@end defmac 3882 3883@defmac FUNCTION_ARG_BOUNDARY (@var{mode}, @var{type}) 3884If defined, a C expression that gives the alignment boundary, in bits, 3885of an argument with the specified mode and type. If it is not defined, 3886@code{PARM_BOUNDARY} is used for all arguments. 3887@end defmac 3888 3889@defmac FUNCTION_ARG_REGNO_P (@var{regno}) 3890A C expression that is nonzero if @var{regno} is the number of a hard 3891register in which function arguments are sometimes passed. This does 3892@emph{not} include implicit arguments such as the static chain and 3893the structure-value address. On many machines, no registers can be 3894used for this purpose since all function arguments are pushed on the 3895stack. 3896@end defmac 3897 3898@deftypefn {Target Hook} bool TARGET_SPLIT_COMPLEX_ARG (tree @var{type}) 3899This hook should return true if parameter of type @var{type} are passed 3900as two scalar parameters. By default, GCC will attempt to pack complex 3901arguments into the target's word size. Some ABIs require complex arguments 3902to be split and treated as their individual components. For example, on 3903AIX64, complex floats should be passed in a pair of floating point 3904registers, even though a complex float would fit in one 64-bit floating 3905point register. 3906 3907The default value of this hook is @code{NULL}, which is treated as always 3908false. 3909@end deftypefn 3910 3911@node Scalar Return 3912@subsection How Scalar Function Values Are Returned 3913@cindex return values in registers 3914@cindex values, returned by functions 3915@cindex scalars, returned as values 3916 3917This section discusses the macros that control returning scalars as 3918values---values that can fit in registers. 3919 3920@defmac FUNCTION_VALUE (@var{valtype}, @var{func}) 3921A C expression to create an RTX representing the place where a 3922function returns a value of data type @var{valtype}. @var{valtype} is 3923a tree node representing a data type. Write @code{TYPE_MODE 3924(@var{valtype})} to get the machine mode used to represent that type. 3925On many machines, only the mode is relevant. (Actually, on most 3926machines, scalar values are returned in the same place regardless of 3927mode). 3928 3929The value of the expression is usually a @code{reg} RTX for the hard 3930register where the return value is stored. The value can also be a 3931@code{parallel} RTX, if the return value is in multiple places. See 3932@code{FUNCTION_ARG} for an explanation of the @code{parallel} form. 3933 3934If @code{TARGET_PROMOTE_FUNCTION_RETURN} returns true, you must apply the same 3935promotion rules specified in @code{PROMOTE_MODE} if @var{valtype} is a 3936scalar type. 3937 3938If the precise function being called is known, @var{func} is a tree 3939node (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null 3940pointer. This makes it possible to use a different value-returning 3941convention for specific functions when all their calls are 3942known. 3943 3944@code{FUNCTION_VALUE} is not used for return vales with aggregate data 3945types, because these are returned in another way. See 3946@code{TARGET_STRUCT_VALUE_RTX} and related macros, below. 3947@end defmac 3948 3949@defmac FUNCTION_OUTGOING_VALUE (@var{valtype}, @var{func}) 3950Define this macro if the target machine has ``register windows'' 3951so that the register in which a function returns its value is not 3952the same as the one in which the caller sees the value. 3953 3954For such machines, @code{FUNCTION_VALUE} computes the register in which 3955the caller will see the value. @code{FUNCTION_OUTGOING_VALUE} should be 3956defined in a similar fashion to tell the function where to put the 3957value. 3958 3959If @code{FUNCTION_OUTGOING_VALUE} is not defined, 3960@code{FUNCTION_VALUE} serves both purposes. 3961 3962@code{FUNCTION_OUTGOING_VALUE} is not used for return vales with 3963aggregate data types, because these are returned in another way. See 3964@code{TARGET_STRUCT_VALUE_RTX} and related macros, below. 3965@end defmac 3966 3967@defmac LIBCALL_VALUE (@var{mode}) 3968A C expression to create an RTX representing the place where a library 3969function returns a value of mode @var{mode}. If the precise function 3970being called is known, @var{func} is a tree node 3971(@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null 3972pointer. This makes it possible to use a different value-returning 3973convention for specific functions when all their calls are 3974known. 3975 3976Note that ``library function'' in this context means a compiler 3977support routine, used to perform arithmetic, whose name is known 3978specially by the compiler and was not mentioned in the C code being 3979compiled. 3980 3981The definition of @code{LIBRARY_VALUE} need not be concerned aggregate 3982data types, because none of the library functions returns such types. 3983@end defmac 3984 3985@defmac FUNCTION_VALUE_REGNO_P (@var{regno}) 3986A C expression that is nonzero if @var{regno} is the number of a hard 3987register in which the values of called function may come back. 3988 3989A register whose use for returning values is limited to serving as the 3990second of a pair (for a value of type @code{double}, say) need not be 3991recognized by this macro. So for most machines, this definition 3992suffices: 3993 3994@smallexample 3995#define FUNCTION_VALUE_REGNO_P(N) ((N) == 0) 3996@end smallexample 3997 3998If the machine has register windows, so that the caller and the called 3999function use different registers for the return value, this macro 4000should recognize only the caller's register numbers. 4001@end defmac 4002 4003@defmac APPLY_RESULT_SIZE 4004Define this macro if @samp{untyped_call} and @samp{untyped_return} 4005need more space than is implied by @code{FUNCTION_VALUE_REGNO_P} for 4006saving and restoring an arbitrary return value. 4007@end defmac 4008 4009@deftypefn {Target Hook} bool TARGET_RETURN_IN_MSB (tree @var{type}) 4010This hook should return true if values of type @var{type} are returned 4011at the most significant end of a register (in other words, if they are 4012padded at the least significant end). You can assume that @var{type} 4013is returned in a register; the caller is required to check this. 4014 4015Note that the register provided by @code{FUNCTION_VALUE} must be able 4016to hold the complete return value. For example, if a 1-, 2- or 3-byte 4017structure is returned at the most significant end of a 4-byte register, 4018@code{FUNCTION_VALUE} should provide an @code{SImode} rtx. 4019@end deftypefn 4020 4021@node Aggregate Return 4022@subsection How Large Values Are Returned 4023@cindex aggregates as return values 4024@cindex large return values 4025@cindex returning aggregate values 4026@cindex structure value address 4027 4028When a function value's mode is @code{BLKmode} (and in some other 4029cases), the value is not returned according to @code{FUNCTION_VALUE} 4030(@pxref{Scalar Return}). Instead, the caller passes the address of a 4031block of memory in which the value should be stored. This address 4032is called the @dfn{structure value address}. 4033 4034This section describes how to control returning structure values in 4035memory. 4036 4037@deftypefn {Target Hook} bool TARGET_RETURN_IN_MEMORY (tree @var{type}, tree @var{fntype}) 4038This target hook should return a nonzero value to say to return the 4039function value in memory, just as large structures are always returned. 4040Here @var{type} will be the data type of the value, and @var{fntype} 4041will be the type of the function doing the returning, or @code{NULL} for 4042libcalls. 4043 4044Note that values of mode @code{BLKmode} must be explicitly handled 4045by this function. Also, the option @option{-fpcc-struct-return} 4046takes effect regardless of this macro. On most systems, it is 4047possible to leave the hook undefined; this causes a default 4048definition to be used, whose value is the constant 1 for @code{BLKmode} 4049values, and 0 otherwise. 4050 4051Do not use this hook to indicate that structures and unions should always 4052be returned in memory. You should instead use @code{DEFAULT_PCC_STRUCT_RETURN} 4053to indicate this. 4054@end deftypefn 4055 4056@defmac DEFAULT_PCC_STRUCT_RETURN 4057Define this macro to be 1 if all structure and union return values must be 4058in memory. Since this results in slower code, this should be defined 4059only if needed for compatibility with other compilers or with an ABI@. 4060If you define this macro to be 0, then the conventions used for structure 4061and union return values are decided by the @code{TARGET_RETURN_IN_MEMORY} 4062target hook. 4063 4064If not defined, this defaults to the value 1. 4065@end defmac 4066 4067@deftypefn {Target Hook} rtx TARGET_STRUCT_VALUE_RTX (tree @var{fndecl}, int @var{incoming}) 4068This target hook should return the location of the structure value 4069address (normally a @code{mem} or @code{reg}), or 0 if the address is 4070passed as an ``invisible'' first argument. Note that @var{fndecl} may 4071be @code{NULL}, for libcalls. 4072 4073On some architectures the place where the structure value address 4074is found by the called function is not the same place that the 4075caller put it. This can be due to register windows, or it could 4076be because the function prologue moves it to a different place. 4077@var{incoming} is @code{true} when the location is needed in 4078the context of the called function, and @code{false} in the context of 4079the caller. 4080 4081If @var{incoming} is @code{true} and the address is to be found on the 4082stack, return a @code{mem} which refers to the frame pointer. 4083@end deftypefn 4084 4085@defmac PCC_STATIC_STRUCT_RETURN 4086Define this macro if the usual system convention on the target machine 4087for returning structures and unions is for the called function to return 4088the address of a static variable containing the value. 4089 4090Do not define this if the usual system convention is for the caller to 4091pass an address to the subroutine. 4092 4093This macro has effect in @option{-fpcc-struct-return} mode, but it does 4094nothing when you use @option{-freg-struct-return} mode. 4095@end defmac 4096 4097@node Caller Saves 4098@subsection Caller-Saves Register Allocation 4099 4100If you enable it, GCC can save registers around function calls. This 4101makes it possible to use call-clobbered registers to hold variables that 4102must live across calls. 4103 4104@defmac CALLER_SAVE_PROFITABLE (@var{refs}, @var{calls}) 4105A C expression to determine whether it is worthwhile to consider placing 4106a pseudo-register in a call-clobbered hard register and saving and 4107restoring it around each function call. The expression should be 1 when 4108this is worth doing, and 0 otherwise. 4109 4110If you don't define this macro, a default is used which is good on most 4111machines: @code{4 * @var{calls} < @var{refs}}. 4112@end defmac 4113 4114@defmac HARD_REGNO_CALLER_SAVE_MODE (@var{regno}, @var{nregs}) 4115A C expression specifying which mode is required for saving @var{nregs} 4116of a pseudo-register in call-clobbered hard register @var{regno}. If 4117@var{regno} is unsuitable for caller save, @code{VOIDmode} should be 4118returned. For most machines this macro need not be defined since GCC 4119will select the smallest suitable mode. 4120@end defmac 4121 4122@node Function Entry 4123@subsection Function Entry and Exit 4124@cindex function entry and exit 4125@cindex prologue 4126@cindex epilogue 4127 4128This section describes the macros that output function entry 4129(@dfn{prologue}) and exit (@dfn{epilogue}) code. 4130 4131@deftypefn {Target Hook} void TARGET_ASM_FUNCTION_PROLOGUE (FILE *@var{file}, HOST_WIDE_INT @var{size}) 4132If defined, a function that outputs the assembler code for entry to a 4133function. The prologue is responsible for setting up the stack frame, 4134initializing the frame pointer register, saving registers that must be 4135saved, and allocating @var{size} additional bytes of storage for the 4136local variables. @var{size} is an integer. @var{file} is a stdio 4137stream to which the assembler code should be output. 4138 4139The label for the beginning of the function need not be output by this 4140macro. That has already been done when the macro is run. 4141 4142@findex regs_ever_live 4143To determine which registers to save, the macro can refer to the array 4144@code{regs_ever_live}: element @var{r} is nonzero if hard register 4145@var{r} is used anywhere within the function. This implies the function 4146prologue should save register @var{r}, provided it is not one of the 4147call-used registers. (@code{TARGET_ASM_FUNCTION_EPILOGUE} must likewise use 4148@code{regs_ever_live}.) 4149 4150On machines that have ``register windows'', the function entry code does 4151not save on the stack the registers that are in the windows, even if 4152they are supposed to be preserved by function calls; instead it takes 4153appropriate steps to ``push'' the register stack, if any non-call-used 4154registers are used in the function. 4155 4156@findex frame_pointer_needed 4157On machines where functions may or may not have frame-pointers, the 4158function entry code must vary accordingly; it must set up the frame 4159pointer if one is wanted, and not otherwise. To determine whether a 4160frame pointer is in wanted, the macro can refer to the variable 4161@code{frame_pointer_needed}. The variable's value will be 1 at run 4162time in a function that needs a frame pointer. @xref{Elimination}. 4163 4164The function entry code is responsible for allocating any stack space 4165required for the function. This stack space consists of the regions 4166listed below. In most cases, these regions are allocated in the 4167order listed, with the last listed region closest to the top of the 4168stack (the lowest address if @code{STACK_GROWS_DOWNWARD} is defined, and 4169the highest address if it is not defined). You can use a different order 4170for a machine if doing so is more convenient or required for 4171compatibility reasons. Except in cases where required by standard 4172or by a debugger, there is no reason why the stack layout used by GCC 4173need agree with that used by other compilers for a machine. 4174@end deftypefn 4175 4176@deftypefn {Target Hook} void TARGET_ASM_FUNCTION_END_PROLOGUE (FILE *@var{file}) 4177If defined, a function that outputs assembler code at the end of a 4178prologue. This should be used when the function prologue is being 4179emitted as RTL, and you have some extra assembler that needs to be 4180emitted. @xref{prologue instruction pattern}. 4181@end deftypefn 4182 4183@deftypefn {Target Hook} void TARGET_ASM_FUNCTION_BEGIN_EPILOGUE (FILE *@var{file}) 4184If defined, a function that outputs assembler code at the start of an 4185epilogue. This should be used when the function epilogue is being 4186emitted as RTL, and you have some extra assembler that needs to be 4187emitted. @xref{epilogue instruction pattern}. 4188@end deftypefn 4189 4190@deftypefn {Target Hook} void TARGET_ASM_FUNCTION_EPILOGUE (FILE *@var{file}, HOST_WIDE_INT @var{size}) 4191If defined, a function that outputs the assembler code for exit from a 4192function. The epilogue is responsible for restoring the saved 4193registers and stack pointer to their values when the function was 4194called, and returning control to the caller. This macro takes the 4195same arguments as the macro @code{TARGET_ASM_FUNCTION_PROLOGUE}, and the 4196registers to restore are determined from @code{regs_ever_live} and 4197@code{CALL_USED_REGISTERS} in the same way. 4198 4199On some machines, there is a single instruction that does all the work 4200of returning from the function. On these machines, give that 4201instruction the name @samp{return} and do not define the macro 4202@code{TARGET_ASM_FUNCTION_EPILOGUE} at all. 4203 4204Do not define a pattern named @samp{return} if you want the 4205@code{TARGET_ASM_FUNCTION_EPILOGUE} to be used. If you want the target 4206switches to control whether return instructions or epilogues are used, 4207define a @samp{return} pattern with a validity condition that tests the 4208target switches appropriately. If the @samp{return} pattern's validity 4209condition is false, epilogues will be used. 4210 4211On machines where functions may or may not have frame-pointers, the 4212function exit code must vary accordingly. Sometimes the code for these 4213two cases is completely different. To determine whether a frame pointer 4214is wanted, the macro can refer to the variable 4215@code{frame_pointer_needed}. The variable's value will be 1 when compiling 4216a function that needs a frame pointer. 4217 4218Normally, @code{TARGET_ASM_FUNCTION_PROLOGUE} and 4219@code{TARGET_ASM_FUNCTION_EPILOGUE} must treat leaf functions specially. 4220The C variable @code{current_function_is_leaf} is nonzero for such a 4221function. @xref{Leaf Functions}. 4222 4223On some machines, some functions pop their arguments on exit while 4224others leave that for the caller to do. For example, the 68020 when 4225given @option{-mrtd} pops arguments in functions that take a fixed 4226number of arguments. 4227 4228@findex current_function_pops_args 4229Your definition of the macro @code{RETURN_POPS_ARGS} decides which 4230functions pop their own arguments. @code{TARGET_ASM_FUNCTION_EPILOGUE} 4231needs to know what was decided. The variable that is called 4232@code{current_function_pops_args} is the number of bytes of its 4233arguments that a function should pop. @xref{Scalar Return}. 4234@c what is the "its arguments" in the above sentence referring to, pray 4235@c tell? --mew 5feb93 4236@end deftypefn 4237 4238@itemize @bullet 4239@item 4240@findex current_function_pretend_args_size 4241A region of @code{current_function_pretend_args_size} bytes of 4242uninitialized space just underneath the first argument arriving on the 4243stack. (This may not be at the very start of the allocated stack region 4244if the calling sequence has pushed anything else since pushing the stack 4245arguments. But usually, on such machines, nothing else has been pushed 4246yet, because the function prologue itself does all the pushing.) This 4247region is used on machines where an argument may be passed partly in 4248registers and partly in memory, and, in some cases to support the 4249features in @code{<stdarg.h>}. 4250 4251@item 4252An area of memory used to save certain registers used by the function. 4253The size of this area, which may also include space for such things as 4254the return address and pointers to previous stack frames, is 4255machine-specific and usually depends on which registers have been used 4256in the function. Machines with register windows often do not require 4257a save area. 4258 4259@item 4260A region of at least @var{size} bytes, possibly rounded up to an allocation 4261boundary, to contain the local variables of the function. On some machines, 4262this region and the save area may occur in the opposite order, with the 4263save area closer to the top of the stack. 4264 4265@item 4266@cindex @code{ACCUMULATE_OUTGOING_ARGS} and stack frames 4267Optionally, when @code{ACCUMULATE_OUTGOING_ARGS} is defined, a region of 4268@code{current_function_outgoing_args_size} bytes to be used for outgoing 4269argument lists of the function. @xref{Stack Arguments}. 4270@end itemize 4271 4272Normally, it is necessary for the macros 4273@code{TARGET_ASM_FUNCTION_PROLOGUE} and 4274@code{TARGET_ASM_FUNCTION_EPILOGUE} to treat leaf functions specially. 4275The C variable @code{current_function_is_leaf} is nonzero for such a 4276function. 4277 4278@defmac EXIT_IGNORE_STACK 4279Define this macro as a C expression that is nonzero if the return 4280instruction or the function epilogue ignores the value of the stack 4281pointer; in other words, if it is safe to delete an instruction to 4282adjust the stack pointer before a return from the function. The 4283default is 0. 4284 4285Note that this macro's value is relevant only for functions for which 4286frame pointers are maintained. It is never safe to delete a final 4287stack adjustment in a function that has no frame pointer, and the 4288compiler knows this regardless of @code{EXIT_IGNORE_STACK}. 4289@end defmac 4290 4291@defmac EPILOGUE_USES (@var{regno}) 4292Define this macro as a C expression that is nonzero for registers that are 4293used by the epilogue or the @samp{return} pattern. The stack and frame 4294pointer registers are already be assumed to be used as needed. 4295@end defmac 4296 4297@defmac EH_USES (@var{regno}) 4298Define this macro as a C expression that is nonzero for registers that are 4299used by the exception handling mechanism, and so should be considered live 4300on entry to an exception edge. 4301@end defmac 4302 4303@defmac DELAY_SLOTS_FOR_EPILOGUE 4304Define this macro if the function epilogue contains delay slots to which 4305instructions from the rest of the function can be ``moved''. The 4306definition should be a C expression whose value is an integer 4307representing the number of delay slots there. 4308@end defmac 4309 4310@defmac ELIGIBLE_FOR_EPILOGUE_DELAY (@var{insn}, @var{n}) 4311A C expression that returns 1 if @var{insn} can be placed in delay 4312slot number @var{n} of the epilogue. 4313 4314The argument @var{n} is an integer which identifies the delay slot now 4315being considered (since different slots may have different rules of 4316eligibility). It is never negative and is always less than the number 4317of epilogue delay slots (what @code{DELAY_SLOTS_FOR_EPILOGUE} returns). 4318If you reject a particular insn for a given delay slot, in principle, it 4319may be reconsidered for a subsequent delay slot. Also, other insns may 4320(at least in principle) be considered for the so far unfilled delay 4321slot. 4322 4323@findex current_function_epilogue_delay_list 4324@findex final_scan_insn 4325The insns accepted to fill the epilogue delay slots are put in an RTL 4326list made with @code{insn_list} objects, stored in the variable 4327@code{current_function_epilogue_delay_list}. The insn for the first 4328delay slot comes first in the list. Your definition of the macro 4329@code{TARGET_ASM_FUNCTION_EPILOGUE} should fill the delay slots by 4330outputting the insns in this list, usually by calling 4331@code{final_scan_insn}. 4332 4333You need not define this macro if you did not define 4334@code{DELAY_SLOTS_FOR_EPILOGUE}. 4335@end defmac 4336 4337@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}) 4338A function that outputs the assembler code for a thunk 4339function, used to implement C++ virtual function calls with multiple 4340inheritance. The thunk acts as a wrapper around a virtual function, 4341adjusting the implicit object parameter before handing control off to 4342the real function. 4343 4344First, emit code to add the integer @var{delta} to the location that 4345contains the incoming first argument. Assume that this argument 4346contains a pointer, and is the one used to pass the @code{this} pointer 4347in C++. This is the incoming argument @emph{before} the function prologue, 4348e.g.@: @samp{%o0} on a sparc. The addition must preserve the values of 4349all other incoming arguments. 4350 4351Then, if @var{vcall_offset} is nonzero, an additional adjustment should be 4352made after adding @code{delta}. In particular, if @var{p} is the 4353adjusted pointer, the following adjustment should be made: 4354 4355@smallexample 4356p += (*((ptrdiff_t **)p))[vcall_offset/sizeof(ptrdiff_t)] 4357@end smallexample 4358 4359After the additions, emit code to jump to @var{function}, which is a 4360@code{FUNCTION_DECL}. This is a direct pure jump, not a call, and does 4361not touch the return address. Hence returning from @var{FUNCTION} will 4362return to whoever called the current @samp{thunk}. 4363 4364The effect must be as if @var{function} had been called directly with 4365the adjusted first argument. This macro is responsible for emitting all 4366of the code for a thunk function; @code{TARGET_ASM_FUNCTION_PROLOGUE} 4367and @code{TARGET_ASM_FUNCTION_EPILOGUE} are not invoked. 4368 4369The @var{thunk_fndecl} is redundant. (@var{delta} and @var{function} 4370have already been extracted from it.) It might possibly be useful on 4371some targets, but probably not. 4372 4373If you do not define this macro, the target-independent code in the C++ 4374front end will generate a less efficient heavyweight thunk that calls 4375@var{function} instead of jumping to it. The generic approach does 4376not support varargs. 4377@end deftypefn 4378 4379@deftypefn {Target Hook} bool TARGET_ASM_CAN_OUTPUT_MI_THUNK (tree @var{thunk_fndecl}, HOST_WIDE_INT @var{delta}, HOST_WIDE_INT @var{vcall_offset}, tree @var{function}) 4380A function that returns true if TARGET_ASM_OUTPUT_MI_THUNK would be able 4381to output the assembler code for the thunk function specified by the 4382arguments it is passed, and false otherwise. In the latter case, the 4383generic approach will be used by the C++ front end, with the limitations 4384previously exposed. 4385@end deftypefn 4386 4387@node Profiling 4388@subsection Generating Code for Profiling 4389@cindex profiling, code generation 4390 4391These macros will help you generate code for profiling. 4392 4393@defmac FUNCTION_PROFILER (@var{file}, @var{labelno}) 4394A C statement or compound statement to output to @var{file} some 4395assembler code to call the profiling subroutine @code{mcount}. 4396 4397@findex mcount 4398The details of how @code{mcount} expects to be called are determined by 4399your operating system environment, not by GCC@. To figure them out, 4400compile a small program for profiling using the system's installed C 4401compiler and look at the assembler code that results. 4402 4403Older implementations of @code{mcount} expect the address of a counter 4404variable to be loaded into some register. The name of this variable is 4405@samp{LP} followed by the number @var{labelno}, so you would generate 4406the name using @samp{LP%d} in a @code{fprintf}. 4407@end defmac 4408 4409@defmac PROFILE_HOOK 4410A C statement or compound statement to output to @var{file} some assembly 4411code to call the profiling subroutine @code{mcount} even the target does 4412not support profiling. 4413@end defmac 4414 4415@defmac NO_PROFILE_COUNTERS 4416Define this macro if the @code{mcount} subroutine on your system does 4417not need a counter variable allocated for each function. This is true 4418for almost all modern implementations. If you define this macro, you 4419must not use the @var{labelno} argument to @code{FUNCTION_PROFILER}. 4420@end defmac 4421 4422@defmac PROFILE_BEFORE_PROLOGUE 4423Define this macro if the code for function profiling should come before 4424the function prologue. Normally, the profiling code comes after. 4425@end defmac 4426 4427@node Tail Calls 4428@subsection Permitting tail calls 4429@cindex tail calls 4430 4431@deftypefn {Target Hook} bool TARGET_FUNCTION_OK_FOR_SIBCALL (tree @var{decl}, tree @var{exp}) 4432True if it is ok to do sibling call optimization for the specified 4433call expression @var{exp}. @var{decl} will be the called function, 4434or @code{NULL} if this is an indirect call. 4435 4436It is not uncommon for limitations of calling conventions to prevent 4437tail calls to functions outside the current unit of translation, or 4438during PIC compilation. The hook is used to enforce these restrictions, 4439as the @code{sibcall} md pattern can not fail, or fall over to a 4440``normal'' call. The criteria for successful sibling call optimization 4441may vary greatly between different architectures. 4442@end deftypefn 4443 4444@node Varargs 4445@section Implementing the Varargs Macros 4446@cindex varargs implementation 4447 4448GCC comes with an implementation of @code{<varargs.h>} and 4449@code{<stdarg.h>} that work without change on machines that pass arguments 4450on the stack. Other machines require their own implementations of 4451varargs, and the two machine independent header files must have 4452conditionals to include it. 4453 4454ISO @code{<stdarg.h>} differs from traditional @code{<varargs.h>} mainly in 4455the calling convention for @code{va_start}. The traditional 4456implementation takes just one argument, which is the variable in which 4457to store the argument pointer. The ISO implementation of 4458@code{va_start} takes an additional second argument. The user is 4459supposed to write the last named argument of the function here. 4460 4461However, @code{va_start} should not use this argument. The way to find 4462the end of the named arguments is with the built-in functions described 4463below. 4464 4465@defmac __builtin_saveregs () 4466Use this built-in function to save the argument registers in memory so 4467that the varargs mechanism can access them. Both ISO and traditional 4468versions of @code{va_start} must use @code{__builtin_saveregs}, unless 4469you use @code{SETUP_INCOMING_VARARGS} (see below) instead. 4470 4471On some machines, @code{__builtin_saveregs} is open-coded under the 4472control of the macro @code{EXPAND_BUILTIN_SAVEREGS}. On other machines, 4473it calls a routine written in assembler language, found in 4474@file{libgcc2.c}. 4475 4476Code generated for the call to @code{__builtin_saveregs} appears at the 4477beginning of the function, as opposed to where the call to 4478@code{__builtin_saveregs} is written, regardless of what the code is. 4479This is because the registers must be saved before the function starts 4480to use them for its own purposes. 4481@c i rewrote the first sentence above to fix an overfull hbox. --mew 4482@c 10feb93 4483@end defmac 4484 4485@defmac __builtin_args_info (@var{category}) 4486Use this built-in function to find the first anonymous arguments in 4487registers. 4488 4489In general, a machine may have several categories of registers used for 4490arguments, each for a particular category of data types. (For example, 4491on some machines, floating-point registers are used for floating-point 4492arguments while other arguments are passed in the general registers.) 4493To make non-varargs functions use the proper calling convention, you 4494have defined the @code{CUMULATIVE_ARGS} data type to record how many 4495registers in each category have been used so far 4496 4497@code{__builtin_args_info} accesses the same data structure of type 4498@code{CUMULATIVE_ARGS} after the ordinary argument layout is finished 4499with it, with @var{category} specifying which word to access. Thus, the 4500value indicates the first unused register in a given category. 4501 4502Normally, you would use @code{__builtin_args_info} in the implementation 4503of @code{va_start}, accessing each category just once and storing the 4504value in the @code{va_list} object. This is because @code{va_list} will 4505have to update the values, and there is no way to alter the 4506values accessed by @code{__builtin_args_info}. 4507@end defmac 4508 4509@defmac __builtin_next_arg (@var{lastarg}) 4510This is the equivalent of @code{__builtin_args_info}, for stack 4511arguments. It returns the address of the first anonymous stack 4512argument, as type @code{void *}. If @code{ARGS_GROW_DOWNWARD}, it 4513returns the address of the location above the first anonymous stack 4514argument. Use it in @code{va_start} to initialize the pointer for 4515fetching arguments from the stack. Also use it in @code{va_start} to 4516verify that the second parameter @var{lastarg} is the last named argument 4517of the current function. 4518@end defmac 4519 4520@defmac __builtin_classify_type (@var{object}) 4521Since each machine has its own conventions for which data types are 4522passed in which kind of register, your implementation of @code{va_arg} 4523has to embody these conventions. The easiest way to categorize the 4524specified data type is to use @code{__builtin_classify_type} together 4525with @code{sizeof} and @code{__alignof__}. 4526 4527@code{__builtin_classify_type} ignores the value of @var{object}, 4528considering only its data type. It returns an integer describing what 4529kind of type that is---integer, floating, pointer, structure, and so on. 4530 4531The file @file{typeclass.h} defines an enumeration that you can use to 4532interpret the values of @code{__builtin_classify_type}. 4533@end defmac 4534 4535These machine description macros help implement varargs: 4536 4537@deftypefn {Target Hook} rtx TARGET_EXPAND_BUILTIN_SAVEREGS (void) 4538If defined, this hook produces the machine-specific code for a call to 4539@code{__builtin_saveregs}. This code will be moved to the very 4540beginning of the function, before any parameter access are made. The 4541return value of this function should be an RTX that contains the value 4542to use as the return of @code{__builtin_saveregs}. 4543@end deftypefn 4544 4545@deftypefn {Target Hook} void TARGET_SETUP_INCOMING_VARARGS (CUMULATIVE_ARGS *@var{args_so_far}, enum machine_mode @var{mode}, tree @var{type}, int *@var{pretend_args_size}, int @var{second_time}) 4546This target hook offers an alternative to using 4547@code{__builtin_saveregs} and defining the hook 4548@code{TARGET_EXPAND_BUILTIN_SAVEREGS}. Use it to store the anonymous 4549register arguments into the stack so that all the arguments appear to 4550have been passed consecutively on the stack. Once this is done, you can 4551use the standard implementation of varargs that works for machines that 4552pass all their arguments on the stack. 4553 4554The argument @var{args_so_far} points to the @code{CUMULATIVE_ARGS} data 4555structure, containing the values that are obtained after processing the 4556named arguments. The arguments @var{mode} and @var{type} describe the 4557last named argument---its machine mode and its data type as a tree node. 4558 4559The target hook should do two things: first, push onto the stack all the 4560argument registers @emph{not} used for the named arguments, and second, 4561store the size of the data thus pushed into the @code{int}-valued 4562variable pointed to by @var{pretend_args_size}. The value that you 4563store here will serve as additional offset for setting up the stack 4564frame. 4565 4566Because you must generate code to push the anonymous arguments at 4567compile time without knowing their data types, 4568@code{TARGET_SETUP_INCOMING_VARARGS} is only useful on machines that 4569have just a single category of argument register and use it uniformly 4570for all data types. 4571 4572If the argument @var{second_time} is nonzero, it means that the 4573arguments of the function are being analyzed for the second time. This 4574happens for an inline function, which is not actually compiled until the 4575end of the source file. The hook @code{TARGET_SETUP_INCOMING_VARARGS} should 4576not generate any instructions in this case. 4577@end deftypefn 4578 4579@deftypefn {Target Hook} bool TARGET_STRICT_ARGUMENT_NAMING (CUMULATIVE_ARGS *@var{ca}) 4580Define this hook to return @code{true} if the location where a function 4581argument is passed depends on whether or not it is a named argument. 4582 4583This hook controls how the @var{named} argument to @code{FUNCTION_ARG} 4584is set for varargs and stdarg functions. If this hook returns 4585@code{true}, the @var{named} argument is always true for named 4586arguments, and false for unnamed arguments. If it returns @code{false}, 4587but @code{TARGET_PRETEND_OUTOGOING_VARARGS_NAMED} returns @code{true}, 4588then all arguments are treated as named. Otherwise, all named arguments 4589except the last are treated as named. 4590 4591You need not define this hook if it always returns zero. 4592@end deftypefn 4593 4594@deftypefn {Target Hook} bool TARGET_PRETEND_OUTGOING_VARARGS_NAMED 4595If you need to conditionally change ABIs so that one works with 4596@code{TARGET_SETUP_INCOMING_VARARGS}, but the other works like neither 4597@code{TARGET_SETUP_INCOMING_VARARGS} nor @code{TARGET_STRICT_ARGUMENT_NAMING} was 4598defined, then define this hook to return @code{true} if 4599@code{SETUP_INCOMING_VARARGS} is used, @code{false} otherwise. 4600Otherwise, you should not define this hook. 4601@end deftypefn 4602 4603@node Trampolines 4604@section Trampolines for Nested Functions 4605@cindex trampolines for nested functions 4606@cindex nested functions, trampolines for 4607 4608A @dfn{trampoline} is a small piece of code that is created at run time 4609when the address of a nested function is taken. It normally resides on 4610the stack, in the stack frame of the containing function. These macros 4611tell GCC how to generate code to allocate and initialize a 4612trampoline. 4613 4614The instructions in the trampoline must do two things: load a constant 4615address into the static chain register, and jump to the real address of 4616the nested function. On CISC machines such as the m68k, this requires 4617two instructions, a move immediate and a jump. Then the two addresses 4618exist in the trampoline as word-long immediate operands. On RISC 4619machines, it is often necessary to load each address into a register in 4620two parts. Then pieces of each address form separate immediate 4621operands. 4622 4623The code generated to initialize the trampoline must store the variable 4624parts---the static chain value and the function address---into the 4625immediate operands of the instructions. On a CISC machine, this is 4626simply a matter of copying each address to a memory reference at the 4627proper offset from the start of the trampoline. On a RISC machine, it 4628may be necessary to take out pieces of the address and store them 4629separately. 4630 4631@defmac TRAMPOLINE_TEMPLATE (@var{file}) 4632A C statement to output, on the stream @var{file}, assembler code for a 4633block of data that contains the constant parts of a trampoline. This 4634code should not include a label---the label is taken care of 4635automatically. 4636 4637If you do not define this macro, it means no template is needed 4638for the target. Do not define this macro on systems where the block move 4639code to copy the trampoline into place would be larger than the code 4640to generate it on the spot. 4641@end defmac 4642 4643@defmac TRAMPOLINE_SECTION 4644The name of a subroutine to switch to the section in which the 4645trampoline template is to be placed (@pxref{Sections}). The default is 4646a value of @samp{readonly_data_section}, which places the trampoline in 4647the section containing read-only data. 4648@end defmac 4649 4650@defmac TRAMPOLINE_SIZE 4651A C expression for the size in bytes of the trampoline, as an integer. 4652@end defmac 4653 4654@defmac TRAMPOLINE_ALIGNMENT 4655Alignment required for trampolines, in bits. 4656 4657If you don't define this macro, the value of @code{BIGGEST_ALIGNMENT} 4658is used for aligning trampolines. 4659@end defmac 4660 4661@defmac INITIALIZE_TRAMPOLINE (@var{addr}, @var{fnaddr}, @var{static_chain}) 4662A C statement to initialize the variable parts of a trampoline. 4663@var{addr} is an RTX for the address of the trampoline; @var{fnaddr} is 4664an RTX for the address of the nested function; @var{static_chain} is an 4665RTX for the static chain value that should be passed to the function 4666when it is called. 4667@end defmac 4668 4669@defmac TRAMPOLINE_ADJUST_ADDRESS (@var{addr}) 4670A C statement that should perform any machine-specific adjustment in 4671the address of the trampoline. Its argument contains the address that 4672was passed to @code{INITIALIZE_TRAMPOLINE}. In case the address to be 4673used for a function call should be different from the address in which 4674the template was stored, the different address should be assigned to 4675@var{addr}. If this macro is not defined, @var{addr} will be used for 4676function calls. 4677 4678@cindex @code{TARGET_ASM_FUNCTION_EPILOGUE} and trampolines 4679@cindex @code{TARGET_ASM_FUNCTION_PROLOGUE} and trampolines 4680If this macro is not defined, by default the trampoline is allocated as 4681a stack slot. This default is right for most machines. The exceptions 4682are machines where it is impossible to execute instructions in the stack 4683area. On such machines, you may have to implement a separate stack, 4684using this macro in conjunction with @code{TARGET_ASM_FUNCTION_PROLOGUE} 4685and @code{TARGET_ASM_FUNCTION_EPILOGUE}. 4686 4687@var{fp} points to a data structure, a @code{struct function}, which 4688describes the compilation status of the immediate containing function of 4689the function which the trampoline is for. The stack slot for the 4690trampoline is in the stack frame of this containing function. Other 4691allocation strategies probably must do something analogous with this 4692information. 4693@end defmac 4694 4695Implementing trampolines is difficult on many machines because they have 4696separate instruction and data caches. Writing into a stack location 4697fails to clear the memory in the instruction cache, so when the program 4698jumps to that location, it executes the old contents. 4699 4700Here are two possible solutions. One is to clear the relevant parts of 4701the instruction cache whenever a trampoline is set up. The other is to 4702make all trampolines identical, by having them jump to a standard 4703subroutine. The former technique makes trampoline execution faster; the 4704latter makes initialization faster. 4705 4706To clear the instruction cache when a trampoline is initialized, define 4707the following macro. 4708 4709@defmac CLEAR_INSN_CACHE (@var{beg}, @var{end}) 4710If defined, expands to a C expression clearing the @emph{instruction 4711cache} in the specified interval. The definition of this macro would 4712typically be a series of @code{asm} statements. Both @var{beg} and 4713@var{end} are both pointer expressions. 4714@end defmac 4715 4716The operating system may also require the stack to be made executable 4717before calling the trampoline. To implement this requirement, define 4718the following macro. 4719 4720@defmac ENABLE_EXECUTE_STACK 4721Define this macro if certain operations must be performed before executing 4722code located on the stack. The macro should expand to a series of C 4723file-scope constructs (e.g. functions) and provide a unique entry point 4724named @code{__enable_execute_stack}. The target is responsible for 4725emitting calls to the entry point in the code, for example from the 4726@code{INITIALIZE_TRAMPOLINE} macro. 4727@end defmac 4728 4729To use a standard subroutine, define the following macro. In addition, 4730you must make sure that the instructions in a trampoline fill an entire 4731cache line with identical instructions, or else ensure that the 4732beginning of the trampoline code is always aligned at the same point in 4733its cache line. Look in @file{m68k.h} as a guide. 4734 4735@defmac TRANSFER_FROM_TRAMPOLINE 4736Define this macro if trampolines need a special subroutine to do their 4737work. The macro should expand to a series of @code{asm} statements 4738which will be compiled with GCC@. They go in a library function named 4739@code{__transfer_from_trampoline}. 4740 4741If you need to avoid executing the ordinary prologue code of a compiled 4742C function when you jump to the subroutine, you can do so by placing a 4743special label of your own in the assembler code. Use one @code{asm} 4744statement to generate an assembler label, and another to make the label 4745global. Then trampolines can use that label to jump directly to your 4746special assembler code. 4747@end defmac 4748 4749@node Library Calls 4750@section Implicit Calls to Library Routines 4751@cindex library subroutine names 4752@cindex @file{libgcc.a} 4753 4754@c prevent bad page break with this line 4755Here is an explanation of implicit calls to library routines. 4756 4757@defmac DECLARE_LIBRARY_RENAMES 4758This macro, if defined, should expand to a piece of C code that will get 4759expanded when compiling functions for libgcc.a. It can be used to 4760provide alternate names for GCC's internal library functions if there 4761are ABI-mandated names that the compiler should provide. 4762@end defmac 4763 4764@findex init_one_libfunc 4765@findex set_optab_libfunc 4766@deftypefn {Target Hook} void TARGET_INIT_LIBFUNCS (void) 4767This hook should declare additional library routines or rename 4768existing ones, using the functions @code{set_optab_libfunc} and 4769@code{init_one_libfunc} defined in @file{optabs.c}. 4770@code{init_optabs} calls this macro after initializing all the normal 4771library routines. 4772 4773The default is to do nothing. Most ports don't need to define this hook. 4774@end deftypefn 4775 4776@defmac TARGET_FLOAT_LIB_COMPARE_RETURNS_BOOL (@var{mode}, @var{comparison}) 4777This macro should return @code{true} if the library routine that 4778implements the floating point comparison operator @var{comparison} in 4779mode @var{mode} will return a boolean, and @var{false} if it will 4780return a tristate. 4781 4782GCC's own floating point libraries return tristates from the 4783comparison operators, so the default returns false always. Most ports 4784don't need to define this macro. 4785@end defmac 4786 4787@cindex US Software GOFAST, floating point emulation library 4788@cindex floating point emulation library, US Software GOFAST 4789@cindex GOFAST, floating point emulation library 4790@findex gofast_maybe_init_libfuncs 4791@defmac US_SOFTWARE_GOFAST 4792Define this macro if your system C library uses the US Software GOFAST 4793library to provide floating point emulation. 4794 4795In addition to defining this macro, your architecture must set 4796@code{TARGET_INIT_LIBFUNCS} to @code{gofast_maybe_init_libfuncs}, or 4797else call that function from its version of that hook. It is defined 4798in @file{config/gofast.h}, which must be included by your 4799architecture's @file{@var{cpu}.c} file. See @file{sparc/sparc.c} for 4800an example. 4801 4802If this macro is defined, the 4803@code{TARGET_FLOAT_LIB_COMPARE_RETURNS_BOOL} target hook must return 4804false for @code{SFmode} and @code{DFmode} comparisons. 4805@end defmac 4806 4807@cindex @code{EDOM}, implicit usage 4808@findex matherr 4809@defmac TARGET_EDOM 4810The value of @code{EDOM} on the target machine, as a C integer constant 4811expression. If you don't define this macro, GCC does not attempt to 4812deposit the value of @code{EDOM} into @code{errno} directly. Look in 4813@file{/usr/include/errno.h} to find the value of @code{EDOM} on your 4814system. 4815 4816If you do not define @code{TARGET_EDOM}, then compiled code reports 4817domain errors by calling the library function and letting it report the 4818error. If mathematical functions on your system use @code{matherr} when 4819there is an error, then you should leave @code{TARGET_EDOM} undefined so 4820that @code{matherr} is used normally. 4821@end defmac 4822 4823@cindex @code{errno}, implicit usage 4824@defmac GEN_ERRNO_RTX 4825Define this macro as a C expression to create an rtl expression that 4826refers to the global ``variable'' @code{errno}. (On certain systems, 4827@code{errno} may not actually be a variable.) If you don't define this 4828macro, a reasonable default is used. 4829@end defmac 4830 4831@cindex @code{bcopy}, implicit usage 4832@cindex @code{memcpy}, implicit usage 4833@cindex @code{memmove}, implicit usage 4834@cindex @code{bzero}, implicit usage 4835@cindex @code{memset}, implicit usage 4836@defmac TARGET_MEM_FUNCTIONS 4837Define this macro if GCC should generate calls to the ISO C 4838(and System V) library functions @code{memcpy}, @code{memmove} and 4839@code{memset} rather than the BSD functions @code{bcopy} and @code{bzero}. 4840@end defmac 4841 4842@cindex C99 math functions, implicit usage 4843@defmac TARGET_C99_FUNCTIONS 4844When this macro is nonzero, GCC will implicitly optimize @code{sin} calls into 4845@code{sinf} and similarly for other functions defined by C99 standard. The 4846default is nonzero that should be proper value for most modern systems, however 4847number of existing systems lacks support for these functions in the runtime so 4848they needs this macro to be redefined to 0. 4849@end defmac 4850 4851@defmac NEXT_OBJC_RUNTIME 4852Define this macro to generate code for Objective-C message sending using 4853the calling convention of the NeXT system. This calling convention 4854involves passing the object, the selector and the method arguments all 4855at once to the method-lookup library function. 4856 4857The default calling convention passes just the object and the selector 4858to the lookup function, which returns a pointer to the method. 4859@end defmac 4860 4861@node Addressing Modes 4862@section Addressing Modes 4863@cindex addressing modes 4864 4865@c prevent bad page break with this line 4866This is about addressing modes. 4867 4868@defmac HAVE_PRE_INCREMENT 4869@defmacx HAVE_PRE_DECREMENT 4870@defmacx HAVE_POST_INCREMENT 4871@defmacx HAVE_POST_DECREMENT 4872A C expression that is nonzero if the machine supports pre-increment, 4873pre-decrement, post-increment, or post-decrement addressing respectively. 4874@end defmac 4875 4876@defmac HAVE_PRE_MODIFY_DISP 4877@defmacx HAVE_POST_MODIFY_DISP 4878A C expression that is nonzero if the machine supports pre- or 4879post-address side-effect generation involving constants other than 4880the size of the memory operand. 4881@end defmac 4882 4883@defmac HAVE_PRE_MODIFY_REG 4884@defmacx HAVE_POST_MODIFY_REG 4885A C expression that is nonzero if the machine supports pre- or 4886post-address side-effect generation involving a register displacement. 4887@end defmac 4888 4889@defmac CONSTANT_ADDRESS_P (@var{x}) 4890A C expression that is 1 if the RTX @var{x} is a constant which 4891is a valid address. On most machines, this can be defined as 4892@code{CONSTANT_P (@var{x})}, but a few machines are more restrictive 4893in which constant addresses are supported. 4894@end defmac 4895 4896@defmac CONSTANT_P (@var{x}) 4897@code{CONSTANT_P}, which is defined by target-independent code, 4898accepts integer-values expressions whose values are not explicitly 4899known, such as @code{symbol_ref}, @code{label_ref}, and @code{high} 4900expressions and @code{const} arithmetic expressions, in addition to 4901@code{const_int} and @code{const_double} expressions. 4902@end defmac 4903 4904@defmac MAX_REGS_PER_ADDRESS 4905A number, the maximum number of registers that can appear in a valid 4906memory address. Note that it is up to you to specify a value equal to 4907the maximum number that @code{GO_IF_LEGITIMATE_ADDRESS} would ever 4908accept. 4909@end defmac 4910 4911@defmac GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{label}) 4912A C compound statement with a conditional @code{goto @var{label};} 4913executed if @var{x} (an RTX) is a legitimate memory address on the 4914target machine for a memory operand of mode @var{mode}. 4915 4916It usually pays to define several simpler macros to serve as 4917subroutines for this one. Otherwise it may be too complicated to 4918understand. 4919 4920This macro must exist in two variants: a strict variant and a 4921non-strict one. The strict variant is used in the reload pass. It 4922must be defined so that any pseudo-register that has not been 4923allocated a hard register is considered a memory reference. In 4924contexts where some kind of register is required, a pseudo-register 4925with no hard register must be rejected. 4926 4927The non-strict variant is used in other passes. It must be defined to 4928accept all pseudo-registers in every context where some kind of 4929register is required. 4930 4931@findex REG_OK_STRICT 4932Compiler source files that want to use the strict variant of this 4933macro define the macro @code{REG_OK_STRICT}. You should use an 4934@code{#ifdef REG_OK_STRICT} conditional to define the strict variant 4935in that case and the non-strict variant otherwise. 4936 4937Subroutines to check for acceptable registers for various purposes (one 4938for base registers, one for index registers, and so on) are typically 4939among the subroutines used to define @code{GO_IF_LEGITIMATE_ADDRESS}. 4940Then only these subroutine macros need have two variants; the higher 4941levels of macros may be the same whether strict or not. 4942 4943Normally, constant addresses which are the sum of a @code{symbol_ref} 4944and an integer are stored inside a @code{const} RTX to mark them as 4945constant. Therefore, there is no need to recognize such sums 4946specifically as legitimate addresses. Normally you would simply 4947recognize any @code{const} as legitimate. 4948 4949Usually @code{PRINT_OPERAND_ADDRESS} is not prepared to handle constant 4950sums that are not marked with @code{const}. It assumes that a naked 4951@code{plus} indicates indexing. If so, then you @emph{must} reject such 4952naked constant sums as illegitimate addresses, so that none of them will 4953be given to @code{PRINT_OPERAND_ADDRESS}. 4954 4955@cindex @code{TARGET_ENCODE_SECTION_INFO} and address validation 4956On some machines, whether a symbolic address is legitimate depends on 4957the section that the address refers to. On these machines, define the 4958target hook @code{TARGET_ENCODE_SECTION_INFO} to store the information 4959into the @code{symbol_ref}, and then check for it here. When you see a 4960@code{const}, you will have to look inside it to find the 4961@code{symbol_ref} in order to determine the section. @xref{Assembler 4962Format}. 4963@end defmac 4964 4965@defmac REG_OK_FOR_BASE_P (@var{x}) 4966A C expression that is nonzero if @var{x} (assumed to be a @code{reg} 4967RTX) is valid for use as a base register. For hard registers, it 4968should always accept those which the hardware permits and reject the 4969others. Whether the macro accepts or rejects pseudo registers must be 4970controlled by @code{REG_OK_STRICT} as described above. This usually 4971requires two variant definitions, of which @code{REG_OK_STRICT} 4972controls the one actually used. 4973@end defmac 4974 4975@defmac REG_MODE_OK_FOR_BASE_P (@var{x}, @var{mode}) 4976A C expression that is just like @code{REG_OK_FOR_BASE_P}, except that 4977that expression may examine the mode of the memory reference in 4978@var{mode}. You should define this macro if the mode of the memory 4979reference affects whether a register may be used as a base register. If 4980you define this macro, the compiler will use it instead of 4981@code{REG_OK_FOR_BASE_P}. 4982@end defmac 4983 4984@defmac REG_OK_FOR_INDEX_P (@var{x}) 4985A C expression that is nonzero if @var{x} (assumed to be a @code{reg} 4986RTX) is valid for use as an index register. 4987 4988The difference between an index register and a base register is that 4989the index register may be scaled. If an address involves the sum of 4990two registers, neither one of them scaled, then either one may be 4991labeled the ``base'' and the other the ``index''; but whichever 4992labeling is used must fit the machine's constraints of which registers 4993may serve in each capacity. The compiler will try both labelings, 4994looking for one that is valid, and will reload one or both registers 4995only if neither labeling works. 4996@end defmac 4997 4998@defmac FIND_BASE_TERM (@var{x}) 4999A C expression to determine the base term of address @var{x}. 5000This macro is used in only one place: `find_base_term' in alias.c. 5001 5002It is always safe for this macro to not be defined. It exists so 5003that alias analysis can understand machine-dependent addresses. 5004 5005The typical use of this macro is to handle addresses containing 5006a label_ref or symbol_ref within an UNSPEC@. 5007@end defmac 5008 5009@defmac LEGITIMIZE_ADDRESS (@var{x}, @var{oldx}, @var{mode}, @var{win}) 5010A C compound statement that attempts to replace @var{x} with a valid 5011memory address for an operand of mode @var{mode}. @var{win} will be a 5012C statement label elsewhere in the code; the macro definition may use 5013 5014@smallexample 5015GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{win}); 5016@end smallexample 5017 5018@noindent 5019to avoid further processing if the address has become legitimate. 5020 5021@findex break_out_memory_refs 5022@var{x} will always be the result of a call to @code{break_out_memory_refs}, 5023and @var{oldx} will be the operand that was given to that function to produce 5024@var{x}. 5025 5026The code generated by this macro should not alter the substructure of 5027@var{x}. If it transforms @var{x} into a more legitimate form, it 5028should assign @var{x} (which will always be a C variable) a new value. 5029 5030It is not necessary for this macro to come up with a legitimate 5031address. The compiler has standard ways of doing so in all cases. In 5032fact, it is safe for this macro to do nothing. But often a 5033machine-dependent strategy can generate better code. 5034@end defmac 5035 5036@defmac LEGITIMIZE_RELOAD_ADDRESS (@var{x}, @var{mode}, @var{opnum}, @var{type}, @var{ind_levels}, @var{win}) 5037A C compound statement that attempts to replace @var{x}, which is an address 5038that needs reloading, with a valid memory address for an operand of mode 5039@var{mode}. @var{win} will be a C statement label elsewhere in the code. 5040It is not necessary to define this macro, but it might be useful for 5041performance reasons. 5042 5043For example, on the i386, it is sometimes possible to use a single 5044reload register instead of two by reloading a sum of two pseudo 5045registers into a register. On the other hand, for number of RISC 5046processors offsets are limited so that often an intermediate address 5047needs to be generated in order to address a stack slot. By defining 5048@code{LEGITIMIZE_RELOAD_ADDRESS} appropriately, the intermediate addresses 5049generated for adjacent some stack slots can be made identical, and thus 5050be shared. 5051 5052@emph{Note}: This macro should be used with caution. It is necessary 5053to know something of how reload works in order to effectively use this, 5054and it is quite easy to produce macros that build in too much knowledge 5055of reload internals. 5056 5057@emph{Note}: This macro must be able to reload an address created by a 5058previous invocation of this macro. If it fails to handle such addresses 5059then the compiler may generate incorrect code or abort. 5060 5061@findex push_reload 5062The macro definition should use @code{push_reload} to indicate parts that 5063need reloading; @var{opnum}, @var{type} and @var{ind_levels} are usually 5064suitable to be passed unaltered to @code{push_reload}. 5065 5066The code generated by this macro must not alter the substructure of 5067@var{x}. If it transforms @var{x} into a more legitimate form, it 5068should assign @var{x} (which will always be a C variable) a new value. 5069This also applies to parts that you change indirectly by calling 5070@code{push_reload}. 5071 5072@findex strict_memory_address_p 5073The macro definition may use @code{strict_memory_address_p} to test if 5074the address has become legitimate. 5075 5076@findex copy_rtx 5077If you want to change only a part of @var{x}, one standard way of doing 5078this is to use @code{copy_rtx}. Note, however, that is unshares only a 5079single level of rtl. Thus, if the part to be changed is not at the 5080top level, you'll need to replace first the top level. 5081It is not necessary for this macro to come up with a legitimate 5082address; but often a machine-dependent strategy can generate better code. 5083@end defmac 5084 5085@defmac GO_IF_MODE_DEPENDENT_ADDRESS (@var{addr}, @var{label}) 5086A C statement or compound statement with a conditional @code{goto 5087@var{label};} executed if memory address @var{x} (an RTX) can have 5088different meanings depending on the machine mode of the memory 5089reference it is used for or if the address is valid for some modes 5090but not others. 5091 5092Autoincrement and autodecrement addresses typically have mode-dependent 5093effects because the amount of the increment or decrement is the size 5094of the operand being addressed. Some machines have other mode-dependent 5095addresses. Many RISC machines have no mode-dependent addresses. 5096 5097You may assume that @var{addr} is a valid address for the machine. 5098@end defmac 5099 5100@defmac LEGITIMATE_CONSTANT_P (@var{x}) 5101A C expression that is nonzero if @var{x} is a legitimate constant for 5102an immediate operand on the target machine. You can assume that 5103@var{x} satisfies @code{CONSTANT_P}, so you need not check this. In fact, 5104@samp{1} is a suitable definition for this macro on machines where 5105anything @code{CONSTANT_P} is valid. 5106@end defmac 5107 5108@node Condition Code 5109@section Condition Code Status 5110@cindex condition code status 5111 5112@c prevent bad page break with this line 5113This describes the condition code status. 5114 5115@findex cc_status 5116The file @file{conditions.h} defines a variable @code{cc_status} to 5117describe how the condition code was computed (in case the interpretation of 5118the condition code depends on the instruction that it was set by). This 5119variable contains the RTL expressions on which the condition code is 5120currently based, and several standard flags. 5121 5122Sometimes additional machine-specific flags must be defined in the machine 5123description header file. It can also add additional machine-specific 5124information by defining @code{CC_STATUS_MDEP}. 5125 5126@defmac CC_STATUS_MDEP 5127C code for a data type which is used for declaring the @code{mdep} 5128component of @code{cc_status}. It defaults to @code{int}. 5129 5130This macro is not used on machines that do not use @code{cc0}. 5131@end defmac 5132 5133@defmac CC_STATUS_MDEP_INIT 5134A C expression to initialize the @code{mdep} field to ``empty''. 5135The default definition does nothing, since most machines don't use 5136the field anyway. If you want to use the field, you should probably 5137define this macro to initialize it. 5138 5139This macro is not used on machines that do not use @code{cc0}. 5140@end defmac 5141 5142@defmac NOTICE_UPDATE_CC (@var{exp}, @var{insn}) 5143A C compound statement to set the components of @code{cc_status} 5144appropriately for an insn @var{insn} whose body is @var{exp}. It is 5145this macro's responsibility to recognize insns that set the condition 5146code as a byproduct of other activity as well as those that explicitly 5147set @code{(cc0)}. 5148 5149This macro is not used on machines that do not use @code{cc0}. 5150 5151If there are insns that do not set the condition code but do alter 5152other machine registers, this macro must check to see whether they 5153invalidate the expressions that the condition code is recorded as 5154reflecting. For example, on the 68000, insns that store in address 5155registers do not set the condition code, which means that usually 5156@code{NOTICE_UPDATE_CC} can leave @code{cc_status} unaltered for such 5157insns. But suppose that the previous insn set the condition code 5158based on location @samp{a4@@(102)} and the current insn stores a new 5159value in @samp{a4}. Although the condition code is not changed by 5160this, it will no longer be true that it reflects the contents of 5161@samp{a4@@(102)}. Therefore, @code{NOTICE_UPDATE_CC} must alter 5162@code{cc_status} in this case to say that nothing is known about the 5163condition code value. 5164 5165The definition of @code{NOTICE_UPDATE_CC} must be prepared to deal 5166with the results of peephole optimization: insns whose patterns are 5167@code{parallel} RTXs containing various @code{reg}, @code{mem} or 5168constants which are just the operands. The RTL structure of these 5169insns is not sufficient to indicate what the insns actually do. What 5170@code{NOTICE_UPDATE_CC} should do when it sees one is just to run 5171@code{CC_STATUS_INIT}. 5172 5173A possible definition of @code{NOTICE_UPDATE_CC} is to call a function 5174that looks at an attribute (@pxref{Insn Attributes}) named, for example, 5175@samp{cc}. This avoids having detailed information about patterns in 5176two places, the @file{md} file and in @code{NOTICE_UPDATE_CC}. 5177@end defmac 5178 5179@defmac SELECT_CC_MODE (@var{op}, @var{x}, @var{y}) 5180Returns a mode from class @code{MODE_CC} to be used when comparison 5181operation code @var{op} is applied to rtx @var{x} and @var{y}. For 5182example, on the SPARC, @code{SELECT_CC_MODE} is defined as (see 5183@pxref{Jump Patterns} for a description of the reason for this 5184definition) 5185 5186@smallexample 5187#define SELECT_CC_MODE(OP,X,Y) \ 5188 (GET_MODE_CLASS (GET_MODE (X)) == MODE_FLOAT \ 5189 ? ((OP == EQ || OP == NE) ? CCFPmode : CCFPEmode) \ 5190 : ((GET_CODE (X) == PLUS || GET_CODE (X) == MINUS \ 5191 || GET_CODE (X) == NEG) \ 5192 ? CC_NOOVmode : CCmode)) 5193@end smallexample 5194 5195You should define this macro if and only if you define extra CC modes 5196in @file{@var{machine}-modes.def}. 5197@end defmac 5198 5199@defmac CANONICALIZE_COMPARISON (@var{code}, @var{op0}, @var{op1}) 5200On some machines not all possible comparisons are defined, but you can 5201convert an invalid comparison into a valid one. For example, the Alpha 5202does not have a @code{GT} comparison, but you can use an @code{LT} 5203comparison instead and swap the order of the operands. 5204 5205On such machines, define this macro to be a C statement to do any 5206required conversions. @var{code} is the initial comparison code 5207and @var{op0} and @var{op1} are the left and right operands of the 5208comparison, respectively. You should modify @var{code}, @var{op0}, and 5209@var{op1} as required. 5210 5211GCC will not assume that the comparison resulting from this macro is 5212valid but will see if the resulting insn matches a pattern in the 5213@file{md} file. 5214 5215You need not define this macro if it would never change the comparison 5216code or operands. 5217@end defmac 5218 5219@defmac REVERSIBLE_CC_MODE (@var{mode}) 5220A C expression whose value is one if it is always safe to reverse a 5221comparison whose mode is @var{mode}. If @code{SELECT_CC_MODE} 5222can ever return @var{mode} for a floating-point inequality comparison, 5223then @code{REVERSIBLE_CC_MODE (@var{mode})} must be zero. 5224 5225You need not define this macro if it would always returns zero or if the 5226floating-point format is anything other than @code{IEEE_FLOAT_FORMAT}. 5227For example, here is the definition used on the SPARC, where floating-point 5228inequality comparisons are always given @code{CCFPEmode}: 5229 5230@smallexample 5231#define REVERSIBLE_CC_MODE(MODE) ((MODE) != CCFPEmode) 5232@end smallexample 5233@end defmac 5234 5235@defmac REVERSE_CONDITION (@var{code}, @var{mode}) 5236A C expression whose value is reversed condition code of the @var{code} for 5237comparison done in CC_MODE @var{mode}. The macro is used only in case 5238@code{REVERSIBLE_CC_MODE (@var{mode})} is nonzero. Define this macro in case 5239machine has some non-standard way how to reverse certain conditionals. For 5240instance in case all floating point conditions are non-trapping, compiler may 5241freely convert unordered compares to ordered one. Then definition may look 5242like: 5243 5244@smallexample 5245#define REVERSE_CONDITION(CODE, MODE) \ 5246 ((MODE) != CCFPmode ? reverse_condition (CODE) \ 5247 : reverse_condition_maybe_unordered (CODE)) 5248@end smallexample 5249@end defmac 5250 5251@defmac REVERSE_CONDEXEC_PREDICATES_P (@var{code1}, @var{code2}) 5252A C expression that returns true if the conditional execution predicate 5253@var{code1} is the inverse of @var{code2} and vice versa. Define this to 5254return 0 if the target has conditional execution predicates that cannot be 5255reversed safely. If no expansion is specified, this macro is defined as 5256follows: 5257 5258@smallexample 5259#define REVERSE_CONDEXEC_PREDICATES_P (x, y) \ 5260 ((x) == reverse_condition (y)) 5261@end smallexample 5262@end defmac 5263 5264@deftypefn {Target Hook} bool TARGET_FIXED_CONDITION_CODE_REGS (unsigned int *, unsigned int *) 5265On targets which do not use @code{(cc0)}, and which use a hard 5266register rather than a pseudo-register to hold condition codes, the 5267regular CSE passes are often not able to identify cases in which the 5268hard register is set to a common value. Use this hook to enable a 5269small pass which optimizes such cases. This hook should return true 5270to enable this pass, and it should set the integers to which its 5271arguments point to the hard register numbers used for condition codes. 5272When there is only one such register, as is true on most systems, the 5273integer pointed to by the second argument should be set to 5274@code{INVALID_REGNUM}. 5275 5276The default version of this hook returns false. 5277@end deftypefn 5278 5279@deftypefn {Target Hook} enum machine_mode TARGET_CC_MODES_COMPATIBLE (enum machine_mode, enum machine_mode) 5280On targets which use multiple condition code modes in class 5281@code{MODE_CC}, it is sometimes the case that a comparison can be 5282validly done in more than one mode. On such a system, define this 5283target hook to take two mode arguments and to return a mode in which 5284both comparisons may be validly done. If there is no such mode, 5285return @code{VOIDmode}. 5286 5287The default version of this hook checks whether the modes are the 5288same. If they are, it returns that mode. If they are different, it 5289returns @code{VOIDmode}. 5290@end deftypefn 5291 5292@node Costs 5293@section Describing Relative Costs of Operations 5294@cindex costs of instructions 5295@cindex relative costs 5296@cindex speed of instructions 5297 5298These macros let you describe the relative speed of various operations 5299on the target machine. 5300 5301@defmac REGISTER_MOVE_COST (@var{mode}, @var{from}, @var{to}) 5302A C expression for the cost of moving data of mode @var{mode} from a 5303register in class @var{from} to one in class @var{to}. The classes are 5304expressed using the enumeration values such as @code{GENERAL_REGS}. A 5305value of 2 is the default; other values are interpreted relative to 5306that. 5307 5308It is not required that the cost always equal 2 when @var{from} is the 5309same as @var{to}; on some machines it is expensive to move between 5310registers if they are not general registers. 5311 5312If reload sees an insn consisting of a single @code{set} between two 5313hard registers, and if @code{REGISTER_MOVE_COST} applied to their 5314classes returns a value of 2, reload does not check to ensure that the 5315constraints of the insn are met. Setting a cost of other than 2 will 5316allow reload to verify that the constraints are met. You should do this 5317if the @samp{mov@var{m}} pattern's constraints do not allow such copying. 5318@end defmac 5319 5320@defmac MEMORY_MOVE_COST (@var{mode}, @var{class}, @var{in}) 5321A C expression for the cost of moving data of mode @var{mode} between a 5322register of class @var{class} and memory; @var{in} is zero if the value 5323is to be written to memory, nonzero if it is to be read in. This cost 5324is relative to those in @code{REGISTER_MOVE_COST}. If moving between 5325registers and memory is more expensive than between two registers, you 5326should define this macro to express the relative cost. 5327 5328If you do not define this macro, GCC uses a default cost of 4 plus 5329the cost of copying via a secondary reload register, if one is 5330needed. If your machine requires a secondary reload register to copy 5331between memory and a register of @var{class} but the reload mechanism is 5332more complex than copying via an intermediate, define this macro to 5333reflect the actual cost of the move. 5334 5335GCC defines the function @code{memory_move_secondary_cost} if 5336secondary reloads are needed. It computes the costs due to copying via 5337a secondary register. If your machine copies from memory using a 5338secondary register in the conventional way but the default base value of 53394 is not correct for your machine, define this macro to add some other 5340value to the result of that function. The arguments to that function 5341are the same as to this macro. 5342@end defmac 5343 5344@defmac BRANCH_COST 5345A C expression for the cost of a branch instruction. A value of 1 is 5346the default; other values are interpreted relative to that. 5347@end defmac 5348 5349Here are additional macros which do not specify precise relative costs, 5350but only that certain actions are more expensive than GCC would 5351ordinarily expect. 5352 5353@defmac SLOW_BYTE_ACCESS 5354Define this macro as a C expression which is nonzero if accessing less 5355than a word of memory (i.e.@: a @code{char} or a @code{short}) is no 5356faster than accessing a word of memory, i.e., if such access 5357require more than one instruction or if there is no difference in cost 5358between byte and (aligned) word loads. 5359 5360When this macro is not defined, the compiler will access a field by 5361finding the smallest containing object; when it is defined, a fullword 5362load will be used if alignment permits. Unless bytes accesses are 5363faster than word accesses, using word accesses is preferable since it 5364may eliminate subsequent memory access if subsequent accesses occur to 5365other fields in the same word of the structure, but to different bytes. 5366@end defmac 5367 5368@defmac SLOW_UNALIGNED_ACCESS (@var{mode}, @var{alignment}) 5369Define this macro to be the value 1 if memory accesses described by the 5370@var{mode} and @var{alignment} parameters have a cost many times greater 5371than aligned accesses, for example if they are emulated in a trap 5372handler. 5373 5374When this macro is nonzero, the compiler will act as if 5375@code{STRICT_ALIGNMENT} were nonzero when generating code for block 5376moves. This can cause significantly more instructions to be produced. 5377Therefore, do not set this macro nonzero if unaligned accesses only add a 5378cycle or two to the time for a memory access. 5379 5380If the value of this macro is always zero, it need not be defined. If 5381this macro is defined, it should produce a nonzero value when 5382@code{STRICT_ALIGNMENT} is nonzero. 5383@end defmac 5384 5385@defmac MOVE_RATIO 5386The threshold of number of scalar memory-to-memory move insns, @emph{below} 5387which a sequence of insns should be generated instead of a 5388string move insn or a library call. Increasing the value will always 5389make code faster, but eventually incurs high cost in increased code size. 5390 5391Note that on machines where the corresponding move insn is a 5392@code{define_expand} that emits a sequence of insns, this macro counts 5393the number of such sequences. 5394 5395If you don't define this, a reasonable default is used. 5396@end defmac 5397 5398@defmac MOVE_BY_PIECES_P (@var{size}, @var{alignment}) 5399A C expression used to determine whether @code{move_by_pieces} will be used to 5400copy a chunk of memory, or whether some other block move mechanism 5401will be used. Defaults to 1 if @code{move_by_pieces_ninsns} returns less 5402than @code{MOVE_RATIO}. 5403@end defmac 5404 5405@defmac MOVE_MAX_PIECES 5406A C expression used by @code{move_by_pieces} to determine the largest unit 5407a load or store used to copy memory is. Defaults to @code{MOVE_MAX}. 5408@end defmac 5409 5410@defmac CLEAR_RATIO 5411The threshold of number of scalar move insns, @emph{below} which a sequence 5412of insns should be generated to clear memory instead of a string clear insn 5413or a library call. Increasing the value will always make code faster, but 5414eventually incurs high cost in increased code size. 5415 5416If you don't define this, a reasonable default is used. 5417@end defmac 5418 5419@defmac CLEAR_BY_PIECES_P (@var{size}, @var{alignment}) 5420A C expression used to determine whether @code{clear_by_pieces} will be used 5421to clear a chunk of memory, or whether some other block clear mechanism 5422will be used. Defaults to 1 if @code{move_by_pieces_ninsns} returns less 5423than @code{CLEAR_RATIO}. 5424@end defmac 5425 5426@defmac STORE_BY_PIECES_P (@var{size}, @var{alignment}) 5427A C expression used to determine whether @code{store_by_pieces} will be 5428used to set a chunk of memory to a constant value, or whether some other 5429mechanism will be used. Used by @code{__builtin_memset} when storing 5430values other than constant zero and by @code{__builtin_strcpy} when 5431when called with a constant source string. 5432Defaults to @code{MOVE_BY_PIECES_P}. 5433@end defmac 5434 5435@defmac USE_LOAD_POST_INCREMENT (@var{mode}) 5436A C expression used to determine whether a load postincrement is a good 5437thing to use for a given mode. Defaults to the value of 5438@code{HAVE_POST_INCREMENT}. 5439@end defmac 5440 5441@defmac USE_LOAD_POST_DECREMENT (@var{mode}) 5442A C expression used to determine whether a load postdecrement is a good 5443thing to use for a given mode. Defaults to the value of 5444@code{HAVE_POST_DECREMENT}. 5445@end defmac 5446 5447@defmac USE_LOAD_PRE_INCREMENT (@var{mode}) 5448A C expression used to determine whether a load preincrement is a good 5449thing to use for a given mode. Defaults to the value of 5450@code{HAVE_PRE_INCREMENT}. 5451@end defmac 5452 5453@defmac USE_LOAD_PRE_DECREMENT (@var{mode}) 5454A C expression used to determine whether a load predecrement is a good 5455thing to use for a given mode. Defaults to the value of 5456@code{HAVE_PRE_DECREMENT}. 5457@end defmac 5458 5459@defmac USE_STORE_POST_INCREMENT (@var{mode}) 5460A C expression used to determine whether a store postincrement is a good 5461thing to use for a given mode. Defaults to the value of 5462@code{HAVE_POST_INCREMENT}. 5463@end defmac 5464 5465@defmac USE_STORE_POST_DECREMENT (@var{mode}) 5466A C expression used to determine whether a store postdecrement is a good 5467thing to use for a given mode. Defaults to the value of 5468@code{HAVE_POST_DECREMENT}. 5469@end defmac 5470 5471@defmac USE_STORE_PRE_INCREMENT (@var{mode}) 5472This macro is used to determine whether a store preincrement is a good 5473thing to use for a given mode. Defaults to the value of 5474@code{HAVE_PRE_INCREMENT}. 5475@end defmac 5476 5477@defmac USE_STORE_PRE_DECREMENT (@var{mode}) 5478This macro is used to determine whether a store predecrement is a good 5479thing to use for a given mode. Defaults to the value of 5480@code{HAVE_PRE_DECREMENT}. 5481@end defmac 5482 5483@defmac NO_FUNCTION_CSE 5484Define this macro if it is as good or better to call a constant 5485function address than to call an address kept in a register. 5486@end defmac 5487 5488@defmac NO_RECURSIVE_FUNCTION_CSE 5489Define this macro if it is as good or better for a function to call 5490itself with an explicit address than to call an address kept in a 5491register. 5492@end defmac 5493 5494@defmac RANGE_TEST_NON_SHORT_CIRCUIT 5495Define this macro if a non-short-circuit operation produced by 5496@samp{fold_range_test ()} is optimal. This macro defaults to true if 5497@code{BRANCH_COST} is greater than or equal to the value 2. 5498@end defmac 5499 5500@deftypefn {Target Hook} bool TARGET_RTX_COSTS (rtx @var{x}, int @var{code}, int @var{outer_code}, int *@var{total}) 5501This target hook describes the relative costs of RTL expressions. 5502 5503The cost may depend on the precise form of the expression, which is 5504available for examination in @var{x}, and the rtx code of the expression 5505in which it is contained, found in @var{outer_code}. @var{code} is the 5506expression code---redundant, since it can be obtained with 5507@code{GET_CODE (@var{x})}. 5508 5509In implementing this hook, you can use the construct 5510@code{COSTS_N_INSNS (@var{n})} to specify a cost equal to @var{n} fast 5511instructions. 5512 5513On entry to the hook, @code{*@var{total}} contains a default estimate 5514for the cost of the expression. The hook should modify this value as 5515necessary. 5516 5517The hook returns true when all subexpressions of @var{x} have been 5518processed, and false when @code{rtx_cost} should recurse. 5519@end deftypefn 5520 5521@deftypefn {Target Hook} int TARGET_ADDRESS_COST (rtx @var{address}) 5522This hook computes the cost of an addressing mode that contains 5523@var{address}. If not defined, the cost is computed from 5524the @var{address} expression and the @code{TARGET_RTX_COST} hook. 5525 5526For most CISC machines, the default cost is a good approximation of the 5527true cost of the addressing mode. However, on RISC machines, all 5528instructions normally have the same length and execution time. Hence 5529all addresses will have equal costs. 5530 5531In cases where more than one form of an address is known, the form with 5532the lowest cost will be used. If multiple forms have the same, lowest, 5533cost, the one that is the most complex will be used. 5534 5535For example, suppose an address that is equal to the sum of a register 5536and a constant is used twice in the same basic block. When this macro 5537is not defined, the address will be computed in a register and memory 5538references will be indirect through that register. On machines where 5539the cost of the addressing mode containing the sum is no higher than 5540that of a simple indirect reference, this will produce an additional 5541instruction and possibly require an additional register. Proper 5542specification of this macro eliminates this overhead for such machines. 5543 5544This hook is never called with an invalid address. 5545 5546On machines where an address involving more than one register is as 5547cheap as an address computation involving only one register, defining 5548@code{TARGET_ADDRESS_COST} to reflect this can cause two registers to 5549be live over a region of code where only one would have been if 5550@code{TARGET_ADDRESS_COST} were not defined in that manner. This effect 5551should be considered in the definition of this macro. Equivalent costs 5552should probably only be given to addresses with different numbers of 5553registers on machines with lots of registers. 5554@end deftypefn 5555 5556@node Scheduling 5557@section Adjusting the Instruction Scheduler 5558 5559The instruction scheduler may need a fair amount of machine-specific 5560adjustment in order to produce good code. GCC provides several target 5561hooks for this purpose. It is usually enough to define just a few of 5562them: try the first ones in this list first. 5563 5564@deftypefn {Target Hook} int TARGET_SCHED_ISSUE_RATE (void) 5565This hook returns the maximum number of instructions that can ever 5566issue at the same time on the target machine. The default is one. 5567Although the insn scheduler can define itself the possibility of issue 5568an insn on the same cycle, the value can serve as an additional 5569constraint to issue insns on the same simulated processor cycle (see 5570hooks @samp{TARGET_SCHED_REORDER} and @samp{TARGET_SCHED_REORDER2}). 5571This value must be constant over the entire compilation. If you need 5572it to vary depending on what the instructions are, you must use 5573@samp{TARGET_SCHED_VARIABLE_ISSUE}. 5574 5575For the automaton based pipeline interface, you could define this hook 5576to return the value of the macro @code{MAX_DFA_ISSUE_RATE}. 5577@end deftypefn 5578 5579@deftypefn {Target Hook} int TARGET_SCHED_VARIABLE_ISSUE (FILE *@var{file}, int @var{verbose}, rtx @var{insn}, int @var{more}) 5580This hook is executed by the scheduler after it has scheduled an insn 5581from the ready list. It should return the number of insns which can 5582still be issued in the current cycle. The default is 5583@samp{@w{@var{more} - 1}} for insns other than @code{CLOBBER} and 5584@code{USE}, which normally are not counted against the issue rate. 5585You should define this hook if some insns take more machine resources 5586than others, so that fewer insns can follow them in the same cycle. 5587@var{file} is either a null pointer, or a stdio stream to write any 5588debug output to. @var{verbose} is the verbose level provided by 5589@option{-fsched-verbose-@var{n}}. @var{insn} is the instruction that 5590was scheduled. 5591@end deftypefn 5592 5593@deftypefn {Target Hook} int TARGET_SCHED_ADJUST_COST (rtx @var{insn}, rtx @var{link}, rtx @var{dep_insn}, int @var{cost}) 5594This function corrects the value of @var{cost} based on the 5595relationship between @var{insn} and @var{dep_insn} through the 5596dependence @var{link}. It should return the new value. The default 5597is to make no adjustment to @var{cost}. This can be used for example 5598to specify to the scheduler using the traditional pipeline description 5599that an output- or anti-dependence does not incur the same cost as a 5600data-dependence. If the scheduler using the automaton based pipeline 5601description, the cost of anti-dependence is zero and the cost of 5602output-dependence is maximum of one and the difference of latency 5603times of the first and the second insns. If these values are not 5604acceptable, you could use the hook to modify them too. See also 5605@pxref{Automaton pipeline description}. 5606@end deftypefn 5607 5608@deftypefn {Target Hook} int TARGET_SCHED_ADJUST_PRIORITY (rtx @var{insn}, int @var{priority}) 5609This hook adjusts the integer scheduling priority @var{priority} of 5610@var{insn}. It should return the new priority. Reduce the priority to 5611execute @var{insn} earlier, increase the priority to execute @var{insn} 5612later. Do not define this hook if you do not need to adjust the 5613scheduling priorities of insns. 5614@end deftypefn 5615 5616@deftypefn {Target Hook} int TARGET_SCHED_REORDER (FILE *@var{file}, int @var{verbose}, rtx *@var{ready}, int *@var{n_readyp}, int @var{clock}) 5617This hook is executed by the scheduler after it has scheduled the ready 5618list, to allow the machine description to reorder it (for example to 5619combine two small instructions together on @samp{VLIW} machines). 5620@var{file} is either a null pointer, or a stdio stream to write any 5621debug output to. @var{verbose} is the verbose level provided by 5622@option{-fsched-verbose-@var{n}}. @var{ready} is a pointer to the ready 5623list of instructions that are ready to be scheduled. @var{n_readyp} is 5624a pointer to the number of elements in the ready list. The scheduler 5625reads the ready list in reverse order, starting with 5626@var{ready}[@var{*n_readyp}-1] and going to @var{ready}[0]. @var{clock} 5627is the timer tick of the scheduler. You may modify the ready list and 5628the number of ready insns. The return value is the number of insns that 5629can issue this cycle; normally this is just @code{issue_rate}. See also 5630@samp{TARGET_SCHED_REORDER2}. 5631@end deftypefn 5632 5633@deftypefn {Target Hook} int TARGET_SCHED_REORDER2 (FILE *@var{file}, int @var{verbose}, rtx *@var{ready}, int *@var{n_ready}, @var{clock}) 5634Like @samp{TARGET_SCHED_REORDER}, but called at a different time. That 5635function is called whenever the scheduler starts a new cycle. This one 5636is called once per iteration over a cycle, immediately after 5637@samp{TARGET_SCHED_VARIABLE_ISSUE}; it can reorder the ready list and 5638return the number of insns to be scheduled in the same cycle. Defining 5639this hook can be useful if there are frequent situations where 5640scheduling one insn causes other insns to become ready in the same 5641cycle. These other insns can then be taken into account properly. 5642@end deftypefn 5643 5644@deftypefn {Target Hook} void TARGET_SCHED_DEPENDENCIES_EVALUATION_HOOK (rtx @var{head}, rtx @var{tail}) 5645This hook is called after evaluation forward dependencies of insns in 5646chain given by two parameter values (@var{head} and @var{tail} 5647correspondingly) but before insns scheduling of the insn chain. For 5648example, it can be used for better insn classification if it requires 5649analysis of dependencies. This hook can use backward and forward 5650dependencies of the insn scheduler because they are already 5651calculated. 5652@end deftypefn 5653 5654@deftypefn {Target Hook} void TARGET_SCHED_INIT (FILE *@var{file}, int @var{verbose}, int @var{max_ready}) 5655This hook is executed by the scheduler at the beginning of each block of 5656instructions that are to be scheduled. @var{file} is either a null 5657pointer, or a stdio stream to write any debug output to. @var{verbose} 5658is the verbose level provided by @option{-fsched-verbose-@var{n}}. 5659@var{max_ready} is the maximum number of insns in the current scheduling 5660region that can be live at the same time. This can be used to allocate 5661scratch space if it is needed, e.g. by @samp{TARGET_SCHED_REORDER}. 5662@end deftypefn 5663 5664@deftypefn {Target Hook} void TARGET_SCHED_FINISH (FILE *@var{file}, int @var{verbose}) 5665This hook is executed by the scheduler at the end of each block of 5666instructions that are to be scheduled. It can be used to perform 5667cleanup of any actions done by the other scheduling hooks. @var{file} 5668is either a null pointer, or a stdio stream to write any debug output 5669to. @var{verbose} is the verbose level provided by 5670@option{-fsched-verbose-@var{n}}. 5671@end deftypefn 5672 5673@deftypefn {Target Hook} int TARGET_SCHED_USE_DFA_PIPELINE_INTERFACE (void) 5674This hook is called many times during insn scheduling. If the hook 5675returns nonzero, the automaton based pipeline description is used for 5676insn scheduling. Otherwise the traditional pipeline description is 5677used. The default is usage of the traditional pipeline description. 5678 5679You should also remember that to simplify the insn scheduler sources 5680an empty traditional pipeline description interface is generated even 5681if there is no a traditional pipeline description in the @file{.md} 5682file. The same is true for the automaton based pipeline description. 5683That means that you should be accurate in defining the hook. 5684@end deftypefn 5685 5686@deftypefn {Target Hook} int TARGET_SCHED_DFA_PRE_CYCLE_INSN (void) 5687The hook returns an RTL insn. The automaton state used in the 5688pipeline hazard recognizer is changed as if the insn were scheduled 5689when the new simulated processor cycle starts. Usage of the hook may 5690simplify the automaton pipeline description for some @acronym{VLIW} 5691processors. If the hook is defined, it is used only for the automaton 5692based pipeline description. The default is not to change the state 5693when the new simulated processor cycle starts. 5694@end deftypefn 5695 5696@deftypefn {Target Hook} void TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN (void) 5697The hook can be used to initialize data used by the previous hook. 5698@end deftypefn 5699 5700@deftypefn {Target Hook} int TARGET_SCHED_DFA_POST_CYCLE_INSN (void) 5701The hook is analogous to @samp{TARGET_SCHED_DFA_PRE_CYCLE_INSN} but used 5702to changed the state as if the insn were scheduled when the new 5703simulated processor cycle finishes. 5704@end deftypefn 5705 5706@deftypefn {Target Hook} void TARGET_SCHED_INIT_DFA_POST_CYCLE_INSN (void) 5707The hook is analogous to @samp{TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN} but 5708used to initialize data used by the previous hook. 5709@end deftypefn 5710 5711@deftypefn {Target Hook} int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD (void) 5712This hook controls better choosing an insn from the ready insn queue 5713for the @acronym{DFA}-based insn scheduler. Usually the scheduler 5714chooses the first insn from the queue. If the hook returns a positive 5715value, an additional scheduler code tries all permutations of 5716@samp{TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD ()} 5717subsequent ready insns to choose an insn whose issue will result in 5718maximal number of issued insns on the same cycle. For the 5719@acronym{VLIW} processor, the code could actually solve the problem of 5720packing simple insns into the @acronym{VLIW} insn. Of course, if the 5721rules of @acronym{VLIW} packing are described in the automaton. 5722 5723This code also could be used for superscalar @acronym{RISC} 5724processors. Let us consider a superscalar @acronym{RISC} processor 5725with 3 pipelines. Some insns can be executed in pipelines @var{A} or 5726@var{B}, some insns can be executed only in pipelines @var{B} or 5727@var{C}, and one insn can be executed in pipeline @var{B}. The 5728processor may issue the 1st insn into @var{A} and the 2nd one into 5729@var{B}. In this case, the 3rd insn will wait for freeing @var{B} 5730until the next cycle. If the scheduler issues the 3rd insn the first, 5731the processor could issue all 3 insns per cycle. 5732 5733Actually this code demonstrates advantages of the automaton based 5734pipeline hazard recognizer. We try quickly and easy many insn 5735schedules to choose the best one. 5736 5737The default is no multipass scheduling. 5738@end deftypefn 5739 5740@deftypefn {Target Hook} int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD (rtx) 5741 5742This hook controls what insns from the ready insn queue will be 5743considered for the multipass insn scheduling. If the hook returns 5744zero for insn passed as the parameter, the insn will be not chosen to 5745be issued. 5746 5747The default is that any ready insns can be chosen to be issued. 5748@end deftypefn 5749 5750@deftypefn {Target Hook} int TARGET_SCHED_DFA_NEW_CYCLE (FILE *, int, rtx, int, int, int *) 5751 5752This hook is called by the insn scheduler before issuing insn passed 5753as the third parameter on given cycle. If the hook returns nonzero, 5754the insn is not issued on given processors cycle. Instead of that, 5755the processor cycle is advanced. If the value passed through the last 5756parameter is zero, the insn ready queue is not sorted on the new cycle 5757start as usually. The first parameter passes file for debugging 5758output. The second one passes the scheduler verbose level of the 5759debugging output. The forth and the fifth parameter values are 5760correspondingly processor cycle on which the previous insn has been 5761issued and the current processor cycle. 5762@end deftypefn 5763 5764@deftypefn {Target Hook} void TARGET_SCHED_INIT_DFA_BUBBLES (void) 5765The @acronym{DFA}-based scheduler could take the insertion of nop 5766operations for better insn scheduling into account. It can be done 5767only if the multi-pass insn scheduling works (see hook 5768@samp{TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD}). 5769 5770Let us consider a @acronym{VLIW} processor insn with 3 slots. Each 5771insn can be placed only in one of the three slots. We have 3 ready 5772insns @var{A}, @var{B}, and @var{C}. @var{A} and @var{C} can be 5773placed only in the 1st slot, @var{B} can be placed only in the 3rd 5774slot. We described the automaton which does not permit empty slot 5775gaps between insns (usually such description is simpler). Without 5776this code the scheduler would place each insn in 3 separate 5777@acronym{VLIW} insns. If the scheduler places a nop insn into the 2nd 5778slot, it could place the 3 insns into 2 @acronym{VLIW} insns. What is 5779the nop insn is returned by hook @samp{TARGET_SCHED_DFA_BUBBLE}. Hook 5780@samp{TARGET_SCHED_INIT_DFA_BUBBLES} can be used to initialize or 5781create the nop insns. 5782 5783You should remember that the scheduler does not insert the nop insns. 5784It is not wise because of the following optimizations. The scheduler 5785only considers such possibility to improve the result schedule. The 5786nop insns should be inserted lately, e.g. on the final phase. 5787@end deftypefn 5788 5789@deftypefn {Target Hook} rtx TARGET_SCHED_DFA_BUBBLE (int @var{index}) 5790This hook @samp{FIRST_CYCLE_MULTIPASS_SCHEDULING} is used to insert 5791nop operations for better insn scheduling when @acronym{DFA}-based 5792scheduler makes multipass insn scheduling (see also description of 5793hook @samp{TARGET_SCHED_INIT_DFA_BUBBLES}). This hook 5794returns a nop insn with given @var{index}. The indexes start with 5795zero. The hook should return @code{NULL} if there are no more nop 5796insns with indexes greater than given index. 5797@end deftypefn 5798 5799@deftypefn {Target Hook} bool IS_COSTLY_DEPENDENCE (rtx @var{insn1}, rtx @var{insn2}, rtx @var{dep_link}, int @var{dep_cost}, int @var{distance}) 5800This hook is used to define which dependences are considered costly by 5801the target, so costly that it is not advisable to schedule the insns that 5802are involved in the dependence too close to one another. The parameters 5803to this hook are as follows: The second parameter @var{insn2} is dependent 5804upon the first parameter @var{insn1}. The dependence between @var{insn1} 5805and @var{insn2} is represented by the third parameter @var{dep_link}. The 5806fourth parameter @var{cost} is the cost of the dependence, and the fifth 5807parameter @var{distance} is the distance in cycles between the two insns. 5808The hook returns @code{true} if considering the distance between the two 5809insns the dependence between them is considered costly by the target, 5810and @code{false} otherwise. 5811 5812Defining this hook can be useful in multiple-issue out-of-order machines, 5813where (a) it's practically hopeless to predict the actual data/resource 5814delays, however: (b) there's a better chance to predict the actual grouping 5815that will be formed, and (c) correctly emulating the grouping can be very 5816important. In such targets one may want to allow issuing dependent insns 5817closer to one another - i.e, closer than the dependence distance; however, 5818not in cases of "costly dependences", which this hooks allows to define. 5819@end deftypefn 5820 5821Macros in the following table are generated by the program 5822@file{genattr} and can be useful for writing the hooks. 5823 5824@defmac TRADITIONAL_PIPELINE_INTERFACE 5825The macro definition is generated if there is a traditional pipeline 5826description in @file{.md} file. You should also remember that to 5827simplify the insn scheduler sources an empty traditional pipeline 5828description interface is generated even if there is no a traditional 5829pipeline description in the @file{.md} file. The macro can be used to 5830distinguish the two types of the traditional interface. 5831@end defmac 5832 5833@defmac DFA_PIPELINE_INTERFACE 5834The macro definition is generated if there is an automaton pipeline 5835description in @file{.md} file. You should also remember that to 5836simplify the insn scheduler sources an empty automaton pipeline 5837description interface is generated even if there is no an automaton 5838pipeline description in the @file{.md} file. The macro can be used to 5839distinguish the two types of the automaton interface. 5840@end defmac 5841 5842@defmac MAX_DFA_ISSUE_RATE 5843The macro definition is generated in the automaton based pipeline 5844description interface. Its value is calculated from the automaton 5845based pipeline description and is equal to maximal number of all insns 5846described in constructions @samp{define_insn_reservation} which can be 5847issued on the same processor cycle. 5848@end defmac 5849 5850@node Sections 5851@section Dividing the Output into Sections (Texts, Data, @dots{}) 5852@c the above section title is WAY too long. maybe cut the part between 5853@c the (...)? --mew 10feb93 5854 5855An object file is divided into sections containing different types of 5856data. In the most common case, there are three sections: the @dfn{text 5857section}, which holds instructions and read-only data; the @dfn{data 5858section}, which holds initialized writable data; and the @dfn{bss 5859section}, which holds uninitialized data. Some systems have other kinds 5860of sections. 5861 5862The compiler must tell the assembler when to switch sections. These 5863macros control what commands to output to tell the assembler this. You 5864can also define additional sections. 5865 5866@defmac TEXT_SECTION_ASM_OP 5867A C expression whose value is a string, including spacing, containing the 5868assembler operation that should precede instructions and read-only data. 5869Normally @code{"\t.text"} is right. 5870@end defmac 5871 5872@defmac HOT_TEXT_SECTION_NAME 5873If defined, a C string constant for the name of the section containing most 5874frequently executed functions of the program. If not defined, GCC will provide 5875a default definition if the target supports named sections. 5876@end defmac 5877 5878@defmac UNLIKELY_EXECUTED_TEXT_SECTION_NAME 5879If defined, a C string constant for the name of the section containing unlikely 5880executed functions in the program. 5881@end defmac 5882 5883@defmac DATA_SECTION_ASM_OP 5884A C expression whose value is a string, including spacing, containing the 5885assembler operation to identify the following data as writable initialized 5886data. Normally @code{"\t.data"} is right. 5887@end defmac 5888 5889@defmac READONLY_DATA_SECTION_ASM_OP 5890A C expression whose value is a string, including spacing, containing the 5891assembler operation to identify the following data as read-only initialized 5892data. 5893@end defmac 5894 5895@defmac READONLY_DATA_SECTION 5896A macro naming a function to call to switch to the proper section for 5897read-only data. The default is to use @code{READONLY_DATA_SECTION_ASM_OP} 5898if defined, else fall back to @code{text_section}. 5899 5900The most common definition will be @code{data_section}, if the target 5901does not have a special read-only data section, and does not put data 5902in the text section. 5903@end defmac 5904 5905@defmac SHARED_SECTION_ASM_OP 5906If defined, a C expression whose value is a string, including spacing, 5907containing the assembler operation to identify the following data as 5908shared data. If not defined, @code{DATA_SECTION_ASM_OP} will be used. 5909@end defmac 5910 5911@defmac BSS_SECTION_ASM_OP 5912If defined, a C expression whose value is a string, including spacing, 5913containing the assembler operation to identify the following data as 5914uninitialized global data. If not defined, and neither 5915@code{ASM_OUTPUT_BSS} nor @code{ASM_OUTPUT_ALIGNED_BSS} are defined, 5916uninitialized global data will be output in the data section if 5917@option{-fno-common} is passed, otherwise @code{ASM_OUTPUT_COMMON} will be 5918used. 5919@end defmac 5920 5921@defmac INIT_SECTION_ASM_OP 5922If defined, a C expression whose value is a string, including spacing, 5923containing the assembler operation to identify the following data as 5924initialization code. If not defined, GCC will assume such a section does 5925not exist. 5926@end defmac 5927 5928@defmac FINI_SECTION_ASM_OP 5929If defined, a C expression whose value is a string, including spacing, 5930containing the assembler operation to identify the following data as 5931finalization code. If not defined, GCC will assume such a section does 5932not exist. 5933@end defmac 5934 5935@defmac CRT_CALL_STATIC_FUNCTION (@var{section_op}, @var{function}) 5936If defined, an ASM statement that switches to a different section 5937via @var{section_op}, calls @var{function}, and switches back to 5938the text section. This is used in @file{crtstuff.c} if 5939@code{INIT_SECTION_ASM_OP} or @code{FINI_SECTION_ASM_OP} to calls 5940to initialization and finalization functions from the init and fini 5941sections. By default, this macro uses a simple function call. Some 5942ports need hand-crafted assembly code to avoid dependencies on 5943registers initialized in the function prologue or to ensure that 5944constant pools don't end up too far way in the text section. 5945@end defmac 5946 5947@defmac FORCE_CODE_SECTION_ALIGN 5948If defined, an ASM statement that aligns a code section to some 5949arbitrary boundary. This is used to force all fragments of the 5950@code{.init} and @code{.fini} sections to have to same alignment 5951and thus prevent the linker from having to add any padding. 5952@end defmac 5953 5954@findex in_text 5955@findex in_data 5956@defmac EXTRA_SECTIONS 5957A list of names for sections other than the standard two, which are 5958@code{in_text} and @code{in_data}. You need not define this macro 5959on a system with no other sections (that GCC needs to use). 5960@end defmac 5961 5962@findex text_section 5963@findex data_section 5964@defmac EXTRA_SECTION_FUNCTIONS 5965One or more functions to be defined in @file{varasm.c}. These 5966functions should do jobs analogous to those of @code{text_section} and 5967@code{data_section}, for your additional sections. Do not define this 5968macro if you do not define @code{EXTRA_SECTIONS}. 5969@end defmac 5970 5971@defmac JUMP_TABLES_IN_TEXT_SECTION 5972Define this macro to be an expression with a nonzero value if jump 5973tables (for @code{tablejump} insns) should be output in the text 5974section, along with the assembler instructions. Otherwise, the 5975readonly data section is used. 5976 5977This macro is irrelevant if there is no separate readonly data section. 5978@end defmac 5979 5980@deftypefn {Target Hook} void TARGET_ASM_SELECT_SECTION (tree @var{exp}, int @var{reloc}, unsigned HOST_WIDE_INT @var{align}) 5981Switches to the appropriate section for output of @var{exp}. You can 5982assume that @var{exp} is either a @code{VAR_DECL} node or a constant of 5983some sort. @var{reloc} indicates whether the initial value of @var{exp} 5984requires link-time relocations. Bit 0 is set when variable contains 5985local relocations only, while bit 1 is set for global relocations. 5986Select the section by calling @code{data_section} or one of the 5987alternatives for other sections. @var{align} is the constant alignment 5988in bits. 5989 5990The default version of this function takes care of putting read-only 5991variables in @code{readonly_data_section}. 5992@end deftypefn 5993 5994@deftypefn {Target Hook} void TARGET_ASM_UNIQUE_SECTION (tree @var{decl}, int @var{reloc}) 5995Build up a unique section name, expressed as a @code{STRING_CST} node, 5996and assign it to @samp{DECL_SECTION_NAME (@var{decl})}. 5997As with @code{TARGET_ASM_SELECT_SECTION}, @var{reloc} indicates whether 5998the initial value of @var{exp} requires link-time relocations. 5999 6000The default version of this function appends the symbol name to the 6001ELF section name that would normally be used for the symbol. For 6002example, the function @code{foo} would be placed in @code{.text.foo}. 6003Whatever the actual target object format, this is often good enough. 6004@end deftypefn 6005 6006@deftypefn {Target Hook} void TARGET_ASM_SELECT_RTX_SECTION (enum machine_mode @var{mode}, rtx @var{x}, unsigned HOST_WIDE_INT @var{align}) 6007Switches to the appropriate section for output of constant pool entry 6008@var{x} in @var{mode}. You can assume that @var{x} is some kind of 6009constant in RTL@. The argument @var{mode} is redundant except in the 6010case of a @code{const_int} rtx. Select the section by calling 6011@code{readonly_data_section} or one of the alternatives for other 6012sections. @var{align} is the constant alignment in bits. 6013 6014The default version of this function takes care of putting symbolic 6015constants in @code{flag_pic} mode in @code{data_section} and everything 6016else in @code{readonly_data_section}. 6017@end deftypefn 6018 6019@deftypefn {Target Hook} void TARGET_ENCODE_SECTION_INFO (tree @var{decl}, rtx @var{rtl}, int @var{new_decl_p}) 6020Define this hook if references to a symbol or a constant must be 6021treated differently depending on something about the variable or 6022function named by the symbol (such as what section it is in). 6023 6024The hook is executed immediately after rtl has been created for 6025@var{decl}, which may be a variable or function declaration or 6026an entry in the constant pool. In either case, @var{rtl} is the 6027rtl in question. Do @emph{not} use @code{DECL_RTL (@var{decl})} 6028in this hook; that field may not have been initialized yet. 6029 6030In the case of a constant, it is safe to assume that the rtl is 6031a @code{mem} whose address is a @code{symbol_ref}. Most decls 6032will also have this form, but that is not guaranteed. Global 6033register variables, for instance, will have a @code{reg} for their 6034rtl. (Normally the right thing to do with such unusual rtl is 6035leave it alone.) 6036 6037The @var{new_decl_p} argument will be true if this is the first time 6038that @code{TARGET_ENCODE_SECTION_INFO} has been invoked on this decl. It will 6039be false for subsequent invocations, which will happen for duplicate 6040declarations. Whether or not anything must be done for the duplicate 6041declaration depends on whether the hook examines @code{DECL_ATTRIBUTES}. 6042@var{new_decl_p} is always true when the hook is called for a constant. 6043 6044@cindex @code{SYMBOL_REF_FLAG}, in @code{TARGET_ENCODE_SECTION_INFO} 6045The usual thing for this hook to do is to record flags in the 6046@code{symbol_ref}, using @code{SYMBOL_REF_FLAG} or @code{SYMBOL_REF_FLAGS}. 6047Historically, the name string was modified if it was necessary to 6048encode more than one bit of information, but this practice is now 6049discouraged; use @code{SYMBOL_REF_FLAGS}. 6050 6051The default definition of this hook, @code{default_encode_section_info} 6052in @file{varasm.c}, sets a number of commonly-useful bits in 6053@code{SYMBOL_REF_FLAGS}. Check whether the default does what you need 6054before overriding it. 6055@end deftypefn 6056 6057@deftypefn {Target Hook} const char *TARGET_STRIP_NAME_ENCODING (const char *name) 6058Decode @var{name} and return the real name part, sans 6059the characters that @code{TARGET_ENCODE_SECTION_INFO} 6060may have added. 6061@end deftypefn 6062 6063@deftypefn {Target Hook} bool TARGET_IN_SMALL_DATA_P (tree @var{exp}) 6064Returns true if @var{exp} should be placed into a ``small data'' section. 6065The default version of this hook always returns false. 6066@end deftypefn 6067 6068@deftypevar {Target Hook} bool TARGET_HAVE_SRODATA_SECTION 6069Contains the value true if the target places read-only 6070``small data'' into a separate section. The default value is false. 6071@end deftypevar 6072 6073@deftypefn {Target Hook} bool TARGET_BINDS_LOCAL_P (tree @var{exp}) 6074Returns true if @var{exp} names an object for which name resolution 6075rules must resolve to the current ``module'' (dynamic shared library 6076or executable image). 6077 6078The default version of this hook implements the name resolution rules 6079for ELF, which has a looser model of global name binding than other 6080currently supported object file formats. 6081@end deftypefn 6082 6083@deftypevar {Target Hook} bool TARGET_HAVE_TLS 6084Contains the value true if the target supports thread-local storage. 6085The default value is false. 6086@end deftypevar 6087 6088 6089@node PIC 6090@section Position Independent Code 6091@cindex position independent code 6092@cindex PIC 6093 6094This section describes macros that help implement generation of position 6095independent code. Simply defining these macros is not enough to 6096generate valid PIC; you must also add support to the macros 6097@code{GO_IF_LEGITIMATE_ADDRESS} and @code{PRINT_OPERAND_ADDRESS}, as 6098well as @code{LEGITIMIZE_ADDRESS}. You must modify the definition of 6099@samp{movsi} to do something appropriate when the source operand 6100contains a symbolic address. You may also need to alter the handling of 6101switch statements so that they use relative addresses. 6102@c i rearranged the order of the macros above to try to force one of 6103@c them to the next line, to eliminate an overfull hbox. --mew 10feb93 6104 6105@defmac PIC_OFFSET_TABLE_REGNUM 6106The register number of the register used to address a table of static 6107data addresses in memory. In some cases this register is defined by a 6108processor's ``application binary interface'' (ABI)@. When this macro 6109is defined, RTL is generated for this register once, as with the stack 6110pointer and frame pointer registers. If this macro is not defined, it 6111is up to the machine-dependent files to allocate such a register (if 6112necessary). Note that this register must be fixed when in use (e.g.@: 6113when @code{flag_pic} is true). 6114@end defmac 6115 6116@defmac PIC_OFFSET_TABLE_REG_CALL_CLOBBERED 6117Define this macro if the register defined by 6118@code{PIC_OFFSET_TABLE_REGNUM} is clobbered by calls. Do not define 6119this macro if @code{PIC_OFFSET_TABLE_REGNUM} is not defined. 6120@end defmac 6121 6122@defmac FINALIZE_PIC 6123By generating position-independent code, when two different programs (A 6124and B) share a common library (libC.a), the text of the library can be 6125shared whether or not the library is linked at the same address for both 6126programs. In some of these environments, position-independent code 6127requires not only the use of different addressing modes, but also 6128special code to enable the use of these addressing modes. 6129 6130The @code{FINALIZE_PIC} macro serves as a hook to emit these special 6131codes once the function is being compiled into assembly code, but not 6132before. (It is not done before, because in the case of compiling an 6133inline function, it would lead to multiple PIC prologues being 6134included in functions which used inline functions and were compiled to 6135assembly language.) 6136@end defmac 6137 6138@defmac LEGITIMATE_PIC_OPERAND_P (@var{x}) 6139A C expression that is nonzero if @var{x} is a legitimate immediate 6140operand on the target machine when generating position independent code. 6141You can assume that @var{x} satisfies @code{CONSTANT_P}, so you need not 6142check this. You can also assume @var{flag_pic} is true, so you need not 6143check it either. You need not define this macro if all constants 6144(including @code{SYMBOL_REF}) can be immediate operands when generating 6145position independent code. 6146@end defmac 6147 6148@node Assembler Format 6149@section Defining the Output Assembler Language 6150 6151This section describes macros whose principal purpose is to describe how 6152to write instructions in assembler language---rather than what the 6153instructions do. 6154 6155@menu 6156* File Framework:: Structural information for the assembler file. 6157* Data Output:: Output of constants (numbers, strings, addresses). 6158* Uninitialized Data:: Output of uninitialized variables. 6159* Label Output:: Output and generation of labels. 6160* Initialization:: General principles of initialization 6161 and termination routines. 6162* Macros for Initialization:: 6163 Specific macros that control the handling of 6164 initialization and termination routines. 6165* Instruction Output:: Output of actual instructions. 6166* Dispatch Tables:: Output of jump tables. 6167* Exception Region Output:: Output of exception region code. 6168* Alignment Output:: Pseudo ops for alignment and skipping data. 6169@end menu 6170 6171@node File Framework 6172@subsection The Overall Framework of an Assembler File 6173@cindex assembler format 6174@cindex output of assembler code 6175 6176@c prevent bad page break with this line 6177This describes the overall framework of an assembly file. 6178 6179@deftypefn {Target Hook} void TARGET_ASM_FILE_START () 6180@findex default_file_start 6181Output to @code{asm_out_file} any text which the assembler expects to 6182find at the beginning of a file. The default behavior is controlled 6183by two flags, documented below. Unless your target's assembler is 6184quite unusual, if you override the default, you should call 6185@code{default_file_start} at some point in your target hook. This 6186lets other target files rely on these variables. 6187@end deftypefn 6188 6189@deftypevr {Target Hook} bool TARGET_ASM_FILE_START_APP_OFF 6190If this flag is true, the text of the macro @code{ASM_APP_OFF} will be 6191printed as the very first line in the assembly file, unless 6192@option{-fverbose-asm} is in effect. (If that macro has been defined 6193to the empty string, this variable has no effect.) With the normal 6194definition of @code{ASM_APP_OFF}, the effect is to notify the GNU 6195assembler that it need not bother stripping comments or extra 6196whitespace from its input. This allows it to work a bit faster. 6197 6198The default is false. You should not set it to true unless you have 6199verified that your port does not generate any extra whitespace or 6200comments that will cause GAS to issue errors in NO_APP mode. 6201@end deftypevr 6202 6203@deftypevr {Target Hook} bool TARGET_ASM_FILE_START_FILE_DIRECTIVE 6204If this flag is true, @code{output_file_directive} will be called 6205for the primary source file, immediately after printing 6206@code{ASM_APP_OFF} (if that is enabled). Most ELF assemblers expect 6207this to be done. The default is false. 6208@end deftypevr 6209 6210@deftypefn {Target Hook} void TARGET_ASM_FILE_END () 6211Output to @code{asm_out_file} any text which the assembler expects 6212to find at the end of a file. The default is to output nothing. 6213@end deftypefn 6214 6215@deftypefun void file_end_indicate_exec_stack () 6216Some systems use a common convention, the @samp{.note.GNU-stack} 6217special section, to indicate whether or not an object file relies on 6218the stack being executable. If your system uses this convention, you 6219should define @code{TARGET_ASM_FILE_END} to this function. If you 6220need to do other things in that hook, have your hook function call 6221this function. 6222@end deftypefun 6223 6224@defmac ASM_COMMENT_START 6225A C string constant describing how to begin a comment in the target 6226assembler language. The compiler assumes that the comment will end at 6227the end of the line. 6228@end defmac 6229 6230@defmac ASM_APP_ON 6231A C string constant for text to be output before each @code{asm} 6232statement or group of consecutive ones. Normally this is 6233@code{"#APP"}, which is a comment that has no effect on most 6234assemblers but tells the GNU assembler that it must check the lines 6235that follow for all valid assembler constructs. 6236@end defmac 6237 6238@defmac ASM_APP_OFF 6239A C string constant for text to be output after each @code{asm} 6240statement or group of consecutive ones. Normally this is 6241@code{"#NO_APP"}, which tells the GNU assembler to resume making the 6242time-saving assumptions that are valid for ordinary compiler output. 6243@end defmac 6244 6245@defmac ASM_OUTPUT_SOURCE_FILENAME (@var{stream}, @var{name}) 6246A C statement to output COFF information or DWARF debugging information 6247which indicates that filename @var{name} is the current source file to 6248the stdio stream @var{stream}. 6249 6250This macro need not be defined if the standard form of output 6251for the file format in use is appropriate. 6252@end defmac 6253 6254@defmac OUTPUT_QUOTED_STRING (@var{stream}, @var{string}) 6255A C statement to output the string @var{string} to the stdio stream 6256@var{stream}. If you do not call the function @code{output_quoted_string} 6257in your config files, GCC will only call it to output filenames to 6258the assembler source. So you can use it to canonicalize the format 6259of the filename using this macro. 6260@end defmac 6261 6262@defmac ASM_OUTPUT_SOURCE_LINE (@var{stream}, @var{line}, @var{counter}) 6263A C statement to output DBX or SDB debugging information before code 6264for line number @var{line} of the current source file to the 6265stdio stream @var{stream}. @var{counter} is the number of time the 6266macro was invoked, including the current invocation; it is intended 6267to generate unique labels in the assembly output. 6268 6269This macro need not be defined if the standard form of debugging 6270information for the debugger in use is appropriate. 6271@end defmac 6272 6273@defmac ASM_OUTPUT_IDENT (@var{stream}, @var{string}) 6274A C statement to output something to the assembler file to handle a 6275@samp{#ident} directive containing the text @var{string}. If this 6276macro is not defined, nothing is output for a @samp{#ident} directive. 6277@end defmac 6278 6279@deftypefn {Target Hook} void TARGET_ASM_NAMED_SECTION (const char *@var{name}, unsigned int @var{flags}, unsigned int @var{align}) 6280Output assembly directives to switch to section @var{name}. The section 6281should have attributes as specified by @var{flags}, which is a bit mask 6282of the @code{SECTION_*} flags defined in @file{output.h}. If @var{align} 6283is nonzero, it contains an alignment in bytes to be used for the section, 6284otherwise some target default should be used. Only targets that must 6285specify an alignment within the section directive need pay attention to 6286@var{align} -- we will still use @code{ASM_OUTPUT_ALIGN}. 6287@end deftypefn 6288 6289@deftypefn {Target Hook} bool TARGET_HAVE_NAMED_SECTIONS 6290This flag is true if the target supports @code{TARGET_ASM_NAMED_SECTION}. 6291@end deftypefn 6292 6293@deftypefn {Target Hook} {unsigned int} TARGET_SECTION_TYPE_FLAGS (tree @var{decl}, const char *@var{name}, int @var{reloc}) 6294Choose a set of section attributes for use by @code{TARGET_ASM_NAMED_SECTION} 6295based on a variable or function decl, a section name, and whether or not the 6296declaration's initializer may contain runtime relocations. @var{decl} may be 6297 null, in which case read-write data should be assumed. 6298 6299The default version if this function handles choosing code vs data, 6300read-only vs read-write data, and @code{flag_pic}. You should only 6301need to override this if your target has special flags that might be 6302set via @code{__attribute__}. 6303@end deftypefn 6304 6305@need 2000 6306@node Data Output 6307@subsection Output of Data 6308 6309 6310@deftypevr {Target Hook} {const char *} TARGET_ASM_BYTE_OP 6311@deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_HI_OP 6312@deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_SI_OP 6313@deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_DI_OP 6314@deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_TI_OP 6315@deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_HI_OP 6316@deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_SI_OP 6317@deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_DI_OP 6318@deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_TI_OP 6319These hooks specify assembly directives for creating certain kinds 6320of integer object. The @code{TARGET_ASM_BYTE_OP} directive creates a 6321byte-sized object, the @code{TARGET_ASM_ALIGNED_HI_OP} one creates an 6322aligned two-byte object, and so on. Any of the hooks may be 6323@code{NULL}, indicating that no suitable directive is available. 6324 6325The compiler will print these strings at the start of a new line, 6326followed immediately by the object's initial value. In most cases, 6327the string should contain a tab, a pseudo-op, and then another tab. 6328@end deftypevr 6329 6330@deftypefn {Target Hook} bool TARGET_ASM_INTEGER (rtx @var{x}, unsigned int @var{size}, int @var{aligned_p}) 6331The @code{assemble_integer} function uses this hook to output an 6332integer object. @var{x} is the object's value, @var{size} is its size 6333in bytes and @var{aligned_p} indicates whether it is aligned. The 6334function should return @code{true} if it was able to output the 6335object. If it returns false, @code{assemble_integer} will try to 6336split the object into smaller parts. 6337 6338The default implementation of this hook will use the 6339@code{TARGET_ASM_BYTE_OP} family of strings, returning @code{false} 6340when the relevant string is @code{NULL}. 6341@end deftypefn 6342 6343@defmac OUTPUT_ADDR_CONST_EXTRA (@var{stream}, @var{x}, @var{fail}) 6344A C statement to recognize @var{rtx} patterns that 6345@code{output_addr_const} can't deal with, and output assembly code to 6346@var{stream} corresponding to the pattern @var{x}. This may be used to 6347allow machine-dependent @code{UNSPEC}s to appear within constants. 6348 6349If @code{OUTPUT_ADDR_CONST_EXTRA} fails to recognize a pattern, it must 6350@code{goto fail}, so that a standard error message is printed. If it 6351prints an error message itself, by calling, for example, 6352@code{output_operand_lossage}, it may just complete normally. 6353@end defmac 6354 6355@defmac ASM_OUTPUT_ASCII (@var{stream}, @var{ptr}, @var{len}) 6356A C statement to output to the stdio stream @var{stream} an assembler 6357instruction to assemble a string constant containing the @var{len} 6358bytes at @var{ptr}. @var{ptr} will be a C expression of type 6359@code{char *} and @var{len} a C expression of type @code{int}. 6360 6361If the assembler has a @code{.ascii} pseudo-op as found in the 6362Berkeley Unix assembler, do not define the macro 6363@code{ASM_OUTPUT_ASCII}. 6364@end defmac 6365 6366@defmac ASM_OUTPUT_FDESC (@var{stream}, @var{decl}, @var{n}) 6367A C statement to output word @var{n} of a function descriptor for 6368@var{decl}. This must be defined if @code{TARGET_VTABLE_USES_DESCRIPTORS} 6369is defined, and is otherwise unused. 6370@end defmac 6371 6372@defmac CONSTANT_POOL_BEFORE_FUNCTION 6373You may define this macro as a C expression. You should define the 6374expression to have a nonzero value if GCC should output the constant 6375pool for a function before the code for the function, or a zero value if 6376GCC should output the constant pool after the function. If you do 6377not define this macro, the usual case, GCC will output the constant 6378pool before the function. 6379@end defmac 6380 6381@defmac ASM_OUTPUT_POOL_PROLOGUE (@var{file}, @var{funname}, @var{fundecl}, @var{size}) 6382A C statement to output assembler commands to define the start of the 6383constant pool for a function. @var{funname} is a string giving 6384the name of the function. Should the return type of the function 6385be required, it can be obtained via @var{fundecl}. @var{size} 6386is the size, in bytes, of the constant pool that will be written 6387immediately after this call. 6388 6389If no constant-pool prefix is required, the usual case, this macro need 6390not be defined. 6391@end defmac 6392 6393@defmac ASM_OUTPUT_SPECIAL_POOL_ENTRY (@var{file}, @var{x}, @var{mode}, @var{align}, @var{labelno}, @var{jumpto}) 6394A C statement (with or without semicolon) to output a constant in the 6395constant pool, if it needs special treatment. (This macro need not do 6396anything for RTL expressions that can be output normally.) 6397 6398The argument @var{file} is the standard I/O stream to output the 6399assembler code on. @var{x} is the RTL expression for the constant to 6400output, and @var{mode} is the machine mode (in case @var{x} is a 6401@samp{const_int}). @var{align} is the required alignment for the value 6402@var{x}; you should output an assembler directive to force this much 6403alignment. 6404 6405The argument @var{labelno} is a number to use in an internal label for 6406the address of this pool entry. The definition of this macro is 6407responsible for outputting the label definition at the proper place. 6408Here is how to do this: 6409 6410@smallexample 6411@code{(*targetm.asm_out.internal_label)} (@var{file}, "LC", @var{labelno}); 6412@end smallexample 6413 6414When you output a pool entry specially, you should end with a 6415@code{goto} to the label @var{jumpto}. This will prevent the same pool 6416entry from being output a second time in the usual manner. 6417 6418You need not define this macro if it would do nothing. 6419@end defmac 6420 6421@defmac ASM_OUTPUT_POOL_EPILOGUE (@var{file} @var{funname} @var{fundecl} @var{size}) 6422A C statement to output assembler commands to at the end of the constant 6423pool for a function. @var{funname} is a string giving the name of the 6424function. Should the return type of the function be required, you can 6425obtain it via @var{fundecl}. @var{size} is the size, in bytes, of the 6426constant pool that GCC wrote immediately before this call. 6427 6428If no constant-pool epilogue is required, the usual case, you need not 6429define this macro. 6430@end defmac 6431 6432@defmac IS_ASM_LOGICAL_LINE_SEPARATOR (@var{C}) 6433Define this macro as a C expression which is nonzero if @var{C} is 6434used as a logical line separator by the assembler. 6435 6436If you do not define this macro, the default is that only 6437the character @samp{;} is treated as a logical line separator. 6438@end defmac 6439 6440@deftypevr {Target Hook} {const char *} TARGET_ASM_OPEN_PAREN 6441@deftypevrx {Target Hook} {const char *} TARGET_ASM_CLOSE_PAREN 6442These target hooks are C string constants, describing the syntax in the 6443assembler for grouping arithmetic expressions. If not overridden, they 6444default to normal parentheses, which is correct for most assemblers. 6445@end deftypevr 6446 6447 These macros are provided by @file{real.h} for writing the definitions 6448of @code{ASM_OUTPUT_DOUBLE} and the like: 6449 6450@defmac REAL_VALUE_TO_TARGET_SINGLE (@var{x}, @var{l}) 6451@defmacx REAL_VALUE_TO_TARGET_DOUBLE (@var{x}, @var{l}) 6452@defmacx REAL_VALUE_TO_TARGET_LONG_DOUBLE (@var{x}, @var{l}) 6453These translate @var{x}, of type @code{REAL_VALUE_TYPE}, to the target's 6454floating point representation, and store its bit pattern in the variable 6455@var{l}. For @code{REAL_VALUE_TO_TARGET_SINGLE}, this variable should 6456be a simple @code{long int}. For the others, it should be an array of 6457@code{long int}. The number of elements in this array is determined by 6458the size of the desired target floating point data type: 32 bits of it 6459go in each @code{long int} array element. Each array element holds 32 6460bits of the result, even if @code{long int} is wider than 32 bits on the 6461host machine. 6462 6463The array element values are designed so that you can print them out 6464using @code{fprintf} in the order they should appear in the target 6465machine's memory. 6466@end defmac 6467 6468@node Uninitialized Data 6469@subsection Output of Uninitialized Variables 6470 6471Each of the macros in this section is used to do the whole job of 6472outputting a single uninitialized variable. 6473 6474@defmac ASM_OUTPUT_COMMON (@var{stream}, @var{name}, @var{size}, @var{rounded}) 6475A C statement (sans semicolon) to output to the stdio stream 6476@var{stream} the assembler definition of a common-label named 6477@var{name} whose size is @var{size} bytes. The variable @var{rounded} 6478is the size rounded up to whatever alignment the caller wants. 6479 6480Use the expression @code{assemble_name (@var{stream}, @var{name})} to 6481output the name itself; before and after that, output the additional 6482assembler syntax for defining the name, and a newline. 6483 6484This macro controls how the assembler definitions of uninitialized 6485common global variables are output. 6486@end defmac 6487 6488@defmac ASM_OUTPUT_ALIGNED_COMMON (@var{stream}, @var{name}, @var{size}, @var{alignment}) 6489Like @code{ASM_OUTPUT_COMMON} except takes the required alignment as a 6490separate, explicit argument. If you define this macro, it is used in 6491place of @code{ASM_OUTPUT_COMMON}, and gives you more flexibility in 6492handling the required alignment of the variable. The alignment is specified 6493as the number of bits. 6494@end defmac 6495 6496@defmac ASM_OUTPUT_ALIGNED_DECL_COMMON (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment}) 6497Like @code{ASM_OUTPUT_ALIGNED_COMMON} except that @var{decl} of the 6498variable to be output, if there is one, or @code{NULL_TREE} if there 6499is no corresponding variable. If you define this macro, GCC will use it 6500in place of both @code{ASM_OUTPUT_COMMON} and 6501@code{ASM_OUTPUT_ALIGNED_COMMON}. Define this macro when you need to see 6502the variable's decl in order to chose what to output. 6503@end defmac 6504 6505@defmac ASM_OUTPUT_SHARED_COMMON (@var{stream}, @var{name}, @var{size}, @var{rounded}) 6506If defined, it is similar to @code{ASM_OUTPUT_COMMON}, except that it 6507is used when @var{name} is shared. If not defined, @code{ASM_OUTPUT_COMMON} 6508will be used. 6509@end defmac 6510 6511@defmac ASM_OUTPUT_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{rounded}) 6512A C statement (sans semicolon) to output to the stdio stream 6513@var{stream} the assembler definition of uninitialized global @var{decl} named 6514@var{name} whose size is @var{size} bytes. The variable @var{rounded} 6515is the size rounded up to whatever alignment the caller wants. 6516 6517Try to use function @code{asm_output_bss} defined in @file{varasm.c} when 6518defining this macro. If unable, use the expression 6519@code{assemble_name (@var{stream}, @var{name})} to output the name itself; 6520before and after that, output the additional assembler syntax for defining 6521the name, and a newline. 6522 6523This macro controls how the assembler definitions of uninitialized global 6524variables are output. This macro exists to properly support languages like 6525C++ which do not have @code{common} data. However, this macro currently 6526is not defined for all targets. If this macro and 6527@code{ASM_OUTPUT_ALIGNED_BSS} are not defined then @code{ASM_OUTPUT_COMMON} 6528or @code{ASM_OUTPUT_ALIGNED_COMMON} or 6529@code{ASM_OUTPUT_ALIGNED_DECL_COMMON} is used. 6530@end defmac 6531 6532@defmac ASM_OUTPUT_ALIGNED_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment}) 6533Like @code{ASM_OUTPUT_BSS} except takes the required alignment as a 6534separate, explicit argument. If you define this macro, it is used in 6535place of @code{ASM_OUTPUT_BSS}, and gives you more flexibility in 6536handling the required alignment of the variable. The alignment is specified 6537as the number of bits. 6538 6539Try to use function @code{asm_output_aligned_bss} defined in file 6540@file{varasm.c} when defining this macro. 6541@end defmac 6542 6543@defmac ASM_OUTPUT_SHARED_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{rounded}) 6544If defined, it is similar to @code{ASM_OUTPUT_BSS}, except that it 6545is used when @var{name} is shared. If not defined, @code{ASM_OUTPUT_BSS} 6546will be used. 6547@end defmac 6548 6549@defmac ASM_OUTPUT_LOCAL (@var{stream}, @var{name}, @var{size}, @var{rounded}) 6550A C statement (sans semicolon) to output to the stdio stream 6551@var{stream} the assembler definition of a local-common-label named 6552@var{name} whose size is @var{size} bytes. The variable @var{rounded} 6553is the size rounded up to whatever alignment the caller wants. 6554 6555Use the expression @code{assemble_name (@var{stream}, @var{name})} to 6556output the name itself; before and after that, output the additional 6557assembler syntax for defining the name, and a newline. 6558 6559This macro controls how the assembler definitions of uninitialized 6560static variables are output. 6561@end defmac 6562 6563@defmac ASM_OUTPUT_ALIGNED_LOCAL (@var{stream}, @var{name}, @var{size}, @var{alignment}) 6564Like @code{ASM_OUTPUT_LOCAL} except takes the required alignment as a 6565separate, explicit argument. If you define this macro, it is used in 6566place of @code{ASM_OUTPUT_LOCAL}, and gives you more flexibility in 6567handling the required alignment of the variable. The alignment is specified 6568as the number of bits. 6569@end defmac 6570 6571@defmac ASM_OUTPUT_ALIGNED_DECL_LOCAL (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment}) 6572Like @code{ASM_OUTPUT_ALIGNED_DECL} except that @var{decl} of the 6573variable to be output, if there is one, or @code{NULL_TREE} if there 6574is no corresponding variable. If you define this macro, GCC will use it 6575in place of both @code{ASM_OUTPUT_DECL} and 6576@code{ASM_OUTPUT_ALIGNED_DECL}. Define this macro when you need to see 6577the variable's decl in order to chose what to output. 6578@end defmac 6579 6580@defmac ASM_OUTPUT_SHARED_LOCAL (@var{stream}, @var{name}, @var{size}, @var{rounded}) 6581If defined, it is similar to @code{ASM_OUTPUT_LOCAL}, except that it 6582is used when @var{name} is shared. If not defined, @code{ASM_OUTPUT_LOCAL} 6583will be used. 6584@end defmac 6585 6586@node Label Output 6587@subsection Output and Generation of Labels 6588 6589@c prevent bad page break with this line 6590This is about outputting labels. 6591 6592@findex assemble_name 6593@defmac ASM_OUTPUT_LABEL (@var{stream}, @var{name}) 6594A C statement (sans semicolon) to output to the stdio stream 6595@var{stream} the assembler definition of a label named @var{name}. 6596Use the expression @code{assemble_name (@var{stream}, @var{name})} to 6597output the name itself; before and after that, output the additional 6598assembler syntax for defining the name, and a newline. A default 6599definition of this macro is provided which is correct for most systems. 6600@end defmac 6601 6602@defmac SIZE_ASM_OP 6603A C string containing the appropriate assembler directive to specify the 6604size of a symbol, without any arguments. On systems that use ELF, the 6605default (in @file{config/elfos.h}) is @samp{"\t.size\t"}; on other 6606systems, the default is not to define this macro. 6607 6608Define this macro only if it is correct to use the default definitions 6609of @code{ASM_OUTPUT_SIZE_DIRECTIVE} and @code{ASM_OUTPUT_MEASURED_SIZE} 6610for your system. If you need your own custom definitions of those 6611macros, or if you do not need explicit symbol sizes at all, do not 6612define this macro. 6613@end defmac 6614 6615@defmac ASM_OUTPUT_SIZE_DIRECTIVE (@var{stream}, @var{name}, @var{size}) 6616A C statement (sans semicolon) to output to the stdio stream 6617@var{stream} a directive telling the assembler that the size of the 6618symbol @var{name} is @var{size}. @var{size} is a @code{HOST_WIDE_INT}. 6619If you define @code{SIZE_ASM_OP}, a default definition of this macro is 6620provided. 6621@end defmac 6622 6623@defmac ASM_OUTPUT_MEASURED_SIZE (@var{stream}, @var{name}) 6624A C statement (sans semicolon) to output to the stdio stream 6625@var{stream} a directive telling the assembler to calculate the size of 6626the symbol @var{name} by subtracting its address from the current 6627address. 6628 6629If you define @code{SIZE_ASM_OP}, a default definition of this macro is 6630provided. The default assumes that the assembler recognizes a special 6631@samp{.} symbol as referring to the current address, and can calculate 6632the difference between this and another symbol. If your assembler does 6633not recognize @samp{.} or cannot do calculations with it, you will need 6634to redefine @code{ASM_OUTPUT_MEASURED_SIZE} to use some other technique. 6635@end defmac 6636 6637@defmac TYPE_ASM_OP 6638A C string containing the appropriate assembler directive to specify the 6639type of a symbol, without any arguments. On systems that use ELF, the 6640default (in @file{config/elfos.h}) is @samp{"\t.type\t"}; on other 6641systems, the default is not to define this macro. 6642 6643Define this macro only if it is correct to use the default definition of 6644@code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own 6645custom definition of this macro, or if you do not need explicit symbol 6646types at all, do not define this macro. 6647@end defmac 6648 6649@defmac TYPE_OPERAND_FMT 6650A C string which specifies (using @code{printf} syntax) the format of 6651the second operand to @code{TYPE_ASM_OP}. On systems that use ELF, the 6652default (in @file{config/elfos.h}) is @samp{"@@%s"}; on other systems, 6653the default is not to define this macro. 6654 6655Define this macro only if it is correct to use the default definition of 6656@code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own 6657custom definition of this macro, or if you do not need explicit symbol 6658types at all, do not define this macro. 6659@end defmac 6660 6661@defmac ASM_OUTPUT_TYPE_DIRECTIVE (@var{stream}, @var{type}) 6662A C statement (sans semicolon) to output to the stdio stream 6663@var{stream} a directive telling the assembler that the type of the 6664symbol @var{name} is @var{type}. @var{type} is a C string; currently, 6665that string is always either @samp{"function"} or @samp{"object"}, but 6666you should not count on this. 6667 6668If you define @code{TYPE_ASM_OP} and @code{TYPE_OPERAND_FMT}, a default 6669definition of this macro is provided. 6670@end defmac 6671 6672@defmac ASM_DECLARE_FUNCTION_NAME (@var{stream}, @var{name}, @var{decl}) 6673A C statement (sans semicolon) to output to the stdio stream 6674@var{stream} any text necessary for declaring the name @var{name} of a 6675function which is being defined. This macro is responsible for 6676outputting the label definition (perhaps using 6677@code{ASM_OUTPUT_LABEL}). The argument @var{decl} is the 6678@code{FUNCTION_DECL} tree node representing the function. 6679 6680If this macro is not defined, then the function name is defined in the 6681usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}). 6682 6683You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in the definition 6684of this macro. 6685@end defmac 6686 6687@defmac ASM_DECLARE_FUNCTION_SIZE (@var{stream}, @var{name}, @var{decl}) 6688A C statement (sans semicolon) to output to the stdio stream 6689@var{stream} any text necessary for declaring the size of a function 6690which is being defined. The argument @var{name} is the name of the 6691function. The argument @var{decl} is the @code{FUNCTION_DECL} tree node 6692representing the function. 6693 6694If this macro is not defined, then the function size is not defined. 6695 6696You may wish to use @code{ASM_OUTPUT_MEASURED_SIZE} in the definition 6697of this macro. 6698@end defmac 6699 6700@defmac ASM_DECLARE_OBJECT_NAME (@var{stream}, @var{name}, @var{decl}) 6701A C statement (sans semicolon) to output to the stdio stream 6702@var{stream} any text necessary for declaring the name @var{name} of an 6703initialized variable which is being defined. This macro must output the 6704label definition (perhaps using @code{ASM_OUTPUT_LABEL}). The argument 6705@var{decl} is the @code{VAR_DECL} tree node representing the variable. 6706 6707If this macro is not defined, then the variable name is defined in the 6708usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}). 6709 6710You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} and/or 6711@code{ASM_OUTPUT_SIZE_DIRECTIVE} in the definition of this macro. 6712@end defmac 6713 6714@defmac ASM_DECLARE_CONSTANT_NAME (@var{stream}, @var{name}, @var{exp}, @var{size}) 6715A C statement (sans semicolon) to output to the stdio stream 6716@var{stream} any text necessary for declaring the name @var{name} of a 6717constant which is being defined. This macro is responsible for 6718outputting the label definition (perhaps using 6719@code{ASM_OUTPUT_LABEL}). The argument @var{exp} is the 6720value of the constant, and @var{size} is the size of the constant 6721in bytes. @var{name} will be an internal label. 6722 6723If this macro is not defined, then the @var{name} is defined in the 6724usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}). 6725 6726You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in the definition 6727of this macro. 6728@end defmac 6729 6730@defmac ASM_DECLARE_REGISTER_GLOBAL (@var{stream}, @var{decl}, @var{regno}, @var{name}) 6731A C statement (sans semicolon) to output to the stdio stream 6732@var{stream} any text necessary for claiming a register @var{regno} 6733for a global variable @var{decl} with name @var{name}. 6734 6735If you don't define this macro, that is equivalent to defining it to do 6736nothing. 6737@end defmac 6738 6739@defmac ASM_FINISH_DECLARE_OBJECT (@var{stream}, @var{decl}, @var{toplevel}, @var{atend}) 6740A C statement (sans semicolon) to finish up declaring a variable name 6741once the compiler has processed its initializer fully and thus has had a 6742chance to determine the size of an array when controlled by an 6743initializer. This is used on systems where it's necessary to declare 6744something about the size of the object. 6745 6746If you don't define this macro, that is equivalent to defining it to do 6747nothing. 6748 6749You may wish to use @code{ASM_OUTPUT_SIZE_DIRECTIVE} and/or 6750@code{ASM_OUTPUT_MEASURED_SIZE} in the definition of this macro. 6751@end defmac 6752 6753@deftypefn {Target Hook} void TARGET_ASM_GLOBALIZE_LABEL (FILE *@var{stream}, const char *@var{name}) 6754This target hook is a function to output to the stdio stream 6755@var{stream} some commands that will make the label @var{name} global; 6756that is, available for reference from other files. 6757 6758The default implementation relies on a proper definition of 6759@code{GLOBAL_ASM_OP}. 6760@end deftypefn 6761 6762@defmac ASM_WEAKEN_LABEL (@var{stream}, @var{name}) 6763A C statement (sans semicolon) to output to the stdio stream 6764@var{stream} some commands that will make the label @var{name} weak; 6765that is, available for reference from other files but only used if 6766no other definition is available. Use the expression 6767@code{assemble_name (@var{stream}, @var{name})} to output the name 6768itself; before and after that, output the additional assembler syntax 6769for making that name weak, and a newline. 6770 6771If you don't define this macro or @code{ASM_WEAKEN_DECL}, GCC will not 6772support weak symbols and you should not define the @code{SUPPORTS_WEAK} 6773macro. 6774@end defmac 6775 6776@defmac ASM_WEAKEN_DECL (@var{stream}, @var{decl}, @var{name}, @var{value}) 6777Combines (and replaces) the function of @code{ASM_WEAKEN_LABEL} and 6778@code{ASM_OUTPUT_WEAK_ALIAS}, allowing access to the associated function 6779or variable decl. If @var{value} is not @code{NULL}, this C statement 6780should output to the stdio stream @var{stream} assembler code which 6781defines (equates) the weak symbol @var{name} to have the value 6782@var{value}. If @var{value} is @code{NULL}, it should output commands 6783to make @var{name} weak. 6784@end defmac 6785 6786@defmac SUPPORTS_WEAK 6787A C expression which evaluates to true if the target supports weak symbols. 6788 6789If you don't define this macro, @file{defaults.h} provides a default 6790definition. If either @code{ASM_WEAKEN_LABEL} or @code{ASM_WEAKEN_DECL} 6791is defined, the default definition is @samp{1}; otherwise, it is 6792@samp{0}. Define this macro if you want to control weak symbol support 6793with a compiler flag such as @option{-melf}. 6794@end defmac 6795 6796@defmac MAKE_DECL_ONE_ONLY (@var{decl}) 6797A C statement (sans semicolon) to mark @var{decl} to be emitted as a 6798public symbol such that extra copies in multiple translation units will 6799be discarded by the linker. Define this macro if your object file 6800format provides support for this concept, such as the @samp{COMDAT} 6801section flags in the Microsoft Windows PE/COFF format, and this support 6802requires changes to @var{decl}, such as putting it in a separate section. 6803@end defmac 6804 6805@defmac SUPPORTS_ONE_ONLY 6806A C expression which evaluates to true if the target supports one-only 6807semantics. 6808 6809If you don't define this macro, @file{varasm.c} provides a default 6810definition. If @code{MAKE_DECL_ONE_ONLY} is defined, the default 6811definition is @samp{1}; otherwise, it is @samp{0}. Define this macro if 6812you want to control one-only symbol support with a compiler flag, or if 6813setting the @code{DECL_ONE_ONLY} flag is enough to mark a declaration to 6814be emitted as one-only. 6815@end defmac 6816 6817@deftypefn {Target Hook} void TARGET_ASM_ASSEMBLE_VISIBILITY (tree @var{decl}, const char *@var{visibility}) 6818This target hook is a function to output to @var{asm_out_file} some 6819commands that will make the symbol(s) associated with @var{decl} have 6820hidden, protected or internal visibility as specified by @var{visibility}. 6821@end deftypefn 6822 6823@defmac ASM_OUTPUT_EXTERNAL (@var{stream}, @var{decl}, @var{name}) 6824A C statement (sans semicolon) to output to the stdio stream 6825@var{stream} any text necessary for declaring the name of an external 6826symbol named @var{name} which is referenced in this compilation but 6827not defined. The value of @var{decl} is the tree node for the 6828declaration. 6829 6830This macro need not be defined if it does not need to output anything. 6831The GNU assembler and most Unix assemblers don't require anything. 6832@end defmac 6833 6834@deftypefn {Target Hook} void TARGET_ASM_EXTERNAL_LIBCALL (rtx @var{symref}) 6835This target hook is a function to output to @var{asm_out_file} an assembler 6836pseudo-op to declare a library function name external. The name of the 6837library function is given by @var{symref}, which is a @code{symbol_ref}. 6838@end deftypefn 6839 6840@defmac ASM_OUTPUT_LABELREF (@var{stream}, @var{name}) 6841A C statement (sans semicolon) to output to the stdio stream 6842@var{stream} a reference in assembler syntax to a label named 6843@var{name}. This should add @samp{_} to the front of the name, if that 6844is customary on your operating system, as it is in most Berkeley Unix 6845systems. This macro is used in @code{assemble_name}. 6846@end defmac 6847 6848@defmac ASM_OUTPUT_SYMBOL_REF (@var{stream}, @var{sym}) 6849A C statement (sans semicolon) to output a reference to 6850@code{SYMBOL_REF} @var{sym}. If not defined, @code{assemble_name} 6851will be used to output the name of the symbol. This macro may be used 6852to modify the way a symbol is referenced depending on information 6853encoded by @code{TARGET_ENCODE_SECTION_INFO}. 6854@end defmac 6855 6856@defmac ASM_OUTPUT_LABEL_REF (@var{stream}, @var{buf}) 6857A C statement (sans semicolon) to output a reference to @var{buf}, the 6858result of @code{ASM_GENERATE_INTERNAL_LABEL}. If not defined, 6859@code{assemble_name} will be used to output the name of the symbol. 6860This macro is not used by @code{output_asm_label}, or the @code{%l} 6861specifier that calls it; the intention is that this macro should be set 6862when it is necessary to output a label differently when its address is 6863being taken. 6864@end defmac 6865 6866@deftypefn {Target Hook} void TARGET_ASM_INTERNAL_LABEL (FILE *@var{stream}, const char *@var{prefix}, unsigned long @var{labelno}) 6867A function to output to the stdio stream @var{stream} a label whose 6868name is made from the string @var{prefix} and the number @var{labelno}. 6869 6870It is absolutely essential that these labels be distinct from the labels 6871used for user-level functions and variables. Otherwise, certain programs 6872will have name conflicts with internal labels. 6873 6874It is desirable to exclude internal labels from the symbol table of the 6875object file. Most assemblers have a naming convention for labels that 6876should be excluded; on many systems, the letter @samp{L} at the 6877beginning of a label has this effect. You should find out what 6878convention your system uses, and follow it. 6879 6880The default version of this function utilizes ASM_GENERATE_INTERNAL_LABEL. 6881@end deftypefn 6882 6883@defmac ASM_OUTPUT_DEBUG_LABEL (@var{stream}, @var{prefix}, @var{num}) 6884A C statement to output to the stdio stream @var{stream} a debug info 6885label whose name is made from the string @var{prefix} and the number 6886@var{num}. This is useful for VLIW targets, where debug info labels 6887may need to be treated differently than branch target labels. On some 6888systems, branch target labels must be at the beginning of instruction 6889bundles, but debug info labels can occur in the middle of instruction 6890bundles. 6891 6892If this macro is not defined, then @code{(*targetm.asm_out.internal_label)} will be 6893used. 6894@end defmac 6895 6896@defmac ASM_GENERATE_INTERNAL_LABEL (@var{string}, @var{prefix}, @var{num}) 6897A C statement to store into the string @var{string} a label whose name 6898is made from the string @var{prefix} and the number @var{num}. 6899 6900This string, when output subsequently by @code{assemble_name}, should 6901produce the output that @code{(*targetm.asm_out.internal_label)} would produce 6902with the same @var{prefix} and @var{num}. 6903 6904If the string begins with @samp{*}, then @code{assemble_name} will 6905output the rest of the string unchanged. It is often convenient for 6906@code{ASM_GENERATE_INTERNAL_LABEL} to use @samp{*} in this way. If the 6907string doesn't start with @samp{*}, then @code{ASM_OUTPUT_LABELREF} gets 6908to output the string, and may change it. (Of course, 6909@code{ASM_OUTPUT_LABELREF} is also part of your machine description, so 6910you should know what it does on your machine.) 6911@end defmac 6912 6913@defmac ASM_FORMAT_PRIVATE_NAME (@var{outvar}, @var{name}, @var{number}) 6914A C expression to assign to @var{outvar} (which is a variable of type 6915@code{char *}) a newly allocated string made from the string 6916@var{name} and the number @var{number}, with some suitable punctuation 6917added. Use @code{alloca} to get space for the string. 6918 6919The string will be used as an argument to @code{ASM_OUTPUT_LABELREF} to 6920produce an assembler label for an internal static variable whose name is 6921@var{name}. Therefore, the string must be such as to result in valid 6922assembler code. The argument @var{number} is different each time this 6923macro is executed; it prevents conflicts between similarly-named 6924internal static variables in different scopes. 6925 6926Ideally this string should not be a valid C identifier, to prevent any 6927conflict with the user's own symbols. Most assemblers allow periods 6928or percent signs in assembler symbols; putting at least one of these 6929between the name and the number will suffice. 6930 6931If this macro is not defined, a default definition will be provided 6932which is correct for most systems. 6933@end defmac 6934 6935@defmac ASM_OUTPUT_DEF (@var{stream}, @var{name}, @var{value}) 6936A C statement to output to the stdio stream @var{stream} assembler code 6937which defines (equates) the symbol @var{name} to have the value @var{value}. 6938 6939@findex SET_ASM_OP 6940If @code{SET_ASM_OP} is defined, a default definition is provided which is 6941correct for most systems. 6942@end defmac 6943 6944@defmac ASM_OUTPUT_DEF_FROM_DECLS (@var{stream}, @var{decl_of_name}, @var{decl_of_value}) 6945A C statement to output to the stdio stream @var{stream} assembler code 6946which defines (equates) the symbol whose tree node is @var{decl_of_name} 6947to have the value of the tree node @var{decl_of_value}. This macro will 6948be used in preference to @samp{ASM_OUTPUT_DEF} if it is defined and if 6949the tree nodes are available. 6950 6951@findex SET_ASM_OP 6952If @code{SET_ASM_OP} is defined, a default definition is provided which is 6953correct for most systems. 6954@end defmac 6955 6956@defmac TARGET_DEFERRED_OUTPUT_DEFS (@var{decl_of_name}, @var{decl_of_value}) 6957A C statement that evaluates to true if the assembler code which defines 6958(equates) the symbol whose tree node is @var{decl_of_name} to have the value 6959of the tree node @var{decl_of_value} should be emitted near the end of the 6960current compilation unit. The default is to not defer output of defines. 6961This macro affects defines output by @samp{ASM_OUTPUT_DEF} and 6962@samp{ASM_OUTPUT_DEF_FROM_DECLS}. 6963@end defmac 6964 6965@defmac ASM_OUTPUT_WEAK_ALIAS (@var{stream}, @var{name}, @var{value}) 6966A C statement to output to the stdio stream @var{stream} assembler code 6967which defines (equates) the weak symbol @var{name} to have the value 6968@var{value}. If @var{value} is @code{NULL}, it defines @var{name} as 6969an undefined weak symbol. 6970 6971Define this macro if the target only supports weak aliases; define 6972@code{ASM_OUTPUT_DEF} instead if possible. 6973@end defmac 6974 6975@defmac OBJC_GEN_METHOD_LABEL (@var{buf}, @var{is_inst}, @var{class_name}, @var{cat_name}, @var{sel_name}) 6976Define this macro to override the default assembler names used for 6977Objective-C methods. 6978 6979The default name is a unique method number followed by the name of the 6980class (e.g.@: @samp{_1_Foo}). For methods in categories, the name of 6981the category is also included in the assembler name (e.g.@: 6982@samp{_1_Foo_Bar}). 6983 6984These names are safe on most systems, but make debugging difficult since 6985the method's selector is not present in the name. Therefore, particular 6986systems define other ways of computing names. 6987 6988@var{buf} is an expression of type @code{char *} which gives you a 6989buffer in which to store the name; its length is as long as 6990@var{class_name}, @var{cat_name} and @var{sel_name} put together, plus 699150 characters extra. 6992 6993The argument @var{is_inst} specifies whether the method is an instance 6994method or a class method; @var{class_name} is the name of the class; 6995@var{cat_name} is the name of the category (or @code{NULL} if the method is not 6996in a category); and @var{sel_name} is the name of the selector. 6997 6998On systems where the assembler can handle quoted names, you can use this 6999macro to provide more human-readable names. 7000@end defmac 7001 7002@defmac ASM_DECLARE_CLASS_REFERENCE (@var{stream}, @var{name}) 7003A C statement (sans semicolon) to output to the stdio stream 7004@var{stream} commands to declare that the label @var{name} is an 7005Objective-C class reference. This is only needed for targets whose 7006linkers have special support for NeXT-style runtimes. 7007@end defmac 7008 7009@defmac ASM_DECLARE_UNRESOLVED_REFERENCE (@var{stream}, @var{name}) 7010A C statement (sans semicolon) to output to the stdio stream 7011@var{stream} commands to declare that the label @var{name} is an 7012unresolved Objective-C class reference. This is only needed for targets 7013whose linkers have special support for NeXT-style runtimes. 7014@end defmac 7015 7016@node Initialization 7017@subsection How Initialization Functions Are Handled 7018@cindex initialization routines 7019@cindex termination routines 7020@cindex constructors, output of 7021@cindex destructors, output of 7022 7023The compiled code for certain languages includes @dfn{constructors} 7024(also called @dfn{initialization routines})---functions to initialize 7025data in the program when the program is started. These functions need 7026to be called before the program is ``started''---that is to say, before 7027@code{main} is called. 7028 7029Compiling some languages generates @dfn{destructors} (also called 7030@dfn{termination routines}) that should be called when the program 7031terminates. 7032 7033To make the initialization and termination functions work, the compiler 7034must output something in the assembler code to cause those functions to 7035be called at the appropriate time. When you port the compiler to a new 7036system, you need to specify how to do this. 7037 7038There are two major ways that GCC currently supports the execution of 7039initialization and termination functions. Each way has two variants. 7040Much of the structure is common to all four variations. 7041 7042@findex __CTOR_LIST__ 7043@findex __DTOR_LIST__ 7044The linker must build two lists of these functions---a list of 7045initialization functions, called @code{__CTOR_LIST__}, and a list of 7046termination functions, called @code{__DTOR_LIST__}. 7047 7048Each list always begins with an ignored function pointer (which may hold 70490, @minus{}1, or a count of the function pointers after it, depending on 7050the environment). This is followed by a series of zero or more function 7051pointers to constructors (or destructors), followed by a function 7052pointer containing zero. 7053 7054Depending on the operating system and its executable file format, either 7055@file{crtstuff.c} or @file{libgcc2.c} traverses these lists at startup 7056time and exit time. Constructors are called in reverse order of the 7057list; destructors in forward order. 7058 7059The best way to handle static constructors works only for object file 7060formats which provide arbitrarily-named sections. A section is set 7061aside for a list of constructors, and another for a list of destructors. 7062Traditionally these are called @samp{.ctors} and @samp{.dtors}. Each 7063object file that defines an initialization function also puts a word in 7064the constructor section to point to that function. The linker 7065accumulates all these words into one contiguous @samp{.ctors} section. 7066Termination functions are handled similarly. 7067 7068This method will be chosen as the default by @file{target-def.h} if 7069@code{TARGET_ASM_NAMED_SECTION} is defined. A target that does not 7070support arbitrary sections, but does support special designated 7071constructor and destructor sections may define @code{CTORS_SECTION_ASM_OP} 7072and @code{DTORS_SECTION_ASM_OP} to achieve the same effect. 7073 7074When arbitrary sections are available, there are two variants, depending 7075upon how the code in @file{crtstuff.c} is called. On systems that 7076support a @dfn{.init} section which is executed at program startup, 7077parts of @file{crtstuff.c} are compiled into that section. The 7078program is linked by the @command{gcc} driver like this: 7079 7080@smallexample 7081ld -o @var{output_file} crti.o crtbegin.o @dots{} -lgcc crtend.o crtn.o 7082@end smallexample 7083 7084The prologue of a function (@code{__init}) appears in the @code{.init} 7085section of @file{crti.o}; the epilogue appears in @file{crtn.o}. Likewise 7086for the function @code{__fini} in the @dfn{.fini} section. Normally these 7087files are provided by the operating system or by the GNU C library, but 7088are provided by GCC for a few targets. 7089 7090The objects @file{crtbegin.o} and @file{crtend.o} are (for most targets) 7091compiled from @file{crtstuff.c}. They contain, among other things, code 7092fragments within the @code{.init} and @code{.fini} sections that branch 7093to routines in the @code{.text} section. The linker will pull all parts 7094of a section together, which results in a complete @code{__init} function 7095that invokes the routines we need at startup. 7096 7097To use this variant, you must define the @code{INIT_SECTION_ASM_OP} 7098macro properly. 7099 7100If no init section is available, when GCC compiles any function called 7101@code{main} (or more accurately, any function designated as a program 7102entry point by the language front end calling @code{expand_main_function}), 7103it inserts a procedure call to @code{__main} as the first executable code 7104after the function prologue. The @code{__main} function is defined 7105in @file{libgcc2.c} and runs the global constructors. 7106 7107In file formats that don't support arbitrary sections, there are again 7108two variants. In the simplest variant, the GNU linker (GNU @code{ld}) 7109and an `a.out' format must be used. In this case, 7110@code{TARGET_ASM_CONSTRUCTOR} is defined to produce a @code{.stabs} 7111entry of type @samp{N_SETT}, referencing the name @code{__CTOR_LIST__}, 7112and with the address of the void function containing the initialization 7113code as its value. The GNU linker recognizes this as a request to add 7114the value to a @dfn{set}; the values are accumulated, and are eventually 7115placed in the executable as a vector in the format described above, with 7116a leading (ignored) count and a trailing zero element. 7117@code{TARGET_ASM_DESTRUCTOR} is handled similarly. Since no init 7118section is available, the absence of @code{INIT_SECTION_ASM_OP} causes 7119the compilation of @code{main} to call @code{__main} as above, starting 7120the initialization process. 7121 7122The last variant uses neither arbitrary sections nor the GNU linker. 7123This is preferable when you want to do dynamic linking and when using 7124file formats which the GNU linker does not support, such as `ECOFF'@. In 7125this case, @code{TARGET_HAVE_CTORS_DTORS} is false, initialization and 7126termination functions are recognized simply by their names. This requires 7127an extra program in the linkage step, called @command{collect2}. This program 7128pretends to be the linker, for use with GCC; it does its job by running 7129the ordinary linker, but also arranges to include the vectors of 7130initialization and termination functions. These functions are called 7131via @code{__main} as described above. In order to use this method, 7132@code{use_collect2} must be defined in the target in @file{config.gcc}. 7133 7134@ifinfo 7135The following section describes the specific macros that control and 7136customize the handling of initialization and termination functions. 7137@end ifinfo 7138 7139@node Macros for Initialization 7140@subsection Macros Controlling Initialization Routines 7141 7142Here are the macros that control how the compiler handles initialization 7143and termination functions: 7144 7145@defmac INIT_SECTION_ASM_OP 7146If defined, a C string constant, including spacing, for the assembler 7147operation to identify the following data as initialization code. If not 7148defined, GCC will assume such a section does not exist. When you are 7149using special sections for initialization and termination functions, this 7150macro also controls how @file{crtstuff.c} and @file{libgcc2.c} arrange to 7151run the initialization functions. 7152@end defmac 7153 7154@defmac HAS_INIT_SECTION 7155If defined, @code{main} will not call @code{__main} as described above. 7156This macro should be defined for systems that control start-up code 7157on a symbol-by-symbol basis, such as OSF/1, and should not 7158be defined explicitly for systems that support @code{INIT_SECTION_ASM_OP}. 7159@end defmac 7160 7161@defmac LD_INIT_SWITCH 7162If defined, a C string constant for a switch that tells the linker that 7163the following symbol is an initialization routine. 7164@end defmac 7165 7166@defmac LD_FINI_SWITCH 7167If defined, a C string constant for a switch that tells the linker that 7168the following symbol is a finalization routine. 7169@end defmac 7170 7171@defmac COLLECT_SHARED_INIT_FUNC (@var{stream}, @var{func}) 7172If defined, a C statement that will write a function that can be 7173automatically called when a shared library is loaded. The function 7174should call @var{func}, which takes no arguments. If not defined, and 7175the object format requires an explicit initialization function, then a 7176function called @code{_GLOBAL__DI} will be generated. 7177 7178This function and the following one are used by collect2 when linking a 7179shared library that needs constructors or destructors, or has DWARF2 7180exception tables embedded in the code. 7181@end defmac 7182 7183@defmac COLLECT_SHARED_FINI_FUNC (@var{stream}, @var{func}) 7184If defined, a C statement that will write a function that can be 7185automatically called when a shared library is unloaded. The function 7186should call @var{func}, which takes no arguments. If not defined, and 7187the object format requires an explicit finalization function, then a 7188function called @code{_GLOBAL__DD} will be generated. 7189@end defmac 7190 7191@defmac INVOKE__main 7192If defined, @code{main} will call @code{__main} despite the presence of 7193@code{INIT_SECTION_ASM_OP}. This macro should be defined for systems 7194where the init section is not actually run automatically, but is still 7195useful for collecting the lists of constructors and destructors. 7196@end defmac 7197 7198@defmac SUPPORTS_INIT_PRIORITY 7199If nonzero, the C++ @code{init_priority} attribute is supported and the 7200compiler should emit instructions to control the order of initialization 7201of objects. If zero, the compiler will issue an error message upon 7202encountering an @code{init_priority} attribute. 7203@end defmac 7204 7205@deftypefn {Target Hook} bool TARGET_HAVE_CTORS_DTORS 7206This value is true if the target supports some ``native'' method of 7207collecting constructors and destructors to be run at startup and exit. 7208It is false if we must use @command{collect2}. 7209@end deftypefn 7210 7211@deftypefn {Target Hook} void TARGET_ASM_CONSTRUCTOR (rtx @var{symbol}, int @var{priority}) 7212If defined, a function that outputs assembler code to arrange to call 7213the function referenced by @var{symbol} at initialization time. 7214 7215Assume that @var{symbol} is a @code{SYMBOL_REF} for a function taking 7216no arguments and with no return value. If the target supports initialization 7217priorities, @var{priority} is a value between 0 and @code{MAX_INIT_PRIORITY}; 7218otherwise it must be @code{DEFAULT_INIT_PRIORITY}. 7219 7220If this macro is not defined by the target, a suitable default will 7221be chosen if (1) the target supports arbitrary section names, (2) the 7222target defines @code{CTORS_SECTION_ASM_OP}, or (3) @code{USE_COLLECT2} 7223is not defined. 7224@end deftypefn 7225 7226@deftypefn {Target Hook} void TARGET_ASM_DESTRUCTOR (rtx @var{symbol}, int @var{priority}) 7227This is like @code{TARGET_ASM_CONSTRUCTOR} but used for termination 7228functions rather than initialization functions. 7229@end deftypefn 7230 7231If @code{TARGET_HAVE_CTORS_DTORS} is true, the initialization routine 7232generated for the generated object file will have static linkage. 7233 7234If your system uses @command{collect2} as the means of processing 7235constructors, then that program normally uses @command{nm} to scan 7236an object file for constructor functions to be called. 7237 7238On certain kinds of systems, you can define this macro to make 7239@command{collect2} work faster (and, in some cases, make it work at all): 7240 7241@defmac OBJECT_FORMAT_COFF 7242Define this macro if the system uses COFF (Common Object File Format) 7243object files, so that @command{collect2} can assume this format and scan 7244object files directly for dynamic constructor/destructor functions. 7245 7246This macro is effective only in a native compiler; @command{collect2} as 7247part of a cross compiler always uses @command{nm} for the target machine. 7248@end defmac 7249 7250@defmac COLLECT_PARSE_FLAG (@var{flag}) 7251Define this macro to be C code that examines @command{collect2} command 7252line option @var{flag} and performs special actions if 7253@command{collect2} needs to behave differently depending on @var{flag}. 7254@end defmac 7255 7256@defmac REAL_NM_FILE_NAME 7257Define this macro as a C string constant containing the file name to use 7258to execute @command{nm}. The default is to search the path normally for 7259@command{nm}. 7260 7261If your system supports shared libraries and has a program to list the 7262dynamic dependencies of a given library or executable, you can define 7263these macros to enable support for running initialization and 7264termination functions in shared libraries: 7265@end defmac 7266 7267@defmac LDD_SUFFIX 7268Define this macro to a C string constant containing the name of the program 7269which lists dynamic dependencies, like @command{"ldd"} under SunOS 4. 7270@end defmac 7271 7272@defmac PARSE_LDD_OUTPUT (@var{ptr}) 7273Define this macro to be C code that extracts filenames from the output 7274of the program denoted by @code{LDD_SUFFIX}. @var{ptr} is a variable 7275of type @code{char *} that points to the beginning of a line of output 7276from @code{LDD_SUFFIX}. If the line lists a dynamic dependency, the 7277code must advance @var{ptr} to the beginning of the filename on that 7278line. Otherwise, it must set @var{ptr} to @code{NULL}. 7279@end defmac 7280 7281@node Instruction Output 7282@subsection Output of Assembler Instructions 7283 7284@c prevent bad page break with this line 7285This describes assembler instruction output. 7286 7287@defmac REGISTER_NAMES 7288A C initializer containing the assembler's names for the machine 7289registers, each one as a C string constant. This is what translates 7290register numbers in the compiler into assembler language. 7291@end defmac 7292 7293@defmac ADDITIONAL_REGISTER_NAMES 7294If defined, a C initializer for an array of structures containing a name 7295and a register number. This macro defines additional names for hard 7296registers, thus allowing the @code{asm} option in declarations to refer 7297to registers using alternate names. 7298@end defmac 7299 7300@defmac ASM_OUTPUT_OPCODE (@var{stream}, @var{ptr}) 7301Define this macro if you are using an unusual assembler that 7302requires different names for the machine instructions. 7303 7304The definition is a C statement or statements which output an 7305assembler instruction opcode to the stdio stream @var{stream}. The 7306macro-operand @var{ptr} is a variable of type @code{char *} which 7307points to the opcode name in its ``internal'' form---the form that is 7308written in the machine description. The definition should output the 7309opcode name to @var{stream}, performing any translation you desire, and 7310increment the variable @var{ptr} to point at the end of the opcode 7311so that it will not be output twice. 7312 7313In fact, your macro definition may process less than the entire opcode 7314name, or more than the opcode name; but if you want to process text 7315that includes @samp{%}-sequences to substitute operands, you must take 7316care of the substitution yourself. Just be sure to increment 7317@var{ptr} over whatever text should not be output normally. 7318 7319@findex recog_data.operand 7320If you need to look at the operand values, they can be found as the 7321elements of @code{recog_data.operand}. 7322 7323If the macro definition does nothing, the instruction is output 7324in the usual way. 7325@end defmac 7326 7327@defmac FINAL_PRESCAN_INSN (@var{insn}, @var{opvec}, @var{noperands}) 7328If defined, a C statement to be executed just prior to the output of 7329assembler code for @var{insn}, to modify the extracted operands so 7330they will be output differently. 7331 7332Here the argument @var{opvec} is the vector containing the operands 7333extracted from @var{insn}, and @var{noperands} is the number of 7334elements of the vector which contain meaningful data for this insn. 7335The contents of this vector are what will be used to convert the insn 7336template into assembler code, so you can change the assembler output 7337by changing the contents of the vector. 7338 7339This macro is useful when various assembler syntaxes share a single 7340file of instruction patterns; by defining this macro differently, you 7341can cause a large class of instructions to be output differently (such 7342as with rearranged operands). Naturally, variations in assembler 7343syntax affecting individual insn patterns ought to be handled by 7344writing conditional output routines in those patterns. 7345 7346If this macro is not defined, it is equivalent to a null statement. 7347@end defmac 7348 7349@defmac PRINT_OPERAND (@var{stream}, @var{x}, @var{code}) 7350A C compound statement to output to stdio stream @var{stream} the 7351assembler syntax for an instruction operand @var{x}. @var{x} is an 7352RTL expression. 7353 7354@var{code} is a value that can be used to specify one of several ways 7355of printing the operand. It is used when identical operands must be 7356printed differently depending on the context. @var{code} comes from 7357the @samp{%} specification that was used to request printing of the 7358operand. If the specification was just @samp{%@var{digit}} then 7359@var{code} is 0; if the specification was @samp{%@var{ltr} 7360@var{digit}} then @var{code} is the ASCII code for @var{ltr}. 7361 7362@findex reg_names 7363If @var{x} is a register, this macro should print the register's name. 7364The names can be found in an array @code{reg_names} whose type is 7365@code{char *[]}. @code{reg_names} is initialized from 7366@code{REGISTER_NAMES}. 7367 7368When the machine description has a specification @samp{%@var{punct}} 7369(a @samp{%} followed by a punctuation character), this macro is called 7370with a null pointer for @var{x} and the punctuation character for 7371@var{code}. 7372@end defmac 7373 7374@defmac PRINT_OPERAND_PUNCT_VALID_P (@var{code}) 7375A C expression which evaluates to true if @var{code} is a valid 7376punctuation character for use in the @code{PRINT_OPERAND} macro. If 7377@code{PRINT_OPERAND_PUNCT_VALID_P} is not defined, it means that no 7378punctuation characters (except for the standard one, @samp{%}) are used 7379in this way. 7380@end defmac 7381 7382@defmac PRINT_OPERAND_ADDRESS (@var{stream}, @var{x}) 7383A C compound statement to output to stdio stream @var{stream} the 7384assembler syntax for an instruction operand that is a memory reference 7385whose address is @var{x}. @var{x} is an RTL expression. 7386 7387@cindex @code{TARGET_ENCODE_SECTION_INFO} usage 7388On some machines, the syntax for a symbolic address depends on the 7389section that the address refers to. On these machines, define the hook 7390@code{TARGET_ENCODE_SECTION_INFO} to store the information into the 7391@code{symbol_ref}, and then check for it here. @xref{Assembler 7392Format}. 7393@end defmac 7394 7395@findex dbr_sequence_length 7396@defmac DBR_OUTPUT_SEQEND (@var{file}) 7397A C statement, to be executed after all slot-filler instructions have 7398been output. If necessary, call @code{dbr_sequence_length} to 7399determine the number of slots filled in a sequence (zero if not 7400currently outputting a sequence), to decide how many no-ops to output, 7401or whatever. 7402 7403Don't define this macro if it has nothing to do, but it is helpful in 7404reading assembly output if the extent of the delay sequence is made 7405explicit (e.g.@: with white space). 7406@end defmac 7407 7408@findex final_sequence 7409Note that output routines for instructions with delay slots must be 7410prepared to deal with not being output as part of a sequence 7411(i.e.@: when the scheduling pass is not run, or when no slot fillers could be 7412found.) The variable @code{final_sequence} is null when not 7413processing a sequence, otherwise it contains the @code{sequence} rtx 7414being output. 7415 7416@findex asm_fprintf 7417@defmac REGISTER_PREFIX 7418@defmacx LOCAL_LABEL_PREFIX 7419@defmacx USER_LABEL_PREFIX 7420@defmacx IMMEDIATE_PREFIX 7421If defined, C string expressions to be used for the @samp{%R}, @samp{%L}, 7422@samp{%U}, and @samp{%I} options of @code{asm_fprintf} (see 7423@file{final.c}). These are useful when a single @file{md} file must 7424support multiple assembler formats. In that case, the various @file{tm.h} 7425files can define these macros differently. 7426@end defmac 7427 7428@defmac ASM_FPRINTF_EXTENSIONS (@var{file}, @var{argptr}, @var{format}) 7429If defined this macro should expand to a series of @code{case} 7430statements which will be parsed inside the @code{switch} statement of 7431the @code{asm_fprintf} function. This allows targets to define extra 7432printf formats which may useful when generating their assembler 7433statements. Note that uppercase letters are reserved for future 7434generic extensions to asm_fprintf, and so are not available to target 7435specific code. The output file is given by the parameter @var{file}. 7436The varargs input pointer is @var{argptr} and the rest of the format 7437string, starting the character after the one that is being switched 7438upon, is pointed to by @var{format}. 7439@end defmac 7440 7441@defmac ASSEMBLER_DIALECT 7442If your target supports multiple dialects of assembler language (such as 7443different opcodes), define this macro as a C expression that gives the 7444numeric index of the assembler language dialect to use, with zero as the 7445first variant. 7446 7447If this macro is defined, you may use constructs of the form 7448@smallexample 7449@samp{@{option0|option1|option2@dots{}@}} 7450@end smallexample 7451@noindent 7452in the output templates of patterns (@pxref{Output Template}) or in the 7453first argument of @code{asm_fprintf}. This construct outputs 7454@samp{option0}, @samp{option1}, @samp{option2}, etc., if the value of 7455@code{ASSEMBLER_DIALECT} is zero, one, two, etc. Any special characters 7456within these strings retain their usual meaning. If there are fewer 7457alternatives within the braces than the value of 7458@code{ASSEMBLER_DIALECT}, the construct outputs nothing. 7459 7460If you do not define this macro, the characters @samp{@{}, @samp{|} and 7461@samp{@}} do not have any special meaning when used in templates or 7462operands to @code{asm_fprintf}. 7463 7464Define the macros @code{REGISTER_PREFIX}, @code{LOCAL_LABEL_PREFIX}, 7465@code{USER_LABEL_PREFIX} and @code{IMMEDIATE_PREFIX} if you can express 7466the variations in assembler language syntax with that mechanism. Define 7467@code{ASSEMBLER_DIALECT} and use the @samp{@{option0|option1@}} syntax 7468if the syntax variant are larger and involve such things as different 7469opcodes or operand order. 7470@end defmac 7471 7472@defmac ASM_OUTPUT_REG_PUSH (@var{stream}, @var{regno}) 7473A C expression to output to @var{stream} some assembler code 7474which will push hard register number @var{regno} onto the stack. 7475The code need not be optimal, since this macro is used only when 7476profiling. 7477@end defmac 7478 7479@defmac ASM_OUTPUT_REG_POP (@var{stream}, @var{regno}) 7480A C expression to output to @var{stream} some assembler code 7481which will pop hard register number @var{regno} off of the stack. 7482The code need not be optimal, since this macro is used only when 7483profiling. 7484@end defmac 7485 7486@node Dispatch Tables 7487@subsection Output of Dispatch Tables 7488 7489@c prevent bad page break with this line 7490This concerns dispatch tables. 7491 7492@cindex dispatch table 7493@defmac ASM_OUTPUT_ADDR_DIFF_ELT (@var{stream}, @var{body}, @var{value}, @var{rel}) 7494A C statement to output to the stdio stream @var{stream} an assembler 7495pseudo-instruction to generate a difference between two labels. 7496@var{value} and @var{rel} are the numbers of two internal labels. The 7497definitions of these labels are output using 7498@code{(*targetm.asm_out.internal_label)}, and they must be printed in the same 7499way here. For example, 7500 7501@smallexample 7502fprintf (@var{stream}, "\t.word L%d-L%d\n", 7503 @var{value}, @var{rel}) 7504@end smallexample 7505 7506You must provide this macro on machines where the addresses in a 7507dispatch table are relative to the table's own address. If defined, GCC 7508will also use this macro on all machines when producing PIC@. 7509@var{body} is the body of the @code{ADDR_DIFF_VEC}; it is provided so that the 7510mode and flags can be read. 7511@end defmac 7512 7513@defmac ASM_OUTPUT_ADDR_VEC_ELT (@var{stream}, @var{value}) 7514This macro should be provided on machines where the addresses 7515in a dispatch table are absolute. 7516 7517The definition should be a C statement to output to the stdio stream 7518@var{stream} an assembler pseudo-instruction to generate a reference to 7519a label. @var{value} is the number of an internal label whose 7520definition is output using @code{(*targetm.asm_out.internal_label)}. 7521For example, 7522 7523@smallexample 7524fprintf (@var{stream}, "\t.word L%d\n", @var{value}) 7525@end smallexample 7526@end defmac 7527 7528@defmac ASM_OUTPUT_CASE_LABEL (@var{stream}, @var{prefix}, @var{num}, @var{table}) 7529Define this if the label before a jump-table needs to be output 7530specially. The first three arguments are the same as for 7531@code{(*targetm.asm_out.internal_label)}; the fourth argument is the 7532jump-table which follows (a @code{jump_insn} containing an 7533@code{addr_vec} or @code{addr_diff_vec}). 7534 7535This feature is used on system V to output a @code{swbeg} statement 7536for the table. 7537 7538If this macro is not defined, these labels are output with 7539@code{(*targetm.asm_out.internal_label)}. 7540@end defmac 7541 7542@defmac ASM_OUTPUT_CASE_END (@var{stream}, @var{num}, @var{table}) 7543Define this if something special must be output at the end of a 7544jump-table. The definition should be a C statement to be executed 7545after the assembler code for the table is written. It should write 7546the appropriate code to stdio stream @var{stream}. The argument 7547@var{table} is the jump-table insn, and @var{num} is the label-number 7548of the preceding label. 7549 7550If this macro is not defined, nothing special is output at the end of 7551the jump-table. 7552@end defmac 7553 7554@node Exception Region Output 7555@subsection Assembler Commands for Exception Regions 7556 7557@c prevent bad page break with this line 7558 7559This describes commands marking the start and the end of an exception 7560region. 7561 7562@defmac EH_FRAME_SECTION_NAME 7563If defined, a C string constant for the name of the section containing 7564exception handling frame unwind information. If not defined, GCC will 7565provide a default definition if the target supports named sections. 7566@file{crtstuff.c} uses this macro to switch to the appropriate section. 7567 7568You should define this symbol if your target supports DWARF 2 frame 7569unwind information and the default definition does not work. 7570@end defmac 7571 7572@defmac EH_FRAME_IN_DATA_SECTION 7573If defined, DWARF 2 frame unwind information will be placed in the 7574data section even though the target supports named sections. This 7575might be necessary, for instance, if the system linker does garbage 7576collection and sections cannot be marked as not to be collected. 7577 7578Do not define this macro unless @code{TARGET_ASM_NAMED_SECTION} is 7579also defined. 7580@end defmac 7581 7582@defmac MASK_RETURN_ADDR 7583An rtx used to mask the return address found via @code{RETURN_ADDR_RTX}, so 7584that it does not contain any extraneous set bits in it. 7585@end defmac 7586 7587@defmac DWARF2_UNWIND_INFO 7588Define this macro to 0 if your target supports DWARF 2 frame unwind 7589information, but it does not yet work with exception handling. 7590Otherwise, if your target supports this information (if it defines 7591@samp{INCOMING_RETURN_ADDR_RTX} and either @samp{UNALIGNED_INT_ASM_OP} 7592or @samp{OBJECT_FORMAT_ELF}), GCC will provide a default definition of 75931. 7594 7595If this macro is defined to 1, the DWARF 2 unwinder will be the default 7596exception handling mechanism; otherwise, @code{setjmp}/@code{longjmp} will be used by 7597default. 7598 7599If this macro is defined to anything, the DWARF 2 unwinder will be used 7600instead of inline unwinders and @code{__unwind_function} in the non-@code{setjmp} case. 7601@end defmac 7602 7603@defmac MUST_USE_SJLJ_EXCEPTIONS 7604This macro need only be defined if @code{DWARF2_UNWIND_INFO} is 7605runtime-variable. In that case, @file{except.h} cannot correctly 7606determine the corresponding definition of 7607@code{MUST_USE_SJLJ_EXCEPTIONS}, so the target must provide it directly. 7608@end defmac 7609 7610@defmac DWARF_CIE_DATA_ALIGNMENT 7611This macro need only be defined if the target might save registers in the 7612function prologue at an offset to the stack pointer that is not aligned to 7613@code{UNITS_PER_WORD}. The definition should be the negative minimum 7614alignment if @code{STACK_GROWS_DOWNWARD} is defined, and the positive 7615minimum alignment otherwise. @xref{SDB and DWARF}. Only applicable if 7616the target supports DWARF 2 frame unwind information. 7617@end defmac 7618 7619@deftypefn {Target Hook} void TARGET_ASM_EXCEPTION_SECTION () 7620If defined, a function that switches to the section in which the main 7621exception table is to be placed (@pxref{Sections}). The default is a 7622function that switches to a section named @code{.gcc_except_table} on 7623machines that support named sections via 7624@code{TARGET_ASM_NAMED_SECTION}, otherwise if @option{-fpic} or 7625@option{-fPIC} is in effect, the @code{data_section}, otherwise the 7626@code{readonly_data_section}. 7627@end deftypefn 7628 7629@deftypefn {Target Hook} void TARGET_ASM_EH_FRAME_SECTION () 7630If defined, a function that switches to the section in which the DWARF 2 7631frame unwind information to be placed (@pxref{Sections}). The default 7632is a function that outputs a standard GAS section directive, if 7633@code{EH_FRAME_SECTION_NAME} is defined, or else a data section 7634directive followed by a synthetic label. 7635@end deftypefn 7636 7637@deftypevar {Target Hook} bool TARGET_TERMINATE_DW2_EH_FRAME_INFO 7638Contains the value true if the target should add a zero word onto the 7639end of a Dwarf-2 frame info section when used for exception handling. 7640Default value is false if @code{EH_FRAME_SECTION_NAME} is defined, and 7641true otherwise. 7642@end deftypevar 7643 7644@deftypefn {Target Hook} rtx TARGET_DWARF_REGISTER_SPAN (rtx @var{reg}) 7645Given a register, this hook should return a parallel of registers to 7646represent where to find the register pieces. Define this hook if the 7647register and its mode are represented in Dwarf in non-contiguous 7648locations, or if the register should be represented in more than one 7649register in Dwarf. Otherwise, this hook should return @code{NULL_RTX}. 7650If not defined, the default is to return @code{NULL_RTX}. 7651@end deftypefn 7652 7653@node Alignment Output 7654@subsection Assembler Commands for Alignment 7655 7656@c prevent bad page break with this line 7657This describes commands for alignment. 7658 7659@defmac JUMP_ALIGN (@var{label}) 7660The alignment (log base 2) to put in front of @var{label}, which is 7661a common destination of jumps and has no fallthru incoming edge. 7662 7663This macro need not be defined if you don't want any special alignment 7664to be done at such a time. Most machine descriptions do not currently 7665define the macro. 7666 7667Unless it's necessary to inspect the @var{label} parameter, it is better 7668to set the variable @var{align_jumps} in the target's 7669@code{OVERRIDE_OPTIONS}. Otherwise, you should try to honor the user's 7670selection in @var{align_jumps} in a @code{JUMP_ALIGN} implementation. 7671@end defmac 7672 7673@defmac LABEL_ALIGN_AFTER_BARRIER (@var{label}) 7674The alignment (log base 2) to put in front of @var{label}, which follows 7675a @code{BARRIER}. 7676 7677This macro need not be defined if you don't want any special alignment 7678to be done at such a time. Most machine descriptions do not currently 7679define the macro. 7680@end defmac 7681 7682@defmac LABEL_ALIGN_AFTER_BARRIER_MAX_SKIP 7683The maximum number of bytes to skip when applying 7684@code{LABEL_ALIGN_AFTER_BARRIER}. This works only if 7685@code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined. 7686@end defmac 7687 7688@defmac LOOP_ALIGN (@var{label}) 7689The alignment (log base 2) to put in front of @var{label}, which follows 7690a @code{NOTE_INSN_LOOP_BEG} note. 7691 7692This macro need not be defined if you don't want any special alignment 7693to be done at such a time. Most machine descriptions do not currently 7694define the macro. 7695 7696Unless it's necessary to inspect the @var{label} parameter, it is better 7697to set the variable @code{align_loops} in the target's 7698@code{OVERRIDE_OPTIONS}. Otherwise, you should try to honor the user's 7699selection in @code{align_loops} in a @code{LOOP_ALIGN} implementation. 7700@end defmac 7701 7702@defmac LOOP_ALIGN_MAX_SKIP 7703The maximum number of bytes to skip when applying @code{LOOP_ALIGN}. 7704This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined. 7705@end defmac 7706 7707@defmac LABEL_ALIGN (@var{label}) 7708The alignment (log base 2) to put in front of @var{label}. 7709If @code{LABEL_ALIGN_AFTER_BARRIER} / @code{LOOP_ALIGN} specify a different alignment, 7710the maximum of the specified values is used. 7711 7712Unless it's necessary to inspect the @var{label} parameter, it is better 7713to set the variable @code{align_labels} in the target's 7714@code{OVERRIDE_OPTIONS}. Otherwise, you should try to honor the user's 7715selection in @code{align_labels} in a @code{LABEL_ALIGN} implementation. 7716@end defmac 7717 7718@defmac LABEL_ALIGN_MAX_SKIP 7719The maximum number of bytes to skip when applying @code{LABEL_ALIGN}. 7720This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined. 7721@end defmac 7722 7723@defmac ASM_OUTPUT_SKIP (@var{stream}, @var{nbytes}) 7724A C statement to output to the stdio stream @var{stream} an assembler 7725instruction to advance the location counter by @var{nbytes} bytes. 7726Those bytes should be zero when loaded. @var{nbytes} will be a C 7727expression of type @code{int}. 7728@end defmac 7729 7730@defmac ASM_NO_SKIP_IN_TEXT 7731Define this macro if @code{ASM_OUTPUT_SKIP} should not be used in the 7732text section because it fails to put zeros in the bytes that are skipped. 7733This is true on many Unix systems, where the pseudo--op to skip bytes 7734produces no-op instructions rather than zeros when used in the text 7735section. 7736@end defmac 7737 7738@defmac ASM_OUTPUT_ALIGN (@var{stream}, @var{power}) 7739A C statement to output to the stdio stream @var{stream} an assembler 7740command to advance the location counter to a multiple of 2 to the 7741@var{power} bytes. @var{power} will be a C expression of type @code{int}. 7742@end defmac 7743 7744@defmac ASM_OUTPUT_ALIGN_WITH_NOP (@var{stream}, @var{power}) 7745Like @code{ASM_OUTPUT_ALIGN}, except that the ``nop'' instruction is used 7746for padding, if necessary. 7747@end defmac 7748 7749@defmac ASM_OUTPUT_MAX_SKIP_ALIGN (@var{stream}, @var{power}, @var{max_skip}) 7750A C statement to output to the stdio stream @var{stream} an assembler 7751command to advance the location counter to a multiple of 2 to the 7752@var{power} bytes, but only if @var{max_skip} or fewer bytes are needed to 7753satisfy the alignment request. @var{power} and @var{max_skip} will be 7754a C expression of type @code{int}. 7755@end defmac 7756 7757@need 3000 7758@node Debugging Info 7759@section Controlling Debugging Information Format 7760 7761@c prevent bad page break with this line 7762This describes how to specify debugging information. 7763 7764@menu 7765* All Debuggers:: Macros that affect all debugging formats uniformly. 7766* DBX Options:: Macros enabling specific options in DBX format. 7767* DBX Hooks:: Hook macros for varying DBX format. 7768* File Names and DBX:: Macros controlling output of file names in DBX format. 7769* SDB and DWARF:: Macros for SDB (COFF) and DWARF formats. 7770* VMS Debug:: Macros for VMS debug format. 7771@end menu 7772 7773@node All Debuggers 7774@subsection Macros Affecting All Debugging Formats 7775 7776@c prevent bad page break with this line 7777These macros affect all debugging formats. 7778 7779@defmac DBX_REGISTER_NUMBER (@var{regno}) 7780A C expression that returns the DBX register number for the compiler 7781register number @var{regno}. In the default macro provided, the value 7782of this expression will be @var{regno} itself. But sometimes there are 7783some registers that the compiler knows about and DBX does not, or vice 7784versa. In such cases, some register may need to have one number in the 7785compiler and another for DBX@. 7786 7787If two registers have consecutive numbers inside GCC, and they can be 7788used as a pair to hold a multiword value, then they @emph{must} have 7789consecutive numbers after renumbering with @code{DBX_REGISTER_NUMBER}. 7790Otherwise, debuggers will be unable to access such a pair, because they 7791expect register pairs to be consecutive in their own numbering scheme. 7792 7793If you find yourself defining @code{DBX_REGISTER_NUMBER} in way that 7794does not preserve register pairs, then what you must do instead is 7795redefine the actual register numbering scheme. 7796@end defmac 7797 7798@defmac DEBUGGER_AUTO_OFFSET (@var{x}) 7799A C expression that returns the integer offset value for an automatic 7800variable having address @var{x} (an RTL expression). The default 7801computation assumes that @var{x} is based on the frame-pointer and 7802gives the offset from the frame-pointer. This is required for targets 7803that produce debugging output for DBX or COFF-style debugging output 7804for SDB and allow the frame-pointer to be eliminated when the 7805@option{-g} options is used. 7806@end defmac 7807 7808@defmac DEBUGGER_ARG_OFFSET (@var{offset}, @var{x}) 7809A C expression that returns the integer offset value for an argument 7810having address @var{x} (an RTL expression). The nominal offset is 7811@var{offset}. 7812@end defmac 7813 7814@defmac PREFERRED_DEBUGGING_TYPE 7815A C expression that returns the type of debugging output GCC should 7816produce when the user specifies just @option{-g}. Define 7817this if you have arranged for GCC to support more than one format of 7818debugging output. Currently, the allowable values are @code{DBX_DEBUG}, 7819@code{SDB_DEBUG}, @code{DWARF_DEBUG}, @code{DWARF2_DEBUG}, 7820@code{XCOFF_DEBUG}, @code{VMS_DEBUG}, and @code{VMS_AND_DWARF2_DEBUG}. 7821 7822When the user specifies @option{-ggdb}, GCC normally also uses the 7823value of this macro to select the debugging output format, but with two 7824exceptions. If @code{DWARF2_DEBUGGING_INFO} is defined, GCC uses the 7825value @code{DWARF2_DEBUG}. Otherwise, if @code{DBX_DEBUGGING_INFO} is 7826defined, GCC uses @code{DBX_DEBUG}. 7827 7828The value of this macro only affects the default debugging output; the 7829user can always get a specific type of output by using @option{-gstabs}, 7830@option{-gcoff}, @option{-gdwarf-2}, @option{-gxcoff}, or @option{-gvms}. 7831@end defmac 7832 7833@node DBX Options 7834@subsection Specific Options for DBX Output 7835 7836@c prevent bad page break with this line 7837These are specific options for DBX output. 7838 7839@defmac DBX_DEBUGGING_INFO 7840Define this macro if GCC should produce debugging output for DBX 7841in response to the @option{-g} option. 7842@end defmac 7843 7844@defmac XCOFF_DEBUGGING_INFO 7845Define this macro if GCC should produce XCOFF format debugging output 7846in response to the @option{-g} option. This is a variant of DBX format. 7847@end defmac 7848 7849@defmac DEFAULT_GDB_EXTENSIONS 7850Define this macro to control whether GCC should by default generate 7851GDB's extended version of DBX debugging information (assuming DBX-format 7852debugging information is enabled at all). If you don't define the 7853macro, the default is 1: always generate the extended information 7854if there is any occasion to. 7855@end defmac 7856 7857@defmac DEBUG_SYMS_TEXT 7858Define this macro if all @code{.stabs} commands should be output while 7859in the text section. 7860@end defmac 7861 7862@defmac ASM_STABS_OP 7863A C string constant, including spacing, naming the assembler pseudo op to 7864use instead of @code{"\t.stabs\t"} to define an ordinary debugging symbol. 7865If you don't define this macro, @code{"\t.stabs\t"} is used. This macro 7866applies only to DBX debugging information format. 7867@end defmac 7868 7869@defmac ASM_STABD_OP 7870A C string constant, including spacing, naming the assembler pseudo op to 7871use instead of @code{"\t.stabd\t"} to define a debugging symbol whose 7872value is the current location. If you don't define this macro, 7873@code{"\t.stabd\t"} is used. This macro applies only to DBX debugging 7874information format. 7875@end defmac 7876 7877@defmac ASM_STABN_OP 7878A C string constant, including spacing, naming the assembler pseudo op to 7879use instead of @code{"\t.stabn\t"} to define a debugging symbol with no 7880name. If you don't define this macro, @code{"\t.stabn\t"} is used. This 7881macro applies only to DBX debugging information format. 7882@end defmac 7883 7884@defmac DBX_NO_XREFS 7885Define this macro if DBX on your system does not support the construct 7886@samp{xs@var{tagname}}. On some systems, this construct is used to 7887describe a forward reference to a structure named @var{tagname}. 7888On other systems, this construct is not supported at all. 7889@end defmac 7890 7891@defmac DBX_CONTIN_LENGTH 7892A symbol name in DBX-format debugging information is normally 7893continued (split into two separate @code{.stabs} directives) when it 7894exceeds a certain length (by default, 80 characters). On some 7895operating systems, DBX requires this splitting; on others, splitting 7896must not be done. You can inhibit splitting by defining this macro 7897with the value zero. You can override the default splitting-length by 7898defining this macro as an expression for the length you desire. 7899@end defmac 7900 7901@defmac DBX_CONTIN_CHAR 7902Normally continuation is indicated by adding a @samp{\} character to 7903the end of a @code{.stabs} string when a continuation follows. To use 7904a different character instead, define this macro as a character 7905constant for the character you want to use. Do not define this macro 7906if backslash is correct for your system. 7907@end defmac 7908 7909@defmac DBX_STATIC_STAB_DATA_SECTION 7910Define this macro if it is necessary to go to the data section before 7911outputting the @samp{.stabs} pseudo-op for a non-global static 7912variable. 7913@end defmac 7914 7915@defmac DBX_TYPE_DECL_STABS_CODE 7916The value to use in the ``code'' field of the @code{.stabs} directive 7917for a typedef. The default is @code{N_LSYM}. 7918@end defmac 7919 7920@defmac DBX_STATIC_CONST_VAR_CODE 7921The value to use in the ``code'' field of the @code{.stabs} directive 7922for a static variable located in the text section. DBX format does not 7923provide any ``right'' way to do this. The default is @code{N_FUN}. 7924@end defmac 7925 7926@defmac DBX_REGPARM_STABS_CODE 7927The value to use in the ``code'' field of the @code{.stabs} directive 7928for a parameter passed in registers. DBX format does not provide any 7929``right'' way to do this. The default is @code{N_RSYM}. 7930@end defmac 7931 7932@defmac DBX_REGPARM_STABS_LETTER 7933The letter to use in DBX symbol data to identify a symbol as a parameter 7934passed in registers. DBX format does not customarily provide any way to 7935do this. The default is @code{'P'}. 7936@end defmac 7937 7938@defmac DBX_MEMPARM_STABS_LETTER 7939The letter to use in DBX symbol data to identify a symbol as a stack 7940parameter. The default is @code{'p'}. 7941@end defmac 7942 7943@defmac DBX_FUNCTION_FIRST 7944Define this macro if the DBX information for a function and its 7945arguments should precede the assembler code for the function. Normally, 7946in DBX format, the debugging information entirely follows the assembler 7947code. 7948@end defmac 7949 7950@defmac DBX_BLOCKS_FUNCTION_RELATIVE 7951Define this macro if the value of a symbol describing the scope of a 7952block (@code{N_LBRAC} or @code{N_RBRAC}) should be relative to the start 7953of the enclosing function. Normally, GCC uses an absolute address. 7954@end defmac 7955 7956@defmac DBX_USE_BINCL 7957Define this macro if GCC should generate @code{N_BINCL} and 7958@code{N_EINCL} stabs for included header files, as on Sun systems. This 7959macro also directs GCC to output a type number as a pair of a file 7960number and a type number within the file. Normally, GCC does not 7961generate @code{N_BINCL} or @code{N_EINCL} stabs, and it outputs a single 7962number for a type number. 7963@end defmac 7964 7965@node DBX Hooks 7966@subsection Open-Ended Hooks for DBX Format 7967 7968@c prevent bad page break with this line 7969These are hooks for DBX format. 7970 7971@defmac DBX_OUTPUT_LBRAC (@var{stream}, @var{name}) 7972Define this macro to say how to output to @var{stream} the debugging 7973information for the start of a scope level for variable names. The 7974argument @var{name} is the name of an assembler symbol (for use with 7975@code{assemble_name}) whose value is the address where the scope begins. 7976@end defmac 7977 7978@defmac DBX_OUTPUT_RBRAC (@var{stream}, @var{name}) 7979Like @code{DBX_OUTPUT_LBRAC}, but for the end of a scope level. 7980@end defmac 7981 7982@defmac DBX_OUTPUT_NFUN (@var{stream}, @var{lscope_label}, @var{decl}) 7983Define this macro if the target machine requires special handling to 7984output an @code{N_FUN} entry for the function @var{decl}. 7985@end defmac 7986 7987@defmac DBX_OUTPUT_FUNCTION_END (@var{stream}, @var{function}) 7988Define this macro if the target machine requires special output at the 7989end of the debugging information for a function. The definition should 7990be a C statement (sans semicolon) to output the appropriate information 7991to @var{stream}. @var{function} is the @code{FUNCTION_DECL} node for 7992the function. 7993@end defmac 7994 7995@defmac DBX_OUTPUT_STANDARD_TYPES (@var{syms}) 7996Define this macro if you need to control the order of output of the 7997standard data types at the beginning of compilation. The argument 7998@var{syms} is a @code{tree} which is a chain of all the predefined 7999global symbols, including names of data types. 8000 8001Normally, DBX output starts with definitions of the types for integers 8002and characters, followed by all the other predefined types of the 8003particular language in no particular order. 8004 8005On some machines, it is necessary to output different particular types 8006first. To do this, define @code{DBX_OUTPUT_STANDARD_TYPES} to output 8007those symbols in the necessary order. Any predefined types that you 8008don't explicitly output will be output afterward in no particular order. 8009 8010Be careful not to define this macro so that it works only for C@. There 8011are no global variables to access most of the built-in types, because 8012another language may have another set of types. The way to output a 8013particular type is to look through @var{syms} to see if you can find it. 8014Here is an example: 8015 8016@smallexample 8017@{ 8018 tree decl; 8019 for (decl = syms; decl; decl = TREE_CHAIN (decl)) 8020 if (!strcmp (IDENTIFIER_POINTER (DECL_NAME (decl)), 8021 "long int")) 8022 dbxout_symbol (decl); 8023 @dots{} 8024@} 8025@end smallexample 8026 8027@noindent 8028This does nothing if the expected type does not exist. 8029 8030See the function @code{init_decl_processing} in @file{c-decl.c} to find 8031the names to use for all the built-in C types. 8032 8033Here is another way of finding a particular type: 8034 8035@c this is still overfull. --mew 10feb93 8036@smallexample 8037@{ 8038 tree decl; 8039 for (decl = syms; decl; decl = TREE_CHAIN (decl)) 8040 if (TREE_CODE (decl) == TYPE_DECL 8041 && (TREE_CODE (TREE_TYPE (decl)) 8042 == INTEGER_CST) 8043 && TYPE_PRECISION (TREE_TYPE (decl)) == 16 8044 && TYPE_UNSIGNED (TREE_TYPE (decl))) 8045@group 8046 /* @r{This must be @code{unsigned short}.} */ 8047 dbxout_symbol (decl); 8048 @dots{} 8049@} 8050@end group 8051@end smallexample 8052@end defmac 8053 8054@defmac NO_DBX_FUNCTION_END 8055Some stabs encapsulation formats (in particular ECOFF), cannot handle the 8056@code{.stabs "",N_FUN,,0,0,Lscope-function-1} gdb dbx extension construct. 8057On those machines, define this macro to turn this feature off without 8058disturbing the rest of the gdb extensions. 8059@end defmac 8060 8061@node File Names and DBX 8062@subsection File Names in DBX Format 8063 8064@c prevent bad page break with this line 8065This describes file names in DBX format. 8066 8067@defmac DBX_OUTPUT_MAIN_SOURCE_FILENAME (@var{stream}, @var{name}) 8068A C statement to output DBX debugging information to the stdio stream 8069@var{stream} which indicates that file @var{name} is the main source 8070file---the file specified as the input file for compilation. 8071This macro is called only once, at the beginning of compilation. 8072 8073This macro need not be defined if the standard form of output 8074for DBX debugging information is appropriate. 8075@end defmac 8076 8077@defmac DBX_OUTPUT_MAIN_SOURCE_DIRECTORY (@var{stream}, @var{name}) 8078A C statement to output DBX debugging information to the stdio stream 8079@var{stream} which indicates that the current directory during 8080compilation is named @var{name}. 8081 8082This macro need not be defined if the standard form of output 8083for DBX debugging information is appropriate. 8084@end defmac 8085 8086@defmac DBX_OUTPUT_MAIN_SOURCE_FILE_END (@var{stream}, @var{name}) 8087A C statement to output DBX debugging information at the end of 8088compilation of the main source file @var{name}. 8089 8090If you don't define this macro, nothing special is output at the end 8091of compilation, which is correct for most machines. 8092@end defmac 8093 8094@need 2000 8095@node SDB and DWARF 8096@subsection Macros for SDB and DWARF Output 8097 8098@c prevent bad page break with this line 8099Here are macros for SDB and DWARF output. 8100 8101@defmac SDB_DEBUGGING_INFO 8102Define this macro if GCC should produce COFF-style debugging output 8103for SDB in response to the @option{-g} option. 8104@end defmac 8105 8106@defmac DWARF2_DEBUGGING_INFO 8107Define this macro if GCC should produce dwarf version 2 format 8108debugging output in response to the @option{-g} option. 8109 8110To support optional call frame debugging information, you must also 8111define @code{INCOMING_RETURN_ADDR_RTX} and either set 8112@code{RTX_FRAME_RELATED_P} on the prologue insns if you use RTL for the 8113prologue, or call @code{dwarf2out_def_cfa} and @code{dwarf2out_reg_save} 8114as appropriate from @code{TARGET_ASM_FUNCTION_PROLOGUE} if you don't. 8115@end defmac 8116 8117@defmac DWARF2_FRAME_INFO 8118Define this macro to a nonzero value if GCC should always output 8119Dwarf 2 frame information. If @code{DWARF2_UNWIND_INFO} 8120(@pxref{Exception Region Output} is nonzero, GCC will output this 8121information not matter how you define @code{DWARF2_FRAME_INFO}. 8122@end defmac 8123 8124@defmac DWARF2_GENERATE_TEXT_SECTION_LABEL 8125By default, the Dwarf 2 debugging information generator will generate a 8126label to mark the beginning of the text section. If it is better simply 8127to use the name of the text section itself, rather than an explicit label, 8128to indicate the beginning of the text section, define this macro to zero. 8129@end defmac 8130 8131@defmac DWARF2_ASM_LINE_DEBUG_INFO 8132Define this macro to be a nonzero value if the assembler can generate Dwarf 2 8133line debug info sections. This will result in much more compact line number 8134tables, and hence is desirable if it works. 8135@end defmac 8136 8137@defmac ASM_OUTPUT_DWARF_DELTA (@var{stream}, @var{size}, @var{label1}, @var{label2}) 8138A C statement to issue assembly directives that create a difference 8139between the two given labels, using an integer of the given size. 8140@end defmac 8141 8142@defmac ASM_OUTPUT_DWARF_OFFSET (@var{stream}, @var{size}, @var{label}) 8143A C statement to issue assembly directives that create a 8144section-relative reference to the given label, using an integer of the 8145given size. 8146@end defmac 8147 8148@defmac ASM_OUTPUT_DWARF_PCREL (@var{stream}, @var{size}, @var{label}) 8149A C statement to issue assembly directives that create a self-relative 8150reference to the given label, using an integer of the given size. 8151@end defmac 8152 8153@defmac PUT_SDB_@dots{} 8154Define these macros to override the assembler syntax for the special 8155SDB assembler directives. See @file{sdbout.c} for a list of these 8156macros and their arguments. If the standard syntax is used, you need 8157not define them yourself. 8158@end defmac 8159 8160@defmac SDB_DELIM 8161Some assemblers do not support a semicolon as a delimiter, even between 8162SDB assembler directives. In that case, define this macro to be the 8163delimiter to use (usually @samp{\n}). It is not necessary to define 8164a new set of @code{PUT_SDB_@var{op}} macros if this is the only change 8165required. 8166@end defmac 8167 8168@defmac SDB_GENERATE_FAKE 8169Define this macro to override the usual method of constructing a dummy 8170name for anonymous structure and union types. See @file{sdbout.c} for 8171more information. 8172@end defmac 8173 8174@defmac SDB_ALLOW_UNKNOWN_REFERENCES 8175Define this macro to allow references to unknown structure, 8176union, or enumeration tags to be emitted. Standard COFF does not 8177allow handling of unknown references, MIPS ECOFF has support for 8178it. 8179@end defmac 8180 8181@defmac SDB_ALLOW_FORWARD_REFERENCES 8182Define this macro to allow references to structure, union, or 8183enumeration tags that have not yet been seen to be handled. Some 8184assemblers choke if forward tags are used, while some require it. 8185@end defmac 8186 8187@need 2000 8188@node VMS Debug 8189@subsection Macros for VMS Debug Format 8190 8191@c prevent bad page break with this line 8192Here are macros for VMS debug format. 8193 8194@defmac VMS_DEBUGGING_INFO 8195Define this macro if GCC should produce debugging output for VMS 8196in response to the @option{-g} option. The default behavior for VMS 8197is to generate minimal debug info for a traceback in the absence of 8198@option{-g} unless explicitly overridden with @option{-g0}. This 8199behavior is controlled by @code{OPTIMIZATION_OPTIONS} and 8200@code{OVERRIDE_OPTIONS}. 8201@end defmac 8202 8203@node Floating Point 8204@section Cross Compilation and Floating Point 8205@cindex cross compilation and floating point 8206@cindex floating point and cross compilation 8207 8208While all modern machines use twos-complement representation for integers, 8209there are a variety of representations for floating point numbers. This 8210means that in a cross-compiler the representation of floating point numbers 8211in the compiled program may be different from that used in the machine 8212doing the compilation. 8213 8214Because different representation systems may offer different amounts of 8215range and precision, all floating point constants must be represented in 8216the target machine's format. Therefore, the cross compiler cannot 8217safely use the host machine's floating point arithmetic; it must emulate 8218the target's arithmetic. To ensure consistency, GCC always uses 8219emulation to work with floating point values, even when the host and 8220target floating point formats are identical. 8221 8222The following macros are provided by @file{real.h} for the compiler to 8223use. All parts of the compiler which generate or optimize 8224floating-point calculations must use these macros. They may evaluate 8225their operands more than once, so operands must not have side effects. 8226 8227@defmac REAL_VALUE_TYPE 8228The C data type to be used to hold a floating point value in the target 8229machine's format. Typically this is a @code{struct} containing an 8230array of @code{HOST_WIDE_INT}, but all code should treat it as an opaque 8231quantity. 8232@end defmac 8233 8234@deftypefn Macro int REAL_VALUES_EQUAL (REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y}) 8235Compares for equality the two values, @var{x} and @var{y}. If the target 8236floating point format supports negative zeroes and/or NaNs, 8237@samp{REAL_VALUES_EQUAL (-0.0, 0.0)} is true, and 8238@samp{REAL_VALUES_EQUAL (NaN, NaN)} is false. 8239@end deftypefn 8240 8241@deftypefn Macro int REAL_VALUES_LESS (REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y}) 8242Tests whether @var{x} is less than @var{y}. 8243@end deftypefn 8244 8245@deftypefn Macro HOST_WIDE_INT REAL_VALUE_FIX (REAL_VALUE_TYPE @var{x}) 8246Truncates @var{x} to a signed integer, rounding toward zero. 8247@end deftypefn 8248 8249@deftypefn Macro {unsigned HOST_WIDE_INT} REAL_VALUE_UNSIGNED_FIX (REAL_VALUE_TYPE @var{x}) 8250Truncates @var{x} to an unsigned integer, rounding toward zero. If 8251@var{x} is negative, returns zero. 8252@end deftypefn 8253 8254@deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ATOF (const char *@var{string}, enum machine_mode @var{mode}) 8255Converts @var{string} into a floating point number in the target machine's 8256representation for mode @var{mode}. This routine can handle both 8257decimal and hexadecimal floating point constants, using the syntax 8258defined by the C language for both. 8259@end deftypefn 8260 8261@deftypefn Macro int REAL_VALUE_NEGATIVE (REAL_VALUE_TYPE @var{x}) 8262Returns 1 if @var{x} is negative (including negative zero), 0 otherwise. 8263@end deftypefn 8264 8265@deftypefn Macro int REAL_VALUE_ISINF (REAL_VALUE_TYPE @var{x}) 8266Determines whether @var{x} represents infinity (positive or negative). 8267@end deftypefn 8268 8269@deftypefn Macro int REAL_VALUE_ISNAN (REAL_VALUE_TYPE @var{x}) 8270Determines whether @var{x} represents a ``NaN'' (not-a-number). 8271@end deftypefn 8272 8273@deftypefn Macro void REAL_ARITHMETIC (REAL_VALUE_TYPE @var{output}, enum tree_code @var{code}, REAL_VALUE_TYPE @var{x}, REAL_VALUE_TYPE @var{y}) 8274Calculates an arithmetic operation on the two floating point values 8275@var{x} and @var{y}, storing the result in @var{output} (which must be a 8276variable). 8277 8278The operation to be performed is specified by @var{code}. Only the 8279following codes are supported: @code{PLUS_EXPR}, @code{MINUS_EXPR}, 8280@code{MULT_EXPR}, @code{RDIV_EXPR}, @code{MAX_EXPR}, @code{MIN_EXPR}. 8281 8282If @code{REAL_ARITHMETIC} is asked to evaluate division by zero and the 8283target's floating point format cannot represent infinity, it will call 8284@code{abort}. Callers should check for this situation first, using 8285@code{MODE_HAS_INFINITIES}. @xref{Storage Layout}. 8286@end deftypefn 8287 8288@deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_NEGATE (REAL_VALUE_TYPE @var{x}) 8289Returns the negative of the floating point value @var{x}. 8290@end deftypefn 8291 8292@deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ABS (REAL_VALUE_TYPE @var{x}) 8293Returns the absolute value of @var{x}. 8294@end deftypefn 8295 8296@deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_TRUNCATE (REAL_VALUE_TYPE @var{mode}, enum machine_mode @var{x}) 8297Truncates the floating point value @var{x} to fit in @var{mode}. The 8298return value is still a full-size @code{REAL_VALUE_TYPE}, but it has an 8299appropriate bit pattern to be output asa floating constant whose 8300precision accords with mode @var{mode}. 8301@end deftypefn 8302 8303@deftypefn Macro void REAL_VALUE_TO_INT (HOST_WIDE_INT @var{low}, HOST_WIDE_INT @var{high}, REAL_VALUE_TYPE @var{x}) 8304Converts a floating point value @var{x} into a double-precision integer 8305which is then stored into @var{low} and @var{high}. If the value is not 8306integral, it is truncated. 8307@end deftypefn 8308 8309@deftypefn Macro void REAL_VALUE_FROM_INT (REAL_VALUE_TYPE @var{x}, HOST_WIDE_INT @var{low}, HOST_WIDE_INT @var{high}, enum machine_mode @var{mode}) 8310Converts a double-precision integer found in @var{low} and @var{high}, 8311into a floating point value which is then stored into @var{x}. The 8312value is truncated to fit in mode @var{mode}. 8313@end deftypefn 8314 8315@node Mode Switching 8316@section Mode Switching Instructions 8317@cindex mode switching 8318The following macros control mode switching optimizations: 8319 8320@defmac OPTIMIZE_MODE_SWITCHING (@var{entity}) 8321Define this macro if the port needs extra instructions inserted for mode 8322switching in an optimizing compilation. 8323 8324For an example, the SH4 can perform both single and double precision 8325floating point operations, but to perform a single precision operation, 8326the FPSCR PR bit has to be cleared, while for a double precision 8327operation, this bit has to be set. Changing the PR bit requires a general 8328purpose register as a scratch register, hence these FPSCR sets have to 8329be inserted before reload, i.e.@: you can't put this into instruction emitting 8330or @code{TARGET_MACHINE_DEPENDENT_REORG}. 8331 8332You can have multiple entities that are mode-switched, and select at run time 8333which entities actually need it. @code{OPTIMIZE_MODE_SWITCHING} should 8334return nonzero for any @var{entity} that needs mode-switching. 8335If you define this macro, you also have to define 8336@code{NUM_MODES_FOR_MODE_SWITCHING}, @code{MODE_NEEDED}, 8337@code{MODE_PRIORITY_TO_MODE} and @code{EMIT_MODE_SET}. 8338@code{MODE_AFTER}, @code{MODE_ENTRY}, and @code{MODE_EXIT} 8339are optional. 8340@end defmac 8341 8342@defmac NUM_MODES_FOR_MODE_SWITCHING 8343If you define @code{OPTIMIZE_MODE_SWITCHING}, you have to define this as 8344initializer for an array of integers. Each initializer element 8345N refers to an entity that needs mode switching, and specifies the number 8346of different modes that might need to be set for this entity. 8347The position of the initializer in the initializer - starting counting at 8348zero - determines the integer that is used to refer to the mode-switched 8349entity in question. 8350In macros that take mode arguments / yield a mode result, modes are 8351represented as numbers 0 @dots{} N @minus{} 1. N is used to specify that no mode 8352switch is needed / supplied. 8353@end defmac 8354 8355@defmac MODE_NEEDED (@var{entity}, @var{insn}) 8356@var{entity} is an integer specifying a mode-switched entity. If 8357@code{OPTIMIZE_MODE_SWITCHING} is defined, you must define this macro to 8358return an integer value not larger than the corresponding element in 8359@code{NUM_MODES_FOR_MODE_SWITCHING}, to denote the mode that @var{entity} must 8360be switched into prior to the execution of @var{insn}. 8361@end defmac 8362 8363@defmac MODE_AFTER (@var{mode}, @var{insn}) 8364If this macro is defined, it is evaluated for every @var{insn} during 8365mode switching. It determines the mode that an insn results in (if 8366different from the incoming mode). 8367@end defmac 8368 8369@defmac MODE_ENTRY (@var{entity}) 8370If this macro is defined, it is evaluated for every @var{entity} that needs 8371mode switching. It should evaluate to an integer, which is a mode that 8372@var{entity} is assumed to be switched to at function entry. If @code{MODE_ENTRY} 8373is defined then @code{MODE_EXIT} must be defined. 8374@end defmac 8375 8376@defmac MODE_EXIT (@var{entity}) 8377If this macro is defined, it is evaluated for every @var{entity} that needs 8378mode switching. It should evaluate to an integer, which is a mode that 8379@var{entity} is assumed to be switched to at function exit. If @code{MODE_EXIT} 8380is defined then @code{MODE_ENTRY} must be defined. 8381@end defmac 8382 8383@defmac MODE_PRIORITY_TO_MODE (@var{entity}, @var{n}) 8384This macro specifies the order in which modes for @var{entity} are processed. 83850 is the highest priority, @code{NUM_MODES_FOR_MODE_SWITCHING[@var{entity}] - 1} the 8386lowest. The value of the macro should be an integer designating a mode 8387for @var{entity}. For any fixed @var{entity}, @code{mode_priority_to_mode} 8388(@var{entity}, @var{n}) shall be a bijection in 0 @dots{} 8389@code{num_modes_for_mode_switching[@var{entity}] - 1}. 8390@end defmac 8391 8392@defmac EMIT_MODE_SET (@var{entity}, @var{mode}, @var{hard_regs_live}) 8393Generate one or more insns to set @var{entity} to @var{mode}. 8394@var{hard_reg_live} is the set of hard registers live at the point where 8395the insn(s) are to be inserted. 8396@end defmac 8397 8398@node Target Attributes 8399@section Defining target-specific uses of @code{__attribute__} 8400@cindex target attributes 8401@cindex machine attributes 8402@cindex attributes, target-specific 8403 8404Target-specific attributes may be defined for functions, data and types. 8405These are described using the following target hooks; they also need to 8406be documented in @file{extend.texi}. 8407 8408@deftypevr {Target Hook} {const struct attribute_spec *} TARGET_ATTRIBUTE_TABLE 8409If defined, this target hook points to an array of @samp{struct 8410attribute_spec} (defined in @file{tree.h}) specifying the machine 8411specific attributes for this target and some of the restrictions on the 8412entities to which these attributes are applied and the arguments they 8413take. 8414@end deftypevr 8415 8416@deftypefn {Target Hook} int TARGET_COMP_TYPE_ATTRIBUTES (tree @var{type1}, tree @var{type2}) 8417If defined, this target hook is a function which returns zero if the attributes on 8418@var{type1} and @var{type2} are incompatible, one if they are compatible, 8419and two if they are nearly compatible (which causes a warning to be 8420generated). If this is not defined, machine-specific attributes are 8421supposed always to be compatible. 8422@end deftypefn 8423 8424@deftypefn {Target Hook} void TARGET_SET_DEFAULT_TYPE_ATTRIBUTES (tree @var{type}) 8425If defined, this target hook is a function which assigns default attributes to 8426newly defined @var{type}. 8427@end deftypefn 8428 8429@deftypefn {Target Hook} tree TARGET_MERGE_TYPE_ATTRIBUTES (tree @var{type1}, tree @var{type2}) 8430Define this target hook if the merging of type attributes needs special 8431handling. If defined, the result is a list of the combined 8432@code{TYPE_ATTRIBUTES} of @var{type1} and @var{type2}. It is assumed 8433that @code{comptypes} has already been called and returned 1. This 8434function may call @code{merge_attributes} to handle machine-independent 8435merging. 8436@end deftypefn 8437 8438@deftypefn {Target Hook} tree TARGET_MERGE_DECL_ATTRIBUTES (tree @var{olddecl}, tree @var{newdecl}) 8439Define this target hook if the merging of decl attributes needs special 8440handling. If defined, the result is a list of the combined 8441@code{DECL_ATTRIBUTES} of @var{olddecl} and @var{newdecl}. 8442@var{newdecl} is a duplicate declaration of @var{olddecl}. Examples of 8443when this is needed are when one attribute overrides another, or when an 8444attribute is nullified by a subsequent definition. This function may 8445call @code{merge_attributes} to handle machine-independent merging. 8446 8447@findex TARGET_DLLIMPORT_DECL_ATTRIBUTES 8448If the only target-specific handling you require is @samp{dllimport} for 8449Microsoft Windows targets, you should define the macro 8450@code{TARGET_DLLIMPORT_DECL_ATTRIBUTES}. This links in a function 8451called @code{merge_dllimport_decl_attributes} which can then be defined 8452as the expansion of @code{TARGET_MERGE_DECL_ATTRIBUTES}. This is done 8453in @file{i386/cygwin.h} and @file{i386/i386.c}, for example. 8454@end deftypefn 8455 8456@deftypefn {Target Hook} void TARGET_INSERT_ATTRIBUTES (tree @var{node}, tree *@var{attr_ptr}) 8457Define this target hook if you want to be able to add attributes to a decl 8458when it is being created. This is normally useful for back ends which 8459wish to implement a pragma by using the attributes which correspond to 8460the pragma's effect. The @var{node} argument is the decl which is being 8461created. The @var{attr_ptr} argument is a pointer to the attribute list 8462for this decl. The list itself should not be modified, since it may be 8463shared with other decls, but attributes may be chained on the head of 8464the list and @code{*@var{attr_ptr}} modified to point to the new 8465attributes, or a copy of the list may be made if further changes are 8466needed. 8467@end deftypefn 8468 8469@deftypefn {Target Hook} bool TARGET_FUNCTION_ATTRIBUTE_INLINABLE_P (tree @var{fndecl}) 8470@cindex inlining 8471This target hook returns @code{true} if it is ok to inline @var{fndecl} 8472into the current function, despite its having target-specific 8473attributes, @code{false} otherwise. By default, if a function has a 8474target specific attribute attached to it, it will not be inlined. 8475@end deftypefn 8476 8477@node MIPS Coprocessors 8478@section Defining coprocessor specifics for MIPS targets. 8479@cindex MIPS coprocessor-definition macros 8480 8481The MIPS specification allows MIPS implementations to have as many as 4 8482coprocessors, each with as many as 32 private registers. GCC supports 8483accessing these registers and transferring values between the registers 8484and memory using asm-ized variables. For example: 8485 8486@smallexample 8487 register unsigned int cp0count asm ("c0r1"); 8488 unsigned int d; 8489 8490 d = cp0count + 3; 8491@end smallexample 8492 8493(``c0r1'' is the default name of register 1 in coprocessor 0; alternate 8494names may be added as described below, or the default names may be 8495overridden entirely in @code{SUBTARGET_CONDITIONAL_REGISTER_USAGE}.) 8496 8497Coprocessor registers are assumed to be epilogue-used; sets to them will 8498be preserved even if it does not appear that the register is used again 8499later in the function. 8500 8501Another note: according to the MIPS spec, coprocessor 1 (if present) is 8502the FPU. One accesses COP1 registers through standard mips 8503floating-point support; they are not included in this mechanism. 8504 8505There is one macro used in defining the MIPS coprocessor interface which 8506you may want to override in subtargets; it is described below. 8507 8508@defmac ALL_COP_ADDITIONAL_REGISTER_NAMES 8509A comma-separated list (with leading comma) of pairs describing the 8510alternate names of coprocessor registers. The format of each entry should be 8511@smallexample 8512@{ @var{alternatename}, @var{register_number}@} 8513@end smallexample 8514Default: empty. 8515@end defmac 8516 8517@node PCH Target 8518@section Parameters for Precompiled Header Validity Checking 8519@cindex parameters, precompiled headers 8520 8521@deftypefn {Target Hook} void * TARGET_GET_PCH_VALIDITY (size_t * @var{sz}) 8522Define this hook if your target needs to check a different collection 8523of flags than the default, which is every flag defined by 8524@code{TARGET_SWITCHES} and @code{TARGET_OPTIONS}. It should return 8525some data which will be saved in the PCH file and presented to 8526@code{TARGET_PCH_VALID_P} later; it should set @code{SZ} to the size 8527of the data. 8528@end deftypefn 8529 8530@deftypefn {Target Hook} const char * TARGET_PCH_VALID_P (const void * @var{data}, size_t @var{sz}) 8531Define this hook if your target needs to check a different collection of 8532flags than the default, which is every flag defined by @code{TARGET_SWITCHES} 8533and @code{TARGET_OPTIONS}. It is given data which came from 8534@code{TARGET_GET_PCH_VALIDITY} (in this version of this compiler, so there 8535is no need for extensive validity checking). It returns @code{NULL} if 8536it is safe to load a PCH file with this data, or a suitable error message 8537if not. The error message will be presented to the user, so it should 8538be localized. 8539@end deftypefn 8540 8541@node Misc 8542@section Miscellaneous Parameters 8543@cindex parameters, miscellaneous 8544 8545@c prevent bad page break with this line 8546Here are several miscellaneous parameters. 8547 8548@defmac PREDICATE_CODES 8549Define this if you have defined special-purpose predicates in the file 8550@file{@var{machine}.c}. This macro is called within an initializer of an 8551array of structures. The first field in the structure is the name of a 8552predicate and the second field is an array of rtl codes. For each 8553predicate, list all rtl codes that can be in expressions matched by the 8554predicate. The list should have a trailing comma. Here is an example 8555of two entries in the list for a typical RISC machine: 8556 8557@smallexample 8558#define PREDICATE_CODES \ 8559 @{"gen_reg_rtx_operand", @{SUBREG, REG@}@}, \ 8560 @{"reg_or_short_cint_operand", @{SUBREG, REG, CONST_INT@}@}, 8561@end smallexample 8562 8563Defining this macro does not affect the generated code (however, 8564incorrect definitions that omit an rtl code that may be matched by the 8565predicate can cause the compiler to malfunction). Instead, it allows 8566the table built by @file{genrecog} to be more compact and efficient, 8567thus speeding up the compiler. The most important predicates to include 8568in the list specified by this macro are those used in the most insn 8569patterns. 8570 8571For each predicate function named in @code{PREDICATE_CODES}, a 8572declaration will be generated in @file{insn-codes.h}. 8573@end defmac 8574 8575@defmac SPECIAL_MODE_PREDICATES 8576Define this if you have special predicates that know special things 8577about modes. Genrecog will warn about certain forms of 8578@code{match_operand} without a mode; if the operand predicate is 8579listed in @code{SPECIAL_MODE_PREDICATES}, the warning will be 8580suppressed. 8581 8582Here is an example from the IA-32 port (@code{ext_register_operand} 8583specially checks for @code{HImode} or @code{SImode} in preparation 8584for a byte extraction from @code{%ah} etc.). 8585 8586@smallexample 8587#define SPECIAL_MODE_PREDICATES \ 8588 "ext_register_operand", 8589@end smallexample 8590@end defmac 8591 8592@defmac CASE_VECTOR_MODE 8593An alias for a machine mode name. This is the machine mode that 8594elements of a jump-table should have. 8595@end defmac 8596 8597@defmac CASE_VECTOR_SHORTEN_MODE (@var{min_offset}, @var{max_offset}, @var{body}) 8598Optional: return the preferred mode for an @code{addr_diff_vec} 8599when the minimum and maximum offset are known. If you define this, 8600it enables extra code in branch shortening to deal with @code{addr_diff_vec}. 8601To make this work, you also have to define @code{INSN_ALIGN} and 8602make the alignment for @code{addr_diff_vec} explicit. 8603The @var{body} argument is provided so that the offset_unsigned and scale 8604flags can be updated. 8605@end defmac 8606 8607@defmac CASE_VECTOR_PC_RELATIVE 8608Define this macro to be a C expression to indicate when jump-tables 8609should contain relative addresses. You need not define this macro if 8610jump-tables never contain relative addresses, or jump-tables should 8611contain relative addresses only when @option{-fPIC} or @option{-fPIC} 8612is in effect. 8613@end defmac 8614 8615@defmac CASE_DROPS_THROUGH 8616Define this if control falls through a @code{case} insn when the index 8617value is out of range. This means the specified default-label is 8618actually ignored by the @code{case} insn proper. 8619@end defmac 8620 8621@defmac CASE_VALUES_THRESHOLD 8622Define this to be the smallest number of different values for which it 8623is best to use a jump-table instead of a tree of conditional branches. 8624The default is four for machines with a @code{casesi} instruction and 8625five otherwise. This is best for most machines. 8626@end defmac 8627 8628@defmac CASE_USE_BIT_TESTS 8629Define this macro to be a C expression to indicate whether C switch 8630statements may be implemented by a sequence of bit tests. This is 8631advantageous on processors that can efficiently implement left shift 8632of 1 by the number of bits held in a register, but inappropriate on 8633targets that would require a loop. By default, this macro returns 8634@code{true} if the target defines an @code{ashlsi3} pattern, and 8635@code{false} otherwise. 8636@end defmac 8637 8638@defmac WORD_REGISTER_OPERATIONS 8639Define this macro if operations between registers with integral mode 8640smaller than a word are always performed on the entire register. 8641Most RISC machines have this property and most CISC machines do not. 8642@end defmac 8643 8644@defmac LOAD_EXTEND_OP (@var{mem_mode}) 8645Define this macro to be a C expression indicating when insns that read 8646memory in @var{mem_mode}, an integral mode narrower than a word, set the 8647bits outside of @var{mem_mode} to be either the sign-extension or the 8648zero-extension of the data read. Return @code{SIGN_EXTEND} for values 8649of @var{mem_mode} for which the 8650insn sign-extends, @code{ZERO_EXTEND} for which it zero-extends, and 8651@code{NIL} for other modes. 8652 8653This macro is not called with @var{mem_mode} non-integral or with a width 8654greater than or equal to @code{BITS_PER_WORD}, so you may return any 8655value in this case. Do not define this macro if it would always return 8656@code{NIL}. On machines where this macro is defined, you will normally 8657define it as the constant @code{SIGN_EXTEND} or @code{ZERO_EXTEND}. 8658 8659You may return a non-@code{NIL} value even if for some hard registers 8660the sign extension is not performed, if for the @code{REGNO_REG_CLASS} 8661of these hard registers @code{CANNOT_CHANGE_MODE_CLASS} returns nonzero 8662when the @var{from} mode is @var{mem_mode} and the @var{to} mode is any 8663integral mode larger than this but not larger than @code{word_mode}. 8664 8665You must return @code{NIL} if for some hard registers that allow this 8666mode, @code{CANNOT_CHANGE_MODE_CLASS} says that they cannot change to 8667@code{word_mode}, but that they can change to another integral mode that 8668is larger then @var{mem_mode} but still smaller than @code{word_mode}. 8669@end defmac 8670 8671@defmac SHORT_IMMEDIATES_SIGN_EXTEND 8672Define this macro if loading short immediate values into registers sign 8673extends. 8674@end defmac 8675 8676@defmac FIXUNS_TRUNC_LIKE_FIX_TRUNC 8677Define this macro if the same instructions that convert a floating 8678point number to a signed fixed point number also convert validly to an 8679unsigned one. 8680@end defmac 8681 8682@defmac MOVE_MAX 8683The maximum number of bytes that a single instruction can move quickly 8684between memory and registers or between two memory locations. 8685@end defmac 8686 8687@defmac MAX_MOVE_MAX 8688The maximum number of bytes that a single instruction can move quickly 8689between memory and registers or between two memory locations. If this 8690is undefined, the default is @code{MOVE_MAX}. Otherwise, it is the 8691constant value that is the largest value that @code{MOVE_MAX} can have 8692at run-time. 8693@end defmac 8694 8695@defmac SHIFT_COUNT_TRUNCATED 8696A C expression that is nonzero if on this machine the number of bits 8697actually used for the count of a shift operation is equal to the number 8698of bits needed to represent the size of the object being shifted. When 8699this macro is nonzero, the compiler will assume that it is safe to omit 8700a sign-extend, zero-extend, and certain bitwise `and' instructions that 8701truncates the count of a shift operation. On machines that have 8702instructions that act on bit-fields at variable positions, which may 8703include `bit test' instructions, a nonzero @code{SHIFT_COUNT_TRUNCATED} 8704also enables deletion of truncations of the values that serve as 8705arguments to bit-field instructions. 8706 8707If both types of instructions truncate the count (for shifts) and 8708position (for bit-field operations), or if no variable-position bit-field 8709instructions exist, you should define this macro. 8710 8711However, on some machines, such as the 80386 and the 680x0, truncation 8712only applies to shift operations and not the (real or pretended) 8713bit-field operations. Define @code{SHIFT_COUNT_TRUNCATED} to be zero on 8714such machines. Instead, add patterns to the @file{md} file that include 8715the implied truncation of the shift instructions. 8716 8717You need not define this macro if it would always have the value of zero. 8718@end defmac 8719 8720@defmac TRULY_NOOP_TRUNCATION (@var{outprec}, @var{inprec}) 8721A C expression which is nonzero if on this machine it is safe to 8722``convert'' an integer of @var{inprec} bits to one of @var{outprec} 8723bits (where @var{outprec} is smaller than @var{inprec}) by merely 8724operating on it as if it had only @var{outprec} bits. 8725 8726On many machines, this expression can be 1. 8727 8728@c rearranged this, removed the phrase "it is reported that". this was 8729@c to fix an overfull hbox. --mew 10feb93 8730When @code{TRULY_NOOP_TRUNCATION} returns 1 for a pair of sizes for 8731modes for which @code{MODES_TIEABLE_P} is 0, suboptimal code can result. 8732If this is the case, making @code{TRULY_NOOP_TRUNCATION} return 0 in 8733such cases may improve things. 8734@end defmac 8735 8736@defmac STORE_FLAG_VALUE 8737A C expression describing the value returned by a comparison operator 8738with an integral mode and stored by a store-flag instruction 8739(@samp{s@var{cond}}) when the condition is true. This description must 8740apply to @emph{all} the @samp{s@var{cond}} patterns and all the 8741comparison operators whose results have a @code{MODE_INT} mode. 8742 8743A value of 1 or @minus{}1 means that the instruction implementing the 8744comparison operator returns exactly 1 or @minus{}1 when the comparison is true 8745and 0 when the comparison is false. Otherwise, the value indicates 8746which bits of the result are guaranteed to be 1 when the comparison is 8747true. This value is interpreted in the mode of the comparison 8748operation, which is given by the mode of the first operand in the 8749@samp{s@var{cond}} pattern. Either the low bit or the sign bit of 8750@code{STORE_FLAG_VALUE} be on. Presently, only those bits are used by 8751the compiler. 8752 8753If @code{STORE_FLAG_VALUE} is neither 1 or @minus{}1, the compiler will 8754generate code that depends only on the specified bits. It can also 8755replace comparison operators with equivalent operations if they cause 8756the required bits to be set, even if the remaining bits are undefined. 8757For example, on a machine whose comparison operators return an 8758@code{SImode} value and where @code{STORE_FLAG_VALUE} is defined as 8759@samp{0x80000000}, saying that just the sign bit is relevant, the 8760expression 8761 8762@smallexample 8763(ne:SI (and:SI @var{x} (const_int @var{power-of-2})) (const_int 0)) 8764@end smallexample 8765 8766@noindent 8767can be converted to 8768 8769@smallexample 8770(ashift:SI @var{x} (const_int @var{n})) 8771@end smallexample 8772 8773@noindent 8774where @var{n} is the appropriate shift count to move the bit being 8775tested into the sign bit. 8776 8777There is no way to describe a machine that always sets the low-order bit 8778for a true value, but does not guarantee the value of any other bits, 8779but we do not know of any machine that has such an instruction. If you 8780are trying to port GCC to such a machine, include an instruction to 8781perform a logical-and of the result with 1 in the pattern for the 8782comparison operators and let us know at @email{gcc@@gcc.gnu.org}. 8783 8784Often, a machine will have multiple instructions that obtain a value 8785from a comparison (or the condition codes). Here are rules to guide the 8786choice of value for @code{STORE_FLAG_VALUE}, and hence the instructions 8787to be used: 8788 8789@itemize @bullet 8790@item 8791Use the shortest sequence that yields a valid definition for 8792@code{STORE_FLAG_VALUE}. It is more efficient for the compiler to 8793``normalize'' the value (convert it to, e.g., 1 or 0) than for the 8794comparison operators to do so because there may be opportunities to 8795combine the normalization with other operations. 8796 8797@item 8798For equal-length sequences, use a value of 1 or @minus{}1, with @minus{}1 being 8799slightly preferred on machines with expensive jumps and 1 preferred on 8800other machines. 8801 8802@item 8803As a second choice, choose a value of @samp{0x80000001} if instructions 8804exist that set both the sign and low-order bits but do not define the 8805others. 8806 8807@item 8808Otherwise, use a value of @samp{0x80000000}. 8809@end itemize 8810 8811Many machines can produce both the value chosen for 8812@code{STORE_FLAG_VALUE} and its negation in the same number of 8813instructions. On those machines, you should also define a pattern for 8814those cases, e.g., one matching 8815 8816@smallexample 8817(set @var{A} (neg:@var{m} (ne:@var{m} @var{B} @var{C}))) 8818@end smallexample 8819 8820Some machines can also perform @code{and} or @code{plus} operations on 8821condition code values with less instructions than the corresponding 8822@samp{s@var{cond}} insn followed by @code{and} or @code{plus}. On those 8823machines, define the appropriate patterns. Use the names @code{incscc} 8824and @code{decscc}, respectively, for the patterns which perform 8825@code{plus} or @code{minus} operations on condition code values. See 8826@file{rs6000.md} for some examples. The GNU Superoptizer can be used to 8827find such instruction sequences on other machines. 8828 8829If this macro is not defined, the default value, 1, is used. You need 8830not define @code{STORE_FLAG_VALUE} if the machine has no store-flag 8831instructions, or if the value generated by these instructions is 1. 8832@end defmac 8833 8834@defmac FLOAT_STORE_FLAG_VALUE (@var{mode}) 8835A C expression that gives a nonzero @code{REAL_VALUE_TYPE} value that is 8836returned when comparison operators with floating-point results are true. 8837Define this macro on machine that have comparison operations that return 8838floating-point values. If there are no such operations, do not define 8839this macro. 8840@end defmac 8841 8842@defmac CLZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value}) 8843@defmacx CTZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value}) 8844A C expression that evaluates to true if the architecture defines a value 8845for @code{clz} or @code{ctz} with a zero operand. If so, @var{value} 8846should be set to this value. If this macro is not defined, the value of 8847@code{clz} or @code{ctz} is assumed to be undefined. 8848 8849This macro must be defined if the target's expansion for @code{ffs} 8850relies on a particular value to get correct results. Otherwise it 8851is not necessary, though it may be used to optimize some corner cases. 8852 8853Note that regardless of this macro the ``definedness'' of @code{clz} 8854and @code{ctz} at zero do @emph{not} extend to the builtin functions 8855visible to the user. Thus one may be free to adjust the value at will 8856to match the target expansion of these operations without fear of 8857breaking the API. 8858@end defmac 8859 8860@defmac Pmode 8861An alias for the machine mode for pointers. On most machines, define 8862this to be the integer mode corresponding to the width of a hardware 8863pointer; @code{SImode} on 32-bit machine or @code{DImode} on 64-bit machines. 8864On some machines you must define this to be one of the partial integer 8865modes, such as @code{PSImode}. 8866 8867The width of @code{Pmode} must be at least as large as the value of 8868@code{POINTER_SIZE}. If it is not equal, you must define the macro 8869@code{POINTERS_EXTEND_UNSIGNED} to specify how pointers are extended 8870to @code{Pmode}. 8871@end defmac 8872 8873@defmac FUNCTION_MODE 8874An alias for the machine mode used for memory references to functions 8875being called, in @code{call} RTL expressions. On most machines this 8876should be @code{QImode}. 8877@end defmac 8878 8879@defmac INTEGRATE_THRESHOLD (@var{decl}) 8880A C expression for the maximum number of instructions above which the 8881function @var{decl} should not be inlined. @var{decl} is a 8882@code{FUNCTION_DECL} node. 8883 8884The default definition of this macro is 64 plus 8 times the number of 8885arguments that the function accepts. Some people think a larger 8886threshold should be used on RISC machines. 8887@end defmac 8888 8889@defmac STDC_0_IN_SYSTEM_HEADERS 8890In normal operation, the preprocessor expands @code{__STDC__} to the 8891constant 1, to signify that GCC conforms to ISO Standard C@. On some 8892hosts, like Solaris, the system compiler uses a different convention, 8893where @code{__STDC__} is normally 0, but is 1 if the user specifies 8894strict conformance to the C Standard. 8895 8896Defining @code{STDC_0_IN_SYSTEM_HEADERS} makes GNU CPP follows the host 8897convention when processing system header files, but when processing user 8898files @code{__STDC__} will always expand to 1. 8899@end defmac 8900 8901@defmac NO_IMPLICIT_EXTERN_C 8902Define this macro if the system header files support C++ as well as C@. 8903This macro inhibits the usual method of using system header files in 8904C++, which is to pretend that the file's contents are enclosed in 8905@samp{extern "C" @{@dots{}@}}. 8906@end defmac 8907 8908@findex #pragma 8909@findex pragma 8910@defmac REGISTER_TARGET_PRAGMAS () 8911Define this macro if you want to implement any target-specific pragmas. 8912If defined, it is a C expression which makes a series of calls to 8913@code{c_register_pragma} for each pragma. The macro may also do any 8914setup required for the pragmas. 8915 8916The primary reason to define this macro is to provide compatibility with 8917other compilers for the same target. In general, we discourage 8918definition of target-specific pragmas for GCC@. 8919 8920If the pragma can be implemented by attributes then you should consider 8921defining the target hook @samp{TARGET_INSERT_ATTRIBUTES} as well. 8922 8923Preprocessor macros that appear on pragma lines are not expanded. All 8924@samp{#pragma} directives that do not match any registered pragma are 8925silently ignored, unless the user specifies @option{-Wunknown-pragmas}. 8926@end defmac 8927 8928@deftypefun void c_register_pragma (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *)) 8929 8930Each call to @code{c_register_pragma} establishes one pragma. The 8931@var{callback} routine will be called when the preprocessor encounters a 8932pragma of the form 8933 8934@smallexample 8935#pragma [@var{space}] @var{name} @dots{} 8936@end smallexample 8937 8938@var{space} is the case-sensitive namespace of the pragma, or 8939@code{NULL} to put the pragma in the global namespace. The callback 8940routine receives @var{pfile} as its first argument, which can be passed 8941on to cpplib's functions if necessary. You can lex tokens after the 8942@var{name} by calling @code{c_lex}. Tokens that are not read by the 8943callback will be silently ignored. The end of the line is indicated by 8944a token of type @code{CPP_EOF} 8945 8946For an example use of this routine, see @file{c4x.h} and the callback 8947routines defined in @file{c4x-c.c}. 8948 8949Note that the use of @code{c_lex} is specific to the C and C++ 8950compilers. It will not work in the Java or Fortran compilers, or any 8951other language compilers for that matter. Thus if @code{c_lex} is going 8952to be called from target-specific code, it must only be done so when 8953building the C and C++ compilers. This can be done by defining the 8954variables @code{c_target_objs} and @code{cxx_target_objs} in the 8955target entry in the @file{config.gcc} file. These variables should name 8956the target-specific, language-specific object file which contains the 8957code that uses @code{c_lex}. Note it will also be necessary to add a 8958rule to the makefile fragment pointed to by @code{tmake_file} that shows 8959how to build this object file. 8960@end deftypefun 8961 8962@findex #pragma 8963@findex pragma 8964@defmac HANDLE_SYSV_PRAGMA 8965Define this macro (to a value of 1) if you want the System V style 8966pragmas @samp{#pragma pack(<n>)} and @samp{#pragma weak <name> 8967[=<value>]} to be supported by gcc. 8968 8969The pack pragma specifies the maximum alignment (in bytes) of fields 8970within a structure, in much the same way as the @samp{__aligned__} and 8971@samp{__packed__} @code{__attribute__}s do. A pack value of zero resets 8972the behavior to the default. 8973 8974A subtlety for Microsoft Visual C/C++ style bit-field packing 8975(e.g. -mms-bitfields) for targets that support it: 8976When a bit-field is inserted into a packed record, the whole size 8977of the underlying type is used by one or more same-size adjacent 8978bit-fields (that is, if its long:3, 32 bits is used in the record, 8979and any additional adjacent long bit-fields are packed into the same 8980chunk of 32 bits. However, if the size changes, a new field of that 8981size is allocated). 8982 8983If both MS bit-fields and @samp{__attribute__((packed))} are used, 8984the latter will take precedence. If @samp{__attribute__((packed))} is 8985used on a single field when MS bit-fields are in use, it will take 8986precedence for that field, but the alignment of the rest of the structure 8987may affect its placement. 8988 8989The weak pragma only works if @code{SUPPORTS_WEAK} and 8990@code{ASM_WEAKEN_LABEL} are defined. If enabled it allows the creation 8991of specifically named weak labels, optionally with a value. 8992@end defmac 8993 8994@findex #pragma 8995@findex pragma 8996@defmac HANDLE_PRAGMA_PACK_PUSH_POP 8997Define this macro (to a value of 1) if you want to support the Win32 8998style pragmas @samp{#pragma pack(push,@var{n})} and @samp{#pragma 8999pack(pop)}. The @samp{pack(push,@var{n})} pragma specifies the maximum alignment 9000(in bytes) of fields within a structure, in much the same way as the 9001@samp{__aligned__} and @samp{__packed__} @code{__attribute__}s do. A 9002pack value of zero resets the behavior to the default. Successive 9003invocations of this pragma cause the previous values to be stacked, so 9004that invocations of @samp{#pragma pack(pop)} will return to the previous 9005value. 9006@end defmac 9007 9008@defmac DOLLARS_IN_IDENTIFIERS 9009Define this macro to control use of the character @samp{$} in 9010identifier names for the C family of languages. 0 means @samp{$} is 9011not allowed by default; 1 means it is allowed. 1 is the default; 9012there is no need to define this macro in that case. 9013@end defmac 9014 9015@defmac NO_DOLLAR_IN_LABEL 9016Define this macro if the assembler does not accept the character 9017@samp{$} in label names. By default constructors and destructors in 9018G++ have @samp{$} in the identifiers. If this macro is defined, 9019@samp{.} is used instead. 9020@end defmac 9021 9022@defmac NO_DOT_IN_LABEL 9023Define this macro if the assembler does not accept the character 9024@samp{.} in label names. By default constructors and destructors in G++ 9025have names that use @samp{.}. If this macro is defined, these names 9026are rewritten to avoid @samp{.}. 9027@end defmac 9028 9029@defmac DEFAULT_MAIN_RETURN 9030Define this macro if the target system expects every program's @code{main} 9031function to return a standard ``success'' value by default (if no other 9032value is explicitly returned). 9033 9034The definition should be a C statement (sans semicolon) to generate the 9035appropriate rtl instructions. It is used only when compiling the end of 9036@code{main}. 9037@end defmac 9038 9039@defmac INSN_SETS_ARE_DELAYED (@var{insn}) 9040Define this macro as a C expression that is nonzero if it is safe for the 9041delay slot scheduler to place instructions in the delay slot of @var{insn}, 9042even if they appear to use a resource set or clobbered in @var{insn}. 9043@var{insn} is always a @code{jump_insn} or an @code{insn}; GCC knows that 9044every @code{call_insn} has this behavior. On machines where some @code{insn} 9045or @code{jump_insn} is really a function call and hence has this behavior, 9046you should define this macro. 9047 9048You need not define this macro if it would always return zero. 9049@end defmac 9050 9051@defmac INSN_REFERENCES_ARE_DELAYED (@var{insn}) 9052Define this macro as a C expression that is nonzero if it is safe for the 9053delay slot scheduler to place instructions in the delay slot of @var{insn}, 9054even if they appear to set or clobber a resource referenced in @var{insn}. 9055@var{insn} is always a @code{jump_insn} or an @code{insn}. On machines where 9056some @code{insn} or @code{jump_insn} is really a function call and its operands 9057are registers whose use is actually in the subroutine it calls, you should 9058define this macro. Doing so allows the delay slot scheduler to move 9059instructions which copy arguments into the argument registers into the delay 9060slot of @var{insn}. 9061 9062You need not define this macro if it would always return zero. 9063@end defmac 9064 9065@defmac MULTIPLE_SYMBOL_SPACES 9066Define this macro if in some cases global symbols from one translation 9067unit may not be bound to undefined symbols in another translation unit 9068without user intervention. For instance, under Microsoft Windows 9069symbols must be explicitly imported from shared libraries (DLLs). 9070@end defmac 9071 9072@defmac MD_ASM_CLOBBERS (@var{clobbers}) 9073A C statement that adds to @var{clobbers} @code{STRING_CST} trees for 9074any hard regs the port wishes to automatically clobber for all asms. 9075@end defmac 9076 9077@defmac MATH_LIBRARY 9078Define this macro as a C string constant for the linker argument to link 9079in the system math library, or @samp{""} if the target does not have a 9080separate math library. 9081 9082You need only define this macro if the default of @samp{"-lm"} is wrong. 9083@end defmac 9084 9085@defmac LIBRARY_PATH_ENV 9086Define this macro as a C string constant for the environment variable that 9087specifies where the linker should look for libraries. 9088 9089You need only define this macro if the default of @samp{"LIBRARY_PATH"} 9090is wrong. 9091@end defmac 9092 9093@defmac TARGET_HAS_F_SETLKW 9094Define this macro if the target supports file locking with fcntl / F_SETLKW@. 9095Note that this functionality is part of POSIX@. 9096Defining @code{TARGET_HAS_F_SETLKW} will enable the test coverage code 9097to use file locking when exiting a program, which avoids race conditions 9098if the program has forked. 9099@end defmac 9100 9101@defmac MAX_CONDITIONAL_EXECUTE 9102 9103A C expression for the maximum number of instructions to execute via 9104conditional execution instructions instead of a branch. A value of 9105@code{BRANCH_COST}+1 is the default if the machine does not use cc0, and 91061 if it does use cc0. 9107@end defmac 9108 9109@defmac IFCVT_MODIFY_TESTS (@var{ce_info}, @var{true_expr}, @var{false_expr}) 9110Used if the target needs to perform machine-dependent modifications on the 9111conditionals used for turning basic blocks into conditionally executed code. 9112@var{ce_info} points to a data structure, @code{struct ce_if_block}, which 9113contains information about the currently processed blocks. @var{true_expr} 9114and @var{false_expr} are the tests that are used for converting the 9115then-block and the else-block, respectively. Set either @var{true_expr} or 9116@var{false_expr} to a null pointer if the tests cannot be converted. 9117@end defmac 9118 9119@defmac IFCVT_MODIFY_MULTIPLE_TESTS (@var{ce_info}, @var{bb}, @var{true_expr}, @var{false_expr}) 9120Like @code{IFCVT_MODIFY_TESTS}, but used when converting more complicated 9121if-statements into conditions combined by @code{and} and @code{or} operations. 9122@var{bb} contains the basic block that contains the test that is currently 9123being processed and about to be turned into a condition. 9124@end defmac 9125 9126@defmac IFCVT_MODIFY_INSN (@var{ce_info}, @var{pattern}, @var{insn}) 9127A C expression to modify the @var{PATTERN} of an @var{INSN} that is to 9128be converted to conditional execution format. @var{ce_info} points to 9129a data structure, @code{struct ce_if_block}, which contains information 9130about the currently processed blocks. 9131@end defmac 9132 9133@defmac IFCVT_MODIFY_FINAL (@var{ce_info}) 9134A C expression to perform any final machine dependent modifications in 9135converting code to conditional execution. The involved basic blocks 9136can be found in the @code{struct ce_if_block} structure that is pointed 9137to by @var{ce_info}. 9138@end defmac 9139 9140@defmac IFCVT_MODIFY_CANCEL (@var{ce_info}) 9141A C expression to cancel any machine dependent modifications in 9142converting code to conditional execution. The involved basic blocks 9143can be found in the @code{struct ce_if_block} structure that is pointed 9144to by @var{ce_info}. 9145@end defmac 9146 9147@defmac IFCVT_INIT_EXTRA_FIELDS (@var{ce_info}) 9148A C expression to initialize any extra fields in a @code{struct ce_if_block} 9149structure, which are defined by the @code{IFCVT_EXTRA_FIELDS} macro. 9150@end defmac 9151 9152@defmac IFCVT_EXTRA_FIELDS 9153If defined, it should expand to a set of field declarations that will be 9154added to the @code{struct ce_if_block} structure. These should be initialized 9155by the @code{IFCVT_INIT_EXTRA_FIELDS} macro. 9156@end defmac 9157 9158@deftypefn {Target Hook} void TARGET_MACHINE_DEPENDENT_REORG () 9159If non-null, this hook performs a target-specific pass over the 9160instruction stream. The compiler will run it at all optimization levels, 9161just before the point at which it normally does delayed-branch scheduling. 9162 9163The exact purpose of the hook varies from target to target. Some use 9164it to do transformations that are necessary for correctness, such as 9165laying out in-function constant pools or avoiding hardware hazards. 9166Others use it as an opportunity to do some machine-dependent optimizations. 9167 9168You need not implement the hook if it has nothing to do. The default 9169definition is null. 9170@end deftypefn 9171 9172@deftypefn {Target Hook} void TARGET_INIT_BUILTINS () 9173Define this hook if you have any machine-specific built-in functions 9174that need to be defined. It should be a function that performs the 9175necessary setup. 9176 9177Machine specific built-in functions can be useful to expand special machine 9178instructions that would otherwise not normally be generated because 9179they have no equivalent in the source language (for example, SIMD vector 9180instructions or prefetch instructions). 9181 9182To create a built-in function, call the function @code{builtin_function} 9183which is defined by the language front end. You can use any type nodes set 9184up by @code{build_common_tree_nodes} and @code{build_common_tree_nodes_2}; 9185only language front ends that use those two functions will call 9186@samp{TARGET_INIT_BUILTINS}. 9187@end deftypefn 9188 9189@deftypefn {Target Hook} rtx TARGET_EXPAND_BUILTIN (tree @var{exp}, rtx @var{target}, rtx @var{subtarget}, enum machine_mode @var{mode}, int @var{ignore}) 9190 9191Expand a call to a machine specific built-in function that was set up by 9192@samp{TARGET_INIT_BUILTINS}. @var{exp} is the expression for the 9193function call; the result should go to @var{target} if that is 9194convenient, and have mode @var{mode} if that is convenient. 9195@var{subtarget} may be used as the target for computing one of 9196@var{exp}'s operands. @var{ignore} is nonzero if the value is to be 9197ignored. This function should return the result of the call to the 9198built-in function. 9199@end deftypefn 9200 9201@defmac MD_CAN_REDIRECT_BRANCH (@var{branch1}, @var{branch2}) 9202 9203Take a branch insn in @var{branch1} and another in @var{branch2}. 9204Return true if redirecting @var{branch1} to the destination of 9205@var{branch2} is possible. 9206 9207On some targets, branches may have a limited range. Optimizing the 9208filling of delay slots can result in branches being redirected, and this 9209may in turn cause a branch offset to overflow. 9210@end defmac 9211 9212@defmac ALLOCATE_INITIAL_VALUE (@var{hard_reg}) 9213 9214When the initial value of a hard register has been copied in a pseudo 9215register, it is often not necessary to actually allocate another register 9216to this pseudo register, because the original hard register or a stack slot 9217it has been saved into can be used. @code{ALLOCATE_INITIAL_VALUE}, if 9218defined, is called at the start of register allocation once for each 9219hard register that had its initial value copied by using 9220@code{get_func_hard_reg_initial_val} or @code{get_hard_reg_initial_val}. 9221Possible values are @code{NULL_RTX}, if you don't want 9222to do any special allocation, a @code{REG} rtx---that would typically be 9223the hard register itself, if it is known not to be clobbered---or a 9224@code{MEM}. 9225If you are returning a @code{MEM}, this is only a hint for the allocator; 9226it might decide to use another register anyways. 9227You may use @code{current_function_leaf_function} in the definition of the 9228macro, functions that use @code{REG_N_SETS}, to determine if the hard 9229register in question will not be clobbered. 9230@end defmac 9231 9232@defmac TARGET_OBJECT_SUFFIX 9233Define this macro to be a C string representing the suffix for object 9234files on your target machine. If you do not define this macro, GCC will 9235use @samp{.o} as the suffix for object files. 9236@end defmac 9237 9238@defmac TARGET_EXECUTABLE_SUFFIX 9239Define this macro to be a C string representing the suffix to be 9240automatically added to executable files on your target machine. If you 9241do not define this macro, GCC will use the null string as the suffix for 9242executable files. 9243@end defmac 9244 9245@defmac COLLECT_EXPORT_LIST 9246If defined, @code{collect2} will scan the individual object files 9247specified on its command line and create an export list for the linker. 9248Define this macro for systems like AIX, where the linker discards 9249object files that are not referenced from @code{main} and uses export 9250lists. 9251@end defmac 9252 9253@defmac MODIFY_JNI_METHOD_CALL (@var{mdecl}) 9254Define this macro to a C expression representing a variant of the 9255method call @var{mdecl}, if Java Native Interface (JNI) methods 9256must be invoked differently from other methods on your target. 9257For example, on 32-bit Microsoft Windows, JNI methods must be invoked using 9258the @code{stdcall} calling convention and this macro is then 9259defined as this expression: 9260 9261@smallexample 9262build_type_attribute_variant (@var{mdecl}, 9263 build_tree_list 9264 (get_identifier ("stdcall"), 9265 NULL)) 9266@end smallexample 9267@end defmac 9268 9269@deftypefn {Target Hook} bool TARGET_CANNOT_MODIFY_JUMPS_P (void) 9270This target hook returns @code{true} past the point in which new jump 9271instructions could be created. On machines that require a register for 9272every jump such as the SHmedia ISA of SH5, this point would typically be 9273reload, so this target hook should be defined to a function such as: 9274 9275@smallexample 9276static bool 9277cannot_modify_jumps_past_reload_p () 9278@{ 9279 return (reload_completed || reload_in_progress); 9280@} 9281@end smallexample 9282@end deftypefn 9283 9284@deftypefn {Target Hook} int TARGET_BRANCH_TARGET_REGISTER_CLASS (void) 9285This target hook returns a register class for which branch target register 9286optimizations should be applied. All registers in this class should be 9287usable interchangeably. After reload, registers in this class will be 9288re-allocated and loads will be hoisted out of loops and be subjected 9289to inter-block scheduling. 9290@end deftypefn 9291 9292@deftypefn {Target Hook} bool TARGET_BRANCH_TARGET_REGISTER_CALLEE_SAVED (bool @var{after_prologue_epilogue_gen}) 9293Branch target register optimization will by default exclude callee-saved 9294registers 9295that are not already live during the current function; if this target hook 9296returns true, they will be included. The target code must than make sure 9297that all target registers in the class returned by 9298@samp{TARGET_BRANCH_TARGET_REGISTER_CLASS} that might need saving are 9299saved. @var{after_prologue_epilogue_gen} indicates if prologues and 9300epilogues have already been generated. Note, even if you only return 9301true when @var{after_prologue_epilogue_gen} is false, you still are likely 9302to have to make special provisions in @code{INITIAL_ELIMINATION_OFFSET} 9303to reserve space for caller-saved target registers. 9304@end deftypefn 9305 9306@defmac POWI_MAX_MULTS 9307If defined, this macro is interpreted as a signed integer C expression 9308that specifies the maximum number of floating point multiplications 9309that should be emitted when expanding exponentiation by an integer 9310constant inline. When this value is defined, exponentiation requiring 9311more than this number of multiplications is implemented by calling the 9312system library's @code{pow}, @code{powf} or @code{powl} routines. 9313The default value places no upper bound on the multiplication count. 9314@end defmac 9315