1@c Copyright (C) 1988-2018 Free Software Foundation, Inc. 2@c This is part of the GCC manual. 3@c For copying conditions, see the file gcc.texi. 4 5@node Target Macros 6@chapter Target Description Macros and Functions 7@cindex machine description macros 8@cindex target description macros 9@cindex macros, target description 10@cindex @file{tm.h} macros 11 12In addition to the file @file{@var{machine}.md}, a machine description 13includes a C header file conventionally given the name 14@file{@var{machine}.h} and a C source file named @file{@var{machine}.c}. 15The header file defines numerous macros that convey the information 16about the target machine that does not fit into the scheme of the 17@file{.md} file. The file @file{tm.h} should be a link to 18@file{@var{machine}.h}. The header file @file{config.h} includes 19@file{tm.h} and most compiler source files include @file{config.h}. The 20source file defines a variable @code{targetm}, which is a structure 21containing pointers to functions and data relating to the target 22machine. @file{@var{machine}.c} should also contain their definitions, 23if they are not defined elsewhere in GCC, and other functions called 24through the macros defined in the @file{.h} file. 25 26@menu 27* Target Structure:: The @code{targetm} variable. 28* Driver:: Controlling how the driver runs the compilation passes. 29* Run-time Target:: Defining @samp{-m} options like @option{-m68000} and @option{-m68020}. 30* Per-Function Data:: Defining data structures for per-function information. 31* Storage Layout:: Defining sizes and alignments of data. 32* Type Layout:: Defining sizes and properties of basic user data types. 33* Registers:: Naming and describing the hardware registers. 34* Register Classes:: Defining the classes of hardware registers. 35* Stack and Calling:: Defining which way the stack grows and by how much. 36* Varargs:: Defining the varargs macros. 37* Trampolines:: Code set up at run time to enter a nested function. 38* Library Calls:: Controlling how library routines are implicitly called. 39* Addressing Modes:: Defining addressing modes valid for memory operands. 40* Anchored Addresses:: Defining how @option{-fsection-anchors} should work. 41* Condition Code:: Defining how insns update the condition code. 42* Costs:: Defining relative costs of different operations. 43* Scheduling:: Adjusting the behavior of the instruction scheduler. 44* Sections:: Dividing storage into text, data, and other sections. 45* PIC:: Macros for position independent code. 46* Assembler Format:: Defining how to write insns and pseudo-ops to output. 47* Debugging Info:: Defining the format of debugging output. 48* Floating Point:: Handling floating point for cross-compilers. 49* Mode Switching:: Insertion of mode-switching instructions. 50* Target Attributes:: Defining target-specific uses of @code{__attribute__}. 51* Emulated TLS:: Emulated TLS support. 52* MIPS Coprocessors:: MIPS coprocessor support and how to customize it. 53* PCH Target:: Validity checking for precompiled headers. 54* C++ ABI:: Controlling C++ ABI changes. 55* Named Address Spaces:: Adding support for named address spaces 56* Misc:: Everything else. 57@end menu 58 59@node Target Structure 60@section The Global @code{targetm} Variable 61@cindex target hooks 62@cindex target functions 63 64@deftypevar {struct gcc_target} targetm 65The target @file{.c} file must define the global @code{targetm} variable 66which contains pointers to functions and data relating to the target 67machine. The variable is declared in @file{target.h}; 68@file{target-def.h} defines the macro @code{TARGET_INITIALIZER} which is 69used to initialize the variable, and macros for the default initializers 70for elements of the structure. The @file{.c} file should override those 71macros for which the default definition is inappropriate. For example: 72@smallexample 73#include "target.h" 74#include "target-def.h" 75 76/* @r{Initialize the GCC target structure.} */ 77 78#undef TARGET_COMP_TYPE_ATTRIBUTES 79#define TARGET_COMP_TYPE_ATTRIBUTES @var{machine}_comp_type_attributes 80 81struct gcc_target targetm = TARGET_INITIALIZER; 82@end smallexample 83@end deftypevar 84 85Where a macro should be defined in the @file{.c} file in this manner to 86form part of the @code{targetm} structure, it is documented below as a 87``Target Hook'' with a prototype. Many macros will change in future 88from being defined in the @file{.h} file to being part of the 89@code{targetm} structure. 90 91Similarly, there is a @code{targetcm} variable for hooks that are 92specific to front ends for C-family languages, documented as ``C 93Target Hook''. This is declared in @file{c-family/c-target.h}, the 94initializer @code{TARGETCM_INITIALIZER} in 95@file{c-family/c-target-def.h}. If targets initialize @code{targetcm} 96themselves, they should set @code{target_has_targetcm=yes} in 97@file{config.gcc}; otherwise a default definition is used. 98 99Similarly, there is a @code{targetm_common} variable for hooks that 100are shared between the compiler driver and the compilers proper, 101documented as ``Common Target Hook''. This is declared in 102@file{common/common-target.h}, the initializer 103@code{TARGETM_COMMON_INITIALIZER} in 104@file{common/common-target-def.h}. If targets initialize 105@code{targetm_common} themselves, they should set 106@code{target_has_targetm_common=yes} in @file{config.gcc}; otherwise a 107default definition is used. 108 109@node Driver 110@section Controlling the Compilation Driver, @file{gcc} 111@cindex driver 112@cindex controlling the compilation driver 113 114@c prevent bad page break with this line 115You can control the compilation driver. 116 117@defmac DRIVER_SELF_SPECS 118A list of specs for the driver itself. It should be a suitable 119initializer for an array of strings, with no surrounding braces. 120 121The driver applies these specs to its own command line between loading 122default @file{specs} files (but not command-line specified ones) and 123choosing the multilib directory or running any subcommands. It 124applies them in the order given, so each spec can depend on the 125options added by earlier ones. It is also possible to remove options 126using @samp{%<@var{option}} in the usual way. 127 128This macro can be useful when a port has several interdependent target 129options. It provides a way of standardizing the command line so 130that the other specs are easier to write. 131 132Do not define this macro if it does not need to do anything. 133@end defmac 134 135@defmac OPTION_DEFAULT_SPECS 136A list of specs used to support configure-time default options (i.e.@: 137@option{--with} options) in the driver. It should be a suitable initializer 138for an array of structures, each containing two strings, without the 139outermost pair of surrounding braces. 140 141The first item in the pair is the name of the default. This must match 142the code in @file{config.gcc} for the target. The second item is a spec 143to apply if a default with this name was specified. The string 144@samp{%(VALUE)} in the spec will be replaced by the value of the default 145everywhere it occurs. 146 147The driver will apply these specs to its own command line between loading 148default @file{specs} files and processing @code{DRIVER_SELF_SPECS}, using 149the same mechanism as @code{DRIVER_SELF_SPECS}. 150 151Do not define this macro if it does not need to do anything. 152@end defmac 153 154@defmac CPP_SPEC 155A C string constant that tells the GCC driver program options to 156pass to CPP@. It can also specify how to translate options you 157give to GCC into options for GCC to pass to the CPP@. 158 159Do not define this macro if it does not need to do anything. 160@end defmac 161 162@defmac CPLUSPLUS_CPP_SPEC 163This macro is just like @code{CPP_SPEC}, but is used for C++, rather 164than C@. If you do not define this macro, then the value of 165@code{CPP_SPEC} (if any) will be used instead. 166@end defmac 167 168@defmac CC1_SPEC 169A C string constant that tells the GCC driver program options to 170pass to @code{cc1}, @code{cc1plus}, @code{f771}, and the other language 171front ends. 172It can also specify how to translate options you give to GCC into options 173for GCC to pass to front ends. 174 175Do not define this macro if it does not need to do anything. 176@end defmac 177 178@defmac CC1PLUS_SPEC 179A C string constant that tells the GCC driver program options to 180pass to @code{cc1plus}. It can also specify how to translate options you 181give to GCC into options for GCC to pass to the @code{cc1plus}. 182 183Do not define this macro if it does not need to do anything. 184Note that everything defined in CC1_SPEC is already passed to 185@code{cc1plus} so there is no need to duplicate the contents of 186CC1_SPEC in CC1PLUS_SPEC@. 187@end defmac 188 189@defmac ASM_SPEC 190A C string constant that tells the GCC driver program options to 191pass to the assembler. It can also specify how to translate options 192you give to GCC into options for GCC to pass to the assembler. 193See the file @file{sun3.h} for an example of this. 194 195Do not define this macro if it does not need to do anything. 196@end defmac 197 198@defmac ASM_FINAL_SPEC 199A C string constant that tells the GCC driver program how to 200run any programs which cleanup after the normal assembler. 201Normally, this is not needed. See the file @file{mips.h} for 202an example of this. 203 204Do not define this macro if it does not need to do anything. 205@end defmac 206 207@defmac AS_NEEDS_DASH_FOR_PIPED_INPUT 208Define this macro, with no value, if the driver should give the assembler 209an argument consisting of a single dash, @option{-}, to instruct it to 210read from its standard input (which will be a pipe connected to the 211output of the compiler proper). This argument is given after any 212@option{-o} option specifying the name of the output file. 213 214If you do not define this macro, the assembler is assumed to read its 215standard input if given no non-option arguments. If your assembler 216cannot read standard input at all, use a @samp{%@{pipe:%e@}} construct; 217see @file{mips.h} for instance. 218@end defmac 219 220@defmac LINK_SPEC 221A C string constant that tells the GCC driver program options to 222pass to the linker. It can also specify how to translate options you 223give to GCC into options for GCC to pass to the linker. 224 225Do not define this macro if it does not need to do anything. 226@end defmac 227 228@defmac LIB_SPEC 229Another C string constant used much like @code{LINK_SPEC}. The difference 230between the two is that @code{LIB_SPEC} is used at the end of the 231command given to the linker. 232 233If this macro is not defined, a default is provided that 234loads the standard C library from the usual place. See @file{gcc.c}. 235@end defmac 236 237@defmac LIBGCC_SPEC 238Another C string constant that tells the GCC driver program 239how and when to place a reference to @file{libgcc.a} into the 240linker command line. This constant is placed both before and after 241the value of @code{LIB_SPEC}. 242 243If this macro is not defined, the GCC driver provides a default that 244passes the string @option{-lgcc} to the linker. 245@end defmac 246 247@defmac REAL_LIBGCC_SPEC 248By default, if @code{ENABLE_SHARED_LIBGCC} is defined, the 249@code{LIBGCC_SPEC} is not directly used by the driver program but is 250instead modified to refer to different versions of @file{libgcc.a} 251depending on the values of the command line flags @option{-static}, 252@option{-shared}, @option{-static-libgcc}, and @option{-shared-libgcc}. On 253targets where these modifications are inappropriate, define 254@code{REAL_LIBGCC_SPEC} instead. @code{REAL_LIBGCC_SPEC} tells the 255driver how to place a reference to @file{libgcc} on the link command 256line, but, unlike @code{LIBGCC_SPEC}, it is used unmodified. 257@end defmac 258 259@defmac USE_LD_AS_NEEDED 260A macro that controls the modifications to @code{LIBGCC_SPEC} 261mentioned in @code{REAL_LIBGCC_SPEC}. If nonzero, a spec will be 262generated that uses @option{--as-needed} or equivalent options and the 263shared @file{libgcc} in place of the 264static exception handler library, when linking without any of 265@code{-static}, @code{-static-libgcc}, or @code{-shared-libgcc}. 266@end defmac 267 268@defmac LINK_EH_SPEC 269If defined, this C string constant is added to @code{LINK_SPEC}. 270When @code{USE_LD_AS_NEEDED} is zero or undefined, it also affects 271the modifications to @code{LIBGCC_SPEC} mentioned in 272@code{REAL_LIBGCC_SPEC}. 273@end defmac 274 275@defmac STARTFILE_SPEC 276Another C string constant used much like @code{LINK_SPEC}. The 277difference between the two is that @code{STARTFILE_SPEC} is used at 278the very beginning of the command given to the linker. 279 280If this macro is not defined, a default is provided that loads the 281standard C startup file from the usual place. See @file{gcc.c}. 282@end defmac 283 284@defmac ENDFILE_SPEC 285Another C string constant used much like @code{LINK_SPEC}. The 286difference between the two is that @code{ENDFILE_SPEC} is used at 287the very end of the command given to the linker. 288 289Do not define this macro if it does not need to do anything. 290@end defmac 291 292@defmac THREAD_MODEL_SPEC 293GCC @code{-v} will print the thread model GCC was configured to use. 294However, this doesn't work on platforms that are multilibbed on thread 295models, such as AIX 4.3. On such platforms, define 296@code{THREAD_MODEL_SPEC} such that it evaluates to a string without 297blanks that names one of the recognized thread models. @code{%*}, the 298default value of this macro, will expand to the value of 299@code{thread_file} set in @file{config.gcc}. 300@end defmac 301 302@defmac SYSROOT_SUFFIX_SPEC 303Define this macro to add a suffix to the target sysroot when GCC is 304configured with a sysroot. This will cause GCC to search for usr/lib, 305et al, within sysroot+suffix. 306@end defmac 307 308@defmac SYSROOT_HEADERS_SUFFIX_SPEC 309Define this macro to add a headers_suffix to the target sysroot when 310GCC is configured with a sysroot. This will cause GCC to pass the 311updated sysroot+headers_suffix to CPP, causing it to search for 312usr/include, et al, within sysroot+headers_suffix. 313@end defmac 314 315@defmac EXTRA_SPECS 316Define this macro to provide additional specifications to put in the 317@file{specs} file that can be used in various specifications like 318@code{CC1_SPEC}. 319 320The definition should be an initializer for an array of structures, 321containing a string constant, that defines the specification name, and a 322string constant that provides the specification. 323 324Do not define this macro if it does not need to do anything. 325 326@code{EXTRA_SPECS} is useful when an architecture contains several 327related targets, which have various @code{@dots{}_SPECS} which are similar 328to each other, and the maintainer would like one central place to keep 329these definitions. 330 331For example, the PowerPC System V.4 targets use @code{EXTRA_SPECS} to 332define either @code{_CALL_SYSV} when the System V calling sequence is 333used or @code{_CALL_AIX} when the older AIX-based calling sequence is 334used. 335 336The @file{config/rs6000/rs6000.h} target file defines: 337 338@smallexample 339#define EXTRA_SPECS \ 340 @{ "cpp_sysv_default", CPP_SYSV_DEFAULT @}, 341 342#define CPP_SYS_DEFAULT "" 343@end smallexample 344 345The @file{config/rs6000/sysv.h} target file defines: 346@smallexample 347#undef CPP_SPEC 348#define CPP_SPEC \ 349"%@{posix: -D_POSIX_SOURCE @} \ 350%@{mcall-sysv: -D_CALL_SYSV @} \ 351%@{!mcall-sysv: %(cpp_sysv_default) @} \ 352%@{msoft-float: -D_SOFT_FLOAT@} %@{mcpu=403: -D_SOFT_FLOAT@}" 353 354#undef CPP_SYSV_DEFAULT 355#define CPP_SYSV_DEFAULT "-D_CALL_SYSV" 356@end smallexample 357 358while the @file{config/rs6000/eabiaix.h} target file defines 359@code{CPP_SYSV_DEFAULT} as: 360 361@smallexample 362#undef CPP_SYSV_DEFAULT 363#define CPP_SYSV_DEFAULT "-D_CALL_AIX" 364@end smallexample 365@end defmac 366 367@defmac LINK_LIBGCC_SPECIAL_1 368Define this macro if the driver program should find the library 369@file{libgcc.a}. If you do not define this macro, the driver program will pass 370the argument @option{-lgcc} to tell the linker to do the search. 371@end defmac 372 373@defmac LINK_GCC_C_SEQUENCE_SPEC 374The sequence in which libgcc and libc are specified to the linker. 375By default this is @code{%G %L %G}. 376@end defmac 377 378@defmac POST_LINK_SPEC 379Define this macro to add additional steps to be executed after linker. 380The default value of this macro is empty string. 381@end defmac 382 383@defmac LINK_COMMAND_SPEC 384A C string constant giving the complete command line need to execute the 385linker. When you do this, you will need to update your port each time a 386change is made to the link command line within @file{gcc.c}. Therefore, 387define this macro only if you need to completely redefine the command 388line for invoking the linker and there is no other way to accomplish 389the effect you need. Overriding this macro may be avoidable by overriding 390@code{LINK_GCC_C_SEQUENCE_SPEC} instead. 391@end defmac 392 393@deftypevr {Common Target Hook} bool TARGET_ALWAYS_STRIP_DOTDOT 394True if @file{..} components should always be removed from directory names computed relative to GCC's internal directories, false (default) if such components should be preserved and directory names containing them passed to other tools such as the linker. 395@end deftypevr 396 397@defmac MULTILIB_DEFAULTS 398Define this macro as a C expression for the initializer of an array of 399string to tell the driver program which options are defaults for this 400target and thus do not need to be handled specially when using 401@code{MULTILIB_OPTIONS}. 402 403Do not define this macro if @code{MULTILIB_OPTIONS} is not defined in 404the target makefile fragment or if none of the options listed in 405@code{MULTILIB_OPTIONS} are set by default. 406@xref{Target Fragment}. 407@end defmac 408 409@defmac RELATIVE_PREFIX_NOT_LINKDIR 410Define this macro to tell @command{gcc} that it should only translate 411a @option{-B} prefix into a @option{-L} linker option if the prefix 412indicates an absolute file name. 413@end defmac 414 415@defmac MD_EXEC_PREFIX 416If defined, this macro is an additional prefix to try after 417@code{STANDARD_EXEC_PREFIX}. @code{MD_EXEC_PREFIX} is not searched 418when the compiler is built as a cross 419compiler. If you define @code{MD_EXEC_PREFIX}, then be sure to add it 420to the list of directories used to find the assembler in @file{configure.ac}. 421@end defmac 422 423@defmac STANDARD_STARTFILE_PREFIX 424Define this macro as a C string constant if you wish to override the 425standard choice of @code{libdir} as the default prefix to 426try when searching for startup files such as @file{crt0.o}. 427@code{STANDARD_STARTFILE_PREFIX} is not searched when the compiler 428is built as a cross compiler. 429@end defmac 430 431@defmac STANDARD_STARTFILE_PREFIX_1 432Define this macro as a C string constant if you wish to override the 433standard choice of @code{/lib} as a prefix to try after the default prefix 434when searching for startup files such as @file{crt0.o}. 435@code{STANDARD_STARTFILE_PREFIX_1} is not searched when the compiler 436is built as a cross compiler. 437@end defmac 438 439@defmac STANDARD_STARTFILE_PREFIX_2 440Define this macro as a C string constant if you wish to override the 441standard choice of @code{/lib} as yet another prefix to try after the 442default prefix when searching for startup files such as @file{crt0.o}. 443@code{STANDARD_STARTFILE_PREFIX_2} is not searched when the compiler 444is built as a cross compiler. 445@end defmac 446 447@defmac MD_STARTFILE_PREFIX 448If defined, this macro supplies an additional prefix to try after the 449standard prefixes. @code{MD_EXEC_PREFIX} is not searched when the 450compiler is built as a cross compiler. 451@end defmac 452 453@defmac MD_STARTFILE_PREFIX_1 454If defined, this macro supplies yet another prefix to try after the 455standard prefixes. It is not searched when the compiler is built as a 456cross compiler. 457@end defmac 458 459@defmac INIT_ENVIRONMENT 460Define this macro as a C string constant if you wish to set environment 461variables for programs called by the driver, such as the assembler and 462loader. The driver passes the value of this macro to @code{putenv} to 463initialize the necessary environment variables. 464@end defmac 465 466@defmac LOCAL_INCLUDE_DIR 467Define this macro as a C string constant if you wish to override the 468standard choice of @file{/usr/local/include} as the default prefix to 469try when searching for local header files. @code{LOCAL_INCLUDE_DIR} 470comes before @code{NATIVE_SYSTEM_HEADER_DIR} (set in 471@file{config.gcc}, normally @file{/usr/include}) in the search order. 472 473Cross compilers do not search either @file{/usr/local/include} or its 474replacement. 475@end defmac 476 477@defmac NATIVE_SYSTEM_HEADER_COMPONENT 478The ``component'' corresponding to @code{NATIVE_SYSTEM_HEADER_DIR}. 479See @code{INCLUDE_DEFAULTS}, below, for the description of components. 480If you do not define this macro, no component is used. 481@end defmac 482 483@defmac INCLUDE_DEFAULTS 484Define this macro if you wish to override the entire default search path 485for include files. For a native compiler, the default search path 486usually consists of @code{GCC_INCLUDE_DIR}, @code{LOCAL_INCLUDE_DIR}, 487@code{GPLUSPLUS_INCLUDE_DIR}, and 488@code{NATIVE_SYSTEM_HEADER_DIR}. In addition, @code{GPLUSPLUS_INCLUDE_DIR} 489and @code{GCC_INCLUDE_DIR} are defined automatically by @file{Makefile}, 490and specify private search areas for GCC@. The directory 491@code{GPLUSPLUS_INCLUDE_DIR} is used only for C++ programs. 492 493The definition should be an initializer for an array of structures. 494Each array element should have four elements: the directory name (a 495string constant), the component name (also a string constant), a flag 496for C++-only directories, 497and a flag showing that the includes in the directory don't need to be 498wrapped in @code{extern @samp{C}} when compiling C++. Mark the end of 499the array with a null element. 500 501The component name denotes what GNU package the include file is part of, 502if any, in all uppercase letters. For example, it might be @samp{GCC} 503or @samp{BINUTILS}. If the package is part of a vendor-supplied 504operating system, code the component name as @samp{0}. 505 506For example, here is the definition used for VAX/VMS: 507 508@smallexample 509#define INCLUDE_DEFAULTS \ 510@{ \ 511 @{ "GNU_GXX_INCLUDE:", "G++", 1, 1@}, \ 512 @{ "GNU_CC_INCLUDE:", "GCC", 0, 0@}, \ 513 @{ "SYS$SYSROOT:[SYSLIB.]", 0, 0, 0@}, \ 514 @{ ".", 0, 0, 0@}, \ 515 @{ 0, 0, 0, 0@} \ 516@} 517@end smallexample 518@end defmac 519 520Here is the order of prefixes tried for exec files: 521 522@enumerate 523@item 524Any prefixes specified by the user with @option{-B}. 525 526@item 527The environment variable @code{GCC_EXEC_PREFIX} or, if @code{GCC_EXEC_PREFIX} 528is not set and the compiler has not been installed in the configure-time 529@var{prefix}, the location in which the compiler has actually been installed. 530 531@item 532The directories specified by the environment variable @code{COMPILER_PATH}. 533 534@item 535The macro @code{STANDARD_EXEC_PREFIX}, if the compiler has been installed 536in the configured-time @var{prefix}. 537 538@item 539The location @file{/usr/libexec/gcc/}, but only if this is a native compiler. 540 541@item 542The location @file{/usr/lib/gcc/}, but only if this is a native compiler. 543 544@item 545The macro @code{MD_EXEC_PREFIX}, if defined, but only if this is a native 546compiler. 547@end enumerate 548 549Here is the order of prefixes tried for startfiles: 550 551@enumerate 552@item 553Any prefixes specified by the user with @option{-B}. 554 555@item 556The environment variable @code{GCC_EXEC_PREFIX} or its automatically determined 557value based on the installed toolchain location. 558 559@item 560The directories specified by the environment variable @code{LIBRARY_PATH} 561(or port-specific name; native only, cross compilers do not use this). 562 563@item 564The macro @code{STANDARD_EXEC_PREFIX}, but only if the toolchain is installed 565in the configured @var{prefix} or this is a native compiler. 566 567@item 568The location @file{/usr/lib/gcc/}, but only if this is a native compiler. 569 570@item 571The macro @code{MD_EXEC_PREFIX}, if defined, but only if this is a native 572compiler. 573 574@item 575The macro @code{MD_STARTFILE_PREFIX}, if defined, but only if this is a 576native compiler, or we have a target system root. 577 578@item 579The macro @code{MD_STARTFILE_PREFIX_1}, if defined, but only if this is a 580native compiler, or we have a target system root. 581 582@item 583The macro @code{STANDARD_STARTFILE_PREFIX}, with any sysroot modifications. 584If this path is relative it will be prefixed by @code{GCC_EXEC_PREFIX} and 585the machine suffix or @code{STANDARD_EXEC_PREFIX} and the machine suffix. 586 587@item 588The macro @code{STANDARD_STARTFILE_PREFIX_1}, but only if this is a native 589compiler, or we have a target system root. The default for this macro is 590@file{/lib/}. 591 592@item 593The macro @code{STANDARD_STARTFILE_PREFIX_2}, but only if this is a native 594compiler, or we have a target system root. The default for this macro is 595@file{/usr/lib/}. 596@end enumerate 597 598@node Run-time Target 599@section Run-time Target Specification 600@cindex run-time target specification 601@cindex predefined macros 602@cindex target specifications 603 604@c prevent bad page break with this line 605Here are run-time target specifications. 606 607@defmac TARGET_CPU_CPP_BUILTINS () 608This function-like macro expands to a block of code that defines 609built-in preprocessor macros and assertions for the target CPU, using 610the functions @code{builtin_define}, @code{builtin_define_std} and 611@code{builtin_assert}. When the front end 612calls this macro it provides a trailing semicolon, and since it has 613finished command line option processing your code can use those 614results freely. 615 616@code{builtin_assert} takes a string in the form you pass to the 617command-line option @option{-A}, such as @code{cpu=mips}, and creates 618the assertion. @code{builtin_define} takes a string in the form 619accepted by option @option{-D} and unconditionally defines the macro. 620 621@code{builtin_define_std} takes a string representing the name of an 622object-like macro. If it doesn't lie in the user's namespace, 623@code{builtin_define_std} defines it unconditionally. Otherwise, it 624defines a version with two leading underscores, and another version 625with two leading and trailing underscores, and defines the original 626only if an ISO standard was not requested on the command line. For 627example, passing @code{unix} defines @code{__unix}, @code{__unix__} 628and possibly @code{unix}; passing @code{_mips} defines @code{__mips}, 629@code{__mips__} and possibly @code{_mips}, and passing @code{_ABI64} 630defines only @code{_ABI64}. 631 632You can also test for the C dialect being compiled. The variable 633@code{c_language} is set to one of @code{clk_c}, @code{clk_cplusplus} 634or @code{clk_objective_c}. Note that if we are preprocessing 635assembler, this variable will be @code{clk_c} but the function-like 636macro @code{preprocessing_asm_p()} will return true, so you might want 637to check for that first. If you need to check for strict ANSI, the 638variable @code{flag_iso} can be used. The function-like macro 639@code{preprocessing_trad_p()} can be used to check for traditional 640preprocessing. 641@end defmac 642 643@defmac TARGET_OS_CPP_BUILTINS () 644Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional 645and is used for the target operating system instead. 646@end defmac 647 648@defmac TARGET_OBJFMT_CPP_BUILTINS () 649Similarly to @code{TARGET_CPU_CPP_BUILTINS} but this macro is optional 650and is used for the target object format. @file{elfos.h} uses this 651macro to define @code{__ELF__}, so you probably do not need to define 652it yourself. 653@end defmac 654 655@deftypevar {extern int} target_flags 656This variable is declared in @file{options.h}, which is included before 657any target-specific headers. 658@end deftypevar 659 660@deftypevr {Common Target Hook} int TARGET_DEFAULT_TARGET_FLAGS 661This variable specifies the initial value of @code{target_flags}. 662Its default setting is 0. 663@end deftypevr 664 665@cindex optional hardware or system features 666@cindex features, optional, in system conventions 667 668@deftypefn {Common Target Hook} bool TARGET_HANDLE_OPTION (struct gcc_options *@var{opts}, struct gcc_options *@var{opts_set}, const struct cl_decoded_option *@var{decoded}, location_t @var{loc}) 669This hook is called whenever the user specifies one of the 670target-specific options described by the @file{.opt} definition files 671(@pxref{Options}). It has the opportunity to do some option-specific 672processing and should return true if the option is valid. The default 673definition does nothing but return true. 674 675@var{decoded} specifies the option and its arguments. @var{opts} and 676@var{opts_set} are the @code{gcc_options} structures to be used for 677storing option state, and @var{loc} is the location at which the 678option was passed (@code{UNKNOWN_LOCATION} except for options passed 679via attributes). 680@end deftypefn 681 682@deftypefn {C Target Hook} bool TARGET_HANDLE_C_OPTION (size_t @var{code}, const char *@var{arg}, int @var{value}) 683This target hook is called whenever the user specifies one of the 684target-specific C language family options described by the @file{.opt} 685definition files(@pxref{Options}). It has the opportunity to do some 686option-specific processing and should return true if the option is 687valid. The arguments are like for @code{TARGET_HANDLE_OPTION}. The 688default definition does nothing but return false. 689 690In general, you should use @code{TARGET_HANDLE_OPTION} to handle 691options. However, if processing an option requires routines that are 692only available in the C (and related language) front ends, then you 693should use @code{TARGET_HANDLE_C_OPTION} instead. 694@end deftypefn 695 696@deftypefn {C Target Hook} tree TARGET_OBJC_CONSTRUCT_STRING_OBJECT (tree @var{string}) 697Targets may provide a string object type that can be used within and between C, C++ and their respective Objective-C dialects. A string object might, for example, embed encoding and length information. These objects are considered opaque to the compiler and handled as references. An ideal implementation makes the composition of the string object match that of the Objective-C @code{NSString} (@code{NXString} for GNUStep), allowing efficient interworking between C-only and Objective-C code. If a target implements string objects then this hook should return a reference to such an object constructed from the normal `C' string representation provided in @var{string}. At present, the hook is used by Objective-C only, to obtain a common-format string object when the target provides one. 698@end deftypefn 699 700@deftypefn {C Target Hook} void TARGET_OBJC_DECLARE_UNRESOLVED_CLASS_REFERENCE (const char *@var{classname}) 701Declare that Objective C class @var{classname} is referenced by the current TU. 702@end deftypefn 703 704@deftypefn {C Target Hook} void TARGET_OBJC_DECLARE_CLASS_DEFINITION (const char *@var{classname}) 705Declare that Objective C class @var{classname} is defined by the current TU. 706@end deftypefn 707 708@deftypefn {C Target Hook} bool TARGET_STRING_OBJECT_REF_TYPE_P (const_tree @var{stringref}) 709If a target implements string objects then this hook should return @code{true} if @var{stringref} is a valid reference to such an object. 710@end deftypefn 711 712@deftypefn {C Target Hook} void TARGET_CHECK_STRING_OBJECT_FORMAT_ARG (tree @var{format_arg}, tree @var{args_list}) 713If a target implements string objects then this hook should should provide a facility to check the function arguments in @var{args_list} against the format specifiers in @var{format_arg} where the type of @var{format_arg} is one recognized as a valid string reference type. 714@end deftypefn 715 716@deftypefn {Target Hook} void TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE (void) 717This target function is similar to the hook @code{TARGET_OPTION_OVERRIDE} 718but is called when the optimize level is changed via an attribute or 719pragma or when it is reset at the end of the code affected by the 720attribute or pragma. It is not called at the beginning of compilation 721when @code{TARGET_OPTION_OVERRIDE} is called so if you want to perform these 722actions then, you should have @code{TARGET_OPTION_OVERRIDE} call 723@code{TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE}. 724@end deftypefn 725 726@defmac C_COMMON_OVERRIDE_OPTIONS 727This is similar to the @code{TARGET_OPTION_OVERRIDE} hook 728but is only used in the C 729language frontends (C, Objective-C, C++, Objective-C++) and so can be 730used to alter option flag variables which only exist in those 731frontends. 732@end defmac 733 734@deftypevr {Common Target Hook} {const struct default_options *} TARGET_OPTION_OPTIMIZATION_TABLE 735Some machines may desire to change what optimizations are performed for 736various optimization levels. This variable, if defined, describes 737options to enable at particular sets of optimization levels. These 738options are processed once 739just after the optimization level is determined and before the remainder 740of the command options have been parsed, so may be overridden by other 741options passed explicitly. 742 743This processing is run once at program startup and when the optimization 744options are changed via @code{#pragma GCC optimize} or by using the 745@code{optimize} attribute. 746@end deftypevr 747 748@deftypefn {Common Target Hook} void TARGET_OPTION_INIT_STRUCT (struct gcc_options *@var{opts}) 749Set target-dependent initial values of fields in @var{opts}. 750@end deftypefn 751 752@deftypefn {Common Target Hook} void TARGET_OPTION_DEFAULT_PARAMS (void) 753Set target-dependent default values for @option{--param} settings, using calls to @code{set_default_param_value}. 754@end deftypefn 755 756@defmac SWITCHABLE_TARGET 757Some targets need to switch between substantially different subtargets 758during compilation. For example, the MIPS target has one subtarget for 759the traditional MIPS architecture and another for MIPS16. Source code 760can switch between these two subarchitectures using the @code{mips16} 761and @code{nomips16} attributes. 762 763Such subtargets can differ in things like the set of available 764registers, the set of available instructions, the costs of various 765operations, and so on. GCC caches a lot of this type of information 766in global variables, and recomputing them for each subtarget takes a 767significant amount of time. The compiler therefore provides a facility 768for maintaining several versions of the global variables and quickly 769switching between them; see @file{target-globals.h} for details. 770 771Define this macro to 1 if your target needs this facility. The default 772is 0. 773@end defmac 774 775@deftypefn {Target Hook} bool TARGET_FLOAT_EXCEPTIONS_ROUNDING_SUPPORTED_P (void) 776Returns true if the target supports IEEE 754 floating-point exceptions and rounding modes, false otherwise. This is intended to relate to the @code{float} and @code{double} types, but not necessarily @code{long double}. By default, returns true if the @code{adddf3} instruction pattern is available and false otherwise, on the assumption that hardware floating point supports exceptions and rounding modes but software floating point does not. 777@end deftypefn 778 779@node Per-Function Data 780@section Defining data structures for per-function information. 781@cindex per-function data 782@cindex data structures 783 784If the target needs to store information on a per-function basis, GCC 785provides a macro and a couple of variables to allow this. Note, just 786using statics to store the information is a bad idea, since GCC supports 787nested functions, so you can be halfway through encoding one function 788when another one comes along. 789 790GCC defines a data structure called @code{struct function} which 791contains all of the data specific to an individual function. This 792structure contains a field called @code{machine} whose type is 793@code{struct machine_function *}, which can be used by targets to point 794to their own specific data. 795 796If a target needs per-function specific data it should define the type 797@code{struct machine_function} and also the macro @code{INIT_EXPANDERS}. 798This macro should be used to initialize the function pointer 799@code{init_machine_status}. This pointer is explained below. 800 801One typical use of per-function, target specific data is to create an 802RTX to hold the register containing the function's return address. This 803RTX can then be used to implement the @code{__builtin_return_address} 804function, for level 0. 805 806Note---earlier implementations of GCC used a single data area to hold 807all of the per-function information. Thus when processing of a nested 808function began the old per-function data had to be pushed onto a 809stack, and when the processing was finished, it had to be popped off the 810stack. GCC used to provide function pointers called 811@code{save_machine_status} and @code{restore_machine_status} to handle 812the saving and restoring of the target specific information. Since the 813single data area approach is no longer used, these pointers are no 814longer supported. 815 816@defmac INIT_EXPANDERS 817Macro called to initialize any target specific information. This macro 818is called once per function, before generation of any RTL has begun. 819The intention of this macro is to allow the initialization of the 820function pointer @code{init_machine_status}. 821@end defmac 822 823@deftypevar {void (*)(struct function *)} init_machine_status 824If this function pointer is non-@code{NULL} it will be called once per 825function, before function compilation starts, in order to allow the 826target to perform any target specific initialization of the 827@code{struct function} structure. It is intended that this would be 828used to initialize the @code{machine} of that structure. 829 830@code{struct machine_function} structures are expected to be freed by GC@. 831Generally, any memory that they reference must be allocated by using 832GC allocation, including the structure itself. 833@end deftypevar 834 835@node Storage Layout 836@section Storage Layout 837@cindex storage layout 838 839Note that the definitions of the macros in this table which are sizes or 840alignments measured in bits do not need to be constant. They can be C 841expressions that refer to static variables, such as the @code{target_flags}. 842@xref{Run-time Target}. 843 844@defmac BITS_BIG_ENDIAN 845Define this macro to have the value 1 if the most significant bit in a 846byte has the lowest number; otherwise define it to have the value zero. 847This means that bit-field instructions count from the most significant 848bit. If the machine has no bit-field instructions, then this must still 849be defined, but it doesn't matter which value it is defined to. This 850macro need not be a constant. 851 852This macro does not affect the way structure fields are packed into 853bytes or words; that is controlled by @code{BYTES_BIG_ENDIAN}. 854@end defmac 855 856@defmac BYTES_BIG_ENDIAN 857Define this macro to have the value 1 if the most significant byte in a 858word has the lowest number. This macro need not be a constant. 859@end defmac 860 861@defmac WORDS_BIG_ENDIAN 862Define this macro to have the value 1 if, in a multiword object, the 863most significant word has the lowest number. This applies to both 864memory locations and registers; see @code{REG_WORDS_BIG_ENDIAN} if the 865order of words in memory is not the same as the order in registers. This 866macro need not be a constant. 867@end defmac 868 869@defmac REG_WORDS_BIG_ENDIAN 870On some machines, the order of words in a multiword object differs between 871registers in memory. In such a situation, define this macro to describe 872the order of words in a register. The macro @code{WORDS_BIG_ENDIAN} controls 873the order of words in memory. 874@end defmac 875 876@defmac FLOAT_WORDS_BIG_ENDIAN 877Define this macro to have the value 1 if @code{DFmode}, @code{XFmode} or 878@code{TFmode} floating point numbers are stored in memory with the word 879containing the sign bit at the lowest address; otherwise define it to 880have the value 0. This macro need not be a constant. 881 882You need not define this macro if the ordering is the same as for 883multi-word integers. 884@end defmac 885 886@defmac BITS_PER_WORD 887Number of bits in a word. If you do not define this macro, the default 888is @code{BITS_PER_UNIT * UNITS_PER_WORD}. 889@end defmac 890 891@defmac MAX_BITS_PER_WORD 892Maximum number of bits in a word. If this is undefined, the default is 893@code{BITS_PER_WORD}. Otherwise, it is the constant value that is the 894largest value that @code{BITS_PER_WORD} can have at run-time. 895@end defmac 896 897@defmac UNITS_PER_WORD 898Number of storage units in a word; normally the size of a general-purpose 899register, a power of two from 1 or 8. 900@end defmac 901 902@defmac MIN_UNITS_PER_WORD 903Minimum number of units in a word. If this is undefined, the default is 904@code{UNITS_PER_WORD}. Otherwise, it is the constant value that is the 905smallest value that @code{UNITS_PER_WORD} can have at run-time. 906@end defmac 907 908@defmac POINTER_SIZE 909Width of a pointer, in bits. You must specify a value no wider than the 910width of @code{Pmode}. If it is not equal to the width of @code{Pmode}, 911you must define @code{POINTERS_EXTEND_UNSIGNED}. If you do not specify 912a value the default is @code{BITS_PER_WORD}. 913@end defmac 914 915@defmac POINTERS_EXTEND_UNSIGNED 916A C expression that determines how pointers should be extended from 917@code{ptr_mode} to either @code{Pmode} or @code{word_mode}. It is 918greater than zero if pointers should be zero-extended, zero if they 919should be sign-extended, and negative if some other sort of conversion 920is needed. In the last case, the extension is done by the target's 921@code{ptr_extend} instruction. 922 923You need not define this macro if the @code{ptr_mode}, @code{Pmode} 924and @code{word_mode} are all the same width. 925@end defmac 926 927@defmac PROMOTE_MODE (@var{m}, @var{unsignedp}, @var{type}) 928A macro to update @var{m} and @var{unsignedp} when an object whose type 929is @var{type} and which has the specified mode and signedness is to be 930stored in a register. This macro is only called when @var{type} is a 931scalar type. 932 933On most RISC machines, which only have operations that operate on a full 934register, define this macro to set @var{m} to @code{word_mode} if 935@var{m} is an integer mode narrower than @code{BITS_PER_WORD}. In most 936cases, only integer modes should be widened because wider-precision 937floating-point operations are usually more expensive than their narrower 938counterparts. 939 940For most machines, the macro definition does not change @var{unsignedp}. 941However, some machines, have instructions that preferentially handle 942either signed or unsigned quantities of certain modes. For example, on 943the DEC Alpha, 32-bit loads from memory and 32-bit add instructions 944sign-extend the result to 64 bits. On such machines, set 945@var{unsignedp} according to which kind of extension is more efficient. 946 947Do not define this macro if it would never modify @var{m}. 948@end defmac 949 950@deftypefn {Target Hook} {enum flt_eval_method} TARGET_C_EXCESS_PRECISION (enum excess_precision_type @var{type}) 951Return a value, with the same meaning as the C99 macro @code{FLT_EVAL_METHOD} that describes which excess precision should be applied. @var{type} is either @code{EXCESS_PRECISION_TYPE_IMPLICIT}, @code{EXCESS_PRECISION_TYPE_FAST}, or @code{EXCESS_PRECISION_TYPE_STANDARD}. For @code{EXCESS_PRECISION_TYPE_IMPLICIT}, the target should return which precision and range operations will be implictly evaluated in regardless of the excess precision explicitly added. For @code{EXCESS_PRECISION_TYPE_STANDARD} and @code{EXCESS_PRECISION_TYPE_FAST}, the target should return the explicit excess precision that should be added depending on the value set for @option{-fexcess-precision=@r{[}standard@r{|}fast@r{]}}. Note that unpredictable explicit excess precision does not make sense, so a target should never return @code{FLT_EVAL_METHOD_UNPREDICTABLE} when @var{type} is @code{EXCESS_PRECISION_TYPE_STANDARD} or @code{EXCESS_PRECISION_TYPE_FAST}. 952@end deftypefn 953 954@deftypefn {Target Hook} machine_mode TARGET_PROMOTE_FUNCTION_MODE (const_tree @var{type}, machine_mode @var{mode}, int *@var{punsignedp}, const_tree @var{funtype}, int @var{for_return}) 955Like @code{PROMOTE_MODE}, but it is applied to outgoing function arguments or 956function return values. The target hook should return the new mode 957and possibly change @code{*@var{punsignedp}} if the promotion should 958change signedness. This function is called only for scalar @emph{or 959pointer} types. 960 961@var{for_return} allows to distinguish the promotion of arguments and 962return values. If it is @code{1}, a return value is being promoted and 963@code{TARGET_FUNCTION_VALUE} must perform the same promotions done here. 964If it is @code{2}, the returned mode should be that of the register in 965which an incoming parameter is copied, or the outgoing result is computed; 966then the hook should return the same mode as @code{promote_mode}, though 967the signedness may be different. 968 969@var{type} can be NULL when promoting function arguments of libcalls. 970 971The default is to not promote arguments and return values. You can 972also define the hook to @code{default_promote_function_mode_always_promote} 973if you would like to apply the same rules given by @code{PROMOTE_MODE}. 974@end deftypefn 975 976@defmac PARM_BOUNDARY 977Normal alignment required for function parameters on the stack, in 978bits. All stack parameters receive at least this much alignment 979regardless of data type. On most machines, this is the same as the 980size of an integer. 981@end defmac 982 983@defmac STACK_BOUNDARY 984Define this macro to the minimum alignment enforced by hardware for the 985stack pointer on this machine. The definition is a C expression for the 986desired alignment (measured in bits). This value is used as a default 987if @code{PREFERRED_STACK_BOUNDARY} is not defined. On most machines, 988this should be the same as @code{PARM_BOUNDARY}. 989@end defmac 990 991@defmac PREFERRED_STACK_BOUNDARY 992Define this macro if you wish to preserve a certain alignment for the 993stack pointer, greater than what the hardware enforces. The definition 994is a C expression for the desired alignment (measured in bits). This 995macro must evaluate to a value equal to or larger than 996@code{STACK_BOUNDARY}. 997@end defmac 998 999@defmac INCOMING_STACK_BOUNDARY 1000Define this macro if the incoming stack boundary may be different 1001from @code{PREFERRED_STACK_BOUNDARY}. This macro must evaluate 1002to a value equal to or larger than @code{STACK_BOUNDARY}. 1003@end defmac 1004 1005@defmac FUNCTION_BOUNDARY 1006Alignment required for a function entry point, in bits. 1007@end defmac 1008 1009@defmac BIGGEST_ALIGNMENT 1010Biggest alignment that any data type can require on this machine, in 1011bits. Note that this is not the biggest alignment that is supported, 1012just the biggest alignment that, when violated, may cause a fault. 1013@end defmac 1014 1015@deftypevr {Target Hook} HOST_WIDE_INT TARGET_ABSOLUTE_BIGGEST_ALIGNMENT 1016If defined, this target hook specifies the absolute biggest alignment 1017that a type or variable can have on this machine, otherwise, 1018@code{BIGGEST_ALIGNMENT} is used. 1019@end deftypevr 1020 1021@defmac MALLOC_ABI_ALIGNMENT 1022Alignment, in bits, a C conformant malloc implementation has to 1023provide. If not defined, the default value is @code{BITS_PER_WORD}. 1024@end defmac 1025 1026@defmac ATTRIBUTE_ALIGNED_VALUE 1027Alignment used by the @code{__attribute__ ((aligned))} construct. If 1028not defined, the default value is @code{BIGGEST_ALIGNMENT}. 1029@end defmac 1030 1031@defmac MINIMUM_ATOMIC_ALIGNMENT 1032If defined, the smallest alignment, in bits, that can be given to an 1033object that can be referenced in one operation, without disturbing any 1034nearby object. Normally, this is @code{BITS_PER_UNIT}, but may be larger 1035on machines that don't have byte or half-word store operations. 1036@end defmac 1037 1038@defmac BIGGEST_FIELD_ALIGNMENT 1039Biggest alignment that any structure or union field can require on this 1040machine, in bits. If defined, this overrides @code{BIGGEST_ALIGNMENT} for 1041structure and union fields only, unless the field alignment has been set 1042by the @code{__attribute__ ((aligned (@var{n})))} construct. 1043@end defmac 1044 1045@defmac ADJUST_FIELD_ALIGN (@var{field}, @var{type}, @var{computed}) 1046An expression for the alignment of a structure field @var{field} of 1047type @var{type} if the alignment computed in the usual way (including 1048applying of @code{BIGGEST_ALIGNMENT} and @code{BIGGEST_FIELD_ALIGNMENT} to the 1049alignment) is @var{computed}. It overrides alignment only if the 1050field alignment has not been set by the 1051@code{__attribute__ ((aligned (@var{n})))} construct. Note that @var{field} 1052may be @code{NULL_TREE} in case we just query for the minimum alignment 1053of a field of type @var{type} in structure context. 1054@end defmac 1055 1056@defmac MAX_STACK_ALIGNMENT 1057Biggest stack alignment guaranteed by the backend. Use this macro 1058to specify the maximum alignment of a variable on stack. 1059 1060If not defined, the default value is @code{STACK_BOUNDARY}. 1061 1062@c FIXME: The default should be @code{PREFERRED_STACK_BOUNDARY}. 1063@c But the fix for PR 32893 indicates that we can only guarantee 1064@c maximum stack alignment on stack up to @code{STACK_BOUNDARY}, not 1065@c @code{PREFERRED_STACK_BOUNDARY}, if stack alignment isn't supported. 1066@end defmac 1067 1068@defmac MAX_OFILE_ALIGNMENT 1069Biggest alignment supported by the object file format of this machine. 1070Use this macro to limit the alignment which can be specified using the 1071@code{__attribute__ ((aligned (@var{n})))} construct. If not defined, 1072the default value is @code{BIGGEST_ALIGNMENT}. 1073 1074On systems that use ELF, the default (in @file{config/elfos.h}) is 1075the largest supported 32-bit ELF section alignment representable on 1076a 32-bit host e.g. @samp{(((uint64_t) 1 << 28) * 8)}. 1077On 32-bit ELF the largest supported section alignment in bits is 1078@samp{(0x80000000 * 8)}, but this is not representable on 32-bit hosts. 1079@end defmac 1080 1081@deftypefn {Target Hook} HOST_WIDE_INT TARGET_STATIC_RTX_ALIGNMENT (machine_mode @var{mode}) 1082This hook returns the preferred alignment in bits for a 1083statically-allocated rtx, such as a constant pool entry. @var{mode} 1084is the mode of the rtx. The default implementation returns 1085@samp{GET_MODE_ALIGNMENT (@var{mode})}. 1086@end deftypefn 1087 1088@defmac DATA_ALIGNMENT (@var{type}, @var{basic-align}) 1089If defined, a C expression to compute the alignment for a variable in 1090the static store. @var{type} is the data type, and @var{basic-align} is 1091the alignment that the object would ordinarily have. The value of this 1092macro is used instead of that alignment to align the object. 1093 1094If this macro is not defined, then @var{basic-align} is used. 1095 1096@findex strcpy 1097One use of this macro is to increase alignment of medium-size data to 1098make it all fit in fewer cache lines. Another is to cause character 1099arrays to be word-aligned so that @code{strcpy} calls that copy 1100constants to character arrays can be done inline. 1101@end defmac 1102 1103@defmac DATA_ABI_ALIGNMENT (@var{type}, @var{basic-align}) 1104Similar to @code{DATA_ALIGNMENT}, but for the cases where the ABI mandates 1105some alignment increase, instead of optimization only purposes. E.g.@ 1106AMD x86-64 psABI says that variables with array type larger than 15 bytes 1107must be aligned to 16 byte boundaries. 1108 1109If this macro is not defined, then @var{basic-align} is used. 1110@end defmac 1111 1112@deftypefn {Target Hook} HOST_WIDE_INT TARGET_CONSTANT_ALIGNMENT (const_tree @var{constant}, HOST_WIDE_INT @var{basic_align}) 1113This hook returns the alignment in bits of a constant that is being 1114placed in memory. @var{constant} is the constant and @var{basic_align} 1115is the alignment that the object would ordinarily have. 1116 1117The default definition just returns @var{basic_align}. 1118 1119The typical use of this hook is to increase alignment for string 1120constants to be word aligned so that @code{strcpy} calls that copy 1121constants can be done inline. The function 1122@code{constant_alignment_word_strings} provides such a definition. 1123@end deftypefn 1124 1125@defmac LOCAL_ALIGNMENT (@var{type}, @var{basic-align}) 1126If defined, a C expression to compute the alignment for a variable in 1127the local store. @var{type} is the data type, and @var{basic-align} is 1128the alignment that the object would ordinarily have. The value of this 1129macro is used instead of that alignment to align the object. 1130 1131If this macro is not defined, then @var{basic-align} is used. 1132 1133One use of this macro is to increase alignment of medium-size data to 1134make it all fit in fewer cache lines. 1135 1136If the value of this macro has a type, it should be an unsigned type. 1137@end defmac 1138 1139@deftypefn {Target Hook} HOST_WIDE_INT TARGET_VECTOR_ALIGNMENT (const_tree @var{type}) 1140This hook can be used to define the alignment for a vector of type 1141@var{type}, in order to comply with a platform ABI. The default is to 1142require natural alignment for vector types. The alignment returned by 1143this hook must be a power-of-two multiple of the default alignment of 1144the vector element type. 1145@end deftypefn 1146 1147@defmac STACK_SLOT_ALIGNMENT (@var{type}, @var{mode}, @var{basic-align}) 1148If defined, a C expression to compute the alignment for stack slot. 1149@var{type} is the data type, @var{mode} is the widest mode available, 1150and @var{basic-align} is the alignment that the slot would ordinarily 1151have. The value of this macro is used instead of that alignment to 1152align the slot. 1153 1154If this macro is not defined, then @var{basic-align} is used when 1155@var{type} is @code{NULL}. Otherwise, @code{LOCAL_ALIGNMENT} will 1156be used. 1157 1158This macro is to set alignment of stack slot to the maximum alignment 1159of all possible modes which the slot may have. 1160 1161If the value of this macro has a type, it should be an unsigned type. 1162@end defmac 1163 1164@defmac LOCAL_DECL_ALIGNMENT (@var{decl}) 1165If defined, a C expression to compute the alignment for a local 1166variable @var{decl}. 1167 1168If this macro is not defined, then 1169@code{LOCAL_ALIGNMENT (TREE_TYPE (@var{decl}), DECL_ALIGN (@var{decl}))} 1170is used. 1171 1172One use of this macro is to increase alignment of medium-size data to 1173make it all fit in fewer cache lines. 1174 1175If the value of this macro has a type, it should be an unsigned type. 1176@end defmac 1177 1178@defmac MINIMUM_ALIGNMENT (@var{exp}, @var{mode}, @var{align}) 1179If defined, a C expression to compute the minimum required alignment 1180for dynamic stack realignment purposes for @var{exp} (a type or decl), 1181@var{mode}, assuming normal alignment @var{align}. 1182 1183If this macro is not defined, then @var{align} will be used. 1184@end defmac 1185 1186@defmac EMPTY_FIELD_BOUNDARY 1187Alignment in bits to be given to a structure bit-field that follows an 1188empty field such as @code{int : 0;}. 1189 1190If @code{PCC_BITFIELD_TYPE_MATTERS} is true, it overrides this macro. 1191@end defmac 1192 1193@defmac STRUCTURE_SIZE_BOUNDARY 1194Number of bits which any structure or union's size must be a multiple of. 1195Each structure or union's size is rounded up to a multiple of this. 1196 1197If you do not define this macro, the default is the same as 1198@code{BITS_PER_UNIT}. 1199@end defmac 1200 1201@defmac STRICT_ALIGNMENT 1202Define this macro to be the value 1 if instructions will fail to work 1203if given data not on the nominal alignment. If instructions will merely 1204go slower in that case, define this macro as 0. 1205@end defmac 1206 1207@defmac PCC_BITFIELD_TYPE_MATTERS 1208Define this if you wish to imitate the way many other C compilers handle 1209alignment of bit-fields and the structures that contain them. 1210 1211The behavior is that the type written for a named bit-field (@code{int}, 1212@code{short}, or other integer type) imposes an alignment for the entire 1213structure, as if the structure really did contain an ordinary field of 1214that type. In addition, the bit-field is placed within the structure so 1215that it would fit within such a field, not crossing a boundary for it. 1216 1217Thus, on most machines, a named bit-field whose type is written as 1218@code{int} would not cross a four-byte boundary, and would force 1219four-byte alignment for the whole structure. (The alignment used may 1220not be four bytes; it is controlled by the other alignment parameters.) 1221 1222An unnamed bit-field will not affect the alignment of the containing 1223structure. 1224 1225If the macro is defined, its definition should be a C expression; 1226a nonzero value for the expression enables this behavior. 1227 1228Note that if this macro is not defined, or its value is zero, some 1229bit-fields may cross more than one alignment boundary. The compiler can 1230support such references if there are @samp{insv}, @samp{extv}, and 1231@samp{extzv} insns that can directly reference memory. 1232 1233The other known way of making bit-fields work is to define 1234@code{STRUCTURE_SIZE_BOUNDARY} as large as @code{BIGGEST_ALIGNMENT}. 1235Then every structure can be accessed with fullwords. 1236 1237Unless the machine has bit-field instructions or you define 1238@code{STRUCTURE_SIZE_BOUNDARY} that way, you must define 1239@code{PCC_BITFIELD_TYPE_MATTERS} to have a nonzero value. 1240 1241If your aim is to make GCC use the same conventions for laying out 1242bit-fields as are used by another compiler, here is how to investigate 1243what the other compiler does. Compile and run this program: 1244 1245@smallexample 1246struct foo1 1247@{ 1248 char x; 1249 char :0; 1250 char y; 1251@}; 1252 1253struct foo2 1254@{ 1255 char x; 1256 int :0; 1257 char y; 1258@}; 1259 1260main () 1261@{ 1262 printf ("Size of foo1 is %d\n", 1263 sizeof (struct foo1)); 1264 printf ("Size of foo2 is %d\n", 1265 sizeof (struct foo2)); 1266 exit (0); 1267@} 1268@end smallexample 1269 1270If this prints 2 and 5, then the compiler's behavior is what you would 1271get from @code{PCC_BITFIELD_TYPE_MATTERS}. 1272@end defmac 1273 1274@defmac BITFIELD_NBYTES_LIMITED 1275Like @code{PCC_BITFIELD_TYPE_MATTERS} except that its effect is limited 1276to aligning a bit-field within the structure. 1277@end defmac 1278 1279@deftypefn {Target Hook} bool TARGET_ALIGN_ANON_BITFIELD (void) 1280When @code{PCC_BITFIELD_TYPE_MATTERS} is true this hook will determine 1281whether unnamed bitfields affect the alignment of the containing 1282structure. The hook should return true if the structure should inherit 1283the alignment requirements of an unnamed bitfield's type. 1284@end deftypefn 1285 1286@deftypefn {Target Hook} bool TARGET_NARROW_VOLATILE_BITFIELD (void) 1287This target hook should return @code{true} if accesses to volatile bitfields 1288should use the narrowest mode possible. It should return @code{false} if 1289these accesses should use the bitfield container type. 1290 1291The default is @code{false}. 1292@end deftypefn 1293 1294@deftypefn {Target Hook} bool TARGET_MEMBER_TYPE_FORCES_BLK (const_tree @var{field}, machine_mode @var{mode}) 1295Return true if a structure, union or array containing @var{field} should 1296be accessed using @code{BLKMODE}. 1297 1298If @var{field} is the only field in the structure, @var{mode} is its 1299mode, otherwise @var{mode} is VOIDmode. @var{mode} is provided in the 1300case where structures of one field would require the structure's mode to 1301retain the field's mode. 1302 1303Normally, this is not needed. 1304@end deftypefn 1305 1306@defmac ROUND_TYPE_ALIGN (@var{type}, @var{computed}, @var{specified}) 1307Define this macro as an expression for the alignment of a type (given 1308by @var{type} as a tree node) if the alignment computed in the usual 1309way is @var{computed} and the alignment explicitly specified was 1310@var{specified}. 1311 1312The default is to use @var{specified} if it is larger; otherwise, use 1313the smaller of @var{computed} and @code{BIGGEST_ALIGNMENT} 1314@end defmac 1315 1316@defmac MAX_FIXED_MODE_SIZE 1317An integer expression for the size in bits of the largest integer 1318machine mode that should actually be used. All integer machine modes of 1319this size or smaller can be used for structures and unions with the 1320appropriate sizes. If this macro is undefined, @code{GET_MODE_BITSIZE 1321(DImode)} is assumed. 1322@end defmac 1323 1324@defmac STACK_SAVEAREA_MODE (@var{save_level}) 1325If defined, an expression of type @code{machine_mode} that 1326specifies the mode of the save area operand of a 1327@code{save_stack_@var{level}} named pattern (@pxref{Standard Names}). 1328@var{save_level} is one of @code{SAVE_BLOCK}, @code{SAVE_FUNCTION}, or 1329@code{SAVE_NONLOCAL} and selects which of the three named patterns is 1330having its mode specified. 1331 1332You need not define this macro if it always returns @code{Pmode}. You 1333would most commonly define this macro if the 1334@code{save_stack_@var{level}} patterns need to support both a 32- and a 133564-bit mode. 1336@end defmac 1337 1338@defmac STACK_SIZE_MODE 1339If defined, an expression of type @code{machine_mode} that 1340specifies the mode of the size increment operand of an 1341@code{allocate_stack} named pattern (@pxref{Standard Names}). 1342 1343You need not define this macro if it always returns @code{word_mode}. 1344You would most commonly define this macro if the @code{allocate_stack} 1345pattern needs to support both a 32- and a 64-bit mode. 1346@end defmac 1347 1348@deftypefn {Target Hook} scalar_int_mode TARGET_LIBGCC_CMP_RETURN_MODE (void) 1349This target hook should return the mode to be used for the return value 1350of compare instructions expanded to libgcc calls. If not defined 1351@code{word_mode} is returned which is the right choice for a majority of 1352targets. 1353@end deftypefn 1354 1355@deftypefn {Target Hook} scalar_int_mode TARGET_LIBGCC_SHIFT_COUNT_MODE (void) 1356This target hook should return the mode to be used for the shift count operand 1357of shift instructions expanded to libgcc calls. If not defined 1358@code{word_mode} is returned which is the right choice for a majority of 1359targets. 1360@end deftypefn 1361 1362@deftypefn {Target Hook} scalar_int_mode TARGET_UNWIND_WORD_MODE (void) 1363Return machine mode to be used for @code{_Unwind_Word} type. 1364The default is to use @code{word_mode}. 1365@end deftypefn 1366 1367@deftypefn {Target Hook} bool TARGET_MS_BITFIELD_LAYOUT_P (const_tree @var{record_type}) 1368This target hook returns @code{true} if bit-fields in the given 1369@var{record_type} are to be laid out following the rules of Microsoft 1370Visual C/C++, namely: (i) a bit-field won't share the same storage 1371unit with the previous bit-field if their underlying types have 1372different sizes, and the bit-field will be aligned to the highest 1373alignment of the underlying types of itself and of the previous 1374bit-field; (ii) a zero-sized bit-field will affect the alignment of 1375the whole enclosing structure, even if it is unnamed; except that 1376(iii) a zero-sized bit-field will be disregarded unless it follows 1377another bit-field of nonzero size. If this hook returns @code{true}, 1378other macros that control bit-field layout are ignored. 1379 1380When a bit-field is inserted into a packed record, the whole size 1381of the underlying type is used by one or more same-size adjacent 1382bit-fields (that is, if its long:3, 32 bits is used in the record, 1383and any additional adjacent long bit-fields are packed into the same 1384chunk of 32 bits. However, if the size changes, a new field of that 1385size is allocated). In an unpacked record, this is the same as using 1386alignment, but not equivalent when packing. 1387 1388If both MS bit-fields and @samp{__attribute__((packed))} are used, 1389the latter will take precedence. If @samp{__attribute__((packed))} is 1390used on a single field when MS bit-fields are in use, it will take 1391precedence for that field, but the alignment of the rest of the structure 1392may affect its placement. 1393@end deftypefn 1394 1395@deftypefn {Target Hook} bool TARGET_DECIMAL_FLOAT_SUPPORTED_P (void) 1396Returns true if the target supports decimal floating point. 1397@end deftypefn 1398 1399@deftypefn {Target Hook} bool TARGET_FIXED_POINT_SUPPORTED_P (void) 1400Returns true if the target supports fixed-point arithmetic. 1401@end deftypefn 1402 1403@deftypefn {Target Hook} void TARGET_EXPAND_TO_RTL_HOOK (void) 1404This hook is called just before expansion into rtl, allowing the target 1405to perform additional initializations or analysis before the expansion. 1406For example, the rs6000 port uses it to allocate a scratch stack slot 1407for use in copying SDmode values between memory and floating point 1408registers whenever the function being expanded has any SDmode 1409usage. 1410@end deftypefn 1411 1412@deftypefn {Target Hook} void TARGET_INSTANTIATE_DECLS (void) 1413This hook allows the backend to perform additional instantiations on rtl 1414that are not actually in any insns yet, but will be later. 1415@end deftypefn 1416 1417@deftypefn {Target Hook} {const char *} TARGET_MANGLE_TYPE (const_tree @var{type}) 1418If your target defines any fundamental types, or any types your target 1419uses should be mangled differently from the default, define this hook 1420to return the appropriate encoding for these types as part of a C++ 1421mangled name. The @var{type} argument is the tree structure representing 1422the type to be mangled. The hook may be applied to trees which are 1423not target-specific fundamental types; it should return @code{NULL} 1424for all such types, as well as arguments it does not recognize. If the 1425return value is not @code{NULL}, it must point to a statically-allocated 1426string constant. 1427 1428Target-specific fundamental types might be new fundamental types or 1429qualified versions of ordinary fundamental types. Encode new 1430fundamental types as @samp{@w{u @var{n} @var{name}}}, where @var{name} 1431is the name used for the type in source code, and @var{n} is the 1432length of @var{name} in decimal. Encode qualified versions of 1433ordinary types as @samp{@w{U @var{n} @var{name} @var{code}}}, where 1434@var{name} is the name used for the type qualifier in source code, 1435@var{n} is the length of @var{name} as above, and @var{code} is the 1436code used to represent the unqualified version of this type. (See 1437@code{write_builtin_type} in @file{cp/mangle.c} for the list of 1438codes.) In both cases the spaces are for clarity; do not include any 1439spaces in your string. 1440 1441This hook is applied to types prior to typedef resolution. If the mangled 1442name for a particular type depends only on that type's main variant, you 1443can perform typedef resolution yourself using @code{TYPE_MAIN_VARIANT} 1444before mangling. 1445 1446The default version of this hook always returns @code{NULL}, which is 1447appropriate for a target that does not define any new fundamental 1448types. 1449@end deftypefn 1450 1451@node Type Layout 1452@section Layout of Source Language Data Types 1453 1454These macros define the sizes and other characteristics of the standard 1455basic data types used in programs being compiled. Unlike the macros in 1456the previous section, these apply to specific features of C and related 1457languages, rather than to fundamental aspects of storage layout. 1458 1459@defmac INT_TYPE_SIZE 1460A C expression for the size in bits of the type @code{int} on the 1461target machine. If you don't define this, the default is one word. 1462@end defmac 1463 1464@defmac SHORT_TYPE_SIZE 1465A C expression for the size in bits of the type @code{short} on the 1466target machine. If you don't define this, the default is half a word. 1467(If this would be less than one storage unit, it is rounded up to one 1468unit.) 1469@end defmac 1470 1471@defmac LONG_TYPE_SIZE 1472A C expression for the size in bits of the type @code{long} on the 1473target machine. If you don't define this, the default is one word. 1474@end defmac 1475 1476@defmac ADA_LONG_TYPE_SIZE 1477On some machines, the size used for the Ada equivalent of the type 1478@code{long} by a native Ada compiler differs from that used by C@. In 1479that situation, define this macro to be a C expression to be used for 1480the size of that type. If you don't define this, the default is the 1481value of @code{LONG_TYPE_SIZE}. 1482@end defmac 1483 1484@defmac LONG_LONG_TYPE_SIZE 1485A C expression for the size in bits of the type @code{long long} on the 1486target machine. If you don't define this, the default is two 1487words. If you want to support GNU Ada on your machine, the value of this 1488macro must be at least 64. 1489@end defmac 1490 1491@defmac CHAR_TYPE_SIZE 1492A C expression for the size in bits of the type @code{char} on the 1493target machine. If you don't define this, the default is 1494@code{BITS_PER_UNIT}. 1495@end defmac 1496 1497@defmac BOOL_TYPE_SIZE 1498A C expression for the size in bits of the C++ type @code{bool} and 1499C99 type @code{_Bool} on the target machine. If you don't define 1500this, and you probably shouldn't, the default is @code{CHAR_TYPE_SIZE}. 1501@end defmac 1502 1503@defmac FLOAT_TYPE_SIZE 1504A C expression for the size in bits of the type @code{float} on the 1505target machine. If you don't define this, the default is one word. 1506@end defmac 1507 1508@defmac DOUBLE_TYPE_SIZE 1509A C expression for the size in bits of the type @code{double} on the 1510target machine. If you don't define this, the default is two 1511words. 1512@end defmac 1513 1514@defmac LONG_DOUBLE_TYPE_SIZE 1515A C expression for the size in bits of the type @code{long double} on 1516the target machine. If you don't define this, the default is two 1517words. 1518@end defmac 1519 1520@defmac SHORT_FRACT_TYPE_SIZE 1521A C expression for the size in bits of the type @code{short _Fract} on 1522the target machine. If you don't define this, the default is 1523@code{BITS_PER_UNIT}. 1524@end defmac 1525 1526@defmac FRACT_TYPE_SIZE 1527A C expression for the size in bits of the type @code{_Fract} on 1528the target machine. If you don't define this, the default is 1529@code{BITS_PER_UNIT * 2}. 1530@end defmac 1531 1532@defmac LONG_FRACT_TYPE_SIZE 1533A C expression for the size in bits of the type @code{long _Fract} on 1534the target machine. If you don't define this, the default is 1535@code{BITS_PER_UNIT * 4}. 1536@end defmac 1537 1538@defmac LONG_LONG_FRACT_TYPE_SIZE 1539A C expression for the size in bits of the type @code{long long _Fract} on 1540the target machine. If you don't define this, the default is 1541@code{BITS_PER_UNIT * 8}. 1542@end defmac 1543 1544@defmac SHORT_ACCUM_TYPE_SIZE 1545A C expression for the size in bits of the type @code{short _Accum} on 1546the target machine. If you don't define this, the default is 1547@code{BITS_PER_UNIT * 2}. 1548@end defmac 1549 1550@defmac ACCUM_TYPE_SIZE 1551A C expression for the size in bits of the type @code{_Accum} on 1552the target machine. If you don't define this, the default is 1553@code{BITS_PER_UNIT * 4}. 1554@end defmac 1555 1556@defmac LONG_ACCUM_TYPE_SIZE 1557A C expression for the size in bits of the type @code{long _Accum} on 1558the target machine. If you don't define this, the default is 1559@code{BITS_PER_UNIT * 8}. 1560@end defmac 1561 1562@defmac LONG_LONG_ACCUM_TYPE_SIZE 1563A C expression for the size in bits of the type @code{long long _Accum} on 1564the target machine. If you don't define this, the default is 1565@code{BITS_PER_UNIT * 16}. 1566@end defmac 1567 1568@defmac LIBGCC2_GNU_PREFIX 1569This macro corresponds to the @code{TARGET_LIBFUNC_GNU_PREFIX} target 1570hook and should be defined if that hook is overriden to be true. It 1571causes function names in libgcc to be changed to use a @code{__gnu_} 1572prefix for their name rather than the default @code{__}. A port which 1573uses this macro should also arrange to use @file{t-gnu-prefix} in 1574the libgcc @file{config.host}. 1575@end defmac 1576 1577@defmac WIDEST_HARDWARE_FP_SIZE 1578A C expression for the size in bits of the widest floating-point format 1579supported by the hardware. If you define this macro, you must specify a 1580value less than or equal to the value of @code{LONG_DOUBLE_TYPE_SIZE}. 1581If you do not define this macro, the value of @code{LONG_DOUBLE_TYPE_SIZE} 1582is the default. 1583@end defmac 1584 1585@defmac DEFAULT_SIGNED_CHAR 1586An expression whose value is 1 or 0, according to whether the type 1587@code{char} should be signed or unsigned by default. The user can 1588always override this default with the options @option{-fsigned-char} 1589and @option{-funsigned-char}. 1590@end defmac 1591 1592@deftypefn {Target Hook} bool TARGET_DEFAULT_SHORT_ENUMS (void) 1593This target hook should return true if the compiler should give an 1594@code{enum} type only as many bytes as it takes to represent the range 1595of possible values of that type. It should return false if all 1596@code{enum} types should be allocated like @code{int}. 1597 1598The default is to return false. 1599@end deftypefn 1600 1601@defmac SIZE_TYPE 1602A C expression for a string describing the name of the data type to use 1603for size values. The typedef name @code{size_t} is defined using the 1604contents of the string. 1605 1606The string can contain more than one keyword. If so, separate them with 1607spaces, and write first any length keyword, then @code{unsigned} if 1608appropriate, and finally @code{int}. The string must exactly match one 1609of the data type names defined in the function 1610@code{c_common_nodes_and_builtins} in the file @file{c-family/c-common.c}. 1611You may not omit @code{int} or change the order---that would cause the 1612compiler to crash on startup. 1613 1614If you don't define this macro, the default is @code{"long unsigned 1615int"}. 1616@end defmac 1617 1618@defmac SIZETYPE 1619GCC defines internal types (@code{sizetype}, @code{ssizetype}, 1620@code{bitsizetype} and @code{sbitsizetype}) for expressions 1621dealing with size. This macro is a C expression for a string describing 1622the name of the data type from which the precision of @code{sizetype} 1623is extracted. 1624 1625The string has the same restrictions as @code{SIZE_TYPE} string. 1626 1627If you don't define this macro, the default is @code{SIZE_TYPE}. 1628@end defmac 1629 1630@defmac PTRDIFF_TYPE 1631A C expression for a string describing the name of the data type to use 1632for the result of subtracting two pointers. The typedef name 1633@code{ptrdiff_t} is defined using the contents of the string. See 1634@code{SIZE_TYPE} above for more information. 1635 1636If you don't define this macro, the default is @code{"long int"}. 1637@end defmac 1638 1639@defmac WCHAR_TYPE 1640A C expression for a string describing the name of the data type to use 1641for wide characters. The typedef name @code{wchar_t} is defined using 1642the contents of the string. See @code{SIZE_TYPE} above for more 1643information. 1644 1645If you don't define this macro, the default is @code{"int"}. 1646@end defmac 1647 1648@defmac WCHAR_TYPE_SIZE 1649A C expression for the size in bits of the data type for wide 1650characters. This is used in @code{cpp}, which cannot make use of 1651@code{WCHAR_TYPE}. 1652@end defmac 1653 1654@defmac WINT_TYPE 1655A C expression for a string describing the name of the data type to 1656use for wide characters passed to @code{printf} and returned from 1657@code{getwc}. The typedef name @code{wint_t} is defined using the 1658contents of the string. See @code{SIZE_TYPE} above for more 1659information. 1660 1661If you don't define this macro, the default is @code{"unsigned int"}. 1662@end defmac 1663 1664@defmac INTMAX_TYPE 1665A C expression for a string describing the name of the data type that 1666can represent any value of any standard or extended signed integer type. 1667The typedef name @code{intmax_t} is defined using the contents of the 1668string. See @code{SIZE_TYPE} above for more information. 1669 1670If you don't define this macro, the default is the first of 1671@code{"int"}, @code{"long int"}, or @code{"long long int"} that has as 1672much precision as @code{long long int}. 1673@end defmac 1674 1675@defmac UINTMAX_TYPE 1676A C expression for a string describing the name of the data type that 1677can represent any value of any standard or extended unsigned integer 1678type. The typedef name @code{uintmax_t} is defined using the contents 1679of the string. See @code{SIZE_TYPE} above for more information. 1680 1681If you don't define this macro, the default is the first of 1682@code{"unsigned int"}, @code{"long unsigned int"}, or @code{"long long 1683unsigned int"} that has as much precision as @code{long long unsigned 1684int}. 1685@end defmac 1686 1687@defmac SIG_ATOMIC_TYPE 1688@defmacx INT8_TYPE 1689@defmacx INT16_TYPE 1690@defmacx INT32_TYPE 1691@defmacx INT64_TYPE 1692@defmacx UINT8_TYPE 1693@defmacx UINT16_TYPE 1694@defmacx UINT32_TYPE 1695@defmacx UINT64_TYPE 1696@defmacx INT_LEAST8_TYPE 1697@defmacx INT_LEAST16_TYPE 1698@defmacx INT_LEAST32_TYPE 1699@defmacx INT_LEAST64_TYPE 1700@defmacx UINT_LEAST8_TYPE 1701@defmacx UINT_LEAST16_TYPE 1702@defmacx UINT_LEAST32_TYPE 1703@defmacx UINT_LEAST64_TYPE 1704@defmacx INT_FAST8_TYPE 1705@defmacx INT_FAST16_TYPE 1706@defmacx INT_FAST32_TYPE 1707@defmacx INT_FAST64_TYPE 1708@defmacx UINT_FAST8_TYPE 1709@defmacx UINT_FAST16_TYPE 1710@defmacx UINT_FAST32_TYPE 1711@defmacx UINT_FAST64_TYPE 1712@defmacx INTPTR_TYPE 1713@defmacx UINTPTR_TYPE 1714C expressions for the standard types @code{sig_atomic_t}, 1715@code{int8_t}, @code{int16_t}, @code{int32_t}, @code{int64_t}, 1716@code{uint8_t}, @code{uint16_t}, @code{uint32_t}, @code{uint64_t}, 1717@code{int_least8_t}, @code{int_least16_t}, @code{int_least32_t}, 1718@code{int_least64_t}, @code{uint_least8_t}, @code{uint_least16_t}, 1719@code{uint_least32_t}, @code{uint_least64_t}, @code{int_fast8_t}, 1720@code{int_fast16_t}, @code{int_fast32_t}, @code{int_fast64_t}, 1721@code{uint_fast8_t}, @code{uint_fast16_t}, @code{uint_fast32_t}, 1722@code{uint_fast64_t}, @code{intptr_t}, and @code{uintptr_t}. See 1723@code{SIZE_TYPE} above for more information. 1724 1725If any of these macros evaluates to a null pointer, the corresponding 1726type is not supported; if GCC is configured to provide 1727@code{<stdint.h>} in such a case, the header provided may not conform 1728to C99, depending on the type in question. The defaults for all of 1729these macros are null pointers. 1730@end defmac 1731 1732@defmac TARGET_PTRMEMFUNC_VBIT_LOCATION 1733The C++ compiler represents a pointer-to-member-function with a struct 1734that looks like: 1735 1736@smallexample 1737 struct @{ 1738 union @{ 1739 void (*fn)(); 1740 ptrdiff_t vtable_index; 1741 @}; 1742 ptrdiff_t delta; 1743 @}; 1744@end smallexample 1745 1746@noindent 1747The C++ compiler must use one bit to indicate whether the function that 1748will be called through a pointer-to-member-function is virtual. 1749Normally, we assume that the low-order bit of a function pointer must 1750always be zero. Then, by ensuring that the vtable_index is odd, we can 1751distinguish which variant of the union is in use. But, on some 1752platforms function pointers can be odd, and so this doesn't work. In 1753that case, we use the low-order bit of the @code{delta} field, and shift 1754the remainder of the @code{delta} field to the left. 1755 1756GCC will automatically make the right selection about where to store 1757this bit using the @code{FUNCTION_BOUNDARY} setting for your platform. 1758However, some platforms such as ARM/Thumb have @code{FUNCTION_BOUNDARY} 1759set such that functions always start at even addresses, but the lowest 1760bit of pointers to functions indicate whether the function at that 1761address is in ARM or Thumb mode. If this is the case of your 1762architecture, you should define this macro to 1763@code{ptrmemfunc_vbit_in_delta}. 1764 1765In general, you should not have to define this macro. On architectures 1766in which function addresses are always even, according to 1767@code{FUNCTION_BOUNDARY}, GCC will automatically define this macro to 1768@code{ptrmemfunc_vbit_in_pfn}. 1769@end defmac 1770 1771@defmac TARGET_VTABLE_USES_DESCRIPTORS 1772Normally, the C++ compiler uses function pointers in vtables. This 1773macro allows the target to change to use ``function descriptors'' 1774instead. Function descriptors are found on targets for whom a 1775function pointer is actually a small data structure. Normally the 1776data structure consists of the actual code address plus a data 1777pointer to which the function's data is relative. 1778 1779If vtables are used, the value of this macro should be the number 1780of words that the function descriptor occupies. 1781@end defmac 1782 1783@defmac TARGET_VTABLE_ENTRY_ALIGN 1784By default, the vtable entries are void pointers, the so the alignment 1785is the same as pointer alignment. The value of this macro specifies 1786the alignment of the vtable entry in bits. It should be defined only 1787when special alignment is necessary. */ 1788@end defmac 1789 1790@defmac TARGET_VTABLE_DATA_ENTRY_DISTANCE 1791There are a few non-descriptor entries in the vtable at offsets below 1792zero. If these entries must be padded (say, to preserve the alignment 1793specified by @code{TARGET_VTABLE_ENTRY_ALIGN}), set this to the number 1794of words in each data entry. 1795@end defmac 1796 1797@node Registers 1798@section Register Usage 1799@cindex register usage 1800 1801This section explains how to describe what registers the target machine 1802has, and how (in general) they can be used. 1803 1804The description of which registers a specific instruction can use is 1805done with register classes; see @ref{Register Classes}. For information 1806on using registers to access a stack frame, see @ref{Frame Registers}. 1807For passing values in registers, see @ref{Register Arguments}. 1808For returning values in registers, see @ref{Scalar Return}. 1809 1810@menu 1811* Register Basics:: Number and kinds of registers. 1812* Allocation Order:: Order in which registers are allocated. 1813* Values in Registers:: What kinds of values each reg can hold. 1814* Leaf Functions:: Renumbering registers for leaf functions. 1815* Stack Registers:: Handling a register stack such as 80387. 1816@end menu 1817 1818@node Register Basics 1819@subsection Basic Characteristics of Registers 1820 1821@c prevent bad page break with this line 1822Registers have various characteristics. 1823 1824@defmac FIRST_PSEUDO_REGISTER 1825Number of hardware registers known to the compiler. They receive 1826numbers 0 through @code{FIRST_PSEUDO_REGISTER-1}; thus, the first 1827pseudo register's number really is assigned the number 1828@code{FIRST_PSEUDO_REGISTER}. 1829@end defmac 1830 1831@defmac FIXED_REGISTERS 1832@cindex fixed register 1833An initializer that says which registers are used for fixed purposes 1834all throughout the compiled code and are therefore not available for 1835general allocation. These would include the stack pointer, the frame 1836pointer (except on machines where that can be used as a general 1837register when no frame pointer is needed), the program counter on 1838machines where that is considered one of the addressable registers, 1839and any other numbered register with a standard use. 1840 1841This information is expressed as a sequence of numbers, separated by 1842commas and surrounded by braces. The @var{n}th number is 1 if 1843register @var{n} is fixed, 0 otherwise. 1844 1845The table initialized from this macro, and the table initialized by 1846the following one, may be overridden at run time either automatically, 1847by the actions of the macro @code{CONDITIONAL_REGISTER_USAGE}, or by 1848the user with the command options @option{-ffixed-@var{reg}}, 1849@option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}}. 1850@end defmac 1851 1852@defmac CALL_USED_REGISTERS 1853@cindex call-used register 1854@cindex call-clobbered register 1855@cindex call-saved register 1856Like @code{FIXED_REGISTERS} but has 1 for each register that is 1857clobbered (in general) by function calls as well as for fixed 1858registers. This macro therefore identifies the registers that are not 1859available for general allocation of values that must live across 1860function calls. 1861 1862If a register has 0 in @code{CALL_USED_REGISTERS}, the compiler 1863automatically saves it on function entry and restores it on function 1864exit, if the register is used within the function. 1865@end defmac 1866 1867@defmac CALL_REALLY_USED_REGISTERS 1868@cindex call-used register 1869@cindex call-clobbered register 1870@cindex call-saved register 1871Like @code{CALL_USED_REGISTERS} except this macro doesn't require 1872that the entire set of @code{FIXED_REGISTERS} be included. 1873(@code{CALL_USED_REGISTERS} must be a superset of @code{FIXED_REGISTERS}). 1874This macro is optional. If not specified, it defaults to the value 1875of @code{CALL_USED_REGISTERS}. 1876@end defmac 1877 1878@cindex call-used register 1879@cindex call-clobbered register 1880@cindex call-saved register 1881@deftypefn {Target Hook} bool TARGET_HARD_REGNO_CALL_PART_CLOBBERED (unsigned int @var{regno}, machine_mode @var{mode}) 1882This hook should return true if @var{regno} is partly call-saved and 1883partly call-clobbered, and if a value of mode @var{mode} would be partly 1884clobbered by a call. For example, if the low 32 bits of @var{regno} are 1885preserved across a call but higher bits are clobbered, this hook should 1886return true for a 64-bit mode but false for a 32-bit mode. 1887 1888The default implementation returns false, which is correct 1889for targets that don't have partly call-clobbered registers. 1890@end deftypefn 1891 1892@findex fixed_regs 1893@findex call_used_regs 1894@findex global_regs 1895@findex reg_names 1896@findex reg_class_contents 1897@deftypefn {Target Hook} void TARGET_CONDITIONAL_REGISTER_USAGE (void) 1898This hook may conditionally modify five variables 1899@code{fixed_regs}, @code{call_used_regs}, @code{global_regs}, 1900@code{reg_names}, and @code{reg_class_contents}, to take into account 1901any dependence of these register sets on target flags. The first three 1902of these are of type @code{char []} (interpreted as boolean vectors). 1903@code{global_regs} is a @code{const char *[]}, and 1904@code{reg_class_contents} is a @code{HARD_REG_SET}. Before the macro is 1905called, @code{fixed_regs}, @code{call_used_regs}, 1906@code{reg_class_contents}, and @code{reg_names} have been initialized 1907from @code{FIXED_REGISTERS}, @code{CALL_USED_REGISTERS}, 1908@code{REG_CLASS_CONTENTS}, and @code{REGISTER_NAMES}, respectively. 1909@code{global_regs} has been cleared, and any @option{-ffixed-@var{reg}}, 1910@option{-fcall-used-@var{reg}} and @option{-fcall-saved-@var{reg}} 1911command options have been applied. 1912 1913@cindex disabling certain registers 1914@cindex controlling register usage 1915If the usage of an entire class of registers depends on the target 1916flags, you may indicate this to GCC by using this macro to modify 1917@code{fixed_regs} and @code{call_used_regs} to 1 for each of the 1918registers in the classes which should not be used by GCC@. Also make 1919@code{define_register_constraint}s return @code{NO_REGS} for constraints 1920that shouldn't be used. 1921 1922(However, if this class is not included in @code{GENERAL_REGS} and all 1923of the insn patterns whose constraints permit this class are 1924controlled by target switches, then GCC will automatically avoid using 1925these registers when the target switches are opposed to them.) 1926@end deftypefn 1927 1928@defmac INCOMING_REGNO (@var{out}) 1929Define this macro if the target machine has register windows. This C 1930expression returns the register number as seen by the called function 1931corresponding to the register number @var{out} as seen by the calling 1932function. Return @var{out} if register number @var{out} is not an 1933outbound register. 1934@end defmac 1935 1936@defmac OUTGOING_REGNO (@var{in}) 1937Define this macro if the target machine has register windows. This C 1938expression returns the register number as seen by the calling function 1939corresponding to the register number @var{in} as seen by the called 1940function. Return @var{in} if register number @var{in} is not an inbound 1941register. 1942@end defmac 1943 1944@defmac LOCAL_REGNO (@var{regno}) 1945Define this macro if the target machine has register windows. This C 1946expression returns true if the register is call-saved but is in the 1947register window. Unlike most call-saved registers, such registers 1948need not be explicitly restored on function exit or during non-local 1949gotos. 1950@end defmac 1951 1952@defmac PC_REGNUM 1953If the program counter has a register number, define this as that 1954register number. Otherwise, do not define it. 1955@end defmac 1956 1957@node Allocation Order 1958@subsection Order of Allocation of Registers 1959@cindex order of register allocation 1960@cindex register allocation order 1961 1962@c prevent bad page break with this line 1963Registers are allocated in order. 1964 1965@defmac REG_ALLOC_ORDER 1966If defined, an initializer for a vector of integers, containing the 1967numbers of hard registers in the order in which GCC should prefer 1968to use them (from most preferred to least). 1969 1970If this macro is not defined, registers are used lowest numbered first 1971(all else being equal). 1972 1973One use of this macro is on machines where the highest numbered 1974registers must always be saved and the save-multiple-registers 1975instruction supports only sequences of consecutive registers. On such 1976machines, define @code{REG_ALLOC_ORDER} to be an initializer that lists 1977the highest numbered allocable register first. 1978@end defmac 1979 1980@defmac ADJUST_REG_ALLOC_ORDER 1981A C statement (sans semicolon) to choose the order in which to allocate 1982hard registers for pseudo-registers local to a basic block. 1983 1984Store the desired register order in the array @code{reg_alloc_order}. 1985Element 0 should be the register to allocate first; element 1, the next 1986register; and so on. 1987 1988The macro body should not assume anything about the contents of 1989@code{reg_alloc_order} before execution of the macro. 1990 1991On most machines, it is not necessary to define this macro. 1992@end defmac 1993 1994@defmac HONOR_REG_ALLOC_ORDER 1995Normally, IRA tries to estimate the costs for saving a register in the 1996prologue and restoring it in the epilogue. This discourages it from 1997using call-saved registers. If a machine wants to ensure that IRA 1998allocates registers in the order given by REG_ALLOC_ORDER even if some 1999call-saved registers appear earlier than call-used ones, then define this 2000macro as a C expression to nonzero. Default is 0. 2001@end defmac 2002 2003@defmac IRA_HARD_REGNO_ADD_COST_MULTIPLIER (@var{regno}) 2004In some case register allocation order is not enough for the 2005Integrated Register Allocator (@acronym{IRA}) to generate a good code. 2006If this macro is defined, it should return a floating point value 2007based on @var{regno}. The cost of using @var{regno} for a pseudo will 2008be increased by approximately the pseudo's usage frequency times the 2009value returned by this macro. Not defining this macro is equivalent 2010to having it always return @code{0.0}. 2011 2012On most machines, it is not necessary to define this macro. 2013@end defmac 2014 2015@node Values in Registers 2016@subsection How Values Fit in Registers 2017 2018This section discusses the macros that describe which kinds of values 2019(specifically, which machine modes) each register can hold, and how many 2020consecutive registers are needed for a given mode. 2021 2022@deftypefn {Target Hook} {unsigned int} TARGET_HARD_REGNO_NREGS (unsigned int @var{regno}, machine_mode @var{mode}) 2023This hook returns the number of consecutive hard registers, starting 2024at register number @var{regno}, required to hold a value of mode 2025@var{mode}. This hook must never return zero, even if a register 2026cannot hold the requested mode - indicate that with 2027@code{TARGET_HARD_REGNO_MODE_OK} and/or 2028@code{TARGET_CAN_CHANGE_MODE_CLASS} instead. 2029 2030The default definition returns the number of words in @var{mode}. 2031@end deftypefn 2032 2033@defmac HARD_REGNO_NREGS_HAS_PADDING (@var{regno}, @var{mode}) 2034A C expression that is nonzero if a value of mode @var{mode}, stored 2035in memory, ends with padding that causes it to take up more space than 2036in registers starting at register number @var{regno} (as determined by 2037multiplying GCC's notion of the size of the register when containing 2038this mode by the number of registers returned by 2039@code{TARGET_HARD_REGNO_NREGS}). By default this is zero. 2040 2041For example, if a floating-point value is stored in three 32-bit 2042registers but takes up 128 bits in memory, then this would be 2043nonzero. 2044 2045This macros only needs to be defined if there are cases where 2046@code{subreg_get_info} 2047would otherwise wrongly determine that a @code{subreg} can be 2048represented by an offset to the register number, when in fact such a 2049@code{subreg} would contain some of the padding not stored in 2050registers and so not be representable. 2051@end defmac 2052 2053@defmac HARD_REGNO_NREGS_WITH_PADDING (@var{regno}, @var{mode}) 2054For values of @var{regno} and @var{mode} for which 2055@code{HARD_REGNO_NREGS_HAS_PADDING} returns nonzero, a C expression 2056returning the greater number of registers required to hold the value 2057including any padding. In the example above, the value would be four. 2058@end defmac 2059 2060@defmac REGMODE_NATURAL_SIZE (@var{mode}) 2061Define this macro if the natural size of registers that hold values 2062of mode @var{mode} is not the word size. It is a C expression that 2063should give the natural size in bytes for the specified mode. It is 2064used by the register allocator to try to optimize its results. This 2065happens for example on SPARC 64-bit where the natural size of 2066floating-point registers is still 32-bit. 2067@end defmac 2068 2069@deftypefn {Target Hook} bool TARGET_HARD_REGNO_MODE_OK (unsigned int @var{regno}, machine_mode @var{mode}) 2070This hook returns true if it is permissible to store a value 2071of mode @var{mode} in hard register number @var{regno} (or in several 2072registers starting with that one). The default definition returns true 2073unconditionally. 2074 2075You need not include code to check for the numbers of fixed registers, 2076because the allocation mechanism considers them to be always occupied. 2077 2078@cindex register pairs 2079On some machines, double-precision values must be kept in even/odd 2080register pairs. You can implement that by defining this hook to reject 2081odd register numbers for such modes. 2082 2083The minimum requirement for a mode to be OK in a register is that the 2084@samp{mov@var{mode}} instruction pattern support moves between the 2085register and other hard register in the same class and that moving a 2086value into the register and back out not alter it. 2087 2088Since the same instruction used to move @code{word_mode} will work for 2089all narrower integer modes, it is not necessary on any machine for 2090this hook to distinguish between these modes, provided you define 2091patterns @samp{movhi}, etc., to take advantage of this. This is 2092useful because of the interaction between @code{TARGET_HARD_REGNO_MODE_OK} 2093and @code{TARGET_MODES_TIEABLE_P}; it is very desirable for all integer 2094modes to be tieable. 2095 2096Many machines have special registers for floating point arithmetic. 2097Often people assume that floating point machine modes are allowed only 2098in floating point registers. This is not true. Any registers that 2099can hold integers can safely @emph{hold} a floating point machine 2100mode, whether or not floating arithmetic can be done on it in those 2101registers. Integer move instructions can be used to move the values. 2102 2103On some machines, though, the converse is true: fixed-point machine 2104modes may not go in floating registers. This is true if the floating 2105registers normalize any value stored in them, because storing a 2106non-floating value there would garble it. In this case, 2107@code{TARGET_HARD_REGNO_MODE_OK} should reject fixed-point machine modes in 2108floating registers. But if the floating registers do not automatically 2109normalize, if you can store any bit pattern in one and retrieve it 2110unchanged without a trap, then any machine mode may go in a floating 2111register, so you can define this hook to say so. 2112 2113The primary significance of special floating registers is rather that 2114they are the registers acceptable in floating point arithmetic 2115instructions. However, this is of no concern to 2116@code{TARGET_HARD_REGNO_MODE_OK}. You handle it by writing the proper 2117constraints for those instructions. 2118 2119On some machines, the floating registers are especially slow to access, 2120so that it is better to store a value in a stack frame than in such a 2121register if floating point arithmetic is not being done. As long as the 2122floating registers are not in class @code{GENERAL_REGS}, they will not 2123be used unless some pattern's constraint asks for one. 2124@end deftypefn 2125 2126@defmac HARD_REGNO_RENAME_OK (@var{from}, @var{to}) 2127A C expression that is nonzero if it is OK to rename a hard register 2128@var{from} to another hard register @var{to}. 2129 2130One common use of this macro is to prevent renaming of a register to 2131another register that is not saved by a prologue in an interrupt 2132handler. 2133 2134The default is always nonzero. 2135@end defmac 2136 2137@deftypefn {Target Hook} bool TARGET_MODES_TIEABLE_P (machine_mode @var{mode1}, machine_mode @var{mode2}) 2138This hook returns true if a value of mode @var{mode1} is accessible 2139in mode @var{mode2} without copying. 2140 2141If @code{TARGET_HARD_REGNO_MODE_OK (@var{r}, @var{mode1})} and 2142@code{TARGET_HARD_REGNO_MODE_OK (@var{r}, @var{mode2})} are always 2143the same for any @var{r}, then 2144@code{TARGET_MODES_TIEABLE_P (@var{mode1}, @var{mode2})} 2145should be true. If they differ for any @var{r}, you should define 2146this hook to return false unless some other mechanism ensures the 2147accessibility of the value in a narrower mode. 2148 2149You should define this hook to return true in as many cases as 2150possible since doing so will allow GCC to perform better register 2151allocation. The default definition returns true unconditionally. 2152@end deftypefn 2153 2154@deftypefn {Target Hook} bool TARGET_HARD_REGNO_SCRATCH_OK (unsigned int @var{regno}) 2155This target hook should return @code{true} if it is OK to use a hard register 2156@var{regno} as scratch reg in peephole2. 2157 2158One common use of this macro is to prevent using of a register that 2159is not saved by a prologue in an interrupt handler. 2160 2161The default version of this hook always returns @code{true}. 2162@end deftypefn 2163 2164@defmac AVOID_CCMODE_COPIES 2165Define this macro if the compiler should avoid copies to/from @code{CCmode} 2166registers. You should only define this macro if support for copying to/from 2167@code{CCmode} is incomplete. 2168@end defmac 2169 2170@node Leaf Functions 2171@subsection Handling Leaf Functions 2172 2173@cindex leaf functions 2174@cindex functions, leaf 2175On some machines, a leaf function (i.e., one which makes no calls) can run 2176more efficiently if it does not make its own register window. Often this 2177means it is required to receive its arguments in the registers where they 2178are passed by the caller, instead of the registers where they would 2179normally arrive. 2180 2181The special treatment for leaf functions generally applies only when 2182other conditions are met; for example, often they may use only those 2183registers for its own variables and temporaries. We use the term ``leaf 2184function'' to mean a function that is suitable for this special 2185handling, so that functions with no calls are not necessarily ``leaf 2186functions''. 2187 2188GCC assigns register numbers before it knows whether the function is 2189suitable for leaf function treatment. So it needs to renumber the 2190registers in order to output a leaf function. The following macros 2191accomplish this. 2192 2193@defmac LEAF_REGISTERS 2194Name of a char vector, indexed by hard register number, which 2195contains 1 for a register that is allowable in a candidate for leaf 2196function treatment. 2197 2198If leaf function treatment involves renumbering the registers, then the 2199registers marked here should be the ones before renumbering---those that 2200GCC would ordinarily allocate. The registers which will actually be 2201used in the assembler code, after renumbering, should not be marked with 1 2202in this vector. 2203 2204Define this macro only if the target machine offers a way to optimize 2205the treatment of leaf functions. 2206@end defmac 2207 2208@defmac LEAF_REG_REMAP (@var{regno}) 2209A C expression whose value is the register number to which @var{regno} 2210should be renumbered, when a function is treated as a leaf function. 2211 2212If @var{regno} is a register number which should not appear in a leaf 2213function before renumbering, then the expression should yield @minus{}1, which 2214will cause the compiler to abort. 2215 2216Define this macro only if the target machine offers a way to optimize the 2217treatment of leaf functions, and registers need to be renumbered to do 2218this. 2219@end defmac 2220 2221@findex current_function_is_leaf 2222@findex current_function_uses_only_leaf_regs 2223@code{TARGET_ASM_FUNCTION_PROLOGUE} and 2224@code{TARGET_ASM_FUNCTION_EPILOGUE} must usually treat leaf functions 2225specially. They can test the C variable @code{current_function_is_leaf} 2226which is nonzero for leaf functions. @code{current_function_is_leaf} is 2227set prior to local register allocation and is valid for the remaining 2228compiler passes. They can also test the C variable 2229@code{current_function_uses_only_leaf_regs} which is nonzero for leaf 2230functions which only use leaf registers. 2231@code{current_function_uses_only_leaf_regs} is valid after all passes 2232that modify the instructions have been run and is only useful if 2233@code{LEAF_REGISTERS} is defined. 2234@c changed this to fix overfull. ALSO: why the "it" at the beginning 2235@c of the next paragraph?! --mew 2feb93 2236 2237@node Stack Registers 2238@subsection Registers That Form a Stack 2239 2240There are special features to handle computers where some of the 2241``registers'' form a stack. Stack registers are normally written by 2242pushing onto the stack, and are numbered relative to the top of the 2243stack. 2244 2245Currently, GCC can only handle one group of stack-like registers, and 2246they must be consecutively numbered. Furthermore, the existing 2247support for stack-like registers is specific to the 80387 floating 2248point coprocessor. If you have a new architecture that uses 2249stack-like registers, you will need to do substantial work on 2250@file{reg-stack.c} and write your machine description to cooperate 2251with it, as well as defining these macros. 2252 2253@defmac STACK_REGS 2254Define this if the machine has any stack-like registers. 2255@end defmac 2256 2257@defmac STACK_REG_COVER_CLASS 2258This is a cover class containing the stack registers. Define this if 2259the machine has any stack-like registers. 2260@end defmac 2261 2262@defmac FIRST_STACK_REG 2263The number of the first stack-like register. This one is the top 2264of the stack. 2265@end defmac 2266 2267@defmac LAST_STACK_REG 2268The number of the last stack-like register. This one is the bottom of 2269the stack. 2270@end defmac 2271 2272@node Register Classes 2273@section Register Classes 2274@cindex register class definitions 2275@cindex class definitions, register 2276 2277On many machines, the numbered registers are not all equivalent. 2278For example, certain registers may not be allowed for indexed addressing; 2279certain registers may not be allowed in some instructions. These machine 2280restrictions are described to the compiler using @dfn{register classes}. 2281 2282You define a number of register classes, giving each one a name and saying 2283which of the registers belong to it. Then you can specify register classes 2284that are allowed as operands to particular instruction patterns. 2285 2286@findex ALL_REGS 2287@findex NO_REGS 2288In general, each register will belong to several classes. In fact, one 2289class must be named @code{ALL_REGS} and contain all the registers. Another 2290class must be named @code{NO_REGS} and contain no registers. Often the 2291union of two classes will be another class; however, this is not required. 2292 2293@findex GENERAL_REGS 2294One of the classes must be named @code{GENERAL_REGS}. There is nothing 2295terribly special about the name, but the operand constraint letters 2296@samp{r} and @samp{g} specify this class. If @code{GENERAL_REGS} is 2297the same as @code{ALL_REGS}, just define it as a macro which expands 2298to @code{ALL_REGS}. 2299 2300Order the classes so that if class @var{x} is contained in class @var{y} 2301then @var{x} has a lower class number than @var{y}. 2302 2303The way classes other than @code{GENERAL_REGS} are specified in operand 2304constraints is through machine-dependent operand constraint letters. 2305You can define such letters to correspond to various classes, then use 2306them in operand constraints. 2307 2308You must define the narrowest register classes for allocatable 2309registers, so that each class either has no subclasses, or that for 2310some mode, the move cost between registers within the class is 2311cheaper than moving a register in the class to or from memory 2312(@pxref{Costs}). 2313 2314You should define a class for the union of two classes whenever some 2315instruction allows both classes. For example, if an instruction allows 2316either a floating point (coprocessor) register or a general register for a 2317certain operand, you should define a class @code{FLOAT_OR_GENERAL_REGS} 2318which includes both of them. Otherwise you will get suboptimal code, 2319or even internal compiler errors when reload cannot find a register in the 2320class computed via @code{reg_class_subunion}. 2321 2322You must also specify certain redundant information about the register 2323classes: for each class, which classes contain it and which ones are 2324contained in it; for each pair of classes, the largest class contained 2325in their union. 2326 2327When a value occupying several consecutive registers is expected in a 2328certain class, all the registers used must belong to that class. 2329Therefore, register classes cannot be used to enforce a requirement for 2330a register pair to start with an even-numbered register. The way to 2331specify this requirement is with @code{TARGET_HARD_REGNO_MODE_OK}. 2332 2333Register classes used for input-operands of bitwise-and or shift 2334instructions have a special requirement: each such class must have, for 2335each fixed-point machine mode, a subclass whose registers can transfer that 2336mode to or from memory. For example, on some machines, the operations for 2337single-byte values (@code{QImode}) are limited to certain registers. When 2338this is so, each register class that is used in a bitwise-and or shift 2339instruction must have a subclass consisting of registers from which 2340single-byte values can be loaded or stored. This is so that 2341@code{PREFERRED_RELOAD_CLASS} can always have a possible value to return. 2342 2343@deftp {Data type} {enum reg_class} 2344An enumerated type that must be defined with all the register class names 2345as enumerated values. @code{NO_REGS} must be first. @code{ALL_REGS} 2346must be the last register class, followed by one more enumerated value, 2347@code{LIM_REG_CLASSES}, which is not a register class but rather 2348tells how many classes there are. 2349 2350Each register class has a number, which is the value of casting 2351the class name to type @code{int}. The number serves as an index 2352in many of the tables described below. 2353@end deftp 2354 2355@defmac N_REG_CLASSES 2356The number of distinct register classes, defined as follows: 2357 2358@smallexample 2359#define N_REG_CLASSES (int) LIM_REG_CLASSES 2360@end smallexample 2361@end defmac 2362 2363@defmac REG_CLASS_NAMES 2364An initializer containing the names of the register classes as C string 2365constants. These names are used in writing some of the debugging dumps. 2366@end defmac 2367 2368@defmac REG_CLASS_CONTENTS 2369An initializer containing the contents of the register classes, as integers 2370which are bit masks. The @var{n}th integer specifies the contents of class 2371@var{n}. The way the integer @var{mask} is interpreted is that 2372register @var{r} is in the class if @code{@var{mask} & (1 << @var{r})} is 1. 2373 2374When the machine has more than 32 registers, an integer does not suffice. 2375Then the integers are replaced by sub-initializers, braced groupings containing 2376several integers. Each sub-initializer must be suitable as an initializer 2377for the type @code{HARD_REG_SET} which is defined in @file{hard-reg-set.h}. 2378In this situation, the first integer in each sub-initializer corresponds to 2379registers 0 through 31, the second integer to registers 32 through 63, and 2380so on. 2381@end defmac 2382 2383@defmac REGNO_REG_CLASS (@var{regno}) 2384A C expression whose value is a register class containing hard register 2385@var{regno}. In general there is more than one such class; choose a class 2386which is @dfn{minimal}, meaning that no smaller class also contains the 2387register. 2388@end defmac 2389 2390@defmac BASE_REG_CLASS 2391A macro whose definition is the name of the class to which a valid 2392base register must belong. A base register is one used in an address 2393which is the register value plus a displacement. 2394@end defmac 2395 2396@defmac MODE_BASE_REG_CLASS (@var{mode}) 2397This is a variation of the @code{BASE_REG_CLASS} macro which allows 2398the selection of a base register in a mode dependent manner. If 2399@var{mode} is VOIDmode then it should return the same value as 2400@code{BASE_REG_CLASS}. 2401@end defmac 2402 2403@defmac MODE_BASE_REG_REG_CLASS (@var{mode}) 2404A C expression whose value is the register class to which a valid 2405base register must belong in order to be used in a base plus index 2406register address. You should define this macro if base plus index 2407addresses have different requirements than other base register uses. 2408@end defmac 2409 2410@defmac MODE_CODE_BASE_REG_CLASS (@var{mode}, @var{address_space}, @var{outer_code}, @var{index_code}) 2411A C expression whose value is the register class to which a valid 2412base register for a memory reference in mode @var{mode} to address 2413space @var{address_space} must belong. @var{outer_code} and @var{index_code} 2414define the context in which the base register occurs. @var{outer_code} is 2415the code of the immediately enclosing expression (@code{MEM} for the top level 2416of an address, @code{ADDRESS} for something that occurs in an 2417@code{address_operand}). @var{index_code} is the code of the corresponding 2418index expression if @var{outer_code} is @code{PLUS}; @code{SCRATCH} otherwise. 2419@end defmac 2420 2421@defmac INDEX_REG_CLASS 2422A macro whose definition is the name of the class to which a valid 2423index register must belong. An index register is one used in an 2424address where its value is either multiplied by a scale factor or 2425added to another register (as well as added to a displacement). 2426@end defmac 2427 2428@defmac REGNO_OK_FOR_BASE_P (@var{num}) 2429A C expression which is nonzero if register number @var{num} is 2430suitable for use as a base register in operand addresses. 2431@end defmac 2432 2433@defmac REGNO_MODE_OK_FOR_BASE_P (@var{num}, @var{mode}) 2434A C expression that is just like @code{REGNO_OK_FOR_BASE_P}, except that 2435that expression may examine the mode of the memory reference in 2436@var{mode}. You should define this macro if the mode of the memory 2437reference affects whether a register may be used as a base register. If 2438you define this macro, the compiler will use it instead of 2439@code{REGNO_OK_FOR_BASE_P}. The mode may be @code{VOIDmode} for 2440addresses that appear outside a @code{MEM}, i.e., as an 2441@code{address_operand}. 2442@end defmac 2443 2444@defmac REGNO_MODE_OK_FOR_REG_BASE_P (@var{num}, @var{mode}) 2445A C expression which is nonzero if register number @var{num} is suitable for 2446use as a base register in base plus index operand addresses, accessing 2447memory in mode @var{mode}. It may be either a suitable hard register or a 2448pseudo register that has been allocated such a hard register. You should 2449define this macro if base plus index addresses have different requirements 2450than other base register uses. 2451 2452Use of this macro is deprecated; please use the more general 2453@code{REGNO_MODE_CODE_OK_FOR_BASE_P}. 2454@end defmac 2455 2456@defmac REGNO_MODE_CODE_OK_FOR_BASE_P (@var{num}, @var{mode}, @var{address_space}, @var{outer_code}, @var{index_code}) 2457A C expression which is nonzero if register number @var{num} is 2458suitable for use as a base register in operand addresses, accessing 2459memory in mode @var{mode} in address space @var{address_space}. 2460This is similar to @code{REGNO_MODE_OK_FOR_BASE_P}, except 2461that that expression may examine the context in which the register 2462appears in the memory reference. @var{outer_code} is the code of the 2463immediately enclosing expression (@code{MEM} if at the top level of the 2464address, @code{ADDRESS} for something that occurs in an 2465@code{address_operand}). @var{index_code} is the code of the 2466corresponding index expression if @var{outer_code} is @code{PLUS}; 2467@code{SCRATCH} otherwise. The mode may be @code{VOIDmode} for addresses 2468that appear outside a @code{MEM}, i.e., as an @code{address_operand}. 2469@end defmac 2470 2471@defmac REGNO_OK_FOR_INDEX_P (@var{num}) 2472A C expression which is nonzero if register number @var{num} is 2473suitable for use as an index register in operand addresses. It may be 2474either a suitable hard register or a pseudo register that has been 2475allocated such a hard register. 2476 2477The difference between an index register and a base register is that 2478the index register may be scaled. If an address involves the sum of 2479two registers, neither one of them scaled, then either one may be 2480labeled the ``base'' and the other the ``index''; but whichever 2481labeling is used must fit the machine's constraints of which registers 2482may serve in each capacity. The compiler will try both labelings, 2483looking for one that is valid, and will reload one or both registers 2484only if neither labeling works. 2485@end defmac 2486 2487@deftypefn {Target Hook} reg_class_t TARGET_PREFERRED_RENAME_CLASS (reg_class_t @var{rclass}) 2488A target hook that places additional preference on the register class to use when it is necessary to rename a register in class @var{rclass} to another class, or perhaps @var{NO_REGS}, if no preferred register class is found or hook @code{preferred_rename_class} is not implemented. Sometimes returning a more restrictive class makes better code. For example, on ARM, thumb-2 instructions using @code{LO_REGS} may be smaller than instructions using @code{GENERIC_REGS}. By returning @code{LO_REGS} from @code{preferred_rename_class}, code size can be reduced. 2489@end deftypefn 2490 2491@deftypefn {Target Hook} reg_class_t TARGET_PREFERRED_RELOAD_CLASS (rtx @var{x}, reg_class_t @var{rclass}) 2492A target hook that places additional restrictions on the register class 2493to use when it is necessary to copy value @var{x} into a register in class 2494@var{rclass}. The value is a register class; perhaps @var{rclass}, or perhaps 2495another, smaller class. 2496 2497The default version of this hook always returns value of @code{rclass} argument. 2498 2499Sometimes returning a more restrictive class makes better code. For 2500example, on the 68000, when @var{x} is an integer constant that is in range 2501for a @samp{moveq} instruction, the value of this macro is always 2502@code{DATA_REGS} as long as @var{rclass} includes the data registers. 2503Requiring a data register guarantees that a @samp{moveq} will be used. 2504 2505One case where @code{TARGET_PREFERRED_RELOAD_CLASS} must not return 2506@var{rclass} is if @var{x} is a legitimate constant which cannot be 2507loaded into some register class. By returning @code{NO_REGS} you can 2508force @var{x} into a memory location. For example, rs6000 can load 2509immediate values into general-purpose registers, but does not have an 2510instruction for loading an immediate value into a floating-point 2511register, so @code{TARGET_PREFERRED_RELOAD_CLASS} returns @code{NO_REGS} when 2512@var{x} is a floating-point constant. If the constant can't be loaded 2513into any kind of register, code generation will be better if 2514@code{TARGET_LEGITIMATE_CONSTANT_P} makes the constant illegitimate instead 2515of using @code{TARGET_PREFERRED_RELOAD_CLASS}. 2516 2517If an insn has pseudos in it after register allocation, reload will go 2518through the alternatives and call repeatedly @code{TARGET_PREFERRED_RELOAD_CLASS} 2519to find the best one. Returning @code{NO_REGS}, in this case, makes 2520reload add a @code{!} in front of the constraint: the x86 back-end uses 2521this feature to discourage usage of 387 registers when math is done in 2522the SSE registers (and vice versa). 2523@end deftypefn 2524 2525@defmac PREFERRED_RELOAD_CLASS (@var{x}, @var{class}) 2526A C expression that places additional restrictions on the register class 2527to use when it is necessary to copy value @var{x} into a register in class 2528@var{class}. The value is a register class; perhaps @var{class}, or perhaps 2529another, smaller class. On many machines, the following definition is 2530safe: 2531 2532@smallexample 2533#define PREFERRED_RELOAD_CLASS(X,CLASS) CLASS 2534@end smallexample 2535 2536Sometimes returning a more restrictive class makes better code. For 2537example, on the 68000, when @var{x} is an integer constant that is in range 2538for a @samp{moveq} instruction, the value of this macro is always 2539@code{DATA_REGS} as long as @var{class} includes the data registers. 2540Requiring a data register guarantees that a @samp{moveq} will be used. 2541 2542One case where @code{PREFERRED_RELOAD_CLASS} must not return 2543@var{class} is if @var{x} is a legitimate constant which cannot be 2544loaded into some register class. By returning @code{NO_REGS} you can 2545force @var{x} into a memory location. For example, rs6000 can load 2546immediate values into general-purpose registers, but does not have an 2547instruction for loading an immediate value into a floating-point 2548register, so @code{PREFERRED_RELOAD_CLASS} returns @code{NO_REGS} when 2549@var{x} is a floating-point constant. If the constant cannot be loaded 2550into any kind of register, code generation will be better if 2551@code{TARGET_LEGITIMATE_CONSTANT_P} makes the constant illegitimate instead 2552of using @code{TARGET_PREFERRED_RELOAD_CLASS}. 2553 2554If an insn has pseudos in it after register allocation, reload will go 2555through the alternatives and call repeatedly @code{PREFERRED_RELOAD_CLASS} 2556to find the best one. Returning @code{NO_REGS}, in this case, makes 2557reload add a @code{!} in front of the constraint: the x86 back-end uses 2558this feature to discourage usage of 387 registers when math is done in 2559the SSE registers (and vice versa). 2560@end defmac 2561 2562@deftypefn {Target Hook} reg_class_t TARGET_PREFERRED_OUTPUT_RELOAD_CLASS (rtx @var{x}, reg_class_t @var{rclass}) 2563Like @code{TARGET_PREFERRED_RELOAD_CLASS}, but for output reloads instead of 2564input reloads. 2565 2566The default version of this hook always returns value of @code{rclass} 2567argument. 2568 2569You can also use @code{TARGET_PREFERRED_OUTPUT_RELOAD_CLASS} to discourage 2570reload from using some alternatives, like @code{TARGET_PREFERRED_RELOAD_CLASS}. 2571@end deftypefn 2572 2573@defmac LIMIT_RELOAD_CLASS (@var{mode}, @var{class}) 2574A C expression that places additional restrictions on the register class 2575to use when it is necessary to be able to hold a value of mode 2576@var{mode} in a reload register for which class @var{class} would 2577ordinarily be used. 2578 2579Unlike @code{PREFERRED_RELOAD_CLASS}, this macro should be used when 2580there are certain modes that simply cannot go in certain reload classes. 2581 2582The value is a register class; perhaps @var{class}, or perhaps another, 2583smaller class. 2584 2585Don't define this macro unless the target machine has limitations which 2586require the macro to do something nontrivial. 2587@end defmac 2588 2589@deftypefn {Target Hook} reg_class_t TARGET_SECONDARY_RELOAD (bool @var{in_p}, rtx @var{x}, reg_class_t @var{reload_class}, machine_mode @var{reload_mode}, secondary_reload_info *@var{sri}) 2590Many machines have some registers that cannot be copied directly to or 2591from memory or even from other types of registers. An example is the 2592@samp{MQ} register, which on most machines, can only be copied to or 2593from general registers, but not memory. Below, we shall be using the 2594term 'intermediate register' when a move operation cannot be performed 2595directly, but has to be done by copying the source into the intermediate 2596register first, and then copying the intermediate register to the 2597destination. An intermediate register always has the same mode as 2598source and destination. Since it holds the actual value being copied, 2599reload might apply optimizations to re-use an intermediate register 2600and eliding the copy from the source when it can determine that the 2601intermediate register still holds the required value. 2602 2603Another kind of secondary reload is required on some machines which 2604allow copying all registers to and from memory, but require a scratch 2605register for stores to some memory locations (e.g., those with symbolic 2606address on the RT, and those with certain symbolic address on the SPARC 2607when compiling PIC)@. Scratch registers need not have the same mode 2608as the value being copied, and usually hold a different value than 2609that being copied. Special patterns in the md file are needed to 2610describe how the copy is performed with the help of the scratch register; 2611these patterns also describe the number, register class(es) and mode(s) 2612of the scratch register(s). 2613 2614In some cases, both an intermediate and a scratch register are required. 2615 2616For input reloads, this target hook is called with nonzero @var{in_p}, 2617and @var{x} is an rtx that needs to be copied to a register of class 2618@var{reload_class} in @var{reload_mode}. For output reloads, this target 2619hook is called with zero @var{in_p}, and a register of class @var{reload_class} 2620needs to be copied to rtx @var{x} in @var{reload_mode}. 2621 2622If copying a register of @var{reload_class} from/to @var{x} requires 2623an intermediate register, the hook @code{secondary_reload} should 2624return the register class required for this intermediate register. 2625If no intermediate register is required, it should return NO_REGS. 2626If more than one intermediate register is required, describe the one 2627that is closest in the copy chain to the reload register. 2628 2629If scratch registers are needed, you also have to describe how to 2630perform the copy from/to the reload register to/from this 2631closest intermediate register. Or if no intermediate register is 2632required, but still a scratch register is needed, describe the 2633copy from/to the reload register to/from the reload operand @var{x}. 2634 2635You do this by setting @code{sri->icode} to the instruction code of a pattern 2636in the md file which performs the move. Operands 0 and 1 are the output 2637and input of this copy, respectively. Operands from operand 2 onward are 2638for scratch operands. These scratch operands must have a mode, and a 2639single-register-class 2640@c [later: or memory] 2641output constraint. 2642 2643When an intermediate register is used, the @code{secondary_reload} 2644hook will be called again to determine how to copy the intermediate 2645register to/from the reload operand @var{x}, so your hook must also 2646have code to handle the register class of the intermediate operand. 2647 2648@c [For later: maybe we'll allow multi-alternative reload patterns - 2649@c the port maintainer could name a mov<mode> pattern that has clobbers - 2650@c and match the constraints of input and output to determine the required 2651@c alternative. A restriction would be that constraints used to match 2652@c against reloads registers would have to be written as register class 2653@c constraints, or we need a new target macro / hook that tells us if an 2654@c arbitrary constraint can match an unknown register of a given class. 2655@c Such a macro / hook would also be useful in other places.] 2656 2657 2658@var{x} might be a pseudo-register or a @code{subreg} of a 2659pseudo-register, which could either be in a hard register or in memory. 2660Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is 2661in memory and the hard register number if it is in a register. 2662 2663Scratch operands in memory (constraint @code{"=m"} / @code{"=&m"}) are 2664currently not supported. For the time being, you will have to continue 2665to use @code{TARGET_SECONDARY_MEMORY_NEEDED} for that purpose. 2666 2667@code{copy_cost} also uses this target hook to find out how values are 2668copied. If you want it to include some extra cost for the need to allocate 2669(a) scratch register(s), set @code{sri->extra_cost} to the additional cost. 2670Or if two dependent moves are supposed to have a lower cost than the sum 2671of the individual moves due to expected fortuitous scheduling and/or special 2672forwarding logic, you can set @code{sri->extra_cost} to a negative amount. 2673@end deftypefn 2674 2675@defmac SECONDARY_RELOAD_CLASS (@var{class}, @var{mode}, @var{x}) 2676@defmacx SECONDARY_INPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x}) 2677@defmacx SECONDARY_OUTPUT_RELOAD_CLASS (@var{class}, @var{mode}, @var{x}) 2678These macros are obsolete, new ports should use the target hook 2679@code{TARGET_SECONDARY_RELOAD} instead. 2680 2681These are obsolete macros, replaced by the @code{TARGET_SECONDARY_RELOAD} 2682target hook. Older ports still define these macros to indicate to the 2683reload phase that it may 2684need to allocate at least one register for a reload in addition to the 2685register to contain the data. Specifically, if copying @var{x} to a 2686register @var{class} in @var{mode} requires an intermediate register, 2687you were supposed to define @code{SECONDARY_INPUT_RELOAD_CLASS} to return the 2688largest register class all of whose registers can be used as 2689intermediate registers or scratch registers. 2690 2691If copying a register @var{class} in @var{mode} to @var{x} requires an 2692intermediate or scratch register, @code{SECONDARY_OUTPUT_RELOAD_CLASS} 2693was supposed to be defined be defined to return the largest register 2694class required. If the 2695requirements for input and output reloads were the same, the macro 2696@code{SECONDARY_RELOAD_CLASS} should have been used instead of defining both 2697macros identically. 2698 2699The values returned by these macros are often @code{GENERAL_REGS}. 2700Return @code{NO_REGS} if no spare register is needed; i.e., if @var{x} 2701can be directly copied to or from a register of @var{class} in 2702@var{mode} without requiring a scratch register. Do not define this 2703macro if it would always return @code{NO_REGS}. 2704 2705If a scratch register is required (either with or without an 2706intermediate register), you were supposed to define patterns for 2707@samp{reload_in@var{m}} or @samp{reload_out@var{m}}, as required 2708(@pxref{Standard Names}. These patterns, which were normally 2709implemented with a @code{define_expand}, should be similar to the 2710@samp{mov@var{m}} patterns, except that operand 2 is the scratch 2711register. 2712 2713These patterns need constraints for the reload register and scratch 2714register that 2715contain a single register class. If the original reload register (whose 2716class is @var{class}) can meet the constraint given in the pattern, the 2717value returned by these macros is used for the class of the scratch 2718register. Otherwise, two additional reload registers are required. 2719Their classes are obtained from the constraints in the insn pattern. 2720 2721@var{x} might be a pseudo-register or a @code{subreg} of a 2722pseudo-register, which could either be in a hard register or in memory. 2723Use @code{true_regnum} to find out; it will return @minus{}1 if the pseudo is 2724in memory and the hard register number if it is in a register. 2725 2726These macros should not be used in the case where a particular class of 2727registers can only be copied to memory and not to another class of 2728registers. In that case, secondary reload registers are not needed and 2729would not be helpful. Instead, a stack location must be used to perform 2730the copy and the @code{mov@var{m}} pattern should use memory as an 2731intermediate storage. This case often occurs between floating-point and 2732general registers. 2733@end defmac 2734 2735@deftypefn {Target Hook} bool TARGET_SECONDARY_MEMORY_NEEDED (machine_mode @var{mode}, reg_class_t @var{class1}, reg_class_t @var{class2}) 2736Certain machines have the property that some registers cannot be copied 2737to some other registers without using memory. Define this hook on 2738those machines to return true if objects of mode @var{m} in registers 2739of @var{class1} can only be copied to registers of class @var{class2} by 2740 storing a register of @var{class1} into memory and loading that memory 2741location into a register of @var{class2}. The default definition returns 2742false for all inputs. 2743@end deftypefn 2744 2745@defmac SECONDARY_MEMORY_NEEDED_RTX (@var{mode}) 2746Normally when @code{TARGET_SECONDARY_MEMORY_NEEDED} is defined, the compiler 2747allocates a stack slot for a memory location needed for register copies. 2748If this macro is defined, the compiler instead uses the memory location 2749defined by this macro. 2750 2751Do not define this macro if you do not define 2752@code{TARGET_SECONDARY_MEMORY_NEEDED}. 2753@end defmac 2754 2755@deftypefn {Target Hook} machine_mode TARGET_SECONDARY_MEMORY_NEEDED_MODE (machine_mode @var{mode}) 2756If @code{TARGET_SECONDARY_MEMORY_NEEDED} tells the compiler to use memory 2757when moving between two particular registers of mode @var{mode}, 2758this hook specifies the mode that the memory should have. 2759 2760The default depends on @code{TARGET_LRA_P}. Without LRA, the default 2761is to use a word-sized mode for integral modes that are smaller than a 2762a word. This is right thing to do on most machines because it ensures 2763that all bits of the register are copied and prevents accesses to the 2764registers in a narrower mode, which some machines prohibit for 2765floating-point registers. 2766 2767However, this default behavior is not correct on some machines, such as 2768the DEC Alpha, that store short integers in floating-point registers 2769differently than in integer registers. On those machines, the default 2770widening will not work correctly and you must define this hook to 2771suppress that widening in some cases. See the file @file{alpha.c} for 2772details. 2773 2774With LRA, the default is to use @var{mode} unmodified. 2775@end deftypefn 2776 2777@deftypefn {Target Hook} void TARGET_SELECT_EARLY_REMAT_MODES (sbitmap @var{modes}) 2778On some targets, certain modes cannot be held in registers around a 2779standard ABI call and are relatively expensive to spill to the stack. 2780The early rematerialization pass can help in such cases by aggressively 2781recomputing values after calls, so that they don't need to be spilled. 2782 2783This hook returns the set of such modes by setting the associated bits 2784in @var{modes}. The default implementation selects no modes, which has 2785the effect of disabling the early rematerialization pass. 2786@end deftypefn 2787 2788@deftypefn {Target Hook} bool TARGET_CLASS_LIKELY_SPILLED_P (reg_class_t @var{rclass}) 2789A target hook which returns @code{true} if pseudos that have been assigned 2790to registers of class @var{rclass} would likely be spilled because 2791registers of @var{rclass} are needed for spill registers. 2792 2793The default version of this target hook returns @code{true} if @var{rclass} 2794has exactly one register and @code{false} otherwise. On most machines, this 2795default should be used. For generally register-starved machines, such as 2796i386, or machines with right register constraints, such as SH, this hook 2797can be used to avoid excessive spilling. 2798 2799This hook is also used by some of the global intra-procedural code 2800transformations to throtle code motion, to avoid increasing register 2801pressure. 2802@end deftypefn 2803 2804@deftypefn {Target Hook} {unsigned char} TARGET_CLASS_MAX_NREGS (reg_class_t @var{rclass}, machine_mode @var{mode}) 2805A target hook returns the maximum number of consecutive registers 2806of class @var{rclass} needed to hold a value of mode @var{mode}. 2807 2808This is closely related to the macro @code{TARGET_HARD_REGNO_NREGS}. 2809In fact, the value returned by @code{TARGET_CLASS_MAX_NREGS (@var{rclass}, 2810@var{mode})} target hook should be the maximum value of 2811@code{TARGET_HARD_REGNO_NREGS (@var{regno}, @var{mode})} for all @var{regno} 2812values in the class @var{rclass}. 2813 2814This target hook helps control the handling of multiple-word values 2815in the reload pass. 2816 2817The default version of this target hook returns the size of @var{mode} 2818in words. 2819@end deftypefn 2820 2821@defmac CLASS_MAX_NREGS (@var{class}, @var{mode}) 2822A C expression for the maximum number of consecutive registers 2823of class @var{class} needed to hold a value of mode @var{mode}. 2824 2825This is closely related to the macro @code{TARGET_HARD_REGNO_NREGS}. In fact, 2826the value of the macro @code{CLASS_MAX_NREGS (@var{class}, @var{mode})} 2827should be the maximum value of @code{TARGET_HARD_REGNO_NREGS (@var{regno}, 2828@var{mode})} for all @var{regno} values in the class @var{class}. 2829 2830This macro helps control the handling of multiple-word values 2831in the reload pass. 2832@end defmac 2833 2834@deftypefn {Target Hook} bool TARGET_CAN_CHANGE_MODE_CLASS (machine_mode @var{from}, machine_mode @var{to}, reg_class_t @var{rclass}) 2835This hook returns true if it is possible to bitcast values held in 2836registers of class @var{rclass} from mode @var{from} to mode @var{to} 2837and if doing so preserves the low-order bits that are common to both modes. 2838The result is only meaningful if @var{rclass} has registers that can hold 2839both @code{from} and @code{to}. The default implementation returns true. 2840 2841As an example of when such bitcasting is invalid, loading 32-bit integer or 2842floating-point objects into floating-point registers on Alpha extends them 2843to 64 bits. Therefore loading a 64-bit object and then storing it as a 284432-bit object does not store the low-order 32 bits, as would be the case 2845for a normal register. Therefore, @file{alpha.h} defines 2846@code{TARGET_CAN_CHANGE_MODE_CLASS} to return: 2847 2848@smallexample 2849(GET_MODE_SIZE (from) == GET_MODE_SIZE (to) 2850 || !reg_classes_intersect_p (FLOAT_REGS, rclass)) 2851@end smallexample 2852 2853Even if storing from a register in mode @var{to} would be valid, 2854if both @var{from} and @code{raw_reg_mode} for @var{rclass} are wider 2855than @code{word_mode}, then we must prevent @var{to} narrowing the 2856mode. This happens when the middle-end assumes that it can load 2857or store pieces of an @var{N}-word pseudo, and that the pseudo will 2858eventually be allocated to @var{N} @code{word_mode} hard registers. 2859Failure to prevent this kind of mode change will result in the 2860entire @code{raw_reg_mode} being modified instead of the partial 2861value that the middle-end intended. 2862@end deftypefn 2863 2864@deftypefn {Target Hook} reg_class_t TARGET_IRA_CHANGE_PSEUDO_ALLOCNO_CLASS (int, @var{reg_class_t}, @var{reg_class_t}) 2865A target hook which can change allocno class for given pseudo from 2866 allocno and best class calculated by IRA. 2867 2868 The default version of this target hook always returns given class. 2869@end deftypefn 2870 2871@deftypefn {Target Hook} bool TARGET_LRA_P (void) 2872A target hook which returns true if we use LRA instead of reload pass. The default version of this target hook returns true. New ports should use LRA, and existing ports are encouraged to convert. 2873@end deftypefn 2874 2875@deftypefn {Target Hook} int TARGET_REGISTER_PRIORITY (int) 2876A target hook which returns the register priority number to which the register @var{hard_regno} belongs to. The bigger the number, the more preferable the hard register usage (when all other conditions are the same). This hook can be used to prefer some hard register over others in LRA. For example, some x86-64 register usage needs additional prefix which makes instructions longer. The hook can return lower priority number for such registers make them less favorable and as result making the generated code smaller. The default version of this target hook returns always zero. 2877@end deftypefn 2878 2879@deftypefn {Target Hook} bool TARGET_REGISTER_USAGE_LEVELING_P (void) 2880A target hook which returns true if we need register usage leveling. That means if a few hard registers are equally good for the assignment, we choose the least used hard register. The register usage leveling may be profitable for some targets. Don't use the usage leveling for targets with conditional execution or targets with big register files as it hurts if-conversion and cross-jumping optimizations. The default version of this target hook returns always false. 2881@end deftypefn 2882 2883@deftypefn {Target Hook} bool TARGET_DIFFERENT_ADDR_DISPLACEMENT_P (void) 2884A target hook which returns true if an address with the same structure can have different maximal legitimate displacement. For example, the displacement can depend on memory mode or on operand combinations in the insn. The default version of this target hook returns always false. 2885@end deftypefn 2886 2887@deftypefn {Target Hook} bool TARGET_CANNOT_SUBSTITUTE_MEM_EQUIV_P (rtx @var{subst}) 2888A target hook which returns @code{true} if @var{subst} can't 2889substitute safely pseudos with equivalent memory values during 2890register allocation. 2891The default version of this target hook returns @code{false}. 2892On most machines, this default should be used. For generally 2893machines with non orthogonal register usage for addressing, such 2894as SH, this hook can be used to avoid excessive spilling. 2895@end deftypefn 2896 2897@deftypefn {Target Hook} bool TARGET_LEGITIMIZE_ADDRESS_DISPLACEMENT (rtx *@var{offset1}, rtx *@var{offset2}, poly_int64 @var{orig_offset}, machine_mode @var{mode}) 2898This hook tries to split address offset @var{orig_offset} into 2899two parts: one that should be added to the base address to create 2900a local anchor point, and an additional offset that can be applied 2901to the anchor to address a value of mode @var{mode}. The idea is that 2902the local anchor could be shared by other accesses to nearby locations. 2903 2904The hook returns true if it succeeds, storing the offset of the 2905anchor from the base in @var{offset1} and the offset of the final address 2906from the anchor in @var{offset2}. The default implementation returns false. 2907@end deftypefn 2908 2909@deftypefn {Target Hook} reg_class_t TARGET_SPILL_CLASS (reg_class_t, @var{machine_mode}) 2910This hook defines a class of registers which could be used for spilling pseudos of the given mode and class, or @code{NO_REGS} if only memory should be used. Not defining this hook is equivalent to returning @code{NO_REGS} for all inputs. 2911@end deftypefn 2912 2913@deftypefn {Target Hook} bool TARGET_ADDITIONAL_ALLOCNO_CLASS_P (reg_class_t) 2914This hook should return @code{true} if given class of registers should be an allocno class in any way. Usually RA uses only one register class from all classes containing the same register set. In some complicated cases, you need to have two or more such classes as allocno ones for RA correct work. Not defining this hook is equivalent to returning @code{false} for all inputs. 2915@end deftypefn 2916 2917@deftypefn {Target Hook} scalar_int_mode TARGET_CSTORE_MODE (enum insn_code @var{icode}) 2918This hook defines the machine mode to use for the boolean result of conditional store patterns. The ICODE argument is the instruction code for the cstore being performed. Not definiting this hook is the same as accepting the mode encoded into operand 0 of the cstore expander patterns. 2919@end deftypefn 2920 2921@deftypefn {Target Hook} int TARGET_COMPUTE_PRESSURE_CLASSES (enum reg_class *@var{pressure_classes}) 2922A target hook which lets a backend compute the set of pressure classes to be used by those optimization passes which take register pressure into account, as opposed to letting IRA compute them. It returns the number of register classes stored in the array @var{pressure_classes}. 2923@end deftypefn 2924 2925@node Stack and Calling 2926@section Stack Layout and Calling Conventions 2927@cindex calling conventions 2928 2929@c prevent bad page break with this line 2930This describes the stack layout and calling conventions. 2931 2932@menu 2933* Frame Layout:: 2934* Exception Handling:: 2935* Stack Checking:: 2936* Frame Registers:: 2937* Elimination:: 2938* Stack Arguments:: 2939* Register Arguments:: 2940* Scalar Return:: 2941* Aggregate Return:: 2942* Caller Saves:: 2943* Function Entry:: 2944* Profiling:: 2945* Tail Calls:: 2946* Shrink-wrapping separate components:: 2947* Stack Smashing Protection:: 2948* Miscellaneous Register Hooks:: 2949@end menu 2950 2951@node Frame Layout 2952@subsection Basic Stack Layout 2953@cindex stack frame layout 2954@cindex frame layout 2955 2956@c prevent bad page break with this line 2957Here is the basic stack layout. 2958 2959@defmac STACK_GROWS_DOWNWARD 2960Define this macro to be true if pushing a word onto the stack moves the stack 2961pointer to a smaller address, and false otherwise. 2962@end defmac 2963 2964@defmac STACK_PUSH_CODE 2965This macro defines the operation used when something is pushed 2966on the stack. In RTL, a push operation will be 2967@code{(set (mem (STACK_PUSH_CODE (reg sp))) @dots{})} 2968 2969The choices are @code{PRE_DEC}, @code{POST_DEC}, @code{PRE_INC}, 2970and @code{POST_INC}. Which of these is correct depends on 2971the stack direction and on whether the stack pointer points 2972to the last item on the stack or whether it points to the 2973space for the next item on the stack. 2974 2975The default is @code{PRE_DEC} when @code{STACK_GROWS_DOWNWARD} is 2976true, which is almost always right, and @code{PRE_INC} otherwise, 2977which is often wrong. 2978@end defmac 2979 2980@defmac FRAME_GROWS_DOWNWARD 2981Define this macro to nonzero value if the addresses of local variable slots 2982are at negative offsets from the frame pointer. 2983@end defmac 2984 2985@defmac ARGS_GROW_DOWNWARD 2986Define this macro if successive arguments to a function occupy decreasing 2987addresses on the stack. 2988@end defmac 2989 2990@deftypefn {Target Hook} HOST_WIDE_INT TARGET_STARTING_FRAME_OFFSET (void) 2991This hook returns the offset from the frame pointer to the first local 2992variable slot to be allocated. If @code{FRAME_GROWS_DOWNWARD}, it is the 2993offset to @emph{end} of the first slot allocated, otherwise it is the 2994offset to @emph{beginning} of the first slot allocated. The default 2995implementation returns 0. 2996@end deftypefn 2997 2998@defmac STACK_ALIGNMENT_NEEDED 2999Define to zero to disable final alignment of the stack during reload. 3000The nonzero default for this macro is suitable for most ports. 3001 3002On ports where @code{TARGET_STARTING_FRAME_OFFSET} is nonzero or where there 3003is a register save block following the local block that doesn't require 3004alignment to @code{STACK_BOUNDARY}, it may be beneficial to disable 3005stack alignment and do it in the backend. 3006@end defmac 3007 3008@defmac STACK_POINTER_OFFSET 3009Offset from the stack pointer register to the first location at which 3010outgoing arguments are placed. If not specified, the default value of 3011zero is used. This is the proper value for most machines. 3012 3013If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above 3014the first location at which outgoing arguments are placed. 3015@end defmac 3016 3017@defmac FIRST_PARM_OFFSET (@var{fundecl}) 3018Offset from the argument pointer register to the first argument's 3019address. On some machines it may depend on the data type of the 3020function. 3021 3022If @code{ARGS_GROW_DOWNWARD}, this is the offset to the location above 3023the first argument's address. 3024@end defmac 3025 3026@defmac STACK_DYNAMIC_OFFSET (@var{fundecl}) 3027Offset from the stack pointer register to an item dynamically allocated 3028on the stack, e.g., by @code{alloca}. 3029 3030The default value for this macro is @code{STACK_POINTER_OFFSET} plus the 3031length of the outgoing arguments. The default is correct for most 3032machines. See @file{function.c} for details. 3033@end defmac 3034 3035@defmac INITIAL_FRAME_ADDRESS_RTX 3036A C expression whose value is RTL representing the address of the initial 3037stack frame. This address is passed to @code{RETURN_ADDR_RTX} and 3038@code{DYNAMIC_CHAIN_ADDRESS}. If you don't define this macro, a reasonable 3039default value will be used. Define this macro in order to make frame pointer 3040elimination work in the presence of @code{__builtin_frame_address (count)} and 3041@code{__builtin_return_address (count)} for @code{count} not equal to zero. 3042@end defmac 3043 3044@defmac DYNAMIC_CHAIN_ADDRESS (@var{frameaddr}) 3045A C expression whose value is RTL representing the address in a stack 3046frame where the pointer to the caller's frame is stored. Assume that 3047@var{frameaddr} is an RTL expression for the address of the stack frame 3048itself. 3049 3050If you don't define this macro, the default is to return the value 3051of @var{frameaddr}---that is, the stack frame address is also the 3052address of the stack word that points to the previous frame. 3053@end defmac 3054 3055@defmac SETUP_FRAME_ADDRESSES 3056A C expression that produces the machine-specific code to 3057setup the stack so that arbitrary frames can be accessed. For example, 3058on the SPARC, we must flush all of the register windows to the stack 3059before we can access arbitrary stack frames. You will seldom need to 3060define this macro. The default is to do nothing. 3061@end defmac 3062 3063@deftypefn {Target Hook} rtx TARGET_BUILTIN_SETJMP_FRAME_VALUE (void) 3064This target hook should return an rtx that is used to store 3065the address of the current frame into the built in @code{setjmp} buffer. 3066The default value, @code{virtual_stack_vars_rtx}, is correct for most 3067machines. One reason you may need to define this target hook is if 3068@code{hard_frame_pointer_rtx} is the appropriate value on your machine. 3069@end deftypefn 3070 3071@defmac FRAME_ADDR_RTX (@var{frameaddr}) 3072A C expression whose value is RTL representing the value of the frame 3073address for the current frame. @var{frameaddr} is the frame pointer 3074of the current frame. This is used for __builtin_frame_address. 3075You need only define this macro if the frame address is not the same 3076as the frame pointer. Most machines do not need to define it. 3077@end defmac 3078 3079@defmac RETURN_ADDR_RTX (@var{count}, @var{frameaddr}) 3080A C expression whose value is RTL representing the value of the return 3081address for the frame @var{count} steps up from the current frame, after 3082the prologue. @var{frameaddr} is the frame pointer of the @var{count} 3083frame, or the frame pointer of the @var{count} @minus{} 1 frame if 3084@code{RETURN_ADDR_IN_PREVIOUS_FRAME} is nonzero. 3085 3086The value of the expression must always be the correct address when 3087@var{count} is zero, but may be @code{NULL_RTX} if there is no way to 3088determine the return address of other frames. 3089@end defmac 3090 3091@defmac RETURN_ADDR_IN_PREVIOUS_FRAME 3092Define this macro to nonzero value if the return address of a particular 3093stack frame is accessed from the frame pointer of the previous stack 3094frame. The zero default for this macro is suitable for most ports. 3095@end defmac 3096 3097@defmac INCOMING_RETURN_ADDR_RTX 3098A C expression whose value is RTL representing the location of the 3099incoming return address at the beginning of any function, before the 3100prologue. This RTL is either a @code{REG}, indicating that the return 3101value is saved in @samp{REG}, or a @code{MEM} representing a location in 3102the stack. 3103 3104You only need to define this macro if you want to support call frame 3105debugging information like that provided by DWARF 2. 3106 3107If this RTL is a @code{REG}, you should also define 3108@code{DWARF_FRAME_RETURN_COLUMN} to @code{DWARF_FRAME_REGNUM (REGNO)}. 3109@end defmac 3110 3111@defmac DWARF_ALT_FRAME_RETURN_COLUMN 3112A C expression whose value is an integer giving a DWARF 2 column 3113number that may be used as an alternative return column. The column 3114must not correspond to any gcc hard register (that is, it must not 3115be in the range of @code{DWARF_FRAME_REGNUM}). 3116 3117This macro can be useful if @code{DWARF_FRAME_RETURN_COLUMN} is set to a 3118general register, but an alternative column needs to be used for signal 3119frames. Some targets have also used different frame return columns 3120over time. 3121@end defmac 3122 3123@defmac DWARF_ZERO_REG 3124A C expression whose value is an integer giving a DWARF 2 register 3125number that is considered to always have the value zero. This should 3126only be defined if the target has an architected zero register, and 3127someone decided it was a good idea to use that register number to 3128terminate the stack backtrace. New ports should avoid this. 3129@end defmac 3130 3131@deftypefn {Target Hook} void TARGET_DWARF_HANDLE_FRAME_UNSPEC (const char *@var{label}, rtx @var{pattern}, int @var{index}) 3132This target hook allows the backend to emit frame-related insns that 3133contain UNSPECs or UNSPEC_VOLATILEs. The DWARF 2 call frame debugging 3134info engine will invoke it on insns of the form 3135@smallexample 3136(set (reg) (unspec [@dots{}] UNSPEC_INDEX)) 3137@end smallexample 3138and 3139@smallexample 3140(set (reg) (unspec_volatile [@dots{}] UNSPECV_INDEX)). 3141@end smallexample 3142to let the backend emit the call frame instructions. @var{label} is 3143the CFI label attached to the insn, @var{pattern} is the pattern of 3144the insn and @var{index} is @code{UNSPEC_INDEX} or @code{UNSPECV_INDEX}. 3145@end deftypefn 3146 3147@deftypefn {Target Hook} {unsigned int} TARGET_DWARF_POLY_INDETERMINATE_VALUE (unsigned int @var{i}, unsigned int *@var{factor}, int *@var{offset}) 3148Express the value of @code{poly_int} indeterminate @var{i} as a DWARF 3149expression, with @var{i} counting from 1. Return the number of a DWARF 3150register @var{R} and set @samp{*@var{factor}} and @samp{*@var{offset}} such 3151that the value of the indeterminate is: 3152@smallexample 3153value_of(@var{R}) / @var{factor} - @var{offset} 3154@end smallexample 3155 3156A target only needs to define this hook if it sets 3157@samp{NUM_POLY_INT_COEFFS} to a value greater than 1. 3158@end deftypefn 3159 3160@defmac INCOMING_FRAME_SP_OFFSET 3161A C expression whose value is an integer giving the offset, in bytes, 3162from the value of the stack pointer register to the top of the stack 3163frame at the beginning of any function, before the prologue. The top of 3164the frame is defined to be the value of the stack pointer in the 3165previous frame, just before the call instruction. 3166 3167You only need to define this macro if you want to support call frame 3168debugging information like that provided by DWARF 2. 3169@end defmac 3170 3171@defmac DEFAULT_INCOMING_FRAME_SP_OFFSET 3172Like @code{INCOMING_FRAME_SP_OFFSET}, but must be the same for all 3173functions of the same ABI, and when using GAS @code{.cfi_*} directives 3174must also agree with the default CFI GAS emits. Define this macro 3175only if @code{INCOMING_FRAME_SP_OFFSET} can have different values 3176between different functions of the same ABI or when 3177@code{INCOMING_FRAME_SP_OFFSET} does not agree with GAS default CFI. 3178@end defmac 3179 3180@defmac ARG_POINTER_CFA_OFFSET (@var{fundecl}) 3181A C expression whose value is an integer giving the offset, in bytes, 3182from the argument pointer to the canonical frame address (cfa). The 3183final value should coincide with that calculated by 3184@code{INCOMING_FRAME_SP_OFFSET}. Which is unfortunately not usable 3185during virtual register instantiation. 3186 3187The default value for this macro is 3188@code{FIRST_PARM_OFFSET (fundecl) + crtl->args.pretend_args_size}, 3189which is correct for most machines; in general, the arguments are found 3190immediately before the stack frame. Note that this is not the case on 3191some targets that save registers into the caller's frame, such as SPARC 3192and rs6000, and so such targets need to define this macro. 3193 3194You only need to define this macro if the default is incorrect, and you 3195want to support call frame debugging information like that provided by 3196DWARF 2. 3197@end defmac 3198 3199@defmac FRAME_POINTER_CFA_OFFSET (@var{fundecl}) 3200If defined, a C expression whose value is an integer giving the offset 3201in bytes from the frame pointer to the canonical frame address (cfa). 3202The final value should coincide with that calculated by 3203@code{INCOMING_FRAME_SP_OFFSET}. 3204 3205Normally the CFA is calculated as an offset from the argument pointer, 3206via @code{ARG_POINTER_CFA_OFFSET}, but if the argument pointer is 3207variable due to the ABI, this may not be possible. If this macro is 3208defined, it implies that the virtual register instantiation should be 3209based on the frame pointer instead of the argument pointer. Only one 3210of @code{FRAME_POINTER_CFA_OFFSET} and @code{ARG_POINTER_CFA_OFFSET} 3211should be defined. 3212@end defmac 3213 3214@defmac CFA_FRAME_BASE_OFFSET (@var{fundecl}) 3215If defined, a C expression whose value is an integer giving the offset 3216in bytes from the canonical frame address (cfa) to the frame base used 3217in DWARF 2 debug information. The default is zero. A different value 3218may reduce the size of debug information on some ports. 3219@end defmac 3220 3221@node Exception Handling 3222@subsection Exception Handling Support 3223@cindex exception handling 3224 3225@defmac EH_RETURN_DATA_REGNO (@var{N}) 3226A C expression whose value is the @var{N}th register number used for 3227data by exception handlers, or @code{INVALID_REGNUM} if fewer than 3228@var{N} registers are usable. 3229 3230The exception handling library routines communicate with the exception 3231handlers via a set of agreed upon registers. Ideally these registers 3232should be call-clobbered; it is possible to use call-saved registers, 3233but may negatively impact code size. The target must support at least 32342 data registers, but should define 4 if there are enough free registers. 3235 3236You must define this macro if you want to support call frame exception 3237handling like that provided by DWARF 2. 3238@end defmac 3239 3240@defmac EH_RETURN_STACKADJ_RTX 3241A C expression whose value is RTL representing a location in which 3242to store a stack adjustment to be applied before function return. 3243This is used to unwind the stack to an exception handler's call frame. 3244It will be assigned zero on code paths that return normally. 3245 3246Typically this is a call-clobbered hard register that is otherwise 3247untouched by the epilogue, but could also be a stack slot. 3248 3249Do not define this macro if the stack pointer is saved and restored 3250by the regular prolog and epilog code in the call frame itself; in 3251this case, the exception handling library routines will update the 3252stack location to be restored in place. Otherwise, you must define 3253this macro if you want to support call frame exception handling like 3254that provided by DWARF 2. 3255@end defmac 3256 3257@defmac EH_RETURN_HANDLER_RTX 3258A C expression whose value is RTL representing a location in which 3259to store the address of an exception handler to which we should 3260return. It will not be assigned on code paths that return normally. 3261 3262Typically this is the location in the call frame at which the normal 3263return address is stored. For targets that return by popping an 3264address off the stack, this might be a memory address just below 3265the @emph{target} call frame rather than inside the current call 3266frame. If defined, @code{EH_RETURN_STACKADJ_RTX} will have already 3267been assigned, so it may be used to calculate the location of the 3268target call frame. 3269 3270Some targets have more complex requirements than storing to an 3271address calculable during initial code generation. In that case 3272the @code{eh_return} instruction pattern should be used instead. 3273 3274If you want to support call frame exception handling, you must 3275define either this macro or the @code{eh_return} instruction pattern. 3276@end defmac 3277 3278@defmac RETURN_ADDR_OFFSET 3279If defined, an integer-valued C expression for which rtl will be generated 3280to add it to the exception handler address before it is searched in the 3281exception handling tables, and to subtract it again from the address before 3282using it to return to the exception handler. 3283@end defmac 3284 3285@defmac ASM_PREFERRED_EH_DATA_FORMAT (@var{code}, @var{global}) 3286This macro chooses the encoding of pointers embedded in the exception 3287handling sections. If at all possible, this should be defined such 3288that the exception handling section will not require dynamic relocations, 3289and so may be read-only. 3290 3291@var{code} is 0 for data, 1 for code labels, 2 for function pointers. 3292@var{global} is true if the symbol may be affected by dynamic relocations. 3293The macro should return a combination of the @code{DW_EH_PE_*} defines 3294as found in @file{dwarf2.h}. 3295 3296If this macro is not defined, pointers will not be encoded but 3297represented directly. 3298@end defmac 3299 3300@defmac ASM_MAYBE_OUTPUT_ENCODED_ADDR_RTX (@var{file}, @var{encoding}, @var{size}, @var{addr}, @var{done}) 3301This macro allows the target to emit whatever special magic is required 3302to represent the encoding chosen by @code{ASM_PREFERRED_EH_DATA_FORMAT}. 3303Generic code takes care of pc-relative and indirect encodings; this must 3304be defined if the target uses text-relative or data-relative encodings. 3305 3306This is a C statement that branches to @var{done} if the format was 3307handled. @var{encoding} is the format chosen, @var{size} is the number 3308of bytes that the format occupies, @var{addr} is the @code{SYMBOL_REF} 3309to be emitted. 3310@end defmac 3311 3312@defmac MD_FALLBACK_FRAME_STATE_FOR (@var{context}, @var{fs}) 3313This macro allows the target to add CPU and operating system specific 3314code to the call-frame unwinder for use when there is no unwind data 3315available. The most common reason to implement this macro is to unwind 3316through signal frames. 3317 3318This macro is called from @code{uw_frame_state_for} in 3319@file{unwind-dw2.c}, @file{unwind-dw2-xtensa.c} and 3320@file{unwind-ia64.c}. @var{context} is an @code{_Unwind_Context}; 3321@var{fs} is an @code{_Unwind_FrameState}. Examine @code{context->ra} 3322for the address of the code being executed and @code{context->cfa} for 3323the stack pointer value. If the frame can be decoded, the register 3324save addresses should be updated in @var{fs} and the macro should 3325evaluate to @code{_URC_NO_REASON}. If the frame cannot be decoded, 3326the macro should evaluate to @code{_URC_END_OF_STACK}. 3327 3328For proper signal handling in Java this macro is accompanied by 3329@code{MAKE_THROW_FRAME}, defined in @file{libjava/include/*-signal.h} headers. 3330@end defmac 3331 3332@defmac MD_HANDLE_UNWABI (@var{context}, @var{fs}) 3333This macro allows the target to add operating system specific code to the 3334call-frame unwinder to handle the IA-64 @code{.unwabi} unwinding directive, 3335usually used for signal or interrupt frames. 3336 3337This macro is called from @code{uw_update_context} in libgcc's 3338@file{unwind-ia64.c}. @var{context} is an @code{_Unwind_Context}; 3339@var{fs} is an @code{_Unwind_FrameState}. Examine @code{fs->unwabi} 3340for the abi and context in the @code{.unwabi} directive. If the 3341@code{.unwabi} directive can be handled, the register save addresses should 3342be updated in @var{fs}. 3343@end defmac 3344 3345@defmac TARGET_USES_WEAK_UNWIND_INFO 3346A C expression that evaluates to true if the target requires unwind 3347info to be given comdat linkage. Define it to be @code{1} if comdat 3348linkage is necessary. The default is @code{0}. 3349@end defmac 3350 3351@node Stack Checking 3352@subsection Specifying How Stack Checking is Done 3353 3354GCC will check that stack references are within the boundaries of the 3355stack, if the option @option{-fstack-check} is specified, in one of 3356three ways: 3357 3358@enumerate 3359@item 3360If the value of the @code{STACK_CHECK_BUILTIN} macro is nonzero, GCC 3361will assume that you have arranged for full stack checking to be done 3362at appropriate places in the configuration files. GCC will not do 3363other special processing. 3364 3365@item 3366If @code{STACK_CHECK_BUILTIN} is zero and the value of the 3367@code{STACK_CHECK_STATIC_BUILTIN} macro is nonzero, GCC will assume 3368that you have arranged for static stack checking (checking of the 3369static stack frame of functions) to be done at appropriate places 3370in the configuration files. GCC will only emit code to do dynamic 3371stack checking (checking on dynamic stack allocations) using the third 3372approach below. 3373 3374@item 3375If neither of the above are true, GCC will generate code to periodically 3376``probe'' the stack pointer using the values of the macros defined below. 3377@end enumerate 3378 3379If neither STACK_CHECK_BUILTIN nor STACK_CHECK_STATIC_BUILTIN is defined, 3380GCC will change its allocation strategy for large objects if the option 3381@option{-fstack-check} is specified: they will always be allocated 3382dynamically if their size exceeds @code{STACK_CHECK_MAX_VAR_SIZE} bytes. 3383 3384@defmac STACK_CHECK_BUILTIN 3385A nonzero value if stack checking is done by the configuration files in a 3386machine-dependent manner. You should define this macro if stack checking 3387is required by the ABI of your machine or if you would like to do stack 3388checking in some more efficient way than the generic approach. The default 3389value of this macro is zero. 3390@end defmac 3391 3392@defmac STACK_CHECK_STATIC_BUILTIN 3393A nonzero value if static stack checking is done by the configuration files 3394in a machine-dependent manner. You should define this macro if you would 3395like to do static stack checking in some more efficient way than the generic 3396approach. The default value of this macro is zero. 3397@end defmac 3398 3399@defmac STACK_CHECK_PROBE_INTERVAL_EXP 3400An integer specifying the interval at which GCC must generate stack probe 3401instructions, defined as 2 raised to this integer. You will normally 3402define this macro so that the interval be no larger than the size of 3403the ``guard pages'' at the end of a stack area. The default value 3404of 12 (4096-byte interval) is suitable for most systems. 3405@end defmac 3406 3407@defmac STACK_CHECK_MOVING_SP 3408An integer which is nonzero if GCC should move the stack pointer page by page 3409when doing probes. This can be necessary on systems where the stack pointer 3410contains the bottom address of the memory area accessible to the executing 3411thread at any point in time. In this situation an alternate signal stack 3412is required in order to be able to recover from a stack overflow. The 3413default value of this macro is zero. 3414@end defmac 3415 3416@defmac STACK_CHECK_PROTECT 3417The number of bytes of stack needed to recover from a stack overflow, for 3418languages where such a recovery is supported. The default value of 4KB/8KB 3419with the @code{setjmp}/@code{longjmp}-based exception handling mechanism and 34208KB/12KB with other exception handling mechanisms should be adequate for most 3421architectures and operating systems. 3422@end defmac 3423 3424The following macros are relevant only if neither STACK_CHECK_BUILTIN 3425nor STACK_CHECK_STATIC_BUILTIN is defined; you can omit them altogether 3426in the opposite case. 3427 3428@defmac STACK_CHECK_MAX_FRAME_SIZE 3429The maximum size of a stack frame, in bytes. GCC will generate probe 3430instructions in non-leaf functions to ensure at least this many bytes of 3431stack are available. If a stack frame is larger than this size, stack 3432checking will not be reliable and GCC will issue a warning. The 3433default is chosen so that GCC only generates one instruction on most 3434systems. You should normally not change the default value of this macro. 3435@end defmac 3436 3437@defmac STACK_CHECK_FIXED_FRAME_SIZE 3438GCC uses this value to generate the above warning message. It 3439represents the amount of fixed frame used by a function, not including 3440space for any callee-saved registers, temporaries and user variables. 3441You need only specify an upper bound for this amount and will normally 3442use the default of four words. 3443@end defmac 3444 3445@defmac STACK_CHECK_MAX_VAR_SIZE 3446The maximum size, in bytes, of an object that GCC will place in the 3447fixed area of the stack frame when the user specifies 3448@option{-fstack-check}. 3449GCC computed the default from the values of the above macros and you will 3450normally not need to override that default. 3451@end defmac 3452 3453@deftypefn {Target Hook} bool TARGET_STACK_CLASH_PROTECTION_FINAL_DYNAMIC_PROBE (rtx @var{residual}) 3454Some targets make optimistic assumptions about the state of stack probing when they emit their prologues. On such targets a probe into the end of any dynamically allocated space is likely required for safety against stack clash style attacks. Define this variable to return nonzero if such a probe is required or zero otherwise. You need not define this macro if it would always have the value zero. 3455@end deftypefn 3456 3457@need 2000 3458@node Frame Registers 3459@subsection Registers That Address the Stack Frame 3460 3461@c prevent bad page break with this line 3462This discusses registers that address the stack frame. 3463 3464@defmac STACK_POINTER_REGNUM 3465The register number of the stack pointer register, which must also be a 3466fixed register according to @code{FIXED_REGISTERS}. On most machines, 3467the hardware determines which register this is. 3468@end defmac 3469 3470@defmac FRAME_POINTER_REGNUM 3471The register number of the frame pointer register, which is used to 3472access automatic variables in the stack frame. On some machines, the 3473hardware determines which register this is. On other machines, you can 3474choose any register you wish for this purpose. 3475@end defmac 3476 3477@defmac HARD_FRAME_POINTER_REGNUM 3478On some machines the offset between the frame pointer and starting 3479offset of the automatic variables is not known until after register 3480allocation has been done (for example, because the saved registers are 3481between these two locations). On those machines, define 3482@code{FRAME_POINTER_REGNUM} the number of a special, fixed register to 3483be used internally until the offset is known, and define 3484@code{HARD_FRAME_POINTER_REGNUM} to be the actual hard register number 3485used for the frame pointer. 3486 3487You should define this macro only in the very rare circumstances when it 3488is not possible to calculate the offset between the frame pointer and 3489the automatic variables until after register allocation has been 3490completed. When this macro is defined, you must also indicate in your 3491definition of @code{ELIMINABLE_REGS} how to eliminate 3492@code{FRAME_POINTER_REGNUM} into either @code{HARD_FRAME_POINTER_REGNUM} 3493or @code{STACK_POINTER_REGNUM}. 3494 3495Do not define this macro if it would be the same as 3496@code{FRAME_POINTER_REGNUM}. 3497@end defmac 3498 3499@defmac ARG_POINTER_REGNUM 3500The register number of the arg pointer register, which is used to access 3501the function's argument list. On some machines, this is the same as the 3502frame pointer register. On some machines, the hardware determines which 3503register this is. On other machines, you can choose any register you 3504wish for this purpose. If this is not the same register as the frame 3505pointer register, then you must mark it as a fixed register according to 3506@code{FIXED_REGISTERS}, or arrange to be able to eliminate it 3507(@pxref{Elimination}). 3508@end defmac 3509 3510@defmac HARD_FRAME_POINTER_IS_FRAME_POINTER 3511Define this to a preprocessor constant that is nonzero if 3512@code{hard_frame_pointer_rtx} and @code{frame_pointer_rtx} should be 3513the same. The default definition is @samp{(HARD_FRAME_POINTER_REGNUM 3514== FRAME_POINTER_REGNUM)}; you only need to define this macro if that 3515definition is not suitable for use in preprocessor conditionals. 3516@end defmac 3517 3518@defmac HARD_FRAME_POINTER_IS_ARG_POINTER 3519Define this to a preprocessor constant that is nonzero if 3520@code{hard_frame_pointer_rtx} and @code{arg_pointer_rtx} should be the 3521same. The default definition is @samp{(HARD_FRAME_POINTER_REGNUM == 3522ARG_POINTER_REGNUM)}; you only need to define this macro if that 3523definition is not suitable for use in preprocessor conditionals. 3524@end defmac 3525 3526@defmac RETURN_ADDRESS_POINTER_REGNUM 3527The register number of the return address pointer register, which is used to 3528access the current function's return address from the stack. On some 3529machines, the return address is not at a fixed offset from the frame 3530pointer or stack pointer or argument pointer. This register can be defined 3531to point to the return address on the stack, and then be converted by 3532@code{ELIMINABLE_REGS} into either the frame pointer or stack pointer. 3533 3534Do not define this macro unless there is no other way to get the return 3535address from the stack. 3536@end defmac 3537 3538@defmac STATIC_CHAIN_REGNUM 3539@defmacx STATIC_CHAIN_INCOMING_REGNUM 3540Register numbers used for passing a function's static chain pointer. If 3541register windows are used, the register number as seen by the called 3542function is @code{STATIC_CHAIN_INCOMING_REGNUM}, while the register 3543number as seen by the calling function is @code{STATIC_CHAIN_REGNUM}. If 3544these registers are the same, @code{STATIC_CHAIN_INCOMING_REGNUM} need 3545not be defined. 3546 3547The static chain register need not be a fixed register. 3548 3549If the static chain is passed in memory, these macros should not be 3550defined; instead, the @code{TARGET_STATIC_CHAIN} hook should be used. 3551@end defmac 3552 3553@deftypefn {Target Hook} rtx TARGET_STATIC_CHAIN (const_tree @var{fndecl_or_type}, bool @var{incoming_p}) 3554This hook replaces the use of @code{STATIC_CHAIN_REGNUM} et al for 3555targets that may use different static chain locations for different 3556nested functions. This may be required if the target has function 3557attributes that affect the calling conventions of the function and 3558those calling conventions use different static chain locations. 3559 3560The default version of this hook uses @code{STATIC_CHAIN_REGNUM} et al. 3561 3562If the static chain is passed in memory, this hook should be used to 3563provide rtx giving @code{mem} expressions that denote where they are stored. 3564Often the @code{mem} expression as seen by the caller will be at an offset 3565from the stack pointer and the @code{mem} expression as seen by the callee 3566will be at an offset from the frame pointer. 3567@findex stack_pointer_rtx 3568@findex frame_pointer_rtx 3569@findex arg_pointer_rtx 3570The variables @code{stack_pointer_rtx}, @code{frame_pointer_rtx}, and 3571@code{arg_pointer_rtx} will have been initialized and should be used 3572to refer to those items. 3573@end deftypefn 3574 3575@defmac DWARF_FRAME_REGISTERS 3576This macro specifies the maximum number of hard registers that can be 3577saved in a call frame. This is used to size data structures used in 3578DWARF2 exception handling. 3579 3580Prior to GCC 3.0, this macro was needed in order to establish a stable 3581exception handling ABI in the face of adding new hard registers for ISA 3582extensions. In GCC 3.0 and later, the EH ABI is insulated from changes 3583in the number of hard registers. Nevertheless, this macro can still be 3584used to reduce the runtime memory requirements of the exception handling 3585routines, which can be substantial if the ISA contains a lot of 3586registers that are not call-saved. 3587 3588If this macro is not defined, it defaults to 3589@code{FIRST_PSEUDO_REGISTER}. 3590@end defmac 3591 3592@defmac PRE_GCC3_DWARF_FRAME_REGISTERS 3593 3594This macro is similar to @code{DWARF_FRAME_REGISTERS}, but is provided 3595for backward compatibility in pre GCC 3.0 compiled code. 3596 3597If this macro is not defined, it defaults to 3598@code{DWARF_FRAME_REGISTERS}. 3599@end defmac 3600 3601@defmac DWARF_REG_TO_UNWIND_COLUMN (@var{regno}) 3602 3603Define this macro if the target's representation for dwarf registers 3604is different than the internal representation for unwind column. 3605Given a dwarf register, this macro should return the internal unwind 3606column number to use instead. 3607@end defmac 3608 3609@defmac DWARF_FRAME_REGNUM (@var{regno}) 3610 3611Define this macro if the target's representation for dwarf registers 3612used in .eh_frame or .debug_frame is different from that used in other 3613debug info sections. Given a GCC hard register number, this macro 3614should return the .eh_frame register number. The default is 3615@code{DBX_REGISTER_NUMBER (@var{regno})}. 3616 3617@end defmac 3618 3619@defmac DWARF2_FRAME_REG_OUT (@var{regno}, @var{for_eh}) 3620 3621Define this macro to map register numbers held in the call frame info 3622that GCC has collected using @code{DWARF_FRAME_REGNUM} to those that 3623should be output in .debug_frame (@code{@var{for_eh}} is zero) and 3624.eh_frame (@code{@var{for_eh}} is nonzero). The default is to 3625return @code{@var{regno}}. 3626 3627@end defmac 3628 3629@defmac REG_VALUE_IN_UNWIND_CONTEXT 3630 3631Define this macro if the target stores register values as 3632@code{_Unwind_Word} type in unwind context. It should be defined if 3633target register size is larger than the size of @code{void *}. The 3634default is to store register values as @code{void *} type. 3635 3636@end defmac 3637 3638@defmac ASSUME_EXTENDED_UNWIND_CONTEXT 3639 3640Define this macro to be 1 if the target always uses extended unwind 3641context with version, args_size and by_value fields. If it is undefined, 3642it will be defined to 1 when @code{REG_VALUE_IN_UNWIND_CONTEXT} is 3643defined and 0 otherwise. 3644 3645@end defmac 3646 3647@defmac DWARF_LAZY_REGISTER_VALUE (@var{regno}, @var{value}) 3648Define this macro if the target has pseudo DWARF registers whose 3649values need to be computed lazily on demand by the unwinder (such as when 3650referenced in a CFA expression). The macro returns true if @var{regno} 3651is such a register and stores its value in @samp{*@var{value}} if so. 3652@end defmac 3653 3654@node Elimination 3655@subsection Eliminating Frame Pointer and Arg Pointer 3656 3657@c prevent bad page break with this line 3658This is about eliminating the frame pointer and arg pointer. 3659 3660@deftypefn {Target Hook} bool TARGET_FRAME_POINTER_REQUIRED (void) 3661This target hook should return @code{true} if a function must have and use 3662a frame pointer. This target hook is called in the reload pass. If its return 3663value is @code{true} the function will have a frame pointer. 3664 3665This target hook can in principle examine the current function and decide 3666according to the facts, but on most machines the constant @code{false} or the 3667constant @code{true} suffices. Use @code{false} when the machine allows code 3668to be generated with no frame pointer, and doing so saves some time or space. 3669Use @code{true} when there is no possible advantage to avoiding a frame 3670pointer. 3671 3672In certain cases, the compiler does not know how to produce valid code 3673without a frame pointer. The compiler recognizes those cases and 3674automatically gives the function a frame pointer regardless of what 3675@code{targetm.frame_pointer_required} returns. You don't need to worry about 3676them. 3677 3678In a function that does not require a frame pointer, the frame pointer 3679register can be allocated for ordinary usage, unless you mark it as a 3680fixed register. See @code{FIXED_REGISTERS} for more information. 3681 3682Default return value is @code{false}. 3683@end deftypefn 3684 3685@defmac ELIMINABLE_REGS 3686This macro specifies a table of register pairs used to eliminate 3687unneeded registers that point into the stack frame. 3688 3689The definition of this macro is a list of structure initializations, each 3690of which specifies an original and replacement register. 3691 3692On some machines, the position of the argument pointer is not known until 3693the compilation is completed. In such a case, a separate hard register 3694must be used for the argument pointer. This register can be eliminated by 3695replacing it with either the frame pointer or the argument pointer, 3696depending on whether or not the frame pointer has been eliminated. 3697 3698In this case, you might specify: 3699@smallexample 3700#define ELIMINABLE_REGS \ 3701@{@{ARG_POINTER_REGNUM, STACK_POINTER_REGNUM@}, \ 3702 @{ARG_POINTER_REGNUM, FRAME_POINTER_REGNUM@}, \ 3703 @{FRAME_POINTER_REGNUM, STACK_POINTER_REGNUM@}@} 3704@end smallexample 3705 3706Note that the elimination of the argument pointer with the stack pointer is 3707specified first since that is the preferred elimination. 3708@end defmac 3709 3710@deftypefn {Target Hook} bool TARGET_CAN_ELIMINATE (const int @var{from_reg}, const int @var{to_reg}) 3711This target hook should return @code{true} if the compiler is allowed to 3712try to replace register number @var{from_reg} with register number 3713@var{to_reg}. This target hook will usually be @code{true}, since most of the 3714cases preventing register elimination are things that the compiler already 3715knows about. 3716 3717Default return value is @code{true}. 3718@end deftypefn 3719 3720@defmac INITIAL_ELIMINATION_OFFSET (@var{from-reg}, @var{to-reg}, @var{offset-var}) 3721This macro returns the initial difference between the specified pair 3722of registers. The value would be computed from information 3723such as the result of @code{get_frame_size ()} and the tables of 3724registers @code{df_regs_ever_live_p} and @code{call_used_regs}. 3725@end defmac 3726 3727@deftypefn {Target Hook} void TARGET_COMPUTE_FRAME_LAYOUT (void) 3728This target hook is called once each time the frame layout needs to be 3729recalculated. The calculations can be cached by the target and can then 3730be used by @code{INITIAL_ELIMINATION_OFFSET} instead of re-computing the 3731layout on every invocation of that hook. This is particularly useful 3732for targets that have an expensive frame layout function. Implementing 3733this callback is optional. 3734@end deftypefn 3735 3736@node Stack Arguments 3737@subsection Passing Function Arguments on the Stack 3738@cindex arguments on stack 3739@cindex stack arguments 3740 3741The macros in this section control how arguments are passed 3742on the stack. See the following section for other macros that 3743control passing certain arguments in registers. 3744 3745@deftypefn {Target Hook} bool TARGET_PROMOTE_PROTOTYPES (const_tree @var{fntype}) 3746This target hook returns @code{true} if an argument declared in a 3747prototype as an integral type smaller than @code{int} should actually be 3748passed as an @code{int}. In addition to avoiding errors in certain 3749cases of mismatch, it also makes for better code on certain machines. 3750The default is to not promote prototypes. 3751@end deftypefn 3752 3753@defmac PUSH_ARGS 3754A C expression. If nonzero, push insns will be used to pass 3755outgoing arguments. 3756If the target machine does not have a push instruction, set it to zero. 3757That directs GCC to use an alternate strategy: to 3758allocate the entire argument block and then store the arguments into 3759it. When @code{PUSH_ARGS} is nonzero, @code{PUSH_ROUNDING} must be defined too. 3760@end defmac 3761 3762@defmac PUSH_ARGS_REVERSED 3763A C expression. If nonzero, function arguments will be evaluated from 3764last to first, rather than from first to last. If this macro is not 3765defined, it defaults to @code{PUSH_ARGS} on targets where the stack 3766and args grow in opposite directions, and 0 otherwise. 3767@end defmac 3768 3769@defmac PUSH_ROUNDING (@var{npushed}) 3770A C expression that is the number of bytes actually pushed onto the 3771stack when an instruction attempts to push @var{npushed} bytes. 3772 3773On some machines, the definition 3774 3775@smallexample 3776#define PUSH_ROUNDING(BYTES) (BYTES) 3777@end smallexample 3778 3779@noindent 3780will suffice. But on other machines, instructions that appear 3781to push one byte actually push two bytes in an attempt to maintain 3782alignment. Then the definition should be 3783 3784@smallexample 3785#define PUSH_ROUNDING(BYTES) (((BYTES) + 1) & ~1) 3786@end smallexample 3787 3788If the value of this macro has a type, it should be an unsigned type. 3789@end defmac 3790 3791@findex outgoing_args_size 3792@findex crtl->outgoing_args_size 3793@defmac ACCUMULATE_OUTGOING_ARGS 3794A C expression. If nonzero, the maximum amount of space required for outgoing arguments 3795will be computed and placed into 3796@code{crtl->outgoing_args_size}. No space will be pushed 3797onto the stack for each call; instead, the function prologue should 3798increase the stack frame size by this amount. 3799 3800Setting both @code{PUSH_ARGS} and @code{ACCUMULATE_OUTGOING_ARGS} 3801is not proper. 3802@end defmac 3803 3804@defmac REG_PARM_STACK_SPACE (@var{fndecl}) 3805Define this macro if functions should assume that stack space has been 3806allocated for arguments even when their values are passed in 3807registers. 3808 3809The value of this macro is the size, in bytes, of the area reserved for 3810arguments passed in registers for the function represented by @var{fndecl}, 3811which can be zero if GCC is calling a library function. 3812The argument @var{fndecl} can be the FUNCTION_DECL, or the type itself 3813of the function. 3814 3815This space can be allocated by the caller, or be a part of the 3816machine-dependent stack frame: @code{OUTGOING_REG_PARM_STACK_SPACE} says 3817which. 3818@end defmac 3819@c above is overfull. not sure what to do. --mew 5feb93 did 3820@c something, not sure if it looks good. --mew 10feb93 3821 3822@defmac INCOMING_REG_PARM_STACK_SPACE (@var{fndecl}) 3823Like @code{REG_PARM_STACK_SPACE}, but for incoming register arguments. 3824Define this macro if space guaranteed when compiling a function body 3825is different to space required when making a call, a situation that 3826can arise with K&R style function definitions. 3827@end defmac 3828 3829@defmac OUTGOING_REG_PARM_STACK_SPACE (@var{fntype}) 3830Define this to a nonzero value if it is the responsibility of the 3831caller to allocate the area reserved for arguments passed in registers 3832when calling a function of @var{fntype}. @var{fntype} may be NULL 3833if the function called is a library function. 3834 3835If @code{ACCUMULATE_OUTGOING_ARGS} is defined, this macro controls 3836whether the space for these arguments counts in the value of 3837@code{crtl->outgoing_args_size}. 3838@end defmac 3839 3840@defmac STACK_PARMS_IN_REG_PARM_AREA 3841Define this macro if @code{REG_PARM_STACK_SPACE} is defined, but the 3842stack parameters don't skip the area specified by it. 3843@c i changed this, makes more sens and it should have taken care of the 3844@c overfull.. not as specific, tho. --mew 5feb93 3845 3846Normally, when a parameter is not passed in registers, it is placed on the 3847stack beyond the @code{REG_PARM_STACK_SPACE} area. Defining this macro 3848suppresses this behavior and causes the parameter to be passed on the 3849stack in its natural location. 3850@end defmac 3851 3852@deftypefn {Target Hook} poly_int64 TARGET_RETURN_POPS_ARGS (tree @var{fundecl}, tree @var{funtype}, poly_int64 @var{size}) 3853This target hook returns the number of bytes of its own arguments that 3854a function pops on returning, or 0 if the function pops no arguments 3855and the caller must therefore pop them all after the function returns. 3856 3857@var{fundecl} is a C variable whose value is a tree node that describes 3858the function in question. Normally it is a node of type 3859@code{FUNCTION_DECL} that describes the declaration of the function. 3860From this you can obtain the @code{DECL_ATTRIBUTES} of the function. 3861 3862@var{funtype} is a C variable whose value is a tree node that 3863describes the function in question. Normally it is a node of type 3864@code{FUNCTION_TYPE} that describes the data type of the function. 3865From this it is possible to obtain the data types of the value and 3866arguments (if known). 3867 3868When a call to a library function is being considered, @var{fundecl} 3869will contain an identifier node for the library function. Thus, if 3870you need to distinguish among various library functions, you can do so 3871by their names. Note that ``library function'' in this context means 3872a function used to perform arithmetic, whose name is known specially 3873in the compiler and was not mentioned in the C code being compiled. 3874 3875@var{size} is the number of bytes of arguments passed on the 3876stack. If a variable number of bytes is passed, it is zero, and 3877argument popping will always be the responsibility of the calling function. 3878 3879On the VAX, all functions always pop their arguments, so the definition 3880of this macro is @var{size}. On the 68000, using the standard 3881calling convention, no functions pop their arguments, so the value of 3882the macro is always 0 in this case. But an alternative calling 3883convention is available in which functions that take a fixed number of 3884arguments pop them but other functions (such as @code{printf}) pop 3885nothing (the caller pops all). When this convention is in use, 3886@var{funtype} is examined to determine whether a function takes a fixed 3887number of arguments. 3888@end deftypefn 3889 3890@defmac CALL_POPS_ARGS (@var{cum}) 3891A C expression that should indicate the number of bytes a call sequence 3892pops off the stack. It is added to the value of @code{RETURN_POPS_ARGS} 3893when compiling a function call. 3894 3895@var{cum} is the variable in which all arguments to the called function 3896have been accumulated. 3897 3898On certain architectures, such as the SH5, a call trampoline is used 3899that pops certain registers off the stack, depending on the arguments 3900that have been passed to the function. Since this is a property of the 3901call site, not of the called function, @code{RETURN_POPS_ARGS} is not 3902appropriate. 3903@end defmac 3904 3905@node Register Arguments 3906@subsection Passing Arguments in Registers 3907@cindex arguments in registers 3908@cindex registers arguments 3909 3910This section describes the macros which let you control how various 3911types of arguments are passed in registers or how they are arranged in 3912the stack. 3913 3914@deftypefn {Target Hook} rtx TARGET_FUNCTION_ARG (cumulative_args_t @var{ca}, machine_mode @var{mode}, const_tree @var{type}, bool @var{named}) 3915Return an RTX indicating whether a function argument is passed in a 3916register and if so, which register. 3917 3918The arguments are @var{ca}, which summarizes all the previous 3919arguments; @var{mode}, the machine mode of the argument; @var{type}, 3920the data type of the argument as a tree node or 0 if that is not known 3921(which happens for C support library functions); and @var{named}, 3922which is @code{true} for an ordinary argument and @code{false} for 3923nameless arguments that correspond to @samp{@dots{}} in the called 3924function's prototype. @var{type} can be an incomplete type if a 3925syntax error has previously occurred. 3926 3927The return value is usually either a @code{reg} RTX for the hard 3928register in which to pass the argument, or zero to pass the argument 3929on the stack. 3930 3931The return value can be a @code{const_int} which means argument is 3932passed in a target specific slot with specified number. Target hooks 3933should be used to store or load argument in such case. See 3934@code{TARGET_STORE_BOUNDS_FOR_ARG} and @code{TARGET_LOAD_BOUNDS_FOR_ARG} 3935for more information. 3936 3937The value of the expression can also be a @code{parallel} RTX@. This is 3938used when an argument is passed in multiple locations. The mode of the 3939@code{parallel} should be the mode of the entire argument. The 3940@code{parallel} holds any number of @code{expr_list} pairs; each one 3941describes where part of the argument is passed. In each 3942@code{expr_list} the first operand must be a @code{reg} RTX for the hard 3943register in which to pass this part of the argument, and the mode of the 3944register RTX indicates how large this part of the argument is. The 3945second operand of the @code{expr_list} is a @code{const_int} which gives 3946the offset in bytes into the entire argument of where this part starts. 3947As a special exception the first @code{expr_list} in the @code{parallel} 3948RTX may have a first operand of zero. This indicates that the entire 3949argument is also stored on the stack. 3950 3951The last time this hook is called, it is called with @code{MODE == 3952VOIDmode}, and its result is passed to the @code{call} or @code{call_value} 3953pattern as operands 2 and 3 respectively. 3954 3955@cindex @file{stdarg.h} and register arguments 3956The usual way to make the ISO library @file{stdarg.h} work on a 3957machine where some arguments are usually passed in registers, is to 3958cause nameless arguments to be passed on the stack instead. This is 3959done by making @code{TARGET_FUNCTION_ARG} return 0 whenever 3960@var{named} is @code{false}. 3961 3962@cindex @code{TARGET_MUST_PASS_IN_STACK}, and @code{TARGET_FUNCTION_ARG} 3963@cindex @code{REG_PARM_STACK_SPACE}, and @code{TARGET_FUNCTION_ARG} 3964You may use the hook @code{targetm.calls.must_pass_in_stack} 3965in the definition of this macro to determine if this argument is of a 3966type that must be passed in the stack. If @code{REG_PARM_STACK_SPACE} 3967is not defined and @code{TARGET_FUNCTION_ARG} returns nonzero for such an 3968argument, the compiler will abort. If @code{REG_PARM_STACK_SPACE} is 3969defined, the argument will be computed in the stack and then loaded into 3970a register. 3971@end deftypefn 3972 3973@deftypefn {Target Hook} bool TARGET_MUST_PASS_IN_STACK (machine_mode @var{mode}, const_tree @var{type}) 3974This target hook should return @code{true} if we should not pass @var{type} 3975solely in registers. The file @file{expr.h} defines a 3976definition that is usually appropriate, refer to @file{expr.h} for additional 3977documentation. 3978@end deftypefn 3979 3980@deftypefn {Target Hook} rtx TARGET_FUNCTION_INCOMING_ARG (cumulative_args_t @var{ca}, machine_mode @var{mode}, const_tree @var{type}, bool @var{named}) 3981Define this hook if the caller and callee on the target have different 3982views of where arguments are passed. Also define this hook if there are 3983functions that are never directly called, but are invoked by the hardware 3984and which have nonstandard calling conventions. 3985 3986In this case @code{TARGET_FUNCTION_ARG} computes the register in 3987which the caller passes the value, and 3988@code{TARGET_FUNCTION_INCOMING_ARG} should be defined in a similar 3989fashion to tell the function being called where the arguments will 3990arrive. 3991 3992@code{TARGET_FUNCTION_INCOMING_ARG} can also return arbitrary address 3993computation using hard register, which can be forced into a register, 3994so that it can be used to pass special arguments. 3995 3996If @code{TARGET_FUNCTION_INCOMING_ARG} is not defined, 3997@code{TARGET_FUNCTION_ARG} serves both purposes. 3998@end deftypefn 3999 4000@deftypefn {Target Hook} bool TARGET_USE_PSEUDO_PIC_REG (void) 4001This hook should return 1 in case pseudo register should be created 4002for pic_offset_table_rtx during function expand. 4003@end deftypefn 4004 4005@deftypefn {Target Hook} void TARGET_INIT_PIC_REG (void) 4006Perform a target dependent initialization of pic_offset_table_rtx. 4007This hook is called at the start of register allocation. 4008@end deftypefn 4009 4010@deftypefn {Target Hook} int TARGET_ARG_PARTIAL_BYTES (cumulative_args_t @var{cum}, machine_mode @var{mode}, tree @var{type}, bool @var{named}) 4011This target hook returns the number of bytes at the beginning of an 4012argument that must be put in registers. The value must be zero for 4013arguments that are passed entirely in registers or that are entirely 4014pushed on the stack. 4015 4016On some machines, certain arguments must be passed partially in 4017registers and partially in memory. On these machines, typically the 4018first few words of arguments are passed in registers, and the rest 4019on the stack. If a multi-word argument (a @code{double} or a 4020structure) crosses that boundary, its first few words must be passed 4021in registers and the rest must be pushed. This macro tells the 4022compiler when this occurs, and how many bytes should go in registers. 4023 4024@code{TARGET_FUNCTION_ARG} for these arguments should return the first 4025register to be used by the caller for this argument; likewise 4026@code{TARGET_FUNCTION_INCOMING_ARG}, for the called function. 4027@end deftypefn 4028 4029@deftypefn {Target Hook} bool TARGET_PASS_BY_REFERENCE (cumulative_args_t @var{cum}, machine_mode @var{mode}, const_tree @var{type}, bool @var{named}) 4030This target hook should return @code{true} if an argument at the 4031position indicated by @var{cum} should be passed by reference. This 4032predicate is queried after target independent reasons for being 4033passed by reference, such as @code{TREE_ADDRESSABLE (type)}. 4034 4035If the hook returns true, a copy of that argument is made in memory and a 4036pointer to the argument is passed instead of the argument itself. 4037The pointer is passed in whatever way is appropriate for passing a pointer 4038to that type. 4039@end deftypefn 4040 4041@deftypefn {Target Hook} bool TARGET_CALLEE_COPIES (cumulative_args_t @var{cum}, machine_mode @var{mode}, const_tree @var{type}, bool @var{named}) 4042The function argument described by the parameters to this hook is 4043known to be passed by reference. The hook should return true if the 4044function argument should be copied by the callee instead of copied 4045by the caller. 4046 4047For any argument for which the hook returns true, if it can be 4048determined that the argument is not modified, then a copy need 4049not be generated. 4050 4051The default version of this hook always returns false. 4052@end deftypefn 4053 4054@defmac CUMULATIVE_ARGS 4055A C type for declaring a variable that is used as the first argument 4056of @code{TARGET_FUNCTION_ARG} and other related values. For some 4057target machines, the type @code{int} suffices and can hold the number 4058of bytes of argument so far. 4059 4060There is no need to record in @code{CUMULATIVE_ARGS} anything about the 4061arguments that have been passed on the stack. The compiler has other 4062variables to keep track of that. For target machines on which all 4063arguments are passed on the stack, there is no need to store anything in 4064@code{CUMULATIVE_ARGS}; however, the data structure must exist and 4065should not be empty, so use @code{int}. 4066@end defmac 4067 4068@defmac OVERRIDE_ABI_FORMAT (@var{fndecl}) 4069If defined, this macro is called before generating any code for a 4070function, but after the @var{cfun} descriptor for the function has been 4071created. The back end may use this macro to update @var{cfun} to 4072reflect an ABI other than that which would normally be used by default. 4073If the compiler is generating code for a compiler-generated function, 4074@var{fndecl} may be @code{NULL}. 4075@end defmac 4076 4077@defmac INIT_CUMULATIVE_ARGS (@var{cum}, @var{fntype}, @var{libname}, @var{fndecl}, @var{n_named_args}) 4078A C statement (sans semicolon) for initializing the variable 4079@var{cum} for the state at the beginning of the argument list. The 4080variable has type @code{CUMULATIVE_ARGS}. The value of @var{fntype} 4081is the tree node for the data type of the function which will receive 4082the args, or 0 if the args are to a compiler support library function. 4083For direct calls that are not libcalls, @var{fndecl} contain the 4084declaration node of the function. @var{fndecl} is also set when 4085@code{INIT_CUMULATIVE_ARGS} is used to find arguments for the function 4086being compiled. @var{n_named_args} is set to the number of named 4087arguments, including a structure return address if it is passed as a 4088parameter, when making a call. When processing incoming arguments, 4089@var{n_named_args} is set to @minus{}1. 4090 4091When processing a call to a compiler support library function, 4092@var{libname} identifies which one. It is a @code{symbol_ref} rtx which 4093contains the name of the function, as a string. @var{libname} is 0 when 4094an ordinary C function call is being processed. Thus, each time this 4095macro is called, either @var{libname} or @var{fntype} is nonzero, but 4096never both of them at once. 4097@end defmac 4098 4099@defmac INIT_CUMULATIVE_LIBCALL_ARGS (@var{cum}, @var{mode}, @var{libname}) 4100Like @code{INIT_CUMULATIVE_ARGS} but only used for outgoing libcalls, 4101it gets a @code{MODE} argument instead of @var{fntype}, that would be 4102@code{NULL}. @var{indirect} would always be zero, too. If this macro 4103is not defined, @code{INIT_CUMULATIVE_ARGS (cum, NULL_RTX, libname, 41040)} is used instead. 4105@end defmac 4106 4107@defmac INIT_CUMULATIVE_INCOMING_ARGS (@var{cum}, @var{fntype}, @var{libname}) 4108Like @code{INIT_CUMULATIVE_ARGS} but overrides it for the purposes of 4109finding the arguments for the function being compiled. If this macro is 4110undefined, @code{INIT_CUMULATIVE_ARGS} is used instead. 4111 4112The value passed for @var{libname} is always 0, since library routines 4113with special calling conventions are never compiled with GCC@. The 4114argument @var{libname} exists for symmetry with 4115@code{INIT_CUMULATIVE_ARGS}. 4116@c could use "this macro" in place of @code{INIT_CUMULATIVE_ARGS}, maybe. 4117@c --mew 5feb93 i switched the order of the sentences. --mew 10feb93 4118@end defmac 4119 4120@deftypefn {Target Hook} void TARGET_FUNCTION_ARG_ADVANCE (cumulative_args_t @var{ca}, machine_mode @var{mode}, const_tree @var{type}, bool @var{named}) 4121This hook updates the summarizer variable pointed to by @var{ca} to 4122advance past an argument in the argument list. The values @var{mode}, 4123@var{type} and @var{named} describe that argument. Once this is done, 4124the variable @var{cum} is suitable for analyzing the @emph{following} 4125argument with @code{TARGET_FUNCTION_ARG}, etc. 4126 4127This hook need not do anything if the argument in question was passed 4128on the stack. The compiler knows how to track the amount of stack space 4129used for arguments without any special help. 4130@end deftypefn 4131 4132@deftypefn {Target Hook} HOST_WIDE_INT TARGET_FUNCTION_ARG_OFFSET (machine_mode @var{mode}, const_tree @var{type}) 4133This hook returns the number of bytes to add to the offset of an 4134argument of type @var{type} and mode @var{mode} when passed in memory. 4135This is needed for the SPU, which passes @code{char} and @code{short} 4136arguments in the preferred slot that is in the middle of the quad word 4137instead of starting at the top. The default implementation returns 0. 4138@end deftypefn 4139 4140@deftypefn {Target Hook} pad_direction TARGET_FUNCTION_ARG_PADDING (machine_mode @var{mode}, const_tree @var{type}) 4141This hook determines whether, and in which direction, to pad out 4142an argument of mode @var{mode} and type @var{type}. It returns 4143@code{PAD_UPWARD} to insert padding above the argument, @code{PAD_DOWNWARD} 4144to insert padding below the argument, or @code{PAD_NONE} to inhibit padding. 4145 4146The @emph{amount} of padding is not controlled by this hook, but by 4147@code{TARGET_FUNCTION_ARG_ROUND_BOUNDARY}. It is always just enough 4148to reach the next multiple of that boundary. 4149 4150This hook has a default definition that is right for most systems. 4151For little-endian machines, the default is to pad upward. For 4152big-endian machines, the default is to pad downward for an argument of 4153constant size shorter than an @code{int}, and upward otherwise. 4154@end deftypefn 4155 4156@defmac PAD_VARARGS_DOWN 4157If defined, a C expression which determines whether the default 4158implementation of va_arg will attempt to pad down before reading the 4159next argument, if that argument is smaller than its aligned space as 4160controlled by @code{PARM_BOUNDARY}. If this macro is not defined, all such 4161arguments are padded down if @code{BYTES_BIG_ENDIAN} is true. 4162@end defmac 4163 4164@defmac BLOCK_REG_PADDING (@var{mode}, @var{type}, @var{first}) 4165Specify padding for the last element of a block move between registers and 4166memory. @var{first} is nonzero if this is the only element. Defining this 4167macro allows better control of register function parameters on big-endian 4168machines, without using @code{PARALLEL} rtl. In particular, 4169@code{MUST_PASS_IN_STACK} need not test padding and mode of types in 4170registers, as there is no longer a "wrong" part of a register; For example, 4171a three byte aggregate may be passed in the high part of a register if so 4172required. 4173@end defmac 4174 4175@deftypefn {Target Hook} {unsigned int} TARGET_FUNCTION_ARG_BOUNDARY (machine_mode @var{mode}, const_tree @var{type}) 4176This hook returns the alignment boundary, in bits, of an argument 4177with the specified mode and type. The default hook returns 4178@code{PARM_BOUNDARY} for all arguments. 4179@end deftypefn 4180 4181@deftypefn {Target Hook} {unsigned int} TARGET_FUNCTION_ARG_ROUND_BOUNDARY (machine_mode @var{mode}, const_tree @var{type}) 4182Normally, the size of an argument is rounded up to @code{PARM_BOUNDARY}, 4183which is the default value for this hook. You can define this hook to 4184return a different value if an argument size must be rounded to a larger 4185value. 4186@end deftypefn 4187 4188@defmac FUNCTION_ARG_REGNO_P (@var{regno}) 4189A C expression that is nonzero if @var{regno} is the number of a hard 4190register in which function arguments are sometimes passed. This does 4191@emph{not} include implicit arguments such as the static chain and 4192the structure-value address. On many machines, no registers can be 4193used for this purpose since all function arguments are pushed on the 4194stack. 4195@end defmac 4196 4197@deftypefn {Target Hook} bool TARGET_SPLIT_COMPLEX_ARG (const_tree @var{type}) 4198This hook should return true if parameter of type @var{type} are passed 4199as two scalar parameters. By default, GCC will attempt to pack complex 4200arguments into the target's word size. Some ABIs require complex arguments 4201to be split and treated as their individual components. For example, on 4202AIX64, complex floats should be passed in a pair of floating point 4203registers, even though a complex float would fit in one 64-bit floating 4204point register. 4205 4206The default value of this hook is @code{NULL}, which is treated as always 4207false. 4208@end deftypefn 4209 4210@deftypefn {Target Hook} tree TARGET_BUILD_BUILTIN_VA_LIST (void) 4211This hook returns a type node for @code{va_list} for the target. 4212The default version of the hook returns @code{void*}. 4213@end deftypefn 4214 4215@deftypefn {Target Hook} int TARGET_ENUM_VA_LIST_P (int @var{idx}, const char **@var{pname}, tree *@var{ptree}) 4216This target hook is used in function @code{c_common_nodes_and_builtins} 4217to iterate through the target specific builtin types for va_list. The 4218variable @var{idx} is used as iterator. @var{pname} has to be a pointer 4219to a @code{const char *} and @var{ptree} a pointer to a @code{tree} typed 4220variable. 4221The arguments @var{pname} and @var{ptree} are used to store the result of 4222this macro and are set to the name of the va_list builtin type and its 4223internal type. 4224If the return value of this macro is zero, then there is no more element. 4225Otherwise the @var{IDX} should be increased for the next call of this 4226macro to iterate through all types. 4227@end deftypefn 4228 4229@deftypefn {Target Hook} tree TARGET_FN_ABI_VA_LIST (tree @var{fndecl}) 4230This hook returns the va_list type of the calling convention specified by 4231@var{fndecl}. 4232The default version of this hook returns @code{va_list_type_node}. 4233@end deftypefn 4234 4235@deftypefn {Target Hook} tree TARGET_CANONICAL_VA_LIST_TYPE (tree @var{type}) 4236This hook returns the va_list type of the calling convention specified by the 4237type of @var{type}. If @var{type} is not a valid va_list type, it returns 4238@code{NULL_TREE}. 4239@end deftypefn 4240 4241@deftypefn {Target Hook} tree TARGET_GIMPLIFY_VA_ARG_EXPR (tree @var{valist}, tree @var{type}, gimple_seq *@var{pre_p}, gimple_seq *@var{post_p}) 4242This hook performs target-specific gimplification of 4243@code{VA_ARG_EXPR}. The first two parameters correspond to the 4244arguments to @code{va_arg}; the latter two are as in 4245@code{gimplify.c:gimplify_expr}. 4246@end deftypefn 4247 4248@deftypefn {Target Hook} bool TARGET_VALID_POINTER_MODE (scalar_int_mode @var{mode}) 4249Define this to return nonzero if the port can handle pointers 4250with machine mode @var{mode}. The default version of this 4251hook returns true for both @code{ptr_mode} and @code{Pmode}. 4252@end deftypefn 4253 4254@deftypefn {Target Hook} bool TARGET_REF_MAY_ALIAS_ERRNO (struct ao_ref *@var{ref}) 4255Define this to return nonzero if the memory reference @var{ref} may alias with the system C library errno location. The default version of this hook assumes the system C library errno location is either a declaration of type int or accessed by dereferencing a pointer to int. 4256@end deftypefn 4257 4258@deftypefn {Target Hook} bool TARGET_SCALAR_MODE_SUPPORTED_P (scalar_mode @var{mode}) 4259Define this to return nonzero if the port is prepared to handle 4260insns involving scalar mode @var{mode}. For a scalar mode to be 4261considered supported, all the basic arithmetic and comparisons 4262must work. 4263 4264The default version of this hook returns true for any mode 4265required to handle the basic C types (as defined by the port). 4266Included here are the double-word arithmetic supported by the 4267code in @file{optabs.c}. 4268@end deftypefn 4269 4270@deftypefn {Target Hook} bool TARGET_VECTOR_MODE_SUPPORTED_P (machine_mode @var{mode}) 4271Define this to return nonzero if the port is prepared to handle 4272insns involving vector mode @var{mode}. At the very least, it 4273must have move patterns for this mode. 4274@end deftypefn 4275 4276@deftypefn {Target Hook} opt_machine_mode TARGET_ARRAY_MODE (machine_mode @var{mode}, unsigned HOST_WIDE_INT @var{nelems}) 4277Return the mode that GCC should use for an array that has 4278@var{nelems} elements, with each element having mode @var{mode}. 4279Return no mode if the target has no special requirements. In the 4280latter case, GCC looks for an integer mode of the appropriate size 4281if available and uses BLKmode otherwise. Usually the search for the 4282integer mode is limited to @code{MAX_FIXED_MODE_SIZE}, but the 4283@code{TARGET_ARRAY_MODE_SUPPORTED_P} hook allows a larger mode to be 4284used in specific cases. 4285 4286The main use of this hook is to specify that an array of vectors should 4287also have a vector mode. The default implementation returns no mode. 4288@end deftypefn 4289 4290@deftypefn {Target Hook} bool TARGET_ARRAY_MODE_SUPPORTED_P (machine_mode @var{mode}, unsigned HOST_WIDE_INT @var{nelems}) 4291Return true if GCC should try to use a scalar mode to store an array 4292of @var{nelems} elements, given that each element has mode @var{mode}. 4293Returning true here overrides the usual @code{MAX_FIXED_MODE} limit 4294and allows GCC to use any defined integer mode. 4295 4296One use of this hook is to support vector load and store operations 4297that operate on several homogeneous vectors. For example, ARM NEON 4298has operations like: 4299 4300@smallexample 4301int8x8x3_t vld3_s8 (const int8_t *) 4302@end smallexample 4303 4304where the return type is defined as: 4305 4306@smallexample 4307typedef struct int8x8x3_t 4308@{ 4309 int8x8_t val[3]; 4310@} int8x8x3_t; 4311@end smallexample 4312 4313If this hook allows @code{val} to have a scalar mode, then 4314@code{int8x8x3_t} can have the same mode. GCC can then store 4315@code{int8x8x3_t}s in registers rather than forcing them onto the stack. 4316@end deftypefn 4317 4318@deftypefn {Target Hook} bool TARGET_LIBGCC_FLOATING_MODE_SUPPORTED_P (scalar_float_mode @var{mode}) 4319Define this to return nonzero if libgcc provides support for the 4320floating-point mode @var{mode}, which is known to pass 4321@code{TARGET_SCALAR_MODE_SUPPORTED_P}. The default version of this 4322hook returns true for all of @code{SFmode}, @code{DFmode}, 4323@code{XFmode} and @code{TFmode}, if such modes exist. 4324@end deftypefn 4325 4326@deftypefn {Target Hook} opt_scalar_float_mode TARGET_FLOATN_MODE (int @var{n}, bool @var{extended}) 4327Define this to return the machine mode to use for the type 4328@code{_Float@var{n}}, if @var{extended} is false, or the type 4329@code{_Float@var{n}x}, if @var{extended} is true. If such a type is not 4330supported, return @code{opt_scalar_float_mode ()}. The default version of 4331this hook returns @code{SFmode} for @code{_Float32}, @code{DFmode} for 4332@code{_Float64} and @code{_Float32x} and @code{TFmode} for 4333@code{_Float128}, if those modes exist and satisfy the requirements for 4334those types and pass @code{TARGET_SCALAR_MODE_SUPPORTED_P} and 4335@code{TARGET_LIBGCC_FLOATING_MODE_SUPPORTED_P}; for @code{_Float64x}, it 4336returns the first of @code{XFmode} and @code{TFmode} that exists and 4337satisfies the same requirements; for other types, it returns 4338@code{opt_scalar_float_mode ()}. The hook is only called for values 4339of @var{n} and @var{extended} that are valid according to 4340ISO/IEC TS 18661-3:2015; that is, @var{n} is one of 32, 64, 128, or, 4341if @var{extended} is false, 16 or greater than 128 and a multiple of 32. 4342@end deftypefn 4343 4344@deftypefn {Target Hook} bool TARGET_FLOATN_BUILTIN_P (int @var{func}) 4345Define this to return true if the @code{_Float@var{n}} and 4346@code{_Float@var{n}x} built-in functions should implicitly enable the 4347built-in function without the @code{__builtin_} prefix in addition to the 4348normal built-in function with the @code{__builtin_} prefix. The default is 4349to only enable built-in functions without the @code{__builtin_} prefix for 4350the GNU C langauge. In strict ANSI/ISO mode, the built-in function without 4351the @code{__builtin_} prefix is not enabled. The argument @code{FUNC} is the 4352@code{enum built_in_function} id of the function to be enabled. 4353@end deftypefn 4354 4355@deftypefn {Target Hook} bool TARGET_SMALL_REGISTER_CLASSES_FOR_MODE_P (machine_mode @var{mode}) 4356Define this to return nonzero for machine modes for which the port has 4357small register classes. If this target hook returns nonzero for a given 4358@var{mode}, the compiler will try to minimize the lifetime of registers 4359in @var{mode}. The hook may be called with @code{VOIDmode} as argument. 4360In this case, the hook is expected to return nonzero if it returns nonzero 4361for any mode. 4362 4363On some machines, it is risky to let hard registers live across arbitrary 4364insns. Typically, these machines have instructions that require values 4365to be in specific registers (like an accumulator), and reload will fail 4366if the required hard register is used for another purpose across such an 4367insn. 4368 4369Passes before reload do not know which hard registers will be used 4370in an instruction, but the machine modes of the registers set or used in 4371the instruction are already known. And for some machines, register 4372classes are small for, say, integer registers but not for floating point 4373registers. For example, the AMD x86-64 architecture requires specific 4374registers for the legacy x86 integer instructions, but there are many 4375SSE registers for floating point operations. On such targets, a good 4376strategy may be to return nonzero from this hook for @code{INTEGRAL_MODE_P} 4377machine modes but zero for the SSE register classes. 4378 4379The default version of this hook returns false for any mode. It is always 4380safe to redefine this hook to return with a nonzero value. But if you 4381unnecessarily define it, you will reduce the amount of optimizations 4382that can be performed in some cases. If you do not define this hook 4383to return a nonzero value when it is required, the compiler will run out 4384of spill registers and print a fatal error message. 4385@end deftypefn 4386 4387@node Scalar Return 4388@subsection How Scalar Function Values Are Returned 4389@cindex return values in registers 4390@cindex values, returned by functions 4391@cindex scalars, returned as values 4392 4393This section discusses the macros that control returning scalars as 4394values---values that can fit in registers. 4395 4396@deftypefn {Target Hook} rtx TARGET_FUNCTION_VALUE (const_tree @var{ret_type}, const_tree @var{fn_decl_or_type}, bool @var{outgoing}) 4397 4398Define this to return an RTX representing the place where a function 4399returns or receives a value of data type @var{ret_type}, a tree node 4400representing a data type. @var{fn_decl_or_type} is a tree node 4401representing @code{FUNCTION_DECL} or @code{FUNCTION_TYPE} of a 4402function being called. If @var{outgoing} is false, the hook should 4403compute the register in which the caller will see the return value. 4404Otherwise, the hook should return an RTX representing the place where 4405a function returns a value. 4406 4407On many machines, only @code{TYPE_MODE (@var{ret_type})} is relevant. 4408(Actually, on most machines, scalar values are returned in the same 4409place regardless of mode.) The value of the expression is usually a 4410@code{reg} RTX for the hard register where the return value is stored. 4411The value can also be a @code{parallel} RTX, if the return value is in 4412multiple places. See @code{TARGET_FUNCTION_ARG} for an explanation of the 4413@code{parallel} form. Note that the callee will populate every 4414location specified in the @code{parallel}, but if the first element of 4415the @code{parallel} contains the whole return value, callers will use 4416that element as the canonical location and ignore the others. The m68k 4417port uses this type of @code{parallel} to return pointers in both 4418@samp{%a0} (the canonical location) and @samp{%d0}. 4419 4420If @code{TARGET_PROMOTE_FUNCTION_RETURN} returns true, you must apply 4421the same promotion rules specified in @code{PROMOTE_MODE} if 4422@var{valtype} is a scalar type. 4423 4424If the precise function being called is known, @var{func} is a tree 4425node (@code{FUNCTION_DECL}) for it; otherwise, @var{func} is a null 4426pointer. This makes it possible to use a different value-returning 4427convention for specific functions when all their calls are 4428known. 4429 4430Some target machines have ``register windows'' so that the register in 4431which a function returns its value is not the same as the one in which 4432the caller sees the value. For such machines, you should return 4433different RTX depending on @var{outgoing}. 4434 4435@code{TARGET_FUNCTION_VALUE} is not used for return values with 4436aggregate data types, because these are returned in another way. See 4437@code{TARGET_STRUCT_VALUE_RTX} and related macros, below. 4438@end deftypefn 4439 4440@defmac FUNCTION_VALUE (@var{valtype}, @var{func}) 4441This macro has been deprecated. Use @code{TARGET_FUNCTION_VALUE} for 4442a new target instead. 4443@end defmac 4444 4445@defmac LIBCALL_VALUE (@var{mode}) 4446A C expression to create an RTX representing the place where a library 4447function returns a value of mode @var{mode}. 4448 4449Note that ``library function'' in this context means a compiler 4450support routine, used to perform arithmetic, whose name is known 4451specially by the compiler and was not mentioned in the C code being 4452compiled. 4453@end defmac 4454 4455@deftypefn {Target Hook} rtx TARGET_LIBCALL_VALUE (machine_mode @var{mode}, const_rtx @var{fun}) 4456Define this hook if the back-end needs to know the name of the libcall 4457function in order to determine where the result should be returned. 4458 4459The mode of the result is given by @var{mode} and the name of the called 4460library function is given by @var{fun}. The hook should return an RTX 4461representing the place where the library function result will be returned. 4462 4463If this hook is not defined, then LIBCALL_VALUE will be used. 4464@end deftypefn 4465 4466@defmac FUNCTION_VALUE_REGNO_P (@var{regno}) 4467A C expression that is nonzero if @var{regno} is the number of a hard 4468register in which the values of called function may come back. 4469 4470A register whose use for returning values is limited to serving as the 4471second of a pair (for a value of type @code{double}, say) need not be 4472recognized by this macro. So for most machines, this definition 4473suffices: 4474 4475@smallexample 4476#define FUNCTION_VALUE_REGNO_P(N) ((N) == 0) 4477@end smallexample 4478 4479If the machine has register windows, so that the caller and the called 4480function use different registers for the return value, this macro 4481should recognize only the caller's register numbers. 4482 4483This macro has been deprecated. Use @code{TARGET_FUNCTION_VALUE_REGNO_P} 4484for a new target instead. 4485@end defmac 4486 4487@deftypefn {Target Hook} bool TARGET_FUNCTION_VALUE_REGNO_P (const unsigned int @var{regno}) 4488A target hook that return @code{true} if @var{regno} is the number of a hard 4489register in which the values of called function may come back. 4490 4491A register whose use for returning values is limited to serving as the 4492second of a pair (for a value of type @code{double}, say) need not be 4493recognized by this target hook. 4494 4495If the machine has register windows, so that the caller and the called 4496function use different registers for the return value, this target hook 4497should recognize only the caller's register numbers. 4498 4499If this hook is not defined, then FUNCTION_VALUE_REGNO_P will be used. 4500@end deftypefn 4501 4502@defmac APPLY_RESULT_SIZE 4503Define this macro if @samp{untyped_call} and @samp{untyped_return} 4504need more space than is implied by @code{FUNCTION_VALUE_REGNO_P} for 4505saving and restoring an arbitrary return value. 4506@end defmac 4507 4508@deftypevr {Target Hook} bool TARGET_OMIT_STRUCT_RETURN_REG 4509Normally, when a function returns a structure by memory, the address 4510is passed as an invisible pointer argument, but the compiler also 4511arranges to return the address from the function like it would a normal 4512pointer return value. Define this to true if that behavior is 4513undesirable on your target. 4514@end deftypevr 4515 4516@deftypefn {Target Hook} bool TARGET_RETURN_IN_MSB (const_tree @var{type}) 4517This hook should return true if values of type @var{type} are returned 4518at the most significant end of a register (in other words, if they are 4519padded at the least significant end). You can assume that @var{type} 4520is returned in a register; the caller is required to check this. 4521 4522Note that the register provided by @code{TARGET_FUNCTION_VALUE} must 4523be able to hold the complete return value. For example, if a 1-, 2- 4524or 3-byte structure is returned at the most significant end of a 45254-byte register, @code{TARGET_FUNCTION_VALUE} should provide an 4526@code{SImode} rtx. 4527@end deftypefn 4528 4529@node Aggregate Return 4530@subsection How Large Values Are Returned 4531@cindex aggregates as return values 4532@cindex large return values 4533@cindex returning aggregate values 4534@cindex structure value address 4535 4536When a function value's mode is @code{BLKmode} (and in some other 4537cases), the value is not returned according to 4538@code{TARGET_FUNCTION_VALUE} (@pxref{Scalar Return}). Instead, the 4539caller passes the address of a block of memory in which the value 4540should be stored. This address is called the @dfn{structure value 4541address}. 4542 4543This section describes how to control returning structure values in 4544memory. 4545 4546@deftypefn {Target Hook} bool TARGET_RETURN_IN_MEMORY (const_tree @var{type}, const_tree @var{fntype}) 4547This target hook should return a nonzero value to say to return the 4548function value in memory, just as large structures are always returned. 4549Here @var{type} will be the data type of the value, and @var{fntype} 4550will be the type of the function doing the returning, or @code{NULL} for 4551libcalls. 4552 4553Note that values of mode @code{BLKmode} must be explicitly handled 4554by this function. Also, the option @option{-fpcc-struct-return} 4555takes effect regardless of this macro. On most systems, it is 4556possible to leave the hook undefined; this causes a default 4557definition to be used, whose value is the constant 1 for @code{BLKmode} 4558values, and 0 otherwise. 4559 4560Do not use this hook to indicate that structures and unions should always 4561be returned in memory. You should instead use @code{DEFAULT_PCC_STRUCT_RETURN} 4562to indicate this. 4563@end deftypefn 4564 4565@defmac DEFAULT_PCC_STRUCT_RETURN 4566Define this macro to be 1 if all structure and union return values must be 4567in memory. Since this results in slower code, this should be defined 4568only if needed for compatibility with other compilers or with an ABI@. 4569If you define this macro to be 0, then the conventions used for structure 4570and union return values are decided by the @code{TARGET_RETURN_IN_MEMORY} 4571target hook. 4572 4573If not defined, this defaults to the value 1. 4574@end defmac 4575 4576@deftypefn {Target Hook} rtx TARGET_STRUCT_VALUE_RTX (tree @var{fndecl}, int @var{incoming}) 4577This target hook should return the location of the structure value 4578address (normally a @code{mem} or @code{reg}), or 0 if the address is 4579passed as an ``invisible'' first argument. Note that @var{fndecl} may 4580be @code{NULL}, for libcalls. You do not need to define this target 4581hook if the address is always passed as an ``invisible'' first 4582argument. 4583 4584On some architectures the place where the structure value address 4585is found by the called function is not the same place that the 4586caller put it. This can be due to register windows, or it could 4587be because the function prologue moves it to a different place. 4588@var{incoming} is @code{1} or @code{2} when the location is needed in 4589the context of the called function, and @code{0} in the context of 4590the caller. 4591 4592If @var{incoming} is nonzero and the address is to be found on the 4593stack, return a @code{mem} which refers to the frame pointer. If 4594@var{incoming} is @code{2}, the result is being used to fetch the 4595structure value address at the beginning of a function. If you need 4596to emit adjusting code, you should do it at this point. 4597@end deftypefn 4598 4599@defmac PCC_STATIC_STRUCT_RETURN 4600Define this macro if the usual system convention on the target machine 4601for returning structures and unions is for the called function to return 4602the address of a static variable containing the value. 4603 4604Do not define this if the usual system convention is for the caller to 4605pass an address to the subroutine. 4606 4607This macro has effect in @option{-fpcc-struct-return} mode, but it does 4608nothing when you use @option{-freg-struct-return} mode. 4609@end defmac 4610 4611@deftypefn {Target Hook} fixed_size_mode TARGET_GET_RAW_RESULT_MODE (int @var{regno}) 4612This target hook returns the mode to be used when accessing raw return registers in @code{__builtin_return}. Define this macro if the value in @var{reg_raw_mode} is not correct. 4613@end deftypefn 4614 4615@deftypefn {Target Hook} fixed_size_mode TARGET_GET_RAW_ARG_MODE (int @var{regno}) 4616This target hook returns the mode to be used when accessing raw argument registers in @code{__builtin_apply_args}. Define this macro if the value in @var{reg_raw_mode} is not correct. 4617@end deftypefn 4618 4619@deftypefn {Target Hook} bool TARGET_EMPTY_RECORD_P (const_tree @var{type}) 4620This target hook returns true if the type is an empty record. The default 4621is to return @code{false}. 4622@end deftypefn 4623 4624@deftypefn {Target Hook} void TARGET_WARN_PARAMETER_PASSING_ABI (cumulative_args_t @var{ca}, tree @var{type}) 4625This target hook warns about the change in empty class parameter passing 4626ABI. 4627@end deftypefn 4628 4629@node Caller Saves 4630@subsection Caller-Saves Register Allocation 4631 4632If you enable it, GCC can save registers around function calls. This 4633makes it possible to use call-clobbered registers to hold variables that 4634must live across calls. 4635 4636@defmac HARD_REGNO_CALLER_SAVE_MODE (@var{regno}, @var{nregs}) 4637A C expression specifying which mode is required for saving @var{nregs} 4638of a pseudo-register in call-clobbered hard register @var{regno}. If 4639@var{regno} is unsuitable for caller save, @code{VOIDmode} should be 4640returned. For most machines this macro need not be defined since GCC 4641will select the smallest suitable mode. 4642@end defmac 4643 4644@node Function Entry 4645@subsection Function Entry and Exit 4646@cindex function entry and exit 4647@cindex prologue 4648@cindex epilogue 4649 4650This section describes the macros that output function entry 4651(@dfn{prologue}) and exit (@dfn{epilogue}) code. 4652 4653@deftypefn {Target Hook} void TARGET_ASM_PRINT_PATCHABLE_FUNCTION_ENTRY (FILE *@var{file}, unsigned HOST_WIDE_INT @var{patch_area_size}, bool @var{record_p}) 4654Generate a patchable area at the function start, consisting of 4655@var{patch_area_size} NOP instructions. If the target supports named 4656sections and if @var{record_p} is true, insert a pointer to the current 4657location in the table of patchable functions. The default implementation 4658of the hook places the table of pointers in the special section named 4659@code{__patchable_function_entries}. 4660@end deftypefn 4661 4662@deftypefn {Target Hook} void TARGET_ASM_FUNCTION_PROLOGUE (FILE *@var{file}) 4663If defined, a function that outputs the assembler code for entry to a 4664function. The prologue is responsible for setting up the stack frame, 4665initializing the frame pointer register, saving registers that must be 4666saved, and allocating @var{size} additional bytes of storage for the 4667local variables. @var{file} is a stdio stream to which the assembler 4668code should be output. 4669 4670The label for the beginning of the function need not be output by this 4671macro. That has already been done when the macro is run. 4672 4673@findex regs_ever_live 4674To determine which registers to save, the macro can refer to the array 4675@code{regs_ever_live}: element @var{r} is nonzero if hard register 4676@var{r} is used anywhere within the function. This implies the function 4677prologue should save register @var{r}, provided it is not one of the 4678call-used registers. (@code{TARGET_ASM_FUNCTION_EPILOGUE} must likewise use 4679@code{regs_ever_live}.) 4680 4681On machines that have ``register windows'', the function entry code does 4682not save on the stack the registers that are in the windows, even if 4683they are supposed to be preserved by function calls; instead it takes 4684appropriate steps to ``push'' the register stack, if any non-call-used 4685registers are used in the function. 4686 4687@findex frame_pointer_needed 4688On machines where functions may or may not have frame-pointers, the 4689function entry code must vary accordingly; it must set up the frame 4690pointer if one is wanted, and not otherwise. To determine whether a 4691frame pointer is in wanted, the macro can refer to the variable 4692@code{frame_pointer_needed}. The variable's value will be 1 at run 4693time in a function that needs a frame pointer. @xref{Elimination}. 4694 4695The function entry code is responsible for allocating any stack space 4696required for the function. This stack space consists of the regions 4697listed below. In most cases, these regions are allocated in the 4698order listed, with the last listed region closest to the top of the 4699stack (the lowest address if @code{STACK_GROWS_DOWNWARD} is defined, and 4700the highest address if it is not defined). You can use a different order 4701for a machine if doing so is more convenient or required for 4702compatibility reasons. Except in cases where required by standard 4703or by a debugger, there is no reason why the stack layout used by GCC 4704need agree with that used by other compilers for a machine. 4705@end deftypefn 4706 4707@deftypefn {Target Hook} void TARGET_ASM_FUNCTION_END_PROLOGUE (FILE *@var{file}) 4708If defined, a function that outputs assembler code at the end of a 4709prologue. This should be used when the function prologue is being 4710emitted as RTL, and you have some extra assembler that needs to be 4711emitted. @xref{prologue instruction pattern}. 4712@end deftypefn 4713 4714@deftypefn {Target Hook} void TARGET_ASM_FUNCTION_BEGIN_EPILOGUE (FILE *@var{file}) 4715If defined, a function that outputs assembler code at the start of an 4716epilogue. This should be used when the function epilogue is being 4717emitted as RTL, and you have some extra assembler that needs to be 4718emitted. @xref{epilogue instruction pattern}. 4719@end deftypefn 4720 4721@deftypefn {Target Hook} void TARGET_ASM_FUNCTION_EPILOGUE (FILE *@var{file}) 4722If defined, a function that outputs the assembler code for exit from a 4723function. The epilogue is responsible for restoring the saved 4724registers and stack pointer to their values when the function was 4725called, and returning control to the caller. This macro takes the 4726same argument as the macro @code{TARGET_ASM_FUNCTION_PROLOGUE}, and the 4727registers to restore are determined from @code{regs_ever_live} and 4728@code{CALL_USED_REGISTERS} in the same way. 4729 4730On some machines, there is a single instruction that does all the work 4731of returning from the function. On these machines, give that 4732instruction the name @samp{return} and do not define the macro 4733@code{TARGET_ASM_FUNCTION_EPILOGUE} at all. 4734 4735Do not define a pattern named @samp{return} if you want the 4736@code{TARGET_ASM_FUNCTION_EPILOGUE} to be used. If you want the target 4737switches to control whether return instructions or epilogues are used, 4738define a @samp{return} pattern with a validity condition that tests the 4739target switches appropriately. If the @samp{return} pattern's validity 4740condition is false, epilogues will be used. 4741 4742On machines where functions may or may not have frame-pointers, the 4743function exit code must vary accordingly. Sometimes the code for these 4744two cases is completely different. To determine whether a frame pointer 4745is wanted, the macro can refer to the variable 4746@code{frame_pointer_needed}. The variable's value will be 1 when compiling 4747a function that needs a frame pointer. 4748 4749Normally, @code{TARGET_ASM_FUNCTION_PROLOGUE} and 4750@code{TARGET_ASM_FUNCTION_EPILOGUE} must treat leaf functions specially. 4751The C variable @code{current_function_is_leaf} is nonzero for such a 4752function. @xref{Leaf Functions}. 4753 4754On some machines, some functions pop their arguments on exit while 4755others leave that for the caller to do. For example, the 68020 when 4756given @option{-mrtd} pops arguments in functions that take a fixed 4757number of arguments. 4758 4759@findex pops_args 4760@findex crtl->args.pops_args 4761Your definition of the macro @code{RETURN_POPS_ARGS} decides which 4762functions pop their own arguments. @code{TARGET_ASM_FUNCTION_EPILOGUE} 4763needs to know what was decided. The number of bytes of the current 4764function's arguments that this function should pop is available in 4765@code{crtl->args.pops_args}. @xref{Scalar Return}. 4766@end deftypefn 4767 4768@itemize @bullet 4769@item 4770@findex pretend_args_size 4771@findex crtl->args.pretend_args_size 4772A region of @code{crtl->args.pretend_args_size} bytes of 4773uninitialized space just underneath the first argument arriving on the 4774stack. (This may not be at the very start of the allocated stack region 4775if the calling sequence has pushed anything else since pushing the stack 4776arguments. But usually, on such machines, nothing else has been pushed 4777yet, because the function prologue itself does all the pushing.) This 4778region is used on machines where an argument may be passed partly in 4779registers and partly in memory, and, in some cases to support the 4780features in @code{<stdarg.h>}. 4781 4782@item 4783An area of memory used to save certain registers used by the function. 4784The size of this area, which may also include space for such things as 4785the return address and pointers to previous stack frames, is 4786machine-specific and usually depends on which registers have been used 4787in the function. Machines with register windows often do not require 4788a save area. 4789 4790@item 4791A region of at least @var{size} bytes, possibly rounded up to an allocation 4792boundary, to contain the local variables of the function. On some machines, 4793this region and the save area may occur in the opposite order, with the 4794save area closer to the top of the stack. 4795 4796@item 4797@cindex @code{ACCUMULATE_OUTGOING_ARGS} and stack frames 4798Optionally, when @code{ACCUMULATE_OUTGOING_ARGS} is defined, a region of 4799@code{crtl->outgoing_args_size} bytes to be used for outgoing 4800argument lists of the function. @xref{Stack Arguments}. 4801@end itemize 4802 4803@defmac EXIT_IGNORE_STACK 4804Define this macro as a C expression that is nonzero if the return 4805instruction or the function epilogue ignores the value of the stack 4806pointer; in other words, if it is safe to delete an instruction to 4807adjust the stack pointer before a return from the function. The 4808default is 0. 4809 4810Note that this macro's value is relevant only for functions for which 4811frame pointers are maintained. It is never safe to delete a final 4812stack adjustment in a function that has no frame pointer, and the 4813compiler knows this regardless of @code{EXIT_IGNORE_STACK}. 4814@end defmac 4815 4816@defmac EPILOGUE_USES (@var{regno}) 4817Define this macro as a C expression that is nonzero for registers that are 4818used by the epilogue or the @samp{return} pattern. The stack and frame 4819pointer registers are already assumed to be used as needed. 4820@end defmac 4821 4822@defmac EH_USES (@var{regno}) 4823Define this macro as a C expression that is nonzero for registers that are 4824used by the exception handling mechanism, and so should be considered live 4825on entry to an exception edge. 4826@end defmac 4827 4828@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}) 4829A function that outputs the assembler code for a thunk 4830function, used to implement C++ virtual function calls with multiple 4831inheritance. The thunk acts as a wrapper around a virtual function, 4832adjusting the implicit object parameter before handing control off to 4833the real function. 4834 4835First, emit code to add the integer @var{delta} to the location that 4836contains the incoming first argument. Assume that this argument 4837contains a pointer, and is the one used to pass the @code{this} pointer 4838in C++. This is the incoming argument @emph{before} the function prologue, 4839e.g.@: @samp{%o0} on a sparc. The addition must preserve the values of 4840all other incoming arguments. 4841 4842Then, if @var{vcall_offset} is nonzero, an additional adjustment should be 4843made after adding @code{delta}. In particular, if @var{p} is the 4844adjusted pointer, the following adjustment should be made: 4845 4846@smallexample 4847p += (*((ptrdiff_t **)p))[vcall_offset/sizeof(ptrdiff_t)] 4848@end smallexample 4849 4850After the additions, emit code to jump to @var{function}, which is a 4851@code{FUNCTION_DECL}. This is a direct pure jump, not a call, and does 4852not touch the return address. Hence returning from @var{FUNCTION} will 4853return to whoever called the current @samp{thunk}. 4854 4855The effect must be as if @var{function} had been called directly with 4856the adjusted first argument. This macro is responsible for emitting all 4857of the code for a thunk function; @code{TARGET_ASM_FUNCTION_PROLOGUE} 4858and @code{TARGET_ASM_FUNCTION_EPILOGUE} are not invoked. 4859 4860The @var{thunk_fndecl} is redundant. (@var{delta} and @var{function} 4861have already been extracted from it.) It might possibly be useful on 4862some targets, but probably not. 4863 4864If you do not define this macro, the target-independent code in the C++ 4865front end will generate a less efficient heavyweight thunk that calls 4866@var{function} instead of jumping to it. The generic approach does 4867not support varargs. 4868@end deftypefn 4869 4870@deftypefn {Target Hook} bool TARGET_ASM_CAN_OUTPUT_MI_THUNK (const_tree @var{thunk_fndecl}, HOST_WIDE_INT @var{delta}, HOST_WIDE_INT @var{vcall_offset}, const_tree @var{function}) 4871A function that returns true if TARGET_ASM_OUTPUT_MI_THUNK would be able 4872to output the assembler code for the thunk function specified by the 4873arguments it is passed, and false otherwise. In the latter case, the 4874generic approach will be used by the C++ front end, with the limitations 4875previously exposed. 4876@end deftypefn 4877 4878@node Profiling 4879@subsection Generating Code for Profiling 4880@cindex profiling, code generation 4881 4882These macros will help you generate code for profiling. 4883 4884@defmac FUNCTION_PROFILER (@var{file}, @var{labelno}) 4885A C statement or compound statement to output to @var{file} some 4886assembler code to call the profiling subroutine @code{mcount}. 4887 4888@findex mcount 4889The details of how @code{mcount} expects to be called are determined by 4890your operating system environment, not by GCC@. To figure them out, 4891compile a small program for profiling using the system's installed C 4892compiler and look at the assembler code that results. 4893 4894Older implementations of @code{mcount} expect the address of a counter 4895variable to be loaded into some register. The name of this variable is 4896@samp{LP} followed by the number @var{labelno}, so you would generate 4897the name using @samp{LP%d} in a @code{fprintf}. 4898@end defmac 4899 4900@defmac PROFILE_HOOK 4901A C statement or compound statement to output to @var{file} some assembly 4902code to call the profiling subroutine @code{mcount} even the target does 4903not support profiling. 4904@end defmac 4905 4906@defmac NO_PROFILE_COUNTERS 4907Define this macro to be an expression with a nonzero value if the 4908@code{mcount} subroutine on your system does not need a counter variable 4909allocated for each function. This is true for almost all modern 4910implementations. If you define this macro, you must not use the 4911@var{labelno} argument to @code{FUNCTION_PROFILER}. 4912@end defmac 4913 4914@defmac PROFILE_BEFORE_PROLOGUE 4915Define this macro if the code for function profiling should come before 4916the function prologue. Normally, the profiling code comes after. 4917@end defmac 4918 4919@deftypefn {Target Hook} bool TARGET_KEEP_LEAF_WHEN_PROFILED (void) 4920This target hook returns true if the target wants the leaf flag for the current function to stay true even if it calls mcount. This might make sense for targets using the leaf flag only to determine whether a stack frame needs to be generated or not and for which the call to mcount is generated before the function prologue. 4921@end deftypefn 4922 4923@node Tail Calls 4924@subsection Permitting tail calls 4925@cindex tail calls 4926 4927@deftypefn {Target Hook} bool TARGET_FUNCTION_OK_FOR_SIBCALL (tree @var{decl}, tree @var{exp}) 4928True if it is OK to do sibling call optimization for the specified 4929call expression @var{exp}. @var{decl} will be the called function, 4930or @code{NULL} if this is an indirect call. 4931 4932It is not uncommon for limitations of calling conventions to prevent 4933tail calls to functions outside the current unit of translation, or 4934during PIC compilation. The hook is used to enforce these restrictions, 4935as the @code{sibcall} md pattern can not fail, or fall over to a 4936``normal'' call. The criteria for successful sibling call optimization 4937may vary greatly between different architectures. 4938@end deftypefn 4939 4940@deftypefn {Target Hook} void TARGET_EXTRA_LIVE_ON_ENTRY (bitmap @var{regs}) 4941Add any hard registers to @var{regs} that are live on entry to the 4942function. This hook only needs to be defined to provide registers that 4943cannot be found by examination of FUNCTION_ARG_REGNO_P, the callee saved 4944registers, STATIC_CHAIN_INCOMING_REGNUM, STATIC_CHAIN_REGNUM, 4945TARGET_STRUCT_VALUE_RTX, FRAME_POINTER_REGNUM, EH_USES, 4946FRAME_POINTER_REGNUM, ARG_POINTER_REGNUM, and the PIC_OFFSET_TABLE_REGNUM. 4947@end deftypefn 4948 4949@deftypefn {Target Hook} void TARGET_SET_UP_BY_PROLOGUE (struct hard_reg_set_container *@var{}) 4950This hook should add additional registers that are computed by the prologue to the hard regset for shrink-wrapping optimization purposes. 4951@end deftypefn 4952 4953@deftypefn {Target Hook} bool TARGET_WARN_FUNC_RETURN (tree) 4954True if a function's return statements should be checked for matching the function's return type. This includes checking for falling off the end of a non-void function. Return false if no such check should be made. 4955@end deftypefn 4956 4957@node Shrink-wrapping separate components 4958@subsection Shrink-wrapping separate components 4959@cindex shrink-wrapping separate components 4960 4961The prologue may perform a variety of target dependent tasks such as 4962saving callee-saved registers, saving the return address, aligning the 4963stack, creating a stack frame, initializing the PIC register, setting 4964up the static chain, etc. 4965 4966On some targets some of these tasks may be independent of others and 4967thus may be shrink-wrapped separately. These independent tasks are 4968referred to as components and are handled generically by the target 4969independent parts of GCC. 4970 4971Using the following hooks those prologue or epilogue components can be 4972shrink-wrapped separately, so that the initialization (and possibly 4973teardown) those components do is not done as frequently on execution 4974paths where this would unnecessary. 4975 4976What exactly those components are is up to the target code; the generic 4977code treats them abstractly, as a bit in an @code{sbitmap}. These 4978@code{sbitmap}s are allocated by the @code{shrink_wrap.get_separate_components} 4979and @code{shrink_wrap.components_for_bb} hooks, and deallocated by the 4980generic code. 4981 4982@deftypefn {Target Hook} sbitmap TARGET_SHRINK_WRAP_GET_SEPARATE_COMPONENTS (void) 4983This hook should return an @code{sbitmap} with the bits set for those 4984components that can be separately shrink-wrapped in the current function. 4985Return @code{NULL} if the current function should not get any separate 4986shrink-wrapping. 4987Don't define this hook if it would always return @code{NULL}. 4988If it is defined, the other hooks in this group have to be defined as well. 4989@end deftypefn 4990 4991@deftypefn {Target Hook} sbitmap TARGET_SHRINK_WRAP_COMPONENTS_FOR_BB (basic_block) 4992This hook should return an @code{sbitmap} with the bits set for those 4993components where either the prologue component has to be executed before 4994the @code{basic_block}, or the epilogue component after it, or both. 4995@end deftypefn 4996 4997@deftypefn {Target Hook} void TARGET_SHRINK_WRAP_DISQUALIFY_COMPONENTS (sbitmap @var{components}, edge @var{e}, sbitmap @var{edge_components}, bool @var{is_prologue}) 4998This hook should clear the bits in the @var{components} bitmap for those 4999components in @var{edge_components} that the target cannot handle on edge 5000@var{e}, where @var{is_prologue} says if this is for a prologue or an 5001epilogue instead. 5002@end deftypefn 5003 5004@deftypefn {Target Hook} void TARGET_SHRINK_WRAP_EMIT_PROLOGUE_COMPONENTS (sbitmap) 5005Emit prologue insns for the components indicated by the parameter. 5006@end deftypefn 5007 5008@deftypefn {Target Hook} void TARGET_SHRINK_WRAP_EMIT_EPILOGUE_COMPONENTS (sbitmap) 5009Emit epilogue insns for the components indicated by the parameter. 5010@end deftypefn 5011 5012@deftypefn {Target Hook} void TARGET_SHRINK_WRAP_SET_HANDLED_COMPONENTS (sbitmap) 5013Mark the components in the parameter as handled, so that the 5014@code{prologue} and @code{epilogue} named patterns know to ignore those 5015components. The target code should not hang on to the @code{sbitmap}, it 5016will be deleted after this call. 5017@end deftypefn 5018 5019@node Stack Smashing Protection 5020@subsection Stack smashing protection 5021@cindex stack smashing protection 5022 5023@deftypefn {Target Hook} tree TARGET_STACK_PROTECT_GUARD (void) 5024This hook returns a @code{DECL} node for the external variable to use 5025for the stack protection guard. This variable is initialized by the 5026runtime to some random value and is used to initialize the guard value 5027that is placed at the top of the local stack frame. The type of this 5028variable must be @code{ptr_type_node}. 5029 5030The default version of this hook creates a variable called 5031@samp{__stack_chk_guard}, which is normally defined in @file{libgcc2.c}. 5032@end deftypefn 5033 5034@deftypefn {Target Hook} tree TARGET_STACK_PROTECT_FAIL (void) 5035This hook returns a @code{CALL_EXPR} that alerts the runtime that the 5036stack protect guard variable has been modified. This expression should 5037involve a call to a @code{noreturn} function. 5038 5039The default version of this hook invokes a function called 5040@samp{__stack_chk_fail}, taking no arguments. This function is 5041normally defined in @file{libgcc2.c}. 5042@end deftypefn 5043 5044@deftypefn {Target Hook} bool TARGET_STACK_PROTECT_RUNTIME_ENABLED_P (void) 5045Returns true if the target wants GCC's default stack protect runtime support, otherwise return false. The default implementation always returns true. 5046@end deftypefn 5047 5048@deftypefn {Common Target Hook} bool TARGET_SUPPORTS_SPLIT_STACK (bool @var{report}, struct gcc_options *@var{opts}) 5049Whether this target supports splitting the stack when the options described in @var{opts} have been passed. This is called after options have been parsed, so the target may reject splitting the stack in some configurations. The default version of this hook returns false. If @var{report} is true, this function may issue a warning or error; if @var{report} is false, it must simply return a value 5050@end deftypefn 5051 5052@node Miscellaneous Register Hooks 5053@subsection Miscellaneous register hooks 5054@cindex miscellaneous register hooks 5055 5056@deftypevr {Target Hook} bool TARGET_CALL_FUSAGE_CONTAINS_NON_CALLEE_CLOBBERS 5057Set to true if each call that binds to a local definition explicitly 5058clobbers or sets all non-fixed registers modified by performing the call. 5059That is, by the call pattern itself, or by code that might be inserted by the 5060linker (e.g. stubs, veneers, branch islands), but not including those 5061modifiable by the callee. The affected registers may be mentioned explicitly 5062in the call pattern, or included as clobbers in CALL_INSN_FUNCTION_USAGE. 5063The default version of this hook is set to false. The purpose of this hook 5064is to enable the fipa-ra optimization. 5065@end deftypevr 5066 5067@node Varargs 5068@section Implementing the Varargs Macros 5069@cindex varargs implementation 5070 5071GCC comes with an implementation of @code{<varargs.h>} and 5072@code{<stdarg.h>} that work without change on machines that pass arguments 5073on the stack. Other machines require their own implementations of 5074varargs, and the two machine independent header files must have 5075conditionals to include it. 5076 5077ISO @code{<stdarg.h>} differs from traditional @code{<varargs.h>} mainly in 5078the calling convention for @code{va_start}. The traditional 5079implementation takes just one argument, which is the variable in which 5080to store the argument pointer. The ISO implementation of 5081@code{va_start} takes an additional second argument. The user is 5082supposed to write the last named argument of the function here. 5083 5084However, @code{va_start} should not use this argument. The way to find 5085the end of the named arguments is with the built-in functions described 5086below. 5087 5088@defmac __builtin_saveregs () 5089Use this built-in function to save the argument registers in memory so 5090that the varargs mechanism can access them. Both ISO and traditional 5091versions of @code{va_start} must use @code{__builtin_saveregs}, unless 5092you use @code{TARGET_SETUP_INCOMING_VARARGS} (see below) instead. 5093 5094On some machines, @code{__builtin_saveregs} is open-coded under the 5095control of the target hook @code{TARGET_EXPAND_BUILTIN_SAVEREGS}. On 5096other machines, it calls a routine written in assembler language, 5097found in @file{libgcc2.c}. 5098 5099Code generated for the call to @code{__builtin_saveregs} appears at the 5100beginning of the function, as opposed to where the call to 5101@code{__builtin_saveregs} is written, regardless of what the code is. 5102This is because the registers must be saved before the function starts 5103to use them for its own purposes. 5104@c i rewrote the first sentence above to fix an overfull hbox. --mew 5105@c 10feb93 5106@end defmac 5107 5108@defmac __builtin_next_arg (@var{lastarg}) 5109This builtin returns the address of the first anonymous stack 5110argument, as type @code{void *}. If @code{ARGS_GROW_DOWNWARD}, it 5111returns the address of the location above the first anonymous stack 5112argument. Use it in @code{va_start} to initialize the pointer for 5113fetching arguments from the stack. Also use it in @code{va_start} to 5114verify that the second parameter @var{lastarg} is the last named argument 5115of the current function. 5116@end defmac 5117 5118@defmac __builtin_classify_type (@var{object}) 5119Since each machine has its own conventions for which data types are 5120passed in which kind of register, your implementation of @code{va_arg} 5121has to embody these conventions. The easiest way to categorize the 5122specified data type is to use @code{__builtin_classify_type} together 5123with @code{sizeof} and @code{__alignof__}. 5124 5125@code{__builtin_classify_type} ignores the value of @var{object}, 5126considering only its data type. It returns an integer describing what 5127kind of type that is---integer, floating, pointer, structure, and so on. 5128 5129The file @file{typeclass.h} defines an enumeration that you can use to 5130interpret the values of @code{__builtin_classify_type}. 5131@end defmac 5132 5133These machine description macros help implement varargs: 5134 5135@deftypefn {Target Hook} rtx TARGET_EXPAND_BUILTIN_SAVEREGS (void) 5136If defined, this hook produces the machine-specific code for a call to 5137@code{__builtin_saveregs}. This code will be moved to the very 5138beginning of the function, before any parameter access are made. The 5139return value of this function should be an RTX that contains the value 5140to use as the return of @code{__builtin_saveregs}. 5141@end deftypefn 5142 5143@deftypefn {Target Hook} void TARGET_SETUP_INCOMING_VARARGS (cumulative_args_t @var{args_so_far}, machine_mode @var{mode}, tree @var{type}, int *@var{pretend_args_size}, int @var{second_time}) 5144This target hook offers an alternative to using 5145@code{__builtin_saveregs} and defining the hook 5146@code{TARGET_EXPAND_BUILTIN_SAVEREGS}. Use it to store the anonymous 5147register arguments into the stack so that all the arguments appear to 5148have been passed consecutively on the stack. Once this is done, you can 5149use the standard implementation of varargs that works for machines that 5150pass all their arguments on the stack. 5151 5152The argument @var{args_so_far} points to the @code{CUMULATIVE_ARGS} data 5153structure, containing the values that are obtained after processing the 5154named arguments. The arguments @var{mode} and @var{type} describe the 5155last named argument---its machine mode and its data type as a tree node. 5156 5157The target hook should do two things: first, push onto the stack all the 5158argument registers @emph{not} used for the named arguments, and second, 5159store the size of the data thus pushed into the @code{int}-valued 5160variable pointed to by @var{pretend_args_size}. The value that you 5161store here will serve as additional offset for setting up the stack 5162frame. 5163 5164Because you must generate code to push the anonymous arguments at 5165compile time without knowing their data types, 5166@code{TARGET_SETUP_INCOMING_VARARGS} is only useful on machines that 5167have just a single category of argument register and use it uniformly 5168for all data types. 5169 5170If the argument @var{second_time} is nonzero, it means that the 5171arguments of the function are being analyzed for the second time. This 5172happens for an inline function, which is not actually compiled until the 5173end of the source file. The hook @code{TARGET_SETUP_INCOMING_VARARGS} should 5174not generate any instructions in this case. 5175@end deftypefn 5176 5177@deftypefn {Target Hook} bool TARGET_STRICT_ARGUMENT_NAMING (cumulative_args_t @var{ca}) 5178Define this hook to return @code{true} if the location where a function 5179argument is passed depends on whether or not it is a named argument. 5180 5181This hook controls how the @var{named} argument to @code{TARGET_FUNCTION_ARG} 5182is set for varargs and stdarg functions. If this hook returns 5183@code{true}, the @var{named} argument is always true for named 5184arguments, and false for unnamed arguments. If it returns @code{false}, 5185but @code{TARGET_PRETEND_OUTGOING_VARARGS_NAMED} returns @code{true}, 5186then all arguments are treated as named. Otherwise, all named arguments 5187except the last are treated as named. 5188 5189You need not define this hook if it always returns @code{false}. 5190@end deftypefn 5191 5192@deftypefn {Target Hook} void TARGET_CALL_ARGS (rtx, @var{tree}) 5193While generating RTL for a function call, this target hook is invoked once 5194for each argument passed to the function, either a register returned by 5195@code{TARGET_FUNCTION_ARG} or a memory location. It is called just 5196before the point where argument registers are stored. The type of the 5197function to be called is also passed as the second argument; it is 5198@code{NULL_TREE} for libcalls. The @code{TARGET_END_CALL_ARGS} hook is 5199invoked just after the code to copy the return reg has been emitted. 5200This functionality can be used to perform special setup of call argument 5201registers if a target needs it. 5202For functions without arguments, the hook is called once with @code{pc_rtx} 5203passed instead of an argument register. 5204Most ports do not need to implement anything for this hook. 5205@end deftypefn 5206 5207@deftypefn {Target Hook} void TARGET_END_CALL_ARGS (void) 5208This target hook is invoked while generating RTL for a function call, 5209just after the point where the return reg is copied into a pseudo. It 5210signals that all the call argument and return registers for the just 5211emitted call are now no longer in use. 5212Most ports do not need to implement anything for this hook. 5213@end deftypefn 5214 5215@deftypefn {Target Hook} bool TARGET_PRETEND_OUTGOING_VARARGS_NAMED (cumulative_args_t @var{ca}) 5216If you need to conditionally change ABIs so that one works with 5217@code{TARGET_SETUP_INCOMING_VARARGS}, but the other works like neither 5218@code{TARGET_SETUP_INCOMING_VARARGS} nor @code{TARGET_STRICT_ARGUMENT_NAMING} was 5219defined, then define this hook to return @code{true} if 5220@code{TARGET_SETUP_INCOMING_VARARGS} is used, @code{false} otherwise. 5221Otherwise, you should not define this hook. 5222@end deftypefn 5223 5224@deftypefn {Target Hook} rtx TARGET_LOAD_BOUNDS_FOR_ARG (rtx @var{slot}, rtx @var{arg}, rtx @var{slot_no}) 5225This hook is used by expand pass to emit insn to load bounds of 5226@var{arg} passed in @var{slot}. Expand pass uses this hook in case 5227bounds of @var{arg} are not passed in register. If @var{slot} is a 5228memory, then bounds are loaded as for regular pointer loaded from 5229memory. If @var{slot} is not a memory then @var{slot_no} is an integer 5230constant holding number of the target dependent special slot which 5231should be used to obtain bounds. Hook returns RTX holding loaded bounds. 5232@end deftypefn 5233 5234@deftypefn {Target Hook} void TARGET_STORE_BOUNDS_FOR_ARG (rtx @var{arg}, rtx @var{slot}, rtx @var{bounds}, rtx @var{slot_no}) 5235This hook is used by expand pass to emit insns to store @var{bounds} of 5236@var{arg} passed in @var{slot}. Expand pass uses this hook in case 5237@var{bounds} of @var{arg} are not passed in register. If @var{slot} is a 5238memory, then @var{bounds} are stored as for regular pointer stored in 5239memory. If @var{slot} is not a memory then @var{slot_no} is an integer 5240constant holding number of the target dependent special slot which 5241should be used to store @var{bounds}. 5242@end deftypefn 5243 5244@deftypefn {Target Hook} rtx TARGET_LOAD_RETURNED_BOUNDS (rtx @var{slot}) 5245This hook is used by expand pass to emit insn to load bounds 5246returned by function call in @var{slot}. Hook returns RTX holding 5247loaded bounds. 5248@end deftypefn 5249 5250@deftypefn {Target Hook} void TARGET_STORE_RETURNED_BOUNDS (rtx @var{slot}, rtx @var{bounds}) 5251This hook is used by expand pass to emit insn to store @var{bounds} 5252returned by function call into @var{slot}. 5253@end deftypefn 5254 5255@deftypefn {Target Hook} rtx TARGET_CHKP_FUNCTION_VALUE_BOUNDS (const_tree @var{ret_type}, const_tree @var{fn_decl_or_type}, bool @var{outgoing}) 5256Define this to return an RTX representing the place where a function 5257returns bounds for returned pointers. Arguments meaning is similar to 5258@code{TARGET_FUNCTION_VALUE}. 5259@end deftypefn 5260 5261@deftypefn {Target Hook} void TARGET_SETUP_INCOMING_VARARG_BOUNDS (cumulative_args_t @var{args_so_far}, machine_mode @var{mode}, tree @var{type}, int *@var{pretend_args_size}, int @var{second_time}) 5262Use it to store bounds for anonymous register arguments stored 5263into the stack. Arguments meaning is similar to 5264@code{TARGET_SETUP_INCOMING_VARARGS}. 5265@end deftypefn 5266 5267@node Trampolines 5268@section Trampolines for Nested Functions 5269@cindex trampolines for nested functions 5270@cindex nested functions, trampolines for 5271 5272A @dfn{trampoline} is a small piece of code that is created at run time 5273when the address of a nested function is taken. It normally resides on 5274the stack, in the stack frame of the containing function. These macros 5275tell GCC how to generate code to allocate and initialize a 5276trampoline. 5277 5278The instructions in the trampoline must do two things: load a constant 5279address into the static chain register, and jump to the real address of 5280the nested function. On CISC machines such as the m68k, this requires 5281two instructions, a move immediate and a jump. Then the two addresses 5282exist in the trampoline as word-long immediate operands. On RISC 5283machines, it is often necessary to load each address into a register in 5284two parts. Then pieces of each address form separate immediate 5285operands. 5286 5287The code generated to initialize the trampoline must store the variable 5288parts---the static chain value and the function address---into the 5289immediate operands of the instructions. On a CISC machine, this is 5290simply a matter of copying each address to a memory reference at the 5291proper offset from the start of the trampoline. On a RISC machine, it 5292may be necessary to take out pieces of the address and store them 5293separately. 5294 5295@deftypefn {Target Hook} void TARGET_ASM_TRAMPOLINE_TEMPLATE (FILE *@var{f}) 5296This hook is called by @code{assemble_trampoline_template} to output, 5297on the stream @var{f}, assembler code for a block of data that contains 5298the constant parts of a trampoline. This code should not include a 5299label---the label is taken care of automatically. 5300 5301If you do not define this hook, it means no template is needed 5302for the target. Do not define this hook on systems where the block move 5303code to copy the trampoline into place would be larger than the code 5304to generate it on the spot. 5305@end deftypefn 5306 5307@defmac TRAMPOLINE_SECTION 5308Return the section into which the trampoline template is to be placed 5309(@pxref{Sections}). The default value is @code{readonly_data_section}. 5310@end defmac 5311 5312@defmac TRAMPOLINE_SIZE 5313A C expression for the size in bytes of the trampoline, as an integer. 5314@end defmac 5315 5316@defmac TRAMPOLINE_ALIGNMENT 5317Alignment required for trampolines, in bits. 5318 5319If you don't define this macro, the value of @code{FUNCTION_ALIGNMENT} 5320is used for aligning trampolines. 5321@end defmac 5322 5323@deftypefn {Target Hook} void TARGET_TRAMPOLINE_INIT (rtx @var{m_tramp}, tree @var{fndecl}, rtx @var{static_chain}) 5324This hook is called to initialize a trampoline. 5325@var{m_tramp} is an RTX for the memory block for the trampoline; @var{fndecl} 5326is the @code{FUNCTION_DECL} for the nested function; @var{static_chain} is an 5327RTX for the static chain value that should be passed to the function 5328when it is called. 5329 5330If the target defines @code{TARGET_ASM_TRAMPOLINE_TEMPLATE}, then the 5331first thing this hook should do is emit a block move into @var{m_tramp} 5332from the memory block returned by @code{assemble_trampoline_template}. 5333Note that the block move need only cover the constant parts of the 5334trampoline. If the target isolates the variable parts of the trampoline 5335to the end, not all @code{TRAMPOLINE_SIZE} bytes need be copied. 5336 5337If the target requires any other actions, such as flushing caches or 5338enabling stack execution, these actions should be performed after 5339initializing the trampoline proper. 5340@end deftypefn 5341 5342@deftypefn {Target Hook} rtx TARGET_TRAMPOLINE_ADJUST_ADDRESS (rtx @var{addr}) 5343This hook should perform any machine-specific adjustment in 5344the address of the trampoline. Its argument contains the address of the 5345memory block that was passed to @code{TARGET_TRAMPOLINE_INIT}. In case 5346the address to be used for a function call should be different from the 5347address at which the template was stored, the different address should 5348be returned; otherwise @var{addr} should be returned unchanged. 5349If this hook is not defined, @var{addr} will be used for function calls. 5350@end deftypefn 5351 5352@deftypevr {Target Hook} int TARGET_CUSTOM_FUNCTION_DESCRIPTORS 5353This hook should be defined to a power of 2 if the target will benefit 5354from the use of custom descriptors for nested functions instead of the 5355standard trampolines. Such descriptors are created at run time on the 5356stack and made up of data only, but they are non-standard so the generated 5357code must be prepared to deal with them. This hook should be defined to 0 5358if the target uses function descriptors for its standard calling sequence, 5359like for example HP-PA or IA-64. Using descriptors for nested functions 5360eliminates the need for trampolines that reside on the stack and require 5361it to be made executable. 5362 5363The value of the macro is used to parameterize the run-time identification 5364scheme implemented to distinguish descriptors from function addresses: it 5365gives the number of bytes by which their address is misaligned compared 5366with function addresses. The value of 1 will generally work, unless it is 5367already reserved by the target for another purpose, like for example on ARM. 5368@end deftypevr 5369 5370Implementing trampolines is difficult on many machines because they have 5371separate instruction and data caches. Writing into a stack location 5372fails to clear the memory in the instruction cache, so when the program 5373jumps to that location, it executes the old contents. 5374 5375Here are two possible solutions. One is to clear the relevant parts of 5376the instruction cache whenever a trampoline is set up. The other is to 5377make all trampolines identical, by having them jump to a standard 5378subroutine. The former technique makes trampoline execution faster; the 5379latter makes initialization faster. 5380 5381To clear the instruction cache when a trampoline is initialized, define 5382the following macro. 5383 5384@defmac CLEAR_INSN_CACHE (@var{beg}, @var{end}) 5385If defined, expands to a C expression clearing the @emph{instruction 5386cache} in the specified interval. The definition of this macro would 5387typically be a series of @code{asm} statements. Both @var{beg} and 5388@var{end} are both pointer expressions. 5389@end defmac 5390 5391To use a standard subroutine, define the following macro. In addition, 5392you must make sure that the instructions in a trampoline fill an entire 5393cache line with identical instructions, or else ensure that the 5394beginning of the trampoline code is always aligned at the same point in 5395its cache line. Look in @file{m68k.h} as a guide. 5396 5397@defmac TRANSFER_FROM_TRAMPOLINE 5398Define this macro if trampolines need a special subroutine to do their 5399work. The macro should expand to a series of @code{asm} statements 5400which will be compiled with GCC@. They go in a library function named 5401@code{__transfer_from_trampoline}. 5402 5403If you need to avoid executing the ordinary prologue code of a compiled 5404C function when you jump to the subroutine, you can do so by placing a 5405special label of your own in the assembler code. Use one @code{asm} 5406statement to generate an assembler label, and another to make the label 5407global. Then trampolines can use that label to jump directly to your 5408special assembler code. 5409@end defmac 5410 5411@node Library Calls 5412@section Implicit Calls to Library Routines 5413@cindex library subroutine names 5414@cindex @file{libgcc.a} 5415 5416@c prevent bad page break with this line 5417Here is an explanation of implicit calls to library routines. 5418 5419@defmac DECLARE_LIBRARY_RENAMES 5420This macro, if defined, should expand to a piece of C code that will get 5421expanded when compiling functions for libgcc.a. It can be used to 5422provide alternate names for GCC's internal library functions if there 5423are ABI-mandated names that the compiler should provide. 5424@end defmac 5425 5426@findex set_optab_libfunc 5427@findex init_one_libfunc 5428@deftypefn {Target Hook} void TARGET_INIT_LIBFUNCS (void) 5429This hook should declare additional library routines or rename 5430existing ones, using the functions @code{set_optab_libfunc} and 5431@code{init_one_libfunc} defined in @file{optabs.c}. 5432@code{init_optabs} calls this macro after initializing all the normal 5433library routines. 5434 5435The default is to do nothing. Most ports don't need to define this hook. 5436@end deftypefn 5437 5438@deftypevr {Target Hook} bool TARGET_LIBFUNC_GNU_PREFIX 5439If false (the default), internal library routines start with two 5440underscores. If set to true, these routines start with @code{__gnu_} 5441instead. E.g., @code{__muldi3} changes to @code{__gnu_muldi3}. This 5442currently only affects functions defined in @file{libgcc2.c}. If this 5443is set to true, the @file{tm.h} file must also 5444@code{#define LIBGCC2_GNU_PREFIX}. 5445@end deftypevr 5446 5447@defmac FLOAT_LIB_COMPARE_RETURNS_BOOL (@var{mode}, @var{comparison}) 5448This macro should return @code{true} if the library routine that 5449implements the floating point comparison operator @var{comparison} in 5450mode @var{mode} will return a boolean, and @var{false} if it will 5451return a tristate. 5452 5453GCC's own floating point libraries return tristates from the 5454comparison operators, so the default returns false always. Most ports 5455don't need to define this macro. 5456@end defmac 5457 5458@defmac TARGET_LIB_INT_CMP_BIASED 5459This macro should evaluate to @code{true} if the integer comparison 5460functions (like @code{__cmpdi2}) return 0 to indicate that the first 5461operand is smaller than the second, 1 to indicate that they are equal, 5462and 2 to indicate that the first operand is greater than the second. 5463If this macro evaluates to @code{false} the comparison functions return 5464@minus{}1, 0, and 1 instead of 0, 1, and 2. If the target uses the routines 5465in @file{libgcc.a}, you do not need to define this macro. 5466@end defmac 5467 5468@defmac TARGET_HAS_NO_HW_DIVIDE 5469This macro should be defined if the target has no hardware divide 5470instructions. If this macro is defined, GCC will use an algorithm which 5471make use of simple logical and arithmetic operations for 64-bit 5472division. If the macro is not defined, GCC will use an algorithm which 5473make use of a 64-bit by 32-bit divide primitive. 5474@end defmac 5475 5476@cindex @code{EDOM}, implicit usage 5477@findex matherr 5478@defmac TARGET_EDOM 5479The value of @code{EDOM} on the target machine, as a C integer constant 5480expression. If you don't define this macro, GCC does not attempt to 5481deposit the value of @code{EDOM} into @code{errno} directly. Look in 5482@file{/usr/include/errno.h} to find the value of @code{EDOM} on your 5483system. 5484 5485If you do not define @code{TARGET_EDOM}, then compiled code reports 5486domain errors by calling the library function and letting it report the 5487error. If mathematical functions on your system use @code{matherr} when 5488there is an error, then you should leave @code{TARGET_EDOM} undefined so 5489that @code{matherr} is used normally. 5490@end defmac 5491 5492@cindex @code{errno}, implicit usage 5493@defmac GEN_ERRNO_RTX 5494Define this macro as a C expression to create an rtl expression that 5495refers to the global ``variable'' @code{errno}. (On certain systems, 5496@code{errno} may not actually be a variable.) If you don't define this 5497macro, a reasonable default is used. 5498@end defmac 5499 5500@deftypefn {Target Hook} bool TARGET_LIBC_HAS_FUNCTION (enum function_class @var{fn_class}) 5501This hook determines whether a function from a class of functions 5502@var{fn_class} is present at the runtime. 5503@end deftypefn 5504 5505@defmac NEXT_OBJC_RUNTIME 5506Set this macro to 1 to use the "NeXT" Objective-C message sending conventions 5507by default. This calling convention involves passing the object, the selector 5508and the method arguments all at once to the method-lookup library function. 5509This is the usual setting when targeting Darwin/Mac OS X systems, which have 5510the NeXT runtime installed. 5511 5512If the macro is set to 0, the "GNU" Objective-C message sending convention 5513will be used by default. This convention passes just the object and the 5514selector to the method-lookup function, which returns a pointer to the method. 5515 5516In either case, it remains possible to select code-generation for the alternate 5517scheme, by means of compiler command line switches. 5518@end defmac 5519 5520@node Addressing Modes 5521@section Addressing Modes 5522@cindex addressing modes 5523 5524@c prevent bad page break with this line 5525This is about addressing modes. 5526 5527@defmac HAVE_PRE_INCREMENT 5528@defmacx HAVE_PRE_DECREMENT 5529@defmacx HAVE_POST_INCREMENT 5530@defmacx HAVE_POST_DECREMENT 5531A C expression that is nonzero if the machine supports pre-increment, 5532pre-decrement, post-increment, or post-decrement addressing respectively. 5533@end defmac 5534 5535@defmac HAVE_PRE_MODIFY_DISP 5536@defmacx HAVE_POST_MODIFY_DISP 5537A C expression that is nonzero if the machine supports pre- or 5538post-address side-effect generation involving constants other than 5539the size of the memory operand. 5540@end defmac 5541 5542@defmac HAVE_PRE_MODIFY_REG 5543@defmacx HAVE_POST_MODIFY_REG 5544A C expression that is nonzero if the machine supports pre- or 5545post-address side-effect generation involving a register displacement. 5546@end defmac 5547 5548@defmac CONSTANT_ADDRESS_P (@var{x}) 5549A C expression that is 1 if the RTX @var{x} is a constant which 5550is a valid address. On most machines the default definition of 5551@code{(CONSTANT_P (@var{x}) && GET_CODE (@var{x}) != CONST_DOUBLE)} 5552is acceptable, but a few machines are more restrictive as to which 5553constant addresses are supported. 5554@end defmac 5555 5556@defmac CONSTANT_P (@var{x}) 5557@code{CONSTANT_P}, which is defined by target-independent code, 5558accepts integer-values expressions whose values are not explicitly 5559known, such as @code{symbol_ref}, @code{label_ref}, and @code{high} 5560expressions and @code{const} arithmetic expressions, in addition to 5561@code{const_int} and @code{const_double} expressions. 5562@end defmac 5563 5564@defmac MAX_REGS_PER_ADDRESS 5565A number, the maximum number of registers that can appear in a valid 5566memory address. Note that it is up to you to specify a value equal to 5567the maximum number that @code{TARGET_LEGITIMATE_ADDRESS_P} would ever 5568accept. 5569@end defmac 5570 5571@deftypefn {Target Hook} bool TARGET_LEGITIMATE_ADDRESS_P (machine_mode @var{mode}, rtx @var{x}, bool @var{strict}) 5572A function that returns whether @var{x} (an RTX) is a legitimate memory 5573address on the target machine for a memory operand of mode @var{mode}. 5574 5575Legitimate addresses are defined in two variants: a strict variant and a 5576non-strict one. The @var{strict} parameter chooses which variant is 5577desired by the caller. 5578 5579The strict variant is used in the reload pass. It must be defined so 5580that any pseudo-register that has not been allocated a hard register is 5581considered a memory reference. This is because in contexts where some 5582kind of register is required, a pseudo-register with no hard register 5583must be rejected. For non-hard registers, the strict variant should look 5584up the @code{reg_renumber} array; it should then proceed using the hard 5585register number in the array, or treat the pseudo as a memory reference 5586if the array holds @code{-1}. 5587 5588The non-strict variant is used in other passes. It must be defined to 5589accept all pseudo-registers in every context where some kind of 5590register is required. 5591 5592Normally, constant addresses which are the sum of a @code{symbol_ref} 5593and an integer are stored inside a @code{const} RTX to mark them as 5594constant. Therefore, there is no need to recognize such sums 5595specifically as legitimate addresses. Normally you would simply 5596recognize any @code{const} as legitimate. 5597 5598Usually @code{PRINT_OPERAND_ADDRESS} is not prepared to handle constant 5599sums that are not marked with @code{const}. It assumes that a naked 5600@code{plus} indicates indexing. If so, then you @emph{must} reject such 5601naked constant sums as illegitimate addresses, so that none of them will 5602be given to @code{PRINT_OPERAND_ADDRESS}. 5603 5604@cindex @code{TARGET_ENCODE_SECTION_INFO} and address validation 5605On some machines, whether a symbolic address is legitimate depends on 5606the section that the address refers to. On these machines, define the 5607target hook @code{TARGET_ENCODE_SECTION_INFO} to store the information 5608into the @code{symbol_ref}, and then check for it here. When you see a 5609@code{const}, you will have to look inside it to find the 5610@code{symbol_ref} in order to determine the section. @xref{Assembler 5611Format}. 5612 5613@cindex @code{GO_IF_LEGITIMATE_ADDRESS} 5614Some ports are still using a deprecated legacy substitute for 5615this hook, the @code{GO_IF_LEGITIMATE_ADDRESS} macro. This macro 5616has this syntax: 5617 5618@example 5619#define GO_IF_LEGITIMATE_ADDRESS (@var{mode}, @var{x}, @var{label}) 5620@end example 5621 5622@noindent 5623and should @code{goto @var{label}} if the address @var{x} is a valid 5624address on the target machine for a memory operand of mode @var{mode}. 5625 5626@findex REG_OK_STRICT 5627Compiler source files that want to use the strict variant of this 5628macro define the macro @code{REG_OK_STRICT}. You should use an 5629@code{#ifdef REG_OK_STRICT} conditional to define the strict variant in 5630that case and the non-strict variant otherwise. 5631 5632Using the hook is usually simpler because it limits the number of 5633files that are recompiled when changes are made. 5634@end deftypefn 5635 5636@defmac TARGET_MEM_CONSTRAINT 5637A single character to be used instead of the default @code{'m'} 5638character for general memory addresses. This defines the constraint 5639letter which matches the memory addresses accepted by 5640@code{TARGET_LEGITIMATE_ADDRESS_P}. Define this macro if you want to 5641support new address formats in your back end without changing the 5642semantics of the @code{'m'} constraint. This is necessary in order to 5643preserve functionality of inline assembly constructs using the 5644@code{'m'} constraint. 5645@end defmac 5646 5647@defmac FIND_BASE_TERM (@var{x}) 5648A C expression to determine the base term of address @var{x}, 5649or to provide a simplified version of @var{x} from which @file{alias.c} 5650can easily find the base term. This macro is used in only two places: 5651@code{find_base_value} and @code{find_base_term} in @file{alias.c}. 5652 5653It is always safe for this macro to not be defined. It exists so 5654that alias analysis can understand machine-dependent addresses. 5655 5656The typical use of this macro is to handle addresses containing 5657a label_ref or symbol_ref within an UNSPEC@. 5658@end defmac 5659 5660@deftypefn {Target Hook} rtx TARGET_LEGITIMIZE_ADDRESS (rtx @var{x}, rtx @var{oldx}, machine_mode @var{mode}) 5661This hook is given an invalid memory address @var{x} for an 5662operand of mode @var{mode} and should try to return a valid memory 5663address. 5664 5665@findex break_out_memory_refs 5666@var{x} will always be the result of a call to @code{break_out_memory_refs}, 5667and @var{oldx} will be the operand that was given to that function to produce 5668@var{x}. 5669 5670The code of the hook should not alter the substructure of 5671@var{x}. If it transforms @var{x} into a more legitimate form, it 5672should return the new @var{x}. 5673 5674It is not necessary for this hook to come up with a legitimate address, 5675with the exception of native TLS addresses (@pxref{Emulated TLS}). 5676The compiler has standard ways of doing so in all cases. In fact, if 5677the target supports only emulated TLS, it 5678is safe to omit this hook or make it return @var{x} if it cannot find 5679a valid way to legitimize the address. But often a machine-dependent 5680strategy can generate better code. 5681@end deftypefn 5682 5683@defmac LEGITIMIZE_RELOAD_ADDRESS (@var{x}, @var{mode}, @var{opnum}, @var{type}, @var{ind_levels}, @var{win}) 5684A C compound statement that attempts to replace @var{x}, which is an address 5685that needs reloading, with a valid memory address for an operand of mode 5686@var{mode}. @var{win} will be a C statement label elsewhere in the code. 5687It is not necessary to define this macro, but it might be useful for 5688performance reasons. 5689 5690For example, on the i386, it is sometimes possible to use a single 5691reload register instead of two by reloading a sum of two pseudo 5692registers into a register. On the other hand, for number of RISC 5693processors offsets are limited so that often an intermediate address 5694needs to be generated in order to address a stack slot. By defining 5695@code{LEGITIMIZE_RELOAD_ADDRESS} appropriately, the intermediate addresses 5696generated for adjacent some stack slots can be made identical, and thus 5697be shared. 5698 5699@emph{Note}: This macro should be used with caution. It is necessary 5700to know something of how reload works in order to effectively use this, 5701and it is quite easy to produce macros that build in too much knowledge 5702of reload internals. 5703 5704@emph{Note}: This macro must be able to reload an address created by a 5705previous invocation of this macro. If it fails to handle such addresses 5706then the compiler may generate incorrect code or abort. 5707 5708@findex push_reload 5709The macro definition should use @code{push_reload} to indicate parts that 5710need reloading; @var{opnum}, @var{type} and @var{ind_levels} are usually 5711suitable to be passed unaltered to @code{push_reload}. 5712 5713The code generated by this macro must not alter the substructure of 5714@var{x}. If it transforms @var{x} into a more legitimate form, it 5715should assign @var{x} (which will always be a C variable) a new value. 5716This also applies to parts that you change indirectly by calling 5717@code{push_reload}. 5718 5719@findex strict_memory_address_p 5720The macro definition may use @code{strict_memory_address_p} to test if 5721the address has become legitimate. 5722 5723@findex copy_rtx 5724If you want to change only a part of @var{x}, one standard way of doing 5725this is to use @code{copy_rtx}. Note, however, that it unshares only a 5726single level of rtl. Thus, if the part to be changed is not at the 5727top level, you'll need to replace first the top level. 5728It is not necessary for this macro to come up with a legitimate 5729address; but often a machine-dependent strategy can generate better code. 5730@end defmac 5731 5732@deftypefn {Target Hook} bool TARGET_MODE_DEPENDENT_ADDRESS_P (const_rtx @var{addr}, addr_space_t @var{addrspace}) 5733This hook returns @code{true} if memory address @var{addr} in address 5734space @var{addrspace} can have 5735different meanings depending on the machine mode of the memory 5736reference it is used for or if the address is valid for some modes 5737but not others. 5738 5739Autoincrement and autodecrement addresses typically have mode-dependent 5740effects because the amount of the increment or decrement is the size 5741of the operand being addressed. Some machines have other mode-dependent 5742addresses. Many RISC machines have no mode-dependent addresses. 5743 5744You may assume that @var{addr} is a valid address for the machine. 5745 5746The default version of this hook returns @code{false}. 5747@end deftypefn 5748 5749@deftypefn {Target Hook} bool TARGET_LEGITIMATE_CONSTANT_P (machine_mode @var{mode}, rtx @var{x}) 5750This hook returns true if @var{x} is a legitimate constant for a 5751@var{mode}-mode immediate operand on the target machine. You can assume that 5752@var{x} satisfies @code{CONSTANT_P}, so you need not check this. 5753 5754The default definition returns true. 5755@end deftypefn 5756 5757@deftypefn {Target Hook} rtx TARGET_DELEGITIMIZE_ADDRESS (rtx @var{x}) 5758This hook is used to undo the possibly obfuscating effects of the 5759@code{LEGITIMIZE_ADDRESS} and @code{LEGITIMIZE_RELOAD_ADDRESS} target 5760macros. Some backend implementations of these macros wrap symbol 5761references inside an @code{UNSPEC} rtx to represent PIC or similar 5762addressing modes. This target hook allows GCC's optimizers to understand 5763the semantics of these opaque @code{UNSPEC}s by converting them back 5764into their original form. 5765@end deftypefn 5766 5767@deftypefn {Target Hook} bool TARGET_CONST_NOT_OK_FOR_DEBUG_P (rtx @var{x}) 5768This hook should return true if @var{x} should not be emitted into 5769debug sections. 5770@end deftypefn 5771 5772@deftypefn {Target Hook} bool TARGET_CANNOT_FORCE_CONST_MEM (machine_mode @var{mode}, rtx @var{x}) 5773This hook should return true if @var{x} is of a form that cannot (or 5774should not) be spilled to the constant pool. @var{mode} is the mode 5775of @var{x}. 5776 5777The default version of this hook returns false. 5778 5779The primary reason to define this hook is to prevent reload from 5780deciding that a non-legitimate constant would be better reloaded 5781from the constant pool instead of spilling and reloading a register 5782holding the constant. This restriction is often true of addresses 5783of TLS symbols for various targets. 5784@end deftypefn 5785 5786@deftypefn {Target Hook} bool TARGET_USE_BLOCKS_FOR_CONSTANT_P (machine_mode @var{mode}, const_rtx @var{x}) 5787This hook should return true if pool entries for constant @var{x} can 5788be placed in an @code{object_block} structure. @var{mode} is the mode 5789of @var{x}. 5790 5791The default version returns false for all constants. 5792@end deftypefn 5793 5794@deftypefn {Target Hook} bool TARGET_USE_BLOCKS_FOR_DECL_P (const_tree @var{decl}) 5795This hook should return true if pool entries for @var{decl} should 5796be placed in an @code{object_block} structure. 5797 5798The default version returns true for all decls. 5799@end deftypefn 5800 5801@deftypefn {Target Hook} tree TARGET_BUILTIN_RECIPROCAL (tree @var{fndecl}) 5802This hook should return the DECL of a function that implements the 5803reciprocal of the machine-specific builtin function @var{fndecl}, or 5804@code{NULL_TREE} if such a function is not available. 5805@end deftypefn 5806 5807@deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_MASK_FOR_LOAD (void) 5808This hook should return the DECL of a function @var{f} that given an 5809address @var{addr} as an argument returns a mask @var{m} that can be 5810used to extract from two vectors the relevant data that resides in 5811@var{addr} in case @var{addr} is not properly aligned. 5812 5813The autovectorizer, when vectorizing a load operation from an address 5814@var{addr} that may be unaligned, will generate two vector loads from 5815the two aligned addresses around @var{addr}. It then generates a 5816@code{REALIGN_LOAD} operation to extract the relevant data from the 5817two loaded vectors. The first two arguments to @code{REALIGN_LOAD}, 5818@var{v1} and @var{v2}, are the two vectors, each of size @var{VS}, and 5819the third argument, @var{OFF}, defines how the data will be extracted 5820from these two vectors: if @var{OFF} is 0, then the returned vector is 5821@var{v2}; otherwise, the returned vector is composed from the last 5822@var{VS}-@var{OFF} elements of @var{v1} concatenated to the first 5823@var{OFF} elements of @var{v2}. 5824 5825If this hook is defined, the autovectorizer will generate a call 5826to @var{f} (using the DECL tree that this hook returns) and will 5827use the return value of @var{f} as the argument @var{OFF} to 5828@code{REALIGN_LOAD}. Therefore, the mask @var{m} returned by @var{f} 5829should comply with the semantics expected by @code{REALIGN_LOAD} 5830described above. 5831If this hook is not defined, then @var{addr} will be used as 5832the argument @var{OFF} to @code{REALIGN_LOAD}, in which case the low 5833log2(@var{VS}) @minus{} 1 bits of @var{addr} will be considered. 5834@end deftypefn 5835 5836@deftypefn {Target Hook} int TARGET_VECTORIZE_BUILTIN_VECTORIZATION_COST (enum vect_cost_for_stmt @var{type_of_cost}, tree @var{vectype}, int @var{misalign}) 5837Returns cost of different scalar or vector statements for vectorization cost model. 5838For vector memory operations the cost may depend on type (@var{vectype}) and 5839misalignment value (@var{misalign}). 5840@end deftypefn 5841 5842@deftypefn {Target Hook} HOST_WIDE_INT TARGET_VECTORIZE_PREFERRED_VECTOR_ALIGNMENT (const_tree @var{type}) 5843This hook returns the preferred alignment in bits for accesses to 5844vectors of type @var{type} in vectorized code. This might be less than 5845or greater than the ABI-defined value returned by 5846@code{TARGET_VECTOR_ALIGNMENT}. It can be equal to the alignment of 5847a single element, in which case the vectorizer will not try to optimize 5848for alignment. 5849 5850The default hook returns @code{TYPE_ALIGN (@var{type})}, which is 5851correct for most targets. 5852@end deftypefn 5853 5854@deftypefn {Target Hook} bool TARGET_VECTORIZE_VECTOR_ALIGNMENT_REACHABLE (const_tree @var{type}, bool @var{is_packed}) 5855Return true if vector alignment is reachable (by peeling N iterations) for the given scalar type @var{type}. @var{is_packed} is false if the scalar access using @var{type} is known to be naturally aligned. 5856@end deftypefn 5857 5858@deftypefn {Target Hook} bool TARGET_VECTORIZE_VEC_PERM_CONST (machine_mode @var{mode}, rtx @var{output}, rtx @var{in0}, rtx @var{in1}, const vec_perm_indices @var{&sel}) 5859This hook is used to test whether the target can permute up to two 5860vectors of mode @var{mode} using the permutation vector @code{sel}, and 5861also to emit such a permutation. In the former case @var{in0}, @var{in1} 5862and @var{out} are all null. In the latter case @var{in0} and @var{in1} are 5863the source vectors and @var{out} is the destination vector; all three are 5864registers of mode @var{mode}. @var{in1} is the same as @var{in0} if 5865@var{sel} describes a permutation on one vector instead of two. 5866 5867Return true if the operation is possible, emitting instructions for it 5868if rtxes are provided. 5869 5870@cindex @code{vec_perm@var{m}} instruction pattern 5871If the hook returns false for a mode with multibyte elements, GCC will 5872try the equivalent byte operation. If that also fails, it will try forcing 5873the selector into a register and using the @var{vec_perm@var{mode}} 5874instruction pattern. There is no need for the hook to handle these two 5875implementation approaches itself. 5876@end deftypefn 5877 5878@deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_CONVERSION (unsigned @var{code}, tree @var{dest_type}, tree @var{src_type}) 5879This hook should return the DECL of a function that implements conversion of the 5880input vector of type @var{src_type} to type @var{dest_type}. 5881The value of @var{code} is one of the enumerators in @code{enum tree_code} and 5882specifies how the conversion is to be applied 5883(truncation, rounding, etc.). 5884 5885If this hook is defined, the autovectorizer will use the 5886@code{TARGET_VECTORIZE_BUILTIN_CONVERSION} target hook when vectorizing 5887conversion. Otherwise, it will return @code{NULL_TREE}. 5888@end deftypefn 5889 5890@deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_VECTORIZED_FUNCTION (unsigned @var{code}, tree @var{vec_type_out}, tree @var{vec_type_in}) 5891This hook should return the decl of a function that implements the 5892vectorized variant of the function with the @code{combined_fn} code 5893@var{code} or @code{NULL_TREE} if such a function is not available. 5894The return type of the vectorized function shall be of vector type 5895@var{vec_type_out} and the argument types should be @var{vec_type_in}. 5896@end deftypefn 5897 5898@deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_MD_VECTORIZED_FUNCTION (tree @var{fndecl}, tree @var{vec_type_out}, tree @var{vec_type_in}) 5899This hook should return the decl of a function that implements the 5900vectorized variant of target built-in function @code{fndecl}. The 5901return type of the vectorized function shall be of vector type 5902@var{vec_type_out} and the argument types should be @var{vec_type_in}. 5903@end deftypefn 5904 5905@deftypefn {Target Hook} bool TARGET_VECTORIZE_SUPPORT_VECTOR_MISALIGNMENT (machine_mode @var{mode}, const_tree @var{type}, int @var{misalignment}, bool @var{is_packed}) 5906This hook should return true if the target supports misaligned vector 5907store/load of a specific factor denoted in the @var{misalignment} 5908parameter. The vector store/load should be of machine mode @var{mode} and 5909the elements in the vectors should be of type @var{type}. @var{is_packed} 5910parameter is true if the memory access is defined in a packed struct. 5911@end deftypefn 5912 5913@deftypefn {Target Hook} machine_mode TARGET_VECTORIZE_PREFERRED_SIMD_MODE (scalar_mode @var{mode}) 5914This hook should return the preferred mode for vectorizing scalar 5915mode @var{mode}. The default is 5916equal to @code{word_mode}, because the vectorizer can do some 5917transformations even in absence of specialized @acronym{SIMD} hardware. 5918@end deftypefn 5919 5920@deftypefn {Target Hook} machine_mode TARGET_VECTORIZE_SPLIT_REDUCTION (machine_mode) 5921This hook should return the preferred mode to split the final reduction 5922step on @var{mode} to. The reduction is then carried out reducing upper 5923against lower halves of vectors recursively until the specified mode is 5924reached. The default is @var{mode} which means no splitting. 5925@end deftypefn 5926 5927@deftypefn {Target Hook} void TARGET_VECTORIZE_AUTOVECTORIZE_VECTOR_SIZES (vector_sizes *@var{sizes}) 5928If the mode returned by @code{TARGET_VECTORIZE_PREFERRED_SIMD_MODE} is not 5929the only one that is worth considering, this hook should add all suitable 5930vector sizes to @var{sizes}, in order of decreasing preference. The first 5931one should be the size of @code{TARGET_VECTORIZE_PREFERRED_SIMD_MODE}. 5932 5933The hook does not need to do anything if the vector returned by 5934@code{TARGET_VECTORIZE_PREFERRED_SIMD_MODE} is the only one relevant 5935for autovectorization. The default implementation does nothing. 5936@end deftypefn 5937 5938@deftypefn {Target Hook} opt_machine_mode TARGET_VECTORIZE_GET_MASK_MODE (poly_uint64 @var{nunits}, poly_uint64 @var{length}) 5939A vector mask is a value that holds one boolean result for every element 5940in a vector. This hook returns the machine mode that should be used to 5941represent such a mask when the vector in question is @var{length} bytes 5942long and contains @var{nunits} elements. The hook returns an empty 5943@code{opt_machine_mode} if no such mode exists. 5944 5945The default implementation returns the mode of an integer vector that 5946is @var{length} bytes long and that contains @var{nunits} elements, 5947if such a mode exists. 5948@end deftypefn 5949 5950@deftypefn {Target Hook} bool TARGET_VECTORIZE_EMPTY_MASK_IS_EXPENSIVE (unsigned @var{ifn}) 5951This hook returns true if masked internal function @var{ifn} (really of 5952type @code{internal_fn}) should be considered expensive when the mask is 5953all zeros. GCC can then try to branch around the instruction instead. 5954@end deftypefn 5955 5956@deftypefn {Target Hook} {void *} TARGET_VECTORIZE_INIT_COST (struct loop *@var{loop_info}) 5957This hook should initialize target-specific data structures in preparation for modeling the costs of vectorizing a loop or basic block. The default allocates three unsigned integers for accumulating costs for the prologue, body, and epilogue of the loop or basic block. If @var{loop_info} is non-NULL, it identifies the loop being vectorized; otherwise a single block is being vectorized. 5958@end deftypefn 5959 5960@deftypefn {Target Hook} unsigned TARGET_VECTORIZE_ADD_STMT_COST (void *@var{data}, int @var{count}, enum vect_cost_for_stmt @var{kind}, struct _stmt_vec_info *@var{stmt_info}, int @var{misalign}, enum vect_cost_model_location @var{where}) 5961This hook should update the target-specific @var{data} in response to adding @var{count} copies of the given @var{kind} of statement to a loop or basic block. The default adds the builtin vectorizer cost for the copies of the statement to the accumulator specified by @var{where}, (the prologue, body, or epilogue) and returns the amount added. The return value should be viewed as a tentative cost that may later be revised. 5962@end deftypefn 5963 5964@deftypefn {Target Hook} void TARGET_VECTORIZE_FINISH_COST (void *@var{data}, unsigned *@var{prologue_cost}, unsigned *@var{body_cost}, unsigned *@var{epilogue_cost}) 5965This hook should complete calculations of the cost of vectorizing a loop or basic block based on @var{data}, and return the prologue, body, and epilogue costs as unsigned integers. The default returns the value of the three accumulators. 5966@end deftypefn 5967 5968@deftypefn {Target Hook} void TARGET_VECTORIZE_DESTROY_COST_DATA (void *@var{data}) 5969This hook should release @var{data} and any related data structures allocated by TARGET_VECTORIZE_INIT_COST. The default releases the accumulator. 5970@end deftypefn 5971 5972@deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_GATHER (const_tree @var{mem_vectype}, const_tree @var{index_type}, int @var{scale}) 5973Target builtin that implements vector gather operation. @var{mem_vectype} 5974is the vector type of the load and @var{index_type} is scalar type of 5975the index, scaled by @var{scale}. 5976The default is @code{NULL_TREE} which means to not vectorize gather 5977loads. 5978@end deftypefn 5979 5980@deftypefn {Target Hook} tree TARGET_VECTORIZE_BUILTIN_SCATTER (const_tree @var{vectype}, const_tree @var{index_type}, int @var{scale}) 5981Target builtin that implements vector scatter operation. @var{vectype} 5982is the vector type of the store and @var{index_type} is scalar type of 5983the index, scaled by @var{scale}. 5984The default is @code{NULL_TREE} which means to not vectorize scatter 5985stores. 5986@end deftypefn 5987 5988@deftypefn {Target Hook} int TARGET_SIMD_CLONE_COMPUTE_VECSIZE_AND_SIMDLEN (struct cgraph_node *@var{}, struct cgraph_simd_clone *@var{}, @var{tree}, @var{int}) 5989This hook should set @var{vecsize_mangle}, @var{vecsize_int}, @var{vecsize_float} 5990fields in @var{simd_clone} structure pointed by @var{clone_info} argument and also 5991@var{simdlen} field if it was previously 0. 5992The hook should return 0 if SIMD clones shouldn't be emitted, 5993or number of @var{vecsize_mangle} variants that should be emitted. 5994@end deftypefn 5995 5996@deftypefn {Target Hook} void TARGET_SIMD_CLONE_ADJUST (struct cgraph_node *@var{}) 5997This hook should add implicit @code{attribute(target("..."))} attribute 5998to SIMD clone @var{node} if needed. 5999@end deftypefn 6000 6001@deftypefn {Target Hook} int TARGET_SIMD_CLONE_USABLE (struct cgraph_node *@var{}) 6002This hook should return -1 if SIMD clone @var{node} shouldn't be used 6003in vectorized loops in current function, or non-negative number if it is 6004usable. In that case, the smaller the number is, the more desirable it is 6005to use it. 6006@end deftypefn 6007 6008@deftypefn {Target Hook} int TARGET_SIMT_VF (void) 6009Return number of threads in SIMT thread group on the target. 6010@end deftypefn 6011 6012@deftypefn {Target Hook} bool TARGET_GOACC_VALIDATE_DIMS (tree @var{decl}, int *@var{dims}, int @var{fn_level}) 6013This hook should check the launch dimensions provided for an OpenACC 6014compute region, or routine. Defaulted values are represented as -1 6015and non-constant values as 0. The @var{fn_level} is negative for the 6016function corresponding to the compute region. For a routine is is the 6017outermost level at which partitioned execution may be spawned. The hook 6018should verify non-default values. If DECL is NULL, global defaults 6019are being validated and unspecified defaults should be filled in. 6020Diagnostics should be issued as appropriate. Return 6021true, if changes have been made. You must override this hook to 6022provide dimensions larger than 1. 6023@end deftypefn 6024 6025@deftypefn {Target Hook} int TARGET_GOACC_DIM_LIMIT (int @var{axis}) 6026This hook should return the maximum size of a particular dimension, 6027or zero if unbounded. 6028@end deftypefn 6029 6030@deftypefn {Target Hook} bool TARGET_GOACC_FORK_JOIN (gcall *@var{call}, const int *@var{dims}, bool @var{is_fork}) 6031This hook can be used to convert IFN_GOACC_FORK and IFN_GOACC_JOIN 6032function calls to target-specific gimple, or indicate whether they 6033should be retained. It is executed during the oacc_device_lower pass. 6034It should return true, if the call should be retained. It should 6035return false, if it is to be deleted (either because target-specific 6036gimple has been inserted before it, or there is no need for it). 6037The default hook returns false, if there are no RTL expanders for them. 6038@end deftypefn 6039 6040@deftypefn {Target Hook} void TARGET_GOACC_REDUCTION (gcall *@var{call}) 6041This hook is used by the oacc_transform pass to expand calls to the 6042@var{GOACC_REDUCTION} internal function, into a sequence of gimple 6043instructions. @var{call} is gimple statement containing the call to 6044the function. This hook removes statement @var{call} after the 6045expanded sequence has been inserted. This hook is also responsible 6046for allocating any storage for reductions when necessary. 6047@end deftypefn 6048 6049@node Anchored Addresses 6050@section Anchored Addresses 6051@cindex anchored addresses 6052@cindex @option{-fsection-anchors} 6053 6054GCC usually addresses every static object as a separate entity. 6055For example, if we have: 6056 6057@smallexample 6058static int a, b, c; 6059int foo (void) @{ return a + b + c; @} 6060@end smallexample 6061 6062the code for @code{foo} will usually calculate three separate symbolic 6063addresses: those of @code{a}, @code{b} and @code{c}. On some targets, 6064it would be better to calculate just one symbolic address and access 6065the three variables relative to it. The equivalent pseudocode would 6066be something like: 6067 6068@smallexample 6069int foo (void) 6070@{ 6071 register int *xr = &x; 6072 return xr[&a - &x] + xr[&b - &x] + xr[&c - &x]; 6073@} 6074@end smallexample 6075 6076(which isn't valid C). We refer to shared addresses like @code{x} as 6077``section anchors''. Their use is controlled by @option{-fsection-anchors}. 6078 6079The hooks below describe the target properties that GCC needs to know 6080in order to make effective use of section anchors. It won't use 6081section anchors at all unless either @code{TARGET_MIN_ANCHOR_OFFSET} 6082or @code{TARGET_MAX_ANCHOR_OFFSET} is set to a nonzero value. 6083 6084@deftypevr {Target Hook} HOST_WIDE_INT TARGET_MIN_ANCHOR_OFFSET 6085The minimum offset that should be applied to a section anchor. 6086On most targets, it should be the smallest offset that can be 6087applied to a base register while still giving a legitimate address 6088for every mode. The default value is 0. 6089@end deftypevr 6090 6091@deftypevr {Target Hook} HOST_WIDE_INT TARGET_MAX_ANCHOR_OFFSET 6092Like @code{TARGET_MIN_ANCHOR_OFFSET}, but the maximum (inclusive) 6093offset that should be applied to section anchors. The default 6094value is 0. 6095@end deftypevr 6096 6097@deftypefn {Target Hook} void TARGET_ASM_OUTPUT_ANCHOR (rtx @var{x}) 6098Write the assembly code to define section anchor @var{x}, which is a 6099@code{SYMBOL_REF} for which @samp{SYMBOL_REF_ANCHOR_P (@var{x})} is true. 6100The hook is called with the assembly output position set to the beginning 6101of @code{SYMBOL_REF_BLOCK (@var{x})}. 6102 6103If @code{ASM_OUTPUT_DEF} is available, the hook's default definition uses 6104it to define the symbol as @samp{. + SYMBOL_REF_BLOCK_OFFSET (@var{x})}. 6105If @code{ASM_OUTPUT_DEF} is not available, the hook's default definition 6106is @code{NULL}, which disables the use of section anchors altogether. 6107@end deftypefn 6108 6109@deftypefn {Target Hook} bool TARGET_USE_ANCHORS_FOR_SYMBOL_P (const_rtx @var{x}) 6110Return true if GCC should attempt to use anchors to access @code{SYMBOL_REF} 6111@var{x}. You can assume @samp{SYMBOL_REF_HAS_BLOCK_INFO_P (@var{x})} and 6112@samp{!SYMBOL_REF_ANCHOR_P (@var{x})}. 6113 6114The default version is correct for most targets, but you might need to 6115intercept this hook to handle things like target-specific attributes 6116or target-specific sections. 6117@end deftypefn 6118 6119@node Condition Code 6120@section Condition Code Status 6121@cindex condition code status 6122 6123The macros in this section can be split in two families, according to the 6124two ways of representing condition codes in GCC. 6125 6126The first representation is the so called @code{(cc0)} representation 6127(@pxref{Jump Patterns}), where all instructions can have an implicit 6128clobber of the condition codes. The second is the condition code 6129register representation, which provides better schedulability for 6130architectures that do have a condition code register, but on which 6131most instructions do not affect it. The latter category includes 6132most RISC machines. 6133 6134The implicit clobbering poses a strong restriction on the placement of 6135the definition and use of the condition code. In the past the definition 6136and use were always adjacent. However, recent changes to support trapping 6137arithmatic may result in the definition and user being in different blocks. 6138Thus, there may be a @code{NOTE_INSN_BASIC_BLOCK} between them. Additionally, 6139the definition may be the source of exception handling edges. 6140 6141These restrictions can prevent important 6142optimizations on some machines. For example, on the IBM RS/6000, there 6143is a delay for taken branches unless the condition code register is set 6144three instructions earlier than the conditional branch. The instruction 6145scheduler cannot perform this optimization if it is not permitted to 6146separate the definition and use of the condition code register. 6147 6148For this reason, it is possible and suggested to use a register to 6149represent the condition code for new ports. If there is a specific 6150condition code register in the machine, use a hard register. If the 6151condition code or comparison result can be placed in any general register, 6152or if there are multiple condition registers, use a pseudo register. 6153Registers used to store the condition code value will usually have a mode 6154that is in class @code{MODE_CC}. 6155 6156Alternatively, you can use @code{BImode} if the comparison operator is 6157specified already in the compare instruction. In this case, you are not 6158interested in most macros in this section. 6159 6160@menu 6161* CC0 Condition Codes:: Old style representation of condition codes. 6162* MODE_CC Condition Codes:: Modern representation of condition codes. 6163@end menu 6164 6165@node CC0 Condition Codes 6166@subsection Representation of condition codes using @code{(cc0)} 6167@findex cc0 6168 6169@findex cc_status 6170The file @file{conditions.h} defines a variable @code{cc_status} to 6171describe how the condition code was computed (in case the interpretation of 6172the condition code depends on the instruction that it was set by). This 6173variable contains the RTL expressions on which the condition code is 6174currently based, and several standard flags. 6175 6176Sometimes additional machine-specific flags must be defined in the machine 6177description header file. It can also add additional machine-specific 6178information by defining @code{CC_STATUS_MDEP}. 6179 6180@defmac CC_STATUS_MDEP 6181C code for a data type which is used for declaring the @code{mdep} 6182component of @code{cc_status}. It defaults to @code{int}. 6183 6184This macro is not used on machines that do not use @code{cc0}. 6185@end defmac 6186 6187@defmac CC_STATUS_MDEP_INIT 6188A C expression to initialize the @code{mdep} field to ``empty''. 6189The default definition does nothing, since most machines don't use 6190the field anyway. If you want to use the field, you should probably 6191define this macro to initialize it. 6192 6193This macro is not used on machines that do not use @code{cc0}. 6194@end defmac 6195 6196@defmac NOTICE_UPDATE_CC (@var{exp}, @var{insn}) 6197A C compound statement to set the components of @code{cc_status} 6198appropriately for an insn @var{insn} whose body is @var{exp}. It is 6199this macro's responsibility to recognize insns that set the condition 6200code as a byproduct of other activity as well as those that explicitly 6201set @code{(cc0)}. 6202 6203This macro is not used on machines that do not use @code{cc0}. 6204 6205If there are insns that do not set the condition code but do alter 6206other machine registers, this macro must check to see whether they 6207invalidate the expressions that the condition code is recorded as 6208reflecting. For example, on the 68000, insns that store in address 6209registers do not set the condition code, which means that usually 6210@code{NOTICE_UPDATE_CC} can leave @code{cc_status} unaltered for such 6211insns. But suppose that the previous insn set the condition code 6212based on location @samp{a4@@(102)} and the current insn stores a new 6213value in @samp{a4}. Although the condition code is not changed by 6214this, it will no longer be true that it reflects the contents of 6215@samp{a4@@(102)}. Therefore, @code{NOTICE_UPDATE_CC} must alter 6216@code{cc_status} in this case to say that nothing is known about the 6217condition code value. 6218 6219The definition of @code{NOTICE_UPDATE_CC} must be prepared to deal 6220with the results of peephole optimization: insns whose patterns are 6221@code{parallel} RTXs containing various @code{reg}, @code{mem} or 6222constants which are just the operands. The RTL structure of these 6223insns is not sufficient to indicate what the insns actually do. What 6224@code{NOTICE_UPDATE_CC} should do when it sees one is just to run 6225@code{CC_STATUS_INIT}. 6226 6227A possible definition of @code{NOTICE_UPDATE_CC} is to call a function 6228that looks at an attribute (@pxref{Insn Attributes}) named, for example, 6229@samp{cc}. This avoids having detailed information about patterns in 6230two places, the @file{md} file and in @code{NOTICE_UPDATE_CC}. 6231@end defmac 6232 6233@node MODE_CC Condition Codes 6234@subsection Representation of condition codes using registers 6235@findex CCmode 6236@findex MODE_CC 6237 6238@defmac SELECT_CC_MODE (@var{op}, @var{x}, @var{y}) 6239On many machines, the condition code may be produced by other instructions 6240than compares, for example the branch can use directly the condition 6241code set by a subtract instruction. However, on some machines 6242when the condition code is set this way some bits (such as the overflow 6243bit) are not set in the same way as a test instruction, so that a different 6244branch instruction must be used for some conditional branches. When 6245this happens, use the machine mode of the condition code register to 6246record different formats of the condition code register. Modes can 6247also be used to record which compare instruction (e.g. a signed or an 6248unsigned comparison) produced the condition codes. 6249 6250If other modes than @code{CCmode} are required, add them to 6251@file{@var{machine}-modes.def} and define @code{SELECT_CC_MODE} to choose 6252a mode given an operand of a compare. This is needed because the modes 6253have to be chosen not only during RTL generation but also, for example, 6254by instruction combination. The result of @code{SELECT_CC_MODE} should 6255be consistent with the mode used in the patterns; for example to support 6256the case of the add on the SPARC discussed above, we have the pattern 6257 6258@smallexample 6259(define_insn "" 6260 [(set (reg:CCNZ 0) 6261 (compare:CCNZ 6262 (plus:SI (match_operand:SI 0 "register_operand" "%r") 6263 (match_operand:SI 1 "arith_operand" "rI")) 6264 (const_int 0)))] 6265 "" 6266 "@dots{}") 6267@end smallexample 6268 6269@noindent 6270together with a @code{SELECT_CC_MODE} that returns @code{CCNZmode} 6271for comparisons whose argument is a @code{plus}: 6272 6273@smallexample 6274#define SELECT_CC_MODE(OP,X,Y) \ 6275 (GET_MODE_CLASS (GET_MODE (X)) == MODE_FLOAT \ 6276 ? ((OP == LT || OP == LE || OP == GT || OP == GE) \ 6277 ? CCFPEmode : CCFPmode) \ 6278 : ((GET_CODE (X) == PLUS || GET_CODE (X) == MINUS \ 6279 || GET_CODE (X) == NEG || GET_CODE (x) == ASHIFT) \ 6280 ? CCNZmode : CCmode)) 6281@end smallexample 6282 6283Another reason to use modes is to retain information on which operands 6284were used by the comparison; see @code{REVERSIBLE_CC_MODE} later in 6285this section. 6286 6287You should define this macro if and only if you define extra CC modes 6288in @file{@var{machine}-modes.def}. 6289@end defmac 6290 6291@deftypefn {Target Hook} void TARGET_CANONICALIZE_COMPARISON (int *@var{code}, rtx *@var{op0}, rtx *@var{op1}, bool @var{op0_preserve_value}) 6292On some machines not all possible comparisons are defined, but you can 6293convert an invalid comparison into a valid one. For example, the Alpha 6294does not have a @code{GT} comparison, but you can use an @code{LT} 6295comparison instead and swap the order of the operands. 6296 6297On such machines, implement this hook to do any required conversions. 6298@var{code} is the initial comparison code and @var{op0} and @var{op1} 6299are the left and right operands of the comparison, respectively. If 6300@var{op0_preserve_value} is @code{true} the implementation is not 6301allowed to change the value of @var{op0} since the value might be used 6302in RTXs which aren't comparisons. E.g. the implementation is not 6303allowed to swap operands in that case. 6304 6305GCC will not assume that the comparison resulting from this macro is 6306valid but will see if the resulting insn matches a pattern in the 6307@file{md} file. 6308 6309You need not to implement this hook if it would never change the 6310comparison code or operands. 6311@end deftypefn 6312 6313@defmac REVERSIBLE_CC_MODE (@var{mode}) 6314A C expression whose value is one if it is always safe to reverse a 6315comparison whose mode is @var{mode}. If @code{SELECT_CC_MODE} 6316can ever return @var{mode} for a floating-point inequality comparison, 6317then @code{REVERSIBLE_CC_MODE (@var{mode})} must be zero. 6318 6319You need not define this macro if it would always returns zero or if the 6320floating-point format is anything other than @code{IEEE_FLOAT_FORMAT}. 6321For example, here is the definition used on the SPARC, where floating-point 6322inequality comparisons are given either @code{CCFPEmode} or @code{CCFPmode}: 6323 6324@smallexample 6325#define REVERSIBLE_CC_MODE(MODE) \ 6326 ((MODE) != CCFPEmode && (MODE) != CCFPmode) 6327@end smallexample 6328@end defmac 6329 6330@defmac REVERSE_CONDITION (@var{code}, @var{mode}) 6331A C expression whose value is reversed condition code of the @var{code} for 6332comparison done in CC_MODE @var{mode}. The macro is used only in case 6333@code{REVERSIBLE_CC_MODE (@var{mode})} is nonzero. Define this macro in case 6334machine has some non-standard way how to reverse certain conditionals. For 6335instance in case all floating point conditions are non-trapping, compiler may 6336freely convert unordered compares to ordered ones. Then definition may look 6337like: 6338 6339@smallexample 6340#define REVERSE_CONDITION(CODE, MODE) \ 6341 ((MODE) != CCFPmode ? reverse_condition (CODE) \ 6342 : reverse_condition_maybe_unordered (CODE)) 6343@end smallexample 6344@end defmac 6345 6346@deftypefn {Target Hook} bool TARGET_FIXED_CONDITION_CODE_REGS (unsigned int *@var{p1}, unsigned int *@var{p2}) 6347On targets which do not use @code{(cc0)}, and which use a hard 6348register rather than a pseudo-register to hold condition codes, the 6349regular CSE passes are often not able to identify cases in which the 6350hard register is set to a common value. Use this hook to enable a 6351small pass which optimizes such cases. This hook should return true 6352to enable this pass, and it should set the integers to which its 6353arguments point to the hard register numbers used for condition codes. 6354When there is only one such register, as is true on most systems, the 6355integer pointed to by @var{p2} should be set to 6356@code{INVALID_REGNUM}. 6357 6358The default version of this hook returns false. 6359@end deftypefn 6360 6361@deftypefn {Target Hook} machine_mode TARGET_CC_MODES_COMPATIBLE (machine_mode @var{m1}, machine_mode @var{m2}) 6362On targets which use multiple condition code modes in class 6363@code{MODE_CC}, it is sometimes the case that a comparison can be 6364validly done in more than one mode. On such a system, define this 6365target hook to take two mode arguments and to return a mode in which 6366both comparisons may be validly done. If there is no such mode, 6367return @code{VOIDmode}. 6368 6369The default version of this hook checks whether the modes are the 6370same. If they are, it returns that mode. If they are different, it 6371returns @code{VOIDmode}. 6372@end deftypefn 6373 6374@deftypevr {Target Hook} {unsigned int} TARGET_FLAGS_REGNUM 6375If the target has a dedicated flags register, and it needs to use the post-reload comparison elimination pass, then this value should be set appropriately. 6376@end deftypevr 6377 6378@node Costs 6379@section Describing Relative Costs of Operations 6380@cindex costs of instructions 6381@cindex relative costs 6382@cindex speed of instructions 6383 6384These macros let you describe the relative speed of various operations 6385on the target machine. 6386 6387@defmac REGISTER_MOVE_COST (@var{mode}, @var{from}, @var{to}) 6388A C expression for the cost of moving data of mode @var{mode} from a 6389register in class @var{from} to one in class @var{to}. The classes are 6390expressed using the enumeration values such as @code{GENERAL_REGS}. A 6391value of 2 is the default; other values are interpreted relative to 6392that. 6393 6394It is not required that the cost always equal 2 when @var{from} is the 6395same as @var{to}; on some machines it is expensive to move between 6396registers if they are not general registers. 6397 6398If reload sees an insn consisting of a single @code{set} between two 6399hard registers, and if @code{REGISTER_MOVE_COST} applied to their 6400classes returns a value of 2, reload does not check to ensure that the 6401constraints of the insn are met. Setting a cost of other than 2 will 6402allow reload to verify that the constraints are met. You should do this 6403if the @samp{mov@var{m}} pattern's constraints do not allow such copying. 6404 6405These macros are obsolete, new ports should use the target hook 6406@code{TARGET_REGISTER_MOVE_COST} instead. 6407@end defmac 6408 6409@deftypefn {Target Hook} int TARGET_REGISTER_MOVE_COST (machine_mode @var{mode}, reg_class_t @var{from}, reg_class_t @var{to}) 6410This target hook should return the cost of moving data of mode @var{mode} 6411from a register in class @var{from} to one in class @var{to}. The classes 6412are expressed using the enumeration values such as @code{GENERAL_REGS}. 6413A value of 2 is the default; other values are interpreted relative to 6414that. 6415 6416It is not required that the cost always equal 2 when @var{from} is the 6417same as @var{to}; on some machines it is expensive to move between 6418registers if they are not general registers. 6419 6420If reload sees an insn consisting of a single @code{set} between two 6421hard registers, and if @code{TARGET_REGISTER_MOVE_COST} applied to their 6422classes returns a value of 2, reload does not check to ensure that the 6423constraints of the insn are met. Setting a cost of other than 2 will 6424allow reload to verify that the constraints are met. You should do this 6425if the @samp{mov@var{m}} pattern's constraints do not allow such copying. 6426 6427The default version of this function returns 2. 6428@end deftypefn 6429 6430@defmac MEMORY_MOVE_COST (@var{mode}, @var{class}, @var{in}) 6431A C expression for the cost of moving data of mode @var{mode} between a 6432register of class @var{class} and memory; @var{in} is zero if the value 6433is to be written to memory, nonzero if it is to be read in. This cost 6434is relative to those in @code{REGISTER_MOVE_COST}. If moving between 6435registers and memory is more expensive than between two registers, you 6436should define this macro to express the relative cost. 6437 6438If you do not define this macro, GCC uses a default cost of 4 plus 6439the cost of copying via a secondary reload register, if one is 6440needed. If your machine requires a secondary reload register to copy 6441between memory and a register of @var{class} but the reload mechanism is 6442more complex than copying via an intermediate, define this macro to 6443reflect the actual cost of the move. 6444 6445GCC defines the function @code{memory_move_secondary_cost} if 6446secondary reloads are needed. It computes the costs due to copying via 6447a secondary register. If your machine copies from memory using a 6448secondary register in the conventional way but the default base value of 64494 is not correct for your machine, define this macro to add some other 6450value to the result of that function. The arguments to that function 6451are the same as to this macro. 6452 6453These macros are obsolete, new ports should use the target hook 6454@code{TARGET_MEMORY_MOVE_COST} instead. 6455@end defmac 6456 6457@deftypefn {Target Hook} int TARGET_MEMORY_MOVE_COST (machine_mode @var{mode}, reg_class_t @var{rclass}, bool @var{in}) 6458This target hook should return the cost of moving data of mode @var{mode} 6459between a register of class @var{rclass} and memory; @var{in} is @code{false} 6460if the value is to be written to memory, @code{true} if it is to be read in. 6461This cost is relative to those in @code{TARGET_REGISTER_MOVE_COST}. 6462If moving between registers and memory is more expensive than between two 6463registers, you should add this target hook to express the relative cost. 6464 6465If you do not add this target hook, GCC uses a default cost of 4 plus 6466the cost of copying via a secondary reload register, if one is 6467needed. If your machine requires a secondary reload register to copy 6468between memory and a register of @var{rclass} but the reload mechanism is 6469more complex than copying via an intermediate, use this target hook to 6470reflect the actual cost of the move. 6471 6472GCC defines the function @code{memory_move_secondary_cost} if 6473secondary reloads are needed. It computes the costs due to copying via 6474a secondary register. If your machine copies from memory using a 6475secondary register in the conventional way but the default base value of 64764 is not correct for your machine, use this target hook to add some other 6477value to the result of that function. The arguments to that function 6478are the same as to this target hook. 6479@end deftypefn 6480 6481@defmac BRANCH_COST (@var{speed_p}, @var{predictable_p}) 6482A C expression for the cost of a branch instruction. A value of 1 is 6483the default; other values are interpreted relative to that. Parameter 6484@var{speed_p} is true when the branch in question should be optimized 6485for speed. When it is false, @code{BRANCH_COST} should return a value 6486optimal for code size rather than performance. @var{predictable_p} is 6487true for well-predicted branches. On many architectures the 6488@code{BRANCH_COST} can be reduced then. 6489@end defmac 6490 6491Here are additional macros which do not specify precise relative costs, 6492but only that certain actions are more expensive than GCC would 6493ordinarily expect. 6494 6495@defmac SLOW_BYTE_ACCESS 6496Define this macro as a C expression which is nonzero if accessing less 6497than a word of memory (i.e.@: a @code{char} or a @code{short}) is no 6498faster than accessing a word of memory, i.e., if such access 6499require more than one instruction or if there is no difference in cost 6500between byte and (aligned) word loads. 6501 6502When this macro is not defined, the compiler will access a field by 6503finding the smallest containing object; when it is defined, a fullword 6504load will be used if alignment permits. Unless bytes accesses are 6505faster than word accesses, using word accesses is preferable since it 6506may eliminate subsequent memory access if subsequent accesses occur to 6507other fields in the same word of the structure, but to different bytes. 6508@end defmac 6509 6510@deftypefn {Target Hook} bool TARGET_SLOW_UNALIGNED_ACCESS (machine_mode @var{mode}, unsigned int @var{align}) 6511This hook returns true if memory accesses described by the 6512@var{mode} and @var{alignment} parameters have a cost many times greater 6513than aligned accesses, for example if they are emulated in a trap handler. 6514This hook is invoked only for unaligned accesses, i.e. when 6515@code{@var{alignment} < GET_MODE_ALIGNMENT (@var{mode})}. 6516 6517When this hook returns true, the compiler will act as if 6518@code{STRICT_ALIGNMENT} were true when generating code for block 6519moves. This can cause significantly more instructions to be produced. 6520Therefore, do not make this hook return true if unaligned accesses only 6521add a cycle or two to the time for a memory access. 6522 6523The hook must return true whenever @code{STRICT_ALIGNMENT} is true. 6524The default implementation returns @code{STRICT_ALIGNMENT}. 6525@end deftypefn 6526 6527@defmac MOVE_RATIO (@var{speed}) 6528The threshold of number of scalar memory-to-memory move insns, @emph{below} 6529which a sequence of insns should be generated instead of a 6530string move insn or a library call. Increasing the value will always 6531make code faster, but eventually incurs high cost in increased code size. 6532 6533Note that on machines where the corresponding move insn is a 6534@code{define_expand} that emits a sequence of insns, this macro counts 6535the number of such sequences. 6536 6537The parameter @var{speed} is true if the code is currently being 6538optimized for speed rather than size. 6539 6540If you don't define this, a reasonable default is used. 6541@end defmac 6542 6543@deftypefn {Target Hook} bool TARGET_USE_BY_PIECES_INFRASTRUCTURE_P (unsigned HOST_WIDE_INT @var{size}, unsigned int @var{alignment}, enum by_pieces_operation @var{op}, bool @var{speed_p}) 6544GCC will attempt several strategies when asked to copy between 6545two areas of memory, or to set, clear or store to memory, for example 6546when copying a @code{struct}. The @code{by_pieces} infrastructure 6547implements such memory operations as a sequence of load, store or move 6548insns. Alternate strategies are to expand the 6549@code{movmem} or @code{setmem} optabs, to emit a library call, or to emit 6550unit-by-unit, loop-based operations. 6551 6552This target hook should return true if, for a memory operation with a 6553given @var{size} and @var{alignment}, using the @code{by_pieces} 6554infrastructure is expected to result in better code generation. 6555Both @var{size} and @var{alignment} are measured in terms of storage 6556units. 6557 6558The parameter @var{op} is one of: @code{CLEAR_BY_PIECES}, 6559@code{MOVE_BY_PIECES}, @code{SET_BY_PIECES}, @code{STORE_BY_PIECES} or 6560@code{COMPARE_BY_PIECES}. These describe the type of memory operation 6561under consideration. 6562 6563The parameter @var{speed_p} is true if the code is currently being 6564optimized for speed rather than size. 6565 6566Returning true for higher values of @var{size} can improve code generation 6567for speed if the target does not provide an implementation of the 6568@code{movmem} or @code{setmem} standard names, if the @code{movmem} or 6569@code{setmem} implementation would be more expensive than a sequence of 6570insns, or if the overhead of a library call would dominate that of 6571the body of the memory operation. 6572 6573Returning true for higher values of @code{size} may also cause an increase 6574in code size, for example where the number of insns emitted to perform a 6575move would be greater than that of a library call. 6576@end deftypefn 6577 6578@deftypefn {Target Hook} int TARGET_COMPARE_BY_PIECES_BRANCH_RATIO (machine_mode @var{mode}) 6579When expanding a block comparison in MODE, gcc can try to reduce the 6580number of branches at the expense of more memory operations. This hook 6581allows the target to override the default choice. It should return the 6582factor by which branches should be reduced over the plain expansion with 6583one comparison per @var{mode}-sized piece. A port can also prevent a 6584particular mode from being used for block comparisons by returning a 6585negative number from this hook. 6586@end deftypefn 6587 6588@defmac MOVE_MAX_PIECES 6589A C expression used by @code{move_by_pieces} to determine the largest unit 6590a load or store used to copy memory is. Defaults to @code{MOVE_MAX}. 6591@end defmac 6592 6593@defmac STORE_MAX_PIECES 6594A C expression used by @code{store_by_pieces} to determine the largest unit 6595a store used to memory is. Defaults to @code{MOVE_MAX_PIECES}, or two times 6596the size of @code{HOST_WIDE_INT}, whichever is smaller. 6597@end defmac 6598 6599@defmac COMPARE_MAX_PIECES 6600A C expression used by @code{compare_by_pieces} to determine the largest unit 6601a load or store used to compare memory is. Defaults to 6602@code{MOVE_MAX_PIECES}. 6603@end defmac 6604 6605@defmac CLEAR_RATIO (@var{speed}) 6606The threshold of number of scalar move insns, @emph{below} which a sequence 6607of insns should be generated to clear memory instead of a string clear insn 6608or a library call. Increasing the value will always make code faster, but 6609eventually incurs high cost in increased code size. 6610 6611The parameter @var{speed} is true if the code is currently being 6612optimized for speed rather than size. 6613 6614If you don't define this, a reasonable default is used. 6615@end defmac 6616 6617@defmac SET_RATIO (@var{speed}) 6618The threshold of number of scalar move insns, @emph{below} which a sequence 6619of insns should be generated to set memory to a constant value, instead of 6620a block set insn or a library call. 6621Increasing the value will always make code faster, but 6622eventually incurs high cost in increased code size. 6623 6624The parameter @var{speed} is true if the code is currently being 6625optimized for speed rather than size. 6626 6627If you don't define this, it defaults to the value of @code{MOVE_RATIO}. 6628@end defmac 6629 6630@defmac USE_LOAD_POST_INCREMENT (@var{mode}) 6631A C expression used to determine whether a load postincrement is a good 6632thing to use for a given mode. Defaults to the value of 6633@code{HAVE_POST_INCREMENT}. 6634@end defmac 6635 6636@defmac USE_LOAD_POST_DECREMENT (@var{mode}) 6637A C expression used to determine whether a load postdecrement is a good 6638thing to use for a given mode. Defaults to the value of 6639@code{HAVE_POST_DECREMENT}. 6640@end defmac 6641 6642@defmac USE_LOAD_PRE_INCREMENT (@var{mode}) 6643A C expression used to determine whether a load preincrement is a good 6644thing to use for a given mode. Defaults to the value of 6645@code{HAVE_PRE_INCREMENT}. 6646@end defmac 6647 6648@defmac USE_LOAD_PRE_DECREMENT (@var{mode}) 6649A C expression used to determine whether a load predecrement is a good 6650thing to use for a given mode. Defaults to the value of 6651@code{HAVE_PRE_DECREMENT}. 6652@end defmac 6653 6654@defmac USE_STORE_POST_INCREMENT (@var{mode}) 6655A C expression used to determine whether a store postincrement is a good 6656thing to use for a given mode. Defaults to the value of 6657@code{HAVE_POST_INCREMENT}. 6658@end defmac 6659 6660@defmac USE_STORE_POST_DECREMENT (@var{mode}) 6661A C expression used to determine whether a store postdecrement is a good 6662thing to use for a given mode. Defaults to the value of 6663@code{HAVE_POST_DECREMENT}. 6664@end defmac 6665 6666@defmac USE_STORE_PRE_INCREMENT (@var{mode}) 6667This macro is used to determine whether a store preincrement is a good 6668thing to use for a given mode. Defaults to the value of 6669@code{HAVE_PRE_INCREMENT}. 6670@end defmac 6671 6672@defmac USE_STORE_PRE_DECREMENT (@var{mode}) 6673This macro is used to determine whether a store predecrement is a good 6674thing to use for a given mode. Defaults to the value of 6675@code{HAVE_PRE_DECREMENT}. 6676@end defmac 6677 6678@defmac NO_FUNCTION_CSE 6679Define this macro to be true if it is as good or better to call a constant 6680function address than to call an address kept in a register. 6681@end defmac 6682 6683@defmac LOGICAL_OP_NON_SHORT_CIRCUIT 6684Define this macro if a non-short-circuit operation produced by 6685@samp{fold_range_test ()} is optimal. This macro defaults to true if 6686@code{BRANCH_COST} is greater than or equal to the value 2. 6687@end defmac 6688 6689@deftypefn {Target Hook} bool TARGET_OPTAB_SUPPORTED_P (int @var{op}, machine_mode @var{mode1}, machine_mode @var{mode2}, optimization_type @var{opt_type}) 6690Return true if the optimizers should use optab @var{op} with 6691modes @var{mode1} and @var{mode2} for optimization type @var{opt_type}. 6692The optab is known to have an associated @file{.md} instruction 6693whose C condition is true. @var{mode2} is only meaningful for conversion 6694optabs; for direct optabs it is a copy of @var{mode1}. 6695 6696For example, when called with @var{op} equal to @code{rint_optab} and 6697@var{mode1} equal to @code{DFmode}, the hook should say whether the 6698optimizers should use optab @code{rintdf2}. 6699 6700The default hook returns true for all inputs. 6701@end deftypefn 6702 6703@deftypefn {Target Hook} bool TARGET_RTX_COSTS (rtx @var{x}, machine_mode @var{mode}, int @var{outer_code}, int @var{opno}, int *@var{total}, bool @var{speed}) 6704This target hook describes the relative costs of RTL expressions. 6705 6706The cost may depend on the precise form of the expression, which is 6707available for examination in @var{x}, and the fact that @var{x} appears 6708as operand @var{opno} of an expression with rtx code @var{outer_code}. 6709That is, the hook can assume that there is some rtx @var{y} such 6710that @samp{GET_CODE (@var{y}) == @var{outer_code}} and such that 6711either (a) @samp{XEXP (@var{y}, @var{opno}) == @var{x}} or 6712(b) @samp{XVEC (@var{y}, @var{opno})} contains @var{x}. 6713 6714@var{mode} is @var{x}'s machine mode, or for cases like @code{const_int} that 6715do not have a mode, the mode in which @var{x} is used. 6716 6717In implementing this hook, you can use the construct 6718@code{COSTS_N_INSNS (@var{n})} to specify a cost equal to @var{n} fast 6719instructions. 6720 6721On entry to the hook, @code{*@var{total}} contains a default estimate 6722for the cost of the expression. The hook should modify this value as 6723necessary. Traditionally, the default costs are @code{COSTS_N_INSNS (5)} 6724for multiplications, @code{COSTS_N_INSNS (7)} for division and modulus 6725operations, and @code{COSTS_N_INSNS (1)} for all other operations. 6726 6727When optimizing for code size, i.e.@: when @code{speed} is 6728false, this target hook should be used to estimate the relative 6729size cost of an expression, again relative to @code{COSTS_N_INSNS}. 6730 6731The hook returns true when all subexpressions of @var{x} have been 6732processed, and false when @code{rtx_cost} should recurse. 6733@end deftypefn 6734 6735@deftypefn {Target Hook} int TARGET_ADDRESS_COST (rtx @var{address}, machine_mode @var{mode}, addr_space_t @var{as}, bool @var{speed}) 6736This hook computes the cost of an addressing mode that contains 6737@var{address}. If not defined, the cost is computed from 6738the @var{address} expression and the @code{TARGET_RTX_COST} hook. 6739 6740For most CISC machines, the default cost is a good approximation of the 6741true cost of the addressing mode. However, on RISC machines, all 6742instructions normally have the same length and execution time. Hence 6743all addresses will have equal costs. 6744 6745In cases where more than one form of an address is known, the form with 6746the lowest cost will be used. If multiple forms have the same, lowest, 6747cost, the one that is the most complex will be used. 6748 6749For example, suppose an address that is equal to the sum of a register 6750and a constant is used twice in the same basic block. When this macro 6751is not defined, the address will be computed in a register and memory 6752references will be indirect through that register. On machines where 6753the cost of the addressing mode containing the sum is no higher than 6754that of a simple indirect reference, this will produce an additional 6755instruction and possibly require an additional register. Proper 6756specification of this macro eliminates this overhead for such machines. 6757 6758This hook is never called with an invalid address. 6759 6760On machines where an address involving more than one register is as 6761cheap as an address computation involving only one register, defining 6762@code{TARGET_ADDRESS_COST} to reflect this can cause two registers to 6763be live over a region of code where only one would have been if 6764@code{TARGET_ADDRESS_COST} were not defined in that manner. This effect 6765should be considered in the definition of this macro. Equivalent costs 6766should probably only be given to addresses with different numbers of 6767registers on machines with lots of registers. 6768@end deftypefn 6769 6770@deftypefn {Target Hook} int TARGET_INSN_COST (rtx_insn *@var{insn}, bool @var{speed}) 6771This target hook describes the relative costs of RTL instructions. 6772 6773In implementing this hook, you can use the construct 6774@code{COSTS_N_INSNS (@var{n})} to specify a cost equal to @var{n} fast 6775instructions. 6776 6777When optimizing for code size, i.e.@: when @code{speed} is 6778false, this target hook should be used to estimate the relative 6779size cost of an expression, again relative to @code{COSTS_N_INSNS}. 6780@end deftypefn 6781 6782@deftypefn {Target Hook} {unsigned int} TARGET_MAX_NOCE_IFCVT_SEQ_COST (edge @var{e}) 6783This hook returns a value in the same units as @code{TARGET_RTX_COSTS}, 6784giving the maximum acceptable cost for a sequence generated by the RTL 6785if-conversion pass when conditional execution is not available. 6786The RTL if-conversion pass attempts to convert conditional operations 6787that would require a branch to a series of unconditional operations and 6788@code{mov@var{mode}cc} insns. This hook returns the maximum cost of the 6789unconditional instructions and the @code{mov@var{mode}cc} insns. 6790RTL if-conversion is cancelled if the cost of the converted sequence 6791is greater than the value returned by this hook. 6792 6793@code{e} is the edge between the basic block containing the conditional 6794branch to the basic block which would be executed if the condition 6795were true. 6796 6797The default implementation of this hook uses the 6798@code{max-rtl-if-conversion-[un]predictable} parameters if they are set, 6799and uses a multiple of @code{BRANCH_COST} otherwise. 6800@end deftypefn 6801 6802@deftypefn {Target Hook} bool TARGET_NOCE_CONVERSION_PROFITABLE_P (rtx_insn *@var{seq}, struct noce_if_info *@var{if_info}) 6803This hook returns true if the instruction sequence @code{seq} is a good 6804candidate as a replacement for the if-convertible sequence described in 6805@code{if_info}. 6806@end deftypefn 6807 6808@deftypefn {Target Hook} bool TARGET_NO_SPECULATION_IN_DELAY_SLOTS_P (void) 6809This predicate controls the use of the eager delay slot filler to disallow 6810speculatively executed instructions being placed in delay slots. Targets 6811such as certain MIPS architectures possess both branches with and without 6812delay slots. As the eager delay slot filler can decrease performance, 6813disabling it is beneficial when ordinary branches are available. Use of 6814delay slot branches filled using the basic filler is often still desirable 6815as the delay slot can hide a pipeline bubble. 6816@end deftypefn 6817 6818@deftypefn {Target Hook} HOST_WIDE_INT TARGET_ESTIMATED_POLY_VALUE (poly_int64 @var{val}) 6819Return an estimate of the runtime value of @var{val}, for use in 6820things like cost calculations or profiling frequencies. The default 6821implementation returns the lowest possible value of @var{val}. 6822@end deftypefn 6823 6824@node Scheduling 6825@section Adjusting the Instruction Scheduler 6826 6827The instruction scheduler may need a fair amount of machine-specific 6828adjustment in order to produce good code. GCC provides several target 6829hooks for this purpose. It is usually enough to define just a few of 6830them: try the first ones in this list first. 6831 6832@deftypefn {Target Hook} int TARGET_SCHED_ISSUE_RATE (void) 6833This hook returns the maximum number of instructions that can ever 6834issue at the same time on the target machine. The default is one. 6835Although the insn scheduler can define itself the possibility of issue 6836an insn on the same cycle, the value can serve as an additional 6837constraint to issue insns on the same simulated processor cycle (see 6838hooks @samp{TARGET_SCHED_REORDER} and @samp{TARGET_SCHED_REORDER2}). 6839This value must be constant over the entire compilation. If you need 6840it to vary depending on what the instructions are, you must use 6841@samp{TARGET_SCHED_VARIABLE_ISSUE}. 6842@end deftypefn 6843 6844@deftypefn {Target Hook} int TARGET_SCHED_VARIABLE_ISSUE (FILE *@var{file}, int @var{verbose}, rtx_insn *@var{insn}, int @var{more}) 6845This hook is executed by the scheduler after it has scheduled an insn 6846from the ready list. It should return the number of insns which can 6847still be issued in the current cycle. The default is 6848@samp{@w{@var{more} - 1}} for insns other than @code{CLOBBER} and 6849@code{USE}, which normally are not counted against the issue rate. 6850You should define this hook if some insns take more machine resources 6851than others, so that fewer insns can follow them in the same cycle. 6852@var{file} is either a null pointer, or a stdio stream to write any 6853debug output to. @var{verbose} is the verbose level provided by 6854@option{-fsched-verbose-@var{n}}. @var{insn} is the instruction that 6855was scheduled. 6856@end deftypefn 6857 6858@deftypefn {Target Hook} int TARGET_SCHED_ADJUST_COST (rtx_insn *@var{insn}, int @var{dep_type1}, rtx_insn *@var{dep_insn}, int @var{cost}, unsigned int @var{dw}) 6859This function corrects the value of @var{cost} based on the 6860relationship between @var{insn} and @var{dep_insn} through a 6861dependence of type dep_type, and strength @var{dw}. It should return the new 6862value. The default is to make no adjustment to @var{cost}. This can be 6863used for example to specify to the scheduler using the traditional pipeline 6864description that an output- or anti-dependence does not incur the same cost 6865as a data-dependence. If the scheduler using the automaton based pipeline 6866description, the cost of anti-dependence is zero and the cost of 6867output-dependence is maximum of one and the difference of latency 6868times of the first and the second insns. If these values are not 6869acceptable, you could use the hook to modify them too. See also 6870@pxref{Processor pipeline description}. 6871@end deftypefn 6872 6873@deftypefn {Target Hook} int TARGET_SCHED_ADJUST_PRIORITY (rtx_insn *@var{insn}, int @var{priority}) 6874This hook adjusts the integer scheduling priority @var{priority} of 6875@var{insn}. It should return the new priority. Increase the priority to 6876execute @var{insn} earlier, reduce the priority to execute @var{insn} 6877later. Do not define this hook if you do not need to adjust the 6878scheduling priorities of insns. 6879@end deftypefn 6880 6881@deftypefn {Target Hook} int TARGET_SCHED_REORDER (FILE *@var{file}, int @var{verbose}, rtx_insn **@var{ready}, int *@var{n_readyp}, int @var{clock}) 6882This hook is executed by the scheduler after it has scheduled the ready 6883list, to allow the machine description to reorder it (for example to 6884combine two small instructions together on @samp{VLIW} machines). 6885@var{file} is either a null pointer, or a stdio stream to write any 6886debug output to. @var{verbose} is the verbose level provided by 6887@option{-fsched-verbose-@var{n}}. @var{ready} is a pointer to the ready 6888list of instructions that are ready to be scheduled. @var{n_readyp} is 6889a pointer to the number of elements in the ready list. The scheduler 6890reads the ready list in reverse order, starting with 6891@var{ready}[@var{*n_readyp} @minus{} 1] and going to @var{ready}[0]. @var{clock} 6892is the timer tick of the scheduler. You may modify the ready list and 6893the number of ready insns. The return value is the number of insns that 6894can issue this cycle; normally this is just @code{issue_rate}. See also 6895@samp{TARGET_SCHED_REORDER2}. 6896@end deftypefn 6897 6898@deftypefn {Target Hook} int TARGET_SCHED_REORDER2 (FILE *@var{file}, int @var{verbose}, rtx_insn **@var{ready}, int *@var{n_readyp}, int @var{clock}) 6899Like @samp{TARGET_SCHED_REORDER}, but called at a different time. That 6900function is called whenever the scheduler starts a new cycle. This one 6901is called once per iteration over a cycle, immediately after 6902@samp{TARGET_SCHED_VARIABLE_ISSUE}; it can reorder the ready list and 6903return the number of insns to be scheduled in the same cycle. Defining 6904this hook can be useful if there are frequent situations where 6905scheduling one insn causes other insns to become ready in the same 6906cycle. These other insns can then be taken into account properly. 6907@end deftypefn 6908 6909@deftypefn {Target Hook} bool TARGET_SCHED_MACRO_FUSION_P (void) 6910This hook is used to check whether target platform supports macro fusion. 6911@end deftypefn 6912 6913@deftypefn {Target Hook} bool TARGET_SCHED_MACRO_FUSION_PAIR_P (rtx_insn *@var{prev}, rtx_insn *@var{curr}) 6914This hook is used to check whether two insns should be macro fused for 6915a target microarchitecture. If this hook returns true for the given insn pair 6916(@var{prev} and @var{curr}), the scheduler will put them into a sched 6917group, and they will not be scheduled apart. The two insns will be either 6918two SET insns or a compare and a conditional jump and this hook should 6919validate any dependencies needed to fuse the two insns together. 6920@end deftypefn 6921 6922@deftypefn {Target Hook} void TARGET_SCHED_DEPENDENCIES_EVALUATION_HOOK (rtx_insn *@var{head}, rtx_insn *@var{tail}) 6923This hook is called after evaluation forward dependencies of insns in 6924chain given by two parameter values (@var{head} and @var{tail} 6925correspondingly) but before insns scheduling of the insn chain. For 6926example, it can be used for better insn classification if it requires 6927analysis of dependencies. This hook can use backward and forward 6928dependencies of the insn scheduler because they are already 6929calculated. 6930@end deftypefn 6931 6932@deftypefn {Target Hook} void TARGET_SCHED_INIT (FILE *@var{file}, int @var{verbose}, int @var{max_ready}) 6933This hook is executed by the scheduler at the beginning of each block of 6934instructions that are to be scheduled. @var{file} is either a null 6935pointer, or a stdio stream to write any debug output to. @var{verbose} 6936is the verbose level provided by @option{-fsched-verbose-@var{n}}. 6937@var{max_ready} is the maximum number of insns in the current scheduling 6938region that can be live at the same time. This can be used to allocate 6939scratch space if it is needed, e.g.@: by @samp{TARGET_SCHED_REORDER}. 6940@end deftypefn 6941 6942@deftypefn {Target Hook} void TARGET_SCHED_FINISH (FILE *@var{file}, int @var{verbose}) 6943This hook is executed by the scheduler at the end of each block of 6944instructions that are to be scheduled. It can be used to perform 6945cleanup of any actions done by the other scheduling hooks. @var{file} 6946is either a null pointer, or a stdio stream to write any debug output 6947to. @var{verbose} is the verbose level provided by 6948@option{-fsched-verbose-@var{n}}. 6949@end deftypefn 6950 6951@deftypefn {Target Hook} void TARGET_SCHED_INIT_GLOBAL (FILE *@var{file}, int @var{verbose}, int @var{old_max_uid}) 6952This hook is executed by the scheduler after function level initializations. 6953@var{file} is either a null pointer, or a stdio stream to write any debug output to. 6954@var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}. 6955@var{old_max_uid} is the maximum insn uid when scheduling begins. 6956@end deftypefn 6957 6958@deftypefn {Target Hook} void TARGET_SCHED_FINISH_GLOBAL (FILE *@var{file}, int @var{verbose}) 6959This is the cleanup hook corresponding to @code{TARGET_SCHED_INIT_GLOBAL}. 6960@var{file} is either a null pointer, or a stdio stream to write any debug output to. 6961@var{verbose} is the verbose level provided by @option{-fsched-verbose-@var{n}}. 6962@end deftypefn 6963 6964@deftypefn {Target Hook} rtx TARGET_SCHED_DFA_PRE_CYCLE_INSN (void) 6965The hook returns an RTL insn. The automaton state used in the 6966pipeline hazard recognizer is changed as if the insn were scheduled 6967when the new simulated processor cycle starts. Usage of the hook may 6968simplify the automaton pipeline description for some @acronym{VLIW} 6969processors. If the hook is defined, it is used only for the automaton 6970based pipeline description. The default is not to change the state 6971when the new simulated processor cycle starts. 6972@end deftypefn 6973 6974@deftypefn {Target Hook} void TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN (void) 6975The hook can be used to initialize data used by the previous hook. 6976@end deftypefn 6977 6978@deftypefn {Target Hook} {rtx_insn *} TARGET_SCHED_DFA_POST_CYCLE_INSN (void) 6979The hook is analogous to @samp{TARGET_SCHED_DFA_PRE_CYCLE_INSN} but used 6980to changed the state as if the insn were scheduled when the new 6981simulated processor cycle finishes. 6982@end deftypefn 6983 6984@deftypefn {Target Hook} void TARGET_SCHED_INIT_DFA_POST_CYCLE_INSN (void) 6985The hook is analogous to @samp{TARGET_SCHED_INIT_DFA_PRE_CYCLE_INSN} but 6986used to initialize data used by the previous hook. 6987@end deftypefn 6988 6989@deftypefn {Target Hook} void TARGET_SCHED_DFA_PRE_ADVANCE_CYCLE (void) 6990The hook to notify target that the current simulated cycle is about to finish. 6991The hook is analogous to @samp{TARGET_SCHED_DFA_PRE_CYCLE_INSN} but used 6992to change the state in more complicated situations - e.g., when advancing 6993state on a single insn is not enough. 6994@end deftypefn 6995 6996@deftypefn {Target Hook} void TARGET_SCHED_DFA_POST_ADVANCE_CYCLE (void) 6997The hook to notify target that new simulated cycle has just started. 6998The hook is analogous to @samp{TARGET_SCHED_DFA_POST_CYCLE_INSN} but used 6999to change the state in more complicated situations - e.g., when advancing 7000state on a single insn is not enough. 7001@end deftypefn 7002 7003@deftypefn {Target Hook} int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD (void) 7004This hook controls better choosing an insn from the ready insn queue 7005for the @acronym{DFA}-based insn scheduler. Usually the scheduler 7006chooses the first insn from the queue. If the hook returns a positive 7007value, an additional scheduler code tries all permutations of 7008@samp{TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD ()} 7009subsequent ready insns to choose an insn whose issue will result in 7010maximal number of issued insns on the same cycle. For the 7011@acronym{VLIW} processor, the code could actually solve the problem of 7012packing simple insns into the @acronym{VLIW} insn. Of course, if the 7013rules of @acronym{VLIW} packing are described in the automaton. 7014 7015This code also could be used for superscalar @acronym{RISC} 7016processors. Let us consider a superscalar @acronym{RISC} processor 7017with 3 pipelines. Some insns can be executed in pipelines @var{A} or 7018@var{B}, some insns can be executed only in pipelines @var{B} or 7019@var{C}, and one insn can be executed in pipeline @var{B}. The 7020processor may issue the 1st insn into @var{A} and the 2nd one into 7021@var{B}. In this case, the 3rd insn will wait for freeing @var{B} 7022until the next cycle. If the scheduler issues the 3rd insn the first, 7023the processor could issue all 3 insns per cycle. 7024 7025Actually this code demonstrates advantages of the automaton based 7026pipeline hazard recognizer. We try quickly and easy many insn 7027schedules to choose the best one. 7028 7029The default is no multipass scheduling. 7030@end deftypefn 7031 7032@deftypefn {Target Hook} int TARGET_SCHED_FIRST_CYCLE_MULTIPASS_DFA_LOOKAHEAD_GUARD (rtx_insn *@var{insn}, int @var{ready_index}) 7033 7034This hook controls what insns from the ready insn queue will be 7035considered for the multipass insn scheduling. If the hook returns 7036zero for @var{insn}, the insn will be considered in multipass scheduling. 7037Positive return values will remove @var{insn} from consideration on 7038the current round of multipass scheduling. 7039Negative return values will remove @var{insn} from consideration for given 7040number of cycles. 7041Backends should be careful about returning non-zero for highest priority 7042instruction at position 0 in the ready list. @var{ready_index} is passed 7043to allow backends make correct judgements. 7044 7045The default is that any ready insns can be chosen to be issued. 7046@end deftypefn 7047 7048@deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_BEGIN (void *@var{data}, signed char *@var{ready_try}, int @var{n_ready}, bool @var{first_cycle_insn_p}) 7049This hook prepares the target backend for a new round of multipass 7050scheduling. 7051@end deftypefn 7052 7053@deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_ISSUE (void *@var{data}, signed char *@var{ready_try}, int @var{n_ready}, rtx_insn *@var{insn}, const void *@var{prev_data}) 7054This hook is called when multipass scheduling evaluates instruction INSN. 7055@end deftypefn 7056 7057@deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_BACKTRACK (const void *@var{data}, signed char *@var{ready_try}, int @var{n_ready}) 7058This is called when multipass scheduling backtracks from evaluation of 7059an instruction. 7060@end deftypefn 7061 7062@deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_END (const void *@var{data}) 7063This hook notifies the target about the result of the concluded current 7064round of multipass scheduling. 7065@end deftypefn 7066 7067@deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_INIT (void *@var{data}) 7068This hook initializes target-specific data used in multipass scheduling. 7069@end deftypefn 7070 7071@deftypefn {Target Hook} void TARGET_SCHED_FIRST_CYCLE_MULTIPASS_FINI (void *@var{data}) 7072This hook finalizes target-specific data used in multipass scheduling. 7073@end deftypefn 7074 7075@deftypefn {Target Hook} int TARGET_SCHED_DFA_NEW_CYCLE (FILE *@var{dump}, int @var{verbose}, rtx_insn *@var{insn}, int @var{last_clock}, int @var{clock}, int *@var{sort_p}) 7076This hook is called by the insn scheduler before issuing @var{insn} 7077on cycle @var{clock}. If the hook returns nonzero, 7078@var{insn} is not issued on this processor cycle. Instead, 7079the processor cycle is advanced. If *@var{sort_p} 7080is zero, the insn ready queue is not sorted on the new cycle 7081start as usually. @var{dump} and @var{verbose} specify the file and 7082verbosity level to use for debugging output. 7083@var{last_clock} and @var{clock} are, respectively, the 7084processor cycle on which the previous insn has been issued, 7085and the current processor cycle. 7086@end deftypefn 7087 7088@deftypefn {Target Hook} bool TARGET_SCHED_IS_COSTLY_DEPENDENCE (struct _dep *@var{_dep}, int @var{cost}, int @var{distance}) 7089This hook is used to define which dependences are considered costly by 7090the target, so costly that it is not advisable to schedule the insns that 7091are involved in the dependence too close to one another. The parameters 7092to this hook are as follows: The first parameter @var{_dep} is the dependence 7093being evaluated. The second parameter @var{cost} is the cost of the 7094dependence as estimated by the scheduler, and the third 7095parameter @var{distance} is the distance in cycles between the two insns. 7096The hook returns @code{true} if considering the distance between the two 7097insns the dependence between them is considered costly by the target, 7098and @code{false} otherwise. 7099 7100Defining this hook can be useful in multiple-issue out-of-order machines, 7101where (a) it's practically hopeless to predict the actual data/resource 7102delays, however: (b) there's a better chance to predict the actual grouping 7103that will be formed, and (c) correctly emulating the grouping can be very 7104important. In such targets one may want to allow issuing dependent insns 7105closer to one another---i.e., closer than the dependence distance; however, 7106not in cases of ``costly dependences'', which this hooks allows to define. 7107@end deftypefn 7108 7109@deftypefn {Target Hook} void TARGET_SCHED_H_I_D_EXTENDED (void) 7110This hook is called by the insn scheduler after emitting a new instruction to 7111the instruction stream. The hook notifies a target backend to extend its 7112per instruction data structures. 7113@end deftypefn 7114 7115@deftypefn {Target Hook} {void *} TARGET_SCHED_ALLOC_SCHED_CONTEXT (void) 7116Return a pointer to a store large enough to hold target scheduling context. 7117@end deftypefn 7118 7119@deftypefn {Target Hook} void TARGET_SCHED_INIT_SCHED_CONTEXT (void *@var{tc}, bool @var{clean_p}) 7120Initialize store pointed to by @var{tc} to hold target scheduling context. 7121It @var{clean_p} is true then initialize @var{tc} as if scheduler is at the 7122beginning of the block. Otherwise, copy the current context into @var{tc}. 7123@end deftypefn 7124 7125@deftypefn {Target Hook} void TARGET_SCHED_SET_SCHED_CONTEXT (void *@var{tc}) 7126Copy target scheduling context pointed to by @var{tc} to the current context. 7127@end deftypefn 7128 7129@deftypefn {Target Hook} void TARGET_SCHED_CLEAR_SCHED_CONTEXT (void *@var{tc}) 7130Deallocate internal data in target scheduling context pointed to by @var{tc}. 7131@end deftypefn 7132 7133@deftypefn {Target Hook} void TARGET_SCHED_FREE_SCHED_CONTEXT (void *@var{tc}) 7134Deallocate a store for target scheduling context pointed to by @var{tc}. 7135@end deftypefn 7136 7137@deftypefn {Target Hook} int TARGET_SCHED_SPECULATE_INSN (rtx_insn *@var{insn}, unsigned int @var{dep_status}, rtx *@var{new_pat}) 7138This hook is called by the insn scheduler when @var{insn} has only 7139speculative dependencies and therefore can be scheduled speculatively. 7140The hook is used to check if the pattern of @var{insn} has a speculative 7141version and, in case of successful check, to generate that speculative 7142pattern. The hook should return 1, if the instruction has a speculative form, 7143or @minus{}1, if it doesn't. @var{request} describes the type of requested 7144speculation. If the return value equals 1 then @var{new_pat} is assigned 7145the generated speculative pattern. 7146@end deftypefn 7147 7148@deftypefn {Target Hook} bool TARGET_SCHED_NEEDS_BLOCK_P (unsigned int @var{dep_status}) 7149This hook is called by the insn scheduler during generation of recovery code 7150for @var{insn}. It should return @code{true}, if the corresponding check 7151instruction should branch to recovery code, or @code{false} otherwise. 7152@end deftypefn 7153 7154@deftypefn {Target Hook} rtx TARGET_SCHED_GEN_SPEC_CHECK (rtx_insn *@var{insn}, rtx_insn *@var{label}, unsigned int @var{ds}) 7155This hook is called by the insn scheduler to generate a pattern for recovery 7156check instruction. If @var{mutate_p} is zero, then @var{insn} is a 7157speculative instruction for which the check should be generated. 7158@var{label} is either a label of a basic block, where recovery code should 7159be emitted, or a null pointer, when requested check doesn't branch to 7160recovery code (a simple check). If @var{mutate_p} is nonzero, then 7161a pattern for a branchy check corresponding to a simple check denoted by 7162@var{insn} should be generated. In this case @var{label} can't be null. 7163@end deftypefn 7164 7165@deftypefn {Target Hook} void TARGET_SCHED_SET_SCHED_FLAGS (struct spec_info_def *@var{spec_info}) 7166This hook is used by the insn scheduler to find out what features should be 7167enabled/used. 7168The structure *@var{spec_info} should be filled in by the target. 7169The structure describes speculation types that can be used in the scheduler. 7170@end deftypefn 7171 7172@deftypefn {Target Hook} bool TARGET_SCHED_CAN_SPECULATE_INSN (rtx_insn *@var{insn}) 7173Some instructions should never be speculated by the schedulers, usually 7174 because the instruction is too expensive to get this wrong. Often such 7175 instructions have long latency, and often they are not fully modeled in the 7176 pipeline descriptions. This hook should return @code{false} if @var{insn} 7177 should not be speculated. 7178@end deftypefn 7179 7180@deftypefn {Target Hook} int TARGET_SCHED_SMS_RES_MII (struct ddg *@var{g}) 7181This hook is called by the swing modulo scheduler to calculate a 7182resource-based lower bound which is based on the resources available in 7183the machine and the resources required by each instruction. The target 7184backend can use @var{g} to calculate such bound. A very simple lower 7185bound will be used in case this hook is not implemented: the total number 7186of instructions divided by the issue rate. 7187@end deftypefn 7188 7189@deftypefn {Target Hook} bool TARGET_SCHED_DISPATCH (rtx_insn *@var{insn}, int @var{x}) 7190This hook is called by Haifa Scheduler. It returns true if dispatch scheduling 7191is supported in hardware and the condition specified in the parameter is true. 7192@end deftypefn 7193 7194@deftypefn {Target Hook} void TARGET_SCHED_DISPATCH_DO (rtx_insn *@var{insn}, int @var{x}) 7195This hook is called by Haifa Scheduler. It performs the operation specified 7196in its second parameter. 7197@end deftypefn 7198 7199@deftypevr {Target Hook} bool TARGET_SCHED_EXPOSED_PIPELINE 7200True if the processor has an exposed pipeline, which means that not just 7201the order of instructions is important for correctness when scheduling, but 7202also the latencies of operations. 7203@end deftypevr 7204 7205@deftypefn {Target Hook} int TARGET_SCHED_REASSOCIATION_WIDTH (unsigned int @var{opc}, machine_mode @var{mode}) 7206This hook is called by tree reassociator to determine a level of 7207parallelism required in output calculations chain. 7208@end deftypefn 7209 7210@deftypefn {Target Hook} void TARGET_SCHED_FUSION_PRIORITY (rtx_insn *@var{insn}, int @var{max_pri}, int *@var{fusion_pri}, int *@var{pri}) 7211This hook is called by scheduling fusion pass. It calculates fusion 7212priorities for each instruction passed in by parameter. The priorities 7213are returned via pointer parameters. 7214 7215@var{insn} is the instruction whose priorities need to be calculated. 7216@var{max_pri} is the maximum priority can be returned in any cases. 7217@var{fusion_pri} is the pointer parameter through which @var{insn}'s 7218fusion priority should be calculated and returned. 7219@var{pri} is the pointer parameter through which @var{insn}'s priority 7220should be calculated and returned. 7221 7222Same @var{fusion_pri} should be returned for instructions which should 7223be scheduled together. Different @var{pri} should be returned for 7224instructions with same @var{fusion_pri}. @var{fusion_pri} is the major 7225sort key, @var{pri} is the minor sort key. All instructions will be 7226scheduled according to the two priorities. All priorities calculated 7227should be between 0 (exclusive) and @var{max_pri} (inclusive). To avoid 7228false dependencies, @var{fusion_pri} of instructions which need to be 7229scheduled together should be smaller than @var{fusion_pri} of irrelevant 7230instructions. 7231 7232Given below example: 7233 7234@smallexample 7235 ldr r10, [r1, 4] 7236 add r4, r4, r10 7237 ldr r15, [r2, 8] 7238 sub r5, r5, r15 7239 ldr r11, [r1, 0] 7240 add r4, r4, r11 7241 ldr r16, [r2, 12] 7242 sub r5, r5, r16 7243@end smallexample 7244 7245On targets like ARM/AArch64, the two pairs of consecutive loads should be 7246merged. Since peephole2 pass can't help in this case unless consecutive 7247loads are actually next to each other in instruction flow. That's where 7248this scheduling fusion pass works. This hook calculates priority for each 7249instruction based on its fustion type, like: 7250 7251@smallexample 7252 ldr r10, [r1, 4] ; fusion_pri=99, pri=96 7253 add r4, r4, r10 ; fusion_pri=100, pri=100 7254 ldr r15, [r2, 8] ; fusion_pri=98, pri=92 7255 sub r5, r5, r15 ; fusion_pri=100, pri=100 7256 ldr r11, [r1, 0] ; fusion_pri=99, pri=100 7257 add r4, r4, r11 ; fusion_pri=100, pri=100 7258 ldr r16, [r2, 12] ; fusion_pri=98, pri=88 7259 sub r5, r5, r16 ; fusion_pri=100, pri=100 7260@end smallexample 7261 7262Scheduling fusion pass then sorts all ready to issue instructions according 7263to the priorities. As a result, instructions of same fusion type will be 7264pushed together in instruction flow, like: 7265 7266@smallexample 7267 ldr r11, [r1, 0] 7268 ldr r10, [r1, 4] 7269 ldr r15, [r2, 8] 7270 ldr r16, [r2, 12] 7271 add r4, r4, r10 7272 sub r5, r5, r15 7273 add r4, r4, r11 7274 sub r5, r5, r16 7275@end smallexample 7276 7277Now peephole2 pass can simply merge the two pairs of loads. 7278 7279Since scheduling fusion pass relies on peephole2 to do real fusion 7280work, it is only enabled by default when peephole2 is in effect. 7281 7282This is firstly introduced on ARM/AArch64 targets, please refer to 7283the hook implementation for how different fusion types are supported. 7284@end deftypefn 7285 7286@deftypefn {Target Hook} void TARGET_EXPAND_DIVMOD_LIBFUNC (rtx @var{libfunc}, machine_mode @var{mode}, rtx @var{op0}, rtx @var{op1}, rtx *@var{quot}, rtx *@var{rem}) 7287Define this hook for enabling divmod transform if the port does not have 7288hardware divmod insn but defines target-specific divmod libfuncs. 7289@end deftypefn 7290 7291@node Sections 7292@section Dividing the Output into Sections (Texts, Data, @dots{}) 7293@c the above section title is WAY too long. maybe cut the part between 7294@c the (...)? --mew 10feb93 7295 7296An object file is divided into sections containing different types of 7297data. In the most common case, there are three sections: the @dfn{text 7298section}, which holds instructions and read-only data; the @dfn{data 7299section}, which holds initialized writable data; and the @dfn{bss 7300section}, which holds uninitialized data. Some systems have other kinds 7301of sections. 7302 7303@file{varasm.c} provides several well-known sections, such as 7304@code{text_section}, @code{data_section} and @code{bss_section}. 7305The normal way of controlling a @code{@var{foo}_section} variable 7306is to define the associated @code{@var{FOO}_SECTION_ASM_OP} macro, 7307as described below. The macros are only read once, when @file{varasm.c} 7308initializes itself, so their values must be run-time constants. 7309They may however depend on command-line flags. 7310 7311@emph{Note:} Some run-time files, such @file{crtstuff.c}, also make 7312use of the @code{@var{FOO}_SECTION_ASM_OP} macros, and expect them 7313to be string literals. 7314 7315Some assemblers require a different string to be written every time a 7316section is selected. If your assembler falls into this category, you 7317should define the @code{TARGET_ASM_INIT_SECTIONS} hook and use 7318@code{get_unnamed_section} to set up the sections. 7319 7320You must always create a @code{text_section}, either by defining 7321@code{TEXT_SECTION_ASM_OP} or by initializing @code{text_section} 7322in @code{TARGET_ASM_INIT_SECTIONS}. The same is true of 7323@code{data_section} and @code{DATA_SECTION_ASM_OP}. If you do not 7324create a distinct @code{readonly_data_section}, the default is to 7325reuse @code{text_section}. 7326 7327All the other @file{varasm.c} sections are optional, and are null 7328if the target does not provide them. 7329 7330@defmac TEXT_SECTION_ASM_OP 7331A C expression whose value is a string, including spacing, containing the 7332assembler operation that should precede instructions and read-only data. 7333Normally @code{"\t.text"} is right. 7334@end defmac 7335 7336@defmac HOT_TEXT_SECTION_NAME 7337If defined, a C string constant for the name of the section containing most 7338frequently executed functions of the program. If not defined, GCC will provide 7339a default definition if the target supports named sections. 7340@end defmac 7341 7342@defmac UNLIKELY_EXECUTED_TEXT_SECTION_NAME 7343If defined, a C string constant for the name of the section containing unlikely 7344executed functions in the program. 7345@end defmac 7346 7347@defmac DATA_SECTION_ASM_OP 7348A C expression whose value is a string, including spacing, containing the 7349assembler operation to identify the following data as writable initialized 7350data. Normally @code{"\t.data"} is right. 7351@end defmac 7352 7353@defmac SDATA_SECTION_ASM_OP 7354If defined, a C expression whose value is a string, including spacing, 7355containing the assembler operation to identify the following data as 7356initialized, writable small data. 7357@end defmac 7358 7359@defmac READONLY_DATA_SECTION_ASM_OP 7360A C expression whose value is a string, including spacing, containing the 7361assembler operation to identify the following data as read-only initialized 7362data. 7363@end defmac 7364 7365@defmac BSS_SECTION_ASM_OP 7366If defined, a C expression whose value is a string, including spacing, 7367containing the assembler operation to identify the following data as 7368uninitialized global data. If not defined, and 7369@code{ASM_OUTPUT_ALIGNED_BSS} not defined, 7370uninitialized global data will be output in the data section if 7371@option{-fno-common} is passed, otherwise @code{ASM_OUTPUT_COMMON} will be 7372used. 7373@end defmac 7374 7375@defmac SBSS_SECTION_ASM_OP 7376If defined, a C expression whose value is a string, including spacing, 7377containing the assembler operation to identify the following data as 7378uninitialized, writable small data. 7379@end defmac 7380 7381@defmac TLS_COMMON_ASM_OP 7382If defined, a C expression whose value is a string containing the 7383assembler operation to identify the following data as thread-local 7384common data. The default is @code{".tls_common"}. 7385@end defmac 7386 7387@defmac TLS_SECTION_ASM_FLAG 7388If defined, a C expression whose value is a character constant 7389containing the flag used to mark a section as a TLS section. The 7390default is @code{'T'}. 7391@end defmac 7392 7393@defmac INIT_SECTION_ASM_OP 7394If defined, a C expression whose value is a string, including spacing, 7395containing the assembler operation to identify the following data as 7396initialization code. If not defined, GCC will assume such a section does 7397not exist. This section has no corresponding @code{init_section} 7398variable; it is used entirely in runtime code. 7399@end defmac 7400 7401@defmac FINI_SECTION_ASM_OP 7402If defined, a C expression whose value is a string, including spacing, 7403containing the assembler operation to identify the following data as 7404finalization code. If not defined, GCC will assume such a section does 7405not exist. This section has no corresponding @code{fini_section} 7406variable; it is used entirely in runtime code. 7407@end defmac 7408 7409@defmac INIT_ARRAY_SECTION_ASM_OP 7410If defined, a C expression whose value is a string, including spacing, 7411containing the assembler operation to identify the following data as 7412part of the @code{.init_array} (or equivalent) section. If not 7413defined, GCC will assume such a section does not exist. Do not define 7414both this macro and @code{INIT_SECTION_ASM_OP}. 7415@end defmac 7416 7417@defmac FINI_ARRAY_SECTION_ASM_OP 7418If defined, a C expression whose value is a string, including spacing, 7419containing the assembler operation to identify the following data as 7420part of the @code{.fini_array} (or equivalent) section. If not 7421defined, GCC will assume such a section does not exist. Do not define 7422both this macro and @code{FINI_SECTION_ASM_OP}. 7423@end defmac 7424 7425@defmac MACH_DEP_SECTION_ASM_FLAG 7426If defined, a C expression whose value is a character constant 7427containing the flag used to mark a machine-dependent section. This 7428corresponds to the @code{SECTION_MACH_DEP} section flag. 7429@end defmac 7430 7431@defmac CRT_CALL_STATIC_FUNCTION (@var{section_op}, @var{function}) 7432If defined, an ASM statement that switches to a different section 7433via @var{section_op}, calls @var{function}, and switches back to 7434the text section. This is used in @file{crtstuff.c} if 7435@code{INIT_SECTION_ASM_OP} or @code{FINI_SECTION_ASM_OP} to calls 7436to initialization and finalization functions from the init and fini 7437sections. By default, this macro uses a simple function call. Some 7438ports need hand-crafted assembly code to avoid dependencies on 7439registers initialized in the function prologue or to ensure that 7440constant pools don't end up too far way in the text section. 7441@end defmac 7442 7443@defmac TARGET_LIBGCC_SDATA_SECTION 7444If defined, a string which names the section into which small 7445variables defined in crtstuff and libgcc should go. This is useful 7446when the target has options for optimizing access to small data, and 7447you want the crtstuff and libgcc routines to be conservative in what 7448they expect of your application yet liberal in what your application 7449expects. For example, for targets with a @code{.sdata} section (like 7450MIPS), you could compile crtstuff with @code{-G 0} so that it doesn't 7451require small data support from your application, but use this macro 7452to put small data into @code{.sdata} so that your application can 7453access these variables whether it uses small data or not. 7454@end defmac 7455 7456@defmac FORCE_CODE_SECTION_ALIGN 7457If defined, an ASM statement that aligns a code section to some 7458arbitrary boundary. This is used to force all fragments of the 7459@code{.init} and @code{.fini} sections to have to same alignment 7460and thus prevent the linker from having to add any padding. 7461@end defmac 7462 7463@defmac JUMP_TABLES_IN_TEXT_SECTION 7464Define this macro to be an expression with a nonzero value if jump 7465tables (for @code{tablejump} insns) should be output in the text 7466section, along with the assembler instructions. Otherwise, the 7467readonly data section is used. 7468 7469This macro is irrelevant if there is no separate readonly data section. 7470@end defmac 7471 7472@deftypefn {Target Hook} void TARGET_ASM_INIT_SECTIONS (void) 7473Define this hook if you need to do something special to set up the 7474@file{varasm.c} sections, or if your target has some special sections 7475of its own that you need to create. 7476 7477GCC calls this hook after processing the command line, but before writing 7478any assembly code, and before calling any of the section-returning hooks 7479described below. 7480@end deftypefn 7481 7482@deftypefn {Target Hook} int TARGET_ASM_RELOC_RW_MASK (void) 7483Return a mask describing how relocations should be treated when 7484selecting sections. Bit 1 should be set if global relocations 7485should be placed in a read-write section; bit 0 should be set if 7486local relocations should be placed in a read-write section. 7487 7488The default version of this function returns 3 when @option{-fpic} 7489is in effect, and 0 otherwise. The hook is typically redefined 7490when the target cannot support (some kinds of) dynamic relocations 7491in read-only sections even in executables. 7492@end deftypefn 7493 7494@deftypefn {Target Hook} {section *} TARGET_ASM_SELECT_SECTION (tree @var{exp}, int @var{reloc}, unsigned HOST_WIDE_INT @var{align}) 7495Return the section into which @var{exp} should be placed. You can 7496assume that @var{exp} is either a @code{VAR_DECL} node or a constant of 7497some sort. @var{reloc} indicates whether the initial value of @var{exp} 7498requires link-time relocations. Bit 0 is set when variable contains 7499local relocations only, while bit 1 is set for global relocations. 7500@var{align} is the constant alignment in bits. 7501 7502The default version of this function takes care of putting read-only 7503variables in @code{readonly_data_section}. 7504 7505See also @var{USE_SELECT_SECTION_FOR_FUNCTIONS}. 7506@end deftypefn 7507 7508@defmac USE_SELECT_SECTION_FOR_FUNCTIONS 7509Define this macro if you wish TARGET_ASM_SELECT_SECTION to be called 7510for @code{FUNCTION_DECL}s as well as for variables and constants. 7511 7512In the case of a @code{FUNCTION_DECL}, @var{reloc} will be zero if the 7513function has been determined to be likely to be called, and nonzero if 7514it is unlikely to be called. 7515@end defmac 7516 7517@deftypefn {Target Hook} void TARGET_ASM_UNIQUE_SECTION (tree @var{decl}, int @var{reloc}) 7518Build up a unique section name, expressed as a @code{STRING_CST} node, 7519and assign it to @samp{DECL_SECTION_NAME (@var{decl})}. 7520As with @code{TARGET_ASM_SELECT_SECTION}, @var{reloc} indicates whether 7521the initial value of @var{exp} requires link-time relocations. 7522 7523The default version of this function appends the symbol name to the 7524ELF section name that would normally be used for the symbol. For 7525example, the function @code{foo} would be placed in @code{.text.foo}. 7526Whatever the actual target object format, this is often good enough. 7527@end deftypefn 7528 7529@deftypefn {Target Hook} {section *} TARGET_ASM_FUNCTION_RODATA_SECTION (tree @var{decl}) 7530Return the readonly data section associated with 7531@samp{DECL_SECTION_NAME (@var{decl})}. 7532The default version of this function selects @code{.gnu.linkonce.r.name} if 7533the function's section is @code{.gnu.linkonce.t.name}, @code{.rodata.name} 7534if function is in @code{.text.name}, and the normal readonly-data section 7535otherwise. 7536@end deftypefn 7537 7538@deftypevr {Target Hook} {const char *} TARGET_ASM_MERGEABLE_RODATA_PREFIX 7539Usually, the compiler uses the prefix @code{".rodata"} to construct 7540section names for mergeable constant data. Define this macro to override 7541the string if a different section name should be used. 7542@end deftypevr 7543 7544@deftypefn {Target Hook} {section *} TARGET_ASM_TM_CLONE_TABLE_SECTION (void) 7545Return the section that should be used for transactional memory clone tables. 7546@end deftypefn 7547 7548@deftypefn {Target Hook} {section *} TARGET_ASM_SELECT_RTX_SECTION (machine_mode @var{mode}, rtx @var{x}, unsigned HOST_WIDE_INT @var{align}) 7549Return the section into which a constant @var{x}, of mode @var{mode}, 7550should be placed. You can assume that @var{x} is some kind of 7551constant in RTL@. The argument @var{mode} is redundant except in the 7552case of a @code{const_int} rtx. @var{align} is the constant alignment 7553in bits. 7554 7555The default version of this function takes care of putting symbolic 7556constants in @code{flag_pic} mode in @code{data_section} and everything 7557else in @code{readonly_data_section}. 7558@end deftypefn 7559 7560@deftypefn {Target Hook} tree TARGET_MANGLE_DECL_ASSEMBLER_NAME (tree @var{decl}, tree @var{id}) 7561Define this hook if you need to postprocess the assembler name generated 7562by target-independent code. The @var{id} provided to this hook will be 7563the computed name (e.g., the macro @code{DECL_NAME} of the @var{decl} in C, 7564or the mangled name of the @var{decl} in C++). The return value of the 7565hook is an @code{IDENTIFIER_NODE} for the appropriate mangled name on 7566your target system. The default implementation of this hook just 7567returns the @var{id} provided. 7568@end deftypefn 7569 7570@deftypefn {Target Hook} void TARGET_ENCODE_SECTION_INFO (tree @var{decl}, rtx @var{rtl}, int @var{new_decl_p}) 7571Define this hook if references to a symbol or a constant must be 7572treated differently depending on something about the variable or 7573function named by the symbol (such as what section it is in). 7574 7575The hook is executed immediately after rtl has been created for 7576@var{decl}, which may be a variable or function declaration or 7577an entry in the constant pool. In either case, @var{rtl} is the 7578rtl in question. Do @emph{not} use @code{DECL_RTL (@var{decl})} 7579in this hook; that field may not have been initialized yet. 7580 7581In the case of a constant, it is safe to assume that the rtl is 7582a @code{mem} whose address is a @code{symbol_ref}. Most decls 7583will also have this form, but that is not guaranteed. Global 7584register variables, for instance, will have a @code{reg} for their 7585rtl. (Normally the right thing to do with such unusual rtl is 7586leave it alone.) 7587 7588The @var{new_decl_p} argument will be true if this is the first time 7589that @code{TARGET_ENCODE_SECTION_INFO} has been invoked on this decl. It will 7590be false for subsequent invocations, which will happen for duplicate 7591declarations. Whether or not anything must be done for the duplicate 7592declaration depends on whether the hook examines @code{DECL_ATTRIBUTES}. 7593@var{new_decl_p} is always true when the hook is called for a constant. 7594 7595@cindex @code{SYMBOL_REF_FLAG}, in @code{TARGET_ENCODE_SECTION_INFO} 7596The usual thing for this hook to do is to record flags in the 7597@code{symbol_ref}, using @code{SYMBOL_REF_FLAG} or @code{SYMBOL_REF_FLAGS}. 7598Historically, the name string was modified if it was necessary to 7599encode more than one bit of information, but this practice is now 7600discouraged; use @code{SYMBOL_REF_FLAGS}. 7601 7602The default definition of this hook, @code{default_encode_section_info} 7603in @file{varasm.c}, sets a number of commonly-useful bits in 7604@code{SYMBOL_REF_FLAGS}. Check whether the default does what you need 7605before overriding it. 7606@end deftypefn 7607 7608@deftypefn {Target Hook} {const char *} TARGET_STRIP_NAME_ENCODING (const char *@var{name}) 7609Decode @var{name} and return the real name part, sans 7610the characters that @code{TARGET_ENCODE_SECTION_INFO} 7611may have added. 7612@end deftypefn 7613 7614@deftypefn {Target Hook} bool TARGET_IN_SMALL_DATA_P (const_tree @var{exp}) 7615Returns true if @var{exp} should be placed into a ``small data'' section. 7616The default version of this hook always returns false. 7617@end deftypefn 7618 7619@deftypevr {Target Hook} bool TARGET_HAVE_SRODATA_SECTION 7620Contains the value true if the target places read-only 7621``small data'' into a separate section. The default value is false. 7622@end deftypevr 7623 7624@deftypefn {Target Hook} bool TARGET_PROFILE_BEFORE_PROLOGUE (void) 7625It returns true if target wants profile code emitted before prologue. 7626 7627The default version of this hook use the target macro 7628@code{PROFILE_BEFORE_PROLOGUE}. 7629@end deftypefn 7630 7631@deftypefn {Target Hook} bool TARGET_BINDS_LOCAL_P (const_tree @var{exp}) 7632Returns true if @var{exp} names an object for which name resolution 7633rules must resolve to the current ``module'' (dynamic shared library 7634or executable image). 7635 7636The default version of this hook implements the name resolution rules 7637for ELF, which has a looser model of global name binding than other 7638currently supported object file formats. 7639@end deftypefn 7640 7641@deftypevr {Target Hook} bool TARGET_HAVE_TLS 7642Contains the value true if the target supports thread-local storage. 7643The default value is false. 7644@end deftypevr 7645 7646 7647@node PIC 7648@section Position Independent Code 7649@cindex position independent code 7650@cindex PIC 7651 7652This section describes macros that help implement generation of position 7653independent code. Simply defining these macros is not enough to 7654generate valid PIC; you must also add support to the hook 7655@code{TARGET_LEGITIMATE_ADDRESS_P} and to the macro 7656@code{PRINT_OPERAND_ADDRESS}, as well as @code{LEGITIMIZE_ADDRESS}. You 7657must modify the definition of @samp{movsi} to do something appropriate 7658when the source operand contains a symbolic address. You may also 7659need to alter the handling of switch statements so that they use 7660relative addresses. 7661@c i rearranged the order of the macros above to try to force one of 7662@c them to the next line, to eliminate an overfull hbox. --mew 10feb93 7663 7664@defmac PIC_OFFSET_TABLE_REGNUM 7665The register number of the register used to address a table of static 7666data addresses in memory. In some cases this register is defined by a 7667processor's ``application binary interface'' (ABI)@. When this macro 7668is defined, RTL is generated for this register once, as with the stack 7669pointer and frame pointer registers. If this macro is not defined, it 7670is up to the machine-dependent files to allocate such a register (if 7671necessary). Note that this register must be fixed when in use (e.g.@: 7672when @code{flag_pic} is true). 7673@end defmac 7674 7675@defmac PIC_OFFSET_TABLE_REG_CALL_CLOBBERED 7676A C expression that is nonzero if the register defined by 7677@code{PIC_OFFSET_TABLE_REGNUM} is clobbered by calls. If not defined, 7678the default is zero. Do not define 7679this macro if @code{PIC_OFFSET_TABLE_REGNUM} is not defined. 7680@end defmac 7681 7682@defmac LEGITIMATE_PIC_OPERAND_P (@var{x}) 7683A C expression that is nonzero if @var{x} is a legitimate immediate 7684operand on the target machine when generating position independent code. 7685You can assume that @var{x} satisfies @code{CONSTANT_P}, so you need not 7686check this. You can also assume @var{flag_pic} is true, so you need not 7687check it either. You need not define this macro if all constants 7688(including @code{SYMBOL_REF}) can be immediate operands when generating 7689position independent code. 7690@end defmac 7691 7692@node Assembler Format 7693@section Defining the Output Assembler Language 7694 7695This section describes macros whose principal purpose is to describe how 7696to write instructions in assembler language---rather than what the 7697instructions do. 7698 7699@menu 7700* File Framework:: Structural information for the assembler file. 7701* Data Output:: Output of constants (numbers, strings, addresses). 7702* Uninitialized Data:: Output of uninitialized variables. 7703* Label Output:: Output and generation of labels. 7704* Initialization:: General principles of initialization 7705 and termination routines. 7706* Macros for Initialization:: 7707 Specific macros that control the handling of 7708 initialization and termination routines. 7709* Instruction Output:: Output of actual instructions. 7710* Dispatch Tables:: Output of jump tables. 7711* Exception Region Output:: Output of exception region code. 7712* Alignment Output:: Pseudo ops for alignment and skipping data. 7713@end menu 7714 7715@node File Framework 7716@subsection The Overall Framework of an Assembler File 7717@cindex assembler format 7718@cindex output of assembler code 7719 7720@c prevent bad page break with this line 7721This describes the overall framework of an assembly file. 7722 7723@findex default_file_start 7724@deftypefn {Target Hook} void TARGET_ASM_FILE_START (void) 7725Output to @code{asm_out_file} any text which the assembler expects to 7726find at the beginning of a file. The default behavior is controlled 7727by two flags, documented below. Unless your target's assembler is 7728quite unusual, if you override the default, you should call 7729@code{default_file_start} at some point in your target hook. This 7730lets other target files rely on these variables. 7731@end deftypefn 7732 7733@deftypevr {Target Hook} bool TARGET_ASM_FILE_START_APP_OFF 7734If this flag is true, the text of the macro @code{ASM_APP_OFF} will be 7735printed as the very first line in the assembly file, unless 7736@option{-fverbose-asm} is in effect. (If that macro has been defined 7737to the empty string, this variable has no effect.) With the normal 7738definition of @code{ASM_APP_OFF}, the effect is to notify the GNU 7739assembler that it need not bother stripping comments or extra 7740whitespace from its input. This allows it to work a bit faster. 7741 7742The default is false. You should not set it to true unless you have 7743verified that your port does not generate any extra whitespace or 7744comments that will cause GAS to issue errors in NO_APP mode. 7745@end deftypevr 7746 7747@deftypevr {Target Hook} bool TARGET_ASM_FILE_START_FILE_DIRECTIVE 7748If this flag is true, @code{output_file_directive} will be called 7749for the primary source file, immediately after printing 7750@code{ASM_APP_OFF} (if that is enabled). Most ELF assemblers expect 7751this to be done. The default is false. 7752@end deftypevr 7753 7754@deftypefn {Target Hook} void TARGET_ASM_FILE_END (void) 7755Output to @code{asm_out_file} any text which the assembler expects 7756to find at the end of a file. The default is to output nothing. 7757@end deftypefn 7758 7759@deftypefun void file_end_indicate_exec_stack () 7760Some systems use a common convention, the @samp{.note.GNU-stack} 7761special section, to indicate whether or not an object file relies on 7762the stack being executable. If your system uses this convention, you 7763should define @code{TARGET_ASM_FILE_END} to this function. If you 7764need to do other things in that hook, have your hook function call 7765this function. 7766@end deftypefun 7767 7768@deftypefn {Target Hook} void TARGET_ASM_LTO_START (void) 7769Output to @code{asm_out_file} any text which the assembler expects 7770to find at the start of an LTO section. The default is to output 7771nothing. 7772@end deftypefn 7773 7774@deftypefn {Target Hook} void TARGET_ASM_LTO_END (void) 7775Output to @code{asm_out_file} any text which the assembler expects 7776to find at the end of an LTO section. The default is to output 7777nothing. 7778@end deftypefn 7779 7780@deftypefn {Target Hook} void TARGET_ASM_CODE_END (void) 7781Output to @code{asm_out_file} any text which is needed before emitting 7782unwind info and debug info at the end of a file. Some targets emit 7783here PIC setup thunks that cannot be emitted at the end of file, 7784because they couldn't have unwind info then. The default is to output 7785nothing. 7786@end deftypefn 7787 7788@defmac ASM_COMMENT_START 7789A C string constant describing how to begin a comment in the target 7790assembler language. The compiler assumes that the comment will end at 7791the end of the line. 7792@end defmac 7793 7794@defmac ASM_APP_ON 7795A C string constant for text to be output before each @code{asm} 7796statement or group of consecutive ones. Normally this is 7797@code{"#APP"}, which is a comment that has no effect on most 7798assemblers but tells the GNU assembler that it must check the lines 7799that follow for all valid assembler constructs. 7800@end defmac 7801 7802@defmac ASM_APP_OFF 7803A C string constant for text to be output after each @code{asm} 7804statement or group of consecutive ones. Normally this is 7805@code{"#NO_APP"}, which tells the GNU assembler to resume making the 7806time-saving assumptions that are valid for ordinary compiler output. 7807@end defmac 7808 7809@defmac ASM_OUTPUT_SOURCE_FILENAME (@var{stream}, @var{name}) 7810A C statement to output COFF information or DWARF debugging information 7811which indicates that filename @var{name} is the current source file to 7812the stdio stream @var{stream}. 7813 7814This macro need not be defined if the standard form of output 7815for the file format in use is appropriate. 7816@end defmac 7817 7818@deftypefn {Target Hook} void TARGET_ASM_OUTPUT_SOURCE_FILENAME (FILE *@var{file}, const char *@var{name}) 7819Output DWARF debugging information which indicates that filename @var{name} is the current source file to the stdio stream @var{file}. 7820 7821 This target hook need not be defined if the standard form of output for the file format in use is appropriate. 7822@end deftypefn 7823 7824@deftypefn {Target Hook} void TARGET_ASM_OUTPUT_IDENT (const char *@var{name}) 7825Output a string based on @var{name}, suitable for the @samp{#ident} directive, or the equivalent directive or pragma in non-C-family languages. If this hook is not defined, nothing is output for the @samp{#ident} directive. 7826@end deftypefn 7827 7828@defmac OUTPUT_QUOTED_STRING (@var{stream}, @var{string}) 7829A C statement to output the string @var{string} to the stdio stream 7830@var{stream}. If you do not call the function @code{output_quoted_string} 7831in your config files, GCC will only call it to output filenames to 7832the assembler source. So you can use it to canonicalize the format 7833of the filename using this macro. 7834@end defmac 7835 7836@deftypefn {Target Hook} void TARGET_ASM_NAMED_SECTION (const char *@var{name}, unsigned int @var{flags}, tree @var{decl}) 7837Output assembly directives to switch to section @var{name}. The section 7838should have attributes as specified by @var{flags}, which is a bit mask 7839of the @code{SECTION_*} flags defined in @file{output.h}. If @var{decl} 7840is non-NULL, it is the @code{VAR_DECL} or @code{FUNCTION_DECL} with which 7841this section is associated. 7842@end deftypefn 7843 7844@deftypefn {Target Hook} bool TARGET_ASM_ELF_FLAGS_NUMERIC (unsigned int @var{flags}, unsigned int *@var{num}) 7845This hook can be used to encode ELF section flags for which no letter 7846code has been defined in the assembler. It is called by 7847@code{default_asm_named_section} whenever the section flags need to be 7848emitted in the assembler output. If the hook returns true, then the 7849numerical value for ELF section flags should be calculated from 7850@var{flags} and saved in @var{*num}; the value is printed out instead of the 7851normal sequence of letter codes. If the hook is not defined, or if it 7852returns false, then @var{num} is ignored and the traditional letter sequence 7853is emitted. 7854@end deftypefn 7855 7856@deftypefn {Target Hook} {section *} TARGET_ASM_FUNCTION_SECTION (tree @var{decl}, enum node_frequency @var{freq}, bool @var{startup}, bool @var{exit}) 7857Return preferred text (sub)section for function @var{decl}. 7858Main purpose of this function is to separate cold, normal and hot 7859functions. @var{startup} is true when function is known to be used only 7860at startup (from static constructors or it is @code{main()}). 7861@var{exit} is true when function is known to be used only at exit 7862(from static destructors). 7863Return NULL if function should go to default text section. 7864@end deftypefn 7865 7866@deftypefn {Target Hook} void TARGET_ASM_FUNCTION_SWITCHED_TEXT_SECTIONS (FILE *@var{file}, tree @var{decl}, bool @var{new_is_cold}) 7867Used by the target to emit any assembler directives or additional labels needed when a function is partitioned between different sections. Output should be written to @var{file}. The function decl is available as @var{decl} and the new section is `cold' if @var{new_is_cold} is @code{true}. 7868@end deftypefn 7869 7870@deftypevr {Common Target Hook} bool TARGET_HAVE_NAMED_SECTIONS 7871This flag is true if the target supports @code{TARGET_ASM_NAMED_SECTION}. 7872It must not be modified by command-line option processing. 7873@end deftypevr 7874 7875@anchor{TARGET_HAVE_SWITCHABLE_BSS_SECTIONS} 7876@deftypevr {Target Hook} bool TARGET_HAVE_SWITCHABLE_BSS_SECTIONS 7877This flag is true if we can create zeroed data by switching to a BSS 7878section and then using @code{ASM_OUTPUT_SKIP} to allocate the space. 7879This is true on most ELF targets. 7880@end deftypevr 7881 7882@deftypefn {Target Hook} {unsigned int} TARGET_SECTION_TYPE_FLAGS (tree @var{decl}, const char *@var{name}, int @var{reloc}) 7883Choose a set of section attributes for use by @code{TARGET_ASM_NAMED_SECTION} 7884based on a variable or function decl, a section name, and whether or not the 7885declaration's initializer may contain runtime relocations. @var{decl} may be 7886null, in which case read-write data should be assumed. 7887 7888The default version of this function handles choosing code vs data, 7889read-only vs read-write data, and @code{flag_pic}. You should only 7890need to override this if your target has special flags that might be 7891set via @code{__attribute__}. 7892@end deftypefn 7893 7894@deftypefn {Target Hook} int TARGET_ASM_RECORD_GCC_SWITCHES (print_switch_type @var{type}, const char *@var{text}) 7895Provides the target with the ability to record the gcc command line 7896switches that have been passed to the compiler, and options that are 7897enabled. The @var{type} argument specifies what is being recorded. 7898It can take the following values: 7899 7900@table @gcctabopt 7901@item SWITCH_TYPE_PASSED 7902@var{text} is a command line switch that has been set by the user. 7903 7904@item SWITCH_TYPE_ENABLED 7905@var{text} is an option which has been enabled. This might be as a 7906direct result of a command line switch, or because it is enabled by 7907default or because it has been enabled as a side effect of a different 7908command line switch. For example, the @option{-O2} switch enables 7909various different individual optimization passes. 7910 7911@item SWITCH_TYPE_DESCRIPTIVE 7912@var{text} is either NULL or some descriptive text which should be 7913ignored. If @var{text} is NULL then it is being used to warn the 7914target hook that either recording is starting or ending. The first 7915time @var{type} is SWITCH_TYPE_DESCRIPTIVE and @var{text} is NULL, the 7916warning is for start up and the second time the warning is for 7917wind down. This feature is to allow the target hook to make any 7918necessary preparations before it starts to record switches and to 7919perform any necessary tidying up after it has finished recording 7920switches. 7921 7922@item SWITCH_TYPE_LINE_START 7923This option can be ignored by this target hook. 7924 7925@item SWITCH_TYPE_LINE_END 7926This option can be ignored by this target hook. 7927@end table 7928 7929The hook's return value must be zero. Other return values may be 7930supported in the future. 7931 7932By default this hook is set to NULL, but an example implementation is 7933provided for ELF based targets. Called @var{elf_record_gcc_switches}, 7934it records the switches as ASCII text inside a new, string mergeable 7935section in the assembler output file. The name of the new section is 7936provided by the @code{TARGET_ASM_RECORD_GCC_SWITCHES_SECTION} target 7937hook. 7938@end deftypefn 7939 7940@deftypevr {Target Hook} {const char *} TARGET_ASM_RECORD_GCC_SWITCHES_SECTION 7941This is the name of the section that will be created by the example 7942ELF implementation of the @code{TARGET_ASM_RECORD_GCC_SWITCHES} target 7943hook. 7944@end deftypevr 7945 7946@need 2000 7947@node Data Output 7948@subsection Output of Data 7949 7950 7951@deftypevr {Target Hook} {const char *} TARGET_ASM_BYTE_OP 7952@deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_HI_OP 7953@deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_SI_OP 7954@deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_DI_OP 7955@deftypevrx {Target Hook} {const char *} TARGET_ASM_ALIGNED_TI_OP 7956@deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_HI_OP 7957@deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_SI_OP 7958@deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_DI_OP 7959@deftypevrx {Target Hook} {const char *} TARGET_ASM_UNALIGNED_TI_OP 7960These hooks specify assembly directives for creating certain kinds 7961of integer object. The @code{TARGET_ASM_BYTE_OP} directive creates a 7962byte-sized object, the @code{TARGET_ASM_ALIGNED_HI_OP} one creates an 7963aligned two-byte object, and so on. Any of the hooks may be 7964@code{NULL}, indicating that no suitable directive is available. 7965 7966The compiler will print these strings at the start of a new line, 7967followed immediately by the object's initial value. In most cases, 7968the string should contain a tab, a pseudo-op, and then another tab. 7969@end deftypevr 7970 7971@deftypefn {Target Hook} bool TARGET_ASM_INTEGER (rtx @var{x}, unsigned int @var{size}, int @var{aligned_p}) 7972The @code{assemble_integer} function uses this hook to output an 7973integer object. @var{x} is the object's value, @var{size} is its size 7974in bytes and @var{aligned_p} indicates whether it is aligned. The 7975function should return @code{true} if it was able to output the 7976object. If it returns false, @code{assemble_integer} will try to 7977split the object into smaller parts. 7978 7979The default implementation of this hook will use the 7980@code{TARGET_ASM_BYTE_OP} family of strings, returning @code{false} 7981when the relevant string is @code{NULL}. 7982@end deftypefn 7983 7984@deftypefn {Target Hook} void TARGET_ASM_DECL_END (void) 7985Define this hook if the target assembler requires a special marker to 7986terminate an initialized variable declaration. 7987@end deftypefn 7988 7989@deftypefn {Target Hook} bool TARGET_ASM_OUTPUT_ADDR_CONST_EXTRA (FILE *@var{file}, rtx @var{x}) 7990A target hook to recognize @var{rtx} patterns that @code{output_addr_const} 7991can't deal with, and output assembly code to @var{file} corresponding to 7992the pattern @var{x}. This may be used to allow machine-dependent 7993@code{UNSPEC}s to appear within constants. 7994 7995If target hook fails to recognize a pattern, it must return @code{false}, 7996so that a standard error message is printed. If it prints an error message 7997itself, by calling, for example, @code{output_operand_lossage}, it may just 7998return @code{true}. 7999@end deftypefn 8000 8001@defmac ASM_OUTPUT_ASCII (@var{stream}, @var{ptr}, @var{len}) 8002A C statement to output to the stdio stream @var{stream} an assembler 8003instruction to assemble a string constant containing the @var{len} 8004bytes at @var{ptr}. @var{ptr} will be a C expression of type 8005@code{char *} and @var{len} a C expression of type @code{int}. 8006 8007If the assembler has a @code{.ascii} pseudo-op as found in the 8008Berkeley Unix assembler, do not define the macro 8009@code{ASM_OUTPUT_ASCII}. 8010@end defmac 8011 8012@defmac ASM_OUTPUT_FDESC (@var{stream}, @var{decl}, @var{n}) 8013A C statement to output word @var{n} of a function descriptor for 8014@var{decl}. This must be defined if @code{TARGET_VTABLE_USES_DESCRIPTORS} 8015is defined, and is otherwise unused. 8016@end defmac 8017 8018@defmac CONSTANT_POOL_BEFORE_FUNCTION 8019You may define this macro as a C expression. You should define the 8020expression to have a nonzero value if GCC should output the constant 8021pool for a function before the code for the function, or a zero value if 8022GCC should output the constant pool after the function. If you do 8023not define this macro, the usual case, GCC will output the constant 8024pool before the function. 8025@end defmac 8026 8027@defmac ASM_OUTPUT_POOL_PROLOGUE (@var{file}, @var{funname}, @var{fundecl}, @var{size}) 8028A C statement to output assembler commands to define the start of the 8029constant pool for a function. @var{funname} is a string giving 8030the name of the function. Should the return type of the function 8031be required, it can be obtained via @var{fundecl}. @var{size} 8032is the size, in bytes, of the constant pool that will be written 8033immediately after this call. 8034 8035If no constant-pool prefix is required, the usual case, this macro need 8036not be defined. 8037@end defmac 8038 8039@defmac ASM_OUTPUT_SPECIAL_POOL_ENTRY (@var{file}, @var{x}, @var{mode}, @var{align}, @var{labelno}, @var{jumpto}) 8040A C statement (with or without semicolon) to output a constant in the 8041constant pool, if it needs special treatment. (This macro need not do 8042anything for RTL expressions that can be output normally.) 8043 8044The argument @var{file} is the standard I/O stream to output the 8045assembler code on. @var{x} is the RTL expression for the constant to 8046output, and @var{mode} is the machine mode (in case @var{x} is a 8047@samp{const_int}). @var{align} is the required alignment for the value 8048@var{x}; you should output an assembler directive to force this much 8049alignment. 8050 8051The argument @var{labelno} is a number to use in an internal label for 8052the address of this pool entry. The definition of this macro is 8053responsible for outputting the label definition at the proper place. 8054Here is how to do this: 8055 8056@smallexample 8057@code{(*targetm.asm_out.internal_label)} (@var{file}, "LC", @var{labelno}); 8058@end smallexample 8059 8060When you output a pool entry specially, you should end with a 8061@code{goto} to the label @var{jumpto}. This will prevent the same pool 8062entry from being output a second time in the usual manner. 8063 8064You need not define this macro if it would do nothing. 8065@end defmac 8066 8067@defmac ASM_OUTPUT_POOL_EPILOGUE (@var{file} @var{funname} @var{fundecl} @var{size}) 8068A C statement to output assembler commands to at the end of the constant 8069pool for a function. @var{funname} is a string giving the name of the 8070function. Should the return type of the function be required, you can 8071obtain it via @var{fundecl}. @var{size} is the size, in bytes, of the 8072constant pool that GCC wrote immediately before this call. 8073 8074If no constant-pool epilogue is required, the usual case, you need not 8075define this macro. 8076@end defmac 8077 8078@defmac IS_ASM_LOGICAL_LINE_SEPARATOR (@var{C}, @var{STR}) 8079Define this macro as a C expression which is nonzero if @var{C} is 8080used as a logical line separator by the assembler. @var{STR} points 8081to the position in the string where @var{C} was found; this can be used if 8082a line separator uses multiple characters. 8083 8084If you do not define this macro, the default is that only 8085the character @samp{;} is treated as a logical line separator. 8086@end defmac 8087 8088@deftypevr {Target Hook} {const char *} TARGET_ASM_OPEN_PAREN 8089@deftypevrx {Target Hook} {const char *} TARGET_ASM_CLOSE_PAREN 8090These target hooks are C string constants, describing the syntax in the 8091assembler for grouping arithmetic expressions. If not overridden, they 8092default to normal parentheses, which is correct for most assemblers. 8093@end deftypevr 8094 8095These macros are provided by @file{real.h} for writing the definitions 8096of @code{ASM_OUTPUT_DOUBLE} and the like: 8097 8098@defmac REAL_VALUE_TO_TARGET_SINGLE (@var{x}, @var{l}) 8099@defmacx REAL_VALUE_TO_TARGET_DOUBLE (@var{x}, @var{l}) 8100@defmacx REAL_VALUE_TO_TARGET_LONG_DOUBLE (@var{x}, @var{l}) 8101@defmacx REAL_VALUE_TO_TARGET_DECIMAL32 (@var{x}, @var{l}) 8102@defmacx REAL_VALUE_TO_TARGET_DECIMAL64 (@var{x}, @var{l}) 8103@defmacx REAL_VALUE_TO_TARGET_DECIMAL128 (@var{x}, @var{l}) 8104These translate @var{x}, of type @code{REAL_VALUE_TYPE}, to the 8105target's floating point representation, and store its bit pattern in 8106the variable @var{l}. For @code{REAL_VALUE_TO_TARGET_SINGLE} and 8107@code{REAL_VALUE_TO_TARGET_DECIMAL32}, this variable should be a 8108simple @code{long int}. For the others, it should be an array of 8109@code{long int}. The number of elements in this array is determined 8110by the size of the desired target floating point data type: 32 bits of 8111it go in each @code{long int} array element. Each array element holds 811232 bits of the result, even if @code{long int} is wider than 32 bits 8113on the host machine. 8114 8115The array element values are designed so that you can print them out 8116using @code{fprintf} in the order they should appear in the target 8117machine's memory. 8118@end defmac 8119 8120@node Uninitialized Data 8121@subsection Output of Uninitialized Variables 8122 8123Each of the macros in this section is used to do the whole job of 8124outputting a single uninitialized variable. 8125 8126@defmac ASM_OUTPUT_COMMON (@var{stream}, @var{name}, @var{size}, @var{rounded}) 8127A C statement (sans semicolon) to output to the stdio stream 8128@var{stream} the assembler definition of a common-label named 8129@var{name} whose size is @var{size} bytes. The variable @var{rounded} 8130is the size rounded up to whatever alignment the caller wants. It is 8131possible that @var{size} may be zero, for instance if a struct with no 8132other member than a zero-length array is defined. In this case, the 8133backend must output a symbol definition that allocates at least one 8134byte, both so that the address of the resulting object does not compare 8135equal to any other, and because some object formats cannot even express 8136the concept of a zero-sized common symbol, as that is how they represent 8137an ordinary undefined external. 8138 8139Use the expression @code{assemble_name (@var{stream}, @var{name})} to 8140output the name itself; before and after that, output the additional 8141assembler syntax for defining the name, and a newline. 8142 8143This macro controls how the assembler definitions of uninitialized 8144common global variables are output. 8145@end defmac 8146 8147@defmac ASM_OUTPUT_ALIGNED_COMMON (@var{stream}, @var{name}, @var{size}, @var{alignment}) 8148Like @code{ASM_OUTPUT_COMMON} except takes the required alignment as a 8149separate, explicit argument. If you define this macro, it is used in 8150place of @code{ASM_OUTPUT_COMMON}, and gives you more flexibility in 8151handling the required alignment of the variable. The alignment is specified 8152as the number of bits. 8153@end defmac 8154 8155@defmac ASM_OUTPUT_ALIGNED_DECL_COMMON (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment}) 8156Like @code{ASM_OUTPUT_ALIGNED_COMMON} except that @var{decl} of the 8157variable to be output, if there is one, or @code{NULL_TREE} if there 8158is no corresponding variable. If you define this macro, GCC will use it 8159in place of both @code{ASM_OUTPUT_COMMON} and 8160@code{ASM_OUTPUT_ALIGNED_COMMON}. Define this macro when you need to see 8161the variable's decl in order to chose what to output. 8162@end defmac 8163 8164@defmac ASM_OUTPUT_ALIGNED_BSS (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment}) 8165A C statement (sans semicolon) to output to the stdio stream 8166@var{stream} the assembler definition of uninitialized global @var{decl} named 8167@var{name} whose size is @var{size} bytes. The variable @var{alignment} 8168is the alignment specified as the number of bits. 8169 8170Try to use function @code{asm_output_aligned_bss} defined in file 8171@file{varasm.c} when defining this macro. If unable, use the expression 8172@code{assemble_name (@var{stream}, @var{name})} to output the name itself; 8173before and after that, output the additional assembler syntax for defining 8174the name, and a newline. 8175 8176There are two ways of handling global BSS@. One is to define this macro. 8177The other is to have @code{TARGET_ASM_SELECT_SECTION} return a 8178switchable BSS section (@pxref{TARGET_HAVE_SWITCHABLE_BSS_SECTIONS}). 8179You do not need to do both. 8180 8181Some languages do not have @code{common} data, and require a 8182non-common form of global BSS in order to handle uninitialized globals 8183efficiently. C++ is one example of this. However, if the target does 8184not support global BSS, the front end may choose to make globals 8185common in order to save space in the object file. 8186@end defmac 8187 8188@defmac ASM_OUTPUT_LOCAL (@var{stream}, @var{name}, @var{size}, @var{rounded}) 8189A C statement (sans semicolon) to output to the stdio stream 8190@var{stream} the assembler definition of a local-common-label named 8191@var{name} whose size is @var{size} bytes. The variable @var{rounded} 8192is the size rounded up to whatever alignment the caller wants. 8193 8194Use the expression @code{assemble_name (@var{stream}, @var{name})} to 8195output the name itself; before and after that, output the additional 8196assembler syntax for defining the name, and a newline. 8197 8198This macro controls how the assembler definitions of uninitialized 8199static variables are output. 8200@end defmac 8201 8202@defmac ASM_OUTPUT_ALIGNED_LOCAL (@var{stream}, @var{name}, @var{size}, @var{alignment}) 8203Like @code{ASM_OUTPUT_LOCAL} except takes the required alignment as a 8204separate, explicit argument. If you define this macro, it is used in 8205place of @code{ASM_OUTPUT_LOCAL}, and gives you more flexibility in 8206handling the required alignment of the variable. The alignment is specified 8207as the number of bits. 8208@end defmac 8209 8210@defmac ASM_OUTPUT_ALIGNED_DECL_LOCAL (@var{stream}, @var{decl}, @var{name}, @var{size}, @var{alignment}) 8211Like @code{ASM_OUTPUT_ALIGNED_DECL} except that @var{decl} of the 8212variable to be output, if there is one, or @code{NULL_TREE} if there 8213is no corresponding variable. If you define this macro, GCC will use it 8214in place of both @code{ASM_OUTPUT_DECL} and 8215@code{ASM_OUTPUT_ALIGNED_DECL}. Define this macro when you need to see 8216the variable's decl in order to chose what to output. 8217@end defmac 8218 8219@node Label Output 8220@subsection Output and Generation of Labels 8221 8222@c prevent bad page break with this line 8223This is about outputting labels. 8224 8225@findex assemble_name 8226@defmac ASM_OUTPUT_LABEL (@var{stream}, @var{name}) 8227A C statement (sans semicolon) to output to the stdio stream 8228@var{stream} the assembler definition of a label named @var{name}. 8229Use the expression @code{assemble_name (@var{stream}, @var{name})} to 8230output the name itself; before and after that, output the additional 8231assembler syntax for defining the name, and a newline. A default 8232definition of this macro is provided which is correct for most systems. 8233@end defmac 8234 8235@defmac ASM_OUTPUT_FUNCTION_LABEL (@var{stream}, @var{name}, @var{decl}) 8236A C statement (sans semicolon) to output to the stdio stream 8237@var{stream} the assembler definition of a label named @var{name} of 8238a function. 8239Use the expression @code{assemble_name (@var{stream}, @var{name})} to 8240output the name itself; before and after that, output the additional 8241assembler syntax for defining the name, and a newline. A default 8242definition of this macro is provided which is correct for most systems. 8243 8244If this macro is not defined, then the function name is defined in the 8245usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}). 8246@end defmac 8247 8248@findex assemble_name_raw 8249@defmac ASM_OUTPUT_INTERNAL_LABEL (@var{stream}, @var{name}) 8250Identical to @code{ASM_OUTPUT_LABEL}, except that @var{name} is known 8251to refer to a compiler-generated label. The default definition uses 8252@code{assemble_name_raw}, which is like @code{assemble_name} except 8253that it is more efficient. 8254@end defmac 8255 8256@defmac SIZE_ASM_OP 8257A C string containing the appropriate assembler directive to specify the 8258size of a symbol, without any arguments. On systems that use ELF, the 8259default (in @file{config/elfos.h}) is @samp{"\t.size\t"}; on other 8260systems, the default is not to define this macro. 8261 8262Define this macro only if it is correct to use the default definitions 8263of @code{ASM_OUTPUT_SIZE_DIRECTIVE} and @code{ASM_OUTPUT_MEASURED_SIZE} 8264for your system. If you need your own custom definitions of those 8265macros, or if you do not need explicit symbol sizes at all, do not 8266define this macro. 8267@end defmac 8268 8269@defmac ASM_OUTPUT_SIZE_DIRECTIVE (@var{stream}, @var{name}, @var{size}) 8270A C statement (sans semicolon) to output to the stdio stream 8271@var{stream} a directive telling the assembler that the size of the 8272symbol @var{name} is @var{size}. @var{size} is a @code{HOST_WIDE_INT}. 8273If you define @code{SIZE_ASM_OP}, a default definition of this macro is 8274provided. 8275@end defmac 8276 8277@defmac ASM_OUTPUT_MEASURED_SIZE (@var{stream}, @var{name}) 8278A C statement (sans semicolon) to output to the stdio stream 8279@var{stream} a directive telling the assembler to calculate the size of 8280the symbol @var{name} by subtracting its address from the current 8281address. 8282 8283If you define @code{SIZE_ASM_OP}, a default definition of this macro is 8284provided. The default assumes that the assembler recognizes a special 8285@samp{.} symbol as referring to the current address, and can calculate 8286the difference between this and another symbol. If your assembler does 8287not recognize @samp{.} or cannot do calculations with it, you will need 8288to redefine @code{ASM_OUTPUT_MEASURED_SIZE} to use some other technique. 8289@end defmac 8290 8291@defmac NO_DOLLAR_IN_LABEL 8292Define this macro if the assembler does not accept the character 8293@samp{$} in label names. By default constructors and destructors in 8294G++ have @samp{$} in the identifiers. If this macro is defined, 8295@samp{.} is used instead. 8296@end defmac 8297 8298@defmac NO_DOT_IN_LABEL 8299Define this macro if the assembler does not accept the character 8300@samp{.} in label names. By default constructors and destructors in G++ 8301have names that use @samp{.}. If this macro is defined, these names 8302are rewritten to avoid @samp{.}. 8303@end defmac 8304 8305@defmac TYPE_ASM_OP 8306A C string containing the appropriate assembler directive to specify the 8307type of a symbol, without any arguments. On systems that use ELF, the 8308default (in @file{config/elfos.h}) is @samp{"\t.type\t"}; on other 8309systems, the default is not to define this macro. 8310 8311Define this macro only if it is correct to use the default definition of 8312@code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own 8313custom definition of this macro, or if you do not need explicit symbol 8314types at all, do not define this macro. 8315@end defmac 8316 8317@defmac TYPE_OPERAND_FMT 8318A C string which specifies (using @code{printf} syntax) the format of 8319the second operand to @code{TYPE_ASM_OP}. On systems that use ELF, the 8320default (in @file{config/elfos.h}) is @samp{"@@%s"}; on other systems, 8321the default is not to define this macro. 8322 8323Define this macro only if it is correct to use the default definition of 8324@code{ASM_OUTPUT_TYPE_DIRECTIVE} for your system. If you need your own 8325custom definition of this macro, or if you do not need explicit symbol 8326types at all, do not define this macro. 8327@end defmac 8328 8329@defmac ASM_OUTPUT_TYPE_DIRECTIVE (@var{stream}, @var{type}) 8330A C statement (sans semicolon) to output to the stdio stream 8331@var{stream} a directive telling the assembler that the type of the 8332symbol @var{name} is @var{type}. @var{type} is a C string; currently, 8333that string is always either @samp{"function"} or @samp{"object"}, but 8334you should not count on this. 8335 8336If you define @code{TYPE_ASM_OP} and @code{TYPE_OPERAND_FMT}, a default 8337definition of this macro is provided. 8338@end defmac 8339 8340@defmac ASM_DECLARE_FUNCTION_NAME (@var{stream}, @var{name}, @var{decl}) 8341A C statement (sans semicolon) to output to the stdio stream 8342@var{stream} any text necessary for declaring the name @var{name} of a 8343function which is being defined. This macro is responsible for 8344outputting the label definition (perhaps using 8345@code{ASM_OUTPUT_FUNCTION_LABEL}). The argument @var{decl} is the 8346@code{FUNCTION_DECL} tree node representing the function. 8347 8348If this macro is not defined, then the function name is defined in the 8349usual manner as a label (by means of @code{ASM_OUTPUT_FUNCTION_LABEL}). 8350 8351You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in the definition 8352of this macro. 8353@end defmac 8354 8355@defmac ASM_DECLARE_FUNCTION_SIZE (@var{stream}, @var{name}, @var{decl}) 8356A C statement (sans semicolon) to output to the stdio stream 8357@var{stream} any text necessary for declaring the size of a function 8358which is being defined. The argument @var{name} is the name of the 8359function. The argument @var{decl} is the @code{FUNCTION_DECL} tree node 8360representing the function. 8361 8362If this macro is not defined, then the function size is not defined. 8363 8364You may wish to use @code{ASM_OUTPUT_MEASURED_SIZE} in the definition 8365of this macro. 8366@end defmac 8367 8368@defmac ASM_DECLARE_COLD_FUNCTION_NAME (@var{stream}, @var{name}, @var{decl}) 8369A C statement (sans semicolon) to output to the stdio stream 8370@var{stream} any text necessary for declaring the name @var{name} of a 8371cold function partition which is being defined. This macro is responsible 8372for outputting the label definition (perhaps using 8373@code{ASM_OUTPUT_FUNCTION_LABEL}). The argument @var{decl} is the 8374@code{FUNCTION_DECL} tree node representing the function. 8375 8376If this macro is not defined, then the cold partition name is defined in the 8377usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}). 8378 8379You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in the definition 8380of this macro. 8381@end defmac 8382 8383@defmac ASM_DECLARE_COLD_FUNCTION_SIZE (@var{stream}, @var{name}, @var{decl}) 8384A C statement (sans semicolon) to output to the stdio stream 8385@var{stream} any text necessary for declaring the size of a cold function 8386partition which is being defined. The argument @var{name} is the name of the 8387cold partition of the function. The argument @var{decl} is the 8388@code{FUNCTION_DECL} tree node representing the function. 8389 8390If this macro is not defined, then the partition size is not defined. 8391 8392You may wish to use @code{ASM_OUTPUT_MEASURED_SIZE} in the definition 8393of this macro. 8394@end defmac 8395 8396@defmac ASM_DECLARE_OBJECT_NAME (@var{stream}, @var{name}, @var{decl}) 8397A C statement (sans semicolon) to output to the stdio stream 8398@var{stream} any text necessary for declaring the name @var{name} of an 8399initialized variable which is being defined. This macro must output the 8400label definition (perhaps using @code{ASM_OUTPUT_LABEL}). The argument 8401@var{decl} is the @code{VAR_DECL} tree node representing the variable. 8402 8403If this macro is not defined, then the variable name is defined in the 8404usual manner as a label (by means of @code{ASM_OUTPUT_LABEL}). 8405 8406You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} and/or 8407@code{ASM_OUTPUT_SIZE_DIRECTIVE} in the definition of this macro. 8408@end defmac 8409 8410@deftypefn {Target Hook} void TARGET_ASM_DECLARE_CONSTANT_NAME (FILE *@var{file}, const char *@var{name}, const_tree @var{expr}, HOST_WIDE_INT @var{size}) 8411A target hook to output to the stdio stream @var{file} any text necessary 8412for declaring the name @var{name} of a constant which is being defined. This 8413target hook is responsible for outputting the label definition (perhaps using 8414@code{assemble_label}). The argument @var{exp} is the value of the constant, 8415and @var{size} is the size of the constant in bytes. The @var{name} 8416will be an internal label. 8417 8418The default version of this target hook, define the @var{name} in the 8419usual manner as a label (by means of @code{assemble_label}). 8420 8421You may wish to use @code{ASM_OUTPUT_TYPE_DIRECTIVE} in this target hook. 8422@end deftypefn 8423 8424@defmac ASM_DECLARE_REGISTER_GLOBAL (@var{stream}, @var{decl}, @var{regno}, @var{name}) 8425A C statement (sans semicolon) to output to the stdio stream 8426@var{stream} any text necessary for claiming a register @var{regno} 8427for a global variable @var{decl} with name @var{name}. 8428 8429If you don't define this macro, that is equivalent to defining it to do 8430nothing. 8431@end defmac 8432 8433@defmac ASM_FINISH_DECLARE_OBJECT (@var{stream}, @var{decl}, @var{toplevel}, @var{atend}) 8434A C statement (sans semicolon) to finish up declaring a variable name 8435once the compiler has processed its initializer fully and thus has had a 8436chance to determine the size of an array when controlled by an 8437initializer. This is used on systems where it's necessary to declare 8438something about the size of the object. 8439 8440If you don't define this macro, that is equivalent to defining it to do 8441nothing. 8442 8443You may wish to use @code{ASM_OUTPUT_SIZE_DIRECTIVE} and/or 8444@code{ASM_OUTPUT_MEASURED_SIZE} in the definition of this macro. 8445@end defmac 8446 8447@deftypefn {Target Hook} void TARGET_ASM_GLOBALIZE_LABEL (FILE *@var{stream}, const char *@var{name}) 8448This target hook is a function to output to the stdio stream 8449@var{stream} some commands that will make the label @var{name} global; 8450that is, available for reference from other files. 8451 8452The default implementation relies on a proper definition of 8453@code{GLOBAL_ASM_OP}. 8454@end deftypefn 8455 8456@deftypefn {Target Hook} void TARGET_ASM_GLOBALIZE_DECL_NAME (FILE *@var{stream}, tree @var{decl}) 8457This target hook is a function to output to the stdio stream 8458@var{stream} some commands that will make the name associated with @var{decl} 8459global; that is, available for reference from other files. 8460 8461The default implementation uses the TARGET_ASM_GLOBALIZE_LABEL target hook. 8462@end deftypefn 8463 8464@deftypefn {Target Hook} void TARGET_ASM_ASSEMBLE_UNDEFINED_DECL (FILE *@var{stream}, const char *@var{name}, const_tree @var{decl}) 8465This target hook is a function to output to the stdio stream 8466@var{stream} some commands that will declare the name associated with 8467@var{decl} which is not defined in the current translation unit. Most 8468assemblers do not require anything to be output in this case. 8469@end deftypefn 8470 8471@defmac ASM_WEAKEN_LABEL (@var{stream}, @var{name}) 8472A C statement (sans semicolon) to output to the stdio stream 8473@var{stream} some commands that will make the label @var{name} weak; 8474that is, available for reference from other files but only used if 8475no other definition is available. Use the expression 8476@code{assemble_name (@var{stream}, @var{name})} to output the name 8477itself; before and after that, output the additional assembler syntax 8478for making that name weak, and a newline. 8479 8480If you don't define this macro or @code{ASM_WEAKEN_DECL}, GCC will not 8481support weak symbols and you should not define the @code{SUPPORTS_WEAK} 8482macro. 8483@end defmac 8484 8485@defmac ASM_WEAKEN_DECL (@var{stream}, @var{decl}, @var{name}, @var{value}) 8486Combines (and replaces) the function of @code{ASM_WEAKEN_LABEL} and 8487@code{ASM_OUTPUT_WEAK_ALIAS}, allowing access to the associated function 8488or variable decl. If @var{value} is not @code{NULL}, this C statement 8489should output to the stdio stream @var{stream} assembler code which 8490defines (equates) the weak symbol @var{name} to have the value 8491@var{value}. If @var{value} is @code{NULL}, it should output commands 8492to make @var{name} weak. 8493@end defmac 8494 8495@defmac ASM_OUTPUT_WEAKREF (@var{stream}, @var{decl}, @var{name}, @var{value}) 8496Outputs a directive that enables @var{name} to be used to refer to 8497symbol @var{value} with weak-symbol semantics. @code{decl} is the 8498declaration of @code{name}. 8499@end defmac 8500 8501@defmac SUPPORTS_WEAK 8502A preprocessor constant expression which evaluates to true if the target 8503supports weak symbols. 8504 8505If you don't define this macro, @file{defaults.h} provides a default 8506definition. If either @code{ASM_WEAKEN_LABEL} or @code{ASM_WEAKEN_DECL} 8507is defined, the default definition is @samp{1}; otherwise, it is @samp{0}. 8508@end defmac 8509 8510@defmac TARGET_SUPPORTS_WEAK 8511A C expression which evaluates to true if the target supports weak symbols. 8512 8513If you don't define this macro, @file{defaults.h} provides a default 8514definition. The default definition is @samp{(SUPPORTS_WEAK)}. Define 8515this macro if you want to control weak symbol support with a compiler 8516flag such as @option{-melf}. 8517@end defmac 8518 8519@defmac MAKE_DECL_ONE_ONLY (@var{decl}) 8520A C statement (sans semicolon) to mark @var{decl} to be emitted as a 8521public symbol such that extra copies in multiple translation units will 8522be discarded by the linker. Define this macro if your object file 8523format provides support for this concept, such as the @samp{COMDAT} 8524section flags in the Microsoft Windows PE/COFF format, and this support 8525requires changes to @var{decl}, such as putting it in a separate section. 8526@end defmac 8527 8528@defmac SUPPORTS_ONE_ONLY 8529A C expression which evaluates to true if the target supports one-only 8530semantics. 8531 8532If you don't define this macro, @file{varasm.c} provides a default 8533definition. If @code{MAKE_DECL_ONE_ONLY} is defined, the default 8534definition is @samp{1}; otherwise, it is @samp{0}. Define this macro if 8535you want to control one-only symbol support with a compiler flag, or if 8536setting the @code{DECL_ONE_ONLY} flag is enough to mark a declaration to 8537be emitted as one-only. 8538@end defmac 8539 8540@deftypefn {Target Hook} void TARGET_ASM_ASSEMBLE_VISIBILITY (tree @var{decl}, int @var{visibility}) 8541This target hook is a function to output to @var{asm_out_file} some 8542commands that will make the symbol(s) associated with @var{decl} have 8543hidden, protected or internal visibility as specified by @var{visibility}. 8544@end deftypefn 8545 8546@defmac TARGET_WEAK_NOT_IN_ARCHIVE_TOC 8547A C expression that evaluates to true if the target's linker expects 8548that weak symbols do not appear in a static archive's table of contents. 8549The default is @code{0}. 8550 8551Leaving weak symbols out of an archive's table of contents means that, 8552if a symbol will only have a definition in one translation unit and 8553will have undefined references from other translation units, that 8554symbol should not be weak. Defining this macro to be nonzero will 8555thus have the effect that certain symbols that would normally be weak 8556(explicit template instantiations, and vtables for polymorphic classes 8557with noninline key methods) will instead be nonweak. 8558 8559The C++ ABI requires this macro to be zero. Define this macro for 8560targets where full C++ ABI compliance is impossible and where linker 8561restrictions require weak symbols to be left out of a static archive's 8562table of contents. 8563@end defmac 8564 8565@defmac ASM_OUTPUT_EXTERNAL (@var{stream}, @var{decl}, @var{name}) 8566A C statement (sans semicolon) to output to the stdio stream 8567@var{stream} any text necessary for declaring the name of an external 8568symbol named @var{name} which is referenced in this compilation but 8569not defined. The value of @var{decl} is the tree node for the 8570declaration. 8571 8572This macro need not be defined if it does not need to output anything. 8573The GNU assembler and most Unix assemblers don't require anything. 8574@end defmac 8575 8576@deftypefn {Target Hook} void TARGET_ASM_EXTERNAL_LIBCALL (rtx @var{symref}) 8577This target hook is a function to output to @var{asm_out_file} an assembler 8578pseudo-op to declare a library function name external. The name of the 8579library function is given by @var{symref}, which is a @code{symbol_ref}. 8580@end deftypefn 8581 8582@deftypefn {Target Hook} void TARGET_ASM_MARK_DECL_PRESERVED (const char *@var{symbol}) 8583This target hook is a function to output to @var{asm_out_file} an assembler 8584directive to annotate @var{symbol} as used. The Darwin target uses the 8585.no_dead_code_strip directive. 8586@end deftypefn 8587 8588@defmac ASM_OUTPUT_LABELREF (@var{stream}, @var{name}) 8589A C statement (sans semicolon) to output to the stdio stream 8590@var{stream} a reference in assembler syntax to a label named 8591@var{name}. This should add @samp{_} to the front of the name, if that 8592is customary on your operating system, as it is in most Berkeley Unix 8593systems. This macro is used in @code{assemble_name}. 8594@end defmac 8595 8596@deftypefn {Target Hook} tree TARGET_MANGLE_ASSEMBLER_NAME (const char *@var{name}) 8597Given a symbol @var{name}, perform same mangling as @code{varasm.c}'s @code{assemble_name}, but in memory rather than to a file stream, returning result as an @code{IDENTIFIER_NODE}. Required for correct LTO symtabs. The default implementation calls the @code{TARGET_STRIP_NAME_ENCODING} hook and then prepends the @code{USER_LABEL_PREFIX}, if any. 8598@end deftypefn 8599 8600@defmac ASM_OUTPUT_SYMBOL_REF (@var{stream}, @var{sym}) 8601A C statement (sans semicolon) to output a reference to 8602@code{SYMBOL_REF} @var{sym}. If not defined, @code{assemble_name} 8603will be used to output the name of the symbol. This macro may be used 8604to modify the way a symbol is referenced depending on information 8605encoded by @code{TARGET_ENCODE_SECTION_INFO}. 8606@end defmac 8607 8608@defmac ASM_OUTPUT_LABEL_REF (@var{stream}, @var{buf}) 8609A C statement (sans semicolon) to output a reference to @var{buf}, the 8610result of @code{ASM_GENERATE_INTERNAL_LABEL}. If not defined, 8611@code{assemble_name} will be used to output the name of the symbol. 8612This macro is not used by @code{output_asm_label}, or the @code{%l} 8613specifier that calls it; the intention is that this macro should be set 8614when it is necessary to output a label differently when its address is 8615being taken. 8616@end defmac 8617 8618@deftypefn {Target Hook} void TARGET_ASM_INTERNAL_LABEL (FILE *@var{stream}, const char *@var{prefix}, unsigned long @var{labelno}) 8619A function to output to the stdio stream @var{stream} a label whose 8620name is made from the string @var{prefix} and the number @var{labelno}. 8621 8622It is absolutely essential that these labels be distinct from the labels 8623used for user-level functions and variables. Otherwise, certain programs 8624will have name conflicts with internal labels. 8625 8626It is desirable to exclude internal labels from the symbol table of the 8627object file. Most assemblers have a naming convention for labels that 8628should be excluded; on many systems, the letter @samp{L} at the 8629beginning of a label has this effect. You should find out what 8630convention your system uses, and follow it. 8631 8632The default version of this function utilizes @code{ASM_GENERATE_INTERNAL_LABEL}. 8633@end deftypefn 8634 8635@defmac ASM_OUTPUT_DEBUG_LABEL (@var{stream}, @var{prefix}, @var{num}) 8636A C statement to output to the stdio stream @var{stream} a debug info 8637label whose name is made from the string @var{prefix} and the number 8638@var{num}. This is useful for VLIW targets, where debug info labels 8639may need to be treated differently than branch target labels. On some 8640systems, branch target labels must be at the beginning of instruction 8641bundles, but debug info labels can occur in the middle of instruction 8642bundles. 8643 8644If this macro is not defined, then @code{(*targetm.asm_out.internal_label)} will be 8645used. 8646@end defmac 8647 8648@defmac ASM_GENERATE_INTERNAL_LABEL (@var{string}, @var{prefix}, @var{num}) 8649A C statement to store into the string @var{string} a label whose name 8650is made from the string @var{prefix} and the number @var{num}. 8651 8652This string, when output subsequently by @code{assemble_name}, should 8653produce the output that @code{(*targetm.asm_out.internal_label)} would produce 8654with the same @var{prefix} and @var{num}. 8655 8656If the string begins with @samp{*}, then @code{assemble_name} will 8657output the rest of the string unchanged. It is often convenient for 8658@code{ASM_GENERATE_INTERNAL_LABEL} to use @samp{*} in this way. If the 8659string doesn't start with @samp{*}, then @code{ASM_OUTPUT_LABELREF} gets 8660to output the string, and may change it. (Of course, 8661@code{ASM_OUTPUT_LABELREF} is also part of your machine description, so 8662you should know what it does on your machine.) 8663@end defmac 8664 8665@defmac ASM_FORMAT_PRIVATE_NAME (@var{outvar}, @var{name}, @var{number}) 8666A C expression to assign to @var{outvar} (which is a variable of type 8667@code{char *}) a newly allocated string made from the string 8668@var{name} and the number @var{number}, with some suitable punctuation 8669added. Use @code{alloca} to get space for the string. 8670 8671The string will be used as an argument to @code{ASM_OUTPUT_LABELREF} to 8672produce an assembler label for an internal static variable whose name is 8673@var{name}. Therefore, the string must be such as to result in valid 8674assembler code. The argument @var{number} is different each time this 8675macro is executed; it prevents conflicts between similarly-named 8676internal static variables in different scopes. 8677 8678Ideally this string should not be a valid C identifier, to prevent any 8679conflict with the user's own symbols. Most assemblers allow periods 8680or percent signs in assembler symbols; putting at least one of these 8681between the name and the number will suffice. 8682 8683If this macro is not defined, a default definition will be provided 8684which is correct for most systems. 8685@end defmac 8686 8687@defmac ASM_OUTPUT_DEF (@var{stream}, @var{name}, @var{value}) 8688A C statement to output to the stdio stream @var{stream} assembler code 8689which defines (equates) the symbol @var{name} to have the value @var{value}. 8690 8691@findex SET_ASM_OP 8692If @code{SET_ASM_OP} is defined, a default definition is provided which is 8693correct for most systems. 8694@end defmac 8695 8696@defmac ASM_OUTPUT_DEF_FROM_DECLS (@var{stream}, @var{decl_of_name}, @var{decl_of_value}) 8697A C statement to output to the stdio stream @var{stream} assembler code 8698which defines (equates) the symbol whose tree node is @var{decl_of_name} 8699to have the value of the tree node @var{decl_of_value}. This macro will 8700be used in preference to @samp{ASM_OUTPUT_DEF} if it is defined and if 8701the tree nodes are available. 8702 8703@findex SET_ASM_OP 8704If @code{SET_ASM_OP} is defined, a default definition is provided which is 8705correct for most systems. 8706@end defmac 8707 8708@defmac TARGET_DEFERRED_OUTPUT_DEFS (@var{decl_of_name}, @var{decl_of_value}) 8709A C statement that evaluates to true if the assembler code which defines 8710(equates) the symbol whose tree node is @var{decl_of_name} to have the value 8711of the tree node @var{decl_of_value} should be emitted near the end of the 8712current compilation unit. The default is to not defer output of defines. 8713This macro affects defines output by @samp{ASM_OUTPUT_DEF} and 8714@samp{ASM_OUTPUT_DEF_FROM_DECLS}. 8715@end defmac 8716 8717@defmac ASM_OUTPUT_WEAK_ALIAS (@var{stream}, @var{name}, @var{value}) 8718A C statement to output to the stdio stream @var{stream} assembler code 8719which defines (equates) the weak symbol @var{name} to have the value 8720@var{value}. If @var{value} is @code{NULL}, it defines @var{name} as 8721an undefined weak symbol. 8722 8723Define this macro if the target only supports weak aliases; define 8724@code{ASM_OUTPUT_DEF} instead if possible. 8725@end defmac 8726 8727@defmac OBJC_GEN_METHOD_LABEL (@var{buf}, @var{is_inst}, @var{class_name}, @var{cat_name}, @var{sel_name}) 8728Define this macro to override the default assembler names used for 8729Objective-C methods. 8730 8731The default name is a unique method number followed by the name of the 8732class (e.g.@: @samp{_1_Foo}). For methods in categories, the name of 8733the category is also included in the assembler name (e.g.@: 8734@samp{_1_Foo_Bar}). 8735 8736These names are safe on most systems, but make debugging difficult since 8737the method's selector is not present in the name. Therefore, particular 8738systems define other ways of computing names. 8739 8740@var{buf} is an expression of type @code{char *} which gives you a 8741buffer in which to store the name; its length is as long as 8742@var{class_name}, @var{cat_name} and @var{sel_name} put together, plus 874350 characters extra. 8744 8745The argument @var{is_inst} specifies whether the method is an instance 8746method or a class method; @var{class_name} is the name of the class; 8747@var{cat_name} is the name of the category (or @code{NULL} if the method is not 8748in a category); and @var{sel_name} is the name of the selector. 8749 8750On systems where the assembler can handle quoted names, you can use this 8751macro to provide more human-readable names. 8752@end defmac 8753 8754@node Initialization 8755@subsection How Initialization Functions Are Handled 8756@cindex initialization routines 8757@cindex termination routines 8758@cindex constructors, output of 8759@cindex destructors, output of 8760 8761The compiled code for certain languages includes @dfn{constructors} 8762(also called @dfn{initialization routines})---functions to initialize 8763data in the program when the program is started. These functions need 8764to be called before the program is ``started''---that is to say, before 8765@code{main} is called. 8766 8767Compiling some languages generates @dfn{destructors} (also called 8768@dfn{termination routines}) that should be called when the program 8769terminates. 8770 8771To make the initialization and termination functions work, the compiler 8772must output something in the assembler code to cause those functions to 8773be called at the appropriate time. When you port the compiler to a new 8774system, you need to specify how to do this. 8775 8776There are two major ways that GCC currently supports the execution of 8777initialization and termination functions. Each way has two variants. 8778Much of the structure is common to all four variations. 8779 8780@findex __CTOR_LIST__ 8781@findex __DTOR_LIST__ 8782The linker must build two lists of these functions---a list of 8783initialization functions, called @code{__CTOR_LIST__}, and a list of 8784termination functions, called @code{__DTOR_LIST__}. 8785 8786Each list always begins with an ignored function pointer (which may hold 87870, @minus{}1, or a count of the function pointers after it, depending on 8788the environment). This is followed by a series of zero or more function 8789pointers to constructors (or destructors), followed by a function 8790pointer containing zero. 8791 8792Depending on the operating system and its executable file format, either 8793@file{crtstuff.c} or @file{libgcc2.c} traverses these lists at startup 8794time and exit time. Constructors are called in reverse order of the 8795list; destructors in forward order. 8796 8797The best way to handle static constructors works only for object file 8798formats which provide arbitrarily-named sections. A section is set 8799aside for a list of constructors, and another for a list of destructors. 8800Traditionally these are called @samp{.ctors} and @samp{.dtors}. Each 8801object file that defines an initialization function also puts a word in 8802the constructor section to point to that function. The linker 8803accumulates all these words into one contiguous @samp{.ctors} section. 8804Termination functions are handled similarly. 8805 8806This method will be chosen as the default by @file{target-def.h} if 8807@code{TARGET_ASM_NAMED_SECTION} is defined. A target that does not 8808support arbitrary sections, but does support special designated 8809constructor and destructor sections may define @code{CTORS_SECTION_ASM_OP} 8810and @code{DTORS_SECTION_ASM_OP} to achieve the same effect. 8811 8812When arbitrary sections are available, there are two variants, depending 8813upon how the code in @file{crtstuff.c} is called. On systems that 8814support a @dfn{.init} section which is executed at program startup, 8815parts of @file{crtstuff.c} are compiled into that section. The 8816program is linked by the @command{gcc} driver like this: 8817 8818@smallexample 8819ld -o @var{output_file} crti.o crtbegin.o @dots{} -lgcc crtend.o crtn.o 8820@end smallexample 8821 8822The prologue of a function (@code{__init}) appears in the @code{.init} 8823section of @file{crti.o}; the epilogue appears in @file{crtn.o}. Likewise 8824for the function @code{__fini} in the @dfn{.fini} section. Normally these 8825files are provided by the operating system or by the GNU C library, but 8826are provided by GCC for a few targets. 8827 8828The objects @file{crtbegin.o} and @file{crtend.o} are (for most targets) 8829compiled from @file{crtstuff.c}. They contain, among other things, code 8830fragments within the @code{.init} and @code{.fini} sections that branch 8831to routines in the @code{.text} section. The linker will pull all parts 8832of a section together, which results in a complete @code{__init} function 8833that invokes the routines we need at startup. 8834 8835To use this variant, you must define the @code{INIT_SECTION_ASM_OP} 8836macro properly. 8837 8838If no init section is available, when GCC compiles any function called 8839@code{main} (or more accurately, any function designated as a program 8840entry point by the language front end calling @code{expand_main_function}), 8841it inserts a procedure call to @code{__main} as the first executable code 8842after the function prologue. The @code{__main} function is defined 8843in @file{libgcc2.c} and runs the global constructors. 8844 8845In file formats that don't support arbitrary sections, there are again 8846two variants. In the simplest variant, the GNU linker (GNU @code{ld}) 8847and an `a.out' format must be used. In this case, 8848@code{TARGET_ASM_CONSTRUCTOR} is defined to produce a @code{.stabs} 8849entry of type @samp{N_SETT}, referencing the name @code{__CTOR_LIST__}, 8850and with the address of the void function containing the initialization 8851code as its value. The GNU linker recognizes this as a request to add 8852the value to a @dfn{set}; the values are accumulated, and are eventually 8853placed in the executable as a vector in the format described above, with 8854a leading (ignored) count and a trailing zero element. 8855@code{TARGET_ASM_DESTRUCTOR} is handled similarly. Since no init 8856section is available, the absence of @code{INIT_SECTION_ASM_OP} causes 8857the compilation of @code{main} to call @code{__main} as above, starting 8858the initialization process. 8859 8860The last variant uses neither arbitrary sections nor the GNU linker. 8861This is preferable when you want to do dynamic linking and when using 8862file formats which the GNU linker does not support, such as `ECOFF'@. In 8863this case, @code{TARGET_HAVE_CTORS_DTORS} is false, initialization and 8864termination functions are recognized simply by their names. This requires 8865an extra program in the linkage step, called @command{collect2}. This program 8866pretends to be the linker, for use with GCC; it does its job by running 8867the ordinary linker, but also arranges to include the vectors of 8868initialization and termination functions. These functions are called 8869via @code{__main} as described above. In order to use this method, 8870@code{use_collect2} must be defined in the target in @file{config.gcc}. 8871 8872@ifinfo 8873The following section describes the specific macros that control and 8874customize the handling of initialization and termination functions. 8875@end ifinfo 8876 8877@node Macros for Initialization 8878@subsection Macros Controlling Initialization Routines 8879 8880Here are the macros that control how the compiler handles initialization 8881and termination functions: 8882 8883@defmac INIT_SECTION_ASM_OP 8884If defined, a C string constant, including spacing, for the assembler 8885operation to identify the following data as initialization code. If not 8886defined, GCC will assume such a section does not exist. When you are 8887using special sections for initialization and termination functions, this 8888macro also controls how @file{crtstuff.c} and @file{libgcc2.c} arrange to 8889run the initialization functions. 8890@end defmac 8891 8892@defmac HAS_INIT_SECTION 8893If defined, @code{main} will not call @code{__main} as described above. 8894This macro should be defined for systems that control start-up code 8895on a symbol-by-symbol basis, such as OSF/1, and should not 8896be defined explicitly for systems that support @code{INIT_SECTION_ASM_OP}. 8897@end defmac 8898 8899@defmac LD_INIT_SWITCH 8900If defined, a C string constant for a switch that tells the linker that 8901the following symbol is an initialization routine. 8902@end defmac 8903 8904@defmac LD_FINI_SWITCH 8905If defined, a C string constant for a switch that tells the linker that 8906the following symbol is a finalization routine. 8907@end defmac 8908 8909@defmac COLLECT_SHARED_INIT_FUNC (@var{stream}, @var{func}) 8910If defined, a C statement that will write a function that can be 8911automatically called when a shared library is loaded. The function 8912should call @var{func}, which takes no arguments. If not defined, and 8913the object format requires an explicit initialization function, then a 8914function called @code{_GLOBAL__DI} will be generated. 8915 8916This function and the following one are used by collect2 when linking a 8917shared library that needs constructors or destructors, or has DWARF2 8918exception tables embedded in the code. 8919@end defmac 8920 8921@defmac COLLECT_SHARED_FINI_FUNC (@var{stream}, @var{func}) 8922If defined, a C statement that will write a function that can be 8923automatically called when a shared library is unloaded. The function 8924should call @var{func}, which takes no arguments. If not defined, and 8925the object format requires an explicit finalization function, then a 8926function called @code{_GLOBAL__DD} will be generated. 8927@end defmac 8928 8929@defmac INVOKE__main 8930If defined, @code{main} will call @code{__main} despite the presence of 8931@code{INIT_SECTION_ASM_OP}. This macro should be defined for systems 8932where the init section is not actually run automatically, but is still 8933useful for collecting the lists of constructors and destructors. 8934@end defmac 8935 8936@defmac SUPPORTS_INIT_PRIORITY 8937If nonzero, the C++ @code{init_priority} attribute is supported and the 8938compiler should emit instructions to control the order of initialization 8939of objects. If zero, the compiler will issue an error message upon 8940encountering an @code{init_priority} attribute. 8941@end defmac 8942 8943@deftypevr {Target Hook} bool TARGET_HAVE_CTORS_DTORS 8944This value is true if the target supports some ``native'' method of 8945collecting constructors and destructors to be run at startup and exit. 8946It is false if we must use @command{collect2}. 8947@end deftypevr 8948 8949@deftypefn {Target Hook} void TARGET_ASM_CONSTRUCTOR (rtx @var{symbol}, int @var{priority}) 8950If defined, a function that outputs assembler code to arrange to call 8951the function referenced by @var{symbol} at initialization time. 8952 8953Assume that @var{symbol} is a @code{SYMBOL_REF} for a function taking 8954no arguments and with no return value. If the target supports initialization 8955priorities, @var{priority} is a value between 0 and @code{MAX_INIT_PRIORITY}; 8956otherwise it must be @code{DEFAULT_INIT_PRIORITY}. 8957 8958If this macro is not defined by the target, a suitable default will 8959be chosen if (1) the target supports arbitrary section names, (2) the 8960target defines @code{CTORS_SECTION_ASM_OP}, or (3) @code{USE_COLLECT2} 8961is not defined. 8962@end deftypefn 8963 8964@deftypefn {Target Hook} void TARGET_ASM_DESTRUCTOR (rtx @var{symbol}, int @var{priority}) 8965This is like @code{TARGET_ASM_CONSTRUCTOR} but used for termination 8966functions rather than initialization functions. 8967@end deftypefn 8968 8969If @code{TARGET_HAVE_CTORS_DTORS} is true, the initialization routine 8970generated for the generated object file will have static linkage. 8971 8972If your system uses @command{collect2} as the means of processing 8973constructors, then that program normally uses @command{nm} to scan 8974an object file for constructor functions to be called. 8975 8976On certain kinds of systems, you can define this macro to make 8977@command{collect2} work faster (and, in some cases, make it work at all): 8978 8979@defmac OBJECT_FORMAT_COFF 8980Define this macro if the system uses COFF (Common Object File Format) 8981object files, so that @command{collect2} can assume this format and scan 8982object files directly for dynamic constructor/destructor functions. 8983 8984This macro is effective only in a native compiler; @command{collect2} as 8985part of a cross compiler always uses @command{nm} for the target machine. 8986@end defmac 8987 8988@defmac REAL_NM_FILE_NAME 8989Define this macro as a C string constant containing the file name to use 8990to execute @command{nm}. The default is to search the path normally for 8991@command{nm}. 8992@end defmac 8993 8994@defmac NM_FLAGS 8995@command{collect2} calls @command{nm} to scan object files for static 8996constructors and destructors and LTO info. By default, @option{-n} is 8997passed. Define @code{NM_FLAGS} to a C string constant if other options 8998are needed to get the same output format as GNU @command{nm -n} 8999produces. 9000@end defmac 9001 9002If your system supports shared libraries and has a program to list the 9003dynamic dependencies of a given library or executable, you can define 9004these macros to enable support for running initialization and 9005termination functions in shared libraries: 9006 9007@defmac LDD_SUFFIX 9008Define this macro to a C string constant containing the name of the program 9009which lists dynamic dependencies, like @command{ldd} under SunOS 4. 9010@end defmac 9011 9012@defmac PARSE_LDD_OUTPUT (@var{ptr}) 9013Define this macro to be C code that extracts filenames from the output 9014of the program denoted by @code{LDD_SUFFIX}. @var{ptr} is a variable 9015of type @code{char *} that points to the beginning of a line of output 9016from @code{LDD_SUFFIX}. If the line lists a dynamic dependency, the 9017code must advance @var{ptr} to the beginning of the filename on that 9018line. Otherwise, it must set @var{ptr} to @code{NULL}. 9019@end defmac 9020 9021@defmac SHLIB_SUFFIX 9022Define this macro to a C string constant containing the default shared 9023library extension of the target (e.g., @samp{".so"}). @command{collect2} 9024strips version information after this suffix when generating global 9025constructor and destructor names. This define is only needed on targets 9026that use @command{collect2} to process constructors and destructors. 9027@end defmac 9028 9029@node Instruction Output 9030@subsection Output of Assembler Instructions 9031 9032@c prevent bad page break with this line 9033This describes assembler instruction output. 9034 9035@defmac REGISTER_NAMES 9036A C initializer containing the assembler's names for the machine 9037registers, each one as a C string constant. This is what translates 9038register numbers in the compiler into assembler language. 9039@end defmac 9040 9041@defmac ADDITIONAL_REGISTER_NAMES 9042If defined, a C initializer for an array of structures containing a name 9043and a register number. This macro defines additional names for hard 9044registers, thus allowing the @code{asm} option in declarations to refer 9045to registers using alternate names. 9046@end defmac 9047 9048@defmac OVERLAPPING_REGISTER_NAMES 9049If defined, a C initializer for an array of structures containing a 9050name, a register number and a count of the number of consecutive 9051machine registers the name overlaps. This macro defines additional 9052names for hard registers, thus allowing the @code{asm} option in 9053declarations to refer to registers using alternate names. Unlike 9054@code{ADDITIONAL_REGISTER_NAMES}, this macro should be used when the 9055register name implies multiple underlying registers. 9056 9057This macro should be used when it is important that a clobber in an 9058@code{asm} statement clobbers all the underlying values implied by the 9059register name. For example, on ARM, clobbering the double-precision 9060VFP register ``d0'' implies clobbering both single-precision registers 9061``s0'' and ``s1''. 9062@end defmac 9063 9064@defmac ASM_OUTPUT_OPCODE (@var{stream}, @var{ptr}) 9065Define this macro if you are using an unusual assembler that 9066requires different names for the machine instructions. 9067 9068The definition is a C statement or statements which output an 9069assembler instruction opcode to the stdio stream @var{stream}. The 9070macro-operand @var{ptr} is a variable of type @code{char *} which 9071points to the opcode name in its ``internal'' form---the form that is 9072written in the machine description. The definition should output the 9073opcode name to @var{stream}, performing any translation you desire, and 9074increment the variable @var{ptr} to point at the end of the opcode 9075so that it will not be output twice. 9076 9077In fact, your macro definition may process less than the entire opcode 9078name, or more than the opcode name; but if you want to process text 9079that includes @samp{%}-sequences to substitute operands, you must take 9080care of the substitution yourself. Just be sure to increment 9081@var{ptr} over whatever text should not be output normally. 9082 9083@findex recog_data.operand 9084If you need to look at the operand values, they can be found as the 9085elements of @code{recog_data.operand}. 9086 9087If the macro definition does nothing, the instruction is output 9088in the usual way. 9089@end defmac 9090 9091@defmac FINAL_PRESCAN_INSN (@var{insn}, @var{opvec}, @var{noperands}) 9092If defined, a C statement to be executed just prior to the output of 9093assembler code for @var{insn}, to modify the extracted operands so 9094they will be output differently. 9095 9096Here the argument @var{opvec} is the vector containing the operands 9097extracted from @var{insn}, and @var{noperands} is the number of 9098elements of the vector which contain meaningful data for this insn. 9099The contents of this vector are what will be used to convert the insn 9100template into assembler code, so you can change the assembler output 9101by changing the contents of the vector. 9102 9103This macro is useful when various assembler syntaxes share a single 9104file of instruction patterns; by defining this macro differently, you 9105can cause a large class of instructions to be output differently (such 9106as with rearranged operands). Naturally, variations in assembler 9107syntax affecting individual insn patterns ought to be handled by 9108writing conditional output routines in those patterns. 9109 9110If this macro is not defined, it is equivalent to a null statement. 9111@end defmac 9112 9113@deftypefn {Target Hook} void TARGET_ASM_FINAL_POSTSCAN_INSN (FILE *@var{file}, rtx_insn *@var{insn}, rtx *@var{opvec}, int @var{noperands}) 9114If defined, this target hook is a function which is executed just after the 9115output of assembler code for @var{insn}, to change the mode of the assembler 9116if necessary. 9117 9118Here the argument @var{opvec} is the vector containing the operands 9119extracted from @var{insn}, and @var{noperands} is the number of 9120elements of the vector which contain meaningful data for this insn. 9121The contents of this vector are what was used to convert the insn 9122template into assembler code, so you can change the assembler mode 9123by checking the contents of the vector. 9124@end deftypefn 9125 9126@defmac PRINT_OPERAND (@var{stream}, @var{x}, @var{code}) 9127A C compound statement to output to stdio stream @var{stream} the 9128assembler syntax for an instruction operand @var{x}. @var{x} is an 9129RTL expression. 9130 9131@var{code} is a value that can be used to specify one of several ways 9132of printing the operand. It is used when identical operands must be 9133printed differently depending on the context. @var{code} comes from 9134the @samp{%} specification that was used to request printing of the 9135operand. If the specification was just @samp{%@var{digit}} then 9136@var{code} is 0; if the specification was @samp{%@var{ltr} 9137@var{digit}} then @var{code} is the ASCII code for @var{ltr}. 9138 9139@findex reg_names 9140If @var{x} is a register, this macro should print the register's name. 9141The names can be found in an array @code{reg_names} whose type is 9142@code{char *[]}. @code{reg_names} is initialized from 9143@code{REGISTER_NAMES}. 9144 9145When the machine description has a specification @samp{%@var{punct}} 9146(a @samp{%} followed by a punctuation character), this macro is called 9147with a null pointer for @var{x} and the punctuation character for 9148@var{code}. 9149@end defmac 9150 9151@defmac PRINT_OPERAND_PUNCT_VALID_P (@var{code}) 9152A C expression which evaluates to true if @var{code} is a valid 9153punctuation character for use in the @code{PRINT_OPERAND} macro. If 9154@code{PRINT_OPERAND_PUNCT_VALID_P} is not defined, it means that no 9155punctuation characters (except for the standard one, @samp{%}) are used 9156in this way. 9157@end defmac 9158 9159@defmac PRINT_OPERAND_ADDRESS (@var{stream}, @var{x}) 9160A C compound statement to output to stdio stream @var{stream} the 9161assembler syntax for an instruction operand that is a memory reference 9162whose address is @var{x}. @var{x} is an RTL expression. 9163 9164@cindex @code{TARGET_ENCODE_SECTION_INFO} usage 9165On some machines, the syntax for a symbolic address depends on the 9166section that the address refers to. On these machines, define the hook 9167@code{TARGET_ENCODE_SECTION_INFO} to store the information into the 9168@code{symbol_ref}, and then check for it here. @xref{Assembler 9169Format}. 9170@end defmac 9171 9172@findex dbr_sequence_length 9173@defmac DBR_OUTPUT_SEQEND (@var{file}) 9174A C statement, to be executed after all slot-filler instructions have 9175been output. If necessary, call @code{dbr_sequence_length} to 9176determine the number of slots filled in a sequence (zero if not 9177currently outputting a sequence), to decide how many no-ops to output, 9178or whatever. 9179 9180Don't define this macro if it has nothing to do, but it is helpful in 9181reading assembly output if the extent of the delay sequence is made 9182explicit (e.g.@: with white space). 9183@end defmac 9184 9185@findex final_sequence 9186Note that output routines for instructions with delay slots must be 9187prepared to deal with not being output as part of a sequence 9188(i.e.@: when the scheduling pass is not run, or when no slot fillers could be 9189found.) The variable @code{final_sequence} is null when not 9190processing a sequence, otherwise it contains the @code{sequence} rtx 9191being output. 9192 9193@findex asm_fprintf 9194@defmac REGISTER_PREFIX 9195@defmacx LOCAL_LABEL_PREFIX 9196@defmacx USER_LABEL_PREFIX 9197@defmacx IMMEDIATE_PREFIX 9198If defined, C string expressions to be used for the @samp{%R}, @samp{%L}, 9199@samp{%U}, and @samp{%I} options of @code{asm_fprintf} (see 9200@file{final.c}). These are useful when a single @file{md} file must 9201support multiple assembler formats. In that case, the various @file{tm.h} 9202files can define these macros differently. 9203@end defmac 9204 9205@defmac ASM_FPRINTF_EXTENSIONS (@var{file}, @var{argptr}, @var{format}) 9206If defined this macro should expand to a series of @code{case} 9207statements which will be parsed inside the @code{switch} statement of 9208the @code{asm_fprintf} function. This allows targets to define extra 9209printf formats which may useful when generating their assembler 9210statements. Note that uppercase letters are reserved for future 9211generic extensions to asm_fprintf, and so are not available to target 9212specific code. The output file is given by the parameter @var{file}. 9213The varargs input pointer is @var{argptr} and the rest of the format 9214string, starting the character after the one that is being switched 9215upon, is pointed to by @var{format}. 9216@end defmac 9217 9218@defmac ASSEMBLER_DIALECT 9219If your target supports multiple dialects of assembler language (such as 9220different opcodes), define this macro as a C expression that gives the 9221numeric index of the assembler language dialect to use, with zero as the 9222first variant. 9223 9224If this macro is defined, you may use constructs of the form 9225@smallexample 9226@samp{@{option0|option1|option2@dots{}@}} 9227@end smallexample 9228@noindent 9229in the output templates of patterns (@pxref{Output Template}) or in the 9230first argument of @code{asm_fprintf}. This construct outputs 9231@samp{option0}, @samp{option1}, @samp{option2}, etc., if the value of 9232@code{ASSEMBLER_DIALECT} is zero, one, two, etc. Any special characters 9233within these strings retain their usual meaning. If there are fewer 9234alternatives within the braces than the value of 9235@code{ASSEMBLER_DIALECT}, the construct outputs nothing. If it's needed 9236to print curly braces or @samp{|} character in assembler output directly, 9237@samp{%@{}, @samp{%@}} and @samp{%|} can be used. 9238 9239If you do not define this macro, the characters @samp{@{}, @samp{|} and 9240@samp{@}} do not have any special meaning when used in templates or 9241operands to @code{asm_fprintf}. 9242 9243Define the macros @code{REGISTER_PREFIX}, @code{LOCAL_LABEL_PREFIX}, 9244@code{USER_LABEL_PREFIX} and @code{IMMEDIATE_PREFIX} if you can express 9245the variations in assembler language syntax with that mechanism. Define 9246@code{ASSEMBLER_DIALECT} and use the @samp{@{option0|option1@}} syntax 9247if the syntax variant are larger and involve such things as different 9248opcodes or operand order. 9249@end defmac 9250 9251@defmac ASM_OUTPUT_REG_PUSH (@var{stream}, @var{regno}) 9252A C expression to output to @var{stream} some assembler code 9253which will push hard register number @var{regno} onto the stack. 9254The code need not be optimal, since this macro is used only when 9255profiling. 9256@end defmac 9257 9258@defmac ASM_OUTPUT_REG_POP (@var{stream}, @var{regno}) 9259A C expression to output to @var{stream} some assembler code 9260which will pop hard register number @var{regno} off of the stack. 9261The code need not be optimal, since this macro is used only when 9262profiling. 9263@end defmac 9264 9265@node Dispatch Tables 9266@subsection Output of Dispatch Tables 9267 9268@c prevent bad page break with this line 9269This concerns dispatch tables. 9270 9271@cindex dispatch table 9272@defmac ASM_OUTPUT_ADDR_DIFF_ELT (@var{stream}, @var{body}, @var{value}, @var{rel}) 9273A C statement to output to the stdio stream @var{stream} an assembler 9274pseudo-instruction to generate a difference between two labels. 9275@var{value} and @var{rel} are the numbers of two internal labels. The 9276definitions of these labels are output using 9277@code{(*targetm.asm_out.internal_label)}, and they must be printed in the same 9278way here. For example, 9279 9280@smallexample 9281fprintf (@var{stream}, "\t.word L%d-L%d\n", 9282 @var{value}, @var{rel}) 9283@end smallexample 9284 9285You must provide this macro on machines where the addresses in a 9286dispatch table are relative to the table's own address. If defined, GCC 9287will also use this macro on all machines when producing PIC@. 9288@var{body} is the body of the @code{ADDR_DIFF_VEC}; it is provided so that the 9289mode and flags can be read. 9290@end defmac 9291 9292@defmac ASM_OUTPUT_ADDR_VEC_ELT (@var{stream}, @var{value}) 9293This macro should be provided on machines where the addresses 9294in a dispatch table are absolute. 9295 9296The definition should be a C statement to output to the stdio stream 9297@var{stream} an assembler pseudo-instruction to generate a reference to 9298a label. @var{value} is the number of an internal label whose 9299definition is output using @code{(*targetm.asm_out.internal_label)}. 9300For example, 9301 9302@smallexample 9303fprintf (@var{stream}, "\t.word L%d\n", @var{value}) 9304@end smallexample 9305@end defmac 9306 9307@defmac ASM_OUTPUT_CASE_LABEL (@var{stream}, @var{prefix}, @var{num}, @var{table}) 9308Define this if the label before a jump-table needs to be output 9309specially. The first three arguments are the same as for 9310@code{(*targetm.asm_out.internal_label)}; the fourth argument is the 9311jump-table which follows (a @code{jump_table_data} containing an 9312@code{addr_vec} or @code{addr_diff_vec}). 9313 9314This feature is used on system V to output a @code{swbeg} statement 9315for the table. 9316 9317If this macro is not defined, these labels are output with 9318@code{(*targetm.asm_out.internal_label)}. 9319@end defmac 9320 9321@defmac ASM_OUTPUT_CASE_END (@var{stream}, @var{num}, @var{table}) 9322Define this if something special must be output at the end of a 9323jump-table. The definition should be a C statement to be executed 9324after the assembler code for the table is written. It should write 9325the appropriate code to stdio stream @var{stream}. The argument 9326@var{table} is the jump-table insn, and @var{num} is the label-number 9327of the preceding label. 9328 9329If this macro is not defined, nothing special is output at the end of 9330the jump-table. 9331@end defmac 9332 9333@deftypefn {Target Hook} void TARGET_ASM_EMIT_UNWIND_LABEL (FILE *@var{stream}, tree @var{decl}, int @var{for_eh}, int @var{empty}) 9334This target hook emits a label at the beginning of each FDE@. It 9335should be defined on targets where FDEs need special labels, and it 9336should write the appropriate label, for the FDE associated with the 9337function declaration @var{decl}, to the stdio stream @var{stream}. 9338The third argument, @var{for_eh}, is a boolean: true if this is for an 9339exception table. The fourth argument, @var{empty}, is a boolean: 9340true if this is a placeholder label for an omitted FDE@. 9341 9342The default is that FDEs are not given nonlocal labels. 9343@end deftypefn 9344 9345@deftypefn {Target Hook} void TARGET_ASM_EMIT_EXCEPT_TABLE_LABEL (FILE *@var{stream}) 9346This target hook emits a label at the beginning of the exception table. 9347It should be defined on targets where it is desirable for the table 9348to be broken up according to function. 9349 9350The default is that no label is emitted. 9351@end deftypefn 9352 9353@deftypefn {Target Hook} void TARGET_ASM_EMIT_EXCEPT_PERSONALITY (rtx @var{personality}) 9354If the target implements @code{TARGET_ASM_UNWIND_EMIT}, this hook may be used to emit a directive to install a personality hook into the unwind info. This hook should not be used if dwarf2 unwind info is used. 9355@end deftypefn 9356 9357@deftypefn {Target Hook} void TARGET_ASM_UNWIND_EMIT (FILE *@var{stream}, rtx_insn *@var{insn}) 9358This target hook emits assembly directives required to unwind the 9359given instruction. This is only used when @code{TARGET_EXCEPT_UNWIND_INFO} 9360returns @code{UI_TARGET}. 9361@end deftypefn 9362 9363@deftypevr {Target Hook} bool TARGET_ASM_UNWIND_EMIT_BEFORE_INSN 9364True if the @code{TARGET_ASM_UNWIND_EMIT} hook should be called before the assembly for @var{insn} has been emitted, false if the hook should be called afterward. 9365@end deftypevr 9366 9367@node Exception Region Output 9368@subsection Assembler Commands for Exception Regions 9369 9370@c prevent bad page break with this line 9371 9372This describes commands marking the start and the end of an exception 9373region. 9374 9375@defmac EH_FRAME_SECTION_NAME 9376If defined, a C string constant for the name of the section containing 9377exception handling frame unwind information. If not defined, GCC will 9378provide a default definition if the target supports named sections. 9379@file{crtstuff.c} uses this macro to switch to the appropriate section. 9380 9381You should define this symbol if your target supports DWARF 2 frame 9382unwind information and the default definition does not work. 9383@end defmac 9384 9385@defmac EH_FRAME_THROUGH_COLLECT2 9386If defined, DWARF 2 frame unwind information will identified by 9387specially named labels. The collect2 process will locate these 9388labels and generate code to register the frames. 9389 9390This might be necessary, for instance, if the system linker will not 9391place the eh_frames in-between the sentinals from @file{crtstuff.c}, 9392or if the system linker does garbage collection and sections cannot 9393be marked as not to be collected. 9394@end defmac 9395 9396@defmac EH_TABLES_CAN_BE_READ_ONLY 9397Define this macro to 1 if your target is such that no frame unwind 9398information encoding used with non-PIC code will ever require a 9399runtime relocation, but the linker may not support merging read-only 9400and read-write sections into a single read-write section. 9401@end defmac 9402 9403@defmac MASK_RETURN_ADDR 9404An rtx used to mask the return address found via @code{RETURN_ADDR_RTX}, so 9405that it does not contain any extraneous set bits in it. 9406@end defmac 9407 9408@defmac DWARF2_UNWIND_INFO 9409Define this macro to 0 if your target supports DWARF 2 frame unwind 9410information, but it does not yet work with exception handling. 9411Otherwise, if your target supports this information (if it defines 9412@code{INCOMING_RETURN_ADDR_RTX} and @code{OBJECT_FORMAT_ELF}), 9413GCC will provide a default definition of 1. 9414@end defmac 9415 9416@deftypefn {Common Target Hook} {enum unwind_info_type} TARGET_EXCEPT_UNWIND_INFO (struct gcc_options *@var{opts}) 9417This hook defines the mechanism that will be used for exception handling 9418by the target. If the target has ABI specified unwind tables, the hook 9419should return @code{UI_TARGET}. If the target is to use the 9420@code{setjmp}/@code{longjmp}-based exception handling scheme, the hook 9421should return @code{UI_SJLJ}. If the target supports DWARF 2 frame unwind 9422information, the hook should return @code{UI_DWARF2}. 9423 9424A target may, if exceptions are disabled, choose to return @code{UI_NONE}. 9425This may end up simplifying other parts of target-specific code. The 9426default implementation of this hook never returns @code{UI_NONE}. 9427 9428Note that the value returned by this hook should be constant. It should 9429not depend on anything except the command-line switches described by 9430@var{opts}. In particular, the 9431setting @code{UI_SJLJ} must be fixed at compiler start-up as C pre-processor 9432macros and builtin functions related to exception handling are set up 9433depending on this setting. 9434 9435The default implementation of the hook first honors the 9436@option{--enable-sjlj-exceptions} configure option, then 9437@code{DWARF2_UNWIND_INFO}, and finally defaults to @code{UI_SJLJ}. If 9438@code{DWARF2_UNWIND_INFO} depends on command-line options, the target 9439must define this hook so that @var{opts} is used correctly. 9440@end deftypefn 9441 9442@deftypevr {Common Target Hook} bool TARGET_UNWIND_TABLES_DEFAULT 9443This variable should be set to @code{true} if the target ABI requires unwinding 9444tables even when exceptions are not used. It must not be modified by 9445command-line option processing. 9446@end deftypevr 9447 9448@defmac DONT_USE_BUILTIN_SETJMP 9449Define this macro to 1 if the @code{setjmp}/@code{longjmp}-based scheme 9450should use the @code{setjmp}/@code{longjmp} functions from the C library 9451instead of the @code{__builtin_setjmp}/@code{__builtin_longjmp} machinery. 9452@end defmac 9453 9454@defmac JMP_BUF_SIZE 9455This macro has no effect unless @code{DONT_USE_BUILTIN_SETJMP} is also 9456defined. Define this macro if the default size of @code{jmp_buf} buffer 9457for the @code{setjmp}/@code{longjmp}-based exception handling mechanism 9458is not large enough, or if it is much too large. 9459The default size is @code{FIRST_PSEUDO_REGISTER * sizeof(void *)}. 9460@end defmac 9461 9462@defmac DWARF_CIE_DATA_ALIGNMENT 9463This macro need only be defined if the target might save registers in the 9464function prologue at an offset to the stack pointer that is not aligned to 9465@code{UNITS_PER_WORD}. The definition should be the negative minimum 9466alignment if @code{STACK_GROWS_DOWNWARD} is true, and the positive 9467minimum alignment otherwise. @xref{DWARF}. Only applicable if 9468the target supports DWARF 2 frame unwind information. 9469@end defmac 9470 9471@deftypevr {Target Hook} bool TARGET_TERMINATE_DW2_EH_FRAME_INFO 9472Contains the value true if the target should add a zero word onto the 9473end of a Dwarf-2 frame info section when used for exception handling. 9474Default value is false if @code{EH_FRAME_SECTION_NAME} is defined, and 9475true otherwise. 9476@end deftypevr 9477 9478@deftypefn {Target Hook} rtx TARGET_DWARF_REGISTER_SPAN (rtx @var{reg}) 9479Given a register, this hook should return a parallel of registers to 9480represent where to find the register pieces. Define this hook if the 9481register and its mode are represented in Dwarf in non-contiguous 9482locations, or if the register should be represented in more than one 9483register in Dwarf. Otherwise, this hook should return @code{NULL_RTX}. 9484If not defined, the default is to return @code{NULL_RTX}. 9485@end deftypefn 9486 9487@deftypefn {Target Hook} machine_mode TARGET_DWARF_FRAME_REG_MODE (int @var{regno}) 9488Given a register, this hook should return the mode which the 9489corresponding Dwarf frame register should have. This is normally 9490used to return a smaller mode than the raw mode to prevent call 9491clobbered parts of a register altering the frame register size 9492@end deftypefn 9493 9494@deftypefn {Target Hook} void TARGET_INIT_DWARF_REG_SIZES_EXTRA (tree @var{address}) 9495If some registers are represented in Dwarf-2 unwind information in 9496multiple pieces, define this hook to fill in information about the 9497sizes of those pieces in the table used by the unwinder at runtime. 9498It will be called by @code{expand_builtin_init_dwarf_reg_sizes} after 9499filling in a single size corresponding to each hard register; 9500@var{address} is the address of the table. 9501@end deftypefn 9502 9503@deftypefn {Target Hook} bool TARGET_ASM_TTYPE (rtx @var{sym}) 9504This hook is used to output a reference from a frame unwinding table to 9505the type_info object identified by @var{sym}. It should return @code{true} 9506if the reference was output. Returning @code{false} will cause the 9507reference to be output using the normal Dwarf2 routines. 9508@end deftypefn 9509 9510@deftypevr {Target Hook} bool TARGET_ARM_EABI_UNWINDER 9511This flag should be set to @code{true} on targets that use an ARM EABI 9512based unwinding library, and @code{false} on other targets. This effects 9513the format of unwinding tables, and how the unwinder in entered after 9514running a cleanup. The default is @code{false}. 9515@end deftypevr 9516 9517@node Alignment Output 9518@subsection Assembler Commands for Alignment 9519 9520@c prevent bad page break with this line 9521This describes commands for alignment. 9522 9523@defmac JUMP_ALIGN (@var{label}) 9524The alignment (log base 2) to put in front of @var{label}, which is 9525a common destination of jumps and has no fallthru incoming edge. 9526 9527This macro need not be defined if you don't want any special alignment 9528to be done at such a time. Most machine descriptions do not currently 9529define the macro. 9530 9531Unless it's necessary to inspect the @var{label} parameter, it is better 9532to set the variable @var{align_jumps} in the target's 9533@code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's 9534selection in @var{align_jumps} in a @code{JUMP_ALIGN} implementation. 9535@end defmac 9536 9537@deftypefn {Target Hook} int TARGET_ASM_JUMP_ALIGN_MAX_SKIP (rtx_insn *@var{label}) 9538The maximum number of bytes to skip before @var{label} when applying 9539@code{JUMP_ALIGN}. This works only if 9540@code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined. 9541@end deftypefn 9542 9543@defmac LABEL_ALIGN_AFTER_BARRIER (@var{label}) 9544The alignment (log base 2) to put in front of @var{label}, which follows 9545a @code{BARRIER}. 9546 9547This macro need not be defined if you don't want any special alignment 9548to be done at such a time. Most machine descriptions do not currently 9549define the macro. 9550@end defmac 9551 9552@deftypefn {Target Hook} int TARGET_ASM_LABEL_ALIGN_AFTER_BARRIER_MAX_SKIP (rtx_insn *@var{label}) 9553The maximum number of bytes to skip before @var{label} when applying 9554@code{LABEL_ALIGN_AFTER_BARRIER}. This works only if 9555@code{ASM_OUTPUT_MAX_SKIP_ALIGN} is defined. 9556@end deftypefn 9557 9558@defmac LOOP_ALIGN (@var{label}) 9559The alignment (log base 2) to put in front of @var{label} that heads 9560a frequently executed basic block (usually the header of a loop). 9561 9562This macro need not be defined if you don't want any special alignment 9563to be done at such a time. Most machine descriptions do not currently 9564define the macro. 9565 9566Unless it's necessary to inspect the @var{label} parameter, it is better 9567to set the variable @code{align_loops} in the target's 9568@code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's 9569selection in @code{align_loops} in a @code{LOOP_ALIGN} implementation. 9570@end defmac 9571 9572@deftypefn {Target Hook} int TARGET_ASM_LOOP_ALIGN_MAX_SKIP (rtx_insn *@var{label}) 9573The maximum number of bytes to skip when applying @code{LOOP_ALIGN} to 9574@var{label}. This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN} is 9575defined. 9576@end deftypefn 9577 9578@defmac LABEL_ALIGN (@var{label}) 9579The alignment (log base 2) to put in front of @var{label}. 9580If @code{LABEL_ALIGN_AFTER_BARRIER} / @code{LOOP_ALIGN} specify a different alignment, 9581the maximum of the specified values is used. 9582 9583Unless it's necessary to inspect the @var{label} parameter, it is better 9584to set the variable @code{align_labels} in the target's 9585@code{TARGET_OPTION_OVERRIDE}. Otherwise, you should try to honor the user's 9586selection in @code{align_labels} in a @code{LABEL_ALIGN} implementation. 9587@end defmac 9588 9589@deftypefn {Target Hook} int TARGET_ASM_LABEL_ALIGN_MAX_SKIP (rtx_insn *@var{label}) 9590The maximum number of bytes to skip when applying @code{LABEL_ALIGN} 9591to @var{label}. This works only if @code{ASM_OUTPUT_MAX_SKIP_ALIGN} 9592is defined. 9593@end deftypefn 9594 9595@defmac ASM_OUTPUT_SKIP (@var{stream}, @var{nbytes}) 9596A C statement to output to the stdio stream @var{stream} an assembler 9597instruction to advance the location counter by @var{nbytes} bytes. 9598Those bytes should be zero when loaded. @var{nbytes} will be a C 9599expression of type @code{unsigned HOST_WIDE_INT}. 9600@end defmac 9601 9602@defmac ASM_NO_SKIP_IN_TEXT 9603Define this macro if @code{ASM_OUTPUT_SKIP} should not be used in the 9604text section because it fails to put zeros in the bytes that are skipped. 9605This is true on many Unix systems, where the pseudo--op to skip bytes 9606produces no-op instructions rather than zeros when used in the text 9607section. 9608@end defmac 9609 9610@defmac ASM_OUTPUT_ALIGN (@var{stream}, @var{power}) 9611A C statement to output to the stdio stream @var{stream} an assembler 9612command to advance the location counter to a multiple of 2 to the 9613@var{power} bytes. @var{power} will be a C expression of type @code{int}. 9614@end defmac 9615 9616@defmac ASM_OUTPUT_ALIGN_WITH_NOP (@var{stream}, @var{power}) 9617Like @code{ASM_OUTPUT_ALIGN}, except that the ``nop'' instruction is used 9618for padding, if necessary. 9619@end defmac 9620 9621@defmac ASM_OUTPUT_MAX_SKIP_ALIGN (@var{stream}, @var{power}, @var{max_skip}) 9622A C statement to output to the stdio stream @var{stream} an assembler 9623command to advance the location counter to a multiple of 2 to the 9624@var{power} bytes, but only if @var{max_skip} or fewer bytes are needed to 9625satisfy the alignment request. @var{power} and @var{max_skip} will be 9626a C expression of type @code{int}. 9627@end defmac 9628 9629@need 3000 9630@node Debugging Info 9631@section Controlling Debugging Information Format 9632 9633@c prevent bad page break with this line 9634This describes how to specify debugging information. 9635 9636@menu 9637* All Debuggers:: Macros that affect all debugging formats uniformly. 9638* DBX Options:: Macros enabling specific options in DBX format. 9639* DBX Hooks:: Hook macros for varying DBX format. 9640* File Names and DBX:: Macros controlling output of file names in DBX format. 9641* DWARF:: Macros for DWARF format. 9642* VMS Debug:: Macros for VMS debug format. 9643@end menu 9644 9645@node All Debuggers 9646@subsection Macros Affecting All Debugging Formats 9647 9648@c prevent bad page break with this line 9649These macros affect all debugging formats. 9650 9651@defmac DBX_REGISTER_NUMBER (@var{regno}) 9652A C expression that returns the DBX register number for the compiler 9653register number @var{regno}. In the default macro provided, the value 9654of this expression will be @var{regno} itself. But sometimes there are 9655some registers that the compiler knows about and DBX does not, or vice 9656versa. In such cases, some register may need to have one number in the 9657compiler and another for DBX@. 9658 9659If two registers have consecutive numbers inside GCC, and they can be 9660used as a pair to hold a multiword value, then they @emph{must} have 9661consecutive numbers after renumbering with @code{DBX_REGISTER_NUMBER}. 9662Otherwise, debuggers will be unable to access such a pair, because they 9663expect register pairs to be consecutive in their own numbering scheme. 9664 9665If you find yourself defining @code{DBX_REGISTER_NUMBER} in way that 9666does not preserve register pairs, then what you must do instead is 9667redefine the actual register numbering scheme. 9668@end defmac 9669 9670@defmac DEBUGGER_AUTO_OFFSET (@var{x}) 9671A C expression that returns the integer offset value for an automatic 9672variable having address @var{x} (an RTL expression). The default 9673computation assumes that @var{x} is based on the frame-pointer and 9674gives the offset from the frame-pointer. This is required for targets 9675that produce debugging output for DBX and allow the frame-pointer to be 9676eliminated when the @option{-g} option is used. 9677@end defmac 9678 9679@defmac DEBUGGER_ARG_OFFSET (@var{offset}, @var{x}) 9680A C expression that returns the integer offset value for an argument 9681having address @var{x} (an RTL expression). The nominal offset is 9682@var{offset}. 9683@end defmac 9684 9685@defmac PREFERRED_DEBUGGING_TYPE 9686A C expression that returns the type of debugging output GCC should 9687produce when the user specifies just @option{-g}. Define 9688this if you have arranged for GCC to support more than one format of 9689debugging output. Currently, the allowable values are @code{DBX_DEBUG}, 9690@code{DWARF2_DEBUG}, @code{XCOFF_DEBUG}, @code{VMS_DEBUG}, 9691and @code{VMS_AND_DWARF2_DEBUG}. 9692 9693When the user specifies @option{-ggdb}, GCC normally also uses the 9694value of this macro to select the debugging output format, but with two 9695exceptions. If @code{DWARF2_DEBUGGING_INFO} is defined, GCC uses the 9696value @code{DWARF2_DEBUG}. Otherwise, if @code{DBX_DEBUGGING_INFO} is 9697defined, GCC uses @code{DBX_DEBUG}. 9698 9699The value of this macro only affects the default debugging output; the 9700user can always get a specific type of output by using @option{-gstabs}, 9701@option{-gdwarf-2}, @option{-gxcoff}, or @option{-gvms}. 9702@end defmac 9703 9704@node DBX Options 9705@subsection Specific Options for DBX Output 9706 9707@c prevent bad page break with this line 9708These are specific options for DBX output. 9709 9710@defmac DBX_DEBUGGING_INFO 9711Define this macro if GCC should produce debugging output for DBX 9712in response to the @option{-g} option. 9713@end defmac 9714 9715@defmac XCOFF_DEBUGGING_INFO 9716Define this macro if GCC should produce XCOFF format debugging output 9717in response to the @option{-g} option. This is a variant of DBX format. 9718@end defmac 9719 9720@defmac DEFAULT_GDB_EXTENSIONS 9721Define this macro to control whether GCC should by default generate 9722GDB's extended version of DBX debugging information (assuming DBX-format 9723debugging information is enabled at all). If you don't define the 9724macro, the default is 1: always generate the extended information 9725if there is any occasion to. 9726@end defmac 9727 9728@defmac DEBUG_SYMS_TEXT 9729Define this macro if all @code{.stabs} commands should be output while 9730in the text section. 9731@end defmac 9732 9733@defmac ASM_STABS_OP 9734A C string constant, including spacing, naming the assembler pseudo op to 9735use instead of @code{"\t.stabs\t"} to define an ordinary debugging symbol. 9736If you don't define this macro, @code{"\t.stabs\t"} is used. This macro 9737applies only to DBX debugging information format. 9738@end defmac 9739 9740@defmac ASM_STABD_OP 9741A C string constant, including spacing, naming the assembler pseudo op to 9742use instead of @code{"\t.stabd\t"} to define a debugging symbol whose 9743value is the current location. If you don't define this macro, 9744@code{"\t.stabd\t"} is used. This macro applies only to DBX debugging 9745information format. 9746@end defmac 9747 9748@defmac ASM_STABN_OP 9749A C string constant, including spacing, naming the assembler pseudo op to 9750use instead of @code{"\t.stabn\t"} to define a debugging symbol with no 9751name. If you don't define this macro, @code{"\t.stabn\t"} is used. This 9752macro applies only to DBX debugging information format. 9753@end defmac 9754 9755@defmac DBX_NO_XREFS 9756Define this macro if DBX on your system does not support the construct 9757@samp{xs@var{tagname}}. On some systems, this construct is used to 9758describe a forward reference to a structure named @var{tagname}. 9759On other systems, this construct is not supported at all. 9760@end defmac 9761 9762@defmac DBX_CONTIN_LENGTH 9763A symbol name in DBX-format debugging information is normally 9764continued (split into two separate @code{.stabs} directives) when it 9765exceeds a certain length (by default, 80 characters). On some 9766operating systems, DBX requires this splitting; on others, splitting 9767must not be done. You can inhibit splitting by defining this macro 9768with the value zero. You can override the default splitting-length by 9769defining this macro as an expression for the length you desire. 9770@end defmac 9771 9772@defmac DBX_CONTIN_CHAR 9773Normally continuation is indicated by adding a @samp{\} character to 9774the end of a @code{.stabs} string when a continuation follows. To use 9775a different character instead, define this macro as a character 9776constant for the character you want to use. Do not define this macro 9777if backslash is correct for your system. 9778@end defmac 9779 9780@defmac DBX_STATIC_STAB_DATA_SECTION 9781Define this macro if it is necessary to go to the data section before 9782outputting the @samp{.stabs} pseudo-op for a non-global static 9783variable. 9784@end defmac 9785 9786@defmac DBX_TYPE_DECL_STABS_CODE 9787The value to use in the ``code'' field of the @code{.stabs} directive 9788for a typedef. The default is @code{N_LSYM}. 9789@end defmac 9790 9791@defmac DBX_STATIC_CONST_VAR_CODE 9792The value to use in the ``code'' field of the @code{.stabs} directive 9793for a static variable located in the text section. DBX format does not 9794provide any ``right'' way to do this. The default is @code{N_FUN}. 9795@end defmac 9796 9797@defmac DBX_REGPARM_STABS_CODE 9798The value to use in the ``code'' field of the @code{.stabs} directive 9799for a parameter passed in registers. DBX format does not provide any 9800``right'' way to do this. The default is @code{N_RSYM}. 9801@end defmac 9802 9803@defmac DBX_REGPARM_STABS_LETTER 9804The letter to use in DBX symbol data to identify a symbol as a parameter 9805passed in registers. DBX format does not customarily provide any way to 9806do this. The default is @code{'P'}. 9807@end defmac 9808 9809@defmac DBX_FUNCTION_FIRST 9810Define this macro if the DBX information for a function and its 9811arguments should precede the assembler code for the function. Normally, 9812in DBX format, the debugging information entirely follows the assembler 9813code. 9814@end defmac 9815 9816@defmac DBX_BLOCKS_FUNCTION_RELATIVE 9817Define this macro, with value 1, if the value of a symbol describing 9818the scope of a block (@code{N_LBRAC} or @code{N_RBRAC}) should be 9819relative to the start of the enclosing function. Normally, GCC uses 9820an absolute address. 9821@end defmac 9822 9823@defmac DBX_LINES_FUNCTION_RELATIVE 9824Define this macro, with value 1, if the value of a symbol indicating 9825the current line number (@code{N_SLINE}) should be relative to the 9826start of the enclosing function. Normally, GCC uses an absolute address. 9827@end defmac 9828 9829@defmac DBX_USE_BINCL 9830Define this macro if GCC should generate @code{N_BINCL} and 9831@code{N_EINCL} stabs for included header files, as on Sun systems. This 9832macro also directs GCC to output a type number as a pair of a file 9833number and a type number within the file. Normally, GCC does not 9834generate @code{N_BINCL} or @code{N_EINCL} stabs, and it outputs a single 9835number for a type number. 9836@end defmac 9837 9838@node DBX Hooks 9839@subsection Open-Ended Hooks for DBX Format 9840 9841@c prevent bad page break with this line 9842These are hooks for DBX format. 9843 9844@defmac DBX_OUTPUT_SOURCE_LINE (@var{stream}, @var{line}, @var{counter}) 9845A C statement to output DBX debugging information before code for line 9846number @var{line} of the current source file to the stdio stream 9847@var{stream}. @var{counter} is the number of time the macro was 9848invoked, including the current invocation; it is intended to generate 9849unique labels in the assembly output. 9850 9851This macro should not be defined if the default output is correct, or 9852if it can be made correct by defining @code{DBX_LINES_FUNCTION_RELATIVE}. 9853@end defmac 9854 9855@defmac NO_DBX_FUNCTION_END 9856Some stabs encapsulation formats (in particular ECOFF), cannot handle the 9857@code{.stabs "",N_FUN,,0,0,Lscope-function-1} gdb dbx extension construct. 9858On those machines, define this macro to turn this feature off without 9859disturbing the rest of the gdb extensions. 9860@end defmac 9861 9862@defmac NO_DBX_BNSYM_ENSYM 9863Some assemblers cannot handle the @code{.stabd BNSYM/ENSYM,0,0} gdb dbx 9864extension construct. On those machines, define this macro to turn this 9865feature off without disturbing the rest of the gdb extensions. 9866@end defmac 9867 9868@node File Names and DBX 9869@subsection File Names in DBX Format 9870 9871@c prevent bad page break with this line 9872This describes file names in DBX format. 9873 9874@defmac DBX_OUTPUT_MAIN_SOURCE_FILENAME (@var{stream}, @var{name}) 9875A C statement to output DBX debugging information to the stdio stream 9876@var{stream}, which indicates that file @var{name} is the main source 9877file---the file specified as the input file for compilation. 9878This macro is called only once, at the beginning of compilation. 9879 9880This macro need not be defined if the standard form of output 9881for DBX debugging information is appropriate. 9882 9883It may be necessary to refer to a label equal to the beginning of the 9884text section. You can use @samp{assemble_name (stream, ltext_label_name)} 9885to do so. If you do this, you must also set the variable 9886@var{used_ltext_label_name} to @code{true}. 9887@end defmac 9888 9889@defmac NO_DBX_MAIN_SOURCE_DIRECTORY 9890Define this macro, with value 1, if GCC should not emit an indication 9891of the current directory for compilation and current source language at 9892the beginning of the file. 9893@end defmac 9894 9895@defmac NO_DBX_GCC_MARKER 9896Define this macro, with value 1, if GCC should not emit an indication 9897that this object file was compiled by GCC@. The default is to emit 9898an @code{N_OPT} stab at the beginning of every source file, with 9899@samp{gcc2_compiled.} for the string and value 0. 9900@end defmac 9901 9902@defmac DBX_OUTPUT_MAIN_SOURCE_FILE_END (@var{stream}, @var{name}) 9903A C statement to output DBX debugging information at the end of 9904compilation of the main source file @var{name}. Output should be 9905written to the stdio stream @var{stream}. 9906 9907If you don't define this macro, nothing special is output at the end 9908of compilation, which is correct for most machines. 9909@end defmac 9910 9911@defmac DBX_OUTPUT_NULL_N_SO_AT_MAIN_SOURCE_FILE_END 9912Define this macro @emph{instead of} defining 9913@code{DBX_OUTPUT_MAIN_SOURCE_FILE_END}, if what needs to be output at 9914the end of compilation is an @code{N_SO} stab with an empty string, 9915whose value is the highest absolute text address in the file. 9916@end defmac 9917 9918@need 2000 9919@node DWARF 9920@subsection Macros for DWARF Output 9921 9922@c prevent bad page break with this line 9923Here are macros for DWARF output. 9924 9925@defmac DWARF2_DEBUGGING_INFO 9926Define this macro if GCC should produce dwarf version 2 format 9927debugging output in response to the @option{-g} option. 9928 9929@deftypefn {Target Hook} int TARGET_DWARF_CALLING_CONVENTION (const_tree @var{function}) 9930Define this to enable the dwarf attribute @code{DW_AT_calling_convention} to 9931be emitted for each function. Instead of an integer return the enum 9932value for the @code{DW_CC_} tag. 9933@end deftypefn 9934 9935To support optional call frame debugging information, you must also 9936define @code{INCOMING_RETURN_ADDR_RTX} and either set 9937@code{RTX_FRAME_RELATED_P} on the prologue insns if you use RTL for the 9938prologue, or call @code{dwarf2out_def_cfa} and @code{dwarf2out_reg_save} 9939as appropriate from @code{TARGET_ASM_FUNCTION_PROLOGUE} if you don't. 9940@end defmac 9941 9942@defmac DWARF2_FRAME_INFO 9943Define this macro to a nonzero value if GCC should always output 9944Dwarf 2 frame information. If @code{TARGET_EXCEPT_UNWIND_INFO} 9945(@pxref{Exception Region Output}) returns @code{UI_DWARF2}, and 9946exceptions are enabled, GCC will output this information not matter 9947how you define @code{DWARF2_FRAME_INFO}. 9948@end defmac 9949 9950@deftypefn {Target Hook} {enum unwind_info_type} TARGET_DEBUG_UNWIND_INFO (void) 9951This hook defines the mechanism that will be used for describing frame 9952unwind information to the debugger. Normally the hook will return 9953@code{UI_DWARF2} if DWARF 2 debug information is enabled, and 9954return @code{UI_NONE} otherwise. 9955 9956A target may return @code{UI_DWARF2} even when DWARF 2 debug information 9957is disabled in order to always output DWARF 2 frame information. 9958 9959A target may return @code{UI_TARGET} if it has ABI specified unwind tables. 9960This will suppress generation of the normal debug frame unwind information. 9961@end deftypefn 9962 9963@defmac DWARF2_ASM_LINE_DEBUG_INFO 9964Define this macro to be a nonzero value if the assembler can generate Dwarf 2 9965line debug info sections. This will result in much more compact line number 9966tables, and hence is desirable if it works. 9967@end defmac 9968 9969@defmac DWARF2_ASM_VIEW_DEBUG_INFO 9970Define this macro to be a nonzero value if the assembler supports view 9971assignment and verification in @code{.loc}. If it does not, but the 9972user enables location views, the compiler may have to fallback to 9973internal line number tables. 9974@end defmac 9975 9976@deftypefn {Target Hook} int TARGET_RESET_LOCATION_VIEW (rtx_insn *@var{}) 9977This hook, if defined, enables -ginternal-reset-location-views, and 9978uses its result to override cases in which the estimated min insn 9979length might be nonzero even when a PC advance (i.e., a view reset) 9980cannot be taken for granted. 9981 9982If the hook is defined, it must return a positive value to indicate 9983the insn definitely advances the PC, and so the view number can be 9984safely assumed to be reset; a negative value to mean the insn 9985definitely does not advance the PC, and os the view number must not 9986be reset; or zero to decide based on the estimated insn length. 9987 9988If insn length is to be regarded as reliable, set the hook to 9989@code{hook_int_rtx_insn_0}. 9990@end deftypefn 9991 9992@deftypevr {Target Hook} bool TARGET_WANT_DEBUG_PUB_SECTIONS 9993True if the @code{.debug_pubtypes} and @code{.debug_pubnames} sections should be emitted. These sections are not used on most platforms, and in particular GDB does not use them. 9994@end deftypevr 9995 9996@deftypevr {Target Hook} bool TARGET_DELAY_SCHED2 9997True if sched2 is not to be run at its normal place. 9998This usually means it will be run as part of machine-specific reorg. 9999@end deftypevr 10000 10001@deftypevr {Target Hook} bool TARGET_DELAY_VARTRACK 10002True if vartrack is not to be run at its normal place. 10003This usually means it will be run as part of machine-specific reorg. 10004@end deftypevr 10005 10006@deftypevr {Target Hook} bool TARGET_NO_REGISTER_ALLOCATION 10007True if register allocation and the passes 10008following it should not be run. Usually true only for virtual assembler 10009targets. 10010@end deftypevr 10011 10012@defmac ASM_OUTPUT_DWARF_DELTA (@var{stream}, @var{size}, @var{label1}, @var{label2}) 10013A C statement to issue assembly directives that create a difference 10014@var{lab1} minus @var{lab2}, using an integer of the given @var{size}. 10015@end defmac 10016 10017@defmac ASM_OUTPUT_DWARF_VMS_DELTA (@var{stream}, @var{size}, @var{label1}, @var{label2}) 10018A C statement to issue assembly directives that create a difference 10019between the two given labels in system defined units, e.g. instruction 10020slots on IA64 VMS, using an integer of the given size. 10021@end defmac 10022 10023@defmac ASM_OUTPUT_DWARF_OFFSET (@var{stream}, @var{size}, @var{label}, @var{offset}, @var{section}) 10024A C statement to issue assembly directives that create a 10025section-relative reference to the given @var{label} plus @var{offset}, using 10026an integer of the given @var{size}. The label is known to be defined in the 10027given @var{section}. 10028@end defmac 10029 10030@defmac ASM_OUTPUT_DWARF_PCREL (@var{stream}, @var{size}, @var{label}) 10031A C statement to issue assembly directives that create a self-relative 10032reference to the given @var{label}, using an integer of the given @var{size}. 10033@end defmac 10034 10035@defmac ASM_OUTPUT_DWARF_DATAREL (@var{stream}, @var{size}, @var{label}) 10036A C statement to issue assembly directives that create a reference to the 10037given @var{label} relative to the dbase, using an integer of the given @var{size}. 10038@end defmac 10039 10040@defmac ASM_OUTPUT_DWARF_TABLE_REF (@var{label}) 10041A C statement to issue assembly directives that create a reference to 10042the DWARF table identifier @var{label} from the current section. This 10043is used on some systems to avoid garbage collecting a DWARF table which 10044is referenced by a function. 10045@end defmac 10046 10047@deftypefn {Target Hook} void TARGET_ASM_OUTPUT_DWARF_DTPREL (FILE *@var{file}, int @var{size}, rtx @var{x}) 10048If defined, this target hook is a function which outputs a DTP-relative 10049reference to the given TLS symbol of the specified size. 10050@end deftypefn 10051 10052@need 2000 10053@node VMS Debug 10054@subsection Macros for VMS Debug Format 10055 10056@c prevent bad page break with this line 10057Here are macros for VMS debug format. 10058 10059@defmac VMS_DEBUGGING_INFO 10060Define this macro if GCC should produce debugging output for VMS 10061in response to the @option{-g} option. The default behavior for VMS 10062is to generate minimal debug info for a traceback in the absence of 10063@option{-g} unless explicitly overridden with @option{-g0}. This 10064behavior is controlled by @code{TARGET_OPTION_OPTIMIZATION} and 10065@code{TARGET_OPTION_OVERRIDE}. 10066@end defmac 10067 10068@node Floating Point 10069@section Cross Compilation and Floating Point 10070@cindex cross compilation and floating point 10071@cindex floating point and cross compilation 10072 10073While all modern machines use twos-complement representation for integers, 10074there are a variety of representations for floating point numbers. This 10075means that in a cross-compiler the representation of floating point numbers 10076in the compiled program may be different from that used in the machine 10077doing the compilation. 10078 10079Because different representation systems may offer different amounts of 10080range and precision, all floating point constants must be represented in 10081the target machine's format. Therefore, the cross compiler cannot 10082safely use the host machine's floating point arithmetic; it must emulate 10083the target's arithmetic. To ensure consistency, GCC always uses 10084emulation to work with floating point values, even when the host and 10085target floating point formats are identical. 10086 10087The following macros are provided by @file{real.h} for the compiler to 10088use. All parts of the compiler which generate or optimize 10089floating-point calculations must use these macros. They may evaluate 10090their operands more than once, so operands must not have side effects. 10091 10092@defmac REAL_VALUE_TYPE 10093The C data type to be used to hold a floating point value in the target 10094machine's format. Typically this is a @code{struct} containing an 10095array of @code{HOST_WIDE_INT}, but all code should treat it as an opaque 10096quantity. 10097@end defmac 10098 10099@deftypefn Macro HOST_WIDE_INT REAL_VALUE_FIX (REAL_VALUE_TYPE @var{x}) 10100Truncates @var{x} to a signed integer, rounding toward zero. 10101@end deftypefn 10102 10103@deftypefn Macro {unsigned HOST_WIDE_INT} REAL_VALUE_UNSIGNED_FIX (REAL_VALUE_TYPE @var{x}) 10104Truncates @var{x} to an unsigned integer, rounding toward zero. If 10105@var{x} is negative, returns zero. 10106@end deftypefn 10107 10108@deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ATOF (const char *@var{string}, machine_mode @var{mode}) 10109Converts @var{string} into a floating point number in the target machine's 10110representation for mode @var{mode}. This routine can handle both 10111decimal and hexadecimal floating point constants, using the syntax 10112defined by the C language for both. 10113@end deftypefn 10114 10115@deftypefn Macro int REAL_VALUE_NEGATIVE (REAL_VALUE_TYPE @var{x}) 10116Returns 1 if @var{x} is negative (including negative zero), 0 otherwise. 10117@end deftypefn 10118 10119@deftypefn Macro int REAL_VALUE_ISINF (REAL_VALUE_TYPE @var{x}) 10120Determines whether @var{x} represents infinity (positive or negative). 10121@end deftypefn 10122 10123@deftypefn Macro int REAL_VALUE_ISNAN (REAL_VALUE_TYPE @var{x}) 10124Determines whether @var{x} represents a ``NaN'' (not-a-number). 10125@end deftypefn 10126 10127@deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_NEGATE (REAL_VALUE_TYPE @var{x}) 10128Returns the negative of the floating point value @var{x}. 10129@end deftypefn 10130 10131@deftypefn Macro REAL_VALUE_TYPE REAL_VALUE_ABS (REAL_VALUE_TYPE @var{x}) 10132Returns the absolute value of @var{x}. 10133@end deftypefn 10134 10135@node Mode Switching 10136@section Mode Switching Instructions 10137@cindex mode switching 10138The following macros control mode switching optimizations: 10139 10140@defmac OPTIMIZE_MODE_SWITCHING (@var{entity}) 10141Define this macro if the port needs extra instructions inserted for mode 10142switching in an optimizing compilation. 10143 10144For an example, the SH4 can perform both single and double precision 10145floating point operations, but to perform a single precision operation, 10146the FPSCR PR bit has to be cleared, while for a double precision 10147operation, this bit has to be set. Changing the PR bit requires a general 10148purpose register as a scratch register, hence these FPSCR sets have to 10149be inserted before reload, i.e.@: you cannot put this into instruction emitting 10150or @code{TARGET_MACHINE_DEPENDENT_REORG}. 10151 10152You can have multiple entities that are mode-switched, and select at run time 10153which entities actually need it. @code{OPTIMIZE_MODE_SWITCHING} should 10154return nonzero for any @var{entity} that needs mode-switching. 10155If you define this macro, you also have to define 10156@code{NUM_MODES_FOR_MODE_SWITCHING}, @code{TARGET_MODE_NEEDED}, 10157@code{TARGET_MODE_PRIORITY} and @code{TARGET_MODE_EMIT}. 10158@code{TARGET_MODE_AFTER}, @code{TARGET_MODE_ENTRY}, and @code{TARGET_MODE_EXIT} 10159are optional. 10160@end defmac 10161 10162@defmac NUM_MODES_FOR_MODE_SWITCHING 10163If you define @code{OPTIMIZE_MODE_SWITCHING}, you have to define this as 10164initializer for an array of integers. Each initializer element 10165N refers to an entity that needs mode switching, and specifies the number 10166of different modes that might need to be set for this entity. 10167The position of the initializer in the initializer---starting counting at 10168zero---determines the integer that is used to refer to the mode-switched 10169entity in question. 10170In macros that take mode arguments / yield a mode result, modes are 10171represented as numbers 0 @dots{} N @minus{} 1. N is used to specify that no mode 10172switch is needed / supplied. 10173@end defmac 10174 10175@deftypefn {Target Hook} void TARGET_MODE_EMIT (int @var{entity}, int @var{mode}, int @var{prev_mode}, HARD_REG_SET @var{regs_live}) 10176Generate one or more insns to set @var{entity} to @var{mode}. @var{hard_reg_live} is the set of hard registers live at the point where the insn(s) are to be inserted. @var{prev_moxde} indicates the mode to switch from. Sets of a lower numbered entity will be emitted before sets of a higher numbered entity to a mode of the same or lower priority. 10177@end deftypefn 10178 10179@deftypefn {Target Hook} int TARGET_MODE_NEEDED (int @var{entity}, rtx_insn *@var{insn}) 10180@var{entity} is an integer specifying a mode-switched entity. If @code{OPTIMIZE_MODE_SWITCHING} is defined, you must define this macro to return an integer value not larger than the corresponding element in @code{NUM_MODES_FOR_MODE_SWITCHING}, to denote the mode that @var{entity} must be switched into prior to the execution of @var{insn}. 10181@end deftypefn 10182 10183@deftypefn {Target Hook} int TARGET_MODE_AFTER (int @var{entity}, int @var{mode}, rtx_insn *@var{insn}) 10184@var{entity} is an integer specifying a mode-switched entity. If this macro is defined, it is evaluated for every @var{insn} during mode switching. It determines the mode that an insn results in (if different from the incoming mode). 10185@end deftypefn 10186 10187@deftypefn {Target Hook} int TARGET_MODE_ENTRY (int @var{entity}) 10188If this macro is defined, it is evaluated for every @var{entity} that needs mode switching. It should evaluate to an integer, which is a mode that @var{entity} is assumed to be switched to at function entry. If @code{TARGET_MODE_ENTRY} is defined then @code{TARGET_MODE_EXIT} must be defined. 10189@end deftypefn 10190 10191@deftypefn {Target Hook} int TARGET_MODE_EXIT (int @var{entity}) 10192If this macro is defined, it is evaluated for every @var{entity} that needs mode switching. It should evaluate to an integer, which is a mode that @var{entity} is assumed to be switched to at function exit. If @code{TARGET_MODE_EXIT} is defined then @code{TARGET_MODE_ENTRY} must be defined. 10193@end deftypefn 10194 10195@deftypefn {Target Hook} int TARGET_MODE_PRIORITY (int @var{entity}, int @var{n}) 10196This macro specifies the order in which modes for @var{entity} are processed. 0 is the highest priority, @code{NUM_MODES_FOR_MODE_SWITCHING[@var{entity}] - 1} the lowest. The value of the macro should be an integer designating a mode for @var{entity}. For any fixed @var{entity}, @code{mode_priority} (@var{entity}, @var{n}) shall be a bijection in 0 @dots{} @code{num_modes_for_mode_switching[@var{entity}] - 1}. 10197@end deftypefn 10198 10199@node Target Attributes 10200@section Defining target-specific uses of @code{__attribute__} 10201@cindex target attributes 10202@cindex machine attributes 10203@cindex attributes, target-specific 10204 10205Target-specific attributes may be defined for functions, data and types. 10206These are described using the following target hooks; they also need to 10207be documented in @file{extend.texi}. 10208 10209@deftypevr {Target Hook} {const struct attribute_spec *} TARGET_ATTRIBUTE_TABLE 10210If defined, this target hook points to an array of @samp{struct 10211attribute_spec} (defined in @file{tree-core.h}) specifying the machine 10212specific attributes for this target and some of the restrictions on the 10213entities to which these attributes are applied and the arguments they 10214take. 10215@end deftypevr 10216 10217@deftypefn {Target Hook} bool TARGET_ATTRIBUTE_TAKES_IDENTIFIER_P (const_tree @var{name}) 10218If defined, this target hook is a function which returns true if the 10219machine-specific attribute named @var{name} expects an identifier 10220given as its first argument to be passed on as a plain identifier, not 10221subjected to name lookup. If this is not defined, the default is 10222false for all machine-specific attributes. 10223@end deftypefn 10224 10225@deftypefn {Target Hook} int TARGET_COMP_TYPE_ATTRIBUTES (const_tree @var{type1}, const_tree @var{type2}) 10226If defined, this target hook is a function which returns zero if the attributes on 10227@var{type1} and @var{type2} are incompatible, one if they are compatible, 10228and two if they are nearly compatible (which causes a warning to be 10229generated). If this is not defined, machine-specific attributes are 10230supposed always to be compatible. 10231@end deftypefn 10232 10233@deftypefn {Target Hook} void TARGET_SET_DEFAULT_TYPE_ATTRIBUTES (tree @var{type}) 10234If defined, this target hook is a function which assigns default attributes to 10235the newly defined @var{type}. 10236@end deftypefn 10237 10238@deftypefn {Target Hook} tree TARGET_MERGE_TYPE_ATTRIBUTES (tree @var{type1}, tree @var{type2}) 10239Define this target hook if the merging of type attributes needs special 10240handling. If defined, the result is a list of the combined 10241@code{TYPE_ATTRIBUTES} of @var{type1} and @var{type2}. It is assumed 10242that @code{comptypes} has already been called and returned 1. This 10243function may call @code{merge_attributes} to handle machine-independent 10244merging. 10245@end deftypefn 10246 10247@deftypefn {Target Hook} tree TARGET_MERGE_DECL_ATTRIBUTES (tree @var{olddecl}, tree @var{newdecl}) 10248Define this target hook if the merging of decl attributes needs special 10249handling. If defined, the result is a list of the combined 10250@code{DECL_ATTRIBUTES} of @var{olddecl} and @var{newdecl}. 10251@var{newdecl} is a duplicate declaration of @var{olddecl}. Examples of 10252when this is needed are when one attribute overrides another, or when an 10253attribute is nullified by a subsequent definition. This function may 10254call @code{merge_attributes} to handle machine-independent merging. 10255 10256@findex TARGET_DLLIMPORT_DECL_ATTRIBUTES 10257If the only target-specific handling you require is @samp{dllimport} 10258for Microsoft Windows targets, you should define the macro 10259@code{TARGET_DLLIMPORT_DECL_ATTRIBUTES} to @code{1}. The compiler 10260will then define a function called 10261@code{merge_dllimport_decl_attributes} which can then be defined as 10262the expansion of @code{TARGET_MERGE_DECL_ATTRIBUTES}. You can also 10263add @code{handle_dll_attribute} in the attribute table for your port 10264to perform initial processing of the @samp{dllimport} and 10265@samp{dllexport} attributes. This is done in @file{i386/cygwin.h} and 10266@file{i386/i386.c}, for example. 10267@end deftypefn 10268 10269@deftypefn {Target Hook} bool TARGET_VALID_DLLIMPORT_ATTRIBUTE_P (const_tree @var{decl}) 10270@var{decl} is a variable or function with @code{__attribute__((dllimport))} specified. Use this hook if the target needs to add extra validation checks to @code{handle_dll_attribute}. 10271@end deftypefn 10272 10273@defmac TARGET_DECLSPEC 10274Define this macro to a nonzero value if you want to treat 10275@code{__declspec(X)} as equivalent to @code{__attribute((X))}. By 10276default, this behavior is enabled only for targets that define 10277@code{TARGET_DLLIMPORT_DECL_ATTRIBUTES}. The current implementation 10278of @code{__declspec} is via a built-in macro, but you should not rely 10279on this implementation detail. 10280@end defmac 10281 10282@deftypefn {Target Hook} void TARGET_INSERT_ATTRIBUTES (tree @var{node}, tree *@var{attr_ptr}) 10283Define this target hook if you want to be able to add attributes to a decl 10284when it is being created. This is normally useful for back ends which 10285wish to implement a pragma by using the attributes which correspond to 10286the pragma's effect. The @var{node} argument is the decl which is being 10287created. The @var{attr_ptr} argument is a pointer to the attribute list 10288for this decl. The list itself should not be modified, since it may be 10289shared with other decls, but attributes may be chained on the head of 10290the list and @code{*@var{attr_ptr}} modified to point to the new 10291attributes, or a copy of the list may be made if further changes are 10292needed. 10293@end deftypefn 10294 10295@deftypefn {Target Hook} bool TARGET_FUNCTION_ATTRIBUTE_INLINABLE_P (const_tree @var{fndecl}) 10296@cindex inlining 10297This target hook returns @code{true} if it is OK to inline @var{fndecl} 10298into the current function, despite its having target-specific 10299attributes, @code{false} otherwise. By default, if a function has a 10300target specific attribute attached to it, it will not be inlined. 10301@end deftypefn 10302 10303@deftypefn {Target Hook} bool TARGET_OPTION_VALID_ATTRIBUTE_P (tree @var{fndecl}, tree @var{name}, tree @var{args}, int @var{flags}) 10304This hook is called to parse @code{attribute(target("..."))}, which 10305allows setting target-specific options on individual functions. 10306These function-specific options may differ 10307from the options specified on the command line. The hook should return 10308@code{true} if the options are valid. 10309 10310The hook should set the @code{DECL_FUNCTION_SPECIFIC_TARGET} field in 10311the function declaration to hold a pointer to a target-specific 10312@code{struct cl_target_option} structure. 10313@end deftypefn 10314 10315@deftypefn {Target Hook} void TARGET_OPTION_SAVE (struct cl_target_option *@var{ptr}, struct gcc_options *@var{opts}) 10316This hook is called to save any additional target-specific information 10317in the @code{struct cl_target_option} structure for function-specific 10318options from the @code{struct gcc_options} structure. 10319@xref{Option file format}. 10320@end deftypefn 10321 10322@deftypefn {Target Hook} void TARGET_OPTION_RESTORE (struct gcc_options *@var{opts}, struct cl_target_option *@var{ptr}) 10323This hook is called to restore any additional target-specific 10324information in the @code{struct cl_target_option} structure for 10325function-specific options to the @code{struct gcc_options} structure. 10326@end deftypefn 10327 10328@deftypefn {Target Hook} void TARGET_OPTION_POST_STREAM_IN (struct cl_target_option *@var{ptr}) 10329This hook is called to update target-specific information in the 10330@code{struct cl_target_option} structure after it is streamed in from 10331LTO bytecode. 10332@end deftypefn 10333 10334@deftypefn {Target Hook} void TARGET_OPTION_PRINT (FILE *@var{file}, int @var{indent}, struct cl_target_option *@var{ptr}) 10335This hook is called to print any additional target-specific 10336information in the @code{struct cl_target_option} structure for 10337function-specific options. 10338@end deftypefn 10339 10340@deftypefn {Target Hook} bool TARGET_OPTION_PRAGMA_PARSE (tree @var{args}, tree @var{pop_target}) 10341This target hook parses the options for @code{#pragma GCC target}, which 10342sets the target-specific options for functions that occur later in the 10343input stream. The options accepted should be the same as those handled by the 10344@code{TARGET_OPTION_VALID_ATTRIBUTE_P} hook. 10345@end deftypefn 10346 10347@deftypefn {Target Hook} void TARGET_OPTION_OVERRIDE (void) 10348Sometimes certain combinations of command options do not make sense on 10349a particular target machine. You can override the hook 10350@code{TARGET_OPTION_OVERRIDE} to take account of this. This hooks is called 10351once just after all the command options have been parsed. 10352 10353Don't use this hook to turn on various extra optimizations for 10354@option{-O}. That is what @code{TARGET_OPTION_OPTIMIZATION} is for. 10355 10356If you need to do something whenever the optimization level is 10357changed via the optimize attribute or pragma, see 10358@code{TARGET_OVERRIDE_OPTIONS_AFTER_CHANGE} 10359@end deftypefn 10360 10361@deftypefn {Target Hook} bool TARGET_OPTION_FUNCTION_VERSIONS (tree @var{decl1}, tree @var{decl2}) 10362This target hook returns @code{true} if @var{DECL1} and @var{DECL2} are 10363versions of the same function. @var{DECL1} and @var{DECL2} are function 10364versions if and only if they have the same function signature and 10365different target specific attributes, that is, they are compiled for 10366different target machines. 10367@end deftypefn 10368 10369@deftypefn {Target Hook} bool TARGET_CAN_INLINE_P (tree @var{caller}, tree @var{callee}) 10370This target hook returns @code{false} if the @var{caller} function 10371cannot inline @var{callee}, based on target specific information. By 10372default, inlining is not allowed if the callee function has function 10373specific target options and the caller does not use the same options. 10374@end deftypefn 10375 10376@deftypefn {Target Hook} void TARGET_RELAYOUT_FUNCTION (tree @var{fndecl}) 10377This target hook fixes function @var{fndecl} after attributes are processed. Default does nothing. On ARM, the default function's alignment is updated with the attribute target. 10378@end deftypefn 10379 10380@node Emulated TLS 10381@section Emulating TLS 10382@cindex Emulated TLS 10383 10384For targets whose psABI does not provide Thread Local Storage via 10385specific relocations and instruction sequences, an emulation layer is 10386used. A set of target hooks allows this emulation layer to be 10387configured for the requirements of a particular target. For instance 10388the psABI may in fact specify TLS support in terms of an emulation 10389layer. 10390 10391The emulation layer works by creating a control object for every TLS 10392object. To access the TLS object, a lookup function is provided 10393which, when given the address of the control object, will return the 10394address of the current thread's instance of the TLS object. 10395 10396@deftypevr {Target Hook} {const char *} TARGET_EMUTLS_GET_ADDRESS 10397Contains the name of the helper function that uses a TLS control 10398object to locate a TLS instance. The default causes libgcc's 10399emulated TLS helper function to be used. 10400@end deftypevr 10401 10402@deftypevr {Target Hook} {const char *} TARGET_EMUTLS_REGISTER_COMMON 10403Contains the name of the helper function that should be used at 10404program startup to register TLS objects that are implicitly 10405initialized to zero. If this is @code{NULL}, all TLS objects will 10406have explicit initializers. The default causes libgcc's emulated TLS 10407registration function to be used. 10408@end deftypevr 10409 10410@deftypevr {Target Hook} {const char *} TARGET_EMUTLS_VAR_SECTION 10411Contains the name of the section in which TLS control variables should 10412be placed. The default of @code{NULL} allows these to be placed in 10413any section. 10414@end deftypevr 10415 10416@deftypevr {Target Hook} {const char *} TARGET_EMUTLS_TMPL_SECTION 10417Contains the name of the section in which TLS initializers should be 10418placed. The default of @code{NULL} allows these to be placed in any 10419section. 10420@end deftypevr 10421 10422@deftypevr {Target Hook} {const char *} TARGET_EMUTLS_VAR_PREFIX 10423Contains the prefix to be prepended to TLS control variable names. 10424The default of @code{NULL} uses a target-specific prefix. 10425@end deftypevr 10426 10427@deftypevr {Target Hook} {const char *} TARGET_EMUTLS_TMPL_PREFIX 10428Contains the prefix to be prepended to TLS initializer objects. The 10429default of @code{NULL} uses a target-specific prefix. 10430@end deftypevr 10431 10432@deftypefn {Target Hook} tree TARGET_EMUTLS_VAR_FIELDS (tree @var{type}, tree *@var{name}) 10433Specifies a function that generates the FIELD_DECLs for a TLS control 10434object type. @var{type} is the RECORD_TYPE the fields are for and 10435@var{name} should be filled with the structure tag, if the default of 10436@code{__emutls_object} is unsuitable. The default creates a type suitable 10437for libgcc's emulated TLS function. 10438@end deftypefn 10439 10440@deftypefn {Target Hook} tree TARGET_EMUTLS_VAR_INIT (tree @var{var}, tree @var{decl}, tree @var{tmpl_addr}) 10441Specifies a function that generates the CONSTRUCTOR to initialize a 10442TLS control object. @var{var} is the TLS control object, @var{decl} 10443is the TLS object and @var{tmpl_addr} is the address of the 10444initializer. The default initializes libgcc's emulated TLS control object. 10445@end deftypefn 10446 10447@deftypevr {Target Hook} bool TARGET_EMUTLS_VAR_ALIGN_FIXED 10448Specifies whether the alignment of TLS control variable objects is 10449fixed and should not be increased as some backends may do to optimize 10450single objects. The default is false. 10451@end deftypevr 10452 10453@deftypevr {Target Hook} bool TARGET_EMUTLS_DEBUG_FORM_TLS_ADDRESS 10454Specifies whether a DWARF @code{DW_OP_form_tls_address} location descriptor 10455may be used to describe emulated TLS control objects. 10456@end deftypevr 10457 10458@node MIPS Coprocessors 10459@section Defining coprocessor specifics for MIPS targets. 10460@cindex MIPS coprocessor-definition macros 10461 10462The MIPS specification allows MIPS implementations to have as many as 4 10463coprocessors, each with as many as 32 private registers. GCC supports 10464accessing these registers and transferring values between the registers 10465and memory using asm-ized variables. For example: 10466 10467@smallexample 10468 register unsigned int cp0count asm ("c0r1"); 10469 unsigned int d; 10470 10471 d = cp0count + 3; 10472@end smallexample 10473 10474(``c0r1'' is the default name of register 1 in coprocessor 0; alternate 10475names may be added as described below, or the default names may be 10476overridden entirely in @code{SUBTARGET_CONDITIONAL_REGISTER_USAGE}.) 10477 10478Coprocessor registers are assumed to be epilogue-used; sets to them will 10479be preserved even if it does not appear that the register is used again 10480later in the function. 10481 10482Another note: according to the MIPS spec, coprocessor 1 (if present) is 10483the FPU@. One accesses COP1 registers through standard mips 10484floating-point support; they are not included in this mechanism. 10485 10486@node PCH Target 10487@section Parameters for Precompiled Header Validity Checking 10488@cindex parameters, precompiled headers 10489 10490@deftypefn {Target Hook} {void *} TARGET_GET_PCH_VALIDITY (size_t *@var{sz}) 10491This hook returns a pointer to the data needed by 10492@code{TARGET_PCH_VALID_P} and sets 10493@samp{*@var{sz}} to the size of the data in bytes. 10494@end deftypefn 10495 10496@deftypefn {Target Hook} {const char *} TARGET_PCH_VALID_P (const void *@var{data}, size_t @var{sz}) 10497This hook checks whether the options used to create a PCH file are 10498compatible with the current settings. It returns @code{NULL} 10499if so and a suitable error message if not. Error messages will 10500be presented to the user and must be localized using @samp{_(@var{msg})}. 10501 10502@var{data} is the data that was returned by @code{TARGET_GET_PCH_VALIDITY} 10503when the PCH file was created and @var{sz} is the size of that data in bytes. 10504It's safe to assume that the data was created by the same version of the 10505compiler, so no format checking is needed. 10506 10507The default definition of @code{default_pch_valid_p} should be 10508suitable for most targets. 10509@end deftypefn 10510 10511@deftypefn {Target Hook} {const char *} TARGET_CHECK_PCH_TARGET_FLAGS (int @var{pch_flags}) 10512If this hook is nonnull, the default implementation of 10513@code{TARGET_PCH_VALID_P} will use it to check for compatible values 10514of @code{target_flags}. @var{pch_flags} specifies the value that 10515@code{target_flags} had when the PCH file was created. The return 10516value is the same as for @code{TARGET_PCH_VALID_P}. 10517@end deftypefn 10518 10519@deftypefn {Target Hook} void TARGET_PREPARE_PCH_SAVE (void) 10520Called before writing out a PCH file. If the target has some 10521garbage-collected data that needs to be in a particular state on PCH loads, 10522it can use this hook to enforce that state. Very few targets need 10523to do anything here. 10524@end deftypefn 10525 10526@node C++ ABI 10527@section C++ ABI parameters 10528@cindex parameters, c++ abi 10529 10530@deftypefn {Target Hook} tree TARGET_CXX_GUARD_TYPE (void) 10531Define this hook to override the integer type used for guard variables. 10532These are used to implement one-time construction of static objects. The 10533default is long_long_integer_type_node. 10534@end deftypefn 10535 10536@deftypefn {Target Hook} bool TARGET_CXX_GUARD_MASK_BIT (void) 10537This hook determines how guard variables are used. It should return 10538@code{false} (the default) if the first byte should be used. A return value of 10539@code{true} indicates that only the least significant bit should be used. 10540@end deftypefn 10541 10542@deftypefn {Target Hook} tree TARGET_CXX_GET_COOKIE_SIZE (tree @var{type}) 10543This hook returns the size of the cookie to use when allocating an array 10544whose elements have the indicated @var{type}. Assumes that it is already 10545known that a cookie is needed. The default is 10546@code{max(sizeof (size_t), alignof(type))}, as defined in section 2.7 of the 10547IA64/Generic C++ ABI@. 10548@end deftypefn 10549 10550@deftypefn {Target Hook} bool TARGET_CXX_COOKIE_HAS_SIZE (void) 10551This hook should return @code{true} if the element size should be stored in 10552array cookies. The default is to return @code{false}. 10553@end deftypefn 10554 10555@deftypefn {Target Hook} int TARGET_CXX_IMPORT_EXPORT_CLASS (tree @var{type}, int @var{import_export}) 10556If defined by a backend this hook allows the decision made to export 10557class @var{type} to be overruled. Upon entry @var{import_export} 10558will contain 1 if the class is going to be exported, @minus{}1 if it is going 10559to be imported and 0 otherwise. This function should return the 10560modified value and perform any other actions necessary to support the 10561backend's targeted operating system. 10562@end deftypefn 10563 10564@deftypefn {Target Hook} bool TARGET_CXX_CDTOR_RETURNS_THIS (void) 10565This hook should return @code{true} if constructors and destructors return 10566the address of the object created/destroyed. The default is to return 10567@code{false}. 10568@end deftypefn 10569 10570@deftypefn {Target Hook} bool TARGET_CXX_KEY_METHOD_MAY_BE_INLINE (void) 10571This hook returns true if the key method for a class (i.e., the method 10572which, if defined in the current translation unit, causes the virtual 10573table to be emitted) may be an inline function. Under the standard 10574Itanium C++ ABI the key method may be an inline function so long as 10575the function is not declared inline in the class definition. Under 10576some variants of the ABI, an inline function can never be the key 10577method. The default is to return @code{true}. 10578@end deftypefn 10579 10580@deftypefn {Target Hook} void TARGET_CXX_DETERMINE_CLASS_DATA_VISIBILITY (tree @var{decl}) 10581@var{decl} is a virtual table, virtual table table, typeinfo object, or other similar implicit class data object that will be emitted with external linkage in this translation unit. No ELF visibility has been explicitly specified. If the target needs to specify a visibility other than that of the containing class, use this hook to set @code{DECL_VISIBILITY} and @code{DECL_VISIBILITY_SPECIFIED}. 10582@end deftypefn 10583 10584@deftypefn {Target Hook} bool TARGET_CXX_CLASS_DATA_ALWAYS_COMDAT (void) 10585This hook returns true (the default) if virtual tables and other 10586similar implicit class data objects are always COMDAT if they have 10587external linkage. If this hook returns false, then class data for 10588classes whose virtual table will be emitted in only one translation 10589unit will not be COMDAT. 10590@end deftypefn 10591 10592@deftypefn {Target Hook} bool TARGET_CXX_LIBRARY_RTTI_COMDAT (void) 10593This hook returns true (the default) if the RTTI information for 10594the basic types which is defined in the C++ runtime should always 10595be COMDAT, false if it should not be COMDAT. 10596@end deftypefn 10597 10598@deftypefn {Target Hook} bool TARGET_CXX_USE_AEABI_ATEXIT (void) 10599This hook returns true if @code{__aeabi_atexit} (as defined by the ARM EABI) 10600should be used to register static destructors when @option{-fuse-cxa-atexit} 10601is in effect. The default is to return false to use @code{__cxa_atexit}. 10602@end deftypefn 10603 10604@deftypefn {Target Hook} bool TARGET_CXX_USE_ATEXIT_FOR_CXA_ATEXIT (void) 10605This hook returns true if the target @code{atexit} function can be used 10606in the same manner as @code{__cxa_atexit} to register C++ static 10607destructors. This requires that @code{atexit}-registered functions in 10608shared libraries are run in the correct order when the libraries are 10609unloaded. The default is to return false. 10610@end deftypefn 10611 10612@deftypefn {Target Hook} void TARGET_CXX_ADJUST_CLASS_AT_DEFINITION (tree @var{type}) 10613@var{type} is a C++ class (i.e., RECORD_TYPE or UNION_TYPE) that has just been defined. Use this hook to make adjustments to the class (eg, tweak visibility or perform any other required target modifications). 10614@end deftypefn 10615 10616@deftypefn {Target Hook} tree TARGET_CXX_DECL_MANGLING_CONTEXT (const_tree @var{decl}) 10617Return target-specific mangling context of @var{decl} or @code{NULL_TREE}. 10618@end deftypefn 10619 10620@node Named Address Spaces 10621@section Adding support for named address spaces 10622@cindex named address spaces 10623 10624The draft technical report of the ISO/IEC JTC1 S22 WG14 N1275 10625standards committee, @cite{Programming Languages - C - Extensions to 10626support embedded processors}, specifies a syntax for embedded 10627processors to specify alternate address spaces. You can configure a 10628GCC port to support section 5.1 of the draft report to add support for 10629address spaces other than the default address space. These address 10630spaces are new keywords that are similar to the @code{volatile} and 10631@code{const} type attributes. 10632 10633Pointers to named address spaces can have a different size than 10634pointers to the generic address space. 10635 10636For example, the SPU port uses the @code{__ea} address space to refer 10637to memory in the host processor, rather than memory local to the SPU 10638processor. Access to memory in the @code{__ea} address space involves 10639issuing DMA operations to move data between the host processor and the 10640local processor memory address space. Pointers in the @code{__ea} 10641address space are either 32 bits or 64 bits based on the 10642@option{-mea32} or @option{-mea64} switches (native SPU pointers are 10643always 32 bits). 10644 10645Internally, address spaces are represented as a small integer in the 10646range 0 to 15 with address space 0 being reserved for the generic 10647address space. 10648 10649To register a named address space qualifier keyword with the C front end, 10650the target may call the @code{c_register_addr_space} routine. For example, 10651the SPU port uses the following to declare @code{__ea} as the keyword for 10652named address space #1: 10653@smallexample 10654#define ADDR_SPACE_EA 1 10655c_register_addr_space ("__ea", ADDR_SPACE_EA); 10656@end smallexample 10657 10658@deftypefn {Target Hook} scalar_int_mode TARGET_ADDR_SPACE_POINTER_MODE (addr_space_t @var{address_space}) 10659Define this to return the machine mode to use for pointers to 10660@var{address_space} if the target supports named address spaces. 10661The default version of this hook returns @code{ptr_mode}. 10662@end deftypefn 10663 10664@deftypefn {Target Hook} scalar_int_mode TARGET_ADDR_SPACE_ADDRESS_MODE (addr_space_t @var{address_space}) 10665Define this to return the machine mode to use for addresses in 10666@var{address_space} if the target supports named address spaces. 10667The default version of this hook returns @code{Pmode}. 10668@end deftypefn 10669 10670@deftypefn {Target Hook} bool TARGET_ADDR_SPACE_VALID_POINTER_MODE (scalar_int_mode @var{mode}, addr_space_t @var{as}) 10671Define this to return nonzero if the port can handle pointers 10672with machine mode @var{mode} to address space @var{as}. This target 10673hook is the same as the @code{TARGET_VALID_POINTER_MODE} target hook, 10674except that it includes explicit named address space support. The default 10675version of this hook returns true for the modes returned by either the 10676@code{TARGET_ADDR_SPACE_POINTER_MODE} or @code{TARGET_ADDR_SPACE_ADDRESS_MODE} 10677target hooks for the given address space. 10678@end deftypefn 10679 10680@deftypefn {Target Hook} bool TARGET_ADDR_SPACE_LEGITIMATE_ADDRESS_P (machine_mode @var{mode}, rtx @var{exp}, bool @var{strict}, addr_space_t @var{as}) 10681Define this to return true if @var{exp} is a valid address for mode 10682@var{mode} in the named address space @var{as}. The @var{strict} 10683parameter says whether strict addressing is in effect after reload has 10684finished. This target hook is the same as the 10685@code{TARGET_LEGITIMATE_ADDRESS_P} target hook, except that it includes 10686explicit named address space support. 10687@end deftypefn 10688 10689@deftypefn {Target Hook} rtx TARGET_ADDR_SPACE_LEGITIMIZE_ADDRESS (rtx @var{x}, rtx @var{oldx}, machine_mode @var{mode}, addr_space_t @var{as}) 10690Define this to modify an invalid address @var{x} to be a valid address 10691with mode @var{mode} in the named address space @var{as}. This target 10692hook is the same as the @code{TARGET_LEGITIMIZE_ADDRESS} target hook, 10693except that it includes explicit named address space support. 10694@end deftypefn 10695 10696@deftypefn {Target Hook} bool TARGET_ADDR_SPACE_SUBSET_P (addr_space_t @var{subset}, addr_space_t @var{superset}) 10697Define this to return whether the @var{subset} named address space is 10698contained within the @var{superset} named address space. Pointers to 10699a named address space that is a subset of another named address space 10700will be converted automatically without a cast if used together in 10701arithmetic operations. Pointers to a superset address space can be 10702converted to pointers to a subset address space via explicit casts. 10703@end deftypefn 10704 10705@deftypefn {Target Hook} bool TARGET_ADDR_SPACE_ZERO_ADDRESS_VALID (addr_space_t @var{as}) 10706Define this to modify the default handling of address 0 for the 10707address space. Return true if 0 should be considered a valid address. 10708@end deftypefn 10709 10710@deftypefn {Target Hook} rtx TARGET_ADDR_SPACE_CONVERT (rtx @var{op}, tree @var{from_type}, tree @var{to_type}) 10711Define this to convert the pointer expression represented by the RTL 10712@var{op} with type @var{from_type} that points to a named address 10713space to a new pointer expression with type @var{to_type} that points 10714to a different named address space. When this hook it called, it is 10715guaranteed that one of the two address spaces is a subset of the other, 10716as determined by the @code{TARGET_ADDR_SPACE_SUBSET_P} target hook. 10717@end deftypefn 10718 10719@deftypefn {Target Hook} int TARGET_ADDR_SPACE_DEBUG (addr_space_t @var{as}) 10720Define this to define how the address space is encoded in dwarf. 10721The result is the value to be used with @code{DW_AT_address_class}. 10722@end deftypefn 10723 10724@deftypefn {Target Hook} void TARGET_ADDR_SPACE_DIAGNOSE_USAGE (addr_space_t @var{as}, location_t @var{loc}) 10725Define this hook if the availability of an address space depends on 10726command line options and some diagnostics should be printed when the 10727address space is used. This hook is called during parsing and allows 10728to emit a better diagnostic compared to the case where the address space 10729was not registered with @code{c_register_addr_space}. @var{as} is 10730the address space as registered with @code{c_register_addr_space}. 10731@var{loc} is the location of the address space qualifier token. 10732The default implementation does nothing. 10733@end deftypefn 10734 10735@node Misc 10736@section Miscellaneous Parameters 10737@cindex parameters, miscellaneous 10738 10739@c prevent bad page break with this line 10740Here are several miscellaneous parameters. 10741 10742@defmac HAS_LONG_COND_BRANCH 10743Define this boolean macro to indicate whether or not your architecture 10744has conditional branches that can span all of memory. It is used in 10745conjunction with an optimization that partitions hot and cold basic 10746blocks into separate sections of the executable. If this macro is 10747set to false, gcc will convert any conditional branches that attempt 10748to cross between sections into unconditional branches or indirect jumps. 10749@end defmac 10750 10751@defmac HAS_LONG_UNCOND_BRANCH 10752Define this boolean macro to indicate whether or not your architecture 10753has unconditional branches that can span all of memory. It is used in 10754conjunction with an optimization that partitions hot and cold basic 10755blocks into separate sections of the executable. If this macro is 10756set to false, gcc will convert any unconditional branches that attempt 10757to cross between sections into indirect jumps. 10758@end defmac 10759 10760@defmac CASE_VECTOR_MODE 10761An alias for a machine mode name. This is the machine mode that 10762elements of a jump-table should have. 10763@end defmac 10764 10765@defmac CASE_VECTOR_SHORTEN_MODE (@var{min_offset}, @var{max_offset}, @var{body}) 10766Optional: return the preferred mode for an @code{addr_diff_vec} 10767when the minimum and maximum offset are known. If you define this, 10768it enables extra code in branch shortening to deal with @code{addr_diff_vec}. 10769To make this work, you also have to define @code{INSN_ALIGN} and 10770make the alignment for @code{addr_diff_vec} explicit. 10771The @var{body} argument is provided so that the offset_unsigned and scale 10772flags can be updated. 10773@end defmac 10774 10775@defmac CASE_VECTOR_PC_RELATIVE 10776Define this macro to be a C expression to indicate when jump-tables 10777should contain relative addresses. You need not define this macro if 10778jump-tables never contain relative addresses, or jump-tables should 10779contain relative addresses only when @option{-fPIC} or @option{-fPIC} 10780is in effect. 10781@end defmac 10782 10783@deftypefn {Target Hook} {unsigned int} TARGET_CASE_VALUES_THRESHOLD (void) 10784This function return the smallest number of different values for which it 10785is best to use a jump-table instead of a tree of conditional branches. 10786The default is four for machines with a @code{casesi} instruction and 10787five otherwise. This is best for most machines. 10788@end deftypefn 10789 10790@defmac WORD_REGISTER_OPERATIONS 10791Define this macro to 1 if operations between registers with integral mode 10792smaller than a word are always performed on the entire register. To be 10793more explicit, if you start with a pair of @code{word_mode} registers with 10794known values and you do a subword, for example @code{QImode}, addition on 10795the low part of the registers, then the compiler may consider that the 10796result has a known value in @code{word_mode} too if the macro is defined 10797to 1. Most RISC machines have this property and most CISC machines do not. 10798@end defmac 10799 10800@deftypefn {Target Hook} {unsigned int} TARGET_MIN_ARITHMETIC_PRECISION (void) 10801On some RISC architectures with 64-bit registers, the processor also 10802maintains 32-bit condition codes that make it possible to do real 32-bit 10803arithmetic, although the operations are performed on the full registers. 10804 10805On such architectures, defining this hook to 32 tells the compiler to try 10806using 32-bit arithmetical operations setting the condition codes instead 10807of doing full 64-bit arithmetic. 10808 10809More generally, define this hook on RISC architectures if you want the 10810compiler to try using arithmetical operations setting the condition codes 10811with a precision lower than the word precision. 10812 10813You need not define this hook if @code{WORD_REGISTER_OPERATIONS} is not 10814defined to 1. 10815@end deftypefn 10816 10817@defmac LOAD_EXTEND_OP (@var{mem_mode}) 10818Define this macro to be a C expression indicating when insns that read 10819memory in @var{mem_mode}, an integral mode narrower than a word, set the 10820bits outside of @var{mem_mode} to be either the sign-extension or the 10821zero-extension of the data read. Return @code{SIGN_EXTEND} for values 10822of @var{mem_mode} for which the 10823insn sign-extends, @code{ZERO_EXTEND} for which it zero-extends, and 10824@code{UNKNOWN} for other modes. 10825 10826This macro is not called with @var{mem_mode} non-integral or with a width 10827greater than or equal to @code{BITS_PER_WORD}, so you may return any 10828value in this case. Do not define this macro if it would always return 10829@code{UNKNOWN}. On machines where this macro is defined, you will normally 10830define it as the constant @code{SIGN_EXTEND} or @code{ZERO_EXTEND}. 10831 10832You may return a non-@code{UNKNOWN} value even if for some hard registers 10833the sign extension is not performed, if for the @code{REGNO_REG_CLASS} 10834of these hard registers @code{TARGET_CAN_CHANGE_MODE_CLASS} returns false 10835when the @var{from} mode is @var{mem_mode} and the @var{to} mode is any 10836integral mode larger than this but not larger than @code{word_mode}. 10837 10838You must return @code{UNKNOWN} if for some hard registers that allow this 10839mode, @code{TARGET_CAN_CHANGE_MODE_CLASS} says that they cannot change to 10840@code{word_mode}, but that they can change to another integral mode that 10841is larger then @var{mem_mode} but still smaller than @code{word_mode}. 10842@end defmac 10843 10844@defmac SHORT_IMMEDIATES_SIGN_EXTEND 10845Define this macro to 1 if loading short immediate values into registers sign 10846extends. 10847@end defmac 10848 10849@deftypefn {Target Hook} {unsigned int} TARGET_MIN_DIVISIONS_FOR_RECIP_MUL (machine_mode @var{mode}) 10850When @option{-ffast-math} is in effect, GCC tries to optimize 10851divisions by the same divisor, by turning them into multiplications by 10852the reciprocal. This target hook specifies the minimum number of divisions 10853that should be there for GCC to perform the optimization for a variable 10854of mode @var{mode}. The default implementation returns 3 if the machine 10855has an instruction for the division, and 2 if it does not. 10856@end deftypefn 10857 10858@defmac MOVE_MAX 10859The maximum number of bytes that a single instruction can move quickly 10860between memory and registers or between two memory locations. 10861@end defmac 10862 10863@defmac MAX_MOVE_MAX 10864The maximum number of bytes that a single instruction can move quickly 10865between memory and registers or between two memory locations. If this 10866is undefined, the default is @code{MOVE_MAX}. Otherwise, it is the 10867constant value that is the largest value that @code{MOVE_MAX} can have 10868at run-time. 10869@end defmac 10870 10871@defmac SHIFT_COUNT_TRUNCATED 10872A C expression that is nonzero if on this machine the number of bits 10873actually used for the count of a shift operation is equal to the number 10874of bits needed to represent the size of the object being shifted. When 10875this macro is nonzero, the compiler will assume that it is safe to omit 10876a sign-extend, zero-extend, and certain bitwise `and' instructions that 10877truncates the count of a shift operation. On machines that have 10878instructions that act on bit-fields at variable positions, which may 10879include `bit test' instructions, a nonzero @code{SHIFT_COUNT_TRUNCATED} 10880also enables deletion of truncations of the values that serve as 10881arguments to bit-field instructions. 10882 10883If both types of instructions truncate the count (for shifts) and 10884position (for bit-field operations), or if no variable-position bit-field 10885instructions exist, you should define this macro. 10886 10887However, on some machines, such as the 80386 and the 680x0, truncation 10888only applies to shift operations and not the (real or pretended) 10889bit-field operations. Define @code{SHIFT_COUNT_TRUNCATED} to be zero on 10890such machines. Instead, add patterns to the @file{md} file that include 10891the implied truncation of the shift instructions. 10892 10893You need not define this macro if it would always have the value of zero. 10894@end defmac 10895 10896@anchor{TARGET_SHIFT_TRUNCATION_MASK} 10897@deftypefn {Target Hook} {unsigned HOST_WIDE_INT} TARGET_SHIFT_TRUNCATION_MASK (machine_mode @var{mode}) 10898This function describes how the standard shift patterns for @var{mode} 10899deal with shifts by negative amounts or by more than the width of the mode. 10900@xref{shift patterns}. 10901 10902On many machines, the shift patterns will apply a mask @var{m} to the 10903shift count, meaning that a fixed-width shift of @var{x} by @var{y} is 10904equivalent to an arbitrary-width shift of @var{x} by @var{y & m}. If 10905this is true for mode @var{mode}, the function should return @var{m}, 10906otherwise it should return 0. A return value of 0 indicates that no 10907particular behavior is guaranteed. 10908 10909Note that, unlike @code{SHIFT_COUNT_TRUNCATED}, this function does 10910@emph{not} apply to general shift rtxes; it applies only to instructions 10911that are generated by the named shift patterns. 10912 10913The default implementation of this function returns 10914@code{GET_MODE_BITSIZE (@var{mode}) - 1} if @code{SHIFT_COUNT_TRUNCATED} 10915and 0 otherwise. This definition is always safe, but if 10916@code{SHIFT_COUNT_TRUNCATED} is false, and some shift patterns 10917nevertheless truncate the shift count, you may get better code 10918by overriding it. 10919@end deftypefn 10920 10921@deftypefn {Target Hook} bool TARGET_TRULY_NOOP_TRUNCATION (poly_uint64 @var{outprec}, poly_uint64 @var{inprec}) 10922This hook returns true if it is safe to ``convert'' a value of 10923@var{inprec} bits to one of @var{outprec} bits (where @var{outprec} is 10924smaller than @var{inprec}) by merely operating on it as if it had only 10925@var{outprec} bits. The default returns true unconditionally, which 10926is correct for most machines. 10927 10928If @code{TARGET_MODES_TIEABLE_P} returns false for a pair of modes, 10929suboptimal code can result if this hook returns true for the corresponding 10930mode sizes. Making this hook return false in such cases may improve things. 10931@end deftypefn 10932 10933@deftypefn {Target Hook} int TARGET_MODE_REP_EXTENDED (scalar_int_mode @var{mode}, scalar_int_mode @var{rep_mode}) 10934The representation of an integral mode can be such that the values 10935are always extended to a wider integral mode. Return 10936@code{SIGN_EXTEND} if values of @var{mode} are represented in 10937sign-extended form to @var{rep_mode}. Return @code{UNKNOWN} 10938otherwise. (Currently, none of the targets use zero-extended 10939representation this way so unlike @code{LOAD_EXTEND_OP}, 10940@code{TARGET_MODE_REP_EXTENDED} is expected to return either 10941@code{SIGN_EXTEND} or @code{UNKNOWN}. Also no target extends 10942@var{mode} to @var{rep_mode} so that @var{rep_mode} is not the next 10943widest integral mode and currently we take advantage of this fact.) 10944 10945Similarly to @code{LOAD_EXTEND_OP} you may return a non-@code{UNKNOWN} 10946value even if the extension is not performed on certain hard registers 10947as long as for the @code{REGNO_REG_CLASS} of these hard registers 10948@code{TARGET_CAN_CHANGE_MODE_CLASS} returns false. 10949 10950Note that @code{TARGET_MODE_REP_EXTENDED} and @code{LOAD_EXTEND_OP} 10951describe two related properties. If you define 10952@code{TARGET_MODE_REP_EXTENDED (mode, word_mode)} you probably also want 10953to define @code{LOAD_EXTEND_OP (mode)} to return the same type of 10954extension. 10955 10956In order to enforce the representation of @code{mode}, 10957@code{TARGET_TRULY_NOOP_TRUNCATION} should return false when truncating to 10958@code{mode}. 10959@end deftypefn 10960 10961@defmac STORE_FLAG_VALUE 10962A C expression describing the value returned by a comparison operator 10963with an integral mode and stored by a store-flag instruction 10964(@samp{cstore@var{mode}4}) when the condition is true. This description must 10965apply to @emph{all} the @samp{cstore@var{mode}4} patterns and all the 10966comparison operators whose results have a @code{MODE_INT} mode. 10967 10968A value of 1 or @minus{}1 means that the instruction implementing the 10969comparison operator returns exactly 1 or @minus{}1 when the comparison is true 10970and 0 when the comparison is false. Otherwise, the value indicates 10971which bits of the result are guaranteed to be 1 when the comparison is 10972true. This value is interpreted in the mode of the comparison 10973operation, which is given by the mode of the first operand in the 10974@samp{cstore@var{mode}4} pattern. Either the low bit or the sign bit of 10975@code{STORE_FLAG_VALUE} be on. Presently, only those bits are used by 10976the compiler. 10977 10978If @code{STORE_FLAG_VALUE} is neither 1 or @minus{}1, the compiler will 10979generate code that depends only on the specified bits. It can also 10980replace comparison operators with equivalent operations if they cause 10981the required bits to be set, even if the remaining bits are undefined. 10982For example, on a machine whose comparison operators return an 10983@code{SImode} value and where @code{STORE_FLAG_VALUE} is defined as 10984@samp{0x80000000}, saying that just the sign bit is relevant, the 10985expression 10986 10987@smallexample 10988(ne:SI (and:SI @var{x} (const_int @var{power-of-2})) (const_int 0)) 10989@end smallexample 10990 10991@noindent 10992can be converted to 10993 10994@smallexample 10995(ashift:SI @var{x} (const_int @var{n})) 10996@end smallexample 10997 10998@noindent 10999where @var{n} is the appropriate shift count to move the bit being 11000tested into the sign bit. 11001 11002There is no way to describe a machine that always sets the low-order bit 11003for a true value, but does not guarantee the value of any other bits, 11004but we do not know of any machine that has such an instruction. If you 11005are trying to port GCC to such a machine, include an instruction to 11006perform a logical-and of the result with 1 in the pattern for the 11007comparison operators and let us know at @email{gcc@@gcc.gnu.org}. 11008 11009Often, a machine will have multiple instructions that obtain a value 11010from a comparison (or the condition codes). Here are rules to guide the 11011choice of value for @code{STORE_FLAG_VALUE}, and hence the instructions 11012to be used: 11013 11014@itemize @bullet 11015@item 11016Use the shortest sequence that yields a valid definition for 11017@code{STORE_FLAG_VALUE}. It is more efficient for the compiler to 11018``normalize'' the value (convert it to, e.g., 1 or 0) than for the 11019comparison operators to do so because there may be opportunities to 11020combine the normalization with other operations. 11021 11022@item 11023For equal-length sequences, use a value of 1 or @minus{}1, with @minus{}1 being 11024slightly preferred on machines with expensive jumps and 1 preferred on 11025other machines. 11026 11027@item 11028As a second choice, choose a value of @samp{0x80000001} if instructions 11029exist that set both the sign and low-order bits but do not define the 11030others. 11031 11032@item 11033Otherwise, use a value of @samp{0x80000000}. 11034@end itemize 11035 11036Many machines can produce both the value chosen for 11037@code{STORE_FLAG_VALUE} and its negation in the same number of 11038instructions. On those machines, you should also define a pattern for 11039those cases, e.g., one matching 11040 11041@smallexample 11042(set @var{A} (neg:@var{m} (ne:@var{m} @var{B} @var{C}))) 11043@end smallexample 11044 11045Some machines can also perform @code{and} or @code{plus} operations on 11046condition code values with less instructions than the corresponding 11047@samp{cstore@var{mode}4} insn followed by @code{and} or @code{plus}. On those 11048machines, define the appropriate patterns. Use the names @code{incscc} 11049and @code{decscc}, respectively, for the patterns which perform 11050@code{plus} or @code{minus} operations on condition code values. See 11051@file{rs6000.md} for some examples. The GNU Superoptimizer can be used to 11052find such instruction sequences on other machines. 11053 11054If this macro is not defined, the default value, 1, is used. You need 11055not define @code{STORE_FLAG_VALUE} if the machine has no store-flag 11056instructions, or if the value generated by these instructions is 1. 11057@end defmac 11058 11059@defmac FLOAT_STORE_FLAG_VALUE (@var{mode}) 11060A C expression that gives a nonzero @code{REAL_VALUE_TYPE} value that is 11061returned when comparison operators with floating-point results are true. 11062Define this macro on machines that have comparison operations that return 11063floating-point values. If there are no such operations, do not define 11064this macro. 11065@end defmac 11066 11067@defmac VECTOR_STORE_FLAG_VALUE (@var{mode}) 11068A C expression that gives a rtx representing the nonzero true element 11069for vector comparisons. The returned rtx should be valid for the inner 11070mode of @var{mode} which is guaranteed to be a vector mode. Define 11071this macro on machines that have vector comparison operations that 11072return a vector result. If there are no such operations, do not define 11073this macro. Typically, this macro is defined as @code{const1_rtx} or 11074@code{constm1_rtx}. This macro may return @code{NULL_RTX} to prevent 11075the compiler optimizing such vector comparison operations for the 11076given mode. 11077@end defmac 11078 11079@defmac CLZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value}) 11080@defmacx CTZ_DEFINED_VALUE_AT_ZERO (@var{mode}, @var{value}) 11081A C expression that indicates whether the architecture defines a value 11082for @code{clz} or @code{ctz} with a zero operand. 11083A result of @code{0} indicates the value is undefined. 11084If the value is defined for only the RTL expression, the macro should 11085evaluate to @code{1}; if the value applies also to the corresponding optab 11086entry (which is normally the case if it expands directly into 11087the corresponding RTL), then the macro should evaluate to @code{2}. 11088In the cases where the value is defined, @var{value} should be set to 11089this value. 11090 11091If this macro is not defined, the value of @code{clz} or 11092@code{ctz} at zero is assumed to be undefined. 11093 11094This macro must be defined if the target's expansion for @code{ffs} 11095relies on a particular value to get correct results. Otherwise it 11096is not necessary, though it may be used to optimize some corner cases, and 11097to provide a default expansion for the @code{ffs} optab. 11098 11099Note that regardless of this macro the ``definedness'' of @code{clz} 11100and @code{ctz} at zero do @emph{not} extend to the builtin functions 11101visible to the user. Thus one may be free to adjust the value at will 11102to match the target expansion of these operations without fear of 11103breaking the API@. 11104@end defmac 11105 11106@defmac Pmode 11107An alias for the machine mode for pointers. On most machines, define 11108this to be the integer mode corresponding to the width of a hardware 11109pointer; @code{SImode} on 32-bit machine or @code{DImode} on 64-bit machines. 11110On some machines you must define this to be one of the partial integer 11111modes, such as @code{PSImode}. 11112 11113The width of @code{Pmode} must be at least as large as the value of 11114@code{POINTER_SIZE}. If it is not equal, you must define the macro 11115@code{POINTERS_EXTEND_UNSIGNED} to specify how pointers are extended 11116to @code{Pmode}. 11117@end defmac 11118 11119@defmac FUNCTION_MODE 11120An alias for the machine mode used for memory references to functions 11121being called, in @code{call} RTL expressions. On most CISC machines, 11122where an instruction can begin at any byte address, this should be 11123@code{QImode}. On most RISC machines, where all instructions have fixed 11124size and alignment, this should be a mode with the same size and alignment 11125as the machine instruction words - typically @code{SImode} or @code{HImode}. 11126@end defmac 11127 11128@defmac STDC_0_IN_SYSTEM_HEADERS 11129In normal operation, the preprocessor expands @code{__STDC__} to the 11130constant 1, to signify that GCC conforms to ISO Standard C@. On some 11131hosts, like Solaris, the system compiler uses a different convention, 11132where @code{__STDC__} is normally 0, but is 1 if the user specifies 11133strict conformance to the C Standard. 11134 11135Defining @code{STDC_0_IN_SYSTEM_HEADERS} makes GNU CPP follows the host 11136convention when processing system header files, but when processing user 11137files @code{__STDC__} will always expand to 1. 11138@end defmac 11139 11140@deftypefn {C Target Hook} {const char *} TARGET_C_PREINCLUDE (void) 11141Define this hook to return the name of a header file to be included at the start of all compilations, as if it had been included with @code{#include <@var{file}>}. If this hook returns @code{NULL}, or is not defined, or the header is not found, or if the user specifies @option{-ffreestanding} or @option{-nostdinc}, no header is included. 11142 11143 This hook can be used together with a header provided by the system C library to implement ISO C requirements for certain macros to be predefined that describe properties of the whole implementation rather than just the compiler. 11144@end deftypefn 11145 11146@deftypefn {C Target Hook} bool TARGET_CXX_IMPLICIT_EXTERN_C (const char*@var{}) 11147Define this hook to add target-specific C++ implicit extern C functions. If this function returns true for the name of a file-scope function, that function implicitly gets extern "C" linkage rather than whatever language linkage the declaration would normally have. An example of such function is WinMain on Win32 targets. 11148@end deftypefn 11149 11150@defmac NO_IMPLICIT_EXTERN_C 11151Define this macro if the system header files support C++ as well as C@. 11152This macro inhibits the usual method of using system header files in 11153C++, which is to pretend that the file's contents are enclosed in 11154@samp{extern "C" @{@dots{}@}}. 11155@end defmac 11156 11157@findex #pragma 11158@findex pragma 11159@defmac REGISTER_TARGET_PRAGMAS () 11160Define this macro if you want to implement any target-specific pragmas. 11161If defined, it is a C expression which makes a series of calls to 11162@code{c_register_pragma} or @code{c_register_pragma_with_expansion} 11163for each pragma. The macro may also do any 11164setup required for the pragmas. 11165 11166The primary reason to define this macro is to provide compatibility with 11167other compilers for the same target. In general, we discourage 11168definition of target-specific pragmas for GCC@. 11169 11170If the pragma can be implemented by attributes then you should consider 11171defining the target hook @samp{TARGET_INSERT_ATTRIBUTES} as well. 11172 11173Preprocessor macros that appear on pragma lines are not expanded. All 11174@samp{#pragma} directives that do not match any registered pragma are 11175silently ignored, unless the user specifies @option{-Wunknown-pragmas}. 11176@end defmac 11177 11178@deftypefun void c_register_pragma (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *)) 11179@deftypefunx void c_register_pragma_with_expansion (const char *@var{space}, const char *@var{name}, void (*@var{callback}) (struct cpp_reader *)) 11180 11181Each call to @code{c_register_pragma} or 11182@code{c_register_pragma_with_expansion} establishes one pragma. The 11183@var{callback} routine will be called when the preprocessor encounters a 11184pragma of the form 11185 11186@smallexample 11187#pragma [@var{space}] @var{name} @dots{} 11188@end smallexample 11189 11190@var{space} is the case-sensitive namespace of the pragma, or 11191@code{NULL} to put the pragma in the global namespace. The callback 11192routine receives @var{pfile} as its first argument, which can be passed 11193on to cpplib's functions if necessary. You can lex tokens after the 11194@var{name} by calling @code{pragma_lex}. Tokens that are not read by the 11195callback will be silently ignored. The end of the line is indicated by 11196a token of type @code{CPP_EOF}. Macro expansion occurs on the 11197arguments of pragmas registered with 11198@code{c_register_pragma_with_expansion} but not on the arguments of 11199pragmas registered with @code{c_register_pragma}. 11200 11201Note that the use of @code{pragma_lex} is specific to the C and C++ 11202compilers. It will not work in the Java or Fortran compilers, or any 11203other language compilers for that matter. Thus if @code{pragma_lex} is going 11204to be called from target-specific code, it must only be done so when 11205building the C and C++ compilers. This can be done by defining the 11206variables @code{c_target_objs} and @code{cxx_target_objs} in the 11207target entry in the @file{config.gcc} file. These variables should name 11208the target-specific, language-specific object file which contains the 11209code that uses @code{pragma_lex}. Note it will also be necessary to add a 11210rule to the makefile fragment pointed to by @code{tmake_file} that shows 11211how to build this object file. 11212@end deftypefun 11213 11214@defmac HANDLE_PRAGMA_PACK_WITH_EXPANSION 11215Define this macro if macros should be expanded in the 11216arguments of @samp{#pragma pack}. 11217@end defmac 11218 11219@defmac TARGET_DEFAULT_PACK_STRUCT 11220If your target requires a structure packing default other than 0 (meaning 11221the machine default), define this macro to the necessary value (in bytes). 11222This must be a value that would also be valid to use with 11223@samp{#pragma pack()} (that is, a small power of two). 11224@end defmac 11225 11226@defmac DOLLARS_IN_IDENTIFIERS 11227Define this macro to control use of the character @samp{$} in 11228identifier names for the C family of languages. 0 means @samp{$} is 11229not allowed by default; 1 means it is allowed. 1 is the default; 11230there is no need to define this macro in that case. 11231@end defmac 11232 11233@defmac INSN_SETS_ARE_DELAYED (@var{insn}) 11234Define this macro as a C expression that is nonzero if it is safe for the 11235delay slot scheduler to place instructions in the delay slot of @var{insn}, 11236even if they appear to use a resource set or clobbered in @var{insn}. 11237@var{insn} is always a @code{jump_insn} or an @code{insn}; GCC knows that 11238every @code{call_insn} has this behavior. On machines where some @code{insn} 11239or @code{jump_insn} is really a function call and hence has this behavior, 11240you should define this macro. 11241 11242You need not define this macro if it would always return zero. 11243@end defmac 11244 11245@defmac INSN_REFERENCES_ARE_DELAYED (@var{insn}) 11246Define this macro as a C expression that is nonzero if it is safe for the 11247delay slot scheduler to place instructions in the delay slot of @var{insn}, 11248even if they appear to set or clobber a resource referenced in @var{insn}. 11249@var{insn} is always a @code{jump_insn} or an @code{insn}. On machines where 11250some @code{insn} or @code{jump_insn} is really a function call and its operands 11251are registers whose use is actually in the subroutine it calls, you should 11252define this macro. Doing so allows the delay slot scheduler to move 11253instructions which copy arguments into the argument registers into the delay 11254slot of @var{insn}. 11255 11256You need not define this macro if it would always return zero. 11257@end defmac 11258 11259@defmac MULTIPLE_SYMBOL_SPACES 11260Define this macro as a C expression that is nonzero if, in some cases, 11261global symbols from one translation unit may not be bound to undefined 11262symbols in another translation unit without user intervention. For 11263instance, under Microsoft Windows symbols must be explicitly imported 11264from shared libraries (DLLs). 11265 11266You need not define this macro if it would always evaluate to zero. 11267@end defmac 11268 11269@deftypefn {Target Hook} {rtx_insn *} TARGET_MD_ASM_ADJUST (vec<rtx>& @var{outputs}, vec<rtx>& @var{inputs}, vec<const char *>& @var{constraints}, vec<rtx>& @var{clobbers}, HARD_REG_SET& @var{clobbered_regs}) 11270This target hook may add @dfn{clobbers} to @var{clobbers} and 11271@var{clobbered_regs} for any hard regs the port wishes to automatically 11272clobber for an asm. The @var{outputs} and @var{inputs} may be inspected 11273to avoid clobbering a register that is already used by the asm. 11274 11275It may modify the @var{outputs}, @var{inputs}, and @var{constraints} 11276as necessary for other pre-processing. In this case the return value is 11277a sequence of insns to emit after the asm. 11278@end deftypefn 11279 11280@defmac MATH_LIBRARY 11281Define this macro as a C string constant for the linker argument to link 11282in the system math library, minus the initial @samp{"-l"}, or 11283@samp{""} if the target does not have a 11284separate math library. 11285 11286You need only define this macro if the default of @samp{"m"} is wrong. 11287@end defmac 11288 11289@defmac LIBRARY_PATH_ENV 11290Define this macro as a C string constant for the environment variable that 11291specifies where the linker should look for libraries. 11292 11293You need only define this macro if the default of @samp{"LIBRARY_PATH"} 11294is wrong. 11295@end defmac 11296 11297@defmac TARGET_POSIX_IO 11298Define this macro if the target supports the following POSIX@ file 11299functions, access, mkdir and file locking with fcntl / F_SETLKW@. 11300Defining @code{TARGET_POSIX_IO} will enable the test coverage code 11301to use file locking when exiting a program, which avoids race conditions 11302if the program has forked. It will also create directories at run-time 11303for cross-profiling. 11304@end defmac 11305 11306@defmac MAX_CONDITIONAL_EXECUTE 11307 11308A C expression for the maximum number of instructions to execute via 11309conditional execution instructions instead of a branch. A value of 11310@code{BRANCH_COST}+1 is the default if the machine does not use cc0, and 113111 if it does use cc0. 11312@end defmac 11313 11314@defmac IFCVT_MODIFY_TESTS (@var{ce_info}, @var{true_expr}, @var{false_expr}) 11315Used if the target needs to perform machine-dependent modifications on the 11316conditionals used for turning basic blocks into conditionally executed code. 11317@var{ce_info} points to a data structure, @code{struct ce_if_block}, which 11318contains information about the currently processed blocks. @var{true_expr} 11319and @var{false_expr} are the tests that are used for converting the 11320then-block and the else-block, respectively. Set either @var{true_expr} or 11321@var{false_expr} to a null pointer if the tests cannot be converted. 11322@end defmac 11323 11324@defmac IFCVT_MODIFY_MULTIPLE_TESTS (@var{ce_info}, @var{bb}, @var{true_expr}, @var{false_expr}) 11325Like @code{IFCVT_MODIFY_TESTS}, but used when converting more complicated 11326if-statements into conditions combined by @code{and} and @code{or} operations. 11327@var{bb} contains the basic block that contains the test that is currently 11328being processed and about to be turned into a condition. 11329@end defmac 11330 11331@defmac IFCVT_MODIFY_INSN (@var{ce_info}, @var{pattern}, @var{insn}) 11332A C expression to modify the @var{PATTERN} of an @var{INSN} that is to 11333be converted to conditional execution format. @var{ce_info} points to 11334a data structure, @code{struct ce_if_block}, which contains information 11335about the currently processed blocks. 11336@end defmac 11337 11338@defmac IFCVT_MODIFY_FINAL (@var{ce_info}) 11339A C expression to perform any final machine dependent modifications in 11340converting code to conditional execution. The involved basic blocks 11341can be found in the @code{struct ce_if_block} structure that is pointed 11342to by @var{ce_info}. 11343@end defmac 11344 11345@defmac IFCVT_MODIFY_CANCEL (@var{ce_info}) 11346A C expression to cancel any machine dependent modifications in 11347converting code to conditional execution. The involved basic blocks 11348can be found in the @code{struct ce_if_block} structure that is pointed 11349to by @var{ce_info}. 11350@end defmac 11351 11352@defmac IFCVT_MACHDEP_INIT (@var{ce_info}) 11353A C expression to initialize any machine specific data for if-conversion 11354of the if-block in the @code{struct ce_if_block} structure that is pointed 11355to by @var{ce_info}. 11356@end defmac 11357 11358@deftypefn {Target Hook} void TARGET_MACHINE_DEPENDENT_REORG (void) 11359If non-null, this hook performs a target-specific pass over the 11360instruction stream. The compiler will run it at all optimization levels, 11361just before the point at which it normally does delayed-branch scheduling. 11362 11363The exact purpose of the hook varies from target to target. Some use 11364it to do transformations that are necessary for correctness, such as 11365laying out in-function constant pools or avoiding hardware hazards. 11366Others use it as an opportunity to do some machine-dependent optimizations. 11367 11368You need not implement the hook if it has nothing to do. The default 11369definition is null. 11370@end deftypefn 11371 11372@deftypefn {Target Hook} void TARGET_INIT_BUILTINS (void) 11373Define this hook if you have any machine-specific built-in functions 11374that need to be defined. It should be a function that performs the 11375necessary setup. 11376 11377Machine specific built-in functions can be useful to expand special machine 11378instructions that would otherwise not normally be generated because 11379they have no equivalent in the source language (for example, SIMD vector 11380instructions or prefetch instructions). 11381 11382To create a built-in function, call the function 11383@code{lang_hooks.builtin_function} 11384which is defined by the language front end. You can use any type nodes set 11385up by @code{build_common_tree_nodes}; 11386only language front ends that use those two functions will call 11387@samp{TARGET_INIT_BUILTINS}. 11388@end deftypefn 11389 11390@deftypefn {Target Hook} tree TARGET_BUILTIN_DECL (unsigned @var{code}, bool @var{initialize_p}) 11391Define this hook if you have any machine-specific built-in functions 11392that need to be defined. It should be a function that returns the 11393builtin function declaration for the builtin function code @var{code}. 11394If there is no such builtin and it cannot be initialized at this time 11395if @var{initialize_p} is true the function should return @code{NULL_TREE}. 11396If @var{code} is out of range the function should return 11397@code{error_mark_node}. 11398@end deftypefn 11399 11400@deftypefn {Target Hook} rtx TARGET_EXPAND_BUILTIN (tree @var{exp}, rtx @var{target}, rtx @var{subtarget}, machine_mode @var{mode}, int @var{ignore}) 11401 11402Expand a call to a machine specific built-in function that was set up by 11403@samp{TARGET_INIT_BUILTINS}. @var{exp} is the expression for the 11404function call; the result should go to @var{target} if that is 11405convenient, and have mode @var{mode} if that is convenient. 11406@var{subtarget} may be used as the target for computing one of 11407@var{exp}'s operands. @var{ignore} is nonzero if the value is to be 11408ignored. This function should return the result of the call to the 11409built-in function. 11410@end deftypefn 11411 11412@deftypefn {Target Hook} tree TARGET_BUILTIN_CHKP_FUNCTION (unsigned @var{fcode}) 11413This hook allows target to redefine built-in functions used by 11414Pointer Bounds Checker for code instrumentation. Hook should return 11415fndecl of function implementing generic builtin whose code is 11416passed in @var{fcode}. Currently following built-in functions are 11417obtained using this hook: 11418@deftypefn {Built-in Function} __bounds_type __chkp_bndmk (const void *@var{lb}, size_t @var{size}) 11419Function code - BUILT_IN_CHKP_BNDMK. This built-in function is used 11420by Pointer Bounds Checker to create bound values. @var{lb} holds low 11421bound of the resulting bounds. @var{size} holds size of created bounds. 11422@end deftypefn 11423 11424@deftypefn {Built-in Function} void __chkp_bndstx (const void *@var{ptr}, __bounds_type @var{b}, const void **@var{loc}) 11425Function code - @code{BUILT_IN_CHKP_BNDSTX}. This built-in function is used 11426by Pointer Bounds Checker to store bounds @var{b} for pointer @var{ptr} 11427when @var{ptr} is stored by address @var{loc}. 11428@end deftypefn 11429 11430@deftypefn {Built-in Function} __bounds_type __chkp_bndldx (const void **@var{loc}, const void *@var{ptr}) 11431Function code - @code{BUILT_IN_CHKP_BNDLDX}. This built-in function is used 11432by Pointer Bounds Checker to get bounds of pointer @var{ptr} loaded by 11433address @var{loc}. 11434@end deftypefn 11435 11436@deftypefn {Built-in Function} void __chkp_bndcl (const void *@var{ptr}, __bounds_type @var{b}) 11437Function code - @code{BUILT_IN_CHKP_BNDCL}. This built-in function is used 11438by Pointer Bounds Checker to perform check for pointer @var{ptr} against 11439lower bound of bounds @var{b}. 11440@end deftypefn 11441 11442@deftypefn {Built-in Function} void __chkp_bndcu (const void *@var{ptr}, __bounds_type @var{b}) 11443Function code - @code{BUILT_IN_CHKP_BNDCU}. This built-in function is used 11444by Pointer Bounds Checker to perform check for pointer @var{ptr} against 11445upper bound of bounds @var{b}. 11446@end deftypefn 11447 11448@deftypefn {Built-in Function} __bounds_type __chkp_bndret (void *@var{ptr}) 11449Function code - @code{BUILT_IN_CHKP_BNDRET}. This built-in function is used 11450by Pointer Bounds Checker to obtain bounds returned by a call statement. 11451@var{ptr} passed to built-in is @code{SSA_NAME} returned by the call. 11452@end deftypefn 11453 11454@deftypefn {Built-in Function} __bounds_type __chkp_intersect (__bounds_type @var{b1}, __bounds_type @var{b2}) 11455Function code - @code{BUILT_IN_CHKP_INTERSECT}. This built-in function 11456returns intersection of bounds @var{b1} and @var{b2}. 11457@end deftypefn 11458 11459@deftypefn {Built-in Function} __bounds_type __chkp_narrow (const void *@var{ptr}, __bounds_type @var{b}, size_t @var{s}) 11460Function code - @code{BUILT_IN_CHKP_NARROW}. This built-in function 11461returns intersection of bounds @var{b} and 11462[@var{ptr}, @var{ptr} + @var{s} - @code{1}]. 11463@end deftypefn 11464 11465@deftypefn {Built-in Function} size_t __chkp_sizeof (const void *@var{ptr}) 11466Function code - @code{BUILT_IN_CHKP_SIZEOF}. This built-in function 11467returns size of object referenced by @var{ptr}. @var{ptr} is always 11468@code{ADDR_EXPR} of @code{VAR_DECL}. This built-in is used by 11469Pointer Bounds Checker when bounds of object cannot be computed statically 11470(e.g. object has incomplete type). 11471@end deftypefn 11472 11473@deftypefn {Built-in Function} const void *__chkp_extract_lower (__bounds_type @var{b}) 11474Function code - @code{BUILT_IN_CHKP_EXTRACT_LOWER}. This built-in function 11475returns lower bound of bounds @var{b}. 11476@end deftypefn 11477 11478@deftypefn {Built-in Function} const void *__chkp_extract_upper (__bounds_type @var{b}) 11479Function code - @code{BUILT_IN_CHKP_EXTRACT_UPPER}. This built-in function 11480returns upper bound of bounds @var{b}. 11481@end deftypefn 11482@end deftypefn 11483@deftypefn {Target Hook} tree TARGET_CHKP_BOUND_TYPE (void) 11484Return type to be used for bounds 11485@end deftypefn 11486@deftypefn {Target Hook} machine_mode TARGET_CHKP_BOUND_MODE (void) 11487Return mode to be used for bounds. 11488@end deftypefn 11489@deftypefn {Target Hook} tree TARGET_CHKP_MAKE_BOUNDS_CONSTANT (HOST_WIDE_INT @var{lb}, HOST_WIDE_INT @var{ub}) 11490Return constant used to statically initialize constant bounds 11491with specified lower bound @var{lb} and upper bounds @var{ub}. 11492@end deftypefn 11493@deftypefn {Target Hook} int TARGET_CHKP_INITIALIZE_BOUNDS (tree @var{var}, tree @var{lb}, tree @var{ub}, tree *@var{stmts}) 11494Generate a list of statements @var{stmts} to initialize pointer 11495bounds variable @var{var} with bounds @var{lb} and @var{ub}. Return 11496the number of generated statements. 11497@end deftypefn 11498 11499@deftypefn {Target Hook} tree TARGET_RESOLVE_OVERLOADED_BUILTIN (unsigned int @var{loc}, tree @var{fndecl}, void *@var{arglist}) 11500Select a replacement for a machine specific built-in function that 11501was set up by @samp{TARGET_INIT_BUILTINS}. This is done 11502@emph{before} regular type checking, and so allows the target to 11503implement a crude form of function overloading. @var{fndecl} is the 11504declaration of the built-in function. @var{arglist} is the list of 11505arguments passed to the built-in function. The result is a 11506complete expression that implements the operation, usually 11507another @code{CALL_EXPR}. 11508@var{arglist} really has type @samp{VEC(tree,gc)*} 11509@end deftypefn 11510 11511@deftypefn {Target Hook} tree TARGET_FOLD_BUILTIN (tree @var{fndecl}, int @var{n_args}, tree *@var{argp}, bool @var{ignore}) 11512Fold a call to a machine specific built-in function that was set up by 11513@samp{TARGET_INIT_BUILTINS}. @var{fndecl} is the declaration of the 11514built-in function. @var{n_args} is the number of arguments passed to 11515the function; the arguments themselves are pointed to by @var{argp}. 11516The result is another tree, valid for both GIMPLE and GENERIC, 11517containing a simplified expression for the call's result. If 11518@var{ignore} is true the value will be ignored. 11519@end deftypefn 11520 11521@deftypefn {Target Hook} bool TARGET_GIMPLE_FOLD_BUILTIN (gimple_stmt_iterator *@var{gsi}) 11522Fold a call to a machine specific built-in function that was set up 11523by @samp{TARGET_INIT_BUILTINS}. @var{gsi} points to the gimple 11524statement holding the function call. Returns true if any change 11525was made to the GIMPLE stream. 11526@end deftypefn 11527 11528@deftypefn {Target Hook} int TARGET_COMPARE_VERSION_PRIORITY (tree @var{decl1}, tree @var{decl2}) 11529This hook is used to compare the target attributes in two functions to 11530determine which function's features get higher priority. This is used 11531during function multi-versioning to figure out the order in which two 11532versions must be dispatched. A function version with a higher priority 11533is checked for dispatching earlier. @var{decl1} and @var{decl2} are 11534 the two function decls that will be compared. 11535@end deftypefn 11536 11537@deftypefn {Target Hook} tree TARGET_GET_FUNCTION_VERSIONS_DISPATCHER (void *@var{decl}) 11538This hook is used to get the dispatcher function for a set of function 11539versions. The dispatcher function is called to invoke the right function 11540version at run-time. @var{decl} is one version from a set of semantically 11541identical versions. 11542@end deftypefn 11543 11544@deftypefn {Target Hook} tree TARGET_GENERATE_VERSION_DISPATCHER_BODY (void *@var{arg}) 11545This hook is used to generate the dispatcher logic to invoke the right 11546function version at run-time for a given set of function versions. 11547@var{arg} points to the callgraph node of the dispatcher function whose 11548body must be generated. 11549@end deftypefn 11550 11551@deftypefn {Target Hook} bool TARGET_CAN_USE_DOLOOP_P (const widest_int @var{&iterations}, const widest_int @var{&iterations_max}, unsigned int @var{loop_depth}, bool @var{entered_at_top}) 11552Return true if it is possible to use low-overhead loops (@code{doloop_end} 11553and @code{doloop_begin}) for a particular loop. @var{iterations} gives the 11554exact number of iterations, or 0 if not known. @var{iterations_max} gives 11555the maximum number of iterations, or 0 if not known. @var{loop_depth} is 11556the nesting depth of the loop, with 1 for innermost loops, 2 for loops that 11557contain innermost loops, and so on. @var{entered_at_top} is true if the 11558loop is only entered from the top. 11559 11560This hook is only used if @code{doloop_end} is available. The default 11561implementation returns true. You can use @code{can_use_doloop_if_innermost} 11562if the loop must be the innermost, and if there are no other restrictions. 11563@end deftypefn 11564 11565@deftypefn {Target Hook} {const char *} TARGET_INVALID_WITHIN_DOLOOP (const rtx_insn *@var{insn}) 11566 11567Take an instruction in @var{insn} and return NULL if it is valid within a 11568low-overhead loop, otherwise return a string explaining why doloop 11569could not be applied. 11570 11571Many targets use special registers for low-overhead looping. For any 11572instruction that clobbers these this function should return a string indicating 11573the reason why the doloop could not be applied. 11574By default, the RTL loop optimizer does not use a present doloop pattern for 11575loops containing function calls or branch on table instructions. 11576@end deftypefn 11577 11578@deftypefn {Target Hook} bool TARGET_LEGITIMATE_COMBINED_INSN (rtx_insn *@var{insn}) 11579Take an instruction in @var{insn} and return @code{false} if the instruction is not appropriate as a combination of two or more instructions. The default is to accept all instructions. 11580@end deftypefn 11581 11582@deftypefn {Target Hook} bool TARGET_CAN_FOLLOW_JUMP (const rtx_insn *@var{follower}, const rtx_insn *@var{followee}) 11583FOLLOWER and FOLLOWEE are JUMP_INSN instructions; return true if FOLLOWER may be modified to follow FOLLOWEE; false, if it can't. For example, on some targets, certain kinds of branches can't be made to follow through a hot/cold partitioning. 11584@end deftypefn 11585 11586@deftypefn {Target Hook} bool TARGET_COMMUTATIVE_P (const_rtx @var{x}, int @var{outer_code}) 11587This target hook returns @code{true} if @var{x} is considered to be commutative. 11588Usually, this is just COMMUTATIVE_P (@var{x}), but the HP PA doesn't consider 11589PLUS to be commutative inside a MEM@. @var{outer_code} is the rtx code 11590of the enclosing rtl, if known, otherwise it is UNKNOWN. 11591@end deftypefn 11592 11593@deftypefn {Target Hook} rtx TARGET_ALLOCATE_INITIAL_VALUE (rtx @var{hard_reg}) 11594 11595When the initial value of a hard register has been copied in a pseudo 11596register, it is often not necessary to actually allocate another register 11597to this pseudo register, because the original hard register or a stack slot 11598it has been saved into can be used. @code{TARGET_ALLOCATE_INITIAL_VALUE} 11599is called at the start of register allocation once for each hard register 11600that had its initial value copied by using 11601@code{get_func_hard_reg_initial_val} or @code{get_hard_reg_initial_val}. 11602Possible values are @code{NULL_RTX}, if you don't want 11603to do any special allocation, a @code{REG} rtx---that would typically be 11604the hard register itself, if it is known not to be clobbered---or a 11605@code{MEM}. 11606If you are returning a @code{MEM}, this is only a hint for the allocator; 11607it might decide to use another register anyways. 11608You may use @code{current_function_is_leaf} or 11609@code{REG_N_SETS} in the hook to determine if the hard 11610register in question will not be clobbered. 11611The default value of this hook is @code{NULL}, which disables any special 11612allocation. 11613@end deftypefn 11614 11615@deftypefn {Target Hook} int TARGET_UNSPEC_MAY_TRAP_P (const_rtx @var{x}, unsigned @var{flags}) 11616This target hook returns nonzero if @var{x}, an @code{unspec} or 11617@code{unspec_volatile} operation, might cause a trap. Targets can use 11618this hook to enhance precision of analysis for @code{unspec} and 11619@code{unspec_volatile} operations. You may call @code{may_trap_p_1} 11620to analyze inner elements of @var{x} in which case @var{flags} should be 11621passed along. 11622@end deftypefn 11623 11624@deftypefn {Target Hook} void TARGET_SET_CURRENT_FUNCTION (tree @var{decl}) 11625The compiler invokes this hook whenever it changes its current function 11626context (@code{cfun}). You can define this function if 11627the back end needs to perform any initialization or reset actions on a 11628per-function basis. For example, it may be used to implement function 11629attributes that affect register usage or code generation patterns. 11630The argument @var{decl} is the declaration for the new function context, 11631and may be null to indicate that the compiler has left a function context 11632and is returning to processing at the top level. 11633The default hook function does nothing. 11634 11635GCC sets @code{cfun} to a dummy function context during initialization of 11636some parts of the back end. The hook function is not invoked in this 11637situation; you need not worry about the hook being invoked recursively, 11638or when the back end is in a partially-initialized state. 11639@code{cfun} might be @code{NULL} to indicate processing at top level, 11640outside of any function scope. 11641@end deftypefn 11642 11643@defmac TARGET_OBJECT_SUFFIX 11644Define this macro to be a C string representing the suffix for object 11645files on your target machine. If you do not define this macro, GCC will 11646use @samp{.o} as the suffix for object files. 11647@end defmac 11648 11649@defmac TARGET_EXECUTABLE_SUFFIX 11650Define this macro to be a C string representing the suffix to be 11651automatically added to executable files on your target machine. If you 11652do not define this macro, GCC will use the null string as the suffix for 11653executable files. 11654@end defmac 11655 11656@defmac COLLECT_EXPORT_LIST 11657If defined, @code{collect2} will scan the individual object files 11658specified on its command line and create an export list for the linker. 11659Define this macro for systems like AIX, where the linker discards 11660object files that are not referenced from @code{main} and uses export 11661lists. 11662@end defmac 11663 11664@defmac MODIFY_JNI_METHOD_CALL (@var{mdecl}) 11665Define this macro to a C expression representing a variant of the 11666method call @var{mdecl}, if Java Native Interface (JNI) methods 11667must be invoked differently from other methods on your target. 11668For example, on 32-bit Microsoft Windows, JNI methods must be invoked using 11669the @code{stdcall} calling convention and this macro is then 11670defined as this expression: 11671 11672@smallexample 11673build_type_attribute_variant (@var{mdecl}, 11674 build_tree_list 11675 (get_identifier ("stdcall"), 11676 NULL)) 11677@end smallexample 11678@end defmac 11679 11680@deftypefn {Target Hook} bool TARGET_CANNOT_MODIFY_JUMPS_P (void) 11681This target hook returns @code{true} past the point in which new jump 11682instructions could be created. On machines that require a register for 11683every jump such as the SHmedia ISA of SH5, this point would typically be 11684reload, so this target hook should be defined to a function such as: 11685 11686@smallexample 11687static bool 11688cannot_modify_jumps_past_reload_p () 11689@{ 11690 return (reload_completed || reload_in_progress); 11691@} 11692@end smallexample 11693@end deftypefn 11694 11695@deftypefn {Target Hook} reg_class_t TARGET_BRANCH_TARGET_REGISTER_CLASS (void) 11696This target hook returns a register class for which branch target register 11697optimizations should be applied. All registers in this class should be 11698usable interchangeably. After reload, registers in this class will be 11699re-allocated and loads will be hoisted out of loops and be subjected 11700to inter-block scheduling. 11701@end deftypefn 11702 11703@deftypefn {Target Hook} bool TARGET_BRANCH_TARGET_REGISTER_CALLEE_SAVED (bool @var{after_prologue_epilogue_gen}) 11704Branch target register optimization will by default exclude callee-saved 11705registers 11706that are not already live during the current function; if this target hook 11707returns true, they will be included. The target code must than make sure 11708that all target registers in the class returned by 11709@samp{TARGET_BRANCH_TARGET_REGISTER_CLASS} that might need saving are 11710saved. @var{after_prologue_epilogue_gen} indicates if prologues and 11711epilogues have already been generated. Note, even if you only return 11712true when @var{after_prologue_epilogue_gen} is false, you still are likely 11713to have to make special provisions in @code{INITIAL_ELIMINATION_OFFSET} 11714to reserve space for caller-saved target registers. 11715@end deftypefn 11716 11717@deftypefn {Target Hook} bool TARGET_HAVE_CONDITIONAL_EXECUTION (void) 11718This target hook returns true if the target supports conditional execution. 11719This target hook is required only when the target has several different 11720modes and they have different conditional execution capability, such as ARM. 11721@end deftypefn 11722 11723@deftypefn {Target Hook} rtx TARGET_GEN_CCMP_FIRST (rtx_insn **@var{prep_seq}, rtx_insn **@var{gen_seq}, int @var{code}, tree @var{op0}, tree @var{op1}) 11724This function prepares to emit a comparison insn for the first compare in a 11725 sequence of conditional comparisions. It returns an appropriate comparison 11726 with @code{CC} for passing to @code{gen_ccmp_next} or @code{cbranch_optab}. 11727 The insns to prepare the compare are saved in @var{prep_seq} and the compare 11728 insns are saved in @var{gen_seq}. They will be emitted when all the 11729 compares in the the conditional comparision are generated without error. 11730 @var{code} is the @code{rtx_code} of the compare for @var{op0} and @var{op1}. 11731@end deftypefn 11732 11733@deftypefn {Target Hook} rtx TARGET_GEN_CCMP_NEXT (rtx_insn **@var{prep_seq}, rtx_insn **@var{gen_seq}, rtx @var{prev}, int @var{cmp_code}, tree @var{op0}, tree @var{op1}, int @var{bit_code}) 11734This function prepares to emit a conditional comparison within a sequence 11735 of conditional comparisons. It returns an appropriate comparison with 11736 @code{CC} for passing to @code{gen_ccmp_next} or @code{cbranch_optab}. 11737 The insns to prepare the compare are saved in @var{prep_seq} and the compare 11738 insns are saved in @var{gen_seq}. They will be emitted when all the 11739 compares in the conditional comparision are generated without error. The 11740 @var{prev} expression is the result of a prior call to @code{gen_ccmp_first} 11741 or @code{gen_ccmp_next}. It may return @code{NULL} if the combination of 11742 @var{prev} and this comparison is not supported, otherwise the result must 11743 be appropriate for passing to @code{gen_ccmp_next} or @code{cbranch_optab}. 11744 @var{code} is the @code{rtx_code} of the compare for @var{op0} and @var{op1}. 11745 @var{bit_code} is @code{AND} or @code{IOR}, which is the op on the compares. 11746@end deftypefn 11747 11748@deftypefn {Target Hook} unsigned TARGET_LOOP_UNROLL_ADJUST (unsigned @var{nunroll}, struct loop *@var{loop}) 11749This target hook returns a new value for the number of times @var{loop} 11750should be unrolled. The parameter @var{nunroll} is the number of times 11751the loop is to be unrolled. The parameter @var{loop} is a pointer to 11752the loop, which is going to be checked for unrolling. This target hook 11753is required only when the target has special constraints like maximum 11754number of memory accesses. 11755@end deftypefn 11756 11757@defmac POWI_MAX_MULTS 11758If defined, this macro is interpreted as a signed integer C expression 11759that specifies the maximum number of floating point multiplications 11760that should be emitted when expanding exponentiation by an integer 11761constant inline. When this value is defined, exponentiation requiring 11762more than this number of multiplications is implemented by calling the 11763system library's @code{pow}, @code{powf} or @code{powl} routines. 11764The default value places no upper bound on the multiplication count. 11765@end defmac 11766 11767@deftypefn Macro void TARGET_EXTRA_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc}) 11768This target hook should register any extra include files for the 11769target. The parameter @var{stdinc} indicates if normal include files 11770are present. The parameter @var{sysroot} is the system root directory. 11771The parameter @var{iprefix} is the prefix for the gcc directory. 11772@end deftypefn 11773 11774@deftypefn Macro void TARGET_EXTRA_PRE_INCLUDES (const char *@var{sysroot}, const char *@var{iprefix}, int @var{stdinc}) 11775This target hook should register any extra include files for the 11776target before any standard headers. The parameter @var{stdinc} 11777indicates if normal include files are present. The parameter 11778@var{sysroot} is the system root directory. The parameter 11779@var{iprefix} is the prefix for the gcc directory. 11780@end deftypefn 11781 11782@deftypefn Macro void TARGET_OPTF (char *@var{path}) 11783This target hook should register special include paths for the target. 11784The parameter @var{path} is the include to register. On Darwin 11785systems, this is used for Framework includes, which have semantics 11786that are different from @option{-I}. 11787@end deftypefn 11788 11789@defmac bool TARGET_USE_LOCAL_THUNK_ALIAS_P (tree @var{fndecl}) 11790This target macro returns @code{true} if it is safe to use a local alias 11791for a virtual function @var{fndecl} when constructing thunks, 11792@code{false} otherwise. By default, the macro returns @code{true} for all 11793functions, if a target supports aliases (i.e.@: defines 11794@code{ASM_OUTPUT_DEF}), @code{false} otherwise, 11795@end defmac 11796 11797@defmac TARGET_FORMAT_TYPES 11798If defined, this macro is the name of a global variable containing 11799target-specific format checking information for the @option{-Wformat} 11800option. The default is to have no target-specific format checks. 11801@end defmac 11802 11803@defmac TARGET_N_FORMAT_TYPES 11804If defined, this macro is the number of entries in 11805@code{TARGET_FORMAT_TYPES}. 11806@end defmac 11807 11808@defmac TARGET_OVERRIDES_FORMAT_ATTRIBUTES 11809If defined, this macro is the name of a global variable containing 11810target-specific format overrides for the @option{-Wformat} option. The 11811default is to have no target-specific format overrides. If defined, 11812@code{TARGET_FORMAT_TYPES} must be defined, too. 11813@end defmac 11814 11815@defmac TARGET_OVERRIDES_FORMAT_ATTRIBUTES_COUNT 11816If defined, this macro specifies the number of entries in 11817@code{TARGET_OVERRIDES_FORMAT_ATTRIBUTES}. 11818@end defmac 11819 11820@defmac TARGET_OVERRIDES_FORMAT_INIT 11821If defined, this macro specifies the optional initialization 11822routine for target specific customizations of the system printf 11823and scanf formatter settings. 11824@end defmac 11825 11826@deftypefn {Target Hook} {const char *} TARGET_INVALID_ARG_FOR_UNPROTOTYPED_FN (const_tree @var{typelist}, const_tree @var{funcdecl}, const_tree @var{val}) 11827If defined, this macro returns the diagnostic message when it is 11828illegal to pass argument @var{val} to function @var{funcdecl} 11829with prototype @var{typelist}. 11830@end deftypefn 11831 11832@deftypefn {Target Hook} {const char *} TARGET_INVALID_CONVERSION (const_tree @var{fromtype}, const_tree @var{totype}) 11833If defined, this macro returns the diagnostic message when it is 11834invalid to convert from @var{fromtype} to @var{totype}, or @code{NULL} 11835if validity should be determined by the front end. 11836@end deftypefn 11837 11838@deftypefn {Target Hook} {const char *} TARGET_INVALID_UNARY_OP (int @var{op}, const_tree @var{type}) 11839If defined, this macro returns the diagnostic message when it is 11840invalid to apply operation @var{op} (where unary plus is denoted by 11841@code{CONVERT_EXPR}) to an operand of type @var{type}, or @code{NULL} 11842if validity should be determined by the front end. 11843@end deftypefn 11844 11845@deftypefn {Target Hook} {const char *} TARGET_INVALID_BINARY_OP (int @var{op}, const_tree @var{type1}, const_tree @var{type2}) 11846If defined, this macro returns the diagnostic message when it is 11847invalid to apply operation @var{op} to operands of types @var{type1} 11848and @var{type2}, or @code{NULL} if validity should be determined by 11849the front end. 11850@end deftypefn 11851 11852@deftypefn {Target Hook} tree TARGET_PROMOTED_TYPE (const_tree @var{type}) 11853If defined, this target hook returns the type to which values of 11854@var{type} should be promoted when they appear in expressions, 11855analogous to the integer promotions, or @code{NULL_TREE} to use the 11856front end's normal promotion rules. This hook is useful when there are 11857target-specific types with special promotion rules. 11858This is currently used only by the C and C++ front ends. 11859@end deftypefn 11860 11861@deftypefn {Target Hook} tree TARGET_CONVERT_TO_TYPE (tree @var{type}, tree @var{expr}) 11862If defined, this hook returns the result of converting @var{expr} to 11863@var{type}. It should return the converted expression, 11864or @code{NULL_TREE} to apply the front end's normal conversion rules. 11865This hook is useful when there are target-specific types with special 11866conversion rules. 11867This is currently used only by the C and C++ front ends. 11868@end deftypefn 11869 11870@defmac OBJC_JBLEN 11871This macro determines the size of the objective C jump buffer for the 11872NeXT runtime. By default, OBJC_JBLEN is defined to an innocuous value. 11873@end defmac 11874 11875@defmac LIBGCC2_UNWIND_ATTRIBUTE 11876Define this macro if any target-specific attributes need to be attached 11877to the functions in @file{libgcc} that provide low-level support for 11878call stack unwinding. It is used in declarations in @file{unwind-generic.h} 11879and the associated definitions of those functions. 11880@end defmac 11881 11882@deftypefn {Target Hook} void TARGET_UPDATE_STACK_BOUNDARY (void) 11883Define this macro to update the current function stack boundary if 11884necessary. 11885@end deftypefn 11886 11887@deftypefn {Target Hook} rtx TARGET_GET_DRAP_RTX (void) 11888This hook should return an rtx for Dynamic Realign Argument Pointer (DRAP) if a 11889different argument pointer register is needed to access the function's 11890argument list due to stack realignment. Return @code{NULL} if no DRAP 11891is needed. 11892@end deftypefn 11893 11894@deftypefn {Target Hook} bool TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS (void) 11895When optimization is disabled, this hook indicates whether or not 11896arguments should be allocated to stack slots. Normally, GCC allocates 11897stacks slots for arguments when not optimizing in order to make 11898debugging easier. However, when a function is declared with 11899@code{__attribute__((naked))}, there is no stack frame, and the compiler 11900cannot safely move arguments from the registers in which they are passed 11901to the stack. Therefore, this hook should return true in general, but 11902false for naked functions. The default implementation always returns true. 11903@end deftypefn 11904 11905@deftypevr {Target Hook} {unsigned HOST_WIDE_INT} TARGET_CONST_ANCHOR 11906On some architectures it can take multiple instructions to synthesize 11907a constant. If there is another constant already in a register that 11908is close enough in value then it is preferable that the new constant 11909is computed from this register using immediate addition or 11910subtraction. We accomplish this through CSE. Besides the value of 11911the constant we also add a lower and an upper constant anchor to the 11912available expressions. These are then queried when encountering new 11913constants. The anchors are computed by rounding the constant up and 11914down to a multiple of the value of @code{TARGET_CONST_ANCHOR}. 11915@code{TARGET_CONST_ANCHOR} should be the maximum positive value 11916accepted by immediate-add plus one. We currently assume that the 11917value of @code{TARGET_CONST_ANCHOR} is a power of 2. For example, on 11918MIPS, where add-immediate takes a 16-bit signed value, 11919@code{TARGET_CONST_ANCHOR} is set to @samp{0x8000}. The default value 11920is zero, which disables this optimization. 11921@end deftypevr 11922 11923@deftypefn {Target Hook} {unsigned HOST_WIDE_INT} TARGET_ASAN_SHADOW_OFFSET (void) 11924Return the offset bitwise ored into shifted address to get corresponding 11925Address Sanitizer shadow memory address. NULL if Address Sanitizer is not 11926supported by the target. 11927@end deftypefn 11928 11929@deftypefn {Target Hook} {unsigned HOST_WIDE_INT} TARGET_MEMMODEL_CHECK (unsigned HOST_WIDE_INT @var{val}) 11930Validate target specific memory model mask bits. When NULL no target specific 11931memory model bits are allowed. 11932@end deftypefn 11933 11934@deftypevr {Target Hook} {unsigned char} TARGET_ATOMIC_TEST_AND_SET_TRUEVAL 11935This value should be set if the result written by @code{atomic_test_and_set} is not exactly 1, i.e. the @code{bool} @code{true}. 11936@end deftypevr 11937 11938@deftypefn {Target Hook} bool TARGET_HAS_IFUNC_P (void) 11939It returns true if the target supports GNU indirect functions. 11940The support includes the assembler, linker and dynamic linker. 11941The default value of this hook is based on target's libc. 11942@end deftypefn 11943 11944@deftypefn {Target Hook} {unsigned int} TARGET_ATOMIC_ALIGN_FOR_MODE (machine_mode @var{mode}) 11945If defined, this function returns an appropriate alignment in bits for an atomic object of machine_mode @var{mode}. If 0 is returned then the default alignment for the specified mode is used. 11946@end deftypefn 11947 11948@deftypefn {Target Hook} void TARGET_ATOMIC_ASSIGN_EXPAND_FENV (tree *@var{hold}, tree *@var{clear}, tree *@var{update}) 11949ISO C11 requires atomic compound assignments that may raise floating-point exceptions to raise exceptions corresponding to the arithmetic operation whose result was successfully stored in a compare-and-exchange sequence. This requires code equivalent to calls to @code{feholdexcept}, @code{feclearexcept} and @code{feupdateenv} to be generated at appropriate points in the compare-and-exchange sequence. This hook should set @code{*@var{hold}} to an expression equivalent to the call to @code{feholdexcept}, @code{*@var{clear}} to an expression equivalent to the call to @code{feclearexcept} and @code{*@var{update}} to an expression equivalent to the call to @code{feupdateenv}. The three expressions are @code{NULL_TREE} on entry to the hook and may be left as @code{NULL_TREE} if no code is required in a particular place. The default implementation leaves all three expressions as @code{NULL_TREE}. The @code{__atomic_feraiseexcept} function from @code{libatomic} may be of use as part of the code generated in @code{*@var{update}}. 11950@end deftypefn 11951 11952@deftypefn {Target Hook} void TARGET_RECORD_OFFLOAD_SYMBOL (tree) 11953Used when offloaded functions are seen in the compilation unit and no named 11954sections are available. It is called once for each symbol that must be 11955recorded in the offload function and variable table. 11956@end deftypefn 11957 11958@deftypefn {Target Hook} {char *} TARGET_OFFLOAD_OPTIONS (void) 11959Used when writing out the list of options into an LTO file. It should 11960translate any relevant target-specific options (such as the ABI in use) 11961into one of the @option{-foffload} options that exist as a common interface 11962to express such options. It should return a string containing these options, 11963separated by spaces, which the caller will free. 11964 11965@end deftypefn 11966 11967@defmac TARGET_SUPPORTS_WIDE_INT 11968 11969On older ports, large integers are stored in @code{CONST_DOUBLE} rtl 11970objects. Newer ports define @code{TARGET_SUPPORTS_WIDE_INT} to be nonzero 11971to indicate that large integers are stored in 11972@code{CONST_WIDE_INT} rtl objects. The @code{CONST_WIDE_INT} allows 11973very large integer constants to be represented. @code{CONST_DOUBLE} 11974is limited to twice the size of the host's @code{HOST_WIDE_INT} 11975representation. 11976 11977Converting a port mostly requires looking for the places where 11978@code{CONST_DOUBLE}s are used with @code{VOIDmode} and replacing that 11979code with code that accesses @code{CONST_WIDE_INT}s. @samp{"grep -i 11980const_double"} at the port level gets you to 95% of the changes that 11981need to be made. There are a few places that require a deeper look. 11982 11983@itemize @bullet 11984@item 11985There is no equivalent to @code{hval} and @code{lval} for 11986@code{CONST_WIDE_INT}s. This would be difficult to express in the md 11987language since there are a variable number of elements. 11988 11989Most ports only check that @code{hval} is either 0 or -1 to see if the 11990value is small. As mentioned above, this will no longer be necessary 11991since small constants are always @code{CONST_INT}. Of course there 11992are still a few exceptions, the alpha's constraint used by the zap 11993instruction certainly requires careful examination by C code. 11994However, all the current code does is pass the hval and lval to C 11995code, so evolving the c code to look at the @code{CONST_WIDE_INT} is 11996not really a large change. 11997 11998@item 11999Because there is no standard template that ports use to materialize 12000constants, there is likely to be some futzing that is unique to each 12001port in this code. 12002 12003@item 12004The rtx costs may have to be adjusted to properly account for larger 12005constants that are represented as @code{CONST_WIDE_INT}. 12006@end itemize 12007 12008All and all it does not take long to convert ports that the 12009maintainer is familiar with. 12010 12011@end defmac 12012 12013@deftypefn {Target Hook} void TARGET_RUN_TARGET_SELFTESTS (void) 12014If selftests are enabled, run any selftests for this target. 12015@end deftypefn 12016