git.ipfire.org Git - thirdparty/gcc.git/blob

   1 ..
   2   Copyright 1988-2022 Free Software Foundation, Inc.
   3   This is part of the GCC manual.
   4   For copying conditions, see the copyright.rst file.
   5
   6 .. _x86-function-attributes:
   7
   8 x86 Function Attributes
   9 ^^^^^^^^^^^^^^^^^^^^^^^
  10
  11 These function attributes are supported by the x86 back end:
  12
  13 .. index:: cdecl function attribute, x86-32, functions that pop the argument stack on x86-32
  14
  15 .. option:: cdecl
  16
  17   On the x86-32 targets, the ``cdecl`` attribute causes the compiler to
  18   assume that the calling function pops off the stack space used to
  19   pass arguments.  This is
  20   useful to override the effects of the :option:`-mrtd` switch.
  21
  22 .. index:: fastcall function attribute, x86-32, functions that pop the argument stack on x86-32
  23
  24 .. x86-fn-attr:: fastcall
  25
  26   On x86-32 targets, the :x86-fn-attr:`fastcall` attribute causes the compiler to
  27   pass the first argument (if of integral type) in the register ECX and
  28   the second argument (if of integral type) in the register EDX.  Subsequent
  29   and other typed arguments are passed on the stack.  The called function
  30   pops the arguments off the stack.  If the number of arguments is variable all
  31   arguments are pushed on the stack.
  32
  33 .. index:: thiscall function attribute, x86-32, functions that pop the argument stack on x86-32
  34
  35 .. x86-fn-attr:: thiscall
  36
  37   On x86-32 targets, the :x86-fn-attr:`thiscall` attribute causes the compiler to
  38   pass the first argument (if of integral type) in the register ECX.
  39   Subsequent and other typed arguments are passed on the stack. The called
  40   function pops the arguments off the stack.
  41   If the number of arguments is variable all arguments are pushed on the
  42   stack.
  43   The :x86-fn-attr:`thiscall` attribute is intended for C++ non-static member functions.
  44   As a GCC extension, this calling convention can be used for C functions
  45   and for static member methods.
  46
  47 .. index:: ms_abi function attribute, x86, sysv_abi function attribute, x86
  48
  49 .. x86-fn-attr:: ms_abi, sysv_abi
  50
  51   On 32-bit and 64-bit x86 targets, you can use an ABI attribute
  52   to indicate which calling convention should be used for a function.  The
  53   :x86-fn-attr:`ms_abi` attribute tells the compiler to use the Microsoft ABI,
  54   while the ``sysv_abi`` attribute tells the compiler to use the System V
  55   ELF ABI, which is used on GNU/Linux and other systems.  The default is to use
  56   the Microsoft ABI when targeting Windows.  On all other systems, the default
  57   is the System V ELF ABI.
  58
  59   Note, the :x86-fn-attr:`ms_abi` attribute for Microsoft Windows 64-bit targets currently
  60   requires the :option:`-maccumulate-outgoing-args` option.
  61
  62 .. index:: callee_pop_aggregate_return function attribute, x86
  63
  64 .. x86-fn-attr:: callee_pop_aggregate_return (number)
  65
  66   On x86-32 targets, you can use this attribute to control how
  67   aggregates are returned in memory.  If the caller is responsible for
  68   popping the hidden pointer together with the rest of the arguments, specify
  69   :samp:`{number}` equal to zero.  If callee is responsible for popping the
  70   hidden pointer, specify :samp:`{number}` equal to one.
  71
  72   The default x86-32 ABI assumes that the callee pops the
  73   stack for hidden pointer.  However, on x86-32 Microsoft Windows targets,
  74   the compiler assumes that the
  75   caller pops the stack for hidden pointer.
  76
  77 .. index:: ms_hook_prologue function attribute, x86
  78
  79 .. x86-fn-attr:: ms_hook_prologue
  80
  81   On 32-bit and 64-bit x86 targets, you can use
  82   this function attribute to make GCC generate the 'hot-patching' function
  83   prologue used in Win32 API functions in Microsoft Windows XP Service Pack 2
  84   and newer.
  85
  86 .. index:: naked function attribute, x86
  87
  88 .. x86-fn-attr:: naked
  89
  90   This attribute allows the compiler to construct the
  91   requisite function declaration, while allowing the body of the
  92   function to be assembly code. The specified function will not have
  93   prologue/epilogue sequences generated by the compiler. Only basic
  94   ``asm`` statements can safely be included in naked functions
  95   (see :ref:`basic-asm`). While using extended ``asm`` or a mixture of
  96   basic ``asm`` and C code may appear to work, they cannot be
  97   depended upon to work reliably and are not supported.
  98
  99 .. index:: regparm function attribute, x86, functions that are passed arguments in registers on x86-32
 100
 101 .. x86-fn-attr:: regparm (number)
 102
 103   On x86-32 targets, the ``regparm`` attribute causes the compiler to
 104   pass arguments number one to :samp:`{number}` if they are of integral type
 105   in registers EAX, EDX, and ECX instead of on the stack.  Functions that
 106   take a variable number of arguments continue to be passed all of their
 107   arguments on the stack.
 108
 109   Beware that on some ELF systems this attribute is unsuitable for
 110   global functions in shared libraries with lazy binding (which is the
 111   default).  Lazy binding sends the first call via resolving code in
 112   the loader, which might assume EAX, EDX and ECX can be clobbered, as
 113   per the standard calling conventions.  Solaris 8 is affected by this.
 114   Systems with the GNU C Library version 2.1 or higher
 115   and FreeBSD are believed to be
 116   safe since the loaders there save EAX, EDX and ECX.  (Lazy binding can be
 117   disabled with the linker or the loader if desired, to avoid the
 118   problem.)
 119
 120 .. index:: sseregparm function attribute, x86
 121
 122 .. x86-fn-attr:: sseregparm
 123
 124   On x86-32 targets with SSE support, the :x86-fn-attr:`sseregparm` attribute
 125   causes the compiler to pass up to 3 floating-point arguments in
 126   SSE registers instead of on the stack.  Functions that take a
 127   variable number of arguments continue to pass all of their
 128   floating-point arguments on the stack.
 129
 130 .. index:: force_align_arg_pointer function attribute, x86
 131
 132 .. x86-fn-attr:: force_align_arg_pointer
 133
 134   On x86 targets, the :x86-fn-attr:`force_align_arg_pointer` attribute may be
 135   applied to individual function definitions, generating an alternate
 136   prologue and epilogue that realigns the run-time stack if necessary.
 137   This supports mixing legacy codes that run with a 4-byte aligned stack
 138   with modern codes that keep a 16-byte stack for SSE compatibility.
 139
 140 .. index:: stdcall function attribute, x86-32, functions that pop the argument stack on x86-32
 141
 142 .. x86-fn-attr:: stdcall
 143
 144   On x86-32 targets, the :x86-fn-attr:`stdcall` attribute causes the compiler to
 145   assume that the called function pops off the stack space used to
 146   pass arguments, unless it takes a variable number of arguments.
 147
 148 .. index:: no_caller_saved_registers function attribute, x86
 149
 150 .. x86-fn-attr:: no_caller_saved_registers
 151
 152   Use this attribute to indicate that the specified function has no
 153   caller-saved registers. That is, all registers are callee-saved. For
 154   example, this attribute can be used for a function called from an
 155   interrupt handler. The compiler generates proper function entry and
 156   exit sequences to save and restore any modified registers, except for
 157   the EFLAGS register.  Since GCC doesn't preserve SSE, MMX nor x87
 158   states, the GCC option :option:`-mgeneral-regs-only` should be used to
 159   compile functions with :x86-fn-attr:`no_caller_saved_registers` attribute.
 160
 161 .. index:: interrupt function attribute, x86
 162
 163 .. x86-fn-attr:: interrupt
 164
 165   Use this attribute to indicate that the specified function is an
 166   interrupt handler or an exception handler (depending on parameters passed
 167   to the function, explained further).  The compiler generates function
 168   entry and exit sequences suitable for use in an interrupt handler when
 169   this attribute is present.  The ``IRET`` instruction, instead of the
 170   ``RET`` instruction, is used to return from interrupt handlers.  All
 171   registers, except for the EFLAGS register which is restored by the
 172   ``IRET`` instruction, are preserved by the compiler.  Since GCC
 173   doesn't preserve SSE, MMX nor x87 states, the GCC option
 174   :option:`-mgeneral-regs-only` should be used to compile interrupt and
 175   exception handlers.
 176
 177   Any interruptible-without-stack-switch code must be compiled with
 178   :option:`-mno-red-zone` since interrupt handlers can and will, because
 179   of the hardware design, touch the red zone.
 180
 181   An interrupt handler must be declared with a mandatory pointer
 182   argument:
 183
 184   .. code-block:: c++
 185
 186     struct interrupt_frame;
 187
 188     __attribute__ ((interrupt))
 189     void
 190     f (struct interrupt_frame *frame)
 191     {
 192     }
 193
 194   and you must define ``struct interrupt_frame`` as described in the
 195   processor's manual.
 196
 197   Exception handlers differ from interrupt handlers because the system
 198   pushes an error code on the stack.  An exception handler declaration is
 199   similar to that for an interrupt handler, but with a different mandatory
 200   function signature.  The compiler arranges to pop the error code off the
 201   stack before the ``IRET`` instruction.
 202
 203   .. code-block:: c++
 204
 205     #ifdef __x86_64__
 206     typedef unsigned long long int uword_t;
 207     #else
 208     typedef unsigned int uword_t;
 209     #endif
 210
 211     struct interrupt_frame;
 212
 213     __attribute__ ((interrupt))
 214     void
 215     f (struct interrupt_frame *frame, uword_t error_code)
 216     {
 217       ...
 218     }
 219
 220   Exception handlers should only be used for exceptions that push an error
 221   code; you should use an interrupt handler in other cases.  The system
 222   will crash if the wrong kind of handler is used.
 223
 224 .. index:: target function attribute
 225
 226 .. x86-fn-attr:: target (options)
 227
 228   As discussed in :ref:`common-function-attributes`, this attribute
 229   allows specification of target-specific compilation options.
 230
 231   On the x86, the following options are allowed:
 232
 233   :samp:`3dnow` :samp:`no-3dnow`
 234
 235     .. index:: target("3dnow") function attribute, x86
 236
 237     Enable/disable the generation of the 3DNow! instructions.
 238
 239   :samp:`3dnowa` :samp:`no-3dnowa`
 240
 241     .. index:: target("3dnowa") function attribute, x86
 242
 243     Enable/disable the generation of the enhanced 3DNow! instructions.
 244
 245   :samp:`abm` :samp:`no-abm`
 246
 247     .. index:: target("abm") function attribute, x86
 248
 249     Enable/disable the generation of the advanced bit instructions.
 250
 251   :samp:`adx` :samp:`no-adx`
 252
 253     .. index:: target("adx") function attribute, x86
 254
 255     Enable/disable the generation of the ADX instructions.
 256
 257   :samp:`aes` :samp:`no-aes`
 258
 259     .. index:: target("aes") function attribute, x86
 260
 261     Enable/disable the generation of the AES instructions.
 262
 263   :samp:`avx` :samp:`no-avx`
 264
 265     .. index:: target("avx") function attribute, x86
 266
 267     Enable/disable the generation of the AVX instructions.
 268
 269   :samp:`avx2` :samp:`no-avx2`
 270
 271     .. index:: target("avx2") function attribute, x86
 272
 273     Enable/disable the generation of the AVX2 instructions.
 274
 275   :samp:`avx5124fmaps` :samp:`no-avx5124fmaps`
 276
 277     .. index:: target("avx5124fmaps") function attribute, x86
 278
 279     Enable/disable the generation of the AVX5124FMAPS instructions.
 280
 281   :samp:`avx5124vnniw` :samp:`no-avx5124vnniw`
 282
 283     .. index:: target("avx5124vnniw") function attribute, x86
 284
 285     Enable/disable the generation of the AVX5124VNNIW instructions.
 286
 287   :samp:`avx512bitalg` :samp:`no-avx512bitalg`
 288
 289     .. index:: target("avx512bitalg") function attribute, x86
 290
 291     Enable/disable the generation of the AVX512BITALG instructions.
 292
 293   :samp:`avx512bw` :samp:`no-avx512bw`
 294
 295     .. index:: target("avx512bw") function attribute, x86
 296
 297     Enable/disable the generation of the AVX512BW instructions.
 298
 299   :samp:`avx512cd` :samp:`no-avx512cd`
 300
 301     .. index:: target("avx512cd") function attribute, x86
 302
 303     Enable/disable the generation of the AVX512CD instructions.
 304
 305   :samp:`avx512dq` :samp:`no-avx512dq`
 306
 307     .. index:: target("avx512dq") function attribute, x86
 308
 309     Enable/disable the generation of the AVX512DQ instructions.
 310
 311   :samp:`avx512er` :samp:`no-avx512er`
 312
 313     .. index:: target("avx512er") function attribute, x86
 314
 315     Enable/disable the generation of the AVX512ER instructions.
 316
 317   :samp:`avx512f` :samp:`no-avx512f`
 318
 319     .. index:: target("avx512f") function attribute, x86
 320
 321     Enable/disable the generation of the AVX512F instructions.
 322
 323   :samp:`avx512ifma` :samp:`no-avx512ifma`
 324
 325     .. index:: target("avx512ifma") function attribute, x86
 326
 327     Enable/disable the generation of the AVX512IFMA instructions.
 328
 329   :samp:`avx512pf` :samp:`no-avx512pf`
 330
 331     .. index:: target("avx512pf") function attribute, x86
 332
 333     Enable/disable the generation of the AVX512PF instructions.
 334
 335   :samp:`avx512vbmi` :samp:`no-avx512vbmi`
 336
 337     .. index:: target("avx512vbmi") function attribute, x86
 338
 339     Enable/disable the generation of the AVX512VBMI instructions.
 340
 341   :samp:`avx512vbmi2` :samp:`no-avx512vbmi2`
 342
 343     .. index:: target("avx512vbmi2") function attribute, x86
 344
 345     Enable/disable the generation of the AVX512VBMI2 instructions.
 346
 347   :samp:`avx512vl` :samp:`no-avx512vl`
 348
 349     .. index:: target("avx512vl") function attribute, x86
 350
 351     Enable/disable the generation of the AVX512VL instructions.
 352
 353   :samp:`avx512vnni` :samp:`no-avx512vnni`
 354
 355     .. index:: target("avx512vnni") function attribute, x86
 356
 357     Enable/disable the generation of the AVX512VNNI instructions.
 358
 359   :samp:`avx512vpopcntdq` :samp:`no-avx512vpopcntdq`
 360
 361     .. index:: target("avx512vpopcntdq") function attribute, x86
 362
 363     Enable/disable the generation of the AVX512VPOPCNTDQ instructions.
 364
 365   :samp:`bmi` :samp:`no-bmi`
 366
 367     .. index:: target("bmi") function attribute, x86
 368
 369     Enable/disable the generation of the BMI instructions.
 370
 371   :samp:`bmi2` :samp:`no-bmi2`
 372
 373     .. index:: target("bmi2") function attribute, x86
 374
 375     Enable/disable the generation of the BMI2 instructions.
 376
 377   :samp:`cldemote` :samp:`no-cldemote`
 378
 379     .. index:: target("cldemote") function attribute, x86
 380
 381     Enable/disable the generation of the CLDEMOTE instructions.
 382
 383   :samp:`clflushopt` :samp:`no-clflushopt`
 384
 385     .. index:: target("clflushopt") function attribute, x86
 386
 387     Enable/disable the generation of the CLFLUSHOPT instructions.
 388
 389   :samp:`clwb` :samp:`no-clwb`
 390
 391     .. index:: target("clwb") function attribute, x86
 392
 393     Enable/disable the generation of the CLWB instructions.
 394
 395   :samp:`clzero` :samp:`no-clzero`
 396
 397     .. index:: target("clzero") function attribute, x86
 398
 399     Enable/disable the generation of the CLZERO instructions.
 400
 401   :samp:`crc32` :samp:`no-crc32`
 402
 403     .. index:: target("crc32") function attribute, x86
 404
 405     Enable/disable the generation of the CRC32 instructions.
 406
 407   :samp:`cx16` :samp:`no-cx16`
 408
 409     .. index:: target("cx16") function attribute, x86
 410
 411     Enable/disable the generation of the CMPXCHG16B instructions.
 412
 413   :samp:`default`
 414
 415     .. index:: target("default") function attribute, x86
 416
 417     See :ref:`function-multiversioning`, where it is used to specify the
 418     default function version.
 419
 420   :samp:`f16c` :samp:`no-f16c`
 421
 422     .. index:: target("f16c") function attribute, x86
 423
 424     Enable/disable the generation of the F16C instructions.
 425
 426   :samp:`fma` :samp:`no-fma`
 427
 428     .. index:: target("fma") function attribute, x86
 429
 430     Enable/disable the generation of the FMA instructions.
 431
 432   :samp:`fma4` :samp:`no-fma4`
 433
 434     .. index:: target("fma4") function attribute, x86
 435
 436     Enable/disable the generation of the FMA4 instructions.
 437
 438   :samp:`fsgsbase` :samp:`no-fsgsbase`
 439
 440     .. index:: target("fsgsbase") function attribute, x86
 441
 442     Enable/disable the generation of the FSGSBASE instructions.
 443
 444   :samp:`fxsr` :samp:`no-fxsr`
 445
 446     .. index:: target("fxsr") function attribute, x86
 447
 448     Enable/disable the generation of the FXSR instructions.
 449
 450   :samp:`gfni` :samp:`no-gfni`
 451
 452     .. index:: target("gfni") function attribute, x86
 453
 454     Enable/disable the generation of the GFNI instructions.
 455
 456   :samp:`hle` :samp:`no-hle`
 457
 458     .. index:: target("hle") function attribute, x86
 459
 460     Enable/disable the generation of the HLE instruction prefixes.
 461
 462   :samp:`lwp` :samp:`no-lwp`
 463
 464     .. index:: target("lwp") function attribute, x86
 465
 466     Enable/disable the generation of the LWP instructions.
 467
 468   :samp:`lzcnt` :samp:`no-lzcnt`
 469
 470     .. index:: target("lzcnt") function attribute, x86
 471
 472     Enable/disable the generation of the LZCNT instructions.
 473
 474   :samp:`mmx` :samp:`no-mmx`
 475
 476     .. index:: target("mmx") function attribute, x86
 477
 478     Enable/disable the generation of the MMX instructions.
 479
 480   :samp:`movbe` :samp:`no-movbe`
 481
 482     .. index:: target("movbe") function attribute, x86
 483
 484     Enable/disable the generation of the MOVBE instructions.
 485
 486   :samp:`movdir64b` :samp:`no-movdir64b`
 487
 488     .. index:: target("movdir64b") function attribute, x86
 489
 490     Enable/disable the generation of the MOVDIR64B instructions.
 491
 492   :samp:`movdiri` :samp:`no-movdiri`
 493
 494     .. index:: target("movdiri") function attribute, x86
 495
 496     Enable/disable the generation of the MOVDIRI instructions.
 497
 498   :samp:`mwait` :samp:`no-mwait`
 499
 500     .. index:: target("mwait") function attribute, x86
 501
 502     Enable/disable the generation of the MWAIT and MONITOR instructions.
 503
 504   :samp:`mwaitx` :samp:`no-mwaitx`
 505
 506     .. index:: target("mwaitx") function attribute, x86
 507
 508     Enable/disable the generation of the MWAITX instructions.
 509
 510   :samp:`pclmul` :samp:`no-pclmul`
 511
 512     .. index:: target("pclmul") function attribute, x86
 513
 514     Enable/disable the generation of the PCLMUL instructions.
 515
 516   :samp:`pconfig` :samp:`no-pconfig`
 517
 518     .. index:: target("pconfig") function attribute, x86
 519
 520     Enable/disable the generation of the PCONFIG instructions.
 521
 522   :samp:`pku` :samp:`no-pku`
 523
 524     .. index:: target("pku") function attribute, x86
 525
 526     Enable/disable the generation of the PKU instructions.
 527
 528   :samp:`popcnt` :samp:`no-popcnt`
 529
 530     .. index:: target("popcnt") function attribute, x86
 531
 532     Enable/disable the generation of the POPCNT instruction.
 533
 534   :samp:`prefetchwt1` :samp:`no-prefetchwt1`
 535
 536     .. index:: target("prefetchwt1") function attribute, x86
 537
 538     Enable/disable the generation of the PREFETCHWT1 instructions.
 539
 540   :samp:`prfchw` :samp:`no-prfchw`
 541
 542     .. index:: target("prfchw") function attribute, x86
 543
 544     Enable/disable the generation of the PREFETCHW instruction.
 545
 546   :samp:`ptwrite` :samp:`no-ptwrite`
 547
 548     .. index:: target("ptwrite") function attribute, x86
 549
 550     Enable/disable the generation of the PTWRITE instructions.
 551
 552   :samp:`rdpid` :samp:`no-rdpid`
 553
 554     .. index:: target("rdpid") function attribute, x86
 555
 556     Enable/disable the generation of the RDPID instructions.
 557
 558   :samp:`rdrnd` :samp:`no-rdrnd`
 559
 560     .. index:: target("rdrnd") function attribute, x86
 561
 562     Enable/disable the generation of the RDRND instructions.
 563
 564   :samp:`rdseed` :samp:`no-rdseed`
 565
 566     .. index:: target("rdseed") function attribute, x86
 567
 568     Enable/disable the generation of the RDSEED instructions.
 569
 570   :samp:`rtm` :samp:`no-rtm`
 571
 572     .. index:: target("rtm") function attribute, x86
 573
 574     Enable/disable the generation of the RTM instructions.
 575
 576   :samp:`sahf` :samp:`no-sahf`
 577
 578     .. index:: target("sahf") function attribute, x86
 579
 580     Enable/disable the generation of the SAHF instructions.
 581
 582   :samp:`sgx` :samp:`no-sgx`
 583
 584     .. index:: target("sgx") function attribute, x86
 585
 586     Enable/disable the generation of the SGX instructions.
 587
 588   :samp:`sha` :samp:`no-sha`
 589
 590     .. index:: target("sha") function attribute, x86
 591
 592     Enable/disable the generation of the SHA instructions.
 593
 594   :samp:`shstk` :samp:`no-shstk`
 595
 596     .. index:: target("shstk") function attribute, x86
 597
 598     Enable/disable the shadow stack built-in functions from CET.
 599
 600   :samp:`sse` :samp:`no-sse`
 601
 602     .. index:: target("sse") function attribute, x86
 603
 604     Enable/disable the generation of the SSE instructions.
 605
 606   :samp:`sse2` :samp:`no-sse2`
 607
 608     .. index:: target("sse2") function attribute, x86
 609
 610     Enable/disable the generation of the SSE2 instructions.
 611
 612   :samp:`sse3` :samp:`no-sse3`
 613
 614     .. index:: target("sse3") function attribute, x86
 615
 616     Enable/disable the generation of the SSE3 instructions.
 617
 618   :samp:`sse4` :samp:`no-sse4`
 619
 620     .. index:: target("sse4") function attribute, x86
 621
 622     Enable/disable the generation of the SSE4 instructions (both SSE4.1
 623     and SSE4.2).
 624
 625   :samp:`sse4.1` :samp:`no-sse4.1`
 626
 627     .. index:: target("sse4.1") function attribute, x86
 628
 629     Enable/disable the generation of the SSE4.1 instructions.
 630
 631   :samp:`sse4.2` :samp:`no-sse4.2`
 632
 633     .. index:: target("sse4.2") function attribute, x86
 634
 635     Enable/disable the generation of the SSE4.2 instructions.
 636
 637   :samp:`sse4a` :samp:`no-sse4a`
 638
 639     .. index:: target("sse4a") function attribute, x86
 640
 641     Enable/disable the generation of the SSE4A instructions.
 642
 643   :samp:`ssse3` :samp:`no-ssse3`
 644
 645     .. index:: target("ssse3") function attribute, x86
 646
 647     Enable/disable the generation of the SSSE3 instructions.
 648
 649   :samp:`tbm` :samp:`no-tbm`
 650
 651     .. index:: target("tbm") function attribute, x86
 652
 653     Enable/disable the generation of the TBM instructions.
 654
 655   :samp:`vaes` :samp:`no-vaes`
 656
 657     .. index:: target("vaes") function attribute, x86
 658
 659     Enable/disable the generation of the VAES instructions.
 660
 661   :samp:`vpclmulqdq` :samp:`no-vpclmulqdq`
 662
 663     .. index:: target("vpclmulqdq") function attribute, x86
 664
 665     Enable/disable the generation of the VPCLMULQDQ instructions.
 666
 667   :samp:`waitpkg` :samp:`no-waitpkg`
 668
 669     .. index:: target("waitpkg") function attribute, x86
 670
 671     Enable/disable the generation of the WAITPKG instructions.
 672
 673   :samp:`wbnoinvd` :samp:`no-wbnoinvd`
 674
 675     .. index:: target("wbnoinvd") function attribute, x86
 676
 677     Enable/disable the generation of the WBNOINVD instructions.
 678
 679   :samp:`xop` :samp:`no-xop`
 680
 681     .. index:: target("xop") function attribute, x86
 682
 683     Enable/disable the generation of the XOP instructions.
 684
 685   :samp:`xsave` :samp:`no-xsave`
 686
 687     .. index:: target("xsave") function attribute, x86
 688
 689     Enable/disable the generation of the XSAVE instructions.
 690
 691   :samp:`xsavec` :samp:`no-xsavec`
 692
 693     .. index:: target("xsavec") function attribute, x86
 694
 695     Enable/disable the generation of the XSAVEC instructions.
 696
 697   :samp:`xsaveopt` :samp:`no-xsaveopt`
 698
 699     .. index:: target("xsaveopt") function attribute, x86
 700
 701     Enable/disable the generation of the XSAVEOPT instructions.
 702
 703   :samp:`xsaves` :samp:`no-xsaves`
 704
 705     .. index:: target("xsaves") function attribute, x86
 706
 707     Enable/disable the generation of the XSAVES instructions.
 708
 709   :samp:`amx-tile` :samp:`no-amx-tile`
 710
 711     .. index:: target("amx-tile") function attribute, x86
 712
 713     Enable/disable the generation of the AMX-TILE instructions.
 714
 715   :samp:`amx-int8` :samp:`no-amx-int8`
 716
 717     .. index:: target("amx-int8") function attribute, x86
 718
 719     Enable/disable the generation of the AMX-INT8 instructions.
 720
 721   :samp:`amx-bf16` :samp:`no-amx-bf16`
 722
 723     .. index:: target("amx-bf16") function attribute, x86
 724
 725     Enable/disable the generation of the AMX-BF16 instructions.
 726
 727   :samp:`uintr` :samp:`no-uintr`
 728
 729     .. index:: target("uintr") function attribute, x86
 730
 731     Enable/disable the generation of the UINTR instructions.
 732
 733   :samp:`hreset` :samp:`no-hreset`
 734
 735     .. index:: target("hreset") function attribute, x86
 736
 737     Enable/disable the generation of the HRESET instruction.
 738
 739   :samp:`kl` :samp:`no-kl`
 740
 741     .. index:: target("kl") function attribute, x86
 742
 743     Enable/disable the generation of the KEYLOCKER instructions.
 744
 745   :samp:`widekl` :samp:`no-widekl`
 746
 747     .. index:: target("widekl") function attribute, x86
 748
 749     Enable/disable the generation of the WIDEKL instructions.
 750
 751   :samp:`avxvnni` :samp:`no-avxvnni`
 752
 753     .. index:: target("avxvnni") function attribute, x86
 754
 755     Enable/disable the generation of the AVXVNNI instructions.
 756
 757   :samp:`avxifma` :samp:`no-avxifma`
 758
 759     .. index:: target("avxifma") function attribute, x86
 760
 761     Enable/disable the generation of the AVXIFMA instructions.
 762
 763   :samp:`avxvnniint8` :samp:`no-avxvnniint8`
 764
 765     .. index:: target("avxvnniint8") function attribute, x86
 766
 767     Enable/disable the generation of the AVXVNNIINT8 instructions.
 768
 769   :samp:`avxneconvert` :samp:`no-avxneconvert`
 770
 771     .. index:: target("avxneconvert") function attribute, x86
 772
 773     Enable/disable the generation of the AVXNECONVERT instructions.
 774
 775   :samp:`cmpccxadd` :samp:`no-cmpccxadd`
 776
 777     .. index:: target("cmpccxadd") function attribute, x86
 778
 779     Enable/disable the generation of the CMPccXADD instructions.
 780
 781   :samp:`amx-fp16` :samp:`no-amx-fp16`
 782
 783     .. index:: target("amx-fp16") function attribute, x86
 784
 785     Enable/disable the generation of the AMX-FP16 instructions.
 786
 787   :samp:`prefetchi` :samp:`no-prefetchi`
 788
 789     .. index:: target("prefetchi") function attribute, x86
 790
 791     Enable/disable the generation of the PREFETCHI instructions.
 792
 793   :samp:`raoint` :samp:`no-raoint`
 794
 795     .. index:: target("raoint") function attribute, x86
 796
 797     Enable/disable the generation of the RAOINT instructions.
 798
 799   :samp:`cld` :samp:`no-cld`
 800
 801     .. index:: target("cld") function attribute, x86
 802
 803     Enable/disable the generation of the CLD before string moves.
 804
 805   :samp:`fancy-math-387` :samp:`no-fancy-math-387`
 806
 807     .. index:: target("fancy-math-387") function attribute, x86
 808
 809     Enable/disable the generation of the ``sin``, ``cos``, and
 810     ``sqrt`` instructions on the 387 floating-point unit.
 811
 812   :samp:`ieee-fp` :samp:`no-ieee-fp`
 813
 814     .. index:: target("ieee-fp") function attribute, x86
 815
 816     Enable/disable the generation of floating point that depends on IEEE arithmetic.
 817
 818   :samp:`inline-all-stringops` :samp:`no-inline-all-stringops`
 819
 820     .. index:: target("inline-all-stringops") function attribute, x86
 821
 822     Enable/disable inlining of string operations.
 823
 824   :samp:`inline-stringops-dynamically` :samp:`no-inline-stringops-dynamically`
 825
 826     .. index:: target("inline-stringops-dynamically") function attribute, x86
 827
 828     Enable/disable the generation of the inline code to do small string
 829     operations and calling the library routines for large operations.
 830
 831   :samp:`align-stringops` :samp:`no-align-stringops`
 832
 833     .. index:: target("align-stringops") function attribute, x86
 834
 835     Do/do not align destination of inlined string operations.
 836
 837   :samp:`recip` :samp:`no-recip`
 838
 839     .. index:: target("recip") function attribute, x86
 840
 841     Enable/disable the generation of RCPSS, RCPPS, RSQRTSS and RSQRTPS
 842     instructions followed an additional Newton-Raphson step instead of
 843     doing a floating-point division.
 844
 845   :samp:`general-regs-only`
 846
 847     .. index:: target("general-regs-only") function attribute, x86
 848
 849     Generate code which uses only the general registers.
 850
 851   :samp:`arch={ARCH}`
 852
 853     .. index:: target("arch=ARCH") function attribute, x86
 854
 855     Specify the architecture to generate code for in compiling the function.
 856
 857   :samp:`tune={TUNE}`
 858
 859     .. index:: target("tune=TUNE") function attribute, x86
 860
 861     Specify the architecture to tune for in compiling the function.
 862
 863   :samp:`fpmath={FPMATH}`
 864
 865     .. index:: target("fpmath=FPMATH") function attribute, x86
 866
 867     Specify which floating-point unit to use.  You must specify the
 868     ``target("fpmath=sse,387")`` option as
 869     ``target("fpmath=sse+387")`` because the comma would separate
 870     different options.
 871
 872   :samp:`prefer-vector-width={OPT}`
 873
 874     .. index:: prefer-vector-width function attribute, x86
 875
 876     On x86 targets, the ``prefer-vector-width`` attribute informs the
 877     compiler to use :samp:`{OPT}` -bit vector width in instructions
 878     instead of the default on the selected platform.
 879
 880     Valid :samp:`{OPT}` values are:
 881
 882     :samp:`none`
 883       No extra limitations applied to GCC other than defined by the selected platform.
 884
 885     :samp:`128`
 886       Prefer 128-bit vector width for instructions.
 887
 888     :samp:`256`
 889       Prefer 256-bit vector width for instructions.
 890
 891     :samp:`512`
 892       Prefer 512-bit vector width for instructions.
 893
 894     On the x86, the inliner does not inline a
 895     function that has different target options than the caller, unless the
 896     callee has a subset of the target options of the caller.  For example
 897     a function declared with ``target("sse3")`` can inline a function
 898     with ``target("sse2")``, since ``-msse3`` implies ``-msse2``.
 899
 900 .. index:: indirect_branch function attribute, x86
 901
 902 .. x86-fn-attr:: indirect_branch("choice")
 903
 904   On x86 targets, the ``indirect_branch`` attribute causes the compiler
 905   to convert indirect call and jump with :samp:`{choice}`.  :samp:`keep`
 906   keeps indirect call and jump unmodified.  :samp:`thunk` converts indirect
 907   call and jump to call and return thunk.  :samp:`thunk-inline` converts
 908   indirect call and jump to inlined call and return thunk.
 909   :samp:`thunk-extern` converts indirect call and jump to external call
 910   and return thunk provided in a separate object file.
 911
 912 .. index:: function_return function attribute, x86
 913
 914 .. x86-fn-attr:: function_return("choice")
 915
 916   On x86 targets, the ``function_return`` attribute causes the compiler
 917   to convert function return with :samp:`{choice}`.  :samp:`keep` keeps function
 918   return unmodified.  :samp:`thunk` converts function return to call and
 919   return thunk.  :samp:`thunk-inline` converts function return to inlined
 920   call and return thunk.  :samp:`thunk-extern` converts function return to
 921   external call and return thunk provided in a separate object file.
 922
 923 .. index:: nocf_check function attribute
 924
 925 .. x86-fn-attr:: nocf_check
 926
 927   The :x86-fn-attr:`nocf_check` attribute on a function is used to inform the
 928   compiler that the function's prologue should not be instrumented when
 929   compiled with the :option:`-fcf-protection=branch` option.  The
 930   compiler assumes that the function's address is a valid target for a
 931   control-flow transfer.
 932
 933   The :x86-fn-attr:`nocf_check` attribute on a type of pointer to function is
 934   used to inform the compiler that a call through the pointer should
 935   not be instrumented when compiled with the
 936   :option:`-fcf-protection=branch` option.  The compiler assumes
 937   that the function's address from the pointer is a valid target for
 938   a control-flow transfer.  A direct function call through a function
 939   name is assumed to be a safe call thus direct calls are not
 940   instrumented by the compiler.
 941
 942   The :x86-fn-attr:`nocf_check` attribute is applied to an object's type.
 943   In case of assignment of a function address or a function pointer to
 944   another pointer, the attribute is not carried over from the right-hand
 945   object's type; the type of left-hand object stays unchanged.  The
 946   compiler checks for :x86-fn-attr:`nocf_check` attribute mismatch and reports
 947   a warning in case of mismatch.
 948
 949   .. code-block:: c++
 950
 951     {
 952     int foo (void) __attribute__(nocf_check);
 953     void (*foo1)(void) __attribute__(nocf_check);
 954     void (*foo2)(void);
 955
 956     /* foo's address is assumed to be valid.  */
 957     int
 958     foo (void)
 959
 960       /* This call site is not checked for control-flow
 961          validity.  */
 962       (*foo1)();
 963
 964       /* A warning is issued about attribute mismatch.  */
 965       foo1 = foo2;
 966
 967       /* This call site is still not checked.  */
 968       (*foo1)();
 969
 970       /* This call site is checked.  */
 971       (*foo2)();
 972
 973       /* A warning is issued about attribute mismatch.  */
 974       foo2 = foo1;
 975
 976       /* This call site is still checked.  */
 977       (*foo2)();
 978
 979       return 0;
 980     }
 981
 982 .. index:: cf_check function attribute, x86
 983
 984 .. x86-fn-attr:: cf_check
 985
 986   The :x86-fn-attr:`cf_check` attribute on a function is used to inform the
 987   compiler that ENDBR instruction should be placed at the function
 988   entry when :option:`-fcf-protection=branch` is enabled.
 989
 990 .. index:: indirect_return function attribute, x86
 991
 992 .. x86-fn-attr:: indirect_return
 993
 994   The :x86-fn-attr:`indirect_return` attribute can be applied to a function,
 995   as well as variable or type of function pointer to inform the
 996   compiler that the function may return via indirect branch.
 997
 998 .. index:: fentry_name function attribute, x86
 999
1000 .. x86-fn-attr:: fentry_name("name")
1001
1002   On x86 targets, the ``fentry_name`` attribute sets the function to
1003   call on function entry when function instrumentation is enabled
1004   with :option:`-pg -mfentry`. When :samp:`{name}` is nop then a 5 byte
1005   nop sequence is generated.
1006
1007 .. index:: fentry_section function attribute, x86
1008
1009 .. x86-fn-attr:: fentry_section("name")
1010
1011   On x86 targets, the ``fentry_section`` attribute sets the name
1012   of the section to record function entry instrumentation calls in when
1013   enabled with :option:`-pg -mrecord-mcount`
1014
1015 .. index:: nodirect_extern_access function attribute
1016
1017 .. option:: nodirect_extern_access
1018
1019   This attribute, attached to a global variable or function, is the
1020   counterpart to option :option:`-mno-direct-extern-access`.