1 This is doc/gcc.info, produced by makeinfo version 4.8 from
2 /scratch/mitchell/gcc-releases/gcc-4.1.2/gcc-4.1.2/gcc/doc/gcc.texi.
4 Copyright (C) 1988, 1989, 1992, 1993, 1994, 1995, 1996, 1997, 1998,
5 1999, 2000, 2001, 2002, 2003, 2004, 2005 Free Software Foundation, Inc.
7 Permission is granted to copy, distribute and/or modify this document
8 under the terms of the GNU Free Documentation License, Version 1.2 or
9 any later version published by the Free Software Foundation; with the
10 Invariant Sections being "GNU General Public License" and "Funding Free
11 Software", the Front-Cover texts being (a) (see below), and with the
12 Back-Cover Texts being (b) (see below). A copy of the license is
13 included in the section entitled "GNU Free Documentation License".
15 (a) The FSF's Front-Cover Text is:
19 (b) The FSF's Back-Cover Text is:
21 You have freedom to copy and modify this GNU Manual, like GNU
22 software. Copies published by the Free Software Foundation raise
23 funds for GNU development.
25 INFO-DIR-SECTION Programming
27 * gcc: (gcc). The GNU Compiler Collection.
29 This file documents the use of the GNU compilers.
31 Copyright (C) 1988, 1989, 1992, 1993, 1994, 1995, 1996, 1997, 1998,
32 1999, 2000, 2001, 2002, 2003, 2004, 2005 Free Software Foundation, Inc.
34 Permission is granted to copy, distribute and/or modify this document
35 under the terms of the GNU Free Documentation License, Version 1.2 or
36 any later version published by the Free Software Foundation; with the
37 Invariant Sections being "GNU General Public License" and "Funding Free
38 Software", the Front-Cover texts being (a) (see below), and with the
39 Back-Cover Texts being (b) (see below). A copy of the license is
40 included in the section entitled "GNU Free Documentation License".
42 (a) The FSF's Front-Cover Text is:
46 (b) The FSF's Back-Cover Text is:
48 You have freedom to copy and modify this GNU Manual, like GNU
49 software. Copies published by the Free Software Foundation raise
50 funds for GNU development.
54 File: gcc.info, Node: Top, Next: G++ and GCC, Up: (DIR)
59 This manual documents how to use the GNU compilers, as well as their
60 features and incompatibilities, and how to report bugs. It corresponds
61 to GCC version 4.1.2. The internals of the GNU compilers, including
62 how to port them to new targets and some information about how to write
63 front ends for new languages, are documented in a separate manual.
64 *Note Introduction: (gccint)Top.
68 * G++ and GCC:: You can compile C or C++ programs.
69 * Standards:: Language standards supported by GCC.
70 * Invoking GCC:: Command options supported by `gcc'.
71 * C Implementation:: How GCC implements the ISO C specification.
72 * C Extensions:: GNU extensions to the C language family.
73 * C++ Extensions:: GNU extensions to the C++ language.
74 * Objective-C:: GNU Objective-C runtime features.
75 * Compatibility:: Binary Compatibility
76 * Gcov:: `gcov'---a test coverage program.
77 * Trouble:: If you have trouble using GCC.
78 * Bugs:: How, why and where to report bugs.
79 * Service:: How to find suppliers of support for GCC.
80 * Contributing:: How to contribute to testing and developing GCC.
82 * Funding:: How to help assure funding for free software.
83 * GNU Project:: The GNU Project and GNU/Linux.
85 * Copying:: GNU General Public License says
86 how you can copy and share GCC.
87 * GNU Free Documentation License:: How you can copy and share this manual.
88 * Contributors:: People who have contributed to GCC.
90 * Option Index:: Index to command line options.
91 * Keyword Index:: Index of concepts and symbol names.
94 File: gcc.info, Node: G++ and GCC, Next: Standards, Prev: Top, Up: Top
96 1 Programming Languages Supported by GCC
97 ****************************************
99 GCC stands for "GNU Compiler Collection". GCC is an integrated
100 distribution of compilers for several major programming languages.
101 These languages currently include C, C++, Objective-C, Objective-C++,
102 Java, Fortran, and Ada.
104 The abbreviation "GCC" has multiple meanings in common use. The
105 current official meaning is "GNU Compiler Collection", which refers
106 generically to the complete suite of tools. The name historically stood
107 for "GNU C Compiler", and this usage is still common when the emphasis
108 is on compiling C programs. Finally, the name is also used when
109 speaking of the "language-independent" component of GCC: code shared
110 among the compilers for all supported languages.
112 The language-independent component of GCC includes the majority of the
113 optimizers, as well as the "back ends" that generate machine code for
116 The part of a compiler that is specific to a particular language is
117 called the "front end". In addition to the front ends that are
118 integrated components of GCC, there are several other front ends that
119 are maintained separately. These support languages such as Pascal,
120 Mercury, and COBOL. To use these, they must be built together with GCC
123 Most of the compilers for languages other than C have their own names.
124 The C++ compiler is G++, the Ada compiler is GNAT, and so on. When we
125 talk about compiling one of those languages, we might refer to that
126 compiler by its own name, or as GCC. Either is correct.
128 Historically, compilers for many languages, including C++ and Fortran,
129 have been implemented as "preprocessors" which emit another high level
130 language such as C. None of the compilers included in GCC are
131 implemented this way; they all generate machine code directly. This
132 sort of preprocessor should not be confused with the "C preprocessor",
133 which is an integral feature of the C, C++, Objective-C and
134 Objective-C++ languages.
137 File: gcc.info, Node: Standards, Next: Invoking GCC, Prev: G++ and GCC, Up: Top
139 2 Language Standards Supported by GCC
140 *************************************
142 For each language compiled by GCC for which there is a standard, GCC
143 attempts to follow one or more versions of that standard, possibly with
144 some exceptions, and possibly with some extensions.
146 GCC supports three versions of the C standard, although support for
147 the most recent version is not yet complete.
149 The original ANSI C standard (X3.159-1989) was ratified in 1989 and
150 published in 1990. This standard was ratified as an ISO standard
151 (ISO/IEC 9899:1990) later in 1990. There were no technical differences
152 between these publications, although the sections of the ANSI standard
153 were renumbered and became clauses in the ISO standard. This standard,
154 in both its forms, is commonly known as "C89", or occasionally as
155 "C90", from the dates of ratification. The ANSI standard, but not the
156 ISO standard, also came with a Rationale document. To select this
157 standard in GCC, use one of the options `-ansi', `-std=c89' or
158 `-std=iso9899:1990'; to obtain all the diagnostics required by the
159 standard, you should also specify `-pedantic' (or `-pedantic-errors' if
160 you want them to be errors rather than warnings). *Note Options
161 Controlling C Dialect: C Dialect Options.
163 Errors in the 1990 ISO C standard were corrected in two Technical
164 Corrigenda published in 1994 and 1996. GCC does not support the
167 An amendment to the 1990 standard was published in 1995. This
168 amendment added digraphs and `__STDC_VERSION__' to the language, but
169 otherwise concerned the library. This amendment is commonly known as
170 "AMD1"; the amended standard is sometimes known as "C94" or "C95". To
171 select this standard in GCC, use the option `-std=iso9899:199409'
172 (with, as for other standard versions, `-pedantic' to receive all
173 required diagnostics).
175 A new edition of the ISO C standard was published in 1999 as ISO/IEC
176 9899:1999, and is commonly known as "C99". GCC has incomplete support
177 for this standard version; see
178 `http://gcc.gnu.org/gcc-4.1/c99status.html' for details. To select this
179 standard, use `-std=c99' or `-std=iso9899:1999'. (While in
180 development, drafts of this standard version were referred to as "C9X".)
182 Errors in the 1999 ISO C standard were corrected in two Technical
183 Corrigenda published in 2001 and 2004. GCC does not support the
186 By default, GCC provides some extensions to the C language that on
187 rare occasions conflict with the C standard. *Note Extensions to the C
188 Language Family: C Extensions. Use of the `-std' options listed above
189 will disable these extensions where they conflict with the C standard
190 version selected. You may also select an extended version of the C
191 language explicitly with `-std=gnu89' (for C89 with GNU extensions) or
192 `-std=gnu99' (for C99 with GNU extensions). The default, if no C
193 language dialect options are given, is `-std=gnu89'; this will change to
194 `-std=gnu99' in some future release when the C99 support is complete.
195 Some features that are part of the C99 standard are accepted as
196 extensions in C89 mode.
198 The ISO C standard defines (in clause 4) two classes of conforming
199 implementation. A "conforming hosted implementation" supports the
200 whole standard including all the library facilities; a "conforming
201 freestanding implementation" is only required to provide certain
202 library facilities: those in `<float.h>', `<limits.h>', `<stdarg.h>',
203 and `<stddef.h>'; since AMD1, also those in `<iso646.h>'; and in C99,
204 also those in `<stdbool.h>' and `<stdint.h>'. In addition, complex
205 types, added in C99, are not required for freestanding implementations.
206 The standard also defines two environments for programs, a
207 "freestanding environment", required of all implementations and which
208 may not have library facilities beyond those required of freestanding
209 implementations, where the handling of program startup and termination
210 are implementation-defined, and a "hosted environment", which is not
211 required, in which all the library facilities are provided and startup
212 is through a function `int main (void)' or `int main (int, char *[])'.
213 An OS kernel would be a freestanding environment; a program using the
214 facilities of an operating system would normally be in a hosted
217 GCC aims towards being usable as a conforming freestanding
218 implementation, or as the compiler for a conforming hosted
219 implementation. By default, it will act as the compiler for a hosted
220 implementation, defining `__STDC_HOSTED__' as `1' and presuming that
221 when the names of ISO C functions are used, they have the semantics
222 defined in the standard. To make it act as a conforming freestanding
223 implementation for a freestanding environment, use the option
224 `-ffreestanding'; it will then define `__STDC_HOSTED__' to `0' and not
225 make assumptions about the meanings of function names from the standard
226 library, with exceptions noted below. To build an OS kernel, you may
227 well still need to make your own arrangements for linking and startup.
228 *Note Options Controlling C Dialect: C Dialect Options.
230 GCC does not provide the library facilities required only of hosted
231 implementations, nor yet all the facilities required by C99 of
232 freestanding implementations; to use the facilities of a hosted
233 environment, you will need to find them elsewhere (for example, in the
234 GNU C library). *Note Standard Libraries: Standard Libraries.
236 Most of the compiler support routines used by GCC are present in
237 `libgcc', but there are a few exceptions. GCC requires the
238 freestanding environment provide `memcpy', `memmove', `memset' and
239 `memcmp'. Finally, if `__builtin_trap' is used, and the target does
240 not implement the `trap' pattern, then GCC will emit a call to `abort'.
242 For references to Technical Corrigenda, Rationale documents and
243 information concerning the history of C that is available online, see
244 `http://gcc.gnu.org/readings.html'
246 There is no formal written standard for Objective-C or Objective-C++.
247 The most authoritative manual is "Object-Oriented Programming and the
248 Objective-C Language", available at a number of web sites:
251 `http://developer.apple.com/documentation/Cocoa/Conceptual/ObjectiveC/'
252 is a recent (and periodically updated) version;
254 * `http://www.toodarkpark.org/computers/objc/' is an older example;
256 * `http://www.gnustep.org' and `http://gcc.gnu.org/readings.html'
257 have additional useful information.
259 There is no standard for treelang, which is a sample language front end
260 for GCC. Its only purpose is as a sample for people wishing to write a
261 new language for GCC. The language is documented in
262 `gcc/treelang/treelang.texi' which can be turned into info or HTML
265 *Note GNAT Reference Manual: (gnat_rm)Top, for information on standard
266 conformance and compatibility of the Ada compiler.
268 *Note Standards: (gfortran)Standards, for details of standards
269 supported by `gfortran'.
271 *Note Compatibility with the Java Platform: (gcj)Compatibility, for
272 details of compatibility between `gcj' and the Java Platform.
275 File: gcc.info, Node: Invoking GCC, Next: C Implementation, Prev: Standards, Up: Top
277 3 GCC Command Options
278 *********************
280 When you invoke GCC, it normally does preprocessing, compilation,
281 assembly and linking. The "overall options" allow you to stop this
282 process at an intermediate stage. For example, the `-c' option says
283 not to run the linker. Then the output consists of object files output
286 Other options are passed on to one stage of processing. Some options
287 control the preprocessor and others the compiler itself. Yet other
288 options control the assembler and linker; most of these are not
289 documented here, since you rarely need to use any of them.
291 Most of the command line options that you can use with GCC are useful
292 for C programs; when an option is only useful with another language
293 (usually C++), the explanation says so explicitly. If the description
294 for a particular option does not mention a source language, you can use
295 that option with all supported languages.
297 *Note Compiling C++ Programs: Invoking G++, for a summary of special
298 options for compiling C++ programs.
300 The `gcc' program accepts options and file names as operands. Many
301 options have multi-letter names; therefore multiple single-letter
302 options may _not_ be grouped: `-dr' is very different from `-d -r'.
304 You can mix options and other arguments. For the most part, the order
305 you use doesn't matter. Order does matter when you use several options
306 of the same kind; for example, if you specify `-L' more than once, the
307 directories are searched in the order specified.
309 Many options have long names starting with `-f' or with `-W'--for
310 example, `-fstrength-reduce', `-Wformat' and so on. Most of these have
311 both positive and negative forms; the negative form of `-ffoo' would be
312 `-fno-foo'. This manual documents only one of these two forms,
313 whichever one is not the default.
315 *Note Option Index::, for an index to GCC's options.
319 * Option Summary:: Brief list of all options, without explanations.
320 * Overall Options:: Controlling the kind of output:
321 an executable, object files, assembler files,
322 or preprocessed source.
323 * Invoking G++:: Compiling C++ programs.
324 * C Dialect Options:: Controlling the variant of C language compiled.
325 * C++ Dialect Options:: Variations on C++.
326 * Objective-C and Objective-C++ Dialect Options:: Variations on Objective-C
328 * Language Independent Options:: Controlling how diagnostics should be
330 * Warning Options:: How picky should the compiler be?
331 * Debugging Options:: Symbol tables, measurements, and debugging dumps.
332 * Optimize Options:: How much optimization?
333 * Preprocessor Options:: Controlling header files and macro definitions.
334 Also, getting dependency information for Make.
335 * Assembler Options:: Passing options to the assembler.
336 * Link Options:: Specifying libraries and so on.
337 * Directory Options:: Where to find header files and libraries.
338 Where to find the compiler executable files.
339 * Spec Files:: How to pass switches to sub-processes.
340 * Target Options:: Running a cross-compiler, or an old version of GCC.
341 * Submodel Options:: Specifying minor hardware or convention variations,
342 such as 68010 vs 68020.
343 * Code Gen Options:: Specifying conventions for function calls, data layout
345 * Environment Variables:: Env vars that affect GCC.
346 * Precompiled Headers:: Compiling a header once, and using it many times.
347 * Running Protoize:: Automatically adding or removing function prototypes.
350 File: gcc.info, Node: Option Summary, Next: Overall Options, Up: Invoking GCC
355 Here is a summary of all the options, grouped by type. Explanations are
356 in the following sections.
359 *Note Options Controlling the Kind of Output: Overall Options.
360 -c -S -E -o FILE -combine -pipe -pass-exit-codes
361 -x LANGUAGE -v -### --help --target-help --version
364 *Note Options Controlling C Dialect: C Dialect Options.
365 -ansi -std=STANDARD -aux-info FILENAME
366 -fno-asm -fno-builtin -fno-builtin-FUNCTION
367 -fhosted -ffreestanding -fms-extensions
368 -trigraphs -no-integrated-cpp -traditional -traditional-cpp
369 -fallow-single-precision -fcond-mismatch
370 -fsigned-bitfields -fsigned-char
371 -funsigned-bitfields -funsigned-char
373 _C++ Language Options_
374 *Note Options Controlling C++ Dialect: C++ Dialect Options.
375 -fabi-version=N -fno-access-control -fcheck-new
376 -fconserve-space -ffriend-injection -fno-const-strings
377 -fno-elide-constructors
378 -fno-enforce-eh-specs
379 -ffor-scope -fno-for-scope -fno-gnu-keywords
380 -fno-implicit-templates
381 -fno-implicit-inline-templates
382 -fno-implement-inlines -fms-extensions
383 -fno-nonansi-builtins -fno-operator-names
384 -fno-optional-diags -fpermissive
385 -frepo -fno-rtti -fstats -ftemplate-depth-N
386 -fno-threadsafe-statics -fuse-cxa-atexit -fno-weak -nostdinc++
387 -fno-default-inline -fvisibility-inlines-hidden
388 -Wabi -Wctor-dtor-privacy
389 -Wnon-virtual-dtor -Wreorder
390 -Weffc++ -Wno-deprecated -Wstrict-null-sentinel
391 -Wno-non-template-friend -Wold-style-cast
392 -Woverloaded-virtual -Wno-pmf-conversions
395 _Objective-C and Objective-C++ Language Options_
396 *Note Options Controlling Objective-C and Objective-C++ Dialects:
397 Objective-C and Objective-C++ Dialect Options.
398 -fconstant-string-class=CLASS-NAME
399 -fgnu-runtime -fnext-runtime
401 -fobjc-call-cxx-cdtors
402 -fobjc-direct-dispatch
405 -freplace-objc-classes
409 -Wno-protocol -Wselector
410 -Wstrict-selector-match
411 -Wundeclared-selector
413 _Language Independent Options_
414 *Note Options to Control Diagnostic Messages Formatting: Language
417 -fdiagnostics-show-location=[once|every-line]
418 -fdiagnostics-show-options
421 *Note Options to Request or Suppress Warnings: Warning Options.
422 -fsyntax-only -pedantic -pedantic-errors
423 -w -Wextra -Wall -Waggregate-return -Wno-attributes
424 -Wc++-compat -Wcast-align -Wcast-qual -Wchar-subscripts -Wcomment
425 -Wconversion -Wno-deprecated-declarations
426 -Wdisabled-optimization -Wno-div-by-zero -Wno-endif-labels
427 -Werror -Werror-implicit-function-declaration
428 -Wfatal-errors -Wfloat-equal -Wformat -Wformat=2
429 -Wno-format-extra-args -Wformat-nonliteral
430 -Wformat-security -Wformat-y2k
431 -Wimplicit -Wimplicit-function-declaration -Wimplicit-int
432 -Wimport -Wno-import -Winit-self -Winline
433 -Wno-int-to-pointer-cast
434 -Wno-invalid-offsetof -Winvalid-pch
435 -Wlarger-than-LEN -Wunsafe-loop-optimizations -Wlong-long
436 -Wmain -Wmissing-braces -Wmissing-field-initializers
437 -Wmissing-format-attribute -Wmissing-include-dirs
439 -Wno-multichar -Wnonnull -Wpacked -Wpadded
440 -Wparentheses -Wpointer-arith -Wno-pointer-to-int-cast
442 -Wreturn-type -Wsequence-point -Wshadow
443 -Wsign-compare -Wstack-protector
444 -Wstrict-aliasing -Wstrict-aliasing=2
445 -Wswitch -Wswitch-default -Wswitch-enum
446 -Wsystem-headers -Wtrigraphs -Wundef -Wuninitialized
447 -Wunknown-pragmas -Wno-pragmas -Wunreachable-code
448 -Wunused -Wunused-function -Wunused-label -Wunused-parameter
449 -Wunused-value -Wunused-variable -Wvariadic-macros
450 -Wvolatile-register-var -Wwrite-strings
452 _C-only Warning Options_
453 -Wbad-function-cast -Wmissing-declarations
454 -Wmissing-prototypes -Wnested-externs -Wold-style-definition
455 -Wstrict-prototypes -Wtraditional
456 -Wdeclaration-after-statement -Wpointer-sign
459 *Note Options for Debugging Your Program or GCC: Debugging Options.
460 -dLETTERS -dumpspecs -dumpmachine -dumpversion
461 -fdump-unnumbered -fdump-translation-unit[-N]
462 -fdump-class-hierarchy[-N]
463 -fdump-ipa-all -fdump-ipa-cgraph
465 -fdump-tree-original[-N]
466 -fdump-tree-optimized[-N]
467 -fdump-tree-inlined[-N]
468 -fdump-tree-cfg -fdump-tree-vcg -fdump-tree-alias
470 -fdump-tree-ssa[-N] -fdump-tree-pre[-N]
471 -fdump-tree-ccp[-N] -fdump-tree-dce[-N]
472 -fdump-tree-gimple[-raw] -fdump-tree-mudflap[-N]
475 -fdump-tree-phiopt[-N]
476 -fdump-tree-forwprop[-N]
477 -fdump-tree-copyrename[-N]
478 -fdump-tree-nrv -fdump-tree-vect
484 -ftree-vectorizer-verbose=N
485 -fdump-tree-storeccp[-N]
486 -feliminate-dwarf2-dups -feliminate-unused-debug-types
487 -feliminate-unused-debug-symbols -fmem-report -fprofile-arcs
488 -frandom-seed=STRING -fsched-verbose=N
489 -ftest-coverage -ftime-report -fvar-tracking
490 -g -gLEVEL -gcoff -gdwarf-2
491 -ggdb -gstabs -gstabs+ -gvms -gxcoff -gxcoff+
492 -p -pg -print-file-name=LIBRARY -print-libgcc-file-name
493 -print-multi-directory -print-multi-lib
494 -print-prog-name=PROGRAM -print-search-dirs -Q
497 _Optimization Options_
498 *Note Options that Control Optimization: Optimize Options.
499 -falign-functions=N -falign-jumps=N
500 -falign-labels=N -falign-loops=N
501 -fbounds-check -fmudflap -fmudflapth -fmudflapir
502 -fbranch-probabilities -fprofile-values -fvpt -fbranch-target-load-optimize
503 -fbranch-target-load-optimize2 -fbtr-bb-exclusive
504 -fcaller-saves -fcprop-registers -fcse-follow-jumps
505 -fcse-skip-blocks -fcx-limited-range -fdata-sections
506 -fdelayed-branch -fdelete-null-pointer-checks -fearly-inlining
507 -fexpensive-optimizations -ffast-math -ffloat-store
508 -fforce-addr -ffunction-sections
509 -fgcse -fgcse-lm -fgcse-sm -fgcse-las -fgcse-after-reload
510 -floop-optimize -fcrossjumping -fif-conversion -fif-conversion2
511 -finline-functions -finline-functions-called-once
512 -finline-limit=N -fkeep-inline-functions
513 -fkeep-static-consts -fmerge-constants -fmerge-all-constants
514 -fmodulo-sched -fno-branch-count-reg
515 -fno-default-inline -fno-defer-pop -floop-optimize2 -fmove-loop-invariants
516 -fno-function-cse -fno-guess-branch-probability
517 -fno-inline -fno-math-errno -fno-peephole -fno-peephole2
518 -funsafe-math-optimizations -funsafe-loop-optimizations -ffinite-math-only
519 -fno-trapping-math -fno-zero-initialized-in-bss
520 -fomit-frame-pointer -foptimize-register-move
521 -foptimize-sibling-calls -fprefetch-loop-arrays
522 -fprofile-generate -fprofile-use
523 -fregmove -frename-registers
524 -freorder-blocks -freorder-blocks-and-partition -freorder-functions
525 -frerun-cse-after-loop -frerun-loop-opt
526 -frounding-math -fschedule-insns -fschedule-insns2
527 -fno-sched-interblock -fno-sched-spec -fsched-spec-load
528 -fsched-spec-load-dangerous
529 -fsched-stalled-insns=N -fsched-stalled-insns-dep=N
530 -fsched2-use-superblocks
531 -fsched2-use-traces -freschedule-modulo-scheduled-loops
532 -fsignaling-nans -fsingle-precision-constant
533 -fstack-protector -fstack-protector-all
534 -fstrength-reduce -fstrict-aliasing -ftracer -fthread-jumps
535 -funroll-all-loops -funroll-loops -fpeel-loops
536 -fsplit-ivs-in-unroller -funswitch-loops
537 -fvariable-expansion-in-unroller
538 -ftree-pre -ftree-ccp -ftree-dce -ftree-loop-optimize
539 -ftree-loop-linear -ftree-loop-im -ftree-loop-ivcanon -fivopts
540 -ftree-dominator-opts -ftree-dse -ftree-copyrename -ftree-sink
541 -ftree-ch -ftree-sra -ftree-ter -ftree-lrs -ftree-fre -ftree-vectorize
542 -ftree-vect-loop-version -ftree-salias -fweb
543 -ftree-copy-prop -ftree-store-ccp -ftree-store-copy-prop -fwhole-program
545 -O -O0 -O1 -O2 -O3 -Os
547 _Preprocessor Options_
548 *Note Options Controlling the Preprocessor: Preprocessor Options.
554 -include FILE -imacros FILE
555 -iprefix FILE -iwithprefix DIR
556 -iwithprefixbefore DIR -isystem DIR
558 -M -MM -MF -MG -MP -MQ -MT -nostdinc
559 -P -fworking-directory -remap
560 -trigraphs -undef -UMACRO -Wp,OPTION
561 -Xpreprocessor OPTION
564 *Note Passing Options to the Assembler: Assembler Options.
565 -Wa,OPTION -Xassembler OPTION
568 *Note Options for Linking: Link Options.
569 OBJECT-FILE-NAME -lLIBRARY
570 -nostartfiles -nodefaultlibs -nostdlib -pie -rdynamic
571 -s -static -static-libgcc -shared -shared-libgcc -symbolic
572 -Wl,OPTION -Xlinker OPTION
576 *Note Options for Directory Search: Directory Options.
577 -BPREFIX -IDIR -iquoteDIR -LDIR
578 -specs=FILE -I- --sysroot=DIR
581 *Note Target Options::.
582 -V VERSION -b MACHINE
584 _Machine Dependent Options_
585 *Note Hardware Models and Configurations: Submodel Options.
589 -mmangle-cpu -mcpu=CPU -mtext=TEXT-SECTION
590 -mdata=DATA-SECTION -mrodata=READONLY-DATA-SECTION
593 -mapcs-frame -mno-apcs-frame
595 -mapcs-stack-check -mno-apcs-stack-check
596 -mapcs-float -mno-apcs-float
597 -mapcs-reentrant -mno-apcs-reentrant
598 -msched-prolog -mno-sched-prolog
599 -mlittle-endian -mbig-endian -mwords-little-endian
600 -mfloat-abi=NAME -msoft-float -mhard-float -mfpe
601 -mthumb-interwork -mno-thumb-interwork
602 -mcpu=NAME -march=NAME -mfpu=NAME
603 -mstructure-size-boundary=N
605 -mlong-calls -mno-long-calls
606 -msingle-pic-base -mno-single-pic-base
609 -mcirrus-fix-invalid-insns -mno-cirrus-fix-invalid-insns
612 -mtpcs-frame -mtpcs-leaf-frame
613 -mcaller-super-interworking -mcallee-super-interworking
617 -mmcu=MCU -msize -minit-stack=N -mno-interrupts
618 -mcall-prologues -mno-tablejump -mtiny-stack -mint8
621 -momit-leaf-frame-pointer -mno-omit-leaf-frame-pointer
622 -mspecld-anomaly -mno-specld-anomaly -mcsync-anomaly -mno-csync-anomaly
623 -mlow-64k -mno-low64k -mid-shared-library
624 -mno-id-shared-library -mshared-library-id=N
625 -mlong-calls -mno-long-calls
628 -mcpu=CPU -march=CPU -mtune=CPU
629 -mmax-stack-frame=N -melinux-stacksize=N
630 -metrax4 -metrax100 -mpdebug -mcc-init -mno-side-effects
631 -mstack-align -mdata-align -mconst-align
632 -m32-bit -m16-bit -m8-bit -mno-prologue-epilogue -mno-gotplt
633 -melf -maout -melinux -mlinux -sim -sim2
634 -mmul-bug-workaround -mno-mul-bug-workaround
640 -all_load -allowable_client -arch -arch_errors_fatal
641 -arch_only -bind_at_load -bundle -bundle_loader
642 -client_name -compatibility_version -current_version
644 -dependency-file -dylib_file -dylinker_install_name
645 -dynamic -dynamiclib -exported_symbols_list
646 -filelist -flat_namespace -force_cpusubtype_ALL
647 -force_flat_namespace -headerpad_max_install_names
648 -image_base -init -install_name -keep_private_externs
649 -multi_module -multiply_defined -multiply_defined_unused
650 -noall_load -no_dead_strip_inits_and_terms
651 -nofixprebinding -nomultidefs -noprebind -noseglinkedit
652 -pagezero_size -prebind -prebind_all_twolevel_modules
653 -private_bundle -read_only_relocs -sectalign
654 -sectobjectsymbols -whyload -seg1addr
655 -sectcreate -sectobjectsymbols -sectorder
656 -segaddr -segs_read_only_addr -segs_read_write_addr
657 -seg_addr_table -seg_addr_table_filename -seglinkedit
658 -segprot -segs_read_only_addr -segs_read_write_addr
659 -single_module -static -sub_library -sub_umbrella
660 -twolevel_namespace -umbrella -undefined
661 -unexported_symbols_list -weak_reference_mismatches
662 -whatsloaded -F -gused -gfull -mmacosx-version-min=VERSION
666 -mno-fp-regs -msoft-float -malpha-as -mgas
667 -mieee -mieee-with-inexact -mieee-conformant
668 -mfp-trap-mode=MODE -mfp-rounding-mode=MODE
669 -mtrap-precision=MODE -mbuild-constants
670 -mcpu=CPU-TYPE -mtune=CPU-TYPE
671 -mbwx -mmax -mfix -mcix
672 -mfloat-vax -mfloat-ieee
673 -mexplicit-relocs -msmall-data -mlarge-data
674 -msmall-text -mlarge-text
675 -mmemory-latency=TIME
677 _DEC Alpha/VMS Options_
681 -mgpr-32 -mgpr-64 -mfpr-32 -mfpr-64
682 -mhard-float -msoft-float
683 -malloc-cc -mfixed-cc -mdword -mno-dword
685 -mmedia -mno-media -mmuladd -mno-muladd
686 -mfdpic -minline-plt -mgprel-ro -multilib-library-pic
687 -mlinked-fp -mlong-calls -malign-labels
688 -mlibrary-pic -macc-4 -macc-8
689 -mpack -mno-pack -mno-eflags -mcond-move -mno-cond-move
690 -moptimize-membar -mno-optimize-membar
691 -mscc -mno-scc -mcond-exec -mno-cond-exec
692 -mvliw-branch -mno-vliw-branch
693 -mmulti-cond-exec -mno-multi-cond-exec -mnested-cond-exec
694 -mno-nested-cond-exec -mtomcat-stats
699 -mrelax -mh -ms -mn -mint32 -malign-300
702 -march=ARCHITECTURE-TYPE
703 -mbig-switch -mdisable-fpregs -mdisable-indexing
704 -mfast-indirect-calls -mgas -mgnu-ld -mhp-ld
705 -mfixed-range=REGISTER-RANGE
706 -mjump-in-delay -mlinker-opt -mlong-calls
707 -mlong-load-store -mno-big-switch -mno-disable-fpregs
708 -mno-disable-indexing -mno-fast-indirect-calls -mno-gas
709 -mno-jump-in-delay -mno-long-load-store
710 -mno-portable-runtime -mno-soft-float
711 -mno-space-regs -msoft-float -mpa-risc-1-0
712 -mpa-risc-1-1 -mpa-risc-2-0 -mportable-runtime
713 -mschedule=CPU-TYPE -mspace-regs -msio -mwsio
714 -munix=UNIX-STD -nolibdld -static -threads
716 _i386 and x86-64 Options_
717 -mtune=CPU-TYPE -march=CPU-TYPE
719 -masm=DIALECT -mno-fancy-math-387
720 -mno-fp-ret-in-387 -msoft-float -msvr3-shlib
721 -mno-wide-multiply -mrtd -malign-double
722 -mpreferred-stack-boundary=NUM
723 -mmmx -msse -msse2 -msse3 -m3dnow
724 -mthreads -mno-align-stringops -minline-all-stringops
725 -mpush-args -maccumulate-outgoing-args -m128bit-long-double
726 -m96bit-long-double -mregparm=NUM -msseregparm
727 -momit-leaf-frame-pointer -mno-red-zone -mno-tls-direct-seg-refs
729 -m32 -m64 -mlarge-data-threshold=NUM
732 -mbig-endian -mlittle-endian -mgnu-as -mgnu-ld -mno-pic
733 -mvolatile-asm-stop -mregister-names -mno-sdata
734 -mconstant-gp -mauto-pic -minline-float-divide-min-latency
735 -minline-float-divide-max-throughput
736 -minline-int-divide-min-latency
737 -minline-int-divide-max-throughput
738 -minline-sqrt-min-latency -minline-sqrt-max-throughput
739 -mno-dwarf2-asm -mearly-stop-bits
740 -mfixed-range=REGISTER-RANGE -mtls-size=TLS-SIZE
741 -mtune=CPU-TYPE -mt -pthread -milp32 -mlp64
746 -malign-loops -mno-align-loops
749 -mmodel=CODE-SIZE-MODEL-TYPE
751 -mno-flush-func -mflush-func=NAME
752 -mno-flush-trap -mflush-trap=NUMBER
756 -mcpu=CPU -msim -memregs=NUMBER
759 -m68000 -m68020 -m68020-40 -m68020-60 -m68030 -m68040
760 -m68060 -mcpu32 -m5200 -m68881 -mbitfield -mc68000 -mc68020
761 -mnobitfield -mrtd -mshort -msoft-float -mpcrel
762 -malign-int -mstrict-align -msep-data -mno-sep-data
763 -mshared-library-id=n -mid-shared-library -mno-id-shared-library
766 -m6811 -m6812 -m68hc11 -m68hc12 -m68hcs12
767 -mauto-incdec -minmax -mlong-calls -mshort
768 -msoft-reg-count=COUNT
771 -mhardlit -mno-hardlit -mdiv -mno-div -mrelax-immediates
772 -mno-relax-immediates -mwide-bitfields -mno-wide-bitfields
773 -m4byte-functions -mno-4byte-functions -mcallgraph-data
774 -mno-callgraph-data -mslow-bytes -mno-slow-bytes -mno-lsim
775 -mlittle-endian -mbig-endian -m210 -m340 -mstack-increment
778 -EL -EB -march=ARCH -mtune=ARCH
779 -mips1 -mips2 -mips3 -mips4 -mips32 -mips32r2 -mips64
780 -mips16 -mno-mips16 -mabi=ABI -mabicalls -mno-abicalls
781 -mxgot -mno-xgot -mgp32 -mgp64 -mfp32 -mfp64
782 -mhard-float -msoft-float -msingle-float -mdouble-float
783 -mdsp -mpaired-single -mips3d
784 -mlong64 -mlong32 -msym32 -mno-sym32
785 -GNUM -membedded-data -mno-embedded-data
786 -muninit-const-in-rodata -mno-uninit-const-in-rodata
787 -msplit-addresses -mno-split-addresses
788 -mexplicit-relocs -mno-explicit-relocs
789 -mcheck-zero-division -mno-check-zero-division
790 -mdivide-traps -mdivide-breaks
791 -mmemcpy -mno-memcpy -mlong-calls -mno-long-calls
792 -mmad -mno-mad -mfused-madd -mno-fused-madd -nocpp
793 -mfix-r4000 -mno-fix-r4000 -mfix-r4400 -mno-fix-r4400
794 -mfix-vr4120 -mno-fix-vr4120 -mfix-vr4130
795 -mfix-sb1 -mno-fix-sb1
796 -mflush-func=FUNC -mno-flush-func
797 -mbranch-likely -mno-branch-likely
798 -mfp-exceptions -mno-fp-exceptions
799 -mvr4130-align -mno-vr4130-align
802 -mlibfuncs -mno-libfuncs -mepsilon -mno-epsilon -mabi=gnu
803 -mabi=mmixware -mzero-extend -mknuthdiv -mtoplevel-symbols
804 -melf -mbranch-predict -mno-branch-predict -mbase-addresses
805 -mno-base-addresses -msingle-exit -mno-single-exit
808 -mmult-bug -mno-mult-bug
811 -mreturn-pointer-on-d0
815 -mno-crt0 -mbacc -msim
819 -mfpu -msoft-float -mac0 -mno-ac0 -m40 -m45 -m10
820 -mbcopy -mbcopy-builtin -mint32 -mno-int16
821 -mint16 -mno-int32 -mfloat32 -mno-float64
822 -mfloat64 -mno-float32 -mabshi -mno-abshi
823 -mbranch-expensive -mbranch-cheap
824 -msplit -mno-split -munix-asm -mdec-asm
826 _PowerPC Options_ See RS/6000 and PowerPC Options.
828 _RS/6000 and PowerPC Options_
831 -mpower -mno-power -mpower2 -mno-power2
832 -mpowerpc -mpowerpc64 -mno-powerpc
833 -maltivec -mno-altivec
834 -mpowerpc-gpopt -mno-powerpc-gpopt
835 -mpowerpc-gfxopt -mno-powerpc-gfxopt
836 -mmfcrf -mno-mfcrf -mpopcntb -mno-popcntb -mfprnd -mno-fprnd
837 -mnew-mnemonics -mold-mnemonics
838 -mfull-toc -mminimal-toc -mno-fp-in-toc -mno-sum-in-toc
839 -m64 -m32 -mxl-compat -mno-xl-compat -mpe
840 -malign-power -malign-natural
841 -msoft-float -mhard-float -mmultiple -mno-multiple
842 -mstring -mno-string -mupdate -mno-update
843 -mfused-madd -mno-fused-madd -mbit-align -mno-bit-align
844 -mstrict-align -mno-strict-align -mrelocatable
845 -mno-relocatable -mrelocatable-lib -mno-relocatable-lib
846 -mtoc -mno-toc -mlittle -mlittle-endian -mbig -mbig-endian
847 -mdynamic-no-pic -maltivec -mswdiv
848 -mprioritize-restricted-insns=PRIORITY
849 -msched-costly-dep=DEPENDENCE_TYPE
850 -minsert-sched-nops=SCHEME
851 -mcall-sysv -mcall-netbsd
852 -maix-struct-return -msvr4-struct-return
853 -mabi=ABI-TYPE -msecure-plt -mbss-plt
859 -mfloat-gprs=yes -mfloat-gprs=no -mfloat-gprs=single -mfloat-gprs=double
860 -mprototype -mno-prototype
861 -msim -mmvme -mads -myellowknife -memb -msdata
862 -msdata=OPT -mvxworks -mwindiss -G NUM -pthread
864 _S/390 and zSeries Options_
865 -mtune=CPU-TYPE -march=CPU-TYPE
866 -mhard-float -msoft-float -mlong-double-64 -mlong-double-128
867 -mbackchain -mno-backchain -mpacked-stack -mno-packed-stack
868 -msmall-exec -mno-small-exec -mmvcle -mno-mvcle
869 -m64 -m31 -mdebug -mno-debug -mesa -mzarch
870 -mtpf-trace -mno-tpf-trace -mfused-madd -mno-fused-madd
871 -mwarn-framesize -mwarn-dynamicstack -mstack-size -mstack-guard
874 -m1 -m2 -m2e -m3 -m3e
875 -m4-nofpu -m4-single-only -m4-single -m4
876 -m4a-nofpu -m4a-single-only -m4a-single -m4a -m4al
877 -m5-64media -m5-64media-nofpu
878 -m5-32media -m5-32media-nofpu
879 -m5-compact -m5-compact-nofpu
880 -mb -ml -mdalign -mrelax
881 -mbigtable -mfmovd -mhitachi -mrenesas -mno-renesas -mnomacsave
882 -mieee -misize -mpadstruct -mspace
883 -mprefergot -musermode -multcost=NUMBER -mdiv=STRATEGY
884 -mdivsi3_libfunc=NAME
885 -madjust-unroll -mindexed-addressing -mgettrcost=NUMBER -mpt-fixed
892 -m32 -m64 -mapp-regs -mno-app-regs
893 -mfaster-structs -mno-faster-structs
894 -mfpu -mno-fpu -mhard-float -msoft-float
895 -mhard-quad-float -msoft-quad-float
896 -mimpure-text -mno-impure-text -mlittle-endian
897 -mstack-bias -mno-stack-bias
898 -munaligned-doubles -mno-unaligned-doubles
899 -mv8plus -mno-v8plus -mvis -mno-vis
900 -threads -pthreads -pthread
903 -Qy -Qn -YP,PATHS -Ym,DIR
905 _TMS320C3x/C4x Options_
906 -mcpu=CPU -mbig -msmall -mregparm -mmemparm
907 -mfast-fix -mmpyi -mbk -mti -mdp-isr-reload
908 -mrpts=COUNT -mrptb -mdb -mloop-unsigned
909 -mparallel-insns -mparallel-mpy -mpreserve-float
912 -mlong-calls -mno-long-calls -mep -mno-ep
913 -mprolog-function -mno-prolog-function -mspace
914 -mtda=N -msda=N -mzda=N
915 -mapp-regs -mno-app-regs
916 -mdisable-callt -mno-disable-callt
924 _x86-64 Options_ See i386 and x86-64 Options.
930 -mconst16 -mno-const16
931 -mfused-madd -mno-fused-madd
932 -mtext-section-literals -mno-text-section-literals
933 -mtarget-align -mno-target-align
934 -mlongcalls -mno-longcalls
936 _zSeries Options_ See S/390 and zSeries Options.
938 _Code Generation Options_
939 *Note Options for Code Generation Conventions: Code Gen Options.
940 -fcall-saved-REG -fcall-used-REG
941 -ffixed-REG -fexceptions
942 -fnon-call-exceptions -funwind-tables
943 -fasynchronous-unwind-tables
944 -finhibit-size-directive -finstrument-functions
945 -fno-common -fno-ident
946 -fpcc-struct-return -fpic -fPIC -fpie -fPIE
948 -freg-struct-return -fshared-data -fshort-enums
949 -fshort-double -fshort-wchar
950 -fverbose-asm -fpack-struct[=N] -fstack-check
951 -fstack-limit-register=REG -fstack-limit-symbol=SYM
952 -fargument-alias -fargument-noalias
953 -fargument-noalias-global -fleading-underscore
955 -ftrapv -fwrapv -fbounds-check
961 * Overall Options:: Controlling the kind of output:
962 an executable, object files, assembler files,
963 or preprocessed source.
964 * C Dialect Options:: Controlling the variant of C language compiled.
965 * C++ Dialect Options:: Variations on C++.
966 * Objective-C and Objective-C++ Dialect Options:: Variations on Objective-C
968 * Language Independent Options:: Controlling how diagnostics should be
970 * Warning Options:: How picky should the compiler be?
971 * Debugging Options:: Symbol tables, measurements, and debugging dumps.
972 * Optimize Options:: How much optimization?
973 * Preprocessor Options:: Controlling header files and macro definitions.
974 Also, getting dependency information for Make.
975 * Assembler Options:: Passing options to the assembler.
976 * Link Options:: Specifying libraries and so on.
977 * Directory Options:: Where to find header files and libraries.
978 Where to find the compiler executable files.
979 * Spec Files:: How to pass switches to sub-processes.
980 * Target Options:: Running a cross-compiler, or an old version of GCC.
983 File: gcc.info, Node: Overall Options, Next: Invoking G++, Prev: Option Summary, Up: Invoking GCC
985 3.2 Options Controlling the Kind of Output
986 ==========================================
988 Compilation can involve up to four stages: preprocessing, compilation
989 proper, assembly and linking, always in that order. GCC is capable of
990 preprocessing and compiling several files either into several assembler
991 input files, or into one assembler input file; then each assembler
992 input file produces an object file, and linking combines all the object
993 files (those newly compiled, and those specified as input) into an
996 For any given input file, the file name suffix determines what kind of
1000 C source code which must be preprocessed.
1003 C source code which should not be preprocessed.
1006 C++ source code which should not be preprocessed.
1009 Objective-C source code. Note that you must link with the
1010 `libobjc' library to make an Objective-C program work.
1013 Objective-C source code which should not be preprocessed.
1017 Objective-C++ source code. Note that you must link with the
1018 `libobjc' library to make an Objective-C++ program work. Note
1019 that `.M' refers to a literal capital M.
1022 Objective-C++ source code which should not be preprocessed.
1025 C, C++, Objective-C or Objective-C++ header file to be turned into
1026 a precompiled header.
1035 C++ source code which must be preprocessed. Note that in `.cxx',
1036 the last two letters must both be literally `x'. Likewise, `.C'
1037 refers to a literal capital C.
1041 Objective-C++ source code which must be preprocessed.
1044 Objective-C++ source code which should not be preprocessed.
1048 C++ header file to be turned into a precompiled header.
1053 Fixed form Fortran source code which should not be preprocessed.
1058 Fixed form Fortran source code which must be preprocessed (with
1059 the traditional preprocessor).
1063 Free form Fortran source code which should not be preprocessed.
1067 Free form Fortran source code which must be preprocessed (with the
1068 traditional preprocessor).
1071 Ada source code file which contains a library unit declaration (a
1072 declaration of a package, subprogram, or generic, or a generic
1073 instantiation), or a library unit renaming declaration (a package,
1074 generic, or subprogram renaming declaration). Such files are also
1078 Ada source code file containing a library unit body (a subprogram
1079 or package body). Such files are also called "bodies".
1085 Assembler code which must be preprocessed.
1088 An object file to be fed straight into linking. Any file name
1089 with no recognized suffix is treated this way.
1091 You can specify the input language explicitly with the `-x' option:
1094 Specify explicitly the LANGUAGE for the following input files
1095 (rather than letting the compiler choose a default based on the
1096 file name suffix). This option applies to all following input
1097 files until the next `-x' option. Possible values for LANGUAGE
1099 c c-header c-cpp-output
1100 c++ c++-header c++-cpp-output
1101 objective-c objective-c-header objective-c-cpp-output
1102 objective-c++ objective-c++-header objective-c++-cpp-output
1103 assembler assembler-with-cpp
1110 Turn off any specification of a language, so that subsequent files
1111 are handled according to their file name suffixes (as they are if
1112 `-x' has not been used at all).
1115 Normally the `gcc' program will exit with the code of 1 if any
1116 phase of the compiler returns a non-success return code. If you
1117 specify `-pass-exit-codes', the `gcc' program will instead return
1118 with numerically highest error produced by any phase that returned
1119 an error indication.
1121 If you only want some of the stages of compilation, you can use `-x'
1122 (or filename suffixes) to tell `gcc' where to start, and one of the
1123 options `-c', `-S', or `-E' to say where `gcc' is to stop. Note that
1124 some combinations (for example, `-x cpp-output -E') instruct `gcc' to
1128 Compile or assemble the source files, but do not link. The linking
1129 stage simply is not done. The ultimate output is in the form of an
1130 object file for each source file.
1132 By default, the object file name for a source file is made by
1133 replacing the suffix `.c', `.i', `.s', etc., with `.o'.
1135 Unrecognized input files, not requiring compilation or assembly,
1139 Stop after the stage of compilation proper; do not assemble. The
1140 output is in the form of an assembler code file for each
1141 non-assembler input file specified.
1143 By default, the assembler file name for a source file is made by
1144 replacing the suffix `.c', `.i', etc., with `.s'.
1146 Input files that don't require compilation are ignored.
1149 Stop after the preprocessing stage; do not run the compiler
1150 proper. The output is in the form of preprocessed source code,
1151 which is sent to the standard output.
1153 Input files which don't require preprocessing are ignored.
1156 Place output in file FILE. This applies regardless to whatever
1157 sort of output is being produced, whether it be an executable file,
1158 an object file, an assembler file or preprocessed C code.
1160 If `-o' is not specified, the default is to put an executable file
1161 in `a.out', the object file for `SOURCE.SUFFIX' in `SOURCE.o', its
1162 assembler file in `SOURCE.s', a precompiled header file in
1163 `SOURCE.SUFFIX.gch', and all preprocessed C source on standard
1167 Print (on standard error output) the commands executed to run the
1168 stages of compilation. Also print the version number of the
1169 compiler driver program and of the preprocessor and the compiler
1173 Like `-v' except the commands are not executed and all command
1174 arguments are quoted. This is useful for shell scripts to capture
1175 the driver-generated command lines.
1178 Use pipes rather than temporary files for communication between the
1179 various stages of compilation. This fails to work on some systems
1180 where the assembler is unable to read from a pipe; but the GNU
1181 assembler has no trouble.
1184 If you are compiling multiple source files, this option tells the
1185 driver to pass all the source files to the compiler at once (for
1186 those languages for which the compiler can handle this). This
1187 will allow intermodule analysis (IMA) to be performed by the
1188 compiler. Currently the only language for which this is supported
1189 is C. If you pass source files for multiple languages to the
1190 driver, using this option, the driver will invoke the compiler(s)
1191 that support IMA once each, passing each compiler all the source
1192 files appropriate for it. For those languages that do not support
1193 IMA this option will be ignored, and the compiler will be invoked
1194 once for each source file in that language. If you use this
1195 option in conjunction with `-save-temps', the compiler will
1196 generate multiple pre-processed files (one for each source file),
1197 but only one (combined) `.o' or `.s' file.
1200 Print (on the standard output) a description of the command line
1201 options understood by `gcc'. If the `-v' option is also specified
1202 then `--help' will also be passed on to the various processes
1203 invoked by `gcc', so that they can display the command line options
1204 they accept. If the `-Wextra' option is also specified then
1205 command line options which have no documentation associated with
1206 them will also be displayed.
1209 Print (on the standard output) a description of target specific
1210 command line options for each tool.
1213 Display the version number and copyrights of the invoked GCC.
1216 File: gcc.info, Node: Invoking G++, Next: C Dialect Options, Prev: Overall Options, Up: Invoking GCC
1218 3.3 Compiling C++ Programs
1219 ==========================
1221 C++ source files conventionally use one of the suffixes `.C', `.cc',
1222 `.cpp', `.CPP', `.c++', `.cp', or `.cxx'; C++ header files often use
1223 `.hh' or `.H'; and preprocessed C++ files use the suffix `.ii'. GCC
1224 recognizes files with these names and compiles them as C++ programs
1225 even if you call the compiler the same way as for compiling C programs
1226 (usually with the name `gcc').
1228 However, C++ programs often require class libraries as well as a
1229 compiler that understands the C++ language--and under some
1230 circumstances, you might want to compile programs or header files from
1231 standard input, or otherwise without a suffix that flags them as C++
1232 programs. You might also like to precompile a C header file with a
1233 `.h' extension to be used in C++ compilations. `g++' is a program that
1234 calls GCC with the default language set to C++, and automatically
1235 specifies linking against the C++ library. On many systems, `g++' is
1236 also installed with the name `c++'.
1238 When you compile C++ programs, you may specify many of the same
1239 command-line options that you use for compiling programs in any
1240 language; or command-line options meaningful for C and related
1241 languages; or options that are meaningful only for C++ programs. *Note
1242 Options Controlling C Dialect: C Dialect Options, for explanations of
1243 options for languages related to C. *Note Options Controlling C++
1244 Dialect: C++ Dialect Options, for explanations of options that are
1245 meaningful only for C++ programs.
1248 File: gcc.info, Node: C Dialect Options, Next: C++ Dialect Options, Prev: Invoking G++, Up: Invoking GCC
1250 3.4 Options Controlling C Dialect
1251 =================================
1253 The following options control the dialect of C (or languages derived
1254 from C, such as C++, Objective-C and Objective-C++) that the compiler
1258 In C mode, support all ISO C90 programs. In C++ mode, remove GNU
1259 extensions that conflict with ISO C++.
1261 This turns off certain features of GCC that are incompatible with
1262 ISO C90 (when compiling C code), or of standard C++ (when
1263 compiling C++ code), such as the `asm' and `typeof' keywords, and
1264 predefined macros such as `unix' and `vax' that identify the type
1265 of system you are using. It also enables the undesirable and
1266 rarely used ISO trigraph feature. For the C compiler, it disables
1267 recognition of C++ style `//' comments as well as the `inline'
1270 The alternate keywords `__asm__', `__extension__', `__inline__'
1271 and `__typeof__' continue to work despite `-ansi'. You would not
1272 want to use them in an ISO C program, of course, but it is useful
1273 to put them in header files that might be included in compilations
1274 done with `-ansi'. Alternate predefined macros such as `__unix__'
1275 and `__vax__' are also available, with or without `-ansi'.
1277 The `-ansi' option does not cause non-ISO programs to be rejected
1278 gratuitously. For that, `-pedantic' is required in addition to
1279 `-ansi'. *Note Warning Options::.
1281 The macro `__STRICT_ANSI__' is predefined when the `-ansi' option
1282 is used. Some header files may notice this macro and refrain from
1283 declaring certain functions or defining certain macros that the
1284 ISO standard doesn't call for; this is to avoid interfering with
1285 any programs that might use these names for other things.
1287 Functions which would normally be built in but do not have
1288 semantics defined by ISO C (such as `alloca' and `ffs') are not
1289 built-in functions with `-ansi' is used. *Note Other built-in
1290 functions provided by GCC: Other Builtins, for details of the
1294 Determine the language standard. This option is currently only
1295 supported when compiling C or C++. A value for this option must be
1296 provided; possible values are
1300 ISO C90 (same as `-ansi').
1303 ISO C90 as modified in amendment 1.
1309 ISO C99. Note that this standard is not yet fully supported;
1310 see `http://gcc.gnu.org/gcc-4.1/c99status.html' for more
1311 information. The names `c9x' and `iso9899:199x' are
1315 Default, ISO C90 plus GNU extensions (including some C99
1320 ISO C99 plus GNU extensions. When ISO C99 is fully
1321 implemented in GCC, this will become the default. The name
1322 `gnu9x' is deprecated.
1325 The 1998 ISO C++ standard plus amendments.
1328 The same as `-std=c++98' plus GNU extensions. This is the
1329 default for C++ code.
1331 Even when this option is not specified, you can still use some of
1332 the features of newer standards in so far as they do not conflict
1333 with previous C standards. For example, you may use
1334 `__restrict__' even when `-std=c99' is not specified.
1336 The `-std' options specifying some version of ISO C have the same
1337 effects as `-ansi', except that features that were not in ISO C90
1338 but are in the specified version (for example, `//' comments and
1339 the `inline' keyword in ISO C99) are not disabled.
1341 *Note Language Standards Supported by GCC: Standards, for details
1342 of these standard versions.
1344 `-aux-info FILENAME'
1345 Output to the given filename prototyped declarations for all
1346 functions declared and/or defined in a translation unit, including
1347 those in header files. This option is silently ignored in any
1348 language other than C.
1350 Besides declarations, the file indicates, in comments, the origin
1351 of each declaration (source file and line), whether the
1352 declaration was implicit, prototyped or unprototyped (`I', `N' for
1353 new or `O' for old, respectively, in the first character after the
1354 line number and the colon), and whether it came from a declaration
1355 or a definition (`C' or `F', respectively, in the following
1356 character). In the case of function definitions, a K&R-style list
1357 of arguments followed by their declarations is also provided,
1358 inside comments, after the declaration.
1361 Do not recognize `asm', `inline' or `typeof' as a keyword, so that
1362 code can use these words as identifiers. You can use the keywords
1363 `__asm__', `__inline__' and `__typeof__' instead. `-ansi' implies
1366 In C++, this switch only affects the `typeof' keyword, since `asm'
1367 and `inline' are standard keywords. You may want to use the
1368 `-fno-gnu-keywords' flag instead, which has the same effect. In
1369 C99 mode (`-std=c99' or `-std=gnu99'), this switch only affects
1370 the `asm' and `typeof' keywords, since `inline' is a standard
1374 `-fno-builtin-FUNCTION'
1375 Don't recognize built-in functions that do not begin with
1376 `__builtin_' as prefix. *Note Other built-in functions provided
1377 by GCC: Other Builtins, for details of the functions affected,
1378 including those which are not built-in functions when `-ansi' or
1379 `-std' options for strict ISO C conformance are used because they
1380 do not have an ISO standard meaning.
1382 GCC normally generates special code to handle certain built-in
1383 functions more efficiently; for instance, calls to `alloca' may
1384 become single instructions that adjust the stack directly, and
1385 calls to `memcpy' may become inline copy loops. The resulting
1386 code is often both smaller and faster, but since the function
1387 calls no longer appear as such, you cannot set a breakpoint on
1388 those calls, nor can you change the behavior of the functions by
1389 linking with a different library. In addition, when a function is
1390 recognized as a built-in function, GCC may use information about
1391 that function to warn about problems with calls to that function,
1392 or to generate more efficient code, even if the resulting code
1393 still contains calls to that function. For example, warnings are
1394 given with `-Wformat' for bad calls to `printf', when `printf' is
1395 built in, and `strlen' is known not to modify global memory.
1397 With the `-fno-builtin-FUNCTION' option only the built-in function
1398 FUNCTION is disabled. FUNCTION must not begin with `__builtin_'.
1399 If a function is named this is not built-in in this version of
1400 GCC, this option is ignored. There is no corresponding
1401 `-fbuiltin-FUNCTION' option; if you wish to enable built-in
1402 functions selectively when using `-fno-builtin' or
1403 `-ffreestanding', you may define macros such as:
1405 #define abs(n) __builtin_abs ((n))
1406 #define strcpy(d, s) __builtin_strcpy ((d), (s))
1409 Assert that compilation takes place in a hosted environment. This
1410 implies `-fbuiltin'. A hosted environment is one in which the
1411 entire standard library is available, and in which `main' has a
1412 return type of `int'. Examples are nearly everything except a
1413 kernel. This is equivalent to `-fno-freestanding'.
1416 Assert that compilation takes place in a freestanding environment.
1417 This implies `-fno-builtin'. A freestanding environment is one
1418 in which the standard library may not exist, and program startup
1419 may not necessarily be at `main'. The most obvious example is an
1420 OS kernel. This is equivalent to `-fno-hosted'.
1422 *Note Language Standards Supported by GCC: Standards, for details
1423 of freestanding and hosted environments.
1426 Accept some non-standard constructs used in Microsoft header files.
1428 Some cases of unnamed fields in structures and unions are only
1429 accepted with this option. *Note Unnamed struct/union fields
1430 within structs/unions: Unnamed Fields, for details.
1433 Support ISO C trigraphs. The `-ansi' option (and `-std' options
1434 for strict ISO C conformance) implies `-trigraphs'.
1436 `-no-integrated-cpp'
1437 Performs a compilation in two passes: preprocessing and compiling.
1438 This option allows a user supplied "cc1", "cc1plus", or "cc1obj"
1439 via the `-B' option. The user supplied compilation step can then
1440 add in an additional preprocessing step after normal preprocessing
1441 but before compiling. The default is to use the integrated cpp
1444 The semantics of this option will change if "cc1", "cc1plus", and
1445 "cc1obj" are merged.
1449 Formerly, these options caused GCC to attempt to emulate a
1450 pre-standard C compiler. They are now only supported with the
1451 `-E' switch. The preprocessor continues to support a pre-standard
1452 mode. See the GNU CPP manual for details.
1455 Allow conditional expressions with mismatched types in the second
1456 and third arguments. The value of such an expression is void.
1457 This option is not supported for C++.
1460 Let the type `char' be unsigned, like `unsigned char'.
1462 Each kind of machine has a default for what `char' should be. It
1463 is either like `unsigned char' by default or like `signed char' by
1466 Ideally, a portable program should always use `signed char' or
1467 `unsigned char' when it depends on the signedness of an object.
1468 But many programs have been written to use plain `char' and expect
1469 it to be signed, or expect it to be unsigned, depending on the
1470 machines they were written for. This option, and its inverse, let
1471 you make such a program work with the opposite default.
1473 The type `char' is always a distinct type from each of `signed
1474 char' or `unsigned char', even though its behavior is always just
1475 like one of those two.
1478 Let the type `char' be signed, like `signed char'.
1480 Note that this is equivalent to `-fno-unsigned-char', which is the
1481 negative form of `-funsigned-char'. Likewise, the option
1482 `-fno-signed-char' is equivalent to `-funsigned-char'.
1484 `-fsigned-bitfields'
1485 `-funsigned-bitfields'
1486 `-fno-signed-bitfields'
1487 `-fno-unsigned-bitfields'
1488 These options control whether a bit-field is signed or unsigned,
1489 when the declaration does not use either `signed' or `unsigned'.
1490 By default, such a bit-field is signed, because this is
1491 consistent: the basic integer types such as `int' are signed types.
1494 File: gcc.info, Node: C++ Dialect Options, Next: Objective-C and Objective-C++ Dialect Options, Prev: C Dialect Options, Up: Invoking GCC
1496 3.5 Options Controlling C++ Dialect
1497 ===================================
1499 This section describes the command-line options that are only meaningful
1500 for C++ programs; but you can also use most of the GNU compiler options
1501 regardless of what language your program is in. For example, you might
1502 compile a file `firstClass.C' like this:
1504 g++ -g -frepo -O -c firstClass.C
1506 In this example, only `-frepo' is an option meant only for C++
1507 programs; you can use the other options with any language supported by
1510 Here is a list of options that are _only_ for compiling C++ programs:
1513 Use version N of the C++ ABI. Version 2 is the version of the C++
1514 ABI that first appeared in G++ 3.4. Version 1 is the version of
1515 the C++ ABI that first appeared in G++ 3.2. Version 0 will always
1516 be the version that conforms most closely to the C++ ABI
1517 specification. Therefore, the ABI obtained using version 0 will
1518 change as ABI bugs are fixed.
1520 The default is version 2.
1522 `-fno-access-control'
1523 Turn off all access checking. This switch is mainly useful for
1524 working around bugs in the access control code.
1527 Check that the pointer returned by `operator new' is non-null
1528 before attempting to modify the storage allocated. This check is
1529 normally unnecessary because the C++ standard specifies that
1530 `operator new' will only return `0' if it is declared `throw()',
1531 in which case the compiler will always check the return value even
1532 without this option. In all other cases, when `operator new' has
1533 a non-empty exception specification, memory exhaustion is
1534 signalled by throwing `std::bad_alloc'. See also `new (nothrow)'.
1537 Put uninitialized or runtime-initialized global variables into the
1538 common segment, as C does. This saves space in the executable at
1539 the cost of not diagnosing duplicate definitions. If you compile
1540 with this flag and your program mysteriously crashes after
1541 `main()' has completed, you may have an object that is being
1542 destroyed twice because two definitions were merged.
1544 This option is no longer useful on most targets, now that support
1545 has been added for putting variables into BSS without making them
1548 `-ffriend-injection'
1549 Inject friend functions into the enclosing namespace, so that they
1550 are visible outside the scope of the class in which they are
1551 declared. Friend functions were documented to work this way in
1552 the old Annotated C++ Reference Manual, and versions of G++ before
1553 4.1 always worked that way. However, in ISO C++ a friend function
1554 which is not declared in an enclosing scope can only be found
1555 using argument dependent lookup. This option causes friends to be
1556 injected as they were in earlier releases.
1558 This option is for compatibility, and may be removed in a future
1561 `-fno-const-strings'
1562 Give string constants type `char *' instead of type `const char
1563 *'. By default, G++ uses type `const char *' as required by the
1564 standard. Even if you use `-fno-const-strings', you cannot
1565 actually modify the value of a string constant.
1567 This option might be removed in a future release of G++. For
1568 maximum portability, you should structure your code so that it
1569 works with string constants that have type `const char *'.
1571 `-fno-elide-constructors'
1572 The C++ standard allows an implementation to omit creating a
1573 temporary which is only used to initialize another object of the
1574 same type. Specifying this option disables that optimization, and
1575 forces G++ to call the copy constructor in all cases.
1577 `-fno-enforce-eh-specs'
1578 Don't generate code to check for violation of exception
1579 specifications at runtime. This option violates the C++ standard,
1580 but may be useful for reducing code size in production builds,
1581 much like defining `NDEBUG'. This does not give user code
1582 permission to throw exceptions in violation of the exception
1583 specifications; the compiler will still optimize based on the
1584 specifications, so throwing an unexpected exception will result in
1589 If `-ffor-scope' is specified, the scope of variables declared in
1590 a for-init-statement is limited to the `for' loop itself, as
1591 specified by the C++ standard. If `-fno-for-scope' is specified,
1592 the scope of variables declared in a for-init-statement extends to
1593 the end of the enclosing scope, as was the case in old versions of
1594 G++, and other (traditional) implementations of C++.
1596 The default if neither flag is given to follow the standard, but
1597 to allow and give a warning for old-style code that would
1598 otherwise be invalid, or have different behavior.
1601 Do not recognize `typeof' as a keyword, so that code can use this
1602 word as an identifier. You can use the keyword `__typeof__'
1603 instead. `-ansi' implies `-fno-gnu-keywords'.
1605 `-fno-implicit-templates'
1606 Never emit code for non-inline templates which are instantiated
1607 implicitly (i.e. by use); only emit code for explicit
1608 instantiations. *Note Template Instantiation::, for more
1611 `-fno-implicit-inline-templates'
1612 Don't emit code for implicit instantiations of inline templates,
1613 either. The default is to handle inlines differently so that
1614 compiles with and without optimization will need the same set of
1615 explicit instantiations.
1617 `-fno-implement-inlines'
1618 To save space, do not emit out-of-line copies of inline functions
1619 controlled by `#pragma implementation'. This will cause linker
1620 errors if these functions are not inlined everywhere they are
1624 Disable pedantic warnings about constructs used in MFC, such as
1625 implicit int and getting a pointer to member function via
1626 non-standard syntax.
1628 `-fno-nonansi-builtins'
1629 Disable built-in declarations of functions that are not mandated by
1630 ANSI/ISO C. These include `ffs', `alloca', `_exit', `index',
1631 `bzero', `conjf', and other related functions.
1633 `-fno-operator-names'
1634 Do not treat the operator name keywords `and', `bitand', `bitor',
1635 `compl', `not', `or' and `xor' as synonyms as keywords.
1637 `-fno-optional-diags'
1638 Disable diagnostics that the standard says a compiler does not
1639 need to issue. Currently, the only such diagnostic issued by G++
1640 is the one for a name having multiple meanings within a class.
1643 Downgrade some diagnostics about nonconformant code from errors to
1644 warnings. Thus, using `-fpermissive' will allow some
1645 nonconforming code to compile.
1648 Enable automatic template instantiation at link time. This option
1649 also implies `-fno-implicit-templates'. *Note Template
1650 Instantiation::, for more information.
1653 Disable generation of information about every class with virtual
1654 functions for use by the C++ runtime type identification features
1655 (`dynamic_cast' and `typeid'). If you don't use those parts of
1656 the language, you can save some space by using this flag. Note
1657 that exception handling uses the same information, but it will
1658 generate it as needed.
1661 Emit statistics about front-end processing at the end of the
1662 compilation. This information is generally only useful to the G++
1665 `-ftemplate-depth-N'
1666 Set the maximum instantiation depth for template classes to N. A
1667 limit on the template instantiation depth is needed to detect
1668 endless recursions during template class instantiation. ANSI/ISO
1669 C++ conforming programs must not rely on a maximum depth greater
1672 `-fno-threadsafe-statics'
1673 Do not emit the extra code to use the routines specified in the C++
1674 ABI for thread-safe initialization of local statics. You can use
1675 this option to reduce code size slightly in code that doesn't need
1679 Register destructors for objects with static storage duration with
1680 the `__cxa_atexit' function rather than the `atexit' function.
1681 This option is required for fully standards-compliant handling of
1682 static destructors, but will only work if your C library supports
1685 `-fvisibility-inlines-hidden'
1686 Causes all inlined methods to be marked with `__attribute__
1687 ((visibility ("hidden")))' so that they do not appear in the
1688 export table of a DSO and do not require a PLT indirection when
1689 used within the DSO. Enabling this option can have a dramatic
1690 effect on load and link times of a DSO as it massively reduces the
1691 size of the dynamic export table when the library makes heavy use
1692 of templates. While it can cause bloating through duplication of
1693 code within each DSO where it is used, often the wastage is less
1694 than the considerable space occupied by a long symbol name in the
1695 export table which is typical when using templates and namespaces.
1696 For even more savings, combine with the `-fvisibility=hidden'
1700 Do not use weak symbol support, even if it is provided by the
1701 linker. By default, G++ will use weak symbols if they are
1702 available. This option exists only for testing, and should not be
1703 used by end-users; it will result in inferior code and has no
1704 benefits. This option may be removed in a future release of G++.
1707 Do not search for header files in the standard directories
1708 specific to C++, but do still search the other standard
1709 directories. (This option is used when building the C++ library.)
1711 In addition, these optimization, warning, and code generation options
1712 have meanings only for C++ programs:
1714 `-fno-default-inline'
1715 Do not assume `inline' for functions defined inside a class scope.
1716 *Note Options That Control Optimization: Optimize Options. Note
1717 that these functions will have linkage like inline functions; they
1718 just won't be inlined by default.
1721 Warn when G++ generates code that is probably not compatible with
1722 the vendor-neutral C++ ABI. Although an effort has been made to
1723 warn about all such cases, there are probably some cases that are
1724 not warned about, even though G++ is generating incompatible code.
1725 There may also be cases where warnings are emitted even though
1726 the code that is generated will be compatible.
1728 You should rewrite your code to avoid these warnings if you are
1729 concerned about the fact that code generated by G++ may not be
1730 binary compatible with code generated by other compilers.
1732 The known incompatibilities at this point include:
1734 * Incorrect handling of tail-padding for bit-fields. G++ may
1735 attempt to pack data into the same byte as a base class. For
1738 struct A { virtual void f(); int f1 : 1; };
1739 struct B : public A { int f2 : 1; };
1741 In this case, G++ will place `B::f2' into the same byte
1742 as`A::f1'; other compilers will not. You can avoid this
1743 problem by explicitly padding `A' so that its size is a
1744 multiple of the byte size on your platform; that will cause
1745 G++ and other compilers to layout `B' identically.
1747 * Incorrect handling of tail-padding for virtual bases. G++
1748 does not use tail padding when laying out virtual bases. For
1751 struct A { virtual void f(); char c1; };
1752 struct B { B(); char c2; };
1753 struct C : public A, public virtual B {};
1755 In this case, G++ will not place `B' into the tail-padding for
1756 `A'; other compilers will. You can avoid this problem by
1757 explicitly padding `A' so that its size is a multiple of its
1758 alignment (ignoring virtual base classes); that will cause
1759 G++ and other compilers to layout `C' identically.
1761 * Incorrect handling of bit-fields with declared widths greater
1762 than that of their underlying types, when the bit-fields
1763 appear in a union. For example:
1765 union U { int i : 4096; };
1767 Assuming that an `int' does not have 4096 bits, G++ will make
1768 the union too small by the number of bits in an `int'.
1770 * Empty classes can be placed at incorrect offsets. For
1780 struct C : public B, public A {};
1782 G++ will place the `A' base class of `C' at a nonzero offset;
1783 it should be placed at offset zero. G++ mistakenly believes
1784 that the `A' data member of `B' is already at offset zero.
1786 * Names of template functions whose types involve `typename' or
1787 template template parameters can be mangled incorrectly.
1789 template <typename Q>
1790 void f(typename Q::X) {}
1792 template <template <typename> class Q>
1793 void f(typename Q<int>::X) {}
1795 Instantiations of these templates may be mangled incorrectly.
1798 `-Wctor-dtor-privacy (C++ only)'
1799 Warn when a class seems unusable because all the constructors or
1800 destructors in that class are private, and it has neither friends
1801 nor public static member functions.
1803 `-Wnon-virtual-dtor (C++ only)'
1804 Warn when a class appears to be polymorphic, thereby requiring a
1805 virtual destructor, yet it declares a non-virtual one. This
1806 warning is enabled by `-Wall'.
1808 `-Wreorder (C++ only)'
1809 Warn when the order of member initializers given in the code does
1810 not match the order in which they must be executed. For instance:
1815 A(): j (0), i (1) { }
1818 The compiler will rearrange the member initializers for `i' and
1819 `j' to match the declaration order of the members, emitting a
1820 warning to that effect. This warning is enabled by `-Wall'.
1822 The following `-W...' options are not affected by `-Wall'.
1824 `-Weffc++ (C++ only)'
1825 Warn about violations of the following style guidelines from Scott
1826 Meyers' `Effective C++' book:
1828 * Item 11: Define a copy constructor and an assignment
1829 operator for classes with dynamically allocated memory.
1831 * Item 12: Prefer initialization to assignment in constructors.
1833 * Item 14: Make destructors virtual in base classes.
1835 * Item 15: Have `operator=' return a reference to `*this'.
1837 * Item 23: Don't try to return a reference when you must
1841 Also warn about violations of the following style guidelines from
1842 Scott Meyers' `More Effective C++' book:
1844 * Item 6: Distinguish between prefix and postfix forms of
1845 increment and decrement operators.
1847 * Item 7: Never overload `&&', `||', or `,'.
1850 When selecting this option, be aware that the standard library
1851 headers do not obey all of these guidelines; use `grep -v' to
1852 filter out those warnings.
1854 `-Wno-deprecated (C++ only)'
1855 Do not warn about usage of deprecated features. *Note Deprecated
1858 `-Wstrict-null-sentinel (C++ only)'
1859 Warn also about the use of an uncasted `NULL' as sentinel. When
1860 compiling only with GCC this is a valid sentinel, as `NULL' is
1861 defined to `__null'. Although it is a null pointer constant not a
1862 null pointer, it is guaranteed to of the same size as a pointer.
1863 But this use is not portable across different compilers.
1865 `-Wno-non-template-friend (C++ only)'
1866 Disable warnings when non-templatized friend functions are declared
1867 within a template. Since the advent of explicit template
1868 specification support in G++, if the name of the friend is an
1869 unqualified-id (i.e., `friend foo(int)'), the C++ language
1870 specification demands that the friend declare or define an
1871 ordinary, nontemplate function. (Section 14.5.3). Before G++
1872 implemented explicit specification, unqualified-ids could be
1873 interpreted as a particular specialization of a templatized
1874 function. Because this non-conforming behavior is no longer the
1875 default behavior for G++, `-Wnon-template-friend' allows the
1876 compiler to check existing code for potential trouble spots and is
1877 on by default. This new compiler behavior can be turned off with
1878 `-Wno-non-template-friend' which keeps the conformant compiler code
1879 but disables the helpful warning.
1881 `-Wold-style-cast (C++ only)'
1882 Warn if an old-style (C-style) cast to a non-void type is used
1883 within a C++ program. The new-style casts (`dynamic_cast',
1884 `static_cast', `reinterpret_cast', and `const_cast') are less
1885 vulnerable to unintended effects and much easier to search for.
1887 `-Woverloaded-virtual (C++ only)'
1888 Warn when a function declaration hides virtual functions from a
1889 base class. For example, in:
1895 struct B: public A {
1899 the `A' class version of `f' is hidden in `B', and code like:
1904 will fail to compile.
1906 `-Wno-pmf-conversions (C++ only)'
1907 Disable the diagnostic for converting a bound pointer to member
1908 function to a plain pointer.
1910 `-Wsign-promo (C++ only)'
1911 Warn when overload resolution chooses a promotion from unsigned or
1912 enumerated type to a signed type, over a conversion to an unsigned
1913 type of the same size. Previous versions of G++ would try to
1914 preserve unsignedness, but the standard mandates the current
1919 A& operator = (int);
1928 In this example, G++ will synthesize a default `A& operator =
1929 (const A&);', while cfront will use the user-defined `operator ='.
1932 File: gcc.info, Node: Objective-C and Objective-C++ Dialect Options, Next: Language Independent Options, Prev: C++ Dialect Options, Up: Invoking GCC
1934 3.6 Options Controlling Objective-C and Objective-C++ Dialects
1935 ==============================================================
1937 (NOTE: This manual does not describe the Objective-C and Objective-C++
1938 languages themselves. See *Note Language Standards Supported by GCC:
1939 Standards, for references.)
1941 This section describes the command-line options that are only
1942 meaningful for Objective-C and Objective-C++ programs, but you can also
1943 use most of the language-independent GNU compiler options. For
1944 example, you might compile a file `some_class.m' like this:
1946 gcc -g -fgnu-runtime -O -c some_class.m
1948 In this example, `-fgnu-runtime' is an option meant only for
1949 Objective-C and Objective-C++ programs; you can use the other options
1950 with any language supported by GCC.
1952 Note that since Objective-C is an extension of the C language,
1953 Objective-C compilations may also use options specific to the C
1954 front-end (e.g., `-Wtraditional'). Similarly, Objective-C++
1955 compilations may use C++-specific options (e.g., `-Wabi').
1957 Here is a list of options that are _only_ for compiling Objective-C
1958 and Objective-C++ programs:
1960 `-fconstant-string-class=CLASS-NAME'
1961 Use CLASS-NAME as the name of the class to instantiate for each
1962 literal string specified with the syntax `@"..."'. The default
1963 class name is `NXConstantString' if the GNU runtime is being used,
1964 and `NSConstantString' if the NeXT runtime is being used (see
1965 below). The `-fconstant-cfstrings' option, if also present, will
1966 override the `-fconstant-string-class' setting and cause `@"..."'
1967 literals to be laid out as constant CoreFoundation strings.
1970 Generate object code compatible with the standard GNU Objective-C
1971 runtime. This is the default for most types of systems.
1974 Generate output compatible with the NeXT runtime. This is the
1975 default for NeXT-based systems, including Darwin and Mac OS X.
1976 The macro `__NEXT_RUNTIME__' is predefined if (and only if) this
1979 `-fno-nil-receivers'
1980 Assume that all Objective-C message dispatches (e.g., `[receiver
1981 message:arg]') in this translation unit ensure that the receiver
1982 is not `nil'. This allows for more efficient entry points in the
1983 runtime to be used. Currently, this option is only available in
1984 conjunction with the NeXT runtime on Mac OS X 10.3 and later.
1986 `-fobjc-call-cxx-cdtors'
1987 For each Objective-C class, check if any of its instance variables
1988 is a C++ object with a non-trivial default constructor. If so,
1989 synthesize a special `- (id) .cxx_construct' instance method that
1990 will run non-trivial default constructors on any such instance
1991 variables, in order, and then return `self'. Similarly, check if
1992 any instance variable is a C++ object with a non-trivial
1993 destructor, and if so, synthesize a special `- (void)
1994 .cxx_destruct' method that will run all such default destructors,
1997 The `- (id) .cxx_construct' and/or `- (void) .cxx_destruct' methods
1998 thusly generated will only operate on instance variables declared
1999 in the current Objective-C class, and not those inherited from
2000 superclasses. It is the responsibility of the Objective-C runtime
2001 to invoke all such methods in an object's inheritance hierarchy.
2002 The `- (id) .cxx_construct' methods will be invoked by the runtime
2003 immediately after a new object instance is allocated; the `-
2004 (void) .cxx_destruct' methods will be invoked immediately before
2005 the runtime deallocates an object instance.
2007 As of this writing, only the NeXT runtime on Mac OS X 10.4 and
2008 later has support for invoking the `- (id) .cxx_construct' and `-
2009 (void) .cxx_destruct' methods.
2011 `-fobjc-direct-dispatch'
2012 Allow fast jumps to the message dispatcher. On Darwin this is
2013 accomplished via the comm page.
2016 Enable syntactic support for structured exception handling in
2017 Objective-C, similar to what is offered by C++ and Java.
2018 Currently, this option is only available in conjunction with the
2019 NeXT runtime on Mac OS X 10.3 and later.
2026 @catch (AnObjCClass *exc) {
2033 @catch (AnotherClass *exc) {
2036 @catch (id allOthers) {
2045 The `@throw' statement may appear anywhere in an Objective-C or
2046 Objective-C++ program; when used inside of a `@catch' block, the
2047 `@throw' may appear without an argument (as shown above), in which
2048 case the object caught by the `@catch' will be rethrown.
2050 Note that only (pointers to) Objective-C objects may be thrown and
2051 caught using this scheme. When an object is thrown, it will be
2052 caught by the nearest `@catch' clause capable of handling objects
2053 of that type, analogously to how `catch' blocks work in C++ and
2054 Java. A `@catch(id ...)' clause (as shown above) may also be
2055 provided to catch any and all Objective-C exceptions not caught by
2056 previous `@catch' clauses (if any).
2058 The `@finally' clause, if present, will be executed upon exit from
2059 the immediately preceding `@try ... @catch' section. This will
2060 happen regardless of whether any exceptions are thrown, caught or
2061 rethrown inside the `@try ... @catch' section, analogously to the
2062 behavior of the `finally' clause in Java.
2064 There are several caveats to using the new exception mechanism:
2066 * Although currently designed to be binary compatible with
2067 `NS_HANDLER'-style idioms provided by the `NSException'
2068 class, the new exceptions can only be used on Mac OS X 10.3
2069 (Panther) and later systems, due to additional functionality
2070 needed in the (NeXT) Objective-C runtime.
2072 * As mentioned above, the new exceptions do not support handling
2073 types other than Objective-C objects. Furthermore, when
2074 used from Objective-C++, the Objective-C exception model does
2075 not interoperate with C++ exceptions at this time. This
2076 means you cannot `@throw' an exception from Objective-C and
2077 `catch' it in C++, or vice versa (i.e., `throw ... @catch').
2079 The `-fobjc-exceptions' switch also enables the use of
2080 synchronization blocks for thread-safe execution:
2082 @synchronized (ObjCClass *guard) {
2086 Upon entering the `@synchronized' block, a thread of execution
2087 shall first check whether a lock has been placed on the
2088 corresponding `guard' object by another thread. If it has, the
2089 current thread shall wait until the other thread relinquishes its
2090 lock. Once `guard' becomes available, the current thread will
2091 place its own lock on it, execute the code contained in the
2092 `@synchronized' block, and finally relinquish the lock (thereby
2093 making `guard' available to other threads).
2095 Unlike Java, Objective-C does not allow for entire methods to be
2096 marked `@synchronized'. Note that throwing exceptions out of
2097 `@synchronized' blocks is allowed, and will cause the guarding
2098 object to be unlocked properly.
2101 Enable garbage collection (GC) in Objective-C and Objective-C++
2104 `-freplace-objc-classes'
2105 Emit a special marker instructing `ld(1)' not to statically link in
2106 the resulting object file, and allow `dyld(1)' to load it in at
2107 run time instead. This is used in conjunction with the
2108 Fix-and-Continue debugging mode, where the object file in question
2109 may be recompiled and dynamically reloaded in the course of
2110 program execution, without the need to restart the program itself.
2111 Currently, Fix-and-Continue functionality is only available in
2112 conjunction with the NeXT runtime on Mac OS X 10.3 and later.
2115 When compiling for the NeXT runtime, the compiler ordinarily
2116 replaces calls to `objc_getClass("...")' (when the name of the
2117 class is known at compile time) with static class references that
2118 get initialized at load time, which improves run-time performance.
2119 Specifying the `-fzero-link' flag suppresses this behavior and
2120 causes calls to `objc_getClass("...")' to be retained. This is
2121 useful in Zero-Link debugging mode, since it allows for individual
2122 class implementations to be modified during program execution.
2125 Dump interface declarations for all classes seen in the source
2126 file to a file named `SOURCENAME.decl'.
2128 `-Wassign-intercept'
2129 Warn whenever an Objective-C assignment is being intercepted by the
2133 If a class is declared to implement a protocol, a warning is
2134 issued for every method in the protocol that is not implemented by
2135 the class. The default behavior is to issue a warning for every
2136 method not explicitly implemented in the class, even if a method
2137 implementation is inherited from the superclass. If you use the
2138 `-Wno-protocol' option, then methods inherited from the superclass
2139 are considered to be implemented, and no warning is issued for
2143 Warn if multiple methods of different types for the same selector
2144 are found during compilation. The check is performed on the list
2145 of methods in the final stage of compilation. Additionally, a
2146 check is performed for each selector appearing in a
2147 `@selector(...)' expression, and a corresponding method for that
2148 selector has been found during compilation. Because these checks
2149 scan the method table only at the end of compilation, these
2150 warnings are not produced if the final stage of compilation is not
2151 reached, for example because an error is found during compilation,
2152 or because the `-fsyntax-only' option is being used.
2154 `-Wstrict-selector-match'
2155 Warn if multiple methods with differing argument and/or return
2156 types are found for a given selector when attempting to send a
2157 message using this selector to a receiver of type `id' or `Class'.
2158 When this flag is off (which is the default behavior), the
2159 compiler will omit such warnings if any differences found are
2160 confined to types which share the same size and alignment.
2162 `-Wundeclared-selector'
2163 Warn if a `@selector(...)' expression referring to an undeclared
2164 selector is found. A selector is considered undeclared if no
2165 method with that name has been declared before the
2166 `@selector(...)' expression, either explicitly in an `@interface'
2167 or `@protocol' declaration, or implicitly in an `@implementation'
2168 section. This option always performs its checks as soon as a
2169 `@selector(...)' expression is found, while `-Wselector' only
2170 performs its checks in the final stage of compilation. This also
2171 enforces the coding style convention that methods and selectors
2172 must be declared before being used.
2174 `-print-objc-runtime-info'
2175 Generate C header describing the largest structure that is passed
2180 File: gcc.info, Node: Language Independent Options, Next: Warning Options, Prev: Objective-C and Objective-C++ Dialect Options, Up: Invoking GCC
2182 3.7 Options to Control Diagnostic Messages Formatting
2183 =====================================================
2185 Traditionally, diagnostic messages have been formatted irrespective of
2186 the output device's aspect (e.g. its width, ...). The options described
2187 below can be used to control the diagnostic messages formatting
2188 algorithm, e.g. how many characters per line, how often source location
2189 information should be reported. Right now, only the C++ front end can
2190 honor these options. However it is expected, in the near future, that
2191 the remaining front ends would be able to digest them correctly.
2193 `-fmessage-length=N'
2194 Try to format error messages so that they fit on lines of about N
2195 characters. The default is 72 characters for `g++' and 0 for the
2196 rest of the front ends supported by GCC. If N is zero, then no
2197 line-wrapping will be done; each error message will appear on a
2200 `-fdiagnostics-show-location=once'
2201 Only meaningful in line-wrapping mode. Instructs the diagnostic
2202 messages reporter to emit _once_ source location information; that
2203 is, in case the message is too long to fit on a single physical
2204 line and has to be wrapped, the source location won't be emitted
2205 (as prefix) again, over and over, in subsequent continuation
2206 lines. This is the default behavior.
2208 `-fdiagnostics-show-location=every-line'
2209 Only meaningful in line-wrapping mode. Instructs the diagnostic
2210 messages reporter to emit the same source location information (as
2211 prefix) for physical lines that result from the process of breaking
2212 a message which is too long to fit on a single line.
2214 `-fdiagnostics-show-options'
2215 This option instructs the diagnostic machinery to add text to each
2216 diagnostic emitted, which indicates which command line option
2217 directly controls that diagnostic, when such an option is known to
2218 the diagnostic machinery.
2222 File: gcc.info, Node: Warning Options, Next: Debugging Options, Prev: Language Independent Options, Up: Invoking GCC
2224 3.8 Options to Request or Suppress Warnings
2225 ===========================================
2227 Warnings are diagnostic messages that report constructions which are
2228 not inherently erroneous but which are risky or suggest there may have
2231 You can request many specific warnings with options beginning `-W',
2232 for example `-Wimplicit' to request warnings on implicit declarations.
2233 Each of these specific warning options also has a negative form
2234 beginning `-Wno-' to turn off warnings; for example, `-Wno-implicit'.
2235 This manual lists only one of the two forms, whichever is not the
2238 The following options control the amount and kinds of warnings produced
2239 by GCC; for further, language-specific options also refer to *Note C++
2240 Dialect Options:: and *Note Objective-C and Objective-C++ Dialect
2244 Check the code for syntax errors, but don't do anything beyond
2248 Issue all the warnings demanded by strict ISO C and ISO C++;
2249 reject all programs that use forbidden extensions, and some other
2250 programs that do not follow ISO C and ISO C++. For ISO C, follows
2251 the version of the ISO C standard specified by any `-std' option
2254 Valid ISO C and ISO C++ programs should compile properly with or
2255 without this option (though a rare few will require `-ansi' or a
2256 `-std' option specifying the required version of ISO C). However,
2257 without this option, certain GNU extensions and traditional C and
2258 C++ features are supported as well. With this option, they are
2261 `-pedantic' does not cause warning messages for use of the
2262 alternate keywords whose names begin and end with `__'. Pedantic
2263 warnings are also disabled in the expression that follows
2264 `__extension__'. However, only system header files should use
2265 these escape routes; application programs should avoid them.
2266 *Note Alternate Keywords::.
2268 Some users try to use `-pedantic' to check programs for strict ISO
2269 C conformance. They soon find that it does not do quite what they
2270 want: it finds some non-ISO practices, but not all--only those for
2271 which ISO C _requires_ a diagnostic, and some others for which
2272 diagnostics have been added.
2274 A feature to report any failure to conform to ISO C might be
2275 useful in some instances, but would require considerable
2276 additional work and would be quite different from `-pedantic'. We
2277 don't have plans to support such a feature in the near future.
2279 Where the standard specified with `-std' represents a GNU extended
2280 dialect of C, such as `gnu89' or `gnu99', there is a corresponding
2281 "base standard", the version of ISO C on which the GNU extended
2282 dialect is based. Warnings from `-pedantic' are given where they
2283 are required by the base standard. (It would not make sense for
2284 such warnings to be given only for features not in the specified
2285 GNU C dialect, since by definition the GNU dialects of C include
2286 all features the compiler supports with the given option, and
2287 there would be nothing to warn about.)
2290 Like `-pedantic', except that errors are produced rather than
2294 Inhibit all warning messages.
2297 Inhibit warning messages about the use of `#import'.
2300 Warn if an array subscript has type `char'. This is a common cause
2301 of error, as programmers often forget that this type is signed on
2302 some machines. This warning is enabled by `-Wall'.
2305 Warn whenever a comment-start sequence `/*' appears in a `/*'
2306 comment, or whenever a Backslash-Newline appears in a `//' comment.
2307 This warning is enabled by `-Wall'.
2310 This option causes the compiler to abort compilation on the first
2311 error occurred rather than trying to keep going and printing
2312 further error messages.
2315 Check calls to `printf' and `scanf', etc., to make sure that the
2316 arguments supplied have types appropriate to the format string
2317 specified, and that the conversions specified in the format string
2318 make sense. This includes standard functions, and others
2319 specified by format attributes (*note Function Attributes::), in
2320 the `printf', `scanf', `strftime' and `strfmon' (an X/Open
2321 extension, not in the C standard) families (or other
2322 target-specific families). Which functions are checked without
2323 format attributes having been specified depends on the standard
2324 version selected, and such checks of functions without the
2325 attribute specified are disabled by `-ffreestanding' or
2328 The formats are checked against the format features supported by
2329 GNU libc version 2.2. These include all ISO C90 and C99 features,
2330 as well as features from the Single Unix Specification and some
2331 BSD and GNU extensions. Other library implementations may not
2332 support all these features; GCC does not support warning about
2333 features that go beyond a particular library's limitations.
2334 However, if `-pedantic' is used with `-Wformat', warnings will be
2335 given about format features not in the selected standard version
2336 (but not for `strfmon' formats, since those are not in any version
2337 of the C standard). *Note Options Controlling C Dialect: C
2340 Since `-Wformat' also checks for null format arguments for several
2341 functions, `-Wformat' also implies `-Wnonnull'.
2343 `-Wformat' is included in `-Wall'. For more control over some
2344 aspects of format checking, the options `-Wformat-y2k',
2345 `-Wno-format-extra-args', `-Wno-format-zero-length',
2346 `-Wformat-nonliteral', `-Wformat-security', and `-Wformat=2' are
2347 available, but are not included in `-Wall'.
2350 If `-Wformat' is specified, also warn about `strftime' formats
2351 which may yield only a two-digit year.
2353 `-Wno-format-extra-args'
2354 If `-Wformat' is specified, do not warn about excess arguments to a
2355 `printf' or `scanf' format function. The C standard specifies
2356 that such arguments are ignored.
2358 Where the unused arguments lie between used arguments that are
2359 specified with `$' operand number specifications, normally
2360 warnings are still given, since the implementation could not know
2361 what type to pass to `va_arg' to skip the unused arguments.
2362 However, in the case of `scanf' formats, this option will suppress
2363 the warning if the unused arguments are all pointers, since the
2364 Single Unix Specification says that such unused arguments are
2367 `-Wno-format-zero-length'
2368 If `-Wformat' is specified, do not warn about zero-length formats.
2369 The C standard specifies that zero-length formats are allowed.
2371 `-Wformat-nonliteral'
2372 If `-Wformat' is specified, also warn if the format string is not a
2373 string literal and so cannot be checked, unless the format function
2374 takes its format arguments as a `va_list'.
2377 If `-Wformat' is specified, also warn about uses of format
2378 functions that represent possible security problems. At present,
2379 this warns about calls to `printf' and `scanf' functions where the
2380 format string is not a string literal and there are no format
2381 arguments, as in `printf (foo);'. This may be a security hole if
2382 the format string came from untrusted input and contains `%n'.
2383 (This is currently a subset of what `-Wformat-nonliteral' warns
2384 about, but in future warnings may be added to `-Wformat-security'
2385 that are not included in `-Wformat-nonliteral'.)
2388 Enable `-Wformat' plus format checks not included in `-Wformat'.
2389 Currently equivalent to `-Wformat -Wformat-nonliteral
2390 -Wformat-security -Wformat-y2k'.
2393 Warn about passing a null pointer for arguments marked as
2394 requiring a non-null value by the `nonnull' function attribute.
2396 `-Wnonnull' is included in `-Wall' and `-Wformat'. It can be
2397 disabled with the `-Wno-nonnull' option.
2399 `-Winit-self (C, C++, Objective-C and Objective-C++ only)'
2400 Warn about uninitialized variables which are initialized with
2401 themselves. Note this option can only be used with the
2402 `-Wuninitialized' option, which in turn only works with `-O1' and
2405 For example, GCC will warn about `i' being uninitialized in the
2406 following snippet only when `-Winit-self' has been specified:
2414 Warn when a declaration does not specify a type. This warning is
2417 `-Wimplicit-function-declaration'
2418 `-Werror-implicit-function-declaration'
2419 Give a warning (or error) whenever a function is used before being
2420 declared. The form `-Wno-error-implicit-function-declaration' is
2421 not supported. This warning is enabled by `-Wall' (as a warning,
2425 Same as `-Wimplicit-int' and `-Wimplicit-function-declaration'.
2426 This warning is enabled by `-Wall'.
2429 Warn if the type of `main' is suspicious. `main' should be a
2430 function with external linkage, returning int, taking either zero
2431 arguments, two, or three arguments of appropriate types. This
2432 warning is enabled by `-Wall'.
2435 Warn if an aggregate or union initializer is not fully bracketed.
2436 In the following example, the initializer for `a' is not fully
2437 bracketed, but that for `b' is fully bracketed.
2439 int a[2][2] = { 0, 1, 2, 3 };
2440 int b[2][2] = { { 0, 1 }, { 2, 3 } };
2442 This warning is enabled by `-Wall'.
2444 `-Wmissing-include-dirs (C, C++, Objective-C and Objective-C++ only)'
2445 Warn if a user-supplied include directory does not exist.
2448 Warn if parentheses are omitted in certain contexts, such as when
2449 there is an assignment in a context where a truth value is
2450 expected, or when operators are nested whose precedence people
2451 often get confused about. Only the warning for an assignment used
2452 as a truth value is supported when compiling C++; the other
2453 warnings are only supported when compiling C.
2455 Also warn if a comparison like `x<=y<=z' appears; this is
2456 equivalent to `(x<=y ? 1 : 0) <= z', which is a different
2457 interpretation from that of ordinary mathematical notation.
2459 Also warn about constructions where there may be confusion to which
2460 `if' statement an `else' branch belongs. Here is an example of
2471 In C, every `else' branch belongs to the innermost possible `if'
2472 statement, which in this example is `if (b)'. This is often not
2473 what the programmer expected, as illustrated in the above example
2474 by indentation the programmer chose. When there is the potential
2475 for this confusion, GCC will issue a warning when this flag is
2476 specified. To eliminate the warning, add explicit braces around
2477 the innermost `if' statement so there is no way the `else' could
2478 belong to the enclosing `if'. The resulting code would look like
2491 This warning is enabled by `-Wall'.
2494 Warn about code that may have undefined semantics because of
2495 violations of sequence point rules in the C standard.
2497 The C standard defines the order in which expressions in a C
2498 program are evaluated in terms of "sequence points", which
2499 represent a partial ordering between the execution of parts of the
2500 program: those executed before the sequence point, and those
2501 executed after it. These occur after the evaluation of a full
2502 expression (one which is not part of a larger expression), after
2503 the evaluation of the first operand of a `&&', `||', `? :' or `,'
2504 (comma) operator, before a function is called (but after the
2505 evaluation of its arguments and the expression denoting the called
2506 function), and in certain other places. Other than as expressed
2507 by the sequence point rules, the order of evaluation of
2508 subexpressions of an expression is not specified. All these rules
2509 describe only a partial order rather than a total order, since,
2510 for example, if two functions are called within one expression
2511 with no sequence point between them, the order in which the
2512 functions are called is not specified. However, the standards
2513 committee have ruled that function calls do not overlap.
2515 It is not specified when between sequence points modifications to
2516 the values of objects take effect. Programs whose behavior
2517 depends on this have undefined behavior; the C standard specifies
2518 that "Between the previous and next sequence point an object shall
2519 have its stored value modified at most once by the evaluation of
2520 an expression. Furthermore, the prior value shall be read only to
2521 determine the value to be stored.". If a program breaks these
2522 rules, the results on any particular implementation are entirely
2525 Examples of code with undefined behavior are `a = a++;', `a[n] =
2526 b[n++]' and `a[i++] = i;'. Some more complicated cases are not
2527 diagnosed by this option, and it may give an occasional false
2528 positive result, but in general it has been found fairly effective
2529 at detecting this sort of problem in programs.
2531 The present implementation of this option only works for C
2532 programs. A future implementation may also work for C++ programs.
2534 The C standard is worded confusingly, therefore there is some
2535 debate over the precise meaning of the sequence point rules in
2536 subtle cases. Links to discussions of the problem, including
2537 proposed formal definitions, may be found on the GCC readings
2538 page, at `http://gcc.gnu.org/readings.html'.
2540 This warning is enabled by `-Wall'.
2543 Warn whenever a function is defined with a return-type that
2544 defaults to `int'. Also warn about any `return' statement with no
2545 return-value in a function whose return-type is not `void'.
2547 For C, also warn if the return type of a function has a type
2548 qualifier such as `const'. Such a type qualifier has no effect,
2549 since the value returned by a function is not an lvalue. ISO C
2550 prohibits qualified `void' return types on function definitions,
2551 so such return types always receive a warning even without this
2554 For C++, a function without return type always produces a
2555 diagnostic message, even when `-Wno-return-type' is specified.
2556 The only exceptions are `main' and functions defined in system
2559 This warning is enabled by `-Wall'.
2562 Warn whenever a `switch' statement has an index of enumerated type
2563 and lacks a `case' for one or more of the named codes of that
2564 enumeration. (The presence of a `default' label prevents this
2565 warning.) `case' labels outside the enumeration range also
2566 provoke warnings when this option is used. This warning is
2570 Warn whenever a `switch' statement does not have a `default' case.
2573 Warn whenever a `switch' statement has an index of enumerated type
2574 and lacks a `case' for one or more of the named codes of that
2575 enumeration. `case' labels outside the enumeration range also
2576 provoke warnings when this option is used.
2579 Warn if any trigraphs are encountered that might change the
2580 meaning of the program (trigraphs within comments are not warned
2581 about). This warning is enabled by `-Wall'.
2584 Warn whenever a static function is declared but not defined or a
2585 non-inline static function is unused. This warning is enabled by
2589 Warn whenever a label is declared but not used. This warning is
2592 To suppress this warning use the `unused' attribute (*note
2593 Variable Attributes::).
2595 `-Wunused-parameter'
2596 Warn whenever a function parameter is unused aside from its
2599 To suppress this warning use the `unused' attribute (*note
2600 Variable Attributes::).
2603 Warn whenever a local variable or non-constant static variable is
2604 unused aside from its declaration. This warning is enabled by
2607 To suppress this warning use the `unused' attribute (*note
2608 Variable Attributes::).
2611 Warn whenever a statement computes a result that is explicitly not
2612 used. This warning is enabled by `-Wall'.
2614 To suppress this warning cast the expression to `void'.
2617 All the above `-Wunused' options combined.
2619 In order to get a warning about an unused function parameter, you
2620 must either specify `-Wextra -Wunused' (note that `-Wall' implies
2621 `-Wunused'), or separately specify `-Wunused-parameter'.
2624 Warn if an automatic variable is used without first being
2625 initialized or if a variable may be clobbered by a `setjmp' call.
2627 These warnings are possible only in optimizing compilation,
2628 because they require data flow information that is computed only
2629 when optimizing. If you don't specify `-O', you simply won't get
2632 If you want to warn about code which uses the uninitialized value
2633 of the variable in its own initializer, use the `-Winit-self'
2636 These warnings occur for individual uninitialized or clobbered
2637 elements of structure, union or array variables as well as for
2638 variables which are uninitialized or clobbered as a whole. They do
2639 not occur for variables or elements declared `volatile'. Because
2640 these warnings depend on optimization, the exact variables or
2641 elements for which there are warnings will depend on the precise
2642 optimization options and version of GCC used.
2644 Note that there may be no warning about a variable that is used
2645 only to compute a value that itself is never used, because such
2646 computations may be deleted by data flow analysis before the
2647 warnings are printed.
2649 These warnings are made optional because GCC is not smart enough
2650 to see all the reasons why the code might be correct despite
2651 appearing to have an error. Here is one example of how this can
2667 If the value of `y' is always 1, 2 or 3, then `x' is always
2668 initialized, but GCC doesn't know this. Here is another common
2673 if (change_y) save_y = y, y = new_y;
2675 if (change_y) y = save_y;
2678 This has no bug because `save_y' is used only if it is set.
2680 This option also warns when a non-volatile automatic variable
2681 might be changed by a call to `longjmp'. These warnings as well
2682 are possible only in optimizing compilation.
2684 The compiler sees only the calls to `setjmp'. It cannot know
2685 where `longjmp' will be called; in fact, a signal handler could
2686 call it at any point in the code. As a result, you may get a
2687 warning even when there is in fact no problem because `longjmp'
2688 cannot in fact be called at the place which would cause a problem.
2690 Some spurious warnings can be avoided if you declare all the
2691 functions you use that never return as `noreturn'. *Note Function
2694 This warning is enabled by `-Wall'.
2697 Warn when a #pragma directive is encountered which is not
2698 understood by GCC. If this command line option is used, warnings
2699 will even be issued for unknown pragmas in system header files.
2700 This is not the case if the warnings were only enabled by the
2701 `-Wall' command line option.
2704 Do not warn about misuses of pragmas, such as incorrect parameters,
2705 invalid syntax, or conflicts between pragmas. See also
2706 `-Wunknown-pragmas'.
2709 This option is only active when `-fstrict-aliasing' is active. It
2710 warns about code which might break the strict aliasing rules that
2711 the compiler is using for optimization. The warning does not
2712 catch all cases, but does attempt to catch the more common
2713 pitfalls. It is included in `-Wall'.
2715 `-Wstrict-aliasing=2'
2716 This option is only active when `-fstrict-aliasing' is active. It
2717 warns about code which might break the strict aliasing rules that
2718 the compiler is using for optimization. This warning catches more
2719 cases than `-Wstrict-aliasing', but it will also give a warning
2720 for some ambiguous cases that are safe.
2723 All of the above `-W' options combined. This enables all the
2724 warnings about constructions that some users consider
2725 questionable, and that are easy to avoid (or modify to prevent the
2726 warning), even in conjunction with macros. This also enables some
2727 language-specific warnings described in *Note C++ Dialect
2728 Options:: and *Note Objective-C and Objective-C++ Dialect
2731 The following `-W...' options are not implied by `-Wall'. Some of
2732 them warn about constructions that users generally do not consider
2733 questionable, but which occasionally you might wish to check for;
2734 others warn about constructions that are necessary or hard to avoid in
2735 some cases, and there is no simple way to modify the code to suppress
2739 (This option used to be called `-W'. The older name is still
2740 supported, but the newer name is more descriptive.) Print extra
2741 warning messages for these events:
2743 * A function can return either with or without a value.
2744 (Falling off the end of the function body is considered
2745 returning without a value.) For example, this function would
2746 evoke such a warning:
2754 * An expression-statement or the left-hand side of a comma
2755 expression contains no side effects. To suppress the
2756 warning, cast the unused expression to void. For example, an
2757 expression such as `x[i,j]' will cause a warning, but
2758 `x[(void)i,j]' will not.
2760 * An unsigned value is compared against zero with `<' or `>='.
2762 * Storage-class specifiers like `static' are not the first
2763 things in a declaration. According to the C Standard, this
2764 usage is obsolescent.
2766 * If `-Wall' or `-Wunused' is also specified, warn about unused
2769 * A comparison between signed and unsigned values could produce
2770 an incorrect result when the signed value is converted to
2771 unsigned. (But don't warn if `-Wno-sign-compare' is also
2774 * An aggregate has an initializer which does not initialize all
2775 members. This warning can be independently controlled by
2776 `-Wmissing-field-initializers'.
2778 * A function parameter is declared without a type specifier in
2779 K&R-style functions:
2783 * An empty body occurs in an `if' or `else' statement.
2785 * A pointer is compared against integer zero with `<', `<=',
2788 * A variable might be changed by `longjmp' or `vfork'.
2790 * Any of several floating-point events that often indicate
2791 errors, such as overflow, underflow, loss of precision, etc.
2793 * (C++ only) An enumerator and a non-enumerator both appear in
2794 a conditional expression.
2796 * (C++ only) A non-static reference or non-static `const'
2797 member appears in a class without constructors.
2799 * (C++ only) Ambiguous virtual bases.
2801 * (C++ only) Subscripting an array which has been declared
2804 * (C++ only) Taking the address of a variable which has been
2805 declared `register'.
2807 * (C++ only) A base class is not initialized in a derived
2808 class' copy constructor.
2811 Do not warn about compile-time integer division by zero. Floating
2812 point division by zero is not warned about, as it can be a
2813 legitimate way of obtaining infinities and NaNs.
2816 Print warning messages for constructs found in system header files.
2817 Warnings from system headers are normally suppressed, on the
2818 assumption that they usually do not indicate real problems and
2819 would only make the compiler output harder to read. Using this
2820 command line option tells GCC to emit warnings from system headers
2821 as if they occurred in user code. However, note that using
2822 `-Wall' in conjunction with this option will _not_ warn about
2823 unknown pragmas in system headers--for that, `-Wunknown-pragmas'
2827 Warn if floating point values are used in equality comparisons.
2829 The idea behind this is that sometimes it is convenient (for the
2830 programmer) to consider floating-point values as approximations to
2831 infinitely precise real numbers. If you are doing this, then you
2832 need to compute (by analyzing the code, or in some other way) the
2833 maximum or likely maximum error that the computation introduces,
2834 and allow for it when performing comparisons (and when producing
2835 output, but that's a different problem). In particular, instead
2836 of testing for equality, you would check to see whether the two
2837 values have ranges that overlap; and this is done with the
2838 relational operators, so equality comparisons are probably
2841 `-Wtraditional (C only)'
2842 Warn about certain constructs that behave differently in
2843 traditional and ISO C. Also warn about ISO C constructs that have
2844 no traditional C equivalent, and/or problematic constructs which
2847 * Macro parameters that appear within string literals in the
2848 macro body. In traditional C macro replacement takes place
2849 within string literals, but does not in ISO C.
2851 * In traditional C, some preprocessor directives did not exist.
2852 Traditional preprocessors would only consider a line to be a
2853 directive if the `#' appeared in column 1 on the line.
2854 Therefore `-Wtraditional' warns about directives that
2855 traditional C understands but would ignore because the `#'
2856 does not appear as the first character on the line. It also
2857 suggests you hide directives like `#pragma' not understood by
2858 traditional C by indenting them. Some traditional
2859 implementations would not recognize `#elif', so it suggests
2860 avoiding it altogether.
2862 * A function-like macro that appears without arguments.
2864 * The unary plus operator.
2866 * The `U' integer constant suffix, or the `F' or `L' floating
2867 point constant suffixes. (Traditional C does support the `L'
2868 suffix on integer constants.) Note, these suffixes appear in
2869 macros defined in the system headers of most modern systems,
2870 e.g. the `_MIN'/`_MAX' macros in `<limits.h>'. Use of these
2871 macros in user code might normally lead to spurious warnings,
2872 however GCC's integrated preprocessor has enough context to
2873 avoid warning in these cases.
2875 * A function declared external in one block and then used after
2876 the end of the block.
2878 * A `switch' statement has an operand of type `long'.
2880 * A non-`static' function declaration follows a `static' one.
2881 This construct is not accepted by some traditional C
2884 * The ISO type of an integer constant has a different width or
2885 signedness from its traditional type. This warning is only
2886 issued if the base of the constant is ten. I.e. hexadecimal
2887 or octal values, which typically represent bit patterns, are
2890 * Usage of ISO string concatenation is detected.
2892 * Initialization of automatic aggregates.
2894 * Identifier conflicts with labels. Traditional C lacks a
2895 separate namespace for labels.
2897 * Initialization of unions. If the initializer is zero, the
2898 warning is omitted. This is done under the assumption that
2899 the zero initializer in user code appears conditioned on e.g.
2900 `__STDC__' to avoid missing initializer warnings and relies
2901 on default initialization to zero in the traditional C case.
2903 * Conversions by prototypes between fixed/floating point values
2904 and vice versa. The absence of these prototypes when
2905 compiling with traditional C would cause serious problems.
2906 This is a subset of the possible conversion warnings, for the
2907 full set use `-Wconversion'.
2909 * Use of ISO C style function definitions. This warning
2910 intentionally is _not_ issued for prototype declarations or
2911 variadic functions because these ISO C features will appear
2912 in your code when using libiberty's traditional C
2913 compatibility macros, `PARAMS' and `VPARAMS'. This warning
2914 is also bypassed for nested functions because that feature is
2915 already a GCC extension and thus not relevant to traditional
2918 `-Wdeclaration-after-statement (C only)'
2919 Warn when a declaration is found after a statement in a block.
2920 This construct, known from C++, was introduced with ISO C99 and is
2921 by default allowed in GCC. It is not supported by ISO C90 and was
2922 not supported by GCC versions before GCC 3.0. *Note Mixed
2926 Warn if an undefined identifier is evaluated in an `#if' directive.
2929 Do not warn whenever an `#else' or an `#endif' are followed by
2933 Warn whenever a local variable shadows another local variable,
2934 parameter or global variable or whenever a built-in function is
2938 Warn whenever an object of larger than LEN bytes is defined.
2940 `-Wunsafe-loop-optimizations'
2941 Warn if the loop cannot be optimized because the compiler could not
2942 assume anything on the bounds of the loop indices. With
2943 `-funsafe-loop-optimizations' warn if the compiler made such
2947 Warn about anything that depends on the "size of" a function type
2948 or of `void'. GNU C assigns these types a size of 1, for
2949 convenience in calculations with `void *' pointers and pointers to
2952 `-Wbad-function-cast (C only)'
2953 Warn whenever a function call is cast to a non-matching type. For
2954 example, warn if `int malloc()' is cast to `anything *'.
2957 Warn about ISO C constructs that are outside of the common subset
2958 of ISO C and ISO C++, e.g. request for implicit conversion from
2959 `void *' to a pointer to non-`void' type.
2962 Warn whenever a pointer is cast so as to remove a type qualifier
2963 from the target type. For example, warn if a `const char *' is
2964 cast to an ordinary `char *'.
2967 Warn whenever a pointer is cast such that the required alignment
2968 of the target is increased. For example, warn if a `char *' is
2969 cast to an `int *' on machines where integers can only be accessed
2970 at two- or four-byte boundaries.
2973 When compiling C, give string constants the type `const
2974 char[LENGTH]' so that copying the address of one into a
2975 non-`const' `char *' pointer will get a warning; when compiling
2976 C++, warn about the deprecated conversion from string constants to
2977 `char *'. These warnings will help you find at compile time code
2978 that can try to write into a string constant, but only if you have
2979 been very careful about using `const' in declarations and
2980 prototypes. Otherwise, it will just be a nuisance; this is why we
2981 did not make `-Wall' request these warnings.
2984 Warn if a prototype causes a type conversion that is different
2985 from what would happen to the same argument in the absence of a
2986 prototype. This includes conversions of fixed point to floating
2987 and vice versa, and conversions changing the width or signedness
2988 of a fixed point argument except when the same as the default
2991 Also, warn if a negative integer constant expression is implicitly
2992 converted to an unsigned type. For example, warn about the
2993 assignment `x = -1' if `x' is unsigned. But do not warn about
2994 explicit casts like `(unsigned) -1'.
2997 Warn when a comparison between signed and unsigned values could
2998 produce an incorrect result when the signed value is converted to
2999 unsigned. This warning is also enabled by `-Wextra'; to get the
3000 other warnings of `-Wextra' without this warning, use `-Wextra
3003 `-Waggregate-return'
3004 Warn if any functions that return structures or unions are defined
3005 or called. (In languages where you can return an array, this also
3009 Do not warn if an unexpected `__attribute__' is used, such as
3010 unrecognized attributes, function attributes applied to variables,
3011 etc. This will not stop errors for incorrect use of supported
3014 `-Wstrict-prototypes (C only)'
3015 Warn if a function is declared or defined without specifying the
3016 argument types. (An old-style function definition is permitted
3017 without a warning if preceded by a declaration which specifies the
3020 `-Wold-style-definition (C only)'
3021 Warn if an old-style function definition is used. A warning is
3022 given even if there is a previous prototype.
3024 `-Wmissing-prototypes (C only)'
3025 Warn if a global function is defined without a previous prototype
3026 declaration. This warning is issued even if the definition itself
3027 provides a prototype. The aim is to detect global functions that
3028 fail to be declared in header files.
3030 `-Wmissing-declarations (C only)'
3031 Warn if a global function is defined without a previous
3032 declaration. Do so even if the definition itself provides a
3033 prototype. Use this option to detect global functions that are
3034 not declared in header files.
3036 `-Wmissing-field-initializers'
3037 Warn if a structure's initializer has some fields missing. For
3038 example, the following code would cause such a warning, because
3039 `x.h' is implicitly zero:
3041 struct s { int f, g, h; };
3042 struct s x = { 3, 4 };
3044 This option does not warn about designated initializers, so the
3045 following modification would not trigger a warning:
3047 struct s { int f, g, h; };
3048 struct s x = { .f = 3, .g = 4 };
3050 This warning is included in `-Wextra'. To get other `-Wextra'
3051 warnings without this one, use `-Wextra
3052 -Wno-missing-field-initializers'.
3054 `-Wmissing-noreturn'
3055 Warn about functions which might be candidates for attribute
3056 `noreturn'. Note these are only possible candidates, not absolute
3057 ones. Care should be taken to manually verify functions actually
3058 do not ever return before adding the `noreturn' attribute,
3059 otherwise subtle code generation bugs could be introduced. You
3060 will not get a warning for `main' in hosted C environments.
3062 `-Wmissing-format-attribute'
3063 Warn about function pointers which might be candidates for `format'
3064 attributes. Note these are only possible candidates, not absolute
3065 ones. GCC will guess that function pointers with `format'
3066 attributes that are used in assignment, initialization, parameter
3067 passing or return statements should have a corresponding `format'
3068 attribute in the resulting type. I.e. the left-hand side of the
3069 assignment or initialization, the type of the parameter variable,
3070 or the return type of the containing function respectively should
3071 also have a `format' attribute to avoid the warning.
3073 GCC will also warn about function definitions which might be
3074 candidates for `format' attributes. Again, these are only
3075 possible candidates. GCC will guess that `format' attributes
3076 might be appropriate for any function that calls a function like
3077 `vprintf' or `vscanf', but this might not always be the case, and
3078 some functions for which `format' attributes are appropriate may
3082 Do not warn if a multicharacter constant (`'FOOF'') is used.
3083 Usually they indicate a typo in the user's code, as they have
3084 implementation-defined values, and should not be used in portable
3087 `-Wnormalized=<none|id|nfc|nfkc>'
3088 In ISO C and ISO C++, two identifiers are different if they are
3089 different sequences of characters. However, sometimes when
3090 characters outside the basic ASCII character set are used, you can
3091 have two different character sequences that look the same. To
3092 avoid confusion, the ISO 10646 standard sets out some
3093 "normalization rules" which when applied ensure that two sequences
3094 that look the same are turned into the same sequence. GCC can
3095 warn you if you are using identifiers which have not been
3096 normalized; this option controls that warning.
3098 There are four levels of warning that GCC supports. The default is
3099 `-Wnormalized=nfc', which warns about any identifier which is not
3100 in the ISO 10646 "C" normalized form, "NFC". NFC is the
3101 recommended form for most uses.
3103 Unfortunately, there are some characters which ISO C and ISO C++
3104 allow in identifiers that when turned into NFC aren't allowable as
3105 identifiers. That is, there's no way to use these symbols in
3106 portable ISO C or C++ and have all your identifiers in NFC.
3107 `-Wnormalized=id' suppresses the warning for these characters. It
3108 is hoped that future versions of the standards involved will
3109 correct this, which is why this option is not the default.
3111 You can switch the warning off for all characters by writing
3112 `-Wnormalized=none'. You would only want to do this if you were
3113 using some other normalization scheme (like "D"), because
3114 otherwise you can easily create bugs that are literally impossible
3117 Some characters in ISO 10646 have distinct meanings but look
3118 identical in some fonts or display methodologies, especially once
3119 formatting has been applied. For instance `\u207F', "SUPERSCRIPT
3120 LATIN SMALL LETTER N", will display just like a regular `n' which
3121 has been placed in a superscript. ISO 10646 defines the "NFKC"
3122 normalisation scheme to convert all these into a standard form as
3123 well, and GCC will warn if your code is not in NFKC if you use
3124 `-Wnormalized=nfkc'. This warning is comparable to warning about
3125 every identifier that contains the letter O because it might be
3126 confused with the digit 0, and so is not the default, but may be
3127 useful as a local coding convention if the programming environment
3128 is unable to be fixed to display these characters distinctly.
3130 `-Wno-deprecated-declarations'
3131 Do not warn about uses of functions, variables, and types marked as
3132 deprecated by using the `deprecated' attribute. (*note Function
3133 Attributes::, *note Variable Attributes::, *note Type
3137 Warn if a structure is given the packed attribute, but the packed
3138 attribute has no effect on the layout or size of the structure.
3139 Such structures may be mis-aligned for little benefit. For
3140 instance, in this code, the variable `f.x' in `struct bar' will be
3141 misaligned even though `struct bar' does not itself have the
3147 } __attribute__((packed));
3154 Warn if padding is included in a structure, either to align an
3155 element of the structure or to align the whole structure.
3156 Sometimes when this happens it is possible to rearrange the fields
3157 of the structure to reduce the padding and so make the structure
3161 Warn if anything is declared more than once in the same scope,
3162 even in cases where multiple declaration is valid and changes
3165 `-Wnested-externs (C only)'
3166 Warn if an `extern' declaration is encountered within a function.
3168 `-Wunreachable-code'
3169 Warn if the compiler detects that code will never be executed.
3171 This option is intended to warn when the compiler detects that at
3172 least a whole line of source code will never be executed, because
3173 some condition is never satisfied or because it is after a
3174 procedure that never returns.
3176 It is possible for this option to produce a warning even though
3177 there are circumstances under which part of the affected line can
3178 be executed, so care should be taken when removing
3179 apparently-unreachable code.
3181 For instance, when a function is inlined, a warning may mean that
3182 the line is unreachable in only one inlined copy of the function.
3184 This option is not made part of `-Wall' because in a debugging
3185 version of a program there is often substantial code which checks
3186 correct functioning of the program and is, hopefully, unreachable
3187 because the program does work. Another common use of unreachable
3188 code is to provide behavior which is selectable at compile-time.
3191 Warn if a function can not be inlined and it was declared as
3192 inline. Even with this option, the compiler will not warn about
3193 failures to inline functions declared in system headers.
3195 The compiler uses a variety of heuristics to determine whether or
3196 not to inline a function. For example, the compiler takes into
3197 account the size of the function being inlined and the amount of
3198 inlining that has already been done in the current function.
3199 Therefore, seemingly insignificant changes in the source program
3200 can cause the warnings produced by `-Winline' to appear or
3203 `-Wno-invalid-offsetof (C++ only)'
3204 Suppress warnings from applying the `offsetof' macro to a non-POD
3205 type. According to the 1998 ISO C++ standard, applying `offsetof'
3206 to a non-POD type is undefined. In existing C++ implementations,
3207 however, `offsetof' typically gives meaningful results even when
3208 applied to certain kinds of non-POD types. (Such as a simple
3209 `struct' that fails to be a POD type only by virtue of having a
3210 constructor.) This flag is for users who are aware that they are
3211 writing nonportable code and who have deliberately chosen to
3212 ignore the warning about it.
3214 The restrictions on `offsetof' may be relaxed in a future version
3215 of the C++ standard.
3217 `-Wno-int-to-pointer-cast (C only)'
3218 Suppress warnings from casts to pointer type of an integer of a
3221 `-Wno-pointer-to-int-cast (C only)'
3222 Suppress warnings from casts from a pointer to an integer type of a
3226 Warn if a precompiled header (*note Precompiled Headers::) is
3227 found in the search path but can't be used.
3230 Warn if `long long' type is used. This is default. To inhibit
3231 the warning messages, use `-Wno-long-long'. Flags `-Wlong-long'
3232 and `-Wno-long-long' are taken into account only when `-pedantic'
3236 Warn if variadic macros are used in pedantic ISO C90 mode, or the
3237 GNU alternate syntax when in pedantic ISO C99 mode. This is
3238 default. To inhibit the warning messages, use
3239 `-Wno-variadic-macros'.
3241 `-Wvolatile-register-var'
3242 Warn if a register variable is declared volatile. The volatile
3243 modifier does not inhibit all optimizations that may eliminate
3244 reads and/or writes to register variables.
3246 `-Wdisabled-optimization'
3247 Warn if a requested optimization pass is disabled. This warning
3248 does not generally indicate that there is anything wrong with your
3249 code; it merely indicates that GCC's optimizers were unable to
3250 handle the code effectively. Often, the problem is that your code
3251 is too big or too complex; GCC will refuse to optimize programs
3252 when the optimization itself is likely to take inordinate amounts
3256 Warn for pointer argument passing or assignment with different
3257 signedness. This option is only supported for C and Objective-C.
3258 It is implied by `-Wall' and by `-pedantic', which can be disabled
3259 with `-Wno-pointer-sign'.
3262 Make all warnings into errors.
3265 This option is only active when `-fstack-protector' is active. It
3266 warns about functions that will not be protected against stack
3271 File: gcc.info, Node: Debugging Options, Next: Optimize Options, Prev: Warning Options, Up: Invoking GCC
3273 3.9 Options for Debugging Your Program or GCC
3274 =============================================
3276 GCC has various special options that are used for debugging either your
3280 Produce debugging information in the operating system's native
3281 format (stabs, COFF, XCOFF, or DWARF 2). GDB can work with this
3282 debugging information.
3284 On most systems that use stabs format, `-g' enables use of extra
3285 debugging information that only GDB can use; this extra information
3286 makes debugging work better in GDB but will probably make other
3287 debuggers crash or refuse to read the program. If you want to
3288 control for certain whether to generate the extra information, use
3289 `-gstabs+', `-gstabs', `-gxcoff+', `-gxcoff', or `-gvms' (see
3292 GCC allows you to use `-g' with `-O'. The shortcuts taken by
3293 optimized code may occasionally produce surprising results: some
3294 variables you declared may not exist at all; flow of control may
3295 briefly move where you did not expect it; some statements may not
3296 be executed because they compute constant results or their values
3297 were already at hand; some statements may execute in different
3298 places because they were moved out of loops.
3300 Nevertheless it proves possible to debug optimized output. This
3301 makes it reasonable to use the optimizer for programs that might
3304 The following options are useful when GCC is generated with the
3305 capability for more than one debugging format.
3308 Produce debugging information for use by GDB. This means to use
3309 the most expressive format available (DWARF 2, stabs, or the
3310 native format if neither of those are supported), including GDB
3311 extensions if at all possible.
3314 Produce debugging information in stabs format (if that is
3315 supported), without GDB extensions. This is the format used by
3316 DBX on most BSD systems. On MIPS, Alpha and System V Release 4
3317 systems this option produces stabs debugging output which is not
3318 understood by DBX or SDB. On System V Release 4 systems this
3319 option requires the GNU assembler.
3321 `-feliminate-unused-debug-symbols'
3322 Produce debugging information in stabs format (if that is
3323 supported), for only symbols that are actually used.
3326 Produce debugging information in stabs format (if that is
3327 supported), using GNU extensions understood only by the GNU
3328 debugger (GDB). The use of these extensions is likely to make
3329 other debuggers crash or refuse to read the program.
3332 Produce debugging information in COFF format (if that is
3333 supported). This is the format used by SDB on most System V
3334 systems prior to System V Release 4.
3337 Produce debugging information in XCOFF format (if that is
3338 supported). This is the format used by the DBX debugger on IBM
3342 Produce debugging information in XCOFF format (if that is
3343 supported), using GNU extensions understood only by the GNU
3344 debugger (GDB). The use of these extensions is likely to make
3345 other debuggers crash or refuse to read the program, and may cause
3346 assemblers other than the GNU assembler (GAS) to fail with an
3350 Produce debugging information in DWARF version 2 format (if that is
3351 supported). This is the format used by DBX on IRIX 6. With this
3352 option, GCC uses features of DWARF version 3 when they are useful;
3353 version 3 is upward compatible with version 2, but may still cause
3354 problems for older debuggers.
3357 Produce debugging information in VMS debug format (if that is
3358 supported). This is the format used by DEBUG on VMS systems.
3366 Request debugging information and also use LEVEL to specify how
3367 much information. The default level is 2.
3369 Level 1 produces minimal information, enough for making backtraces
3370 in parts of the program that you don't plan to debug. This
3371 includes descriptions of functions and external variables, but no
3372 information about local variables and no line numbers.
3374 Level 3 includes extra information, such as all the macro
3375 definitions present in the program. Some debuggers support macro
3376 expansion when you use `-g3'.
3378 `-gdwarf-2' does not accept a concatenated debug level, because
3379 GCC used to support an option `-gdwarf' that meant to generate
3380 debug information in version 1 of the DWARF format (which is very
3381 different from version 2), and it would have been too confusing.
3382 That debug format is long obsolete, but the option cannot be
3383 changed now. Instead use an additional `-gLEVEL' option to change
3384 the debug level for DWARF2.
3386 `-feliminate-dwarf2-dups'
3387 Compress DWARF2 debugging information by eliminating duplicated
3388 information about each symbol. This option only makes sense when
3389 generating DWARF2 debugging information with `-gdwarf-2'.
3392 Generate extra code to write profile information suitable for the
3393 analysis program `prof'. You must use this option when compiling
3394 the source files you want data about, and you must also use it when
3398 Generate extra code to write profile information suitable for the
3399 analysis program `gprof'. You must use this option when compiling
3400 the source files you want data about, and you must also use it when
3404 Makes the compiler print out each function name as it is compiled,
3405 and print some statistics about each pass when it finishes.
3408 Makes the compiler print some statistics about the time consumed
3409 by each pass when it finishes.
3412 Makes the compiler print some statistics about permanent memory
3413 allocation when it finishes.
3416 Add code so that program flow "arcs" are instrumented. During
3417 execution the program records how many times each branch and call
3418 is executed and how many times it is taken or returns. When the
3419 compiled program exits it saves this data to a file called
3420 `AUXNAME.gcda' for each source file. The data may be used for
3421 profile-directed optimizations (`-fbranch-probabilities'), or for
3422 test coverage analysis (`-ftest-coverage'). Each object file's
3423 AUXNAME is generated from the name of the output file, if
3424 explicitly specified and it is not the final executable, otherwise
3425 it is the basename of the source file. In both cases any suffix
3426 is removed (e.g. `foo.gcda' for input file `dir/foo.c', or
3427 `dir/foo.gcda' for output file specified as `-o dir/foo.o').
3428 *Note Cross-profiling::.
3431 This option is used to compile and link code instrumented for
3432 coverage analysis. The option is a synonym for `-fprofile-arcs'
3433 `-ftest-coverage' (when compiling) and `-lgcov' (when linking).
3434 See the documentation for those options for more details.
3436 * Compile the source files with `-fprofile-arcs' plus
3437 optimization and code generation options. For test coverage
3438 analysis, use the additional `-ftest-coverage' option. You
3439 do not need to profile every source file in a program.
3441 * Link your object files with `-lgcov' or `-fprofile-arcs' (the
3442 latter implies the former).
3444 * Run the program on a representative workload to generate the
3445 arc profile information. This may be repeated any number of
3446 times. You can run concurrent instances of your program, and
3447 provided that the file system supports locking, the data
3448 files will be correctly updated. Also `fork' calls are
3449 detected and correctly handled (double counting will not
3452 * For profile-directed optimizations, compile the source files
3453 again with the same optimization and code generation options
3454 plus `-fbranch-probabilities' (*note Options that Control
3455 Optimization: Optimize Options.).
3457 * For test coverage analysis, use `gcov' to produce human
3458 readable information from the `.gcno' and `.gcda' files.
3459 Refer to the `gcov' documentation for further information.
3462 With `-fprofile-arcs', for each function of your program GCC
3463 creates a program flow graph, then finds a spanning tree for the
3464 graph. Only arcs that are not on the spanning tree have to be
3465 instrumented: the compiler adds code to count the number of times
3466 that these arcs are executed. When an arc is the only exit or
3467 only entrance to a block, the instrumentation code can be added to
3468 the block; otherwise, a new basic block must be created to hold
3469 the instrumentation code.
3472 Produce a notes file that the `gcov' code-coverage utility (*note
3473 `gcov'--a Test Coverage Program: Gcov.) can use to show program
3474 coverage. Each source file's note file is called `AUXNAME.gcno'.
3475 Refer to the `-fprofile-arcs' option above for a description of
3476 AUXNAME and instructions on how to generate test coverage data.
3477 Coverage data will match the source files more closely, if you do
3483 Says to make debugging dumps during compilation at times specified
3484 by LETTERS. This is used for debugging the RTL-based passes of
3485 the compiler. The file names for most of the dumps are made by
3486 appending a pass number and a word to the DUMPNAME. DUMPNAME is
3487 generated from the name of the output file, if explicitly
3488 specified and it is not an executable, otherwise it is the
3489 basename of the source file.
3491 Most debug dumps can be enabled either passing a letter to the `-d'
3492 option, or with a long `-fdump-rtl' switch; here are the possible
3493 letters for use in LETTERS and PASS, and their meanings:
3496 Annotate the assembler output with miscellaneous debugging
3501 Dump after computing branch probabilities, to `FILE.09.bp'.
3505 Dump after block reordering, to `FILE.30.bbro'.
3508 `-fdump-rtl-combine'
3509 Dump after instruction combination, to the file
3515 `-dC' and `-fdump-rtl-ce1' enable dumping after the first if
3516 conversion, to the file `FILE.11.ce1'. `-dC' and
3517 `-fdump-rtl-ce2' enable dumping after the second if
3518 conversion, to the file `FILE.18.ce2'.
3523 `-dd' and `-fdump-rtl-btl' enable dumping after branch target
3524 load optimization, to `FILE.31.btl'. `-dd' and
3525 `-fdump-rtl-dbr' enable dumping after delayed branch
3526 scheduling, to `FILE.36.dbr'.
3529 Dump all macro definitions, at the end of preprocessing, in
3530 addition to normal output.
3534 Dump after the third if conversion, to `FILE.28.ce3'.
3539 `-df' and `-fdump-rtl-cfg' enable dumping after control and
3540 data flow analysis, to `FILE.08.cfg'. `-df' and
3541 `-fdump-rtl-cfg' enable dumping dump after life analysis, to
3546 Dump after global register allocation, to `FILE.23.greg'.
3551 `-dG' and `-fdump-rtl-gcse' enable dumping after GCSE, to
3552 `FILE.05.gcse'. `-dG' and `-fdump-rtl-bypass' enable dumping
3553 after jump bypassing and control flow optimizations, to
3558 Dump after finalization of EH handling code, to `FILE.02.eh'.
3561 `-fdump-rtl-sibling'
3562 Dump after sibling call optimizations, to `FILE.01.sibling'.
3566 Dump after the first jump optimization, to `FILE.03.jump'.
3570 Dump after conversion from registers to stack, to
3575 Dump after local register allocation, to `FILE.22.lreg'.
3580 `-dL' and `-fdump-rtl-loop' enable dumping after the first
3581 loop optimization pass, to `FILE.06.loop'. `-dL' and
3582 `-fdump-rtl-loop2' enable dumping after the second pass, to
3587 Dump after modulo scheduling, to `FILE.20.sms'.
3591 Dump after performing the machine dependent reorganization
3592 pass, to `FILE.35.mach'.
3596 Dump after register renumbering, to `FILE.29.rnreg'.
3599 `-fdump-rtl-regmove'
3600 Dump after the register move pass, to `FILE.19.regmove'.
3603 `-fdump-rtl-postreload'
3604 Dump after post-reload optimizations, to `FILE.24.postreload'.
3608 Dump after RTL generation, to `FILE.00.expand'.
3612 Dump after the second scheduling pass, to `FILE.32.sched2'.
3616 Dump after CSE (including the jump optimization that
3617 sometimes follows CSE), to `FILE.04.cse'.
3621 Dump after the first scheduling pass, to `FILE.21.sched'.
3625 Dump after the second CSE pass (including the jump
3626 optimization that sometimes follows CSE), to `FILE.15.cse2'.
3630 Dump after running tracer, to `FILE.12.tracer'.
3634 `-fdump-rtl-vartrack'
3635 `-dV' and `-fdump-rtl-vpt' enable dumping after the value
3636 profile transformations, to `FILE.10.vpt'. `-dV' and
3637 `-fdump-rtl-vartrack' enable dumping after variable tracking,
3638 to `FILE.34.vartrack'.
3642 Dump after the second flow pass, to `FILE.26.flow2'.
3645 `-fdump-rtl-peephole2'
3646 Dump after the peephole pass, to `FILE.27.peephole2'.
3650 Dump after live range splitting, to `FILE.14.web'.
3654 Produce all the dumps listed above.
3657 Produce a core dump whenever an error occurs.
3660 Print statistics on memory usage, at the end of the run, to
3664 Annotate the assembler output with a comment indicating which
3665 pattern and alternative was used. The length of each
3666 instruction is also printed.
3669 Dump the RTL in the assembler output as a comment before each
3670 instruction. Also turns on `-dp' annotation.
3673 For each of the other indicated dump files (either with `-d'
3674 or `-fdump-rtl-PASS'), dump a representation of the control
3675 flow graph suitable for viewing with VCG to `FILE.PASS.vcg'.
3678 Just generate RTL for a function instead of compiling it.
3679 Usually used with `r' (`-fdump-rtl-expand').
3682 Dump debugging information during parsing, to standard error.
3685 When doing debugging dumps (see `-d' option above), suppress
3686 instruction numbers and line number note output. This makes it
3687 more feasible to use diff on debugging dumps for compiler
3688 invocations with different options, in particular with and without
3691 `-fdump-translation-unit (C++ only)'
3692 `-fdump-translation-unit-OPTIONS (C++ only)'
3693 Dump a representation of the tree structure for the entire
3694 translation unit to a file. The file name is made by appending
3695 `.tu' to the source file name. If the `-OPTIONS' form is used,
3696 OPTIONS controls the details of the dump as described for the
3697 `-fdump-tree' options.
3699 `-fdump-class-hierarchy (C++ only)'
3700 `-fdump-class-hierarchy-OPTIONS (C++ only)'
3701 Dump a representation of each class's hierarchy and virtual
3702 function table layout to a file. The file name is made by
3703 appending `.class' to the source file name. If the `-OPTIONS'
3704 form is used, OPTIONS controls the details of the dump as
3705 described for the `-fdump-tree' options.
3708 Control the dumping at various stages of inter-procedural analysis
3709 language tree to a file. The file name is generated by appending
3710 a switch specific suffix to the source file name. The following
3714 Enables all inter-procedural analysis dumps; currently the
3715 only produced dump is the `cgraph' dump.
3718 Dumps information about call-graph optimization, unused
3719 function removal, and inlining decisions.
3721 `-fdump-tree-SWITCH'
3722 `-fdump-tree-SWITCH-OPTIONS'
3723 Control the dumping at various stages of processing the
3724 intermediate language tree to a file. The file name is generated
3725 by appending a switch specific suffix to the source file name. If
3726 the `-OPTIONS' form is used, OPTIONS is a list of `-' separated
3727 options that control the details of the dump. Not all options are
3728 applicable to all dumps, those which are not meaningful will be
3729 ignored. The following options are available
3732 Print the address of each node. Usually this is not
3733 meaningful as it changes according to the environment and
3734 source file. Its primary use is for tying up a dump file
3735 with a debug environment.
3738 Inhibit dumping of members of a scope or body of a function
3739 merely because that scope has been reached. Only dump such
3740 items when they are directly reachable by some other path.
3741 When dumping pretty-printed trees, this option inhibits
3742 dumping the bodies of control structures.
3745 Print a raw representation of the tree. By default, trees are
3746 pretty-printed into a C-like representation.
3749 Enable more detailed dumps (not honored by every dump option).
3752 Enable dumping various statistics about the pass (not honored
3753 by every dump option).
3756 Enable showing basic block boundaries (disabled in raw dumps).
3759 Enable showing virtual operands for every statement.
3762 Enable showing line numbers for statements.
3765 Enable showing the unique ID (`DECL_UID') for each variable.
3768 Turn on all options, except `raw', `slim' and `lineno'.
3770 The following tree dumps are possible:
3772 Dump before any tree based optimization, to `FILE.original'.
3775 Dump after all tree based optimization, to `FILE.optimized'.
3778 Dump after function inlining, to `FILE.inlined'.
3781 Dump each function before and after the gimplification pass
3782 to a file. The file name is made by appending `.gimple' to
3783 the source file name.
3786 Dump the control flow graph of each function to a file. The
3787 file name is made by appending `.cfg' to the source file name.
3790 Dump the control flow graph of each function to a file in VCG
3791 format. The file name is made by appending `.vcg' to the
3792 source file name. Note that if the file contains more than
3793 one function, the generated file cannot be used directly by
3794 VCG. You will need to cut and paste each function's graph
3795 into its own separate file first.
3798 Dump each function after copying loop headers. The file name
3799 is made by appending `.ch' to the source file name.
3802 Dump SSA related information to a file. The file name is
3803 made by appending `.ssa' to the source file name.
3806 Dump structure aliasing variable information to a file. This
3807 file name is made by appending `.salias' to the source file
3811 Dump aliasing information for each function. The file name
3812 is made by appending `.alias' to the source file name.
3815 Dump each function after CCP. The file name is made by
3816 appending `.ccp' to the source file name.
3819 Dump each function after STORE-CCP. The file name is made by
3820 appending `.storeccp' to the source file name.
3823 Dump trees after partial redundancy elimination. The file
3824 name is made by appending `.pre' to the source file name.
3827 Dump trees after full redundancy elimination. The file name
3828 is made by appending `.fre' to the source file name.
3831 Dump trees after copy propagation. The file name is made by
3832 appending `.copyprop' to the source file name.
3835 Dump trees after store copy-propagation. The file name is
3836 made by appending `.store_copyprop' to the source file name.
3839 Dump each function after dead code elimination. The file
3840 name is made by appending `.dce' to the source file name.
3843 Dump each function after adding mudflap instrumentation. The
3844 file name is made by appending `.mudflap' to the source file
3848 Dump each function after performing scalar replacement of
3849 aggregates. The file name is made by appending `.sra' to the
3853 Dump each function after performing code sinking. The file
3854 name is made by appending `.sink' to the source file name.
3857 Dump each function after applying dominator tree
3858 optimizations. The file name is made by appending `.dom' to
3859 the source file name.
3862 Dump each function after applying dead store elimination.
3863 The file name is made by appending `.dse' to the source file
3867 Dump each function after optimizing PHI nodes into
3868 straightline code. The file name is made by appending
3869 `.phiopt' to the source file name.
3872 Dump each function after forward propagating single use
3873 variables. The file name is made by appending `.forwprop' to
3874 the source file name.
3877 Dump each function after applying the copy rename
3878 optimization. The file name is made by appending
3879 `.copyrename' to the source file name.
3882 Dump each function after applying the named return value
3883 optimization on generic trees. The file name is made by
3884 appending `.nrv' to the source file name.
3887 Dump each function after applying vectorization of loops.
3888 The file name is made by appending `.vect' to the source file
3892 Dump each function after Value Range Propagation (VRP). The
3893 file name is made by appending `.vrp' to the source file name.
3896 Enable all the available tree dumps with the flags provided
3899 `-ftree-vectorizer-verbose=N'
3900 This option controls the amount of debugging output the vectorizer
3901 prints. This information is written to standard error, unless
3902 `-fdump-tree-all' or `-fdump-tree-vect' is specified, in which
3903 case it is output to the usual dump listing file, `.vect'.
3905 `-frandom-seed=STRING'
3906 This option provides a seed that GCC uses when it would otherwise
3907 use random numbers. It is used to generate certain symbol names
3908 that have to be different in every compiled file. It is also used
3909 to place unique stamps in coverage data files and the object files
3910 that produce them. You can use the `-frandom-seed' option to
3911 produce reproducibly identical object files.
3913 The STRING should be different for every file you compile.
3916 On targets that use instruction scheduling, this option controls
3917 the amount of debugging output the scheduler prints. This
3918 information is written to standard error, unless `-dS' or `-dR' is
3919 specified, in which case it is output to the usual dump listing
3920 file, `.sched' or `.sched2' respectively. However for N greater
3921 than nine, the output is always printed to standard error.
3923 For N greater than zero, `-fsched-verbose' outputs the same
3924 information as `-dRS'. For N greater than one, it also output
3925 basic block probabilities, detailed ready list information and
3926 unit/insn info. For N greater than two, it includes RTL at abort
3927 point, control-flow and regions info. And for N over four,
3928 `-fsched-verbose' also includes dependence info.
3931 Store the usual "temporary" intermediate files permanently; place
3932 them in the current directory and name them based on the source
3933 file. Thus, compiling `foo.c' with `-c -save-temps' would produce
3934 files `foo.i' and `foo.s', as well as `foo.o'. This creates a
3935 preprocessed `foo.i' output file even though the compiler now
3936 normally uses an integrated preprocessor.
3938 When used in combination with the `-x' command line option,
3939 `-save-temps' is sensible enough to avoid over writing an input
3940 source file with the same extension as an intermediate file. The
3941 corresponding intermediate file may be obtained by renaming the
3942 source file before using `-save-temps'.
3945 Report the CPU time taken by each subprocess in the compilation
3946 sequence. For C source files, this is the compiler proper and
3947 assembler (plus the linker if linking is done). The output looks
3953 The first number on each line is the "user time", that is time
3954 spent executing the program itself. The second number is "system
3955 time", time spent executing operating system routines on behalf of
3956 the program. Both numbers are in seconds.
3959 Run variable tracking pass. It computes where variables are
3960 stored at each position in code. Better debugging information is
3961 then generated (if the debugging information format supports this
3964 It is enabled by default when compiling with optimization (`-Os',
3965 `-O', `-O2', ...), debugging information (`-g') and the debug info
3968 `-print-file-name=LIBRARY'
3969 Print the full absolute name of the library file LIBRARY that
3970 would be used when linking--and don't do anything else. With this
3971 option, GCC does not compile or link anything; it just prints the
3974 `-print-multi-directory'
3975 Print the directory name corresponding to the multilib selected by
3976 any other switches present in the command line. This directory is
3977 supposed to exist in `GCC_EXEC_PREFIX'.
3980 Print the mapping from multilib directory names to compiler
3981 switches that enable them. The directory name is separated from
3982 the switches by `;', and each switch starts with an `@' instead of
3983 the `-', without spaces between multiple switches. This is
3984 supposed to ease shell-processing.
3986 `-print-prog-name=PROGRAM'
3987 Like `-print-file-name', but searches for a program such as `cpp'.
3989 `-print-libgcc-file-name'
3990 Same as `-print-file-name=libgcc.a'.
3992 This is useful when you use `-nostdlib' or `-nodefaultlibs' but
3993 you do want to link with `libgcc.a'. You can do
3995 gcc -nostdlib FILES... `gcc -print-libgcc-file-name`
3997 `-print-search-dirs'
3998 Print the name of the configured installation directory and a list
3999 of program and library directories `gcc' will search--and don't do
4002 This is useful when `gcc' prints the error message `installation
4003 problem, cannot exec cpp0: No such file or directory'. To resolve
4004 this you either need to put `cpp0' and the other compiler
4005 components where `gcc' expects to find them, or you can set the
4006 environment variable `GCC_EXEC_PREFIX' to the directory where you
4007 installed them. Don't forget the trailing `/'. *Note Environment
4011 Print the compiler's target machine (for example,
4012 `i686-pc-linux-gnu')--and don't do anything else.
4015 Print the compiler version (for example, `3.0')--and don't do
4019 Print the compiler's built-in specs--and don't do anything else.
4020 (This is used when GCC itself is being built.) *Note Spec Files::.
4022 `-feliminate-unused-debug-types'
4023 Normally, when producing DWARF2 output, GCC will emit debugging
4024 information for all types declared in a compilation unit,
4025 regardless of whether or not they are actually used in that
4026 compilation unit. Sometimes this is useful, such as if, in the
4027 debugger, you want to cast a value to a type that is not actually
4028 used in your program (but is declared). More often, however, this
4029 results in a significant amount of wasted space. With this
4030 option, GCC will avoid producing debug symbol output for types
4031 that are nowhere used in the source file being compiled.
4034 File: gcc.info, Node: Optimize Options, Next: Preprocessor Options, Prev: Debugging Options, Up: Invoking GCC
4036 3.10 Options That Control Optimization
4037 ======================================
4039 These options control various sorts of optimizations.
4041 Without any optimization option, the compiler's goal is to reduce the
4042 cost of compilation and to make debugging produce the expected results.
4043 Statements are independent: if you stop the program with a breakpoint
4044 between statements, you can then assign a new value to any variable or
4045 change the program counter to any other statement in the function and
4046 get exactly the results you would expect from the source code.
4048 Turning on optimization flags makes the compiler attempt to improve
4049 the performance and/or code size at the expense of compilation time and
4050 possibly the ability to debug the program.
4052 The compiler performs optimization based on the knowledge it has of
4053 the program. Optimization levels `-O' and above, in particular, enable
4054 _unit-at-a-time_ mode, which allows the compiler to consider
4055 information gained from later functions in the file when compiling a
4056 function. Compiling multiple files at once to a single output file in
4057 _unit-at-a-time_ mode allows the compiler to use information gained
4058 from all of the files when compiling each of them.
4060 Not all optimizations are controlled directly by a flag. Only
4061 optimizations that have a flag are listed.
4065 Optimize. Optimizing compilation takes somewhat more time, and a
4066 lot more memory for a large function.
4068 With `-O', the compiler tries to reduce code size and execution
4069 time, without performing any optimizations that take a great deal
4070 of compilation time.
4072 `-O' turns on the following optimization flags:
4075 -fguess-branch-probability
4082 -ftree-dominator-opts
4093 `-O' also turns on `-fomit-frame-pointer' on machines where doing
4094 so does not interfere with debugging.
4096 `-O' doesn't turn on `-ftree-sra' for the Ada compiler. This
4097 option must be explicitly specified on the command line to be
4098 enabled for the Ada compiler.
4101 Optimize even more. GCC performs nearly all supported
4102 optimizations that do not involve a space-speed tradeoff. The
4103 compiler does not perform loop unrolling or function inlining when
4104 you specify `-O2'. As compared to `-O', this option increases
4105 both compilation time and the performance of the generated code.
4107 `-O2' turns on all optimization flags specified by `-O'. It also
4108 turns on the following optimization flags:
4111 -foptimize-sibling-calls
4112 -fcse-follow-jumps -fcse-skip-blocks
4114 -fexpensive-optimizations
4116 -frerun-cse-after-loop -frerun-loop-opt
4119 -fschedule-insns -fschedule-insns2
4120 -fsched-interblock -fsched-spec
4123 -fdelete-null-pointer-checks
4124 -freorder-blocks -freorder-functions
4125 -falign-functions -falign-jumps
4126 -falign-loops -falign-labels
4130 Please note the warning under `-fgcse' about invoking `-O2' on
4131 programs that use computed gotos.
4134 Optimize yet more. `-O3' turns on all optimizations specified by
4135 `-O2' and also turns on the `-finline-functions',
4136 `-funswitch-loops' and `-fgcse-after-reload' options.
4139 Do not optimize. This is the default.
4142 Optimize for size. `-Os' enables all `-O2' optimizations that do
4143 not typically increase code size. It also performs further
4144 optimizations designed to reduce code size.
4146 `-Os' disables the following optimization flags:
4147 -falign-functions -falign-jumps -falign-loops
4148 -falign-labels -freorder-blocks -freorder-blocks-and-partition
4149 -fprefetch-loop-arrays -ftree-vect-loop-version
4151 If you use multiple `-O' options, with or without level numbers,
4152 the last such option is the one that is effective.
4154 Options of the form `-fFLAG' specify machine-independent flags. Most
4155 flags have both positive and negative forms; the negative form of
4156 `-ffoo' would be `-fno-foo'. In the table below, only one of the forms
4157 is listed--the one you typically will use. You can figure out the
4158 other form by either removing `no-' or adding it.
4160 The following options control specific optimizations. They are either
4161 activated by `-O' options or are related to ones that are. You can use
4162 the following flags in the rare cases when "fine-tuning" of
4163 optimizations to be performed is desired.
4165 `-fno-default-inline'
4166 Do not make member functions inline by default merely because they
4167 are defined inside the class scope (C++ only). Otherwise, when
4168 you specify `-O', member functions defined inside class scope are
4169 compiled inline by default; i.e., you don't need to add `inline'
4170 in front of the member function name.
4173 Always pop the arguments to each function call as soon as that
4174 function returns. For machines which must pop arguments after a
4175 function call, the compiler normally lets arguments accumulate on
4176 the stack for several function calls and pops them all at once.
4178 Disabled at levels `-O', `-O2', `-O3', `-Os'.
4181 Force memory operands to be copied into registers before doing
4182 arithmetic on them. This produces better code by making all memory
4183 references potential common subexpressions. When they are not
4184 common subexpressions, instruction combination should eliminate
4185 the separate register-load. This option is now a nop and will be
4189 Force memory address constants to be copied into registers before
4190 doing arithmetic on them.
4192 `-fomit-frame-pointer'
4193 Don't keep the frame pointer in a register for functions that
4194 don't need one. This avoids the instructions to save, set up and
4195 restore frame pointers; it also makes an extra register available
4196 in many functions. *It also makes debugging impossible on some
4199 On some machines, such as the VAX, this flag has no effect, because
4200 the standard calling sequence automatically handles the frame
4201 pointer and nothing is saved by pretending it doesn't exist. The
4202 machine-description macro `FRAME_POINTER_REQUIRED' controls
4203 whether a target machine supports this flag. *Note Register
4204 Usage: (gccint)Registers.
4206 Enabled at levels `-O', `-O2', `-O3', `-Os'.
4208 `-foptimize-sibling-calls'
4209 Optimize sibling and tail recursive calls.
4211 Enabled at levels `-O2', `-O3', `-Os'.
4214 Don't pay attention to the `inline' keyword. Normally this option
4215 is used to keep the compiler from expanding any functions inline.
4216 Note that if you are not optimizing, no functions can be expanded
4219 `-finline-functions'
4220 Integrate all simple functions into their callers. The compiler
4221 heuristically decides which functions are simple enough to be worth
4222 integrating in this way.
4224 If all calls to a given function are integrated, and the function
4225 is declared `static', then the function is normally not output as
4226 assembler code in its own right.
4228 Enabled at level `-O3'.
4230 `-finline-functions-called-once'
4231 Consider all `static' functions called once for inlining into their
4232 caller even if they are not marked `inline'. If a call to a given
4233 function is integrated, then the function is not output as
4234 assembler code in its own right.
4236 Enabled if `-funit-at-a-time' is enabled.
4239 Inline functions marked by `always_inline' and functions whose
4240 body seems smaller than the function call overhead early before
4241 doing `-fprofile-generate' instrumentation and real inlining pass.
4242 Doing so makes profiling significantly cheaper and usually
4243 inlining faster on programs having large chains of nested wrapper
4249 By default, GCC limits the size of functions that can be inlined.
4250 This flag allows the control of this limit for functions that are
4251 explicitly marked as inline (i.e., marked with the inline keyword
4252 or defined within the class definition in c++). N is the size of
4253 functions that can be inlined in number of pseudo instructions
4254 (not counting parameter handling). The default value of N is 600.
4255 Increasing this value can result in more inlined code at the cost
4256 of compilation time and memory consumption. Decreasing usually
4257 makes the compilation faster and less code will be inlined (which
4258 presumably means slower programs). This option is particularly
4259 useful for programs that use inlining heavily such as those based
4260 on recursive templates with C++.
4262 Inlining is actually controlled by a number of parameters, which
4263 may be specified individually by using `--param NAME=VALUE'. The
4264 `-finline-limit=N' option sets some of these parameters as follows:
4266 `max-inline-insns-single'
4269 `max-inline-insns-auto'
4273 is set to 130 or N/4, whichever is smaller.
4275 `max-inline-insns-rtl'
4278 See below for a documentation of the individual parameters
4279 controlling inlining.
4281 _Note:_ pseudo instruction represents, in this particular context,
4282 an abstract measurement of function's size. In no way does it
4283 represent a count of assembly instructions and as such its exact
4284 meaning might change from one release to an another.
4286 `-fkeep-inline-functions'
4287 In C, emit `static' functions that are declared `inline' into the
4288 object file, even if the function has been inlined into all of its
4289 callers. This switch does not affect functions using the `extern
4290 inline' extension in GNU C. In C++, emit any and all inline
4291 functions into the object file.
4293 `-fkeep-static-consts'
4294 Emit variables declared `static const' when optimization isn't
4295 turned on, even if the variables aren't referenced.
4297 GCC enables this option by default. If you want to force the
4298 compiler to check if the variable was referenced, regardless of
4299 whether or not optimization is turned on, use the
4300 `-fno-keep-static-consts' option.
4303 Attempt to merge identical constants (string constants and
4304 floating point constants) across compilation units.
4306 This option is the default for optimized compilation if the
4307 assembler and linker support it. Use `-fno-merge-constants' to
4308 inhibit this behavior.
4310 Enabled at levels `-O', `-O2', `-O3', `-Os'.
4312 `-fmerge-all-constants'
4313 Attempt to merge identical constants and identical variables.
4315 This option implies `-fmerge-constants'. In addition to
4316 `-fmerge-constants' this considers e.g. even constant initialized
4317 arrays or initialized constant variables with integral or floating
4318 point types. Languages like C or C++ require each non-automatic
4319 variable to have distinct location, so using this option will
4320 result in non-conforming behavior.
4323 Perform swing modulo scheduling immediately before the first
4324 scheduling pass. This pass looks at innermost loops and reorders
4325 their instructions by overlapping different iterations.
4327 `-fno-branch-count-reg'
4328 Do not use "decrement and branch" instructions on a count register,
4329 but instead generate a sequence of instructions that decrement a
4330 register, compare it against zero, then branch based upon the
4331 result. This option is only meaningful on architectures that
4332 support such instructions, which include x86, PowerPC, IA-64 and
4335 The default is `-fbranch-count-reg', enabled when
4336 `-fstrength-reduce' is enabled.
4339 Do not put function addresses in registers; make each instruction
4340 that calls a constant function contain the function's address
4343 This option results in less efficient code, but some strange hacks
4344 that alter the assembler output may be confused by the
4345 optimizations performed when this option is not used.
4347 The default is `-ffunction-cse'
4349 `-fno-zero-initialized-in-bss'
4350 If the target supports a BSS section, GCC by default puts
4351 variables that are initialized to zero into BSS. This can save
4352 space in the resulting code.
4354 This option turns off this behavior because some programs
4355 explicitly rely on variables going to the data section. E.g., so
4356 that the resulting executable can find the beginning of that
4357 section and/or make assumptions based on that.
4359 The default is `-fzero-initialized-in-bss'.
4362 For front-ends that support it, generate additional code to check
4363 that indices used to access arrays are within the declared range.
4364 This is currently only supported by the Java and Fortran
4365 front-ends, where this option defaults to true and false
4368 `-fmudflap -fmudflapth -fmudflapir'
4369 For front-ends that support it (C and C++), instrument all risky
4370 pointer/array dereferencing operations, some standard library
4371 string/heap functions, and some other associated constructs with
4372 range/validity tests. Modules so instrumented should be immune to
4373 buffer overflows, invalid heap use, and some other classes of C/C++
4374 programming errors. The instrumentation relies on a separate
4375 runtime library (`libmudflap'), which will be linked into a
4376 program if `-fmudflap' is given at link time. Run-time behavior
4377 of the instrumented program is controlled by the `MUDFLAP_OPTIONS'
4378 environment variable. See `env MUDFLAP_OPTIONS=-help a.out' for
4381 Use `-fmudflapth' instead of `-fmudflap' to compile and to link if
4382 your program is multi-threaded. Use `-fmudflapir', in addition to
4383 `-fmudflap' or `-fmudflapth', if instrumentation should ignore
4384 pointer reads. This produces less instrumentation (and therefore
4385 faster execution) and still provides some protection against
4386 outright memory corrupting writes, but allows erroneously read
4387 data to propagate within a program.
4390 Perform the optimizations of loop strength reduction and
4391 elimination of iteration variables.
4393 Enabled at levels `-O2', `-O3', `-Os'.
4396 Perform optimizations where we check to see if a jump branches to a
4397 location where another comparison subsumed by the first is found.
4398 If so, the first branch is redirected to either the destination of
4399 the second branch or a point immediately following it, depending
4400 on whether the condition is known to be true or false.
4402 Enabled at levels `-O2', `-O3', `-Os'.
4404 `-fcse-follow-jumps'
4405 In common subexpression elimination, scan through jump instructions
4406 when the target of the jump is not reached by any other path. For
4407 example, when CSE encounters an `if' statement with an `else'
4408 clause, CSE will follow the jump when the condition tested is
4411 Enabled at levels `-O2', `-O3', `-Os'.
4414 This is similar to `-fcse-follow-jumps', but causes CSE to follow
4415 jumps which conditionally skip over blocks. When CSE encounters a
4416 simple `if' statement with no else clause, `-fcse-skip-blocks'
4417 causes CSE to follow the jump around the body of the `if'.
4419 Enabled at levels `-O2', `-O3', `-Os'.
4421 `-frerun-cse-after-loop'
4422 Re-run common subexpression elimination after loop optimizations
4425 Enabled at levels `-O2', `-O3', `-Os'.
4428 Run the loop optimizer twice.
4430 Enabled at levels `-O2', `-O3', `-Os'.
4433 Perform a global common subexpression elimination pass. This pass
4434 also performs global constant and copy propagation.
4436 _Note:_ When compiling a program using computed gotos, a GCC
4437 extension, you may get better runtime performance if you disable
4438 the global common subexpression elimination pass by adding
4439 `-fno-gcse' to the command line.
4441 Enabled at levels `-O2', `-O3', `-Os'.
4444 When `-fgcse-lm' is enabled, global common subexpression
4445 elimination will attempt to move loads which are only killed by
4446 stores into themselves. This allows a loop containing a
4447 load/store sequence to be changed to a load outside the loop, and
4448 a copy/store within the loop.
4450 Enabled by default when gcse is enabled.
4453 When `-fgcse-sm' is enabled, a store motion pass is run after
4454 global common subexpression elimination. This pass will attempt
4455 to move stores out of loops. When used in conjunction with
4456 `-fgcse-lm', loops containing a load/store sequence can be changed
4457 to a load before the loop and a store after the loop.
4459 Not enabled at any optimization level.
4462 When `-fgcse-las' is enabled, the global common subexpression
4463 elimination pass eliminates redundant loads that come after stores
4464 to the same memory location (both partial and full redundancies).
4466 Not enabled at any optimization level.
4468 `-fgcse-after-reload'
4469 When `-fgcse-after-reload' is enabled, a redundant load elimination
4470 pass is performed after reload. The purpose of this pass is to
4471 cleanup redundant spilling.
4474 Perform loop optimizations: move constant expressions out of
4475 loops, simplify exit test conditions and optionally do
4476 strength-reduction as well.
4478 Enabled at levels `-O', `-O2', `-O3', `-Os'.
4481 Perform loop optimizations using the new loop optimizer. The
4482 optimizations (loop unrolling, peeling and unswitching, loop
4483 invariant motion) are enabled by separate flags.
4485 `-funsafe-loop-optimizations'
4486 If given, the loop optimizer will assume that loop indices do not
4487 overflow, and that the loops with nontrivial exit condition are not
4488 infinite. This enables a wider range of loop optimizations even if
4489 the loop optimizer itself cannot prove that these assumptions are
4490 valid. Using `-Wunsafe-loop-optimizations', the compiler will
4491 warn you if it finds this kind of loop.
4494 Perform cross-jumping transformation. This transformation unifies
4495 equivalent code and save code size. The resulting code may or may
4496 not perform better than without cross-jumping.
4498 Enabled at levels `-O2', `-O3', `-Os'.
4501 Attempt to transform conditional jumps into branch-less
4502 equivalents. This include use of conditional moves, min, max, set
4503 flags and abs instructions, and some tricks doable by standard
4504 arithmetics. The use of conditional execution on chips where it
4505 is available is controlled by `if-conversion2'.
4507 Enabled at levels `-O', `-O2', `-O3', `-Os'.
4510 Use conditional execution (where available) to transform
4511 conditional jumps into branch-less equivalents.
4513 Enabled at levels `-O', `-O2', `-O3', `-Os'.
4515 `-fdelete-null-pointer-checks'
4516 Use global dataflow analysis to identify and eliminate useless
4517 checks for null pointers. The compiler assumes that dereferencing
4518 a null pointer would have halted the program. If a pointer is
4519 checked after it has already been dereferenced, it cannot be null.
4521 In some environments, this assumption is not true, and programs can
4522 safely dereference null pointers. Use
4523 `-fno-delete-null-pointer-checks' to disable this optimization for
4524 programs which depend on that behavior.
4526 Enabled at levels `-O2', `-O3', `-Os'.
4528 `-fexpensive-optimizations'
4529 Perform a number of minor optimizations that are relatively
4532 Enabled at levels `-O2', `-O3', `-Os'.
4534 `-foptimize-register-move'
4536 Attempt to reassign register numbers in move instructions and as
4537 operands of other simple instructions in order to maximize the
4538 amount of register tying. This is especially helpful on machines
4539 with two-operand instructions.
4541 Note `-fregmove' and `-foptimize-register-move' are the same
4544 Enabled at levels `-O2', `-O3', `-Os'.
4547 If supported for the target machine, attempt to reorder
4548 instructions to exploit instruction slots available after delayed
4549 branch instructions.
4551 Enabled at levels `-O', `-O2', `-O3', `-Os'.
4554 If supported for the target machine, attempt to reorder
4555 instructions to eliminate execution stalls due to required data
4556 being unavailable. This helps machines that have slow floating
4557 point or memory load instructions by allowing other instructions
4558 to be issued until the result of the load or floating point
4559 instruction is required.
4561 Enabled at levels `-O2', `-O3', `-Os'.
4564 Similar to `-fschedule-insns', but requests an additional pass of
4565 instruction scheduling after register allocation has been done.
4566 This is especially useful on machines with a relatively small
4567 number of registers and where memory load instructions take more
4570 Enabled at levels `-O2', `-O3', `-Os'.
4572 `-fno-sched-interblock'
4573 Don't schedule instructions across basic blocks. This is normally
4574 enabled by default when scheduling before register allocation, i.e.
4575 with `-fschedule-insns' or at `-O2' or higher.
4578 Don't allow speculative motion of non-load instructions. This is
4579 normally enabled by default when scheduling before register
4580 allocation, i.e. with `-fschedule-insns' or at `-O2' or higher.
4583 Allow speculative motion of some load instructions. This only
4584 makes sense when scheduling before register allocation, i.e. with
4585 `-fschedule-insns' or at `-O2' or higher.
4587 `-fsched-spec-load-dangerous'
4588 Allow speculative motion of more load instructions. This only
4589 makes sense when scheduling before register allocation, i.e. with
4590 `-fschedule-insns' or at `-O2' or higher.
4592 `-fsched-stalled-insns=N'
4593 Define how many insns (if any) can be moved prematurely from the
4594 queue of stalled insns into the ready list, during the second
4597 `-fsched-stalled-insns-dep=N'
4598 Define how many insn groups (cycles) will be examined for a
4599 dependency on a stalled insn that is candidate for premature
4600 removal from the queue of stalled insns. Has an effect only
4601 during the second scheduling pass, and only if
4602 `-fsched-stalled-insns' is used and its value is not zero.
4604 `-fsched2-use-superblocks'
4605 When scheduling after register allocation, do use superblock
4606 scheduling algorithm. Superblock scheduling allows motion across
4607 basic block boundaries resulting on faster schedules. This option
4608 is experimental, as not all machine descriptions used by GCC model
4609 the CPU closely enough to avoid unreliable results from the
4612 This only makes sense when scheduling after register allocation,
4613 i.e. with `-fschedule-insns2' or at `-O2' or higher.
4615 `-fsched2-use-traces'
4616 Use `-fsched2-use-superblocks' algorithm when scheduling after
4617 register allocation and additionally perform code duplication in
4618 order to increase the size of superblocks using tracer pass. See
4619 `-ftracer' for details on trace formation.
4621 This mode should produce faster but significantly longer programs.
4622 Also without `-fbranch-probabilities' the traces constructed may
4623 not match the reality and hurt the performance. This only makes
4624 sense when scheduling after register allocation, i.e. with
4625 `-fschedule-insns2' or at `-O2' or higher.
4627 `-freschedule-modulo-scheduled-loops'
4628 The modulo scheduling comes before the traditional scheduling, if
4629 a loop was modulo scheduled we may want to prevent the later
4630 scheduling passes from changing its schedule, we use this option
4634 Enable values to be allocated in registers that will be clobbered
4635 by function calls, by emitting extra instructions to save and
4636 restore the registers around such calls. Such allocation is done
4637 only when it seems to result in better code than would otherwise
4640 This option is always enabled by default on certain machines,
4641 usually those which have no call-preserved registers to use
4644 Enabled at levels `-O2', `-O3', `-Os'.
4647 Perform Partial Redundancy Elimination (PRE) on trees. This flag
4648 is enabled by default at `-O2' and `-O3'.
4651 Perform Full Redundancy Elimination (FRE) on trees. The difference
4652 between FRE and PRE is that FRE only considers expressions that
4653 are computed on all paths leading to the redundant computation.
4654 This analysis faster than PRE, though it exposes fewer
4655 redundancies. This flag is enabled by default at `-O' and higher.
4658 Perform copy propagation on trees. This pass eliminates
4659 unnecessary copy operations. This flag is enabled by default at
4662 `-ftree-store-copy-prop'
4663 Perform copy propagation of memory loads and stores. This pass
4664 eliminates unnecessary copy operations in memory references
4665 (structures, global variables, arrays, etc). This flag is enabled
4666 by default at `-O2' and higher.
4669 Perform structural alias analysis on trees. This flag is enabled
4670 by default at `-O' and higher.
4673 Perform forward store motion on trees. This flag is enabled by
4674 default at `-O' and higher.
4677 Perform sparse conditional constant propagation (CCP) on trees.
4678 This pass only operates on local scalar variables and is enabled
4679 by default at `-O' and higher.
4682 Perform sparse conditional constant propagation (CCP) on trees.
4683 This pass operates on both local scalar variables and memory
4684 stores and loads (global variables, structures, arrays, etc).
4685 This flag is enabled by default at `-O2' and higher.
4688 Perform dead code elimination (DCE) on trees. This flag is
4689 enabled by default at `-O' and higher.
4691 `-ftree-dominator-opts'
4692 Perform a variety of simple scalar cleanups (constant/copy
4693 propagation, redundancy elimination, range propagation and
4694 expression simplification) based on a dominator tree traversal.
4695 This also performs jump threading (to reduce jumps to jumps). This
4696 flag is enabled by default at `-O' and higher.
4699 Perform loop header copying on trees. This is beneficial since it
4700 increases effectiveness of code motion optimizations. It also
4701 saves one jump. This flag is enabled by default at `-O' and
4702 higher. It is not enabled for `-Os', since it usually increases
4705 `-ftree-loop-optimize'
4706 Perform loop optimizations on trees. This flag is enabled by
4707 default at `-O' and higher.
4709 `-ftree-loop-linear'
4710 Perform linear loop transformations on tree. This flag can
4711 improve cache performance and allow further loop optimizations to
4715 Perform loop invariant motion on trees. This pass moves only
4716 invariants that would be hard to handle at RTL level (function
4717 calls, operations that expand to nontrivial sequences of insns).
4718 With `-funswitch-loops' it also moves operands of conditions that
4719 are invariant out of the loop, so that we can use just trivial
4720 invariantness analysis in loop unswitching. The pass also includes
4723 `-ftree-loop-ivcanon'
4724 Create a canonical counter for number of iterations in the loop
4725 for that determining number of iterations requires complicated
4726 analysis. Later optimizations then may determine the number
4727 easily. Useful especially in connection with unrolling.
4730 Perform induction variable optimizations (strength reduction,
4731 induction variable merging and induction variable elimination) on
4735 Perform scalar replacement of aggregates. This pass replaces
4736 structure references with scalars to prevent committing structures
4737 to memory too early. This flag is enabled by default at `-O' and
4741 Perform copy renaming on trees. This pass attempts to rename
4742 compiler temporaries to other variables at copy locations, usually
4743 resulting in variable names which more closely resemble the
4744 original variables. This flag is enabled by default at `-O' and
4748 Perform temporary expression replacement during the SSA->normal
4749 phase. Single use/single def temporaries are replaced at their
4750 use location with their defining expression. This results in
4751 non-GIMPLE code, but gives the expanders much more complex trees
4752 to work on resulting in better RTL generation. This is enabled by
4753 default at `-O' and higher.
4756 Perform live range splitting during the SSA->normal phase.
4757 Distinct live ranges of a variable are split into unique
4758 variables, allowing for better optimization later. This is
4759 enabled by default at `-O' and higher.
4762 Perform loop vectorization on trees.
4764 `-ftree-vect-loop-version'
4765 Perform loop versioning when doing loop vectorization on trees.
4766 When a loop appears to be vectorizable except that data alignment
4767 or data dependence cannot be determined at compile time then
4768 vectorized and non-vectorized versions of the loop are generated
4769 along with runtime checks for alignment or dependence to control
4770 which version is executed. This option is enabled by default
4771 except at level `-Os' where it is disabled.
4774 Perform Value Range Propagation on trees. This is similar to the
4775 constant propagation pass, but instead of values, ranges of values
4776 are propagated. This allows the optimizers to remove unnecessary
4777 range checks like array bound checks and null pointer checks.
4778 This is enabled by default at `-O2' and higher. Null pointer check
4779 elimination is only done if `-fdelete-null-pointer-checks' is
4783 Perform tail duplication to enlarge superblock size. This
4784 transformation simplifies the control flow of the function
4785 allowing other optimizations to do better job.
4788 Unroll loops whose number of iterations can be determined at
4789 compile time or upon entry to the loop. `-funroll-loops' implies
4790 both `-fstrength-reduce' and `-frerun-cse-after-loop'. This
4791 option makes code larger, and may or may not make it run faster.
4793 `-funroll-all-loops'
4794 Unroll all loops, even if their number of iterations is uncertain
4795 when the loop is entered. This usually makes programs run more
4796 slowly. `-funroll-all-loops' implies the same options as
4799 `-fsplit-ivs-in-unroller'
4800 Enables expressing of values of induction variables in later
4801 iterations of the unrolled loop using the value in the first
4802 iteration. This breaks long dependency chains, thus improving
4803 efficiency of the scheduling passes.
4805 Combination of `-fweb' and CSE is often sufficient to obtain the
4806 same effect. However in cases the loop body is more complicated
4807 than a single basic block, this is not reliable. It also does not
4808 work at all on some of the architectures due to restrictions in
4811 This optimization is enabled by default.
4813 `-fvariable-expansion-in-unroller'
4814 With this option, the compiler will create multiple copies of some
4815 local variables when unrolling a loop which can result in superior
4818 `-fprefetch-loop-arrays'
4819 If supported by the target machine, generate instructions to
4820 prefetch memory to improve the performance of loops that access
4823 These options may generate better or worse code; results are highly
4824 dependent on the structure of loops within the source code.
4828 Disable any machine-specific peephole optimizations. The
4829 difference between `-fno-peephole' and `-fno-peephole2' is in how
4830 they are implemented in the compiler; some targets use one, some
4831 use the other, a few use both.
4833 `-fpeephole' is enabled by default. `-fpeephole2' enabled at
4834 levels `-O2', `-O3', `-Os'.
4836 `-fno-guess-branch-probability'
4837 Do not guess branch probabilities using heuristics.
4839 GCC will use heuristics to guess branch probabilities if they are
4840 not provided by profiling feedback (`-fprofile-arcs'). These
4841 heuristics are based on the control flow graph. If some branch
4842 probabilities are specified by `__builtin_expect', then the
4843 heuristics will be used to guess branch probabilities for the rest
4844 of the control flow graph, taking the `__builtin_expect' info into
4845 account. The interactions between the heuristics and
4846 `__builtin_expect' can be complex, and in some cases, it may be
4847 useful to disable the heuristics so that the effects of
4848 `__builtin_expect' are easier to understand.
4850 The default is `-fguess-branch-probability' at levels `-O', `-O2',
4854 Reorder basic blocks in the compiled function in order to reduce
4855 number of taken branches and improve code locality.
4857 Enabled at levels `-O2', `-O3'.
4859 `-freorder-blocks-and-partition'
4860 In addition to reordering basic blocks in the compiled function,
4861 in order to reduce number of taken branches, partitions hot and
4862 cold basic blocks into separate sections of the assembly and .o
4863 files, to improve paging and cache locality performance.
4865 This optimization is automatically turned off in the presence of
4866 exception handling, for linkonce sections, for functions with a
4867 user-defined section attribute and on any architecture that does
4868 not support named sections.
4870 `-freorder-functions'
4871 Reorder functions in the object file in order to improve code
4872 locality. This is implemented by using special subsections
4873 `.text.hot' for most frequently executed functions and
4874 `.text.unlikely' for unlikely executed functions. Reordering is
4875 done by the linker so object file format must support named
4876 sections and linker must place them in a reasonable way.
4878 Also profile feedback must be available in to make this option
4879 effective. See `-fprofile-arcs' for details.
4881 Enabled at levels `-O2', `-O3', `-Os'.
4884 Allows the compiler to assume the strictest aliasing rules
4885 applicable to the language being compiled. For C (and C++), this
4886 activates optimizations based on the type of expressions. In
4887 particular, an object of one type is assumed never to reside at
4888 the same address as an object of a different type, unless the
4889 types are almost the same. For example, an `unsigned int' can
4890 alias an `int', but not a `void*' or a `double'. A character type
4891 may alias any other type.
4893 Pay special attention to code like this:
4904 The practice of reading from a different union member than the one
4905 most recently written to (called "type-punning") is common. Even
4906 with `-fstrict-aliasing', type-punning is allowed, provided the
4907 memory is accessed through the union type. So, the code above
4908 will work as expected. However, this code might not:
4917 Every language that wishes to perform language-specific alias
4918 analysis should define a function that computes, given an `tree'
4919 node, an alias set for the node. Nodes in different alias sets
4920 are not allowed to alias. For an example, see the C front-end
4921 function `c_get_alias_set'.
4923 Enabled at levels `-O2', `-O3', `-Os'.
4926 `-falign-functions=N'
4927 Align the start of functions to the next power-of-two greater than
4928 N, skipping up to N bytes. For instance, `-falign-functions=32'
4929 aligns functions to the next 32-byte boundary, but
4930 `-falign-functions=24' would align to the next 32-byte boundary
4931 only if this can be done by skipping 23 bytes or less.
4933 `-fno-align-functions' and `-falign-functions=1' are equivalent
4934 and mean that functions will not be aligned.
4936 Some assemblers only support this flag when N is a power of two;
4937 in that case, it is rounded up.
4939 If N is not specified or is zero, use a machine-dependent default.
4941 Enabled at levels `-O2', `-O3'.
4945 Align all branch targets to a power-of-two boundary, skipping up to
4946 N bytes like `-falign-functions'. This option can easily make
4947 code slower, because it must insert dummy operations for when the
4948 branch target is reached in the usual flow of the code.
4950 `-fno-align-labels' and `-falign-labels=1' are equivalent and mean
4951 that labels will not be aligned.
4953 If `-falign-loops' or `-falign-jumps' are applicable and are
4954 greater than this value, then their values are used instead.
4956 If N is not specified or is zero, use a machine-dependent default
4957 which is very likely to be `1', meaning no alignment.
4959 Enabled at levels `-O2', `-O3'.
4963 Align loops to a power-of-two boundary, skipping up to N bytes
4964 like `-falign-functions'. The hope is that the loop will be
4965 executed many times, which will make up for any execution of the
4968 `-fno-align-loops' and `-falign-loops=1' are equivalent and mean
4969 that loops will not be aligned.
4971 If N is not specified or is zero, use a machine-dependent default.
4973 Enabled at levels `-O2', `-O3'.
4977 Align branch targets to a power-of-two boundary, for branch targets
4978 where the targets can only be reached by jumping, skipping up to N
4979 bytes like `-falign-functions'. In this case, no dummy operations
4982 `-fno-align-jumps' and `-falign-jumps=1' are equivalent and mean
4983 that loops will not be aligned.
4985 If N is not specified or is zero, use a machine-dependent default.
4987 Enabled at levels `-O2', `-O3'.
4990 Parse the whole compilation unit before starting to produce code.
4991 This allows some extra optimizations to take place but consumes
4992 more memory (in general). There are some compatibility issues
4993 with _unit-at-at-time_ mode:
4994 * enabling _unit-at-a-time_ mode may change the order in which
4995 functions, variables, and top-level `asm' statements are
4996 emitted, and will likely break code relying on some particular
4997 ordering. The majority of such top-level `asm' statements,
4998 though, can be replaced by `section' attributes.
5000 * _unit-at-a-time_ mode removes unreferenced static variables
5001 and functions. This may result in undefined references when
5002 an `asm' statement refers directly to variables or functions
5003 that are otherwise unused. In that case either the
5004 variable/function shall be listed as an operand of the `asm'
5005 statement operand or, in the case of top-level `asm'
5006 statements the attribute `used' shall be used on the
5009 * Static functions now can use non-standard passing conventions
5010 that may break `asm' statements calling functions directly.
5011 Again, attribute `used' will prevent this behavior.
5013 As a temporary workaround, `-fno-unit-at-a-time' can be used, but
5014 this scheme may not be supported by future releases of GCC.
5016 Enabled at levels `-O', `-O2', `-O3', `-Os'.
5019 Constructs webs as commonly used for register allocation purposes
5020 and assign each web individual pseudo register. This allows the
5021 register allocation pass to operate on pseudos directly, but also
5022 strengthens several other optimization passes, such as CSE, loop
5023 optimizer and trivial dead code remover. It can, however, make
5024 debugging impossible, since variables will no longer stay in a
5027 Enabled by default with `-funroll-loops'.
5030 Assume that the current compilation unit represents whole program
5031 being compiled. All public functions and variables with the
5032 exception of `main' and those merged by attribute
5033 `externally_visible' become static functions and in a affect gets
5034 more aggressively optimized by interprocedural optimizers. While
5035 this option is equivalent to proper use of `static' keyword for
5036 programs consisting of single file, in combination with option
5037 `--combine' this flag can be used to compile most of smaller scale
5038 C programs since the functions and variables become local for the
5039 whole combined compilation unit, not for the single source file
5042 `-fno-cprop-registers'
5043 After register allocation and post-register allocation instruction
5044 splitting, we perform a copy-propagation pass to try to reduce
5045 scheduling dependencies and occasionally eliminate the copy.
5047 Disabled at levels `-O', `-O2', `-O3', `-Os'.
5049 `-fprofile-generate'
5050 Enable options usually used for instrumenting application to
5051 produce profile useful for later recompilation with profile
5052 feedback based optimization. You must use `-fprofile-generate'
5053 both when compiling and when linking your program.
5055 The following options are enabled: `-fprofile-arcs',
5056 `-fprofile-values', `-fvpt'.
5059 Enable profile feedback directed optimizations, and optimizations
5060 generally profitable only with profile feedback available.
5062 The following options are enabled: `-fbranch-probabilities',
5063 `-fvpt', `-funroll-loops', `-fpeel-loops', `-ftracer',
5064 `-fno-loop-optimize'.
5067 The following options control compiler behavior regarding floating
5068 point arithmetic. These options trade off between speed and
5069 correctness. All must be specifically enabled.
5072 Do not store floating point variables in registers, and inhibit
5073 other options that might change whether a floating point value is
5074 taken from a register or memory.
5076 This option prevents undesirable excess precision on machines such
5077 as the 68000 where the floating registers (of the 68881) keep more
5078 precision than a `double' is supposed to have. Similarly for the
5079 x86 architecture. For most programs, the excess precision does
5080 only good, but a few programs rely on the precise definition of
5081 IEEE floating point. Use `-ffloat-store' for such programs, after
5082 modifying them to store all pertinent intermediate computations
5086 Sets `-fno-math-errno', `-funsafe-math-optimizations',
5087 `-fno-trapping-math', `-ffinite-math-only', `-fno-rounding-math',
5088 `-fno-signaling-nans' and `fcx-limited-range'.
5090 This option causes the preprocessor macro `__FAST_MATH__' to be
5093 This option should never be turned on by any `-O' option since it
5094 can result in incorrect output for programs which depend on an
5095 exact implementation of IEEE or ISO rules/specifications for math
5099 Do not set ERRNO after calling math functions that are executed
5100 with a single instruction, e.g., sqrt. A program that relies on
5101 IEEE exceptions for math error handling may want to use this flag
5102 for speed while maintaining IEEE arithmetic compatibility.
5104 This option should never be turned on by any `-O' option since it
5105 can result in incorrect output for programs which depend on an
5106 exact implementation of IEEE or ISO rules/specifications for math
5109 The default is `-fmath-errno'.
5111 On Darwin systems, the math library never sets `errno'. There is
5112 therefore no reason for the compiler to consider the possibility
5113 that it might, and `-fno-math-errno' is the default.
5115 `-funsafe-math-optimizations'
5116 Allow optimizations for floating-point arithmetic that (a) assume
5117 that arguments and results are valid and (b) may violate IEEE or
5118 ANSI standards. When used at link-time, it may include libraries
5119 or startup files that change the default FPU control word or other
5120 similar optimizations.
5122 This option should never be turned on by any `-O' option since it
5123 can result in incorrect output for programs which depend on an
5124 exact implementation of IEEE or ISO rules/specifications for math
5127 The default is `-fno-unsafe-math-optimizations'.
5129 `-ffinite-math-only'
5130 Allow optimizations for floating-point arithmetic that assume that
5131 arguments and results are not NaNs or +-Infs.
5133 This option should never be turned on by any `-O' option since it
5134 can result in incorrect output for programs which depend on an
5135 exact implementation of IEEE or ISO rules/specifications.
5137 The default is `-fno-finite-math-only'.
5139 `-fno-trapping-math'
5140 Compile code assuming that floating-point operations cannot
5141 generate user-visible traps. These traps include division by
5142 zero, overflow, underflow, inexact result and invalid operation.
5143 This option implies `-fno-signaling-nans'. Setting this option
5144 may allow faster code if one relies on "non-stop" IEEE arithmetic,
5147 This option should never be turned on by any `-O' option since it
5148 can result in incorrect output for programs which depend on an
5149 exact implementation of IEEE or ISO rules/specifications for math
5152 The default is `-ftrapping-math'.
5155 Disable transformations and optimizations that assume default
5156 floating point rounding behavior. This is round-to-zero for all
5157 floating point to integer conversions, and round-to-nearest for
5158 all other arithmetic truncations. This option should be specified
5159 for programs that change the FP rounding mode dynamically, or that
5160 may be executed with a non-default rounding mode. This option
5161 disables constant folding of floating point expressions at
5162 compile-time (which may be affected by rounding mode) and
5163 arithmetic transformations that are unsafe in the presence of
5164 sign-dependent rounding modes.
5166 The default is `-fno-rounding-math'.
5168 This option is experimental and does not currently guarantee to
5169 disable all GCC optimizations that are affected by rounding mode.
5170 Future versions of GCC may provide finer control of this setting
5171 using C99's `FENV_ACCESS' pragma. This command line option will
5172 be used to specify the default state for `FENV_ACCESS'.
5175 Compile code assuming that IEEE signaling NaNs may generate
5176 user-visible traps during floating-point operations. Setting this
5177 option disables optimizations that may change the number of
5178 exceptions visible with signaling NaNs. This option implies
5181 This option causes the preprocessor macro `__SUPPORT_SNAN__' to be
5184 The default is `-fno-signaling-nans'.
5186 This option is experimental and does not currently guarantee to
5187 disable all GCC optimizations that affect signaling NaN behavior.
5189 `-fsingle-precision-constant'
5190 Treat floating point constant as single precision constant instead
5191 of implicitly converting it to double precision constant.
5193 `-fcx-limited-range'
5194 `-fno-cx-limited-range'
5195 When enabled, this option states that a range reduction step is not
5196 needed when performing complex division. The default is
5197 `-fno-cx-limited-range', but is enabled by `-ffast-math'.
5199 This option controls the default setting of the ISO C99
5200 `CX_LIMITED_RANGE' pragma. Nevertheless, the option applies to
5204 The following options control optimizations that may improve
5205 performance, but are not enabled by any `-O' options. This section
5206 includes experimental options that may produce broken code.
5208 `-fbranch-probabilities'
5209 After running a program compiled with `-fprofile-arcs' (*note
5210 Options for Debugging Your Program or `gcc': Debugging Options.),
5211 you can compile it a second time using `-fbranch-probabilities',
5212 to improve optimizations based on the number of times each branch
5213 was taken. When the program compiled with `-fprofile-arcs' exits
5214 it saves arc execution counts to a file called `SOURCENAME.gcda'
5215 for each source file The information in this data file is very
5216 dependent on the structure of the generated code, so you must use
5217 the same source code and the same optimization options for both
5220 With `-fbranch-probabilities', GCC puts a `REG_BR_PROB' note on
5221 each `JUMP_INSN' and `CALL_INSN'. These can be used to improve
5222 optimization. Currently, they are only used in one place: in
5223 `reorg.c', instead of guessing which path a branch is mostly to
5224 take, the `REG_BR_PROB' values are used to exactly determine which
5225 path is taken more often.
5228 If combined with `-fprofile-arcs', it adds code so that some data
5229 about values of expressions in the program is gathered.
5231 With `-fbranch-probabilities', it reads back the data gathered
5232 from profiling values of expressions and adds `REG_VALUE_PROFILE'
5233 notes to instructions for their later usage in optimizations.
5235 Enabled with `-fprofile-generate' and `-fprofile-use'.
5238 If combined with `-fprofile-arcs', it instructs the compiler to add
5239 a code to gather information about values of expressions.
5241 With `-fbranch-probabilities', it reads back the data gathered and
5242 actually performs the optimizations based on them. Currently the
5243 optimizations include specialization of division operation using
5244 the knowledge about the value of the denominator.
5246 `-frename-registers'
5247 Attempt to avoid false dependencies in scheduled code by making use
5248 of registers left over after register allocation. This
5249 optimization will most benefit processors with lots of registers.
5250 Depending on the debug information format adopted by the target,
5251 however, it can make debugging impossible, since variables will no
5252 longer stay in a "home register".
5254 Enabled by default with `-funroll-loops'.
5257 Perform tail duplication to enlarge superblock size. This
5258 transformation simplifies the control flow of the function
5259 allowing other optimizations to do better job.
5261 Enabled with `-fprofile-use'.
5264 Unroll loops whose number of iterations can be determined at
5265 compile time or upon entry to the loop. `-funroll-loops' implies
5266 `-frerun-cse-after-loop', `-fweb' and `-frename-registers'. It
5267 also turns on complete loop peeling (i.e. complete removal of
5268 loops with small constant number of iterations). This option
5269 makes code larger, and may or may not make it run faster.
5271 Enabled with `-fprofile-use'.
5273 `-funroll-all-loops'
5274 Unroll all loops, even if their number of iterations is uncertain
5275 when the loop is entered. This usually makes programs run more
5276 slowly. `-funroll-all-loops' implies the same options as
5280 Peels the loops for that there is enough information that they do
5281 not roll much (from profile feedback). It also turns on complete
5282 loop peeling (i.e. complete removal of loops with small constant
5283 number of iterations).
5285 Enabled with `-fprofile-use'.
5287 `-fmove-loop-invariants'
5288 Enables the loop invariant motion pass in the new loop optimizer.
5289 Enabled at level `-O1'
5292 Move branches with loop invariant conditions out of the loop, with
5293 duplicates of the loop on both branches (modified according to
5294 result of the condition).
5296 `-fprefetch-loop-arrays'
5297 If supported by the target machine, generate instructions to
5298 prefetch memory to improve the performance of loops that access
5301 Disabled at level `-Os'.
5303 `-ffunction-sections'
5305 Place each function or data item into its own section in the output
5306 file if the target supports arbitrary sections. The name of the
5307 function or the name of the data item determines the section's name
5310 Use these options on systems where the linker can perform
5311 optimizations to improve locality of reference in the instruction
5312 space. Most systems using the ELF object format and SPARC
5313 processors running Solaris 2 have linkers with such optimizations.
5314 AIX may have these optimizations in the future.
5316 Only use these options when there are significant benefits from
5317 doing so. When you specify these options, the assembler and
5318 linker will create larger object and executable files and will
5319 also be slower. You will not be able to use `gprof' on all
5320 systems if you specify this option and you may have problems with
5321 debugging if you specify both this option and `-g'.
5323 `-fbranch-target-load-optimize'
5324 Perform branch target register load optimization before prologue /
5325 epilogue threading. The use of target registers can typically be
5326 exposed only during reload, thus hoisting loads out of loops and
5327 doing inter-block scheduling needs a separate optimization pass.
5329 `-fbranch-target-load-optimize2'
5330 Perform branch target register load optimization after prologue /
5333 `-fbtr-bb-exclusive'
5334 When performing branch target register load optimization, don't
5335 reuse branch target registers in within any basic block.
5338 Emit extra code to check for buffer overflows, such as stack
5339 smashing attacks. This is done by adding a guard variable to
5340 functions with vulnerable objects. This includes functions that
5341 call alloca, and functions with buffers larger than 8 bytes. The
5342 guards are initialized when a function is entered and then checked
5343 when the function exits. If a guard check fails, an error message
5344 is printed and the program exits.
5346 `-fstack-protector-all'
5347 Like `-fstack-protector' except that all functions are protected.
5349 `--param NAME=VALUE'
5350 In some places, GCC uses various constants to control the amount of
5351 optimization that is done. For example, GCC will not inline
5352 functions that contain more that a certain number of instructions.
5353 You can control some of these constants on the command-line using
5354 the `--param' option.
5356 The names of specific parameters, and the meaning of the values,
5357 are tied to the internals of the compiler, and are subject to
5358 change without notice in future releases.
5360 In each case, the VALUE is an integer. The allowable choices for
5361 NAME are given in the following table:
5363 `salias-max-implicit-fields'
5364 The maximum number of fields in a variable without direct
5365 structure accesses for which structure aliasing will consider
5366 trying to track each field. The default is 5
5368 `sra-max-structure-size'
5369 The maximum structure size, in bytes, at which the scalar
5370 replacement of aggregates (SRA) optimization will perform
5371 block copies. The default value, 0, implies that GCC will
5372 select the most appropriate size itself.
5374 `sra-field-structure-ratio'
5375 The threshold ratio (as a percentage) between instantiated
5376 fields and the complete structure size. We say that if the
5377 ratio of the number of bytes in instantiated fields to the
5378 number of bytes in the complete structure exceeds this
5379 parameter, then block copies are not used. The default is 75.
5381 `max-crossjump-edges'
5382 The maximum number of incoming edges to consider for
5383 crossjumping. The algorithm used by `-fcrossjumping' is
5384 O(N^2) in the number of edges incoming to each block.
5385 Increasing values mean more aggressive optimization, making
5386 the compile time increase with probably small improvement in
5389 `min-crossjump-insns'
5390 The minimum number of instructions which must be matched at
5391 the end of two blocks before crossjumping will be performed
5392 on them. This value is ignored in the case where all
5393 instructions in the block being crossjumped from are matched.
5394 The default value is 5.
5396 `max-grow-copy-bb-insns'
5397 The maximum code size expansion factor when copying basic
5398 blocks instead of jumping. The expansion is relative to a
5399 jump instruction. The default value is 8.
5401 `max-goto-duplication-insns'
5402 The maximum number of instructions to duplicate to a block
5403 that jumps to a computed goto. To avoid O(N^2) behavior in a
5404 number of passes, GCC factors computed gotos early in the
5405 compilation process, and unfactors them as late as possible.
5406 Only computed jumps at the end of a basic blocks with no more
5407 than max-goto-duplication-insns are unfactored. The default
5410 `max-delay-slot-insn-search'
5411 The maximum number of instructions to consider when looking
5412 for an instruction to fill a delay slot. If more than this
5413 arbitrary number of instructions is searched, the time
5414 savings from filling the delay slot will be minimal so stop
5415 searching. Increasing values mean more aggressive
5416 optimization, making the compile time increase with probably
5417 small improvement in executable run time.
5419 `max-delay-slot-live-search'
5420 When trying to fill delay slots, the maximum number of
5421 instructions to consider when searching for a block with
5422 valid live register information. Increasing this arbitrarily
5423 chosen value means more aggressive optimization, increasing
5424 the compile time. This parameter should be removed when the
5425 delay slot code is rewritten to maintain the control-flow
5429 The approximate maximum amount of memory that will be
5430 allocated in order to perform the global common subexpression
5431 elimination optimization. If more memory than specified is
5432 required, the optimization will not be done.
5435 The maximum number of passes of GCSE to run. The default is
5438 `max-pending-list-length'
5439 The maximum number of pending dependencies scheduling will
5440 allow before flushing the current state and starting over.
5441 Large functions with few branches or calls can create
5442 excessively large lists which needlessly consume memory and
5445 `max-inline-insns-single'
5446 Several parameters control the tree inliner used in gcc.
5447 This number sets the maximum number of instructions (counted
5448 in GCC's internal representation) in a single function that
5449 the tree inliner will consider for inlining. This only
5450 affects functions declared inline and methods implemented in
5451 a class declaration (C++). The default value is 450.
5453 `max-inline-insns-auto'
5454 When you use `-finline-functions' (included in `-O3'), a lot
5455 of functions that would otherwise not be considered for
5456 inlining by the compiler will be investigated. To those
5457 functions, a different (more restrictive) limit compared to
5458 functions declared inline can be applied. The default value
5461 `large-function-insns'
5462 The limit specifying really large functions. For functions
5463 larger than this limit after inlining inlining is constrained
5464 by `--param large-function-growth'. This parameter is useful
5465 primarily to avoid extreme compilation time caused by
5466 non-linear algorithms used by the backend. This parameter is
5467 ignored when `-funit-at-a-time' is not used. The default
5470 `large-function-growth'
5471 Specifies maximal growth of large function caused by inlining
5472 in percents. This parameter is ignored when
5473 `-funit-at-a-time' is not used. The default value is 100
5474 which limits large function growth to 2.0 times the original
5478 The limit specifying large translation unit. Growth caused
5479 by inlining of units larger than this limit is limited by
5480 `--param inline-unit-growth'. For small units this might be
5481 too tight (consider unit consisting of function A that is
5482 inline and B that just calls A three time. If B is small
5483 relative to A, the growth of unit is 300\% and yet such
5484 inlining is very sane. For very large units consisting of
5485 small inlininable functions however the overall unit growth
5486 limit is needed to avoid exponential explosion of code size.
5487 Thus for smaller units, the size is increased to `--param
5488 large-unit-insns' before aplying `--param
5489 inline-unit-growth'. The default is 10000
5491 `inline-unit-growth'
5492 Specifies maximal overall growth of the compilation unit
5493 caused by inlining. This parameter is ignored when
5494 `-funit-at-a-time' is not used. The default value is 50
5495 which limits unit growth to 1.5 times the original size.
5497 `max-inline-insns-recursive'
5498 `max-inline-insns-recursive-auto'
5499 Specifies maximum number of instructions out-of-line copy of
5500 self recursive inline function can grow into by performing
5503 For functions declared inline `--param
5504 max-inline-insns-recursive' is taken into account. For
5505 function not declared inline, recursive inlining happens only
5506 when `-finline-functions' (included in `-O3') is enabled and
5507 `--param max-inline-insns-recursive-auto' is used. The
5508 default value is 450.
5510 `max-inline-recursive-depth'
5511 `max-inline-recursive-depth-auto'
5512 Specifies maximum recursion depth used by the recursive
5515 For functions declared inline `--param
5516 max-inline-recursive-depth' is taken into account. For
5517 function not declared inline, recursive inlining happens only
5518 when `-finline-functions' (included in `-O3') is enabled and
5519 `--param max-inline-recursive-depth-auto' is used. The
5520 default value is 450.
5522 `min-inline-recursive-probability'
5523 Recursive inlining is profitable only for function having
5524 deep recursion in average and can hurt for function having
5525 little recursion depth by increasing the prologue size or
5526 complexity of function body to other optimizers.
5528 When profile feedback is available (see `-fprofile-generate')
5529 the actual recursion depth can be guessed from probability
5530 that function will recurse via given call expression. This
5531 parameter limits inlining only to call expression whose
5532 probability exceeds given threshold (in percents). The
5533 default value is 10.
5536 Specify cost of call instruction relative to simple
5537 arithmetics operations (having cost of 1). Increasing this
5538 cost disqualifies inlining of non-leaf functions and at the
5539 same time increases size of leaf function that is believed to
5540 reduce function size by being inlined. In effect it
5541 increases amount of inlining for code having large
5542 abstraction penalty (many functions that just pass the
5543 arguments to other functions) and decrease inlining for code
5544 with low abstraction penalty. The default value is 16.
5546 `max-unrolled-insns'
5547 The maximum number of instructions that a loop should have if
5548 that loop is unrolled, and if the loop is unrolled, it
5549 determines how many times the loop code is unrolled.
5551 `max-average-unrolled-insns'
5552 The maximum number of instructions biased by probabilities of
5553 their execution that a loop should have if that loop is
5554 unrolled, and if the loop is unrolled, it determines how many
5555 times the loop code is unrolled.
5558 The maximum number of unrollings of a single loop.
5561 The maximum number of instructions that a loop should have if
5562 that loop is peeled, and if the loop is peeled, it determines
5563 how many times the loop code is peeled.
5566 The maximum number of peelings of a single loop.
5568 `max-completely-peeled-insns'
5569 The maximum number of insns of a completely peeled loop.
5571 `max-completely-peel-times'
5572 The maximum number of iterations of a loop to be suitable for
5575 `max-unswitch-insns'
5576 The maximum number of insns of an unswitched loop.
5578 `max-unswitch-level'
5579 The maximum number of branches unswitched in a single loop.
5582 The minimum cost of an expensive expression in the loop
5585 `iv-consider-all-candidates-bound'
5586 Bound on number of candidates for induction variables below
5587 that all candidates are considered for each use in induction
5588 variable optimizations. Only the most relevant candidates
5589 are considered if there are more candidates, to avoid
5590 quadratic time complexity.
5592 `iv-max-considered-uses'
5593 The induction variable optimizations give up on loops that
5594 contain more induction variable uses.
5596 `iv-always-prune-cand-set-bound'
5597 If number of candidates in the set is smaller than this value,
5598 we always try to remove unnecessary ivs from the set during
5599 its optimization when a new iv is added to the set.
5601 `scev-max-expr-size'
5602 Bound on size of expressions used in the scalar evolutions
5603 analyzer. Large expressions slow the analyzer.
5605 `vect-max-version-checks'
5606 The maximum number of runtime checks that can be performed
5607 when doing loop versioning in the vectorizer. See option
5608 ftree-vect-loop-version for more information.
5610 `max-iterations-to-track'
5611 The maximum number of iterations of a loop the brute force
5612 algorithm for analysis of # of iterations of the loop tries
5615 `hot-bb-count-fraction'
5616 Select fraction of the maximal count of repetitions of basic
5617 block in program given basic block needs to have to be
5620 `hot-bb-frequency-fraction'
5621 Select fraction of the maximal frequency of executions of
5622 basic block in function given basic block needs to have to be
5625 `max-predicted-iterations'
5626 The maximum number of loop iterations we predict statically.
5627 This is useful in cases where function contain single loop
5628 with known bound and other loop with unknown. We predict the
5629 known number of iterations correctly, while the unknown
5630 number of iterations average to roughly 10. This means that
5631 the loop without bounds would appear artificially cold
5632 relative to the other one.
5634 `tracer-dynamic-coverage'
5635 `tracer-dynamic-coverage-feedback'
5636 This value is used to limit superblock formation once the
5637 given percentage of executed instructions is covered. This
5638 limits unnecessary code size expansion.
5640 The `tracer-dynamic-coverage-feedback' is used only when
5641 profile feedback is available. The real profiles (as opposed
5642 to statically estimated ones) are much less balanced allowing
5643 the threshold to be larger value.
5645 `tracer-max-code-growth'
5646 Stop tail duplication once code growth has reached given
5647 percentage. This is rather hokey argument, as most of the
5648 duplicates will be eliminated later in cross jumping, so it
5649 may be set to much higher values than is the desired code
5652 `tracer-min-branch-ratio'
5653 Stop reverse growth when the reverse probability of best edge
5654 is less than this threshold (in percent).
5656 `tracer-min-branch-ratio'
5657 `tracer-min-branch-ratio-feedback'
5658 Stop forward growth if the best edge do have probability
5659 lower than this threshold.
5661 Similarly to `tracer-dynamic-coverage' two values are
5662 present, one for compilation for profile feedback and one for
5663 compilation without. The value for compilation with profile
5664 feedback needs to be more conservative (higher) in order to
5665 make tracer effective.
5667 `max-cse-path-length'
5668 Maximum number of basic blocks on path that cse considers.
5672 The maximum instructions CSE process before flushing. The
5675 `global-var-threshold'
5676 Counts the number of function calls (N) and the number of
5677 call-clobbered variables (V). If NxV is larger than this
5678 limit, a single artificial variable will be created to
5679 represent all the call-clobbered variables at function call
5680 sites. This artificial variable will then be made to alias
5681 every call-clobbered variable. (done as `int * size_t' on
5682 the host machine; beware overflow).
5685 Maximum number of virtual operands allowed to represent
5686 aliases before triggering the alias grouping heuristic.
5687 Alias grouping reduces compile times and memory consumption
5688 needed for aliasing at the expense of precision loss in alias
5692 GCC uses a garbage collector to manage its own memory
5693 allocation. This parameter specifies the minimum percentage
5694 by which the garbage collector's heap should be allowed to
5695 expand between collections. Tuning this may improve
5696 compilation speed; it has no effect on code generation.
5698 The default is 30% + 70% * (RAM/1GB) with an upper bound of
5699 100% when RAM >= 1GB. If `getrlimit' is available, the
5700 notion of "RAM" is the smallest of actual RAM and
5701 `RLIMIT_DATA' or `RLIMIT_AS'. If GCC is not able to
5702 calculate RAM on a particular platform, the lower bound of
5703 30% is used. Setting this parameter and `ggc-min-heapsize'
5704 to zero causes a full collection to occur at every
5705 opportunity. This is extremely slow, but can be useful for
5709 Minimum size of the garbage collector's heap before it begins
5710 bothering to collect garbage. The first collection occurs
5711 after the heap expands by `ggc-min-expand'% beyond
5712 `ggc-min-heapsize'. Again, tuning this may improve
5713 compilation speed, and has no effect on code generation.
5715 The default is the smaller of RAM/8, RLIMIT_RSS, or a limit
5716 which tries to ensure that RLIMIT_DATA or RLIMIT_AS are not
5717 exceeded, but with a lower bound of 4096 (four megabytes) and
5718 an upper bound of 131072 (128 megabytes). If GCC is not able
5719 to calculate RAM on a particular platform, the lower bound is
5720 used. Setting this parameter very large effectively disables
5721 garbage collection. Setting this parameter and
5722 `ggc-min-expand' to zero causes a full collection to occur at
5725 `max-reload-search-insns'
5726 The maximum number of instruction reload should look backward
5727 for equivalent register. Increasing values mean more
5728 aggressive optimization, making the compile time increase
5729 with probably slightly better performance. The default value
5732 `max-cselib-memory-location'
5733 The maximum number of memory locations cselib should take
5734 into account. Increasing values mean more aggressive
5735 optimization, making the compile time increase with probably
5736 slightly better performance. The default value is 500.
5738 `max-flow-memory-location'
5739 Similar as `max-cselib-memory-location' but for dataflow
5740 liveness. The default value is 100.
5742 `reorder-blocks-duplicate'
5743 `reorder-blocks-duplicate-feedback'
5744 Used by basic block reordering pass to decide whether to use
5745 unconditional branch or duplicate the code on its
5746 destination. Code is duplicated when its estimated size is
5747 smaller than this value multiplied by the estimated size of
5748 unconditional jump in the hot spots of the program.
5750 The `reorder-block-duplicate-feedback' is used only when
5751 profile feedback is available and may be set to higher values
5752 than `reorder-block-duplicate' since information about the
5753 hot spots is more accurate.
5755 `max-sched-region-blocks'
5756 The maximum number of blocks in a region to be considered for
5757 interblock scheduling. The default value is 10.
5759 `max-sched-region-insns'
5760 The maximum number of insns in a region to be considered for
5761 interblock scheduling. The default value is 100.
5764 The minimum probability of reaching a source block for
5765 interblock speculative scheduling. The default value is 40.
5767 `max-last-value-rtl'
5768 The maximum size measured as number of RTLs that can be
5769 recorded in an expression in combiner for a pseudo register
5770 as last known value of that register. The default is 10000.
5772 `integer-share-limit'
5773 Small integer constants can use a shared data structure,
5774 reducing the compiler's memory usage and increasing its
5775 speed. This sets the maximum value of a shared integer
5776 constant's. The default value is 256.
5778 `min-virtual-mappings'
5779 Specifies the minimum number of virtual mappings in the
5780 incremental SSA updater that should be registered to trigger
5781 the virtual mappings heuristic defined by
5782 virtual-mappings-ratio. The default value is 100.
5784 `virtual-mappings-ratio'
5785 If the number of virtual mappings is virtual-mappings-ratio
5786 bigger than the number of virtual symbols to be updated, then
5787 the incremental SSA updater switches to a full update for
5788 those symbols. The default ratio is 3.
5791 The minimum size of buffers (i.e. arrays) that will receive
5792 stack smashing protection when `-fstack-protection' is used.
5794 `max-jump-thread-duplication-stmts'
5795 Maximum number of statements allowed in a block that needs to
5796 be duplicated when threading jumps.
5798 `max-fields-for-field-sensitive'
5799 Maximum number of fields in a structure we will treat in a
5800 field sensitive manner during pointer analysis.
5804 File: gcc.info, Node: Preprocessor Options, Next: Assembler Options, Prev: Optimize Options, Up: Invoking GCC
5806 3.11 Options Controlling the Preprocessor
5807 =========================================
5809 These options control the C preprocessor, which is run on each C source
5810 file before actual compilation.
5812 If you use the `-E' option, nothing is done except preprocessing.
5813 Some of these options make sense only together with `-E' because they
5814 cause the preprocessor output to be unsuitable for actual compilation.
5816 You can use `-Wp,OPTION' to bypass the compiler driver and pass
5817 OPTION directly through to the preprocessor. If OPTION contains
5818 commas, it is split into multiple options at the commas. However,
5819 many options are modified, translated or interpreted by the
5820 compiler driver before being passed to the preprocessor, and `-Wp'
5821 forcibly bypasses this phase. The preprocessor's direct interface
5822 is undocumented and subject to change, so whenever possible you
5823 should avoid using `-Wp' and let the driver handle the options
5826 `-Xpreprocessor OPTION'
5827 Pass OPTION as an option to the preprocessor. You can use this to
5828 supply system-specific preprocessor options which GCC does not
5829 know how to recognize.
5831 If you want to pass an option that takes an argument, you must use
5832 `-Xpreprocessor' twice, once for the option and once for the
5836 Predefine NAME as a macro, with definition `1'.
5838 `-D NAME=DEFINITION'
5839 The contents of DEFINITION are tokenized and processed as if they
5840 appeared during translation phase three in a `#define' directive.
5841 In particular, the definition will be truncated by embedded
5844 If you are invoking the preprocessor from a shell or shell-like
5845 program you may need to use the shell's quoting syntax to protect
5846 characters such as spaces that have a meaning in the shell syntax.
5848 If you wish to define a function-like macro on the command line,
5849 write its argument list with surrounding parentheses before the
5850 equals sign (if any). Parentheses are meaningful to most shells,
5851 so you will need to quote the option. With `sh' and `csh',
5852 `-D'NAME(ARGS...)=DEFINITION'' works.
5854 `-D' and `-U' options are processed in the order they are given on
5855 the command line. All `-imacros FILE' and `-include FILE' options
5856 are processed after all `-D' and `-U' options.
5859 Cancel any previous definition of NAME, either built in or
5860 provided with a `-D' option.
5863 Do not predefine any system-specific or GCC-specific macros. The
5864 standard predefined macros remain defined.
5867 Add the directory DIR to the list of directories to be searched
5868 for header files. Directories named by `-I' are searched before
5869 the standard system include directories. If the directory DIR is
5870 a standard system include directory, the option is ignored to
5871 ensure that the default search order for system directories and
5872 the special treatment of system headers are not defeated .
5875 Write output to FILE. This is the same as specifying FILE as the
5876 second non-option argument to `cpp'. `gcc' has a different
5877 interpretation of a second non-option argument, so you must use
5878 `-o' to specify the output file.
5881 Turns on all optional warnings which are desirable for normal code.
5882 At present this is `-Wcomment', `-Wtrigraphs', `-Wmultichar' and a
5883 warning about integer promotion causing a change of sign in `#if'
5884 expressions. Note that many of the preprocessor's warnings are on
5885 by default and have no options to control them.
5889 Warn whenever a comment-start sequence `/*' appears in a `/*'
5890 comment, or whenever a backslash-newline appears in a `//' comment.
5891 (Both forms have the same effect.)
5894 Most trigraphs in comments cannot affect the meaning of the
5895 program. However, a trigraph that would form an escaped newline
5896 (`??/' at the end of a line) can, by changing where the comment
5897 begins or ends. Therefore, only trigraphs that would form escaped
5898 newlines produce warnings inside a comment.
5900 This option is implied by `-Wall'. If `-Wall' is not given, this
5901 option is still enabled unless trigraphs are enabled. To get
5902 trigraph conversion without warnings, but get the other `-Wall'
5903 warnings, use `-trigraphs -Wall -Wno-trigraphs'.
5906 Warn about certain constructs that behave differently in
5907 traditional and ISO C. Also warn about ISO C constructs that have
5908 no traditional C equivalent, and problematic constructs which
5912 Warn the first time `#import' is used.
5915 Warn whenever an identifier which is not a macro is encountered in
5916 an `#if' directive, outside of `defined'. Such identifiers are
5920 Warn about macros defined in the main file that are unused. A
5921 macro is "used" if it is expanded or tested for existence at least
5922 once. The preprocessor will also warn if the macro has not been
5923 used at the time it is redefined or undefined.
5925 Built-in macros, macros defined on the command line, and macros
5926 defined in include files are not warned about.
5928 _Note:_ If a macro is actually used, but only used in skipped
5929 conditional blocks, then CPP will report it as unused. To avoid
5930 the warning in such a case, you might improve the scope of the
5931 macro's definition by, for example, moving it into the first
5932 skipped block. Alternatively, you could provide a dummy use with
5935 #if defined the_macro_causing_the_warning
5939 Warn whenever an `#else' or an `#endif' are followed by text.
5940 This usually happens in code of the form
5948 The second and third `FOO' should be in comments, but often are not
5949 in older programs. This warning is on by default.
5952 Make all warnings into hard errors. Source code which triggers
5953 warnings will be rejected.
5956 Issue warnings for code in system headers. These are normally
5957 unhelpful in finding bugs in your own code, therefore suppressed.
5958 If you are responsible for the system library, you may want to see
5962 Suppress all warnings, including those which GNU CPP issues by
5966 Issue all the mandatory diagnostics listed in the C standard.
5967 Some of them are left out by default, since they trigger
5968 frequently on harmless code.
5971 Issue all the mandatory diagnostics, and make all mandatory
5972 diagnostics into errors. This includes mandatory diagnostics that
5973 GCC issues without `-pedantic' but treats as warnings.
5976 Instead of outputting the result of preprocessing, output a rule
5977 suitable for `make' describing the dependencies of the main source
5978 file. The preprocessor outputs one `make' rule containing the
5979 object file name for that source file, a colon, and the names of
5980 all the included files, including those coming from `-include' or
5981 `-imacros' command line options.
5983 Unless specified explicitly (with `-MT' or `-MQ'), the object file
5984 name consists of the basename of the source file with any suffix
5985 replaced with object file suffix. If there are many included
5986 files then the rule is split into several lines using `\'-newline.
5987 The rule has no commands.
5989 This option does not suppress the preprocessor's debug output,
5990 such as `-dM'. To avoid mixing such debug output with the
5991 dependency rules you should explicitly specify the dependency
5992 output file with `-MF', or use an environment variable like
5993 `DEPENDENCIES_OUTPUT' (*note Environment Variables::). Debug
5994 output will still be sent to the regular output stream as normal.
5996 Passing `-M' to the driver implies `-E', and suppresses warnings
5997 with an implicit `-w'.
6000 Like `-M' but do not mention header files that are found in system
6001 header directories, nor header files that are included, directly
6002 or indirectly, from such a header.
6004 This implies that the choice of angle brackets or double quotes in
6005 an `#include' directive does not in itself determine whether that
6006 header will appear in `-MM' dependency output. This is a slight
6007 change in semantics from GCC versions 3.0 and earlier.
6010 When used with `-M' or `-MM', specifies a file to write the
6011 dependencies to. If no `-MF' switch is given the preprocessor
6012 sends the rules to the same place it would have sent preprocessed
6015 When used with the driver options `-MD' or `-MMD', `-MF' overrides
6016 the default dependency output file.
6019 In conjunction with an option such as `-M' requesting dependency
6020 generation, `-MG' assumes missing header files are generated files
6021 and adds them to the dependency list without raising an error.
6022 The dependency filename is taken directly from the `#include'
6023 directive without prepending any path. `-MG' also suppresses
6024 preprocessed output, as a missing header file renders this useless.
6026 This feature is used in automatic updating of makefiles.
6029 This option instructs CPP to add a phony target for each dependency
6030 other than the main file, causing each to depend on nothing. These
6031 dummy rules work around errors `make' gives if you remove header
6032 files without updating the `Makefile' to match.
6034 This is typical output:
6036 test.o: test.c test.h
6041 Change the target of the rule emitted by dependency generation. By
6042 default CPP takes the name of the main input file, including any
6043 path, deletes any file suffix such as `.c', and appends the
6044 platform's usual object suffix. The result is the target.
6046 An `-MT' option will set the target to be exactly the string you
6047 specify. If you want multiple targets, you can specify them as a
6048 single argument to `-MT', or use multiple `-MT' options.
6050 For example, `-MT '$(objpfx)foo.o'' might give
6052 $(objpfx)foo.o: foo.c
6055 Same as `-MT', but it quotes any characters which are special to
6056 Make. `-MQ '$(objpfx)foo.o'' gives
6058 $$(objpfx)foo.o: foo.c
6060 The default target is automatically quoted, as if it were given
6064 `-MD' is equivalent to `-M -MF FILE', except that `-E' is not
6065 implied. The driver determines FILE based on whether an `-o'
6066 option is given. If it is, the driver uses its argument but with
6067 a suffix of `.d', otherwise it take the basename of the input file
6068 and applies a `.d' suffix.
6070 If `-MD' is used in conjunction with `-E', any `-o' switch is
6071 understood to specify the dependency output file (but *note -MF:
6072 dashMF.), but if used without `-E', each `-o' is understood to
6073 specify a target object file.
6075 Since `-E' is not implied, `-MD' can be used to generate a
6076 dependency output file as a side-effect of the compilation process.
6079 Like `-MD' except mention only user header files, not system
6083 When using precompiled headers (*note Precompiled Headers::), this
6084 flag will cause the dependency-output flags to also list the files
6085 from the precompiled header's dependencies. If not specified only
6086 the precompiled header would be listed and not the files that were
6087 used to create it because those files are not consulted when a
6088 precompiled header is used.
6091 This option allows use of a precompiled header (*note Precompiled
6092 Headers::) together with `-E'. It inserts a special `#pragma',
6093 `#pragma GCC pch_preprocess "<filename>"' in the output to mark
6094 the place where the precompiled header was found, and its
6095 filename. When `-fpreprocessed' is in use, GCC recognizes this
6096 `#pragma' and loads the PCH.
6098 This option is off by default, because the resulting preprocessed
6099 output is only really suitable as input to GCC. It is switched on
6102 You should not write this `#pragma' in your own code, but it is
6103 safe to edit the filename if the PCH file is available in a
6104 different location. The filename may be absolute or it may be
6105 relative to GCC's current directory.
6110 `-x assembler-with-cpp'
6111 Specify the source language: C, C++, Objective-C, or assembly.
6112 This has nothing to do with standards conformance or extensions;
6113 it merely selects which base syntax to expect. If you give none
6114 of these options, cpp will deduce the language from the extension
6115 of the source file: `.c', `.cc', `.m', or `.S'. Some other common
6116 extensions for C++ and assembly are also recognized. If cpp does
6117 not recognize the extension, it will treat the file as C; this is
6118 the most generic mode.
6120 _Note:_ Previous versions of cpp accepted a `-lang' option which
6121 selected both the language and the standards conformance level.
6122 This option has been removed, because it conflicts with the `-l'
6127 Specify the standard to which the code should conform. Currently
6128 CPP knows about C and C++ standards; others may be added in the
6131 STANDARD may be one of:
6134 The ISO C standard from 1990. `c89' is the customary
6135 shorthand for this version of the standard.
6137 The `-ansi' option is equivalent to `-std=c89'.
6140 The 1990 C standard, as amended in 1994.
6146 The revised ISO C standard, published in December 1999.
6147 Before publication, this was known as C9X.
6150 The 1990 C standard plus GNU extensions. This is the default.
6154 The 1999 C standard plus GNU extensions.
6157 The 1998 ISO C++ standard plus amendments.
6160 The same as `-std=c++98' plus GNU extensions. This is the
6161 default for C++ code.
6164 Split the include path. Any directories specified with `-I'
6165 options before `-I-' are searched only for headers requested with
6166 `#include "FILE"'; they are not searched for `#include <FILE>'.
6167 If additional directories are specified with `-I' options after
6168 the `-I-', those directories are searched for all `#include'
6171 In addition, `-I-' inhibits the use of the directory of the current
6172 file directory as the first search directory for `#include "FILE"'.
6173 This option has been deprecated.
6176 Do not search the standard system directories for header files.
6177 Only the directories you have specified with `-I' options (and the
6178 directory of the current file, if appropriate) are searched.
6181 Do not search for header files in the C++-specific standard
6182 directories, but do still search the other standard directories.
6183 (This option is used when building the C++ library.)
6186 Process FILE as if `#include "file"' appeared as the first line of
6187 the primary source file. However, the first directory searched
6188 for FILE is the preprocessor's working directory _instead of_ the
6189 directory containing the main source file. If not found there, it
6190 is searched for in the remainder of the `#include "..."' search
6193 If multiple `-include' options are given, the files are included
6194 in the order they appear on the command line.
6197 Exactly like `-include', except that any output produced by
6198 scanning FILE is thrown away. Macros it defines remain defined.
6199 This allows you to acquire all the macros from a header without
6200 also processing its declarations.
6202 All files specified by `-imacros' are processed before all files
6203 specified by `-include'.
6206 Search DIR for header files, but do it _after_ all directories
6207 specified with `-I' and the standard system directories have been
6208 exhausted. DIR is treated as a system include directory.
6211 Specify PREFIX as the prefix for subsequent `-iwithprefix'
6212 options. If the prefix represents a directory, you should include
6216 `-iwithprefixbefore DIR'
6217 Append DIR to the prefix specified previously with `-iprefix', and
6218 add the resulting directory to the include search path.
6219 `-iwithprefixbefore' puts it in the same place `-I' would;
6220 `-iwithprefix' puts it where `-idirafter' would.
6223 This option is like the `--sysroot' option, but applies only to
6224 header files. See the `--sysroot' option for more information.
6227 Search DIR for header files, after all directories specified by
6228 `-I' but before the standard system directories. Mark it as a
6229 system directory, so that it gets the same special treatment as is
6230 applied to the standard system directories.
6233 Search DIR only for header files requested with `#include "FILE"';
6234 they are not searched for `#include <FILE>', before all
6235 directories specified by `-I' and before the standard system
6238 `-fdollars-in-identifiers'
6239 Accept `$' in identifiers.
6241 `-fextended-identifiers'
6242 Accept universal character names in identifiers. This option is
6243 experimental; in a future version of GCC, it will be enabled by
6244 default for C99 and C++.
6247 Indicate to the preprocessor that the input file has already been
6248 preprocessed. This suppresses things like macro expansion,
6249 trigraph conversion, escaped newline splicing, and processing of
6250 most directives. The preprocessor still recognizes and removes
6251 comments, so that you can pass a file preprocessed with `-C' to
6252 the compiler without problems. In this mode the integrated
6253 preprocessor is little more than a tokenizer for the front ends.
6255 `-fpreprocessed' is implicit if the input file has one of the
6256 extensions `.i', `.ii' or `.mi'. These are the extensions that
6257 GCC uses for preprocessed files created by `-save-temps'.
6260 Set the distance between tab stops. This helps the preprocessor
6261 report correct column numbers in warnings or errors, even if tabs
6262 appear on the line. If the value is less than 1 or greater than
6263 100, the option is ignored. The default is 8.
6265 `-fexec-charset=CHARSET'
6266 Set the execution character set, used for string and character
6267 constants. The default is UTF-8. CHARSET can be any encoding
6268 supported by the system's `iconv' library routine.
6270 `-fwide-exec-charset=CHARSET'
6271 Set the wide execution character set, used for wide string and
6272 character constants. The default is UTF-32 or UTF-16, whichever
6273 corresponds to the width of `wchar_t'. As with `-fexec-charset',
6274 CHARSET can be any encoding supported by the system's `iconv'
6275 library routine; however, you will have problems with encodings
6276 that do not fit exactly in `wchar_t'.
6278 `-finput-charset=CHARSET'
6279 Set the input character set, used for translation from the
6280 character set of the input file to the source character set used
6281 by GCC. If the locale does not specify, or GCC cannot get this
6282 information from the locale, the default is UTF-8. This can be
6283 overridden by either the locale or this command line option.
6284 Currently the command line option takes precedence if there's a
6285 conflict. CHARSET can be any encoding supported by the system's
6286 `iconv' library routine.
6288 `-fworking-directory'
6289 Enable generation of linemarkers in the preprocessor output that
6290 will let the compiler know the current working directory at the
6291 time of preprocessing. When this option is enabled, the
6292 preprocessor will emit, after the initial linemarker, a second
6293 linemarker with the current working directory followed by two
6294 slashes. GCC will use this directory, when it's present in the
6295 preprocessed input, as the directory emitted as the current
6296 working directory in some debugging information formats. This
6297 option is implicitly enabled if debugging information is enabled,
6298 but this can be inhibited with the negated form
6299 `-fno-working-directory'. If the `-P' flag is present in the
6300 command line, this option has no effect, since no `#line'
6301 directives are emitted whatsoever.
6304 Do not print column numbers in diagnostics. This may be necessary
6305 if diagnostics are being scanned by a program that does not
6306 understand the column numbers, such as `dejagnu'.
6308 `-A PREDICATE=ANSWER'
6309 Make an assertion with the predicate PREDICATE and answer ANSWER.
6310 This form is preferred to the older form `-A PREDICATE(ANSWER)',
6311 which is still supported, because it does not use shell special
6314 `-A -PREDICATE=ANSWER'
6315 Cancel an assertion with the predicate PREDICATE and answer ANSWER.
6318 CHARS is a sequence of one or more of the following characters,
6319 and must not be preceded by a space. Other characters are
6320 interpreted by the compiler proper, or reserved for future
6321 versions of GCC, and so are silently ignored. If you specify
6322 characters whose behavior conflicts, the result is undefined.
6325 Instead of the normal output, generate a list of `#define'
6326 directives for all the macros defined during the execution of
6327 the preprocessor, including predefined macros. This gives
6328 you a way of finding out what is predefined in your version
6329 of the preprocessor. Assuming you have no file `foo.h', the
6332 touch foo.h; cpp -dM foo.h
6334 will show all the predefined macros.
6337 Like `M' except in two respects: it does _not_ include the
6338 predefined macros, and it outputs _both_ the `#define'
6339 directives and the result of preprocessing. Both kinds of
6340 output go to the standard output file.
6343 Like `D', but emit only the macro names, not their expansions.
6346 Output `#include' directives in addition to the result of
6350 Inhibit generation of linemarkers in the output from the
6351 preprocessor. This might be useful when running the preprocessor
6352 on something that is not C code, and will be sent to a program
6353 which might be confused by the linemarkers.
6356 Do not discard comments. All comments are passed through to the
6357 output file, except for comments in processed directives, which
6358 are deleted along with the directive.
6360 You should be prepared for side effects when using `-C'; it causes
6361 the preprocessor to treat comments as tokens in their own right.
6362 For example, comments appearing at the start of what would be a
6363 directive line have the effect of turning that line into an
6364 ordinary source line, since the first token on the line is no
6368 Do not discard comments, including during macro expansion. This is
6369 like `-C', except that comments contained within macros are also
6370 passed through to the output file where the macro is expanded.
6372 In addition to the side-effects of the `-C' option, the `-CC'
6373 option causes all C++-style comments inside a macro to be
6374 converted to C-style comments. This is to prevent later use of
6375 that macro from inadvertently commenting out the remainder of the
6378 The `-CC' option is generally used to support lint comments.
6381 Try to imitate the behavior of old-fashioned C preprocessors, as
6382 opposed to ISO C preprocessors.
6385 Process trigraph sequences. These are three-character sequences,
6386 all starting with `??', that are defined by ISO C to stand for
6387 single characters. For example, `??/' stands for `\', so `'??/n''
6388 is a character constant for a newline. By default, GCC ignores
6389 trigraphs, but in standard-conforming modes it converts them. See
6390 the `-std' and `-ansi' options.
6392 The nine trigraphs and their replacements are
6394 Trigraph: ??( ??) ??< ??> ??= ??/ ??' ??! ??-
6395 Replacement: [ ] { } # \ ^ | ~
6398 Enable special code to work around file systems which only permit
6399 very short file names, such as MS-DOS.
6403 Print text describing all the command line options instead of
6404 preprocessing anything.
6407 Verbose mode. Print out GNU CPP's version number at the beginning
6408 of execution, and report the final form of the include path.
6411 Print the name of each header file used, in addition to other
6412 normal activities. Each name is indented to show how deep in the
6413 `#include' stack it is. Precompiled header files are also
6414 printed, even if they are found to be invalid; an invalid
6415 precompiled header file is printed with `...x' and a valid one
6420 Print out GNU CPP's version number. With one dash, proceed to
6421 preprocess as normal. With two dashes, exit immediately.
6424 File: gcc.info, Node: Assembler Options, Next: Link Options, Prev: Preprocessor Options, Up: Invoking GCC
6426 3.12 Passing Options to the Assembler
6427 =====================================
6429 You can pass options to the assembler.
6432 Pass OPTION as an option to the assembler. If OPTION contains
6433 commas, it is split into multiple options at the commas.
6435 `-Xassembler OPTION'
6436 Pass OPTION as an option to the assembler. You can use this to
6437 supply system-specific assembler options which GCC does not know
6440 If you want to pass an option that takes an argument, you must use
6441 `-Xassembler' twice, once for the option and once for the argument.
6445 File: gcc.info, Node: Link Options, Next: Directory Options, Prev: Assembler Options, Up: Invoking GCC
6447 3.13 Options for Linking
6448 ========================
6450 These options come into play when the compiler links object files into
6451 an executable output file. They are meaningless if the compiler is not
6455 A file name that does not end in a special recognized suffix is
6456 considered to name an object file or library. (Object files are
6457 distinguished from libraries by the linker according to the file
6458 contents.) If linking is done, these object files are used as
6459 input to the linker.
6464 If any of these options is used, then the linker is not run, and
6465 object file names should not be used as arguments. *Note Overall
6470 Search the library named LIBRARY when linking. (The second
6471 alternative with the library as a separate argument is only for
6472 POSIX compliance and is not recommended.)
6474 It makes a difference where in the command you write this option;
6475 the linker searches and processes libraries and object files in
6476 the order they are specified. Thus, `foo.o -lz bar.o' searches
6477 library `z' after file `foo.o' but before `bar.o'. If `bar.o'
6478 refers to functions in `z', those functions may not be loaded.
6480 The linker searches a standard list of directories for the library,
6481 which is actually a file named `libLIBRARY.a'. The linker then
6482 uses this file as if it had been specified precisely by name.
6484 The directories searched include several standard system
6485 directories plus any that you specify with `-L'.
6487 Normally the files found this way are library files--archive files
6488 whose members are object files. The linker handles an archive
6489 file by scanning through it for members which define symbols that
6490 have so far been referenced but not defined. But if the file that
6491 is found is an ordinary object file, it is linked in the usual
6492 fashion. The only difference between using an `-l' option and
6493 specifying a file name is that `-l' surrounds LIBRARY with `lib'
6494 and `.a' and searches several directories.
6497 You need this special case of the `-l' option in order to link an
6498 Objective-C or Objective-C++ program.
6501 Do not use the standard system startup files when linking. The
6502 standard system libraries are used normally, unless `-nostdlib' or
6503 `-nodefaultlibs' is used.
6506 Do not use the standard system libraries when linking. Only the
6507 libraries you specify will be passed to the linker. The standard
6508 startup files are used normally, unless `-nostartfiles' is used.
6509 The compiler may generate calls to `memcmp', `memset', `memcpy'
6510 and `memmove'. These entries are usually resolved by entries in
6511 libc. These entry points should be supplied through some other
6512 mechanism when this option is specified.
6515 Do not use the standard system startup files or libraries when
6516 linking. No startup files and only the libraries you specify will
6517 be passed to the linker. The compiler may generate calls to
6518 `memcmp', `memset', `memcpy' and `memmove'. These entries are
6519 usually resolved by entries in libc. These entry points should be
6520 supplied through some other mechanism when this option is
6523 One of the standard libraries bypassed by `-nostdlib' and
6524 `-nodefaultlibs' is `libgcc.a', a library of internal subroutines
6525 that GCC uses to overcome shortcomings of particular machines, or
6526 special needs for some languages. (*Note Interfacing to GCC
6527 Output: (gccint)Interface, for more discussion of `libgcc.a'.) In
6528 most cases, you need `libgcc.a' even when you want to avoid other
6529 standard libraries. In other words, when you specify `-nostdlib'
6530 or `-nodefaultlibs' you should usually specify `-lgcc' as well.
6531 This ensures that you have no unresolved references to internal GCC
6532 library subroutines. (For example, `__main', used to ensure C++
6533 constructors will be called; *note `collect2': (gccint)Collect2.)
6536 Produce a position independent executable on targets which support
6537 it. For predictable results, you must also specify the same set
6538 of options that were used to generate code (`-fpie', `-fPIE', or
6539 model suboptions) when you specify this option.
6542 Pass the flag `-export-dynamic' to the ELF linker, on targets that
6543 support it. This instructs the linker to add all symbols, not only
6544 used ones, to the dynamic symbol table. This option is needed for
6545 some uses of `dlopen' or to allow obtaining backtraces from within
6549 Remove all symbol table and relocation information from the
6553 On systems that support dynamic linking, this prevents linking
6554 with the shared libraries. On other systems, this option has no
6558 Produce a shared object which can then be linked with other
6559 objects to form an executable. Not all systems support this
6560 option. For predictable results, you must also specify the same
6561 set of options that were used to generate code (`-fpic', `-fPIC',
6562 or model suboptions) when you specify this option.(1)
6566 On systems that provide `libgcc' as a shared library, these options
6567 force the use of either the shared or static version respectively.
6568 If no shared version of `libgcc' was built when the compiler was
6569 configured, these options have no effect.
6571 There are several situations in which an application should use the
6572 shared `libgcc' instead of the static version. The most common of
6573 these is when the application wishes to throw and catch exceptions
6574 across different shared libraries. In that case, each of the
6575 libraries as well as the application itself should use the shared
6578 Therefore, the G++ and GCJ drivers automatically add
6579 `-shared-libgcc' whenever you build a shared library or a main
6580 executable, because C++ and Java programs typically use
6581 exceptions, so this is the right thing to do.
6583 If, instead, you use the GCC driver to create shared libraries,
6584 you may find that they will not always be linked with the shared
6585 `libgcc'. If GCC finds, at its configuration time, that you have
6586 a non-GNU linker or a GNU linker that does not support option
6587 `--eh-frame-hdr', it will link the shared version of `libgcc' into
6588 shared libraries by default. Otherwise, it will take advantage of
6589 the linker and optimize away the linking with the shared version
6590 of `libgcc', linking with the static version of libgcc by default.
6591 This allows exceptions to propagate through such shared
6592 libraries, without incurring relocation costs at library load time.
6594 However, if a library or main executable is supposed to throw or
6595 catch exceptions, you must link it using the G++ or GCJ driver, as
6596 appropriate for the languages used in the program, or using the
6597 option `-shared-libgcc', such that it is linked with the shared
6601 Bind references to global symbols when building a shared object.
6602 Warn about any unresolved references (unless overridden by the
6603 link editor option `-Xlinker -z -Xlinker defs'). Only a few
6604 systems support this option.
6607 Pass OPTION as an option to the linker. You can use this to
6608 supply system-specific linker options which GCC does not know how
6611 If you want to pass an option that takes an argument, you must use
6612 `-Xlinker' twice, once for the option and once for the argument.
6613 For example, to pass `-assert definitions', you must write
6614 `-Xlinker -assert -Xlinker definitions'. It does not work to write
6615 `-Xlinker "-assert definitions"', because this passes the entire
6616 string as a single argument, which is not what the linker expects.
6619 Pass OPTION as an option to the linker. If OPTION contains
6620 commas, it is split into multiple options at the commas.
6623 Pretend the symbol SYMBOL is undefined, to force linking of
6624 library modules to define it. You can use `-u' multiple times with
6625 different symbols to force loading of additional library modules.
6627 ---------- Footnotes ----------
6629 (1) On some systems, `gcc -shared' needs to build supplementary stub
6630 code for constructors to work. On multi-libbed systems, `gcc -shared'
6631 must select the correct support libraries to link against. Failing to
6632 supply the correct flags may lead to subtle defects. Supplying them in
6633 cases where they are not necessary is innocuous.
6636 File: gcc.info, Node: Directory Options, Next: Spec Files, Prev: Link Options, Up: Invoking GCC
6638 3.14 Options for Directory Search
6639 =================================
6641 These options specify directories to search for header files, for
6642 libraries and for parts of the compiler:
6645 Add the directory DIR to the head of the list of directories to be
6646 searched for header files. This can be used to override a system
6647 header file, substituting your own version, since these
6648 directories are searched before the system header file
6649 directories. However, you should not use this option to add
6650 directories that contain vendor-supplied system header files (use
6651 `-isystem' for that). If you use more than one `-I' option, the
6652 directories are scanned in left-to-right order; the standard
6653 system directories come after.
6655 If a standard system include directory, or a directory specified
6656 with `-isystem', is also specified with `-I', the `-I' option will
6657 be ignored. The directory will still be searched but as a system
6658 directory at its normal position in the system include chain.
6659 This is to ensure that GCC's procedure to fix buggy system headers
6660 and the ordering for the include_next directive are not
6661 inadvertently changed. If you really need to change the search
6662 order for system directories, use the `-nostdinc' and/or
6666 Add the directory DIR to the head of the list of directories to be
6667 searched for header files only for the case of `#include "FILE"';
6668 they are not searched for `#include <FILE>', otherwise just like
6672 Add directory DIR to the list of directories to be searched for
6676 This option specifies where to find the executables, libraries,
6677 include files, and data files of the compiler itself.
6679 The compiler driver program runs one or more of the subprograms
6680 `cpp', `cc1', `as' and `ld'. It tries PREFIX as a prefix for each
6681 program it tries to run, both with and without `MACHINE/VERSION/'
6682 (*note Target Options::).
6684 For each subprogram to be run, the compiler driver first tries the
6685 `-B' prefix, if any. If that name is not found, or if `-B' was
6686 not specified, the driver tries two standard prefixes, which are
6687 `/usr/lib/gcc/' and `/usr/local/lib/gcc/'. If neither of those
6688 results in a file name that is found, the unmodified program name
6689 is searched for using the directories specified in your `PATH'
6690 environment variable.
6692 The compiler will check to see if the path provided by the `-B'
6693 refers to a directory, and if necessary it will add a directory
6694 separator character at the end of the path.
6696 `-B' prefixes that effectively specify directory names also apply
6697 to libraries in the linker, because the compiler translates these
6698 options into `-L' options for the linker. They also apply to
6699 includes files in the preprocessor, because the compiler
6700 translates these options into `-isystem' options for the
6701 preprocessor. In this case, the compiler appends `include' to the
6704 The run-time support file `libgcc.a' can also be searched for using
6705 the `-B' prefix, if needed. If it is not found there, the two
6706 standard prefixes above are tried, and that is all. The file is
6707 left out of the link if it is not found by those means.
6709 Another way to specify a prefix much like the `-B' prefix is to use
6710 the environment variable `GCC_EXEC_PREFIX'. *Note Environment
6713 As a special kludge, if the path provided by `-B' is
6714 `[dir/]stageN/', where N is a number in the range 0 to 9, then it
6715 will be replaced by `[dir/]include'. This is to help with
6716 boot-strapping the compiler.
6719 Process FILE after the compiler reads in the standard `specs'
6720 file, in order to override the defaults that the `gcc' driver
6721 program uses when determining what switches to pass to `cc1',
6722 `cc1plus', `as', `ld', etc. More than one `-specs=FILE' can be
6723 specified on the command line, and they are processed in order,
6727 Use DIR as the logical root directory for headers and libraries.
6728 For example, if the compiler would normally search for headers in
6729 `/usr/include' and libraries in `/usr/lib', it will instead search
6730 `DIR/usr/include' and `DIR/usr/lib'.
6732 If you use both this option and the `-isysroot' option, then the
6733 `--sysroot' option will apply to libraries, but the `-isysroot'
6734 option will apply to header files.
6736 The GNU linker (beginning with version 2.16) has the necessary
6737 support for this option. If your linker does not support this
6738 option, the header file aspect of `--sysroot' will still work, but
6739 the library aspect will not.
6742 This option has been deprecated. Please use `-iquote' instead for
6743 `-I' directories before the `-I-' and remove the `-I-'. Any
6744 directories you specify with `-I' options before the `-I-' option
6745 are searched only for the case of `#include "FILE"'; they are not
6746 searched for `#include <FILE>'.
6748 If additional directories are specified with `-I' options after
6749 the `-I-', these directories are searched for all `#include'
6750 directives. (Ordinarily _all_ `-I' directories are used this way.)
6752 In addition, the `-I-' option inhibits the use of the current
6753 directory (where the current input file came from) as the first
6754 search directory for `#include "FILE"'. There is no way to
6755 override this effect of `-I-'. With `-I.' you can specify
6756 searching the directory which was current when the compiler was
6757 invoked. That is not exactly the same as what the preprocessor
6758 does by default, but it is often satisfactory.
6760 `-I-' does not inhibit the use of the standard system directories
6761 for header files. Thus, `-I-' and `-nostdinc' are independent.
6764 File: gcc.info, Node: Spec Files, Next: Target Options, Prev: Directory Options, Up: Invoking GCC
6766 3.15 Specifying subprocesses and the switches to pass to them
6767 =============================================================
6769 `gcc' is a driver program. It performs its job by invoking a sequence
6770 of other programs to do the work of compiling, assembling and linking.
6771 GCC interprets its command-line parameters and uses these to deduce
6772 which programs it should invoke, and which command-line options it
6773 ought to place on their command lines. This behavior is controlled by
6774 "spec strings". In most cases there is one spec string for each
6775 program that GCC can invoke, but a few programs have multiple spec
6776 strings to control their behavior. The spec strings built into GCC can
6777 be overridden by using the `-specs=' command-line switch to specify a
6780 "Spec files" are plaintext files that are used to construct spec
6781 strings. They consist of a sequence of directives separated by blank
6782 lines. The type of directive is determined by the first non-whitespace
6783 character on the line and it can be one of the following:
6786 Issues a COMMAND to the spec file processor. The commands that can
6790 Search for FILE and insert its text at the current point in
6793 `%include_noerr <FILE>'
6794 Just like `%include', but do not generate an error message if
6795 the include file cannot be found.
6797 `%rename OLD_NAME NEW_NAME'
6798 Rename the spec string OLD_NAME to NEW_NAME.
6802 This tells the compiler to create, override or delete the named
6803 spec string. All lines after this directive up to the next
6804 directive or blank line are considered to be the text for the spec
6805 string. If this results in an empty string then the spec will be
6806 deleted. (Or, if the spec did not exist, then nothing will
6807 happened.) Otherwise, if the spec does not currently exist a new
6808 spec will be created. If the spec does exist then its contents
6809 will be overridden by the text of this directive, unless the first
6810 character of that text is the `+' character, in which case the
6811 text will be appended to the spec.
6814 Creates a new `[SUFFIX] spec' pair. All lines after this directive
6815 and up to the next directive or blank line are considered to make
6816 up the spec string for the indicated suffix. When the compiler
6817 encounters an input file with the named suffix, it will processes
6818 the spec string in order to work out how to compile that file.
6824 This says that any input file whose name ends in `.ZZ' should be
6825 passed to the program `z-compile', which should be invoked with the
6826 command-line switch `-input' and with the result of performing the
6827 `%i' substitution. (See below.)
6829 As an alternative to providing a spec string, the text that
6830 follows a suffix directive can be one of the following:
6833 This says that the suffix is an alias for a known LANGUAGE.
6834 This is similar to using the `-x' command-line switch to GCC
6835 to specify a language explicitly. For example:
6840 Says that .ZZ files are, in fact, C++ source files.
6843 This causes an error messages saying:
6845 NAME compiler not installed on this system.
6847 GCC already has an extensive list of suffixes built into it. This
6848 directive will add an entry to the end of the list of suffixes, but
6849 since the list is searched from the end backwards, it is
6850 effectively possible to override earlier entries using this
6854 GCC has the following spec strings built into it. Spec files can
6855 override these strings or create their own. Note that individual
6856 targets can also add their own spec strings to this list.
6858 asm Options to pass to the assembler
6859 asm_final Options to pass to the assembler post-processor
6860 cpp Options to pass to the C preprocessor
6861 cc1 Options to pass to the C compiler
6862 cc1plus Options to pass to the C++ compiler
6863 endfile Object files to include at the end of the link
6864 link Options to pass to the linker
6865 lib Libraries to include on the command line to the linker
6866 libgcc Decides which GCC support library to pass to the linker
6867 linker Sets the name of the linker
6868 predefines Defines to be passed to the C preprocessor
6869 signed_char Defines to pass to CPP to say whether `char' is signed
6871 startfile Object files to include at the start of the link
6873 Here is a small example of a spec file:
6878 --start-group -lgcc -lc -leval1 --end-group %(old_lib)
6880 This example renames the spec called `lib' to `old_lib' and then
6881 overrides the previous definition of `lib' with a new one. The new
6882 definition adds in some extra command-line options before including the
6883 text of the old definition.
6885 "Spec strings" are a list of command-line options to be passed to their
6886 corresponding program. In addition, the spec strings can contain
6887 `%'-prefixed sequences to substitute variable text or to conditionally
6888 insert text into the command line. Using these constructs it is
6889 possible to generate quite complex command lines.
6891 Here is a table of all defined `%'-sequences for spec strings. Note
6892 that spaces are not generated automatically around the results of
6893 expanding these sequences. Therefore you can concatenate them together
6894 or combine them with constant text in a single argument.
6897 Substitute one `%' into the program name or argument.
6900 Substitute the name of the input file being processed.
6903 Substitute the basename of the input file being processed. This
6904 is the substring up to (and not including) the last period and not
6905 including the directory.
6908 This is the same as `%b', but include the file suffix (text after
6912 Marks the argument containing or following the `%d' as a temporary
6913 file name, so that that file will be deleted if GCC exits
6914 successfully. Unlike `%g', this contributes no text to the
6918 Substitute a file name that has suffix SUFFIX and is chosen once
6919 per compilation, and mark the argument in the same way as `%d'.
6920 To reduce exposure to denial-of-service attacks, the file name is
6921 now chosen in a way that is hard to predict even when previously
6922 chosen file names are known. For example, `%g.s ... %g.o ... %g.s'
6923 might turn into `ccUVUUAU.s ccXYAXZ12.o ccUVUUAU.s'. SUFFIX
6924 matches the regexp `[.A-Za-z]*' or the special string `%O', which
6925 is treated exactly as if `%O' had been preprocessed. Previously,
6926 `%g' was simply substituted with a file name chosen once per
6927 compilation, without regard to any appended suffix (which was
6928 therefore treated just like ordinary text), making such attacks
6929 more likely to succeed.
6932 Like `%g', but generates a new temporary file name even if
6933 `%uSUFFIX' was already seen.
6936 Substitutes the last file name generated with `%uSUFFIX',
6937 generating a new one if there is no such last file name. In the
6938 absence of any `%uSUFFIX', this is just like `%gSUFFIX', except
6939 they don't share the same suffix _space_, so `%g.s ... %U.s ...
6940 %g.s ... %U.s' would involve the generation of two distinct file
6941 names, one for each `%g.s' and another for each `%U.s'.
6942 Previously, `%U' was simply substituted with a file name chosen
6943 for the previous `%u', without regard to any appended suffix.
6946 Substitutes the name of the `HOST_BIT_BUCKET', if any, and if it is
6947 writable, and if save-temps is off; otherwise, substitute the name
6948 of a temporary file, just like `%u'. This temporary file is not
6949 meant for communication between processes, but rather as a junk
6954 Like `%g', except if `-pipe' is in effect. In that case `%|'
6955 substitutes a single dash and `%m' substitutes nothing at all.
6956 These are the two most common ways to instruct a program that it
6957 should read from standard input or write to standard output. If
6958 you need something more elaborate you can use an `%{pipe:`X'}'
6959 construct: see for example `f/lang-specs.h'.
6962 Substitutes .SUFFIX for the suffixes of a matched switch's args
6963 when it is subsequently output with `%*'. SUFFIX is terminated by
6964 the next space or %.
6967 Marks the argument containing or following the `%w' as the
6968 designated output file of this compilation. This puts the argument
6969 into the sequence of arguments that `%o' will substitute later.
6972 Substitutes the names of all the output files, with spaces
6973 automatically placed around them. You should write spaces around
6974 the `%o' as well or the results are undefined. `%o' is for use in
6975 the specs for running the linker. Input files whose names have no
6976 recognized suffix are not compiled at all, but they are included
6977 among the output files, so they will be linked.
6980 Substitutes the suffix for object files. Note that this is
6981 handled specially when it immediately follows `%g, %u, or %U',
6982 because of the need for those to form complete file names. The
6983 handling is such that `%O' is treated exactly as if it had already
6984 been substituted, except that `%g, %u, and %U' do not currently
6985 support additional SUFFIX characters following `%O' as they would
6986 following, for example, `.o'.
6989 Substitutes the standard macro predefinitions for the current
6990 target machine. Use this when running `cpp'.
6993 Like `%p', but puts `__' before and after the name of each
6994 predefined macro, except for macros that start with `__' or with
6995 `_L', where L is an uppercase letter. This is for ISO C.
6998 Substitute any of `-iprefix' (made from `GCC_EXEC_PREFIX'),
6999 `-isysroot' (made from `TARGET_SYSTEM_ROOT'), and `-isystem' (made
7000 from `COMPILER_PATH' and `-B' options) as necessary.
7003 Current argument is the name of a library or startup file of some
7004 sort. Search for that file in a standard list of directories and
7005 substitute the full name found.
7008 Print STR as an error message. STR is terminated by a newline.
7009 Use this when inconsistent options are detected.
7012 Substitute the contents of spec string NAME at this point.
7015 Like `%(...)' but put `__' around `-D' arguments.
7018 Accumulate an option for `%X'.
7021 Output the accumulated linker options specified by `-Wl' or a `%x'
7025 Output the accumulated assembler options specified by `-Wa'.
7028 Output the accumulated preprocessor options specified by `-Wp'.
7031 Process the `asm' spec. This is used to compute the switches to
7032 be passed to the assembler.
7035 Process the `asm_final' spec. This is a spec string for passing
7036 switches to an assembler post-processor, if such a program is
7040 Process the `link' spec. This is the spec for computing the
7041 command line passed to the linker. Typically it will make use of
7042 the `%L %G %S %D and %E' sequences.
7045 Dump out a `-L' option for each directory that GCC believes might
7046 contain startup files. If the target supports multilibs then the
7047 current multilib directory will be prepended to each of these
7051 Process the `lib' spec. This is a spec string for deciding which
7052 libraries should be included on the command line to the linker.
7055 Process the `libgcc' spec. This is a spec string for deciding
7056 which GCC support library should be included on the command line
7060 Process the `startfile' spec. This is a spec for deciding which
7061 object files should be the first ones passed to the linker.
7062 Typically this might be a file named `crt0.o'.
7065 Process the `endfile' spec. This is a spec string that specifies
7066 the last object files that will be passed to the linker.
7069 Process the `cpp' spec. This is used to construct the arguments
7070 to be passed to the C preprocessor.
7073 Process the `cc1' spec. This is used to construct the options to
7074 be passed to the actual C compiler (`cc1').
7077 Process the `cc1plus' spec. This is used to construct the options
7078 to be passed to the actual C++ compiler (`cc1plus').
7081 Substitute the variable part of a matched option. See below.
7082 Note that each comma in the substituted string is replaced by a
7086 Remove all occurrences of `-S' from the command line. Note--this
7087 command is position dependent. `%' commands in the spec string
7088 before this one will see `-S', `%' commands in the spec string
7089 after this one will not.
7092 Call the named function FUNCTION, passing it ARGS. ARGS is first
7093 processed as a nested spec string, then split into an argument
7094 vector in the usual fashion. The function returns a string which
7095 is processed as if it had appeared literally as part of the
7098 The following built-in spec functions are provided:
7101 The `if-exists' spec function takes one argument, an absolute
7102 pathname to a file. If the file exists, `if-exists' returns
7103 the pathname. Here is a small example of its usage:
7106 crt0%O%s %:if-exists(crti%O%s) crtbegin%O%s
7109 The `if-exists-else' spec function is similar to the
7110 `if-exists' spec function, except that it takes two
7111 arguments. The first argument is an absolute pathname to a
7112 file. If the file exists, `if-exists-else' returns the
7113 pathname. If it does not exist, it returns the second
7114 argument. This way, `if-exists-else' can be used to select
7115 one file or another, based on the existence of the first.
7116 Here is a small example of its usage:
7119 crt0%O%s %:if-exists(crti%O%s) \
7120 %:if-exists-else(crtbeginT%O%s crtbegin%O%s)
7123 The `replace-outfile' spec function takes two arguments. It
7124 looks for the first argument in the outfiles array and
7125 replaces it with the second argument. Here is a small
7126 example of its usage:
7128 %{fgnu-runtime:%:replace-outfile(-lobjc -lobjc-gnu)}
7132 Substitutes the `-S' switch, if that switch was given to GCC. If
7133 that switch was not specified, this substitutes nothing. Note that
7134 the leading dash is omitted when specifying this option, and it is
7135 automatically inserted if the substitution is performed. Thus the
7136 spec string `%{foo}' would match the command-line option `-foo'
7137 and would output the command line option `-foo'.
7140 Like %{`S'} but mark last argument supplied within as a file to be
7144 Substitutes all the switches specified to GCC whose names start
7145 with `-S', but which also take an argument. This is used for
7146 switches like `-o', `-D', `-I', etc. GCC considers `-o foo' as
7147 being one switch whose names starts with `o'. %{o*} would
7148 substitute this text, including the space. Thus two arguments
7152 Like %{`S'*}, but preserve order of `S' and `T' options (the order
7153 of `S' and `T' in the spec is not significant). There can be any
7154 number of ampersand-separated variables; for each the wild card is
7155 optional. Useful for CPP as `%{D*&U*&A*}'.
7158 Substitutes `X', if the `-S' switch was given to GCC.
7161 Substitutes `X', if the `-S' switch was _not_ given to GCC.
7164 Substitutes `X' if one or more switches whose names start with
7165 `-S' are specified to GCC. Normally `X' is substituted only once,
7166 no matter how many such switches appeared. However, if `%*'
7167 appears somewhere in `X', then `X' will be substituted once for
7168 each matching switch, with the `%*' replaced by the part of that
7169 switch that matched the `*'.
7172 Substitutes `X', if processing a file with suffix `S'.
7175 Substitutes `X', if _not_ processing a file with suffix `S'.
7178 Substitutes `X' if either `-S' or `-P' was given to GCC. This may
7179 be combined with `!', `.', and `*' sequences as well, although
7180 they have a stronger binding than the `|'. If `%*' appears in
7181 `X', all of the alternatives must be starred, and only the first
7182 matching alternative is substituted.
7184 For example, a spec string like this:
7186 %{.c:-foo} %{!.c:-bar} %{.c|d:-baz} %{!.c|d:-boggle}
7188 will output the following command-line options from the following
7189 input command-line options:
7193 -d fred.c -foo -baz -boggle
7194 -d jim.d -bar -baz -boggle
7197 If `S' was given to GCC, substitutes `X'; else if `T' was given to
7198 GCC, substitutes `Y'; else substitutes `D'. There can be as many
7199 clauses as you need. This may be combined with `.', `!', `|', and
7203 The conditional text `X' in a %{`S':`X'} or similar construct may
7204 contain other nested `%' constructs or spaces, or even newlines. They
7205 are processed as usual, as described above. Trailing white space in
7206 `X' is ignored. White space may also appear anywhere on the left side
7207 of the colon in these constructs, except between `.' or `*' and the
7210 The `-O', `-f', `-m', and `-W' switches are handled specifically in
7211 these constructs. If another value of `-O' or the negated form of a
7212 `-f', `-m', or `-W' switch is found later in the command line, the
7213 earlier switch value is ignored, except with {`S'*} where `S' is just
7214 one letter, which passes all matching options.
7216 The character `|' at the beginning of the predicate text is used to
7217 indicate that a command should be piped to the following command, but
7218 only if `-pipe' is specified.
7220 It is built into GCC which switches take arguments and which do not.
7221 (You might think it would be useful to generalize this to allow each
7222 compiler's spec to say which switches take arguments. But this cannot
7223 be done in a consistent fashion. GCC cannot even decide which input
7224 files have been specified without knowing which switches take arguments,
7225 and it must know which input files to compile in order to tell which
7228 GCC also knows implicitly that arguments starting in `-l' are to be
7229 treated as compiler output files, and passed to the linker in their
7230 proper position among the other output files.
7233 File: gcc.info, Node: Target Options, Next: Submodel Options, Prev: Spec Files, Up: Invoking GCC
7235 3.16 Specifying Target Machine and Compiler Version
7236 ===================================================
7238 The usual way to run GCC is to run the executable called `gcc', or
7239 `<machine>-gcc' when cross-compiling, or `<machine>-gcc-<version>' to
7240 run a version other than the one that was installed last. Sometimes
7241 this is inconvenient, so GCC provides options that will switch to
7242 another cross-compiler or version.
7245 The argument MACHINE specifies the target machine for compilation.
7247 The value to use for MACHINE is the same as was specified as the
7248 machine type when configuring GCC as a cross-compiler. For
7249 example, if a cross-compiler was configured with `configure
7250 arm-elf', meaning to compile for an arm processor with elf
7251 binaries, then you would specify `-b arm-elf' to run that cross
7252 compiler. Because there are other options beginning with `-b', the
7253 configuration must contain a hyphen.
7256 The argument VERSION specifies which version of GCC to run. This
7257 is useful when multiple versions are installed. For example,
7258 VERSION might be `4.0', meaning to run GCC version 4.0.
7260 The `-V' and `-b' options work by running the
7261 `<machine>-gcc-<version>' executable, so there's no real reason to use
7262 them if you can just run that directly.
7265 File: gcc.info, Node: Submodel Options, Next: Code Gen Options, Prev: Target Options, Up: Invoking GCC
7267 3.17 Hardware Models and Configurations
7268 =======================================
7270 Earlier we discussed the standard option `-b' which chooses among
7271 different installed compilers for completely different target machines,
7272 such as VAX vs. 68000 vs. 80386.
7274 In addition, each of these target machine types can have its own
7275 special options, starting with `-m', to choose among various hardware
7276 models or configurations--for example, 68010 vs 68020, floating
7277 coprocessor or none. A single installed version of the compiler can
7278 compile for any model or configuration, according to the options
7281 Some configurations of the compiler also support additional special
7282 options, usually for compatibility with other compilers on the same
7290 * Blackfin Options::
7294 * DEC Alpha Options::
7295 * DEC Alpha/VMS Options::
7299 * i386 and x86-64 Options::
7312 * RS/6000 and PowerPC Options::
7313 * S/390 and zSeries Options::
7316 * System V Options::
7317 * TMS320C3x/C4x Options::
7321 * Xstormy16 Options::
7326 File: gcc.info, Node: ARC Options, Next: ARM Options, Up: Submodel Options
7331 These options are defined for ARC implementations:
7334 Compile code for little endian mode. This is the default.
7337 Compile code for big endian mode.
7340 Prepend the name of the cpu to all public symbol names. In
7341 multiple-processor systems, there are many ARC variants with
7342 different instruction and register set characteristics. This flag
7343 prevents code compiled for one cpu to be linked with code compiled
7344 for another. No facility exists for handling variants that are
7345 "almost identical". This is an all or nothing option.
7348 Compile code for ARC variant CPU. Which variants are supported
7349 depend on the configuration. All variants support `-mcpu=base',
7350 this is the default.
7352 `-mtext=TEXT-SECTION'
7353 `-mdata=DATA-SECTION'
7354 `-mrodata=READONLY-DATA-SECTION'
7355 Put functions, data, and readonly data in TEXT-SECTION,
7356 DATA-SECTION, and READONLY-DATA-SECTION respectively by default.
7357 This can be overridden with the `section' attribute. *Note
7358 Variable Attributes::.
7362 File: gcc.info, Node: ARM Options, Next: AVR Options, Prev: ARC Options, Up: Submodel Options
7367 These `-m' options are defined for Advanced RISC Machines (ARM)
7371 Generate code for the specified ABI. Permissible values are:
7372 `apcs-gnu', `atpcs', `aapcs', `aapcs-linux' and `iwmmxt'.
7375 Generate a stack frame that is compliant with the ARM Procedure
7376 Call Standard for all functions, even if this is not strictly
7377 necessary for correct execution of the code. Specifying
7378 `-fomit-frame-pointer' with this option will cause the stack
7379 frames not to be generated for leaf functions. The default is
7383 This is a synonym for `-mapcs-frame'.
7386 Generate code which supports calling between the ARM and Thumb
7387 instruction sets. Without this option the two instruction sets
7388 cannot be reliably used inside one program. The default is
7389 `-mno-thumb-interwork', since slightly larger code is generated
7390 when `-mthumb-interwork' is specified.
7393 Prevent the reordering of instructions in the function prolog, or
7394 the merging of those instruction with the instructions in the
7395 function's body. This means that all functions will start with a
7396 recognizable set of instructions (or in fact one of a choice from
7397 a small set of different function prologues), and this information
7398 can be used to locate the start if functions inside an executable
7399 piece of code. The default is `-msched-prolog'.
7402 Generate output containing floating point instructions. This is
7406 Generate output containing library calls for floating point.
7407 *Warning:* the requisite libraries are not available for all ARM
7408 targets. Normally the facilities of the machine's usual C
7409 compiler are used, but this cannot be done directly in
7410 cross-compilation. You must make your own arrangements to provide
7411 suitable library functions for cross-compilation.
7413 `-msoft-float' changes the calling convention in the output file;
7414 therefore, it is only useful if you compile _all_ of a program with
7415 this option. In particular, you need to compile `libgcc.a', the
7416 library that comes with GCC, with `-msoft-float' in order for this
7420 Specifies which ABI to use for floating point values. Permissible
7421 values are: `soft', `softfp' and `hard'.
7423 `soft' and `hard' are equivalent to `-msoft-float' and
7424 `-mhard-float' respectively. `softfp' allows the generation of
7425 floating point instructions, but still uses the soft-float calling
7429 Generate code for a processor running in little-endian mode. This
7430 is the default for all standard configurations.
7433 Generate code for a processor running in big-endian mode; the
7434 default is to compile code for a little-endian processor.
7436 `-mwords-little-endian'
7437 This option only applies when generating code for big-endian
7438 processors. Generate code for a little-endian word order but a
7439 big-endian byte order. That is, a byte order of the form
7440 `32107654'. Note: this option should only be used if you require
7441 compatibility with code for big-endian ARM processors generated by
7442 versions of the compiler prior to 2.8.
7445 This specifies the name of the target ARM processor. GCC uses
7446 this name to determine what kind of instructions it can emit when
7447 generating assembly code. Permissible names are: `arm2', `arm250',
7448 `arm3', `arm6', `arm60', `arm600', `arm610', `arm620', `arm7',
7449 `arm7m', `arm7d', `arm7dm', `arm7di', `arm7dmi', `arm70', `arm700',
7450 `arm700i', `arm710', `arm710c', `arm7100', `arm7500', `arm7500fe',
7451 `arm7tdmi', `arm7tdmi-s', `arm8', `strongarm', `strongarm110',
7452 `strongarm1100', `arm8', `arm810', `arm9', `arm9e', `arm920',
7453 `arm920t', `arm922t', `arm946e-s', `arm966e-s', `arm968e-s',
7454 `arm926ej-s', `arm940t', `arm9tdmi', `arm10tdmi', `arm1020t',
7455 `arm1026ej-s', `arm10e', `arm1020e', `arm1022e', `arm1136j-s',
7456 `arm1136jf-s', `mpcore', `mpcorenovfp', `arm1176jz-s',
7457 `arm1176jzf-s', `xscale', `iwmmxt', `ep9312'.
7460 This option is very similar to the `-mcpu=' option, except that
7461 instead of specifying the actual target processor type, and hence
7462 restricting which instructions can be used, it specifies that GCC
7463 should tune the performance of the code as if the target were of
7464 the type specified in this option, but still choosing the
7465 instructions that it will generate based on the cpu specified by a
7466 `-mcpu=' option. For some ARM implementations better performance
7467 can be obtained by using this option.
7470 This specifies the name of the target ARM architecture. GCC uses
7471 this name to determine what kind of instructions it can emit when
7472 generating assembly code. This option can be used in conjunction
7473 with or instead of the `-mcpu=' option. Permissible names are:
7474 `armv2', `armv2a', `armv3', `armv3m', `armv4', `armv4t', `armv5',
7475 `armv5t', `armv5te', `armv6', `armv6j', `iwmmxt', `ep9312'.
7480 This specifies what floating point hardware (or hardware
7481 emulation) is available on the target. Permissible names are:
7482 `fpa', `fpe2', `fpe3', `maverick', `vfp'. `-mfp' and `-mfpe' are
7483 synonyms for `-mfpu'=`fpe'NUMBER, for compatibility with older
7486 If `-msoft-float' is specified this specifies the format of
7487 floating point values.
7489 `-mstructure-size-boundary=N'
7490 The size of all structures and unions will be rounded up to a
7491 multiple of the number of bits set by this option. Permissible
7492 values are 8, 32 and 64. The default value varies for different
7493 toolchains. For the COFF targeted toolchain the default value is
7494 8. A value of 64 is only allowed if the underlying ABI supports
7497 Specifying the larger number can produce faster, more efficient
7498 code, but can also increase the size of the program. Different
7499 values are potentially incompatible. Code compiled with one value
7500 cannot necessarily expect to work with code or libraries compiled
7501 with another value, if they exchange information using structures
7504 `-mabort-on-noreturn'
7505 Generate a call to the function `abort' at the end of a `noreturn'
7506 function. It will be executed if the function tries to return.
7510 Tells the compiler to perform function calls by first loading the
7511 address of the function into a register and then performing a
7512 subroutine call on this register. This switch is needed if the
7513 target function will lie outside of the 64 megabyte addressing
7514 range of the offset based version of subroutine call instruction.
7516 Even if this switch is enabled, not all function calls will be
7517 turned into long calls. The heuristic is that static functions,
7518 functions which have the `short-call' attribute, functions that
7519 are inside the scope of a `#pragma no_long_calls' directive and
7520 functions whose definitions have already been compiled within the
7521 current compilation unit, will not be turned into long calls. The
7522 exception to this rule is that weak function definitions,
7523 functions with the `long-call' attribute or the `section'
7524 attribute, and functions that are within the scope of a `#pragma
7525 long_calls' directive, will always be turned into long calls.
7527 This feature is not enabled by default. Specifying
7528 `-mno-long-calls' will restore the default behavior, as will
7529 placing the function calls within the scope of a `#pragma
7530 long_calls_off' directive. Note these switches have no effect on
7531 how the compiler generates code to handle function calls via
7534 `-mnop-fun-dllimport'
7535 Disable support for the `dllimport' attribute.
7538 Treat the register used for PIC addressing as read-only, rather
7539 than loading it in the prologue for each function. The run-time
7540 system is responsible for initializing this register with an
7541 appropriate value before execution begins.
7543 `-mpic-register=REG'
7544 Specify the register to be used for PIC addressing. The default
7545 is R10 unless stack-checking is enabled, when R9 is used.
7547 `-mcirrus-fix-invalid-insns'
7548 Insert NOPs into the instruction stream to in order to work around
7549 problems with invalid Maverick instruction combinations. This
7550 option is only valid if the `-mcpu=ep9312' option has been used to
7551 enable generation of instructions for the Cirrus Maverick floating
7552 point co-processor. This option is not enabled by default, since
7553 the problem is only present in older Maverick implementations.
7554 The default can be re-enabled by use of the
7555 `-mno-cirrus-fix-invalid-insns' switch.
7557 `-mpoke-function-name'
7558 Write the name of each function into the text section, directly
7559 preceding the function prologue. The generated code is similar to
7563 .ascii "arm_poke_function_name", 0
7566 .word 0xff000000 + (t1 - t0)
7567 arm_poke_function_name
7569 stmfd sp!, {fp, ip, lr, pc}
7572 When performing a stack backtrace, code can inspect the value of
7573 `pc' stored at `fp + 0'. If the trace function then looks at
7574 location `pc - 12' and the top 8 bits are set, then we know that
7575 there is a function name embedded immediately preceding this
7576 location and has length `((pc[-3]) & 0xff000000)'.
7579 Generate code for the 16-bit Thumb instruction set. The default
7580 is to use the 32-bit ARM instruction set.
7583 Generate a stack frame that is compliant with the Thumb Procedure
7584 Call Standard for all non-leaf functions. (A leaf function is one
7585 that does not call any other functions.) The default is
7589 Generate a stack frame that is compliant with the Thumb Procedure
7590 Call Standard for all leaf functions. (A leaf function is one
7591 that does not call any other functions.) The default is
7592 `-mno-apcs-leaf-frame'.
7594 `-mcallee-super-interworking'
7595 Gives all externally visible functions in the file being compiled
7596 an ARM instruction set header which switches to Thumb mode before
7597 executing the rest of the function. This allows these functions
7598 to be called from non-interworking code.
7600 `-mcaller-super-interworking'
7601 Allows calls via function pointers (including virtual functions) to
7602 execute correctly regardless of whether the target code has been
7603 compiled for interworking or not. There is a small overhead in
7604 the cost of executing a function pointer if this option is enabled.
7607 Specify the access model for the thread local storage pointer.
7608 The valid models are `soft', which generates calls to
7609 `__aeabi_read_tp', `cp15', which fetches the thread pointer from
7610 `cp15' directly (supported in the arm6k architecture), and `auto',
7611 which uses the best available method for the selected processor.
7612 The default setting is `auto'.
7616 File: gcc.info, Node: AVR Options, Next: Blackfin Options, Prev: ARM Options, Up: Submodel Options
7621 These options are defined for AVR implementations:
7624 Specify ATMEL AVR instruction set or MCU type.
7626 Instruction set avr1 is for the minimal AVR core, not supported by
7627 the C compiler, only for assembler programs (MCU types: at90s1200,
7628 attiny10, attiny11, attiny12, attiny15, attiny28).
7630 Instruction set avr2 (default) is for the classic AVR core with up
7631 to 8K program memory space (MCU types: at90s2313, at90s2323,
7632 attiny22, at90s2333, at90s2343, at90s4414, at90s4433, at90s4434,
7633 at90s8515, at90c8534, at90s8535).
7635 Instruction set avr3 is for the classic AVR core with up to 128K
7636 program memory space (MCU types: atmega103, atmega603, at43usb320,
7639 Instruction set avr4 is for the enhanced AVR core with up to 8K
7640 program memory space (MCU types: atmega8, atmega83, atmega85).
7642 Instruction set avr5 is for the enhanced AVR core with up to 128K
7643 program memory space (MCU types: atmega16, atmega161, atmega163,
7644 atmega32, atmega323, atmega64, atmega128, at43usb355, at94k).
7647 Output instruction sizes to the asm file.
7650 Specify the initial stack address, which may be a symbol or
7651 numeric value, `__stack' is the default.
7654 Generated code is not compatible with hardware interrupts. Code
7655 size will be smaller.
7658 Functions prologues/epilogues expanded as call to appropriate
7659 subroutines. Code size will be smaller.
7662 Do not generate tablejump insns which sometimes increase code size.
7665 Change only the low 8 bits of the stack pointer.
7668 Assume int to be 8 bit integer. This affects the sizes of all
7669 types: A char will be 1 byte, an int will be 1 byte, an long will
7670 be 2 bytes and long long will be 4 bytes. Please note that this
7671 option does not comply to the C standards, but it will provide you
7672 with smaller code size.
7675 File: gcc.info, Node: Blackfin Options, Next: CRIS Options, Prev: AVR Options, Up: Submodel Options
7677 3.17.4 Blackfin Options
7678 -----------------------
7680 `-momit-leaf-frame-pointer'
7681 Don't keep the frame pointer in a register for leaf functions.
7682 This avoids the instructions to save, set up and restore frame
7683 pointers and makes an extra register available in leaf functions.
7684 The option `-fomit-frame-pointer' removes the frame pointer for
7685 all functions which might make debugging harder.
7688 When enabled, the compiler will ensure that the generated code
7689 does not contain speculative loads after jump instructions. This
7690 option is enabled by default.
7692 `-mno-specld-anomaly'
7693 Don't generate extra code to prevent speculative loads from
7697 When enabled, the compiler will ensure that the generated code
7698 does not contain CSYNC or SSYNC instructions too soon after
7699 conditional branches. This option is enabled by default.
7701 `-mno-csync-anomaly'
7702 Don't generate extra code to prevent CSYNC or SSYNC instructions
7703 from occurring too soon after a conditional branch.
7706 When enabled, the compiler is free to take advantage of the
7707 knowledge that the entire program fits into the low 64k of memory.
7710 Assume that the program is arbitrarily large. This is the default.
7712 `-mid-shared-library'
7713 Generate code that supports shared libraries via the library ID
7714 method. This allows for execute in place and shared libraries in
7715 an environment without virtual memory management. This option
7718 `-mno-id-shared-library'
7719 Generate code that doesn't assume ID based shared libraries are
7720 being used. This is the default.
7722 `-mshared-library-id=n'
7723 Specified the identification number of the ID based shared library
7724 being compiled. Specifying a value of 0 will generate more
7725 compact code, specifying other values will force the allocation of
7726 that number to the current library but is no more space or time
7727 efficient than omitting this option.
7731 Tells the compiler to perform function calls by first loading the
7732 address of the function into a register and then performing a
7733 subroutine call on this register. This switch is needed if the
7734 target function will lie outside of the 24 bit addressing range of
7735 the offset based version of subroutine call instruction.
7737 This feature is not enabled by default. Specifying
7738 `-mno-long-calls' will restore the default behavior. Note these
7739 switches have no effect on how the compiler generates code to
7740 handle function calls via function pointers.
7743 File: gcc.info, Node: CRIS Options, Next: CRX Options, Prev: Blackfin Options, Up: Submodel Options
7748 These options are defined specifically for the CRIS ports.
7750 `-march=ARCHITECTURE-TYPE'
7751 `-mcpu=ARCHITECTURE-TYPE'
7752 Generate code for the specified architecture. The choices for
7753 ARCHITECTURE-TYPE are `v3', `v8' and `v10' for respectively
7754 ETRAX 4, ETRAX 100, and ETRAX 100 LX. Default is `v0' except for
7755 cris-axis-linux-gnu, where the default is `v10'.
7757 `-mtune=ARCHITECTURE-TYPE'
7758 Tune to ARCHITECTURE-TYPE everything applicable about the generated
7759 code, except for the ABI and the set of available instructions.
7760 The choices for ARCHITECTURE-TYPE are the same as for
7761 `-march=ARCHITECTURE-TYPE'.
7763 `-mmax-stack-frame=N'
7764 Warn when the stack frame of a function exceeds N bytes.
7766 `-melinux-stacksize=N'
7767 Only available with the `cris-axis-aout' target. Arranges for
7768 indications in the program to the kernel loader that the stack of
7769 the program should be set to N bytes.
7773 The options `-metrax4' and `-metrax100' are synonyms for
7774 `-march=v3' and `-march=v8' respectively.
7776 `-mmul-bug-workaround'
7777 `-mno-mul-bug-workaround'
7778 Work around a bug in the `muls' and `mulu' instructions for CPU
7779 models where it applies. This option is active by default.
7782 Enable CRIS-specific verbose debug-related information in the
7783 assembly code. This option also has the effect to turn off the
7784 `#NO_APP' formatted-code indicator to the assembler at the
7785 beginning of the assembly file.
7788 Do not use condition-code results from previous instruction;
7789 always emit compare and test instructions before use of condition
7793 Do not emit instructions with side-effects in addressing modes
7794 other than post-increment.
7802 These options (no-options) arranges (eliminate arrangements) for
7803 the stack-frame, individual data and constants to be aligned for
7804 the maximum single data access size for the chosen CPU model. The
7805 default is to arrange for 32-bit alignment. ABI details such as
7806 structure layout are not affected by these options.
7811 Similar to the stack- data- and const-align options above, these
7812 options arrange for stack-frame, writable data and constants to
7813 all be 32-bit, 16-bit or 8-bit aligned. The default is 32-bit
7816 `-mno-prologue-epilogue'
7817 `-mprologue-epilogue'
7818 With `-mno-prologue-epilogue', the normal function prologue and
7819 epilogue that sets up the stack-frame are omitted and no return
7820 instructions or return sequences are generated in the code. Use
7821 this option only together with visual inspection of the compiled
7822 code: no warnings or errors are generated when call-saved
7823 registers must be saved, or storage for local variable needs to be
7828 With `-fpic' and `-fPIC', don't generate (do generate) instruction
7829 sequences that load addresses for functions from the PLT part of
7830 the GOT rather than (traditional on other architectures) calls to
7831 the PLT. The default is `-mgotplt'.
7834 Legacy no-op option only recognized with the cris-axis-aout target.
7837 Legacy no-op option only recognized with the cris-axis-elf and
7838 cris-axis-linux-gnu targets.
7841 Only recognized with the cris-axis-aout target, where it selects a
7842 GNU/linux-like multilib, include files and instruction set for
7846 Legacy no-op option only recognized with the cris-axis-linux-gnu
7850 This option, recognized for the cris-axis-aout and cris-axis-elf
7851 arranges to link with input-output functions from a simulator
7852 library. Code, initialized data and zero-initialized data are
7853 allocated consecutively.
7856 Like `-sim', but pass linker options to locate initialized data at
7857 0x40000000 and zero-initialized data at 0x80000000.
7860 File: gcc.info, Node: CRX Options, Next: Darwin Options, Prev: CRIS Options, Up: Submodel Options
7865 These options are defined specifically for the CRX ports.
7868 Enable the use of multiply-accumulate instructions. Disabled by
7872 Push instructions will be used to pass outgoing arguments when
7873 functions are called. Enabled by default.
7876 File: gcc.info, Node: Darwin Options, Next: DEC Alpha Options, Prev: CRX Options, Up: Submodel Options
7878 3.17.7 Darwin Options
7879 ---------------------
7881 These options are defined for all architectures running the Darwin
7884 FSF GCC on Darwin does not create "fat" object files; it will create
7885 an object file for the single architecture that it was built to target.
7886 Apple's GCC on Darwin does create "fat" files if multiple `-arch'
7887 options are used; it does so by running the compiler or linker multiple
7888 times and joining the results together with `lipo'.
7890 The subtype of the file created (like `ppc7400' or `ppc970' or `i686')
7891 is determined by the flags that specify the ISA that GCC is targetting,
7892 like `-mcpu' or `-march'. The `-force_cpusubtype_ALL' option can be
7893 used to override this.
7895 The Darwin tools vary in their behavior when presented with an ISA
7896 mismatch. The assembler, `as', will only permit instructions to be
7897 used that are valid for the subtype of the file it is generating, so
7898 you cannot put 64-bit instructions in an `ppc750' object file. The
7899 linker for shared libraries, `/usr/bin/libtool', will fail and print an
7900 error if asked to create a shared library with a less restrictive
7901 subtype than its input files (for instance, trying to put a `ppc970'
7902 object file in a `ppc7400' library). The linker for executables, `ld',
7903 will quietly give the executable the most restrictive subtype of any of
7907 Add the framework directory DIR to the head of the list of
7908 directories to be searched for header files. These directories are
7909 interleaved with those specified by `-I' options and are scanned
7910 in a left-to-right order.
7912 A framework directory is a directory with frameworks in it. A
7913 framework is a directory with a `"Headers"' and/or
7914 `"PrivateHeaders"' directory contained directly in it that ends in
7915 `".framework"'. The name of a framework is the name of this
7916 directory excluding the `".framework"'. Headers associated with
7917 the framework are found in one of those two directories, with
7918 `"Headers"' being searched first. A subframework is a framework
7919 directory that is in a framework's `"Frameworks"' directory.
7920 Includes of subframework headers can only appear in a header of a
7921 framework that contains the subframework, or in a sibling
7922 subframework header. Two subframeworks are siblings if they occur
7923 in the same framework. A subframework should not have the same
7924 name as a framework, a warning will be issued if this is violated.
7925 Currently a subframework cannot have subframeworks, in the
7926 future, the mechanism may be extended to support this. The
7927 standard frameworks can be found in `"/System/Library/Frameworks"'
7928 and `"/Library/Frameworks"'. An example include looks like
7929 `#include <Framework/header.h>', where `Framework' denotes the
7930 name of the framework and header.h is found in the
7931 `"PrivateHeaders"' or `"Headers"' directory.
7934 Emit debugging information for symbols that are used. For STABS
7935 debugging format, this enables `-feliminate-unused-debug-symbols'.
7936 This is by default ON.
7939 Emit debugging information for all symbols and types.
7941 `-mmacosx-version-min=VERSION'
7942 The earliest version of MacOS X that this executable will run on
7943 is VERSION. Typical values of VERSION include `10.1', `10.2', and
7946 The default for this option is to make choices that seem to be most
7950 Override the defaults for `bool' so that `sizeof(bool)==1'. By
7951 default `sizeof(bool)' is `4' when compiling for Darwin/PowerPC
7952 and `1' when compiling for Darwin/x86, so this option has no
7955 *Warning:* The `-mone-byte-bool' switch causes GCC to generate
7956 code that is not binary compatible with code generated without
7957 that switch. Using this switch may require recompiling all other
7958 modules in a program, including system libraries. Use this switch
7959 to conform to a non-default data model.
7961 `-mfix-and-continue'
7962 `-ffix-and-continue'
7964 Generate code suitable for fast turn around development. Needed to
7965 enable gdb to dynamically load `.o' files into already running
7966 programs. `-findirect-data' and `-ffix-and-continue' are provided
7967 for backwards compatibility.
7970 Loads all members of static archive libraries. See man ld(1) for
7973 `-arch_errors_fatal'
7974 Cause the errors having to do with files that have the wrong
7975 architecture to be fatal.
7978 Causes the output file to be marked such that the dynamic linker
7979 will bind all undefined references when the file is loaded or
7983 Produce a Mach-o bundle format file. See man ld(1) for more
7986 `-bundle_loader EXECUTABLE'
7987 This option specifies the EXECUTABLE that will be loading the build
7988 output file being linked. See man ld(1) for more information.
7991 When passed this option, GCC will produce a dynamic library
7992 instead of an executable when linking, using the Darwin `libtool'
7995 `-force_cpusubtype_ALL'
7996 This causes GCC's output file to have the ALL subtype, instead of
7997 one controlled by the `-mcpu' or `-march' option.
7999 `-allowable_client CLIENT_NAME'
8001 `-compatibility_version'
8006 `-dylinker_install_name'
8008 `-exported_symbols_list'
8011 `-force_flat_namespace'
8012 `-headerpad_max_install_names'
8016 `-keep_private_externs'
8019 `-multiply_defined_unused'
8021 `-no_dead_strip_inits_and_terms'
8028 `-prebind_all_twolevel_modules'
8032 `-sectobjectsymbols'
8036 `-sectobjectsymbols'
8039 `-segs_read_only_addr'
8040 `-segs_read_write_addr'
8042 `-seg_addr_table_filename'
8045 `-segs_read_only_addr'
8046 `-segs_read_write_addr'
8051 `-twolevel_namespace'
8054 `-unexported_symbols_list'
8055 `-weak_reference_mismatches'
8057 These options are passed to the Darwin linker. The Darwin linker
8058 man page describes them in detail.
8061 File: gcc.info, Node: DEC Alpha Options, Next: DEC Alpha/VMS Options, Prev: Darwin Options, Up: Submodel Options
8063 3.17.8 DEC Alpha Options
8064 ------------------------
8066 These `-m' options are defined for the DEC Alpha implementations:
8070 Use (do not use) the hardware floating-point instructions for
8071 floating-point operations. When `-msoft-float' is specified,
8072 functions in `libgcc.a' will be used to perform floating-point
8073 operations. Unless they are replaced by routines that emulate the
8074 floating-point operations, or compiled in such a way as to call
8075 such emulations routines, these routines will issue floating-point
8076 operations. If you are compiling for an Alpha without
8077 floating-point operations, you must ensure that the library is
8078 built so as not to call them.
8080 Note that Alpha implementations without floating-point operations
8081 are required to have floating-point registers.
8085 Generate code that uses (does not use) the floating-point register
8086 set. `-mno-fp-regs' implies `-msoft-float'. If the floating-point
8087 register set is not used, floating point operands are passed in
8088 integer registers as if they were integers and floating-point
8089 results are passed in `$0' instead of `$f0'. This is a
8090 non-standard calling sequence, so any function with a
8091 floating-point argument or return value called by code compiled
8092 with `-mno-fp-regs' must also be compiled with that option.
8094 A typical use of this option is building a kernel that does not
8095 use, and hence need not save and restore, any floating-point
8099 The Alpha architecture implements floating-point hardware
8100 optimized for maximum performance. It is mostly compliant with
8101 the IEEE floating point standard. However, for full compliance,
8102 software assistance is required. This option generates code fully
8103 IEEE compliant code _except_ that the INEXACT-FLAG is not
8104 maintained (see below). If this option is turned on, the
8105 preprocessor macro `_IEEE_FP' is defined during compilation. The
8106 resulting code is less efficient but is able to correctly support
8107 denormalized numbers and exceptional IEEE values such as
8108 not-a-number and plus/minus infinity. Other Alpha compilers call
8109 this option `-ieee_with_no_inexact'.
8111 `-mieee-with-inexact'
8112 This is like `-mieee' except the generated code also maintains the
8113 IEEE INEXACT-FLAG. Turning on this option causes the generated
8114 code to implement fully-compliant IEEE math. In addition to
8115 `_IEEE_FP', `_IEEE_FP_EXACT' is defined as a preprocessor macro.
8116 On some Alpha implementations the resulting code may execute
8117 significantly slower than the code generated by default. Since
8118 there is very little code that depends on the INEXACT-FLAG, you
8119 should normally not specify this option. Other Alpha compilers
8120 call this option `-ieee_with_inexact'.
8122 `-mfp-trap-mode=TRAP-MODE'
8123 This option controls what floating-point related traps are enabled.
8124 Other Alpha compilers call this option `-fptm TRAP-MODE'. The
8125 trap mode can be set to one of four values:
8128 This is the default (normal) setting. The only traps that
8129 are enabled are the ones that cannot be disabled in software
8130 (e.g., division by zero trap).
8133 In addition to the traps enabled by `n', underflow traps are
8137 Like `su', but the instructions are marked to be safe for
8138 software completion (see Alpha architecture manual for
8142 Like `su', but inexact traps are enabled as well.
8144 `-mfp-rounding-mode=ROUNDING-MODE'
8145 Selects the IEEE rounding mode. Other Alpha compilers call this
8146 option `-fprm ROUNDING-MODE'. The ROUNDING-MODE can be one of:
8149 Normal IEEE rounding mode. Floating point numbers are
8150 rounded towards the nearest machine number or towards the
8151 even machine number in case of a tie.
8154 Round towards minus infinity.
8157 Chopped rounding mode. Floating point numbers are rounded
8161 Dynamic rounding mode. A field in the floating point control
8162 register (FPCR, see Alpha architecture reference manual)
8163 controls the rounding mode in effect. The C library
8164 initializes this register for rounding towards plus infinity.
8165 Thus, unless your program modifies the FPCR, `d' corresponds
8166 to round towards plus infinity.
8168 `-mtrap-precision=TRAP-PRECISION'
8169 In the Alpha architecture, floating point traps are imprecise.
8170 This means without software assistance it is impossible to recover
8171 from a floating trap and program execution normally needs to be
8172 terminated. GCC can generate code that can assist operating
8173 system trap handlers in determining the exact location that caused
8174 a floating point trap. Depending on the requirements of an
8175 application, different levels of precisions can be selected:
8178 Program precision. This option is the default and means a
8179 trap handler can only identify which program caused a
8180 floating point exception.
8183 Function precision. The trap handler can determine the
8184 function that caused a floating point exception.
8187 Instruction precision. The trap handler can determine the
8188 exact instruction that caused a floating point exception.
8190 Other Alpha compilers provide the equivalent options called
8191 `-scope_safe' and `-resumption_safe'.
8194 This option marks the generated code as IEEE conformant. You must
8195 not use this option unless you also specify `-mtrap-precision=i'
8196 and either `-mfp-trap-mode=su' or `-mfp-trap-mode=sui'. Its only
8197 effect is to emit the line `.eflag 48' in the function prologue of
8198 the generated assembly file. Under DEC Unix, this has the effect
8199 that IEEE-conformant math library routines will be linked in.
8202 Normally GCC examines a 32- or 64-bit integer constant to see if
8203 it can construct it from smaller constants in two or three
8204 instructions. If it cannot, it will output the constant as a
8205 literal and generate code to load it from the data segment at
8208 Use this option to require GCC to construct _all_ integer constants
8209 using code, even if it takes more instructions (the maximum is
8212 You would typically use this option to build a shared library
8213 dynamic loader. Itself a shared library, it must relocate itself
8214 in memory before it can find the variables and constants in its
8219 Select whether to generate code to be assembled by the
8220 vendor-supplied assembler (`-malpha-as') or by the GNU assembler
8231 Indicate whether GCC should generate code to use the optional BWX,
8232 CIX, FIX and MAX instruction sets. The default is to use the
8233 instruction sets supported by the CPU type specified via `-mcpu='
8234 option or that of the CPU on which GCC was built if none was
8239 Generate code that uses (does not use) VAX F and G floating point
8240 arithmetic instead of IEEE single and double precision.
8243 `-mno-explicit-relocs'
8244 Older Alpha assemblers provided no way to generate symbol
8245 relocations except via assembler macros. Use of these macros does
8246 not allow optimal instruction scheduling. GNU binutils as of
8247 version 2.12 supports a new syntax that allows the compiler to
8248 explicitly mark which relocations should apply to which
8249 instructions. This option is mostly useful for debugging, as GCC
8250 detects the capabilities of the assembler when it is built and
8251 sets the default accordingly.
8255 When `-mexplicit-relocs' is in effect, static data is accessed via
8256 "gp-relative" relocations. When `-msmall-data' is used, objects 8
8257 bytes long or smaller are placed in a "small data area" (the
8258 `.sdata' and `.sbss' sections) and are accessed via 16-bit
8259 relocations off of the `$gp' register. This limits the size of
8260 the small data area to 64KB, but allows the variables to be
8261 directly accessed via a single instruction.
8263 The default is `-mlarge-data'. With this option the data area is
8264 limited to just below 2GB. Programs that require more than 2GB of
8265 data must use `malloc' or `mmap' to allocate the data in the heap
8266 instead of in the program's data segment.
8268 When generating code for shared libraries, `-fpic' implies
8269 `-msmall-data' and `-fPIC' implies `-mlarge-data'.
8273 When `-msmall-text' is used, the compiler assumes that the code of
8274 the entire program (or shared library) fits in 4MB, and is thus
8275 reachable with a branch instruction. When `-msmall-data' is used,
8276 the compiler can assume that all local symbols share the same
8277 `$gp' value, and thus reduce the number of instructions required
8278 for a function call from 4 to 1.
8280 The default is `-mlarge-text'.
8283 Set the instruction set and instruction scheduling parameters for
8284 machine type CPU_TYPE. You can specify either the `EV' style name
8285 or the corresponding chip number. GCC supports scheduling
8286 parameters for the EV4, EV5 and EV6 family of processors and will
8287 choose the default values for the instruction set from the
8288 processor you specify. If you do not specify a processor type,
8289 GCC will default to the processor on which the compiler was built.
8291 Supported values for CPU_TYPE are
8296 Schedules as an EV4 and has no instruction set extensions.
8300 Schedules as an EV5 and has no instruction set extensions.
8304 Schedules as an EV5 and supports the BWX extension.
8309 Schedules as an EV5 and supports the BWX and MAX extensions.
8313 Schedules as an EV6 and supports the BWX, FIX, and MAX
8318 Schedules as an EV6 and supports the BWX, CIX, FIX, and MAX
8322 Set only the instruction scheduling parameters for machine type
8323 CPU_TYPE. The instruction set is not changed.
8325 `-mmemory-latency=TIME'
8326 Sets the latency the scheduler should assume for typical memory
8327 references as seen by the application. This number is highly
8328 dependent on the memory access patterns used by the application
8329 and the size of the external cache on the machine.
8331 Valid options for TIME are
8334 A decimal number representing clock cycles.
8340 The compiler contains estimates of the number of clock cycles
8341 for "typical" EV4 & EV5 hardware for the Level 1, 2 & 3 caches
8342 (also called Dcache, Scache, and Bcache), as well as to main
8343 memory. Note that L3 is only valid for EV5.
8347 File: gcc.info, Node: DEC Alpha/VMS Options, Next: FRV Options, Prev: DEC Alpha Options, Up: Submodel Options
8349 3.17.9 DEC Alpha/VMS Options
8350 ----------------------------
8352 These `-m' options are defined for the DEC Alpha/VMS implementations:
8354 `-mvms-return-codes'
8355 Return VMS condition codes from main. The default is to return
8356 POSIX style condition (e.g. error) codes.
8359 File: gcc.info, Node: FRV Options, Next: H8/300 Options, Prev: DEC Alpha/VMS Options, Up: Submodel Options
8365 Only use the first 32 general purpose registers.
8368 Use all 64 general purpose registers.
8371 Use only the first 32 floating point registers.
8374 Use all 64 floating point registers
8377 Use hardware instructions for floating point operations.
8380 Use library routines for floating point operations.
8383 Dynamically allocate condition code registers.
8386 Do not try to dynamically allocate condition code registers, only
8387 use `icc0' and `fcc0'.
8390 Change ABI to use double word insns.
8393 Do not use double word instructions.
8396 Use floating point double instructions.
8399 Do not use floating point double instructions.
8402 Use media instructions.
8405 Do not use media instructions.
8408 Use multiply and add/subtract instructions.
8411 Do not use multiply and add/subtract instructions.
8414 Select the FDPIC ABI, that uses function descriptors to represent
8415 pointers to functions. Without any PIC/PIE-related options, it
8416 implies `-fPIE'. With `-fpic' or `-fpie', it assumes GOT entries
8417 and small data are within a 12-bit range from the GOT base
8418 address; with `-fPIC' or `-fPIE', GOT offsets are computed with 32
8422 Enable inlining of PLT entries in function calls to functions that
8423 are not known to bind locally. It has no effect without `-mfdpic'.
8424 It's enabled by default if optimizing for speed and compiling for
8425 shared libraries (i.e., `-fPIC' or `-fpic'), or when an
8426 optimization option such as `-O3' or above is present in the
8430 Assume a large TLS segment when generating thread-local code.
8433 Do not assume a large TLS segment when generating thread-local
8437 Enable the use of `GPREL' relocations in the FDPIC ABI for data
8438 that is known to be in read-only sections. It's enabled by
8439 default, except for `-fpic' or `-fpie': even though it may help
8440 make the global offset table smaller, it trades 1 instruction for
8441 4. With `-fPIC' or `-fPIE', it trades 3 instructions for 4, one
8442 of which may be shared by multiple symbols, and it avoids the need
8443 for a GOT entry for the referenced symbol, so it's more likely to
8444 be a win. If it is not, `-mno-gprel-ro' can be used to disable it.
8446 `-multilib-library-pic'
8447 Link with the (library, not FD) pic libraries. It's implied by
8448 `-mlibrary-pic', as well as by `-fPIC' and `-fpic' without
8449 `-mfdpic'. You should never have to use it explicitly.
8452 Follow the EABI requirement of always creating a frame pointer
8453 whenever a stack frame is allocated. This option is enabled by
8454 default and can be disabled with `-mno-linked-fp'.
8457 Use indirect addressing to call functions outside the current
8458 compilation unit. This allows the functions to be placed anywhere
8459 within the 32-bit address space.
8462 Try to align labels to an 8-byte boundary by inserting nops into
8463 the previous packet. This option only has an effect when VLIW
8464 packing is enabled. It doesn't create new packets; it merely adds
8465 nops to existing ones.
8468 Generate position-independent EABI code.
8471 Use only the first four media accumulator registers.
8474 Use all eight media accumulator registers.
8477 Pack VLIW instructions.
8480 Do not pack VLIW instructions.
8483 Do not mark ABI switches in e_flags.
8486 Enable the use of conditional-move instructions (default).
8488 This switch is mainly for debugging the compiler and will likely
8489 be removed in a future version.
8492 Disable the use of conditional-move instructions.
8494 This switch is mainly for debugging the compiler and will likely
8495 be removed in a future version.
8498 Enable the use of conditional set instructions (default).
8500 This switch is mainly for debugging the compiler and will likely
8501 be removed in a future version.
8504 Disable the use of conditional set instructions.
8506 This switch is mainly for debugging the compiler and will likely
8507 be removed in a future version.
8510 Enable the use of conditional execution (default).
8512 This switch is mainly for debugging the compiler and will likely
8513 be removed in a future version.
8516 Disable the use of conditional execution.
8518 This switch is mainly for debugging the compiler and will likely
8519 be removed in a future version.
8522 Run a pass to pack branches into VLIW instructions (default).
8524 This switch is mainly for debugging the compiler and will likely
8525 be removed in a future version.
8528 Do not run a pass to pack branches into VLIW instructions.
8530 This switch is mainly for debugging the compiler and will likely
8531 be removed in a future version.
8534 Enable optimization of `&&' and `||' in conditional execution
8537 This switch is mainly for debugging the compiler and will likely
8538 be removed in a future version.
8540 `-mno-multi-cond-exec'
8541 Disable optimization of `&&' and `||' in conditional execution.
8543 This switch is mainly for debugging the compiler and will likely
8544 be removed in a future version.
8546 `-mnested-cond-exec'
8547 Enable nested conditional execution optimizations (default).
8549 This switch is mainly for debugging the compiler and will likely
8550 be removed in a future version.
8552 `-mno-nested-cond-exec'
8553 Disable nested conditional execution optimizations.
8555 This switch is mainly for debugging the compiler and will likely
8556 be removed in a future version.
8559 This switch removes redundant `membar' instructions from the
8560 compiler generated code. It is enabled by default.
8562 `-mno-optimize-membar'
8563 This switch disables the automatic removal of redundant `membar'
8564 instructions from the generated code.
8567 Cause gas to print out tomcat statistics.
8570 Select the processor type for which to generate code. Possible
8571 values are `frv', `fr550', `tomcat', `fr500', `fr450', `fr405',
8572 `fr400', `fr300' and `simple'.
8576 File: gcc.info, Node: H8/300 Options, Next: HPPA Options, Prev: FRV Options, Up: Submodel Options
8578 3.17.11 H8/300 Options
8579 ----------------------
8581 These `-m' options are defined for the H8/300 implementations:
8584 Shorten some address references at link time, when possible; uses
8585 the linker option `-relax'. *Note `ld' and the H8/300:
8586 (ld)H8/300, for a fuller description.
8589 Generate code for the H8/300H.
8592 Generate code for the H8S.
8595 Generate code for the H8S and H8/300H in the normal mode. This
8596 switch must be used either with `-mh' or `-ms'.
8599 Generate code for the H8S/2600. This switch must be used with
8603 Make `int' data 32 bits by default.
8606 On the H8/300H and H8S, use the same alignment rules as for the
8607 H8/300. The default for the H8/300H and H8S is to align longs and
8608 floats on 4 byte boundaries. `-malign-300' causes them to be
8609 aligned on 2 byte boundaries. This option has no effect on the
8613 File: gcc.info, Node: HPPA Options, Next: i386 and x86-64 Options, Prev: H8/300 Options, Up: Submodel Options
8615 3.17.12 HPPA Options
8616 --------------------
8618 These `-m' options are defined for the HPPA family of computers:
8620 `-march=ARCHITECTURE-TYPE'
8621 Generate code for the specified architecture. The choices for
8622 ARCHITECTURE-TYPE are `1.0' for PA 1.0, `1.1' for PA 1.1, and
8623 `2.0' for PA 2.0 processors. Refer to `/usr/lib/sched.models' on
8624 an HP-UX system to determine the proper architecture option for
8625 your machine. Code compiled for lower numbered architectures will
8626 run on higher numbered architectures, but not the other way around.
8631 Synonyms for `-march=1.0', `-march=1.1', and `-march=2.0'
8635 Generate code suitable for big switch tables. Use this option
8636 only if the assembler/linker complain about out of range branches
8637 within a switch table.
8640 Fill delay slots of function calls with unconditional jump
8641 instructions by modifying the return pointer for the function call
8642 to be the target of the conditional jump.
8645 Prevent floating point registers from being used in any manner.
8646 This is necessary for compiling kernels which perform lazy context
8647 switching of floating point registers. If you use this option and
8648 attempt to perform floating point operations, the compiler will
8651 `-mdisable-indexing'
8652 Prevent the compiler from using indexing address modes. This
8653 avoids some rather obscure problems when compiling MIG generated
8657 Generate code that assumes the target has no space registers.
8658 This allows GCC to generate faster indirect calls and use unscaled
8659 index address modes.
8661 Such code is suitable for level 0 PA systems and kernels.
8663 `-mfast-indirect-calls'
8664 Generate code that assumes calls never cross space boundaries.
8665 This allows GCC to emit code which performs faster indirect calls.
8667 This option will not work in the presence of shared libraries or
8670 `-mfixed-range=REGISTER-RANGE'
8671 Generate code treating the given register range as fixed registers.
8672 A fixed register is one that the register allocator can not use.
8673 This is useful when compiling kernel code. A register range is
8674 specified as two registers separated by a dash. Multiple register
8675 ranges can be specified separated by a comma.
8678 Generate 3-instruction load and store sequences as sometimes
8679 required by the HP-UX 10 linker. This is equivalent to the `+k'
8680 option to the HP compilers.
8682 `-mportable-runtime'
8683 Use the portable calling conventions proposed by HP for ELF
8687 Enable the use of assembler directives only GAS understands.
8689 `-mschedule=CPU-TYPE'
8690 Schedule code according to the constraints for the machine type
8691 CPU-TYPE. The choices for CPU-TYPE are `700' `7100', `7100LC',
8692 `7200', `7300' and `8000'. Refer to `/usr/lib/sched.models' on an
8693 HP-UX system to determine the proper scheduling option for your
8694 machine. The default scheduling is `8000'.
8697 Enable the optimization pass in the HP-UX linker. Note this makes
8698 symbolic debugging impossible. It also triggers a bug in the
8699 HP-UX 8 and HP-UX 9 linkers in which they give bogus error
8700 messages when linking some programs.
8703 Generate output containing library calls for floating point.
8704 *Warning:* the requisite libraries are not available for all HPPA
8705 targets. Normally the facilities of the machine's usual C
8706 compiler are used, but this cannot be done directly in
8707 cross-compilation. You must make your own arrangements to provide
8708 suitable library functions for cross-compilation. The embedded
8709 target `hppa1.1-*-pro' does provide software floating point
8712 `-msoft-float' changes the calling convention in the output file;
8713 therefore, it is only useful if you compile _all_ of a program with
8714 this option. In particular, you need to compile `libgcc.a', the
8715 library that comes with GCC, with `-msoft-float' in order for this
8719 Generate the predefine, `_SIO', for server IO. The default is
8720 `-mwsio'. This generates the predefines, `__hp9000s700',
8721 `__hp9000s700__' and `_WSIO', for workstation IO. These options
8722 are available under HP-UX and HI-UX.
8725 Use GNU ld specific options. This passes `-shared' to ld when
8726 building a shared library. It is the default when GCC is
8727 configured, explicitly or implicitly, with the GNU linker. This
8728 option does not have any affect on which ld is called, it only
8729 changes what parameters are passed to that ld. The ld that is
8730 called is determined by the `--with-ld' configure option, GCC's
8731 program search path, and finally by the user's `PATH'. The linker
8732 used by GCC can be printed using `which `gcc
8733 -print-prog-name=ld`'. This option is only available on the 64
8734 bit HP-UX GCC, i.e. configured with `hppa*64*-*-hpux*'.
8737 Use HP ld specific options. This passes `-b' to ld when building
8738 a shared library and passes `+Accept TypeMismatch' to ld on all
8739 links. It is the default when GCC is configured, explicitly or
8740 implicitly, with the HP linker. This option does not have any
8741 affect on which ld is called, it only changes what parameters are
8742 passed to that ld. The ld that is called is determined by the
8743 `--with-ld' configure option, GCC's program search path, and
8744 finally by the user's `PATH'. The linker used by GCC can be
8745 printed using `which `gcc -print-prog-name=ld`'. This option is
8746 only available on the 64 bit HP-UX GCC, i.e. configured with
8750 Generate code that uses long call sequences. This ensures that a
8751 call is always able to reach linker generated stubs. The default
8752 is to generate long calls only when the distance from the call
8753 site to the beginning of the function or translation unit, as the
8754 case may be, exceeds a predefined limit set by the branch type
8755 being used. The limits for normal calls are 7,600,000 and 240,000
8756 bytes, respectively for the PA 2.0 and PA 1.X architectures.
8757 Sibcalls are always limited at 240,000 bytes.
8759 Distances are measured from the beginning of functions when using
8760 the `-ffunction-sections' option, or when using the `-mgas' and
8761 `-mno-portable-runtime' options together under HP-UX with the SOM
8764 It is normally not desirable to use this option as it will degrade
8765 performance. However, it may be useful in large applications,
8766 particularly when partial linking is used to build the application.
8768 The types of long calls used depends on the capabilities of the
8769 assembler and linker, and the type of code being generated. The
8770 impact on systems that support long absolute calls, and long pic
8771 symbol-difference or pc-relative calls should be relatively small.
8772 However, an indirect call is used on 32-bit ELF systems in pic code
8773 and it is quite long.
8776 Generate compiler predefines and select a startfile for the
8777 specified UNIX standard. The choices for UNIX-STD are `93', `95'
8778 and `98'. `93' is supported on all HP-UX versions. `95' is
8779 available on HP-UX 10.10 and later. `98' is available on HP-UX
8780 11.11 and later. The default values are `93' for HP-UX 10.00,
8781 `95' for HP-UX 10.10 though to 11.00, and `98' for HP-UX 11.11 and
8784 `-munix=93' provides the same predefines as GCC 3.3 and 3.4.
8785 `-munix=95' provides additional predefines for `XOPEN_UNIX' and
8786 `_XOPEN_SOURCE_EXTENDED', and the startfile `unix95.o'.
8787 `-munix=98' provides additional predefines for `_XOPEN_UNIX',
8788 `_XOPEN_SOURCE_EXTENDED', `_INCLUDE__STDC_A1_SOURCE' and
8789 `_INCLUDE_XOPEN_SOURCE_500', and the startfile `unix98.o'.
8791 It is _important_ to note that this option changes the interfaces
8792 for various library routines. It also affects the operational
8793 behavior of the C library. Thus, _extreme_ care is needed in
8796 Library code that is intended to operate with more than one UNIX
8797 standard must test, set and restore the variable
8798 __XPG4_EXTENDED_MASK as appropriate. Most GNU software doesn't
8799 provide this capability.
8802 Suppress the generation of link options to search libdld.sl when
8803 the `-static' option is specified on HP-UX 10 and later.
8806 The HP-UX implementation of setlocale in libc has a dependency on
8807 libdld.sl. There isn't an archive version of libdld.sl. Thus,
8808 when the `-static' option is specified, special link options are
8809 needed to resolve this dependency.
8811 On HP-UX 10 and later, the GCC driver adds the necessary options to
8812 link with libdld.sl when the `-static' option is specified. This
8813 causes the resulting binary to be dynamic. On the 64-bit port,
8814 the linkers generate dynamic binaries by default in any case. The
8815 `-nolibdld' option can be used to prevent the GCC driver from
8816 adding these link options.
8819 Add support for multithreading with the "dce thread" library under
8820 HP-UX. This option sets flags for both the preprocessor and
8824 File: gcc.info, Node: i386 and x86-64 Options, Next: IA-64 Options, Prev: HPPA Options, Up: Submodel Options
8826 3.17.13 Intel 386 and AMD x86-64 Options
8827 ----------------------------------------
8829 These `-m' options are defined for the i386 and x86-64 family of
8833 Tune to CPU-TYPE everything applicable about the generated code,
8834 except for the ABI and the set of available instructions. The
8835 choices for CPU-TYPE are:
8837 Original Intel's i386 CPU.
8840 Intel's i486 CPU. (No scheduling is implemented for this
8844 Intel Pentium CPU with no MMX support.
8847 Intel PentiumMMX CPU based on Pentium core with MMX
8848 instruction set support.
8851 Intel PentiumPro CPU.
8854 Intel Pentium2 CPU based on PentiumPro core with MMX
8855 instruction set support.
8857 _pentium3, pentium3m_
8858 Intel Pentium3 CPU based on PentiumPro core with MMX and SSE
8859 instruction set support.
8862 Low power version of Intel Pentium3 CPU with MMX, SSE and
8863 SSE2 instruction set support. Used by Centrino notebooks.
8865 _pentium4, pentium4m_
8866 Intel Pentium4 CPU with MMX, SSE and SSE2 instruction set
8870 Improved version of Intel Pentium4 CPU with MMX, SSE, SSE2
8871 and SSE3 instruction set support.
8874 Improved version of Intel Pentium4 CPU with 64-bit
8875 extensions, MMX, SSE, SSE2 and SSE3 instruction set support.
8878 AMD K6 CPU with MMX instruction set support.
8881 Improved versions of AMD K6 CPU with MMX and 3dNOW!
8882 instruction set support.
8884 _athlon, athlon-tbird_
8885 AMD Athlon CPU with MMX, 3dNOW!, enhanced 3dNOW! and SSE
8886 prefetch instructions support.
8888 _athlon-4, athlon-xp, athlon-mp_
8889 Improved AMD Athlon CPU with MMX, 3dNOW!, enhanced 3dNOW! and
8890 full SSE instruction set support.
8892 _k8, opteron, athlon64, athlon-fx_
8893 AMD K8 core based CPUs with x86-64 instruction set support.
8894 (This supersets MMX, SSE, SSE2, 3dNOW!, enhanced 3dNOW! and
8895 64-bit instruction set extensions.)
8898 IDT Winchip C6 CPU, dealt in same way as i486 with additional
8899 MMX instruction set support.
8902 IDT Winchip2 CPU, dealt in same way as i486 with additional
8903 MMX and 3dNOW! instruction set support.
8906 Via C3 CPU with MMX and 3dNOW! instruction set support. (No
8907 scheduling is implemented for this chip.)
8910 Via C3-2 CPU with MMX and SSE instruction set support. (No
8911 scheduling is implemented for this chip.)
8913 While picking a specific CPU-TYPE will schedule things
8914 appropriately for that particular chip, the compiler will not
8915 generate any code that does not run on the i386 without the
8916 `-march=CPU-TYPE' option being used.
8919 Generate instructions for the machine type CPU-TYPE. The choices
8920 for CPU-TYPE are the same as for `-mtune'. Moreover, specifying
8921 `-march=CPU-TYPE' implies `-mtune=CPU-TYPE'.
8924 A deprecated synonym for `-mtune'.
8930 These options are synonyms for `-mtune=i386', `-mtune=i486',
8931 `-mtune=pentium', and `-mtune=pentiumpro' respectively. These
8932 synonyms are deprecated.
8935 Generate floating point arithmetics for selected unit UNIT. The
8936 choices for UNIT are:
8939 Use the standard 387 floating point coprocessor present
8940 majority of chips and emulated otherwise. Code compiled with
8941 this option will run almost everywhere. The temporary
8942 results are computed in 80bit precision instead of precision
8943 specified by the type resulting in slightly different results
8944 compared to most of other chips. See `-ffloat-store' for
8945 more detailed description.
8947 This is the default choice for i386 compiler.
8950 Use scalar floating point instructions present in the SSE
8951 instruction set. This instruction set is supported by
8952 Pentium3 and newer chips, in the AMD line by Athlon-4,
8953 Athlon-xp and Athlon-mp chips. The earlier version of SSE
8954 instruction set supports only single precision arithmetics,
8955 thus the double and extended precision arithmetics is still
8956 done using 387. Later version, present only in Pentium4 and
8957 the future AMD x86-64 chips supports double precision
8960 For the i386 compiler, you need to use `-march=CPU-TYPE',
8961 `-msse' or `-msse2' switches to enable SSE extensions and
8962 make this option effective. For the x86-64 compiler, these
8963 extensions are enabled by default.
8965 The resulting code should be considerably faster in the
8966 majority of cases and avoid the numerical instability
8967 problems of 387 code, but may break some existing code that
8968 expects temporaries to be 80bit.
8970 This is the default choice for the x86-64 compiler.
8973 Attempt to utilize both instruction sets at once. This
8974 effectively double the amount of available registers and on
8975 chips with separate execution units for 387 and SSE the
8976 execution resources too. Use this option with care, as it is
8977 still experimental, because the GCC register allocator does
8978 not model separate functional units well resulting in
8979 instable performance.
8982 Output asm instructions using selected DIALECT. Supported choices
8983 are `intel' or `att' (the default one). Darwin does not support
8988 Control whether or not the compiler uses IEEE floating point
8989 comparisons. These handle correctly the case where the result of a
8990 comparison is unordered.
8993 Generate output containing library calls for floating point.
8994 *Warning:* the requisite libraries are not part of GCC. Normally
8995 the facilities of the machine's usual C compiler are used, but
8996 this can't be done directly in cross-compilation. You must make
8997 your own arrangements to provide suitable library functions for
9000 On machines where a function returns floating point results in the
9001 80387 register stack, some floating point opcodes may be emitted
9002 even if `-msoft-float' is used.
9004 `-mno-fp-ret-in-387'
9005 Do not use the FPU registers for return values of functions.
9007 The usual calling convention has functions return values of types
9008 `float' and `double' in an FPU register, even if there is no FPU.
9009 The idea is that the operating system should emulate an FPU.
9011 The option `-mno-fp-ret-in-387' causes such values to be returned
9012 in ordinary CPU registers instead.
9014 `-mno-fancy-math-387'
9015 Some 387 emulators do not support the `sin', `cos' and `sqrt'
9016 instructions for the 387. Specify this option to avoid generating
9017 those instructions. This option is the default on FreeBSD,
9018 OpenBSD and NetBSD. This option is overridden when `-march'
9019 indicates that the target cpu will always have an FPU and so the
9020 instruction will not need emulation. As of revision 2.6.1, these
9021 instructions are not generated unless you also use the
9022 `-funsafe-math-optimizations' switch.
9026 Control whether GCC aligns `double', `long double', and `long
9027 long' variables on a two word boundary or a one word boundary.
9028 Aligning `double' variables on a two word boundary will produce
9029 code that runs somewhat faster on a `Pentium' at the expense of
9032 On x86-64, `-malign-double' is enabled by default.
9034 *Warning:* if you use the `-malign-double' switch, structures
9035 containing the above types will be aligned differently than the
9036 published application binary interface specifications for the 386
9037 and will not be binary compatible with structures in code compiled
9038 without that switch.
9040 `-m96bit-long-double'
9041 `-m128bit-long-double'
9042 These switches control the size of `long double' type. The i386
9043 application binary interface specifies the size to be 96 bits, so
9044 `-m96bit-long-double' is the default in 32 bit mode.
9046 Modern architectures (Pentium and newer) would prefer `long double'
9047 to be aligned to an 8 or 16 byte boundary. In arrays or structures
9048 conforming to the ABI, this would not be possible. So specifying a
9049 `-m128bit-long-double' will align `long double' to a 16 byte
9050 boundary by padding the `long double' with an additional 32 bit
9053 In the x86-64 compiler, `-m128bit-long-double' is the default
9054 choice as its ABI specifies that `long double' is to be aligned on
9057 Notice that neither of these options enable any extra precision
9058 over the x87 standard of 80 bits for a `long double'.
9060 *Warning:* if you override the default value for your target ABI,
9061 the structures and arrays containing `long double' variables will
9062 change their size as well as function calling convention for
9063 function taking `long double' will be modified. Hence they will
9064 not be binary compatible with arrays or structures in code
9065 compiled without that switch.
9067 `-mmlarge-data-threshold=NUMBER'
9068 When `-mcmodel=medium' is specified, the data greater than
9069 THRESHOLD are placed in large data section. This value must be the
9070 same across all object linked into the binary and defaults to
9075 Control whether GCC places uninitialized local variables into the
9076 `bss' or `data' segments. `-msvr3-shlib' places them into `bss'.
9077 These options are meaningful only on System V Release 3.
9080 Use a different function-calling convention, in which functions
9081 that take a fixed number of arguments return with the `ret' NUM
9082 instruction, which pops their arguments while returning. This
9083 saves one instruction in the caller since there is no need to pop
9084 the arguments there.
9086 You can specify that an individual function is called with this
9087 calling sequence with the function attribute `stdcall'. You can
9088 also override the `-mrtd' option by using the function attribute
9089 `cdecl'. *Note Function Attributes::.
9091 *Warning:* this calling convention is incompatible with the one
9092 normally used on Unix, so you cannot use it if you need to call
9093 libraries compiled with the Unix compiler.
9095 Also, you must provide function prototypes for all functions that
9096 take variable numbers of arguments (including `printf'); otherwise
9097 incorrect code will be generated for calls to those functions.
9099 In addition, seriously incorrect code will result if you call a
9100 function with too many arguments. (Normally, extra arguments are
9101 harmlessly ignored.)
9104 Control how many registers are used to pass integer arguments. By
9105 default, no registers are used to pass arguments, and at most 3
9106 registers can be used. You can control this behavior for a
9107 specific function by using the function attribute `regparm'.
9108 *Note Function Attributes::.
9110 *Warning:* if you use this switch, and NUM is nonzero, then you
9111 must build all modules with the same value, including any
9112 libraries. This includes the system libraries and startup modules.
9115 Use SSE register passing conventions for float and double arguments
9116 and return values. You can control this behavior for a specific
9117 function by using the function attribute `sseregparm'. *Note
9118 Function Attributes::.
9120 *Warning:* if you use this switch then you must build all modules
9121 with the same value, including any libraries. This includes the
9122 system libraries and startup modules.
9124 `-mpreferred-stack-boundary=NUM'
9125 Attempt to keep the stack boundary aligned to a 2 raised to NUM
9126 byte boundary. If `-mpreferred-stack-boundary' is not specified,
9127 the default is 4 (16 bytes or 128 bits).
9129 On Pentium and PentiumPro, `double' and `long double' values
9130 should be aligned to an 8 byte boundary (see `-malign-double') or
9131 suffer significant run time performance penalties. On Pentium
9132 III, the Streaming SIMD Extension (SSE) data type `__m128' may not
9133 work properly if it is not 16 byte aligned.
9135 To ensure proper alignment of this values on the stack, the stack
9136 boundary must be as aligned as that required by any value stored
9137 on the stack. Further, every function must be generated such that
9138 it keeps the stack aligned. Thus calling a function compiled with
9139 a higher preferred stack boundary from a function compiled with a
9140 lower preferred stack boundary will most likely misalign the
9141 stack. It is recommended that libraries that use callbacks always
9142 use the default setting.
9144 This extra alignment does consume extra stack space, and generally
9145 increases code size. Code that is sensitive to stack space usage,
9146 such as embedded systems and operating system kernels, may want to
9147 reduce the preferred alignment to `-mpreferred-stack-boundary=2'.
9163 These switches enable or disable the use of instructions in the
9164 MMX, SSE, SSE2 or 3DNow! extended instruction sets. These
9165 extensions are also available as built-in functions: see *Note X86
9166 Built-in Functions::, for details of the functions enabled and
9167 disabled by these switches.
9169 To have SSE/SSE2 instructions generated automatically from
9170 floating-point code (as opposed to 387 instructions), see
9173 These options will enable GCC to use these extended instructions in
9174 generated code, even without `-mfpmath=sse'. Applications which
9175 perform runtime CPU detection must compile separate files for each
9176 supported architecture, using the appropriate flags. In
9177 particular, the file containing the CPU detection code should be
9178 compiled without these options.
9182 Use PUSH operations to store outgoing parameters. This method is
9183 shorter and usually equally fast as method using SUB/MOV
9184 operations and is enabled by default. In some cases disabling it
9185 may improve performance because of improved scheduling and reduced
9188 `-maccumulate-outgoing-args'
9189 If enabled, the maximum amount of space required for outgoing
9190 arguments will be computed in the function prologue. This is
9191 faster on most modern CPUs because of reduced dependencies,
9192 improved scheduling and reduced stack usage when preferred stack
9193 boundary is not equal to 2. The drawback is a notable increase in
9194 code size. This switch implies `-mno-push-args'.
9197 Support thread-safe exception handling on `Mingw32'. Code that
9198 relies on thread-safe exception handling must compile and link all
9199 code with the `-mthreads' option. When compiling, `-mthreads'
9200 defines `-D_MT'; when linking, it links in a special thread helper
9201 library `-lmingwthrd' which cleans up per thread exception
9204 `-mno-align-stringops'
9205 Do not align destination of inlined string operations. This
9206 switch reduces code size and improves performance in case the
9207 destination is already aligned, but GCC doesn't know about it.
9209 `-minline-all-stringops'
9210 By default GCC inlines string operations only when destination is
9211 known to be aligned at least to 4 byte boundary. This enables
9212 more inlining, increase code size, but may improve performance of
9213 code that depends on fast memcpy, strlen and memset for short
9216 `-momit-leaf-frame-pointer'
9217 Don't keep the frame pointer in a register for leaf functions.
9218 This avoids the instructions to save, set up and restore frame
9219 pointers and makes an extra register available in leaf functions.
9220 The option `-fomit-frame-pointer' removes the frame pointer for
9221 all functions which might make debugging harder.
9223 `-mtls-direct-seg-refs'
9224 `-mno-tls-direct-seg-refs'
9225 Controls whether TLS variables may be accessed with offsets from
9226 the TLS segment register (`%gs' for 32-bit, `%fs' for 64-bit), or
9227 whether the thread base pointer must be added. Whether or not this
9228 is legal depends on the operating system, and whether it maps the
9229 segment to cover the entire TLS area.
9231 For systems that use GNU libc, the default is on.
9233 These `-m' switches are supported in addition to the above on AMD
9234 x86-64 processors in 64-bit environments.
9238 Generate code for a 32-bit or 64-bit environment. The 32-bit
9239 environment sets int, long and pointer to 32 bits and generates
9240 code that runs on any i386 system. The 64-bit environment sets
9241 int to 32 bits and long and pointer to 64 bits and generates code
9242 for AMD's x86-64 architecture.
9245 Do not use a so called red zone for x86-64 code. The red zone is
9246 mandated by the x86-64 ABI, it is a 128-byte area beyond the
9247 location of the stack pointer that will not be modified by signal
9248 or interrupt handlers and therefore can be used for temporary data
9249 without adjusting the stack pointer. The flag `-mno-red-zone'
9250 disables this red zone.
9253 Generate code for the small code model: the program and its
9254 symbols must be linked in the lower 2 GB of the address space.
9255 Pointers are 64 bits. Programs can be statically or dynamically
9256 linked. This is the default code model.
9259 Generate code for the kernel code model. The kernel runs in the
9260 negative 2 GB of the address space. This model has to be used for
9264 Generate code for the medium model: The program is linked in the
9265 lower 2 GB of the address space but symbols can be located
9266 anywhere in the address space. Programs can be statically or
9267 dynamically linked, but building of shared libraries are not
9268 supported with the medium model.
9271 Generate code for the large model: This model makes no assumptions
9272 about addresses and sizes of sections. Currently GCC does not
9273 implement this model.
9276 File: gcc.info, Node: IA-64 Options, Next: M32C Options, Prev: i386 and x86-64 Options, Up: Submodel Options
9278 3.17.14 IA-64 Options
9279 ---------------------
9281 These are the `-m' options defined for the Intel IA-64 architecture.
9284 Generate code for a big endian target. This is the default for
9288 Generate code for a little endian target. This is the default for
9293 Generate (or don't) code for the GNU assembler. This is the
9298 Generate (or don't) code for the GNU linker. This is the default.
9301 Generate code that does not use a global pointer register. The
9302 result is not position independent code, and violates the IA-64
9305 `-mvolatile-asm-stop'
9306 `-mno-volatile-asm-stop'
9307 Generate (or don't) a stop bit immediately before and after
9308 volatile asm statements.
9311 `-mno-register-names'
9312 Generate (or don't) `in', `loc', and `out' register names for the
9313 stacked registers. This may make assembler output more readable.
9317 Disable (or enable) optimizations that use the small data section.
9318 This may be useful for working around optimizer bugs.
9321 Generate code that uses a single constant global pointer value.
9322 This is useful when compiling kernel code.
9325 Generate code that is self-relocatable. This implies
9326 `-mconstant-gp'. This is useful when compiling firmware code.
9328 `-minline-float-divide-min-latency'
9329 Generate code for inline divides of floating point values using
9330 the minimum latency algorithm.
9332 `-minline-float-divide-max-throughput'
9333 Generate code for inline divides of floating point values using
9334 the maximum throughput algorithm.
9336 `-minline-int-divide-min-latency'
9337 Generate code for inline divides of integer values using the
9338 minimum latency algorithm.
9340 `-minline-int-divide-max-throughput'
9341 Generate code for inline divides of integer values using the
9342 maximum throughput algorithm.
9344 `-minline-sqrt-min-latency'
9345 Generate code for inline square roots using the minimum latency
9348 `-minline-sqrt-max-throughput'
9349 Generate code for inline square roots using the maximum throughput
9354 Don't (or do) generate assembler code for the DWARF2 line number
9355 debugging info. This may be useful when not using the GNU
9359 `-mno-early-stop-bits'
9360 Allow stop bits to be placed earlier than immediately preceding the
9361 instruction that triggered the stop bit. This can improve
9362 instruction scheduling, but does not always do so.
9364 `-mfixed-range=REGISTER-RANGE'
9365 Generate code treating the given register range as fixed registers.
9366 A fixed register is one that the register allocator can not use.
9367 This is useful when compiling kernel code. A register range is
9368 specified as two registers separated by a dash. Multiple register
9369 ranges can be specified separated by a comma.
9371 `-mtls-size=TLS-SIZE'
9372 Specify bit size of immediate TLS offsets. Valid values are 14,
9376 Tune the instruction scheduling for a particular CPU, Valid values
9377 are itanium, itanium1, merced, itanium2, and mckinley.
9381 Add support for multithreading using the POSIX threads library.
9382 This option sets flags for both the preprocessor and linker. It
9383 does not affect the thread safety of object code produced by the
9384 compiler or that of libraries supplied with it. These are HP-UX
9389 Generate code for a 32-bit or 64-bit environment. The 32-bit
9390 environment sets int, long and pointer to 32 bits. The 64-bit
9391 environment sets int to 32 bits and long and pointer to 64 bits.
9392 These are HP-UX specific flags.
9396 File: gcc.info, Node: M32C Options, Next: M32R/D Options, Prev: IA-64 Options, Up: Submodel Options
9398 3.17.15 M32C Options
9399 --------------------
9402 Select the CPU for which code is generated. NAME may be one of
9403 `r8c' for the R8C/Tiny series, `m16c' for the M16C (up to /60)
9404 series, `m32cm' for the M16C/80 series, or `m32c' for the M32C/80
9408 Specifies that the program will be run on the simulator. This
9409 causes an alternate runtime library to be linked in which
9410 supports, for example, file I/O. You must not use this option
9411 when generating programs that will run on real hardware; you must
9412 provide your own runtime library for whatever I/O functions are
9416 Specifies the number of memory-based pseudo-registers GCC will use
9417 during code generation. These pseudo-registers will be used like
9418 real registers, so there is a tradeoff between GCC's ability to
9419 fit the code into available registers, and the performance penalty
9420 of using memory instead of registers. Note that all modules in a
9421 program must be compiled with the same value for this option.
9422 Because of that, you must not use this option with the default
9423 runtime libraries gcc builds.
9427 File: gcc.info, Node: M32R/D Options, Next: M680x0 Options, Prev: M32C Options, Up: Submodel Options
9429 3.17.16 M32R/D Options
9430 ----------------------
9432 These `-m' options are defined for Renesas M32R/D architectures:
9435 Generate code for the M32R/2.
9438 Generate code for the M32R/X.
9441 Generate code for the M32R. This is the default.
9444 Assume all objects live in the lower 16MB of memory (so that their
9445 addresses can be loaded with the `ld24' instruction), and assume
9446 all subroutines are reachable with the `bl' instruction. This is
9449 The addressability of a particular object can be set with the
9453 Assume objects may be anywhere in the 32-bit address space (the
9454 compiler will generate `seth/add3' instructions to load their
9455 addresses), and assume all subroutines are reachable with the `bl'
9459 Assume objects may be anywhere in the 32-bit address space (the
9460 compiler will generate `seth/add3' instructions to load their
9461 addresses), and assume subroutines may not be reachable with the
9462 `bl' instruction (the compiler will generate the much slower
9463 `seth/add3/jl' instruction sequence).
9466 Disable use of the small data area. Variables will be put into
9467 one of `.data', `bss', or `.rodata' (unless the `section'
9468 attribute has been specified). This is the default.
9470 The small data area consists of sections `.sdata' and `.sbss'.
9471 Objects may be explicitly put in the small data area with the
9472 `section' attribute using one of these sections.
9475 Put small global and static data in the small data area, but do not
9476 generate special code to reference them.
9479 Put small global and static data in the small data area, and
9480 generate special instructions to reference them.
9483 Put global and static objects less than or equal to NUM bytes into
9484 the small data or bss sections instead of the normal data or bss
9485 sections. The default value of NUM is 8. The `-msdata' option
9486 must be set to one of `sdata' or `use' for this option to have any
9489 All modules should be compiled with the same `-G NUM' value.
9490 Compiling with different values of NUM may or may not work; if it
9491 doesn't the linker will give an error message--incorrect code will
9495 Makes the M32R specific code in the compiler display some
9496 statistics that might help in debugging programs.
9499 Align all loops to a 32-byte boundary.
9502 Do not enforce a 32-byte alignment for loops. This is the default.
9504 `-missue-rate=NUMBER'
9505 Issue NUMBER instructions per cycle. NUMBER can only be 1 or 2.
9507 `-mbranch-cost=NUMBER'
9508 NUMBER can only be 1 or 2. If it is 1 then branches will be
9509 preferred over conditional code, if it is 2, then the opposite will
9512 `-mflush-trap=NUMBER'
9513 Specifies the trap number to use to flush the cache. The default
9514 is 12. Valid numbers are between 0 and 15 inclusive.
9517 Specifies that the cache cannot be flushed by using a trap.
9520 Specifies the name of the operating system function to call to
9521 flush the cache. The default is __flush_cache_, but a function
9522 call will only be used if a trap is not available.
9525 Indicates that there is no OS function for flushing the cache.
9529 File: gcc.info, Node: M680x0 Options, Next: M68hc1x Options, Prev: M32R/D Options, Up: Submodel Options
9531 3.17.17 M680x0 Options
9532 ----------------------
9534 These are the `-m' options defined for the 68000 series. The default
9535 values for these options depends on which style of 68000 was selected
9536 when the compiler was configured; the defaults for the most common
9537 choices are given below.
9541 Generate output for a 68000. This is the default when the
9542 compiler is configured for 68000-based systems.
9544 Use this option for microcontrollers with a 68000 or EC000 core,
9545 including the 68008, 68302, 68306, 68307, 68322, 68328 and 68356.
9549 Generate output for a 68020. This is the default when the
9550 compiler is configured for 68020-based systems.
9553 Generate output containing 68881 instructions for floating point.
9554 This is the default for most 68020 systems unless `--nfp' was
9555 specified when the compiler was configured.
9558 Generate output for a 68030. This is the default when the
9559 compiler is configured for 68030-based systems.
9562 Generate output for a 68040. This is the default when the
9563 compiler is configured for 68040-based systems.
9565 This option inhibits the use of 68881/68882 instructions that have
9566 to be emulated by software on the 68040. Use this option if your
9567 68040 does not have code to emulate those instructions.
9570 Generate output for a 68060. This is the default when the
9571 compiler is configured for 68060-based systems.
9573 This option inhibits the use of 68020 and 68881/68882 instructions
9574 that have to be emulated by software on the 68060. Use this
9575 option if your 68060 does not have code to emulate those
9579 Generate output for a CPU32. This is the default when the
9580 compiler is configured for CPU32-based systems.
9582 Use this option for microcontrollers with a CPU32 or CPU32+ core,
9583 including the 68330, 68331, 68332, 68333, 68334, 68336, 68340,
9584 68341, 68349 and 68360.
9587 Generate output for a 520X "coldfire" family cpu. This is the
9588 default when the compiler is configured for 520X-based systems.
9590 Use this option for microcontroller with a 5200 core, including
9591 the MCF5202, MCF5203, MCF5204 and MCF5202.
9594 Generate output for a 68040, without using any of the new
9595 instructions. This results in code which can run relatively
9596 efficiently on either a 68020/68881 or a 68030 or a 68040. The
9597 generated code does use the 68881 instructions that are emulated
9601 Generate output for a 68060, without using any of the new
9602 instructions. This results in code which can run relatively
9603 efficiently on either a 68020/68881 or a 68030 or a 68040. The
9604 generated code does use the 68881 instructions that are emulated
9608 Generate output containing library calls for floating point.
9609 *Warning:* the requisite libraries are not available for all m68k
9610 targets. Normally the facilities of the machine's usual C
9611 compiler are used, but this can't be done directly in
9612 cross-compilation. You must make your own arrangements to provide
9613 suitable library functions for cross-compilation. The embedded
9614 targets `m68k-*-aout' and `m68k-*-coff' do provide software
9615 floating point support.
9618 Consider type `int' to be 16 bits wide, like `short int'.
9619 Additionally, parameters passed on the stack are also aligned to a
9620 16-bit boundary even on targets whose API mandates promotion to
9624 Do not use the bit-field instructions. The `-m68000', `-mcpu32'
9625 and `-m5200' options imply `-mnobitfield'.
9628 Do use the bit-field instructions. The `-m68020' option implies
9629 `-mbitfield'. This is the default if you use a configuration
9630 designed for a 68020.
9633 Use a different function-calling convention, in which functions
9634 that take a fixed number of arguments return with the `rtd'
9635 instruction, which pops their arguments while returning. This
9636 saves one instruction in the caller since there is no need to pop
9637 the arguments there.
9639 This calling convention is incompatible with the one normally used
9640 on Unix, so you cannot use it if you need to call libraries
9641 compiled with the Unix compiler.
9643 Also, you must provide function prototypes for all functions that
9644 take variable numbers of arguments (including `printf'); otherwise
9645 incorrect code will be generated for calls to those functions.
9647 In addition, seriously incorrect code will result if you call a
9648 function with too many arguments. (Normally, extra arguments are
9649 harmlessly ignored.)
9651 The `rtd' instruction is supported by the 68010, 68020, 68030,
9652 68040, 68060 and CPU32 processors, but not by the 68000 or 5200.
9656 Control whether GCC aligns `int', `long', `long long', `float',
9657 `double', and `long double' variables on a 32-bit boundary
9658 (`-malign-int') or a 16-bit boundary (`-mno-align-int'). Aligning
9659 variables on 32-bit boundaries produces code that runs somewhat
9660 faster on processors with 32-bit busses at the expense of more
9663 *Warning:* if you use the `-malign-int' switch, GCC will align
9664 structures containing the above types differently than most
9665 published application binary interface specifications for the m68k.
9668 Use the pc-relative addressing mode of the 68000 directly, instead
9669 of using a global offset table. At present, this option implies
9670 `-fpic', allowing at most a 16-bit offset for pc-relative
9671 addressing. `-fPIC' is not presently supported with `-mpcrel',
9672 though this could be supported for 68020 and higher processors.
9676 Do not (do) assume that unaligned memory references will be
9677 handled by the system.
9680 Generate code that allows the data segment to be located in a
9681 different area of memory from the text segment. This allows for
9682 execute in place in an environment without virtual memory
9683 management. This option implies `-fPIC'.
9686 Generate code that assumes that the data segment follows the text
9687 segment. This is the default.
9689 `-mid-shared-library'
9690 Generate code that supports shared libraries via the library ID
9691 method. This allows for execute in place and shared libraries in
9692 an environment without virtual memory management. This option
9695 `-mno-id-shared-library'
9696 Generate code that doesn't assume ID based shared libraries are
9697 being used. This is the default.
9699 `-mshared-library-id=n'
9700 Specified the identification number of the ID based shared library
9701 being compiled. Specifying a value of 0 will generate more
9702 compact code, specifying other values will force the allocation of
9703 that number to the current library but is no more space or time
9704 efficient than omitting this option.
9708 File: gcc.info, Node: M68hc1x Options, Next: MCore Options, Prev: M680x0 Options, Up: Submodel Options
9710 3.17.18 M68hc1x Options
9711 -----------------------
9713 These are the `-m' options defined for the 68hc11 and 68hc12
9714 microcontrollers. The default values for these options depends on
9715 which style of microcontroller was selected when the compiler was
9716 configured; the defaults for the most common choices are given below.
9720 Generate output for a 68HC11. This is the default when the
9721 compiler is configured for 68HC11-based systems.
9725 Generate output for a 68HC12. This is the default when the
9726 compiler is configured for 68HC12-based systems.
9730 Generate output for a 68HCS12.
9733 Enable the use of 68HC12 pre and post auto-increment and
9734 auto-decrement addressing modes.
9738 Enable the use of 68HC12 min and max instructions.
9742 Treat all calls as being far away (near). If calls are assumed to
9743 be far away, the compiler will use the `call' instruction to call
9744 a function and the `rtc' instruction for returning.
9747 Consider type `int' to be 16 bits wide, like `short int'.
9749 `-msoft-reg-count=COUNT'
9750 Specify the number of pseudo-soft registers which are used for the
9751 code generation. The maximum number is 32. Using more pseudo-soft
9752 register may or may not result in better code depending on the
9753 program. The default is 4 for 68HC11 and 2 for 68HC12.
9757 File: gcc.info, Node: MCore Options, Next: MIPS Options, Prev: M68hc1x Options, Up: Submodel Options
9759 3.17.19 MCore Options
9760 ---------------------
9762 These are the `-m' options defined for the Motorola M*Core processors.
9766 Inline constants into the code stream if it can be done in two
9767 instructions or less.
9771 Use the divide instruction. (Enabled by default).
9774 `-mno-relax-immediate'
9775 Allow arbitrary sized immediates in bit operations.
9778 `-mno-wide-bitfields'
9779 Always treat bit-fields as int-sized.
9782 `-mno-4byte-functions'
9783 Force all functions to be aligned to a four byte boundary.
9786 `-mno-callgraph-data'
9787 Emit callgraph information.
9791 Prefer word access when reading byte quantities.
9795 Generate code for a little endian target.
9799 Generate code for the 210 processor.
9802 File: gcc.info, Node: MIPS Options, Next: MMIX Options, Prev: MCore Options, Up: Submodel Options
9804 3.17.20 MIPS Options
9805 --------------------
9808 Generate big-endian code.
9811 Generate little-endian code. This is the default for `mips*el-*-*'
9815 Generate code that will run on ARCH, which can be the name of a
9816 generic MIPS ISA, or the name of a particular processor. The ISA
9817 names are: `mips1', `mips2', `mips3', `mips4', `mips32',
9818 `mips32r2', and `mips64'. The processor names are: `4kc', `4km',
9819 `4kp', `5kc', `5kf', `20kc', `24k', `24kc', `24kf', `24kx', `m4k',
9820 `orion', `r2000', `r3000', `r3900', `r4000', `r4400', `r4600',
9821 `r4650', `r6000', `r8000', `rm7000', `rm9000', `sb1', `sr71000',
9822 `vr4100', `vr4111', `vr4120', `vr4130', `vr4300', `vr5000',
9823 `vr5400' and `vr5500'. The special value `from-abi' selects the
9824 most compatible architecture for the selected ABI (that is,
9825 `mips1' for 32-bit ABIs and `mips3' for 64-bit ABIs).
9827 In processor names, a final `000' can be abbreviated as `k' (for
9828 example, `-march=r2k'). Prefixes are optional, and `vr' may be
9831 GCC defines two macros based on the value of this option. The
9832 first is `_MIPS_ARCH', which gives the name of target
9833 architecture, as a string. The second has the form
9834 `_MIPS_ARCH_FOO', where FOO is the capitalized value of
9835 `_MIPS_ARCH'. For example, `-march=r2000' will set `_MIPS_ARCH'
9836 to `"r2000"' and define the macro `_MIPS_ARCH_R2000'.
9838 Note that the `_MIPS_ARCH' macro uses the processor names given
9839 above. In other words, it will have the full prefix and will not
9840 abbreviate `000' as `k'. In the case of `from-abi', the macro
9841 names the resolved architecture (either `"mips1"' or `"mips3"').
9842 It names the default architecture when no `-march' option is given.
9845 Optimize for ARCH. Among other things, this option controls the
9846 way instructions are scheduled, and the perceived cost of
9847 arithmetic operations. The list of ARCH values is the same as for
9850 When this option is not used, GCC will optimize for the processor
9851 specified by `-march'. By using `-march' and `-mtune' together,
9852 it is possible to generate code that will run on a family of
9853 processors, but optimize the code for one particular member of
9856 `-mtune' defines the macros `_MIPS_TUNE' and `_MIPS_TUNE_FOO',
9857 which work in the same way as the `-march' ones described above.
9860 Equivalent to `-march=mips1'.
9863 Equivalent to `-march=mips2'.
9866 Equivalent to `-march=mips3'.
9869 Equivalent to `-march=mips4'.
9872 Equivalent to `-march=mips32'.
9875 Equivalent to `-march=mips32r2'.
9878 Equivalent to `-march=mips64'.
9882 Generate (do not generate) MIPS16 code. If GCC is targetting a
9883 MIPS32 or MIPS64 architecture, it will make use of the MIPS16e ASE.
9890 Generate code for the given ABI.
9892 Note that the EABI has a 32-bit and a 64-bit variant. GCC normally
9893 generates 64-bit code when you select a 64-bit architecture, but
9894 you can use `-mgp32' to get 32-bit code instead.
9896 For information about the O64 ABI, see
9897 `http://gcc.gnu.org/projects/mipso64-abi.html'.
9901 Generate (do not generate) SVR4-style position-independent code.
9902 `-mabicalls' is the default for SVR4-based systems.
9906 Lift (do not lift) the usual restrictions on the size of the global
9909 GCC normally uses a single instruction to load values from the GOT.
9910 While this is relatively efficient, it will only work if the GOT
9911 is smaller than about 64k. Anything larger will cause the linker
9912 to report an error such as:
9914 relocation truncated to fit: R_MIPS_GOT16 foobar
9916 If this happens, you should recompile your code with `-mxgot'. It
9917 should then work with very large GOTs, although it will also be
9918 less efficient, since it will take three instructions to fetch the
9919 value of a global symbol.
9921 Note that some linkers can create multiple GOTs. If you have such
9922 a linker, you should only need to use `-mxgot' when a single object
9923 file accesses more than 64k's worth of GOT entries. Very few do.
9925 These options have no effect unless GCC is generating position
9929 Assume that general-purpose registers are 32 bits wide.
9932 Assume that general-purpose registers are 64 bits wide.
9935 Assume that floating-point registers are 32 bits wide.
9938 Assume that floating-point registers are 64 bits wide.
9941 Use floating-point coprocessor instructions.
9944 Do not use floating-point coprocessor instructions. Implement
9945 floating-point calculations using library calls instead.
9948 Assume that the floating-point coprocessor only supports
9949 single-precision operations.
9952 Assume that the floating-point coprocessor supports
9953 double-precision operations. This is the default.
9957 Use (do not use) the MIPS DSP ASE. *Note MIPS DSP Built-in
9961 `-mno-paired-single'
9962 Use (do not use) paired-single floating-point instructions. *Note
9963 MIPS Paired-Single Support::. This option can only be used when
9964 generating 64-bit code and requires hardware floating-point
9965 support to be enabled.
9969 Use (do not use) the MIPS-3D ASE. *Note MIPS-3D Built-in
9970 Functions::. The option `-mips3d' implies `-mpaired-single'.
9973 Force `long' types to be 64 bits wide. See `-mlong32' for an
9974 explanation of the default and the way that the pointer size is
9978 Force `long', `int', and pointer types to be 32 bits wide.
9980 The default size of `int's, `long's and pointers depends on the
9981 ABI. All the supported ABIs use 32-bit `int's. The n64 ABI uses
9982 64-bit `long's, as does the 64-bit EABI; the others use 32-bit
9983 `long's. Pointers are the same size as `long's, or the same size
9984 as integer registers, whichever is smaller.
9988 Assume (do not assume) that all symbols have 32-bit values,
9989 regardless of the selected ABI. This option is useful in
9990 combination with `-mabi=64' and `-mno-abicalls' because it allows
9991 GCC to generate shorter and faster references to symbolic
9995 Put global and static items less than or equal to NUM bytes into
9996 the small data or bss section instead of the normal data or bss
9997 section. This allows the data to be accessed using a single
10000 All modules should be compiled with the same `-G NUM' value.
10003 `-mno-embedded-data'
10004 Allocate variables to the read-only data section first if
10005 possible, then next in the small data section if possible,
10006 otherwise in data. This gives slightly slower code than the
10007 default, but reduces the amount of RAM required when executing,
10008 and thus may be preferred for some embedded systems.
10010 `-muninit-const-in-rodata'
10011 `-mno-uninit-const-in-rodata'
10012 Put uninitialized `const' variables in the read-only data section.
10013 This option is only meaningful in conjunction with
10016 `-msplit-addresses'
10017 `-mno-split-addresses'
10018 Enable (disable) use of the `%hi()' and `%lo()' assembler
10019 relocation operators. This option has been superseded by
10020 `-mexplicit-relocs' but is retained for backwards compatibility.
10022 `-mexplicit-relocs'
10023 `-mno-explicit-relocs'
10024 Use (do not use) assembler relocation operators when dealing with
10025 symbolic addresses. The alternative, selected by
10026 `-mno-explicit-relocs', is to use assembler macros instead.
10028 `-mexplicit-relocs' is the default if GCC was configured to use an
10029 assembler that supports relocation operators.
10031 `-mcheck-zero-division'
10032 `-mno-check-zero-division'
10033 Trap (do not trap) on integer division by zero. The default is
10034 `-mcheck-zero-division'.
10038 MIPS systems check for division by zero by generating either a
10039 conditional trap or a break instruction. Using traps results in
10040 smaller code, but is only supported on MIPS II and later. Also,
10041 some versions of the Linux kernel have a bug that prevents trap
10042 from generating the proper signal (`SIGFPE'). Use
10043 `-mdivide-traps' to allow conditional traps on architectures that
10044 support them and `-mdivide-breaks' to force the use of breaks.
10046 The default is usually `-mdivide-traps', but this can be
10047 overridden at configure time using `--with-divide=breaks'.
10048 Divide-by-zero checks can be completely disabled using
10049 `-mno-check-zero-division'.
10053 Force (do not force) the use of `memcpy()' for non-trivial block
10054 moves. The default is `-mno-memcpy', which allows GCC to inline
10055 most constant-sized copies.
10059 Disable (do not disable) use of the `jal' instruction. Calling
10060 functions using `jal' is more efficient but requires the caller
10061 and callee to be in the same 256 megabyte segment.
10063 This option has no effect on abicalls code. The default is
10068 Enable (disable) use of the `mad', `madu' and `mul' instructions,
10069 as provided by the R4650 ISA.
10073 Enable (disable) use of the floating point multiply-accumulate
10074 instructions, when they are available. The default is
10077 When multiply-accumulate instructions are used, the intermediate
10078 product is calculated to infinite precision and is not subject to
10079 the FCSR Flush to Zero bit. This may be undesirable in some
10083 Tell the MIPS assembler to not run its preprocessor over user
10084 assembler files (with a `.s' suffix) when assembling them.
10088 Work around certain R4000 CPU errata:
10089 - A double-word or a variable shift may give an incorrect
10090 result if executed immediately after starting an integer
10093 - A double-word or a variable shift may give an incorrect
10094 result if executed while an integer multiplication is in
10097 - An integer division may give an incorrect result if started
10098 in a delay slot of a taken branch or a jump.
10102 Work around certain R4400 CPU errata:
10103 - A double-word or a variable shift may give an incorrect
10104 result if executed immediately after starting an integer
10109 Work around certain VR4120 errata:
10110 - `dmultu' does not always produce the correct result.
10112 - `div' and `ddiv' do not always produce the correct result if
10113 one of the operands is negative.
10114 The workarounds for the division errata rely on special functions
10115 in `libgcc.a'. At present, these functions are only provided by
10116 the `mips64vr*-elf' configurations.
10118 Other VR4120 errata require a nop to be inserted between certain
10119 pairs of instructions. These errata are handled by the assembler,
10123 Work around the VR4130 `mflo'/`mfhi' errata. The workarounds are
10124 implemented by the assembler rather than by GCC, although GCC will
10125 avoid using `mflo' and `mfhi' if the VR4130 `macc', `macchi',
10126 `dmacc' and `dmacchi' instructions are available instead.
10130 Work around certain SB-1 CPU core errata. (This flag currently
10131 works around the SB-1 revision 2 "F1" and "F2" floating point
10134 `-mflush-func=FUNC'
10136 Specifies the function to call to flush the I and D caches, or to
10137 not call any such function. If called, the function must take the
10138 same arguments as the common `_flush_func()', that is, the address
10139 of the memory range for which the cache is being flushed, the size
10140 of the memory range, and the number 3 (to flush both caches). The
10141 default depends on the target GCC was configured for, but commonly
10142 is either `_flush_func' or `__cpu_flush'.
10145 `-mno-branch-likely'
10146 Enable or disable use of Branch Likely instructions, regardless of
10147 the default for the selected architecture. By default, Branch
10148 Likely instructions may be generated if they are supported by the
10149 selected architecture. An exception is for the MIPS32 and MIPS64
10150 architectures and processors which implement those architectures;
10151 for those, Branch Likely instructions will not be generated by
10152 default because the MIPS32 and MIPS64 architectures specifically
10153 deprecate their use.
10156 `-mno-fp-exceptions'
10157 Specifies whether FP exceptions are enabled. This affects how we
10158 schedule FP instructions for some processors. The default is that
10159 FP exceptions are enabled.
10161 For instance, on the SB-1, if FP exceptions are disabled, and we
10162 are emitting 64-bit code, then we can use both FP pipes.
10163 Otherwise, we can only use one FP pipe.
10166 `-mno-vr4130-align'
10167 The VR4130 pipeline is two-way superscalar, but can only issue two
10168 instructions together if the first one is 8-byte aligned. When
10169 this option is enabled, GCC will align pairs of instructions that
10170 it thinks should execute in parallel.
10172 This option only has an effect when optimizing for the VR4130. It
10173 normally makes code faster, but at the expense of making it bigger.
10174 It is enabled by default at optimization level `-O3'.
10177 File: gcc.info, Node: MMIX Options, Next: MN10300 Options, Prev: MIPS Options, Up: Submodel Options
10179 3.17.21 MMIX Options
10180 --------------------
10182 These options are defined for the MMIX:
10186 Specify that intrinsic library functions are being compiled,
10187 passing all values in registers, no matter the size.
10191 Generate floating-point comparison instructions that compare with
10192 respect to the `rE' epsilon register.
10196 Generate code that passes function parameters and return values
10197 that (in the called function) are seen as registers `$0' and up,
10198 as opposed to the GNU ABI which uses global registers `$231' and
10203 When reading data from memory in sizes shorter than 64 bits, use
10204 (do not use) zero-extending load instructions by default, rather
10205 than sign-extending ones.
10209 Make the result of a division yielding a remainder have the same
10210 sign as the divisor. With the default, `-mno-knuthdiv', the sign
10211 of the remainder follows the sign of the dividend. Both methods
10212 are arithmetically valid, the latter being almost exclusively used.
10214 `-mtoplevel-symbols'
10215 `-mno-toplevel-symbols'
10216 Prepend (do not prepend) a `:' to all global symbols, so the
10217 assembly code can be used with the `PREFIX' assembly directive.
10220 Generate an executable in the ELF format, rather than the default
10221 `mmo' format used by the `mmix' simulator.
10224 `-mno-branch-predict'
10225 Use (do not use) the probable-branch instructions, when static
10226 branch prediction indicates a probable branch.
10229 `-mno-base-addresses'
10230 Generate (do not generate) code that uses _base addresses_. Using
10231 a base address automatically generates a request (handled by the
10232 assembler and the linker) for a constant to be set up in a global
10233 register. The register is used for one or more base address
10234 requests within the range 0 to 255 from the value held in the
10235 register. The generally leads to short and fast code, but the
10236 number of different data items that can be addressed is limited.
10237 This means that a program that uses lots of static data may
10238 require `-mno-base-addresses'.
10242 Force (do not force) generated code to have a single exit point in
10246 File: gcc.info, Node: MN10300 Options, Next: MT Options, Prev: MMIX Options, Up: Submodel Options
10248 3.17.22 MN10300 Options
10249 -----------------------
10251 These `-m' options are defined for Matsushita MN10300 architectures:
10254 Generate code to avoid bugs in the multiply instructions for the
10255 MN10300 processors. This is the default.
10258 Do not generate code to avoid bugs in the multiply instructions
10259 for the MN10300 processors.
10262 Generate code which uses features specific to the AM33 processor.
10265 Do not generate code which uses features specific to the AM33
10266 processor. This is the default.
10268 `-mreturn-pointer-on-d0'
10269 When generating a function which returns a pointer, return the
10270 pointer in both `a0' and `d0'. Otherwise, the pointer is returned
10271 only in a0, and attempts to call such functions without a prototype
10272 would result in errors. Note that this option is on by default;
10273 use `-mno-return-pointer-on-d0' to disable it.
10276 Do not link in the C run-time initialization object file.
10279 Indicate to the linker that it should perform a relaxation
10280 optimization pass to shorten branches, calls and absolute memory
10281 addresses. This option only has an effect when used on the
10282 command line for the final link step.
10284 This option makes symbolic debugging impossible.
10287 File: gcc.info, Node: MT Options, Next: PDP-11 Options, Prev: MN10300 Options, Up: Submodel Options
10292 These `-m' options are defined for Morpho MT architectures:
10295 Generate code that will run on CPU-TYPE, which is the name of a
10296 system representing a certain processor type. Possible values for
10297 CPU-TYPE are `ms1-64-001', `ms1-16-002', `ms1-16-003' and `ms2'.
10299 When this option is not used, the default is `-march=ms1-16-002'.
10302 Use byte loads and stores when generating code.
10305 Do not use byte loads and stores when generating code.
10308 Use simulator runtime
10311 Do not link in the C run-time initialization object file `crti.o'.
10312 Other run-time initialization and termination files such as
10313 `startup.o' and `exit.o' are still included on the linker command
10318 File: gcc.info, Node: PDP-11 Options, Next: PowerPC Options, Prev: MT Options, Up: Submodel Options
10320 3.17.24 PDP-11 Options
10321 ----------------------
10323 These options are defined for the PDP-11:
10326 Use hardware FPP floating point. This is the default. (FIS
10327 floating point on the PDP-11/40 is not supported.)
10330 Do not use hardware floating point.
10333 Return floating-point results in ac0 (fr0 in Unix assembler
10337 Return floating-point results in memory. This is the default.
10340 Generate code for a PDP-11/40.
10343 Generate code for a PDP-11/45. This is the default.
10346 Generate code for a PDP-11/10.
10349 Use inline `movmemhi' patterns for copying memory. This is the
10353 Do not use inline `movmemhi' patterns for copying memory.
10357 Use 16-bit `int'. This is the default.
10365 Use 64-bit `float'. This is the default.
10369 Use 32-bit `float'.
10372 Use `abshi2' pattern. This is the default.
10375 Do not use `abshi2' pattern.
10377 `-mbranch-expensive'
10378 Pretend that branches are expensive. This is for experimenting
10379 with code generation only.
10382 Do not pretend that branches are expensive. This is the default.
10385 Generate code for a system with split I&D.
10388 Generate code for a system without split I&D. This is the default.
10391 Use Unix assembler syntax. This is the default when configured for
10395 Use DEC assembler syntax. This is the default when configured for
10396 any PDP-11 target other than `pdp11-*-bsd'.
10399 File: gcc.info, Node: PowerPC Options, Next: RS/6000 and PowerPC Options, Prev: PDP-11 Options, Up: Submodel Options
10401 3.17.25 PowerPC Options
10402 -----------------------
10404 These are listed under *Note RS/6000 and PowerPC Options::.
10407 File: gcc.info, Node: RS/6000 and PowerPC Options, Next: S/390 and zSeries Options, Prev: PowerPC Options, Up: Submodel Options
10409 3.17.26 IBM RS/6000 and PowerPC Options
10410 ---------------------------------------
10412 These `-m' options are defined for the IBM RS/6000 and PowerPC:
10420 `-mno-powerpc-gpopt'
10422 `-mno-powerpc-gfxopt'
10431 GCC supports two related instruction set architectures for the
10432 RS/6000 and PowerPC. The "POWER" instruction set are those
10433 instructions supported by the `rios' chip set used in the original
10434 RS/6000 systems and the "PowerPC" instruction set is the
10435 architecture of the Freescale MPC5xx, MPC6xx, MPC8xx
10436 microprocessors, and the IBM 4xx, 6xx, and follow-on
10439 Neither architecture is a subset of the other. However there is a
10440 large common subset of instructions supported by both. An MQ
10441 register is included in processors supporting the POWER
10444 You use these options to specify which instructions are available
10445 on the processor you are using. The default value of these
10446 options is determined when configuring GCC. Specifying the
10447 `-mcpu=CPU_TYPE' overrides the specification of these options. We
10448 recommend you use the `-mcpu=CPU_TYPE' option rather than the
10449 options listed above.
10451 The `-mpower' option allows GCC to generate instructions that are
10452 found only in the POWER architecture and to use the MQ register.
10453 Specifying `-mpower2' implies `-power' and also allows GCC to
10454 generate instructions that are present in the POWER2 architecture
10455 but not the original POWER architecture.
10457 The `-mpowerpc' option allows GCC to generate instructions that
10458 are found only in the 32-bit subset of the PowerPC architecture.
10459 Specifying `-mpowerpc-gpopt' implies `-mpowerpc' and also allows
10460 GCC to use the optional PowerPC architecture instructions in the
10461 General Purpose group, including floating-point square root.
10462 Specifying `-mpowerpc-gfxopt' implies `-mpowerpc' and also allows
10463 GCC to use the optional PowerPC architecture instructions in the
10464 Graphics group, including floating-point select.
10466 The `-mmfcrf' option allows GCC to generate the move from
10467 condition register field instruction implemented on the POWER4
10468 processor and other processors that support the PowerPC V2.01
10469 architecture. The `-mpopcntb' option allows GCC to generate the
10470 popcount and double precision FP reciprocal estimate instruction
10471 implemented on the POWER5 processor and other processors that
10472 support the PowerPC V2.02 architecture. The `-mfprnd' option
10473 allows GCC to generate the FP round to integer instructions
10474 implemented on the POWER5+ processor and other processors that
10475 support the PowerPC V2.03 architecture.
10477 The `-mpowerpc64' option allows GCC to generate the additional
10478 64-bit instructions that are found in the full PowerPC64
10479 architecture and to treat GPRs as 64-bit, doubleword quantities.
10480 GCC defaults to `-mno-powerpc64'.
10482 If you specify both `-mno-power' and `-mno-powerpc', GCC will use
10483 only the instructions in the common subset of both architectures
10484 plus some special AIX common-mode calls, and will not use the MQ
10485 register. Specifying both `-mpower' and `-mpowerpc' permits GCC
10486 to use any instruction from either architecture and to allow use
10487 of the MQ register; specify this for the Motorola MPC601.
10491 Select which mnemonics to use in the generated assembler code.
10492 With `-mnew-mnemonics', GCC uses the assembler mnemonics defined
10493 for the PowerPC architecture. With `-mold-mnemonics' it uses the
10494 assembler mnemonics defined for the POWER architecture.
10495 Instructions defined in only one architecture have only one
10496 mnemonic; GCC uses that mnemonic irrespective of which of these
10497 options is specified.
10499 GCC defaults to the mnemonics appropriate for the architecture in
10500 use. Specifying `-mcpu=CPU_TYPE' sometimes overrides the value of
10501 these option. Unless you are building a cross-compiler, you
10502 should normally not specify either `-mnew-mnemonics' or
10503 `-mold-mnemonics', but should instead accept the default.
10506 Set architecture type, register usage, choice of mnemonics, and
10507 instruction scheduling parameters for machine type CPU_TYPE.
10508 Supported values for CPU_TYPE are `401', `403', `405', `405fp',
10509 `440', `440fp', `505', `601', `602', `603', `603e', `604', `604e',
10510 `620', `630', `740', `7400', `7450', `750', `801', `821', `823',
10511 `860', `970', `8540', `ec603e', `G3', `G4', `G5', `power',
10512 `power2', `power3', `power4', `power5', `power5+', `power6',
10513 `common', `powerpc', `powerpc64', `rios', `rios1', `rios2', `rsc',
10516 `-mcpu=common' selects a completely generic processor. Code
10517 generated under this option will run on any POWER or PowerPC
10518 processor. GCC will use only the instructions in the common
10519 subset of both architectures, and will not use the MQ register.
10520 GCC assumes a generic processor model for scheduling purposes.
10522 `-mcpu=power', `-mcpu=power2', `-mcpu=powerpc', and
10523 `-mcpu=powerpc64' specify generic POWER, POWER2, pure 32-bit
10524 PowerPC (i.e., not MPC601), and 64-bit PowerPC architecture machine
10525 types, with an appropriate, generic processor model assumed for
10526 scheduling purposes.
10528 The other options specify a specific processor. Code generated
10529 under those options will run best on that processor, and may not
10530 run at all on others.
10532 The `-mcpu' options automatically enable or disable the following
10533 options: `-maltivec', `-mfprnd', `-mhard-float', `-mmfcrf',
10534 `-mmultiple', `-mnew-mnemonics', `-mpopcntb', `-mpower',
10535 `-mpower2', `-mpowerpc64', `-mpowerpc-gpopt', `-mpowerpc-gfxopt',
10536 `-mstring'. The particular options set for any particular CPU
10537 will vary between compiler versions, depending on what setting
10538 seems to produce optimal code for that CPU; it doesn't necessarily
10539 reflect the actual hardware's capabilities. If you wish to set an
10540 individual option to a particular value, you may specify it after
10541 the `-mcpu' option, like `-mcpu=970 -mno-altivec'.
10543 On AIX, the `-maltivec' and `-mpowerpc64' options are not enabled
10544 or disabled by the `-mcpu' option at present because AIX does not
10545 have full support for these options. You may still enable or
10546 disable them individually if you're sure it'll work in your
10550 Set the instruction scheduling parameters for machine type
10551 CPU_TYPE, but do not set the architecture type, register usage, or
10552 choice of mnemonics, as `-mcpu=CPU_TYPE' would. The same values
10553 for CPU_TYPE are used for `-mtune' as for `-mcpu'. If both are
10554 specified, the code generated will use the architecture,
10555 registers, and mnemonics set by `-mcpu', but the scheduling
10556 parameters set by `-mtune'.
10560 Generate code to compute division as reciprocal estimate and
10561 iterative refinement, creating opportunities for increased
10562 throughput. This feature requires: optional PowerPC Graphics
10563 instruction set for single precision and FRE instruction for
10564 double precision, assuming divides cannot generate user-visible
10565 traps, and the domain values not include Infinities, denormals or
10570 Generate code that uses (does not use) AltiVec instructions, and
10571 also enable the use of built-in functions that allow more direct
10572 access to the AltiVec instruction set. You may also need to set
10573 `-mabi=altivec' to adjust the current ABI with AltiVec ABI
10579 Generate VRSAVE instructions when generating AltiVec code.
10582 Generate code that allows ld and ld.so to build executables and
10583 shared libraries with non-exec .plt and .got sections. This is a
10584 PowerPC 32-bit SYSV ABI option.
10587 Generate code that uses a BSS .plt section that ld.so fills in, and
10588 requires .plt and .got sections that are both writable and
10589 executable. This is a PowerPC 32-bit SYSV ABI option.
10593 This switch enables or disables the generation of ISEL
10597 This switch has been deprecated. Use `-misel' and `-mno-isel'
10602 This switch enables or disables the generation of SPE simd
10606 This option has been deprecated. Use `-mspe' and `-mno-spe'
10609 `-mfloat-gprs=YES/SINGLE/DOUBLE/NO'
10611 This switch enables or disables the generation of floating point
10612 operations on the general purpose registers for architectures that
10615 The argument YES or SINGLE enables the use of single-precision
10616 floating point operations.
10618 The argument DOUBLE enables the use of single and double-precision
10619 floating point operations.
10621 The argument NO disables floating point operations on the general
10624 This option is currently only available on the MPC854x.
10628 Generate code for 32-bit or 64-bit environments of Darwin and SVR4
10629 targets (including GNU/Linux). The 32-bit environment sets int,
10630 long and pointer to 32 bits and generates code that runs on any
10631 PowerPC variant. The 64-bit environment sets int to 32 bits and
10632 long and pointer to 64 bits, and generates code for PowerPC64, as
10639 Modify generation of the TOC (Table Of Contents), which is created
10640 for every executable file. The `-mfull-toc' option is selected by
10641 default. In that case, GCC will allocate at least one TOC entry
10642 for each unique non-automatic variable reference in your program.
10643 GCC will also place floating-point constants in the TOC. However,
10644 only 16,384 entries are available in the TOC.
10646 If you receive a linker error message that saying you have
10647 overflowed the available TOC space, you can reduce the amount of
10648 TOC space used with the `-mno-fp-in-toc' and `-mno-sum-in-toc'
10649 options. `-mno-fp-in-toc' prevents GCC from putting floating-point
10650 constants in the TOC and `-mno-sum-in-toc' forces GCC to generate
10651 code to calculate the sum of an address and a constant at run-time
10652 instead of putting that sum into the TOC. You may specify one or
10653 both of these options. Each causes GCC to produce very slightly
10654 slower and larger code at the expense of conserving TOC space.
10656 If you still run out of space in the TOC even when you specify
10657 both of these options, specify `-mminimal-toc' instead. This
10658 option causes GCC to make only one TOC entry for every file. When
10659 you specify this option, GCC will produce code that is slower and
10660 larger but which uses extremely little TOC space. You may wish to
10661 use this option only on files that contain less frequently
10666 Enable 64-bit AIX ABI and calling convention: 64-bit pointers,
10667 64-bit `long' type, and the infrastructure needed to support them.
10668 Specifying `-maix64' implies `-mpowerpc64' and `-mpowerpc', while
10669 `-maix32' disables the 64-bit ABI and implies `-mno-powerpc64'.
10670 GCC defaults to `-maix32'.
10674 Produce code that conforms more closely to IBM XL compiler
10675 semantics when using AIX-compatible ABI. Pass floating-point
10676 arguments to prototyped functions beyond the register save area
10677 (RSA) on the stack in addition to argument FPRs. Do not assume
10678 that most significant double in 128-bit long double value is
10679 properly rounded when comparing values and converting to double.
10680 Use XL symbol names for long double support routines.
10682 The AIX calling convention was extended but not initially
10683 documented to handle an obscure K&R C case of calling a function
10684 that takes the address of its arguments with fewer arguments than
10685 declared. IBM XL compilers access floating point arguments which
10686 do not fit in the RSA from the stack when a subroutine is compiled
10687 without optimization. Because always storing floating-point
10688 arguments on the stack is inefficient and rarely needed, this
10689 option is not enabled by default and only is necessary when
10690 calling subroutines compiled by IBM XL compilers without
10694 Support "IBM RS/6000 SP" "Parallel Environment" (PE). Link an
10695 application written to use message passing with special startup
10696 code to enable the application to run. The system must have PE
10697 installed in the standard location (`/usr/lpp/ppe.poe/'), or the
10698 `specs' file must be overridden with the `-specs=' option to
10699 specify the appropriate directory location. The Parallel
10700 Environment does not support threads, so the `-mpe' option and the
10701 `-pthread' option are incompatible.
10705 On AIX, 32-bit Darwin, and 64-bit PowerPC GNU/Linux, the option
10706 `-malign-natural' overrides the ABI-defined alignment of larger
10707 types, such as floating-point doubles, on their natural size-based
10708 boundary. The option `-malign-power' instructs GCC to follow the
10709 ABI-specified alignment rules. GCC defaults to the standard
10710 alignment defined in the ABI.
10712 On 64-bit Darwin, natural alignment is the default, and
10713 `-malign-power' is not supported.
10717 Generate code that does not use (uses) the floating-point register
10718 set. Software floating point emulation is provided if you use the
10719 `-msoft-float' option, and pass the option to GCC when linking.
10723 Generate code that uses (does not use) the load multiple word
10724 instructions and the store multiple word instructions. These
10725 instructions are generated by default on POWER systems, and not
10726 generated on PowerPC systems. Do not use `-mmultiple' on little
10727 endian PowerPC systems, since those instructions do not work when
10728 the processor is in little endian mode. The exceptions are PPC740
10729 and PPC750 which permit the instructions usage in little endian
10734 Generate code that uses (does not use) the load string instructions
10735 and the store string word instructions to save multiple registers
10736 and do small block moves. These instructions are generated by
10737 default on POWER systems, and not generated on PowerPC systems.
10738 Do not use `-mstring' on little endian PowerPC systems, since those
10739 instructions do not work when the processor is in little endian
10740 mode. The exceptions are PPC740 and PPC750 which permit the
10741 instructions usage in little endian mode.
10745 Generate code that uses (does not use) the load or store
10746 instructions that update the base register to the address of the
10747 calculated memory location. These instructions are generated by
10748 default. If you use `-mno-update', there is a small window
10749 between the time that the stack pointer is updated and the address
10750 of the previous frame is stored, which means code that walks the
10751 stack frame across interrupts or signals may get corrupted data.
10755 Generate code that uses (does not use) the floating point multiply
10756 and accumulate instructions. These instructions are generated by
10757 default if hardware floating is used.
10761 On System V.4 and embedded PowerPC systems do not (do) force
10762 structures and unions that contain bit-fields to be aligned to the
10763 base type of the bit-field.
10765 For example, by default a structure containing nothing but 8
10766 `unsigned' bit-fields of length 1 would be aligned to a 4 byte
10767 boundary and have a size of 4 bytes. By using `-mno-bit-align',
10768 the structure would be aligned to a 1 byte boundary and be one
10771 `-mno-strict-align'
10773 On System V.4 and embedded PowerPC systems do not (do) assume that
10774 unaligned memory references will be handled by the system.
10778 On embedded PowerPC systems generate code that allows (does not
10779 allow) the program to be relocated to a different address at
10780 runtime. If you use `-mrelocatable' on any module, all objects
10781 linked together must be compiled with `-mrelocatable' or
10782 `-mrelocatable-lib'.
10784 `-mrelocatable-lib'
10785 `-mno-relocatable-lib'
10786 On embedded PowerPC systems generate code that allows (does not
10787 allow) the program to be relocated to a different address at
10788 runtime. Modules compiled with `-mrelocatable-lib' can be linked
10789 with either modules compiled without `-mrelocatable' and
10790 `-mrelocatable-lib' or with modules compiled with the
10791 `-mrelocatable' options.
10795 On System V.4 and embedded PowerPC systems do not (do) assume that
10796 register 2 contains a pointer to a global area pointing to the
10797 addresses used in the program.
10801 On System V.4 and embedded PowerPC systems compile code for the
10802 processor in little endian mode. The `-mlittle-endian' option is
10803 the same as `-mlittle'.
10807 On System V.4 and embedded PowerPC systems compile code for the
10808 processor in big endian mode. The `-mbig-endian' option is the
10812 On Darwin and Mac OS X systems, compile code so that it is not
10813 relocatable, but that its external references are relocatable. The
10814 resulting code is suitable for applications, but not shared
10817 `-mprioritize-restricted-insns=PRIORITY'
10818 This option controls the priority that is assigned to
10819 dispatch-slot restricted instructions during the second scheduling
10820 pass. The argument PRIORITY takes the value 0/1/2 to assign
10821 NO/HIGHEST/SECOND-HIGHEST priority to dispatch slot restricted
10824 `-msched-costly-dep=DEPENDENCE_TYPE'
10825 This option controls which dependences are considered costly by
10826 the target during instruction scheduling. The argument
10827 DEPENDENCE_TYPE takes one of the following values: NO: no
10828 dependence is costly, ALL: all dependences are costly,
10829 TRUE_STORE_TO_LOAD: a true dependence from store to load is costly,
10830 STORE_TO_LOAD: any dependence from store to load is costly,
10831 NUMBER: any dependence which latency >= NUMBER is costly.
10833 `-minsert-sched-nops=SCHEME'
10834 This option controls which nop insertion scheme will be used during
10835 the second scheduling pass. The argument SCHEME takes one of the
10836 following values: NO: Don't insert nops. PAD: Pad with nops any
10837 dispatch group which has vacant issue slots, according to the
10838 scheduler's grouping. REGROUP_EXACT: Insert nops to force costly
10839 dependent insns into separate groups. Insert exactly as many nops
10840 as needed to force an insn to a new group, according to the
10841 estimated processor grouping. NUMBER: Insert nops to force costly
10842 dependent insns into separate groups. Insert NUMBER nops to force
10843 an insn to a new group.
10846 On System V.4 and embedded PowerPC systems compile code using
10847 calling conventions that adheres to the March 1995 draft of the
10848 System V Application Binary Interface, PowerPC processor
10849 supplement. This is the default unless you configured GCC using
10850 `powerpc-*-eabiaix'.
10853 Specify both `-mcall-sysv' and `-meabi' options.
10855 `-mcall-sysv-noeabi'
10856 Specify both `-mcall-sysv' and `-mno-eabi' options.
10859 On System V.4 and embedded PowerPC systems compile code for the
10860 Solaris operating system.
10863 On System V.4 and embedded PowerPC systems compile code for the
10864 Linux-based GNU system.
10867 On System V.4 and embedded PowerPC systems compile code for the
10868 Hurd-based GNU system.
10871 On System V.4 and embedded PowerPC systems compile code for the
10872 NetBSD operating system.
10874 `-maix-struct-return'
10875 Return all structures in memory (as specified by the AIX ABI).
10877 `-msvr4-struct-return'
10878 Return structures smaller than 8 bytes in registers (as specified
10882 Extend the current ABI with a particular extension, or remove such
10883 extension. Valid values are ALTIVEC, NO-ALTIVEC, SPE, NO-SPE,
10884 IBMLONGDOUBLE, IEEELONGDOUBLE.
10887 Extend the current ABI with SPE ABI extensions. This does not
10888 change the default ABI, instead it adds the SPE ABI extensions to
10892 Disable Booke SPE ABI extensions for the current ABI.
10894 `-mabi=ibmlongdouble'
10895 Change the current ABI to use IBM extended precision long double.
10896 This is a PowerPC 32-bit SYSV ABI option.
10898 `-mabi=ieeelongdouble'
10899 Change the current ABI to use IEEE extended precision long double.
10900 This is a PowerPC 32-bit Linux ABI option.
10904 On System V.4 and embedded PowerPC systems assume that all calls to
10905 variable argument functions are properly prototyped. Otherwise,
10906 the compiler must insert an instruction before every non
10907 prototyped call to set or clear bit 6 of the condition code
10908 register (CR) to indicate whether floating point values were
10909 passed in the floating point registers in case the function takes
10910 a variable arguments. With `-mprototype', only calls to
10911 prototyped variable argument functions will set or clear the bit.
10914 On embedded PowerPC systems, assume that the startup module is
10915 called `sim-crt0.o' and that the standard C libraries are
10916 `libsim.a' and `libc.a'. This is the default for
10917 `powerpc-*-eabisim'. configurations.
10920 On embedded PowerPC systems, assume that the startup module is
10921 called `crt0.o' and the standard C libraries are `libmvme.a' and
10925 On embedded PowerPC systems, assume that the startup module is
10926 called `crt0.o' and the standard C libraries are `libads.a' and
10930 On embedded PowerPC systems, assume that the startup module is
10931 called `crt0.o' and the standard C libraries are `libyk.a' and
10935 On System V.4 and embedded PowerPC systems, specify that you are
10936 compiling for a VxWorks system.
10939 Specify that you are compiling for the WindISS simulation
10943 On embedded PowerPC systems, set the PPC_EMB bit in the ELF flags
10944 header to indicate that `eabi' extended relocations are used.
10948 On System V.4 and embedded PowerPC systems do (do not) adhere to
10949 the Embedded Applications Binary Interface (eabi) which is a set of
10950 modifications to the System V.4 specifications. Selecting `-meabi'
10951 means that the stack is aligned to an 8 byte boundary, a function
10952 `__eabi' is called to from `main' to set up the eabi environment,
10953 and the `-msdata' option can use both `r2' and `r13' to point to
10954 two separate small data areas. Selecting `-mno-eabi' means that
10955 the stack is aligned to a 16 byte boundary, do not call an
10956 initialization function from `main', and the `-msdata' option will
10957 only use `r13' to point to a single small data area. The `-meabi'
10958 option is on by default if you configured GCC using one of the
10959 `powerpc*-*-eabi*' options.
10962 On System V.4 and embedded PowerPC systems, put small initialized
10963 `const' global and static data in the `.sdata2' section, which is
10964 pointed to by register `r2'. Put small initialized non-`const'
10965 global and static data in the `.sdata' section, which is pointed
10966 to by register `r13'. Put small uninitialized global and static
10967 data in the `.sbss' section, which is adjacent to the `.sdata'
10968 section. The `-msdata=eabi' option is incompatible with the
10969 `-mrelocatable' option. The `-msdata=eabi' option also sets the
10973 On System V.4 and embedded PowerPC systems, put small global and
10974 static data in the `.sdata' section, which is pointed to by
10975 register `r13'. Put small uninitialized global and static data in
10976 the `.sbss' section, which is adjacent to the `.sdata' section.
10977 The `-msdata=sysv' option is incompatible with the `-mrelocatable'
10982 On System V.4 and embedded PowerPC systems, if `-meabi' is used,
10983 compile code the same as `-msdata=eabi', otherwise compile code the
10984 same as `-msdata=sysv'.
10987 On System V.4 and embedded PowerPC systems, put small global and
10988 static data in the `.sdata' section. Put small uninitialized
10989 global and static data in the `.sbss' section. Do not use
10990 register `r13' to address small data however. This is the default
10991 behavior unless other `-msdata' options are used.
10995 On embedded PowerPC systems, put all initialized global and static
10996 data in the `.data' section, and all uninitialized data in the
11000 On embedded PowerPC systems, put global and static items less than
11001 or equal to NUM bytes into the small data or bss sections instead
11002 of the normal data or bss section. By default, NUM is 8. The `-G
11003 NUM' switch is also passed to the linker. All modules should be
11004 compiled with the same `-G NUM' value.
11008 On System V.4 and embedded PowerPC systems do (do not) emit
11009 register names in the assembly language output using symbolic
11014 Default to making all function calls indirectly, using a register,
11015 so that functions which reside further than 32 megabytes
11016 (33,554,432 bytes) from the current location can be called. This
11017 setting can be overridden by the `shortcall' function attribute,
11018 or by `#pragma longcall(0)'.
11020 Some linkers are capable of detecting out-of-range calls and
11021 generating glue code on the fly. On these systems, long calls are
11022 unnecessary and generate slower code. As of this writing, the AIX
11023 linker can do this, as can the GNU linker for PowerPC/64. It is
11024 planned to add this feature to the GNU linker for 32-bit PowerPC
11027 On Darwin/PPC systems, `#pragma longcall' will generate "jbsr
11028 callee, L42", plus a "branch island" (glue code). The two target
11029 addresses represent the callee and the "branch island". The
11030 Darwin/PPC linker will prefer the first address and generate a "bl
11031 callee" if the PPC "bl" instruction will reach the callee directly;
11032 otherwise, the linker will generate "bl L42" to call the "branch
11033 island". The "branch island" is appended to the body of the
11034 calling function; it computes the full 32-bit address of the callee
11037 On Mach-O (Darwin) systems, this option directs the compiler emit
11038 to the glue for every direct call, and the Darwin linker decides
11039 whether to use or discard it.
11041 In the future, we may cause GCC to ignore all longcall
11042 specifications when the linker is known to generate glue.
11045 Adds support for multithreading with the "pthreads" library. This
11046 option sets flags for both the preprocessor and linker.
11050 File: gcc.info, Node: S/390 and zSeries Options, Next: SH Options, Prev: RS/6000 and PowerPC Options, Up: Submodel Options
11052 3.17.27 S/390 and zSeries Options
11053 ---------------------------------
11055 These are the `-m' options defined for the S/390 and zSeries
11060 Use (do not use) the hardware floating-point instructions and
11061 registers for floating-point operations. When `-msoft-float' is
11062 specified, functions in `libgcc.a' will be used to perform
11063 floating-point operations. When `-mhard-float' is specified, the
11064 compiler generates IEEE floating-point instructions. This is the
11068 `-mlong-double-128'
11069 These switches control the size of `long double' type. A size of
11070 64bit makes the `long double' type equivalent to the `double'
11071 type. This is the default.
11075 Store (do not store) the address of the caller's frame as
11076 backchain pointer into the callee's stack frame. A backchain may
11077 be needed to allow debugging using tools that do not understand
11078 DWARF-2 call frame information. When `-mno-packed-stack' is in
11079 effect, the backchain pointer is stored at the bottom of the stack
11080 frame; when `-mpacked-stack' is in effect, the backchain is placed
11081 into the topmost word of the 96/160 byte register save area.
11083 In general, code compiled with `-mbackchain' is call-compatible
11084 with code compiled with `-mmo-backchain'; however, use of the
11085 backchain for debugging purposes usually requires that the whole
11086 binary is built with `-mbackchain'. Note that the combination of
11087 `-mbackchain', `-mpacked-stack' and `-mhard-float' is not
11088 supported. In order to build a linux kernel use `-msoft-float'.
11090 The default is to not maintain the backchain.
11094 `-mno-packed-stack'
11095 Use (do not use) the packed stack layout. When
11096 `-mno-packed-stack' is specified, the compiler uses the all fields
11097 of the 96/160 byte register save area only for their default
11098 purpose; unused fields still take up stack space. When
11099 `-mpacked-stack' is specified, register save slots are densely
11100 packed at the top of the register save area; unused space is
11101 reused for other purposes, allowing for more efficient use of the
11102 available stack space. However, when `-mbackchain' is also in
11103 effect, the topmost word of the save area is always used to store
11104 the backchain, and the return address register is always saved two
11105 words below the backchain.
11107 As long as the stack frame backchain is not used, code generated
11108 with `-mpacked-stack' is call-compatible with code generated with
11109 `-mno-packed-stack'. Note that some non-FSF releases of GCC 2.95
11110 for S/390 or zSeries generated code that uses the stack frame
11111 backchain at run time, not just for debugging purposes. Such code
11112 is not call-compatible with code compiled with `-mpacked-stack'.
11113 Also, note that the combination of `-mbackchain', `-mpacked-stack'
11114 and `-mhard-float' is not supported. In order to build a linux
11115 kernel use `-msoft-float'.
11117 The default is to not use the packed stack layout.
11121 Generate (or do not generate) code using the `bras' instruction to
11122 do subroutine calls. This only works reliably if the total
11123 executable size does not exceed 64k. The default is to use the
11124 `basr' instruction instead, which does not have this limitation.
11128 When `-m31' is specified, generate code compliant to the GNU/Linux
11129 for S/390 ABI. When `-m64' is specified, generate code compliant
11130 to the GNU/Linux for zSeries ABI. This allows GCC in particular
11131 to generate 64-bit instructions. For the `s390' targets, the
11132 default is `-m31', while the `s390x' targets default to `-m64'.
11136 When `-mzarch' is specified, generate code using the instructions
11137 available on z/Architecture. When `-mesa' is specified, generate
11138 code using the instructions available on ESA/390. Note that
11139 `-mesa' is not possible with `-m64'. When generating code
11140 compliant to the GNU/Linux for S/390 ABI, the default is `-mesa'.
11141 When generating code compliant to the GNU/Linux for zSeries ABI,
11142 the default is `-mzarch'.
11146 Generate (or do not generate) code using the `mvcle' instruction
11147 to perform block moves. When `-mno-mvcle' is specified, use a
11148 `mvc' loop instead. This is the default unless optimizing for
11153 Print (or do not print) additional debug information when
11154 compiling. The default is to not print debug information.
11157 Generate code that will run on CPU-TYPE, which is the name of a
11158 system representing a certain processor type. Possible values for
11159 CPU-TYPE are `g5', `g6', `z900', and `z990'. When generating code
11160 using the instructions available on z/Architecture, the default is
11161 `-march=z900'. Otherwise, the default is `-march=g5'.
11164 Tune to CPU-TYPE everything applicable about the generated code,
11165 except for the ABI and the set of available instructions. The
11166 list of CPU-TYPE values is the same as for `-march'. The default
11167 is the value used for `-march'.
11171 Generate code that adds (does not add) in TPF OS specific branches
11172 to trace routines in the operating system. This option is off by
11173 default, even when compiling for the TPF OS.
11177 Generate code that uses (does not use) the floating point multiply
11178 and accumulate instructions. These instructions are generated by
11179 default if hardware floating point is used.
11181 `-mwarn-framesize=FRAMESIZE'
11182 Emit a warning if the current function exceeds the given frame
11183 size. Because this is a compile time check it doesn't need to be
11184 a real problem when the program runs. It is intended to identify
11185 functions which most probably cause a stack overflow. It is
11186 useful to be used in an environment with limited stack size e.g.
11189 `-mwarn-dynamicstack'
11190 Emit a warning if the function calls alloca or uses dynamically
11191 sized arrays. This is generally a bad idea with a limited stack
11194 `-mstack-guard=STACK-GUARD'
11196 `-mstack-size=STACK-SIZE'
11197 These arguments always have to be used in conjunction. If they
11198 are present the s390 back end emits additional instructions in the
11199 function prologue which trigger a trap if the stack size is
11200 STACK-GUARD bytes above the STACK-SIZE (remember that the stack on
11201 s390 grows downward). These options are intended to be used to
11202 help debugging stack overflow problems. The additionally emitted
11203 code causes only little overhead and hence can also be used in
11204 production like systems without greater performance degradation.
11205 The given values have to be exact powers of 2 and STACK-SIZE has
11206 to be greater than STACK-GUARD without exceeding 64k. In order to
11207 be efficient the extra code makes the assumption that the stack
11208 starts at an address aligned to the value given by STACK-SIZE.
11211 File: gcc.info, Node: SH Options, Next: SPARC Options, Prev: S/390 and zSeries Options, Up: Submodel Options
11216 These `-m' options are defined for the SH implementations:
11219 Generate code for the SH1.
11222 Generate code for the SH2.
11225 Generate code for the SH2e.
11228 Generate code for the SH3.
11231 Generate code for the SH3e.
11234 Generate code for the SH4 without a floating-point unit.
11237 Generate code for the SH4 with a floating-point unit that only
11238 supports single-precision arithmetic.
11241 Generate code for the SH4 assuming the floating-point unit is in
11242 single-precision mode by default.
11245 Generate code for the SH4.
11248 Generate code for the SH4al-dsp, or for a SH4a in such a way that
11249 the floating-point unit is not used.
11252 Generate code for the SH4a, in such a way that no double-precision
11253 floating point operations are used.
11256 Generate code for the SH4a assuming the floating-point unit is in
11257 single-precision mode by default.
11260 Generate code for the SH4a.
11263 Same as `-m4a-nofpu', except that it implicitly passes `-dsp' to
11264 the assembler. GCC doesn't generate any DSP instructions at the
11268 Compile code for the processor in big endian mode.
11271 Compile code for the processor in little endian mode.
11274 Align doubles at 64-bit boundaries. Note that this changes the
11275 calling conventions, and thus some functions from the standard C
11276 library will not work unless you recompile it first with
11280 Shorten some address references at link time, when possible; uses
11281 the linker option `-relax'.
11284 Use 32-bit offsets in `switch' tables. The default is to use
11288 Enable the use of the instruction `fmovd'.
11291 Comply with the calling conventions defined by Renesas.
11294 Comply with the calling conventions defined by Renesas.
11297 Comply with the calling conventions defined for GCC before the
11298 Renesas conventions were available. This option is the default
11299 for all targets of the SH toolchain except for `sh-symbianelf'.
11302 Mark the `MAC' register as call-clobbered, even if `-mhitachi' is
11306 Increase IEEE-compliance of floating-point code. At the moment,
11307 this is equivalent to `-fno-finite-math-only'. When generating 16
11308 bit SH opcodes, getting IEEE-conforming results for comparisons of
11309 NANs / infinities incurs extra overhead in every floating point
11310 comparison, therefore the default is set to `-ffinite-math-only'.
11313 Dump instruction size and location in the assembly code.
11316 This option is deprecated. It pads structures to multiple of 4
11317 bytes, which is incompatible with the SH ABI.
11320 Optimize for space instead of speed. Implied by `-Os'.
11323 When generating position-independent code, emit function calls
11324 using the Global Offset Table instead of the Procedure Linkage
11328 Generate a library function call to invalidate instruction cache
11329 entries, after fixing up a trampoline. This library function call
11330 doesn't assume it can write to the whole memory address space.
11331 This is the default when the target is `sh-*-linux*'.
11334 Set the cost to assume for a multiply insn.
11337 Set the division strategy to use for SHmedia code. STRATEGY must
11338 be one of: call, call2, fp, inv, inv:minlat, inv20u, inv20l,
11339 inv:call, inv:call2, inv:fp . "fp" performs the operation in
11340 floating point. This has a very high latency, but needs only a
11341 few instructions, so it might be a good choice if your code has
11342 enough easily exploitable ILP to allow the compiler to schedule
11343 the floating point instructions together with other instructions.
11344 Division by zero causes a floating point exception. "inv" uses
11345 integer operations to calculate the inverse of the divisor, and
11346 then multiplies the dividend with the inverse. This strategy
11347 allows cse and hoisting of the inverse calculation. Division by
11348 zero calculates an unspecified result, but does not trap.
11349 "inv:minlat" is a variant of "inv" where if no cse / hoisting
11350 opportunities have been found, or if the entire operation has been
11351 hoisted to the same place, the last stages of the inverse
11352 calculation are intertwined with the final multiply to reduce the
11353 overall latency, at the expense of using a few more instructions,
11354 and thus offering fewer scheduling opportunities with other code.
11355 "call" calls a library function that usually implements the
11356 inv:minlat strategy. This gives high code density for
11357 m5-*media-nofpu compilations. "call2" uses a different entry
11358 point of the same library function, where it assumes that a
11359 pointer to a lookup table has already been set up, which exposes
11360 the pointer load to cse / code hoisting optimizations.
11361 "inv:call", "inv:call2" and "inv:fp" all use the "inv" algorithm
11362 for initial code generation, but if the code stays unoptimized,
11363 revert to the "call", "call2", or "fp" strategies, respectively.
11364 Note that the potentially-trapping side effect of division by zero
11365 is carried by a separate instruction, so it is possible that all
11366 the integer instructions are hoisted out, but the marker for the
11367 side effect stays where it is. A recombination to fp operations
11368 or a call is not possible in that case. "inv20u" and "inv20l" are
11369 variants of the "inv:minlat" strategy. In the case that the
11370 inverse calculation was nor separated from the multiply, they speed
11371 up division where the dividend fits into 20 bits (plus sign where
11372 applicable), by inserting a test to skip a number of operations in
11373 this case; this test slows down the case of larger dividends.
11374 inv20u assumes the case of a such a small dividend to be unlikely,
11375 and inv20l assumes it to be likely.
11377 `-mdivsi3_libfunc=NAME'
11378 Set the name of the library function used for 32 bit signed
11379 division to NAME. This only affect the name used in the call and
11380 inv:call division strategies, and the compiler will still expect
11381 the same sets of input/output/clobbered registers as if this
11382 option was not present.
11385 Throttle unrolling to avoid thrashing target registers. This
11386 option only has an effect if the gcc code base supports the
11387 TARGET_ADJUST_UNROLL_MAX target hook.
11389 `-mindexed-addressing'
11390 Enable the use of the indexed addressing mode for
11391 SHmedia32/SHcompact. This is only safe if the hardware and/or OS
11392 implement 32 bit wrap-around semantics for the indexed addressing
11393 mode. The architecture allows the implementation of processors
11394 with 64 bit MMU, which the OS could use to get 32 bit addressing,
11395 but since no current hardware implementation supports this or any
11396 other way to make the indexed addressing mode safe to use in the
11397 32 bit ABI, the default is -mno-indexed-addressing.
11399 `-mgettrcost=NUMBER'
11400 Set the cost assumed for the gettr instruction to NUMBER. The
11401 default is 2 if `-mpt-fixed' is in effect, 100 otherwise.
11404 Assume pt* instructions won't trap. This will generally generate
11405 better scheduled code, but is unsafe on current hardware. The
11406 current architecture definition says that ptabs and ptrel trap
11407 when the target anded with 3 is 3. This has the unintentional
11408 effect of making it unsafe to schedule ptabs / ptrel before a
11409 branch, or hoist it out of a loop. For example,
11410 __do_global_ctors, a part of libgcc that runs constructors at
11411 program startup, calls functions in a list which is delimited by
11412 -1. With the -mpt-fixed option, the ptabs will be done before
11413 testing against -1. That means that all the constructors will be
11414 run a bit quicker, but when the loop comes to the end of the list,
11415 the program crashes because ptabs loads -1 into a target register.
11416 Since this option is unsafe for any hardware implementing the
11417 current architecture specification, the default is -mno-pt-fixed.
11418 Unless the user specifies a specific cost with `-mgettrcost',
11419 -mno-pt-fixed also implies `-mgettrcost=100'; this deters register
11420 allocation using target registers for storing ordinary integers.
11422 `-minvalid-symbols'
11423 Assume symbols might be invalid. Ordinary function symbols
11424 generated by the compiler will always be valid to load with
11425 movi/shori/ptabs or movi/shori/ptrel, but with assembler and/or
11426 linker tricks it is possible to generate symbols that will cause
11427 ptabs / ptrel to trap. This option is only meaningful when
11428 `-mno-pt-fixed' is in effect. It will then prevent
11429 cross-basic-block cse, hoisting and most scheduling of symbol
11430 loads. The default is `-mno-invalid-symbols'.
11433 File: gcc.info, Node: SPARC Options, Next: System V Options, Prev: SH Options, Up: Submodel Options
11435 3.17.29 SPARC Options
11436 ---------------------
11438 These `-m' options are supported on the SPARC:
11442 Specify `-mapp-regs' to generate output using the global registers
11443 2 through 4, which the SPARC SVR4 ABI reserves for applications.
11444 This is the default.
11446 To be fully SVR4 ABI compliant at the cost of some performance
11447 loss, specify `-mno-app-regs'. You should compile libraries and
11448 system software with this option.
11452 Generate output containing floating point instructions. This is
11457 Generate output containing library calls for floating point.
11458 *Warning:* the requisite libraries are not available for all SPARC
11459 targets. Normally the facilities of the machine's usual C
11460 compiler are used, but this cannot be done directly in
11461 cross-compilation. You must make your own arrangements to provide
11462 suitable library functions for cross-compilation. The embedded
11463 targets `sparc-*-aout' and `sparclite-*-*' do provide software
11464 floating point support.
11466 `-msoft-float' changes the calling convention in the output file;
11467 therefore, it is only useful if you compile _all_ of a program with
11468 this option. In particular, you need to compile `libgcc.a', the
11469 library that comes with GCC, with `-msoft-float' in order for this
11472 `-mhard-quad-float'
11473 Generate output containing quad-word (long double) floating point
11476 `-msoft-quad-float'
11477 Generate output containing library calls for quad-word (long
11478 double) floating point instructions. The functions called are
11479 those specified in the SPARC ABI. This is the default.
11481 As of this writing, there are no SPARC implementations that have
11482 hardware support for the quad-word floating point instructions.
11483 They all invoke a trap handler for one of these instructions, and
11484 then the trap handler emulates the effect of the instruction.
11485 Because of the trap handler overhead, this is much slower than
11486 calling the ABI library routines. Thus the `-msoft-quad-float'
11487 option is the default.
11489 `-mno-unaligned-doubles'
11490 `-munaligned-doubles'
11491 Assume that doubles have 8 byte alignment. This is the default.
11493 With `-munaligned-doubles', GCC assumes that doubles have 8 byte
11494 alignment only if they are contained in another type, or if they
11495 have an absolute address. Otherwise, it assumes they have 4 byte
11496 alignment. Specifying this option avoids some rare compatibility
11497 problems with code generated by other compilers. It is not the
11498 default because it results in a performance loss, especially for
11499 floating point code.
11501 `-mno-faster-structs'
11503 With `-mfaster-structs', the compiler assumes that structures
11504 should have 8 byte alignment. This enables the use of pairs of
11505 `ldd' and `std' instructions for copies in structure assignment,
11506 in place of twice as many `ld' and `st' pairs. However, the use
11507 of this changed alignment directly violates the SPARC ABI. Thus,
11508 it's intended only for use on targets where the developer
11509 acknowledges that their resulting code will not be directly in
11510 line with the rules of the ABI.
11513 `-mimpure-text', used in addition to `-shared', tells the compiler
11514 to not pass `-z text' to the linker when linking a shared object.
11515 Using this option, you can link position-dependent code into a
11518 `-mimpure-text' suppresses the "relocations remain against
11519 allocatable but non-writable sections" linker error message.
11520 However, the necessary relocations will trigger copy-on-write, and
11521 the shared object is not actually shared across processes.
11522 Instead of using `-mimpure-text', you should compile all source
11523 code with `-fpic' or `-fPIC'.
11525 This option is only available on SunOS and Solaris.
11528 Set the instruction set, register set, and instruction scheduling
11529 parameters for machine type CPU_TYPE. Supported values for
11530 CPU_TYPE are `v7', `cypress', `v8', `supersparc', `sparclite',
11531 `f930', `f934', `hypersparc', `sparclite86x', `sparclet',
11532 `tsc701', `v9', `ultrasparc', and `ultrasparc3'.
11534 Default instruction scheduling parameters are used for values that
11535 select an architecture and not an implementation. These are `v7',
11536 `v8', `sparclite', `sparclet', `v9'.
11538 Here is a list of each supported architecture and their supported
11542 v8: supersparc, hypersparc
11543 sparclite: f930, f934, sparclite86x
11545 v9: ultrasparc, ultrasparc3
11547 By default (unless configured otherwise), GCC generates code for
11548 the V7 variant of the SPARC architecture. With `-mcpu=cypress',
11549 the compiler additionally optimizes it for the Cypress CY7C602
11550 chip, as used in the SPARCStation/SPARCServer 3xx series. This is
11551 also appropriate for the older SPARCStation 1, 2, IPX etc.
11553 With `-mcpu=v8', GCC generates code for the V8 variant of the SPARC
11554 architecture. The only difference from V7 code is that the
11555 compiler emits the integer multiply and integer divide
11556 instructions which exist in SPARC-V8 but not in SPARC-V7. With
11557 `-mcpu=supersparc', the compiler additionally optimizes it for the
11558 SuperSPARC chip, as used in the SPARCStation 10, 1000 and 2000
11561 With `-mcpu=sparclite', GCC generates code for the SPARClite
11562 variant of the SPARC architecture. This adds the integer
11563 multiply, integer divide step and scan (`ffs') instructions which
11564 exist in SPARClite but not in SPARC-V7. With `-mcpu=f930', the
11565 compiler additionally optimizes it for the Fujitsu MB86930 chip,
11566 which is the original SPARClite, with no FPU. With `-mcpu=f934',
11567 the compiler additionally optimizes it for the Fujitsu MB86934
11568 chip, which is the more recent SPARClite with FPU.
11570 With `-mcpu=sparclet', GCC generates code for the SPARClet variant
11571 of the SPARC architecture. This adds the integer multiply,
11572 multiply/accumulate, integer divide step and scan (`ffs')
11573 instructions which exist in SPARClet but not in SPARC-V7. With
11574 `-mcpu=tsc701', the compiler additionally optimizes it for the
11575 TEMIC SPARClet chip.
11577 With `-mcpu=v9', GCC generates code for the V9 variant of the SPARC
11578 architecture. This adds 64-bit integer and floating-point move
11579 instructions, 3 additional floating-point condition code registers
11580 and conditional move instructions. With `-mcpu=ultrasparc', the
11581 compiler additionally optimizes it for the Sun UltraSPARC I/II
11582 chips. With `-mcpu=ultrasparc3', the compiler additionally
11583 optimizes it for the Sun UltraSPARC III chip.
11586 Set the instruction scheduling parameters for machine type
11587 CPU_TYPE, but do not set the instruction set or register set that
11588 the option `-mcpu=CPU_TYPE' would.
11590 The same values for `-mcpu=CPU_TYPE' can be used for
11591 `-mtune=CPU_TYPE', but the only useful values are those that
11592 select a particular cpu implementation. Those are `cypress',
11593 `supersparc', `hypersparc', `f930', `f934', `sparclite86x',
11594 `tsc701', `ultrasparc', and `ultrasparc3'.
11598 With `-mv8plus', GCC generates code for the SPARC-V8+ ABI. The
11599 difference from the V8 ABI is that the global and out registers are
11600 considered 64-bit wide. This is enabled by default on Solaris in
11601 32-bit mode for all SPARC-V9 processors.
11605 With `-mvis', GCC generates code that takes advantage of the
11606 UltraSPARC Visual Instruction Set extensions. The default is
11609 These `-m' options are supported in addition to the above on SPARC-V9
11610 processors in 64-bit environments:
11613 Generate code for a processor running in little-endian mode. It
11614 is only available for a few configurations and most notably not on
11619 Generate code for a 32-bit or 64-bit environment. The 32-bit
11620 environment sets int, long and pointer to 32 bits. The 64-bit
11621 environment sets int to 32 bits and long and pointer to 64 bits.
11624 Generate code for the Medium/Low code model: 64-bit addresses,
11625 programs must be linked in the low 32 bits of memory. Programs
11626 can be statically or dynamically linked.
11629 Generate code for the Medium/Middle code model: 64-bit addresses,
11630 programs must be linked in the low 44 bits of memory, the text and
11631 data segments must be less than 2GB in size and the data segment
11632 must be located within 2GB of the text segment.
11635 Generate code for the Medium/Anywhere code model: 64-bit
11636 addresses, programs may be linked anywhere in memory, the text and
11637 data segments must be less than 2GB in size and the data segment
11638 must be located within 2GB of the text segment.
11640 `-mcmodel=embmedany'
11641 Generate code for the Medium/Anywhere code model for embedded
11642 systems: 64-bit addresses, the text and data segments must be less
11643 than 2GB in size, both starting anywhere in memory (determined at
11644 link time). The global register %g4 points to the base of the
11645 data segment. Programs are statically linked and PIC is not
11650 With `-mstack-bias', GCC assumes that the stack pointer, and frame
11651 pointer if present, are offset by -2047 which must be added back
11652 when making stack frame references. This is the default in 64-bit
11653 mode. Otherwise, assume no such offset is present.
11655 These switches are supported in addition to the above on Solaris:
11658 Add support for multithreading using the Solaris threads library.
11659 This option sets flags for both the preprocessor and linker. This
11660 option does not affect the thread safety of object code produced
11661 by the compiler or that of libraries supplied with it.
11664 Add support for multithreading using the POSIX threads library.
11665 This option sets flags for both the preprocessor and linker. This
11666 option does not affect the thread safety of object code produced
11667 by the compiler or that of libraries supplied with it.
11670 This is a synonym for `-pthreads'.
11673 File: gcc.info, Node: System V Options, Next: TMS320C3x/C4x Options, Prev: SPARC Options, Up: Submodel Options
11675 3.17.30 Options for System V
11676 ----------------------------
11678 These additional options are available on System V Release 4 for
11679 compatibility with other compilers on those systems:
11682 Create a shared object. It is recommended that `-symbolic' or
11683 `-shared' be used instead.
11686 Identify the versions of each tool used by the compiler, in a
11687 `.ident' assembler directive in the output.
11690 Refrain from adding `.ident' directives to the output file (this is
11694 Search the directories DIRS, and no others, for libraries
11695 specified with `-l'.
11698 Look in the directory DIR to find the M4 preprocessor. The
11699 assembler uses this option.
11702 File: gcc.info, Node: TMS320C3x/C4x Options, Next: V850 Options, Prev: System V Options, Up: Submodel Options
11704 3.17.31 TMS320C3x/C4x Options
11705 -----------------------------
11707 These `-m' options are defined for TMS320C3x/C4x implementations:
11710 Set the instruction set, register set, and instruction scheduling
11711 parameters for machine type CPU_TYPE. Supported values for
11712 CPU_TYPE are `c30', `c31', `c32', `c40', and `c44'. The default
11713 is `c40' to generate code for the TMS320C40.
11719 Generates code for the big or small memory model. The small memory
11720 model assumed that all data fits into one 64K word page. At
11721 run-time the data page (DP) register must be set to point to the
11722 64K page containing the .bss and .data program sections. The big
11723 memory model is the default and requires reloading of the DP
11724 register for every direct memory access.
11728 Allow (disallow) allocation of general integer operands into the
11729 block count register BK.
11733 Enable (disable) generation of code using decrement and branch,
11734 DBcond(D), instructions. This is enabled by default for the C4x.
11735 To be on the safe side, this is disabled for the C3x, since the
11736 maximum iteration count on the C3x is 2^23 + 1 (but who iterates
11737 loops more than 2^23 times on the C3x?). Note that GCC will try
11738 to reverse a loop so that it can utilize the decrement and branch
11739 instruction, but will give up if there is more than one memory
11740 reference in the loop. Thus a loop where the loop counter is
11741 decremented can generate slightly more efficient code, in cases
11742 where the RPTB instruction cannot be utilized.
11746 Force the DP register to be saved on entry to an interrupt service
11747 routine (ISR), reloaded to point to the data section, and restored
11748 on exit from the ISR. This should not be required unless someone
11749 has violated the small memory model by modifying the DP register,
11750 say within an object library.
11754 For the C3x use the 24-bit MPYI instruction for integer multiplies
11755 instead of a library call to guarantee 32-bit results. Note that
11756 if one of the operands is a constant, then the multiplication will
11757 be performed using shifts and adds. If the `-mmpyi' option is not
11758 specified for the C3x, then squaring operations are performed
11759 inline instead of a library call.
11763 The C3x/C4x FIX instruction to convert a floating point value to an
11764 integer value chooses the nearest integer less than or equal to the
11765 floating point value rather than to the nearest integer. Thus if
11766 the floating point number is negative, the result will be
11767 incorrectly truncated an additional code is necessary to detect
11768 and correct this case. This option can be used to disable
11769 generation of the additional code required to correct the result.
11773 Enable (disable) generation of repeat block sequences using the
11774 RPTB instruction for zero overhead looping. The RPTB construct is
11775 only used for innermost loops that do not call functions or jump
11776 across the loop boundaries. There is no advantage having nested
11777 RPTB loops due to the overhead required to save and restore the
11778 RC, RS, and RE registers. This is enabled by default with `-O2'.
11782 Enable (disable) the use of the single instruction repeat
11783 instruction RPTS. If a repeat block contains a single
11784 instruction, and the loop count can be guaranteed to be less than
11785 the value COUNT, GCC will emit a RPTS instruction instead of a
11786 RPTB. If no value is specified, then a RPTS will be emitted even
11787 if the loop count cannot be determined at compile time. Note that
11788 the repeated instruction following RPTS does not have to be
11789 reloaded from memory each iteration, thus freeing up the CPU buses
11790 for operands. However, since interrupts are blocked by this
11791 instruction, it is disabled by default.
11794 `-mno-loop-unsigned'
11795 The maximum iteration count when using RPTS and RPTB (and DB on
11796 the C40) is 2^31 + 1 since these instructions test if the
11797 iteration count is negative to terminate the loop. If the
11798 iteration count is unsigned there is a possibility than the 2^31 +
11799 1 maximum iteration count may be exceeded. This switch allows an
11800 unsigned iteration count.
11803 Try to emit an assembler syntax that the TI assembler (asm30) is
11804 happy with. This also enforces compatibility with the API
11805 employed by the TI C3x C compiler. For example, long doubles are
11806 passed as structures rather than in floating point registers.
11810 Generate code that uses registers (stack) for passing arguments to
11811 functions. By default, arguments are passed in registers where
11812 possible rather than by pushing arguments on to the stack.
11815 `-mno-parallel-insns'
11816 Allow the generation of parallel instructions. This is enabled by
11817 default with `-O2'.
11820 `-mno-parallel-mpy'
11821 Allow the generation of MPY||ADD and MPY||SUB parallel
11822 instructions, provided `-mparallel-insns' is also specified.
11823 These instructions have tight register constraints which can
11824 pessimize the code generation of large functions.
11828 File: gcc.info, Node: V850 Options, Next: VAX Options, Prev: TMS320C3x/C4x Options, Up: Submodel Options
11830 3.17.32 V850 Options
11831 --------------------
11833 These `-m' options are defined for V850 implementations:
11837 Treat all calls as being far away (near). If calls are assumed to
11838 be far away, the compiler will always load the functions address
11839 up into a register, and call indirect through the pointer.
11843 Do not optimize (do optimize) basic blocks that use the same index
11844 pointer 4 or more times to copy pointer into the `ep' register, and
11845 use the shorter `sld' and `sst' instructions. The `-mep' option
11846 is on by default if you optimize.
11848 `-mno-prolog-function'
11849 `-mprolog-function'
11850 Do not use (do use) external functions to save and restore
11851 registers at the prologue and epilogue of a function. The
11852 external functions are slower, but use less code space if more
11853 than one function saves the same number of registers. The
11854 `-mprolog-function' option is on by default if you optimize.
11857 Try to make the code as small as possible. At present, this just
11858 turns on the `-mep' and `-mprolog-function' options.
11861 Put static or global variables whose size is N bytes or less into
11862 the tiny data area that register `ep' points to. The tiny data
11863 area can hold up to 256 bytes in total (128 bytes for byte
11867 Put static or global variables whose size is N bytes or less into
11868 the small data area that register `gp' points to. The small data
11869 area can hold up to 64 kilobytes.
11872 Put static or global variables whose size is N bytes or less into
11873 the first 32 kilobytes of memory.
11876 Specify that the target processor is the V850.
11879 Generate code suitable for big switch tables. Use this option
11880 only if the assembler/linker complain about out of range branches
11881 within a switch table.
11884 This option will cause r2 and r5 to be used in the code generated
11885 by the compiler. This setting is the default.
11888 This option will cause r2 and r5 to be treated as fixed registers.
11891 Specify that the target processor is the V850E1. The preprocessor
11892 constants `__v850e1__' and `__v850e__' will be defined if this
11896 Specify that the target processor is the V850E. The preprocessor
11897 constant `__v850e__' will be defined if this option is used.
11899 If neither `-mv850' nor `-mv850e' nor `-mv850e1' are defined then
11900 a default target processor will be chosen and the relevant
11901 `__v850*__' preprocessor constant will be defined.
11903 The preprocessor constants `__v850' and `__v851__' are always
11904 defined, regardless of which processor variant is the target.
11907 This option will suppress generation of the CALLT instruction for
11908 the v850e and v850e1 flavors of the v850 architecture. The
11909 default is `-mno-disable-callt' which allows the CALLT instruction
11914 File: gcc.info, Node: VAX Options, Next: x86-64 Options, Prev: V850 Options, Up: Submodel Options
11916 3.17.33 VAX Options
11917 -------------------
11919 These `-m' options are defined for the VAX:
11922 Do not output certain jump instructions (`aobleq' and so on) that
11923 the Unix assembler for the VAX cannot handle across long ranges.
11926 Do output those jump instructions, on the assumption that you will
11927 assemble with the GNU assembler.
11930 Output code for g-format floating point numbers instead of
11934 File: gcc.info, Node: x86-64 Options, Next: Xstormy16 Options, Prev: VAX Options, Up: Submodel Options
11936 3.17.34 x86-64 Options
11937 ----------------------
11939 These are listed under *Note i386 and x86-64 Options::.
11942 File: gcc.info, Node: Xstormy16 Options, Next: Xtensa Options, Prev: x86-64 Options, Up: Submodel Options
11944 3.17.35 Xstormy16 Options
11945 -------------------------
11947 These options are defined for Xstormy16:
11950 Choose startup files and linker script suitable for the simulator.
11953 File: gcc.info, Node: Xtensa Options, Next: zSeries Options, Prev: Xstormy16 Options, Up: Submodel Options
11955 3.17.36 Xtensa Options
11956 ----------------------
11958 These options are supported for Xtensa targets:
11962 Enable or disable use of `CONST16' instructions for loading
11963 constant values. The `CONST16' instruction is currently not a
11964 standard option from Tensilica. When enabled, `CONST16'
11965 instructions are always used in place of the standard `L32R'
11966 instructions. The use of `CONST16' is enabled by default only if
11967 the `L32R' instruction is not available.
11971 Enable or disable use of fused multiply/add and multiply/subtract
11972 instructions in the floating-point option. This has no effect if
11973 the floating-point option is not also enabled. Disabling fused
11974 multiply/add and multiply/subtract instructions forces the
11975 compiler to use separate instructions for the multiply and
11976 add/subtract operations. This may be desirable in some cases
11977 where strict IEEE 754-compliant results are required: the fused
11978 multiply add/subtract instructions do not round the intermediate
11979 result, thereby producing results with _more_ bits of precision
11980 than specified by the IEEE standard. Disabling fused multiply
11981 add/subtract instructions also ensures that the program output is
11982 not sensitive to the compiler's ability to combine multiply and
11983 add/subtract operations.
11985 `-mtext-section-literals'
11986 `-mno-text-section-literals'
11987 Control the treatment of literal pools. The default is
11988 `-mno-text-section-literals', which places literals in a separate
11989 section in the output file. This allows the literal pool to be
11990 placed in a data RAM/ROM, and it also allows the linker to combine
11991 literal pools from separate object files to remove redundant
11992 literals and improve code size. With `-mtext-section-literals',
11993 the literals are interspersed in the text section in order to keep
11994 them as close as possible to their references. This may be
11995 necessary for large assembly files.
11998 `-mno-target-align'
11999 When this option is enabled, GCC instructs the assembler to
12000 automatically align instructions to reduce branch penalties at the
12001 expense of some code density. The assembler attempts to widen
12002 density instructions to align branch targets and the instructions
12003 following call instructions. If there are not enough preceding
12004 safe density instructions to align a target, no widening will be
12005 performed. The default is `-mtarget-align'. These options do not
12006 affect the treatment of auto-aligned instructions like `LOOP',
12007 which the assembler will always align, either by widening density
12008 instructions or by inserting no-op instructions.
12012 When this option is enabled, GCC instructs the assembler to
12013 translate direct calls to indirect calls unless it can determine
12014 that the target of a direct call is in the range allowed by the
12015 call instruction. This translation typically occurs for calls to
12016 functions in other source files. Specifically, the assembler
12017 translates a direct `CALL' instruction into an `L32R' followed by
12018 a `CALLX' instruction. The default is `-mno-longcalls'. This
12019 option should be used in programs where the call target can
12020 potentially be out of range. This option is implemented in the
12021 assembler, not the compiler, so the assembly code generated by GCC
12022 will still show direct call instructions--look at the disassembled
12023 object code to see the actual instructions. Note that the
12024 assembler will use an indirect call for every cross-file call, not
12025 just those that really will be out of range.
12028 File: gcc.info, Node: zSeries Options, Prev: Xtensa Options, Up: Submodel Options
12030 3.17.37 zSeries Options
12031 -----------------------
12033 These are listed under *Note S/390 and zSeries Options::.
12036 File: gcc.info, Node: Code Gen Options, Next: Environment Variables, Prev: Submodel Options, Up: Invoking GCC
12038 3.18 Options for Code Generation Conventions
12039 ============================================
12041 These machine-independent options control the interface conventions
12042 used in code generation.
12044 Most of them have both positive and negative forms; the negative form
12045 of `-ffoo' would be `-fno-foo'. In the table below, only one of the
12046 forms is listed--the one which is not the default. You can figure out
12047 the other form by either removing `no-' or adding it.
12050 For front-ends that support it, generate additional code to check
12051 that indices used to access arrays are within the declared range.
12052 This is currently only supported by the Java and Fortran
12053 front-ends, where this option defaults to true and false
12057 This option generates traps for signed overflow on addition,
12058 subtraction, multiplication operations.
12061 This option instructs the compiler to assume that signed arithmetic
12062 overflow of addition, subtraction and multiplication wraps around
12063 using twos-complement representation. This flag enables some
12064 optimizations and disables others. This option is enabled by
12065 default for the Java front-end, as required by the Java language
12069 Enable exception handling. Generates extra code needed to
12070 propagate exceptions. For some targets, this implies GCC will
12071 generate frame unwind information for all functions, which can
12072 produce significant data size overhead, although it does not
12073 affect execution. If you do not specify this option, GCC will
12074 enable it by default for languages like C++ which normally require
12075 exception handling, and disable it for languages like C that do
12076 not normally require it. However, you may need to enable this
12077 option when compiling C code that needs to interoperate properly
12078 with exception handlers written in C++. You may also wish to
12079 disable this option if you are compiling older C++ programs that
12080 don't use exception handling.
12082 `-fnon-call-exceptions'
12083 Generate code that allows trapping instructions to throw
12084 exceptions. Note that this requires platform-specific runtime
12085 support that does not exist everywhere. Moreover, it only allows
12086 _trapping_ instructions to throw exceptions, i.e. memory
12087 references or floating point instructions. It does not allow
12088 exceptions to be thrown from arbitrary signal handlers such as
12092 Similar to `-fexceptions', except that it will just generate any
12093 needed static data, but will not affect the generated code in any
12094 other way. You will normally not enable this option; instead, a
12095 language processor that needs this handling would enable it on
12098 `-fasynchronous-unwind-tables'
12099 Generate unwind table in dwarf2 format, if supported by target
12100 machine. The table is exact at each instruction boundary, so it
12101 can be used for stack unwinding from asynchronous events (such as
12102 debugger or garbage collector).
12104 `-fpcc-struct-return'
12105 Return "short" `struct' and `union' values in memory like longer
12106 ones, rather than in registers. This convention is less
12107 efficient, but it has the advantage of allowing intercallability
12108 between GCC-compiled files and files compiled with other
12109 compilers, particularly the Portable C Compiler (pcc).
12111 The precise convention for returning structures in memory depends
12112 on the target configuration macros.
12114 Short structures and unions are those whose size and alignment
12115 match that of some integer type.
12117 *Warning:* code compiled with the `-fpcc-struct-return' switch is
12118 not binary compatible with code compiled with the
12119 `-freg-struct-return' switch. Use it to conform to a non-default
12120 application binary interface.
12122 `-freg-struct-return'
12123 Return `struct' and `union' values in registers when possible.
12124 This is more efficient for small structures than
12125 `-fpcc-struct-return'.
12127 If you specify neither `-fpcc-struct-return' nor
12128 `-freg-struct-return', GCC defaults to whichever convention is
12129 standard for the target. If there is no standard convention, GCC
12130 defaults to `-fpcc-struct-return', except on targets where GCC is
12131 the principal compiler. In those cases, we can choose the
12132 standard, and we chose the more efficient register return
12135 *Warning:* code compiled with the `-freg-struct-return' switch is
12136 not binary compatible with code compiled with the
12137 `-fpcc-struct-return' switch. Use it to conform to a non-default
12138 application binary interface.
12141 Allocate to an `enum' type only as many bytes as it needs for the
12142 declared range of possible values. Specifically, the `enum' type
12143 will be equivalent to the smallest integer type which has enough
12146 *Warning:* the `-fshort-enums' switch causes GCC to generate code
12147 that is not binary compatible with code generated without that
12148 switch. Use it to conform to a non-default application binary
12152 Use the same size for `double' as for `float'.
12154 *Warning:* the `-fshort-double' switch causes GCC to generate code
12155 that is not binary compatible with code generated without that
12156 switch. Use it to conform to a non-default application binary
12160 Override the underlying type for `wchar_t' to be `short unsigned
12161 int' instead of the default for the target. This option is useful
12162 for building programs to run under WINE.
12164 *Warning:* the `-fshort-wchar' switch causes GCC to generate code
12165 that is not binary compatible with code generated without that
12166 switch. Use it to conform to a non-default application binary
12170 Requests that the data and non-`const' variables of this
12171 compilation be shared data rather than private data. The
12172 distinction makes sense only on certain operating systems, where
12173 shared data is shared between processes running the same program,
12174 while private data exists in one copy per process.
12177 In C, allocate even uninitialized global variables in the data
12178 section of the object file, rather than generating them as common
12179 blocks. This has the effect that if the same variable is declared
12180 (without `extern') in two different compilations, you will get an
12181 error when you link them. The only reason this might be useful is
12182 if you wish to verify that the program will work on other systems
12183 which always work this way.
12186 Ignore the `#ident' directive.
12188 `-finhibit-size-directive'
12189 Don't output a `.size' assembler directive, or anything else that
12190 would cause trouble if the function is split in the middle, and the
12191 two halves are placed at locations far apart in memory. This
12192 option is used when compiling `crtstuff.c'; you should not need to
12193 use it for anything else.
12196 Put extra commentary information in the generated assembly code to
12197 make it more readable. This option is generally only of use to
12198 those who actually need to read the generated assembly code
12199 (perhaps while debugging the compiler itself).
12201 `-fno-verbose-asm', the default, causes the extra information to
12202 be omitted and is useful when comparing two assembler files.
12205 Generate position-independent code (PIC) suitable for use in a
12206 shared library, if supported for the target machine. Such code
12207 accesses all constant addresses through a global offset table
12208 (GOT). The dynamic loader resolves the GOT entries when the
12209 program starts (the dynamic loader is not part of GCC; it is part
12210 of the operating system). If the GOT size for the linked
12211 executable exceeds a machine-specific maximum size, you get an
12212 error message from the linker indicating that `-fpic' does not
12213 work; in that case, recompile with `-fPIC' instead. (These
12214 maximums are 8k on the SPARC and 32k on the m68k and RS/6000. The
12215 386 has no such limit.)
12217 Position-independent code requires special support, and therefore
12218 works only on certain machines. For the 386, GCC supports PIC for
12219 System V but not for the Sun 386i. Code generated for the IBM
12220 RS/6000 is always position-independent.
12223 If supported for the target machine, emit position-independent
12224 code, suitable for dynamic linking and avoiding any limit on the
12225 size of the global offset table. This option makes a difference
12226 on the m68k, PowerPC and SPARC.
12228 Position-independent code requires special support, and therefore
12229 works only on certain machines.
12233 These options are similar to `-fpic' and `-fPIC', but generated
12234 position independent code can be only linked into executables.
12235 Usually these options are used when `-pie' GCC option will be used
12239 Do not use jump tables for switch statements even where it would be
12240 more efficient than other code generation strategies. This option
12241 is of use in conjunction with `-fpic' or `-fPIC' for building code
12242 which forms part of a dynamic linker and cannot reference the
12243 address of a jump table. On some targets, jump tables do not
12244 require a GOT and this option is not needed.
12247 Treat the register named REG as a fixed register; generated code
12248 should never refer to it (except perhaps as a stack pointer, frame
12249 pointer or in some other fixed role).
12251 REG must be the name of a register. The register names accepted
12252 are machine-specific and are defined in the `REGISTER_NAMES' macro
12253 in the machine description macro file.
12255 This flag does not have a negative form, because it specifies a
12259 Treat the register named REG as an allocable register that is
12260 clobbered by function calls. It may be allocated for temporaries
12261 or variables that do not live across a call. Functions compiled
12262 this way will not save and restore the register REG.
12264 It is an error to used this flag with the frame pointer or stack
12265 pointer. Use of this flag for other registers that have fixed
12266 pervasive roles in the machine's execution model will produce
12267 disastrous results.
12269 This flag does not have a negative form, because it specifies a
12273 Treat the register named REG as an allocable register saved by
12274 functions. It may be allocated even for temporaries or variables
12275 that live across a call. Functions compiled this way will save
12276 and restore the register REG if they use it.
12278 It is an error to used this flag with the frame pointer or stack
12279 pointer. Use of this flag for other registers that have fixed
12280 pervasive roles in the machine's execution model will produce
12281 disastrous results.
12283 A different sort of disaster will result from the use of this flag
12284 for a register in which function values may be returned.
12286 This flag does not have a negative form, because it specifies a
12289 `-fpack-struct[=N]'
12290 Without a value specified, pack all structure members together
12291 without holes. When a value is specified (which must be a small
12292 power of two), pack structure members according to this value,
12293 representing the maximum alignment (that is, objects with default
12294 alignment requirements larger than this will be output potentially
12295 unaligned at the next fitting location.
12297 *Warning:* the `-fpack-struct' switch causes GCC to generate code
12298 that is not binary compatible with code generated without that
12299 switch. Additionally, it makes the code suboptimal. Use it to
12300 conform to a non-default application binary interface.
12302 `-finstrument-functions'
12303 Generate instrumentation calls for entry and exit to functions.
12304 Just after function entry and just before function exit, the
12305 following profiling functions will be called with the address of
12306 the current function and its call site. (On some platforms,
12307 `__builtin_return_address' does not work beyond the current
12308 function, so the call site information may not be available to the
12309 profiling functions otherwise.)
12311 void __cyg_profile_func_enter (void *this_fn,
12313 void __cyg_profile_func_exit (void *this_fn,
12316 The first argument is the address of the start of the current
12317 function, which may be looked up exactly in the symbol table.
12319 This instrumentation is also done for functions expanded inline in
12320 other functions. The profiling calls will indicate where,
12321 conceptually, the inline function is entered and exited. This
12322 means that addressable versions of such functions must be
12323 available. If all your uses of a function are expanded inline,
12324 this may mean an additional expansion of code size. If you use
12325 `extern inline' in your C code, an addressable version of such
12326 functions must be provided. (This is normally the case anyways,
12327 but if you get lucky and the optimizer always expands the
12328 functions inline, you might have gotten away without providing
12331 A function may be given the attribute `no_instrument_function', in
12332 which case this instrumentation will not be done. This can be
12333 used, for example, for the profiling functions listed above,
12334 high-priority interrupt routines, and any functions from which the
12335 profiling functions cannot safely be called (perhaps signal
12336 handlers, if the profiling routines generate output or allocate
12340 Generate code to verify that you do not go beyond the boundary of
12341 the stack. You should specify this flag if you are running in an
12342 environment with multiple threads, but only rarely need to specify
12343 it in a single-threaded environment since stack overflow is
12344 automatically detected on nearly all systems if there is only one
12347 Note that this switch does not actually cause checking to be done;
12348 the operating system must do that. The switch causes generation
12349 of code to ensure that the operating system sees the stack being
12352 `-fstack-limit-register=REG'
12353 `-fstack-limit-symbol=SYM'
12355 Generate code to ensure that the stack does not grow beyond a
12356 certain value, either the value of a register or the address of a
12357 symbol. If the stack would grow beyond the value, a signal is
12358 raised. For most targets, the signal is raised before the stack
12359 overruns the boundary, so it is possible to catch the signal
12360 without taking special precautions.
12362 For instance, if the stack starts at absolute address `0x80000000'
12363 and grows downwards, you can use the flags
12364 `-fstack-limit-symbol=__stack_limit' and
12365 `-Wl,--defsym,__stack_limit=0x7ffe0000' to enforce a stack limit
12366 of 128KB. Note that this may only work with the GNU linker.
12369 `-fargument-noalias'
12370 `-fargument-noalias-global'
12371 Specify the possible relationships among parameters and between
12372 parameters and global data.
12374 `-fargument-alias' specifies that arguments (parameters) may alias
12375 each other and may alias global storage.
12376 `-fargument-noalias' specifies that arguments do not alias each
12377 other, but may alias global storage.
12378 `-fargument-noalias-global' specifies that arguments do not alias
12379 each other and do not alias global storage.
12381 Each language will automatically use whatever option is required by
12382 the language standard. You should not need to use these options
12385 `-fleading-underscore'
12386 This option and its counterpart, `-fno-leading-underscore',
12387 forcibly change the way C symbols are represented in the object
12388 file. One use is to help link with legacy assembly code.
12390 *Warning:* the `-fleading-underscore' switch causes GCC to
12391 generate code that is not binary compatible with code generated
12392 without that switch. Use it to conform to a non-default
12393 application binary interface. Not all targets provide complete
12394 support for this switch.
12396 `-ftls-model=MODEL'
12397 Alter the thread-local storage model to be used (*note
12398 Thread-Local::). The MODEL argument should be one of
12399 `global-dynamic', `local-dynamic', `initial-exec' or `local-exec'.
12401 The default without `-fpic' is `initial-exec'; with `-fpic' the
12402 default is `global-dynamic'.
12404 `-fvisibility=DEFAULT|INTERNAL|HIDDEN|PROTECTED'
12405 Set the default ELF image symbol visibility to the specified
12406 option--all symbols will be marked with this unless overridden
12407 within the code. Using this feature can very substantially
12408 improve linking and load times of shared object libraries, produce
12409 more optimized code, provide near-perfect API export and prevent
12410 symbol clashes. It is *strongly* recommended that you use this in
12411 any shared objects you distribute.
12413 Despite the nomenclature, `default' always means public ie;
12414 available to be linked against from outside the shared object.
12415 `protected' and `internal' are pretty useless in real-world usage
12416 so the only other commonly used option will be `hidden'. The
12417 default if `-fvisibility' isn't specified is `default', i.e., make
12418 every symbol public--this causes the same behavior as previous
12421 A good explanation of the benefits offered by ensuring ELF symbols
12422 have the correct visibility is given by "How To Write Shared
12423 Libraries" by Ulrich Drepper (which can be found at
12424 `http://people.redhat.com/~drepper/')--however a superior solution
12425 made possible by this option to marking things hidden when the
12426 default is public is to make the default hidden and mark things
12427 public. This is the norm with DLL's on Windows and with
12428 `-fvisibility=hidden' and `__attribute__
12429 ((visibility("default")))' instead of `__declspec(dllexport)' you
12430 get almost identical semantics with identical syntax. This is a
12431 great boon to those working with cross-platform projects.
12433 For those adding visibility support to existing code, you may find
12434 `#pragma GCC visibility' of use. This works by you enclosing the
12435 declarations you wish to set visibility for with (for example)
12436 `#pragma GCC visibility push(hidden)' and `#pragma GCC visibility
12437 pop'. Bear in mind that symbol visibility should be viewed *as
12438 part of the API interface contract* and thus all new code should
12439 always specify visibility when it is not the default ie;
12440 declarations only for use within the local DSO should *always* be
12441 marked explicitly as hidden as so to avoid PLT indirection
12442 overheads--making this abundantly clear also aids readability and
12443 self-documentation of the code. Note that due to ISO C++
12444 specification requirements, operator new and operator delete must
12445 always be of default visibility.
12447 An overview of these techniques, their benefits and how to use them
12448 is at `http://gcc.gnu.org/wiki/Visibility'.
12452 File: gcc.info, Node: Environment Variables, Next: Precompiled Headers, Prev: Code Gen Options, Up: Invoking GCC
12454 3.19 Environment Variables Affecting GCC
12455 ========================================
12457 This section describes several environment variables that affect how GCC
12458 operates. Some of them work by specifying directories or prefixes to
12459 use when searching for various kinds of files. Some are used to
12460 specify other aspects of the compilation environment.
12462 Note that you can also specify places to search using options such as
12463 `-B', `-I' and `-L' (*note Directory Options::). These take precedence
12464 over places specified using environment variables, which in turn take
12465 precedence over those specified by the configuration of GCC. *Note
12466 Controlling the Compilation Driver `gcc': (gccint)Driver.
12472 These environment variables control the way that GCC uses
12473 localization information that allow GCC to work with different
12474 national conventions. GCC inspects the locale categories
12475 `LC_CTYPE' and `LC_MESSAGES' if it has been configured to do so.
12476 These locale categories can be set to any value supported by your
12477 installation. A typical value is `en_GB.UTF-8' for English in the
12478 United Kingdom encoded in UTF-8.
12480 The `LC_CTYPE' environment variable specifies character
12481 classification. GCC uses it to determine the character boundaries
12482 in a string; this is needed for some multibyte encodings that
12483 contain quote and escape characters that would otherwise be
12484 interpreted as a string end or escape.
12486 The `LC_MESSAGES' environment variable specifies the language to
12487 use in diagnostic messages.
12489 If the `LC_ALL' environment variable is set, it overrides the value
12490 of `LC_CTYPE' and `LC_MESSAGES'; otherwise, `LC_CTYPE' and
12491 `LC_MESSAGES' default to the value of the `LANG' environment
12492 variable. If none of these variables are set, GCC defaults to
12493 traditional C English behavior.
12496 If `TMPDIR' is set, it specifies the directory to use for temporary
12497 files. GCC uses temporary files to hold the output of one stage of
12498 compilation which is to be used as input to the next stage: for
12499 example, the output of the preprocessor, which is the input to the
12503 If `GCC_EXEC_PREFIX' is set, it specifies a prefix to use in the
12504 names of the subprograms executed by the compiler. No slash is
12505 added when this prefix is combined with the name of a subprogram,
12506 but you can specify a prefix that ends with a slash if you wish.
12508 If `GCC_EXEC_PREFIX' is not set, GCC will attempt to figure out an
12509 appropriate prefix to use based on the pathname it was invoked
12512 If GCC cannot find the subprogram using the specified prefix, it
12513 tries looking in the usual places for the subprogram.
12515 The default value of `GCC_EXEC_PREFIX' is `PREFIX/lib/gcc/' where
12516 PREFIX is the value of `prefix' when you ran the `configure'
12519 Other prefixes specified with `-B' take precedence over this
12522 This prefix is also used for finding files such as `crt0.o' that
12523 are used for linking.
12525 In addition, the prefix is used in an unusual way in finding the
12526 directories to search for header files. For each of the standard
12527 directories whose name normally begins with `/usr/local/lib/gcc'
12528 (more precisely, with the value of `GCC_INCLUDE_DIR'), GCC tries
12529 replacing that beginning with the specified prefix to produce an
12530 alternate directory name. Thus, with `-Bfoo/', GCC will search
12531 `foo/bar' where it would normally search `/usr/local/lib/bar'.
12532 These alternate directories are searched first; the standard
12533 directories come next.
12536 The value of `COMPILER_PATH' is a colon-separated list of
12537 directories, much like `PATH'. GCC tries the directories thus
12538 specified when searching for subprograms, if it can't find the
12539 subprograms using `GCC_EXEC_PREFIX'.
12542 The value of `LIBRARY_PATH' is a colon-separated list of
12543 directories, much like `PATH'. When configured as a native
12544 compiler, GCC tries the directories thus specified when searching
12545 for special linker files, if it can't find them using
12546 `GCC_EXEC_PREFIX'. Linking using GCC also uses these directories
12547 when searching for ordinary libraries for the `-l' option (but
12548 directories specified with `-L' come first).
12551 This variable is used to pass locale information to the compiler.
12552 One way in which this information is used is to determine the
12553 character set to be used when character literals, string literals
12554 and comments are parsed in C and C++. When the compiler is
12555 configured to allow multibyte characters, the following values for
12556 `LANG' are recognized:
12559 Recognize JIS characters.
12562 Recognize SJIS characters.
12565 Recognize EUCJP characters.
12567 If `LANG' is not defined, or if it has some other value, then the
12568 compiler will use mblen and mbtowc as defined by the default
12569 locale to recognize and translate multibyte characters.
12571 Some additional environments variables affect the behavior of the
12576 `CPLUS_INCLUDE_PATH'
12577 `OBJC_INCLUDE_PATH'
12578 Each variable's value is a list of directories separated by a
12579 special character, much like `PATH', in which to look for header
12580 files. The special character, `PATH_SEPARATOR', is
12581 target-dependent and determined at GCC build time. For Microsoft
12582 Windows-based targets it is a semicolon, and for almost all other
12583 targets it is a colon.
12585 `CPATH' specifies a list of directories to be searched as if
12586 specified with `-I', but after any paths given with `-I' options
12587 on the command line. This environment variable is used regardless
12588 of which language is being preprocessed.
12590 The remaining environment variables apply only when preprocessing
12591 the particular language indicated. Each specifies a list of
12592 directories to be searched as if specified with `-isystem', but
12593 after any paths given with `-isystem' options on the command line.
12595 In all these variables, an empty element instructs the compiler to
12596 search its current working directory. Empty elements can appear
12597 at the beginning or end of a path. For instance, if the value of
12598 `CPATH' is `:/special/include', that has the same effect as
12599 `-I. -I/special/include'.
12601 `DEPENDENCIES_OUTPUT'
12602 If this variable is set, its value specifies how to output
12603 dependencies for Make based on the non-system header files
12604 processed by the compiler. System header files are ignored in the
12607 The value of `DEPENDENCIES_OUTPUT' can be just a file name, in
12608 which case the Make rules are written to that file, guessing the
12609 target name from the source file name. Or the value can have the
12610 form `FILE TARGET', in which case the rules are written to file
12611 FILE using TARGET as the target name.
12613 In other words, this environment variable is equivalent to
12614 combining the options `-MM' and `-MF' (*note Preprocessor
12615 Options::), with an optional `-MT' switch too.
12617 `SUNPRO_DEPENDENCIES'
12618 This variable is the same as `DEPENDENCIES_OUTPUT' (see above),
12619 except that system header files are not ignored, so it implies
12620 `-M' rather than `-MM'. However, the dependence on the main input
12621 file is omitted. *Note Preprocessor Options::.
12624 File: gcc.info, Node: Precompiled Headers, Next: Running Protoize, Prev: Environment Variables, Up: Invoking GCC
12626 3.20 Using Precompiled Headers
12627 ==============================
12629 Often large projects have many header files that are included in every
12630 source file. The time the compiler takes to process these header files
12631 over and over again can account for nearly all of the time required to
12632 build the project. To make builds faster, GCC allows users to
12633 `precompile' a header file; then, if builds can use the precompiled
12634 header file they will be much faster.
12636 To create a precompiled header file, simply compile it as you would any
12637 other file, if necessary using the `-x' option to make the driver treat
12638 it as a C or C++ header file. You will probably want to use a tool
12639 like `make' to keep the precompiled header up-to-date when the headers
12640 it contains change.
12642 A precompiled header file will be searched for when `#include' is seen
12643 in the compilation. As it searches for the included file (*note Search
12644 Path: (cpp)Search Path.) the compiler looks for a precompiled header in
12645 each directory just before it looks for the include file in that
12646 directory. The name searched for is the name specified in the
12647 `#include' with `.gch' appended. If the precompiled header file can't
12648 be used, it is ignored.
12650 For instance, if you have `#include "all.h"', and you have `all.h.gch'
12651 in the same directory as `all.h', then the precompiled header file will
12652 be used if possible, and the original header will be used otherwise.
12654 Alternatively, you might decide to put the precompiled header file in a
12655 directory and use `-I' to ensure that directory is searched before (or
12656 instead of) the directory containing the original header. Then, if you
12657 want to check that the precompiled header file is always used, you can
12658 put a file of the same name as the original header in this directory
12659 containing an `#error' command.
12661 This also works with `-include'. So yet another way to use
12662 precompiled headers, good for projects not designed with precompiled
12663 header files in mind, is to simply take most of the header files used by
12664 a project, include them from another header file, precompile that header
12665 file, and `-include' the precompiled header. If the header files have
12666 guards against multiple inclusion, they will be skipped because they've
12667 already been included (in the precompiled header).
12669 If you need to precompile the same header file for different
12670 languages, targets, or compiler options, you can instead make a
12671 _directory_ named like `all.h.gch', and put each precompiled header in
12672 the directory, perhaps using `-o'. It doesn't matter what you call the
12673 files in the directory, every precompiled header in the directory will
12674 be considered. The first precompiled header encountered in the
12675 directory that is valid for this compilation will be used; they're
12676 searched in no particular order.
12678 There are many other possibilities, limited only by your imagination,
12679 good sense, and the constraints of your build system.
12681 A precompiled header file can be used only when these conditions apply:
12683 * Only one precompiled header can be used in a particular
12686 * A precompiled header can't be used once the first C token is seen.
12687 You can have preprocessor directives before a precompiled header;
12688 you can even include a precompiled header from inside another
12689 header, so long as there are no C tokens before the `#include'.
12691 * The precompiled header file must be produced for the same language
12692 as the current compilation. You can't use a C precompiled header
12693 for a C++ compilation.
12695 * The precompiled header file must have been produced by the same
12696 compiler binary as the current compilation is using.
12698 * Any macros defined before the precompiled header is included must
12699 either be defined in the same way as when the precompiled header
12700 was generated, or must not affect the precompiled header, which
12701 usually means that they don't appear in the precompiled header at
12704 The `-D' option is one way to define a macro before a precompiled
12705 header is included; using a `#define' can also do it. There are
12706 also some options that define macros implicitly, like `-O' and
12707 `-Wdeprecated'; the same rule applies to macros defined this way.
12709 * If debugging information is output when using the precompiled
12710 header, using `-g' or similar, the same kind of debugging
12711 information must have been output when building the precompiled
12712 header. However, a precompiled header built using `-g' can be
12713 used in a compilation when no debugging information is being
12716 * The same `-m' options must generally be used when building and
12717 using the precompiled header. *Note Submodel Options::, for any
12718 cases where this rule is relaxed.
12720 * Each of the following options must be the same when building and
12721 using the precompiled header:
12723 -fexceptions -funit-at-a-time
12725 * Some other command-line options starting with `-f', `-p', or `-O'
12726 must be defined in the same way as when the precompiled header was
12727 generated. At present, it's not clear which options are safe to
12728 change and which are not; the safest choice is to use exactly the
12729 same options when generating and using the precompiled header.
12730 The following are known to be safe:
12732 -fmessage-length= -fpreprocessed
12733 -fsched-interblock -fsched-spec -fsched-spec-load -fsched-spec-load-dangerous
12734 -fsched-verbose=<number> -fschedule-insns -fvisibility=
12738 For all of these except the last, the compiler will automatically
12739 ignore the precompiled header if the conditions aren't met. If you
12740 find an option combination that doesn't work and doesn't cause the
12741 precompiled header to be ignored, please consider filing a bug report,
12744 If you do use differing options when generating and using the
12745 precompiled header, the actual behavior will be a mixture of the
12746 behavior for the options. For instance, if you use `-g' to generate
12747 the precompiled header but not when using it, you may or may not get
12748 debugging information for routines in the precompiled header.
12751 File: gcc.info, Node: Running Protoize, Prev: Precompiled Headers, Up: Invoking GCC
12753 3.21 Running Protoize
12754 =====================
12756 The program `protoize' is an optional part of GCC. You can use it to
12757 add prototypes to a program, thus converting the program to ISO C in
12758 one respect. The companion program `unprotoize' does the reverse: it
12759 removes argument types from any prototypes that are found.
12761 When you run these programs, you must specify a set of source files as
12762 command line arguments. The conversion programs start out by compiling
12763 these files to see what functions they define. The information gathered
12764 about a file FOO is saved in a file named `FOO.X'.
12766 After scanning comes actual conversion. The specified files are all
12767 eligible to be converted; any files they include (whether sources or
12768 just headers) are eligible as well.
12770 But not all the eligible files are converted. By default, `protoize'
12771 and `unprotoize' convert only source and header files in the current
12772 directory. You can specify additional directories whose files should
12773 be converted with the `-d DIRECTORY' option. You can also specify
12774 particular files to exclude with the `-x FILE' option. A file is
12775 converted if it is eligible, its directory name matches one of the
12776 specified directory names, and its name within the directory has not
12779 Basic conversion with `protoize' consists of rewriting most function
12780 definitions and function declarations to specify the types of the
12781 arguments. The only ones not rewritten are those for varargs functions.
12783 `protoize' optionally inserts prototype declarations at the beginning
12784 of the source file, to make them available for any calls that precede
12785 the function's definition. Or it can insert prototype declarations
12786 with block scope in the blocks where undeclared functions are called.
12788 Basic conversion with `unprotoize' consists of rewriting most function
12789 declarations to remove any argument types, and rewriting function
12790 definitions to the old-style pre-ISO form.
12792 Both conversion programs print a warning for any function declaration
12793 or definition that they can't convert. You can suppress these warnings
12796 The output from `protoize' or `unprotoize' replaces the original
12797 source file. The original file is renamed to a name ending with
12798 `.save' (for DOS, the saved filename ends in `.sav' without the
12799 original `.c' suffix). If the `.save' (`.sav' for DOS) file already
12800 exists, then the source file is simply discarded.
12802 `protoize' and `unprotoize' both depend on GCC itself to scan the
12803 program and collect information about the functions it uses. So
12804 neither of these programs will work until GCC is installed.
12806 Here is a table of the options you can use with `protoize' and
12807 `unprotoize'. Each option works with both programs unless otherwise
12811 Look for the file `SYSCALLS.c.X' in DIRECTORY, instead of the
12812 usual directory (normally `/usr/local/lib'). This file contains
12813 prototype information about standard system functions. This option
12814 applies only to `protoize'.
12816 `-c COMPILATION-OPTIONS'
12817 Use COMPILATION-OPTIONS as the options when running `gcc' to
12818 produce the `.X' files. The special option `-aux-info' is always
12819 passed in addition, to tell `gcc' to write a `.X' file.
12821 Note that the compilation options must be given as a single
12822 argument to `protoize' or `unprotoize'. If you want to specify
12823 several `gcc' options, you must quote the entire set of
12824 compilation options to make them a single word in the shell.
12826 There are certain `gcc' arguments that you cannot use, because they
12827 would produce the wrong kind of output. These include `-g', `-O',
12828 `-c', `-S', and `-o' If you include these in the
12829 COMPILATION-OPTIONS, they are ignored.
12832 Rename files to end in `.C' (`.cc' for DOS-based file systems)
12833 instead of `.c'. This is convenient if you are converting a C
12834 program to C++. This option applies only to `protoize'.
12837 Add explicit global declarations. This means inserting explicit
12838 declarations at the beginning of each source file for each function
12839 that is called in the file and was not declared. These
12840 declarations precede the first function definition that contains a
12841 call to an undeclared function. This option applies only to
12845 Indent old-style parameter declarations with the string STRING.
12846 This option applies only to `protoize'.
12848 `unprotoize' converts prototyped function definitions to old-style
12849 function definitions, where the arguments are declared between the
12850 argument list and the initial `{'. By default, `unprotoize' uses
12851 five spaces as the indentation. If you want to indent with just
12852 one space instead, use `-i " "'.
12855 Keep the `.X' files. Normally, they are deleted after conversion
12859 Add explicit local declarations. `protoize' with `-l' inserts a
12860 prototype declaration for each function in each block which calls
12861 the function without any declaration. This option applies only to
12865 Make no real changes. This mode just prints information about the
12866 conversions that would have been done without `-n'.
12869 Make no `.save' files. The original files are simply deleted.
12870 Use this option with caution.
12873 Use the program PROGRAM as the compiler. Normally, the name `gcc'
12877 Work quietly. Most warnings are suppressed.
12880 Print the version number, just like `-v' for `gcc'.
12882 If you need special compiler options to compile one of your program's
12883 source files, then you should generate that file's `.X' file specially,
12884 by running `gcc' on that source file with the appropriate options and
12885 the option `-aux-info'. Then run `protoize' on the entire set of
12886 files. `protoize' will use the existing `.X' file because it is newer
12887 than the source file. For example:
12889 gcc -Dfoo=bar file1.c -aux-info file1.X
12892 You need to include the special files along with the rest in the
12893 `protoize' command, even though their `.X' files already exist, because
12894 otherwise they won't get converted.
12896 *Note Protoize Caveats::, for more information on how to use
12897 `protoize' successfully.
12900 File: gcc.info, Node: C Implementation, Next: C Extensions, Prev: Invoking GCC, Up: Top
12902 4 C Implementation-defined behavior
12903 ***********************************
12905 A conforming implementation of ISO C is required to document its choice
12906 of behavior in each of the areas that are designated "implementation
12907 defined". The following lists all such areas, along with the section
12908 numbers from the ISO/IEC 9899:1990 and ISO/IEC 9899:1999 standards.
12909 Some areas are only implementation-defined in one version of the
12912 Some choices depend on the externally determined ABI for the platform
12913 (including standard character encodings) which GCC follows; these are
12914 listed as "determined by ABI" below. *Note Binary Compatibility:
12915 Compatibility, and `http://gcc.gnu.org/readings.html'. Some choices
12916 are documented in the preprocessor manual. *Note
12917 Implementation-defined behavior: (cpp)Implementation-defined behavior.
12918 Some choices are made by the library and operating system (or other
12919 environment when compiling for a freestanding environment); refer to
12920 their documentation for details.
12924 * Translation implementation::
12925 * Environment implementation::
12926 * Identifiers implementation::
12927 * Characters implementation::
12928 * Integers implementation::
12929 * Floating point implementation::
12930 * Arrays and pointers implementation::
12931 * Hints implementation::
12932 * Structures unions enumerations and bit-fields implementation::
12933 * Qualifiers implementation::
12934 * Declarators implementation::
12935 * Statements implementation::
12936 * Preprocessing directives implementation::
12937 * Library functions implementation::
12938 * Architecture implementation::
12939 * Locale-specific behavior implementation::
12942 File: gcc.info, Node: Translation implementation, Next: Environment implementation, Up: C Implementation
12947 * `How a diagnostic is identified (C90 3.7, C99 3.10, C90 and C99
12950 Diagnostics consist of all the output sent to stderr by GCC.
12952 * `Whether each nonempty sequence of white-space characters other
12953 than new-line is retained or replaced by one space character in
12954 translation phase 3 (C90 and C99 5.1.1.2).'
12956 *Note Implementation-defined behavior: (cpp)Implementation-defined
12961 File: gcc.info, Node: Environment implementation, Next: Identifiers implementation, Prev: Translation implementation, Up: C Implementation
12966 The behavior of most of these points are dependent on the implementation
12967 of the C library, and are not defined by GCC itself.
12969 * `The mapping between physical source file multibyte characters and
12970 the source character set in translation phase 1 (C90 and C99
12973 *Note Implementation-defined behavior: (cpp)Implementation-defined
12978 File: gcc.info, Node: Identifiers implementation, Next: Characters implementation, Prev: Environment implementation, Up: C Implementation
12983 * `Which additional multibyte characters may appear in identifiers
12984 and their correspondence to universal character names (C99 6.4.2).'
12986 *Note Implementation-defined behavior: (cpp)Implementation-defined
12989 * `The number of significant initial characters in an identifier
12990 (C90 6.1.2, C90 and C99 5.2.4.1, C99 6.4.2).'
12992 For internal names, all characters are significant. For external
12993 names, the number of significant characters are defined by the
12994 linker; for almost all targets, all characters are significant.
12996 * `Whether case distinctions are significant in an identifier with
12997 external linkage (C90 6.1.2).'
12999 This is a property of the linker. C99 requires that case
13000 distinctions are always significant in identifiers with external
13001 linkage and systems without this property are not supported by GCC.
13005 File: gcc.info, Node: Characters implementation, Next: Integers implementation, Prev: Identifiers implementation, Up: C Implementation
13010 * `The number of bits in a byte (C90 3.4, C99 3.6).'
13014 * `The values of the members of the execution character set (C90 and
13019 * `The unique value of the member of the execution character set
13020 produced for each of the standard alphabetic escape sequences (C90
13025 * `The value of a `char' object into which has been stored any
13026 character other than a member of the basic execution character set
13027 (C90 6.1.2.5, C99 6.2.5).'
13031 * `Which of `signed char' or `unsigned char' has the same range,
13032 representation, and behavior as "plain" `char' (C90 6.1.2.5, C90
13033 6.2.1.1, C99 6.2.5, C99 6.3.1.1).'
13035 Determined by ABI. The options `-funsigned-char' and
13036 `-fsigned-char' change the default. *Note Options Controlling C
13037 Dialect: C Dialect Options.
13039 * `The mapping of members of the source character set (in character
13040 constants and string literals) to members of the execution
13041 character set (C90 6.1.3.4, C99 6.4.4.4, C90 and C99 5.1.1.2).'
13045 * `The value of an integer character constant containing more than
13046 one character or containing a character or escape sequence that
13047 does not map to a single-byte execution character (C90 6.1.3.4,
13050 *Note Implementation-defined behavior: (cpp)Implementation-defined
13053 * `The value of a wide character constant containing more than one
13054 multibyte character, or containing a multibyte character or escape
13055 sequence not represented in the extended execution character set
13056 (C90 6.1.3.4, C99 6.4.4.4).'
13058 *Note Implementation-defined behavior: (cpp)Implementation-defined
13061 * `The current locale used to convert a wide character constant
13062 consisting of a single multibyte character that maps to a member
13063 of the extended execution character set into a corresponding wide
13064 character code (C90 6.1.3.4, C99 6.4.4.4).'
13066 *Note Implementation-defined behavior: (cpp)Implementation-defined
13069 * `The current locale used to convert a wide string literal into
13070 corresponding wide character codes (C90 6.1.4, C99 6.4.5).'
13072 *Note Implementation-defined behavior: (cpp)Implementation-defined
13075 * `The value of a string literal containing a multibyte character or
13076 escape sequence not represented in the execution character set
13077 (C90 6.1.4, C99 6.4.5).'
13079 *Note Implementation-defined behavior: (cpp)Implementation-defined
13083 File: gcc.info, Node: Integers implementation, Next: Floating point implementation, Prev: Characters implementation, Up: C Implementation
13088 * `Any extended integer types that exist in the implementation (C99
13091 GCC does not support any extended integer types.
13093 * `Whether signed integer types are represented using sign and
13094 magnitude, two's complement, or one's complement, and whether the
13095 extraordinary value is a trap representation or an ordinary value
13098 GCC supports only two's complement integer types, and all bit
13099 patterns are ordinary values.
13101 * `The rank of any extended integer type relative to another extended
13102 integer type with the same precision (C99 6.3.1.1).'
13104 GCC does not support any extended integer types.
13106 * `The result of, or the signal raised by, converting an integer to a
13107 signed integer type when the value cannot be represented in an
13108 object of that type (C90 6.2.1.2, C99 6.3.1.3).'
13110 For conversion to a type of width N, the value is reduced modulo
13111 2^N to be within range of the type; no signal is raised.
13113 * `The results of some bitwise operations on signed integers (C90
13116 Bitwise operators act on the representation of the value including
13117 both the sign and value bits, where the sign bit is considered
13118 immediately above the highest-value value bit. Signed `>>' acts
13119 on negative numbers by sign extension.
13121 GCC does not use the latitude given in C99 only to treat certain
13122 aspects of signed `<<' as undefined, but this is subject to change.
13124 * `The sign of the remainder on integer division (C90 6.3.5).'
13126 GCC always follows the C99 requirement that the result of division
13127 is truncated towards zero.
13131 File: gcc.info, Node: Floating point implementation, Next: Arrays and pointers implementation, Prev: Integers implementation, Up: C Implementation
13136 * `The accuracy of the floating-point operations and of the library
13137 functions in `<math.h>' and `<complex.h>' that return
13138 floating-point results (C90 and C99 5.2.4.2.2).'
13140 The accuracy is unknown.
13142 * `The rounding behaviors characterized by non-standard values of
13143 `FLT_ROUNDS' (C90 and C99 5.2.4.2.2).'
13145 GCC does not use such values.
13147 * `The evaluation methods characterized by non-standard negative
13148 values of `FLT_EVAL_METHOD' (C99 5.2.4.2.2).'
13150 GCC does not use such values.
13152 * `The direction of rounding when an integer is converted to a
13153 floating-point number that cannot exactly represent the original
13154 value (C90 6.2.1.3, C99 6.3.1.4).'
13156 C99 Annex F is followed.
13158 * `The direction of rounding when a floating-point number is
13159 converted to a narrower floating-point number (C90 6.2.1.4, C99
13162 C99 Annex F is followed.
13164 * `How the nearest representable value or the larger or smaller
13165 representable value immediately adjacent to the nearest
13166 representable value is chosen for certain floating constants (C90
13167 6.1.3.1, C99 6.4.4.2).'
13169 C99 Annex F is followed.
13171 * `Whether and how floating expressions are contracted when not
13172 disallowed by the `FP_CONTRACT' pragma (C99 6.5).'
13174 Expressions are currently only contracted if
13175 `-funsafe-math-optimizations' or `-ffast-math' are used. This is
13178 * `The default state for the `FENV_ACCESS' pragma (C99 7.6.1).'
13180 This pragma is not implemented, but the default is to "off" unless
13181 `-frounding-math' is used in which case it is "on".
13183 * `Additional floating-point exceptions, rounding modes,
13184 environments, and classifications, and their macro names (C99 7.6,
13187 This is dependent on the implementation of the C library, and is
13188 not defined by GCC itself.
13190 * `The default state for the `FP_CONTRACT' pragma (C99 7.12.2).'
13192 This pragma is not implemented. Expressions are currently only
13193 contracted if `-funsafe-math-optimizations' or `-ffast-math' are
13194 used. This is subject to change.
13196 * `Whether the "inexact" floating-point exception can be raised when
13197 the rounded result actually does equal the mathematical result in
13198 an IEC 60559 conformant implementation (C99 F.9).'
13200 This is dependent on the implementation of the C library, and is
13201 not defined by GCC itself.
13203 * `Whether the "underflow" (and "inexact") floating-point exception
13204 can be raised when a result is tiny but not inexact in an IEC
13205 60559 conformant implementation (C99 F.9).'
13207 This is dependent on the implementation of the C library, and is
13208 not defined by GCC itself.
13212 File: gcc.info, Node: Arrays and pointers implementation, Next: Hints implementation, Prev: Floating point implementation, Up: C Implementation
13214 4.7 Arrays and pointers
13215 =======================
13217 * `The result of converting a pointer to an integer or vice versa
13218 (C90 6.3.4, C99 6.3.2.3).'
13220 A cast from pointer to integer discards most-significant bits if
13221 the pointer representation is larger than the integer type,
13222 sign-extends(1) if the pointer representation is smaller than the
13223 integer type, otherwise the bits are unchanged.
13225 A cast from integer to pointer discards most-significant bits if
13226 the pointer representation is smaller than the integer type,
13227 extends according to the signedness of the integer type if the
13228 pointer representation is larger than the integer type, otherwise
13229 the bits are unchanged.
13231 When casting from pointer to integer and back again, the resulting
13232 pointer must reference the same object as the original pointer,
13233 otherwise the behavior is undefined. That is, one may not use
13234 integer arithmetic to avoid the undefined behavior of pointer
13235 arithmetic as proscribed in C99 6.5.6/8.
13237 * `The size of the result of subtracting two pointers to elements of
13238 the same array (C90 6.3.6, C99 6.5.6).'
13240 The value is as specified in the standard and the type is
13241 determined by the ABI.
13244 ---------- Footnotes ----------
13246 (1) Future versions of GCC may zero-extend, or use a target-defined
13247 `ptr_extend' pattern. Do not rely on sign extension.
13250 File: gcc.info, Node: Hints implementation, Next: Structures unions enumerations and bit-fields implementation, Prev: Arrays and pointers implementation, Up: C Implementation
13255 * `The extent to which suggestions made by using the `register'
13256 storage-class specifier are effective (C90 6.5.1, C99 6.7.1).'
13258 The `register' specifier affects code generation only in these
13261 * When used as part of the register variable extension, see
13262 *Note Explicit Reg Vars::.
13264 * When `-O0' is in use, the compiler allocates distinct stack
13265 memory for all variables that do not have the `register'
13266 storage-class specifier; if `register' is specified, the
13267 variable may have a shorter lifespan than the code would
13268 indicate and may never be placed in memory.
13270 * On some rare x86 targets, `setjmp' doesn't save the registers
13271 in all circumstances. In those cases, GCC doesn't allocate
13272 any variables in registers unless they are marked `register'.
13275 * `The extent to which suggestions made by using the inline function
13276 specifier are effective (C99 6.7.4).'
13278 GCC will not inline any functions if the `-fno-inline' option is
13279 used or if `-O0' is used. Otherwise, GCC may still be unable to
13280 inline a function for many reasons; the `-Winline' option may be
13281 used to determine if a function has not been inlined and why not.
13285 File: gcc.info, Node: Structures unions enumerations and bit-fields implementation, Next: Qualifiers implementation, Prev: Hints implementation, Up: C Implementation
13287 4.9 Structures, unions, enumerations, and bit-fields
13288 ====================================================
13290 * `A member of a union object is accessed using a member of a
13291 different type (C90 6.3.2.3).'
13293 The relevant bytes of the representation of the object are treated
13294 as an object of the type used for the access. This may be a trap
13297 * `Whether a "plain" `int' bit-field is treated as a `signed int'
13298 bit-field or as an `unsigned int' bit-field (C90 6.5.2, C90
13299 6.5.2.1, C99 6.7.2, C99 6.7.2.1).'
13301 By default it is treated as `signed int' but this may be changed
13302 by the `-funsigned-bitfields' option.
13304 * `Allowable bit-field types other than `_Bool', `signed int', and
13305 `unsigned int' (C99 6.7.2.1).'
13307 No other types are permitted in strictly conforming mode.
13309 * `Whether a bit-field can straddle a storage-unit boundary (C90
13310 6.5.2.1, C99 6.7.2.1).'
13314 * `The order of allocation of bit-fields within a unit (C90 6.5.2.1,
13319 * `The alignment of non-bit-field members of structures (C90
13320 6.5.2.1, C99 6.7.2.1).'
13324 * `The integer type compatible with each enumerated type (C90
13325 6.5.2.2, C99 6.7.2.2).'
13327 Normally, the type is `unsigned int' if there are no negative
13328 values in the enumeration, otherwise `int'. If `-fshort-enums' is
13329 specified, then if there are negative values it is the first of
13330 `signed char', `short' and `int' that can represent all the
13331 values, otherwise it is the first of `unsigned char', `unsigned
13332 short' and `unsigned int' that can represent all the values.
13334 On some targets, `-fshort-enums' is the default; this is
13335 determined by the ABI.
13339 File: gcc.info, Node: Qualifiers implementation, Next: Declarators implementation, Prev: Structures unions enumerations and bit-fields implementation, Up: C Implementation
13344 * `What constitutes an access to an object that has
13345 volatile-qualified type (C90 6.5.3, C99 6.7.3).'
13347 Such an object is normally accessed by pointers and used for
13348 accessing hardware. In most expressions, it is intuitively
13349 obvious what is a read and what is a write. For example
13351 volatile int *dst = SOMEVALUE;
13352 volatile int *src = SOMEOTHERVALUE;
13355 will cause a read of the volatile object pointed to by SRC and
13356 store the value into the volatile object pointed to by DST. There
13357 is no guarantee that these reads and writes are atomic, especially
13358 for objects larger than `int'.
13360 However, if the volatile storage is not being modified, and the
13361 value of the volatile storage is not used, then the situation is
13362 less obvious. For example
13364 volatile int *src = SOMEVALUE;
13367 According to the C standard, such an expression is an rvalue whose
13368 type is the unqualified version of its original type, i.e. `int'.
13369 Whether GCC interprets this as a read of the volatile object being
13370 pointed to or only as a request to evaluate the expression for its
13371 side-effects depends on this type.
13373 If it is a scalar type, or on most targets an aggregate type whose
13374 only member object is of a scalar type, or a union type whose
13375 member objects are of scalar types, the expression is interpreted
13376 by GCC as a read of the volatile object; in the other cases, the
13377 expression is only evaluated for its side-effects.
13381 File: gcc.info, Node: Declarators implementation, Next: Statements implementation, Prev: Qualifiers implementation, Up: C Implementation
13386 * `The maximum number of declarators that may modify an arithmetic,
13387 structure or union type (C90 6.5.4).'
13389 GCC is only limited by available memory.
13393 File: gcc.info, Node: Statements implementation, Next: Preprocessing directives implementation, Prev: Declarators implementation, Up: C Implementation
13398 * `The maximum number of `case' values in a `switch' statement (C90
13401 GCC is only limited by available memory.
13405 File: gcc.info, Node: Preprocessing directives implementation, Next: Library functions implementation, Prev: Statements implementation, Up: C Implementation
13407 4.13 Preprocessing directives
13408 =============================
13410 *Note Implementation-defined behavior: (cpp)Implementation-defined
13411 behavior, for details of these aspects of implementation-defined
13414 * `How sequences in both forms of header names are mapped to headers
13415 or external source file names (C90 6.1.7, C99 6.4.7).'
13417 * `Whether the value of a character constant in a constant expression
13418 that controls conditional inclusion matches the value of the same
13419 character constant in the execution character set (C90 6.8.1, C99
13422 * `Whether the value of a single-character character constant in a
13423 constant expression that controls conditional inclusion may have a
13424 negative value (C90 6.8.1, C99 6.10.1).'
13426 * `The places that are searched for an included `<>' delimited
13427 header, and how the places are specified or the header is
13428 identified (C90 6.8.2, C99 6.10.2).'
13430 * `How the named source file is searched for in an included `""'
13431 delimited header (C90 6.8.2, C99 6.10.2).'
13433 * `The method by which preprocessing tokens (possibly resulting from
13434 macro expansion) in a `#include' directive are combined into a
13435 header name (C90 6.8.2, C99 6.10.2).'
13437 * `The nesting limit for `#include' processing (C90 6.8.2, C99
13440 * `Whether the `#' operator inserts a `\' character before the `\'
13441 character that begins a universal character name in a character
13442 constant or string literal (C99 6.10.3.2).'
13444 * `The behavior on each recognized non-`STDC #pragma' directive (C90
13445 6.8.6, C99 6.10.6).'
13447 *Note Pragmas: (cpp)Pragmas, for details of pragmas accepted by
13448 GCC on all targets. *Note Pragmas Accepted by GCC: Pragmas, for
13449 details of target-specific pragmas.
13451 * `The definitions for `__DATE__' and `__TIME__' when respectively,
13452 the date and time of translation are not available (C90 6.8.8, C99
13457 File: gcc.info, Node: Library functions implementation, Next: Architecture implementation, Prev: Preprocessing directives implementation, Up: C Implementation
13459 4.14 Library functions
13460 ======================
13462 The behavior of most of these points are dependent on the implementation
13463 of the C library, and are not defined by GCC itself.
13465 * `The null pointer constant to which the macro `NULL' expands (C90
13468 In `<stddef.h>', `NULL' expands to `((void *)0)'. GCC does not
13469 provide the other headers which define `NULL' and some library
13470 implementations may use other definitions in those headers.
13474 File: gcc.info, Node: Architecture implementation, Next: Locale-specific behavior implementation, Prev: Library functions implementation, Up: C Implementation
13479 * `The values or expressions assigned to the macros specified in the
13480 headers `<float.h>', `<limits.h>', and `<stdint.h>' (C90 and C99
13481 5.2.4.2, C99 7.18.2, C99 7.18.3).'
13485 * `The number, order, and encoding of bytes in any object (when not
13486 explicitly specified in this International Standard) (C99
13491 * `The value of the result of the `sizeof' operator (C90 6.3.3.4,
13498 File: gcc.info, Node: Locale-specific behavior implementation, Prev: Architecture implementation, Up: C Implementation
13500 4.16 Locale-specific behavior
13501 =============================
13503 The behavior of these points are dependent on the implementation of the
13504 C library, and are not defined by GCC itself.
13507 File: gcc.info, Node: C Extensions, Next: C++ Extensions, Prev: C Implementation, Up: Top
13509 5 Extensions to the C Language Family
13510 *************************************
13512 GNU C provides several language features not found in ISO standard C.
13513 (The `-pedantic' option directs GCC to print a warning message if any
13514 of these features is used.) To test for the availability of these
13515 features in conditional compilation, check for a predefined macro
13516 `__GNUC__', which is always defined under GCC.
13518 These extensions are available in C and Objective-C. Most of them are
13519 also available in C++. *Note Extensions to the C++ Language: C++
13520 Extensions, for extensions that apply _only_ to C++.
13522 Some features that are in ISO C99 but not C89 or C++ are also, as
13523 extensions, accepted by GCC in C89 mode and in C++.
13527 * Statement Exprs:: Putting statements and declarations inside expressions.
13528 * Local Labels:: Labels local to a block.
13529 * Labels as Values:: Getting pointers to labels, and computed gotos.
13530 * Nested Functions:: As in Algol and Pascal, lexical scoping of functions.
13531 * Constructing Calls:: Dispatching a call to another function.
13532 * Typeof:: `typeof': referring to the type of an expression.
13533 * Conditionals:: Omitting the middle operand of a `?:' expression.
13534 * Long Long:: Double-word integers---`long long int'.
13535 * Complex:: Data types for complex numbers.
13536 * Hex Floats:: Hexadecimal floating-point constants.
13537 * Zero Length:: Zero-length arrays.
13538 * Variable Length:: Arrays whose length is computed at run time.
13539 * Empty Structures:: Structures with no members.
13540 * Variadic Macros:: Macros with a variable number of arguments.
13541 * Escaped Newlines:: Slightly looser rules for escaped newlines.
13542 * Subscripting:: Any array can be subscripted, even if not an lvalue.
13543 * Pointer Arith:: Arithmetic on `void'-pointers and function pointers.
13544 * Initializers:: Non-constant initializers.
13545 * Compound Literals:: Compound literals give structures, unions
13546 or arrays as values.
13547 * Designated Inits:: Labeling elements of initializers.
13548 * Cast to Union:: Casting to union type from any member of the union.
13549 * Case Ranges:: `case 1 ... 9' and such.
13550 * Mixed Declarations:: Mixing declarations and code.
13551 * Function Attributes:: Declaring that functions have no side effects,
13552 or that they can never return.
13553 * Attribute Syntax:: Formal syntax for attributes.
13554 * Function Prototypes:: Prototype declarations and old-style definitions.
13555 * C++ Comments:: C++ comments are recognized.
13556 * Dollar Signs:: Dollar sign is allowed in identifiers.
13557 * Character Escapes:: `\e' stands for the character <ESC>.
13558 * Variable Attributes:: Specifying attributes of variables.
13559 * Type Attributes:: Specifying attributes of types.
13560 * Alignment:: Inquiring about the alignment of a type or variable.
13561 * Inline:: Defining inline functions (as fast as macros).
13562 * Extended Asm:: Assembler instructions with C expressions as operands.
13563 (With them you can define ``built-in'' functions.)
13564 * Constraints:: Constraints for asm operands
13565 * Asm Labels:: Specifying the assembler name to use for a C symbol.
13566 * Explicit Reg Vars:: Defining variables residing in specified registers.
13567 * Alternate Keywords:: `__const__', `__asm__', etc., for header files.
13568 * Incomplete Enums:: `enum foo;', with details to follow.
13569 * Function Names:: Printable strings which are the name of the current
13571 * Return Address:: Getting the return or frame address of a function.
13572 * Vector Extensions:: Using vector instructions through built-in functions.
13573 * Offsetof:: Special syntax for implementing `offsetof'.
13574 * Atomic Builtins:: Built-in functions for atomic memory access.
13575 * Object Size Checking:: Built-in functions for limited buffer overflow
13577 * Other Builtins:: Other built-in functions.
13578 * Target Builtins:: Built-in functions specific to particular targets.
13579 * Target Format Checks:: Format checks specific to particular targets.
13580 * Pragmas:: Pragmas accepted by GCC.
13581 * Unnamed Fields:: Unnamed struct/union fields within structs/unions.
13582 * Thread-Local:: Per-thread variables.
13585 File: gcc.info, Node: Statement Exprs, Next: Local Labels, Up: C Extensions
13587 5.1 Statements and Declarations in Expressions
13588 ==============================================
13590 A compound statement enclosed in parentheses may appear as an expression
13591 in GNU C. This allows you to use loops, switches, and local variables
13592 within an expression.
13594 Recall that a compound statement is a sequence of statements surrounded
13595 by braces; in this construct, parentheses go around the braces. For
13598 ({ int y = foo (); int z;
13603 is a valid (though slightly more complex than necessary) expression for
13604 the absolute value of `foo ()'.
13606 The last thing in the compound statement should be an expression
13607 followed by a semicolon; the value of this subexpression serves as the
13608 value of the entire construct. (If you use some other kind of statement
13609 last within the braces, the construct has type `void', and thus
13610 effectively no value.)
13612 This feature is especially useful in making macro definitions "safe"
13613 (so that they evaluate each operand exactly once). For example, the
13614 "maximum" function is commonly defined as a macro in standard C as
13617 #define max(a,b) ((a) > (b) ? (a) : (b))
13619 But this definition computes either A or B twice, with bad results if
13620 the operand has side effects. In GNU C, if you know the type of the
13621 operands (here taken as `int'), you can define the macro safely as
13624 #define maxint(a,b) \
13625 ({int _a = (a), _b = (b); _a > _b ? _a : _b; })
13627 Embedded statements are not allowed in constant expressions, such as
13628 the value of an enumeration constant, the width of a bit-field, or the
13629 initial value of a static variable.
13631 If you don't know the type of the operand, you can still do this, but
13632 you must use `typeof' (*note Typeof::).
13634 In G++, the result value of a statement expression undergoes array and
13635 function pointer decay, and is returned by value to the enclosing
13636 expression. For instance, if `A' is a class, then
13642 will construct a temporary `A' object to hold the result of the
13643 statement expression, and that will be used to invoke `Foo'. Therefore
13644 the `this' pointer observed by `Foo' will not be the address of `a'.
13646 Any temporaries created within a statement within a statement
13647 expression will be destroyed at the statement's end. This makes
13648 statement expressions inside macros slightly different from function
13649 calls. In the latter case temporaries introduced during argument
13650 evaluation will be destroyed at the end of the statement that includes
13651 the function call. In the statement expression case they will be
13652 destroyed during the statement expression. For instance,
13654 #define macro(a) ({__typeof__(a) b = (a); b + 3; })
13655 template<typename T> T function(T a) { T b = a; return b + 3; }
13663 will have different places where temporaries are destroyed. For the
13664 `macro' case, the temporary `X' will be destroyed just after the
13665 initialization of `b'. In the `function' case that temporary will be
13666 destroyed when the function returns.
13668 These considerations mean that it is probably a bad idea to use
13669 statement-expressions of this form in header files that are designed to
13670 work with C++. (Note that some versions of the GNU C Library contained
13671 header files using statement-expression that lead to precisely this
13674 Jumping into a statement expression with `goto' or using a `switch'
13675 statement outside the statement expression with a `case' or `default'
13676 label inside the statement expression is not permitted. Jumping into a
13677 statement expression with a computed `goto' (*note Labels as Values::)
13678 yields undefined behavior. Jumping out of a statement expression is
13679 permitted, but if the statement expression is part of a larger
13680 expression then it is unspecified which other subexpressions of that
13681 expression have been evaluated except where the language definition
13682 requires certain subexpressions to be evaluated before or after the
13683 statement expression. In any case, as with a function call the
13684 evaluation of a statement expression is not interleaved with the
13685 evaluation of other parts of the containing expression. For example,
13687 foo (), (({ bar1 (); goto a; 0; }) + bar2 ()), baz();
13689 will call `foo' and `bar1' and will not call `baz' but may or may not
13690 call `bar2'. If `bar2' is called, it will be called after `foo' and
13694 File: gcc.info, Node: Local Labels, Next: Labels as Values, Prev: Statement Exprs, Up: C Extensions
13696 5.2 Locally Declared Labels
13697 ===========================
13699 GCC allows you to declare "local labels" in any nested block scope. A
13700 local label is just like an ordinary label, but you can only reference
13701 it (with a `goto' statement, or by taking its address) within the block
13702 in which it was declared.
13704 A local label declaration looks like this:
13710 __label__ LABEL1, LABEL2, /* ... */;
13712 Local label declarations must come at the beginning of the block,
13713 before any ordinary declarations or statements.
13715 The label declaration defines the label _name_, but does not define
13716 the label itself. You must do this in the usual way, with `LABEL:',
13717 within the statements of the statement expression.
13719 The local label feature is useful for complex macros. If a macro
13720 contains nested loops, a `goto' can be useful for breaking out of them.
13721 However, an ordinary label whose scope is the whole function cannot be
13722 used: if the macro can be expanded several times in one function, the
13723 label will be multiply defined in that function. A local label avoids
13724 this problem. For example:
13726 #define SEARCH(value, array, target) \
13729 typeof (target) _SEARCH_target = (target); \
13730 typeof (*(array)) *_SEARCH_array = (array); \
13733 for (i = 0; i < max; i++) \
13734 for (j = 0; j < max; j++) \
13735 if (_SEARCH_array[i][j] == _SEARCH_target) \
13736 { (value) = i; goto found; } \
13741 This could also be written using a statement-expression:
13743 #define SEARCH(array, target) \
13746 typeof (target) _SEARCH_target = (target); \
13747 typeof (*(array)) *_SEARCH_array = (array); \
13750 for (i = 0; i < max; i++) \
13751 for (j = 0; j < max; j++) \
13752 if (_SEARCH_array[i][j] == _SEARCH_target) \
13753 { value = i; goto found; } \
13759 Local label declarations also make the labels they declare visible to
13760 nested functions, if there are any. *Note Nested Functions::, for
13764 File: gcc.info, Node: Labels as Values, Next: Nested Functions, Prev: Local Labels, Up: C Extensions
13766 5.3 Labels as Values
13767 ====================
13769 You can get the address of a label defined in the current function (or
13770 a containing function) with the unary operator `&&'. The value has
13771 type `void *'. This value is a constant and can be used wherever a
13772 constant of that type is valid. For example:
13778 To use these values, you need to be able to jump to one. This is done
13779 with the computed goto statement(1), `goto *EXP;'. For example,
13783 Any expression of type `void *' is allowed.
13785 One way of using these constants is in initializing a static array that
13786 will serve as a jump table:
13788 static void *array[] = { &&foo, &&bar, &&hack };
13790 Then you can select a label with indexing, like this:
13794 Note that this does not check whether the subscript is in bounds--array
13795 indexing in C never does that.
13797 Such an array of label values serves a purpose much like that of the
13798 `switch' statement. The `switch' statement is cleaner, so use that
13799 rather than an array unless the problem does not fit a `switch'
13800 statement very well.
13802 Another use of label values is in an interpreter for threaded code.
13803 The labels within the interpreter function can be stored in the
13804 threaded code for super-fast dispatching.
13806 You may not use this mechanism to jump to code in a different function.
13807 If you do that, totally unpredictable things will happen. The best way
13808 to avoid this is to store the label address only in automatic variables
13809 and never pass it as an argument.
13811 An alternate way to write the above example is
13813 static const int array[] = { &&foo - &&foo, &&bar - &&foo,
13815 goto *(&&foo + array[i]);
13817 This is more friendly to code living in shared libraries, as it reduces
13818 the number of dynamic relocations that are needed, and by consequence,
13819 allows the data to be read-only.
13821 ---------- Footnotes ----------
13823 (1) The analogous feature in Fortran is called an assigned goto, but
13824 that name seems inappropriate in C, where one can do more than simply
13825 store label addresses in label variables.
13828 File: gcc.info, Node: Nested Functions, Next: Constructing Calls, Prev: Labels as Values, Up: C Extensions
13830 5.4 Nested Functions
13831 ====================
13833 A "nested function" is a function defined inside another function.
13834 (Nested functions are not supported for GNU C++.) The nested function's
13835 name is local to the block where it is defined. For example, here we
13836 define a nested function named `square', and call it twice:
13838 foo (double a, double b)
13840 double square (double z) { return z * z; }
13842 return square (a) + square (b);
13845 The nested function can access all the variables of the containing
13846 function that are visible at the point of its definition. This is
13847 called "lexical scoping". For example, here we show a nested function
13848 which uses an inherited variable named `offset':
13850 bar (int *array, int offset, int size)
13852 int access (int *array, int index)
13853 { return array[index + offset]; }
13856 for (i = 0; i < size; i++)
13857 /* ... */ access (array, i) /* ... */
13860 Nested function definitions are permitted within functions in the
13861 places where variable definitions are allowed; that is, in any block,
13862 mixed with the other declarations and statements in the block.
13864 It is possible to call the nested function from outside the scope of
13865 its name by storing its address or passing the address to another
13868 hack (int *array, int size)
13870 void store (int index, int value)
13871 { array[index] = value; }
13873 intermediate (store, size);
13876 Here, the function `intermediate' receives the address of `store' as
13877 an argument. If `intermediate' calls `store', the arguments given to
13878 `store' are used to store into `array'. But this technique works only
13879 so long as the containing function (`hack', in this example) does not
13882 If you try to call the nested function through its address after the
13883 containing function has exited, all hell will break loose. If you try
13884 to call it after a containing scope level has exited, and if it refers
13885 to some of the variables that are no longer in scope, you may be lucky,
13886 but it's not wise to take the risk. If, however, the nested function
13887 does not refer to anything that has gone out of scope, you should be
13890 GCC implements taking the address of a nested function using a
13891 technique called "trampolines". A paper describing them is available as
13893 `http://people.debian.org/~aaronl/Usenix88-lexic.pdf'.
13895 A nested function can jump to a label inherited from a containing
13896 function, provided the label was explicitly declared in the containing
13897 function (*note Local Labels::). Such a jump returns instantly to the
13898 containing function, exiting the nested function which did the `goto'
13899 and any intermediate functions as well. Here is an example:
13901 bar (int *array, int offset, int size)
13904 int access (int *array, int index)
13908 return array[index + offset];
13912 for (i = 0; i < size; i++)
13913 /* ... */ access (array, i) /* ... */
13917 /* Control comes here from `access'
13918 if it detects an error. */
13923 A nested function always has no linkage. Declaring one with `extern'
13924 or `static' is erroneous. If you need to declare the nested function
13925 before its definition, use `auto' (which is otherwise meaningless for
13926 function declarations).
13928 bar (int *array, int offset, int size)
13931 auto int access (int *, int);
13933 int access (int *array, int index)
13937 return array[index + offset];
13943 File: gcc.info, Node: Constructing Calls, Next: Typeof, Prev: Nested Functions, Up: C Extensions
13945 5.5 Constructing Function Calls
13946 ===============================
13948 Using the built-in functions described below, you can record the
13949 arguments a function received, and call another function with the same
13950 arguments, without knowing the number or types of the arguments.
13952 You can also record the return value of that function call, and later
13953 return that value, without knowing what data type the function tried to
13954 return (as long as your caller expects that data type).
13956 However, these built-in functions may interact badly with some
13957 sophisticated features or other extensions of the language. It is,
13958 therefore, not recommended to use them outside very simple functions
13959 acting as mere forwarders for their arguments.
13961 -- Built-in Function: void * __builtin_apply_args ()
13962 This built-in function returns a pointer to data describing how to
13963 perform a call with the same arguments as were passed to the
13966 The function saves the arg pointer register, structure value
13967 address, and all registers that might be used to pass arguments to
13968 a function into a block of memory allocated on the stack. Then it
13969 returns the address of that block.
13971 -- Built-in Function: void * __builtin_apply (void (*FUNCTION)(), void
13972 *ARGUMENTS, size_t SIZE)
13973 This built-in function invokes FUNCTION with a copy of the
13974 parameters described by ARGUMENTS and SIZE.
13976 The value of ARGUMENTS should be the value returned by
13977 `__builtin_apply_args'. The argument SIZE specifies the size of
13978 the stack argument data, in bytes.
13980 This function returns a pointer to data describing how to return
13981 whatever value was returned by FUNCTION. The data is saved in a
13982 block of memory allocated on the stack.
13984 It is not always simple to compute the proper value for SIZE. The
13985 value is used by `__builtin_apply' to compute the amount of data
13986 that should be pushed on the stack and copied from the incoming
13989 -- Built-in Function: void __builtin_return (void *RESULT)
13990 This built-in function returns the value described by RESULT from
13991 the containing function. You should specify, for RESULT, a value
13992 returned by `__builtin_apply'.
13995 File: gcc.info, Node: Typeof, Next: Conditionals, Prev: Constructing Calls, Up: C Extensions
13997 5.6 Referring to a Type with `typeof'
13998 =====================================
14000 Another way to refer to the type of an expression is with `typeof'.
14001 The syntax of using of this keyword looks like `sizeof', but the
14002 construct acts semantically like a type name defined with `typedef'.
14004 There are two ways of writing the argument to `typeof': with an
14005 expression or with a type. Here is an example with an expression:
14009 This assumes that `x' is an array of pointers to functions; the type
14010 described is that of the values of the functions.
14012 Here is an example with a typename as the argument:
14016 Here the type described is that of pointers to `int'.
14018 If you are writing a header file that must work when included in ISO C
14019 programs, write `__typeof__' instead of `typeof'. *Note Alternate
14022 A `typeof'-construct can be used anywhere a typedef name could be
14023 used. For example, you can use it in a declaration, in a cast, or
14024 inside of `sizeof' or `typeof'.
14026 `typeof' is often useful in conjunction with the
14027 statements-within-expressions feature. Here is how the two together can
14028 be used to define a safe "maximum" macro that operates on any
14029 arithmetic type and evaluates each of its arguments exactly once:
14032 ({ typeof (a) _a = (a); \
14033 typeof (b) _b = (b); \
14034 _a > _b ? _a : _b; })
14036 The reason for using names that start with underscores for the local
14037 variables is to avoid conflicts with variable names that occur within
14038 the expressions that are substituted for `a' and `b'. Eventually we
14039 hope to design a new form of declaration syntax that allows you to
14040 declare variables whose scopes start only after their initializers;
14041 this will be a more reliable way to prevent such conflicts.
14043 Some more examples of the use of `typeof':
14045 * This declares `y' with the type of what `x' points to.
14049 * This declares `y' as an array of such values.
14053 * This declares `y' as an array of pointers to characters:
14055 typeof (typeof (char *)[4]) y;
14057 It is equivalent to the following traditional C declaration:
14061 To see the meaning of the declaration using `typeof', and why it
14062 might be a useful way to write, rewrite it with these macros:
14064 #define pointer(T) typeof(T *)
14065 #define array(T, N) typeof(T [N])
14067 Now the declaration can be rewritten this way:
14069 array (pointer (char), 4) y;
14071 Thus, `array (pointer (char), 4)' is the type of arrays of 4
14072 pointers to `char'.
14074 _Compatibility Note:_ In addition to `typeof', GCC 2 supported a more
14075 limited extension which permitted one to write
14079 with the effect of declaring T to have the type of the expression EXPR.
14080 This extension does not work with GCC 3 (versions between 3.0 and 3.2
14081 will crash; 3.2.1 and later give an error). Code which relies on it
14082 should be rewritten to use `typeof':
14084 typedef typeof(EXPR) T;
14086 This will work with all versions of GCC.
14089 File: gcc.info, Node: Conditionals, Next: Long Long, Prev: Typeof, Up: C Extensions
14091 5.7 Conditionals with Omitted Operands
14092 ======================================
14094 The middle operand in a conditional expression may be omitted. Then if
14095 the first operand is nonzero, its value is the value of the conditional
14098 Therefore, the expression
14102 has the value of `x' if that is nonzero; otherwise, the value of `y'.
14104 This example is perfectly equivalent to
14108 In this simple case, the ability to omit the middle operand is not
14109 especially useful. When it becomes useful is when the first operand
14110 does, or may (if it is a macro argument), contain a side effect. Then
14111 repeating the operand in the middle would perform the side effect
14112 twice. Omitting the middle operand uses the value already computed
14113 without the undesirable effects of recomputing it.
14116 File: gcc.info, Node: Long Long, Next: Complex, Prev: Conditionals, Up: C Extensions
14118 5.8 Double-Word Integers
14119 ========================
14121 ISO C99 supports data types for integers that are at least 64 bits wide,
14122 and as an extension GCC supports them in C89 mode and in C++. Simply
14123 write `long long int' for a signed integer, or `unsigned long long int'
14124 for an unsigned integer. To make an integer constant of type `long
14125 long int', add the suffix `LL' to the integer. To make an integer
14126 constant of type `unsigned long long int', add the suffix `ULL' to the
14129 You can use these types in arithmetic like any other integer types.
14130 Addition, subtraction, and bitwise boolean operations on these types
14131 are open-coded on all types of machines. Multiplication is open-coded
14132 if the machine supports fullword-to-doubleword a widening multiply
14133 instruction. Division and shifts are open-coded only on machines that
14134 provide special support. The operations that are not open-coded use
14135 special library routines that come with GCC.
14137 There may be pitfalls when you use `long long' types for function
14138 arguments, unless you declare function prototypes. If a function
14139 expects type `int' for its argument, and you pass a value of type `long
14140 long int', confusion will result because the caller and the subroutine
14141 will disagree about the number of bytes for the argument. Likewise, if
14142 the function expects `long long int' and you pass `int'. The best way
14143 to avoid such problems is to use prototypes.
14146 File: gcc.info, Node: Complex, Next: Hex Floats, Prev: Long Long, Up: C Extensions
14148 5.9 Complex Numbers
14149 ===================
14151 ISO C99 supports complex floating data types, and as an extension GCC
14152 supports them in C89 mode and in C++, and supports complex integer data
14153 types which are not part of ISO C99. You can declare complex types
14154 using the keyword `_Complex'. As an extension, the older GNU keyword
14155 `__complex__' is also supported.
14157 For example, `_Complex double x;' declares `x' as a variable whose
14158 real part and imaginary part are both of type `double'. `_Complex
14159 short int y;' declares `y' to have real and imaginary parts of type
14160 `short int'; this is not likely to be useful, but it shows that the set
14161 of complex types is complete.
14163 To write a constant with a complex data type, use the suffix `i' or
14164 `j' (either one; they are equivalent). For example, `2.5fi' has type
14165 `_Complex float' and `3i' has type `_Complex int'. Such a constant
14166 always has a pure imaginary value, but you can form any complex value
14167 you like by adding one to a real constant. This is a GNU extension; if
14168 you have an ISO C99 conforming C library (such as GNU libc), and want
14169 to construct complex constants of floating type, you should include
14170 `<complex.h>' and use the macros `I' or `_Complex_I' instead.
14172 To extract the real part of a complex-valued expression EXP, write
14173 `__real__ EXP'. Likewise, use `__imag__' to extract the imaginary
14174 part. This is a GNU extension; for values of floating type, you should
14175 use the ISO C99 functions `crealf', `creal', `creall', `cimagf',
14176 `cimag' and `cimagl', declared in `<complex.h>' and also provided as
14177 built-in functions by GCC.
14179 The operator `~' performs complex conjugation when used on a value
14180 with a complex type. This is a GNU extension; for values of floating
14181 type, you should use the ISO C99 functions `conjf', `conj' and `conjl',
14182 declared in `<complex.h>' and also provided as built-in functions by
14185 GCC can allocate complex automatic variables in a noncontiguous
14186 fashion; it's even possible for the real part to be in a register while
14187 the imaginary part is on the stack (or vice-versa). Only the DWARF2
14188 debug info format can represent this, so use of DWARF2 is recommended.
14189 If you are using the stabs debug info format, GCC describes a
14190 noncontiguous complex variable as if it were two separate variables of
14191 noncomplex type. If the variable's actual name is `foo', the two
14192 fictitious variables are named `foo$real' and `foo$imag'. You can
14193 examine and set these two fictitious variables with your debugger.
14196 File: gcc.info, Node: Hex Floats, Next: Zero Length, Prev: Complex, Up: C Extensions
14201 ISO C99 supports floating-point numbers written not only in the usual
14202 decimal notation, such as `1.55e1', but also numbers such as `0x1.fp3'
14203 written in hexadecimal format. As a GNU extension, GCC supports this
14204 in C89 mode (except in some cases when strictly conforming) and in C++.
14205 In that format the `0x' hex introducer and the `p' or `P' exponent
14206 field are mandatory. The exponent is a decimal number that indicates
14207 the power of 2 by which the significant part will be multiplied. Thus
14208 `0x1.f' is 1 15/16, `p3' multiplies it by 8, and the value of `0x1.fp3'
14209 is the same as `1.55e1'.
14211 Unlike for floating-point numbers in the decimal notation the exponent
14212 is always required in the hexadecimal notation. Otherwise the compiler
14213 would not be able to resolve the ambiguity of, e.g., `0x1.f'. This
14214 could mean `1.0f' or `1.9375' since `f' is also the extension for
14215 floating-point constants of type `float'.
14218 File: gcc.info, Node: Zero Length, Next: Variable Length, Prev: Hex Floats, Up: C Extensions
14220 5.11 Arrays of Length Zero
14221 ==========================
14223 Zero-length arrays are allowed in GNU C. They are very useful as the
14224 last element of a structure which is really a header for a
14225 variable-length object:
14232 struct line *thisline = (struct line *)
14233 malloc (sizeof (struct line) + this_length);
14234 thisline->length = this_length;
14236 In ISO C90, you would have to give `contents' a length of 1, which
14237 means either you waste space or complicate the argument to `malloc'.
14239 In ISO C99, you would use a "flexible array member", which is slightly
14240 different in syntax and semantics:
14242 * Flexible array members are written as `contents[]' without the `0'.
14244 * Flexible array members have incomplete type, and so the `sizeof'
14245 operator may not be applied. As a quirk of the original
14246 implementation of zero-length arrays, `sizeof' evaluates to zero.
14248 * Flexible array members may only appear as the last member of a
14249 `struct' that is otherwise non-empty.
14251 * A structure containing a flexible array member, or a union
14252 containing such a structure (possibly recursively), may not be a
14253 member of a structure or an element of an array. (However, these
14254 uses are permitted by GCC as extensions.)
14256 GCC versions before 3.0 allowed zero-length arrays to be statically
14257 initialized, as if they were flexible arrays. In addition to those
14258 cases that were useful, it also allowed initializations in situations
14259 that would corrupt later data. Non-empty initialization of zero-length
14260 arrays is now treated like any case where there are more initializer
14261 elements than the array holds, in that a suitable warning about "excess
14262 elements in array" is given, and the excess elements (all of them, in
14263 this case) are ignored.
14265 Instead GCC allows static initialization of flexible array members.
14266 This is equivalent to defining a new structure containing the original
14267 structure followed by an array of sufficient size to contain the data.
14268 I.e. in the following, `f1' is constructed as if it were declared like
14273 } f1 = { 1, { 2, 3, 4 } };
14276 struct f1 f1; int data[3];
14277 } f2 = { { 1 }, { 2, 3, 4 } };
14279 The convenience of this extension is that `f1' has the desired type,
14280 eliminating the need to consistently refer to `f2.f1'.
14282 This has symmetry with normal static arrays, in that an array of
14283 unknown size is also written with `[]'.
14285 Of course, this extension only makes sense if the extra data comes at
14286 the end of a top-level object, as otherwise we would be overwriting
14287 data at subsequent offsets. To avoid undue complication and confusion
14288 with initialization of deeply nested arrays, we simply disallow any
14289 non-empty initialization except when the structure is the top-level
14290 object. For example:
14292 struct foo { int x; int y[]; };
14293 struct bar { struct foo z; };
14295 struct foo a = { 1, { 2, 3, 4 } }; // Valid.
14296 struct bar b = { { 1, { 2, 3, 4 } } }; // Invalid.
14297 struct bar c = { { 1, { } } }; // Valid.
14298 struct foo d[1] = { { 1 { 2, 3, 4 } } }; // Invalid.
14301 File: gcc.info, Node: Empty Structures, Next: Variadic Macros, Prev: Variable Length, Up: C Extensions
14303 5.12 Structures With No Members
14304 ===============================
14306 GCC permits a C structure to have no members:
14311 The structure will have size zero. In C++, empty structures are part
14312 of the language. G++ treats empty structures as if they had a single
14313 member of type `char'.
14316 File: gcc.info, Node: Variable Length, Next: Empty Structures, Prev: Zero Length, Up: C Extensions
14318 5.13 Arrays of Variable Length
14319 ==============================
14321 Variable-length automatic arrays are allowed in ISO C99, and as an
14322 extension GCC accepts them in C89 mode and in C++. (However, GCC's
14323 implementation of variable-length arrays does not yet conform in detail
14324 to the ISO C99 standard.) These arrays are declared like any other
14325 automatic arrays, but with a length that is not a constant expression.
14326 The storage is allocated at the point of declaration and deallocated
14327 when the brace-level is exited. For example:
14330 concat_fopen (char *s1, char *s2, char *mode)
14332 char str[strlen (s1) + strlen (s2) + 1];
14335 return fopen (str, mode);
14338 Jumping or breaking out of the scope of the array name deallocates the
14339 storage. Jumping into the scope is not allowed; you get an error
14342 You can use the function `alloca' to get an effect much like
14343 variable-length arrays. The function `alloca' is available in many
14344 other C implementations (but not in all). On the other hand,
14345 variable-length arrays are more elegant.
14347 There are other differences between these two methods. Space allocated
14348 with `alloca' exists until the containing _function_ returns. The
14349 space for a variable-length array is deallocated as soon as the array
14350 name's scope ends. (If you use both variable-length arrays and
14351 `alloca' in the same function, deallocation of a variable-length array
14352 will also deallocate anything more recently allocated with `alloca'.)
14354 You can also use variable-length arrays as arguments to functions:
14357 tester (int len, char data[len][len])
14362 The length of an array is computed once when the storage is allocated
14363 and is remembered for the scope of the array in case you access it with
14366 If you want to pass the array first and the length afterward, you can
14367 use a forward declaration in the parameter list--another GNU extension.
14370 tester (int len; char data[len][len], int len)
14375 The `int len' before the semicolon is a "parameter forward
14376 declaration", and it serves the purpose of making the name `len' known
14377 when the declaration of `data' is parsed.
14379 You can write any number of such parameter forward declarations in the
14380 parameter list. They can be separated by commas or semicolons, but the
14381 last one must end with a semicolon, which is followed by the "real"
14382 parameter declarations. Each forward declaration must match a "real"
14383 declaration in parameter name and data type. ISO C99 does not support
14384 parameter forward declarations.
14387 File: gcc.info, Node: Variadic Macros, Next: Escaped Newlines, Prev: Empty Structures, Up: C Extensions
14389 5.14 Macros with a Variable Number of Arguments.
14390 ================================================
14392 In the ISO C standard of 1999, a macro can be declared to accept a
14393 variable number of arguments much as a function can. The syntax for
14394 defining the macro is similar to that of a function. Here is an
14397 #define debug(format, ...) fprintf (stderr, format, __VA_ARGS__)
14399 Here `...' is a "variable argument". In the invocation of such a
14400 macro, it represents the zero or more tokens until the closing
14401 parenthesis that ends the invocation, including any commas. This set of
14402 tokens replaces the identifier `__VA_ARGS__' in the macro body wherever
14403 it appears. See the CPP manual for more information.
14405 GCC has long supported variadic macros, and used a different syntax
14406 that allowed you to give a name to the variable arguments just like any
14407 other argument. Here is an example:
14409 #define debug(format, args...) fprintf (stderr, format, args)
14411 This is in all ways equivalent to the ISO C example above, but arguably
14412 more readable and descriptive.
14414 GNU CPP has two further variadic macro extensions, and permits them to
14415 be used with either of the above forms of macro definition.
14417 In standard C, you are not allowed to leave the variable argument out
14418 entirely; but you are allowed to pass an empty argument. For example,
14419 this invocation is invalid in ISO C, because there is no comma after
14422 debug ("A message")
14424 GNU CPP permits you to completely omit the variable arguments in this
14425 way. In the above examples, the compiler would complain, though since
14426 the expansion of the macro still has the extra comma after the format
14429 To help solve this problem, CPP behaves specially for variable
14430 arguments used with the token paste operator, `##'. If instead you
14433 #define debug(format, ...) fprintf (stderr, format, ## __VA_ARGS__)
14435 and if the variable arguments are omitted or empty, the `##' operator
14436 causes the preprocessor to remove the comma before it. If you do
14437 provide some variable arguments in your macro invocation, GNU CPP does
14438 not complain about the paste operation and instead places the variable
14439 arguments after the comma. Just like any other pasted macro argument,
14440 these arguments are not macro expanded.
14443 File: gcc.info, Node: Escaped Newlines, Next: Subscripting, Prev: Variadic Macros, Up: C Extensions
14445 5.15 Slightly Looser Rules for Escaped Newlines
14446 ===============================================
14448 Recently, the preprocessor has relaxed its treatment of escaped
14449 newlines. Previously, the newline had to immediately follow a
14450 backslash. The current implementation allows whitespace in the form of
14451 spaces, horizontal and vertical tabs, and form feeds between the
14452 backslash and the subsequent newline. The preprocessor issues a
14453 warning, but treats it as a valid escaped newline and combines the two
14454 lines to form a single logical line. This works within comments and
14455 tokens, as well as between tokens. Comments are _not_ treated as
14456 whitespace for the purposes of this relaxation, since they have not yet
14457 been replaced with spaces.
14460 File: gcc.info, Node: Subscripting, Next: Pointer Arith, Prev: Escaped Newlines, Up: C Extensions
14462 5.16 Non-Lvalue Arrays May Have Subscripts
14463 ==========================================
14465 In ISO C99, arrays that are not lvalues still decay to pointers, and
14466 may be subscripted, although they may not be modified or used after the
14467 next sequence point and the unary `&' operator may not be applied to
14468 them. As an extension, GCC allows such arrays to be subscripted in C89
14469 mode, though otherwise they do not decay to pointers outside C99 mode.
14470 For example, this is valid in GNU C though not valid in C89:
14472 struct foo {int a[4];};
14478 return f().a[index];
14482 File: gcc.info, Node: Pointer Arith, Next: Initializers, Prev: Subscripting, Up: C Extensions
14484 5.17 Arithmetic on `void'- and Function-Pointers
14485 ================================================
14487 In GNU C, addition and subtraction operations are supported on pointers
14488 to `void' and on pointers to functions. This is done by treating the
14489 size of a `void' or of a function as 1.
14491 A consequence of this is that `sizeof' is also allowed on `void' and
14492 on function types, and returns 1.
14494 The option `-Wpointer-arith' requests a warning if these extensions
14498 File: gcc.info, Node: Initializers, Next: Compound Literals, Prev: Pointer Arith, Up: C Extensions
14500 5.18 Non-Constant Initializers
14501 ==============================
14503 As in standard C++ and ISO C99, the elements of an aggregate
14504 initializer for an automatic variable are not required to be constant
14505 expressions in GNU C. Here is an example of an initializer with
14506 run-time varying elements:
14508 foo (float f, float g)
14510 float beat_freqs[2] = { f-g, f+g };
14515 File: gcc.info, Node: Compound Literals, Next: Designated Inits, Prev: Initializers, Up: C Extensions
14517 5.19 Compound Literals
14518 ======================
14520 ISO C99 supports compound literals. A compound literal looks like a
14521 cast containing an initializer. Its value is an object of the type
14522 specified in the cast, containing the elements specified in the
14523 initializer; it is an lvalue. As an extension, GCC supports compound
14524 literals in C89 mode and in C++.
14526 Usually, the specified type is a structure. Assume that `struct foo'
14527 and `structure' are declared as shown:
14529 struct foo {int a; char b[2];} structure;
14531 Here is an example of constructing a `struct foo' with a compound
14534 structure = ((struct foo) {x + y, 'a', 0});
14536 This is equivalent to writing the following:
14539 struct foo temp = {x + y, 'a', 0};
14543 You can also construct an array. If all the elements of the compound
14544 literal are (made up of) simple constant expressions, suitable for use
14545 in initializers of objects of static storage duration, then the compound
14546 literal can be coerced to a pointer to its first element and used in
14547 such an initializer, as shown here:
14549 char **foo = (char *[]) { "x", "y", "z" };
14551 Compound literals for scalar types and union types are is also
14552 allowed, but then the compound literal is equivalent to a cast.
14554 As a GNU extension, GCC allows initialization of objects with static
14555 storage duration by compound literals (which is not possible in ISO
14556 C99, because the initializer is not a constant). It is handled as if
14557 the object was initialized only with the bracket enclosed list if the
14558 types of the compound literal and the object match. The initializer
14559 list of the compound literal must be constant. If the object being
14560 initialized has array type of unknown size, the size is determined by
14561 compound literal size.
14563 static struct foo x = (struct foo) {1, 'a', 'b'};
14564 static int y[] = (int []) {1, 2, 3};
14565 static int z[] = (int [3]) {1};
14567 The above lines are equivalent to the following:
14568 static struct foo x = {1, 'a', 'b'};
14569 static int y[] = {1, 2, 3};
14570 static int z[] = {1, 0, 0};
14573 File: gcc.info, Node: Designated Inits, Next: Cast to Union, Prev: Compound Literals, Up: C Extensions
14575 5.20 Designated Initializers
14576 ============================
14578 Standard C89 requires the elements of an initializer to appear in a
14579 fixed order, the same as the order of the elements in the array or
14580 structure being initialized.
14582 In ISO C99 you can give the elements in any order, specifying the array
14583 indices or structure field names they apply to, and GNU C allows this as
14584 an extension in C89 mode as well. This extension is not implemented in
14587 To specify an array index, write `[INDEX] =' before the element value.
14590 int a[6] = { [4] = 29, [2] = 15 };
14594 int a[6] = { 0, 0, 15, 0, 29, 0 };
14596 The index values must be constant expressions, even if the array being
14597 initialized is automatic.
14599 An alternative syntax for this which has been obsolete since GCC 2.5
14600 but GCC still accepts is to write `[INDEX]' before the element value,
14603 To initialize a range of elements to the same value, write `[FIRST ...
14604 LAST] = VALUE'. This is a GNU extension. For example,
14606 int widths[] = { [0 ... 9] = 1, [10 ... 99] = 2, [100] = 3 };
14608 If the value in it has side-effects, the side-effects will happen only
14609 once, not for each initialized field by the range initializer.
14611 Note that the length of the array is the highest value specified plus
14614 In a structure initializer, specify the name of a field to initialize
14615 with `.FIELDNAME =' before the element value. For example, given the
14616 following structure,
14618 struct point { int x, y; };
14620 the following initialization
14622 struct point p = { .y = yvalue, .x = xvalue };
14626 struct point p = { xvalue, yvalue };
14628 Another syntax which has the same meaning, obsolete since GCC 2.5, is
14629 `FIELDNAME:', as shown here:
14631 struct point p = { y: yvalue, x: xvalue };
14633 The `[INDEX]' or `.FIELDNAME' is known as a "designator". You can
14634 also use a designator (or the obsolete colon syntax) when initializing
14635 a union, to specify which element of the union should be used. For
14638 union foo { int i; double d; };
14640 union foo f = { .d = 4 };
14642 will convert 4 to a `double' to store it in the union using the second
14643 element. By contrast, casting 4 to type `union foo' would store it
14644 into the union as the integer `i', since it is an integer. (*Note Cast
14647 You can combine this technique of naming elements with ordinary C
14648 initialization of successive elements. Each initializer element that
14649 does not have a designator applies to the next consecutive element of
14650 the array or structure. For example,
14652 int a[6] = { [1] = v1, v2, [4] = v4 };
14656 int a[6] = { 0, v1, v2, 0, v4, 0 };
14658 Labeling the elements of an array initializer is especially useful
14659 when the indices are characters or belong to an `enum' type. For
14662 int whitespace[256]
14663 = { [' '] = 1, ['\t'] = 1, ['\h'] = 1,
14664 ['\f'] = 1, ['\n'] = 1, ['\r'] = 1 };
14666 You can also write a series of `.FIELDNAME' and `[INDEX]' designators
14667 before an `=' to specify a nested subobject to initialize; the list is
14668 taken relative to the subobject corresponding to the closest
14669 surrounding brace pair. For example, with the `struct point'
14672 struct point ptarray[10] = { [2].y = yv2, [2].x = xv2, [0].x = xv0 };
14674 If the same field is initialized multiple times, it will have value from
14675 the last initialization. If any such overridden initialization has
14676 side-effect, it is unspecified whether the side-effect happens or not.
14677 Currently, GCC will discard them and issue a warning.
14680 File: gcc.info, Node: Case Ranges, Next: Mixed Declarations, Prev: Cast to Union, Up: C Extensions
14685 You can specify a range of consecutive values in a single `case' label,
14690 This has the same effect as the proper number of individual `case'
14691 labels, one for each integer value from LOW to HIGH, inclusive.
14693 This feature is especially useful for ranges of ASCII character codes:
14697 *Be careful:* Write spaces around the `...', for otherwise it may be
14698 parsed wrong when you use it with integer values. For example, write
14708 File: gcc.info, Node: Cast to Union, Next: Case Ranges, Prev: Designated Inits, Up: C Extensions
14710 5.22 Cast to a Union Type
14711 =========================
14713 A cast to union type is similar to other casts, except that the type
14714 specified is a union type. You can specify the type either with `union
14715 TAG' or with a typedef name. A cast to union is actually a constructor
14716 though, not a cast, and hence does not yield an lvalue like normal
14717 casts. (*Note Compound Literals::.)
14719 The types that may be cast to the union type are those of the members
14720 of the union. Thus, given the following union and variables:
14722 union foo { int i; double d; };
14726 both `x' and `y' can be cast to type `union foo'.
14728 Using the cast as the right-hand side of an assignment to a variable of
14729 union type is equivalent to storing in a member of the union:
14733 u = (union foo) x == u.i = x
14734 u = (union foo) y == u.d = y
14736 You can also use the union cast as a function argument:
14738 void hack (union foo);
14740 hack ((union foo) x);
14743 File: gcc.info, Node: Mixed Declarations, Next: Function Attributes, Prev: Case Ranges, Up: C Extensions
14745 5.23 Mixed Declarations and Code
14746 ================================
14748 ISO C99 and ISO C++ allow declarations and code to be freely mixed
14749 within compound statements. As an extension, GCC also allows this in
14750 C89 mode. For example, you could do:
14757 Each identifier is visible from where it is declared until the end of
14758 the enclosing block.
14761 File: gcc.info, Node: Function Attributes, Next: Attribute Syntax, Prev: Mixed Declarations, Up: C Extensions
14763 5.24 Declaring Attributes of Functions
14764 ======================================
14766 In GNU C, you declare certain things about functions called in your
14767 program which help the compiler optimize function calls and check your
14768 code more carefully.
14770 The keyword `__attribute__' allows you to specify special attributes
14771 when making a declaration. This keyword is followed by an attribute
14772 specification inside double parentheses. The following attributes are
14773 currently defined for functions on all targets: `noreturn',
14774 `returns_twice', `noinline', `always_inline', `flatten', `pure',
14775 `const', `nothrow', `sentinel', `format', `format_arg',
14776 `no_instrument_function', `section', `constructor', `destructor',
14777 `used', `unused', `deprecated', `weak', `malloc', `alias',
14778 `warn_unused_result', `nonnull' and `externally_visible'. Several other
14779 attributes are defined for functions on particular target systems.
14780 Other attributes, including `section' are supported for variables
14781 declarations (*note Variable Attributes::) and for types (*note Type
14784 You may also specify attributes with `__' preceding and following each
14785 keyword. This allows you to use them in header files without being
14786 concerned about a possible macro of the same name. For example, you
14787 may use `__noreturn__' instead of `noreturn'.
14789 *Note Attribute Syntax::, for details of the exact syntax for using
14793 The `alias' attribute causes the declaration to be emitted as an
14794 alias for another symbol, which must be specified. For instance,
14796 void __f () { /* Do something. */; }
14797 void f () __attribute__ ((weak, alias ("__f")));
14799 declares `f' to be a weak alias for `__f'. In C++, the mangled
14800 name for the target must be used. It is an error if `__f' is not
14801 defined in the same translation unit.
14803 Not all target machines support this attribute.
14806 Generally, functions are not inlined unless optimization is
14807 specified. For functions declared inline, this attribute inlines
14808 the function even if no optimization level was specified.
14811 Generally, inlining into a function is limited. For a function
14812 marked with this attribute, every call inside this function will
14813 be inlined, if possible. Whether the function itself is
14814 considered for inlining depends on its size and the current
14815 inlining parameters. The `flatten' attribute only works reliably
14816 in unit-at-a-time mode.
14819 On the Intel 386, the `cdecl' attribute causes the compiler to
14820 assume that the calling function will pop off the stack space used
14821 to pass arguments. This is useful to override the effects of the
14825 Many functions do not examine any values except their arguments,
14826 and have no effects except the return value. Basically this is
14827 just slightly more strict class than the `pure' attribute below,
14828 since function is not allowed to read global memory.
14830 Note that a function that has pointer arguments and examines the
14831 data pointed to must _not_ be declared `const'. Likewise, a
14832 function that calls a non-`const' function usually must not be
14833 `const'. It does not make sense for a `const' function to return
14836 The attribute `const' is not implemented in GCC versions earlier
14837 than 2.5. An alternative way to declare that a function has no
14838 side effects, which works in the current version and in some older
14839 versions, is as follows:
14841 typedef int intfn ();
14843 extern const intfn square;
14845 This approach does not work in GNU C++ from 2.6.0 on, since the
14846 language specifies that the `const' must be attached to the return
14851 The `constructor' attribute causes the function to be called
14852 automatically before execution enters `main ()'. Similarly, the
14853 `destructor' attribute causes the function to be called
14854 automatically after `main ()' has completed or `exit ()' has been
14855 called. Functions with these attributes are useful for
14856 initializing data that will be used implicitly during the
14857 execution of the program.
14859 These attributes are not currently implemented for Objective-C.
14862 The `deprecated' attribute results in a warning if the function is
14863 used anywhere in the source file. This is useful when identifying
14864 functions that are expected to be removed in a future version of a
14865 program. The warning also includes the location of the declaration
14866 of the deprecated function, to enable users to easily find further
14867 information about why the function is deprecated, or what they
14868 should do instead. Note that the warnings only occurs for uses:
14870 int old_fn () __attribute__ ((deprecated));
14872 int (*fn_ptr)() = old_fn;
14874 results in a warning on line 3 but not line 2.
14876 The `deprecated' attribute can also be used for variables and
14877 types (*note Variable Attributes::, *note Type Attributes::.)
14880 On Microsoft Windows targets and Symbian OS targets the
14881 `dllexport' attribute causes the compiler to provide a global
14882 pointer to a pointer in a DLL, so that it can be referenced with
14883 the `dllimport' attribute. On Microsoft Windows targets, the
14884 pointer name is formed by combining `_imp__' and the function or
14887 You can use `__declspec(dllexport)' as a synonym for
14888 `__attribute__ ((dllexport))' for compatibility with other
14891 On systems that support the `visibility' attribute, this attribute
14892 also implies "default" visibility, unless a `visibility' attribute
14893 is explicitly specified. You should avoid the use of `dllexport'
14894 with "hidden" or "internal" visibility; in the future GCC may
14895 issue an error for those cases.
14897 Currently, the `dllexport' attribute is ignored for inlined
14898 functions, unless the `-fkeep-inline-functions' flag has been
14899 used. The attribute is also ignored for undefined symbols.
14901 When applied to C++ classes, the attribute marks defined
14902 non-inlined member functions and static data members as exports.
14903 Static consts initialized in-class are not marked unless they are
14904 also defined out-of-class.
14906 For Microsoft Windows targets there are alternative methods for
14907 including the symbol in the DLL's export table such as using a
14908 `.def' file with an `EXPORTS' section or, with GNU ld, using the
14909 `--export-all' linker flag.
14912 On Microsoft Windows and Symbian OS targets, the `dllimport'
14913 attribute causes the compiler to reference a function or variable
14914 via a global pointer to a pointer that is set up by the DLL
14915 exporting the symbol. The attribute implies `extern' storage. On
14916 Microsoft Windows targets, the pointer name is formed by combining
14917 `_imp__' and the function or variable name.
14919 You can use `__declspec(dllimport)' as a synonym for
14920 `__attribute__ ((dllimport))' for compatibility with other
14923 Currently, the attribute is ignored for inlined functions. If the
14924 attribute is applied to a symbol _definition_, an error is
14925 reported. If a symbol previously declared `dllimport' is later
14926 defined, the attribute is ignored in subsequent references, and a
14927 warning is emitted. The attribute is also overridden by a
14928 subsequent declaration as `dllexport'.
14930 When applied to C++ classes, the attribute marks non-inlined
14931 member functions and static data members as imports. However, the
14932 attribute is ignored for virtual methods to allow creation of
14933 vtables using thunks.
14935 On the SH Symbian OS target the `dllimport' attribute also has
14936 another affect--it can cause the vtable and run-time type
14937 information for a class to be exported. This happens when the
14938 class has a dllimport'ed constructor or a non-inline, non-pure
14939 virtual function and, for either of those two conditions, the
14940 class also has a inline constructor or destructor and has a key
14941 function that is defined in the current translation unit.
14943 For Microsoft Windows based targets the use of the `dllimport'
14944 attribute on functions is not necessary, but provides a small
14945 performance benefit by eliminating a thunk in the DLL. The use of
14946 the `dllimport' attribute on imported variables was required on
14947 older versions of the GNU linker, but can now be avoided by
14948 passing the `--enable-auto-import' switch to the GNU linker. As
14949 with functions, using the attribute for a variable eliminates a
14952 One drawback to using this attribute is that a pointer to a
14953 function or variable marked as `dllimport' cannot be used as a
14954 constant address. On Microsoft Windows targets, the attribute can
14955 be disabled for functions by setting the `-mnop-fun-dllimport'
14959 Use this attribute on the H8/300, H8/300H, and H8S to indicate
14960 that the specified variable should be placed into the eight bit
14961 data section. The compiler will generate more efficient code for
14962 certain operations on data in the eight bit data area. Note the
14963 eight bit data area is limited to 256 bytes of data.
14965 You must use GAS and GLD from GNU binutils version 2.7 or later for
14966 this attribute to work correctly.
14968 `exception_handler'
14969 Use this attribute on the Blackfin to indicate that the specified
14970 function is an exception handler. The compiler will generate
14971 function entry and exit sequences suitable for use in an exception
14972 handler when this attribute is present.
14975 On 68HC11 and 68HC12 the `far' attribute causes the compiler to
14976 use a calling convention that takes care of switching memory banks
14977 when entering and leaving a function. This calling convention is
14978 also the default when using the `-mlong-calls' option.
14980 On 68HC12 the compiler will use the `call' and `rtc' instructions
14981 to call and return from a function.
14983 On 68HC11 the compiler will generate a sequence of instructions to
14984 invoke a board-specific routine to switch the memory bank and call
14985 the real function. The board-specific routine simulates a `call'.
14986 At the end of a function, it will jump to a board-specific routine
14987 instead of using `rts'. The board-specific return routine
14988 simulates the `rtc'.
14991 On the Intel 386, the `fastcall' attribute causes the compiler to
14992 pass the first argument (if of integral type) in the register ECX
14993 and the second argument (if of integral type) in the register EDX.
14994 Subsequent and other typed arguments are passed on the stack.
14995 The called function will pop the arguments off the stack. If the
14996 number of arguments is variable all arguments are pushed on the
14999 `format (ARCHETYPE, STRING-INDEX, FIRST-TO-CHECK)'
15000 The `format' attribute specifies that a function takes `printf',
15001 `scanf', `strftime' or `strfmon' style arguments which should be
15002 type-checked against a format string. For example, the
15006 my_printf (void *my_object, const char *my_format, ...)
15007 __attribute__ ((format (printf, 2, 3)));
15009 causes the compiler to check the arguments in calls to `my_printf'
15010 for consistency with the `printf' style format string argument
15013 The parameter ARCHETYPE determines how the format string is
15014 interpreted, and should be `printf', `scanf', `strftime' or
15015 `strfmon'. (You can also use `__printf__', `__scanf__',
15016 `__strftime__' or `__strfmon__'.) The parameter STRING-INDEX
15017 specifies which argument is the format string argument (starting
15018 from 1), while FIRST-TO-CHECK is the number of the first argument
15019 to check against the format string. For functions where the
15020 arguments are not available to be checked (such as `vprintf'),
15021 specify the third parameter as zero. In this case the compiler
15022 only checks the format string for consistency. For `strftime'
15023 formats, the third parameter is required to be zero. Since
15024 non-static C++ methods have an implicit `this' argument, the
15025 arguments of such methods should be counted from two, not one, when
15026 giving values for STRING-INDEX and FIRST-TO-CHECK.
15028 In the example above, the format string (`my_format') is the second
15029 argument of the function `my_print', and the arguments to check
15030 start with the third argument, so the correct parameters for the
15031 format attribute are 2 and 3.
15033 The `format' attribute allows you to identify your own functions
15034 which take format strings as arguments, so that GCC can check the
15035 calls to these functions for errors. The compiler always (unless
15036 `-ffreestanding' or `-fno-builtin' is used) checks formats for the
15037 standard library functions `printf', `fprintf', `sprintf',
15038 `scanf', `fscanf', `sscanf', `strftime', `vprintf', `vfprintf' and
15039 `vsprintf' whenever such warnings are requested (using
15040 `-Wformat'), so there is no need to modify the header file
15041 `stdio.h'. In C99 mode, the functions `snprintf', `vsnprintf',
15042 `vscanf', `vfscanf' and `vsscanf' are also checked. Except in
15043 strictly conforming C standard modes, the X/Open function
15044 `strfmon' is also checked as are `printf_unlocked' and
15045 `fprintf_unlocked'. *Note Options Controlling C Dialect: C
15048 The target may provide additional types of format checks. *Note
15049 Format Checks Specific to Particular Target Machines: Target
15052 `format_arg (STRING-INDEX)'
15053 The `format_arg' attribute specifies that a function takes a format
15054 string for a `printf', `scanf', `strftime' or `strfmon' style
15055 function and modifies it (for example, to translate it into
15056 another language), so the result can be passed to a `printf',
15057 `scanf', `strftime' or `strfmon' style function (with the
15058 remaining arguments to the format function the same as they would
15059 have been for the unmodified string). For example, the
15063 my_dgettext (char *my_domain, const char *my_format)
15064 __attribute__ ((format_arg (2)));
15066 causes the compiler to check the arguments in calls to a `printf',
15067 `scanf', `strftime' or `strfmon' type function, whose format
15068 string argument is a call to the `my_dgettext' function, for
15069 consistency with the format string argument `my_format'. If the
15070 `format_arg' attribute had not been specified, all the compiler
15071 could tell in such calls to format functions would be that the
15072 format string argument is not constant; this would generate a
15073 warning when `-Wformat-nonliteral' is used, but the calls could
15074 not be checked without the attribute.
15076 The parameter STRING-INDEX specifies which argument is the format
15077 string argument (starting from one). Since non-static C++ methods
15078 have an implicit `this' argument, the arguments of such methods
15079 should be counted from two.
15081 The `format-arg' attribute allows you to identify your own
15082 functions which modify format strings, so that GCC can check the
15083 calls to `printf', `scanf', `strftime' or `strfmon' type function
15084 whose operands are a call to one of your own function. The
15085 compiler always treats `gettext', `dgettext', and `dcgettext' in
15086 this manner except when strict ISO C support is requested by
15087 `-ansi' or an appropriate `-std' option, or `-ffreestanding' or
15088 `-fno-builtin' is used. *Note Options Controlling C Dialect: C
15092 Use this attribute on the H8/300, H8/300H, and H8S to indicate
15093 that the specified function should be called through the function
15094 vector. Calling a function through the function vector will
15095 reduce code size, however; the function vector has a limited size
15096 (maximum 128 entries on the H8/300 and 64 entries on the H8/300H
15097 and H8S) and shares space with the interrupt vector.
15099 You must use GAS and GLD from GNU binutils version 2.7 or later for
15100 this attribute to work correctly.
15103 Use this attribute on the ARM, AVR, C4x, CRX, M32C, M32R/D, MS1,
15104 and Xstormy16 ports to indicate that the specified function is an
15105 interrupt handler. The compiler will generate function entry and
15106 exit sequences suitable for use in an interrupt handler when this
15107 attribute is present.
15109 Note, interrupt handlers for the Blackfin, m68k, H8/300, H8/300H,
15110 H8S, and SH processors can be specified via the
15111 `interrupt_handler' attribute.
15113 Note, on the AVR, interrupts will be enabled inside the function.
15115 Note, for the ARM, you can specify the kind of interrupt to be
15116 handled by adding an optional parameter to the interrupt attribute
15119 void f () __attribute__ ((interrupt ("IRQ")));
15121 Permissible values for this parameter are: IRQ, FIQ, SWI, ABORT
15124 `interrupt_handler'
15125 Use this attribute on the Blackfin, m68k, H8/300, H8/300H, H8S,
15126 and SH to indicate that the specified function is an interrupt
15127 handler. The compiler will generate function entry and exit
15128 sequences suitable for use in an interrupt handler when this
15129 attribute is present.
15132 When used together with `interrupt_handler', `exception_handler'
15133 or `nmi_handler', code will be generated to load the stack pointer
15134 from the USP register in the function prologue.
15136 `long_call/short_call'
15137 This attribute specifies how a particular function is called on
15138 ARM. Both attributes override the `-mlong-calls' (*note ARM
15139 Options::) command line switch and `#pragma long_calls' settings.
15140 The `long_call' attribute causes the compiler to always call the
15141 function by first loading its address into a register and then
15142 using the contents of that register. The `short_call' attribute
15143 always places the offset to the function from the call site into
15144 the `BL' instruction directly.
15146 `longcall/shortcall'
15147 On the Blackfin, RS/6000 and PowerPC, the `longcall' attribute
15148 causes the compiler to always call this function via a pointer,
15149 just as it would if the `-mlongcall' option had been specified.
15150 The `shortcall' attribute causes the compiler not to do this.
15151 These attributes override both the `-mlongcall' switch and, on the
15152 RS/6000 and PowerPC, the `#pragma longcall' setting.
15154 *Note RS/6000 and PowerPC Options::, for more information on
15155 whether long calls are necessary.
15158 This attribute specifies how a particular function is called on
15159 MIPS. The attribute overrides the `-mlong-calls' (*note MIPS
15160 Options::) command line switch. This attribute causes the
15161 compiler to always call the function by first loading its address
15162 into a register, and then using the contents of that register.
15165 The `malloc' attribute is used to tell the compiler that a function
15166 may be treated as if any non-`NULL' pointer it returns cannot
15167 alias any other pointer valid when the function returns. This
15168 will often improve optimization. Standard functions with this
15169 property include `malloc' and `calloc'. `realloc'-like functions
15170 have this property as long as the old pointer is never referred to
15171 (including comparing it to the new pointer) after the function
15172 returns a non-`NULL' value.
15174 `model (MODEL-NAME)'
15175 On the M32R/D, use this attribute to set the addressability of an
15176 object, and of the code generated for a function. The identifier
15177 MODEL-NAME is one of `small', `medium', or `large', representing
15178 each of the code models.
15180 Small model objects live in the lower 16MB of memory (so that their
15181 addresses can be loaded with the `ld24' instruction), and are
15182 callable with the `bl' instruction.
15184 Medium model objects may live anywhere in the 32-bit address space
15185 (the compiler will generate `seth/add3' instructions to load their
15186 addresses), and are callable with the `bl' instruction.
15188 Large model objects may live anywhere in the 32-bit address space
15189 (the compiler will generate `seth/add3' instructions to load their
15190 addresses), and may not be reachable with the `bl' instruction
15191 (the compiler will generate the much slower `seth/add3/jl'
15192 instruction sequence).
15194 On IA-64, use this attribute to set the addressability of an
15195 object. At present, the only supported identifier for MODEL-NAME
15196 is `small', indicating addressability via "small" (22-bit)
15197 addresses (so that their addresses can be loaded with the `addl'
15198 instruction). Caveat: such addressing is by definition not
15199 position independent and hence this attribute must not be used for
15200 objects defined by shared libraries.
15203 Use this attribute on the ARM, AVR, C4x and IP2K ports to indicate
15204 that the specified function does not need prologue/epilogue
15205 sequences generated by the compiler. It is up to the programmer
15206 to provide these sequences.
15209 On 68HC11 and 68HC12 the `near' attribute causes the compiler to
15210 use the normal calling convention based on `jsr' and `rts'. This
15211 attribute can be used to cancel the effect of the `-mlong-calls'
15215 Use this attribute together with `interrupt_handler',
15216 `exception_handler' or `nmi_handler' to indicate that the function
15217 entry code should enable nested interrupts or exceptions.
15220 Use this attribute on the Blackfin to indicate that the specified
15221 function is an NMI handler. The compiler will generate function
15222 entry and exit sequences suitable for use in an NMI handler when
15223 this attribute is present.
15225 `no_instrument_function'
15226 If `-finstrument-functions' is given, profiling function calls will
15227 be generated at entry and exit of most user-compiled functions.
15228 Functions with this attribute will not be so instrumented.
15231 This function attribute prevents a function from being considered
15234 `nonnull (ARG-INDEX, ...)'
15235 The `nonnull' attribute specifies that some function parameters
15236 should be non-null pointers. For instance, the declaration:
15239 my_memcpy (void *dest, const void *src, size_t len)
15240 __attribute__((nonnull (1, 2)));
15242 causes the compiler to check that, in calls to `my_memcpy',
15243 arguments DEST and SRC are non-null. If the compiler determines
15244 that a null pointer is passed in an argument slot marked as
15245 non-null, and the `-Wnonnull' option is enabled, a warning is
15246 issued. The compiler may also choose to make optimizations based
15247 on the knowledge that certain function arguments will not be null.
15249 If no argument index list is given to the `nonnull' attribute, all
15250 pointer arguments are marked as non-null. To illustrate, the
15251 following declaration is equivalent to the previous example:
15254 my_memcpy (void *dest, const void *src, size_t len)
15255 __attribute__((nonnull));
15258 A few standard library functions, such as `abort' and `exit',
15259 cannot return. GCC knows this automatically. Some programs define
15260 their own functions that never return. You can declare them
15261 `noreturn' to tell the compiler this fact. For example,
15263 void fatal () __attribute__ ((noreturn));
15268 /* ... */ /* Print error message. */ /* ... */
15272 The `noreturn' keyword tells the compiler to assume that `fatal'
15273 cannot return. It can then optimize without regard to what would
15274 happen if `fatal' ever did return. This makes slightly better
15275 code. More importantly, it helps avoid spurious warnings of
15276 uninitialized variables.
15278 The `noreturn' keyword does not affect the exceptional path when
15279 that applies: a `noreturn'-marked function may still return to the
15280 caller by throwing an exception or calling `longjmp'.
15282 Do not assume that registers saved by the calling function are
15283 restored before calling the `noreturn' function.
15285 It does not make sense for a `noreturn' function to have a return
15286 type other than `void'.
15288 The attribute `noreturn' is not implemented in GCC versions
15289 earlier than 2.5. An alternative way to declare that a function
15290 does not return, which works in the current version and in some
15291 older versions, is as follows:
15293 typedef void voidfn ();
15295 volatile voidfn fatal;
15297 This approach does not work in GNU C++.
15300 The `nothrow' attribute is used to inform the compiler that a
15301 function cannot throw an exception. For example, most functions in
15302 the standard C library can be guaranteed not to throw an exception
15303 with the notable exceptions of `qsort' and `bsearch' that take
15304 function pointer arguments. The `nothrow' attribute is not
15305 implemented in GCC versions earlier than 3.3.
15308 Many functions have no effects except the return value and their
15309 return value depends only on the parameters and/or global
15310 variables. Such a function can be subject to common subexpression
15311 elimination and loop optimization just as an arithmetic operator
15312 would be. These functions should be declared with the attribute
15313 `pure'. For example,
15315 int square (int) __attribute__ ((pure));
15317 says that the hypothetical function `square' is safe to call fewer
15318 times than the program says.
15320 Some of common examples of pure functions are `strlen' or `memcmp'.
15321 Interesting non-pure functions are functions with infinite loops
15322 or those depending on volatile memory or other system resource,
15323 that may change between two consecutive calls (such as `feof' in a
15324 multithreading environment).
15326 The attribute `pure' is not implemented in GCC versions earlier
15330 On the Intel 386, the `regparm' attribute causes the compiler to
15331 pass arguments number one to NUMBER if they are of integral type
15332 in registers EAX, EDX, and ECX instead of on the stack. Functions
15333 that take a variable number of arguments will continue to be
15334 passed all of their arguments on the stack.
15336 Beware that on some ELF systems this attribute is unsuitable for
15337 global functions in shared libraries with lazy binding (which is
15338 the default). Lazy binding will send the first call via resolving
15339 code in the loader, which might assume EAX, EDX and ECX can be
15340 clobbered, as per the standard calling conventions. Solaris 8 is
15341 affected by this. GNU systems with GLIBC 2.1 or higher, and
15342 FreeBSD, are believed to be safe since the loaders there save all
15343 registers. (Lazy binding can be disabled with the linker or the
15344 loader if desired, to avoid the problem.)
15347 On the Intel 386 with SSE support, the `sseregparm' attribute
15348 causes the compiler to pass up to 3 floating point arguments in
15349 SSE registers instead of on the stack. Functions that take a
15350 variable number of arguments will continue to pass all of their
15351 floating point arguments on the stack.
15354 The `returns_twice' attribute tells the compiler that a function
15355 may return more than one time. The compiler will ensure that all
15356 registers are dead before calling such a function and will emit a
15357 warning about the variables that may be clobbered after the second
15358 return from the function. Examples of such functions are `setjmp'
15359 and `vfork'. The `longjmp'-like counterpart of such function, if
15360 any, might need to be marked with the `noreturn' attribute.
15363 Use this attribute on the Blackfin, H8/300, H8/300H, and H8S to
15364 indicate that all registers except the stack pointer should be
15365 saved in the prologue regardless of whether they are used or not.
15367 `section ("SECTION-NAME")'
15368 Normally, the compiler places the code it generates in the `text'
15369 section. Sometimes, however, you need additional sections, or you
15370 need certain particular functions to appear in special sections.
15371 The `section' attribute specifies that a function lives in a
15372 particular section. For example, the declaration:
15374 extern void foobar (void) __attribute__ ((section ("bar")));
15376 puts the function `foobar' in the `bar' section.
15378 Some file formats do not support arbitrary sections so the
15379 `section' attribute is not available on all platforms. If you
15380 need to map the entire contents of a module to a particular
15381 section, consider using the facilities of the linker instead.
15384 This function attribute ensures that a parameter in a function
15385 call is an explicit `NULL'. The attribute is only valid on
15386 variadic functions. By default, the sentinel is located at
15387 position zero, the last parameter of the function call. If an
15388 optional integer position argument P is supplied to the attribute,
15389 the sentinel must be located at position P counting backwards from
15390 the end of the argument list.
15392 __attribute__ ((sentinel))
15394 __attribute__ ((sentinel(0)))
15396 The attribute is automatically set with a position of 0 for the
15397 built-in functions `execl' and `execlp'. The built-in function
15398 `execle' has the attribute set with a position of 1.
15400 A valid `NULL' in this context is defined as zero with any pointer
15401 type. If your system defines the `NULL' macro with an integer type
15402 then you need to add an explicit cast. GCC replaces `stddef.h'
15403 with a copy that redefines NULL appropriately.
15405 The warnings for missing or incorrect sentinels are enabled with
15409 See long_call/short_call.
15412 See longcall/shortcall.
15415 Use this attribute on the AVR to indicate that the specified
15416 function is a signal handler. The compiler will generate function
15417 entry and exit sequences suitable for use in a signal handler when
15418 this attribute is present. Interrupts will be disabled inside the
15422 Use this attribute on the SH to indicate an `interrupt_handler'
15423 function should switch to an alternate stack. It expects a string
15424 argument that names a global variable holding the address of the
15428 void f () __attribute__ ((interrupt_handler,
15429 sp_switch ("alt_stack")));
15432 On the Intel 386, the `stdcall' attribute causes the compiler to
15433 assume that the called function will pop off the stack space used
15434 to pass arguments, unless it takes a variable number of arguments.
15437 Use this attribute on the H8/300H and H8S to indicate that the
15438 specified variable should be placed into the tiny data section.
15439 The compiler will generate more efficient code for loads and stores
15440 on data in the tiny data section. Note the tiny data area is
15441 limited to slightly under 32kbytes of data.
15444 Use this attribute on the SH for an `interrupt_handler' to return
15445 using `trapa' instead of `rte'. This attribute expects an integer
15446 argument specifying the trap number to be used.
15449 This attribute, attached to a function, means that the function is
15450 meant to be possibly unused. GCC will not produce a warning for
15454 This attribute, attached to a function, means that code must be
15455 emitted for the function even if it appears that the function is
15456 not referenced. This is useful, for example, when the function is
15457 referenced only in inline assembly.
15459 `visibility ("VISIBILITY_TYPE")'
15460 The `visibility' attribute on ELF targets causes the declaration
15461 to be emitted with default, hidden, protected or internal
15464 void __attribute__ ((visibility ("protected")))
15465 f () { /* Do something. */; }
15466 int i __attribute__ ((visibility ("hidden")));
15468 See the ELF gABI for complete details, but the short story is:
15471 Default visibility is the normal case for ELF. This value is
15472 available for the visibility attribute to override other
15473 options that may change the assumed visibility of symbols.
15476 Hidden visibility indicates that the symbol will not be
15477 placed into the dynamic symbol table, so no other "module"
15478 (executable or shared library) can reference it directly.
15481 Internal visibility is like hidden visibility, but with
15482 additional processor specific semantics. Unless otherwise
15483 specified by the psABI, GCC defines internal visibility to
15484 mean that the function is _never_ called from another module.
15485 Note that hidden symbols, while they cannot be referenced
15486 directly by other modules, can be referenced indirectly via
15487 function pointers. By indicating that a symbol cannot be
15488 called from outside the module, GCC may for instance omit the
15489 load of a PIC register since it is known that the calling
15490 function loaded the correct value.
15493 Protected visibility indicates that the symbol will be placed
15494 in the dynamic symbol table, but that references within the
15495 defining module will bind to the local symbol. That is, the
15496 symbol cannot be overridden by another module.
15499 Not all ELF targets support this attribute.
15501 `warn_unused_result'
15502 The `warn_unused_result' attribute causes a warning to be emitted
15503 if a caller of the function with this attribute does not use its
15504 return value. This is useful for functions where not checking the
15505 result is either a security problem or always a bug, such as
15508 int fn () __attribute__ ((warn_unused_result));
15511 if (fn () < 0) return -1;
15516 results in warning on line 5.
15519 The `weak' attribute causes the declaration to be emitted as a weak
15520 symbol rather than a global. This is primarily useful in defining
15521 library functions which can be overridden in user code, though it
15522 can also be used with non-function declarations. Weak symbols are
15523 supported for ELF targets, and also for a.out targets when using
15524 the GNU assembler and linker.
15527 `weakref ("TARGET")'
15528 The `weakref' attribute marks a declaration as a weak reference.
15529 Without arguments, it should be accompanied by an `alias' attribute
15530 naming the target symbol. Optionally, the TARGET may be given as
15531 an argument to `weakref' itself. In either case, `weakref'
15532 implicitly marks the declaration as `weak'. Without a TARGET,
15533 given as an argument to `weakref' or to `alias', `weakref' is
15534 equivalent to `weak'.
15536 extern int x() __attribute__ ((weakref ("y")));
15537 /* is equivalent to... */
15538 extern int x() __attribute__ ((weak, weakref, alias ("y")));
15540 extern int x() __attribute__ ((weakref));
15541 extern int x() __attribute__ ((alias ("y")));
15543 A weak reference is an alias that does not by itself require a
15544 definition to be given for the target symbol. If the target
15545 symbol is only referenced through weak references, then the
15546 becomes a `weak' undefined symbol. If it is directly referenced,
15547 however, then such strong references prevail, and a definition
15548 will be required for the symbol, not necessarily in the same
15551 The effect is equivalent to moving all references to the alias to a
15552 separate translation unit, renaming the alias to the aliased
15553 symbol, declaring it as weak, compiling the two separate
15554 translation units and performing a reloadable link on them.
15556 `externally_visible'
15557 This attribute, attached to a global variable or function nullify
15558 effect of `-fwhole-program' command line option, so the object
15559 remain visible outside the current compilation unit
15562 You can specify multiple attributes in a declaration by separating them
15563 by commas within the double parentheses or by immediately following an
15564 attribute declaration with another attribute declaration.
15566 Some people object to the `__attribute__' feature, suggesting that ISO
15567 C's `#pragma' should be used instead. At the time `__attribute__' was
15568 designed, there were two reasons for not doing this.
15570 1. It is impossible to generate `#pragma' commands from a macro.
15572 2. There is no telling what the same `#pragma' might mean in another
15575 These two reasons applied to almost any application that might have
15576 been proposed for `#pragma'. It was basically a mistake to use
15577 `#pragma' for _anything_.
15579 The ISO C99 standard includes `_Pragma', which now allows pragmas to
15580 be generated from macros. In addition, a `#pragma GCC' namespace is
15581 now in use for GCC-specific pragmas. However, it has been found
15582 convenient to use `__attribute__' to achieve a natural attachment of
15583 attributes to their corresponding declarations, whereas `#pragma GCC'
15584 is of use for constructs that do not naturally form part of the
15585 grammar. *Note Miscellaneous Preprocessing Directives: (cpp)Other
15589 File: gcc.info, Node: Attribute Syntax, Next: Function Prototypes, Prev: Function Attributes, Up: C Extensions
15591 5.25 Attribute Syntax
15592 =====================
15594 This section describes the syntax with which `__attribute__' may be
15595 used, and the constructs to which attribute specifiers bind, for the C
15596 language. Some details may vary for C++ and Objective-C. Because of
15597 infelicities in the grammar for attributes, some forms described here
15598 may not be successfully parsed in all cases.
15600 There are some problems with the semantics of attributes in C++. For
15601 example, there are no manglings for attributes, although they may affect
15602 code generation, so problems may arise when attributed types are used in
15603 conjunction with templates or overloading. Similarly, `typeid' does
15604 not distinguish between types with different attributes. Support for
15605 attributes in C++ may be restricted in future to attributes on
15606 declarations only, but not on nested declarators.
15608 *Note Function Attributes::, for details of the semantics of attributes
15609 applying to functions. *Note Variable Attributes::, for details of the
15610 semantics of attributes applying to variables. *Note Type Attributes::,
15611 for details of the semantics of attributes applying to structure, union
15612 and enumerated types.
15614 An "attribute specifier" is of the form `__attribute__
15615 ((ATTRIBUTE-LIST))'. An "attribute list" is a possibly empty
15616 comma-separated sequence of "attributes", where each attribute is one
15619 * Empty. Empty attributes are ignored.
15621 * A word (which may be an identifier such as `unused', or a reserved
15622 word such as `const').
15624 * A word, followed by, in parentheses, parameters for the attribute.
15625 These parameters take one of the following forms:
15627 * An identifier. For example, `mode' attributes use this form.
15629 * An identifier followed by a comma and a non-empty
15630 comma-separated list of expressions. For example, `format'
15631 attributes use this form.
15633 * A possibly empty comma-separated list of expressions. For
15634 example, `format_arg' attributes use this form with the list
15635 being a single integer constant expression, and `alias'
15636 attributes use this form with the list being a single string
15639 An "attribute specifier list" is a sequence of one or more attribute
15640 specifiers, not separated by any other tokens.
15642 In GNU C, an attribute specifier list may appear after the colon
15643 following a label, other than a `case' or `default' label. The only
15644 attribute it makes sense to use after a label is `unused'. This
15645 feature is intended for code generated by programs which contains labels
15646 that may be unused but which is compiled with `-Wall'. It would not
15647 normally be appropriate to use in it human-written code, though it
15648 could be useful in cases where the code that jumps to the label is
15649 contained within an `#ifdef' conditional. GNU C++ does not permit such
15650 placement of attribute lists, as it is permissible for a declaration,
15651 which could begin with an attribute list, to be labelled in C++.
15652 Declarations cannot be labelled in C90 or C99, so the ambiguity does
15655 An attribute specifier list may appear as part of a `struct', `union'
15656 or `enum' specifier. It may go either immediately after the `struct',
15657 `union' or `enum' keyword, or after the closing brace. It is ignored
15658 if the content of the structure, union or enumerated type is not
15659 defined in the specifier in which the attribute specifier list is
15660 used--that is, in usages such as `struct __attribute__((foo)) bar' with
15661 no following opening brace. Where attribute specifiers follow the
15662 closing brace, they are considered to relate to the structure, union or
15663 enumerated type defined, not to any enclosing declaration the type
15664 specifier appears in, and the type defined is not complete until after
15665 the attribute specifiers.
15667 Otherwise, an attribute specifier appears as part of a declaration,
15668 counting declarations of unnamed parameters and type names, and relates
15669 to that declaration (which may be nested in another declaration, for
15670 example in the case of a parameter declaration), or to a particular
15671 declarator within a declaration. Where an attribute specifier is
15672 applied to a parameter declared as a function or an array, it should
15673 apply to the function or array rather than the pointer to which the
15674 parameter is implicitly converted, but this is not yet correctly
15677 Any list of specifiers and qualifiers at the start of a declaration may
15678 contain attribute specifiers, whether or not such a list may in that
15679 context contain storage class specifiers. (Some attributes, however,
15680 are essentially in the nature of storage class specifiers, and only make
15681 sense where storage class specifiers may be used; for example,
15682 `section'.) There is one necessary limitation to this syntax: the
15683 first old-style parameter declaration in a function definition cannot
15684 begin with an attribute specifier, because such an attribute applies to
15685 the function instead by syntax described below (which, however, is not
15686 yet implemented in this case). In some other cases, attribute
15687 specifiers are permitted by this grammar but not yet supported by the
15688 compiler. All attribute specifiers in this place relate to the
15689 declaration as a whole. In the obsolescent usage where a type of `int'
15690 is implied by the absence of type specifiers, such a list of specifiers
15691 and qualifiers may be an attribute specifier list with no other
15692 specifiers or qualifiers.
15694 At present, the first parameter in a function prototype must have some
15695 type specifier which is not an attribute specifier; this resolves an
15696 ambiguity in the interpretation of `void f(int (__attribute__((foo))
15697 x))', but is subject to change. At present, if the parentheses of a
15698 function declarator contain only attributes then those attributes are
15699 ignored, rather than yielding an error or warning or implying a single
15700 parameter of type int, but this is subject to change.
15702 An attribute specifier list may appear immediately before a declarator
15703 (other than the first) in a comma-separated list of declarators in a
15704 declaration of more than one identifier using a single list of
15705 specifiers and qualifiers. Such attribute specifiers apply only to the
15706 identifier before whose declarator they appear. For example, in
15708 __attribute__((noreturn)) void d0 (void),
15709 __attribute__((format(printf, 1, 2))) d1 (const char *, ...),
15712 the `noreturn' attribute applies to all the functions declared; the
15713 `format' attribute only applies to `d1'.
15715 An attribute specifier list may appear immediately before the comma,
15716 `=' or semicolon terminating the declaration of an identifier other
15717 than a function definition. At present, such attribute specifiers apply
15718 to the declared object or function, but in future they may attach to the
15719 outermost adjacent declarator. In simple cases there is no difference,
15720 but, for example, in
15722 void (****f)(void) __attribute__((noreturn));
15724 at present the `noreturn' attribute applies to `f', which causes a
15725 warning since `f' is not a function, but in future it may apply to the
15726 function `****f'. The precise semantics of what attributes in such
15727 cases will apply to are not yet specified. Where an assembler name for
15728 an object or function is specified (*note Asm Labels::), at present the
15729 attribute must follow the `asm' specification; in future, attributes
15730 before the `asm' specification may apply to the adjacent declarator,
15731 and those after it to the declared object or function.
15733 An attribute specifier list may, in future, be permitted to appear
15734 after the declarator in a function definition (before any old-style
15735 parameter declarations or the function body).
15737 Attribute specifiers may be mixed with type qualifiers appearing inside
15738 the `[]' of a parameter array declarator, in the C99 construct by which
15739 such qualifiers are applied to the pointer to which the array is
15740 implicitly converted. Such attribute specifiers apply to the pointer,
15741 not to the array, but at present this is not implemented and they are
15744 An attribute specifier list may appear at the start of a nested
15745 declarator. At present, there are some limitations in this usage: the
15746 attributes correctly apply to the declarator, but for most individual
15747 attributes the semantics this implies are not implemented. When
15748 attribute specifiers follow the `*' of a pointer declarator, they may
15749 be mixed with any type qualifiers present. The following describes the
15750 formal semantics of this syntax. It will make the most sense if you
15751 are familiar with the formal specification of declarators in the ISO C
15754 Consider (as in C99 subclause 6.7.5 paragraph 4) a declaration `T D1',
15755 where `T' contains declaration specifiers that specify a type TYPE
15756 (such as `int') and `D1' is a declarator that contains an identifier
15757 IDENT. The type specified for IDENT for derived declarators whose type
15758 does not include an attribute specifier is as in the ISO C standard.
15760 If `D1' has the form `( ATTRIBUTE-SPECIFIER-LIST D )', and the
15761 declaration `T D' specifies the type "DERIVED-DECLARATOR-TYPE-LIST
15762 TYPE" for IDENT, then `T D1' specifies the type
15763 "DERIVED-DECLARATOR-TYPE-LIST ATTRIBUTE-SPECIFIER-LIST TYPE" for IDENT.
15765 If `D1' has the form `* TYPE-QUALIFIER-AND-ATTRIBUTE-SPECIFIER-LIST
15766 D', and the declaration `T D' specifies the type
15767 "DERIVED-DECLARATOR-TYPE-LIST TYPE" for IDENT, then `T D1' specifies
15768 the type "DERIVED-DECLARATOR-TYPE-LIST
15769 TYPE-QUALIFIER-AND-ATTRIBUTE-SPECIFIER-LIST TYPE" for IDENT.
15773 void (__attribute__((noreturn)) ****f) (void);
15775 specifies the type "pointer to pointer to pointer to pointer to
15776 non-returning function returning `void'". As another example,
15778 char *__attribute__((aligned(8))) *f;
15780 specifies the type "pointer to 8-byte-aligned pointer to `char'". Note
15781 again that this does not work with most attributes; for example, the
15782 usage of `aligned' and `noreturn' attributes given above is not yet
15785 For compatibility with existing code written for compiler versions that
15786 did not implement attributes on nested declarators, some laxity is
15787 allowed in the placing of attributes. If an attribute that only applies
15788 to types is applied to a declaration, it will be treated as applying to
15789 the type of that declaration. If an attribute that only applies to
15790 declarations is applied to the type of a declaration, it will be treated
15791 as applying to that declaration; and, for compatibility with code
15792 placing the attributes immediately before the identifier declared, such
15793 an attribute applied to a function return type will be treated as
15794 applying to the function type, and such an attribute applied to an array
15795 element type will be treated as applying to the array type. If an
15796 attribute that only applies to function types is applied to a
15797 pointer-to-function type, it will be treated as applying to the pointer
15798 target type; if such an attribute is applied to a function return type
15799 that is not a pointer-to-function type, it will be treated as applying
15800 to the function type.
15803 File: gcc.info, Node: Function Prototypes, Next: C++ Comments, Prev: Attribute Syntax, Up: C Extensions
15805 5.26 Prototypes and Old-Style Function Definitions
15806 ==================================================
15808 GNU C extends ISO C to allow a function prototype to override a later
15809 old-style non-prototype definition. Consider the following example:
15811 /* Use prototypes unless the compiler is old-fashioned. */
15818 /* Prototype function declaration. */
15819 int isroot P((uid_t));
15821 /* Old-style function definition. */
15823 isroot (x) /* ??? lossage here ??? */
15829 Suppose the type `uid_t' happens to be `short'. ISO C does not allow
15830 this example, because subword arguments in old-style non-prototype
15831 definitions are promoted. Therefore in this example the function
15832 definition's argument is really an `int', which does not match the
15833 prototype argument type of `short'.
15835 This restriction of ISO C makes it hard to write code that is portable
15836 to traditional C compilers, because the programmer does not know
15837 whether the `uid_t' type is `short', `int', or `long'. Therefore, in
15838 cases like these GNU C allows a prototype to override a later old-style
15839 definition. More precisely, in GNU C, a function prototype argument
15840 type overrides the argument type specified by a later old-style
15841 definition if the former type is the same as the latter type before
15842 promotion. Thus in GNU C the above example is equivalent to the
15845 int isroot (uid_t);
15853 GNU C++ does not support old-style function definitions, so this
15854 extension is irrelevant.
15857 File: gcc.info, Node: C++ Comments, Next: Dollar Signs, Prev: Function Prototypes, Up: C Extensions
15859 5.27 C++ Style Comments
15860 =======================
15862 In GNU C, you may use C++ style comments, which start with `//' and
15863 continue until the end of the line. Many other C implementations allow
15864 such comments, and they are included in the 1999 C standard. However,
15865 C++ style comments are not recognized if you specify an `-std' option
15866 specifying a version of ISO C before C99, or `-ansi' (equivalent to
15870 File: gcc.info, Node: Dollar Signs, Next: Character Escapes, Prev: C++ Comments, Up: C Extensions
15872 5.28 Dollar Signs in Identifier Names
15873 =====================================
15875 In GNU C, you may normally use dollar signs in identifier names. This
15876 is because many traditional C implementations allow such identifiers.
15877 However, dollar signs in identifiers are not supported on a few target
15878 machines, typically because the target assembler does not allow them.
15881 File: gcc.info, Node: Character Escapes, Next: Variable Attributes, Prev: Dollar Signs, Up: C Extensions
15883 5.29 The Character <ESC> in Constants
15884 =====================================
15886 You can use the sequence `\e' in a string or character constant to
15887 stand for the ASCII character <ESC>.
15890 File: gcc.info, Node: Alignment, Next: Inline, Prev: Type Attributes, Up: C Extensions
15892 5.30 Inquiring on Alignment of Types or Variables
15893 =================================================
15895 The keyword `__alignof__' allows you to inquire about how an object is
15896 aligned, or the minimum alignment usually required by a type. Its
15897 syntax is just like `sizeof'.
15899 For example, if the target machine requires a `double' value to be
15900 aligned on an 8-byte boundary, then `__alignof__ (double)' is 8. This
15901 is true on many RISC machines. On more traditional machine designs,
15902 `__alignof__ (double)' is 4 or even 2.
15904 Some machines never actually require alignment; they allow reference
15905 to any data type even at an odd address. For these machines,
15906 `__alignof__' reports the _recommended_ alignment of a type.
15908 If the operand of `__alignof__' is an lvalue rather than a type, its
15909 value is the required alignment for its type, taking into account any
15910 minimum alignment specified with GCC's `__attribute__' extension (*note
15911 Variable Attributes::). For example, after this declaration:
15913 struct foo { int x; char y; } foo1;
15915 the value of `__alignof__ (foo1.y)' is 1, even though its actual
15916 alignment is probably 2 or 4, the same as `__alignof__ (int)'.
15918 It is an error to ask for the alignment of an incomplete type.
15921 File: gcc.info, Node: Variable Attributes, Next: Type Attributes, Prev: Character Escapes, Up: C Extensions
15923 5.31 Specifying Attributes of Variables
15924 =======================================
15926 The keyword `__attribute__' allows you to specify special attributes of
15927 variables or structure fields. This keyword is followed by an
15928 attribute specification inside double parentheses. Some attributes are
15929 currently defined generically for variables. Other attributes are
15930 defined for variables on particular target systems. Other attributes
15931 are available for functions (*note Function Attributes::) and for types
15932 (*note Type Attributes::). Other front ends might define more
15933 attributes (*note Extensions to the C++ Language: C++ Extensions.).
15935 You may also specify attributes with `__' preceding and following each
15936 keyword. This allows you to use them in header files without being
15937 concerned about a possible macro of the same name. For example, you
15938 may use `__aligned__' instead of `aligned'.
15940 *Note Attribute Syntax::, for details of the exact syntax for using
15943 `aligned (ALIGNMENT)'
15944 This attribute specifies a minimum alignment for the variable or
15945 structure field, measured in bytes. For example, the declaration:
15947 int x __attribute__ ((aligned (16))) = 0;
15949 causes the compiler to allocate the global variable `x' on a
15950 16-byte boundary. On a 68040, this could be used in conjunction
15951 with an `asm' expression to access the `move16' instruction which
15952 requires 16-byte aligned operands.
15954 You can also specify the alignment of structure fields. For
15955 example, to create a double-word aligned `int' pair, you could
15958 struct foo { int x[2] __attribute__ ((aligned (8))); };
15960 This is an alternative to creating a union with a `double' member
15961 that forces the union to be double-word aligned.
15963 As in the preceding examples, you can explicitly specify the
15964 alignment (in bytes) that you wish the compiler to use for a given
15965 variable or structure field. Alternatively, you can leave out the
15966 alignment factor and just ask the compiler to align a variable or
15967 field to the maximum useful alignment for the target machine you
15968 are compiling for. For example, you could write:
15970 short array[3] __attribute__ ((aligned));
15972 Whenever you leave out the alignment factor in an `aligned'
15973 attribute specification, the compiler automatically sets the
15974 alignment for the declared variable or field to the largest
15975 alignment which is ever used for any data type on the target
15976 machine you are compiling for. Doing this can often make copy
15977 operations more efficient, because the compiler can use whatever
15978 instructions copy the biggest chunks of memory when performing
15979 copies to or from the variables or fields that you have aligned
15982 The `aligned' attribute can only increase the alignment; but you
15983 can decrease it by specifying `packed' as well. See below.
15985 Note that the effectiveness of `aligned' attributes may be limited
15986 by inherent limitations in your linker. On many systems, the
15987 linker is only able to arrange for variables to be aligned up to a
15988 certain maximum alignment. (For some linkers, the maximum
15989 supported alignment may be very very small.) If your linker is
15990 only able to align variables up to a maximum of 8 byte alignment,
15991 then specifying `aligned(16)' in an `__attribute__' will still
15992 only provide you with 8 byte alignment. See your linker
15993 documentation for further information.
15995 `cleanup (CLEANUP_FUNCTION)'
15996 The `cleanup' attribute runs a function when the variable goes out
15997 of scope. This attribute can only be applied to auto function
15998 scope variables; it may not be applied to parameters or variables
15999 with static storage duration. The function must take one
16000 parameter, a pointer to a type compatible with the variable. The
16001 return value of the function (if any) is ignored.
16003 If `-fexceptions' is enabled, then CLEANUP_FUNCTION will be run
16004 during the stack unwinding that happens during the processing of
16005 the exception. Note that the `cleanup' attribute does not allow
16006 the exception to be caught, only to perform an action. It is
16007 undefined what happens if CLEANUP_FUNCTION does not return
16012 The `common' attribute requests GCC to place a variable in
16013 "common" storage. The `nocommon' attribute requests the
16014 opposite--to allocate space for it directly.
16016 These attributes override the default chosen by the `-fno-common'
16017 and `-fcommon' flags respectively.
16020 The `deprecated' attribute results in a warning if the variable is
16021 used anywhere in the source file. This is useful when identifying
16022 variables that are expected to be removed in a future version of a
16023 program. The warning also includes the location of the declaration
16024 of the deprecated variable, to enable users to easily find further
16025 information about why the variable is deprecated, or what they
16026 should do instead. Note that the warning only occurs for uses:
16028 extern int old_var __attribute__ ((deprecated));
16029 extern int old_var;
16030 int new_fn () { return old_var; }
16032 results in a warning on line 3 but not line 2.
16034 The `deprecated' attribute can also be used for functions and
16035 types (*note Function Attributes::, *note Type Attributes::.)
16038 This attribute specifies the data type for the
16039 declaration--whichever type corresponds to the mode MODE. This in
16040 effect lets you request an integer or floating point type
16041 according to its width.
16043 You may also specify a mode of `byte' or `__byte__' to indicate
16044 the mode corresponding to a one-byte integer, `word' or `__word__'
16045 for the mode of a one-word integer, and `pointer' or `__pointer__'
16046 for the mode used to represent pointers.
16049 The `packed' attribute specifies that a variable or structure field
16050 should have the smallest possible alignment--one byte for a
16051 variable, and one bit for a field, unless you specify a larger
16052 value with the `aligned' attribute.
16054 Here is a structure in which the field `x' is packed, so that it
16055 immediately follows `a':
16060 int x[2] __attribute__ ((packed));
16063 `section ("SECTION-NAME")'
16064 Normally, the compiler places the objects it generates in sections
16065 like `data' and `bss'. Sometimes, however, you need additional
16066 sections, or you need certain particular variables to appear in
16067 special sections, for example to map to special hardware. The
16068 `section' attribute specifies that a variable (or function) lives
16069 in a particular section. For example, this small program uses
16070 several specific section names:
16072 struct duart a __attribute__ ((section ("DUART_A"))) = { 0 };
16073 struct duart b __attribute__ ((section ("DUART_B"))) = { 0 };
16074 char stack[10000] __attribute__ ((section ("STACK"))) = { 0 };
16075 int init_data __attribute__ ((section ("INITDATA"))) = 0;
16079 /* Initialize stack pointer */
16080 init_sp (stack + sizeof (stack));
16082 /* Initialize initialized data */
16083 memcpy (&init_data, &data, &edata - &data);
16085 /* Turn on the serial ports */
16090 Use the `section' attribute with an _initialized_ definition of a
16091 _global_ variable, as shown in the example. GCC issues a warning
16092 and otherwise ignores the `section' attribute in uninitialized
16093 variable declarations.
16095 You may only use the `section' attribute with a fully initialized
16096 global definition because of the way linkers work. The linker
16097 requires each object be defined once, with the exception that
16098 uninitialized variables tentatively go in the `common' (or `bss')
16099 section and can be multiply "defined". You can force a variable
16100 to be initialized with the `-fno-common' flag or the `nocommon'
16103 Some file formats do not support arbitrary sections so the
16104 `section' attribute is not available on all platforms. If you
16105 need to map the entire contents of a module to a particular
16106 section, consider using the facilities of the linker instead.
16109 On Microsoft Windows, in addition to putting variable definitions
16110 in a named section, the section can also be shared among all
16111 running copies of an executable or DLL. For example, this small
16112 program defines shared data by putting it in a named section
16113 `shared' and marking the section shareable:
16115 int foo __attribute__((section ("shared"), shared)) = 0;
16120 /* Read and write foo. All running
16121 copies see the same value. */
16125 You may only use the `shared' attribute along with `section'
16126 attribute with a fully initialized global definition because of
16127 the way linkers work. See `section' attribute for more
16130 The `shared' attribute is only available on Microsoft Windows.
16132 `tls_model ("TLS_MODEL")'
16133 The `tls_model' attribute sets thread-local storage model (*note
16134 Thread-Local::) of a particular `__thread' variable, overriding
16135 `-ftls-model=' command line switch on a per-variable basis. The
16136 TLS_MODEL argument should be one of `global-dynamic',
16137 `local-dynamic', `initial-exec' or `local-exec'.
16139 Not all targets support this attribute.
16142 This attribute, attached to a variable, means that the variable is
16143 meant to be possibly unused. GCC will not produce a warning for
16146 `vector_size (BYTES)'
16147 This attribute specifies the vector size for the variable,
16148 measured in bytes. For example, the declaration:
16150 int foo __attribute__ ((vector_size (16)));
16152 causes the compiler to set the mode for `foo', to be 16 bytes,
16153 divided into `int' sized units. Assuming a 32-bit int (a vector of
16154 4 units of 4 bytes), the corresponding mode of `foo' will be V4SI.
16156 This attribute is only applicable to integral and float scalars,
16157 although arrays, pointers, and function return values are allowed
16158 in conjunction with this construct.
16160 Aggregates with this attribute are invalid, even if they are of
16161 the same size as a corresponding scalar. For example, the
16164 struct S { int a; };
16165 struct S __attribute__ ((vector_size (16))) foo;
16167 is invalid even if the size of the structure is the same as the
16171 The `selectany' attribute causes an initialized global variable to
16172 have link-once semantics. When multiple definitions of the
16173 variable are encountered by the linker, the first is selected and
16174 the remainder are discarded. Following usage by the Microsoft
16175 compiler, the linker is told _not_ to warn about size or content
16176 differences of the multiple definitions.
16178 Although the primary usage of this attribute is for POD types, the
16179 attribute can also be applied to global C++ objects that are
16180 initialized by a constructor. In this case, the static
16181 initialization and destruction code for the object is emitted in
16182 each translation defining the object, but the calls to the
16183 constructor and destructor are protected by a link-once guard
16186 The `selectany' attribute is only available on Microsoft Windows
16187 targets. You can use `__declspec (selectany)' as a synonym for
16188 `__attribute__ ((selectany))' for compatibility with other
16192 The `weak' attribute is described in *Note Function Attributes::.
16195 The `dllimport' attribute is described in *Note Function
16199 The `dllexport' attribute is described in *Note Function
16203 5.31.1 M32R/D Variable Attributes
16204 ---------------------------------
16206 One attribute is currently defined for the M32R/D.
16208 `model (MODEL-NAME)'
16209 Use this attribute on the M32R/D to set the addressability of an
16210 object. The identifier MODEL-NAME is one of `small', `medium', or
16211 `large', representing each of the code models.
16213 Small model objects live in the lower 16MB of memory (so that their
16214 addresses can be loaded with the `ld24' instruction).
16216 Medium and large model objects may live anywhere in the 32-bit
16217 address space (the compiler will generate `seth/add3' instructions
16218 to load their addresses).
16220 5.31.2 i386 Variable Attributes
16221 -------------------------------
16223 Two attributes are currently defined for i386 configurations:
16224 `ms_struct' and `gcc_struct'
16228 If `packed' is used on a structure, or if bit-fields are used it
16229 may be that the Microsoft ABI packs them differently than GCC
16230 would normally pack them. Particularly when moving packed data
16231 between functions compiled with GCC and the native Microsoft
16232 compiler (either via function call or as data in a file), it may
16233 be necessary to access either format.
16235 Currently `-m[no-]ms-bitfields' is provided for the Microsoft
16236 Windows X86 compilers to match the native Microsoft compiler.
16238 5.31.3 Xstormy16 Variable Attributes
16239 ------------------------------------
16241 One attribute is currently defined for xstormy16 configurations:
16245 If a variable has the `below100' attribute (`BELOW100' is allowed
16246 also), GCC will place the variable in the first 0x100 bytes of
16247 memory and use special opcodes to access it. Such variables will
16248 be placed in either the `.bss_below100' section or the
16249 `.data_below100' section.
16253 File: gcc.info, Node: Type Attributes, Next: Alignment, Prev: Variable Attributes, Up: C Extensions
16255 5.32 Specifying Attributes of Types
16256 ===================================
16258 The keyword `__attribute__' allows you to specify special attributes of
16259 `struct' and `union' types when you define such types. This keyword is
16260 followed by an attribute specification inside double parentheses. Six
16261 attributes are currently defined for types: `aligned', `packed',
16262 `transparent_union', `unused', `deprecated' and `may_alias'. Other
16263 attributes are defined for functions (*note Function Attributes::) and
16264 for variables (*note Variable Attributes::).
16266 You may also specify any one of these attributes with `__' preceding
16267 and following its keyword. This allows you to use these attributes in
16268 header files without being concerned about a possible macro of the same
16269 name. For example, you may use `__aligned__' instead of `aligned'.
16271 You may specify the `aligned' and `transparent_union' attributes
16272 either in a `typedef' declaration or just past the closing curly brace
16273 of a complete enum, struct or union type _definition_ and the `packed'
16274 attribute only past the closing brace of a definition.
16276 You may also specify attributes between the enum, struct or union tag
16277 and the name of the type rather than after the closing brace.
16279 *Note Attribute Syntax::, for details of the exact syntax for using
16282 `aligned (ALIGNMENT)'
16283 This attribute specifies a minimum alignment (in bytes) for
16284 variables of the specified type. For example, the declarations:
16286 struct S { short f[3]; } __attribute__ ((aligned (8)));
16287 typedef int more_aligned_int __attribute__ ((aligned (8)));
16289 force the compiler to insure (as far as it can) that each variable
16290 whose type is `struct S' or `more_aligned_int' will be allocated
16291 and aligned _at least_ on a 8-byte boundary. On a SPARC, having
16292 all variables of type `struct S' aligned to 8-byte boundaries
16293 allows the compiler to use the `ldd' and `std' (doubleword load and
16294 store) instructions when copying one variable of type `struct S' to
16295 another, thus improving run-time efficiency.
16297 Note that the alignment of any given `struct' or `union' type is
16298 required by the ISO C standard to be at least a perfect multiple of
16299 the lowest common multiple of the alignments of all of the members
16300 of the `struct' or `union' in question. This means that you _can_
16301 effectively adjust the alignment of a `struct' or `union' type by
16302 attaching an `aligned' attribute to any one of the members of such
16303 a type, but the notation illustrated in the example above is a
16304 more obvious, intuitive, and readable way to request the compiler
16305 to adjust the alignment of an entire `struct' or `union' type.
16307 As in the preceding example, you can explicitly specify the
16308 alignment (in bytes) that you wish the compiler to use for a given
16309 `struct' or `union' type. Alternatively, you can leave out the
16310 alignment factor and just ask the compiler to align a type to the
16311 maximum useful alignment for the target machine you are compiling
16312 for. For example, you could write:
16314 struct S { short f[3]; } __attribute__ ((aligned));
16316 Whenever you leave out the alignment factor in an `aligned'
16317 attribute specification, the compiler automatically sets the
16318 alignment for the type to the largest alignment which is ever used
16319 for any data type on the target machine you are compiling for.
16320 Doing this can often make copy operations more efficient, because
16321 the compiler can use whatever instructions copy the biggest chunks
16322 of memory when performing copies to or from the variables which
16323 have types that you have aligned this way.
16325 In the example above, if the size of each `short' is 2 bytes, then
16326 the size of the entire `struct S' type is 6 bytes. The smallest
16327 power of two which is greater than or equal to that is 8, so the
16328 compiler sets the alignment for the entire `struct S' type to 8
16331 Note that although you can ask the compiler to select a
16332 time-efficient alignment for a given type and then declare only
16333 individual stand-alone objects of that type, the compiler's
16334 ability to select a time-efficient alignment is primarily useful
16335 only when you plan to create arrays of variables having the
16336 relevant (efficiently aligned) type. If you declare or use arrays
16337 of variables of an efficiently-aligned type, then it is likely
16338 that your program will also be doing pointer arithmetic (or
16339 subscripting, which amounts to the same thing) on pointers to the
16340 relevant type, and the code that the compiler generates for these
16341 pointer arithmetic operations will often be more efficient for
16342 efficiently-aligned types than for other types.
16344 The `aligned' attribute can only increase the alignment; but you
16345 can decrease it by specifying `packed' as well. See below.
16347 Note that the effectiveness of `aligned' attributes may be limited
16348 by inherent limitations in your linker. On many systems, the
16349 linker is only able to arrange for variables to be aligned up to a
16350 certain maximum alignment. (For some linkers, the maximum
16351 supported alignment may be very very small.) If your linker is
16352 only able to align variables up to a maximum of 8 byte alignment,
16353 then specifying `aligned(16)' in an `__attribute__' will still
16354 only provide you with 8 byte alignment. See your linker
16355 documentation for further information.
16358 This attribute, attached to `struct' or `union' type definition,
16359 specifies that each member (other than zero-width bitfields) of
16360 the structure or union is placed to minimize the memory required.
16361 When attached to an `enum' definition, it indicates that the
16362 smallest integral type should be used.
16364 Specifying this attribute for `struct' and `union' types is
16365 equivalent to specifying the `packed' attribute on each of the
16366 structure or union members. Specifying the `-fshort-enums' flag
16367 on the line is equivalent to specifying the `packed' attribute on
16368 all `enum' definitions.
16370 In the following example `struct my_packed_struct''s members are
16371 packed closely together, but the internal layout of its `s' member
16372 is not packed--to do that, `struct my_unpacked_struct' would need
16375 struct my_unpacked_struct
16381 struct __attribute__ ((__packed__)) my_packed_struct
16385 struct my_unpacked_struct s;
16388 You may only specify this attribute on the definition of a `enum',
16389 `struct' or `union', not on a `typedef' which does not also define
16390 the enumerated type, structure or union.
16392 `transparent_union'
16393 This attribute, attached to a `union' type definition, indicates
16394 that any function parameter having that union type causes calls to
16395 that function to be treated in a special way.
16397 First, the argument corresponding to a transparent union type can
16398 be of any type in the union; no cast is required. Also, if the
16399 union contains a pointer type, the corresponding argument can be a
16400 null pointer constant or a void pointer expression; and if the
16401 union contains a void pointer type, the corresponding argument can
16402 be any pointer expression. If the union member type is a pointer,
16403 qualifiers like `const' on the referenced type must be respected,
16404 just as with normal pointer conversions.
16406 Second, the argument is passed to the function using the calling
16407 conventions of the first member of the transparent union, not the
16408 calling conventions of the union itself. All members of the union
16409 must have the same machine representation; this is necessary for
16410 this argument passing to work properly.
16412 Transparent unions are designed for library functions that have
16413 multiple interfaces for compatibility reasons. For example,
16414 suppose the `wait' function must accept either a value of type
16415 `int *' to comply with Posix, or a value of type `union wait *' to
16416 comply with the 4.1BSD interface. If `wait''s parameter were
16417 `void *', `wait' would accept both kinds of arguments, but it
16418 would also accept any other pointer type and this would make
16419 argument type checking less useful. Instead, `<sys/wait.h>' might
16420 define the interface as follows:
16426 } wait_status_ptr_t __attribute__ ((__transparent_union__));
16428 pid_t wait (wait_status_ptr_t);
16430 This interface allows either `int *' or `union wait *' arguments
16431 to be passed, using the `int *' calling convention. The program
16432 can call `wait' with arguments of either type:
16434 int w1 () { int w; return wait (&w); }
16435 int w2 () { union wait w; return wait (&w); }
16437 With this interface, `wait''s implementation might look like this:
16439 pid_t wait (wait_status_ptr_t p)
16441 return waitpid (-1, p.__ip, 0);
16445 When attached to a type (including a `union' or a `struct'), this
16446 attribute means that variables of that type are meant to appear
16447 possibly unused. GCC will not produce a warning for any variables
16448 of that type, even if the variable appears to do nothing. This is
16449 often the case with lock or thread classes, which are usually
16450 defined and then not referenced, but contain constructors and
16451 destructors that have nontrivial bookkeeping functions.
16454 The `deprecated' attribute results in a warning if the type is
16455 used anywhere in the source file. This is useful when identifying
16456 types that are expected to be removed in a future version of a
16457 program. If possible, the warning also includes the location of
16458 the declaration of the deprecated type, to enable users to easily
16459 find further information about why the type is deprecated, or what
16460 they should do instead. Note that the warnings only occur for
16461 uses and then only if the type is being applied to an identifier
16462 that itself is not being declared as deprecated.
16464 typedef int T1 __attribute__ ((deprecated));
16468 typedef T1 T3 __attribute__ ((deprecated));
16469 T3 z __attribute__ ((deprecated));
16471 results in a warning on line 2 and 3 but not lines 4, 5, or 6. No
16472 warning is issued for line 4 because T2 is not explicitly
16473 deprecated. Line 5 has no warning because T3 is explicitly
16474 deprecated. Similarly for line 6.
16476 The `deprecated' attribute can also be used for functions and
16477 variables (*note Function Attributes::, *note Variable
16481 Accesses to objects with types with this attribute are not
16482 subjected to type-based alias analysis, but are instead assumed to
16483 be able to alias any other type of objects, just like the `char'
16484 type. See `-fstrict-aliasing' for more information on aliasing
16489 typedef short __attribute__((__may_alias__)) short_a;
16494 int a = 0x12345678;
16495 short_a *b = (short_a *) &a;
16499 if (a == 0x12345678)
16505 If you replaced `short_a' with `short' in the variable
16506 declaration, the above program would abort when compiled with
16507 `-fstrict-aliasing', which is on by default at `-O2' or above in
16508 recent GCC versions.
16510 5.32.1 ARM Type Attributes
16511 --------------------------
16513 On those ARM targets that support `dllimport' (such as Symbian
16514 OS), you can use the `notshared' attribute to indicate that the virtual
16515 table and other similar data for a class should not be exported from a
16518 class __declspec(notshared) C {
16520 __declspec(dllimport) C();
16524 __declspec(dllexport)
16527 In this code, `C::C' is exported from the current DLL, but the
16528 virtual table for `C' is not exported. (You can use `__attribute__'
16529 instead of `__declspec' if you prefer, but most Symbian OS code uses
16532 5.32.2 i386 Type Attributes
16533 ---------------------------
16535 Two attributes are currently defined for i386 configurations:
16536 `ms_struct' and `gcc_struct'
16540 If `packed' is used on a structure, or if bit-fields are used it
16541 may be that the Microsoft ABI packs them differently than GCC
16542 would normally pack them. Particularly when moving packed data
16543 between functions compiled with GCC and the native Microsoft
16544 compiler (either via function call or as data in a file), it may
16545 be necessary to access either format.
16547 Currently `-m[no-]ms-bitfields' is provided for the Microsoft
16548 Windows X86 compilers to match the native Microsoft compiler.
16550 To specify multiple attributes, separate them by commas within the
16551 double parentheses: for example, `__attribute__ ((aligned (16),
16555 File: gcc.info, Node: Inline, Next: Extended Asm, Prev: Alignment, Up: C Extensions
16557 5.33 An Inline Function is As Fast As a Macro
16558 =============================================
16560 By declaring a function `inline', you can direct GCC to integrate that
16561 function's code into the code for its callers. This makes execution
16562 faster by eliminating the function-call overhead; in addition, if any
16563 of the actual argument values are constant, their known values may
16564 permit simplifications at compile time so that not all of the inline
16565 function's code needs to be included. The effect on code size is less
16566 predictable; object code may be larger or smaller with function
16567 inlining, depending on the particular case. Inlining of functions is an
16568 optimization and it really "works" only in optimizing compilation. If
16569 you don't use `-O', no function is really inline.
16571 Inline functions are included in the ISO C99 standard, but there are
16572 currently substantial differences between what GCC implements and what
16573 the ISO C99 standard requires.
16575 To declare a function inline, use the `inline' keyword in its
16576 declaration, like this:
16584 (If you are writing a header file to be included in ISO C programs,
16585 write `__inline__' instead of `inline'. *Note Alternate Keywords::.)
16586 You can also make all "simple enough" functions inline with the option
16587 `-finline-functions'.
16589 Note that certain usages in a function definition can make it
16590 unsuitable for inline substitution. Among these usages are: use of
16591 varargs, use of alloca, use of variable sized data types (*note
16592 Variable Length::), use of computed goto (*note Labels as Values::),
16593 use of nonlocal goto, and nested functions (*note Nested Functions::).
16594 Using `-Winline' will warn when a function marked `inline' could not be
16595 substituted, and will give the reason for the failure.
16597 Note that in C and Objective-C, unlike C++, the `inline' keyword does
16598 not affect the linkage of the function.
16600 GCC automatically inlines member functions defined within the class
16601 body of C++ programs even if they are not explicitly declared `inline'.
16602 (You can override this with `-fno-default-inline'; *note Options
16603 Controlling C++ Dialect: C++ Dialect Options.)
16605 When a function is both inline and `static', if all calls to the
16606 function are integrated into the caller, and the function's address is
16607 never used, then the function's own assembler code is never referenced.
16608 In this case, GCC does not actually output assembler code for the
16609 function, unless you specify the option `-fkeep-inline-functions'.
16610 Some calls cannot be integrated for various reasons (in particular,
16611 calls that precede the function's definition cannot be integrated, and
16612 neither can recursive calls within the definition). If there is a
16613 nonintegrated call, then the function is compiled to assembler code as
16614 usual. The function must also be compiled as usual if the program
16615 refers to its address, because that can't be inlined.
16617 When an inline function is not `static', then the compiler must assume
16618 that there may be calls from other source files; since a global symbol
16619 can be defined only once in any program, the function must not be
16620 defined in the other source files, so the calls therein cannot be
16621 integrated. Therefore, a non-`static' inline function is always
16622 compiled on its own in the usual fashion.
16624 If you specify both `inline' and `extern' in the function definition,
16625 then the definition is used only for inlining. In no case is the
16626 function compiled on its own, not even if you refer to its address
16627 explicitly. Such an address becomes an external reference, as if you
16628 had only declared the function, and had not defined it.
16630 This combination of `inline' and `extern' has almost the effect of a
16631 macro. The way to use it is to put a function definition in a header
16632 file with these keywords, and put another copy of the definition
16633 (lacking `inline' and `extern') in a library file. The definition in
16634 the header file will cause most calls to the function to be inlined.
16635 If any uses of the function remain, they will refer to the single copy
16638 Since GCC eventually will implement ISO C99 semantics for inline
16639 functions, it is best to use `static inline' only to guarantee
16640 compatibility. (The existing semantics will remain available when
16641 `-std=gnu89' is specified, but eventually the default will be
16642 `-std=gnu99' and that will implement the C99 semantics, though it does
16645 GCC does not inline any functions when not optimizing unless you
16646 specify the `always_inline' attribute for the function, like this:
16649 inline void foo (const char) __attribute__((always_inline));
16652 File: gcc.info, Node: Extended Asm, Next: Constraints, Prev: Inline, Up: C Extensions
16654 5.34 Assembler Instructions with C Expression Operands
16655 ======================================================
16657 In an assembler instruction using `asm', you can specify the operands
16658 of the instruction using C expressions. This means you need not guess
16659 which registers or memory locations will contain the data you want to
16662 You must specify an assembler instruction template much like what
16663 appears in a machine description, plus an operand constraint string for
16666 For example, here is how to use the 68881's `fsinx' instruction:
16668 asm ("fsinx %1,%0" : "=f" (result) : "f" (angle));
16670 Here `angle' is the C expression for the input operand while `result'
16671 is that of the output operand. Each has `"f"' as its operand
16672 constraint, saying that a floating point register is required. The `='
16673 in `=f' indicates that the operand is an output; all output operands'
16674 constraints must use `='. The constraints use the same language used
16675 in the machine description (*note Constraints::).
16677 Each operand is described by an operand-constraint string followed by
16678 the C expression in parentheses. A colon separates the assembler
16679 template from the first output operand and another separates the last
16680 output operand from the first input, if any. Commas separate the
16681 operands within each group. The total number of operands is currently
16682 limited to 30; this limitation may be lifted in some future version of
16685 If there are no output operands but there are input operands, you must
16686 place two consecutive colons surrounding the place where the output
16689 As of GCC version 3.1, it is also possible to specify input and output
16690 operands using symbolic names which can be referenced within the
16691 assembler code. These names are specified inside square brackets
16692 preceding the constraint string, and can be referenced inside the
16693 assembler code using `%[NAME]' instead of a percentage sign followed by
16694 the operand number. Using named operands the above example could look
16697 asm ("fsinx %[angle],%[output]"
16698 : [output] "=f" (result)
16699 : [angle] "f" (angle));
16701 Note that the symbolic operand names have no relation whatsoever to
16702 other C identifiers. You may use any name you like, even those of
16703 existing C symbols, but you must ensure that no two operands within the
16704 same assembler construct use the same symbolic name.
16706 Output operand expressions must be lvalues; the compiler can check
16707 this. The input operands need not be lvalues. The compiler cannot
16708 check whether the operands have data types that are reasonable for the
16709 instruction being executed. It does not parse the assembler instruction
16710 template and does not know what it means or even whether it is valid
16711 assembler input. The extended `asm' feature is most often used for
16712 machine instructions the compiler itself does not know exist. If the
16713 output expression cannot be directly addressed (for example, it is a
16714 bit-field), your constraint must allow a register. In that case, GCC
16715 will use the register as the output of the `asm', and then store that
16716 register into the output.
16718 The ordinary output operands must be write-only; GCC will assume that
16719 the values in these operands before the instruction are dead and need
16720 not be generated. Extended asm supports input-output or read-write
16721 operands. Use the constraint character `+' to indicate such an operand
16722 and list it with the output operands. You should only use read-write
16723 operands when the constraints for the operand (or the operand in which
16724 only some of the bits are to be changed) allow a register.
16726 You may, as an alternative, logically split its function into two
16727 separate operands, one input operand and one write-only output operand.
16728 The connection between them is expressed by constraints which say they
16729 need to be in the same location when the instruction executes. You can
16730 use the same C expression for both operands, or different expressions.
16731 For example, here we write the (fictitious) `combine' instruction with
16732 `bar' as its read-only source operand and `foo' as its read-write
16735 asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar));
16737 The constraint `"0"' for operand 1 says that it must occupy the same
16738 location as operand 0. A number in constraint is allowed only in an
16739 input operand and it must refer to an output operand.
16741 Only a number in the constraint can guarantee that one operand will be
16742 in the same place as another. The mere fact that `foo' is the value of
16743 both operands is not enough to guarantee that they will be in the same
16744 place in the generated assembler code. The following would not work
16747 asm ("combine %2,%0" : "=r" (foo) : "r" (foo), "g" (bar));
16749 Various optimizations or reloading could cause operands 0 and 1 to be
16750 in different registers; GCC knows no reason not to do so. For example,
16751 the compiler might find a copy of the value of `foo' in one register and
16752 use it for operand 1, but generate the output operand 0 in a different
16753 register (copying it afterward to `foo''s own address). Of course,
16754 since the register for operand 1 is not even mentioned in the assembler
16755 code, the result will not work, but GCC can't tell that.
16757 As of GCC version 3.1, one may write `[NAME]' instead of the operand
16758 number for a matching constraint. For example:
16760 asm ("cmoveq %1,%2,%[result]"
16761 : [result] "=r"(result)
16762 : "r" (test), "r"(new), "[result]"(old));
16764 Sometimes you need to make an `asm' operand be a specific register,
16765 but there's no matching constraint letter for that register _by
16766 itself_. To force the operand into that register, use a local variable
16767 for the operand and specify the register in the variable declaration.
16768 *Note Explicit Reg Vars::. Then for the `asm' operand, use any
16769 register constraint letter that matches the register:
16771 register int *p1 asm ("r0") = ...;
16772 register int *p2 asm ("r1") = ...;
16773 register int *result asm ("r0");
16774 asm ("sysint" : "=r" (result) : "0" (p1), "r" (p2));
16776 In the above example, beware that a register that is call-clobbered by
16777 the target ABI will be overwritten by any function call in the
16778 assignment, including library calls for arithmetic operators. Assuming
16779 it is a call-clobbered register, this may happen to `r0' above by the
16780 assignment to `p2'. If you have to use such a register, use temporary
16781 variables for expressions between the register assignment and use:
16784 register int *p1 asm ("r0") = ...;
16785 register int *p2 asm ("r1") = t1;
16786 register int *result asm ("r0");
16787 asm ("sysint" : "=r" (result) : "0" (p1), "r" (p2));
16789 Some instructions clobber specific hard registers. To describe this,
16790 write a third colon after the input operands, followed by the names of
16791 the clobbered hard registers (given as strings). Here is a realistic
16792 example for the VAX:
16794 asm volatile ("movc3 %0,%1,%2"
16796 : "g" (from), "g" (to), "g" (count)
16797 : "r0", "r1", "r2", "r3", "r4", "r5");
16799 You may not write a clobber description in a way that overlaps with an
16800 input or output operand. For example, you may not have an operand
16801 describing a register class with one member if you mention that register
16802 in the clobber list. Variables declared to live in specific registers
16803 (*note Explicit Reg Vars::), and used as asm input or output operands
16804 must have no part mentioned in the clobber description. There is no
16805 way for you to specify that an input operand is modified without also
16806 specifying it as an output operand. Note that if all the output
16807 operands you specify are for this purpose (and hence unused), you will
16808 then also need to specify `volatile' for the `asm' construct, as
16809 described below, to prevent GCC from deleting the `asm' statement as
16812 If you refer to a particular hardware register from the assembler code,
16813 you will probably have to list the register after the third colon to
16814 tell the compiler the register's value is modified. In some assemblers,
16815 the register names begin with `%'; to produce one `%' in the assembler
16816 code, you must write `%%' in the input.
16818 If your assembler instruction can alter the condition code register,
16819 add `cc' to the list of clobbered registers. GCC on some machines
16820 represents the condition codes as a specific hardware register; `cc'
16821 serves to name this register. On other machines, the condition code is
16822 handled differently, and specifying `cc' has no effect. But it is
16823 valid no matter what the machine.
16825 If your assembler instructions access memory in an unpredictable
16826 fashion, add `memory' to the list of clobbered registers. This will
16827 cause GCC to not keep memory values cached in registers across the
16828 assembler instruction and not optimize stores or loads to that memory.
16829 You will also want to add the `volatile' keyword if the memory affected
16830 is not listed in the inputs or outputs of the `asm', as the `memory'
16831 clobber does not count as a side-effect of the `asm'. If you know how
16832 large the accessed memory is, you can add it as input or output but if
16833 this is not known, you should add `memory'. As an example, if you
16834 access ten bytes of a string, you can use a memory input like:
16836 {"m"( ({ struct { char x[10]; } *p = (void *)ptr ; *p; }) )}.
16838 Note that in the following example the memory input is necessary,
16839 otherwise GCC might optimize the store to `x' away:
16845 asm ("magic stuff accessing an 'int' pointed to by '%1'"
16846 "=&d" (r) : "a" (y), "m" (*y));
16850 You can put multiple assembler instructions together in a single `asm'
16851 template, separated by the characters normally used in assembly code
16852 for the system. A combination that works in most places is a newline
16853 to break the line, plus a tab character to move to the instruction field
16854 (written as `\n\t'). Sometimes semicolons can be used, if the
16855 assembler allows semicolons as a line-breaking character. Note that
16856 some assembler dialects use semicolons to start a comment. The input
16857 operands are guaranteed not to use any of the clobbered registers, and
16858 neither will the output operands' addresses, so you can read and write
16859 the clobbered registers as many times as you like. Here is an example
16860 of multiple instructions in a template; it assumes the subroutine
16861 `_foo' accepts arguments in registers 9 and 10:
16863 asm ("movl %0,r9\n\tmovl %1,r10\n\tcall _foo"
16865 : "g" (from), "g" (to)
16868 Unless an output operand has the `&' constraint modifier, GCC may
16869 allocate it in the same register as an unrelated input operand, on the
16870 assumption the inputs are consumed before the outputs are produced.
16871 This assumption may be false if the assembler code actually consists of
16872 more than one instruction. In such a case, use `&' for each output
16873 operand that may not overlap an input. *Note Modifiers::.
16875 If you want to test the condition code produced by an assembler
16876 instruction, you must include a branch and a label in the `asm'
16877 construct, as follows:
16879 asm ("clr %0\n\tfrob %1\n\tbeq 0f\n\tmov #1,%0\n0:"
16883 This assumes your assembler supports local labels, as the GNU assembler
16884 and most Unix assemblers do.
16886 Speaking of labels, jumps from one `asm' to another are not supported.
16887 The compiler's optimizers do not know about these jumps, and therefore
16888 they cannot take account of them when deciding how to optimize.
16890 Usually the most convenient way to use these `asm' instructions is to
16891 encapsulate them in macros that look like functions. For example,
16894 ({ double __value, __arg = (x); \
16895 asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \
16898 Here the variable `__arg' is used to make sure that the instruction
16899 operates on a proper `double' value, and to accept only those arguments
16900 `x' which can convert automatically to a `double'.
16902 Another way to make sure the instruction operates on the correct data
16903 type is to use a cast in the `asm'. This is different from using a
16904 variable `__arg' in that it converts more different types. For
16905 example, if the desired type were `int', casting the argument to `int'
16906 would accept a pointer with no complaint, while assigning the argument
16907 to an `int' variable named `__arg' would warn about using a pointer
16908 unless the caller explicitly casts it.
16910 If an `asm' has output operands, GCC assumes for optimization purposes
16911 the instruction has no side effects except to change the output
16912 operands. This does not mean instructions with a side effect cannot be
16913 used, but you must be careful, because the compiler may eliminate them
16914 if the output operands aren't used, or move them out of loops, or
16915 replace two with one if they constitute a common subexpression. Also,
16916 if your instruction does have a side effect on a variable that otherwise
16917 appears not to change, the old value of the variable may be reused later
16918 if it happens to be found in a register.
16920 You can prevent an `asm' instruction from being deleted by writing the
16921 keyword `volatile' after the `asm'. For example:
16923 #define get_and_set_priority(new) \
16925 asm volatile ("get_and_set_priority %0, %1" \
16926 : "=g" (__old) : "g" (new)); \
16929 The `volatile' keyword indicates that the instruction has important
16930 side-effects. GCC will not delete a volatile `asm' if it is reachable.
16931 (The instruction can still be deleted if GCC can prove that
16932 control-flow will never reach the location of the instruction.) Note
16933 that even a volatile `asm' instruction can be moved relative to other
16934 code, including across jump instructions. For example, on many targets
16935 there is a system register which can be set to control the rounding
16936 mode of floating point operations. You might try setting it with a
16937 volatile `asm', like this PowerPC example:
16939 asm volatile("mtfsf 255,%0" : : "f" (fpenv));
16942 This will not work reliably, as the compiler may move the addition back
16943 before the volatile `asm'. To make it work you need to add an
16944 artificial dependency to the `asm' referencing a variable in the code
16945 you don't want moved, for example:
16947 asm volatile ("mtfsf 255,%1" : "=X"(sum): "f"(fpenv));
16950 Similarly, you can't expect a sequence of volatile `asm' instructions
16951 to remain perfectly consecutive. If you want consecutive output, use a
16952 single `asm'. Also, GCC will perform some optimizations across a
16953 volatile `asm' instruction; GCC does not "forget everything" when it
16954 encounters a volatile `asm' instruction the way some other compilers do.
16956 An `asm' instruction without any output operands will be treated
16957 identically to a volatile `asm' instruction.
16959 It is a natural idea to look for a way to give access to the condition
16960 code left by the assembler instruction. However, when we attempted to
16961 implement this, we found no way to make it work reliably. The problem
16962 is that output operands might need reloading, which would result in
16963 additional following "store" instructions. On most machines, these
16964 instructions would alter the condition code before there was time to
16965 test it. This problem doesn't arise for ordinary "test" and "compare"
16966 instructions because they don't have any output operands.
16968 For reasons similar to those described above, it is not possible to
16969 give an assembler instruction access to the condition code left by
16970 previous instructions.
16972 If you are writing a header file that should be includable in ISO C
16973 programs, write `__asm__' instead of `asm'. *Note Alternate Keywords::.
16975 5.34.1 Size of an `asm'
16976 -----------------------
16978 Some targets require that GCC track the size of each instruction used in
16979 order to generate correct code. Because the final length of an `asm'
16980 is only known by the assembler, GCC must make an estimate as to how big
16981 it will be. The estimate is formed by counting the number of
16982 statements in the pattern of the `asm' and multiplying that by the
16983 length of the longest instruction on that processor. Statements in the
16984 `asm' are identified by newline characters and whatever statement
16985 separator characters are supported by the assembler; on most processors
16986 this is the ``;'' character.
16988 Normally, GCC's estimate is perfectly adequate to ensure that correct
16989 code is generated, but it is possible to confuse the compiler if you use
16990 pseudo instructions or assembler macros that expand into multiple real
16991 instructions or if you use assembler directives that expand to more
16992 space in the object file than would be needed for a single instruction.
16993 If this happens then the assembler will produce a diagnostic saying that
16994 a label is unreachable.
16996 5.34.2 i386 floating point asm operands
16997 ---------------------------------------
16999 There are several rules on the usage of stack-like regs in asm_operands
17000 insns. These rules apply only to the operands that are stack-like regs:
17002 1. Given a set of input regs that die in an asm_operands, it is
17003 necessary to know which are implicitly popped by the asm, and
17004 which must be explicitly popped by gcc.
17006 An input reg that is implicitly popped by the asm must be
17007 explicitly clobbered, unless it is constrained to match an output
17010 2. For any input reg that is implicitly popped by an asm, it is
17011 necessary to know how to adjust the stack to compensate for the
17012 pop. If any non-popped input is closer to the top of the
17013 reg-stack than the implicitly popped reg, it would not be possible
17014 to know what the stack looked like--it's not clear how the rest of
17015 the stack "slides up".
17017 All implicitly popped input regs must be closer to the top of the
17018 reg-stack than any input that is not implicitly popped.
17020 It is possible that if an input dies in an insn, reload might use
17021 the input reg for an output reload. Consider this example:
17023 asm ("foo" : "=t" (a) : "f" (b));
17025 This asm says that input B is not popped by the asm, and that the
17026 asm pushes a result onto the reg-stack, i.e., the stack is one
17027 deeper after the asm than it was before. But, it is possible that
17028 reload will think that it can use the same reg for both the input
17029 and the output, if input B dies in this insn.
17031 If any input operand uses the `f' constraint, all output reg
17032 constraints must use the `&' earlyclobber.
17034 The asm above would be written as
17036 asm ("foo" : "=&t" (a) : "f" (b));
17038 3. Some operands need to be in particular places on the stack. All
17039 output operands fall in this category--there is no other way to
17040 know which regs the outputs appear in unless the user indicates
17041 this in the constraints.
17043 Output operands must specifically indicate which reg an output
17044 appears in after an asm. `=f' is not allowed: the operand
17045 constraints must select a class with a single reg.
17047 4. Output operands may not be "inserted" between existing stack regs.
17048 Since no 387 opcode uses a read/write operand, all output operands
17049 are dead before the asm_operands, and are pushed by the
17050 asm_operands. It makes no sense to push anywhere but the top of
17053 Output operands must start at the top of the reg-stack: output
17054 operands may not "skip" a reg.
17056 5. Some asm statements may need extra stack space for internal
17057 calculations. This can be guaranteed by clobbering stack registers
17058 unrelated to the inputs and outputs.
17061 Here are a couple of reasonable asms to want to write. This asm takes
17062 one input, which is internally popped, and produces two outputs.
17064 asm ("fsincos" : "=t" (cos), "=u" (sin) : "0" (inp));
17066 This asm takes two inputs, which are popped by the `fyl2xp1' opcode,
17067 and replaces them with one output. The user must code the `st(1)'
17068 clobber for reg-stack.c to know that `fyl2xp1' pops both inputs.
17070 asm ("fyl2xp1" : "=t" (result) : "0" (x), "u" (y) : "st(1)");
17073 File: gcc.info, Node: Constraints, Next: Asm Labels, Prev: Extended Asm, Up: C Extensions
17075 5.35 Constraints for `asm' Operands
17076 ===================================
17078 Here are specific details on what constraint letters you can use with
17079 `asm' operands. Constraints can say whether an operand may be in a
17080 register, and which kinds of register; whether the operand can be a
17081 memory reference, and which kinds of address; whether the operand may
17082 be an immediate constant, and which possible values it may have.
17083 Constraints can also require two operands to match.
17087 * Simple Constraints:: Basic use of constraints.
17088 * Multi-Alternative:: When an insn has two alternative constraint-patterns.
17089 * Modifiers:: More precise control over effects of constraints.
17090 * Machine Constraints:: Special constraints for some particular machines.
17093 File: gcc.info, Node: Simple Constraints, Next: Multi-Alternative, Up: Constraints
17095 5.35.1 Simple Constraints
17096 -------------------------
17098 The simplest kind of constraint is a string full of letters, each of
17099 which describes one kind of operand that is permitted. Here are the
17100 letters that are allowed:
17103 Whitespace characters are ignored and can be inserted at any
17104 position except the first. This enables each alternative for
17105 different operands to be visually aligned in the machine
17106 description even if they have different number of constraints and
17110 A memory operand is allowed, with any kind of address that the
17111 machine supports in general.
17114 A memory operand is allowed, but only if the address is
17115 "offsettable". This means that adding a small integer (actually,
17116 the width in bytes of the operand, as determined by its machine
17117 mode) may be added to the address and the result is also a valid
17120 For example, an address which is constant is offsettable; so is an
17121 address that is the sum of a register and a constant (as long as a
17122 slightly larger constant is also within the range of
17123 address-offsets supported by the machine); but an autoincrement or
17124 autodecrement address is not offsettable. More complicated
17125 indirect/indexed addresses may or may not be offsettable depending
17126 on the other addressing modes that the machine supports.
17128 Note that in an output operand which can be matched by another
17129 operand, the constraint letter `o' is valid only when accompanied
17130 by both `<' (if the target machine has predecrement addressing)
17131 and `>' (if the target machine has preincrement addressing).
17134 A memory operand that is not offsettable. In other words,
17135 anything that would fit the `m' constraint but not the `o'
17139 A memory operand with autodecrement addressing (either
17140 predecrement or postdecrement) is allowed.
17143 A memory operand with autoincrement addressing (either
17144 preincrement or postincrement) is allowed.
17147 A register operand is allowed provided that it is in a general
17151 An immediate integer operand (one with constant value) is allowed.
17152 This includes symbolic constants whose values will be known only at
17153 assembly time or later.
17156 An immediate integer operand with a known numeric value is allowed.
17157 Many systems cannot support assembly-time constants for operands
17158 less than a word wide. Constraints for these operands should use
17159 `n' rather than `i'.
17161 `I', `J', `K', ... `P'
17162 Other letters in the range `I' through `P' may be defined in a
17163 machine-dependent fashion to permit immediate integer operands with
17164 explicit integer values in specified ranges. For example, on the
17165 68000, `I' is defined to stand for the range of values 1 to 8.
17166 This is the range permitted as a shift count in the shift
17170 An immediate floating operand (expression code `const_double') is
17171 allowed, but only if the target floating point format is the same
17172 as that of the host machine (on which the compiler is running).
17175 An immediate floating operand (expression code `const_double' or
17176 `const_vector') is allowed.
17179 `G' and `H' may be defined in a machine-dependent fashion to
17180 permit immediate floating operands in particular ranges of values.
17183 An immediate integer operand whose value is not an explicit
17184 integer is allowed.
17186 This might appear strange; if an insn allows a constant operand
17187 with a value not known at compile time, it certainly must allow
17188 any known value. So why use `s' instead of `i'? Sometimes it
17189 allows better code to be generated.
17191 For example, on the 68000 in a fullword instruction it is possible
17192 to use an immediate operand; but if the immediate value is between
17193 -128 and 127, better code results from loading the value into a
17194 register and using the register. This is because the load into
17195 the register can be done with a `moveq' instruction. We arrange
17196 for this to happen by defining the letter `K' to mean "any integer
17197 outside the range -128 to 127", and then specifying `Ks' in the
17198 operand constraints.
17201 Any register, memory or immediate integer operand is allowed,
17202 except for registers that are not general registers.
17205 Any operand whatsoever is allowed.
17207 `0', `1', `2', ... `9'
17208 An operand that matches the specified operand number is allowed.
17209 If a digit is used together with letters within the same
17210 alternative, the digit should come last.
17212 This number is allowed to be more than a single digit. If multiple
17213 digits are encountered consecutively, they are interpreted as a
17214 single decimal integer. There is scant chance for ambiguity,
17215 since to-date it has never been desirable that `10' be interpreted
17216 as matching either operand 1 _or_ operand 0. Should this be
17217 desired, one can use multiple alternatives instead.
17219 This is called a "matching constraint" and what it really means is
17220 that the assembler has only a single operand that fills two roles
17221 which `asm' distinguishes. For example, an add instruction uses
17222 two input operands and an output operand, but on most CISC
17223 machines an add instruction really has only two operands, one of
17224 them an input-output operand:
17228 Matching constraints are used in these circumstances. More
17229 precisely, the two operands that match must include one input-only
17230 operand and one output-only operand. Moreover, the digit must be a
17231 smaller number than the number of the operand that uses it in the
17235 An operand that is a valid memory address is allowed. This is for
17236 "load address" and "push address" instructions.
17238 `p' in the constraint must be accompanied by `address_operand' as
17239 the predicate in the `match_operand'. This predicate interprets
17240 the mode specified in the `match_operand' as the mode of the memory
17241 reference for which the address would be valid.
17244 Other letters can be defined in machine-dependent fashion to stand
17245 for particular classes of registers or other arbitrary operand
17246 types. `d', `a' and `f' are defined on the 68000/68020 to stand
17247 for data, address and floating point registers.
17251 File: gcc.info, Node: Multi-Alternative, Next: Modifiers, Prev: Simple Constraints, Up: Constraints
17253 5.35.2 Multiple Alternative Constraints
17254 ---------------------------------------
17256 Sometimes a single instruction has multiple alternative sets of possible
17257 operands. For example, on the 68000, a logical-or instruction can
17258 combine register or an immediate value into memory, or it can combine
17259 any kind of operand into a register; but it cannot combine one memory
17260 location into another.
17262 These constraints are represented as multiple alternatives. An
17263 alternative can be described by a series of letters for each operand.
17264 The overall constraint for an operand is made from the letters for this
17265 operand from the first alternative, a comma, the letters for this
17266 operand from the second alternative, a comma, and so on until the last
17269 If all the operands fit any one alternative, the instruction is valid.
17270 Otherwise, for each alternative, the compiler counts how many
17271 instructions must be added to copy the operands so that that
17272 alternative applies. The alternative requiring the least copying is
17273 chosen. If two alternatives need the same amount of copying, the one
17274 that comes first is chosen. These choices can be altered with the `?'
17275 and `!' characters:
17278 Disparage slightly the alternative that the `?' appears in, as a
17279 choice when no alternative applies exactly. The compiler regards
17280 this alternative as one unit more costly for each `?' that appears
17284 Disparage severely the alternative that the `!' appears in. This
17285 alternative can still be used if it fits without reloading, but if
17286 reloading is needed, some other alternative will be used.
17289 File: gcc.info, Node: Modifiers, Next: Machine Constraints, Prev: Multi-Alternative, Up: Constraints
17291 5.35.3 Constraint Modifier Characters
17292 -------------------------------------
17294 Here are constraint modifier characters.
17297 Means that this operand is write-only for this instruction: the
17298 previous value is discarded and replaced by output data.
17301 Means that this operand is both read and written by the
17304 When the compiler fixes up the operands to satisfy the constraints,
17305 it needs to know which operands are inputs to the instruction and
17306 which are outputs from it. `=' identifies an output; `+'
17307 identifies an operand that is both input and output; all other
17308 operands are assumed to be input only.
17310 If you specify `=' or `+' in a constraint, you put it in the first
17311 character of the constraint string.
17314 Means (in a particular alternative) that this operand is an
17315 "earlyclobber" operand, which is modified before the instruction is
17316 finished using the input operands. Therefore, this operand may
17317 not lie in a register that is used as an input operand or as part
17318 of any memory address.
17320 `&' applies only to the alternative in which it is written. In
17321 constraints with multiple alternatives, sometimes one alternative
17322 requires `&' while others do not. See, for example, the `movdf'
17325 An input operand can be tied to an earlyclobber operand if its only
17326 use as an input occurs before the early result is written. Adding
17327 alternatives of this form often allows GCC to produce better code
17328 when only some of the inputs can be affected by the earlyclobber.
17329 See, for example, the `mulsi3' insn of the ARM.
17331 `&' does not obviate the need to write `='.
17334 Declares the instruction to be commutative for this operand and the
17335 following operand. This means that the compiler may interchange
17336 the two operands if that is the cheapest way to make all operands
17337 fit the constraints. GCC can only handle one commutative pair in
17338 an asm; if you use more, the compiler may fail. Note that you
17339 need not use the modifier if the two alternatives are strictly
17340 identical; this would only waste time in the reload pass. The
17341 modifier is not operational after register allocation, so the
17342 result of `define_peephole2' and `define_split's performed after
17343 reload cannot rely on `%' to make the intended insn match.
17346 Says that all following characters, up to the next comma, are to be
17347 ignored as a constraint. They are significant only for choosing
17348 register preferences.
17351 Says that the following character should be ignored when choosing
17352 register preferences. `*' has no effect on the meaning of the
17353 constraint as a constraint, and no effect on reloading.
17357 File: gcc.info, Node: Machine Constraints, Prev: Modifiers, Up: Constraints
17359 5.35.4 Constraints for Particular Machines
17360 ------------------------------------------
17362 Whenever possible, you should use the general-purpose constraint letters
17363 in `asm' arguments, since they will convey meaning more readily to
17364 people reading your code. Failing that, use the constraint letters
17365 that usually have very similar meanings across architectures. The most
17366 commonly used constraints are `m' and `r' (for memory and
17367 general-purpose registers respectively; *note Simple Constraints::), and
17368 `I', usually the letter indicating the most common immediate-constant
17371 For each machine architecture, the `config/MACHINE/MACHINE.h' file
17372 defines additional constraints. These constraints are used by the
17373 compiler itself for instruction generation, as well as for `asm'
17374 statements; therefore, some of the constraints are not particularly
17375 interesting for `asm'. The constraints are defined through these
17378 `REG_CLASS_FROM_LETTER'
17379 Register class constraints (usually lowercase).
17381 `CONST_OK_FOR_LETTER_P'
17382 Immediate constant constraints, for non-floating point constants of
17383 word size or smaller precision (usually uppercase).
17385 `CONST_DOUBLE_OK_FOR_LETTER_P'
17386 Immediate constant constraints, for all floating point constants
17387 and for constants of greater than word size precision (usually
17391 Special cases of registers or memory. This macro is not required,
17392 and is only defined for some machines.
17394 Inspecting these macro definitions in the compiler source for your
17395 machine is the best way to be certain you have the right constraints.
17396 However, here is a summary of the machine-dependent constraints
17397 available on some particular machines.
17399 _ARM family--`arm.h'_
17402 Floating-point register
17405 VFP floating-point register
17408 One of the floating-point constants 0.0, 0.5, 1.0, 2.0, 3.0,
17412 Floating-point constant that would satisfy the constraint `F'
17416 Integer that is valid as an immediate operand in a data
17417 processing instruction. That is, an integer in the range 0
17418 to 255 rotated by a multiple of 2
17421 Integer in the range -4095 to 4095
17424 Integer that satisfies constraint `I' when inverted (ones
17428 Integer that satisfies constraint `I' when negated (twos
17432 Integer in the range 0 to 32
17435 A memory reference where the exact address is in a single
17436 register (``m'' is preferable for `asm' statements)
17439 An item in the constant pool
17442 A symbol in the text segment of the current file
17445 A memory reference suitable for VFP load/store insns
17446 (reg+constant offset)
17449 A memory reference suitable for iWMMXt load/store
17453 A memory reference suitable for the ARMv4 ldrsb instruction.
17455 _AVR family--`avr.h'_
17458 Registers from r0 to r15
17461 Registers from r16 to r23
17464 Registers from r16 to r31
17467 Registers from r24 to r31. These registers can be used in
17471 Pointer register (r26-r31)
17474 Base pointer register (r28-r31)
17477 Stack pointer register (SPH:SPL)
17480 Temporary register r0
17483 Register pair X (r27:r26)
17486 Register pair Y (r29:r28)
17489 Register pair Z (r31:r30)
17492 Constant greater than -1, less than 64
17495 Constant greater than -64, less than 1
17504 Constant that fits in 8 bits
17507 Constant integer -1
17510 Constant integer 8, 16, or 24
17516 A floating point constant 0.0
17518 _CRX Architecture--`crx.h'_
17521 Registers from r0 to r14 (registers without stack pointer)
17524 Register r16 (64-bit accumulator lo register)
17527 Register r17 (64-bit accumulator hi register)
17530 Register pair r16-r17. (64-bit accumulator lo-hi pair)
17533 Constant that fits in 3 bits
17536 Constant that fits in 4 bits
17539 Constant that fits in 5 bits
17542 Constant that is one of -1, 4, -4, 7, 8, 12, 16, 20, 32, 48
17545 Floating point constant that is legal for store immediate
17547 _PowerPC and IBM RS6000--`rs6000.h'_
17550 Address base register
17553 Floating point register
17559 `MQ', `CTR', or `LINK' register
17571 `CR' register (condition register) number 0
17574 `CR' register (condition register)
17577 `FPMEM' stack memory for FPR-GPR transfers
17580 Signed 16-bit constant
17583 Unsigned 16-bit constant shifted left 16 bits (use `L'
17584 instead for `SImode' constants)
17587 Unsigned 16-bit constant
17590 Signed 16-bit constant shifted left 16 bits
17593 Constant larger than 31
17602 Constant whose negation is a signed 16-bit constant
17605 Floating point constant that can be loaded into a register
17606 with one instruction per word
17609 Memory operand that is an offset from a register (`m' is
17610 preferable for `asm' statements)
17616 Constant suitable as a 64-bit mask operand
17619 Constant suitable as a 32-bit mask operand
17622 System V Release 4 small data area reference
17624 _MorphoTech family--`mt.h'_
17627 Constant for an arithmetic insn (16-bit signed integer).
17633 Constant for a logical insn (16-bit zero-extended integer).
17636 A constant that can be loaded with `lui' (i.e. the bottom 16
17640 A constant that takes two words to load (i.e. not matched by
17644 Negative 16-bit constants other than -65536.
17647 A 15-bit signed integer constant.
17650 A positive 16-bit constant.
17652 _Intel 386--`i386.h'_
17655 `a', `b', `c', or `d' register for the i386. For x86-64 it
17656 is equivalent to `r' class (for 8-bit instructions that do
17657 not use upper halves).
17660 `a', `b', `c', or `d' register (for 8-bit instructions, that
17661 do use upper halves).
17664 Legacy register--equivalent to `r' class in i386 mode. (for
17665 non-8-bit registers used together with 8-bit upper halves in
17666 a single instruction)
17669 Specifies the `a' or `d' registers. This is primarily useful
17670 for 64-bit integer values (when in 32-bit mode) intended to
17671 be returned with the `d' register holding the most
17672 significant bits and the `a' register holding the least
17676 Floating point register
17679 First (top of stack) floating point register
17682 Second floating point register
17694 Specifies constant that can be easily constructed in SSE
17695 register without loading it from memory.
17713 Constant in range 0 to 31 (for 32-bit shifts)
17716 Constant in range 0 to 63 (for 64-bit shifts)
17725 0, 1, 2, or 3 (shifts for `lea' instruction)
17728 Constant in range 0 to 255 (for `out' instruction)
17731 Constant in range 0 to `0xffffffff' or symbolic reference
17732 known to fit specified range. (for using immediates in zero
17733 extending 32-bit to 64-bit x86-64 instructions)
17736 Constant in range -2147483648 to 2147483647 or symbolic
17737 reference known to fit specified range. (for using
17738 immediates in 64-bit x86-64 instructions)
17741 Standard 80387 floating point constant
17743 _Intel IA-64--`ia64.h'_
17746 General register `r0' to `r3' for `addl' instruction
17752 Predicate register (`c' as in "conditional")
17755 Application register residing in M-unit
17758 Application register residing in I-unit
17761 Floating-point register
17764 Memory operand. Remember that `m' allows postincrement and
17765 postdecrement which require printing with `%Pn' on IA-64.
17766 Use `S' to disallow postincrement and postdecrement.
17769 Floating-point constant 0.0 or 1.0
17772 14-bit signed integer constant
17775 22-bit signed integer constant
17778 8-bit signed integer constant for logical instructions
17781 8-bit adjusted signed integer constant for compare pseudo-ops
17784 6-bit unsigned integer constant for shift counts
17787 9-bit signed integer constant for load and store
17794 0 or -1 for `dep' instruction
17797 Non-volatile memory for floating-point loads and stores
17800 Integer constant in the range 1 to 4 for `shladd' instruction
17803 Memory operand except postincrement and postdecrement
17808 Register in the class `ACC_REGS' (`acc0' to `acc7').
17811 Register in the class `EVEN_ACC_REGS' (`acc0' to `acc7').
17814 Register in the class `CC_REGS' (`fcc0' to `fcc3' and `icc0'
17818 Register in the class `GPR_REGS' (`gr0' to `gr63').
17821 Register in the class `EVEN_REGS' (`gr0' to `gr63'). Odd
17822 registers are excluded not in the class but through the use
17823 of a machine mode larger than 4 bytes.
17826 Register in the class `FPR_REGS' (`fr0' to `fr63').
17829 Register in the class `FEVEN_REGS' (`fr0' to `fr63'). Odd
17830 registers are excluded not in the class but through the use
17831 of a machine mode larger than 4 bytes.
17834 Register in the class `LR_REG' (the `lr' register).
17837 Register in the class `QUAD_REGS' (`gr2' to `gr63').
17838 Register numbers not divisible by 4 are excluded not in the
17839 class but through the use of a machine mode larger than 8
17843 Register in the class `ICC_REGS' (`icc0' to `icc3').
17846 Register in the class `FCC_REGS' (`fcc0' to `fcc3').
17849 Register in the class `ICR_REGS' (`cc4' to `cc7').
17852 Register in the class `FCR_REGS' (`cc0' to `cc3').
17855 Register in the class `QUAD_FPR_REGS' (`fr0' to `fr63').
17856 Register numbers not divisible by 4 are excluded not in the
17857 class but through the use of a machine mode larger than 8
17861 Register in the class `SPR_REGS' (`lcr' and `lr').
17864 Register in the class `QUAD_ACC_REGS' (`acc0' to `acc7').
17867 Register in the class `ACCG_REGS' (`accg0' to `accg7').
17870 Register in the class `CR_REGS' (`cc0' to `cc7').
17873 Floating point constant zero
17876 6-bit signed integer constant
17879 10-bit signed integer constant
17882 16-bit signed integer constant
17885 16-bit unsigned integer constant
17888 12-bit signed integer constant that is negative--i.e. in the
17889 range of -2048 to -1
17895 12-bit signed integer constant that is greater than
17896 zero--i.e. in the range of 1 to 2047.
17899 _Blackfin family--`bfin.h'_
17908 A call clobbered P register.
17911 Even-numbered D register
17914 Odd-numbered D register
17917 Accumulator register.
17920 Even-numbered accumulator register.
17923 Odd-numbered accumulator register.
17935 Registers used for circular buffering, i.e. I, B, or L
17942 Any D, P, B, M, I or L register.
17945 Additional registers typically used only in prologues and
17946 epilogues: RETS, RETN, RETI, RETX, RETE, ASTAT, SEQSTAT and
17950 Any register except accumulators or CC.
17953 Signed 16 bit integer (in the range -32768 to 32767)
17956 Unsigned 16 bit integer (in the range 0 to 65535)
17959 Signed 7 bit integer (in the range -64 to 63)
17962 Unsigned 7 bit integer (in the range 0 to 127)
17965 Unsigned 5 bit integer (in the range 0 to 31)
17968 Signed 4 bit integer (in the range -8 to 7)
17971 Signed 3 bit integer (in the range -3 to 4)
17974 Unsigned 3 bit integer (in the range 0 to 7)
17977 Constant N, where N is a single-digit constant in the range 0
17987 An integer constant with exactly a single bit set.
17990 An integer constant with all bits set except exactly one.
18002 `$sp', `$fb', `$sb'.
18005 Any control register, when they're 16 bits wide (nothing if control
18006 registers are 24 bits wide)
18009 Any control register, when they're 24 bits wide.
18015 $r0, $r1, $r2, $r3.
18018 $r0 or $r2, or $r2r0 for 32 bit values.
18021 $r1 or $r3, or $r3r1 for 32 bit values.
18024 A register that can hold a 64 bit value.
18027 $r0 or $r1 (registers with addressable high/low bytes)
18036 Address registers when they're 16 bits wide.
18039 Address registers when they're 24 bits wide.
18042 Registers that can hold QI values.
18045 Registers that can be used with displacements ($a0, $a1, $sb).
18048 Registers that can hold 32 bit values.
18051 Registers that can hold 16 bit values.
18054 Registers chat can hold 16 bit values, including all control
18058 $r0 through R1, plus $a0 and $a1.
18061 The flags register.
18064 The memory-based pseudo-registers $mem0 through $mem15.
18067 Registers that can hold pointers (16 bit registers for r8c, m16c;
18068 24 bit registers for m32cm, m32c).
18071 Matches multiple registers in a PARALLEL to form a larger register.
18072 Used to match function return values.
18087 -8 ... -1 or 1 ... 8
18090 -16 ... -1 or 1 ... 16
18093 -8 ... -1 or 1 ... 8
18099 An 8 bit value with exactly one bit set.
18102 A 16 bit value with exactly one bit set.
18105 The common src/dest memory addressing modes.
18108 Memory addressed using $a0 or $a1.
18111 Memory addressed with immediate addresses.
18114 Memory addressed using the stack pointer ($sp).
18117 Memory addressed using the frame base register ($fb).
18120 Memory addressed using the small base register ($sb).
18128 General-purpose integer register
18131 Floating-point register (if available)
18140 `Hi' or `Lo' register
18143 General-purpose integer register
18146 Floating-point status register
18149 Signed 16-bit constant (for arithmetic instructions)
18155 Zero-extended 16-bit constant (for logic instructions)
18158 Constant with low 16 bits zero (can be loaded with `lui')
18161 32-bit constant which requires two instructions to load (a
18162 constant which is not `I', `K', or `L')
18165 Negative 16-bit constant
18171 Positive 16-bit constant
18174 Floating point zero
18177 Memory reference that can be loaded with more than one
18178 instruction (`m' is preferable for `asm' statements)
18181 Memory reference that can be loaded with one instruction (`m'
18182 is preferable for `asm' statements)
18185 Memory reference in external OSF/rose PIC format (`m' is
18186 preferable for `asm' statements)
18188 _Motorola 680x0--`m68k.h'_
18197 68881 floating-point register, if available
18200 Integer in the range 1 to 8
18203 16-bit signed number
18206 Signed number whose magnitude is greater than 0x80
18209 Integer in the range -8 to -1
18212 Signed number whose magnitude is greater than 0x100
18215 Floating point constant that is not a 68881 constant
18217 _Motorola 68HC11 & 68HC12 families--`m68hc11.h'_
18232 Temporary soft register _.tmp
18235 A soft register _.d1 to _.d31
18238 Stack pointer register
18247 Pseudo register `z' (replaced by `x' or `y' at the end)
18250 An address register: x, y or z
18253 An address register: x or y
18256 Register pair (x:d) to form a 32-bit value
18259 Constants in the range -65536 to 65535
18262 Constants whose 16-bit low part is zero
18265 Constant integer 1 or -1
18268 Constant integer 16
18271 Constants in the range -8 to 2
18277 Floating-point register on the SPARC-V8 architecture and
18278 lower floating-point register on the SPARC-V9 architecture.
18281 Floating-point register. It is equivalent to `f' on the
18282 SPARC-V8 architecture and contains both lower and upper
18283 floating-point registers on the SPARC-V9 architecture.
18286 Floating-point condition code register.
18289 Lower floating-point register. It is only valid on the
18290 SPARC-V9 architecture when the Visual Instruction Set is
18294 Floating-point register. It is only valid on the SPARC-V9
18295 architecture when the Visual Instruction Set is available.
18298 64-bit global or out register for the SPARC-V8+ architecture.
18301 Signed 13-bit constant
18307 32-bit constant with the low 12 bits clear (a constant that
18308 can be loaded with the `sethi' instruction)
18311 A constant in the range supported by `movcc' instructions
18314 A constant in the range supported by `movrcc' instructions
18317 Same as `K', except that it verifies that bits that are not
18318 in the lower 32-bit range are all zero. Must be used instead
18319 of `K' for modes wider than `SImode'
18325 Floating-point zero
18328 Signed 13-bit constant, sign-extended to 32 or 64 bits
18331 Floating-point constant whose integral representation can be
18332 moved into an integer register using a single sethi
18336 Floating-point constant whose integral representation can be
18337 moved into an integer register using a single mov instruction
18340 Floating-point constant whose integral representation can be
18341 moved into an integer register using a high/lo_sum
18342 instruction sequence
18345 Memory address aligned to an 8-byte boundary
18351 Memory address for `e' constraint registers
18357 _TMS320C3x/C4x--`c4x.h'_
18360 Auxiliary (address) register (ar0-ar7)
18363 Stack pointer register (sp)
18366 Standard (32-bit) precision integer register
18369 Extended (40-bit) precision register (r0-r11)
18372 Block count register (bk)
18375 Extended (40-bit) precision low register (r0-r7)
18378 Extended (40-bit) precision register (r0-r1)
18381 Extended (40-bit) precision register (r2-r3)
18384 Repeat count register (rc)
18387 Index register (ir0-ir1)
18390 Status (condition code) register (st)
18393 Data page register (dp)
18396 Floating-point zero
18399 Immediate 16-bit floating-point constant
18402 Signed 16-bit constant
18405 Signed 8-bit constant
18408 Signed 5-bit constant
18411 Unsigned 16-bit constant
18414 Unsigned 8-bit constant
18417 Ones complement of unsigned 16-bit constant
18420 High 16-bit constant (32-bit constant with 16 LSBs zero)
18423 Indirect memory reference with signed 8-bit or index register
18427 Indirect memory reference with unsigned 5-bit displacement
18430 Indirect memory reference with 1 bit or index register
18434 Direct memory reference
18440 _S/390 and zSeries--`s390.h'_
18443 Address register (general purpose register except r0)
18446 Condition code register
18449 Data register (arbitrary general purpose register)
18452 Floating-point register
18455 Unsigned 8-bit constant (0-255)
18458 Unsigned 12-bit constant (0-4095)
18461 Signed 16-bit constant (-32768-32767)
18464 Value appropriate as displacement.
18466 for short displacement
18468 `(-524288..524287)'
18469 for long displacement
18472 Constant integer with a value of 0x7fffffff.
18475 Multiple letter constraint followed by 4 parameter letters.
18477 number of the part counting from most to least
18484 mode of the containing operand
18487 value of the other parts (F--all bits set)
18488 The constraint matches if the specified part of a constant
18489 has a value different from it's other parts.
18492 Memory reference without index register and with short
18496 Memory reference with index register and short displacement.
18499 Memory reference without index register but with long
18503 Memory reference with index register and long displacement.
18506 Pointer with short displacement.
18509 Pointer with long displacement.
18512 Shift count operand.
18515 _Xstormy16--`stormy16.h'_
18530 Registers r0 through r7.
18533 Registers r0 and r1.
18536 The carry register.
18539 Registers r8 and r9.
18542 A constant between 0 and 3 inclusive.
18545 A constant that has exactly one bit set.
18548 A constant that has exactly one bit clear.
18551 A constant between 0 and 255 inclusive.
18554 A constant between -255 and 0 inclusive.
18557 A constant between -3 and 0 inclusive.
18560 A constant between 1 and 4 inclusive.
18563 A constant between -4 and -1 inclusive.
18566 A memory reference that is a stack push.
18569 A memory reference that is a stack pop.
18572 A memory reference that refers to a constant address of known
18576 The register indicated by Rx (not implemented yet).
18579 A constant that is not between 2 and 15 inclusive.
18585 _Xtensa--`xtensa.h'_
18588 General-purpose 32-bit register
18591 One-bit boolean register
18594 MAC16 40-bit accumulator register
18597 Signed 12-bit integer constant, for use in MOVI instructions
18600 Signed 8-bit integer constant, for use in ADDI instructions
18603 Integer constant valid for BccI instructions
18606 Unsigned constant valid for BccUI instructions
18611 File: gcc.info, Node: Asm Labels, Next: Explicit Reg Vars, Prev: Constraints, Up: C Extensions
18613 5.36 Controlling Names Used in Assembler Code
18614 =============================================
18616 You can specify the name to be used in the assembler code for a C
18617 function or variable by writing the `asm' (or `__asm__') keyword after
18618 the declarator as follows:
18620 int foo asm ("myfoo") = 2;
18622 This specifies that the name to be used for the variable `foo' in the
18623 assembler code should be `myfoo' rather than the usual `_foo'.
18625 On systems where an underscore is normally prepended to the name of a C
18626 function or variable, this feature allows you to define names for the
18627 linker that do not start with an underscore.
18629 It does not make sense to use this feature with a non-static local
18630 variable since such variables do not have assembler names. If you are
18631 trying to put the variable in a particular register, see *Note Explicit
18632 Reg Vars::. GCC presently accepts such code with a warning, but will
18633 probably be changed to issue an error, rather than a warning, in the
18636 You cannot use `asm' in this way in a function _definition_; but you
18637 can get the same effect by writing a declaration for the function
18638 before its definition and putting `asm' there, like this:
18640 extern func () asm ("FUNC");
18646 It is up to you to make sure that the assembler names you choose do not
18647 conflict with any other assembler symbols. Also, you must not use a
18648 register name; that would produce completely invalid assembler code.
18649 GCC does not as yet have the ability to store static variables in
18650 registers. Perhaps that will be added.
18653 File: gcc.info, Node: Explicit Reg Vars, Next: Alternate Keywords, Prev: Asm Labels, Up: C Extensions
18655 5.37 Variables in Specified Registers
18656 =====================================
18658 GNU C allows you to put a few global variables into specified hardware
18659 registers. You can also specify the register in which an ordinary
18660 register variable should be allocated.
18662 * Global register variables reserve registers throughout the program.
18663 This may be useful in programs such as programming language
18664 interpreters which have a couple of global variables that are
18665 accessed very often.
18667 * Local register variables in specific registers do not reserve the
18668 registers, except at the point where they are used as input or
18669 output operands in an `asm' statement and the `asm' statement
18670 itself is not deleted. The compiler's data flow analysis is
18671 capable of determining where the specified registers contain live
18672 values, and where they are available for other uses. Stores into
18673 local register variables may be deleted when they appear to be
18674 dead according to dataflow analysis. References to local register
18675 variables may be deleted or moved or simplified.
18677 These local variables are sometimes convenient for use with the
18678 extended `asm' feature (*note Extended Asm::), if you want to
18679 write one output of the assembler instruction directly into a
18680 particular register. (This will work provided the register you
18681 specify fits the constraints specified for that operand in the
18686 * Global Reg Vars::
18690 File: gcc.info, Node: Global Reg Vars, Next: Local Reg Vars, Up: Explicit Reg Vars
18692 5.37.1 Defining Global Register Variables
18693 -----------------------------------------
18695 You can define a global register variable in GNU C like this:
18697 register int *foo asm ("a5");
18699 Here `a5' is the name of the register which should be used. Choose a
18700 register which is normally saved and restored by function calls on your
18701 machine, so that library routines will not clobber it.
18703 Naturally the register name is cpu-dependent, so you would need to
18704 conditionalize your program according to cpu type. The register `a5'
18705 would be a good choice on a 68000 for a variable of pointer type. On
18706 machines with register windows, be sure to choose a "global" register
18707 that is not affected magically by the function call mechanism.
18709 In addition, operating systems on one type of cpu may differ in how
18710 they name the registers; then you would need additional conditionals.
18711 For example, some 68000 operating systems call this register `%a5'.
18713 Eventually there may be a way of asking the compiler to choose a
18714 register automatically, but first we need to figure out how it should
18715 choose and how to enable you to guide the choice. No solution is
18718 Defining a global register variable in a certain register reserves that
18719 register entirely for this use, at least within the current compilation.
18720 The register will not be allocated for any other purpose in the
18721 functions in the current compilation. The register will not be saved
18722 and restored by these functions. Stores into this register are never
18723 deleted even if they would appear to be dead, but references may be
18724 deleted or moved or simplified.
18726 It is not safe to access the global register variables from signal
18727 handlers, or from more than one thread of control, because the system
18728 library routines may temporarily use the register for other things
18729 (unless you recompile them specially for the task at hand).
18731 It is not safe for one function that uses a global register variable to
18732 call another such function `foo' by way of a third function `lose' that
18733 was compiled without knowledge of this variable (i.e. in a different
18734 source file in which the variable wasn't declared). This is because
18735 `lose' might save the register and put some other value there. For
18736 example, you can't expect a global register variable to be available in
18737 the comparison-function that you pass to `qsort', since `qsort' might
18738 have put something else in that register. (If you are prepared to
18739 recompile `qsort' with the same global register variable, you can solve
18742 If you want to recompile `qsort' or other source files which do not
18743 actually use your global register variable, so that they will not use
18744 that register for any other purpose, then it suffices to specify the
18745 compiler option `-ffixed-REG'. You need not actually add a global
18746 register declaration to their source code.
18748 A function which can alter the value of a global register variable
18749 cannot safely be called from a function compiled without this variable,
18750 because it could clobber the value the caller expects to find there on
18751 return. Therefore, the function which is the entry point into the part
18752 of the program that uses the global register variable must explicitly
18753 save and restore the value which belongs to its caller.
18755 On most machines, `longjmp' will restore to each global register
18756 variable the value it had at the time of the `setjmp'. On some
18757 machines, however, `longjmp' will not change the value of global
18758 register variables. To be portable, the function that called `setjmp'
18759 should make other arrangements to save the values of the global register
18760 variables, and to restore them in a `longjmp'. This way, the same
18761 thing will happen regardless of what `longjmp' does.
18763 All global register variable declarations must precede all function
18764 definitions. If such a declaration could appear after function
18765 definitions, the declaration would be too late to prevent the register
18766 from being used for other purposes in the preceding functions.
18768 Global register variables may not have initial values, because an
18769 executable file has no means to supply initial contents for a register.
18771 On the SPARC, there are reports that g3 ... g7 are suitable registers,
18772 but certain library functions, such as `getwd', as well as the
18773 subroutines for division and remainder, modify g3 and g4. g1 and g2
18774 are local temporaries.
18776 On the 68000, a2 ... a5 should be suitable, as should d2 ... d7. Of
18777 course, it will not do to use more than a few of those.
18780 File: gcc.info, Node: Local Reg Vars, Prev: Global Reg Vars, Up: Explicit Reg Vars
18782 5.37.2 Specifying Registers for Local Variables
18783 -----------------------------------------------
18785 You can define a local register variable with a specified register like
18788 register int *foo asm ("a5");
18790 Here `a5' is the name of the register which should be used. Note that
18791 this is the same syntax used for defining global register variables,
18792 but for a local variable it would appear within a function.
18794 Naturally the register name is cpu-dependent, but this is not a
18795 problem, since specific registers are most often useful with explicit
18796 assembler instructions (*note Extended Asm::). Both of these things
18797 generally require that you conditionalize your program according to cpu
18800 In addition, operating systems on one type of cpu may differ in how
18801 they name the registers; then you would need additional conditionals.
18802 For example, some 68000 operating systems call this register `%a5'.
18804 Defining such a register variable does not reserve the register; it
18805 remains available for other uses in places where flow control determines
18806 the variable's value is not live.
18808 This option does not guarantee that GCC will generate code that has
18809 this variable in the register you specify at all times. You may not
18810 code an explicit reference to this register in the _assembler
18811 instruction template_ part of an `asm' statement and assume it will
18812 always refer to this variable. However, using the variable as an `asm'
18813 _operand_ guarantees that the specified register is used for the
18816 Stores into local register variables may be deleted when they appear
18817 to be dead according to dataflow analysis. References to local
18818 register variables may be deleted or moved or simplified.
18820 As for global register variables, it's recommended that you choose a
18821 register which is normally saved and restored by function calls on your
18822 machine, so that library routines will not clobber it. A common
18823 pitfall is to initialize multiple call-clobbered registers with
18824 arbitrary expressions, where a function call or library call for an
18825 arithmetic operator will overwrite a register value from a previous
18826 assignment, for example `r0' below:
18827 register int *p1 asm ("r0") = ...;
18828 register int *p2 asm ("r1") = ...;
18829 In those cases, a solution is to use a temporary variable for each
18830 arbitrary expression. *Note Example of asm with clobbered asm reg::.
18833 File: gcc.info, Node: Alternate Keywords, Next: Incomplete Enums, Prev: Explicit Reg Vars, Up: C Extensions
18835 5.38 Alternate Keywords
18836 =======================
18838 `-ansi' and the various `-std' options disable certain keywords. This
18839 causes trouble when you want to use GNU C extensions, or a
18840 general-purpose header file that should be usable by all programs,
18841 including ISO C programs. The keywords `asm', `typeof' and `inline'
18842 are not available in programs compiled with `-ansi' or `-std' (although
18843 `inline' can be used in a program compiled with `-std=c99'). The ISO
18844 C99 keyword `restrict' is only available when `-std=gnu99' (which will
18845 eventually be the default) or `-std=c99' (or the equivalent
18846 `-std=iso9899:1999') is used.
18848 The way to solve these problems is to put `__' at the beginning and
18849 end of each problematical keyword. For example, use `__asm__' instead
18850 of `asm', and `__inline__' instead of `inline'.
18852 Other C compilers won't accept these alternative keywords; if you want
18853 to compile with another compiler, you can define the alternate keywords
18854 as macros to replace them with the customary keywords. It looks like
18858 #define __asm__ asm
18861 `-pedantic' and other options cause warnings for many GNU C extensions.
18862 You can prevent such warnings within one expression by writing
18863 `__extension__' before the expression. `__extension__' has no effect
18867 File: gcc.info, Node: Incomplete Enums, Next: Function Names, Prev: Alternate Keywords, Up: C Extensions
18869 5.39 Incomplete `enum' Types
18870 ============================
18872 You can define an `enum' tag without specifying its possible values.
18873 This results in an incomplete type, much like what you get if you write
18874 `struct foo' without describing the elements. A later declaration
18875 which does specify the possible values completes the type.
18877 You can't allocate variables or storage using the type while it is
18878 incomplete. However, you can work with pointers to that type.
18880 This extension may not be very useful, but it makes the handling of
18881 `enum' more consistent with the way `struct' and `union' are handled.
18883 This extension is not supported by GNU C++.
18886 File: gcc.info, Node: Function Names, Next: Return Address, Prev: Incomplete Enums, Up: C Extensions
18888 5.40 Function Names as Strings
18889 ==============================
18891 GCC provides three magic variables which hold the name of the current
18892 function, as a string. The first of these is `__func__', which is part
18893 of the C99 standard:
18895 The identifier `__func__' is implicitly declared by the translator
18896 as if, immediately following the opening brace of each function
18897 definition, the declaration
18898 static const char __func__[] = "function-name";
18900 appeared, where function-name is the name of the lexically-enclosing
18901 function. This name is the unadorned name of the function.
18903 `__FUNCTION__' is another name for `__func__'. Older versions of GCC
18904 recognize only this name. However, it is not standardized. For
18905 maximum portability, we recommend you use `__func__', but provide a
18906 fallback definition with the preprocessor:
18908 #if __STDC_VERSION__ < 199901L
18910 # define __func__ __FUNCTION__
18912 # define __func__ "<unknown>"
18916 In C, `__PRETTY_FUNCTION__' is yet another name for `__func__'.
18917 However, in C++, `__PRETTY_FUNCTION__' contains the type signature of
18918 the function as well as its bare name. For example, this program:
18921 extern int printf (char *, ...);
18928 printf ("__FUNCTION__ = %s\n", __FUNCTION__);
18929 printf ("__PRETTY_FUNCTION__ = %s\n", __PRETTY_FUNCTION__);
18944 __PRETTY_FUNCTION__ = void a::sub(int)
18946 These identifiers are not preprocessor macros. In GCC 3.3 and
18947 earlier, in C only, `__FUNCTION__' and `__PRETTY_FUNCTION__' were
18948 treated as string literals; they could be used to initialize `char'
18949 arrays, and they could be concatenated with other string literals. GCC
18950 3.4 and later treat them as variables, like `__func__'. In C++,
18951 `__FUNCTION__' and `__PRETTY_FUNCTION__' have always been variables.
18954 File: gcc.info, Node: Return Address, Next: Vector Extensions, Prev: Function Names, Up: C Extensions
18956 5.41 Getting the Return or Frame Address of a Function
18957 ======================================================
18959 These functions may be used to get information about the callers of a
18962 -- Built-in Function: void * __builtin_return_address (unsigned int
18964 This function returns the return address of the current function,
18965 or of one of its callers. The LEVEL argument is number of frames
18966 to scan up the call stack. A value of `0' yields the return
18967 address of the current function, a value of `1' yields the return
18968 address of the caller of the current function, and so forth. When
18969 inlining the expected behavior is that the function will return
18970 the address of the function that will be returned to. To work
18971 around this behavior use the `noinline' function attribute.
18973 The LEVEL argument must be a constant integer.
18975 On some machines it may be impossible to determine the return
18976 address of any function other than the current one; in such cases,
18977 or when the top of the stack has been reached, this function will
18978 return `0' or a random value. In addition,
18979 `__builtin_frame_address' may be used to determine if the top of
18980 the stack has been reached.
18982 This function should only be used with a nonzero argument for
18983 debugging purposes.
18985 -- Built-in Function: void * __builtin_frame_address (unsigned int
18987 This function is similar to `__builtin_return_address', but it
18988 returns the address of the function frame rather than the return
18989 address of the function. Calling `__builtin_frame_address' with a
18990 value of `0' yields the frame address of the current function, a
18991 value of `1' yields the frame address of the caller of the current
18992 function, and so forth.
18994 The frame is the area on the stack which holds local variables and
18995 saved registers. The frame address is normally the address of the
18996 first word pushed on to the stack by the function. However, the
18997 exact definition depends upon the processor and the calling
18998 convention. If the processor has a dedicated frame pointer
18999 register, and the function has a frame, then
19000 `__builtin_frame_address' will return the value of the frame
19003 On some machines it may be impossible to determine the frame
19004 address of any function other than the current one; in such cases,
19005 or when the top of the stack has been reached, this function will
19006 return `0' if the first frame pointer is properly initialized by
19009 This function should only be used with a nonzero argument for
19010 debugging purposes.
19013 File: gcc.info, Node: Vector Extensions, Next: Offsetof, Prev: Return Address, Up: C Extensions
19015 5.42 Using vector instructions through built-in functions
19016 =========================================================
19018 On some targets, the instruction set contains SIMD vector instructions
19019 that operate on multiple values contained in one large register at the
19020 same time. For example, on the i386 the MMX, 3Dnow! and SSE extensions
19021 can be used this way.
19023 The first step in using these extensions is to provide the necessary
19024 data types. This should be done using an appropriate `typedef':
19026 typedef int v4si __attribute__ ((vector_size (16)));
19028 The `int' type specifies the base type, while the attribute specifies
19029 the vector size for the variable, measured in bytes. For example, the
19030 declaration above causes the compiler to set the mode for the `v4si'
19031 type to be 16 bytes wide and divided into `int' sized units. For a
19032 32-bit `int' this means a vector of 4 units of 4 bytes, and the
19033 corresponding mode of `foo' will be V4SI.
19035 The `vector_size' attribute is only applicable to integral and float
19036 scalars, although arrays, pointers, and function return values are
19037 allowed in conjunction with this construct.
19039 All the basic integer types can be used as base types, both as signed
19040 and as unsigned: `char', `short', `int', `long', `long long'. In
19041 addition, `float' and `double' can be used to build floating-point
19044 Specifying a combination that is not valid for the current architecture
19045 will cause GCC to synthesize the instructions using a narrower mode.
19046 For example, if you specify a variable of type `V4SI' and your
19047 architecture does not allow for this specific SIMD type, GCC will
19048 produce code that uses 4 `SIs'.
19050 The types defined in this manner can be used with a subset of normal C
19051 operations. Currently, GCC will allow using the following operators on
19052 these types: `+, -, *, /, unary minus, ^, |, &, ~'.
19054 The operations behave like C++ `valarrays'. Addition is defined as
19055 the addition of the corresponding elements of the operands. For
19056 example, in the code below, each of the 4 elements in A will be added
19057 to the corresponding 4 elements in B and the resulting vector will be
19060 typedef int v4si __attribute__ ((vector_size (16)));
19066 Subtraction, multiplication, division, and the logical operations
19067 operate in a similar manner. Likewise, the result of using the unary
19068 minus or complement operators on a vector type is a vector whose
19069 elements are the negative or complemented values of the corresponding
19070 elements in the operand.
19072 You can declare variables and use them in function calls and returns,
19073 as well as in assignments and some casts. You can specify a vector
19074 type as a return type for a function. Vector types can also be used as
19075 function arguments. It is possible to cast from one vector type to
19076 another, provided they are of the same size (in fact, you can also cast
19077 vectors to and from other datatypes of the same size).
19079 You cannot operate between vectors of different lengths or different
19080 signedness without a cast.
19082 A port that supports hardware vector operations, usually provides a set
19083 of built-in functions that can be used to operate on vectors. For
19084 example, a function to add two vectors and multiply the result by a
19085 third could look like this:
19087 v4si f (v4si a, v4si b, v4si c)
19089 v4si tmp = __builtin_addv4si (a, b);
19090 return __builtin_mulv4si (tmp, c);
19094 File: gcc.info, Node: Offsetof, Next: Atomic Builtins, Prev: Vector Extensions, Up: C Extensions
19099 GCC implements for both C and C++ a syntactic extension to implement
19100 the `offsetof' macro.
19103 "__builtin_offsetof" "(" `typename' "," offsetof_member_designator ")"
19105 offsetof_member_designator:
19107 | offsetof_member_designator "." `identifier'
19108 | offsetof_member_designator "[" `expr' "]"
19110 This extension is sufficient such that
19112 #define offsetof(TYPE, MEMBER) __builtin_offsetof (TYPE, MEMBER)
19114 is a suitable definition of the `offsetof' macro. In C++, TYPE may be
19115 dependent. In either case, MEMBER may consist of a single identifier,
19116 or a sequence of member accesses and array references.
19119 File: gcc.info, Node: Atomic Builtins, Next: Object Size Checking, Prev: Offsetof, Up: C Extensions
19121 5.44 Built-in functions for atomic memory access
19122 ================================================
19124 The following builtins are intended to be compatible with those
19125 described in the `Intel Itanium Processor-specific Application Binary
19126 Interface', section 7.4. As such, they depart from the normal GCC
19127 practice of using the "__builtin_" prefix, and further that they are
19128 overloaded such that they work on multiple types.
19130 The definition given in the Intel documentation allows only for the
19131 use of the types `int', `long', `long long' as well as their unsigned
19132 counterparts. GCC will allow any integral scalar or pointer type that
19133 is 1, 2, 4 or 8 bytes in length.
19135 Not all operations are supported by all target processors. If a
19136 particular operation cannot be implemented on the target processor, a
19137 warning will be generated and a call an external function will be
19138 generated. The external function will carry the same name as the
19139 builtin, with an additional suffix `_N' where N is the size of the data
19142 In most cases, these builtins are considered a "full barrier". That
19143 is, no memory operand will be moved across the operation, either
19144 forward or backward. Further, instructions will be issued as necessary
19145 to prevent the processor from speculating loads across the operation
19146 and from queuing stores after the operation.
19148 All of the routines are are described in the Intel documentation to
19149 take "an optional list of variables protected by the memory barrier".
19150 It's not clear what is meant by that; it could mean that _only_ the
19151 following variables are protected, or it could mean that these variables
19152 should in addition be protected. At present GCC ignores this list and
19153 protects all variables which are globally accessible. If in the future
19154 we make some use of this list, an empty list will continue to mean all
19155 globally accessible variables.
19157 `TYPE __sync_fetch_and_add (TYPE *ptr, TYPE value, ...)'
19158 `TYPE __sync_fetch_and_sub (TYPE *ptr, TYPE value, ...)'
19159 `TYPE __sync_fetch_and_or (TYPE *ptr, TYPE value, ...)'
19160 `TYPE __sync_fetch_and_and (TYPE *ptr, TYPE value, ...)'
19161 `TYPE __sync_fetch_and_xor (TYPE *ptr, TYPE value, ...)'
19162 `TYPE __sync_fetch_and_nand (TYPE *ptr, TYPE value, ...)'
19163 These builtins perform the operation suggested by the name, and
19164 returns the value that had previously been in memory. That is,
19166 { tmp = *ptr; *ptr OP= value; return tmp; }
19167 { tmp = *ptr; *ptr = ~tmp & value; return tmp; } // nand
19169 `TYPE __sync_add_and_fetch (TYPE *ptr, TYPE value, ...)'
19170 `TYPE __sync_sub_and_fetch (TYPE *ptr, TYPE value, ...)'
19171 `TYPE __sync_or_and_fetch (TYPE *ptr, TYPE value, ...)'
19172 `TYPE __sync_and_and_fetch (TYPE *ptr, TYPE value, ...)'
19173 `TYPE __sync_xor_and_fetch (TYPE *ptr, TYPE value, ...)'
19174 `TYPE __sync_nand_and_fetch (TYPE *ptr, TYPE value, ...)'
19175 These builtins perform the operation suggested by the name, and
19176 return the new value. That is,
19178 { *ptr OP= value; return *ptr; }
19179 { *ptr = ~*ptr & value; return *ptr; } // nand
19181 `bool __sync_bool_compare_and_swap (TYPE *ptr, TYPE oldval TYPE newval, ...)'
19182 `TYPE __sync_val_compare_and_swap (TYPE *ptr, TYPE oldval TYPE newval, ...)'
19183 These builtins perform an atomic compare and swap. That is, if
19184 the current value of `*PTR' is OLDVAL, then write NEWVAL into
19187 The "bool" version returns true if the comparison is successful and
19188 NEWVAL was written. The "val" version returns the contents of
19189 `*PTR' before the operation.
19191 `__sync_synchronize (...)'
19192 This builtin issues a full memory barrier.
19194 `TYPE __sync_lock_test_and_set (TYPE *ptr, TYPE value, ...)'
19195 This builtin, as described by Intel, is not a traditional
19196 test-and-set operation, but rather an atomic exchange operation.
19197 It writes VALUE into `*PTR', and returns the previous contents of
19200 Many targets have only minimal support for such locks, and do not
19201 support a full exchange operation. In this case, a target may
19202 support reduced functionality here by which the _only_ valid value
19203 to store is the immediate constant 1. The exact value actually
19204 stored in `*PTR' is implementation defined.
19206 This builtin is not a full barrier, but rather an "acquire
19207 barrier". This means that references after the builtin cannot
19208 move to (or be speculated to) before the builtin, but previous
19209 memory stores may not be globally visible yet, and previous memory
19210 loads may not yet be satisfied.
19212 `void __sync_lock_release (TYPE *ptr, ...)'
19213 This builtin releases the lock acquired by
19214 `__sync_lock_test_and_set'. Normally this means writing the
19215 constant 0 to `*PTR'.
19217 This builtin is not a full barrier, but rather a "release barrier".
19218 This means that all previous memory stores are globally visible,
19219 and all previous memory loads have been satisfied, but following
19220 memory reads are not prevented from being speculated to before the
19224 File: gcc.info, Node: Object Size Checking, Next: Other Builtins, Prev: Atomic Builtins, Up: C Extensions
19226 5.45 Object Size Checking Builtins
19227 ==================================
19229 GCC implements a limited buffer overflow protection mechanism that can
19230 prevent some buffer overflow attacks.
19232 -- Built-in Function: size_t __builtin_object_size (void * PTR, int
19234 is a built-in construct that returns a constant number of bytes
19235 from PTR to the end of the object PTR pointer points to (if known
19236 at compile time). `__builtin_object_size' never evaluates its
19237 arguments for side-effects. If there are any side-effects in
19238 them, it returns `(size_t) -1' for TYPE 0 or 1 and `(size_t) 0'
19239 for TYPE 2 or 3. If there are multiple objects PTR can point to
19240 and all of them are known at compile time, the returned number is
19241 the maximum of remaining byte counts in those objects if TYPE & 2
19242 is 0 and minimum if nonzero. If it is not possible to determine
19243 which objects PTR points to at compile time,
19244 `__builtin_object_size' should return `(size_t) -1' for TYPE 0 or
19245 1 and `(size_t) 0' for TYPE 2 or 3.
19247 TYPE is an integer constant from 0 to 3. If the least significant
19248 bit is clear, objects are whole variables, if it is set, a closest
19249 surrounding subobject is considered the object a pointer points to.
19250 The second bit determines if maximum or minimum of remaining bytes
19253 struct V { char buf1[10]; int b; char buf2[10]; } var;
19254 char *p = &var.buf1[1], *q = &var.b;
19256 /* Here the object p points to is var. */
19257 assert (__builtin_object_size (p, 0) == sizeof (var) - 1);
19258 /* The subobject p points to is var.buf1. */
19259 assert (__builtin_object_size (p, 1) == sizeof (var.buf1) - 1);
19260 /* The object q points to is var. */
19261 assert (__builtin_object_size (q, 0)
19262 == (char *) (&var + 1) - (char *) &var.b);
19263 /* The subobject q points to is var.b. */
19264 assert (__builtin_object_size (q, 1) == sizeof (var.b));
19266 There are built-in functions added for many common string operation
19267 functions, e.g. for `memcpy' `__builtin___memcpy_chk' built-in is
19268 provided. This built-in has an additional last argument, which is the
19269 number of bytes remaining in object the DEST argument points to or
19270 `(size_t) -1' if the size is not known.
19272 The built-in functions are optimized into the normal string functions
19273 like `memcpy' if the last argument is `(size_t) -1' or if it is known
19274 at compile time that the destination object will not be overflown. If
19275 the compiler can determine at compile time the object will be always
19276 overflown, it issues a warning.
19278 The intended use can be e.g.
19281 #define bos0(dest) __builtin_object_size (dest, 0)
19282 #define memcpy(dest, src, n) \
19283 __builtin___memcpy_chk (dest, src, n, bos0 (dest))
19287 /* It is unknown what object p points to, so this is optimized
19288 into plain memcpy - no checking is possible. */
19289 memcpy (p, "abcde", n);
19290 /* Destination is known and length too. It is known at compile
19291 time there will be no overflow. */
19292 memcpy (&buf[5], "abcde", 5);
19293 /* Destination is known, but the length is not known at compile time.
19294 This will result in __memcpy_chk call that can check for overflow
19296 memcpy (&buf[5], "abcde", n);
19297 /* Destination is known and it is known at compile time there will
19298 be overflow. There will be a warning and __memcpy_chk call that
19299 will abort the program at runtime. */
19300 memcpy (&buf[6], "abcde", 5);
19302 Such built-in functions are provided for `memcpy', `mempcpy',
19303 `memmove', `memset', `strcpy', `stpcpy', `strncpy', `strcat' and
19306 There are also checking built-in functions for formatted output
19308 int __builtin___sprintf_chk (char *s, int flag, size_t os, const char *fmt, ...);
19309 int __builtin___snprintf_chk (char *s, size_t maxlen, int flag, size_t os,
19310 const char *fmt, ...);
19311 int __builtin___vsprintf_chk (char *s, int flag, size_t os, const char *fmt,
19313 int __builtin___vsnprintf_chk (char *s, size_t maxlen, int flag, size_t os,
19314 const char *fmt, va_list ap);
19316 The added FLAG argument is passed unchanged to `__sprintf_chk' etc.
19317 functions and can contain implementation specific flags on what
19318 additional security measures the checking function might take, such as
19319 handling `%n' differently.
19321 The OS argument is the object size S points to, like in the other
19322 built-in functions. There is a small difference in the behavior
19323 though, if OS is `(size_t) -1', the built-in functions are optimized
19324 into the non-checking functions only if FLAG is 0, otherwise the
19325 checking function is called with OS argument set to `(size_t) -1'.
19327 In addition to this, there are checking built-in functions
19328 `__builtin___printf_chk', `__builtin___vprintf_chk',
19329 `__builtin___fprintf_chk' and `__builtin___vfprintf_chk'. These have
19330 just one additional argument, FLAG, right before format string FMT. If
19331 the compiler is able to optimize them to `fputc' etc. functions, it
19332 will, otherwise the checking function should be called and the FLAG
19333 argument passed to it.
19336 File: gcc.info, Node: Other Builtins, Next: Target Builtins, Prev: Object Size Checking, Up: C Extensions
19338 5.46 Other built-in functions provided by GCC
19339 =============================================
19341 GCC provides a large number of built-in functions other than the ones
19342 mentioned above. Some of these are for internal use in the processing
19343 of exceptions or variable-length argument lists and will not be
19344 documented here because they may change from time to time; we do not
19345 recommend general use of these functions.
19347 The remaining functions are provided for optimization purposes.
19349 GCC includes built-in versions of many of the functions in the standard
19350 C library. The versions prefixed with `__builtin_' will always be
19351 treated as having the same meaning as the C library function even if you
19352 specify the `-fno-builtin' option. (*note C Dialect Options::) Many of
19353 these functions are only optimized in certain cases; if they are not
19354 optimized in a particular case, a call to the library function will be
19357 Outside strict ISO C mode (`-ansi', `-std=c89' or `-std=c99'), the
19358 functions `_exit', `alloca', `bcmp', `bzero', `dcgettext', `dgettext',
19359 `dremf', `dreml', `drem', `exp10f', `exp10l', `exp10', `ffsll', `ffsl',
19360 `ffs', `fprintf_unlocked', `fputs_unlocked', `gammaf', `gammal',
19361 `gamma', `gettext', `index', `isascii', `j0f', `j0l', `j0', `j1f',
19362 `j1l', `j1', `jnf', `jnl', `jn', `mempcpy', `pow10f', `pow10l', `pow10',
19363 `printf_unlocked', `rindex', `scalbf', `scalbl', `scalb', `signbit',
19364 `signbitf', `signbitl', `significandf', `significandl', `significand',
19365 `sincosf', `sincosl', `sincos', `stpcpy', `stpncpy', `strcasecmp',
19366 `strdup', `strfmon', `strncasecmp', `strndup', `toascii', `y0f', `y0l',
19367 `y0', `y1f', `y1l', `y1', `ynf', `ynl' and `yn' may be handled as
19368 built-in functions. All these functions have corresponding versions
19369 prefixed with `__builtin_', which may be used even in strict C89 mode.
19371 The ISO C99 functions `_Exit', `acoshf', `acoshl', `acosh', `asinhf',
19372 `asinhl', `asinh', `atanhf', `atanhl', `atanh', `cabsf', `cabsl',
19373 `cabs', `cacosf', `cacoshf', `cacoshl', `cacosh', `cacosl', `cacos',
19374 `cargf', `cargl', `carg', `casinf', `casinhf', `casinhl', `casinh',
19375 `casinl', `casin', `catanf', `catanhf', `catanhl', `catanh', `catanl',
19376 `catan', `cbrtf', `cbrtl', `cbrt', `ccosf', `ccoshf', `ccoshl',
19377 `ccosh', `ccosl', `ccos', `cexpf', `cexpl', `cexp', `cimagf', `cimagl',
19378 `cimag', `clogf', `clogl', `clog', `conjf', `conjl', `conj',
19379 `copysignf', `copysignl', `copysign', `cpowf', `cpowl', `cpow',
19380 `cprojf', `cprojl', `cproj', `crealf', `creall', `creal', `csinf',
19381 `csinhf', `csinhl', `csinh', `csinl', `csin', `csqrtf', `csqrtl',
19382 `csqrt', `ctanf', `ctanhf', `ctanhl', `ctanh', `ctanl', `ctan',
19383 `erfcf', `erfcl', `erfc', `erff', `erfl', `erf', `exp2f', `exp2l',
19384 `exp2', `expm1f', `expm1l', `expm1', `fdimf', `fdiml', `fdim', `fmaf',
19385 `fmal', `fmaxf', `fmaxl', `fmax', `fma', `fminf', `fminl', `fmin',
19386 `hypotf', `hypotl', `hypot', `ilogbf', `ilogbl', `ilogb', `imaxabs',
19387 `isblank', `iswblank', `lgammaf', `lgammal', `lgamma', `llabs',
19388 `llrintf', `llrintl', `llrint', `llroundf', `llroundl', `llround',
19389 `log1pf', `log1pl', `log1p', `log2f', `log2l', `log2', `logbf',
19390 `logbl', `logb', `lrintf', `lrintl', `lrint', `lroundf', `lroundl',
19391 `lround', `nearbyintf', `nearbyintl', `nearbyint', `nextafterf',
19392 `nextafterl', `nextafter', `nexttowardf', `nexttowardl', `nexttoward',
19393 `remainderf', `remainderl', `remainder', `remquof', `remquol',
19394 `remquo', `rintf', `rintl', `rint', `roundf', `roundl', `round',
19395 `scalblnf', `scalblnl', `scalbln', `scalbnf', `scalbnl', `scalbn',
19396 `snprintf', `tgammaf', `tgammal', `tgamma', `truncf', `truncl', `trunc',
19397 `vfscanf', `vscanf', `vsnprintf' and `vsscanf' are handled as built-in
19398 functions except in strict ISO C90 mode (`-ansi' or `-std=c89').
19400 There are also built-in versions of the ISO C99 functions `acosf',
19401 `acosl', `asinf', `asinl', `atan2f', `atan2l', `atanf', `atanl',
19402 `ceilf', `ceill', `cosf', `coshf', `coshl', `cosl', `expf', `expl',
19403 `fabsf', `fabsl', `floorf', `floorl', `fmodf', `fmodl', `frexpf',
19404 `frexpl', `ldexpf', `ldexpl', `log10f', `log10l', `logf', `logl',
19405 `modfl', `modf', `powf', `powl', `sinf', `sinhf', `sinhl', `sinl',
19406 `sqrtf', `sqrtl', `tanf', `tanhf', `tanhl' and `tanl' that are
19407 recognized in any mode since ISO C90 reserves these names for the
19408 purpose to which ISO C99 puts them. All these functions have
19409 corresponding versions prefixed with `__builtin_'.
19411 The ISO C94 functions `iswalnum', `iswalpha', `iswcntrl', `iswdigit',
19412 `iswgraph', `iswlower', `iswprint', `iswpunct', `iswspace', `iswupper',
19413 `iswxdigit', `towlower' and `towupper' are handled as built-in functions
19414 except in strict ISO C90 mode (`-ansi' or `-std=c89').
19416 The ISO C90 functions `abort', `abs', `acos', `asin', `atan2', `atan',
19417 `calloc', `ceil', `cosh', `cos', `exit', `exp', `fabs', `floor', `fmod',
19418 `fprintf', `fputs', `frexp', `fscanf', `isalnum', `isalpha', `iscntrl',
19419 `isdigit', `isgraph', `islower', `isprint', `ispunct', `isspace',
19420 `isupper', `isxdigit', `tolower', `toupper', `labs', `ldexp', `log10',
19421 `log', `malloc', `memcmp', `memcpy', `memset', `modf', `pow', `printf',
19422 `putchar', `puts', `scanf', `sinh', `sin', `snprintf', `sprintf',
19423 `sqrt', `sscanf', `strcat', `strchr', `strcmp', `strcpy', `strcspn',
19424 `strlen', `strncat', `strncmp', `strncpy', `strpbrk', `strrchr',
19425 `strspn', `strstr', `tanh', `tan', `vfprintf', `vprintf' and `vsprintf'
19426 are all recognized as built-in functions unless `-fno-builtin' is
19427 specified (or `-fno-builtin-FUNCTION' is specified for an individual
19428 function). All of these functions have corresponding versions prefixed
19431 GCC provides built-in versions of the ISO C99 floating point comparison
19432 macros that avoid raising exceptions for unordered operands. They have
19433 the same names as the standard macros ( `isgreater', `isgreaterequal',
19434 `isless', `islessequal', `islessgreater', and `isunordered') , with
19435 `__builtin_' prefixed. We intend for a library implementor to be able
19436 to simply `#define' each standard macro to its built-in equivalent.
19438 -- Built-in Function: int __builtin_types_compatible_p (TYPE1, TYPE2)
19439 You can use the built-in function `__builtin_types_compatible_p' to
19440 determine whether two types are the same.
19442 This built-in function returns 1 if the unqualified versions of the
19443 types TYPE1 and TYPE2 (which are types, not expressions) are
19444 compatible, 0 otherwise. The result of this built-in function can
19445 be used in integer constant expressions.
19447 This built-in function ignores top level qualifiers (e.g., `const',
19448 `volatile'). For example, `int' is equivalent to `const int'.
19450 The type `int[]' and `int[5]' are compatible. On the other hand,
19451 `int' and `char *' are not compatible, even if the size of their
19452 types, on the particular architecture are the same. Also, the
19453 amount of pointer indirection is taken into account when
19454 determining similarity. Consequently, `short *' is not similar to
19455 `short **'. Furthermore, two types that are typedefed are
19456 considered compatible if their underlying types are compatible.
19458 An `enum' type is not considered to be compatible with another
19459 `enum' type even if both are compatible with the same integer
19460 type; this is what the C standard specifies. For example, `enum
19461 {foo, bar}' is not similar to `enum {hot, dog}'.
19463 You would typically use this function in code whose execution
19464 varies depending on the arguments' types. For example:
19469 if (__builtin_types_compatible_p (typeof (x), long double)) \
19470 tmp = foo_long_double (tmp); \
19471 else if (__builtin_types_compatible_p (typeof (x), double)) \
19472 tmp = foo_double (tmp); \
19473 else if (__builtin_types_compatible_p (typeof (x), float)) \
19474 tmp = foo_float (tmp); \
19480 _Note:_ This construct is only available for C.
19483 -- Built-in Function: TYPE __builtin_choose_expr (CONST_EXP, EXP1,
19485 You can use the built-in function `__builtin_choose_expr' to
19486 evaluate code depending on the value of a constant expression.
19487 This built-in function returns EXP1 if CONST_EXP, which is a
19488 constant expression that must be able to be determined at compile
19489 time, is nonzero. Otherwise it returns 0.
19491 This built-in function is analogous to the `? :' operator in C,
19492 except that the expression returned has its type unaltered by
19493 promotion rules. Also, the built-in function does not evaluate
19494 the expression that was not chosen. For example, if CONST_EXP
19495 evaluates to true, EXP2 is not evaluated even if it has
19498 This built-in function can return an lvalue if the chosen argument
19501 If EXP1 is returned, the return type is the same as EXP1's type.
19502 Similarly, if EXP2 is returned, its return type is the same as
19508 __builtin_choose_expr ( \
19509 __builtin_types_compatible_p (typeof (x), double), \
19511 __builtin_choose_expr ( \
19512 __builtin_types_compatible_p (typeof (x), float), \
19514 /* The void expression results in a compile-time error \
19515 when assigning the result to something. */ \
19518 _Note:_ This construct is only available for C. Furthermore, the
19519 unused expression (EXP1 or EXP2 depending on the value of
19520 CONST_EXP) may still generate syntax errors. This may change in
19524 -- Built-in Function: int __builtin_constant_p (EXP)
19525 You can use the built-in function `__builtin_constant_p' to
19526 determine if a value is known to be constant at compile-time and
19527 hence that GCC can perform constant-folding on expressions
19528 involving that value. The argument of the function is the value
19529 to test. The function returns the integer 1 if the argument is
19530 known to be a compile-time constant and 0 if it is not known to be
19531 a compile-time constant. A return of 0 does not indicate that the
19532 value is _not_ a constant, but merely that GCC cannot prove it is
19533 a constant with the specified value of the `-O' option.
19535 You would typically use this function in an embedded application
19536 where memory was a critical resource. If you have some complex
19537 calculation, you may want it to be folded if it involves
19538 constants, but need to call a function if it does not. For
19541 #define Scale_Value(X) \
19542 (__builtin_constant_p (X) \
19543 ? ((X) * SCALE + OFFSET) : Scale (X))
19545 You may use this built-in function in either a macro or an inline
19546 function. However, if you use it in an inlined function and pass
19547 an argument of the function as the argument to the built-in, GCC
19548 will never return 1 when you call the inline function with a
19549 string constant or compound literal (*note Compound Literals::)
19550 and will not return 1 when you pass a constant numeric value to
19551 the inline function unless you specify the `-O' option.
19553 You may also use `__builtin_constant_p' in initializers for static
19554 data. For instance, you can write
19556 static const int table[] = {
19557 __builtin_constant_p (EXPRESSION) ? (EXPRESSION) : -1,
19561 This is an acceptable initializer even if EXPRESSION is not a
19562 constant expression. GCC must be more conservative about
19563 evaluating the built-in in this case, because it has no
19564 opportunity to perform optimization.
19566 Previous versions of GCC did not accept this built-in in data
19567 initializers. The earliest version where it is completely safe is
19570 -- Built-in Function: long __builtin_expect (long EXP, long C)
19571 You may use `__builtin_expect' to provide the compiler with branch
19572 prediction information. In general, you should prefer to use
19573 actual profile feedback for this (`-fprofile-arcs'), as
19574 programmers are notoriously bad at predicting how their programs
19575 actually perform. However, there are applications in which this
19576 data is hard to collect.
19578 The return value is the value of EXP, which should be an integral
19579 expression. The value of C must be a compile-time constant. The
19580 semantics of the built-in are that it is expected that EXP == C.
19583 if (__builtin_expect (x, 0))
19586 would indicate that we do not expect to call `foo', since we
19587 expect `x' to be zero. Since you are limited to integral
19588 expressions for EXP, you should use constructions such as
19590 if (__builtin_expect (ptr != NULL, 1))
19593 when testing pointer or floating-point values.
19595 -- Built-in Function: void __builtin_prefetch (const void *ADDR, ...)
19596 This function is used to minimize cache-miss latency by moving
19597 data into a cache before it is accessed. You can insert calls to
19598 `__builtin_prefetch' into code for which you know addresses of
19599 data in memory that is likely to be accessed soon. If the target
19600 supports them, data prefetch instructions will be generated. If
19601 the prefetch is done early enough before the access then the data
19602 will be in the cache by the time it is accessed.
19604 The value of ADDR is the address of the memory to prefetch. There
19605 are two optional arguments, RW and LOCALITY. The value of RW is a
19606 compile-time constant one or zero; one means that the prefetch is
19607 preparing for a write to the memory address and zero, the default,
19608 means that the prefetch is preparing for a read. The value
19609 LOCALITY must be a compile-time constant integer between zero and
19610 three. A value of zero means that the data has no temporal
19611 locality, so it need not be left in the cache after the access. A
19612 value of three means that the data has a high degree of temporal
19613 locality and should be left in all levels of cache possible.
19614 Values of one and two mean, respectively, a low or moderate degree
19615 of temporal locality. The default is three.
19617 for (i = 0; i < n; i++)
19619 a[i] = a[i] + b[i];
19620 __builtin_prefetch (&a[i+j], 1, 1);
19621 __builtin_prefetch (&b[i+j], 0, 1);
19625 Data prefetch does not generate faults if ADDR is invalid, but the
19626 address expression itself must be valid. For example, a prefetch
19627 of `p->next' will not fault if `p->next' is not a valid address,
19628 but evaluation will fault if `p' is not a valid address.
19630 If the target does not support data prefetch, the address
19631 expression is evaluated if it includes side effects but no other
19632 code is generated and GCC does not issue a warning.
19634 -- Built-in Function: double __builtin_huge_val (void)
19635 Returns a positive infinity, if supported by the floating-point
19636 format, else `DBL_MAX'. This function is suitable for
19637 implementing the ISO C macro `HUGE_VAL'.
19639 -- Built-in Function: float __builtin_huge_valf (void)
19640 Similar to `__builtin_huge_val', except the return type is `float'.
19642 -- Built-in Function: long double __builtin_huge_vall (void)
19643 Similar to `__builtin_huge_val', except the return type is `long
19646 -- Built-in Function: double __builtin_inf (void)
19647 Similar to `__builtin_huge_val', except a warning is generated if
19648 the target floating-point format does not support infinities.
19650 -- Built-in Function: float __builtin_inff (void)
19651 Similar to `__builtin_inf', except the return type is `float'.
19652 This function is suitable for implementing the ISO C99 macro
19655 -- Built-in Function: long double __builtin_infl (void)
19656 Similar to `__builtin_inf', except the return type is `long
19659 -- Built-in Function: double __builtin_nan (const char *str)
19660 This is an implementation of the ISO C99 function `nan'.
19662 Since ISO C99 defines this function in terms of `strtod', which we
19663 do not implement, a description of the parsing is in order. The
19664 string is parsed as by `strtol'; that is, the base is recognized by
19665 leading `0' or `0x' prefixes. The number parsed is placed in the
19666 significand such that the least significant bit of the number is
19667 at the least significant bit of the significand. The number is
19668 truncated to fit the significand field provided. The significand
19669 is forced to be a quiet NaN.
19671 This function, if given a string literal, is evaluated early enough
19672 that it is considered a compile-time constant.
19674 -- Built-in Function: float __builtin_nanf (const char *str)
19675 Similar to `__builtin_nan', except the return type is `float'.
19677 -- Built-in Function: long double __builtin_nanl (const char *str)
19678 Similar to `__builtin_nan', except the return type is `long
19681 -- Built-in Function: double __builtin_nans (const char *str)
19682 Similar to `__builtin_nan', except the significand is forced to be
19683 a signaling NaN. The `nans' function is proposed by WG14 N965.
19685 -- Built-in Function: float __builtin_nansf (const char *str)
19686 Similar to `__builtin_nans', except the return type is `float'.
19688 -- Built-in Function: long double __builtin_nansl (const char *str)
19689 Similar to `__builtin_nans', except the return type is `long
19692 -- Built-in Function: int __builtin_ffs (unsigned int x)
19693 Returns one plus the index of the least significant 1-bit of X, or
19694 if X is zero, returns zero.
19696 -- Built-in Function: int __builtin_clz (unsigned int x)
19697 Returns the number of leading 0-bits in X, starting at the most
19698 significant bit position. If X is 0, the result is undefined.
19700 -- Built-in Function: int __builtin_ctz (unsigned int x)
19701 Returns the number of trailing 0-bits in X, starting at the least
19702 significant bit position. If X is 0, the result is undefined.
19704 -- Built-in Function: int __builtin_popcount (unsigned int x)
19705 Returns the number of 1-bits in X.
19707 -- Built-in Function: int __builtin_parity (unsigned int x)
19708 Returns the parity of X, i.e. the number of 1-bits in X modulo 2.
19710 -- Built-in Function: int __builtin_ffsl (unsigned long)
19711 Similar to `__builtin_ffs', except the argument type is `unsigned
19714 -- Built-in Function: int __builtin_clzl (unsigned long)
19715 Similar to `__builtin_clz', except the argument type is `unsigned
19718 -- Built-in Function: int __builtin_ctzl (unsigned long)
19719 Similar to `__builtin_ctz', except the argument type is `unsigned
19722 -- Built-in Function: int __builtin_popcountl (unsigned long)
19723 Similar to `__builtin_popcount', except the argument type is
19726 -- Built-in Function: int __builtin_parityl (unsigned long)
19727 Similar to `__builtin_parity', except the argument type is
19730 -- Built-in Function: int __builtin_ffsll (unsigned long long)
19731 Similar to `__builtin_ffs', except the argument type is `unsigned
19734 -- Built-in Function: int __builtin_clzll (unsigned long long)
19735 Similar to `__builtin_clz', except the argument type is `unsigned
19738 -- Built-in Function: int __builtin_ctzll (unsigned long long)
19739 Similar to `__builtin_ctz', except the argument type is `unsigned
19742 -- Built-in Function: int __builtin_popcountll (unsigned long long)
19743 Similar to `__builtin_popcount', except the argument type is
19744 `unsigned long long'.
19746 -- Built-in Function: int __builtin_parityll (unsigned long long)
19747 Similar to `__builtin_parity', except the argument type is
19748 `unsigned long long'.
19750 -- Built-in Function: double __builtin_powi (double, int)
19751 Returns the first argument raised to the power of the second.
19752 Unlike the `pow' function no guarantees about precision and
19755 -- Built-in Function: float __builtin_powif (float, int)
19756 Similar to `__builtin_powi', except the argument and return types
19759 -- Built-in Function: long double __builtin_powil (long double, int)
19760 Similar to `__builtin_powi', except the argument and return types
19764 File: gcc.info, Node: Target Builtins, Next: Target Format Checks, Prev: Other Builtins, Up: C Extensions
19766 5.47 Built-in Functions Specific to Particular Target Machines
19767 ==============================================================
19769 On some target machines, GCC supports many built-in functions specific
19770 to those machines. Generally these generate calls to specific machine
19771 instructions, but allow the compiler to schedule those calls.
19775 * Alpha Built-in Functions::
19776 * ARM Built-in Functions::
19777 * Blackfin Built-in Functions::
19778 * FR-V Built-in Functions::
19779 * X86 Built-in Functions::
19780 * MIPS DSP Built-in Functions::
19781 * MIPS Paired-Single Support::
19782 * PowerPC AltiVec Built-in Functions::
19783 * SPARC VIS Built-in Functions::
19786 File: gcc.info, Node: Alpha Built-in Functions, Next: ARM Built-in Functions, Up: Target Builtins
19788 5.47.1 Alpha Built-in Functions
19789 -------------------------------
19791 These built-in functions are available for the Alpha family of
19792 processors, depending on the command-line switches used.
19794 The following built-in functions are always available. They all
19795 generate the machine instruction that is part of the name.
19797 long __builtin_alpha_implver (void)
19798 long __builtin_alpha_rpcc (void)
19799 long __builtin_alpha_amask (long)
19800 long __builtin_alpha_cmpbge (long, long)
19801 long __builtin_alpha_extbl (long, long)
19802 long __builtin_alpha_extwl (long, long)
19803 long __builtin_alpha_extll (long, long)
19804 long __builtin_alpha_extql (long, long)
19805 long __builtin_alpha_extwh (long, long)
19806 long __builtin_alpha_extlh (long, long)
19807 long __builtin_alpha_extqh (long, long)
19808 long __builtin_alpha_insbl (long, long)
19809 long __builtin_alpha_inswl (long, long)
19810 long __builtin_alpha_insll (long, long)
19811 long __builtin_alpha_insql (long, long)
19812 long __builtin_alpha_inswh (long, long)
19813 long __builtin_alpha_inslh (long, long)
19814 long __builtin_alpha_insqh (long, long)
19815 long __builtin_alpha_mskbl (long, long)
19816 long __builtin_alpha_mskwl (long, long)
19817 long __builtin_alpha_mskll (long, long)
19818 long __builtin_alpha_mskql (long, long)
19819 long __builtin_alpha_mskwh (long, long)
19820 long __builtin_alpha_msklh (long, long)
19821 long __builtin_alpha_mskqh (long, long)
19822 long __builtin_alpha_umulh (long, long)
19823 long __builtin_alpha_zap (long, long)
19824 long __builtin_alpha_zapnot (long, long)
19826 The following built-in functions are always with `-mmax' or
19827 `-mcpu=CPU' where CPU is `pca56' or later. They all generate the
19828 machine instruction that is part of the name.
19830 long __builtin_alpha_pklb (long)
19831 long __builtin_alpha_pkwb (long)
19832 long __builtin_alpha_unpkbl (long)
19833 long __builtin_alpha_unpkbw (long)
19834 long __builtin_alpha_minub8 (long, long)
19835 long __builtin_alpha_minsb8 (long, long)
19836 long __builtin_alpha_minuw4 (long, long)
19837 long __builtin_alpha_minsw4 (long, long)
19838 long __builtin_alpha_maxub8 (long, long)
19839 long __builtin_alpha_maxsb8 (long, long)
19840 long __builtin_alpha_maxuw4 (long, long)
19841 long __builtin_alpha_maxsw4 (long, long)
19842 long __builtin_alpha_perr (long, long)
19844 The following built-in functions are always with `-mcix' or
19845 `-mcpu=CPU' where CPU is `ev67' or later. They all generate the
19846 machine instruction that is part of the name.
19848 long __builtin_alpha_cttz (long)
19849 long __builtin_alpha_ctlz (long)
19850 long __builtin_alpha_ctpop (long)
19852 The following builtins are available on systems that use the OSF/1
19853 PALcode. Normally they invoke the `rduniq' and `wruniq' PAL calls, but
19854 when invoked with `-mtls-kernel', they invoke `rdval' and `wrval'.
19856 void *__builtin_thread_pointer (void)
19857 void __builtin_set_thread_pointer (void *)
19860 File: gcc.info, Node: ARM Built-in Functions, Next: Blackfin Built-in Functions, Prev: Alpha Built-in Functions, Up: Target Builtins
19862 5.47.2 ARM Built-in Functions
19863 -----------------------------
19865 These built-in functions are available for the ARM family of
19866 processors, when the `-mcpu=iwmmxt' switch is used:
19868 typedef int v2si __attribute__ ((vector_size (8)));
19869 typedef short v4hi __attribute__ ((vector_size (8)));
19870 typedef char v8qi __attribute__ ((vector_size (8)));
19872 int __builtin_arm_getwcx (int)
19873 void __builtin_arm_setwcx (int, int)
19874 int __builtin_arm_textrmsb (v8qi, int)
19875 int __builtin_arm_textrmsh (v4hi, int)
19876 int __builtin_arm_textrmsw (v2si, int)
19877 int __builtin_arm_textrmub (v8qi, int)
19878 int __builtin_arm_textrmuh (v4hi, int)
19879 int __builtin_arm_textrmuw (v2si, int)
19880 v8qi __builtin_arm_tinsrb (v8qi, int)
19881 v4hi __builtin_arm_tinsrh (v4hi, int)
19882 v2si __builtin_arm_tinsrw (v2si, int)
19883 long long __builtin_arm_tmia (long long, int, int)
19884 long long __builtin_arm_tmiabb (long long, int, int)
19885 long long __builtin_arm_tmiabt (long long, int, int)
19886 long long __builtin_arm_tmiaph (long long, int, int)
19887 long long __builtin_arm_tmiatb (long long, int, int)
19888 long long __builtin_arm_tmiatt (long long, int, int)
19889 int __builtin_arm_tmovmskb (v8qi)
19890 int __builtin_arm_tmovmskh (v4hi)
19891 int __builtin_arm_tmovmskw (v2si)
19892 long long __builtin_arm_waccb (v8qi)
19893 long long __builtin_arm_wacch (v4hi)
19894 long long __builtin_arm_waccw (v2si)
19895 v8qi __builtin_arm_waddb (v8qi, v8qi)
19896 v8qi __builtin_arm_waddbss (v8qi, v8qi)
19897 v8qi __builtin_arm_waddbus (v8qi, v8qi)
19898 v4hi __builtin_arm_waddh (v4hi, v4hi)
19899 v4hi __builtin_arm_waddhss (v4hi, v4hi)
19900 v4hi __builtin_arm_waddhus (v4hi, v4hi)
19901 v2si __builtin_arm_waddw (v2si, v2si)
19902 v2si __builtin_arm_waddwss (v2si, v2si)
19903 v2si __builtin_arm_waddwus (v2si, v2si)
19904 v8qi __builtin_arm_walign (v8qi, v8qi, int)
19905 long long __builtin_arm_wand(long long, long long)
19906 long long __builtin_arm_wandn (long long, long long)
19907 v8qi __builtin_arm_wavg2b (v8qi, v8qi)
19908 v8qi __builtin_arm_wavg2br (v8qi, v8qi)
19909 v4hi __builtin_arm_wavg2h (v4hi, v4hi)
19910 v4hi __builtin_arm_wavg2hr (v4hi, v4hi)
19911 v8qi __builtin_arm_wcmpeqb (v8qi, v8qi)
19912 v4hi __builtin_arm_wcmpeqh (v4hi, v4hi)
19913 v2si __builtin_arm_wcmpeqw (v2si, v2si)
19914 v8qi __builtin_arm_wcmpgtsb (v8qi, v8qi)
19915 v4hi __builtin_arm_wcmpgtsh (v4hi, v4hi)
19916 v2si __builtin_arm_wcmpgtsw (v2si, v2si)
19917 v8qi __builtin_arm_wcmpgtub (v8qi, v8qi)
19918 v4hi __builtin_arm_wcmpgtuh (v4hi, v4hi)
19919 v2si __builtin_arm_wcmpgtuw (v2si, v2si)
19920 long long __builtin_arm_wmacs (long long, v4hi, v4hi)
19921 long long __builtin_arm_wmacsz (v4hi, v4hi)
19922 long long __builtin_arm_wmacu (long long, v4hi, v4hi)
19923 long long __builtin_arm_wmacuz (v4hi, v4hi)
19924 v4hi __builtin_arm_wmadds (v4hi, v4hi)
19925 v4hi __builtin_arm_wmaddu (v4hi, v4hi)
19926 v8qi __builtin_arm_wmaxsb (v8qi, v8qi)
19927 v4hi __builtin_arm_wmaxsh (v4hi, v4hi)
19928 v2si __builtin_arm_wmaxsw (v2si, v2si)
19929 v8qi __builtin_arm_wmaxub (v8qi, v8qi)
19930 v4hi __builtin_arm_wmaxuh (v4hi, v4hi)
19931 v2si __builtin_arm_wmaxuw (v2si, v2si)
19932 v8qi __builtin_arm_wminsb (v8qi, v8qi)
19933 v4hi __builtin_arm_wminsh (v4hi, v4hi)
19934 v2si __builtin_arm_wminsw (v2si, v2si)
19935 v8qi __builtin_arm_wminub (v8qi, v8qi)
19936 v4hi __builtin_arm_wminuh (v4hi, v4hi)
19937 v2si __builtin_arm_wminuw (v2si, v2si)
19938 v4hi __builtin_arm_wmulsm (v4hi, v4hi)
19939 v4hi __builtin_arm_wmulul (v4hi, v4hi)
19940 v4hi __builtin_arm_wmulum (v4hi, v4hi)
19941 long long __builtin_arm_wor (long long, long long)
19942 v2si __builtin_arm_wpackdss (long long, long long)
19943 v2si __builtin_arm_wpackdus (long long, long long)
19944 v8qi __builtin_arm_wpackhss (v4hi, v4hi)
19945 v8qi __builtin_arm_wpackhus (v4hi, v4hi)
19946 v4hi __builtin_arm_wpackwss (v2si, v2si)
19947 v4hi __builtin_arm_wpackwus (v2si, v2si)
19948 long long __builtin_arm_wrord (long long, long long)
19949 long long __builtin_arm_wrordi (long long, int)
19950 v4hi __builtin_arm_wrorh (v4hi, long long)
19951 v4hi __builtin_arm_wrorhi (v4hi, int)
19952 v2si __builtin_arm_wrorw (v2si, long long)
19953 v2si __builtin_arm_wrorwi (v2si, int)
19954 v2si __builtin_arm_wsadb (v8qi, v8qi)
19955 v2si __builtin_arm_wsadbz (v8qi, v8qi)
19956 v2si __builtin_arm_wsadh (v4hi, v4hi)
19957 v2si __builtin_arm_wsadhz (v4hi, v4hi)
19958 v4hi __builtin_arm_wshufh (v4hi, int)
19959 long long __builtin_arm_wslld (long long, long long)
19960 long long __builtin_arm_wslldi (long long, int)
19961 v4hi __builtin_arm_wsllh (v4hi, long long)
19962 v4hi __builtin_arm_wsllhi (v4hi, int)
19963 v2si __builtin_arm_wsllw (v2si, long long)
19964 v2si __builtin_arm_wsllwi (v2si, int)
19965 long long __builtin_arm_wsrad (long long, long long)
19966 long long __builtin_arm_wsradi (long long, int)
19967 v4hi __builtin_arm_wsrah (v4hi, long long)
19968 v4hi __builtin_arm_wsrahi (v4hi, int)
19969 v2si __builtin_arm_wsraw (v2si, long long)
19970 v2si __builtin_arm_wsrawi (v2si, int)
19971 long long __builtin_arm_wsrld (long long, long long)
19972 long long __builtin_arm_wsrldi (long long, int)
19973 v4hi __builtin_arm_wsrlh (v4hi, long long)
19974 v4hi __builtin_arm_wsrlhi (v4hi, int)
19975 v2si __builtin_arm_wsrlw (v2si, long long)
19976 v2si __builtin_arm_wsrlwi (v2si, int)
19977 v8qi __builtin_arm_wsubb (v8qi, v8qi)
19978 v8qi __builtin_arm_wsubbss (v8qi, v8qi)
19979 v8qi __builtin_arm_wsubbus (v8qi, v8qi)
19980 v4hi __builtin_arm_wsubh (v4hi, v4hi)
19981 v4hi __builtin_arm_wsubhss (v4hi, v4hi)
19982 v4hi __builtin_arm_wsubhus (v4hi, v4hi)
19983 v2si __builtin_arm_wsubw (v2si, v2si)
19984 v2si __builtin_arm_wsubwss (v2si, v2si)
19985 v2si __builtin_arm_wsubwus (v2si, v2si)
19986 v4hi __builtin_arm_wunpckehsb (v8qi)
19987 v2si __builtin_arm_wunpckehsh (v4hi)
19988 long long __builtin_arm_wunpckehsw (v2si)
19989 v4hi __builtin_arm_wunpckehub (v8qi)
19990 v2si __builtin_arm_wunpckehuh (v4hi)
19991 long long __builtin_arm_wunpckehuw (v2si)
19992 v4hi __builtin_arm_wunpckelsb (v8qi)
19993 v2si __builtin_arm_wunpckelsh (v4hi)
19994 long long __builtin_arm_wunpckelsw (v2si)
19995 v4hi __builtin_arm_wunpckelub (v8qi)
19996 v2si __builtin_arm_wunpckeluh (v4hi)
19997 long long __builtin_arm_wunpckeluw (v2si)
19998 v8qi __builtin_arm_wunpckihb (v8qi, v8qi)
19999 v4hi __builtin_arm_wunpckihh (v4hi, v4hi)
20000 v2si __builtin_arm_wunpckihw (v2si, v2si)
20001 v8qi __builtin_arm_wunpckilb (v8qi, v8qi)
20002 v4hi __builtin_arm_wunpckilh (v4hi, v4hi)
20003 v2si __builtin_arm_wunpckilw (v2si, v2si)
20004 long long __builtin_arm_wxor (long long, long long)
20005 long long __builtin_arm_wzero ()
20008 File: gcc.info, Node: Blackfin Built-in Functions, Next: FR-V Built-in Functions, Prev: ARM Built-in Functions, Up: Target Builtins
20010 5.47.3 Blackfin Built-in Functions
20011 ----------------------------------
20013 Currently, there are two Blackfin-specific built-in functions. These
20014 are used for generating `CSYNC' and `SSYNC' machine insns without using
20015 inline assembly; by using these built-in functions the compiler can
20016 automatically add workarounds for hardware errata involving these
20017 instructions. These functions are named as follows:
20019 void __builtin_bfin_csync (void)
20020 void __builtin_bfin_ssync (void)
20023 File: gcc.info, Node: FR-V Built-in Functions, Next: X86 Built-in Functions, Prev: Blackfin Built-in Functions, Up: Target Builtins
20025 5.47.4 FR-V Built-in Functions
20026 ------------------------------
20028 GCC provides many FR-V-specific built-in functions. In general, these
20029 functions are intended to be compatible with those described by `FR-V
20030 Family, Softune C/C++ Compiler Manual (V6), Fujitsu Semiconductor'.
20031 The two exceptions are `__MDUNPACKH' and `__MBTOHE', the gcc forms of
20032 which pass 128-bit values by pointer rather than by value.
20034 Most of the functions are named after specific FR-V instructions.
20035 Such functions are said to be "directly mapped" and are summarized here
20041 * Directly-mapped Integer Functions::
20042 * Directly-mapped Media Functions::
20043 * Raw read/write Functions::
20044 * Other Built-in Functions::
20047 File: gcc.info, Node: Argument Types, Next: Directly-mapped Integer Functions, Up: FR-V Built-in Functions
20049 5.47.4.1 Argument Types
20050 .......................
20052 The arguments to the built-in functions can be divided into three
20053 groups: register numbers, compile-time constants and run-time values.
20054 In order to make this classification clear at a glance, the arguments
20055 and return values are given the following pseudo types:
20057 Pseudo type Real C type Constant? Description
20058 `uh' `unsigned short' No an unsigned halfword
20059 `uw1' `unsigned int' No an unsigned word
20060 `sw1' `int' No a signed word
20061 `uw2' `unsigned long long' No an unsigned doubleword
20062 `sw2' `long long' No a signed doubleword
20063 `const' `int' Yes an integer constant
20064 `acc' `int' Yes an ACC register number
20065 `iacc' `int' Yes an IACC register number
20067 These pseudo types are not defined by GCC, they are simply a notational
20068 convenience used in this manual.
20070 Arguments of type `uh', `uw1', `sw1', `uw2' and `sw2' are evaluated at
20071 run time. They correspond to register operands in the underlying FR-V
20074 `const' arguments represent immediate operands in the underlying FR-V
20075 instructions. They must be compile-time constants.
20077 `acc' arguments are evaluated at compile time and specify the number
20078 of an accumulator register. For example, an `acc' argument of 2 will
20079 select the ACC2 register.
20081 `iacc' arguments are similar to `acc' arguments but specify the number
20082 of an IACC register. See *note Other Built-in Functions:: for more
20086 File: gcc.info, Node: Directly-mapped Integer Functions, Next: Directly-mapped Media Functions, Prev: Argument Types, Up: FR-V Built-in Functions
20088 5.47.4.2 Directly-mapped Integer Functions
20089 ..........................................
20091 The functions listed below map directly to FR-V I-type instructions.
20093 Function prototype Example usage Assembly output
20094 `sw1 __ADDSS (sw1, sw1)' `C = __ADDSS (A, B)' `ADDSS A,B,C'
20095 `sw1 __SCAN (sw1, sw1)' `C = __SCAN (A, B)' `SCAN A,B,C'
20096 `sw1 __SCUTSS (sw1)' `B = __SCUTSS (A)' `SCUTSS A,B'
20097 `sw1 __SLASS (sw1, sw1)' `C = __SLASS (A, B)' `SLASS A,B,C'
20098 `void __SMASS (sw1, sw1)' `__SMASS (A, B)' `SMASS A,B'
20099 `void __SMSSS (sw1, sw1)' `__SMSSS (A, B)' `SMSSS A,B'
20100 `void __SMU (sw1, sw1)' `__SMU (A, B)' `SMU A,B'
20101 `sw2 __SMUL (sw1, sw1)' `C = __SMUL (A, B)' `SMUL A,B,C'
20102 `sw1 __SUBSS (sw1, sw1)' `C = __SUBSS (A, B)' `SUBSS A,B,C'
20103 `uw2 __UMUL (uw1, uw1)' `C = __UMUL (A, B)' `UMUL A,B,C'
20106 File: gcc.info, Node: Directly-mapped Media Functions, Next: Raw read/write Functions, Prev: Directly-mapped Integer Functions, Up: FR-V Built-in Functions
20108 5.47.4.3 Directly-mapped Media Functions
20109 ........................................
20111 The functions listed below map directly to FR-V M-type instructions.
20113 Function prototype Example usage Assembly output
20114 `uw1 __MABSHS (sw1)' `B = __MABSHS (A)' `MABSHS A,B'
20115 `void __MADDACCS (acc, acc)' `__MADDACCS (B, A)' `MADDACCS A,B'
20116 `sw1 __MADDHSS (sw1, sw1)' `C = __MADDHSS (A, B)' `MADDHSS A,B,C'
20117 `uw1 __MADDHUS (uw1, uw1)' `C = __MADDHUS (A, B)' `MADDHUS A,B,C'
20118 `uw1 __MAND (uw1, uw1)' `C = __MAND (A, B)' `MAND A,B,C'
20119 `void __MASACCS (acc, acc)' `__MASACCS (B, A)' `MASACCS A,B'
20120 `uw1 __MAVEH (uw1, uw1)' `C = __MAVEH (A, B)' `MAVEH A,B,C'
20121 `uw2 __MBTOH (uw1)' `B = __MBTOH (A)' `MBTOH A,B'
20122 `void __MBTOHE (uw1 *, uw1)' `__MBTOHE (&B, A)' `MBTOHE A,B'
20123 `void __MCLRACC (acc)' `__MCLRACC (A)' `MCLRACC A'
20124 `void __MCLRACCA (void)' `__MCLRACCA ()' `MCLRACCA'
20125 `uw1 __Mcop1 (uw1, uw1)' `C = __Mcop1 (A, B)' `Mcop1 A,B,C'
20126 `uw1 __Mcop2 (uw1, uw1)' `C = __Mcop2 (A, B)' `Mcop2 A,B,C'
20127 `uw1 __MCPLHI (uw2, const)' `C = __MCPLHI (A, B)' `MCPLHI A,#B,C'
20128 `uw1 __MCPLI (uw2, const)' `C = __MCPLI (A, B)' `MCPLI A,#B,C'
20129 `void __MCPXIS (acc, sw1, sw1)' `__MCPXIS (C, A, B)' `MCPXIS A,B,C'
20130 `void __MCPXIU (acc, uw1, uw1)' `__MCPXIU (C, A, B)' `MCPXIU A,B,C'
20131 `void __MCPXRS (acc, sw1, sw1)' `__MCPXRS (C, A, B)' `MCPXRS A,B,C'
20132 `void __MCPXRU (acc, uw1, uw1)' `__MCPXRU (C, A, B)' `MCPXRU A,B,C'
20133 `uw1 __MCUT (acc, uw1)' `C = __MCUT (A, B)' `MCUT A,B,C'
20134 `uw1 __MCUTSS (acc, sw1)' `C = __MCUTSS (A, B)' `MCUTSS A,B,C'
20135 `void __MDADDACCS (acc, acc)' `__MDADDACCS (B, A)' `MDADDACCS A,B'
20136 `void __MDASACCS (acc, acc)' `__MDASACCS (B, A)' `MDASACCS A,B'
20137 `uw2 __MDCUTSSI (acc, const)' `C = __MDCUTSSI (A, B)' `MDCUTSSI A,#B,C'
20138 `uw2 __MDPACKH (uw2, uw2)' `C = __MDPACKH (A, B)' `MDPACKH A,B,C'
20139 `uw2 __MDROTLI (uw2, const)' `C = __MDROTLI (A, B)' `MDROTLI A,#B,C'
20140 `void __MDSUBACCS (acc, acc)' `__MDSUBACCS (B, A)' `MDSUBACCS A,B'
20141 `void __MDUNPACKH (uw1 *, uw2)' `__MDUNPACKH (&B, A)' `MDUNPACKH A,B'
20142 `uw2 __MEXPDHD (uw1, const)' `C = __MEXPDHD (A, B)' `MEXPDHD A,#B,C'
20143 `uw1 __MEXPDHW (uw1, const)' `C = __MEXPDHW (A, B)' `MEXPDHW A,#B,C'
20144 `uw1 __MHDSETH (uw1, const)' `C = __MHDSETH (A, B)' `MHDSETH A,#B,C'
20145 `sw1 __MHDSETS (const)' `B = __MHDSETS (A)' `MHDSETS #A,B'
20146 `uw1 __MHSETHIH (uw1, const)' `B = __MHSETHIH (B, A)' `MHSETHIH #A,B'
20147 `sw1 __MHSETHIS (sw1, const)' `B = __MHSETHIS (B, A)' `MHSETHIS #A,B'
20148 `uw1 __MHSETLOH (uw1, const)' `B = __MHSETLOH (B, A)' `MHSETLOH #A,B'
20149 `sw1 __MHSETLOS (sw1, const)' `B = __MHSETLOS (B, A)' `MHSETLOS #A,B'
20150 `uw1 __MHTOB (uw2)' `B = __MHTOB (A)' `MHTOB A,B'
20151 `void __MMACHS (acc, sw1, sw1)' `__MMACHS (C, A, B)' `MMACHS A,B,C'
20152 `void __MMACHU (acc, uw1, uw1)' `__MMACHU (C, A, B)' `MMACHU A,B,C'
20153 `void __MMRDHS (acc, sw1, sw1)' `__MMRDHS (C, A, B)' `MMRDHS A,B,C'
20154 `void __MMRDHU (acc, uw1, uw1)' `__MMRDHU (C, A, B)' `MMRDHU A,B,C'
20155 `void __MMULHS (acc, sw1, sw1)' `__MMULHS (C, A, B)' `MMULHS A,B,C'
20156 `void __MMULHU (acc, uw1, uw1)' `__MMULHU (C, A, B)' `MMULHU A,B,C'
20157 `void __MMULXHS (acc, sw1, sw1)' `__MMULXHS (C, A, B)' `MMULXHS A,B,C'
20158 `void __MMULXHU (acc, uw1, uw1)' `__MMULXHU (C, A, B)' `MMULXHU A,B,C'
20159 `uw1 __MNOT (uw1)' `B = __MNOT (A)' `MNOT A,B'
20160 `uw1 __MOR (uw1, uw1)' `C = __MOR (A, B)' `MOR A,B,C'
20161 `uw1 __MPACKH (uh, uh)' `C = __MPACKH (A, B)' `MPACKH A,B,C'
20162 `sw2 __MQADDHSS (sw2, sw2)' `C = __MQADDHSS (A, B)' `MQADDHSS A,B,C'
20163 `uw2 __MQADDHUS (uw2, uw2)' `C = __MQADDHUS (A, B)' `MQADDHUS A,B,C'
20164 `void __MQCPXIS (acc, sw2, sw2)' `__MQCPXIS (C, A, B)' `MQCPXIS A,B,C'
20165 `void __MQCPXIU (acc, uw2, uw2)' `__MQCPXIU (C, A, B)' `MQCPXIU A,B,C'
20166 `void __MQCPXRS (acc, sw2, sw2)' `__MQCPXRS (C, A, B)' `MQCPXRS A,B,C'
20167 `void __MQCPXRU (acc, uw2, uw2)' `__MQCPXRU (C, A, B)' `MQCPXRU A,B,C'
20168 `sw2 __MQLCLRHS (sw2, sw2)' `C = __MQLCLRHS (A, B)' `MQLCLRHS A,B,C'
20169 `sw2 __MQLMTHS (sw2, sw2)' `C = __MQLMTHS (A, B)' `MQLMTHS A,B,C'
20170 `void __MQMACHS (acc, sw2, sw2)' `__MQMACHS (C, A, B)' `MQMACHS A,B,C'
20171 `void __MQMACHU (acc, uw2, uw2)' `__MQMACHU (C, A, B)' `MQMACHU A,B,C'
20172 `void __MQMACXHS (acc, sw2, `__MQMACXHS (C, A, B)' `MQMACXHS A,B,C'
20174 `void __MQMULHS (acc, sw2, sw2)' `__MQMULHS (C, A, B)' `MQMULHS A,B,C'
20175 `void __MQMULHU (acc, uw2, uw2)' `__MQMULHU (C, A, B)' `MQMULHU A,B,C'
20176 `void __MQMULXHS (acc, sw2, `__MQMULXHS (C, A, B)' `MQMULXHS A,B,C'
20178 `void __MQMULXHU (acc, uw2, `__MQMULXHU (C, A, B)' `MQMULXHU A,B,C'
20180 `sw2 __MQSATHS (sw2, sw2)' `C = __MQSATHS (A, B)' `MQSATHS A,B,C'
20181 `uw2 __MQSLLHI (uw2, int)' `C = __MQSLLHI (A, B)' `MQSLLHI A,B,C'
20182 `sw2 __MQSRAHI (sw2, int)' `C = __MQSRAHI (A, B)' `MQSRAHI A,B,C'
20183 `sw2 __MQSUBHSS (sw2, sw2)' `C = __MQSUBHSS (A, B)' `MQSUBHSS A,B,C'
20184 `uw2 __MQSUBHUS (uw2, uw2)' `C = __MQSUBHUS (A, B)' `MQSUBHUS A,B,C'
20185 `void __MQXMACHS (acc, sw2, `__MQXMACHS (C, A, B)' `MQXMACHS A,B,C'
20187 `void __MQXMACXHS (acc, sw2, `__MQXMACXHS (C, A, B)' `MQXMACXHS A,B,C'
20189 `uw1 __MRDACC (acc)' `B = __MRDACC (A)' `MRDACC A,B'
20190 `uw1 __MRDACCG (acc)' `B = __MRDACCG (A)' `MRDACCG A,B'
20191 `uw1 __MROTLI (uw1, const)' `C = __MROTLI (A, B)' `MROTLI A,#B,C'
20192 `uw1 __MROTRI (uw1, const)' `C = __MROTRI (A, B)' `MROTRI A,#B,C'
20193 `sw1 __MSATHS (sw1, sw1)' `C = __MSATHS (A, B)' `MSATHS A,B,C'
20194 `uw1 __MSATHU (uw1, uw1)' `C = __MSATHU (A, B)' `MSATHU A,B,C'
20195 `uw1 __MSLLHI (uw1, const)' `C = __MSLLHI (A, B)' `MSLLHI A,#B,C'
20196 `sw1 __MSRAHI (sw1, const)' `C = __MSRAHI (A, B)' `MSRAHI A,#B,C'
20197 `uw1 __MSRLHI (uw1, const)' `C = __MSRLHI (A, B)' `MSRLHI A,#B,C'
20198 `void __MSUBACCS (acc, acc)' `__MSUBACCS (B, A)' `MSUBACCS A,B'
20199 `sw1 __MSUBHSS (sw1, sw1)' `C = __MSUBHSS (A, B)' `MSUBHSS A,B,C'
20200 `uw1 __MSUBHUS (uw1, uw1)' `C = __MSUBHUS (A, B)' `MSUBHUS A,B,C'
20201 `void __MTRAP (void)' `__MTRAP ()' `MTRAP'
20202 `uw2 __MUNPACKH (uw1)' `B = __MUNPACKH (A)' `MUNPACKH A,B'
20203 `uw1 __MWCUT (uw2, uw1)' `C = __MWCUT (A, B)' `MWCUT A,B,C'
20204 `void __MWTACC (acc, uw1)' `__MWTACC (B, A)' `MWTACC A,B'
20205 `void __MWTACCG (acc, uw1)' `__MWTACCG (B, A)' `MWTACCG A,B'
20206 `uw1 __MXOR (uw1, uw1)' `C = __MXOR (A, B)' `MXOR A,B,C'
20209 File: gcc.info, Node: Raw read/write Functions, Next: Other Built-in Functions, Prev: Directly-mapped Media Functions, Up: FR-V Built-in Functions
20211 5.47.4.4 Raw read/write Functions
20212 .................................
20214 This sections describes built-in functions related to read and write
20215 instructions to access memory. These functions generate `membar'
20216 instructions to flush the I/O load and stores where appropriate, as
20217 described in Fujitsu's manual described above.
20219 `unsigned char __builtin_read8 (void *DATA)'
20221 `unsigned short __builtin_read16 (void *DATA)'
20223 `unsigned long __builtin_read32 (void *DATA)'
20225 `unsigned long long __builtin_read64 (void *DATA)'
20227 `void __builtin_write8 (void *DATA, unsigned char DATUM)'
20229 `void __builtin_write16 (void *DATA, unsigned short DATUM)'
20231 `void __builtin_write32 (void *DATA, unsigned long DATUM)'
20233 `void __builtin_write64 (void *DATA, unsigned long long DATUM)'
20236 File: gcc.info, Node: Other Built-in Functions, Prev: Raw read/write Functions, Up: FR-V Built-in Functions
20238 5.47.4.5 Other Built-in Functions
20239 .................................
20241 This section describes built-in functions that are not named after a
20242 specific FR-V instruction.
20244 `sw2 __IACCreadll (iacc REG)'
20245 Return the full 64-bit value of IACC0. The REG argument is
20246 reserved for future expansion and must be 0.
20248 `sw1 __IACCreadl (iacc REG)'
20249 Return the value of IACC0H if REG is 0 and IACC0L if REG is 1.
20250 Other values of REG are rejected as invalid.
20252 `void __IACCsetll (iacc REG, sw2 X)'
20253 Set the full 64-bit value of IACC0 to X. The REG argument is
20254 reserved for future expansion and must be 0.
20256 `void __IACCsetl (iacc REG, sw1 X)'
20257 Set IACC0H to X if REG is 0 and IACC0L to X if REG is 1. Other
20258 values of REG are rejected as invalid.
20260 `void __data_prefetch0 (const void *X)'
20261 Use the `dcpl' instruction to load the contents of address X into
20264 `void __data_prefetch (const void *X)'
20265 Use the `nldub' instruction to load the contents of address X into
20266 the data cache. The instruction will be issued in slot I1.
20269 File: gcc.info, Node: X86 Built-in Functions, Next: MIPS DSP Built-in Functions, Prev: FR-V Built-in Functions, Up: Target Builtins
20271 5.47.5 X86 Built-in Functions
20272 -----------------------------
20274 These built-in functions are available for the i386 and x86-64 family
20275 of computers, depending on the command-line switches used.
20277 Note that, if you specify command-line switches such as `-msse', the
20278 compiler could use the extended instruction sets even if the built-ins
20279 are not used explicitly in the program. For this reason, applications
20280 which perform runtime CPU detection must compile separate files for each
20281 supported architecture, using the appropriate flags. In particular,
20282 the file containing the CPU detection code should be compiled without
20285 The following machine modes are available for use with MMX built-in
20286 functions (*note Vector Extensions::): `V2SI' for a vector of two
20287 32-bit integers, `V4HI' for a vector of four 16-bit integers, and
20288 `V8QI' for a vector of eight 8-bit integers. Some of the built-in
20289 functions operate on MMX registers as a whole 64-bit entity, these use
20290 `DI' as their mode.
20292 If 3Dnow extensions are enabled, `V2SF' is used as a mode for a vector
20293 of two 32-bit floating point values.
20295 If SSE extensions are enabled, `V4SF' is used for a vector of four
20296 32-bit floating point values. Some instructions use a vector of four
20297 32-bit integers, these use `V4SI'. Finally, some instructions operate
20298 on an entire vector register, interpreting it as a 128-bit integer,
20299 these use mode `TI'.
20301 The following built-in functions are made available by `-mmmx'. All
20302 of them generate the machine instruction that is part of the name.
20304 v8qi __builtin_ia32_paddb (v8qi, v8qi)
20305 v4hi __builtin_ia32_paddw (v4hi, v4hi)
20306 v2si __builtin_ia32_paddd (v2si, v2si)
20307 v8qi __builtin_ia32_psubb (v8qi, v8qi)
20308 v4hi __builtin_ia32_psubw (v4hi, v4hi)
20309 v2si __builtin_ia32_psubd (v2si, v2si)
20310 v8qi __builtin_ia32_paddsb (v8qi, v8qi)
20311 v4hi __builtin_ia32_paddsw (v4hi, v4hi)
20312 v8qi __builtin_ia32_psubsb (v8qi, v8qi)
20313 v4hi __builtin_ia32_psubsw (v4hi, v4hi)
20314 v8qi __builtin_ia32_paddusb (v8qi, v8qi)
20315 v4hi __builtin_ia32_paddusw (v4hi, v4hi)
20316 v8qi __builtin_ia32_psubusb (v8qi, v8qi)
20317 v4hi __builtin_ia32_psubusw (v4hi, v4hi)
20318 v4hi __builtin_ia32_pmullw (v4hi, v4hi)
20319 v4hi __builtin_ia32_pmulhw (v4hi, v4hi)
20320 di __builtin_ia32_pand (di, di)
20321 di __builtin_ia32_pandn (di,di)
20322 di __builtin_ia32_por (di, di)
20323 di __builtin_ia32_pxor (di, di)
20324 v8qi __builtin_ia32_pcmpeqb (v8qi, v8qi)
20325 v4hi __builtin_ia32_pcmpeqw (v4hi, v4hi)
20326 v2si __builtin_ia32_pcmpeqd (v2si, v2si)
20327 v8qi __builtin_ia32_pcmpgtb (v8qi, v8qi)
20328 v4hi __builtin_ia32_pcmpgtw (v4hi, v4hi)
20329 v2si __builtin_ia32_pcmpgtd (v2si, v2si)
20330 v8qi __builtin_ia32_punpckhbw (v8qi, v8qi)
20331 v4hi __builtin_ia32_punpckhwd (v4hi, v4hi)
20332 v2si __builtin_ia32_punpckhdq (v2si, v2si)
20333 v8qi __builtin_ia32_punpcklbw (v8qi, v8qi)
20334 v4hi __builtin_ia32_punpcklwd (v4hi, v4hi)
20335 v2si __builtin_ia32_punpckldq (v2si, v2si)
20336 v8qi __builtin_ia32_packsswb (v4hi, v4hi)
20337 v4hi __builtin_ia32_packssdw (v2si, v2si)
20338 v8qi __builtin_ia32_packuswb (v4hi, v4hi)
20340 The following built-in functions are made available either with
20341 `-msse', or with a combination of `-m3dnow' and `-march=athlon'. All
20342 of them generate the machine instruction that is part of the name.
20344 v4hi __builtin_ia32_pmulhuw (v4hi, v4hi)
20345 v8qi __builtin_ia32_pavgb (v8qi, v8qi)
20346 v4hi __builtin_ia32_pavgw (v4hi, v4hi)
20347 v4hi __builtin_ia32_psadbw (v8qi, v8qi)
20348 v8qi __builtin_ia32_pmaxub (v8qi, v8qi)
20349 v4hi __builtin_ia32_pmaxsw (v4hi, v4hi)
20350 v8qi __builtin_ia32_pminub (v8qi, v8qi)
20351 v4hi __builtin_ia32_pminsw (v4hi, v4hi)
20352 int __builtin_ia32_pextrw (v4hi, int)
20353 v4hi __builtin_ia32_pinsrw (v4hi, int, int)
20354 int __builtin_ia32_pmovmskb (v8qi)
20355 void __builtin_ia32_maskmovq (v8qi, v8qi, char *)
20356 void __builtin_ia32_movntq (di *, di)
20357 void __builtin_ia32_sfence (void)
20359 The following built-in functions are available when `-msse' is used.
20360 All of them generate the machine instruction that is part of the name.
20362 int __builtin_ia32_comieq (v4sf, v4sf)
20363 int __builtin_ia32_comineq (v4sf, v4sf)
20364 int __builtin_ia32_comilt (v4sf, v4sf)
20365 int __builtin_ia32_comile (v4sf, v4sf)
20366 int __builtin_ia32_comigt (v4sf, v4sf)
20367 int __builtin_ia32_comige (v4sf, v4sf)
20368 int __builtin_ia32_ucomieq (v4sf, v4sf)
20369 int __builtin_ia32_ucomineq (v4sf, v4sf)
20370 int __builtin_ia32_ucomilt (v4sf, v4sf)
20371 int __builtin_ia32_ucomile (v4sf, v4sf)
20372 int __builtin_ia32_ucomigt (v4sf, v4sf)
20373 int __builtin_ia32_ucomige (v4sf, v4sf)
20374 v4sf __builtin_ia32_addps (v4sf, v4sf)
20375 v4sf __builtin_ia32_subps (v4sf, v4sf)
20376 v4sf __builtin_ia32_mulps (v4sf, v4sf)
20377 v4sf __builtin_ia32_divps (v4sf, v4sf)
20378 v4sf __builtin_ia32_addss (v4sf, v4sf)
20379 v4sf __builtin_ia32_subss (v4sf, v4sf)
20380 v4sf __builtin_ia32_mulss (v4sf, v4sf)
20381 v4sf __builtin_ia32_divss (v4sf, v4sf)
20382 v4si __builtin_ia32_cmpeqps (v4sf, v4sf)
20383 v4si __builtin_ia32_cmpltps (v4sf, v4sf)
20384 v4si __builtin_ia32_cmpleps (v4sf, v4sf)
20385 v4si __builtin_ia32_cmpgtps (v4sf, v4sf)
20386 v4si __builtin_ia32_cmpgeps (v4sf, v4sf)
20387 v4si __builtin_ia32_cmpunordps (v4sf, v4sf)
20388 v4si __builtin_ia32_cmpneqps (v4sf, v4sf)
20389 v4si __builtin_ia32_cmpnltps (v4sf, v4sf)
20390 v4si __builtin_ia32_cmpnleps (v4sf, v4sf)
20391 v4si __builtin_ia32_cmpngtps (v4sf, v4sf)
20392 v4si __builtin_ia32_cmpngeps (v4sf, v4sf)
20393 v4si __builtin_ia32_cmpordps (v4sf, v4sf)
20394 v4si __builtin_ia32_cmpeqss (v4sf, v4sf)
20395 v4si __builtin_ia32_cmpltss (v4sf, v4sf)
20396 v4si __builtin_ia32_cmpless (v4sf, v4sf)
20397 v4si __builtin_ia32_cmpunordss (v4sf, v4sf)
20398 v4si __builtin_ia32_cmpneqss (v4sf, v4sf)
20399 v4si __builtin_ia32_cmpnlts (v4sf, v4sf)
20400 v4si __builtin_ia32_cmpnless (v4sf, v4sf)
20401 v4si __builtin_ia32_cmpordss (v4sf, v4sf)
20402 v4sf __builtin_ia32_maxps (v4sf, v4sf)
20403 v4sf __builtin_ia32_maxss (v4sf, v4sf)
20404 v4sf __builtin_ia32_minps (v4sf, v4sf)
20405 v4sf __builtin_ia32_minss (v4sf, v4sf)
20406 v4sf __builtin_ia32_andps (v4sf, v4sf)
20407 v4sf __builtin_ia32_andnps (v4sf, v4sf)
20408 v4sf __builtin_ia32_orps (v4sf, v4sf)
20409 v4sf __builtin_ia32_xorps (v4sf, v4sf)
20410 v4sf __builtin_ia32_movss (v4sf, v4sf)
20411 v4sf __builtin_ia32_movhlps (v4sf, v4sf)
20412 v4sf __builtin_ia32_movlhps (v4sf, v4sf)
20413 v4sf __builtin_ia32_unpckhps (v4sf, v4sf)
20414 v4sf __builtin_ia32_unpcklps (v4sf, v4sf)
20415 v4sf __builtin_ia32_cvtpi2ps (v4sf, v2si)
20416 v4sf __builtin_ia32_cvtsi2ss (v4sf, int)
20417 v2si __builtin_ia32_cvtps2pi (v4sf)
20418 int __builtin_ia32_cvtss2si (v4sf)
20419 v2si __builtin_ia32_cvttps2pi (v4sf)
20420 int __builtin_ia32_cvttss2si (v4sf)
20421 v4sf __builtin_ia32_rcpps (v4sf)
20422 v4sf __builtin_ia32_rsqrtps (v4sf)
20423 v4sf __builtin_ia32_sqrtps (v4sf)
20424 v4sf __builtin_ia32_rcpss (v4sf)
20425 v4sf __builtin_ia32_rsqrtss (v4sf)
20426 v4sf __builtin_ia32_sqrtss (v4sf)
20427 v4sf __builtin_ia32_shufps (v4sf, v4sf, int)
20428 void __builtin_ia32_movntps (float *, v4sf)
20429 int __builtin_ia32_movmskps (v4sf)
20431 The following built-in functions are available when `-msse' is used.
20433 `v4sf __builtin_ia32_loadaps (float *)'
20434 Generates the `movaps' machine instruction as a load from memory.
20436 `void __builtin_ia32_storeaps (float *, v4sf)'
20437 Generates the `movaps' machine instruction as a store to memory.
20439 `v4sf __builtin_ia32_loadups (float *)'
20440 Generates the `movups' machine instruction as a load from memory.
20442 `void __builtin_ia32_storeups (float *, v4sf)'
20443 Generates the `movups' machine instruction as a store to memory.
20445 `v4sf __builtin_ia32_loadsss (float *)'
20446 Generates the `movss' machine instruction as a load from memory.
20448 `void __builtin_ia32_storess (float *, v4sf)'
20449 Generates the `movss' machine instruction as a store to memory.
20451 `v4sf __builtin_ia32_loadhps (v4sf, v2si *)'
20452 Generates the `movhps' machine instruction as a load from memory.
20454 `v4sf __builtin_ia32_loadlps (v4sf, v2si *)'
20455 Generates the `movlps' machine instruction as a load from memory
20457 `void __builtin_ia32_storehps (v4sf, v2si *)'
20458 Generates the `movhps' machine instruction as a store to memory.
20460 `void __builtin_ia32_storelps (v4sf, v2si *)'
20461 Generates the `movlps' machine instruction as a store to memory.
20463 The following built-in functions are available when `-msse2' is used.
20464 All of them generate the machine instruction that is part of the name.
20466 int __builtin_ia32_comisdeq (v2df, v2df)
20467 int __builtin_ia32_comisdlt (v2df, v2df)
20468 int __builtin_ia32_comisdle (v2df, v2df)
20469 int __builtin_ia32_comisdgt (v2df, v2df)
20470 int __builtin_ia32_comisdge (v2df, v2df)
20471 int __builtin_ia32_comisdneq (v2df, v2df)
20472 int __builtin_ia32_ucomisdeq (v2df, v2df)
20473 int __builtin_ia32_ucomisdlt (v2df, v2df)
20474 int __builtin_ia32_ucomisdle (v2df, v2df)
20475 int __builtin_ia32_ucomisdgt (v2df, v2df)
20476 int __builtin_ia32_ucomisdge (v2df, v2df)
20477 int __builtin_ia32_ucomisdneq (v2df, v2df)
20478 v2df __builtin_ia32_cmpeqpd (v2df, v2df)
20479 v2df __builtin_ia32_cmpltpd (v2df, v2df)
20480 v2df __builtin_ia32_cmplepd (v2df, v2df)
20481 v2df __builtin_ia32_cmpgtpd (v2df, v2df)
20482 v2df __builtin_ia32_cmpgepd (v2df, v2df)
20483 v2df __builtin_ia32_cmpunordpd (v2df, v2df)
20484 v2df __builtin_ia32_cmpneqpd (v2df, v2df)
20485 v2df __builtin_ia32_cmpnltpd (v2df, v2df)
20486 v2df __builtin_ia32_cmpnlepd (v2df, v2df)
20487 v2df __builtin_ia32_cmpngtpd (v2df, v2df)
20488 v2df __builtin_ia32_cmpngepd (v2df, v2df)
20489 v2df __builtin_ia32_cmpordpd (v2df, v2df)
20490 v2df __builtin_ia32_cmpeqsd (v2df, v2df)
20491 v2df __builtin_ia32_cmpltsd (v2df, v2df)
20492 v2df __builtin_ia32_cmplesd (v2df, v2df)
20493 v2df __builtin_ia32_cmpunordsd (v2df, v2df)
20494 v2df __builtin_ia32_cmpneqsd (v2df, v2df)
20495 v2df __builtin_ia32_cmpnltsd (v2df, v2df)
20496 v2df __builtin_ia32_cmpnlesd (v2df, v2df)
20497 v2df __builtin_ia32_cmpordsd (v2df, v2df)
20498 v2di __builtin_ia32_paddq (v2di, v2di)
20499 v2di __builtin_ia32_psubq (v2di, v2di)
20500 v2df __builtin_ia32_addpd (v2df, v2df)
20501 v2df __builtin_ia32_subpd (v2df, v2df)
20502 v2df __builtin_ia32_mulpd (v2df, v2df)
20503 v2df __builtin_ia32_divpd (v2df, v2df)
20504 v2df __builtin_ia32_addsd (v2df, v2df)
20505 v2df __builtin_ia32_subsd (v2df, v2df)
20506 v2df __builtin_ia32_mulsd (v2df, v2df)
20507 v2df __builtin_ia32_divsd (v2df, v2df)
20508 v2df __builtin_ia32_minpd (v2df, v2df)
20509 v2df __builtin_ia32_maxpd (v2df, v2df)
20510 v2df __builtin_ia32_minsd (v2df, v2df)
20511 v2df __builtin_ia32_maxsd (v2df, v2df)
20512 v2df __builtin_ia32_andpd (v2df, v2df)
20513 v2df __builtin_ia32_andnpd (v2df, v2df)
20514 v2df __builtin_ia32_orpd (v2df, v2df)
20515 v2df __builtin_ia32_xorpd (v2df, v2df)
20516 v2df __builtin_ia32_movsd (v2df, v2df)
20517 v2df __builtin_ia32_unpckhpd (v2df, v2df)
20518 v2df __builtin_ia32_unpcklpd (v2df, v2df)
20519 v16qi __builtin_ia32_paddb128 (v16qi, v16qi)
20520 v8hi __builtin_ia32_paddw128 (v8hi, v8hi)
20521 v4si __builtin_ia32_paddd128 (v4si, v4si)
20522 v2di __builtin_ia32_paddq128 (v2di, v2di)
20523 v16qi __builtin_ia32_psubb128 (v16qi, v16qi)
20524 v8hi __builtin_ia32_psubw128 (v8hi, v8hi)
20525 v4si __builtin_ia32_psubd128 (v4si, v4si)
20526 v2di __builtin_ia32_psubq128 (v2di, v2di)
20527 v8hi __builtin_ia32_pmullw128 (v8hi, v8hi)
20528 v8hi __builtin_ia32_pmulhw128 (v8hi, v8hi)
20529 v2di __builtin_ia32_pand128 (v2di, v2di)
20530 v2di __builtin_ia32_pandn128 (v2di, v2di)
20531 v2di __builtin_ia32_por128 (v2di, v2di)
20532 v2di __builtin_ia32_pxor128 (v2di, v2di)
20533 v16qi __builtin_ia32_pavgb128 (v16qi, v16qi)
20534 v8hi __builtin_ia32_pavgw128 (v8hi, v8hi)
20535 v16qi __builtin_ia32_pcmpeqb128 (v16qi, v16qi)
20536 v8hi __builtin_ia32_pcmpeqw128 (v8hi, v8hi)
20537 v4si __builtin_ia32_pcmpeqd128 (v4si, v4si)
20538 v16qi __builtin_ia32_pcmpgtb128 (v16qi, v16qi)
20539 v8hi __builtin_ia32_pcmpgtw128 (v8hi, v8hi)
20540 v4si __builtin_ia32_pcmpgtd128 (v4si, v4si)
20541 v16qi __builtin_ia32_pmaxub128 (v16qi, v16qi)
20542 v8hi __builtin_ia32_pmaxsw128 (v8hi, v8hi)
20543 v16qi __builtin_ia32_pminub128 (v16qi, v16qi)
20544 v8hi __builtin_ia32_pminsw128 (v8hi, v8hi)
20545 v16qi __builtin_ia32_punpckhbw128 (v16qi, v16qi)
20546 v8hi __builtin_ia32_punpckhwd128 (v8hi, v8hi)
20547 v4si __builtin_ia32_punpckhdq128 (v4si, v4si)
20548 v2di __builtin_ia32_punpckhqdq128 (v2di, v2di)
20549 v16qi __builtin_ia32_punpcklbw128 (v16qi, v16qi)
20550 v8hi __builtin_ia32_punpcklwd128 (v8hi, v8hi)
20551 v4si __builtin_ia32_punpckldq128 (v4si, v4si)
20552 v2di __builtin_ia32_punpcklqdq128 (v2di, v2di)
20553 v16qi __builtin_ia32_packsswb128 (v16qi, v16qi)
20554 v8hi __builtin_ia32_packssdw128 (v8hi, v8hi)
20555 v16qi __builtin_ia32_packuswb128 (v16qi, v16qi)
20556 v8hi __builtin_ia32_pmulhuw128 (v8hi, v8hi)
20557 void __builtin_ia32_maskmovdqu (v16qi, v16qi)
20558 v2df __builtin_ia32_loadupd (double *)
20559 void __builtin_ia32_storeupd (double *, v2df)
20560 v2df __builtin_ia32_loadhpd (v2df, double *)
20561 v2df __builtin_ia32_loadlpd (v2df, double *)
20562 int __builtin_ia32_movmskpd (v2df)
20563 int __builtin_ia32_pmovmskb128 (v16qi)
20564 void __builtin_ia32_movnti (int *, int)
20565 void __builtin_ia32_movntpd (double *, v2df)
20566 void __builtin_ia32_movntdq (v2df *, v2df)
20567 v4si __builtin_ia32_pshufd (v4si, int)
20568 v8hi __builtin_ia32_pshuflw (v8hi, int)
20569 v8hi __builtin_ia32_pshufhw (v8hi, int)
20570 v2di __builtin_ia32_psadbw128 (v16qi, v16qi)
20571 v2df __builtin_ia32_sqrtpd (v2df)
20572 v2df __builtin_ia32_sqrtsd (v2df)
20573 v2df __builtin_ia32_shufpd (v2df, v2df, int)
20574 v2df __builtin_ia32_cvtdq2pd (v4si)
20575 v4sf __builtin_ia32_cvtdq2ps (v4si)
20576 v4si __builtin_ia32_cvtpd2dq (v2df)
20577 v2si __builtin_ia32_cvtpd2pi (v2df)
20578 v4sf __builtin_ia32_cvtpd2ps (v2df)
20579 v4si __builtin_ia32_cvttpd2dq (v2df)
20580 v2si __builtin_ia32_cvttpd2pi (v2df)
20581 v2df __builtin_ia32_cvtpi2pd (v2si)
20582 int __builtin_ia32_cvtsd2si (v2df)
20583 int __builtin_ia32_cvttsd2si (v2df)
20584 long long __builtin_ia32_cvtsd2si64 (v2df)
20585 long long __builtin_ia32_cvttsd2si64 (v2df)
20586 v4si __builtin_ia32_cvtps2dq (v4sf)
20587 v2df __builtin_ia32_cvtps2pd (v4sf)
20588 v4si __builtin_ia32_cvttps2dq (v4sf)
20589 v2df __builtin_ia32_cvtsi2sd (v2df, int)
20590 v2df __builtin_ia32_cvtsi642sd (v2df, long long)
20591 v4sf __builtin_ia32_cvtsd2ss (v4sf, v2df)
20592 v2df __builtin_ia32_cvtss2sd (v2df, v4sf)
20593 void __builtin_ia32_clflush (const void *)
20594 void __builtin_ia32_lfence (void)
20595 void __builtin_ia32_mfence (void)
20596 v16qi __builtin_ia32_loaddqu (const char *)
20597 void __builtin_ia32_storedqu (char *, v16qi)
20598 unsigned long long __builtin_ia32_pmuludq (v2si, v2si)
20599 v2di __builtin_ia32_pmuludq128 (v4si, v4si)
20600 v8hi __builtin_ia32_psllw128 (v8hi, v2di)
20601 v4si __builtin_ia32_pslld128 (v4si, v2di)
20602 v2di __builtin_ia32_psllq128 (v4si, v2di)
20603 v8hi __builtin_ia32_psrlw128 (v8hi, v2di)
20604 v4si __builtin_ia32_psrld128 (v4si, v2di)
20605 v2di __builtin_ia32_psrlq128 (v2di, v2di)
20606 v8hi __builtin_ia32_psraw128 (v8hi, v2di)
20607 v4si __builtin_ia32_psrad128 (v4si, v2di)
20608 v2di __builtin_ia32_pslldqi128 (v2di, int)
20609 v8hi __builtin_ia32_psllwi128 (v8hi, int)
20610 v4si __builtin_ia32_pslldi128 (v4si, int)
20611 v2di __builtin_ia32_psllqi128 (v2di, int)
20612 v2di __builtin_ia32_psrldqi128 (v2di, int)
20613 v8hi __builtin_ia32_psrlwi128 (v8hi, int)
20614 v4si __builtin_ia32_psrldi128 (v4si, int)
20615 v2di __builtin_ia32_psrlqi128 (v2di, int)
20616 v8hi __builtin_ia32_psrawi128 (v8hi, int)
20617 v4si __builtin_ia32_psradi128 (v4si, int)
20618 v4si __builtin_ia32_pmaddwd128 (v8hi, v8hi)
20620 The following built-in functions are available when `-msse3' is used.
20621 All of them generate the machine instruction that is part of the name.
20623 v2df __builtin_ia32_addsubpd (v2df, v2df)
20624 v4sf __builtin_ia32_addsubps (v4sf, v4sf)
20625 v2df __builtin_ia32_haddpd (v2df, v2df)
20626 v4sf __builtin_ia32_haddps (v4sf, v4sf)
20627 v2df __builtin_ia32_hsubpd (v2df, v2df)
20628 v4sf __builtin_ia32_hsubps (v4sf, v4sf)
20629 v16qi __builtin_ia32_lddqu (char const *)
20630 void __builtin_ia32_monitor (void *, unsigned int, unsigned int)
20631 v2df __builtin_ia32_movddup (v2df)
20632 v4sf __builtin_ia32_movshdup (v4sf)
20633 v4sf __builtin_ia32_movsldup (v4sf)
20634 void __builtin_ia32_mwait (unsigned int, unsigned int)
20636 The following built-in functions are available when `-msse3' is used.
20638 `v2df __builtin_ia32_loadddup (double const *)'
20639 Generates the `movddup' machine instruction as a load from memory.
20641 The following built-in functions are available when `-m3dnow' is used.
20642 All of them generate the machine instruction that is part of the name.
20644 void __builtin_ia32_femms (void)
20645 v8qi __builtin_ia32_pavgusb (v8qi, v8qi)
20646 v2si __builtin_ia32_pf2id (v2sf)
20647 v2sf __builtin_ia32_pfacc (v2sf, v2sf)
20648 v2sf __builtin_ia32_pfadd (v2sf, v2sf)
20649 v2si __builtin_ia32_pfcmpeq (v2sf, v2sf)
20650 v2si __builtin_ia32_pfcmpge (v2sf, v2sf)
20651 v2si __builtin_ia32_pfcmpgt (v2sf, v2sf)
20652 v2sf __builtin_ia32_pfmax (v2sf, v2sf)
20653 v2sf __builtin_ia32_pfmin (v2sf, v2sf)
20654 v2sf __builtin_ia32_pfmul (v2sf, v2sf)
20655 v2sf __builtin_ia32_pfrcp (v2sf)
20656 v2sf __builtin_ia32_pfrcpit1 (v2sf, v2sf)
20657 v2sf __builtin_ia32_pfrcpit2 (v2sf, v2sf)
20658 v2sf __builtin_ia32_pfrsqrt (v2sf)
20659 v2sf __builtin_ia32_pfrsqrtit1 (v2sf, v2sf)
20660 v2sf __builtin_ia32_pfsub (v2sf, v2sf)
20661 v2sf __builtin_ia32_pfsubr (v2sf, v2sf)
20662 v2sf __builtin_ia32_pi2fd (v2si)
20663 v4hi __builtin_ia32_pmulhrw (v4hi, v4hi)
20665 The following built-in functions are available when both `-m3dnow' and
20666 `-march=athlon' are used. All of them generate the machine instruction
20667 that is part of the name.
20669 v2si __builtin_ia32_pf2iw (v2sf)
20670 v2sf __builtin_ia32_pfnacc (v2sf, v2sf)
20671 v2sf __builtin_ia32_pfpnacc (v2sf, v2sf)
20672 v2sf __builtin_ia32_pi2fw (v2si)
20673 v2sf __builtin_ia32_pswapdsf (v2sf)
20674 v2si __builtin_ia32_pswapdsi (v2si)
20677 File: gcc.info, Node: MIPS DSP Built-in Functions, Next: MIPS Paired-Single Support, Prev: X86 Built-in Functions, Up: Target Builtins
20679 5.47.6 MIPS DSP Built-in Functions
20680 ----------------------------------
20682 The MIPS DSP Application-Specific Extension (ASE) includes new
20683 instructions that are designed to improve the performance of DSP and
20684 media applications. It provides instructions that operate on packed
20685 8-bit integer data, Q15 fractional data and Q31 fractional data.
20687 GCC supports MIPS DSP operations using both the generic vector
20688 extensions (*note Vector Extensions::) and a collection of
20689 MIPS-specific built-in functions. Both kinds of support are enabled by
20690 the `-mdsp' command-line option.
20692 At present, GCC only provides support for operations on 32-bit
20693 vectors. The vector type associated with 8-bit integer data is usually
20694 called `v4i8' and the vector type associated with Q15 is usually called
20695 `v2q15'. They can be defined in C as follows:
20697 typedef char v4i8 __attribute__ ((vector_size(4)));
20698 typedef short v2q15 __attribute__ ((vector_size(4)));
20700 `v4i8' and `v2q15' values are initialized in the same way as
20701 aggregates. For example:
20703 v4i8 a = {1, 2, 3, 4};
20705 b = (v4i8) {5, 6, 7, 8};
20707 v2q15 c = {0x0fcb, 0x3a75};
20709 d = (v2q15) {0.1234 * 0x1.0p15, 0.4567 * 0x1.0p15};
20711 _Note:_ The CPU's endianness determines the order in which values are
20712 packed. On little-endian targets, the first value is the least
20713 significant and the last value is the most significant. The opposite
20714 order applies to big-endian targets. For example, the code above will
20715 set the lowest byte of `a' to `1' on little-endian targets and `4' on
20716 big-endian targets.
20718 _Note:_ Q15 and Q31 values must be initialized with their integer
20719 representation. As shown in this example, the integer representation
20720 of a Q15 value can be obtained by multiplying the fractional value by
20721 `0x1.0p15'. The equivalent for Q31 values is to multiply by `0x1.0p31'.
20723 The table below lists the `v4i8' and `v2q15' operations for which
20724 hardware support exists. `a' and `b' are `v4i8' values, and `c' and
20725 `d' are `v2q15' values.
20727 C code MIPS instruction
20733 It is easier to describe the DSP built-in functions if we first define
20734 the following types:
20738 typedef long long a64;
20740 `q31' and `i32' are actually the same as `int', but we use `q31' to
20741 indicate a Q31 fractional value and `i32' to indicate a 32-bit integer
20742 value. Similarly, `a64' is the same as `long long', but we use `a64'
20743 to indicate values that will be placed in one of the four DSP
20744 accumulators (`$ac0', `$ac1', `$ac2' or `$ac3').
20746 Also, some built-in functions prefer or require immediate numbers as
20747 parameters, because the corresponding DSP instructions accept both
20748 immediate numbers and register operands, or accept immediate numbers
20749 only. The immediate parameters are listed as follows.
20755 imm0_255: 0 to 255.
20756 imm_n32_31: -32 to 31.
20757 imm_n512_511: -512 to 511.
20759 The following built-in functions map directly to a particular MIPS DSP
20760 instruction. Please refer to the architecture specification for
20761 details on what each instruction does.
20763 v2q15 __builtin_mips_addq_ph (v2q15, v2q15)
20764 v2q15 __builtin_mips_addq_s_ph (v2q15, v2q15)
20765 q31 __builtin_mips_addq_s_w (q31, q31)
20766 v4i8 __builtin_mips_addu_qb (v4i8, v4i8)
20767 v4i8 __builtin_mips_addu_s_qb (v4i8, v4i8)
20768 v2q15 __builtin_mips_subq_ph (v2q15, v2q15)
20769 v2q15 __builtin_mips_subq_s_ph (v2q15, v2q15)
20770 q31 __builtin_mips_subq_s_w (q31, q31)
20771 v4i8 __builtin_mips_subu_qb (v4i8, v4i8)
20772 v4i8 __builtin_mips_subu_s_qb (v4i8, v4i8)
20773 i32 __builtin_mips_addsc (i32, i32)
20774 i32 __builtin_mips_addwc (i32, i32)
20775 i32 __builtin_mips_modsub (i32, i32)
20776 i32 __builtin_mips_raddu_w_qb (v4i8)
20777 v2q15 __builtin_mips_absq_s_ph (v2q15)
20778 q31 __builtin_mips_absq_s_w (q31)
20779 v4i8 __builtin_mips_precrq_qb_ph (v2q15, v2q15)
20780 v2q15 __builtin_mips_precrq_ph_w (q31, q31)
20781 v2q15 __builtin_mips_precrq_rs_ph_w (q31, q31)
20782 v4i8 __builtin_mips_precrqu_s_qb_ph (v2q15, v2q15)
20783 q31 __builtin_mips_preceq_w_phl (v2q15)
20784 q31 __builtin_mips_preceq_w_phr (v2q15)
20785 v2q15 __builtin_mips_precequ_ph_qbl (v4i8)
20786 v2q15 __builtin_mips_precequ_ph_qbr (v4i8)
20787 v2q15 __builtin_mips_precequ_ph_qbla (v4i8)
20788 v2q15 __builtin_mips_precequ_ph_qbra (v4i8)
20789 v2q15 __builtin_mips_preceu_ph_qbl (v4i8)
20790 v2q15 __builtin_mips_preceu_ph_qbr (v4i8)
20791 v2q15 __builtin_mips_preceu_ph_qbla (v4i8)
20792 v2q15 __builtin_mips_preceu_ph_qbra (v4i8)
20793 v4i8 __builtin_mips_shll_qb (v4i8, imm0_7)
20794 v4i8 __builtin_mips_shll_qb (v4i8, i32)
20795 v2q15 __builtin_mips_shll_ph (v2q15, imm0_15)
20796 v2q15 __builtin_mips_shll_ph (v2q15, i32)
20797 v2q15 __builtin_mips_shll_s_ph (v2q15, imm0_15)
20798 v2q15 __builtin_mips_shll_s_ph (v2q15, i32)
20799 q31 __builtin_mips_shll_s_w (q31, imm0_31)
20800 q31 __builtin_mips_shll_s_w (q31, i32)
20801 v4i8 __builtin_mips_shrl_qb (v4i8, imm0_7)
20802 v4i8 __builtin_mips_shrl_qb (v4i8, i32)
20803 v2q15 __builtin_mips_shra_ph (v2q15, imm0_15)
20804 v2q15 __builtin_mips_shra_ph (v2q15, i32)
20805 v2q15 __builtin_mips_shra_r_ph (v2q15, imm0_15)
20806 v2q15 __builtin_mips_shra_r_ph (v2q15, i32)
20807 q31 __builtin_mips_shra_r_w (q31, imm0_31)
20808 q31 __builtin_mips_shra_r_w (q31, i32)
20809 v2q15 __builtin_mips_muleu_s_ph_qbl (v4i8, v2q15)
20810 v2q15 __builtin_mips_muleu_s_ph_qbr (v4i8, v2q15)
20811 v2q15 __builtin_mips_mulq_rs_ph (v2q15, v2q15)
20812 q31 __builtin_mips_muleq_s_w_phl (v2q15, v2q15)
20813 q31 __builtin_mips_muleq_s_w_phr (v2q15, v2q15)
20814 a64 __builtin_mips_dpau_h_qbl (a64, v4i8, v4i8)
20815 a64 __builtin_mips_dpau_h_qbr (a64, v4i8, v4i8)
20816 a64 __builtin_mips_dpsu_h_qbl (a64, v4i8, v4i8)
20817 a64 __builtin_mips_dpsu_h_qbr (a64, v4i8, v4i8)
20818 a64 __builtin_mips_dpaq_s_w_ph (a64, v2q15, v2q15)
20819 a64 __builtin_mips_dpaq_sa_l_w (a64, q31, q31)
20820 a64 __builtin_mips_dpsq_s_w_ph (a64, v2q15, v2q15)
20821 a64 __builtin_mips_dpsq_sa_l_w (a64, q31, q31)
20822 a64 __builtin_mips_mulsaq_s_w_ph (a64, v2q15, v2q15)
20823 a64 __builtin_mips_maq_s_w_phl (a64, v2q15, v2q15)
20824 a64 __builtin_mips_maq_s_w_phr (a64, v2q15, v2q15)
20825 a64 __builtin_mips_maq_sa_w_phl (a64, v2q15, v2q15)
20826 a64 __builtin_mips_maq_sa_w_phr (a64, v2q15, v2q15)
20827 i32 __builtin_mips_bitrev (i32)
20828 i32 __builtin_mips_insv (i32, i32)
20829 v4i8 __builtin_mips_repl_qb (imm0_255)
20830 v4i8 __builtin_mips_repl_qb (i32)
20831 v2q15 __builtin_mips_repl_ph (imm_n512_511)
20832 v2q15 __builtin_mips_repl_ph (i32)
20833 void __builtin_mips_cmpu_eq_qb (v4i8, v4i8)
20834 void __builtin_mips_cmpu_lt_qb (v4i8, v4i8)
20835 void __builtin_mips_cmpu_le_qb (v4i8, v4i8)
20836 i32 __builtin_mips_cmpgu_eq_qb (v4i8, v4i8)
20837 i32 __builtin_mips_cmpgu_lt_qb (v4i8, v4i8)
20838 i32 __builtin_mips_cmpgu_le_qb (v4i8, v4i8)
20839 void __builtin_mips_cmp_eq_ph (v2q15, v2q15)
20840 void __builtin_mips_cmp_lt_ph (v2q15, v2q15)
20841 void __builtin_mips_cmp_le_ph (v2q15, v2q15)
20842 v4i8 __builtin_mips_pick_qb (v4i8, v4i8)
20843 v2q15 __builtin_mips_pick_ph (v2q15, v2q15)
20844 v2q15 __builtin_mips_packrl_ph (v2q15, v2q15)
20845 i32 __builtin_mips_extr_w (a64, imm0_31)
20846 i32 __builtin_mips_extr_w (a64, i32)
20847 i32 __builtin_mips_extr_r_w (a64, imm0_31)
20848 i32 __builtin_mips_extr_s_h (a64, i32)
20849 i32 __builtin_mips_extr_rs_w (a64, imm0_31)
20850 i32 __builtin_mips_extr_rs_w (a64, i32)
20851 i32 __builtin_mips_extr_s_h (a64, imm0_31)
20852 i32 __builtin_mips_extr_r_w (a64, i32)
20853 i32 __builtin_mips_extp (a64, imm0_31)
20854 i32 __builtin_mips_extp (a64, i32)
20855 i32 __builtin_mips_extpdp (a64, imm0_31)
20856 i32 __builtin_mips_extpdp (a64, i32)
20857 a64 __builtin_mips_shilo (a64, imm_n32_31)
20858 a64 __builtin_mips_shilo (a64, i32)
20859 a64 __builtin_mips_mthlip (a64, i32)
20860 void __builtin_mips_wrdsp (i32, imm0_63)
20861 i32 __builtin_mips_rddsp (imm0_63)
20862 i32 __builtin_mips_lbux (void *, i32)
20863 i32 __builtin_mips_lhx (void *, i32)
20864 i32 __builtin_mips_lwx (void *, i32)
20865 i32 __builtin_mips_bposge32 (void)
20868 File: gcc.info, Node: MIPS Paired-Single Support, Next: PowerPC AltiVec Built-in Functions, Prev: MIPS DSP Built-in Functions, Up: Target Builtins
20870 5.47.7 MIPS Paired-Single Support
20871 ---------------------------------
20873 The MIPS64 architecture includes a number of instructions that operate
20874 on pairs of single-precision floating-point values. Each pair is
20875 packed into a 64-bit floating-point register, with one element being
20876 designated the "upper half" and the other being designated the "lower
20879 GCC supports paired-single operations using both the generic vector
20880 extensions (*note Vector Extensions::) and a collection of
20881 MIPS-specific built-in functions. Both kinds of support are enabled by
20882 the `-mpaired-single' command-line option.
20884 The vector type associated with paired-single values is usually called
20885 `v2sf'. It can be defined in C as follows:
20887 typedef float v2sf __attribute__ ((vector_size (8)));
20889 `v2sf' values are initialized in the same way as aggregates. For
20892 v2sf a = {1.5, 9.1};
20897 _Note:_ The CPU's endianness determines which value is stored in the
20898 upper half of a register and which value is stored in the lower half.
20899 On little-endian targets, the first value is the lower one and the
20900 second value is the upper one. The opposite order applies to
20901 big-endian targets. For example, the code above will set the lower
20902 half of `a' to `1.5' on little-endian targets and `9.1' on big-endian
20907 * Paired-Single Arithmetic::
20908 * Paired-Single Built-in Functions::
20909 * MIPS-3D Built-in Functions::
20912 File: gcc.info, Node: Paired-Single Arithmetic, Next: Paired-Single Built-in Functions, Up: MIPS Paired-Single Support
20914 5.47.7.1 Paired-Single Arithmetic
20915 .................................
20917 The table below lists the `v2sf' operations for which hardware support
20918 exists. `a', `b' and `c' are `v2sf' values and `x' is an integral
20921 C code MIPS instruction
20926 `a * b + c' `madd.ps'
20927 `a * b - c' `msub.ps'
20928 `-(a * b + c)' `nmadd.ps'
20929 `-(a * b - c)' `nmsub.ps'
20930 `x ? a : b' `movn.ps'/`movz.ps'
20932 Note that the multiply-accumulate instructions can be disabled using
20933 the command-line option `-mno-fused-madd'.
20936 File: gcc.info, Node: Paired-Single Built-in Functions, Next: MIPS-3D Built-in Functions, Prev: Paired-Single Arithmetic, Up: MIPS Paired-Single Support
20938 5.47.7.2 Paired-Single Built-in Functions
20939 .........................................
20941 The following paired-single functions map directly to a particular MIPS
20942 instruction. Please refer to the architecture specification for
20943 details on what each instruction does.
20945 `v2sf __builtin_mips_pll_ps (v2sf, v2sf)'
20946 Pair lower lower (`pll.ps').
20948 `v2sf __builtin_mips_pul_ps (v2sf, v2sf)'
20949 Pair upper lower (`pul.ps').
20951 `v2sf __builtin_mips_plu_ps (v2sf, v2sf)'
20952 Pair lower upper (`plu.ps').
20954 `v2sf __builtin_mips_puu_ps (v2sf, v2sf)'
20955 Pair upper upper (`puu.ps').
20957 `v2sf __builtin_mips_cvt_ps_s (float, float)'
20958 Convert pair to paired single (`cvt.ps.s').
20960 `float __builtin_mips_cvt_s_pl (v2sf)'
20961 Convert pair lower to single (`cvt.s.pl').
20963 `float __builtin_mips_cvt_s_pu (v2sf)'
20964 Convert pair upper to single (`cvt.s.pu').
20966 `v2sf __builtin_mips_abs_ps (v2sf)'
20967 Absolute value (`abs.ps').
20969 `v2sf __builtin_mips_alnv_ps (v2sf, v2sf, int)'
20970 Align variable (`alnv.ps').
20972 _Note:_ The value of the third parameter must be 0 or 4 modulo 8,
20973 otherwise the result will be unpredictable. Please read the
20974 instruction description for details.
20976 The following multi-instruction functions are also available. In each
20977 case, COND can be any of the 16 floating-point conditions: `f', `un',
20978 `eq', `ueq', `olt', `ult', `ole', `ule', `sf', `ngle', `seq', `ngl',
20979 `lt', `nge', `le' or `ngt'.
20981 `v2sf __builtin_mips_movt_c_COND_ps (v2sf A, v2sf B, v2sf C, v2sf D)'
20982 `v2sf __builtin_mips_movf_c_COND_ps (v2sf A, v2sf B, v2sf C, v2sf D)'
20983 Conditional move based on floating point comparison (`c.COND.ps',
20984 `movt.ps'/`movf.ps').
20986 The `movt' functions return the value X computed by:
20992 The `movf' functions are similar but use `movf.ps' instead of
20995 `int __builtin_mips_upper_c_COND_ps (v2sf A, v2sf B)'
20996 `int __builtin_mips_lower_c_COND_ps (v2sf A, v2sf B)'
20997 Comparison of two paired-single values (`c.COND.ps',
21000 These functions compare A and B using `c.COND.ps' and return
21001 either the upper or lower half of the result. For example:
21004 if (__builtin_mips_upper_c_eq_ps (a, b))
21005 upper_halves_are_equal ();
21007 upper_halves_are_unequal ();
21009 if (__builtin_mips_lower_c_eq_ps (a, b))
21010 lower_halves_are_equal ();
21012 lower_halves_are_unequal ();
21015 File: gcc.info, Node: MIPS-3D Built-in Functions, Prev: Paired-Single Built-in Functions, Up: MIPS Paired-Single Support
21017 5.47.7.3 MIPS-3D Built-in Functions
21018 ...................................
21020 The MIPS-3D Application-Specific Extension (ASE) includes additional
21021 paired-single instructions that are designed to improve the performance
21022 of 3D graphics operations. Support for these instructions is controlled
21023 by the `-mips3d' command-line option.
21025 The functions listed below map directly to a particular MIPS-3D
21026 instruction. Please refer to the architecture specification for more
21027 details on what each instruction does.
21029 `v2sf __builtin_mips_addr_ps (v2sf, v2sf)'
21030 Reduction add (`addr.ps').
21032 `v2sf __builtin_mips_mulr_ps (v2sf, v2sf)'
21033 Reduction multiply (`mulr.ps').
21035 `v2sf __builtin_mips_cvt_pw_ps (v2sf)'
21036 Convert paired single to paired word (`cvt.pw.ps').
21038 `v2sf __builtin_mips_cvt_ps_pw (v2sf)'
21039 Convert paired word to paired single (`cvt.ps.pw').
21041 `float __builtin_mips_recip1_s (float)'
21042 `double __builtin_mips_recip1_d (double)'
21043 `v2sf __builtin_mips_recip1_ps (v2sf)'
21044 Reduced precision reciprocal (sequence step 1) (`recip1.FMT').
21046 `float __builtin_mips_recip2_s (float, float)'
21047 `double __builtin_mips_recip2_d (double, double)'
21048 `v2sf __builtin_mips_recip2_ps (v2sf, v2sf)'
21049 Reduced precision reciprocal (sequence step 2) (`recip2.FMT').
21051 `float __builtin_mips_rsqrt1_s (float)'
21052 `double __builtin_mips_rsqrt1_d (double)'
21053 `v2sf __builtin_mips_rsqrt1_ps (v2sf)'
21054 Reduced precision reciprocal square root (sequence step 1)
21057 `float __builtin_mips_rsqrt2_s (float, float)'
21058 `double __builtin_mips_rsqrt2_d (double, double)'
21059 `v2sf __builtin_mips_rsqrt2_ps (v2sf, v2sf)'
21060 Reduced precision reciprocal square root (sequence step 2)
21063 The following multi-instruction functions are also available. In each
21064 case, COND can be any of the 16 floating-point conditions: `f', `un',
21065 `eq', `ueq', `olt', `ult', `ole', `ule', `sf', `ngle', `seq', `ngl',
21066 `lt', `nge', `le' or `ngt'.
21068 `int __builtin_mips_cabs_COND_s (float A, float B)'
21069 `int __builtin_mips_cabs_COND_d (double A, double B)'
21070 Absolute comparison of two scalar values (`cabs.COND.FMT',
21073 These functions compare A and B using `cabs.COND.s' or
21074 `cabs.COND.d' and return the result as a boolean value. For
21078 if (__builtin_mips_cabs_eq_s (a, b))
21083 `int __builtin_mips_upper_cabs_COND_ps (v2sf A, v2sf B)'
21084 `int __builtin_mips_lower_cabs_COND_ps (v2sf A, v2sf B)'
21085 Absolute comparison of two paired-single values (`cabs.COND.ps',
21088 These functions compare A and B using `cabs.COND.ps' and return
21089 either the upper or lower half of the result. For example:
21092 if (__builtin_mips_upper_cabs_eq_ps (a, b))
21093 upper_halves_are_equal ();
21095 upper_halves_are_unequal ();
21097 if (__builtin_mips_lower_cabs_eq_ps (a, b))
21098 lower_halves_are_equal ();
21100 lower_halves_are_unequal ();
21102 `v2sf __builtin_mips_movt_cabs_COND_ps (v2sf A, v2sf B, v2sf C, v2sf D)'
21103 `v2sf __builtin_mips_movf_cabs_COND_ps (v2sf A, v2sf B, v2sf C, v2sf D)'
21104 Conditional move based on absolute comparison (`cabs.COND.ps',
21105 `movt.ps'/`movf.ps').
21107 The `movt' functions return the value X computed by:
21109 cabs.COND.ps CC,A,B
21113 The `movf' functions are similar but use `movf.ps' instead of
21116 `int __builtin_mips_any_c_COND_ps (v2sf A, v2sf B)'
21117 `int __builtin_mips_all_c_COND_ps (v2sf A, v2sf B)'
21118 `int __builtin_mips_any_cabs_COND_ps (v2sf A, v2sf B)'
21119 `int __builtin_mips_all_cabs_COND_ps (v2sf A, v2sf B)'
21120 Comparison of two paired-single values (`c.COND.ps'/`cabs.COND.ps',
21121 `bc1any2t'/`bc1any2f').
21123 These functions compare A and B using `c.COND.ps' or
21124 `cabs.COND.ps'. The `any' forms return true if either result is
21125 true and the `all' forms return true if both results are true.
21129 if (__builtin_mips_any_c_eq_ps (a, b))
21134 if (__builtin_mips_all_c_eq_ps (a, b))
21139 `int __builtin_mips_any_c_COND_4s (v2sf A, v2sf B, v2sf C, v2sf D)'
21140 `int __builtin_mips_all_c_COND_4s (v2sf A, v2sf B, v2sf C, v2sf D)'
21141 `int __builtin_mips_any_cabs_COND_4s (v2sf A, v2sf B, v2sf C, v2sf D)'
21142 `int __builtin_mips_all_cabs_COND_4s (v2sf A, v2sf B, v2sf C, v2sf D)'
21143 Comparison of four paired-single values
21144 (`c.COND.ps'/`cabs.COND.ps', `bc1any4t'/`bc1any4f').
21146 These functions use `c.COND.ps' or `cabs.COND.ps' to compare A
21147 with B and to compare C with D. The `any' forms return true if
21148 any of the four results are true and the `all' forms return true
21149 if all four results are true. For example:
21152 if (__builtin_mips_any_c_eq_4s (a, b, c, d))
21157 if (__builtin_mips_all_c_eq_4s (a, b, c, d))
21163 File: gcc.info, Node: PowerPC AltiVec Built-in Functions, Next: SPARC VIS Built-in Functions, Prev: MIPS Paired-Single Support, Up: Target Builtins
21165 5.47.8 PowerPC AltiVec Built-in Functions
21166 -----------------------------------------
21168 GCC provides an interface for the PowerPC family of processors to access
21169 the AltiVec operations described in Motorola's AltiVec Programming
21170 Interface Manual. The interface is made available by including
21171 `<altivec.h>' and using `-maltivec' and `-mabi=altivec'. The interface
21172 supports the following vector types.
21174 vector unsigned char
21178 vector unsigned short
21179 vector signed short
21183 vector unsigned int
21188 GCC's implementation of the high-level language interface available
21189 from C and C++ code differs from Motorola's documentation in several
21192 * A vector constant is a list of constant expressions within curly
21195 * A vector initializer requires no cast if the vector constant is of
21196 the same type as the variable it is initializing.
21198 * If `signed' or `unsigned' is omitted, the signedness of the vector
21199 type is the default signedness of the base type. The default
21200 varies depending on the operating system, so a portable program
21201 should always specify the signedness.
21203 * Compiling with `-maltivec' adds keywords `__vector', `__pixel',
21204 and `__bool'. Macros `vector', `pixel', and `bool' are defined in
21205 `<altivec.h>' and can be undefined.
21207 * GCC allows using a `typedef' name as the type specifier for a
21210 * For C, overloaded functions are implemented with macros so the
21211 following does not work:
21213 vec_add ((vector signed int){1, 2, 3, 4}, foo);
21215 Since `vec_add' is a macro, the vector constant in the example is
21216 treated as four separate arguments. Wrap the entire argument in
21217 parentheses for this to work.
21219 _Note:_ Only the `<altivec.h>' interface is supported. Internally,
21220 GCC uses built-in functions to achieve the functionality in the
21221 aforementioned header file, but they are not supported and are subject
21222 to change without notice.
21224 The following interfaces are supported for the generic and specific
21225 AltiVec operations and the AltiVec predicates. In cases where there is
21226 a direct mapping between generic and specific operations, only the
21227 generic names are shown here, although the specific operations can also
21230 Arguments that are documented as `const int' require literal integral
21231 values within the range required for that operation.
21233 vector signed char vec_abs (vector signed char);
21234 vector signed short vec_abs (vector signed short);
21235 vector signed int vec_abs (vector signed int);
21236 vector float vec_abs (vector float);
21238 vector signed char vec_abss (vector signed char);
21239 vector signed short vec_abss (vector signed short);
21240 vector signed int vec_abss (vector signed int);
21242 vector signed char vec_add (vector bool char, vector signed char);
21243 vector signed char vec_add (vector signed char, vector bool char);
21244 vector signed char vec_add (vector signed char, vector signed char);
21245 vector unsigned char vec_add (vector bool char, vector unsigned char);
21246 vector unsigned char vec_add (vector unsigned char, vector bool char);
21247 vector unsigned char vec_add (vector unsigned char,
21248 vector unsigned char);
21249 vector signed short vec_add (vector bool short, vector signed short);
21250 vector signed short vec_add (vector signed short, vector bool short);
21251 vector signed short vec_add (vector signed short, vector signed short);
21252 vector unsigned short vec_add (vector bool short,
21253 vector unsigned short);
21254 vector unsigned short vec_add (vector unsigned short,
21255 vector bool short);
21256 vector unsigned short vec_add (vector unsigned short,
21257 vector unsigned short);
21258 vector signed int vec_add (vector bool int, vector signed int);
21259 vector signed int vec_add (vector signed int, vector bool int);
21260 vector signed int vec_add (vector signed int, vector signed int);
21261 vector unsigned int vec_add (vector bool int, vector unsigned int);
21262 vector unsigned int vec_add (vector unsigned int, vector bool int);
21263 vector unsigned int vec_add (vector unsigned int, vector unsigned int);
21264 vector float vec_add (vector float, vector float);
21266 vector float vec_vaddfp (vector float, vector float);
21268 vector signed int vec_vadduwm (vector bool int, vector signed int);
21269 vector signed int vec_vadduwm (vector signed int, vector bool int);
21270 vector signed int vec_vadduwm (vector signed int, vector signed int);
21271 vector unsigned int vec_vadduwm (vector bool int, vector unsigned int);
21272 vector unsigned int vec_vadduwm (vector unsigned int, vector bool int);
21273 vector unsigned int vec_vadduwm (vector unsigned int,
21274 vector unsigned int);
21276 vector signed short vec_vadduhm (vector bool short,
21277 vector signed short);
21278 vector signed short vec_vadduhm (vector signed short,
21279 vector bool short);
21280 vector signed short vec_vadduhm (vector signed short,
21281 vector signed short);
21282 vector unsigned short vec_vadduhm (vector bool short,
21283 vector unsigned short);
21284 vector unsigned short vec_vadduhm (vector unsigned short,
21285 vector bool short);
21286 vector unsigned short vec_vadduhm (vector unsigned short,
21287 vector unsigned short);
21289 vector signed char vec_vaddubm (vector bool char, vector signed char);
21290 vector signed char vec_vaddubm (vector signed char, vector bool char);
21291 vector signed char vec_vaddubm (vector signed char, vector signed char);
21292 vector unsigned char vec_vaddubm (vector bool char,
21293 vector unsigned char);
21294 vector unsigned char vec_vaddubm (vector unsigned char,
21296 vector unsigned char vec_vaddubm (vector unsigned char,
21297 vector unsigned char);
21299 vector unsigned int vec_addc (vector unsigned int, vector unsigned int);
21301 vector unsigned char vec_adds (vector bool char, vector unsigned char);
21302 vector unsigned char vec_adds (vector unsigned char, vector bool char);
21303 vector unsigned char vec_adds (vector unsigned char,
21304 vector unsigned char);
21305 vector signed char vec_adds (vector bool char, vector signed char);
21306 vector signed char vec_adds (vector signed char, vector bool char);
21307 vector signed char vec_adds (vector signed char, vector signed char);
21308 vector unsigned short vec_adds (vector bool short,
21309 vector unsigned short);
21310 vector unsigned short vec_adds (vector unsigned short,
21311 vector bool short);
21312 vector unsigned short vec_adds (vector unsigned short,
21313 vector unsigned short);
21314 vector signed short vec_adds (vector bool short, vector signed short);
21315 vector signed short vec_adds (vector signed short, vector bool short);
21316 vector signed short vec_adds (vector signed short, vector signed short);
21317 vector unsigned int vec_adds (vector bool int, vector unsigned int);
21318 vector unsigned int vec_adds (vector unsigned int, vector bool int);
21319 vector unsigned int vec_adds (vector unsigned int, vector unsigned int);
21320 vector signed int vec_adds (vector bool int, vector signed int);
21321 vector signed int vec_adds (vector signed int, vector bool int);
21322 vector signed int vec_adds (vector signed int, vector signed int);
21324 vector signed int vec_vaddsws (vector bool int, vector signed int);
21325 vector signed int vec_vaddsws (vector signed int, vector bool int);
21326 vector signed int vec_vaddsws (vector signed int, vector signed int);
21328 vector unsigned int vec_vadduws (vector bool int, vector unsigned int);
21329 vector unsigned int vec_vadduws (vector unsigned int, vector bool int);
21330 vector unsigned int vec_vadduws (vector unsigned int,
21331 vector unsigned int);
21333 vector signed short vec_vaddshs (vector bool short,
21334 vector signed short);
21335 vector signed short vec_vaddshs (vector signed short,
21336 vector bool short);
21337 vector signed short vec_vaddshs (vector signed short,
21338 vector signed short);
21340 vector unsigned short vec_vadduhs (vector bool short,
21341 vector unsigned short);
21342 vector unsigned short vec_vadduhs (vector unsigned short,
21343 vector bool short);
21344 vector unsigned short vec_vadduhs (vector unsigned short,
21345 vector unsigned short);
21347 vector signed char vec_vaddsbs (vector bool char, vector signed char);
21348 vector signed char vec_vaddsbs (vector signed char, vector bool char);
21349 vector signed char vec_vaddsbs (vector signed char, vector signed char);
21351 vector unsigned char vec_vaddubs (vector bool char,
21352 vector unsigned char);
21353 vector unsigned char vec_vaddubs (vector unsigned char,
21355 vector unsigned char vec_vaddubs (vector unsigned char,
21356 vector unsigned char);
21358 vector float vec_and (vector float, vector float);
21359 vector float vec_and (vector float, vector bool int);
21360 vector float vec_and (vector bool int, vector float);
21361 vector bool int vec_and (vector bool int, vector bool int);
21362 vector signed int vec_and (vector bool int, vector signed int);
21363 vector signed int vec_and (vector signed int, vector bool int);
21364 vector signed int vec_and (vector signed int, vector signed int);
21365 vector unsigned int vec_and (vector bool int, vector unsigned int);
21366 vector unsigned int vec_and (vector unsigned int, vector bool int);
21367 vector unsigned int vec_and (vector unsigned int, vector unsigned int);
21368 vector bool short vec_and (vector bool short, vector bool short);
21369 vector signed short vec_and (vector bool short, vector signed short);
21370 vector signed short vec_and (vector signed short, vector bool short);
21371 vector signed short vec_and (vector signed short, vector signed short);
21372 vector unsigned short vec_and (vector bool short,
21373 vector unsigned short);
21374 vector unsigned short vec_and (vector unsigned short,
21375 vector bool short);
21376 vector unsigned short vec_and (vector unsigned short,
21377 vector unsigned short);
21378 vector signed char vec_and (vector bool char, vector signed char);
21379 vector bool char vec_and (vector bool char, vector bool char);
21380 vector signed char vec_and (vector signed char, vector bool char);
21381 vector signed char vec_and (vector signed char, vector signed char);
21382 vector unsigned char vec_and (vector bool char, vector unsigned char);
21383 vector unsigned char vec_and (vector unsigned char, vector bool char);
21384 vector unsigned char vec_and (vector unsigned char,
21385 vector unsigned char);
21387 vector float vec_andc (vector float, vector float);
21388 vector float vec_andc (vector float, vector bool int);
21389 vector float vec_andc (vector bool int, vector float);
21390 vector bool int vec_andc (vector bool int, vector bool int);
21391 vector signed int vec_andc (vector bool int, vector signed int);
21392 vector signed int vec_andc (vector signed int, vector bool int);
21393 vector signed int vec_andc (vector signed int, vector signed int);
21394 vector unsigned int vec_andc (vector bool int, vector unsigned int);
21395 vector unsigned int vec_andc (vector unsigned int, vector bool int);
21396 vector unsigned int vec_andc (vector unsigned int, vector unsigned int);
21397 vector bool short vec_andc (vector bool short, vector bool short);
21398 vector signed short vec_andc (vector bool short, vector signed short);
21399 vector signed short vec_andc (vector signed short, vector bool short);
21400 vector signed short vec_andc (vector signed short, vector signed short);
21401 vector unsigned short vec_andc (vector bool short,
21402 vector unsigned short);
21403 vector unsigned short vec_andc (vector unsigned short,
21404 vector bool short);
21405 vector unsigned short vec_andc (vector unsigned short,
21406 vector unsigned short);
21407 vector signed char vec_andc (vector bool char, vector signed char);
21408 vector bool char vec_andc (vector bool char, vector bool char);
21409 vector signed char vec_andc (vector signed char, vector bool char);
21410 vector signed char vec_andc (vector signed char, vector signed char);
21411 vector unsigned char vec_andc (vector bool char, vector unsigned char);
21412 vector unsigned char vec_andc (vector unsigned char, vector bool char);
21413 vector unsigned char vec_andc (vector unsigned char,
21414 vector unsigned char);
21416 vector unsigned char vec_avg (vector unsigned char,
21417 vector unsigned char);
21418 vector signed char vec_avg (vector signed char, vector signed char);
21419 vector unsigned short vec_avg (vector unsigned short,
21420 vector unsigned short);
21421 vector signed short vec_avg (vector signed short, vector signed short);
21422 vector unsigned int vec_avg (vector unsigned int, vector unsigned int);
21423 vector signed int vec_avg (vector signed int, vector signed int);
21425 vector signed int vec_vavgsw (vector signed int, vector signed int);
21427 vector unsigned int vec_vavguw (vector unsigned int,
21428 vector unsigned int);
21430 vector signed short vec_vavgsh (vector signed short,
21431 vector signed short);
21433 vector unsigned short vec_vavguh (vector unsigned short,
21434 vector unsigned short);
21436 vector signed char vec_vavgsb (vector signed char, vector signed char);
21438 vector unsigned char vec_vavgub (vector unsigned char,
21439 vector unsigned char);
21441 vector float vec_ceil (vector float);
21443 vector signed int vec_cmpb (vector float, vector float);
21445 vector bool char vec_cmpeq (vector signed char, vector signed char);
21446 vector bool char vec_cmpeq (vector unsigned char, vector unsigned char);
21447 vector bool short vec_cmpeq (vector signed short, vector signed short);
21448 vector bool short vec_cmpeq (vector unsigned short,
21449 vector unsigned short);
21450 vector bool int vec_cmpeq (vector signed int, vector signed int);
21451 vector bool int vec_cmpeq (vector unsigned int, vector unsigned int);
21452 vector bool int vec_cmpeq (vector float, vector float);
21454 vector bool int vec_vcmpeqfp (vector float, vector float);
21456 vector bool int vec_vcmpequw (vector signed int, vector signed int);
21457 vector bool int vec_vcmpequw (vector unsigned int, vector unsigned int);
21459 vector bool short vec_vcmpequh (vector signed short,
21460 vector signed short);
21461 vector bool short vec_vcmpequh (vector unsigned short,
21462 vector unsigned short);
21464 vector bool char vec_vcmpequb (vector signed char, vector signed char);
21465 vector bool char vec_vcmpequb (vector unsigned char,
21466 vector unsigned char);
21468 vector bool int vec_cmpge (vector float, vector float);
21470 vector bool char vec_cmpgt (vector unsigned char, vector unsigned char);
21471 vector bool char vec_cmpgt (vector signed char, vector signed char);
21472 vector bool short vec_cmpgt (vector unsigned short,
21473 vector unsigned short);
21474 vector bool short vec_cmpgt (vector signed short, vector signed short);
21475 vector bool int vec_cmpgt (vector unsigned int, vector unsigned int);
21476 vector bool int vec_cmpgt (vector signed int, vector signed int);
21477 vector bool int vec_cmpgt (vector float, vector float);
21479 vector bool int vec_vcmpgtfp (vector float, vector float);
21481 vector bool int vec_vcmpgtsw (vector signed int, vector signed int);
21483 vector bool int vec_vcmpgtuw (vector unsigned int, vector unsigned int);
21485 vector bool short vec_vcmpgtsh (vector signed short,
21486 vector signed short);
21488 vector bool short vec_vcmpgtuh (vector unsigned short,
21489 vector unsigned short);
21491 vector bool char vec_vcmpgtsb (vector signed char, vector signed char);
21493 vector bool char vec_vcmpgtub (vector unsigned char,
21494 vector unsigned char);
21496 vector bool int vec_cmple (vector float, vector float);
21498 vector bool char vec_cmplt (vector unsigned char, vector unsigned char);
21499 vector bool char vec_cmplt (vector signed char, vector signed char);
21500 vector bool short vec_cmplt (vector unsigned short,
21501 vector unsigned short);
21502 vector bool short vec_cmplt (vector signed short, vector signed short);
21503 vector bool int vec_cmplt (vector unsigned int, vector unsigned int);
21504 vector bool int vec_cmplt (vector signed int, vector signed int);
21505 vector bool int vec_cmplt (vector float, vector float);
21507 vector float vec_ctf (vector unsigned int, const int);
21508 vector float vec_ctf (vector signed int, const int);
21510 vector float vec_vcfsx (vector signed int, const int);
21512 vector float vec_vcfux (vector unsigned int, const int);
21514 vector signed int vec_cts (vector float, const int);
21516 vector unsigned int vec_ctu (vector float, const int);
21518 void vec_dss (const int);
21520 void vec_dssall (void);
21522 void vec_dst (const vector unsigned char *, int, const int);
21523 void vec_dst (const vector signed char *, int, const int);
21524 void vec_dst (const vector bool char *, int, const int);
21525 void vec_dst (const vector unsigned short *, int, const int);
21526 void vec_dst (const vector signed short *, int, const int);
21527 void vec_dst (const vector bool short *, int, const int);
21528 void vec_dst (const vector pixel *, int, const int);
21529 void vec_dst (const vector unsigned int *, int, const int);
21530 void vec_dst (const vector signed int *, int, const int);
21531 void vec_dst (const vector bool int *, int, const int);
21532 void vec_dst (const vector float *, int, const int);
21533 void vec_dst (const unsigned char *, int, const int);
21534 void vec_dst (const signed char *, int, const int);
21535 void vec_dst (const unsigned short *, int, const int);
21536 void vec_dst (const short *, int, const int);
21537 void vec_dst (const unsigned int *, int, const int);
21538 void vec_dst (const int *, int, const int);
21539 void vec_dst (const unsigned long *, int, const int);
21540 void vec_dst (const long *, int, const int);
21541 void vec_dst (const float *, int, const int);
21543 void vec_dstst (const vector unsigned char *, int, const int);
21544 void vec_dstst (const vector signed char *, int, const int);
21545 void vec_dstst (const vector bool char *, int, const int);
21546 void vec_dstst (const vector unsigned short *, int, const int);
21547 void vec_dstst (const vector signed short *, int, const int);
21548 void vec_dstst (const vector bool short *, int, const int);
21549 void vec_dstst (const vector pixel *, int, const int);
21550 void vec_dstst (const vector unsigned int *, int, const int);
21551 void vec_dstst (const vector signed int *, int, const int);
21552 void vec_dstst (const vector bool int *, int, const int);
21553 void vec_dstst (const vector float *, int, const int);
21554 void vec_dstst (const unsigned char *, int, const int);
21555 void vec_dstst (const signed char *, int, const int);
21556 void vec_dstst (const unsigned short *, int, const int);
21557 void vec_dstst (const short *, int, const int);
21558 void vec_dstst (const unsigned int *, int, const int);
21559 void vec_dstst (const int *, int, const int);
21560 void vec_dstst (const unsigned long *, int, const int);
21561 void vec_dstst (const long *, int, const int);
21562 void vec_dstst (const float *, int, const int);
21564 void vec_dststt (const vector unsigned char *, int, const int);
21565 void vec_dststt (const vector signed char *, int, const int);
21566 void vec_dststt (const vector bool char *, int, const int);
21567 void vec_dststt (const vector unsigned short *, int, const int);
21568 void vec_dststt (const vector signed short *, int, const int);
21569 void vec_dststt (const vector bool short *, int, const int);
21570 void vec_dststt (const vector pixel *, int, const int);
21571 void vec_dststt (const vector unsigned int *, int, const int);
21572 void vec_dststt (const vector signed int *, int, const int);
21573 void vec_dststt (const vector bool int *, int, const int);
21574 void vec_dststt (const vector float *, int, const int);
21575 void vec_dststt (const unsigned char *, int, const int);
21576 void vec_dststt (const signed char *, int, const int);
21577 void vec_dststt (const unsigned short *, int, const int);
21578 void vec_dststt (const short *, int, const int);
21579 void vec_dststt (const unsigned int *, int, const int);
21580 void vec_dststt (const int *, int, const int);
21581 void vec_dststt (const unsigned long *, int, const int);
21582 void vec_dststt (const long *, int, const int);
21583 void vec_dststt (const float *, int, const int);
21585 void vec_dstt (const vector unsigned char *, int, const int);
21586 void vec_dstt (const vector signed char *, int, const int);
21587 void vec_dstt (const vector bool char *, int, const int);
21588 void vec_dstt (const vector unsigned short *, int, const int);
21589 void vec_dstt (const vector signed short *, int, const int);
21590 void vec_dstt (const vector bool short *, int, const int);
21591 void vec_dstt (const vector pixel *, int, const int);
21592 void vec_dstt (const vector unsigned int *, int, const int);
21593 void vec_dstt (const vector signed int *, int, const int);
21594 void vec_dstt (const vector bool int *, int, const int);
21595 void vec_dstt (const vector float *, int, const int);
21596 void vec_dstt (const unsigned char *, int, const int);
21597 void vec_dstt (const signed char *, int, const int);
21598 void vec_dstt (const unsigned short *, int, const int);
21599 void vec_dstt (const short *, int, const int);
21600 void vec_dstt (const unsigned int *, int, const int);
21601 void vec_dstt (const int *, int, const int);
21602 void vec_dstt (const unsigned long *, int, const int);
21603 void vec_dstt (const long *, int, const int);
21604 void vec_dstt (const float *, int, const int);
21606 vector float vec_expte (vector float);
21608 vector float vec_floor (vector float);
21610 vector float vec_ld (int, const vector float *);
21611 vector float vec_ld (int, const float *);
21612 vector bool int vec_ld (int, const vector bool int *);
21613 vector signed int vec_ld (int, const vector signed int *);
21614 vector signed int vec_ld (int, const int *);
21615 vector signed int vec_ld (int, const long *);
21616 vector unsigned int vec_ld (int, const vector unsigned int *);
21617 vector unsigned int vec_ld (int, const unsigned int *);
21618 vector unsigned int vec_ld (int, const unsigned long *);
21619 vector bool short vec_ld (int, const vector bool short *);
21620 vector pixel vec_ld (int, const vector pixel *);
21621 vector signed short vec_ld (int, const vector signed short *);
21622 vector signed short vec_ld (int, const short *);
21623 vector unsigned short vec_ld (int, const vector unsigned short *);
21624 vector unsigned short vec_ld (int, const unsigned short *);
21625 vector bool char vec_ld (int, const vector bool char *);
21626 vector signed char vec_ld (int, const vector signed char *);
21627 vector signed char vec_ld (int, const signed char *);
21628 vector unsigned char vec_ld (int, const vector unsigned char *);
21629 vector unsigned char vec_ld (int, const unsigned char *);
21631 vector signed char vec_lde (int, const signed char *);
21632 vector unsigned char vec_lde (int, const unsigned char *);
21633 vector signed short vec_lde (int, const short *);
21634 vector unsigned short vec_lde (int, const unsigned short *);
21635 vector float vec_lde (int, const float *);
21636 vector signed int vec_lde (int, const int *);
21637 vector unsigned int vec_lde (int, const unsigned int *);
21638 vector signed int vec_lde (int, const long *);
21639 vector unsigned int vec_lde (int, const unsigned long *);
21641 vector float vec_lvewx (int, float *);
21642 vector signed int vec_lvewx (int, int *);
21643 vector unsigned int vec_lvewx (int, unsigned int *);
21644 vector signed int vec_lvewx (int, long *);
21645 vector unsigned int vec_lvewx (int, unsigned long *);
21647 vector signed short vec_lvehx (int, short *);
21648 vector unsigned short vec_lvehx (int, unsigned short *);
21650 vector signed char vec_lvebx (int, char *);
21651 vector unsigned char vec_lvebx (int, unsigned char *);
21653 vector float vec_ldl (int, const vector float *);
21654 vector float vec_ldl (int, const float *);
21655 vector bool int vec_ldl (int, const vector bool int *);
21656 vector signed int vec_ldl (int, const vector signed int *);
21657 vector signed int vec_ldl (int, const int *);
21658 vector signed int vec_ldl (int, const long *);
21659 vector unsigned int vec_ldl (int, const vector unsigned int *);
21660 vector unsigned int vec_ldl (int, const unsigned int *);
21661 vector unsigned int vec_ldl (int, const unsigned long *);
21662 vector bool short vec_ldl (int, const vector bool short *);
21663 vector pixel vec_ldl (int, const vector pixel *);
21664 vector signed short vec_ldl (int, const vector signed short *);
21665 vector signed short vec_ldl (int, const short *);
21666 vector unsigned short vec_ldl (int, const vector unsigned short *);
21667 vector unsigned short vec_ldl (int, const unsigned short *);
21668 vector bool char vec_ldl (int, const vector bool char *);
21669 vector signed char vec_ldl (int, const vector signed char *);
21670 vector signed char vec_ldl (int, const signed char *);
21671 vector unsigned char vec_ldl (int, const vector unsigned char *);
21672 vector unsigned char vec_ldl (int, const unsigned char *);
21674 vector float vec_loge (vector float);
21676 vector unsigned char vec_lvsl (int, const volatile unsigned char *);
21677 vector unsigned char vec_lvsl (int, const volatile signed char *);
21678 vector unsigned char vec_lvsl (int, const volatile unsigned short *);
21679 vector unsigned char vec_lvsl (int, const volatile short *);
21680 vector unsigned char vec_lvsl (int, const volatile unsigned int *);
21681 vector unsigned char vec_lvsl (int, const volatile int *);
21682 vector unsigned char vec_lvsl (int, const volatile unsigned long *);
21683 vector unsigned char vec_lvsl (int, const volatile long *);
21684 vector unsigned char vec_lvsl (int, const volatile float *);
21686 vector unsigned char vec_lvsr (int, const volatile unsigned char *);
21687 vector unsigned char vec_lvsr (int, const volatile signed char *);
21688 vector unsigned char vec_lvsr (int, const volatile unsigned short *);
21689 vector unsigned char vec_lvsr (int, const volatile short *);
21690 vector unsigned char vec_lvsr (int, const volatile unsigned int *);
21691 vector unsigned char vec_lvsr (int, const volatile int *);
21692 vector unsigned char vec_lvsr (int, const volatile unsigned long *);
21693 vector unsigned char vec_lvsr (int, const volatile long *);
21694 vector unsigned char vec_lvsr (int, const volatile float *);
21696 vector float vec_madd (vector float, vector float, vector float);
21698 vector signed short vec_madds (vector signed short,
21699 vector signed short,
21700 vector signed short);
21702 vector unsigned char vec_max (vector bool char, vector unsigned char);
21703 vector unsigned char vec_max (vector unsigned char, vector bool char);
21704 vector unsigned char vec_max (vector unsigned char,
21705 vector unsigned char);
21706 vector signed char vec_max (vector bool char, vector signed char);
21707 vector signed char vec_max (vector signed char, vector bool char);
21708 vector signed char vec_max (vector signed char, vector signed char);
21709 vector unsigned short vec_max (vector bool short,
21710 vector unsigned short);
21711 vector unsigned short vec_max (vector unsigned short,
21712 vector bool short);
21713 vector unsigned short vec_max (vector unsigned short,
21714 vector unsigned short);
21715 vector signed short vec_max (vector bool short, vector signed short);
21716 vector signed short vec_max (vector signed short, vector bool short);
21717 vector signed short vec_max (vector signed short, vector signed short);
21718 vector unsigned int vec_max (vector bool int, vector unsigned int);
21719 vector unsigned int vec_max (vector unsigned int, vector bool int);
21720 vector unsigned int vec_max (vector unsigned int, vector unsigned int);
21721 vector signed int vec_max (vector bool int, vector signed int);
21722 vector signed int vec_max (vector signed int, vector bool int);
21723 vector signed int vec_max (vector signed int, vector signed int);
21724 vector float vec_max (vector float, vector float);
21726 vector float vec_vmaxfp (vector float, vector float);
21728 vector signed int vec_vmaxsw (vector bool int, vector signed int);
21729 vector signed int vec_vmaxsw (vector signed int, vector bool int);
21730 vector signed int vec_vmaxsw (vector signed int, vector signed int);
21732 vector unsigned int vec_vmaxuw (vector bool int, vector unsigned int);
21733 vector unsigned int vec_vmaxuw (vector unsigned int, vector bool int);
21734 vector unsigned int vec_vmaxuw (vector unsigned int,
21735 vector unsigned int);
21737 vector signed short vec_vmaxsh (vector bool short, vector signed short);
21738 vector signed short vec_vmaxsh (vector signed short, vector bool short);
21739 vector signed short vec_vmaxsh (vector signed short,
21740 vector signed short);
21742 vector unsigned short vec_vmaxuh (vector bool short,
21743 vector unsigned short);
21744 vector unsigned short vec_vmaxuh (vector unsigned short,
21745 vector bool short);
21746 vector unsigned short vec_vmaxuh (vector unsigned short,
21747 vector unsigned short);
21749 vector signed char vec_vmaxsb (vector bool char, vector signed char);
21750 vector signed char vec_vmaxsb (vector signed char, vector bool char);
21751 vector signed char vec_vmaxsb (vector signed char, vector signed char);
21753 vector unsigned char vec_vmaxub (vector bool char,
21754 vector unsigned char);
21755 vector unsigned char vec_vmaxub (vector unsigned char,
21757 vector unsigned char vec_vmaxub (vector unsigned char,
21758 vector unsigned char);
21760 vector bool char vec_mergeh (vector bool char, vector bool char);
21761 vector signed char vec_mergeh (vector signed char, vector signed char);
21762 vector unsigned char vec_mergeh (vector unsigned char,
21763 vector unsigned char);
21764 vector bool short vec_mergeh (vector bool short, vector bool short);
21765 vector pixel vec_mergeh (vector pixel, vector pixel);
21766 vector signed short vec_mergeh (vector signed short,
21767 vector signed short);
21768 vector unsigned short vec_mergeh (vector unsigned short,
21769 vector unsigned short);
21770 vector float vec_mergeh (vector float, vector float);
21771 vector bool int vec_mergeh (vector bool int, vector bool int);
21772 vector signed int vec_mergeh (vector signed int, vector signed int);
21773 vector unsigned int vec_mergeh (vector unsigned int,
21774 vector unsigned int);
21776 vector float vec_vmrghw (vector float, vector float);
21777 vector bool int vec_vmrghw (vector bool int, vector bool int);
21778 vector signed int vec_vmrghw (vector signed int, vector signed int);
21779 vector unsigned int vec_vmrghw (vector unsigned int,
21780 vector unsigned int);
21782 vector bool short vec_vmrghh (vector bool short, vector bool short);
21783 vector signed short vec_vmrghh (vector signed short,
21784 vector signed short);
21785 vector unsigned short vec_vmrghh (vector unsigned short,
21786 vector unsigned short);
21787 vector pixel vec_vmrghh (vector pixel, vector pixel);
21789 vector bool char vec_vmrghb (vector bool char, vector bool char);
21790 vector signed char vec_vmrghb (vector signed char, vector signed char);
21791 vector unsigned char vec_vmrghb (vector unsigned char,
21792 vector unsigned char);
21794 vector bool char vec_mergel (vector bool char, vector bool char);
21795 vector signed char vec_mergel (vector signed char, vector signed char);
21796 vector unsigned char vec_mergel (vector unsigned char,
21797 vector unsigned char);
21798 vector bool short vec_mergel (vector bool short, vector bool short);
21799 vector pixel vec_mergel (vector pixel, vector pixel);
21800 vector signed short vec_mergel (vector signed short,
21801 vector signed short);
21802 vector unsigned short vec_mergel (vector unsigned short,
21803 vector unsigned short);
21804 vector float vec_mergel (vector float, vector float);
21805 vector bool int vec_mergel (vector bool int, vector bool int);
21806 vector signed int vec_mergel (vector signed int, vector signed int);
21807 vector unsigned int vec_mergel (vector unsigned int,
21808 vector unsigned int);
21810 vector float vec_vmrglw (vector float, vector float);
21811 vector signed int vec_vmrglw (vector signed int, vector signed int);
21812 vector unsigned int vec_vmrglw (vector unsigned int,
21813 vector unsigned int);
21814 vector bool int vec_vmrglw (vector bool int, vector bool int);
21816 vector bool short vec_vmrglh (vector bool short, vector bool short);
21817 vector signed short vec_vmrglh (vector signed short,
21818 vector signed short);
21819 vector unsigned short vec_vmrglh (vector unsigned short,
21820 vector unsigned short);
21821 vector pixel vec_vmrglh (vector pixel, vector pixel);
21823 vector bool char vec_vmrglb (vector bool char, vector bool char);
21824 vector signed char vec_vmrglb (vector signed char, vector signed char);
21825 vector unsigned char vec_vmrglb (vector unsigned char,
21826 vector unsigned char);
21828 vector unsigned short vec_mfvscr (void);
21830 vector unsigned char vec_min (vector bool char, vector unsigned char);
21831 vector unsigned char vec_min (vector unsigned char, vector bool char);
21832 vector unsigned char vec_min (vector unsigned char,
21833 vector unsigned char);
21834 vector signed char vec_min (vector bool char, vector signed char);
21835 vector signed char vec_min (vector signed char, vector bool char);
21836 vector signed char vec_min (vector signed char, vector signed char);
21837 vector unsigned short vec_min (vector bool short,
21838 vector unsigned short);
21839 vector unsigned short vec_min (vector unsigned short,
21840 vector bool short);
21841 vector unsigned short vec_min (vector unsigned short,
21842 vector unsigned short);
21843 vector signed short vec_min (vector bool short, vector signed short);
21844 vector signed short vec_min (vector signed short, vector bool short);
21845 vector signed short vec_min (vector signed short, vector signed short);
21846 vector unsigned int vec_min (vector bool int, vector unsigned int);
21847 vector unsigned int vec_min (vector unsigned int, vector bool int);
21848 vector unsigned int vec_min (vector unsigned int, vector unsigned int);
21849 vector signed int vec_min (vector bool int, vector signed int);
21850 vector signed int vec_min (vector signed int, vector bool int);
21851 vector signed int vec_min (vector signed int, vector signed int);
21852 vector float vec_min (vector float, vector float);
21854 vector float vec_vminfp (vector float, vector float);
21856 vector signed int vec_vminsw (vector bool int, vector signed int);
21857 vector signed int vec_vminsw (vector signed int, vector bool int);
21858 vector signed int vec_vminsw (vector signed int, vector signed int);
21860 vector unsigned int vec_vminuw (vector bool int, vector unsigned int);
21861 vector unsigned int vec_vminuw (vector unsigned int, vector bool int);
21862 vector unsigned int vec_vminuw (vector unsigned int,
21863 vector unsigned int);
21865 vector signed short vec_vminsh (vector bool short, vector signed short);
21866 vector signed short vec_vminsh (vector signed short, vector bool short);
21867 vector signed short vec_vminsh (vector signed short,
21868 vector signed short);
21870 vector unsigned short vec_vminuh (vector bool short,
21871 vector unsigned short);
21872 vector unsigned short vec_vminuh (vector unsigned short,
21873 vector bool short);
21874 vector unsigned short vec_vminuh (vector unsigned short,
21875 vector unsigned short);
21877 vector signed char vec_vminsb (vector bool char, vector signed char);
21878 vector signed char vec_vminsb (vector signed char, vector bool char);
21879 vector signed char vec_vminsb (vector signed char, vector signed char);
21881 vector unsigned char vec_vminub (vector bool char,
21882 vector unsigned char);
21883 vector unsigned char vec_vminub (vector unsigned char,
21885 vector unsigned char vec_vminub (vector unsigned char,
21886 vector unsigned char);
21888 vector signed short vec_mladd (vector signed short,
21889 vector signed short,
21890 vector signed short);
21891 vector signed short vec_mladd (vector signed short,
21892 vector unsigned short,
21893 vector unsigned short);
21894 vector signed short vec_mladd (vector unsigned short,
21895 vector signed short,
21896 vector signed short);
21897 vector unsigned short vec_mladd (vector unsigned short,
21898 vector unsigned short,
21899 vector unsigned short);
21901 vector signed short vec_mradds (vector signed short,
21902 vector signed short,
21903 vector signed short);
21905 vector unsigned int vec_msum (vector unsigned char,
21906 vector unsigned char,
21907 vector unsigned int);
21908 vector signed int vec_msum (vector signed char,
21909 vector unsigned char,
21910 vector signed int);
21911 vector unsigned int vec_msum (vector unsigned short,
21912 vector unsigned short,
21913 vector unsigned int);
21914 vector signed int vec_msum (vector signed short,
21915 vector signed short,
21916 vector signed int);
21918 vector signed int vec_vmsumshm (vector signed short,
21919 vector signed short,
21920 vector signed int);
21922 vector unsigned int vec_vmsumuhm (vector unsigned short,
21923 vector unsigned short,
21924 vector unsigned int);
21926 vector signed int vec_vmsummbm (vector signed char,
21927 vector unsigned char,
21928 vector signed int);
21930 vector unsigned int vec_vmsumubm (vector unsigned char,
21931 vector unsigned char,
21932 vector unsigned int);
21934 vector unsigned int vec_msums (vector unsigned short,
21935 vector unsigned short,
21936 vector unsigned int);
21937 vector signed int vec_msums (vector signed short,
21938 vector signed short,
21939 vector signed int);
21941 vector signed int vec_vmsumshs (vector signed short,
21942 vector signed short,
21943 vector signed int);
21945 vector unsigned int vec_vmsumuhs (vector unsigned short,
21946 vector unsigned short,
21947 vector unsigned int);
21949 void vec_mtvscr (vector signed int);
21950 void vec_mtvscr (vector unsigned int);
21951 void vec_mtvscr (vector bool int);
21952 void vec_mtvscr (vector signed short);
21953 void vec_mtvscr (vector unsigned short);
21954 void vec_mtvscr (vector bool short);
21955 void vec_mtvscr (vector pixel);
21956 void vec_mtvscr (vector signed char);
21957 void vec_mtvscr (vector unsigned char);
21958 void vec_mtvscr (vector bool char);
21960 vector unsigned short vec_mule (vector unsigned char,
21961 vector unsigned char);
21962 vector signed short vec_mule (vector signed char,
21963 vector signed char);
21964 vector unsigned int vec_mule (vector unsigned short,
21965 vector unsigned short);
21966 vector signed int vec_mule (vector signed short, vector signed short);
21968 vector signed int vec_vmulesh (vector signed short,
21969 vector signed short);
21971 vector unsigned int vec_vmuleuh (vector unsigned short,
21972 vector unsigned short);
21974 vector signed short vec_vmulesb (vector signed char,
21975 vector signed char);
21977 vector unsigned short vec_vmuleub (vector unsigned char,
21978 vector unsigned char);
21980 vector unsigned short vec_mulo (vector unsigned char,
21981 vector unsigned char);
21982 vector signed short vec_mulo (vector signed char, vector signed char);
21983 vector unsigned int vec_mulo (vector unsigned short,
21984 vector unsigned short);
21985 vector signed int vec_mulo (vector signed short, vector signed short);
21987 vector signed int vec_vmulosh (vector signed short,
21988 vector signed short);
21990 vector unsigned int vec_vmulouh (vector unsigned short,
21991 vector unsigned short);
21993 vector signed short vec_vmulosb (vector signed char,
21994 vector signed char);
21996 vector unsigned short vec_vmuloub (vector unsigned char,
21997 vector unsigned char);
21999 vector float vec_nmsub (vector float, vector float, vector float);
22001 vector float vec_nor (vector float, vector float);
22002 vector signed int vec_nor (vector signed int, vector signed int);
22003 vector unsigned int vec_nor (vector unsigned int, vector unsigned int);
22004 vector bool int vec_nor (vector bool int, vector bool int);
22005 vector signed short vec_nor (vector signed short, vector signed short);
22006 vector unsigned short vec_nor (vector unsigned short,
22007 vector unsigned short);
22008 vector bool short vec_nor (vector bool short, vector bool short);
22009 vector signed char vec_nor (vector signed char, vector signed char);
22010 vector unsigned char vec_nor (vector unsigned char,
22011 vector unsigned char);
22012 vector bool char vec_nor (vector bool char, vector bool char);
22014 vector float vec_or (vector float, vector float);
22015 vector float vec_or (vector float, vector bool int);
22016 vector float vec_or (vector bool int, vector float);
22017 vector bool int vec_or (vector bool int, vector bool int);
22018 vector signed int vec_or (vector bool int, vector signed int);
22019 vector signed int vec_or (vector signed int, vector bool int);
22020 vector signed int vec_or (vector signed int, vector signed int);
22021 vector unsigned int vec_or (vector bool int, vector unsigned int);
22022 vector unsigned int vec_or (vector unsigned int, vector bool int);
22023 vector unsigned int vec_or (vector unsigned int, vector unsigned int);
22024 vector bool short vec_or (vector bool short, vector bool short);
22025 vector signed short vec_or (vector bool short, vector signed short);
22026 vector signed short vec_or (vector signed short, vector bool short);
22027 vector signed short vec_or (vector signed short, vector signed short);
22028 vector unsigned short vec_or (vector bool short, vector unsigned short);
22029 vector unsigned short vec_or (vector unsigned short, vector bool short);
22030 vector unsigned short vec_or (vector unsigned short,
22031 vector unsigned short);
22032 vector signed char vec_or (vector bool char, vector signed char);
22033 vector bool char vec_or (vector bool char, vector bool char);
22034 vector signed char vec_or (vector signed char, vector bool char);
22035 vector signed char vec_or (vector signed char, vector signed char);
22036 vector unsigned char vec_or (vector bool char, vector unsigned char);
22037 vector unsigned char vec_or (vector unsigned char, vector bool char);
22038 vector unsigned char vec_or (vector unsigned char,
22039 vector unsigned char);
22041 vector signed char vec_pack (vector signed short, vector signed short);
22042 vector unsigned char vec_pack (vector unsigned short,
22043 vector unsigned short);
22044 vector bool char vec_pack (vector bool short, vector bool short);
22045 vector signed short vec_pack (vector signed int, vector signed int);
22046 vector unsigned short vec_pack (vector unsigned int,
22047 vector unsigned int);
22048 vector bool short vec_pack (vector bool int, vector bool int);
22050 vector bool short vec_vpkuwum (vector bool int, vector bool int);
22051 vector signed short vec_vpkuwum (vector signed int, vector signed int);
22052 vector unsigned short vec_vpkuwum (vector unsigned int,
22053 vector unsigned int);
22055 vector bool char vec_vpkuhum (vector bool short, vector bool short);
22056 vector signed char vec_vpkuhum (vector signed short,
22057 vector signed short);
22058 vector unsigned char vec_vpkuhum (vector unsigned short,
22059 vector unsigned short);
22061 vector pixel vec_packpx (vector unsigned int, vector unsigned int);
22063 vector unsigned char vec_packs (vector unsigned short,
22064 vector unsigned short);
22065 vector signed char vec_packs (vector signed short, vector signed short);
22066 vector unsigned short vec_packs (vector unsigned int,
22067 vector unsigned int);
22068 vector signed short vec_packs (vector signed int, vector signed int);
22070 vector signed short vec_vpkswss (vector signed int, vector signed int);
22072 vector unsigned short vec_vpkuwus (vector unsigned int,
22073 vector unsigned int);
22075 vector signed char vec_vpkshss (vector signed short,
22076 vector signed short);
22078 vector unsigned char vec_vpkuhus (vector unsigned short,
22079 vector unsigned short);
22081 vector unsigned char vec_packsu (vector unsigned short,
22082 vector unsigned short);
22083 vector unsigned char vec_packsu (vector signed short,
22084 vector signed short);
22085 vector unsigned short vec_packsu (vector unsigned int,
22086 vector unsigned int);
22087 vector unsigned short vec_packsu (vector signed int, vector signed int);
22089 vector unsigned short vec_vpkswus (vector signed int,
22090 vector signed int);
22092 vector unsigned char vec_vpkshus (vector signed short,
22093 vector signed short);
22095 vector float vec_perm (vector float,
22097 vector unsigned char);
22098 vector signed int vec_perm (vector signed int,
22100 vector unsigned char);
22101 vector unsigned int vec_perm (vector unsigned int,
22102 vector unsigned int,
22103 vector unsigned char);
22104 vector bool int vec_perm (vector bool int,
22106 vector unsigned char);
22107 vector signed short vec_perm (vector signed short,
22108 vector signed short,
22109 vector unsigned char);
22110 vector unsigned short vec_perm (vector unsigned short,
22111 vector unsigned short,
22112 vector unsigned char);
22113 vector bool short vec_perm (vector bool short,
22115 vector unsigned char);
22116 vector pixel vec_perm (vector pixel,
22118 vector unsigned char);
22119 vector signed char vec_perm (vector signed char,
22120 vector signed char,
22121 vector unsigned char);
22122 vector unsigned char vec_perm (vector unsigned char,
22123 vector unsigned char,
22124 vector unsigned char);
22125 vector bool char vec_perm (vector bool char,
22127 vector unsigned char);
22129 vector float vec_re (vector float);
22131 vector signed char vec_rl (vector signed char,
22132 vector unsigned char);
22133 vector unsigned char vec_rl (vector unsigned char,
22134 vector unsigned char);
22135 vector signed short vec_rl (vector signed short, vector unsigned short);
22136 vector unsigned short vec_rl (vector unsigned short,
22137 vector unsigned short);
22138 vector signed int vec_rl (vector signed int, vector unsigned int);
22139 vector unsigned int vec_rl (vector unsigned int, vector unsigned int);
22141 vector signed int vec_vrlw (vector signed int, vector unsigned int);
22142 vector unsigned int vec_vrlw (vector unsigned int, vector unsigned int);
22144 vector signed short vec_vrlh (vector signed short,
22145 vector unsigned short);
22146 vector unsigned short vec_vrlh (vector unsigned short,
22147 vector unsigned short);
22149 vector signed char vec_vrlb (vector signed char, vector unsigned char);
22150 vector unsigned char vec_vrlb (vector unsigned char,
22151 vector unsigned char);
22153 vector float vec_round (vector float);
22155 vector float vec_rsqrte (vector float);
22157 vector float vec_sel (vector float, vector float, vector bool int);
22158 vector float vec_sel (vector float, vector float, vector unsigned int);
22159 vector signed int vec_sel (vector signed int,
22162 vector signed int vec_sel (vector signed int,
22164 vector unsigned int);
22165 vector unsigned int vec_sel (vector unsigned int,
22166 vector unsigned int,
22168 vector unsigned int vec_sel (vector unsigned int,
22169 vector unsigned int,
22170 vector unsigned int);
22171 vector bool int vec_sel (vector bool int,
22174 vector bool int vec_sel (vector bool int,
22176 vector unsigned int);
22177 vector signed short vec_sel (vector signed short,
22178 vector signed short,
22179 vector bool short);
22180 vector signed short vec_sel (vector signed short,
22181 vector signed short,
22182 vector unsigned short);
22183 vector unsigned short vec_sel (vector unsigned short,
22184 vector unsigned short,
22185 vector bool short);
22186 vector unsigned short vec_sel (vector unsigned short,
22187 vector unsigned short,
22188 vector unsigned short);
22189 vector bool short vec_sel (vector bool short,
22191 vector bool short);
22192 vector bool short vec_sel (vector bool short,
22194 vector unsigned short);
22195 vector signed char vec_sel (vector signed char,
22196 vector signed char,
22198 vector signed char vec_sel (vector signed char,
22199 vector signed char,
22200 vector unsigned char);
22201 vector unsigned char vec_sel (vector unsigned char,
22202 vector unsigned char,
22204 vector unsigned char vec_sel (vector unsigned char,
22205 vector unsigned char,
22206 vector unsigned char);
22207 vector bool char vec_sel (vector bool char,
22210 vector bool char vec_sel (vector bool char,
22212 vector unsigned char);
22214 vector signed char vec_sl (vector signed char,
22215 vector unsigned char);
22216 vector unsigned char vec_sl (vector unsigned char,
22217 vector unsigned char);
22218 vector signed short vec_sl (vector signed short, vector unsigned short);
22219 vector unsigned short vec_sl (vector unsigned short,
22220 vector unsigned short);
22221 vector signed int vec_sl (vector signed int, vector unsigned int);
22222 vector unsigned int vec_sl (vector unsigned int, vector unsigned int);
22224 vector signed int vec_vslw (vector signed int, vector unsigned int);
22225 vector unsigned int vec_vslw (vector unsigned int, vector unsigned int);
22227 vector signed short vec_vslh (vector signed short,
22228 vector unsigned short);
22229 vector unsigned short vec_vslh (vector unsigned short,
22230 vector unsigned short);
22232 vector signed char vec_vslb (vector signed char, vector unsigned char);
22233 vector unsigned char vec_vslb (vector unsigned char,
22234 vector unsigned char);
22236 vector float vec_sld (vector float, vector float, const int);
22237 vector signed int vec_sld (vector signed int,
22240 vector unsigned int vec_sld (vector unsigned int,
22241 vector unsigned int,
22243 vector bool int vec_sld (vector bool int,
22246 vector signed short vec_sld (vector signed short,
22247 vector signed short,
22249 vector unsigned short vec_sld (vector unsigned short,
22250 vector unsigned short,
22252 vector bool short vec_sld (vector bool short,
22255 vector pixel vec_sld (vector pixel,
22258 vector signed char vec_sld (vector signed char,
22259 vector signed char,
22261 vector unsigned char vec_sld (vector unsigned char,
22262 vector unsigned char,
22264 vector bool char vec_sld (vector bool char,
22268 vector signed int vec_sll (vector signed int,
22269 vector unsigned int);
22270 vector signed int vec_sll (vector signed int,
22271 vector unsigned short);
22272 vector signed int vec_sll (vector signed int,
22273 vector unsigned char);
22274 vector unsigned int vec_sll (vector unsigned int,
22275 vector unsigned int);
22276 vector unsigned int vec_sll (vector unsigned int,
22277 vector unsigned short);
22278 vector unsigned int vec_sll (vector unsigned int,
22279 vector unsigned char);
22280 vector bool int vec_sll (vector bool int,
22281 vector unsigned int);
22282 vector bool int vec_sll (vector bool int,
22283 vector unsigned short);
22284 vector bool int vec_sll (vector bool int,
22285 vector unsigned char);
22286 vector signed short vec_sll (vector signed short,
22287 vector unsigned int);
22288 vector signed short vec_sll (vector signed short,
22289 vector unsigned short);
22290 vector signed short vec_sll (vector signed short,
22291 vector unsigned char);
22292 vector unsigned short vec_sll (vector unsigned short,
22293 vector unsigned int);
22294 vector unsigned short vec_sll (vector unsigned short,
22295 vector unsigned short);
22296 vector unsigned short vec_sll (vector unsigned short,
22297 vector unsigned char);
22298 vector bool short vec_sll (vector bool short, vector unsigned int);
22299 vector bool short vec_sll (vector bool short, vector unsigned short);
22300 vector bool short vec_sll (vector bool short, vector unsigned char);
22301 vector pixel vec_sll (vector pixel, vector unsigned int);
22302 vector pixel vec_sll (vector pixel, vector unsigned short);
22303 vector pixel vec_sll (vector pixel, vector unsigned char);
22304 vector signed char vec_sll (vector signed char, vector unsigned int);
22305 vector signed char vec_sll (vector signed char, vector unsigned short);
22306 vector signed char vec_sll (vector signed char, vector unsigned char);
22307 vector unsigned char vec_sll (vector unsigned char,
22308 vector unsigned int);
22309 vector unsigned char vec_sll (vector unsigned char,
22310 vector unsigned short);
22311 vector unsigned char vec_sll (vector unsigned char,
22312 vector unsigned char);
22313 vector bool char vec_sll (vector bool char, vector unsigned int);
22314 vector bool char vec_sll (vector bool char, vector unsigned short);
22315 vector bool char vec_sll (vector bool char, vector unsigned char);
22317 vector float vec_slo (vector float, vector signed char);
22318 vector float vec_slo (vector float, vector unsigned char);
22319 vector signed int vec_slo (vector signed int, vector signed char);
22320 vector signed int vec_slo (vector signed int, vector unsigned char);
22321 vector unsigned int vec_slo (vector unsigned int, vector signed char);
22322 vector unsigned int vec_slo (vector unsigned int, vector unsigned char);
22323 vector signed short vec_slo (vector signed short, vector signed char);
22324 vector signed short vec_slo (vector signed short, vector unsigned char);
22325 vector unsigned short vec_slo (vector unsigned short,
22326 vector signed char);
22327 vector unsigned short vec_slo (vector unsigned short,
22328 vector unsigned char);
22329 vector pixel vec_slo (vector pixel, vector signed char);
22330 vector pixel vec_slo (vector pixel, vector unsigned char);
22331 vector signed char vec_slo (vector signed char, vector signed char);
22332 vector signed char vec_slo (vector signed char, vector unsigned char);
22333 vector unsigned char vec_slo (vector unsigned char, vector signed char);
22334 vector unsigned char vec_slo (vector unsigned char,
22335 vector unsigned char);
22337 vector signed char vec_splat (vector signed char, const int);
22338 vector unsigned char vec_splat (vector unsigned char, const int);
22339 vector bool char vec_splat (vector bool char, const int);
22340 vector signed short vec_splat (vector signed short, const int);
22341 vector unsigned short vec_splat (vector unsigned short, const int);
22342 vector bool short vec_splat (vector bool short, const int);
22343 vector pixel vec_splat (vector pixel, const int);
22344 vector float vec_splat (vector float, const int);
22345 vector signed int vec_splat (vector signed int, const int);
22346 vector unsigned int vec_splat (vector unsigned int, const int);
22347 vector bool int vec_splat (vector bool int, const int);
22349 vector float vec_vspltw (vector float, const int);
22350 vector signed int vec_vspltw (vector signed int, const int);
22351 vector unsigned int vec_vspltw (vector unsigned int, const int);
22352 vector bool int vec_vspltw (vector bool int, const int);
22354 vector bool short vec_vsplth (vector bool short, const int);
22355 vector signed short vec_vsplth (vector signed short, const int);
22356 vector unsigned short vec_vsplth (vector unsigned short, const int);
22357 vector pixel vec_vsplth (vector pixel, const int);
22359 vector signed char vec_vspltb (vector signed char, const int);
22360 vector unsigned char vec_vspltb (vector unsigned char, const int);
22361 vector bool char vec_vspltb (vector bool char, const int);
22363 vector signed char vec_splat_s8 (const int);
22365 vector signed short vec_splat_s16 (const int);
22367 vector signed int vec_splat_s32 (const int);
22369 vector unsigned char vec_splat_u8 (const int);
22371 vector unsigned short vec_splat_u16 (const int);
22373 vector unsigned int vec_splat_u32 (const int);
22375 vector signed char vec_sr (vector signed char, vector unsigned char);
22376 vector unsigned char vec_sr (vector unsigned char,
22377 vector unsigned char);
22378 vector signed short vec_sr (vector signed short,
22379 vector unsigned short);
22380 vector unsigned short vec_sr (vector unsigned short,
22381 vector unsigned short);
22382 vector signed int vec_sr (vector signed int, vector unsigned int);
22383 vector unsigned int vec_sr (vector unsigned int, vector unsigned int);
22385 vector signed int vec_vsrw (vector signed int, vector unsigned int);
22386 vector unsigned int vec_vsrw (vector unsigned int, vector unsigned int);
22388 vector signed short vec_vsrh (vector signed short,
22389 vector unsigned short);
22390 vector unsigned short vec_vsrh (vector unsigned short,
22391 vector unsigned short);
22393 vector signed char vec_vsrb (vector signed char, vector unsigned char);
22394 vector unsigned char vec_vsrb (vector unsigned char,
22395 vector unsigned char);
22397 vector signed char vec_sra (vector signed char, vector unsigned char);
22398 vector unsigned char vec_sra (vector unsigned char,
22399 vector unsigned char);
22400 vector signed short vec_sra (vector signed short,
22401 vector unsigned short);
22402 vector unsigned short vec_sra (vector unsigned short,
22403 vector unsigned short);
22404 vector signed int vec_sra (vector signed int, vector unsigned int);
22405 vector unsigned int vec_sra (vector unsigned int, vector unsigned int);
22407 vector signed int vec_vsraw (vector signed int, vector unsigned int);
22408 vector unsigned int vec_vsraw (vector unsigned int,
22409 vector unsigned int);
22411 vector signed short vec_vsrah (vector signed short,
22412 vector unsigned short);
22413 vector unsigned short vec_vsrah (vector unsigned short,
22414 vector unsigned short);
22416 vector signed char vec_vsrab (vector signed char, vector unsigned char);
22417 vector unsigned char vec_vsrab (vector unsigned char,
22418 vector unsigned char);
22420 vector signed int vec_srl (vector signed int, vector unsigned int);
22421 vector signed int vec_srl (vector signed int, vector unsigned short);
22422 vector signed int vec_srl (vector signed int, vector unsigned char);
22423 vector unsigned int vec_srl (vector unsigned int, vector unsigned int);
22424 vector unsigned int vec_srl (vector unsigned int,
22425 vector unsigned short);
22426 vector unsigned int vec_srl (vector unsigned int, vector unsigned char);
22427 vector bool int vec_srl (vector bool int, vector unsigned int);
22428 vector bool int vec_srl (vector bool int, vector unsigned short);
22429 vector bool int vec_srl (vector bool int, vector unsigned char);
22430 vector signed short vec_srl (vector signed short, vector unsigned int);
22431 vector signed short vec_srl (vector signed short,
22432 vector unsigned short);
22433 vector signed short vec_srl (vector signed short, vector unsigned char);
22434 vector unsigned short vec_srl (vector unsigned short,
22435 vector unsigned int);
22436 vector unsigned short vec_srl (vector unsigned short,
22437 vector unsigned short);
22438 vector unsigned short vec_srl (vector unsigned short,
22439 vector unsigned char);
22440 vector bool short vec_srl (vector bool short, vector unsigned int);
22441 vector bool short vec_srl (vector bool short, vector unsigned short);
22442 vector bool short vec_srl (vector bool short, vector unsigned char);
22443 vector pixel vec_srl (vector pixel, vector unsigned int);
22444 vector pixel vec_srl (vector pixel, vector unsigned short);
22445 vector pixel vec_srl (vector pixel, vector unsigned char);
22446 vector signed char vec_srl (vector signed char, vector unsigned int);
22447 vector signed char vec_srl (vector signed char, vector unsigned short);
22448 vector signed char vec_srl (vector signed char, vector unsigned char);
22449 vector unsigned char vec_srl (vector unsigned char,
22450 vector unsigned int);
22451 vector unsigned char vec_srl (vector unsigned char,
22452 vector unsigned short);
22453 vector unsigned char vec_srl (vector unsigned char,
22454 vector unsigned char);
22455 vector bool char vec_srl (vector bool char, vector unsigned int);
22456 vector bool char vec_srl (vector bool char, vector unsigned short);
22457 vector bool char vec_srl (vector bool char, vector unsigned char);
22459 vector float vec_sro (vector float, vector signed char);
22460 vector float vec_sro (vector float, vector unsigned char);
22461 vector signed int vec_sro (vector signed int, vector signed char);
22462 vector signed int vec_sro (vector signed int, vector unsigned char);
22463 vector unsigned int vec_sro (vector unsigned int, vector signed char);
22464 vector unsigned int vec_sro (vector unsigned int, vector unsigned char);
22465 vector signed short vec_sro (vector signed short, vector signed char);
22466 vector signed short vec_sro (vector signed short, vector unsigned char);
22467 vector unsigned short vec_sro (vector unsigned short,
22468 vector signed char);
22469 vector unsigned short vec_sro (vector unsigned short,
22470 vector unsigned char);
22471 vector pixel vec_sro (vector pixel, vector signed char);
22472 vector pixel vec_sro (vector pixel, vector unsigned char);
22473 vector signed char vec_sro (vector signed char, vector signed char);
22474 vector signed char vec_sro (vector signed char, vector unsigned char);
22475 vector unsigned char vec_sro (vector unsigned char, vector signed char);
22476 vector unsigned char vec_sro (vector unsigned char,
22477 vector unsigned char);
22479 void vec_st (vector float, int, vector float *);
22480 void vec_st (vector float, int, float *);
22481 void vec_st (vector signed int, int, vector signed int *);
22482 void vec_st (vector signed int, int, int *);
22483 void vec_st (vector unsigned int, int, vector unsigned int *);
22484 void vec_st (vector unsigned int, int, unsigned int *);
22485 void vec_st (vector bool int, int, vector bool int *);
22486 void vec_st (vector bool int, int, unsigned int *);
22487 void vec_st (vector bool int, int, int *);
22488 void vec_st (vector signed short, int, vector signed short *);
22489 void vec_st (vector signed short, int, short *);
22490 void vec_st (vector unsigned short, int, vector unsigned short *);
22491 void vec_st (vector unsigned short, int, unsigned short *);
22492 void vec_st (vector bool short, int, vector bool short *);
22493 void vec_st (vector bool short, int, unsigned short *);
22494 void vec_st (vector pixel, int, vector pixel *);
22495 void vec_st (vector pixel, int, unsigned short *);
22496 void vec_st (vector pixel, int, short *);
22497 void vec_st (vector bool short, int, short *);
22498 void vec_st (vector signed char, int, vector signed char *);
22499 void vec_st (vector signed char, int, signed char *);
22500 void vec_st (vector unsigned char, int, vector unsigned char *);
22501 void vec_st (vector unsigned char, int, unsigned char *);
22502 void vec_st (vector bool char, int, vector bool char *);
22503 void vec_st (vector bool char, int, unsigned char *);
22504 void vec_st (vector bool char, int, signed char *);
22506 void vec_ste (vector signed char, int, signed char *);
22507 void vec_ste (vector unsigned char, int, unsigned char *);
22508 void vec_ste (vector bool char, int, signed char *);
22509 void vec_ste (vector bool char, int, unsigned char *);
22510 void vec_ste (vector signed short, int, short *);
22511 void vec_ste (vector unsigned short, int, unsigned short *);
22512 void vec_ste (vector bool short, int, short *);
22513 void vec_ste (vector bool short, int, unsigned short *);
22514 void vec_ste (vector pixel, int, short *);
22515 void vec_ste (vector pixel, int, unsigned short *);
22516 void vec_ste (vector float, int, float *);
22517 void vec_ste (vector signed int, int, int *);
22518 void vec_ste (vector unsigned int, int, unsigned int *);
22519 void vec_ste (vector bool int, int, int *);
22520 void vec_ste (vector bool int, int, unsigned int *);
22522 void vec_stvewx (vector float, int, float *);
22523 void vec_stvewx (vector signed int, int, int *);
22524 void vec_stvewx (vector unsigned int, int, unsigned int *);
22525 void vec_stvewx (vector bool int, int, int *);
22526 void vec_stvewx (vector bool int, int, unsigned int *);
22528 void vec_stvehx (vector signed short, int, short *);
22529 void vec_stvehx (vector unsigned short, int, unsigned short *);
22530 void vec_stvehx (vector bool short, int, short *);
22531 void vec_stvehx (vector bool short, int, unsigned short *);
22532 void vec_stvehx (vector pixel, int, short *);
22533 void vec_stvehx (vector pixel, int, unsigned short *);
22535 void vec_stvebx (vector signed char, int, signed char *);
22536 void vec_stvebx (vector unsigned char, int, unsigned char *);
22537 void vec_stvebx (vector bool char, int, signed char *);
22538 void vec_stvebx (vector bool char, int, unsigned char *);
22540 void vec_stl (vector float, int, vector float *);
22541 void vec_stl (vector float, int, float *);
22542 void vec_stl (vector signed int, int, vector signed int *);
22543 void vec_stl (vector signed int, int, int *);
22544 void vec_stl (vector unsigned int, int, vector unsigned int *);
22545 void vec_stl (vector unsigned int, int, unsigned int *);
22546 void vec_stl (vector bool int, int, vector bool int *);
22547 void vec_stl (vector bool int, int, unsigned int *);
22548 void vec_stl (vector bool int, int, int *);
22549 void vec_stl (vector signed short, int, vector signed short *);
22550 void vec_stl (vector signed short, int, short *);
22551 void vec_stl (vector unsigned short, int, vector unsigned short *);
22552 void vec_stl (vector unsigned short, int, unsigned short *);
22553 void vec_stl (vector bool short, int, vector bool short *);
22554 void vec_stl (vector bool short, int, unsigned short *);
22555 void vec_stl (vector bool short, int, short *);
22556 void vec_stl (vector pixel, int, vector pixel *);
22557 void vec_stl (vector pixel, int, unsigned short *);
22558 void vec_stl (vector pixel, int, short *);
22559 void vec_stl (vector signed char, int, vector signed char *);
22560 void vec_stl (vector signed char, int, signed char *);
22561 void vec_stl (vector unsigned char, int, vector unsigned char *);
22562 void vec_stl (vector unsigned char, int, unsigned char *);
22563 void vec_stl (vector bool char, int, vector bool char *);
22564 void vec_stl (vector bool char, int, unsigned char *);
22565 void vec_stl (vector bool char, int, signed char *);
22567 vector signed char vec_sub (vector bool char, vector signed char);
22568 vector signed char vec_sub (vector signed char, vector bool char);
22569 vector signed char vec_sub (vector signed char, vector signed char);
22570 vector unsigned char vec_sub (vector bool char, vector unsigned char);
22571 vector unsigned char vec_sub (vector unsigned char, vector bool char);
22572 vector unsigned char vec_sub (vector unsigned char,
22573 vector unsigned char);
22574 vector signed short vec_sub (vector bool short, vector signed short);
22575 vector signed short vec_sub (vector signed short, vector bool short);
22576 vector signed short vec_sub (vector signed short, vector signed short);
22577 vector unsigned short vec_sub (vector bool short,
22578 vector unsigned short);
22579 vector unsigned short vec_sub (vector unsigned short,
22580 vector bool short);
22581 vector unsigned short vec_sub (vector unsigned short,
22582 vector unsigned short);
22583 vector signed int vec_sub (vector bool int, vector signed int);
22584 vector signed int vec_sub (vector signed int, vector bool int);
22585 vector signed int vec_sub (vector signed int, vector signed int);
22586 vector unsigned int vec_sub (vector bool int, vector unsigned int);
22587 vector unsigned int vec_sub (vector unsigned int, vector bool int);
22588 vector unsigned int vec_sub (vector unsigned int, vector unsigned int);
22589 vector float vec_sub (vector float, vector float);
22591 vector float vec_vsubfp (vector float, vector float);
22593 vector signed int vec_vsubuwm (vector bool int, vector signed int);
22594 vector signed int vec_vsubuwm (vector signed int, vector bool int);
22595 vector signed int vec_vsubuwm (vector signed int, vector signed int);
22596 vector unsigned int vec_vsubuwm (vector bool int, vector unsigned int);
22597 vector unsigned int vec_vsubuwm (vector unsigned int, vector bool int);
22598 vector unsigned int vec_vsubuwm (vector unsigned int,
22599 vector unsigned int);
22601 vector signed short vec_vsubuhm (vector bool short,
22602 vector signed short);
22603 vector signed short vec_vsubuhm (vector signed short,
22604 vector bool short);
22605 vector signed short vec_vsubuhm (vector signed short,
22606 vector signed short);
22607 vector unsigned short vec_vsubuhm (vector bool short,
22608 vector unsigned short);
22609 vector unsigned short vec_vsubuhm (vector unsigned short,
22610 vector bool short);
22611 vector unsigned short vec_vsubuhm (vector unsigned short,
22612 vector unsigned short);
22614 vector signed char vec_vsububm (vector bool char, vector signed char);
22615 vector signed char vec_vsububm (vector signed char, vector bool char);
22616 vector signed char vec_vsububm (vector signed char, vector signed char);
22617 vector unsigned char vec_vsububm (vector bool char,
22618 vector unsigned char);
22619 vector unsigned char vec_vsububm (vector unsigned char,
22621 vector unsigned char vec_vsububm (vector unsigned char,
22622 vector unsigned char);
22624 vector unsigned int vec_subc (vector unsigned int, vector unsigned int);
22626 vector unsigned char vec_subs (vector bool char, vector unsigned char);
22627 vector unsigned char vec_subs (vector unsigned char, vector bool char);
22628 vector unsigned char vec_subs (vector unsigned char,
22629 vector unsigned char);
22630 vector signed char vec_subs (vector bool char, vector signed char);
22631 vector signed char vec_subs (vector signed char, vector bool char);
22632 vector signed char vec_subs (vector signed char, vector signed char);
22633 vector unsigned short vec_subs (vector bool short,
22634 vector unsigned short);
22635 vector unsigned short vec_subs (vector unsigned short,
22636 vector bool short);
22637 vector unsigned short vec_subs (vector unsigned short,
22638 vector unsigned short);
22639 vector signed short vec_subs (vector bool short, vector signed short);
22640 vector signed short vec_subs (vector signed short, vector bool short);
22641 vector signed short vec_subs (vector signed short, vector signed short);
22642 vector unsigned int vec_subs (vector bool int, vector unsigned int);
22643 vector unsigned int vec_subs (vector unsigned int, vector bool int);
22644 vector unsigned int vec_subs (vector unsigned int, vector unsigned int);
22645 vector signed int vec_subs (vector bool int, vector signed int);
22646 vector signed int vec_subs (vector signed int, vector bool int);
22647 vector signed int vec_subs (vector signed int, vector signed int);
22649 vector signed int vec_vsubsws (vector bool int, vector signed int);
22650 vector signed int vec_vsubsws (vector signed int, vector bool int);
22651 vector signed int vec_vsubsws (vector signed int, vector signed int);
22653 vector unsigned int vec_vsubuws (vector bool int, vector unsigned int);
22654 vector unsigned int vec_vsubuws (vector unsigned int, vector bool int);
22655 vector unsigned int vec_vsubuws (vector unsigned int,
22656 vector unsigned int);
22658 vector signed short vec_vsubshs (vector bool short,
22659 vector signed short);
22660 vector signed short vec_vsubshs (vector signed short,
22661 vector bool short);
22662 vector signed short vec_vsubshs (vector signed short,
22663 vector signed short);
22665 vector unsigned short vec_vsubuhs (vector bool short,
22666 vector unsigned short);
22667 vector unsigned short vec_vsubuhs (vector unsigned short,
22668 vector bool short);
22669 vector unsigned short vec_vsubuhs (vector unsigned short,
22670 vector unsigned short);
22672 vector signed char vec_vsubsbs (vector bool char, vector signed char);
22673 vector signed char vec_vsubsbs (vector signed char, vector bool char);
22674 vector signed char vec_vsubsbs (vector signed char, vector signed char);
22676 vector unsigned char vec_vsububs (vector bool char,
22677 vector unsigned char);
22678 vector unsigned char vec_vsububs (vector unsigned char,
22680 vector unsigned char vec_vsububs (vector unsigned char,
22681 vector unsigned char);
22683 vector unsigned int vec_sum4s (vector unsigned char,
22684 vector unsigned int);
22685 vector signed int vec_sum4s (vector signed char, vector signed int);
22686 vector signed int vec_sum4s (vector signed short, vector signed int);
22688 vector signed int vec_vsum4shs (vector signed short, vector signed int);
22690 vector signed int vec_vsum4sbs (vector signed char, vector signed int);
22692 vector unsigned int vec_vsum4ubs (vector unsigned char,
22693 vector unsigned int);
22695 vector signed int vec_sum2s (vector signed int, vector signed int);
22697 vector signed int vec_sums (vector signed int, vector signed int);
22699 vector float vec_trunc (vector float);
22701 vector signed short vec_unpackh (vector signed char);
22702 vector bool short vec_unpackh (vector bool char);
22703 vector signed int vec_unpackh (vector signed short);
22704 vector bool int vec_unpackh (vector bool short);
22705 vector unsigned int vec_unpackh (vector pixel);
22707 vector bool int vec_vupkhsh (vector bool short);
22708 vector signed int vec_vupkhsh (vector signed short);
22710 vector unsigned int vec_vupkhpx (vector pixel);
22712 vector bool short vec_vupkhsb (vector bool char);
22713 vector signed short vec_vupkhsb (vector signed char);
22715 vector signed short vec_unpackl (vector signed char);
22716 vector bool short vec_unpackl (vector bool char);
22717 vector unsigned int vec_unpackl (vector pixel);
22718 vector signed int vec_unpackl (vector signed short);
22719 vector bool int vec_unpackl (vector bool short);
22721 vector unsigned int vec_vupklpx (vector pixel);
22723 vector bool int vec_vupklsh (vector bool short);
22724 vector signed int vec_vupklsh (vector signed short);
22726 vector bool short vec_vupklsb (vector bool char);
22727 vector signed short vec_vupklsb (vector signed char);
22729 vector float vec_xor (vector float, vector float);
22730 vector float vec_xor (vector float, vector bool int);
22731 vector float vec_xor (vector bool int, vector float);
22732 vector bool int vec_xor (vector bool int, vector bool int);
22733 vector signed int vec_xor (vector bool int, vector signed int);
22734 vector signed int vec_xor (vector signed int, vector bool int);
22735 vector signed int vec_xor (vector signed int, vector signed int);
22736 vector unsigned int vec_xor (vector bool int, vector unsigned int);
22737 vector unsigned int vec_xor (vector unsigned int, vector bool int);
22738 vector unsigned int vec_xor (vector unsigned int, vector unsigned int);
22739 vector bool short vec_xor (vector bool short, vector bool short);
22740 vector signed short vec_xor (vector bool short, vector signed short);
22741 vector signed short vec_xor (vector signed short, vector bool short);
22742 vector signed short vec_xor (vector signed short, vector signed short);
22743 vector unsigned short vec_xor (vector bool short,
22744 vector unsigned short);
22745 vector unsigned short vec_xor (vector unsigned short,
22746 vector bool short);
22747 vector unsigned short vec_xor (vector unsigned short,
22748 vector unsigned short);
22749 vector signed char vec_xor (vector bool char, vector signed char);
22750 vector bool char vec_xor (vector bool char, vector bool char);
22751 vector signed char vec_xor (vector signed char, vector bool char);
22752 vector signed char vec_xor (vector signed char, vector signed char);
22753 vector unsigned char vec_xor (vector bool char, vector unsigned char);
22754 vector unsigned char vec_xor (vector unsigned char, vector bool char);
22755 vector unsigned char vec_xor (vector unsigned char,
22756 vector unsigned char);
22758 int vec_all_eq (vector signed char, vector bool char);
22759 int vec_all_eq (vector signed char, vector signed char);
22760 int vec_all_eq (vector unsigned char, vector bool char);
22761 int vec_all_eq (vector unsigned char, vector unsigned char);
22762 int vec_all_eq (vector bool char, vector bool char);
22763 int vec_all_eq (vector bool char, vector unsigned char);
22764 int vec_all_eq (vector bool char, vector signed char);
22765 int vec_all_eq (vector signed short, vector bool short);
22766 int vec_all_eq (vector signed short, vector signed short);
22767 int vec_all_eq (vector unsigned short, vector bool short);
22768 int vec_all_eq (vector unsigned short, vector unsigned short);
22769 int vec_all_eq (vector bool short, vector bool short);
22770 int vec_all_eq (vector bool short, vector unsigned short);
22771 int vec_all_eq (vector bool short, vector signed short);
22772 int vec_all_eq (vector pixel, vector pixel);
22773 int vec_all_eq (vector signed int, vector bool int);
22774 int vec_all_eq (vector signed int, vector signed int);
22775 int vec_all_eq (vector unsigned int, vector bool int);
22776 int vec_all_eq (vector unsigned int, vector unsigned int);
22777 int vec_all_eq (vector bool int, vector bool int);
22778 int vec_all_eq (vector bool int, vector unsigned int);
22779 int vec_all_eq (vector bool int, vector signed int);
22780 int vec_all_eq (vector float, vector float);
22782 int vec_all_ge (vector bool char, vector unsigned char);
22783 int vec_all_ge (vector unsigned char, vector bool char);
22784 int vec_all_ge (vector unsigned char, vector unsigned char);
22785 int vec_all_ge (vector bool char, vector signed char);
22786 int vec_all_ge (vector signed char, vector bool char);
22787 int vec_all_ge (vector signed char, vector signed char);
22788 int vec_all_ge (vector bool short, vector unsigned short);
22789 int vec_all_ge (vector unsigned short, vector bool short);
22790 int vec_all_ge (vector unsigned short, vector unsigned short);
22791 int vec_all_ge (vector signed short, vector signed short);
22792 int vec_all_ge (vector bool short, vector signed short);
22793 int vec_all_ge (vector signed short, vector bool short);
22794 int vec_all_ge (vector bool int, vector unsigned int);
22795 int vec_all_ge (vector unsigned int, vector bool int);
22796 int vec_all_ge (vector unsigned int, vector unsigned int);
22797 int vec_all_ge (vector bool int, vector signed int);
22798 int vec_all_ge (vector signed int, vector bool int);
22799 int vec_all_ge (vector signed int, vector signed int);
22800 int vec_all_ge (vector float, vector float);
22802 int vec_all_gt (vector bool char, vector unsigned char);
22803 int vec_all_gt (vector unsigned char, vector bool char);
22804 int vec_all_gt (vector unsigned char, vector unsigned char);
22805 int vec_all_gt (vector bool char, vector signed char);
22806 int vec_all_gt (vector signed char, vector bool char);
22807 int vec_all_gt (vector signed char, vector signed char);
22808 int vec_all_gt (vector bool short, vector unsigned short);
22809 int vec_all_gt (vector unsigned short, vector bool short);
22810 int vec_all_gt (vector unsigned short, vector unsigned short);
22811 int vec_all_gt (vector bool short, vector signed short);
22812 int vec_all_gt (vector signed short, vector bool short);
22813 int vec_all_gt (vector signed short, vector signed short);
22814 int vec_all_gt (vector bool int, vector unsigned int);
22815 int vec_all_gt (vector unsigned int, vector bool int);
22816 int vec_all_gt (vector unsigned int, vector unsigned int);
22817 int vec_all_gt (vector bool int, vector signed int);
22818 int vec_all_gt (vector signed int, vector bool int);
22819 int vec_all_gt (vector signed int, vector signed int);
22820 int vec_all_gt (vector float, vector float);
22822 int vec_all_in (vector float, vector float);
22824 int vec_all_le (vector bool char, vector unsigned char);
22825 int vec_all_le (vector unsigned char, vector bool char);
22826 int vec_all_le (vector unsigned char, vector unsigned char);
22827 int vec_all_le (vector bool char, vector signed char);
22828 int vec_all_le (vector signed char, vector bool char);
22829 int vec_all_le (vector signed char, vector signed char);
22830 int vec_all_le (vector bool short, vector unsigned short);
22831 int vec_all_le (vector unsigned short, vector bool short);
22832 int vec_all_le (vector unsigned short, vector unsigned short);
22833 int vec_all_le (vector bool short, vector signed short);
22834 int vec_all_le (vector signed short, vector bool short);
22835 int vec_all_le (vector signed short, vector signed short);
22836 int vec_all_le (vector bool int, vector unsigned int);
22837 int vec_all_le (vector unsigned int, vector bool int);
22838 int vec_all_le (vector unsigned int, vector unsigned int);
22839 int vec_all_le (vector bool int, vector signed int);
22840 int vec_all_le (vector signed int, vector bool int);
22841 int vec_all_le (vector signed int, vector signed int);
22842 int vec_all_le (vector float, vector float);
22844 int vec_all_lt (vector bool char, vector unsigned char);
22845 int vec_all_lt (vector unsigned char, vector bool char);
22846 int vec_all_lt (vector unsigned char, vector unsigned char);
22847 int vec_all_lt (vector bool char, vector signed char);
22848 int vec_all_lt (vector signed char, vector bool char);
22849 int vec_all_lt (vector signed char, vector signed char);
22850 int vec_all_lt (vector bool short, vector unsigned short);
22851 int vec_all_lt (vector unsigned short, vector bool short);
22852 int vec_all_lt (vector unsigned short, vector unsigned short);
22853 int vec_all_lt (vector bool short, vector signed short);
22854 int vec_all_lt (vector signed short, vector bool short);
22855 int vec_all_lt (vector signed short, vector signed short);
22856 int vec_all_lt (vector bool int, vector unsigned int);
22857 int vec_all_lt (vector unsigned int, vector bool int);
22858 int vec_all_lt (vector unsigned int, vector unsigned int);
22859 int vec_all_lt (vector bool int, vector signed int);
22860 int vec_all_lt (vector signed int, vector bool int);
22861 int vec_all_lt (vector signed int, vector signed int);
22862 int vec_all_lt (vector float, vector float);
22864 int vec_all_nan (vector float);
22866 int vec_all_ne (vector signed char, vector bool char);
22867 int vec_all_ne (vector signed char, vector signed char);
22868 int vec_all_ne (vector unsigned char, vector bool char);
22869 int vec_all_ne (vector unsigned char, vector unsigned char);
22870 int vec_all_ne (vector bool char, vector bool char);
22871 int vec_all_ne (vector bool char, vector unsigned char);
22872 int vec_all_ne (vector bool char, vector signed char);
22873 int vec_all_ne (vector signed short, vector bool short);
22874 int vec_all_ne (vector signed short, vector signed short);
22875 int vec_all_ne (vector unsigned short, vector bool short);
22876 int vec_all_ne (vector unsigned short, vector unsigned short);
22877 int vec_all_ne (vector bool short, vector bool short);
22878 int vec_all_ne (vector bool short, vector unsigned short);
22879 int vec_all_ne (vector bool short, vector signed short);
22880 int vec_all_ne (vector pixel, vector pixel);
22881 int vec_all_ne (vector signed int, vector bool int);
22882 int vec_all_ne (vector signed int, vector signed int);
22883 int vec_all_ne (vector unsigned int, vector bool int);
22884 int vec_all_ne (vector unsigned int, vector unsigned int);
22885 int vec_all_ne (vector bool int, vector bool int);
22886 int vec_all_ne (vector bool int, vector unsigned int);
22887 int vec_all_ne (vector bool int, vector signed int);
22888 int vec_all_ne (vector float, vector float);
22890 int vec_all_nge (vector float, vector float);
22892 int vec_all_ngt (vector float, vector float);
22894 int vec_all_nle (vector float, vector float);
22896 int vec_all_nlt (vector float, vector float);
22898 int vec_all_numeric (vector float);
22900 int vec_any_eq (vector signed char, vector bool char);
22901 int vec_any_eq (vector signed char, vector signed char);
22902 int vec_any_eq (vector unsigned char, vector bool char);
22903 int vec_any_eq (vector unsigned char, vector unsigned char);
22904 int vec_any_eq (vector bool char, vector bool char);
22905 int vec_any_eq (vector bool char, vector unsigned char);
22906 int vec_any_eq (vector bool char, vector signed char);
22907 int vec_any_eq (vector signed short, vector bool short);
22908 int vec_any_eq (vector signed short, vector signed short);
22909 int vec_any_eq (vector unsigned short, vector bool short);
22910 int vec_any_eq (vector unsigned short, vector unsigned short);
22911 int vec_any_eq (vector bool short, vector bool short);
22912 int vec_any_eq (vector bool short, vector unsigned short);
22913 int vec_any_eq (vector bool short, vector signed short);
22914 int vec_any_eq (vector pixel, vector pixel);
22915 int vec_any_eq (vector signed int, vector bool int);
22916 int vec_any_eq (vector signed int, vector signed int);
22917 int vec_any_eq (vector unsigned int, vector bool int);
22918 int vec_any_eq (vector unsigned int, vector unsigned int);
22919 int vec_any_eq (vector bool int, vector bool int);
22920 int vec_any_eq (vector bool int, vector unsigned int);
22921 int vec_any_eq (vector bool int, vector signed int);
22922 int vec_any_eq (vector float, vector float);
22924 int vec_any_ge (vector signed char, vector bool char);
22925 int vec_any_ge (vector unsigned char, vector bool char);
22926 int vec_any_ge (vector unsigned char, vector unsigned char);
22927 int vec_any_ge (vector signed char, vector signed char);
22928 int vec_any_ge (vector bool char, vector unsigned char);
22929 int vec_any_ge (vector bool char, vector signed char);
22930 int vec_any_ge (vector unsigned short, vector bool short);
22931 int vec_any_ge (vector unsigned short, vector unsigned short);
22932 int vec_any_ge (vector signed short, vector signed short);
22933 int vec_any_ge (vector signed short, vector bool short);
22934 int vec_any_ge (vector bool short, vector unsigned short);
22935 int vec_any_ge (vector bool short, vector signed short);
22936 int vec_any_ge (vector signed int, vector bool int);
22937 int vec_any_ge (vector unsigned int, vector bool int);
22938 int vec_any_ge (vector unsigned int, vector unsigned int);
22939 int vec_any_ge (vector signed int, vector signed int);
22940 int vec_any_ge (vector bool int, vector unsigned int);
22941 int vec_any_ge (vector bool int, vector signed int);
22942 int vec_any_ge (vector float, vector float);
22944 int vec_any_gt (vector bool char, vector unsigned char);
22945 int vec_any_gt (vector unsigned char, vector bool char);
22946 int vec_any_gt (vector unsigned char, vector unsigned char);
22947 int vec_any_gt (vector bool char, vector signed char);
22948 int vec_any_gt (vector signed char, vector bool char);
22949 int vec_any_gt (vector signed char, vector signed char);
22950 int vec_any_gt (vector bool short, vector unsigned short);
22951 int vec_any_gt (vector unsigned short, vector bool short);
22952 int vec_any_gt (vector unsigned short, vector unsigned short);
22953 int vec_any_gt (vector bool short, vector signed short);
22954 int vec_any_gt (vector signed short, vector bool short);
22955 int vec_any_gt (vector signed short, vector signed short);
22956 int vec_any_gt (vector bool int, vector unsigned int);
22957 int vec_any_gt (vector unsigned int, vector bool int);
22958 int vec_any_gt (vector unsigned int, vector unsigned int);
22959 int vec_any_gt (vector bool int, vector signed int);
22960 int vec_any_gt (vector signed int, vector bool int);
22961 int vec_any_gt (vector signed int, vector signed int);
22962 int vec_any_gt (vector float, vector float);
22964 int vec_any_le (vector bool char, vector unsigned char);
22965 int vec_any_le (vector unsigned char, vector bool char);
22966 int vec_any_le (vector unsigned char, vector unsigned char);
22967 int vec_any_le (vector bool char, vector signed char);
22968 int vec_any_le (vector signed char, vector bool char);
22969 int vec_any_le (vector signed char, vector signed char);
22970 int vec_any_le (vector bool short, vector unsigned short);
22971 int vec_any_le (vector unsigned short, vector bool short);
22972 int vec_any_le (vector unsigned short, vector unsigned short);
22973 int vec_any_le (vector bool short, vector signed short);
22974 int vec_any_le (vector signed short, vector bool short);
22975 int vec_any_le (vector signed short, vector signed short);
22976 int vec_any_le (vector bool int, vector unsigned int);
22977 int vec_any_le (vector unsigned int, vector bool int);
22978 int vec_any_le (vector unsigned int, vector unsigned int);
22979 int vec_any_le (vector bool int, vector signed int);
22980 int vec_any_le (vector signed int, vector bool int);
22981 int vec_any_le (vector signed int, vector signed int);
22982 int vec_any_le (vector float, vector float);
22984 int vec_any_lt (vector bool char, vector unsigned char);
22985 int vec_any_lt (vector unsigned char, vector bool char);
22986 int vec_any_lt (vector unsigned char, vector unsigned char);
22987 int vec_any_lt (vector bool char, vector signed char);
22988 int vec_any_lt (vector signed char, vector bool char);
22989 int vec_any_lt (vector signed char, vector signed char);
22990 int vec_any_lt (vector bool short, vector unsigned short);
22991 int vec_any_lt (vector unsigned short, vector bool short);
22992 int vec_any_lt (vector unsigned short, vector unsigned short);
22993 int vec_any_lt (vector bool short, vector signed short);
22994 int vec_any_lt (vector signed short, vector bool short);
22995 int vec_any_lt (vector signed short, vector signed short);
22996 int vec_any_lt (vector bool int, vector unsigned int);
22997 int vec_any_lt (vector unsigned int, vector bool int);
22998 int vec_any_lt (vector unsigned int, vector unsigned int);
22999 int vec_any_lt (vector bool int, vector signed int);
23000 int vec_any_lt (vector signed int, vector bool int);
23001 int vec_any_lt (vector signed int, vector signed int);
23002 int vec_any_lt (vector float, vector float);
23004 int vec_any_nan (vector float);
23006 int vec_any_ne (vector signed char, vector bool char);
23007 int vec_any_ne (vector signed char, vector signed char);
23008 int vec_any_ne (vector unsigned char, vector bool char);
23009 int vec_any_ne (vector unsigned char, vector unsigned char);
23010 int vec_any_ne (vector bool char, vector bool char);
23011 int vec_any_ne (vector bool char, vector unsigned char);
23012 int vec_any_ne (vector bool char, vector signed char);
23013 int vec_any_ne (vector signed short, vector bool short);
23014 int vec_any_ne (vector signed short, vector signed short);
23015 int vec_any_ne (vector unsigned short, vector bool short);
23016 int vec_any_ne (vector unsigned short, vector unsigned short);
23017 int vec_any_ne (vector bool short, vector bool short);
23018 int vec_any_ne (vector bool short, vector unsigned short);
23019 int vec_any_ne (vector bool short, vector signed short);
23020 int vec_any_ne (vector pixel, vector pixel);
23021 int vec_any_ne (vector signed int, vector bool int);
23022 int vec_any_ne (vector signed int, vector signed int);
23023 int vec_any_ne (vector unsigned int, vector bool int);
23024 int vec_any_ne (vector unsigned int, vector unsigned int);
23025 int vec_any_ne (vector bool int, vector bool int);
23026 int vec_any_ne (vector bool int, vector unsigned int);
23027 int vec_any_ne (vector bool int, vector signed int);
23028 int vec_any_ne (vector float, vector float);
23030 int vec_any_nge (vector float, vector float);
23032 int vec_any_ngt (vector float, vector float);
23034 int vec_any_nle (vector float, vector float);
23036 int vec_any_nlt (vector float, vector float);
23038 int vec_any_numeric (vector float);
23040 int vec_any_out (vector float, vector float);
23043 File: gcc.info, Node: SPARC VIS Built-in Functions, Prev: PowerPC AltiVec Built-in Functions, Up: Target Builtins
23045 5.47.9 SPARC VIS Built-in Functions
23046 -----------------------------------
23048 GCC supports SIMD operations on the SPARC using both the generic vector
23049 extensions (*note Vector Extensions::) as well as built-in functions for
23050 the SPARC Visual Instruction Set (VIS). When you use the `-mvis'
23051 switch, the VIS extension is exposed as the following built-in
23054 typedef int v2si __attribute__ ((vector_size (8)));
23055 typedef short v4hi __attribute__ ((vector_size (8)));
23056 typedef short v2hi __attribute__ ((vector_size (4)));
23057 typedef char v8qi __attribute__ ((vector_size (8)));
23058 typedef char v4qi __attribute__ ((vector_size (4)));
23060 void * __builtin_vis_alignaddr (void *, long);
23061 int64_t __builtin_vis_faligndatadi (int64_t, int64_t);
23062 v2si __builtin_vis_faligndatav2si (v2si, v2si);
23063 v4hi __builtin_vis_faligndatav4hi (v4si, v4si);
23064 v8qi __builtin_vis_faligndatav8qi (v8qi, v8qi);
23066 v4hi __builtin_vis_fexpand (v4qi);
23068 v4hi __builtin_vis_fmul8x16 (v4qi, v4hi);
23069 v4hi __builtin_vis_fmul8x16au (v4qi, v4hi);
23070 v4hi __builtin_vis_fmul8x16al (v4qi, v4hi);
23071 v4hi __builtin_vis_fmul8sux16 (v8qi, v4hi);
23072 v4hi __builtin_vis_fmul8ulx16 (v8qi, v4hi);
23073 v2si __builtin_vis_fmuld8sux16 (v4qi, v2hi);
23074 v2si __builtin_vis_fmuld8ulx16 (v4qi, v2hi);
23076 v4qi __builtin_vis_fpack16 (v4hi);
23077 v8qi __builtin_vis_fpack32 (v2si, v2si);
23078 v2hi __builtin_vis_fpackfix (v2si);
23079 v8qi __builtin_vis_fpmerge (v4qi, v4qi);
23081 int64_t __builtin_vis_pdist (v8qi, v8qi, int64_t);
23084 File: gcc.info, Node: Target Format Checks, Next: Pragmas, Prev: Target Builtins, Up: C Extensions
23086 5.48 Format Checks Specific to Particular Target Machines
23087 =========================================================
23089 For some target machines, GCC supports additional options to the format
23090 attribute (*note Declaring Attributes of Functions: Function
23095 * Solaris Format Checks::
23098 File: gcc.info, Node: Solaris Format Checks, Up: Target Format Checks
23100 5.48.1 Solaris Format Checks
23101 ----------------------------
23103 Solaris targets support the `cmn_err' (or `__cmn_err__') format check.
23104 `cmn_err' accepts a subset of the standard `printf' conversions, and
23105 the two-argument `%b' conversion for displaying bit-fields. See the
23106 Solaris man page for `cmn_err' for more information.
23109 File: gcc.info, Node: Pragmas, Next: Unnamed Fields, Prev: Target Format Checks, Up: C Extensions
23111 5.49 Pragmas Accepted by GCC
23112 ============================
23114 GCC supports several types of pragmas, primarily in order to compile
23115 code originally written for other compilers. Note that in general we
23116 do not recommend the use of pragmas; *Note Function Attributes::, for
23117 further explanation.
23123 * RS/6000 and PowerPC Pragmas::
23125 * Solaris Pragmas::
23126 * Symbol-Renaming Pragmas::
23127 * Structure-Packing Pragmas::
23131 File: gcc.info, Node: ARM Pragmas, Next: M32C Pragmas, Up: Pragmas
23136 The ARM target defines pragmas for controlling the default addition of
23137 `long_call' and `short_call' attributes to functions. *Note Function
23138 Attributes::, for information about the effects of these attributes.
23141 Set all subsequent functions to have the `long_call' attribute.
23144 Set all subsequent functions to have the `short_call' attribute.
23147 Do not affect the `long_call' or `short_call' attributes of
23148 subsequent functions.
23151 File: gcc.info, Node: M32C Pragmas, Next: RS/6000 and PowerPC Pragmas, Prev: ARM Pragmas, Up: Pragmas
23153 5.49.2 M32C Pragmas
23154 -------------------
23157 Overrides the command line option `-memregs=' for the current
23158 file. Use with care! This pragma must be before any function in
23159 the file, and mixing different memregs values in different objects
23160 may make them incompatible. This pragma is useful when a
23161 performance-critical function uses a memreg for temporary values,
23162 as it may allow you to reduce the number of memregs used.
23166 File: gcc.info, Node: RS/6000 and PowerPC Pragmas, Next: Darwin Pragmas, Prev: M32C Pragmas, Up: Pragmas
23168 5.49.3 RS/6000 and PowerPC Pragmas
23169 ----------------------------------
23171 The RS/6000 and PowerPC targets define one pragma for controlling
23172 whether or not the `longcall' attribute is added to function
23173 declarations by default. This pragma overrides the `-mlongcall'
23174 option, but not the `longcall' and `shortcall' attributes. *Note
23175 RS/6000 and PowerPC Options::, for more information about when long
23176 calls are and are not necessary.
23179 Apply the `longcall' attribute to all subsequent function
23183 Do not apply the `longcall' attribute to subsequent function
23187 File: gcc.info, Node: Darwin Pragmas, Next: Solaris Pragmas, Prev: RS/6000 and PowerPC Pragmas, Up: Pragmas
23189 5.49.4 Darwin Pragmas
23190 ---------------------
23192 The following pragmas are available for all architectures running the
23193 Darwin operating system. These are useful for compatibility with other
23197 This pragma is accepted, but has no effect.
23199 `options align=ALIGNMENT'
23200 This pragma sets the alignment of fields in structures. The
23201 values of ALIGNMENT may be `mac68k', to emulate m68k alignment, or
23202 `power', to emulate PowerPC alignment. Uses of this pragma nest
23203 properly; to restore the previous setting, use `reset' for the
23206 `segment TOKENS...'
23207 This pragma is accepted, but has no effect.
23209 `unused (VAR [, VAR]...)'
23210 This pragma declares variables to be possibly unused. GCC will not
23211 produce warnings for the listed variables. The effect is similar
23212 to that of the `unused' attribute, except that this pragma may
23213 appear anywhere within the variables' scopes.
23216 File: gcc.info, Node: Solaris Pragmas, Next: Symbol-Renaming Pragmas, Prev: Darwin Pragmas, Up: Pragmas
23218 5.49.5 Solaris Pragmas
23219 ----------------------
23221 The Solaris target supports `#pragma redefine_extname' (*note
23222 Symbol-Renaming Pragmas::). It also supports additional `#pragma'
23223 directives for compatibility with the system compiler.
23225 `align ALIGNMENT (VARIABLE [, VARIABLE]...)'
23226 Increase the minimum alignment of each VARIABLE to ALIGNMENT.
23227 This is the same as GCC's `aligned' attribute *note Variable
23228 Attributes::). Macro expansion occurs on the arguments to this
23229 pragma when compiling C and Objective-C. It does not currently
23230 occur when compiling C++, but this is a bug which may be fixed in
23233 `fini (FUNCTION [, FUNCTION]...)'
23234 This pragma causes each listed FUNCTION to be called after main,
23235 or during shared module unloading, by adding a call to the `.fini'
23238 `init (FUNCTION [, FUNCTION]...)'
23239 This pragma causes each listed FUNCTION to be called during
23240 initialization (before `main') or during shared module loading, by
23241 adding a call to the `.init' section.
23245 File: gcc.info, Node: Symbol-Renaming Pragmas, Next: Structure-Packing Pragmas, Prev: Solaris Pragmas, Up: Pragmas
23247 5.49.6 Symbol-Renaming Pragmas
23248 ------------------------------
23250 For compatibility with the Solaris and Tru64 UNIX system headers, GCC
23251 supports two `#pragma' directives which change the name used in
23252 assembly for a given declaration. These pragmas are only available on
23253 platforms whose system headers need them. To get this effect on all
23254 platforms supported by GCC, use the asm labels extension (*note Asm
23257 `redefine_extname OLDNAME NEWNAME'
23258 This pragma gives the C function OLDNAME the assembly symbol
23259 NEWNAME. The preprocessor macro `__PRAGMA_REDEFINE_EXTNAME' will
23260 be defined if this pragma is available (currently only on Solaris).
23262 `extern_prefix STRING'
23263 This pragma causes all subsequent external function and variable
23264 declarations to have STRING prepended to their assembly symbols.
23265 This effect may be terminated with another `extern_prefix' pragma
23266 whose argument is an empty string. The preprocessor macro
23267 `__PRAGMA_EXTERN_PREFIX' will be defined if this pragma is
23268 available (currently only on Tru64 UNIX).
23270 These pragmas and the asm labels extension interact in a complicated
23271 manner. Here are some corner cases you may want to be aware of.
23273 1. Both pragmas silently apply only to declarations with external
23274 linkage. Asm labels do not have this restriction.
23276 2. In C++, both pragmas silently apply only to declarations with "C"
23277 linkage. Again, asm labels do not have this restriction.
23279 3. If any of the three ways of changing the assembly name of a
23280 declaration is applied to a declaration whose assembly name has
23281 already been determined (either by a previous use of one of these
23282 features, or because the compiler needed the assembly name in
23283 order to generate code), and the new name is different, a warning
23284 issues and the name does not change.
23286 4. The OLDNAME used by `#pragma redefine_extname' is always the
23289 5. If `#pragma extern_prefix' is in effect, and a declaration occurs
23290 with an asm label attached, the prefix is silently ignored for
23293 6. If `#pragma extern_prefix' and `#pragma redefine_extname' apply to
23294 the same declaration, whichever triggered first wins, and a
23295 warning issues if they contradict each other. (We would like to
23296 have `#pragma redefine_extname' always win, for consistency with
23297 asm labels, but if `#pragma extern_prefix' triggers first we have
23298 no way of knowing that that happened.)
23301 File: gcc.info, Node: Structure-Packing Pragmas, Next: Weak Pragmas, Prev: Symbol-Renaming Pragmas, Up: Pragmas
23303 5.49.7 Structure-Packing Pragmas
23304 --------------------------------
23306 For compatibility with Win32, GCC supports a set of `#pragma'
23307 directives which change the maximum alignment of members of structures
23308 (other than zero-width bitfields), unions, and classes subsequently
23309 defined. The N value below always is required to be a small power of
23310 two and specifies the new alignment in bytes.
23312 1. `#pragma pack(N)' simply sets the new alignment.
23314 2. `#pragma pack()' sets the alignment to the one that was in effect
23315 when compilation started (see also command line option
23316 `-fpack-struct[=<n>]' *note Code Gen Options::).
23318 3. `#pragma pack(push[,N])' pushes the current alignment setting on
23319 an internal stack and then optionally sets the new alignment.
23321 4. `#pragma pack(pop)' restores the alignment setting to the one
23322 saved at the top of the internal stack (and removes that stack
23323 entry). Note that `#pragma pack([N])' does not influence this
23324 internal stack; thus it is possible to have `#pragma pack(push)'
23325 followed by multiple `#pragma pack(N)' instances and finalized by
23326 a single `#pragma pack(pop)'.
23329 File: gcc.info, Node: Weak Pragmas, Prev: Structure-Packing Pragmas, Up: Pragmas
23331 5.49.8 Weak Pragmas
23332 -------------------
23334 For compatibility with SVR4, GCC supports a set of `#pragma' directives
23335 for declaring symbols to be weak, and defining weak aliases.
23337 `#pragma weak SYMBOL'
23338 This pragma declares SYMBOL to be weak, as if the declaration had
23339 the attribute of the same name. The pragma may appear before or
23340 after the declaration of SYMBOL, but must appear before either its
23341 first use or its definition. It is not an error for SYMBOL to
23342 never be defined at all.
23344 `#pragma weak SYMBOL1 = SYMBOL2'
23345 This pragma declares SYMBOL1 to be a weak alias of SYMBOL2. It is
23346 an error if SYMBOL2 is not defined in the current translation unit.
23349 File: gcc.info, Node: Unnamed Fields, Next: Thread-Local, Prev: Pragmas, Up: C Extensions
23351 5.50 Unnamed struct/union fields within structs/unions
23352 ======================================================
23354 For compatibility with other compilers, GCC allows you to define a
23355 structure or union that contains, as fields, structures and unions
23356 without names. For example:
23367 In this example, the user would be able to access members of the
23368 unnamed union with code like `foo.b'. Note that only unnamed structs
23369 and unions are allowed, you may not have, for example, an unnamed `int'.
23371 You must never create such structures that cause ambiguous field
23372 definitions. For example, this structure:
23381 It is ambiguous which `a' is being referred to with `foo.a'. Such
23382 constructs are not supported and must be avoided. In the future, such
23383 constructs may be detected and treated as compilation errors.
23385 Unless `-fms-extensions' is used, the unnamed field must be a
23386 structure or union definition without a tag (for example, `struct { int
23387 a; };'). If `-fms-extensions' is used, the field may also be a
23388 definition with a tag such as `struct foo { int a; };', a reference to
23389 a previously defined structure or union such as `struct foo;', or a
23390 reference to a `typedef' name for a previously defined structure or
23394 File: gcc.info, Node: Thread-Local, Prev: Unnamed Fields, Up: C Extensions
23396 5.51 Thread-Local Storage
23397 =========================
23399 Thread-local storage (TLS) is a mechanism by which variables are
23400 allocated such that there is one instance of the variable per extant
23401 thread. The run-time model GCC uses to implement this originates in
23402 the IA-64 processor-specific ABI, but has since been migrated to other
23403 processors as well. It requires significant support from the linker
23404 (`ld'), dynamic linker (`ld.so'), and system libraries (`libc.so' and
23405 `libpthread.so'), so it is not available everywhere.
23407 At the user level, the extension is visible with a new storage class
23408 keyword: `__thread'. For example:
23411 extern __thread struct state s;
23412 static __thread char *p;
23414 The `__thread' specifier may be used alone, with the `extern' or
23415 `static' specifiers, but with no other storage class specifier. When
23416 used with `extern' or `static', `__thread' must appear immediately
23417 after the other storage class specifier.
23419 The `__thread' specifier may be applied to any global, file-scoped
23420 static, function-scoped static, or static data member of a class. It
23421 may not be applied to block-scoped automatic or non-static data member.
23423 When the address-of operator is applied to a thread-local variable, it
23424 is evaluated at run-time and returns the address of the current thread's
23425 instance of that variable. An address so obtained may be used by any
23426 thread. When a thread terminates, any pointers to thread-local
23427 variables in that thread become invalid.
23429 No static initialization may refer to the address of a thread-local
23432 In C++, if an initializer is present for a thread-local variable, it
23433 must be a CONSTANT-EXPRESSION, as defined in 5.19.2 of the ANSI/ISO C++
23436 See ELF Handling For Thread-Local Storage
23437 (http://people.redhat.com/drepper/tls.pdf) for a detailed explanation of
23438 the four thread-local storage addressing models, and how the run-time
23439 is expected to function.
23443 * C99 Thread-Local Edits::
23444 * C++98 Thread-Local Edits::
23447 File: gcc.info, Node: C99 Thread-Local Edits, Next: C++98 Thread-Local Edits, Up: Thread-Local
23449 5.51.1 ISO/IEC 9899:1999 Edits for Thread-Local Storage
23450 -------------------------------------------------------
23452 The following are a set of changes to ISO/IEC 9899:1999 (aka C99) that
23453 document the exact semantics of the language extension.
23455 * `5.1.2 Execution environments'
23457 Add new text after paragraph 1
23459 Within either execution environment, a "thread" is a flow of
23460 control within a program. It is implementation defined
23461 whether or not there may be more than one thread associated
23462 with a program. It is implementation defined how threads
23463 beyond the first are created, the name and type of the
23464 function called at thread startup, and how threads may be
23465 terminated. However, objects with thread storage duration
23466 shall be initialized before thread startup.
23468 * `6.2.4 Storage durations of objects'
23470 Add new text before paragraph 3
23472 An object whose identifier is declared with the storage-class
23473 specifier `__thread' has "thread storage duration". Its
23474 lifetime is the entire execution of the thread, and its
23475 stored value is initialized only once, prior to thread
23482 * `6.7.1 Storage-class specifiers'
23484 Add `__thread' to the list of storage class specifiers in
23487 Change paragraph 2 to
23489 With the exception of `__thread', at most one storage-class
23490 specifier may be given [...]. The `__thread' specifier may
23491 be used alone, or immediately following `extern' or `static'.
23493 Add new text after paragraph 6
23495 The declaration of an identifier for a variable that has
23496 block scope that specifies `__thread' shall also specify
23497 either `extern' or `static'.
23499 The `__thread' specifier shall be used only with variables.
23502 File: gcc.info, Node: C++98 Thread-Local Edits, Prev: C99 Thread-Local Edits, Up: Thread-Local
23504 5.51.2 ISO/IEC 14882:1998 Edits for Thread-Local Storage
23505 --------------------------------------------------------
23507 The following are a set of changes to ISO/IEC 14882:1998 (aka C++98)
23508 that document the exact semantics of the language extension.
23510 * [intro.execution]
23512 New text after paragraph 4
23514 A "thread" is a flow of control within the abstract machine.
23515 It is implementation defined whether or not there may be more
23518 New text after paragraph 7
23520 It is unspecified whether additional action must be taken to
23521 ensure when and whether side effects are visible to other
23528 * [basic.start.main]
23530 Add after paragraph 5
23532 The thread that begins execution at the `main' function is
23533 called the "main thread". It is implementation defined how
23534 functions beginning threads other than the main thread are
23535 designated or typed. A function so designated, as well as
23536 the `main' function, is called a "thread startup function".
23537 It is implementation defined what happens if a thread startup
23538 function returns. It is implementation defined what happens
23539 to other threads when any thread calls `exit'.
23541 * [basic.start.init]
23543 Add after paragraph 4
23545 The storage for an object of thread storage duration shall be
23546 statically initialized before the first statement of the
23547 thread startup function. An object of thread storage
23548 duration shall not require dynamic initialization.
23550 * [basic.start.term]
23552 Add after paragraph 3
23554 The type of an object with thread storage duration shall not
23555 have a non-trivial destructor, nor shall it be an array type
23556 whose elements (directly or indirectly) have non-trivial
23561 Add "thread storage duration" to the list in paragraph 1.
23565 Thread, static, and automatic storage durations are
23566 associated with objects introduced by declarations [...].
23568 Add `__thread' to the list of specifiers in paragraph 3.
23570 * [basic.stc.thread]
23572 New section before [basic.stc.static]
23574 The keyword `__thread' applied to a non-local object gives the
23575 object thread storage duration.
23577 A local variable or class data member declared both `static'
23578 and `__thread' gives the variable or member thread storage
23581 * [basic.stc.static]
23585 All objects which have neither thread storage duration,
23586 dynamic storage duration nor are local [...].
23590 Add `__thread' to the list in paragraph 1.
23594 With the exception of `__thread', at most one
23595 STORAGE-CLASS-SPECIFIER shall appear in a given
23596 DECL-SPECIFIER-SEQ. The `__thread' specifier may be used
23597 alone, or immediately following the `extern' or `static'
23600 Add after paragraph 5
23602 The `__thread' specifier can be applied only to the names of
23603 objects and to anonymous unions.
23607 Add after paragraph 6
23609 Non-`static' members shall not be `__thread'.
23612 File: gcc.info, Node: C++ Extensions, Next: Objective-C, Prev: C Extensions, Up: Top
23614 6 Extensions to the C++ Language
23615 ********************************
23617 The GNU compiler provides these extensions to the C++ language (and you
23618 can also use most of the C language extensions in your C++ programs).
23619 If you want to write code that checks whether these features are
23620 available, you can test for the GNU compiler the same way as for C
23621 programs: check for a predefined macro `__GNUC__'. You can also use
23622 `__GNUG__' to test specifically for GNU C++ (*note Predefined Macros:
23623 (cpp)Common Predefined Macros.).
23627 * Volatiles:: What constitutes an access to a volatile object.
23628 * Restricted Pointers:: C99 restricted pointers and references.
23629 * Vague Linkage:: Where G++ puts inlines, vtables and such.
23630 * C++ Interface:: You can use a single C++ header file for both
23631 declarations and definitions.
23632 * Template Instantiation:: Methods for ensuring that exactly one copy of
23633 each needed template instantiation is emitted.
23634 * Bound member functions:: You can extract a function pointer to the
23635 method denoted by a `->*' or `.*' expression.
23636 * C++ Attributes:: Variable, function, and type attributes for C++ only.
23637 * Strong Using:: Strong using-directives for namespace composition.
23638 * Java Exceptions:: Tweaking exception handling to work with Java.
23639 * Deprecated Features:: Things will disappear from g++.
23640 * Backwards Compatibility:: Compatibilities with earlier definitions of C++.
23643 File: gcc.info, Node: Volatiles, Next: Restricted Pointers, Up: C++ Extensions
23645 6.1 When is a Volatile Object Accessed?
23646 =======================================
23648 Both the C and C++ standard have the concept of volatile objects. These
23649 are normally accessed by pointers and used for accessing hardware. The
23650 standards encourage compilers to refrain from optimizations concerning
23651 accesses to volatile objects. The C standard leaves it implementation
23652 defined as to what constitutes a volatile access. The C++ standard
23653 omits to specify this, except to say that C++ should behave in a
23654 similar manner to C with respect to volatiles, where possible. The
23655 minimum either standard specifies is that at a sequence point all
23656 previous accesses to volatile objects have stabilized and no subsequent
23657 accesses have occurred. Thus an implementation is free to reorder and
23658 combine volatile accesses which occur between sequence points, but
23659 cannot do so for accesses across a sequence point. The use of
23660 volatiles does not allow you to violate the restriction on updating
23661 objects multiple times within a sequence point.
23663 *Note Volatile qualifier and the C compiler: Qualifiers implementation.
23665 The behavior differs slightly between C and C++ in the non-obvious
23668 volatile int *src = SOMEVALUE;
23671 With C, such expressions are rvalues, and GCC interprets this either
23672 as a read of the volatile object being pointed to or only as request to
23673 evaluate the side-effects. The C++ standard specifies that such
23674 expressions do not undergo lvalue to rvalue conversion, and that the
23675 type of the dereferenced object may be incomplete. The C++ standard
23676 does not specify explicitly that it is this lvalue to rvalue conversion
23677 which may be responsible for causing an access. However, there is
23678 reason to believe that it is, because otherwise certain simple
23679 expressions become undefined. However, because it would surprise most
23680 programmers, G++ treats dereferencing a pointer to volatile object of
23681 complete type when the value is unused as GCC would do for an
23682 equivalent type in C. When the object has incomplete type, G++ issues
23683 a warning; if you wish to force an error, you must force a conversion
23684 to rvalue with, for instance, a static cast.
23686 When using a reference to volatile, G++ does not treat equivalent
23687 expressions as accesses to volatiles, but instead issues a warning that
23688 no volatile is accessed. The rationale for this is that otherwise it
23689 becomes difficult to determine where volatile access occur, and not
23690 possible to ignore the return value from functions returning volatile
23691 references. Again, if you wish to force a read, cast the reference to
23695 File: gcc.info, Node: Restricted Pointers, Next: Vague Linkage, Prev: Volatiles, Up: C++ Extensions
23697 6.2 Restricting Pointer Aliasing
23698 ================================
23700 As with the C front end, G++ understands the C99 feature of restricted
23701 pointers, specified with the `__restrict__', or `__restrict' type
23702 qualifier. Because you cannot compile C++ by specifying the `-std=c99'
23703 language flag, `restrict' is not a keyword in C++.
23705 In addition to allowing restricted pointers, you can specify restricted
23706 references, which indicate that the reference is not aliased in the
23709 void fn (int *__restrict__ rptr, int &__restrict__ rref)
23714 In the body of `fn', RPTR points to an unaliased integer and RREF
23715 refers to a (different) unaliased integer.
23717 You may also specify whether a member function's THIS pointer is
23718 unaliased by using `__restrict__' as a member function qualifier.
23720 void T::fn () __restrict__
23725 Within the body of `T::fn', THIS will have the effective definition `T
23726 *__restrict__ const this'. Notice that the interpretation of a
23727 `__restrict__' member function qualifier is different to that of
23728 `const' or `volatile' qualifier, in that it is applied to the pointer
23729 rather than the object. This is consistent with other compilers which
23730 implement restricted pointers.
23732 As with all outermost parameter qualifiers, `__restrict__' is ignored
23733 in function definition matching. This means you only need to specify
23734 `__restrict__' in a function definition, rather than in a function
23738 File: gcc.info, Node: Vague Linkage, Next: C++ Interface, Prev: Restricted Pointers, Up: C++ Extensions
23743 There are several constructs in C++ which require space in the object
23744 file but are not clearly tied to a single translation unit. We say that
23745 these constructs have "vague linkage". Typically such constructs are
23746 emitted wherever they are needed, though sometimes we can be more
23750 Inline functions are typically defined in a header file which can
23751 be included in many different compilations. Hopefully they can
23752 usually be inlined, but sometimes an out-of-line copy is
23753 necessary, if the address of the function is taken or if inlining
23754 fails. In general, we emit an out-of-line copy in all translation
23755 units where one is needed. As an exception, we only emit inline
23756 virtual functions with the vtable, since it will always require a
23759 Local static variables and string constants used in an inline
23760 function are also considered to have vague linkage, since they
23761 must be shared between all inlined and out-of-line instances of
23765 C++ virtual functions are implemented in most compilers using a
23766 lookup table, known as a vtable. The vtable contains pointers to
23767 the virtual functions provided by a class, and each object of the
23768 class contains a pointer to its vtable (or vtables, in some
23769 multiple-inheritance situations). If the class declares any
23770 non-inline, non-pure virtual functions, the first one is chosen as
23771 the "key method" for the class, and the vtable is only emitted in
23772 the translation unit where the key method is defined.
23774 _Note:_ If the chosen key method is later defined as inline, the
23775 vtable will still be emitted in every translation unit which
23776 defines it. Make sure that any inline virtuals are declared
23777 inline in the class body, even if they are not defined there.
23780 C++ requires information about types to be written out in order to
23781 implement `dynamic_cast', `typeid' and exception handling. For
23782 polymorphic classes (classes with virtual functions), the type_info
23783 object is written out along with the vtable so that `dynamic_cast'
23784 can determine the dynamic type of a class object at runtime. For
23785 all other types, we write out the type_info object when it is
23786 used: when applying `typeid' to an expression, throwing an object,
23787 or referring to a type in a catch clause or exception
23790 Template Instantiations
23791 Most everything in this section also applies to template
23792 instantiations, but there are other options as well. *Note
23793 Where's the Template?: Template Instantiation.
23796 When used with GNU ld version 2.8 or later on an ELF system such as
23797 GNU/Linux or Solaris 2, or on Microsoft Windows, duplicate copies of
23798 these constructs will be discarded at link time. This is known as
23801 On targets that don't support COMDAT, but do support weak symbols, GCC
23802 will use them. This way one copy will override all the others, but the
23803 unused copies will still take up space in the executable.
23805 For targets which do not support either COMDAT or weak symbols, most
23806 entities with vague linkage will be emitted as local symbols to avoid
23807 duplicate definition errors from the linker. This will not happen for
23808 local statics in inlines, however, as having multiple copies will
23809 almost certainly break things.
23811 *Note Declarations and Definitions in One Header: C++ Interface, for
23812 another way to control placement of these constructs.
23815 File: gcc.info, Node: C++ Interface, Next: Template Instantiation, Prev: Vague Linkage, Up: C++ Extensions
23817 6.4 #pragma interface and implementation
23818 ========================================
23820 `#pragma interface' and `#pragma implementation' provide the user with
23821 a way of explicitly directing the compiler to emit entities with vague
23822 linkage (and debugging information) in a particular translation unit.
23824 _Note:_ As of GCC 2.7.2, these `#pragma's are not useful in most
23825 cases, because of COMDAT support and the "key method" heuristic
23826 mentioned in *Note Vague Linkage::. Using them can actually cause your
23827 program to grow due to unnecessary out-of-line copies of inline
23828 functions. Currently (3.4) the only benefit of these `#pragma's is
23829 reduced duplication of debugging information, and that should be
23830 addressed soon on DWARF 2 targets with the use of COMDAT groups.
23832 `#pragma interface'
23833 `#pragma interface "SUBDIR/OBJECTS.h"'
23834 Use this directive in _header files_ that define object classes,
23835 to save space in most of the object files that use those classes.
23836 Normally, local copies of certain information (backup copies of
23837 inline member functions, debugging information, and the internal
23838 tables that implement virtual functions) must be kept in each
23839 object file that includes class definitions. You can use this
23840 pragma to avoid such duplication. When a header file containing
23841 `#pragma interface' is included in a compilation, this auxiliary
23842 information will not be generated (unless the main input source
23843 file itself uses `#pragma implementation'). Instead, the object
23844 files will contain references to be resolved at link time.
23846 The second form of this directive is useful for the case where you
23847 have multiple headers with the same name in different directories.
23848 If you use this form, you must specify the same string to `#pragma
23851 `#pragma implementation'
23852 `#pragma implementation "OBJECTS.h"'
23853 Use this pragma in a _main input file_, when you want full output
23854 from included header files to be generated (and made globally
23855 visible). The included header file, in turn, should use `#pragma
23856 interface'. Backup copies of inline member functions, debugging
23857 information, and the internal tables used to implement virtual
23858 functions are all generated in implementation files.
23860 If you use `#pragma implementation' with no argument, it applies to
23861 an include file with the same basename(1) as your source file.
23862 For example, in `allclass.cc', giving just `#pragma implementation'
23863 by itself is equivalent to `#pragma implementation "allclass.h"'.
23865 In versions of GNU C++ prior to 2.6.0 `allclass.h' was treated as
23866 an implementation file whenever you would include it from
23867 `allclass.cc' even if you never specified `#pragma
23868 implementation'. This was deemed to be more trouble than it was
23869 worth, however, and disabled.
23871 Use the string argument if you want a single implementation file to
23872 include code from multiple header files. (You must also use
23873 `#include' to include the header file; `#pragma implementation'
23874 only specifies how to use the file--it doesn't actually include
23877 There is no way to split up the contents of a single header file
23878 into multiple implementation files.
23880 `#pragma implementation' and `#pragma interface' also have an effect
23881 on function inlining.
23883 If you define a class in a header file marked with `#pragma
23884 interface', the effect on an inline function defined in that class is
23885 similar to an explicit `extern' declaration--the compiler emits no code
23886 at all to define an independent version of the function. Its
23887 definition is used only for inlining with its callers.
23889 Conversely, when you include the same header file in a main source file
23890 that declares it as `#pragma implementation', the compiler emits code
23891 for the function itself; this defines a version of the function that
23892 can be found via pointers (or by callers compiled without inlining).
23893 If all calls to the function can be inlined, you can avoid emitting the
23894 function by compiling with `-fno-implement-inlines'. If any calls were
23895 not inlined, you will get linker errors.
23897 ---------- Footnotes ----------
23899 (1) A file's "basename" was the name stripped of all leading path
23900 information and of trailing suffixes, such as `.h' or `.C' or `.cc'.
23903 File: gcc.info, Node: Template Instantiation, Next: Bound member functions, Prev: C++ Interface, Up: C++ Extensions
23905 6.5 Where's the Template?
23906 =========================
23908 C++ templates are the first language feature to require more
23909 intelligence from the environment than one usually finds on a UNIX
23910 system. Somehow the compiler and linker have to make sure that each
23911 template instance occurs exactly once in the executable if it is needed,
23912 and not at all otherwise. There are two basic approaches to this
23913 problem, which are referred to as the Borland model and the Cfront
23917 Borland C++ solved the template instantiation problem by adding
23918 the code equivalent of common blocks to their linker; the compiler
23919 emits template instances in each translation unit that uses them,
23920 and the linker collapses them together. The advantage of this
23921 model is that the linker only has to consider the object files
23922 themselves; there is no external complexity to worry about. This
23923 disadvantage is that compilation time is increased because the
23924 template code is being compiled repeatedly. Code written for this
23925 model tends to include definitions of all templates in the header
23926 file, since they must be seen to be instantiated.
23929 The AT&T C++ translator, Cfront, solved the template instantiation
23930 problem by creating the notion of a template repository, an
23931 automatically maintained place where template instances are
23932 stored. A more modern version of the repository works as follows:
23933 As individual object files are built, the compiler places any
23934 template definitions and instantiations encountered in the
23935 repository. At link time, the link wrapper adds in the objects in
23936 the repository and compiles any needed instances that were not
23937 previously emitted. The advantages of this model are more optimal
23938 compilation speed and the ability to use the system linker; to
23939 implement the Borland model a compiler vendor also needs to
23940 replace the linker. The disadvantages are vastly increased
23941 complexity, and thus potential for error; for some code this can be
23942 just as transparent, but in practice it can been very difficult to
23943 build multiple programs in one directory and one program in
23944 multiple directories. Code written for this model tends to
23945 separate definitions of non-inline member templates into a
23946 separate file, which should be compiled separately.
23948 When used with GNU ld version 2.8 or later on an ELF system such as
23949 GNU/Linux or Solaris 2, or on Microsoft Windows, G++ supports the
23950 Borland model. On other systems, G++ implements neither automatic
23953 A future version of G++ will support a hybrid model whereby the
23954 compiler will emit any instantiations for which the template definition
23955 is included in the compile, and store template definitions and
23956 instantiation context information into the object file for the rest.
23957 The link wrapper will extract that information as necessary and invoke
23958 the compiler to produce the remaining instantiations. The linker will
23959 then combine duplicate instantiations.
23961 In the mean time, you have the following options for dealing with
23962 template instantiations:
23964 1. Compile your template-using code with `-frepo'. The compiler will
23965 generate files with the extension `.rpo' listing all of the
23966 template instantiations used in the corresponding object files
23967 which could be instantiated there; the link wrapper, `collect2',
23968 will then update the `.rpo' files to tell the compiler where to
23969 place those instantiations and rebuild any affected object files.
23970 The link-time overhead is negligible after the first pass, as the
23971 compiler will continue to place the instantiations in the same
23974 This is your best option for application code written for the
23975 Borland model, as it will just work. Code written for the Cfront
23976 model will need to be modified so that the template definitions
23977 are available at one or more points of instantiation; usually this
23978 is as simple as adding `#include <tmethods.cc>' to the end of each
23981 For library code, if you want the library to provide all of the
23982 template instantiations it needs, just try to link all of its
23983 object files together; the link will fail, but cause the
23984 instantiations to be generated as a side effect. Be warned,
23985 however, that this may cause conflicts if multiple libraries try
23986 to provide the same instantiations. For greater control, use
23987 explicit instantiation as described in the next option.
23989 2. Compile your code with `-fno-implicit-templates' to disable the
23990 implicit generation of template instances, and explicitly
23991 instantiate all the ones you use. This approach requires more
23992 knowledge of exactly which instances you need than do the others,
23993 but it's less mysterious and allows greater control. You can
23994 scatter the explicit instantiations throughout your program,
23995 perhaps putting them in the translation units where the instances
23996 are used or the translation units that define the templates
23997 themselves; you can put all of the explicit instantiations you
23998 need into one big file; or you can create small files like
24003 template class Foo<int>;
24004 template ostream& operator <<
24005 (ostream&, const Foo<int>&);
24007 for each of the instances you need, and create a template
24008 instantiation library from those.
24010 If you are using Cfront-model code, you can probably get away with
24011 not using `-fno-implicit-templates' when compiling files that don't
24012 `#include' the member template definitions.
24014 If you use one big file to do the instantiations, you may want to
24015 compile it without `-fno-implicit-templates' so you get all of the
24016 instances required by your explicit instantiations (but not by any
24017 other files) without having to specify them as well.
24019 G++ has extended the template instantiation syntax given in the ISO
24020 standard to allow forward declaration of explicit instantiations
24021 (with `extern'), instantiation of the compiler support data for a
24022 template class (i.e. the vtable) without instantiating any of its
24023 members (with `inline'), and instantiation of only the static data
24024 members of a template class, without the support data or member
24025 functions (with (`static'):
24027 extern template int max (int, int);
24028 inline template class Foo<int>;
24029 static template class Foo<int>;
24031 3. Do nothing. Pretend G++ does implement automatic instantiation
24032 management. Code written for the Borland model will work fine, but
24033 each translation unit will contain instances of each of the
24034 templates it uses. In a large program, this can lead to an
24035 unacceptable amount of code duplication.
24038 File: gcc.info, Node: Bound member functions, Next: C++ Attributes, Prev: Template Instantiation, Up: C++ Extensions
24040 6.6 Extracting the function pointer from a bound pointer to member function
24041 ===========================================================================
24043 In C++, pointer to member functions (PMFs) are implemented using a wide
24044 pointer of sorts to handle all the possible call mechanisms; the PMF
24045 needs to store information about how to adjust the `this' pointer, and
24046 if the function pointed to is virtual, where to find the vtable, and
24047 where in the vtable to look for the member function. If you are using
24048 PMFs in an inner loop, you should really reconsider that decision. If
24049 that is not an option, you can extract the pointer to the function that
24050 would be called for a given object/PMF pair and call it directly inside
24051 the inner loop, to save a bit of time.
24053 Note that you will still be paying the penalty for the call through a
24054 function pointer; on most modern architectures, such a call defeats the
24055 branch prediction features of the CPU. This is also true of normal
24056 virtual function calls.
24058 The syntax for this extension is
24061 extern int (A::*fp)();
24062 typedef int (*fptr)(A *);
24064 fptr p = (fptr)(a.*fp);
24066 For PMF constants (i.e. expressions of the form `&Klasse::Member'), no
24067 object is needed to obtain the address of the function. They can be
24068 converted to function pointers directly:
24070 fptr p1 = (fptr)(&A::foo);
24072 You must specify `-Wno-pmf-conversions' to use this extension.
24075 File: gcc.info, Node: C++ Attributes, Next: Strong Using, Prev: Bound member functions, Up: C++ Extensions
24077 6.7 C++-Specific Variable, Function, and Type Attributes
24078 ========================================================
24080 Some attributes only make sense for C++ programs.
24082 `init_priority (PRIORITY)'
24083 In Standard C++, objects defined at namespace scope are guaranteed
24084 to be initialized in an order in strict accordance with that of
24085 their definitions _in a given translation unit_. No guarantee is
24086 made for initializations across translation units. However, GNU
24087 C++ allows users to control the order of initialization of objects
24088 defined at namespace scope with the `init_priority' attribute by
24089 specifying a relative PRIORITY, a constant integral expression
24090 currently bounded between 101 and 65535 inclusive. Lower numbers
24091 indicate a higher priority.
24093 In the following example, `A' would normally be created before
24094 `B', but the `init_priority' attribute has reversed that order:
24096 Some_Class A __attribute__ ((init_priority (2000)));
24097 Some_Class B __attribute__ ((init_priority (543)));
24099 Note that the particular values of PRIORITY do not matter; only
24100 their relative ordering.
24103 This type attribute informs C++ that the class is a Java
24104 interface. It may only be applied to classes declared within an
24105 `extern "Java"' block. Calls to methods declared in this
24106 interface will be dispatched using GCJ's interface table
24107 mechanism, instead of regular virtual table dispatch.
24110 See also *Note Strong Using::.
24113 File: gcc.info, Node: Strong Using, Next: Java Exceptions, Prev: C++ Attributes, Up: C++ Extensions
24118 *Caution:* The semantics of this extension are not fully defined.
24119 Users should refrain from using this extension as its semantics may
24120 change subtly over time. It is possible that this extension wil be
24121 removed in future versions of G++.
24123 A using-directive with `__attribute ((strong))' is stronger than a
24124 normal using-directive in two ways:
24126 * Templates from the used namespace can be specialized as though
24127 they were members of the using namespace.
24129 * The using namespace is considered an associated namespace of all
24130 templates in the used namespace for purposes of argument-dependent
24133 This is useful for composing a namespace transparently from
24134 implementation namespaces. For example:
24138 template <class T> struct A { };
24140 using namespace debug __attribute ((__strong__));
24141 template <> struct A<int> { }; // ok to specialize
24143 template <class T> void f (A<T>);
24148 f (std::A<float>()); // lookup finds std::f
24153 File: gcc.info, Node: Java Exceptions, Next: Deprecated Features, Prev: Strong Using, Up: C++ Extensions
24155 6.9 Java Exceptions
24156 ===================
24158 The Java language uses a slightly different exception handling model
24159 from C++. Normally, GNU C++ will automatically detect when you are
24160 writing C++ code that uses Java exceptions, and handle them
24161 appropriately. However, if C++ code only needs to execute destructors
24162 when Java exceptions are thrown through it, GCC will guess incorrectly.
24163 Sample problematic code is:
24165 struct S { ~S(); };
24166 extern void bar(); // is written in Java, and may throw exceptions
24173 The usual effect of an incorrect guess is a link failure, complaining of
24174 a missing routine called `__gxx_personality_v0'.
24176 You can inform the compiler that Java exceptions are to be used in a
24177 translation unit, irrespective of what it might think, by writing
24178 `#pragma GCC java_exceptions' at the head of the file. This `#pragma'
24179 must appear before any functions that throw or catch exceptions, or run
24180 destructors when exceptions are thrown through them.
24182 You cannot mix Java and C++ exceptions in the same translation unit.
24183 It is believed to be safe to throw a C++ exception from one file through
24184 another file compiled for the Java exception model, or vice versa, but
24185 there may be bugs in this area.
24188 File: gcc.info, Node: Deprecated Features, Next: Backwards Compatibility, Prev: Java Exceptions, Up: C++ Extensions
24190 6.10 Deprecated Features
24191 ========================
24193 In the past, the GNU C++ compiler was extended to experiment with new
24194 features, at a time when the C++ language was still evolving. Now that
24195 the C++ standard is complete, some of those features are superseded by
24196 superior alternatives. Using the old features might cause a warning in
24197 some cases that the feature will be dropped in the future. In other
24198 cases, the feature might be gone already.
24200 While the list below is not exhaustive, it documents some of the
24201 options that are now deprecated:
24203 `-fexternal-templates'
24204 `-falt-external-templates'
24205 These are two of the many ways for G++ to implement template
24206 instantiation. *Note Template Instantiation::. The C++ standard
24207 clearly defines how template definitions have to be organized
24208 across implementation units. G++ has an implicit instantiation
24209 mechanism that should work just fine for standard-conforming code.
24211 `-fstrict-prototype'
24212 `-fno-strict-prototype'
24213 Previously it was possible to use an empty prototype parameter
24214 list to indicate an unspecified number of parameters (like C),
24215 rather than no parameters, as C++ demands. This feature has been
24216 removed, except where it is required for backwards compatibility
24217 *Note Backwards Compatibility::.
24219 G++ allows a virtual function returning `void *' to be overridden by
24220 one returning a different pointer type. This extension to the
24221 covariant return type rules is now deprecated and will be removed from a
24224 The G++ minimum and maximum operators (`<?' and `>?') and their
24225 compound forms (`<?=') and `>?=') have been deprecated and will be
24226 removed in a future version. Code using these operators should be
24227 modified to use `std::min' and `std::max' instead.
24229 The named return value extension has been deprecated, and is now
24232 The use of initializer lists with new expressions has been deprecated,
24233 and is now removed from G++.
24235 Floating and complex non-type template parameters have been deprecated,
24236 and are now removed from G++.
24238 The implicit typename extension has been deprecated and is now removed
24241 The use of default arguments in function pointers, function typedefs
24242 and other places where they are not permitted by the standard is
24243 deprecated and will be removed from a future version of G++.
24245 G++ allows floating-point literals to appear in integral constant
24246 expressions, e.g. ` enum E { e = int(2.2 * 3.7) } ' This extension is
24247 deprecated and will be removed from a future version.
24249 G++ allows static data members of const floating-point type to be
24250 declared with an initializer in a class definition. The standard only
24251 allows initializers for static members of const integral types and const
24252 enumeration types so this extension has been deprecated and will be
24253 removed from a future version.
24256 File: gcc.info, Node: Backwards Compatibility, Prev: Deprecated Features, Up: C++ Extensions
24258 6.11 Backwards Compatibility
24259 ============================
24261 Now that there is a definitive ISO standard C++, G++ has a specification
24262 to adhere to. The C++ language evolved over time, and features that
24263 used to be acceptable in previous drafts of the standard, such as the
24264 ARM [Annotated C++ Reference Manual], are no longer accepted. In order
24265 to allow compilation of C++ written to such drafts, G++ contains some
24266 backwards compatibilities. _All such backwards compatibility features
24267 are liable to disappear in future versions of G++._ They should be
24268 considered deprecated *Note Deprecated Features::.
24271 If a variable is declared at for scope, it used to remain in scope
24272 until the end of the scope which contained the for statement
24273 (rather than just within the for scope). G++ retains this, but
24274 issues a warning, if such a variable is accessed outside the for
24277 `Implicit C language'
24278 Old C system header files did not contain an `extern "C" {...}'
24279 scope to set the language. On such systems, all header files are
24280 implicitly scoped inside a C language scope. Also, an empty
24281 prototype `()' will be treated as an unspecified number of
24282 arguments, rather than no arguments, as C++ demands.
24285 File: gcc.info, Node: Objective-C, Next: Compatibility, Prev: C++ Extensions, Up: Top
24287 7 GNU Objective-C runtime features
24288 **********************************
24290 This document is meant to describe some of the GNU Objective-C runtime
24291 features. It is not intended to teach you Objective-C, there are
24292 several resources on the Internet that present the language. Questions
24293 and comments about this document to Ovidiu Predescu <ovidiu@cup.hp.com>.
24297 * Executing code before main::
24299 * Garbage Collection::
24300 * Constant string objects::
24301 * compatibility_alias::
24304 File: gcc.info, Node: Executing code before main, Next: Type encoding, Prev: Objective-C, Up: Objective-C
24306 7.1 `+load': Executing code before main
24307 =======================================
24309 The GNU Objective-C runtime provides a way that allows you to execute
24310 code before the execution of the program enters the `main' function.
24311 The code is executed on a per-class and a per-category basis, through a
24312 special class method `+load'.
24314 This facility is very useful if you want to initialize global variables
24315 which can be accessed by the program directly, without sending a message
24316 to the class first. The usual way to initialize global variables, in
24317 the `+initialize' method, might not be useful because `+initialize' is
24318 only called when the first message is sent to a class object, which in
24319 some cases could be too late.
24321 Suppose for example you have a `FileStream' class that declares
24322 `Stdin', `Stdout' and `Stderr' as global variables, like below:
24325 FileStream *Stdin = nil;
24326 FileStream *Stdout = nil;
24327 FileStream *Stderr = nil;
24329 @implementation FileStream
24333 Stdin = [[FileStream new] initWithFd:0];
24334 Stdout = [[FileStream new] initWithFd:1];
24335 Stderr = [[FileStream new] initWithFd:2];
24338 /* Other methods here */
24341 In this example, the initialization of `Stdin', `Stdout' and `Stderr'
24342 in `+initialize' occurs too late. The programmer can send a message to
24343 one of these objects before the variables are actually initialized,
24344 thus sending messages to the `nil' object. The `+initialize' method
24345 which actually initializes the global variables is not invoked until
24346 the first message is sent to the class object. The solution would
24347 require these variables to be initialized just before entering `main'.
24349 The correct solution of the above problem is to use the `+load' method
24350 instead of `+initialize':
24353 @implementation FileStream
24357 Stdin = [[FileStream new] initWithFd:0];
24358 Stdout = [[FileStream new] initWithFd:1];
24359 Stderr = [[FileStream new] initWithFd:2];
24362 /* Other methods here */
24365 The `+load' is a method that is not overridden by categories. If a
24366 class and a category of it both implement `+load', both methods are
24367 invoked. This allows some additional initializations to be performed in
24370 This mechanism is not intended to be a replacement for `+initialize'.
24371 You should be aware of its limitations when you decide to use it
24372 instead of `+initialize'.
24376 * What you can and what you cannot do in +load::
24379 File: gcc.info, Node: What you can and what you cannot do in +load, Prev: Executing code before main, Up: Executing code before main
24381 7.1.1 What you can and what you cannot do in `+load'
24382 ----------------------------------------------------
24384 The `+load' implementation in the GNU runtime guarantees you the
24387 * you can write whatever C code you like;
24389 * you can send messages to Objective-C constant strings (`@"this is a
24390 constant string"');
24392 * you can allocate and send messages to objects whose class is
24393 implemented in the same file;
24395 * the `+load' implementation of all super classes of a class are
24396 executed before the `+load' of that class is executed;
24398 * the `+load' implementation of a class is executed before the
24399 `+load' implementation of any category.
24402 In particular, the following things, even if they can work in a
24403 particular case, are not guaranteed:
24405 * allocation of or sending messages to arbitrary objects;
24407 * allocation of or sending messages to objects whose classes have a
24408 category implemented in the same file;
24411 You should make no assumptions about receiving `+load' in sibling
24412 classes when you write `+load' of a class. The order in which sibling
24413 classes receive `+load' is not guaranteed.
24415 The order in which `+load' and `+initialize' are called could be
24416 problematic if this matters. If you don't allocate objects inside
24417 `+load', it is guaranteed that `+load' is called before `+initialize'.
24418 If you create an object inside `+load' the `+initialize' method of
24419 object's class is invoked even if `+load' was not invoked. Note if you
24420 explicitly call `+load' on a class, `+initialize' will be called first.
24421 To avoid possible problems try to implement only one of these methods.
24423 The `+load' method is also invoked when a bundle is dynamically loaded
24424 into your running program. This happens automatically without any
24425 intervening operation from you. When you write bundles and you need to
24426 write `+load' you can safely create and send messages to objects whose
24427 classes already exist in the running program. The same restrictions as
24428 above apply to classes defined in bundle.
24431 File: gcc.info, Node: Type encoding, Next: Garbage Collection, Prev: Executing code before main, Up: Objective-C
24436 The Objective-C compiler generates type encodings for all the types.
24437 These type encodings are used at runtime to find out information about
24438 selectors and methods and about objects and classes.
24440 The types are encoded in the following way:
24443 `unsigned char' `C'
24445 `unsigned short' `S'
24449 `unsigned long' `L'
24461 bit-fields `b' followed by the starting position of the
24462 bit-field, the type of the bit-field and the size of
24463 the bit-field (the bit-fields encoding was changed
24464 from the NeXT's compiler encoding, see below)
24466 The encoding of bit-fields has changed to allow bit-fields to be
24467 properly handled by the runtime functions that compute sizes and
24468 alignments of types that contain bit-fields. The previous encoding
24469 contained only the size of the bit-field. Using only this information
24470 it is not possible to reliably compute the size occupied by the
24471 bit-field. This is very important in the presence of the Boehm's
24472 garbage collector because the objects are allocated using the typed
24473 memory facility available in this collector. The typed memory
24474 allocation requires information about where the pointers are located
24477 The position in the bit-field is the position, counting in bits, of the
24478 bit closest to the beginning of the structure.
24480 The non-atomic types are encoded as follows:
24482 pointers `^' followed by the pointed type.
24483 arrays `[' followed by the number of elements in the array
24484 followed by the type of the elements followed by `]'
24485 structures `{' followed by the name of the structure (or `?' if the
24486 structure is unnamed), the `=' sign, the type of the
24488 unions `(' followed by the name of the structure (or `?' if the
24489 union is unnamed), the `=' sign, the type of the members
24492 Here are some types and their encodings, as they are generated by the
24493 compiler on an i386 machine:
24496 Objective-C type Compiler encoding
24498 struct { `{?=i[3f]b128i3b131i2c}'
24507 In addition to the types the compiler also encodes the type
24508 specifiers. The table below describes the encoding of the current
24509 Objective-C type specifiers:
24521 The type specifiers are encoded just before the type. Unlike types
24522 however, the type specifiers are only encoded when they appear in method
24526 File: gcc.info, Node: Garbage Collection, Next: Constant string objects, Prev: Type encoding, Up: Objective-C
24528 7.3 Garbage Collection
24529 ======================
24531 Support for a new memory management policy has been added by using a
24532 powerful conservative garbage collector, known as the
24533 Boehm-Demers-Weiser conservative garbage collector. It is available
24534 from `http://www.hpl.hp.com/personal/Hans_Boehm/gc/'.
24536 To enable the support for it you have to configure the compiler using
24537 an additional argument, `--enable-objc-gc'. You need to have garbage
24538 collector installed before building the compiler. This will build an
24539 additional runtime library which has several enhancements to support
24540 the garbage collector. The new library has a new name, `libobjc_gc.a'
24541 to not conflict with the non-garbage-collected library.
24543 When the garbage collector is used, the objects are allocated using the
24544 so-called typed memory allocation mechanism available in the
24545 Boehm-Demers-Weiser collector. This mode requires precise information
24546 on where pointers are located inside objects. This information is
24547 computed once per class, immediately after the class has been
24550 There is a new runtime function `class_ivar_set_gcinvisible()' which
24551 can be used to declare a so-called "weak pointer" reference. Such a
24552 pointer is basically hidden for the garbage collector; this can be
24553 useful in certain situations, especially when you want to keep track of
24554 the allocated objects, yet allow them to be collected. This kind of
24555 pointers can only be members of objects, you cannot declare a global
24556 pointer as a weak reference. Every type which is a pointer type can be
24557 declared a weak pointer, including `id', `Class' and `SEL'.
24559 Here is an example of how to use this feature. Suppose you want to
24560 implement a class whose instances hold a weak pointer reference; the
24561 following class does this:
24564 @interface WeakPointer : Object
24566 const void* weakPointer;
24569 - initWithPointer:(const void*)p;
24570 - (const void*)weakPointer;
24574 @implementation WeakPointer
24578 class_ivar_set_gcinvisible (self, "weakPointer", YES);
24581 - initWithPointer:(const void*)p
24587 - (const void*)weakPointer
24589 return weakPointer;
24594 Weak pointers are supported through a new type character specifier
24595 represented by the `!' character. The `class_ivar_set_gcinvisible()'
24596 function adds or removes this specifier to the string type description
24597 of the instance variable named as argument.
24600 File: gcc.info, Node: Constant string objects, Next: compatibility_alias, Prev: Garbage Collection, Up: Objective-C
24602 7.4 Constant string objects
24603 ===========================
24605 GNU Objective-C provides constant string objects that are generated
24606 directly by the compiler. You declare a constant string object by
24607 prefixing a C constant string with the character `@':
24609 id myString = @"this is a constant string object";
24611 The constant string objects are by default instances of the
24612 `NXConstantString' class which is provided by the GNU Objective-C
24613 runtime. To get the definition of this class you must include the
24614 `objc/NXConstStr.h' header file.
24616 User defined libraries may want to implement their own constant string
24617 class. To be able to support them, the GNU Objective-C compiler
24618 provides a new command line options
24619 `-fconstant-string-class=CLASS-NAME'. The provided class should adhere
24620 to a strict structure, the same as `NXConstantString''s structure:
24623 @interface MyConstantStringClass
24631 `NXConstantString' inherits from `Object'; user class libraries may
24632 choose to inherit the customized constant string class from a different
24633 class than `Object'. There is no requirement in the methods the
24634 constant string class has to implement, but the final ivar layout of
24635 the class must be the compatible with the given structure.
24637 When the compiler creates the statically allocated constant string
24638 object, the `c_string' field will be filled by the compiler with the
24639 string; the `length' field will be filled by the compiler with the
24640 string length; the `isa' pointer will be filled with `NULL' by the
24641 compiler, and it will later be fixed up automatically at runtime by the
24642 GNU Objective-C runtime library to point to the class which was set by
24643 the `-fconstant-string-class' option when the object file is loaded (if
24644 you wonder how it works behind the scenes, the name of the class to
24645 use, and the list of static objects to fixup, are stored by the
24646 compiler in the object file in a place where the GNU runtime library
24647 will find them at runtime).
24649 As a result, when a file is compiled with the
24650 `-fconstant-string-class' option, all the constant string objects will
24651 be instances of the class specified as argument to this option. It is
24652 possible to have multiple compilation units referring to different
24653 constant string classes, neither the compiler nor the linker impose any
24654 restrictions in doing this.
24657 File: gcc.info, Node: compatibility_alias, Prev: Constant string objects, Up: Objective-C
24659 7.5 compatibility_alias
24660 =======================
24662 This is a feature of the Objective-C compiler rather than of the
24663 runtime, anyway since it is documented nowhere and its existence was
24664 forgotten, we are documenting it here.
24666 The keyword `@compatibility_alias' allows you to define a class name
24667 as equivalent to another class name. For example:
24669 @compatibility_alias WOApplication GSWApplication;
24671 tells the compiler that each time it encounters `WOApplication' as a
24672 class name, it should replace it with `GSWApplication' (that is,
24673 `WOApplication' is just an alias for `GSWApplication').
24675 There are some constraints on how this can be used--
24677 * `WOApplication' (the alias) must not be an existing class;
24679 * `GSWApplication' (the real class) must be an existing class.
24683 File: gcc.info, Node: Compatibility, Next: Gcov, Prev: Objective-C, Up: Top
24685 8 Binary Compatibility
24686 **********************
24688 Binary compatibility encompasses several related concepts:
24690 "application binary interface (ABI)"
24691 The set of runtime conventions followed by all of the tools that
24692 deal with binary representations of a program, including
24693 compilers, assemblers, linkers, and language runtime support.
24694 Some ABIs are formal with a written specification, possibly
24695 designed by multiple interested parties. Others are simply the
24696 way things are actually done by a particular set of tools.
24699 A compiler conforms to an ABI if it generates code that follows
24700 all of the specifications enumerated by that ABI. A library
24701 conforms to an ABI if it is implemented according to that ABI. An
24702 application conforms to an ABI if it is built using tools that
24703 conform to that ABI and does not contain source code that
24704 specifically changes behavior specified by the ABI.
24706 "calling conventions"
24707 Calling conventions are a subset of an ABI that specify of how
24708 arguments are passed and function results are returned.
24711 Different sets of tools are interoperable if they generate files
24712 that can be used in the same program. The set of tools includes
24713 compilers, assemblers, linkers, libraries, header files, startup
24714 files, and debuggers. Binaries produced by different sets of
24715 tools are not interoperable unless they implement the same ABI.
24716 This applies to different versions of the same tools as well as
24717 tools from different vendors.
24720 Whether a function in a binary built by one set of tools can call a
24721 function in a binary built by a different set of tools is a subset
24722 of interoperability.
24724 "implementation-defined features"
24725 Language standards include lists of implementation-defined
24726 features whose behavior can vary from one implementation to
24727 another. Some of these features are normally covered by a
24728 platform's ABI and others are not. The features that are not
24729 covered by an ABI generally affect how a program behaves, but not
24733 Conformance to the same ABI and the same behavior of
24734 implementation-defined features are both relevant for
24737 The application binary interface implemented by a C or C++ compiler
24738 affects code generation and runtime support for:
24740 * size and alignment of data types
24742 * layout of structured types
24744 * calling conventions
24746 * register usage conventions
24748 * interfaces for runtime arithmetic support
24750 * object file formats
24752 In addition, the application binary interface implemented by a C++
24753 compiler affects code generation and runtime support for:
24756 * exception handling
24758 * invoking constructors and destructors
24760 * layout, alignment, and padding of classes
24762 * layout and alignment of virtual tables
24764 Some GCC compilation options cause the compiler to generate code that
24765 does not conform to the platform's default ABI. Other options cause
24766 different program behavior for implementation-defined features that are
24767 not covered by an ABI. These options are provided for consistency with
24768 other compilers that do not follow the platform's default ABI or the
24769 usual behavior of implementation-defined features for the platform. Be
24770 very careful about using such options.
24772 Most platforms have a well-defined ABI that covers C code, but ABIs
24773 that cover C++ functionality are not yet common.
24775 Starting with GCC 3.2, GCC binary conventions for C++ are based on a
24776 written, vendor-neutral C++ ABI that was designed to be specific to
24777 64-bit Itanium but also includes generic specifications that apply to
24778 any platform. This C++ ABI is also implemented by other compiler
24779 vendors on some platforms, notably GNU/Linux and BSD systems. We have
24780 tried hard to provide a stable ABI that will be compatible with future
24781 GCC releases, but it is possible that we will encounter problems that
24782 make this difficult. Such problems could include different
24783 interpretations of the C++ ABI by different vendors, bugs in the ABI, or
24784 bugs in the implementation of the ABI in different compilers. GCC's
24785 `-Wabi' switch warns when G++ generates code that is probably not
24786 compatible with the C++ ABI.
24788 The C++ library used with a C++ compiler includes the Standard C++
24789 Library, with functionality defined in the C++ Standard, plus language
24790 runtime support. The runtime support is included in a C++ ABI, but
24791 there is no formal ABI for the Standard C++ Library. Two
24792 implementations of that library are interoperable if one follows the
24793 de-facto ABI of the other and if they are both built with the same
24794 compiler, or with compilers that conform to the same ABI for C++
24795 compiler and runtime support.
24797 When G++ and another C++ compiler conform to the same C++ ABI, but the
24798 implementations of the Standard C++ Library that they normally use do
24799 not follow the same ABI for the Standard C++ Library, object files
24800 built with those compilers can be used in the same program only if they
24801 use the same C++ library. This requires specifying the location of the
24802 C++ library header files when invoking the compiler whose usual library
24803 is not being used. The location of GCC's C++ header files depends on
24804 how the GCC build was configured, but can be seen by using the G++ `-v'
24805 option. With default configuration options for G++ 3.3 the compile
24806 line for a different C++ compiler needs to include
24808 -IGCC_INSTALL_DIRECTORY/include/c++/3.3
24810 Similarly, compiling code with G++ that must use a C++ library other
24811 than the GNU C++ library requires specifying the location of the header
24812 files for that other library.
24814 The most straightforward way to link a program to use a particular C++
24815 library is to use a C++ driver that specifies that C++ library by
24816 default. The `g++' driver, for example, tells the linker where to find
24817 GCC's C++ library (`libstdc++') plus the other libraries and startup
24818 files it needs, in the proper order.
24820 If a program must use a different C++ library and it's not possible to
24821 do the final link using a C++ driver that uses that library by default,
24822 it is necessary to tell `g++' the location and name of that library.
24823 It might also be necessary to specify different startup files and other
24824 runtime support libraries, and to suppress the use of GCC's support
24825 libraries with one or more of the options `-nostdlib', `-nostartfiles',
24826 and `-nodefaultlibs'.
24829 File: gcc.info, Node: Gcov, Next: Trouble, Prev: Compatibility, Up: Top
24831 9 `gcov'--a Test Coverage Program
24832 *********************************
24834 `gcov' is a tool you can use in conjunction with GCC to test code
24835 coverage in your programs.
24839 * Gcov Intro:: Introduction to gcov.
24840 * Invoking Gcov:: How to use gcov.
24841 * Gcov and Optimization:: Using gcov with GCC optimization.
24842 * Gcov Data Files:: The files used by gcov.
24843 * Cross-profiling:: Data file relocation.
24846 File: gcc.info, Node: Gcov Intro, Next: Invoking Gcov, Up: Gcov
24848 9.1 Introduction to `gcov'
24849 ==========================
24851 `gcov' is a test coverage program. Use it in concert with GCC to
24852 analyze your programs to help create more efficient, faster running
24853 code and to discover untested parts of your program. You can use
24854 `gcov' as a profiling tool to help discover where your optimization
24855 efforts will best affect your code. You can also use `gcov' along with
24856 the other profiling tool, `gprof', to assess which parts of your code
24857 use the greatest amount of computing time.
24859 Profiling tools help you analyze your code's performance. Using a
24860 profiler such as `gcov' or `gprof', you can find out some basic
24861 performance statistics, such as:
24863 * how often each line of code executes
24865 * what lines of code are actually executed
24867 * how much computing time each section of code uses
24869 Once you know these things about how your code works when compiled, you
24870 can look at each module to see which modules should be optimized.
24871 `gcov' helps you determine where to work on optimization.
24873 Software developers also use coverage testing in concert with
24874 testsuites, to make sure software is actually good enough for a release.
24875 Testsuites can verify that a program works as expected; a coverage
24876 program tests to see how much of the program is exercised by the
24877 testsuite. Developers can then determine what kinds of test cases need
24878 to be added to the testsuites to create both better testing and a better
24881 You should compile your code without optimization if you plan to use
24882 `gcov' because the optimization, by combining some lines of code into
24883 one function, may not give you as much information as you need to look
24884 for `hot spots' where the code is using a great deal of computer time.
24885 Likewise, because `gcov' accumulates statistics by line (at the lowest
24886 resolution), it works best with a programming style that places only
24887 one statement on each line. If you use complicated macros that expand
24888 to loops or to other control structures, the statistics are less
24889 helpful--they only report on the line where the macro call appears. If
24890 your complex macros behave like functions, you can replace them with
24891 inline functions to solve this problem.
24893 `gcov' creates a logfile called `SOURCEFILE.gcov' which indicates how
24894 many times each line of a source file `SOURCEFILE.c' has executed. You
24895 can use these logfiles along with `gprof' to aid in fine-tuning the
24896 performance of your programs. `gprof' gives timing information you can
24897 use along with the information you get from `gcov'.
24899 `gcov' works only on code compiled with GCC. It is not compatible
24900 with any other profiling or test coverage mechanism.
24903 File: gcc.info, Node: Invoking Gcov, Next: Gcov and Optimization, Prev: Gcov Intro, Up: Gcov
24908 gcov [OPTIONS] SOURCEFILE
24910 `gcov' accepts the following options:
24914 Display help about using `gcov' (on the standard output), and exit
24915 without doing any further processing.
24919 Display the `gcov' version number (on the standard output), and
24920 exit without doing any further processing.
24924 Write individual execution counts for every basic block. Normally
24925 gcov outputs execution counts only for the main blocks of a line.
24926 With this option you can determine if blocks within a single line
24927 are not being executed.
24930 `--branch-probabilities'
24931 Write branch frequencies to the output file, and write branch
24932 summary info to the standard output. This option allows you to
24933 see how often each branch in your program was taken.
24934 Unconditional branches will not be shown, unless the `-u' option
24939 Write branch frequencies as the number of branches taken, rather
24940 than the percentage of branches taken.
24944 Do not create the `gcov' output file.
24947 `--long-file-names'
24948 Create long file names for included source files. For example, if
24949 the header file `x.h' contains code, and was included in the file
24950 `a.c', then running `gcov' on the file `a.c' will produce an
24951 output file called `a.c##x.h.gcov' instead of `x.h.gcov'. This
24952 can be useful if `x.h' is included in multiple source files. If
24953 you use the `-p' option, both the including and included file
24954 names will be complete path names.
24958 Preserve complete path information in the names of generated
24959 `.gcov' files. Without this option, just the filename component is
24960 used. With this option, all directories are used, with `/'
24961 characters translated to `#' characters, `.' directory components
24962 removed and `..' components renamed to `^'. This is useful if
24963 sourcefiles are in several different directories. It also affects
24967 `--function-summaries'
24968 Output summaries for each function in addition to the file level
24971 `-o DIRECTORY|FILE'
24972 `--object-directory DIRECTORY'
24973 `--object-file FILE'
24974 Specify either the directory containing the gcov data files, or the
24975 object path name. The `.gcno', and `.gcda' data files are
24976 searched for using this option. If a directory is specified, the
24977 data files are in that directory and named after the source file
24978 name, without its extension. If a file is specified here, the
24979 data files are named after that file, without its extension. If
24980 this option is not supplied, it defaults to the current directory.
24983 `--unconditional-branches'
24984 When branch probabilities are given, include those of
24985 unconditional branches. Unconditional branches are normally not
24989 `gcov' should be run with the current directory the same as that when
24990 you invoked the compiler. Otherwise it will not be able to locate the
24991 source files. `gcov' produces files called `MANGLEDNAME.gcov' in the
24992 current directory. These contain the coverage information of the
24993 source file they correspond to. One `.gcov' file is produced for each
24994 source file containing code, which was compiled to produce the data
24995 files. The MANGLEDNAME part of the output file name is usually simply
24996 the source file name, but can be something more complicated if the `-l'
24997 or `-p' options are given. Refer to those options for details.
24999 The `.gcov' files contain the `:' separated fields along with program
25000 source code. The format is
25002 EXECUTION_COUNT:LINE_NUMBER:SOURCE LINE TEXT
25004 Additional block information may succeed each line, when requested by
25005 command line option. The EXECUTION_COUNT is `-' for lines containing
25006 no code and `#####' for lines which were never executed. Some lines of
25007 information at the start have LINE_NUMBER of zero.
25009 The preamble lines are of the form
25013 The ordering and number of these preamble lines will be augmented as
25014 `gcov' development progresses -- do not rely on them remaining
25015 unchanged. Use TAG to locate a particular preamble line.
25017 The additional block information is of the form
25021 The INFORMATION is human readable, but designed to be simple enough
25022 for machine parsing too.
25024 When printing percentages, 0% and 100% are only printed when the values
25025 are _exactly_ 0% and 100% respectively. Other values which would
25026 conventionally be rounded to 0% or 100% are instead printed as the
25027 nearest non-boundary value.
25029 When using `gcov', you must first compile your program with two
25030 special GCC options: `-fprofile-arcs -ftest-coverage'. This tells the
25031 compiler to generate additional information needed by gcov (basically a
25032 flow graph of the program) and also includes additional code in the
25033 object files for generating the extra profiling information needed by
25034 gcov. These additional files are placed in the directory where the
25035 object file is located.
25037 Running the program will cause profile output to be generated. For
25038 each source file compiled with `-fprofile-arcs', an accompanying
25039 `.gcda' file will be placed in the object file directory.
25041 Running `gcov' with your program's source file names as arguments will
25042 now produce a listing of the code along with frequency of execution for
25043 each line. For example, if your program is called `tmp.c', this is
25044 what you see when you use the basic `gcov' facility:
25046 $ gcc -fprofile-arcs -ftest-coverage tmp.c
25049 90.00% of 10 source lines executed in file tmp.c
25050 Creating tmp.c.gcov.
25052 The file `tmp.c.gcov' contains output from `gcov'. Here is a sample:
25055 -: 0:Graph:tmp.gcno
25059 -: 1:#include <stdio.h>
25061 -: 3:int main (void)
25063 1: 5: int i, total;
25067 11: 9: for (i = 0; i < 10; i++)
25068 10: 10: total += i;
25070 1: 12: if (total != 45)
25071 #####: 13: printf ("Failure\n");
25073 1: 15: printf ("Success\n");
25077 When you use the `-a' option, you will get individual block counts,
25078 and the output looks like this:
25081 -: 0:Graph:tmp.gcno
25085 -: 1:#include <stdio.h>
25087 -: 3:int main (void)
25090 1: 5: int i, total;
25094 11: 9: for (i = 0; i < 10; i++)
25096 10: 10: total += i;
25099 1: 12: if (total != 45)
25101 #####: 13: printf ("Failure\n");
25104 1: 15: printf ("Success\n");
25110 In this mode, each basic block is only shown on one line - the last
25111 line of the block. A multi-line block will only contribute to the
25112 execution count of that last line, and other lines will not be shown to
25113 contain code, unless previous blocks end on those lines. The total
25114 execution count of a line is shown and subsequent lines show the
25115 execution counts for individual blocks that end on that line. After
25116 each block, the branch and call counts of the block will be shown, if
25117 the `-b' option is given.
25119 Because of the way GCC instruments calls, a call count can be shown
25120 after a line with no individual blocks. As you can see, line 13
25121 contains a basic block that was not executed.
25123 When you use the `-b' option, your output looks like this:
25126 90.00% of 10 source lines executed in file tmp.c
25127 80.00% of 5 branches executed in file tmp.c
25128 80.00% of 5 branches taken at least once in file tmp.c
25129 50.00% of 2 calls executed in file tmp.c
25130 Creating tmp.c.gcov.
25132 Here is a sample of a resulting `tmp.c.gcov' file:
25135 -: 0:Graph:tmp.gcno
25139 -: 1:#include <stdio.h>
25141 -: 3:int main (void)
25142 function main called 1 returned 1 blocks executed 75%
25144 1: 5: int i, total;
25148 11: 9: for (i = 0; i < 10; i++)
25149 branch 0 taken 91% (fallthrough)
25151 10: 10: total += i;
25153 1: 12: if (total != 45)
25154 branch 0 taken 0% (fallthrough)
25155 branch 1 taken 100%
25156 #####: 13: printf ("Failure\n");
25157 call 0 never executed
25159 1: 15: printf ("Success\n");
25160 call 0 called 1 returned 100%
25164 For each function, a line is printed showing how many times the
25165 function is called, how many times it returns and what percentage of the
25166 function's blocks were executed.
25168 For each basic block, a line is printed after the last line of the
25169 basic block describing the branch or call that ends the basic block.
25170 There can be multiple branches and calls listed for a single source
25171 line if there are multiple basic blocks that end on that line. In this
25172 case, the branches and calls are each given a number. There is no
25173 simple way to map these branches and calls back to source constructs.
25174 In general, though, the lowest numbered branch or call will correspond
25175 to the leftmost construct on the source line.
25177 For a branch, if it was executed at least once, then a percentage
25178 indicating the number of times the branch was taken divided by the
25179 number of times the branch was executed will be printed. Otherwise, the
25180 message "never executed" is printed.
25182 For a call, if it was executed at least once, then a percentage
25183 indicating the number of times the call returned divided by the number
25184 of times the call was executed will be printed. This will usually be
25185 100%, but may be less for functions that call `exit' or `longjmp', and
25186 thus may not return every time they are called.
25188 The execution counts are cumulative. If the example program were
25189 executed again without removing the `.gcda' file, the count for the
25190 number of times each line in the source was executed would be added to
25191 the results of the previous run(s). This is potentially useful in
25192 several ways. For example, it could be used to accumulate data over a
25193 number of program runs as part of a test verification suite, or to
25194 provide more accurate long-term information over a large number of
25197 The data in the `.gcda' files is saved immediately before the program
25198 exits. For each source file compiled with `-fprofile-arcs', the
25199 profiling code first attempts to read in an existing `.gcda' file; if
25200 the file doesn't match the executable (differing number of basic block
25201 counts) it will ignore the contents of the file. It then adds in the
25202 new execution counts and finally writes the data to the file.
25205 File: gcc.info, Node: Gcov and Optimization, Next: Gcov Data Files, Prev: Invoking Gcov, Up: Gcov
25207 9.3 Using `gcov' with GCC Optimization
25208 ======================================
25210 If you plan to use `gcov' to help optimize your code, you must first
25211 compile your program with two special GCC options: `-fprofile-arcs
25212 -ftest-coverage'. Aside from that, you can use any other GCC options;
25213 but if you want to prove that every single line in your program was
25214 executed, you should not compile with optimization at the same time.
25215 On some machines the optimizer can eliminate some simple code lines by
25216 combining them with other lines. For example, code like this:
25223 can be compiled into one instruction on some machines. In this case,
25224 there is no way for `gcov' to calculate separate execution counts for
25225 each line because there isn't separate code for each line. Hence the
25226 `gcov' output looks like this if you compiled the program with
25229 100: 12:if (a != b)
25234 The output shows that this block of code, combined by optimization,
25235 executed 100 times. In one sense this result is correct, because there
25236 was only one instruction representing all four of these lines. However,
25237 the output does not indicate how many times the result was 0 and how
25238 many times the result was 1.
25240 Inlineable functions can create unexpected line counts. Line counts
25241 are shown for the source code of the inlineable function, but what is
25242 shown depends on where the function is inlined, or if it is not inlined
25245 If the function is not inlined, the compiler must emit an out of line
25246 copy of the function, in any object file that needs it. If `fileA.o'
25247 and `fileB.o' both contain out of line bodies of a particular
25248 inlineable function, they will also both contain coverage counts for
25249 that function. When `fileA.o' and `fileB.o' are linked together, the
25250 linker will, on many systems, select one of those out of line bodies
25251 for all calls to that function, and remove or ignore the other.
25252 Unfortunately, it will not remove the coverage counters for the unused
25253 function body. Hence when instrumented, all but one use of that
25254 function will show zero counts.
25256 If the function is inlined in several places, the block structure in
25257 each location might not be the same. For instance, a condition might
25258 now be calculable at compile time in some instances. Because the
25259 coverage of all the uses of the inline function will be shown for the
25260 same source lines, the line counts themselves might seem inconsistent.
25263 File: gcc.info, Node: Gcov Data Files, Next: Cross-profiling, Prev: Gcov and Optimization, Up: Gcov
25265 9.4 Brief description of `gcov' data files
25266 ==========================================
25268 `gcov' uses two files for profiling. The names of these files are
25269 derived from the original _object_ file by substituting the file suffix
25270 with either `.gcno', or `.gcda'. All of these files are placed in the
25271 same directory as the object file, and contain data stored in a
25272 platform-independent format.
25274 The `.gcno' file is generated when the source file is compiled with
25275 the GCC `-ftest-coverage' option. It contains information to
25276 reconstruct the basic block graphs and assign source line numbers to
25279 The `.gcda' file is generated when a program containing object files
25280 built with the GCC `-fprofile-arcs' option is executed. A separate
25281 `.gcda' file is created for each object file compiled with this option.
25282 It contains arc transition counts, and some summary information.
25284 The full details of the file format is specified in `gcov-io.h', and
25285 functions provided in that header file should be used to access the
25289 File: gcc.info, Node: Cross-profiling, Prev: Gcov Data Files, Up: Gcov
25291 9.5 Data file relocation to support cross-profiling
25292 ===================================================
25294 Running the program will cause profile output to be generated. For each
25295 source file compiled with `-fprofile-arcs', an accompanying `.gcda'
25296 file will be placed in the object file directory. That implicitly
25297 requires running the program on the same system as it was built or
25298 having the same absolute directory structure on the target system. The
25299 program will try to create the needed directory structure, if it is not
25302 To support cross-profiling, a program compiled with `-fprofile-arcs'
25303 can relocate the data files based on two environment variables:
25305 * GCOV_PREFIX contains the prefix to add to the absolute paths in
25306 the object file. Prefix must be absolute as well, otherwise its
25307 value is ignored. The default is no prefix.
25309 * GCOV_PREFIX_STRIP indicates the how many initial directory names
25310 to strip off the hardwired absolute paths. Default value is 0.
25312 _Note:_ GCOV_PREFIX_STRIP has no effect if GCOV_PREFIX is
25313 undefined, empty or non-absolute.
25315 For example, if the object file `/user/build/foo.o' was built with
25316 `-fprofile-arcs', the final executable will try to create the data file
25317 `/user/build/foo.gcda' when running on the target system. This will
25318 fail if the corresponding directory does not exist and it is unable to
25319 create it. This can be overcome by, for example, setting the
25320 environment as `GCOV_PREFIX=/target/run' and `GCOV_PREFIX_STRIP=1'.
25321 Such a setting will name the data file `/target/run/build/foo.gcda'.
25323 You must move the data files to the expected directory tree in order to
25324 use them for profile directed optimizations (`--use-profile'), or to
25325 use the `gcov' tool.
25328 File: gcc.info, Node: Trouble, Next: Bugs, Prev: Gcov, Up: Top
25330 10 Known Causes of Trouble with GCC
25331 ***********************************
25333 This section describes known problems that affect users of GCC. Most
25334 of these are not GCC bugs per se--if they were, we would fix them. But
25335 the result for a user may be like the result of a bug.
25337 Some of these problems are due to bugs in other software, some are
25338 missing features that are too much work to add, and some are places
25339 where people's opinions differ as to what is best.
25343 * Actual Bugs:: Bugs we will fix later.
25344 * Cross-Compiler Problems:: Common problems of cross compiling with GCC.
25345 * Interoperation:: Problems using GCC with other compilers,
25346 and with certain linkers, assemblers and debuggers.
25347 * Incompatibilities:: GCC is incompatible with traditional C.
25348 * Fixed Headers:: GCC uses corrected versions of system header files.
25349 This is necessary, but doesn't always work smoothly.
25350 * Standard Libraries:: GCC uses the system C library, which might not be
25351 compliant with the ISO C standard.
25352 * Disappointments:: Regrettable things we can't change, but not quite bugs.
25353 * C++ Misunderstandings:: Common misunderstandings with GNU C++.
25354 * Protoize Caveats:: Things to watch out for when using `protoize'.
25355 * Non-bugs:: Things we think are right, but some others disagree.
25356 * Warnings and Errors:: Which problems in your code get warnings,
25357 and which get errors.
25360 File: gcc.info, Node: Actual Bugs, Next: Cross-Compiler Problems, Up: Trouble
25362 10.1 Actual Bugs We Haven't Fixed Yet
25363 =====================================
25365 * The `fixincludes' script interacts badly with automounters; if the
25366 directory of system header files is automounted, it tends to be
25367 unmounted while `fixincludes' is running. This would seem to be a
25368 bug in the automounter. We don't know any good way to work around
25371 * The `fixproto' script will sometimes add prototypes for the
25372 `sigsetjmp' and `siglongjmp' functions that reference the
25373 `jmp_buf' type before that type is defined. To work around this,
25374 edit the offending file and place the typedef in front of the
25378 File: gcc.info, Node: Cross-Compiler Problems, Next: Interoperation, Prev: Actual Bugs, Up: Trouble
25380 10.2 Cross-Compiler Problems
25381 ============================
25383 You may run into problems with cross compilation on certain machines,
25384 for several reasons.
25386 * At present, the program `mips-tfile' which adds debug support to
25387 object files on MIPS systems does not work in a cross compile
25391 File: gcc.info, Node: Interoperation, Next: Incompatibilities, Prev: Cross-Compiler Problems, Up: Trouble
25393 10.3 Interoperation
25394 ===================
25396 This section lists various difficulties encountered in using GCC
25397 together with other compilers or with the assemblers, linkers,
25398 libraries and debuggers on certain systems.
25400 * On many platforms, GCC supports a different ABI for C++ than do
25401 other compilers, so the object files compiled by GCC cannot be
25402 used with object files generated by another C++ compiler.
25404 An area where the difference is most apparent is name mangling.
25405 The use of different name mangling is intentional, to protect you
25406 from more subtle problems. Compilers differ as to many internal
25407 details of C++ implementation, including: how class instances are
25408 laid out, how multiple inheritance is implemented, and how virtual
25409 function calls are handled. If the name encoding were made the
25410 same, your programs would link against libraries provided from
25411 other compilers--but the programs would then crash when run.
25412 Incompatible libraries are then detected at link time, rather than
25415 * On some BSD systems, including some versions of Ultrix, use of
25416 profiling causes static variable destructors (currently used only
25417 in C++) not to be run.
25419 * On some SGI systems, when you use `-lgl_s' as an option, it gets
25420 translated magically to `-lgl_s -lX11_s -lc_s'. Naturally, this
25421 does not happen when you use GCC. You must specify all three
25422 options explicitly.
25424 * On a SPARC, GCC aligns all values of type `double' on an 8-byte
25425 boundary, and it expects every `double' to be so aligned. The Sun
25426 compiler usually gives `double' values 8-byte alignment, with one
25427 exception: function arguments of type `double' may not be aligned.
25429 As a result, if a function compiled with Sun CC takes the address
25430 of an argument of type `double' and passes this pointer of type
25431 `double *' to a function compiled with GCC, dereferencing the
25432 pointer may cause a fatal signal.
25434 One way to solve this problem is to compile your entire program
25435 with GCC. Another solution is to modify the function that is
25436 compiled with Sun CC to copy the argument into a local variable;
25437 local variables are always properly aligned. A third solution is
25438 to modify the function that uses the pointer to dereference it via
25439 the following function `access_double' instead of directly with
25443 access_double (double *unaligned_ptr)
25445 union d2i { double d; int i[2]; };
25447 union d2i *p = (union d2i *) unaligned_ptr;
25456 Storing into the pointer can be done likewise with the same union.
25458 * On Solaris, the `malloc' function in the `libmalloc.a' library may
25459 allocate memory that is only 4 byte aligned. Since GCC on the
25460 SPARC assumes that doubles are 8 byte aligned, this may result in a
25461 fatal signal if doubles are stored in memory allocated by the
25462 `libmalloc.a' library.
25464 The solution is to not use the `libmalloc.a' library. Use instead
25465 `malloc' and related functions from `libc.a'; they do not have
25468 * On the HP PA machine, ADB sometimes fails to work on functions
25469 compiled with GCC. Specifically, it fails to work on functions
25470 that use `alloca' or variable-size arrays. This is because GCC
25471 doesn't generate HP-UX unwind descriptors for such functions. It
25472 may even be impossible to generate them.
25474 * Debugging (`-g') is not supported on the HP PA machine, unless you
25475 use the preliminary GNU tools.
25477 * Taking the address of a label may generate errors from the HP-UX
25478 PA assembler. GAS for the PA does not have this problem.
25480 * Using floating point parameters for indirect calls to static
25481 functions will not work when using the HP assembler. There simply
25482 is no way for GCC to specify what registers hold arguments for
25483 static functions when using the HP assembler. GAS for the PA does
25484 not have this problem.
25486 * In extremely rare cases involving some very large functions you may
25487 receive errors from the HP linker complaining about an out of
25488 bounds unconditional branch offset. This used to occur more often
25489 in previous versions of GCC, but is now exceptionally rare. If
25490 you should run into it, you can work around by making your
25493 * GCC compiled code sometimes emits warnings from the HP-UX
25494 assembler of the form:
25496 (warning) Use of GR3 when
25497 frame >= 8192 may cause conflict.
25499 These warnings are harmless and can be safely ignored.
25501 * In extremely rare cases involving some very large functions you may
25502 receive errors from the AIX Assembler complaining about a
25503 displacement that is too large. If you should run into it, you
25504 can work around by making your function smaller.
25506 * The `libstdc++.a' library in GCC relies on the SVR4 dynamic linker
25507 semantics which merges global symbols between libraries and
25508 applications, especially necessary for C++ streams functionality.
25509 This is not the default behavior of AIX shared libraries and
25510 dynamic linking. `libstdc++.a' is built on AIX with
25511 "runtime-linking" enabled so that symbol merging can occur. To
25512 utilize this feature, the application linked with `libstdc++.a'
25513 must include the `-Wl,-brtl' flag on the link line. G++ cannot
25514 impose this because this option may interfere with the semantics
25515 of the user program and users may not always use `g++' to link his
25516 or her application. Applications are not required to use the
25517 `-Wl,-brtl' flag on the link line--the rest of the `libstdc++.a'
25518 library which is not dependent on the symbol merging semantics
25519 will continue to function correctly.
25521 * An application can interpose its own definition of functions for
25522 functions invoked by `libstdc++.a' with "runtime-linking" enabled
25523 on AIX. To accomplish this the application must be linked with
25524 "runtime-linking" option and the functions explicitly must be
25525 exported by the application (`-Wl,-brtl,-bE:exportfile').
25527 * AIX on the RS/6000 provides support (NLS) for environments outside
25528 of the United States. Compilers and assemblers use NLS to support
25529 locale-specific representations of various objects including
25530 floating-point numbers (`.' vs `,' for separating decimal
25531 fractions). There have been problems reported where the library
25532 linked with GCC does not produce the same floating-point formats
25533 that the assembler accepts. If you have this problem, set the
25534 `LANG' environment variable to `C' or `En_US'.
25536 * Even if you specify `-fdollars-in-identifiers', you cannot
25537 successfully use `$' in identifiers on the RS/6000 due to a
25538 restriction in the IBM assembler. GAS supports these identifiers.
25540 * On Ultrix, the Fortran compiler expects registers 2 through 5 to
25541 be saved by function calls. However, the C compiler uses
25542 conventions compatible with BSD Unix: registers 2 through 5 may be
25543 clobbered by function calls.
25545 GCC uses the same convention as the Ultrix C compiler. You can use
25546 these options to produce code compatible with the Fortran compiler:
25548 -fcall-saved-r2 -fcall-saved-r3 -fcall-saved-r4 -fcall-saved-r5
25551 File: gcc.info, Node: Incompatibilities, Next: Fixed Headers, Prev: Interoperation, Up: Trouble
25553 10.4 Incompatibilities of GCC
25554 =============================
25556 There are several noteworthy incompatibilities between GNU C and K&R
25557 (non-ISO) versions of C.
25559 * GCC normally makes string constants read-only. If several
25560 identical-looking string constants are used, GCC stores only one
25561 copy of the string.
25563 One consequence is that you cannot call `mktemp' with a string
25564 constant argument. The function `mktemp' always alters the string
25565 its argument points to.
25567 Another consequence is that `sscanf' does not work on some very
25568 old systems when passed a string constant as its format control
25569 string or input. This is because `sscanf' incorrectly tries to
25570 write into the string constant. Likewise `fscanf' and `scanf'.
25572 The solution to these problems is to change the program to use
25573 `char'-array variables with initialization strings for these
25574 purposes instead of string constants.
25576 * `-2147483648' is positive.
25578 This is because 2147483648 cannot fit in the type `int', so
25579 (following the ISO C rules) its data type is `unsigned long int'.
25580 Negating this value yields 2147483648 again.
25582 * GCC does not substitute macro arguments when they appear inside of
25583 string constants. For example, the following macro in GCC
25587 will produce output `"a"' regardless of what the argument A is.
25589 * When you use `setjmp' and `longjmp', the only automatic variables
25590 guaranteed to remain valid are those declared `volatile'. This is
25591 a consequence of automatic register allocation. Consider this
25605 /* `longjmp (j)' may occur in `fun3'. */
25606 return a + fun3 ();
25609 Here `a' may or may not be restored to its first value when the
25610 `longjmp' occurs. If `a' is allocated in a register, then its
25611 first value is restored; otherwise, it keeps the last value stored
25614 If you use the `-W' option with the `-O' option, you will get a
25615 warning when GCC thinks such a problem might be possible.
25617 * Programs that use preprocessing directives in the middle of macro
25618 arguments do not work with GCC. For example, a program like this
25625 ISO C does not permit such a construct.
25627 * K&R compilers allow comments to cross over an inclusion boundary
25628 (i.e. started in an include file and ended in the including file).
25630 * Declarations of external variables and functions within a block
25631 apply only to the block containing the declaration. In other
25632 words, they have the same scope as any other declaration in the
25635 In some other C compilers, a `extern' declaration affects all the
25636 rest of the file even if it happens within a block.
25638 * In traditional C, you can combine `long', etc., with a typedef
25639 name, as shown here:
25642 typedef long foo bar;
25644 In ISO C, this is not allowed: `long' and other type modifiers
25645 require an explicit `int'.
25647 * PCC allows typedef names to be used as function parameters.
25649 * Traditional C allows the following erroneous pair of declarations
25650 to appear together in a given scope:
25655 * GCC treats all characters of identifiers as significant.
25656 According to K&R-1 (2.2), "No more than the first eight characters
25657 are significant, although more may be used.". Also according to
25658 K&R-1 (2.2), "An identifier is a sequence of letters and digits;
25659 the first character must be a letter. The underscore _ counts as
25660 a letter.", but GCC also allows dollar signs in identifiers.
25662 * PCC allows whitespace in the middle of compound assignment
25663 operators such as `+='. GCC, following the ISO standard, does not
25666 * GCC complains about unterminated character constants inside of
25667 preprocessing conditionals that fail. Some programs have English
25668 comments enclosed in conditionals that are guaranteed to fail; if
25669 these comments contain apostrophes, GCC will probably report an
25670 error. For example, this code would produce an error:
25673 You can't expect this to work.
25676 The best solution to such a problem is to put the text into an
25677 actual C comment delimited by `/*...*/'.
25679 * Many user programs contain the declaration `long time ();'. In the
25680 past, the system header files on many systems did not actually
25681 declare `time', so it did not matter what type your program
25682 declared it to return. But in systems with ISO C headers, `time'
25683 is declared to return `time_t', and if that is not the same as
25684 `long', then `long time ();' is erroneous.
25686 The solution is to change your program to use appropriate system
25687 headers (`<time.h>' on systems with ISO C headers) and not to
25688 declare `time' if the system header files declare it, or failing
25689 that to use `time_t' as the return type of `time'.
25691 * When compiling functions that return `float', PCC converts it to a
25692 double. GCC actually returns a `float'. If you are concerned
25693 with PCC compatibility, you should declare your functions to return
25694 `double'; you might as well say what you mean.
25696 * When compiling functions that return structures or unions, GCC
25697 output code normally uses a method different from that used on most
25698 versions of Unix. As a result, code compiled with GCC cannot call
25699 a structure-returning function compiled with PCC, and vice versa.
25701 The method used by GCC is as follows: a structure or union which is
25702 1, 2, 4 or 8 bytes long is returned like a scalar. A structure or
25703 union with any other size is stored into an address supplied by
25704 the caller (usually in a special, fixed register, but on some
25705 machines it is passed on the stack). The target hook
25706 `TARGET_STRUCT_VALUE_RTX' tells GCC where to pass this address.
25708 By contrast, PCC on most target machines returns structures and
25709 unions of any size by copying the data into an area of static
25710 storage, and then returning the address of that storage as if it
25711 were a pointer value. The caller must copy the data from that
25712 memory area to the place where the value is wanted. GCC does not
25713 use this method because it is slower and nonreentrant.
25715 On some newer machines, PCC uses a reentrant convention for all
25716 structure and union returning. GCC on most of these machines uses
25717 a compatible convention when returning structures and unions in
25718 memory, but still returns small structures and unions in registers.
25720 You can tell GCC to use a compatible convention for all structure
25721 and union returning with the option `-fpcc-struct-return'.
25723 * GCC complains about program fragments such as `0x74ae-0x4000'
25724 which appear to be two hexadecimal constants separated by the minus
25725 operator. Actually, this string is a single "preprocessing token".
25726 Each such token must correspond to one token in C. Since this
25727 does not, GCC prints an error message. Although it may appear
25728 obvious that what is meant is an operator and two values, the ISO
25729 C standard specifically requires that this be treated as erroneous.
25731 A "preprocessing token" is a "preprocessing number" if it begins
25732 with a digit and is followed by letters, underscores, digits,
25733 periods and `e+', `e-', `E+', `E-', `p+', `p-', `P+', or `P-'
25734 character sequences. (In strict C89 mode, the sequences `p+',
25735 `p-', `P+' and `P-' cannot appear in preprocessing numbers.)
25737 To make the above program fragment valid, place whitespace in
25738 front of the minus sign. This whitespace will end the
25739 preprocessing number.
25742 File: gcc.info, Node: Fixed Headers, Next: Standard Libraries, Prev: Incompatibilities, Up: Trouble
25744 10.5 Fixed Header Files
25745 =======================
25747 GCC needs to install corrected versions of some system header files.
25748 This is because most target systems have some header files that won't
25749 work with GCC unless they are changed. Some have bugs, some are
25750 incompatible with ISO C, and some depend on special features of other
25753 Installing GCC automatically creates and installs the fixed header
25754 files, by running a program called `fixincludes'. Normally, you don't
25755 need to pay attention to this. But there are cases where it doesn't do
25756 the right thing automatically.
25758 * If you update the system's header files, such as by installing a
25759 new system version, the fixed header files of GCC are not
25760 automatically updated. They can be updated using the `mkheaders'
25761 script installed in `LIBEXECDIR/gcc/TARGET/VERSION/install-tools/'.
25763 * On some systems, header file directories contain machine-specific
25764 symbolic links in certain places. This makes it possible to share
25765 most of the header files among hosts running the same version of
25766 the system on different machine models.
25768 The programs that fix the header files do not understand this
25769 special way of using symbolic links; therefore, the directory of
25770 fixed header files is good only for the machine model used to
25773 It is possible to make separate sets of fixed header files for the
25774 different machine models, and arrange a structure of symbolic
25775 links so as to use the proper set, but you'll have to do this by
25779 File: gcc.info, Node: Standard Libraries, Next: Disappointments, Prev: Fixed Headers, Up: Trouble
25781 10.6 Standard Libraries
25782 =======================
25784 GCC by itself attempts to be a conforming freestanding implementation.
25785 *Note Language Standards Supported by GCC: Standards, for details of
25786 what this means. Beyond the library facilities required of such an
25787 implementation, the rest of the C library is supplied by the vendor of
25788 the operating system. If that C library doesn't conform to the C
25789 standards, then your programs might get warnings (especially when using
25790 `-Wall') that you don't expect.
25792 For example, the `sprintf' function on SunOS 4.1.3 returns `char *'
25793 while the C standard says that `sprintf' returns an `int'. The
25794 `fixincludes' program could make the prototype for this function match
25795 the Standard, but that would be wrong, since the function will still
25798 If you need a Standard compliant library, then you need to find one, as
25799 GCC does not provide one. The GNU C library (called `glibc') provides
25800 ISO C, POSIX, BSD, SystemV and X/Open compatibility for GNU/Linux and
25801 HURD-based GNU systems; no recent version of it supports other systems,
25802 though some very old versions did. Version 2.2 of the GNU C library
25803 includes nearly complete C99 support. You could also ask your
25804 operating system vendor if newer libraries are available.
25807 File: gcc.info, Node: Disappointments, Next: C++ Misunderstandings, Prev: Standard Libraries, Up: Trouble
25809 10.7 Disappointments and Misunderstandings
25810 ==========================================
25812 These problems are perhaps regrettable, but we don't know any practical
25815 * Certain local variables aren't recognized by debuggers when you
25816 compile with optimization.
25818 This occurs because sometimes GCC optimizes the variable out of
25819 existence. There is no way to tell the debugger how to compute the
25820 value such a variable "would have had", and it is not clear that
25821 would be desirable anyway. So GCC simply does not mention the
25822 eliminated variable when it writes debugging information.
25824 You have to expect a certain amount of disagreement between the
25825 executable and your source code, when you use optimization.
25827 * Users often think it is a bug when GCC reports an error for code
25830 int foo (struct mumble *);
25832 struct mumble { ... };
25834 int foo (struct mumble *x)
25837 This code really is erroneous, because the scope of `struct
25838 mumble' in the prototype is limited to the argument list
25839 containing it. It does not refer to the `struct mumble' defined
25840 with file scope immediately below--they are two unrelated types
25841 with similar names in different scopes.
25843 But in the definition of `foo', the file-scope type is used
25844 because that is available to be inherited. Thus, the definition
25845 and the prototype do not match, and you get an error.
25847 This behavior may seem silly, but it's what the ISO standard
25848 specifies. It is easy enough for you to make your code work by
25849 moving the definition of `struct mumble' above the prototype.
25850 It's not worth being incompatible with ISO C just to avoid an
25851 error for the example shown above.
25853 * Accesses to bit-fields even in volatile objects works by accessing
25854 larger objects, such as a byte or a word. You cannot rely on what
25855 size of object is accessed in order to read or write the
25856 bit-field; it may even vary for a given bit-field according to the
25859 If you care about controlling the amount of memory that is
25860 accessed, use volatile but do not use bit-fields.
25862 * GCC comes with shell scripts to fix certain known problems in
25863 system header files. They install corrected copies of various
25864 header files in a special directory where only GCC will normally
25865 look for them. The scripts adapt to various systems by searching
25866 all the system header files for the problem cases that we know
25869 If new system header files are installed, nothing automatically
25870 arranges to update the corrected header files. They can be
25871 updated using the `mkheaders' script installed in
25872 `LIBEXECDIR/gcc/TARGET/VERSION/install-tools/'.
25874 * On 68000 and x86 systems, for instance, you can get paradoxical
25875 results if you test the precise values of floating point numbers.
25876 For example, you can find that a floating point value which is not
25877 a NaN is not equal to itself. This results from the fact that the
25878 floating point registers hold a few more bits of precision than
25879 fit in a `double' in memory. Compiled code moves values between
25880 memory and floating point registers at its convenience, and moving
25881 them into memory truncates them.
25883 You can partially avoid this problem by using the `-ffloat-store'
25884 option (*note Optimize Options::).
25886 * On AIX and other platforms without weak symbol support, templates
25887 need to be instantiated explicitly and symbols for static members
25888 of templates will not be generated.
25890 * On AIX, GCC scans object files and library archives for static
25891 constructors and destructors when linking an application before the
25892 linker prunes unreferenced symbols. This is necessary to prevent
25893 the AIX linker from mistakenly assuming that static constructor or
25894 destructor are unused and removing them before the scanning can
25895 occur. All static constructors and destructors found will be
25896 referenced even though the modules in which they occur may not be
25897 used by the program. This may lead to both increased executable
25898 size and unexpected symbol references.
25901 File: gcc.info, Node: C++ Misunderstandings, Next: Protoize Caveats, Prev: Disappointments, Up: Trouble
25903 10.8 Common Misunderstandings with GNU C++
25904 ==========================================
25906 C++ is a complex language and an evolving one, and its standard
25907 definition (the ISO C++ standard) was only recently completed. As a
25908 result, your C++ compiler may occasionally surprise you, even when its
25909 behavior is correct. This section discusses some areas that frequently
25910 give rise to questions of this sort.
25914 * Static Definitions:: Static member declarations are not definitions
25915 * Name lookup:: Name lookup, templates, and accessing members of base classes
25916 * Temporaries:: Temporaries may vanish before you expect
25917 * Copy Assignment:: Copy Assignment operators copy virtual bases twice
25920 File: gcc.info, Node: Static Definitions, Next: Name lookup, Up: C++ Misunderstandings
25922 10.8.1 Declare _and_ Define Static Members
25923 ------------------------------------------
25925 When a class has static data members, it is not enough to _declare_ the
25926 static member; you must also _define_ it. For example:
25935 This declaration only establishes that the class `Foo' has an `int'
25936 named `Foo::bar', and a member function named `Foo::method'. But you
25937 still need to define _both_ `method' and `bar' elsewhere. According to
25938 the ISO standard, you must supply an initializer in one (and only one)
25939 source file, such as:
25943 Other C++ compilers may not correctly implement the standard behavior.
25944 As a result, when you switch to `g++' from one of these compilers, you
25945 may discover that a program that appeared to work correctly in fact
25946 does not conform to the standard: `g++' reports as undefined symbols
25947 any static data members that lack definitions.
25950 File: gcc.info, Node: Name lookup, Next: Temporaries, Prev: Static Definitions, Up: C++ Misunderstandings
25952 10.8.2 Name lookup, templates, and accessing members of base classes
25953 --------------------------------------------------------------------
25955 The C++ standard prescribes that all names that are not dependent on
25956 template parameters are bound to their present definitions when parsing
25957 a template function or class.(1) Only names that are dependent are
25958 looked up at the point of instantiation. For example, consider
25963 template <typename T>
25972 static const int N;
25975 Here, the names `foo' and `N' appear in a context that does not depend
25976 on the type of `T'. The compiler will thus require that they are
25977 defined in the context of use in the template, not only before the
25978 point of instantiation, and will here use `::foo(double)' and `A::N',
25979 respectively. In particular, it will convert the integer value to a
25980 `double' when passing it to `::foo(double)'.
25982 Conversely, `bar' and the call to `foo' in the fourth marked line are
25983 used in contexts that do depend on the type of `T', so they are only
25984 looked up at the point of instantiation, and you can provide
25985 declarations for them after declaring the template, but before
25986 instantiating it. In particular, if you instantiate `A::f<int>', the
25987 last line will call an overloaded `::foo(int)' if one was provided,
25988 even if after the declaration of `struct A'.
25990 This distinction between lookup of dependent and non-dependent names is
25991 called two-stage (or dependent) name lookup. G++ implements it since
25994 Two-stage name lookup sometimes leads to situations with behavior
25995 different from non-template codes. The most common is probably this:
25997 template <typename T> struct Base {
26001 template <typename T> struct Derived : public Base<T> {
26002 int get_i() { return i; }
26005 In `get_i()', `i' is not used in a dependent context, so the compiler
26006 will look for a name declared at the enclosing namespace scope (which
26007 is the global scope here). It will not look into the base class, since
26008 that is dependent and you may declare specializations of `Base' even
26009 after declaring `Derived', so the compiler can't really know what `i'
26010 would refer to. If there is no global variable `i', then you will get
26013 In order to make it clear that you want the member of the base class,
26014 you need to defer lookup until instantiation time, at which the base
26015 class is known. For this, you need to access `i' in a dependent
26016 context, by either using `this->i' (remember that `this' is of type
26017 `Derived<T>*', so is obviously dependent), or using `Base<T>::i'.
26018 Alternatively, `Base<T>::i' might be brought into scope by a
26019 `using'-declaration.
26021 Another, similar example involves calling member functions of a base
26024 template <typename T> struct Base {
26028 template <typename T> struct Derived : Base<T> {
26029 int g() { return f(); };
26032 Again, the call to `f()' is not dependent on template arguments (there
26033 are no arguments that depend on the type `T', and it is also not
26034 otherwise specified that the call should be in a dependent context).
26035 Thus a global declaration of such a function must be available, since
26036 the one in the base class is not visible until instantiation time. The
26037 compiler will consequently produce the following error message:
26039 x.cc: In member function `int Derived<T>::g()':
26040 x.cc:6: error: there are no arguments to `f' that depend on a template
26041 parameter, so a declaration of `f' must be available
26042 x.cc:6: error: (if you use `-fpermissive', G++ will accept your code, but
26043 allowing the use of an undeclared name is deprecated)
26045 To make the code valid either use `this->f()', or `Base<T>::f()'.
26046 Using the `-fpermissive' flag will also let the compiler accept the
26047 code, by marking all function calls for which no declaration is visible
26048 at the time of definition of the template for later lookup at
26049 instantiation time, as if it were a dependent call. We do not
26050 recommend using `-fpermissive' to work around invalid code, and it will
26051 also only catch cases where functions in base classes are called, not
26052 where variables in base classes are used (as in the example above).
26054 Note that some compilers (including G++ versions prior to 3.4) get
26055 these examples wrong and accept above code without an error. Those
26056 compilers do not implement two-stage name lookup correctly.
26058 ---------- Footnotes ----------
26060 (1) The C++ standard just uses the term "dependent" for names that
26061 depend on the type or value of template parameters. This shorter term
26062 will also be used in the rest of this section.
26065 File: gcc.info, Node: Temporaries, Next: Copy Assignment, Prev: Name lookup, Up: C++ Misunderstandings
26067 10.8.3 Temporaries May Vanish Before You Expect
26068 -----------------------------------------------
26070 It is dangerous to use pointers or references to _portions_ of a
26071 temporary object. The compiler may very well delete the object before
26072 you expect it to, leaving a pointer to garbage. The most common place
26073 where this problem crops up is in classes like string classes,
26074 especially ones that define a conversion function to type `char *' or
26075 `const char *'--which is one reason why the standard `string' class
26076 requires you to call the `c_str' member function. However, any class
26077 that returns a pointer to some internal structure is potentially
26078 subject to this problem.
26080 For example, a program may use a function `strfunc' that returns
26081 `string' objects, and another function `charfunc' that operates on
26082 pointers to `char':
26085 void charfunc (const char *);
26090 const char *p = strfunc().c_str();
26097 In this situation, it may seem reasonable to save a pointer to the C
26098 string returned by the `c_str' member function and use that rather than
26099 call `c_str' repeatedly. However, the temporary string created by the
26100 call to `strfunc' is destroyed after `p' is initialized, at which point
26101 `p' is left pointing to freed memory.
26103 Code like this may run successfully under some other compilers,
26104 particularly obsolete cfront-based compilers that delete temporaries
26105 along with normal local variables. However, the GNU C++ behavior is
26106 standard-conforming, so if your program depends on late destruction of
26107 temporaries it is not portable.
26109 The safe way to write such code is to give the temporary a name, which
26110 forces it to remain until the end of the scope of the name. For
26113 const string& tmp = strfunc ();
26114 charfunc (tmp.c_str ());
26117 File: gcc.info, Node: Copy Assignment, Prev: Temporaries, Up: C++ Misunderstandings
26119 10.8.4 Implicit Copy-Assignment for Virtual Bases
26120 -------------------------------------------------
26122 When a base class is virtual, only one subobject of the base class
26123 belongs to each full object. Also, the constructors and destructors are
26124 invoked only once, and called from the most-derived class. However,
26125 such objects behave unspecified when being assigned. For example:
26129 Base(char *n) : name(strdup(n)){}
26130 Base& operator= (const Base& other){
26132 name = strdup (other.name);
26136 struct A:virtual Base{
26141 struct B:virtual Base{
26146 struct Derived:public A, public B{
26147 Derived():Base("Derived"){}
26150 void func(Derived &d1, Derived &d2)
26155 The C++ standard specifies that `Base::Base' is only called once when
26156 constructing or copy-constructing a Derived object. It is unspecified
26157 whether `Base::operator=' is called more than once when the implicit
26158 copy-assignment for Derived objects is invoked (as it is inside `func'
26161 G++ implements the "intuitive" algorithm for copy-assignment: assign
26162 all direct bases, then assign all members. In that algorithm, the
26163 virtual base subobject can be encountered more than once. In the
26164 example, copying proceeds in the following order: `val', `name' (via
26165 `strdup'), `bval', and `name' again.
26167 If application code relies on copy-assignment, a user-defined
26168 copy-assignment operator removes any uncertainties. With such an
26169 operator, the application can define whether and how the virtual base
26170 subobject is assigned.
26173 File: gcc.info, Node: Protoize Caveats, Next: Non-bugs, Prev: C++ Misunderstandings, Up: Trouble
26175 10.9 Caveats of using `protoize'
26176 ================================
26178 The conversion programs `protoize' and `unprotoize' can sometimes
26179 change a source file in a way that won't work unless you rearrange it.
26181 * `protoize' can insert references to a type name or type tag before
26182 the definition, or in a file where they are not defined.
26184 If this happens, compiler error messages should show you where the
26185 new references are, so fixing the file by hand is straightforward.
26187 * There are some C constructs which `protoize' cannot figure out.
26188 For example, it can't determine argument types for declaring a
26189 pointer-to-function variable; this you must do by hand. `protoize'
26190 inserts a comment containing `???' each time it finds such a
26191 variable; so you can find all such variables by searching for this
26192 string. ISO C does not require declaring the argument types of
26193 pointer-to-function types.
26195 * Using `unprotoize' can easily introduce bugs. If the program
26196 relied on prototypes to bring about conversion of arguments, these
26197 conversions will not take place in the program without prototypes.
26198 One case in which you can be sure `unprotoize' is safe is when you
26199 are removing prototypes that were made with `protoize'; if the
26200 program worked before without any prototypes, it will work again
26203 You can find all the places where this problem might occur by
26204 compiling the program with the `-Wconversion' option. It prints a
26205 warning whenever an argument is converted.
26207 * Both conversion programs can be confused if there are macro calls
26208 in and around the text to be converted. In other words, the
26209 standard syntax for a declaration or definition must not result
26210 from expanding a macro. This problem is inherent in the design of
26211 C and cannot be fixed. If only a few functions have confusing
26212 macro calls, you can easily convert them manually.
26214 * `protoize' cannot get the argument types for a function whose
26215 definition was not actually compiled due to preprocessing
26216 conditionals. When this happens, `protoize' changes nothing in
26217 regard to such a function. `protoize' tries to detect such
26218 instances and warn about them.
26220 You can generally work around this problem by using `protoize' step
26221 by step, each time specifying a different set of `-D' options for
26222 compilation, until all of the functions have been converted.
26223 There is no automatic way to verify that you have got them all,
26226 * Confusion may result if there is an occasion to convert a function
26227 declaration or definition in a region of source code where there
26228 is more than one formal parameter list present. Thus, attempts to
26229 convert code containing multiple (conditionally compiled) versions
26230 of a single function header (in the same vicinity) may not produce
26231 the desired (or expected) results.
26233 If you plan on converting source files which contain such code, it
26234 is recommended that you first make sure that each conditionally
26235 compiled region of source code which contains an alternative
26236 function header also contains at least one additional follower
26237 token (past the final right parenthesis of the function header).
26238 This should circumvent the problem.
26240 * `unprotoize' can become confused when trying to convert a function
26241 definition or declaration which contains a declaration for a
26242 pointer-to-function formal argument which has the same name as the
26243 function being defined or declared. We recommend you avoid such
26244 choices of formal parameter names.
26246 * You might also want to correct some of the indentation by hand and
26247 break long lines. (The conversion programs don't write lines
26248 longer than eighty characters in any case.)
26251 File: gcc.info, Node: Non-bugs, Next: Warnings and Errors, Prev: Protoize Caveats, Up: Trouble
26253 10.10 Certain Changes We Don't Want to Make
26254 ===========================================
26256 This section lists changes that people frequently request, but which we
26257 do not make because we think GCC is better without them.
26259 * Checking the number and type of arguments to a function which has
26260 an old-fashioned definition and no prototype.
26262 Such a feature would work only occasionally--only for calls that
26263 appear in the same file as the called function, following the
26264 definition. The only way to check all calls reliably is to add a
26265 prototype for the function. But adding a prototype eliminates the
26266 motivation for this feature. So the feature is not worthwhile.
26268 * Warning about using an expression whose type is signed as a shift
26271 Shift count operands are probably signed more often than unsigned.
26272 Warning about this would cause far more annoyance than good.
26274 * Warning about assigning a signed value to an unsigned variable.
26276 Such assignments must be very common; warning about them would
26277 cause more annoyance than good.
26279 * Warning when a non-void function value is ignored.
26281 C contains many standard functions that return a value that most
26282 programs choose to ignore. One obvious example is `printf'.
26283 Warning about this practice only leads the defensive programmer to
26284 clutter programs with dozens of casts to `void'. Such casts are
26285 required so frequently that they become visual noise. Writing
26286 those casts becomes so automatic that they no longer convey useful
26287 information about the intentions of the programmer. For functions
26288 where the return value should never be ignored, use the
26289 `warn_unused_result' function attribute (*note Function
26292 * Making `-fshort-enums' the default.
26294 This would cause storage layout to be incompatible with most other
26295 C compilers. And it doesn't seem very important, given that you
26296 can get the same result in other ways. The case where it matters
26297 most is when the enumeration-valued object is inside a structure,
26298 and in that case you can specify a field width explicitly.
26300 * Making bit-fields unsigned by default on particular machines where
26301 "the ABI standard" says to do so.
26303 The ISO C standard leaves it up to the implementation whether a
26304 bit-field declared plain `int' is signed or not. This in effect
26305 creates two alternative dialects of C.
26307 The GNU C compiler supports both dialects; you can specify the
26308 signed dialect with `-fsigned-bitfields' and the unsigned dialect
26309 with `-funsigned-bitfields'. However, this leaves open the
26310 question of which dialect to use by default.
26312 Currently, the preferred dialect makes plain bit-fields signed,
26313 because this is simplest. Since `int' is the same as `signed int'
26314 in every other context, it is cleanest for them to be the same in
26315 bit-fields as well.
26317 Some computer manufacturers have published Application Binary
26318 Interface standards which specify that plain bit-fields should be
26319 unsigned. It is a mistake, however, to say anything about this
26320 issue in an ABI. This is because the handling of plain bit-fields
26321 distinguishes two dialects of C. Both dialects are meaningful on
26322 every type of machine. Whether a particular object file was
26323 compiled using signed bit-fields or unsigned is of no concern to
26324 other object files, even if they access the same bit-fields in the
26325 same data structures.
26327 A given program is written in one or the other of these two
26328 dialects. The program stands a chance to work on most any machine
26329 if it is compiled with the proper dialect. It is unlikely to work
26330 at all if compiled with the wrong dialect.
26332 Many users appreciate the GNU C compiler because it provides an
26333 environment that is uniform across machines. These users would be
26334 inconvenienced if the compiler treated plain bit-fields
26335 differently on certain machines.
26337 Occasionally users write programs intended only for a particular
26338 machine type. On these occasions, the users would benefit if the
26339 GNU C compiler were to support by default the same dialect as the
26340 other compilers on that machine. But such applications are rare.
26341 And users writing a program to run on more than one type of
26342 machine cannot possibly benefit from this kind of compatibility.
26344 This is why GCC does and will treat plain bit-fields in the same
26345 fashion on all types of machines (by default).
26347 There are some arguments for making bit-fields unsigned by default
26348 on all machines. If, for example, this becomes a universal de
26349 facto standard, it would make sense for GCC to go along with it.
26350 This is something to be considered in the future.
26352 (Of course, users strongly concerned about portability should
26353 indicate explicitly in each bit-field whether it is signed or not.
26354 In this way, they write programs which have the same meaning in
26357 * Undefining `__STDC__' when `-ansi' is not used.
26359 Currently, GCC defines `__STDC__' unconditionally. This provides
26360 good results in practice.
26362 Programmers normally use conditionals on `__STDC__' to ask whether
26363 it is safe to use certain features of ISO C, such as function
26364 prototypes or ISO token concatenation. Since plain `gcc' supports
26365 all the features of ISO C, the correct answer to these questions is
26368 Some users try to use `__STDC__' to check for the availability of
26369 certain library facilities. This is actually incorrect usage in
26370 an ISO C program, because the ISO C standard says that a conforming
26371 freestanding implementation should define `__STDC__' even though it
26372 does not have the library facilities. `gcc -ansi -pedantic' is a
26373 conforming freestanding implementation, and it is therefore
26374 required to define `__STDC__', even though it does not come with
26377 Sometimes people say that defining `__STDC__' in a compiler that
26378 does not completely conform to the ISO C standard somehow violates
26379 the standard. This is illogical. The standard is a standard for
26380 compilers that claim to support ISO C, such as `gcc -ansi'--not
26381 for other compilers such as plain `gcc'. Whatever the ISO C
26382 standard says is relevant to the design of plain `gcc' without
26383 `-ansi' only for pragmatic reasons, not as a requirement.
26385 GCC normally defines `__STDC__' to be 1, and in addition defines
26386 `__STRICT_ANSI__' if you specify the `-ansi' option, or a `-std'
26387 option for strict conformance to some version of ISO C. On some
26388 hosts, system include files use a different convention, where
26389 `__STDC__' is normally 0, but is 1 if the user specifies strict
26390 conformance to the C Standard. GCC follows the host convention
26391 when processing system include files, but when processing user
26392 files it follows the usual GNU C convention.
26394 * Undefining `__STDC__' in C++.
26396 Programs written to compile with C++-to-C translators get the
26397 value of `__STDC__' that goes with the C compiler that is
26398 subsequently used. These programs must test `__STDC__' to
26399 determine what kind of C preprocessor that compiler uses: whether
26400 they should concatenate tokens in the ISO C fashion or in the
26401 traditional fashion.
26403 These programs work properly with GNU C++ if `__STDC__' is defined.
26404 They would not work otherwise.
26406 In addition, many header files are written to provide prototypes
26407 in ISO C but not in traditional C. Many of these header files can
26408 work without change in C++ provided `__STDC__' is defined. If
26409 `__STDC__' is not defined, they will all fail, and will all need
26410 to be changed to test explicitly for C++ as well.
26412 * Deleting "empty" loops.
26414 Historically, GCC has not deleted "empty" loops under the
26415 assumption that the most likely reason you would put one in a
26416 program is to have a delay, so deleting them will not make real
26417 programs run any faster.
26419 However, the rationale here is that optimization of a nonempty loop
26420 cannot produce an empty one. This held for carefully written C
26421 compiled with less powerful optimizers but is not always the case
26422 for carefully written C++ or with more powerful optimizers. Thus
26423 GCC will remove operations from loops whenever it can determine
26424 those operations are not externally visible (apart from the time
26425 taken to execute them, of course). In case the loop can be proved
26426 to be finite, GCC will also remove the loop itself.
26428 Be aware of this when performing timing tests, for instance the
26429 following loop can be completely removed, provided
26430 `some_expression' can provably not change any global state.
26436 for (ix = 0; ix != 10000; ix++)
26437 sum += some_expression;
26440 Even though `sum' is accumulated in the loop, no use is made of
26441 that summation, so the accumulation can be removed.
26443 * Making side effects happen in the same order as in some other
26446 It is never safe to depend on the order of evaluation of side
26447 effects. For example, a function call like this may very well
26448 behave differently from one compiler to another:
26450 void func (int, int);
26455 There is no guarantee (in either the C or the C++ standard language
26456 definitions) that the increments will be evaluated in any
26457 particular order. Either increment might happen first. `func'
26458 might get the arguments `2, 3', or it might get `3, 2', or even
26461 * Making certain warnings into errors by default.
26463 Some ISO C testsuites report failure when the compiler does not
26464 produce an error message for a certain program.
26466 ISO C requires a "diagnostic" message for certain kinds of invalid
26467 programs, but a warning is defined by GCC to count as a
26468 diagnostic. If GCC produces a warning but not an error, that is
26469 correct ISO C support. If testsuites call this "failure", they
26470 should be run with the GCC option `-pedantic-errors', which will
26471 turn these warnings into errors.
26475 File: gcc.info, Node: Warnings and Errors, Prev: Non-bugs, Up: Trouble
26477 10.11 Warning Messages and Error Messages
26478 =========================================
26480 The GNU compiler can produce two kinds of diagnostics: errors and
26481 warnings. Each kind has a different purpose:
26483 "Errors" report problems that make it impossible to compile your
26484 program. GCC reports errors with the source file name and line
26485 number where the problem is apparent.
26487 "Warnings" report other unusual conditions in your code that _may_
26488 indicate a problem, although compilation can (and does) proceed.
26489 Warning messages also report the source file name and line number,
26490 but include the text `warning:' to distinguish them from error
26493 Warnings may indicate danger points where you should check to make sure
26494 that your program really does what you intend; or the use of obsolete
26495 features; or the use of nonstandard features of GNU C or C++. Many
26496 warnings are issued only if you ask for them, with one of the `-W'
26497 options (for instance, `-Wall' requests a variety of useful warnings).
26499 GCC always tries to compile your program if possible; it never
26500 gratuitously rejects a program whose meaning is clear merely because
26501 (for instance) it fails to conform to a standard. In some cases,
26502 however, the C and C++ standards specify that certain extensions are
26503 forbidden, and a diagnostic _must_ be issued by a conforming compiler.
26504 The `-pedantic' option tells GCC to issue warnings in such cases;
26505 `-pedantic-errors' says to make them errors instead. This does not
26506 mean that _all_ non-ISO constructs get warnings or errors.
26508 *Note Options to Request or Suppress Warnings: Warning Options, for
26509 more detail on these and related command-line options.
26512 File: gcc.info, Node: Bugs, Next: Service, Prev: Trouble, Up: Top
26517 Your bug reports play an essential role in making GCC reliable.
26519 When you encounter a problem, the first thing to do is to see if it is
26520 already known. *Note Trouble::. If it isn't known, then you should
26521 report the problem.
26525 * Criteria: Bug Criteria. Have you really found a bug?
26526 * Reporting: Bug Reporting. How to report a bug effectively.
26527 * Known: Trouble. Known problems.
26528 * Help: Service. Where to ask for help.
26531 File: gcc.info, Node: Bug Criteria, Next: Bug Reporting, Up: Bugs
26533 11.1 Have You Found a Bug?
26534 ==========================
26536 If you are not sure whether you have found a bug, here are some
26539 * If the compiler gets a fatal signal, for any input whatever, that
26540 is a compiler bug. Reliable compilers never crash.
26542 * If the compiler produces invalid assembly code, for any input
26543 whatever (except an `asm' statement), that is a compiler bug,
26544 unless the compiler reports errors (not just warnings) which would
26545 ordinarily prevent the assembler from being run.
26547 * If the compiler produces valid assembly code that does not
26548 correctly execute the input source code, that is a compiler bug.
26550 However, you must double-check to make sure, because you may have a
26551 program whose behavior is undefined, which happened by chance to
26552 give the desired results with another C or C++ compiler.
26554 For example, in many nonoptimizing compilers, you can write `x;'
26555 at the end of a function instead of `return x;', with the same
26556 results. But the value of the function is undefined if `return'
26557 is omitted; it is not a bug when GCC produces different results.
26559 Problems often result from expressions with two increment
26560 operators, as in `f (*p++, *p++)'. Your previous compiler might
26561 have interpreted that expression the way you intended; GCC might
26562 interpret it another way. Neither compiler is wrong. The bug is
26565 After you have localized the error to a single source line, it
26566 should be easy to check for these things. If your program is
26567 correct and well defined, you have found a compiler bug.
26569 * If the compiler produces an error message for valid input, that is
26572 * If the compiler does not produce an error message for invalid
26573 input, that is a compiler bug. However, you should note that your
26574 idea of "invalid input" might be someone else's idea of "an
26575 extension" or "support for traditional practice".
26577 * If you are an experienced user of one of the languages GCC
26578 supports, your suggestions for improvement of GCC are welcome in
26582 File: gcc.info, Node: Bug Reporting, Prev: Bug Criteria, Up: Bugs
26584 11.2 How and where to Report Bugs
26585 =================================
26587 Bugs should be reported to the GCC bug database. Please refer to
26588 `http://gcc.gnu.org/bugs.html' for up-to-date instructions how to
26589 submit bug reports. Copies of this file in HTML (`bugs.html') and
26590 plain text (`BUGS') are also part of GCC releases.
26593 File: gcc.info, Node: Service, Next: Contributing, Prev: Bugs, Up: Top
26595 12 How To Get Help with GCC
26596 ***************************
26598 If you need help installing, using or changing GCC, there are two ways
26601 * Send a message to a suitable network mailing list. First try
26602 <gcc-help@gcc.gnu.org> (for help installing or using GCC), and if
26603 that brings no response, try <gcc@gcc.gnu.org>. For help changing
26604 GCC, ask <gcc@gcc.gnu.org>. If you think you have found a bug in
26605 GCC, please report it following the instructions at *note Bug
26608 * Look in the service directory for someone who might help you for a
26609 fee. The service directory is found at
26610 `http://www.gnu.org/prep/service.html'.
26612 For further information, see `http://gcc.gnu.org/faq.html#support'.
26615 File: gcc.info, Node: Contributing, Next: Funding, Prev: Service, Up: Top
26617 13 Contributing to GCC Development
26618 **********************************
26620 If you would like to help pretest GCC releases to assure they work well,
26621 current development sources are available by CVS (see
26622 `http://gcc.gnu.org/cvs.html'). Source and binary snapshots are also
26623 available for FTP; see `http://gcc.gnu.org/snapshots.html'.
26625 If you would like to work on improvements to GCC, please read the
26626 advice at these URLs:
26628 `http://gcc.gnu.org/contribute.html'
26629 `http://gcc.gnu.org/contributewhy.html'
26631 for information on how to make useful contributions and avoid
26632 duplication of effort. Suggested projects are listed at
26633 `http://gcc.gnu.org/projects/'.
26636 File: gcc.info, Node: Funding, Next: GNU Project, Prev: Contributing, Up: Top
26638 Funding Free Software
26639 *********************
26641 If you want to have more free software a few years from now, it makes
26642 sense for you to help encourage people to contribute funds for its
26643 development. The most effective approach known is to encourage
26644 commercial redistributors to donate.
26646 Users of free software systems can boost the pace of development by
26647 encouraging for-a-fee distributors to donate part of their selling price
26648 to free software developers--the Free Software Foundation, and others.
26650 The way to convince distributors to do this is to demand it and expect
26651 it from them. So when you compare distributors, judge them partly by
26652 how much they give to free software development. Show distributors
26653 they must compete to be the one who gives the most.
26655 To make this approach work, you must insist on numbers that you can
26656 compare, such as, "We will donate ten dollars to the Frobnitz project
26657 for each disk sold." Don't be satisfied with a vague promise, such as
26658 "A portion of the profits are donated," since it doesn't give a basis
26661 Even a precise fraction "of the profits from this disk" is not very
26662 meaningful, since creative accounting and unrelated business decisions
26663 can greatly alter what fraction of the sales price counts as profit.
26664 If the price you pay is $50, ten percent of the profit is probably less
26665 than a dollar; it might be a few cents, or nothing at all.
26667 Some redistributors do development work themselves. This is useful
26668 too; but to keep everyone honest, you need to inquire how much they do,
26669 and what kind. Some kinds of development make much more long-term
26670 difference than others. For example, maintaining a separate version of
26671 a program contributes very little; maintaining the standard version of a
26672 program for the whole community contributes much. Easy new ports
26673 contribute little, since someone else would surely do them; difficult
26674 ports such as adding a new CPU to the GNU Compiler Collection
26675 contribute more; major new features or packages contribute the most.
26677 By establishing the idea that supporting further development is "the
26678 proper thing to do" when distributing free software for a fee, we can
26679 assure a steady flow of resources into making more free software.
26681 Copyright (C) 1994 Free Software Foundation, Inc.
26682 Verbatim copying and redistribution of this section is permitted
26683 without royalty; alteration is not permitted.
26686 File: gcc.info, Node: GNU Project, Next: Copying, Prev: Funding, Up: Top
26688 The GNU Project and GNU/Linux
26689 *****************************
26691 The GNU Project was launched in 1984 to develop a complete Unix-like
26692 operating system which is free software: the GNU system. (GNU is a
26693 recursive acronym for "GNU's Not Unix"; it is pronounced "guh-NEW".)
26694 Variants of the GNU operating system, which use the kernel Linux, are
26695 now widely used; though these systems are often referred to as "Linux",
26696 they are more accurately called GNU/Linux systems.
26698 For more information, see:
26699 `http://www.gnu.org/'
26700 `http://www.gnu.org/gnu/linux-and-gnu.html'
26703 File: gcc.info, Node: Copying, Next: GNU Free Documentation License, Prev: GNU Project, Up: Top
26705 GNU GENERAL PUBLIC LICENSE
26706 **************************
26708 Version 2, June 1991
26710 Copyright (C) 1989, 1991 Free Software Foundation, Inc.
26711 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA
26713 Everyone is permitted to copy and distribute verbatim copies
26714 of this license document, but changing it is not allowed.
26719 The licenses for most software are designed to take away your freedom
26720 to share and change it. By contrast, the GNU General Public License is
26721 intended to guarantee your freedom to share and change free
26722 software--to make sure the software is free for all its users. This
26723 General Public License applies to most of the Free Software
26724 Foundation's software and to any other program whose authors commit to
26725 using it. (Some other Free Software Foundation software is covered by
26726 the GNU Library General Public License instead.) You can apply it to
26727 your programs, too.
26729 When we speak of free software, we are referring to freedom, not
26730 price. Our General Public Licenses are designed to make sure that you
26731 have the freedom to distribute copies of free software (and charge for
26732 this service if you wish), that you receive source code or can get it
26733 if you want it, that you can change the software or use pieces of it in
26734 new free programs; and that you know you can do these things.
26736 To protect your rights, we need to make restrictions that forbid
26737 anyone to deny you these rights or to ask you to surrender the rights.
26738 These restrictions translate to certain responsibilities for you if you
26739 distribute copies of the software, or if you modify it.
26741 For example, if you distribute copies of such a program, whether
26742 gratis or for a fee, you must give the recipients all the rights that
26743 you have. You must make sure that they, too, receive or can get the
26744 source code. And you must show them these terms so they know their
26747 We protect your rights with two steps: (1) copyright the software, and
26748 (2) offer you this license which gives you legal permission to copy,
26749 distribute and/or modify the software.
26751 Also, for each author's protection and ours, we want to make certain
26752 that everyone understands that there is no warranty for this free
26753 software. If the software is modified by someone else and passed on, we
26754 want its recipients to know that what they have is not the original, so
26755 that any problems introduced by others will not reflect on the original
26756 authors' reputations.
26758 Finally, any free program is threatened constantly by software
26759 patents. We wish to avoid the danger that redistributors of a free
26760 program will individually obtain patent licenses, in effect making the
26761 program proprietary. To prevent this, we have made it clear that any
26762 patent must be licensed for everyone's free use or not licensed at all.
26764 The precise terms and conditions for copying, distribution and
26765 modification follow.
26767 TERMS AND CONDITIONS FOR COPYING, DISTRIBUTION AND MODIFICATION
26768 0. This License applies to any program or other work which contains a
26769 notice placed by the copyright holder saying it may be distributed
26770 under the terms of this General Public License. The "Program",
26771 below, refers to any such program or work, and a "work based on
26772 the Program" means either the Program or any derivative work under
26773 copyright law: that is to say, a work containing the Program or a
26774 portion of it, either verbatim or with modifications and/or
26775 translated into another language. (Hereinafter, translation is
26776 included without limitation in the term "modification".) Each
26777 licensee is addressed as "you".
26779 Activities other than copying, distribution and modification are
26780 not covered by this License; they are outside its scope. The act
26781 of running the Program is not restricted, and the output from the
26782 Program is covered only if its contents constitute a work based on
26783 the Program (independent of having been made by running the
26784 Program). Whether that is true depends on what the Program does.
26786 1. You may copy and distribute verbatim copies of the Program's
26787 source code as you receive it, in any medium, provided that you
26788 conspicuously and appropriately publish on each copy an appropriate
26789 copyright notice and disclaimer of warranty; keep intact all the
26790 notices that refer to this License and to the absence of any
26791 warranty; and give any other recipients of the Program a copy of
26792 this License along with the Program.
26794 You may charge a fee for the physical act of transferring a copy,
26795 and you may at your option offer warranty protection in exchange
26798 2. You may modify your copy or copies of the Program or any portion
26799 of it, thus forming a work based on the Program, and copy and
26800 distribute such modifications or work under the terms of Section 1
26801 above, provided that you also meet all of these conditions:
26803 a. You must cause the modified files to carry prominent notices
26804 stating that you changed the files and the date of any change.
26806 b. You must cause any work that you distribute or publish, that
26807 in whole or in part contains or is derived from the Program
26808 or any part thereof, to be licensed as a whole at no charge
26809 to all third parties under the terms of this License.
26811 c. If the modified program normally reads commands interactively
26812 when run, you must cause it, when started running for such
26813 interactive use in the most ordinary way, to print or display
26814 an announcement including an appropriate copyright notice and
26815 a notice that there is no warranty (or else, saying that you
26816 provide a warranty) and that users may redistribute the
26817 program under these conditions, and telling the user how to
26818 view a copy of this License. (Exception: if the Program
26819 itself is interactive but does not normally print such an
26820 announcement, your work based on the Program is not required
26821 to print an announcement.)
26823 These requirements apply to the modified work as a whole. If
26824 identifiable sections of that work are not derived from the
26825 Program, and can be reasonably considered independent and separate
26826 works in themselves, then this License, and its terms, do not
26827 apply to those sections when you distribute them as separate
26828 works. But when you distribute the same sections as part of a
26829 whole which is a work based on the Program, the distribution of
26830 the whole must be on the terms of this License, whose permissions
26831 for other licensees extend to the entire whole, and thus to each
26832 and every part regardless of who wrote it.
26834 Thus, it is not the intent of this section to claim rights or
26835 contest your rights to work written entirely by you; rather, the
26836 intent is to exercise the right to control the distribution of
26837 derivative or collective works based on the Program.
26839 In addition, mere aggregation of another work not based on the
26840 Program with the Program (or with a work based on the Program) on
26841 a volume of a storage or distribution medium does not bring the
26842 other work under the scope of this License.
26844 3. You may copy and distribute the Program (or a work based on it,
26845 under Section 2) in object code or executable form under the terms
26846 of Sections 1 and 2 above provided that you also do one of the
26849 a. Accompany it with the complete corresponding machine-readable
26850 source code, which must be distributed under the terms of
26851 Sections 1 and 2 above on a medium customarily used for
26852 software interchange; or,
26854 b. Accompany it with a written offer, valid for at least three
26855 years, to give any third party, for a charge no more than your
26856 cost of physically performing source distribution, a complete
26857 machine-readable copy of the corresponding source code, to be
26858 distributed under the terms of Sections 1 and 2 above on a
26859 medium customarily used for software interchange; or,
26861 c. Accompany it with the information you received as to the offer
26862 to distribute corresponding source code. (This alternative is
26863 allowed only for noncommercial distribution and only if you
26864 received the program in object code or executable form with
26865 such an offer, in accord with Subsection b above.)
26867 The source code for a work means the preferred form of the work for
26868 making modifications to it. For an executable work, complete
26869 source code means all the source code for all modules it contains,
26870 plus any associated interface definition files, plus the scripts
26871 used to control compilation and installation of the executable.
26872 However, as a special exception, the source code distributed need
26873 not include anything that is normally distributed (in either
26874 source or binary form) with the major components (compiler,
26875 kernel, and so on) of the operating system on which the executable
26876 runs, unless that component itself accompanies the executable.
26878 If distribution of executable or object code is made by offering
26879 access to copy from a designated place, then offering equivalent
26880 access to copy the source code from the same place counts as
26881 distribution of the source code, even though third parties are not
26882 compelled to copy the source along with the object code.
26884 4. You may not copy, modify, sublicense, or distribute the Program
26885 except as expressly provided under this License. Any attempt
26886 otherwise to copy, modify, sublicense or distribute the Program is
26887 void, and will automatically terminate your rights under this
26888 License. However, parties who have received copies, or rights,
26889 from you under this License will not have their licenses
26890 terminated so long as such parties remain in full compliance.
26892 5. You are not required to accept this License, since you have not
26893 signed it. However, nothing else grants you permission to modify
26894 or distribute the Program or its derivative works. These actions
26895 are prohibited by law if you do not accept this License.
26896 Therefore, by modifying or distributing the Program (or any work
26897 based on the Program), you indicate your acceptance of this
26898 License to do so, and all its terms and conditions for copying,
26899 distributing or modifying the Program or works based on it.
26901 6. Each time you redistribute the Program (or any work based on the
26902 Program), the recipient automatically receives a license from the
26903 original licensor to copy, distribute or modify the Program
26904 subject to these terms and conditions. You may not impose any
26905 further restrictions on the recipients' exercise of the rights
26906 granted herein. You are not responsible for enforcing compliance
26907 by third parties to this License.
26909 7. If, as a consequence of a court judgment or allegation of patent
26910 infringement or for any other reason (not limited to patent
26911 issues), conditions are imposed on you (whether by court order,
26912 agreement or otherwise) that contradict the conditions of this
26913 License, they do not excuse you from the conditions of this
26914 License. If you cannot distribute so as to satisfy simultaneously
26915 your obligations under this License and any other pertinent
26916 obligations, then as a consequence you may not distribute the
26917 Program at all. For example, if a patent license would not permit
26918 royalty-free redistribution of the Program by all those who
26919 receive copies directly or indirectly through you, then the only
26920 way you could satisfy both it and this License would be to refrain
26921 entirely from distribution of the Program.
26923 If any portion of this section is held invalid or unenforceable
26924 under any particular circumstance, the balance of the section is
26925 intended to apply and the section as a whole is intended to apply
26926 in other circumstances.
26928 It is not the purpose of this section to induce you to infringe any
26929 patents or other property right claims or to contest validity of
26930 any such claims; this section has the sole purpose of protecting
26931 the integrity of the free software distribution system, which is
26932 implemented by public license practices. Many people have made
26933 generous contributions to the wide range of software distributed
26934 through that system in reliance on consistent application of that
26935 system; it is up to the author/donor to decide if he or she is
26936 willing to distribute software through any other system and a
26937 licensee cannot impose that choice.
26939 This section is intended to make thoroughly clear what is believed
26940 to be a consequence of the rest of this License.
26942 8. If the distribution and/or use of the Program is restricted in
26943 certain countries either by patents or by copyrighted interfaces,
26944 the original copyright holder who places the Program under this
26945 License may add an explicit geographical distribution limitation
26946 excluding those countries, so that distribution is permitted only
26947 in or among countries not thus excluded. In such case, this
26948 License incorporates the limitation as if written in the body of
26951 9. The Free Software Foundation may publish revised and/or new
26952 versions of the General Public License from time to time. Such
26953 new versions will be similar in spirit to the present version, but
26954 may differ in detail to address new problems or concerns.
26956 Each version is given a distinguishing version number. If the
26957 Program specifies a version number of this License which applies
26958 to it and "any later version", you have the option of following
26959 the terms and conditions either of that version or of any later
26960 version published by the Free Software Foundation. If the Program
26961 does not specify a version number of this License, you may choose
26962 any version ever published by the Free Software Foundation.
26964 10. If you wish to incorporate parts of the Program into other free
26965 programs whose distribution conditions are different, write to the
26966 author to ask for permission. For software which is copyrighted
26967 by the Free Software Foundation, write to the Free Software
26968 Foundation; we sometimes make exceptions for this. Our decision
26969 will be guided by the two goals of preserving the free status of
26970 all derivatives of our free software and of promoting the sharing
26971 and reuse of software generally.
26974 11. BECAUSE THE PROGRAM IS LICENSED FREE OF CHARGE, THERE IS NO
26975 WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY APPLICABLE
26976 LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT
26977 HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT
26978 WARRANTY OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT
26979 NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
26980 FITNESS FOR A PARTICULAR PURPOSE. THE ENTIRE RISK AS TO THE
26981 QUALITY AND PERFORMANCE OF THE PROGRAM IS WITH YOU. SHOULD THE
26982 PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF ALL NECESSARY
26983 SERVICING, REPAIR OR CORRECTION.
26985 12. IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN
26986 WRITING WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MAY
26987 MODIFY AND/OR REDISTRIBUTE THE PROGRAM AS PERMITTED ABOVE, BE
26988 LIABLE TO YOU FOR DAMAGES, INCLUDING ANY GENERAL, SPECIAL,
26989 INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE USE OR
26990 INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF
26991 DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU
26992 OR THIRD PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY
26993 OTHER PROGRAMS), EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN
26994 ADVISED OF THE POSSIBILITY OF SUCH DAMAGES.
26996 END OF TERMS AND CONDITIONS
26997 Appendix: How to Apply These Terms to Your New Programs
26998 =======================================================
27000 If you develop a new program, and you want it to be of the greatest
27001 possible use to the public, the best way to achieve this is to make it
27002 free software which everyone can redistribute and change under these
27005 To do so, attach the following notices to the program. It is safest
27006 to attach them to the start of each source file to most effectively
27007 convey the exclusion of warranty; and each file should have at least
27008 the "copyright" line and a pointer to where the full notice is found.
27010 ONE LINE TO GIVE THE PROGRAM'S NAME AND A BRIEF IDEA OF WHAT IT DOES.
27011 Copyright (C) YEAR NAME OF AUTHOR
27013 This program is free software; you can redistribute it and/or modify
27014 it under the terms of the GNU General Public License as published by
27015 the Free Software Foundation; either version 2 of the License, or
27016 (at your option) any later version.
27018 This program is distributed in the hope that it will be useful,
27019 but WITHOUT ANY WARRANTY; without even the implied warranty of
27020 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
27021 GNU General Public License for more details.
27023 You should have received a copy of the GNU General Public License
27024 along with this program; if not, write to the Free Software
27025 Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA
27027 Also add information on how to contact you by electronic and paper
27030 If the program is interactive, make it output a short notice like this
27031 when it starts in an interactive mode:
27033 Gnomovision version 69, Copyright (C) YEAR NAME OF AUTHOR
27034 Gnomovision comes with ABSOLUTELY NO WARRANTY; for details
27036 This is free software, and you are welcome to redistribute it
27037 under certain conditions; type `show c' for details.
27039 The hypothetical commands `show w' and `show c' should show the
27040 appropriate parts of the General Public License. Of course, the
27041 commands you use may be called something other than `show w' and `show
27042 c'; they could even be mouse-clicks or menu items--whatever suits your
27045 You should also get your employer (if you work as a programmer) or your
27046 school, if any, to sign a "copyright disclaimer" for the program, if
27047 necessary. Here is a sample; alter the names:
27049 Yoyodyne, Inc., hereby disclaims all copyright interest in the program
27050 `Gnomovision' (which makes passes at compilers) written by James Hacker.
27052 SIGNATURE OF TY COON, 1 April 1989
27053 Ty Coon, President of Vice
27055 This General Public License does not permit incorporating your program
27056 into proprietary programs. If your program is a subroutine library,
27057 you may consider it more useful to permit linking proprietary
27058 applications with the library. If this is what you want to do, use the
27059 GNU Library General Public License instead of this License.
27062 File: gcc.info, Node: GNU Free Documentation License, Next: Contributors, Prev: Copying, Up: Top
27064 GNU Free Documentation License
27065 ******************************
27067 Version 1.2, November 2002
27069 Copyright (C) 2000,2001,2002 Free Software Foundation, Inc.
27070 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA
27072 Everyone is permitted to copy and distribute verbatim copies
27073 of this license document, but changing it is not allowed.
27077 The purpose of this License is to make a manual, textbook, or other
27078 functional and useful document "free" in the sense of freedom: to
27079 assure everyone the effective freedom to copy and redistribute it,
27080 with or without modifying it, either commercially or
27081 noncommercially. Secondarily, this License preserves for the
27082 author and publisher a way to get credit for their work, while not
27083 being considered responsible for modifications made by others.
27085 This License is a kind of "copyleft", which means that derivative
27086 works of the document must themselves be free in the same sense.
27087 It complements the GNU General Public License, which is a copyleft
27088 license designed for free software.
27090 We have designed this License in order to use it for manuals for
27091 free software, because free software needs free documentation: a
27092 free program should come with manuals providing the same freedoms
27093 that the software does. But this License is not limited to
27094 software manuals; it can be used for any textual work, regardless
27095 of subject matter or whether it is published as a printed book.
27096 We recommend this License principally for works whose purpose is
27097 instruction or reference.
27099 1. APPLICABILITY AND DEFINITIONS
27101 This License applies to any manual or other work, in any medium,
27102 that contains a notice placed by the copyright holder saying it
27103 can be distributed under the terms of this License. Such a notice
27104 grants a world-wide, royalty-free license, unlimited in duration,
27105 to use that work under the conditions stated herein. The
27106 "Document", below, refers to any such manual or work. Any member
27107 of the public is a licensee, and is addressed as "you". You
27108 accept the license if you copy, modify or distribute the work in a
27109 way requiring permission under copyright law.
27111 A "Modified Version" of the Document means any work containing the
27112 Document or a portion of it, either copied verbatim, or with
27113 modifications and/or translated into another language.
27115 A "Secondary Section" is a named appendix or a front-matter section
27116 of the Document that deals exclusively with the relationship of the
27117 publishers or authors of the Document to the Document's overall
27118 subject (or to related matters) and contains nothing that could
27119 fall directly within that overall subject. (Thus, if the Document
27120 is in part a textbook of mathematics, a Secondary Section may not
27121 explain any mathematics.) The relationship could be a matter of
27122 historical connection with the subject or with related matters, or
27123 of legal, commercial, philosophical, ethical or political position
27126 The "Invariant Sections" are certain Secondary Sections whose
27127 titles are designated, as being those of Invariant Sections, in
27128 the notice that says that the Document is released under this
27129 License. If a section does not fit the above definition of
27130 Secondary then it is not allowed to be designated as Invariant.
27131 The Document may contain zero Invariant Sections. If the Document
27132 does not identify any Invariant Sections then there are none.
27134 The "Cover Texts" are certain short passages of text that are
27135 listed, as Front-Cover Texts or Back-Cover Texts, in the notice
27136 that says that the Document is released under this License. A
27137 Front-Cover Text may be at most 5 words, and a Back-Cover Text may
27138 be at most 25 words.
27140 A "Transparent" copy of the Document means a machine-readable copy,
27141 represented in a format whose specification is available to the
27142 general public, that is suitable for revising the document
27143 straightforwardly with generic text editors or (for images
27144 composed of pixels) generic paint programs or (for drawings) some
27145 widely available drawing editor, and that is suitable for input to
27146 text formatters or for automatic translation to a variety of
27147 formats suitable for input to text formatters. A copy made in an
27148 otherwise Transparent file format whose markup, or absence of
27149 markup, has been arranged to thwart or discourage subsequent
27150 modification by readers is not Transparent. An image format is
27151 not Transparent if used for any substantial amount of text. A
27152 copy that is not "Transparent" is called "Opaque".
27154 Examples of suitable formats for Transparent copies include plain
27155 ASCII without markup, Texinfo input format, LaTeX input format,
27156 SGML or XML using a publicly available DTD, and
27157 standard-conforming simple HTML, PostScript or PDF designed for
27158 human modification. Examples of transparent image formats include
27159 PNG, XCF and JPG. Opaque formats include proprietary formats that
27160 can be read and edited only by proprietary word processors, SGML or
27161 XML for which the DTD and/or processing tools are not generally
27162 available, and the machine-generated HTML, PostScript or PDF
27163 produced by some word processors for output purposes only.
27165 The "Title Page" means, for a printed book, the title page itself,
27166 plus such following pages as are needed to hold, legibly, the
27167 material this License requires to appear in the title page. For
27168 works in formats which do not have any title page as such, "Title
27169 Page" means the text near the most prominent appearance of the
27170 work's title, preceding the beginning of the body of the text.
27172 A section "Entitled XYZ" means a named subunit of the Document
27173 whose title either is precisely XYZ or contains XYZ in parentheses
27174 following text that translates XYZ in another language. (Here XYZ
27175 stands for a specific section name mentioned below, such as
27176 "Acknowledgements", "Dedications", "Endorsements", or "History".)
27177 To "Preserve the Title" of such a section when you modify the
27178 Document means that it remains a section "Entitled XYZ" according
27179 to this definition.
27181 The Document may include Warranty Disclaimers next to the notice
27182 which states that this License applies to the Document. These
27183 Warranty Disclaimers are considered to be included by reference in
27184 this License, but only as regards disclaiming warranties: any other
27185 implication that these Warranty Disclaimers may have is void and
27186 has no effect on the meaning of this License.
27188 2. VERBATIM COPYING
27190 You may copy and distribute the Document in any medium, either
27191 commercially or noncommercially, provided that this License, the
27192 copyright notices, and the license notice saying this License
27193 applies to the Document are reproduced in all copies, and that you
27194 add no other conditions whatsoever to those of this License. You
27195 may not use technical measures to obstruct or control the reading
27196 or further copying of the copies you make or distribute. However,
27197 you may accept compensation in exchange for copies. If you
27198 distribute a large enough number of copies you must also follow
27199 the conditions in section 3.
27201 You may also lend copies, under the same conditions stated above,
27202 and you may publicly display copies.
27204 3. COPYING IN QUANTITY
27206 If you publish printed copies (or copies in media that commonly
27207 have printed covers) of the Document, numbering more than 100, and
27208 the Document's license notice requires Cover Texts, you must
27209 enclose the copies in covers that carry, clearly and legibly, all
27210 these Cover Texts: Front-Cover Texts on the front cover, and
27211 Back-Cover Texts on the back cover. Both covers must also clearly
27212 and legibly identify you as the publisher of these copies. The
27213 front cover must present the full title with all words of the
27214 title equally prominent and visible. You may add other material
27215 on the covers in addition. Copying with changes limited to the
27216 covers, as long as they preserve the title of the Document and
27217 satisfy these conditions, can be treated as verbatim copying in
27220 If the required texts for either cover are too voluminous to fit
27221 legibly, you should put the first ones listed (as many as fit
27222 reasonably) on the actual cover, and continue the rest onto
27225 If you publish or distribute Opaque copies of the Document
27226 numbering more than 100, you must either include a
27227 machine-readable Transparent copy along with each Opaque copy, or
27228 state in or with each Opaque copy a computer-network location from
27229 which the general network-using public has access to download
27230 using public-standard network protocols a complete Transparent
27231 copy of the Document, free of added material. If you use the
27232 latter option, you must take reasonably prudent steps, when you
27233 begin distribution of Opaque copies in quantity, to ensure that
27234 this Transparent copy will remain thus accessible at the stated
27235 location until at least one year after the last time you
27236 distribute an Opaque copy (directly or through your agents or
27237 retailers) of that edition to the public.
27239 It is requested, but not required, that you contact the authors of
27240 the Document well before redistributing any large number of
27241 copies, to give them a chance to provide you with an updated
27242 version of the Document.
27246 You may copy and distribute a Modified Version of the Document
27247 under the conditions of sections 2 and 3 above, provided that you
27248 release the Modified Version under precisely this License, with
27249 the Modified Version filling the role of the Document, thus
27250 licensing distribution and modification of the Modified Version to
27251 whoever possesses a copy of it. In addition, you must do these
27252 things in the Modified Version:
27254 A. Use in the Title Page (and on the covers, if any) a title
27255 distinct from that of the Document, and from those of
27256 previous versions (which should, if there were any, be listed
27257 in the History section of the Document). You may use the
27258 same title as a previous version if the original publisher of
27259 that version gives permission.
27261 B. List on the Title Page, as authors, one or more persons or
27262 entities responsible for authorship of the modifications in
27263 the Modified Version, together with at least five of the
27264 principal authors of the Document (all of its principal
27265 authors, if it has fewer than five), unless they release you
27266 from this requirement.
27268 C. State on the Title page the name of the publisher of the
27269 Modified Version, as the publisher.
27271 D. Preserve all the copyright notices of the Document.
27273 E. Add an appropriate copyright notice for your modifications
27274 adjacent to the other copyright notices.
27276 F. Include, immediately after the copyright notices, a license
27277 notice giving the public permission to use the Modified
27278 Version under the terms of this License, in the form shown in
27279 the Addendum below.
27281 G. Preserve in that license notice the full lists of Invariant
27282 Sections and required Cover Texts given in the Document's
27285 H. Include an unaltered copy of this License.
27287 I. Preserve the section Entitled "History", Preserve its Title,
27288 and add to it an item stating at least the title, year, new
27289 authors, and publisher of the Modified Version as given on
27290 the Title Page. If there is no section Entitled "History" in
27291 the Document, create one stating the title, year, authors,
27292 and publisher of the Document as given on its Title Page,
27293 then add an item describing the Modified Version as stated in
27294 the previous sentence.
27296 J. Preserve the network location, if any, given in the Document
27297 for public access to a Transparent copy of the Document, and
27298 likewise the network locations given in the Document for
27299 previous versions it was based on. These may be placed in
27300 the "History" section. You may omit a network location for a
27301 work that was published at least four years before the
27302 Document itself, or if the original publisher of the version
27303 it refers to gives permission.
27305 K. For any section Entitled "Acknowledgements" or "Dedications",
27306 Preserve the Title of the section, and preserve in the
27307 section all the substance and tone of each of the contributor
27308 acknowledgements and/or dedications given therein.
27310 L. Preserve all the Invariant Sections of the Document,
27311 unaltered in their text and in their titles. Section numbers
27312 or the equivalent are not considered part of the section
27315 M. Delete any section Entitled "Endorsements". Such a section
27316 may not be included in the Modified Version.
27318 N. Do not retitle any existing section to be Entitled
27319 "Endorsements" or to conflict in title with any Invariant
27322 O. Preserve any Warranty Disclaimers.
27324 If the Modified Version includes new front-matter sections or
27325 appendices that qualify as Secondary Sections and contain no
27326 material copied from the Document, you may at your option
27327 designate some or all of these sections as invariant. To do this,
27328 add their titles to the list of Invariant Sections in the Modified
27329 Version's license notice. These titles must be distinct from any
27330 other section titles.
27332 You may add a section Entitled "Endorsements", provided it contains
27333 nothing but endorsements of your Modified Version by various
27334 parties--for example, statements of peer review or that the text
27335 has been approved by an organization as the authoritative
27336 definition of a standard.
27338 You may add a passage of up to five words as a Front-Cover Text,
27339 and a passage of up to 25 words as a Back-Cover Text, to the end
27340 of the list of Cover Texts in the Modified Version. Only one
27341 passage of Front-Cover Text and one of Back-Cover Text may be
27342 added by (or through arrangements made by) any one entity. If the
27343 Document already includes a cover text for the same cover,
27344 previously added by you or by arrangement made by the same entity
27345 you are acting on behalf of, you may not add another; but you may
27346 replace the old one, on explicit permission from the previous
27347 publisher that added the old one.
27349 The author(s) and publisher(s) of the Document do not by this
27350 License give permission to use their names for publicity for or to
27351 assert or imply endorsement of any Modified Version.
27353 5. COMBINING DOCUMENTS
27355 You may combine the Document with other documents released under
27356 this License, under the terms defined in section 4 above for
27357 modified versions, provided that you include in the combination
27358 all of the Invariant Sections of all of the original documents,
27359 unmodified, and list them all as Invariant Sections of your
27360 combined work in its license notice, and that you preserve all
27361 their Warranty Disclaimers.
27363 The combined work need only contain one copy of this License, and
27364 multiple identical Invariant Sections may be replaced with a single
27365 copy. If there are multiple Invariant Sections with the same name
27366 but different contents, make the title of each such section unique
27367 by adding at the end of it, in parentheses, the name of the
27368 original author or publisher of that section if known, or else a
27369 unique number. Make the same adjustment to the section titles in
27370 the list of Invariant Sections in the license notice of the
27373 In the combination, you must combine any sections Entitled
27374 "History" in the various original documents, forming one section
27375 Entitled "History"; likewise combine any sections Entitled
27376 "Acknowledgements", and any sections Entitled "Dedications". You
27377 must delete all sections Entitled "Endorsements."
27379 6. COLLECTIONS OF DOCUMENTS
27381 You may make a collection consisting of the Document and other
27382 documents released under this License, and replace the individual
27383 copies of this License in the various documents with a single copy
27384 that is included in the collection, provided that you follow the
27385 rules of this License for verbatim copying of each of the
27386 documents in all other respects.
27388 You may extract a single document from such a collection, and
27389 distribute it individually under this License, provided you insert
27390 a copy of this License into the extracted document, and follow
27391 this License in all other respects regarding verbatim copying of
27394 7. AGGREGATION WITH INDEPENDENT WORKS
27396 A compilation of the Document or its derivatives with other
27397 separate and independent documents or works, in or on a volume of
27398 a storage or distribution medium, is called an "aggregate" if the
27399 copyright resulting from the compilation is not used to limit the
27400 legal rights of the compilation's users beyond what the individual
27401 works permit. When the Document is included in an aggregate, this
27402 License does not apply to the other works in the aggregate which
27403 are not themselves derivative works of the Document.
27405 If the Cover Text requirement of section 3 is applicable to these
27406 copies of the Document, then if the Document is less than one half
27407 of the entire aggregate, the Document's Cover Texts may be placed
27408 on covers that bracket the Document within the aggregate, or the
27409 electronic equivalent of covers if the Document is in electronic
27410 form. Otherwise they must appear on printed covers that bracket
27411 the whole aggregate.
27415 Translation is considered a kind of modification, so you may
27416 distribute translations of the Document under the terms of section
27417 4. Replacing Invariant Sections with translations requires special
27418 permission from their copyright holders, but you may include
27419 translations of some or all Invariant Sections in addition to the
27420 original versions of these Invariant Sections. You may include a
27421 translation of this License, and all the license notices in the
27422 Document, and any Warranty Disclaimers, provided that you also
27423 include the original English version of this License and the
27424 original versions of those notices and disclaimers. In case of a
27425 disagreement between the translation and the original version of
27426 this License or a notice or disclaimer, the original version will
27429 If a section in the Document is Entitled "Acknowledgements",
27430 "Dedications", or "History", the requirement (section 4) to
27431 Preserve its Title (section 1) will typically require changing the
27436 You may not copy, modify, sublicense, or distribute the Document
27437 except as expressly provided for under this License. Any other
27438 attempt to copy, modify, sublicense or distribute the Document is
27439 void, and will automatically terminate your rights under this
27440 License. However, parties who have received copies, or rights,
27441 from you under this License will not have their licenses
27442 terminated so long as such parties remain in full compliance.
27444 10. FUTURE REVISIONS OF THIS LICENSE
27446 The Free Software Foundation may publish new, revised versions of
27447 the GNU Free Documentation License from time to time. Such new
27448 versions will be similar in spirit to the present version, but may
27449 differ in detail to address new problems or concerns. See
27450 `http://www.gnu.org/copyleft/'.
27452 Each version of the License is given a distinguishing version
27453 number. If the Document specifies that a particular numbered
27454 version of this License "or any later version" applies to it, you
27455 have the option of following the terms and conditions either of
27456 that specified version or of any later version that has been
27457 published (not as a draft) by the Free Software Foundation. If
27458 the Document does not specify a version number of this License,
27459 you may choose any version ever published (not as a draft) by the
27460 Free Software Foundation.
27462 ADDENDUM: How to use this License for your documents
27463 ====================================================
27465 To use this License in a document you have written, include a copy of
27466 the License in the document and put the following copyright and license
27467 notices just after the title page:
27469 Copyright (C) YEAR YOUR NAME.
27470 Permission is granted to copy, distribute and/or modify this document
27471 under the terms of the GNU Free Documentation License, Version 1.2
27472 or any later version published by the Free Software Foundation;
27473 with no Invariant Sections, no Front-Cover Texts, and no Back-Cover
27474 Texts. A copy of the license is included in the section entitled ``GNU
27475 Free Documentation License''.
27477 If you have Invariant Sections, Front-Cover Texts and Back-Cover Texts,
27478 replace the "with...Texts." line with this:
27480 with the Invariant Sections being LIST THEIR TITLES, with
27481 the Front-Cover Texts being LIST, and with the Back-Cover Texts
27484 If you have Invariant Sections without Cover Texts, or some other
27485 combination of the three, merge those two alternatives to suit the
27488 If your document contains nontrivial examples of program code, we
27489 recommend releasing these examples in parallel under your choice of
27490 free software license, such as the GNU General Public License, to
27491 permit their use in free software.
27494 File: gcc.info, Node: Contributors, Next: Option Index, Prev: GNU Free Documentation License, Up: Top
27496 Contributors to GCC
27497 *******************
27499 The GCC project would like to thank its many contributors. Without
27500 them the project would not have been nearly as successful as it has
27501 been. Any omissions in this list are accidental. Feel free to contact
27502 <law@redhat.com> or <gerald@pfeifer.com> if you have been left out or
27503 some of your contributions are not listed. Please keep this list in
27504 alphabetical order.
27506 * Analog Devices helped implement the support for complex data types
27509 * John David Anglin for threading-related fixes and improvements to
27510 libstdc++-v3, and the HP-UX port.
27512 * James van Artsdalen wrote the code that makes efficient use of the
27513 Intel 80387 register stack.
27515 * Abramo and Roberto Bagnara for the SysV68 Motorola 3300 Delta
27518 * Alasdair Baird for various bug fixes.
27520 * Giovanni Bajo for analyzing lots of complicated C++ problem
27523 * Peter Barada for his work to improve code generation for new
27526 * Gerald Baumgartner added the signature extension to the C++ front
27529 * Godmar Back for his Java improvements and encouragement.
27531 * Scott Bambrough for help porting the Java compiler.
27533 * Wolfgang Bangerth for processing tons of bug reports.
27535 * Jon Beniston for his Microsoft Windows port of Java.
27537 * Daniel Berlin for better DWARF2 support, faster/better
27538 optimizations, improved alias analysis, plus migrating GCC to
27541 * Geoff Berry for his Java object serialization work and various
27544 * Eric Blake for helping to make GCJ and libgcj conform to the
27547 * Janne Blomqvist for contributions to gfortran.
27549 * Segher Boessenkool for various fixes.
27551 * Hans-J. Boehm for his garbage collector, IA-64 libffi port, and
27554 * Neil Booth for work on cpplib, lang hooks, debug hooks and other
27555 miscellaneous clean-ups.
27557 * Steven Bosscher for integrating the gfortran front end into GCC
27558 and for contributing to the tree-ssa branch.
27560 * Eric Botcazou for fixing middle- and backend bugs left and right.
27562 * Per Bothner for his direction via the steering committee and
27563 various improvements to the infrastructure for supporting new
27564 languages. Chill front end implementation. Initial
27565 implementations of cpplib, fix-header, config.guess, libio, and
27566 past C++ library (libg++) maintainer. Dreaming up, designing and
27567 implementing much of GCJ.
27569 * Devon Bowen helped port GCC to the Tahoe.
27571 * Don Bowman for mips-vxworks contributions.
27573 * Dave Brolley for work on cpplib and Chill.
27575 * Paul Brook for work on the ARM architecture and maintaining
27578 * Robert Brown implemented the support for Encore 32000 systems.
27580 * Christian Bruel for improvements to local store elimination.
27582 * Herman A.J. ten Brugge for various fixes.
27584 * Joerg Brunsmann for Java compiler hacking and help with the GCJ
27587 * Joe Buck for his direction via the steering committee.
27589 * Craig Burley for leadership of the G77 Fortran effort.
27591 * Stephan Buys for contributing Doxygen notes for libstdc++.
27593 * Paolo Carlini for libstdc++ work: lots of efficiency improvements
27594 to the C++ strings, streambufs and formatted I/O, hard detective
27595 work on the frustrating localization issues, and keeping up with
27596 the problem reports.
27598 * John Carr for his alias work, SPARC hacking, infrastructure
27599 improvements, previous contributions to the steering committee,
27600 loop optimizations, etc.
27602 * Stephane Carrez for 68HC11 and 68HC12 ports.
27604 * Steve Chamberlain for support for the Renesas SH and H8 processors
27605 and the PicoJava processor, and for GCJ config fixes.
27607 * Glenn Chambers for help with the GCJ FAQ.
27609 * John-Marc Chandonia for various libgcj patches.
27611 * Scott Christley for his Objective-C contributions.
27613 * Eric Christopher for his Java porting help and clean-ups.
27615 * Branko Cibej for more warning contributions.
27617 * The GNU Classpath project for all of their merged runtime code.
27619 * Nick Clifton for arm, mcore, fr30, v850, m32r work, `--help', and
27620 other random hacking.
27622 * Michael Cook for libstdc++ cleanup patches to reduce warnings.
27624 * R. Kelley Cook for making GCC buildable from a read-only directory
27625 as well as other miscellaneous build process and documentation
27628 * Ralf Corsepius for SH testing and minor bugfixing.
27630 * Stan Cox for care and feeding of the x86 port and lots of behind
27631 the scenes hacking.
27633 * Alex Crain provided changes for the 3b1.
27635 * Ian Dall for major improvements to the NS32k port.
27637 * Paul Dale for his work to add uClinux platform support to the m68k
27640 * Dario Dariol contributed the four varieties of sample programs
27641 that print a copy of their source.
27643 * Russell Davidson for fstream and stringstream fixes in libstdc++.
27645 * Bud Davis for work on the G77 and gfortran compilers.
27647 * Mo DeJong for GCJ and libgcj bug fixes.
27649 * DJ Delorie for the DJGPP port, build and libiberty maintenance,
27650 various bug fixes, and the M32C port.
27652 * Arnaud Desitter for helping to debug gfortran.
27654 * Gabriel Dos Reis for contributions to G++, contributions and
27655 maintenance of GCC diagnostics infrastructure, libstdc++-v3,
27656 including `valarray<>', `complex<>', maintaining the numerics
27657 library (including that pesky `<limits>' :-) and keeping
27658 up-to-date anything to do with numbers.
27660 * Ulrich Drepper for his work on glibc, testing of GCC using glibc,
27661 ISO C99 support, CFG dumping support, etc., plus support of the
27662 C++ runtime libraries including for all kinds of C interface
27663 issues, contributing and maintaining `complex<>', sanity checking
27664 and disbursement, configuration architecture, libio maintenance,
27665 and early math work.
27667 * Zdenek Dvorak for a new loop unroller and various fixes.
27669 * Richard Earnshaw for his ongoing work with the ARM.
27671 * David Edelsohn for his direction via the steering committee,
27672 ongoing work with the RS6000/PowerPC port, help cleaning up Haifa
27673 loop changes, doing the entire AIX port of libstdc++ with his bare
27674 hands, and for ensuring GCC properly keeps working on AIX.
27676 * Kevin Ediger for the floating point formatting of num_put::do_put
27679 * Phil Edwards for libstdc++ work including configuration hackery,
27680 documentation maintainer, chief breaker of the web pages, the
27681 occasional iostream bug fix, and work on shared library symbol
27684 * Paul Eggert for random hacking all over GCC.
27686 * Mark Elbrecht for various DJGPP improvements, and for libstdc++
27687 configuration support for locales and fstream-related fixes.
27689 * Vadim Egorov for libstdc++ fixes in strings, streambufs, and
27692 * Christian Ehrhardt for dealing with bug reports.
27694 * Ben Elliston for his work to move the Objective-C runtime into its
27695 own subdirectory and for his work on autoconf.
27697 * Marc Espie for OpenBSD support.
27699 * Doug Evans for much of the global optimization framework, arc,
27700 m32r, and SPARC work.
27702 * Christopher Faylor for his work on the Cygwin port and for caring
27703 and feeding the gcc.gnu.org box and saving its users tons of spam.
27705 * Fred Fish for BeOS support and Ada fixes.
27707 * Ivan Fontes Garcia for the Portuguese translation of the GCJ FAQ.
27709 * Peter Gerwinski for various bug fixes and the Pascal front end.
27711 * Kaveh R. Ghazi for his direction via the steering committee,
27712 amazing work to make `-W -Wall -W* -Werror' useful, and
27713 continuously testing GCC on a plethora of platforms. Kaveh
27714 extends his gratitude to the CAIP Center at Rutgers University for
27715 providing him with computing resources to work on Free Software
27716 since the late 1980s.
27718 * John Gilmore for a donation to the FSF earmarked improving GNU
27721 * Judy Goldberg for c++ contributions.
27723 * Torbjorn Granlund for various fixes and the c-torture testsuite,
27724 multiply- and divide-by-constant optimization, improved long long
27725 support, improved leaf function register allocation, and his
27726 direction via the steering committee.
27728 * Anthony Green for his `-Os' contributions and Java front end work.
27730 * Stu Grossman for gdb hacking, allowing GCJ developers to debug
27733 * Michael K. Gschwind contributed the port to the PDP-11.
27735 * Ron Guilmette implemented the `protoize' and `unprotoize' tools,
27736 the support for Dwarf symbolic debugging information, and much of
27737 the support for System V Release 4. He has also worked heavily on
27738 the Intel 386 and 860 support.
27740 * Mostafa Hagog for Swing Modulo Scheduling (SMS) and post reload
27743 * Bruno Haible for improvements in the runtime overhead for EH, new
27744 warnings and assorted bug fixes.
27746 * Andrew Haley for his amazing Java compiler and library efforts.
27748 * Chris Hanson assisted in making GCC work on HP-UX for the 9000
27751 * Michael Hayes for various thankless work he's done trying to get
27752 the c30/c40 ports functional. Lots of loop and unroll
27753 improvements and fixes.
27755 * Dara Hazeghi for wading through myriads of target-specific bug
27758 * Kate Hedstrom for staking the G77 folks with an initial testsuite.
27760 * Richard Henderson for his ongoing SPARC, alpha, ia32, and ia64
27761 work, loop opts, and generally fixing lots of old problems we've
27762 ignored for years, flow rewrite and lots of further stuff,
27763 including reviewing tons of patches.
27765 * Aldy Hernandez for working on the PowerPC port, SIMD support, and
27768 * Nobuyuki Hikichi of Software Research Associates, Tokyo,
27769 contributed the support for the Sony NEWS machine.
27771 * Kazu Hirata for caring and feeding the Renesas H8/300 port and
27774 * Katherine Holcomb for work on gfortran.
27776 * Manfred Hollstein for his ongoing work to keep the m88k alive, lots
27777 of testing and bug fixing, particularly of GCC configury code.
27779 * Steve Holmgren for MachTen patches.
27781 * Jan Hubicka for his x86 port improvements.
27783 * Falk Hueffner for working on C and optimization bug reports.
27785 * Bernardo Innocenti for his m68k work, including merging of
27786 ColdFire improvements and uClinux support.
27788 * Christian Iseli for various bug fixes.
27790 * Kamil Iskra for general m68k hacking.
27792 * Lee Iverson for random fixes and MIPS testing.
27794 * Andreas Jaeger for testing and benchmarking of GCC and various bug
27797 * Jakub Jelinek for his SPARC work and sibling call optimizations as
27798 well as lots of bug fixes and test cases, and for improving the
27801 * Janis Johnson for ia64 testing and fixes, her quality improvement
27802 sidetracks, and web page maintenance.
27804 * Kean Johnston for SCO OpenServer support and various fixes.
27806 * Tim Josling for the sample language treelang based originally on
27807 Richard Kenner's "toy" language.
27809 * Nicolai Josuttis for additional libstdc++ documentation.
27811 * Klaus Kaempf for his ongoing work to make alpha-vms a viable
27814 * Steven G. Kargl for work on gfortran.
27816 * David Kashtan of SRI adapted GCC to VMS.
27818 * Ryszard Kabatek for many, many libstdc++ bug fixes and
27819 optimizations of strings, especially member functions, and for
27822 * Geoffrey Keating for his ongoing work to make the PPC work for
27823 GNU/Linux and his automatic regression tester.
27825 * Brendan Kehoe for his ongoing work with G++ and for a lot of early
27826 work in just about every part of libstdc++.
27828 * Oliver M. Kellogg of Deutsche Aerospace contributed the port to the
27831 * Richard Kenner of the New York University Ultracomputer Research
27832 Laboratory wrote the machine descriptions for the AMD 29000, the
27833 DEC Alpha, the IBM RT PC, and the IBM RS/6000 as well as the
27834 support for instruction attributes. He also made changes to
27835 better support RISC processors including changes to common
27836 subexpression elimination, strength reduction, function calling
27837 sequence handling, and condition code support, in addition to
27838 generalizing the code for frame pointer elimination and delay slot
27839 scheduling. Richard Kenner was also the head maintainer of GCC
27842 * Mumit Khan for various contributions to the Cygwin and Mingw32
27843 ports and maintaining binary releases for Microsoft Windows hosts,
27844 and for massive libstdc++ porting work to Cygwin/Mingw32.
27846 * Robin Kirkham for cpu32 support.
27848 * Mark Klein for PA improvements.
27850 * Thomas Koenig for various bug fixes.
27852 * Bruce Korb for the new and improved fixincludes code.
27854 * Benjamin Kosnik for his G++ work and for leading the libstdc++-v3
27857 * Charles LaBrec contributed the support for the Integrated Solutions
27860 * Asher Langton and Mike Kumbera for contributing Cray pointer
27861 support to gfortran, and for other gfortran improvements.
27863 * Jeff Law for his direction via the steering committee,
27864 coordinating the entire egcs project and GCC 2.95, rolling out
27865 snapshots and releases, handling merges from GCC2, reviewing tons
27866 of patches that might have fallen through the cracks else, and
27867 random but extensive hacking.
27869 * Marc Lehmann for his direction via the steering committee and
27870 helping with analysis and improvements of x86 performance.
27872 * Victor Leikehman for work on gfortran.
27874 * Ted Lemon wrote parts of the RTL reader and printer.
27876 * Kriang Lerdsuwanakij for C++ improvements including template as
27877 template parameter support, and many C++ fixes.
27879 * Warren Levy for tremendous work on libgcj (Java Runtime Library)
27880 and random work on the Java front end.
27882 * Alain Lichnewsky ported GCC to the MIPS CPU.
27884 * Oskar Liljeblad for hacking on AWT and his many Java bug reports
27887 * Robert Lipe for OpenServer support, new testsuites, testing, etc.
27889 * Weiwen Liu for testing and various bug fixes.
27891 * Dave Love for his ongoing work with the Fortran front end and
27894 * Martin von Lo"wis for internal consistency checking infrastructure,
27895 various C++ improvements including namespace support, and tons of
27896 assistance with libstdc++/compiler merges.
27898 * H.J. Lu for his previous contributions to the steering committee,
27899 many x86 bug reports, prototype patches, and keeping the GNU/Linux
27902 * Greg McGary for random fixes and (someday) bounded pointers.
27904 * Andrew MacLeod for his ongoing work in building a real EH system,
27905 various code generation improvements, work on the global
27908 * Vladimir Makarov for hacking some ugly i960 problems, PowerPC
27909 hacking improvements to compile-time performance, overall
27910 knowledge and direction in the area of instruction scheduling, and
27911 design and implementation of the automaton based instruction
27914 * Bob Manson for his behind the scenes work on dejagnu.
27916 * Philip Martin for lots of libstdc++ string and vector iterator
27917 fixes and improvements, and string clean up and testsuites.
27919 * All of the Mauve project contributors, for Java test code.
27921 * Bryce McKinlay for numerous GCJ and libgcj fixes and improvements.
27923 * Adam Megacz for his work on the Microsoft Windows port of GCJ.
27925 * Michael Meissner for LRS framework, ia32, m32r, v850, m88k, MIPS,
27926 powerpc, haifa, ECOFF debug support, and other assorted hacking.
27928 * Jason Merrill for his direction via the steering committee and
27929 leading the G++ effort.
27931 * David Miller for his direction via the steering committee, lots of
27932 SPARC work, improvements in jump.c and interfacing with the Linux
27935 * Gary Miller ported GCC to Charles River Data Systems machines.
27937 * Alfred Minarik for libstdc++ string and ios bug fixes, and turning
27938 the entire libstdc++ testsuite namespace-compatible.
27940 * Mark Mitchell for his direction via the steering committee,
27941 mountains of C++ work, load/store hoisting out of loops, alias
27942 analysis improvements, ISO C `restrict' support, and serving as
27943 release manager for GCC 3.x.
27945 * Alan Modra for various GNU/Linux bits and testing.
27947 * Toon Moene for his direction via the steering committee, Fortran
27948 maintenance, and his ongoing work to make us make Fortran run fast.
27950 * Jason Molenda for major help in the care and feeding of all the
27951 services on the gcc.gnu.org (formerly egcs.cygnus.com)
27952 machine--mail, web services, ftp services, etc etc. Doing all
27953 this work on scrap paper and the backs of envelopes would have
27956 * Catherine Moore for fixing various ugly problems we have sent her
27957 way, including the haifa bug which was killing the Alpha & PowerPC
27960 * Mike Moreton for his various Java patches.
27962 * David Mosberger-Tang for various Alpha improvements, and for the
27963 initial IA-64 port.
27965 * Stephen Moshier contributed the floating point emulator that
27966 assists in cross-compilation and permits support for floating
27967 point numbers wider than 64 bits and for ISO C99 support.
27969 * Bill Moyer for his behind the scenes work on various issues.
27971 * Philippe De Muyter for his work on the m68k port.
27973 * Joseph S. Myers for his work on the PDP-11 port, format checking
27974 and ISO C99 support, and continuous emphasis on (and contributions
27977 * Nathan Myers for his work on libstdc++-v3: architecture and
27978 authorship through the first three snapshots, including
27979 implementation of locale infrastructure, string, shadow C headers,
27980 and the initial project documentation (DESIGN, CHECKLIST, and so
27981 forth). Later, more work on MT-safe string and shadow headers.
27983 * Felix Natter for documentation on porting libstdc++.
27985 * Nathanael Nerode for cleaning up the configuration/build process.
27987 * NeXT, Inc. donated the front end that supports the Objective-C
27990 * Hans-Peter Nilsson for the CRIS and MMIX ports, improvements to
27991 the search engine setup, various documentation fixes and other
27994 * Geoff Noer for his work on getting cygwin native builds working.
27996 * Diego Novillo for his SPEC performance tracking web pages and
27997 assorted fixes in the middle end and various back ends.
27999 * David O'Brien for the FreeBSD/alpha, FreeBSD/AMD x86-64,
28000 FreeBSD/ARM, FreeBSD/PowerPC, and FreeBSD/SPARC64 ports and
28001 related infrastructure improvements.
28003 * Alexandre Oliva for various build infrastructure improvements,
28004 scripts and amazing testing work, including keeping libtool issues
28007 * Stefan Olsson for work on mt_alloc.
28009 * Melissa O'Neill for various NeXT fixes.
28011 * Rainer Orth for random MIPS work, including improvements to GCC's
28012 o32 ABI support, improvements to dejagnu's MIPS support, Java
28013 configuration clean-ups and porting work, etc.
28015 * Hartmut Penner for work on the s390 port.
28017 * Paul Petersen wrote the machine description for the Alliant FX/8.
28019 * Alexandre Petit-Bianco for implementing much of the Java compiler
28020 and continued Java maintainership.
28022 * Matthias Pfaller for major improvements to the NS32k port.
28024 * Gerald Pfeifer for his direction via the steering committee,
28025 pointing out lots of problems we need to solve, maintenance of the
28026 web pages, and taking care of documentation maintenance in general.
28028 * Andrew Pinski for processing bug reports by the dozen.
28030 * Ovidiu Predescu for his work on the Objective-C front end and
28033 * Jerry Quinn for major performance improvements in C++ formatted
28036 * Ken Raeburn for various improvements to checker, MIPS ports and
28037 various cleanups in the compiler.
28039 * Rolf W. Rasmussen for hacking on AWT.
28041 * David Reese of Sun Microsystems contributed to the Solaris on
28044 * Volker Reichelt for keeping up with the problem reports.
28046 * Joern Rennecke for maintaining the sh port, loop, regmove & reload
28049 * Loren J. Rittle for improvements to libstdc++-v3 including the
28050 FreeBSD port, threading fixes, thread-related configury changes,
28051 critical threading documentation, and solutions to really tricky
28052 I/O problems, as well as keeping GCC properly working on FreeBSD
28053 and continuous testing.
28055 * Craig Rodrigues for processing tons of bug reports.
28057 * Ola Ro"nnerup for work on mt_alloc.
28059 * Gavin Romig-Koch for lots of behind the scenes MIPS work.
28061 * David Ronis inspired and encouraged Craig to rewrite the G77
28062 documentation in texinfo format by contributing a first pass at a
28063 translation of the old `g77-0.5.16/f/DOC' file.
28065 * Ken Rose for fixes to GCC's delay slot filling code.
28067 * Paul Rubin wrote most of the preprocessor.
28069 * Pe'tur Runo'lfsson for major performance improvements in C++
28070 formatted I/O and large file support in C++ filebuf.
28072 * Chip Salzenberg for libstdc++ patches and improvements to locales,
28073 traits, Makefiles, libio, libtool hackery, and "long long" support.
28075 * Juha Sarlin for improvements to the H8 code generator.
28077 * Greg Satz assisted in making GCC work on HP-UX for the 9000 series
28080 * Roger Sayle for improvements to constant folding and GCC's RTL
28081 optimizers as well as for fixing numerous bugs.
28083 * Bradley Schatz for his work on the GCJ FAQ.
28085 * Peter Schauer wrote the code to allow debugging to work on the
28088 * William Schelter did most of the work on the Intel 80386 support.
28090 * Tobias Schlu"ter for work on gfortran.
28092 * Bernd Schmidt for various code generation improvements and major
28093 work in the reload pass as well a serving as release manager for
28096 * Peter Schmid for constant testing of libstdc++--especially
28097 application testing, going above and beyond what was requested for
28098 the release criteria--and libstdc++ header file tweaks.
28100 * Jason Schroeder for jcf-dump patches.
28102 * Andreas Schwab for his work on the m68k port.
28104 * Lars Segerlund for work on gfortran.
28106 * Joel Sherrill for his direction via the steering committee, RTEMS
28107 contributions and RTEMS testing.
28109 * Nathan Sidwell for many C++ fixes/improvements.
28111 * Jeffrey Siegal for helping RMS with the original design of GCC,
28112 some code which handles the parse tree and RTL data structures,
28113 constant folding and help with the original VAX & m68k ports.
28115 * Kenny Simpson for prompting libstdc++ fixes due to defect reports
28116 from the LWG (thereby keeping GCC in line with updates from the
28119 * Franz Sirl for his ongoing work with making the PPC port stable
28122 * Andrey Slepuhin for assorted AIX hacking.
28124 * Christopher Smith did the port for Convex machines.
28126 * Danny Smith for his major efforts on the Mingw (and Cygwin) ports.
28128 * Randy Smith finished the Sun FPA support.
28130 * Scott Snyder for queue, iterator, istream, and string fixes and
28131 libstdc++ testsuite entries. Also for providing the patch to G77
28132 to add rudimentary support for `INTEGER*1', `INTEGER*2', and
28135 * Brad Spencer for contributions to the GLIBCPP_FORCE_NEW technique.
28137 * Richard Stallman, for writing the original GCC and launching the
28140 * Jan Stein of the Chalmers Computer Society provided support for
28141 Genix, as well as part of the 32000 machine description.
28143 * Nigel Stephens for various mips16 related fixes/improvements.
28145 * Jonathan Stone wrote the machine description for the Pyramid
28148 * Graham Stott for various infrastructure improvements.
28150 * John Stracke for his Java HTTP protocol fixes.
28152 * Mike Stump for his Elxsi port, G++ contributions over the years
28153 and more recently his vxworks contributions
28155 * Jeff Sturm for Java porting help, bug fixes, and encouragement.
28157 * Shigeya Suzuki for this fixes for the bsdi platforms.
28159 * Ian Lance Taylor for his mips16 work, general configury hacking,
28162 * Holger Teutsch provided the support for the Clipper CPU.
28164 * Gary Thomas for his ongoing work to make the PPC work for
28167 * Philipp Thomas for random bug fixes throughout the compiler
28169 * Jason Thorpe for thread support in libstdc++ on NetBSD.
28171 * Kresten Krab Thorup wrote the run time support for the Objective-C
28172 language and the fantastic Java bytecode interpreter.
28174 * Michael Tiemann for random bug fixes, the first instruction
28175 scheduler, initial C++ support, function integration, NS32k, SPARC
28176 and M88k machine description work, delay slot scheduling.
28178 * Andreas Tobler for his work porting libgcj to Darwin.
28180 * Teemu Torma for thread safe exception handling support.
28182 * Leonard Tower wrote parts of the parser, RTL generator, and RTL
28183 definitions, and of the VAX machine description.
28185 * Tom Tromey for internationalization support and for his many Java
28186 contributions and libgcj maintainership.
28188 * Lassi Tuura for improvements to config.guess to determine HP
28191 * Petter Urkedal for libstdc++ CXXFLAGS, math, and algorithms fixes.
28193 * Andy Vaught for the design and initial implementation of the
28194 gfortran front end.
28196 * Brent Verner for work with the libstdc++ cshadow files and their
28197 associated configure steps.
28199 * Todd Vierling for contributions for NetBSD ports.
28201 * Jonathan Wakely for contributing libstdc++ Doxygen notes and XHTML
28204 * Dean Wakerley for converting the install documentation from HTML
28205 to texinfo in time for GCC 3.0.
28207 * Krister Walfridsson for random bug fixes.
28209 * Feng Wang for contributions to gfortran.
28211 * Stephen M. Webb for time and effort on making libstdc++ shadow
28212 files work with the tricky Solaris 8+ headers, and for pushing the
28213 build-time header tree.
28215 * John Wehle for various improvements for the x86 code generator,
28216 related infrastructure improvements to help x86 code generation,
28217 value range propagation and other work, WE32k port.
28219 * Ulrich Weigand for work on the s390 port.
28221 * Zack Weinberg for major work on cpplib and various other bug fixes.
28223 * Matt Welsh for help with Linux Threads support in GCJ.
28225 * Urban Widmark for help fixing java.io.
28227 * Mark Wielaard for new Java library code and his work integrating
28230 * Dale Wiles helped port GCC to the Tahoe.
28232 * Bob Wilson from Tensilica, Inc. for the Xtensa port.
28234 * Jim Wilson for his direction via the steering committee, tackling
28235 hard problems in various places that nobody else wanted to work
28236 on, strength reduction and other loop optimizations.
28238 * Paul Woegerer and Tal Agmon for the CRX port.
28240 * Carlo Wood for various fixes.
28242 * Tom Wood for work on the m88k port.
28244 * Canqun Yang for work on gfortran.
28246 * Masanobu Yuhara of Fujitsu Laboratories implemented the machine
28247 description for the Tron architecture (specifically, the Gmicro).
28249 * Kevin Zachmann helped port GCC to the Tahoe.
28251 * Ayal Zaks for Swing Modulo Scheduling (SMS).
28253 * Xiaoqiang Zhang for work on gfortran.
28255 * Gilles Zunino for help porting Java to Irix.
28258 The following people are recognized for their contributions to GNAT,
28259 the Ada front end of GCC:
28262 * Romain Berrendonner
28312 * Hristian Kirtchev
28355 The following people are recognized for their contributions of new
28356 features, bug reports, testing and integration of classpath/libgcj for
28358 * Lillian Angel for `JTree' implementation and lots Free Swing
28359 additions and bugfixes.
28361 * Wolfgang Baer for `GapContent' bugfixes.
28363 * Anthony Balkissoon for `JList', Free Swing 1.5 updates and mouse
28364 event fixes, lots of Free Swing work including `JTable' editing.
28366 * Stuart Ballard for RMI constant fixes.
28368 * Goffredo Baroncelli for `HTTPURLConnection' fixes.
28370 * Gary Benson for `MessageFormat' fixes.
28372 * Daniel Bonniot for `Serialization' fixes.
28374 * Chris Burdess for lots of gnu.xml and http protocol fixes, `StAX'
28375 and `DOM xml:id' support.
28377 * Ka-Hing Cheung for `TreePath' and `TreeSelection' fixes.
28379 * Archie Cobbs for build fixes, VM interface updates,
28380 `URLClassLoader' updates.
28382 * Kelley Cook for build fixes.
28384 * Martin Cordova for Suggestions for better `SocketTimeoutException'.
28386 * David Daney for `BitSet' bugfixes, `HttpURLConnection' rewrite and
28389 * Thomas Fitzsimmons for lots of upgrades to the gtk+ AWT and Cairo
28390 2D support. Lots of imageio framework additions, lots of AWT and
28391 Free Swing bugfixes.
28393 * Jeroen Frijters for `ClassLoader' and nio cleanups, serialization
28394 fixes, better `Proxy' support, bugfixes and IKVM integration.
28396 * Santiago Gala for `AccessControlContext' fixes.
28398 * Nicolas Geoffray for `VMClassLoader' and `AccessController'
28401 * David Gilbert for `basic' and `metal' icon and plaf support and
28402 lots of documenting, Lots of Free Swing and metal theme additions.
28403 `MetalIconFactory' implementation.
28405 * Anthony Green for `MIDI' framework, `ALSA' and `DSSI' providers.
28407 * Andrew Haley for `Serialization' and `URLClassLoader' fixes, gcj
28410 * Kim Ho for `JFileChooser' implementation.
28412 * Andrew John Hughes for `Locale' and net fixes, URI RFC2986
28413 updates, `Serialization' fixes, `Properties' XML support and
28414 generic branch work, VMIntegration guide update.
28416 * Bastiaan Huisman for `TimeZone' bugfixing.
28418 * Andreas Jaeger for mprec updates.
28420 * Paul Jenner for better `-Werror' support.
28422 * Ito Kazumitsu for `NetworkInterface' implementation and updates.
28424 * Roman Kennke for `BoxLayout', `GrayFilter' and `SplitPane', plus
28425 bugfixes all over. Lots of Free Swing work including styled text.
28427 * Simon Kitching for `String' cleanups and optimization suggestions.
28429 * Michael Koch for configuration fixes, `Locale' updates, bug and
28432 * Guilhem Lavaux for configuration, thread and channel fixes and
28433 Kaffe integration. JCL native `Pointer' updates. Logger bugfixes.
28435 * David Lichteblau for JCL support library global/local reference
28438 * Aaron Luchko for JDWP updates and documentation fixes.
28440 * Ziga Mahkovec for `Graphics2D' upgraded to Cairo 0.5 and new regex
28443 * Sven de Marothy for BMP imageio support, CSS and `TextLayout'
28444 fixes. `GtkImage' rewrite, 2D, awt, free swing and date/time fixes
28445 and implementing the Qt4 peers.
28447 * Casey Marshall for crypto algorithm fixes, `FileChannel' lock,
28448 `SystemLogger' and `FileHandler' rotate implementations, NIO
28449 `FileChannel.map' support, security and policy updates.
28451 * Bryce McKinlay for RMI work.
28453 * Audrius Meskauskas for lots of Free Corba, RMI and HTML work plus
28454 testing and documenting.
28456 * Kalle Olavi Niemitalo for build fixes.
28458 * Rainer Orth for build fixes.
28460 * Andrew Overholt for `File' locking fixes.
28462 * Ingo Proetel for `Image', `Logger' and `URLClassLoader' updates.
28464 * Olga Rodimina for `MenuSelectionManager' implemenation.
28466 * Jan Roehrich for `BasicTreeUI' and `JTree' fixes.
28468 * Julian Scheid for documentation updates and gjdoc support.
28470 * Christian Schlichtherle for zip fixes and cleanups.
28472 * Robert Schuster for documentation updates and beans fixes,
28473 `TreeNode' enumerations and `ActionCommand' and various fixes, XML
28474 and URL, AWT and Free Swing bugfixes.
28476 * Keith Seitz for lots of JDWP work.
28478 * Christian Thalinger for 64-bit cleanups, Configuration and VM
28479 interface fixes and `CACAO' integration, `fdlibm' updates.
28481 * Gael Thomas for `VMClassLoader' boot packages support sugestions.
28483 * Andreas Tobler for Darwin and Solaris testing and fixing, `Qt4'
28484 support for Darwin/OS X, `Graphics2D' support, `gtk+' updates.
28486 * Dalibor Topic for better `DEBUG' support, build cleanups and Kaffe
28487 integration. `Qt4' build infrastructure, `SHA1PRNG' and
28488 `GdkPixbugDecoder' updates.
28490 * Tom Tromey for Eclipse integration, generics work, lots of bugfixes
28491 and gcj integration including coordinating The Big Merge.
28493 * Mark Wielaard for bugfixes, packaging and release management,
28494 `Clipboard' implementation, system call interrupts and network
28495 timeouts and `GdkPixpufDecoder' fixes.
28498 In addition to the above, all of which also contributed time and
28499 energy in testing GCC, we would like to thank the following for their
28500 contributions to testing:
28502 * Michael Abd-El-Malek
28512 * David Billinghurst
28516 * Stephane Bortzmeyer
28526 * Bradford Castalia
28546 * Charles-Antoine Gauthier
28568 * Kevin B. Hendricks
28572 * Christian Joensson
28580 * Anand Krishnaswamy
28582 * A. O. V. Le Blanc
28646 * Pedro A. M. Vazquez
28656 And finally we'd like to thank everyone who uses the compiler, submits
28657 bug reports and generally reminds us why we're doing this work in the
28661 File: gcc.info, Node: Option Index, Next: Keyword Index, Prev: Contributors, Up: Top
28666 GCC's command line options are indexed here without any initial `-' or
28667 `--'. Where an option has both positive and negative forms (such as
28668 `-fOPTION' and `-fno-OPTION'), relevant entries in the manual are
28669 indexed under the most appropriate form; it may sometimes be useful to
28670 look up both forms.
28675 * ###: Overall Options. (line 191)
28676 * -dynamiclib: Darwin Options. (line 116)
28677 * -force_cpusubtype_ALL: Darwin Options. (line 121)
28678 * -fsplit-ivs-in-unroller: Optimize Options. (line 767)
28679 * -fvariable-expansion-in-unroller: Optimize Options. (line 781)
28680 * -gfull: Darwin Options. (line 64)
28681 * -gused: Darwin Options. (line 59)
28682 * -mone-byte-bool: Darwin Options. (line 75)
28683 * A: Preprocessor Options.
28685 * all_load: Darwin Options. (line 95)
28686 * allowable_client: Darwin Options. (line 182)
28687 * ansi <1>: Non-bugs. (line 107)
28688 * ansi <2>: Other Builtins. (line 22)
28689 * ansi <3>: Preprocessor Options.
28691 * ansi <4>: C Dialect Options. (line 11)
28692 * ansi: Standards. (line 13)
28693 * arch_errors_fatal: Darwin Options. (line 99)
28694 * aux-info: C Dialect Options. (line 98)
28695 * b: Target Options. (line 13)
28696 * B: Directory Options. (line 41)
28697 * bcopy-builtin: PDP-11 Options. (line 32)
28698 * bind_at_load: Darwin Options. (line 103)
28699 * bundle: Darwin Options. (line 108)
28700 * bundle_loader: Darwin Options. (line 112)
28701 * c: Link Options. (line 20)
28702 * C: Preprocessor Options.
28704 * c: Overall Options. (line 146)
28705 * client_name: Darwin Options. (line 182)
28706 * combine: Overall Options. (line 202)
28707 * compatibility_version: Darwin Options. (line 182)
28708 * coverage: Debugging Options. (line 161)
28709 * crossjumping: Optimize Options. (line 461)
28710 * current_version: Darwin Options. (line 182)
28711 * D: Preprocessor Options.
28713 * d: Debugging Options. (line 213)
28714 * da: Debugging Options. (line 384)
28715 * dA: Debugging Options. (line 226)
28716 * dB: Debugging Options. (line 235)
28717 * db: Debugging Options. (line 231)
28718 * dC: Debugging Options. (line 245)
28719 * dc: Debugging Options. (line 239)
28720 * dD <1>: Preprocessor Options.
28722 * dD: Debugging Options. (line 259)
28723 * dd: Debugging Options. (line 253)
28724 * dE: Debugging Options. (line 264)
28725 * dead_strip: Darwin Options. (line 182)
28726 * dependency-file: Darwin Options. (line 182)
28727 * df: Debugging Options. (line 269)
28728 * dG: Debugging Options. (line 281)
28729 * dg: Debugging Options. (line 276)
28730 * dH: Debugging Options. (line 387)
28731 * dh: Debugging Options. (line 288)
28732 * dI: Preprocessor Options.
28734 * di: Debugging Options. (line 292)
28735 * dj: Debugging Options. (line 296)
28736 * dk: Debugging Options. (line 300)
28737 * dL: Debugging Options. (line 310)
28738 * dl: Debugging Options. (line 305)
28739 * dM: Preprocessor Options.
28741 * dm: Debugging Options. (line 390)
28742 * dM: Debugging Options. (line 321)
28743 * dm: Debugging Options. (line 317)
28744 * dN <1>: Preprocessor Options.
28746 * dN: Debugging Options. (line 330)
28747 * dn: Debugging Options. (line 326)
28748 * do: Debugging Options. (line 334)
28749 * dP: Debugging Options. (line 399)
28750 * dp: Debugging Options. (line 394)
28751 * dR: Debugging Options. (line 342)
28752 * dr: Debugging Options. (line 338)
28753 * dS: Debugging Options. (line 351)
28754 * ds: Debugging Options. (line 346)
28755 * dT: Debugging Options. (line 360)
28756 * dt: Debugging Options. (line 355)
28757 * dumpmachine: Debugging Options. (line 741)
28758 * dumpspecs: Debugging Options. (line 749)
28759 * dumpversion: Debugging Options. (line 745)
28760 * dv: Debugging Options. (line 403)
28761 * dV: Debugging Options. (line 365)
28762 * dw: Debugging Options. (line 372)
28763 * dx: Debugging Options. (line 408)
28764 * dy: Debugging Options. (line 412)
28765 * dylib_file: Darwin Options. (line 182)
28766 * dylinker_install_name: Darwin Options. (line 182)
28767 * dynamic: Darwin Options. (line 182)
28768 * dZ: Debugging Options. (line 380)
28769 * dz: Debugging Options. (line 376)
28770 * E <1>: Link Options. (line 20)
28771 * E: Overall Options. (line 167)
28772 * EB <1>: MIPS Options. (line 7)
28773 * EB: ARC Options. (line 12)
28774 * EL <1>: MIPS Options. (line 10)
28775 * EL: ARC Options. (line 9)
28776 * exported_symbols_list: Darwin Options. (line 182)
28777 * F: Darwin Options. (line 32)
28778 * fabi-version: C++ Dialect Options.
28780 * falign-functions: Optimize Options. (line 894)
28781 * falign-jumps: Optimize Options. (line 944)
28782 * falign-labels: Optimize Options. (line 912)
28783 * falign-loops: Optimize Options. (line 930)
28784 * fargument-alias: Code Gen Options. (line 336)
28785 * fargument-noalias: Code Gen Options. (line 336)
28786 * fargument-noalias-global: Code Gen Options. (line 336)
28787 * fasynchronous-unwind-tables: Code Gen Options. (line 64)
28788 * fbounds-check <1>: Code Gen Options. (line 15)
28789 * fbounds-check: Optimize Options. (line 329)
28790 * fbranch-probabilities: Optimize Options. (line 1176)
28791 * fbranch-target-load-optimize: Optimize Options. (line 1291)
28792 * fbranch-target-load-optimize2: Optimize Options. (line 1297)
28793 * fbtr-bb-exclusive: Optimize Options. (line 1301)
28794 * fcall-saved <1>: Interoperation. (line 150)
28795 * fcall-saved: Code Gen Options. (line 238)
28796 * fcall-used: Code Gen Options. (line 224)
28797 * fcaller-saves: Optimize Options. (line 601)
28798 * fcheck-new: C++ Dialect Options.
28800 * fcommon: Variable Attributes.
28802 * fcond-mismatch: C Dialect Options. (line 208)
28803 * fconserve-space: C++ Dialect Options.
28805 * fconstant-string-class: Objective-C and Objective-C++ Dialect Options.
28807 * fcse-follow-jumps: Optimize Options. (line 372)
28808 * fcse-skip-blocks: Optimize Options. (line 381)
28809 * fcx-limited-range: Optimize Options. (line 1162)
28810 * fdata-sections: Optimize Options. (line 1272)
28811 * fdelayed-branch: Optimize Options. (line 514)
28812 * fdelete-null-pointer-checks: Optimize Options. (line 483)
28813 * fdiagnostics-show-location: Language Independent Options.
28815 * fdiagnostics-show-options: Language Independent Options.
28817 * fdollars-in-identifiers <1>: Interoperation. (line 146)
28818 * fdollars-in-identifiers: Preprocessor Options.
28820 * fdump-class-hierarchy: Debugging Options. (line 431)
28821 * fdump-ipa: Debugging Options. (line 438)
28822 * fdump-rtl-all: Debugging Options. (line 384)
28823 * fdump-rtl-bbro: Debugging Options. (line 235)
28824 * fdump-rtl-bp: Debugging Options. (line 231)
28825 * fdump-rtl-btl: Debugging Options. (line 253)
28826 * fdump-rtl-bypass: Debugging Options. (line 281)
28827 * fdump-rtl-ce1: Debugging Options. (line 245)
28828 * fdump-rtl-ce2: Debugging Options. (line 245)
28829 * fdump-rtl-ce3: Debugging Options. (line 264)
28830 * fdump-rtl-cfg: Debugging Options. (line 269)
28831 * fdump-rtl-combine: Debugging Options. (line 239)
28832 * fdump-rtl-cse: Debugging Options. (line 346)
28833 * fdump-rtl-cse2: Debugging Options. (line 355)
28834 * fdump-rtl-dbr: Debugging Options. (line 253)
28835 * fdump-rtl-eh: Debugging Options. (line 288)
28836 * fdump-rtl-expand: Debugging Options. (line 338)
28837 * fdump-rtl-flow2: Debugging Options. (line 372)
28838 * fdump-rtl-gcse: Debugging Options. (line 281)
28839 * fdump-rtl-greg: Debugging Options. (line 276)
28840 * fdump-rtl-jump: Debugging Options. (line 296)
28841 * fdump-rtl-life: Debugging Options. (line 269)
28842 * fdump-rtl-loop: Debugging Options. (line 310)
28843 * fdump-rtl-loop2: Debugging Options. (line 310)
28844 * fdump-rtl-lreg: Debugging Options. (line 305)
28845 * fdump-rtl-mach: Debugging Options. (line 321)
28846 * fdump-rtl-peephole2: Debugging Options. (line 376)
28847 * fdump-rtl-postreload: Debugging Options. (line 334)
28848 * fdump-rtl-regmove: Debugging Options. (line 330)
28849 * fdump-rtl-rnreg: Debugging Options. (line 326)
28850 * fdump-rtl-sched: Debugging Options. (line 351)
28851 * fdump-rtl-sched2: Debugging Options. (line 342)
28852 * fdump-rtl-sibling: Debugging Options. (line 292)
28853 * fdump-rtl-sms: Debugging Options. (line 317)
28854 * fdump-rtl-stack: Debugging Options. (line 300)
28855 * fdump-rtl-tracer: Debugging Options. (line 360)
28856 * fdump-rtl-vartrack: Debugging Options. (line 365)
28857 * fdump-rtl-vpt: Debugging Options. (line 365)
28858 * fdump-rtl-web: Debugging Options. (line 380)
28859 * fdump-translation-unit: Debugging Options. (line 423)
28860 * fdump-tree: Debugging Options. (line 453)
28861 * fdump-tree-alias: Debugging Options. (line 541)
28862 * fdump-tree-all: Debugging Options. (line 626)
28863 * fdump-tree-ccp: Debugging Options. (line 545)
28864 * fdump-tree-cfg: Debugging Options. (line 516)
28865 * fdump-tree-ch: Debugging Options. (line 528)
28866 * fdump-tree-copyprop: Debugging Options. (line 561)
28867 * fdump-tree-copyrename: Debugging Options. (line 607)
28868 * fdump-tree-dce: Debugging Options. (line 569)
28869 * fdump-tree-dom: Debugging Options. (line 587)
28870 * fdump-tree-dse: Debugging Options. (line 592)
28871 * fdump-tree-forwprop: Debugging Options. (line 602)
28872 * fdump-tree-fre: Debugging Options. (line 557)
28873 * fdump-tree-gimple: Debugging Options. (line 511)
28874 * fdump-tree-mudflap: Debugging Options. (line 573)
28875 * fdump-tree-nrv: Debugging Options. (line 612)
28876 * fdump-tree-phiopt: Debugging Options. (line 597)
28877 * fdump-tree-pre: Debugging Options. (line 553)
28878 * fdump-tree-salias: Debugging Options. (line 536)
28879 * fdump-tree-sink: Debugging Options. (line 583)
28880 * fdump-tree-sra: Debugging Options. (line 578)
28881 * fdump-tree-ssa: Debugging Options. (line 532)
28882 * fdump-tree-store_copyprop: Debugging Options. (line 565)
28883 * fdump-tree-storeccp: Debugging Options. (line 549)
28884 * fdump-tree-vcg: Debugging Options. (line 520)
28885 * fdump-tree-vect: Debugging Options. (line 617)
28886 * fdump-tree-vrp: Debugging Options. (line 622)
28887 * fdump-unnumbered: Debugging Options. (line 415)
28888 * fearly-inlining: Optimize Options. (line 206)
28889 * feliminate-dwarf2-dups: Debugging Options. (line 117)
28890 * feliminate-unused-debug-symbols: Debugging Options. (line 52)
28891 * feliminate-unused-debug-types: Debugging Options. (line 753)
28892 * fexceptions: Code Gen Options. (line 34)
28893 * fexec-charset: Preprocessor Options.
28895 * fexpensive-optimizations: Optimize Options. (line 496)
28896 * fextended-identifiers: Preprocessor Options.
28898 * ffast-math: Optimize Options. (line 1053)
28899 * ffinite-math-only: Optimize Options. (line 1097)
28900 * ffix-and-continue: Darwin Options. (line 89)
28901 * ffixed: Code Gen Options. (line 212)
28902 * ffloat-store <1>: Disappointments. (line 77)
28903 * ffloat-store: Optimize Options. (line 1039)
28904 * ffor-scope: C++ Dialect Options.
28906 * fforce-addr: Optimize Options. (line 156)
28907 * fforce-mem: Optimize Options. (line 148)
28908 * ffreestanding <1>: Function Attributes.
28910 * ffreestanding <2>: Warning Options. (line 94)
28911 * ffreestanding <3>: C Dialect Options. (line 169)
28912 * ffreestanding: Standards. (line 81)
28913 * ffriend-injection: C++ Dialect Options.
28915 * ffunction-sections: Optimize Options. (line 1272)
28916 * fgcse: Optimize Options. (line 400)
28917 * fgcse-after-reload: Optimize Options. (line 436)
28918 * fgcse-las: Optimize Options. (line 429)
28919 * fgcse-lm: Optimize Options. (line 411)
28920 * fgcse-sm: Optimize Options. (line 420)
28921 * fgnu-runtime: Objective-C and Objective-C++ Dialect Options.
28923 * fhosted: C Dialect Options. (line 162)
28924 * filelist: Darwin Options. (line 182)
28925 * findirect-data: Darwin Options. (line 89)
28926 * finhibit-size-directive: Code Gen Options. (line 154)
28927 * finline-functions: Optimize Options. (line 187)
28928 * finline-functions-called-once: Optimize Options. (line 198)
28929 * finline-limit: Optimize Options. (line 216)
28930 * finput-charset: Preprocessor Options.
28932 * finstrument-functions <1>: Function Attributes.
28934 * finstrument-functions: Code Gen Options. (line 268)
28935 * fkeep-inline-functions <1>: Inline. (line 51)
28936 * fkeep-inline-functions: Optimize Options. (line 254)
28937 * fkeep-static-consts: Optimize Options. (line 261)
28938 * flat_namespace: Darwin Options. (line 182)
28939 * fleading-underscore: Code Gen Options. (line 351)
28940 * floop-optimize: Optimize Options. (line 441)
28941 * floop-optimize2: Optimize Options. (line 448)
28942 * fmem-report: Debugging Options. (line 142)
28943 * fmessage-length: Language Independent Options.
28945 * fmodulo-sched: Optimize Options. (line 290)
28946 * fmove-loop-invariants: Optimize Options. (line 1255)
28947 * fms-extensions <1>: Unnamed Fields. (line 37)
28948 * fms-extensions <2>: C++ Dialect Options.
28950 * fms-extensions: C Dialect Options. (line 179)
28951 * fmudflap: Optimize Options. (line 336)
28952 * fmudflapir: Optimize Options. (line 336)
28953 * fmudflapth: Optimize Options. (line 336)
28954 * fnext-runtime: Objective-C and Objective-C++ Dialect Options.
28956 * fno-access-control: C++ Dialect Options.
28958 * fno-asm: C Dialect Options. (line 114)
28959 * fno-branch-count-reg: Optimize Options. (line 295)
28960 * fno-builtin <1>: Other Builtins. (line 14)
28961 * fno-builtin <2>: Function Attributes.
28963 * fno-builtin <3>: Warning Options. (line 94)
28964 * fno-builtin: C Dialect Options. (line 128)
28965 * fno-common <1>: Variable Attributes.
28967 * fno-common: Code Gen Options. (line 142)
28968 * fno-const-strings: C++ Dialect Options.
28970 * fno-cprop-registers: Optimize Options. (line 1010)
28971 * fno-cx-limited-range: Optimize Options. (line 1162)
28972 * fno-default-inline <1>: Inline. (line 46)
28973 * fno-default-inline <2>: Optimize Options. (line 133)
28974 * fno-default-inline: C++ Dialect Options.
28976 * fno-defer-pop: Optimize Options. (line 140)
28977 * fno-elide-constructors: C++ Dialect Options.
28979 * fno-enforce-eh-specs: C++ Dialect Options.
28981 * fno-for-scope: C++ Dialect Options.
28983 * fno-function-cse: Optimize Options. (line 306)
28984 * fno-gnu-keywords: C++ Dialect Options.
28986 * fno-guess-branch-probability: Optimize Options. (line 804)
28987 * fno-ident: Code Gen Options. (line 151)
28988 * fno-implement-inlines <1>: C++ Interface. (line 75)
28989 * fno-implement-inlines: C++ Dialect Options.
28991 * fno-implicit-inline-templates: C++ Dialect Options.
28993 * fno-implicit-templates <1>: Template Instantiation.
28995 * fno-implicit-templates: C++ Dialect Options.
28997 * fno-inline: Optimize Options. (line 181)
28998 * fno-jump-tables: Code Gen Options. (line 204)
28999 * fno-math-errno: Optimize Options. (line 1066)
29000 * fno-nil-receivers: Objective-C and Objective-C++ Dialect Options.
29002 * fno-nonansi-builtins: C++ Dialect Options.
29004 * fno-operator-names: C++ Dialect Options.
29006 * fno-optional-diags: C++ Dialect Options.
29008 * fno-peephole: Optimize Options. (line 795)
29009 * fno-peephole2: Optimize Options. (line 795)
29010 * fno-rtti: C++ Dialect Options.
29012 * fno-sched-interblock: Optimize Options. (line 540)
29013 * fno-sched-spec: Optimize Options. (line 545)
29014 * fno-show-column: Preprocessor Options.
29016 * fno-signed-bitfields: C Dialect Options. (line 241)
29017 * fno-stack-limit: Code Gen Options. (line 320)
29018 * fno-threadsafe-statics: C++ Dialect Options.
29020 * fno-trapping-math: Optimize Options. (line 1107)
29021 * fno-unsigned-bitfields: C Dialect Options. (line 241)
29022 * fno-weak: C++ Dialect Options.
29024 * fno-working-directory: Preprocessor Options.
29026 * fno-zero-initialized-in-bss: Optimize Options. (line 317)
29027 * fnon-call-exceptions: Code Gen Options. (line 48)
29028 * fobjc-call-cxx-cdtors: Objective-C and Objective-C++ Dialect Options.
29030 * fobjc-direct-dispatch: Objective-C and Objective-C++ Dialect Options.
29032 * fobjc-exceptions: Objective-C and Objective-C++ Dialect Options.
29034 * fobjc-gc: Objective-C and Objective-C++ Dialect Options.
29036 * fomit-frame-pointer: Optimize Options. (line 160)
29037 * foptimize-register-move: Optimize Options. (line 503)
29038 * foptimize-sibling-calls: Optimize Options. (line 176)
29039 * force_flat_namespace: Darwin Options. (line 182)
29040 * fpack-struct: Code Gen Options. (line 255)
29041 * fpcc-struct-return <1>: Incompatibilities. (line 170)
29042 * fpcc-struct-return: Code Gen Options. (line 70)
29043 * fpch-deps: Preprocessor Options.
29045 * fpch-preprocess: Preprocessor Options.
29047 * fpeel-loops: Optimize Options. (line 1247)
29048 * fpermissive: C++ Dialect Options.
29050 * fPIC: Code Gen Options. (line 188)
29051 * fpic: Code Gen Options. (line 170)
29052 * fPIE: Code Gen Options. (line 198)
29053 * fpie: Code Gen Options. (line 198)
29054 * fprefetch-loop-arrays: Optimize Options. (line 786)
29055 * fpreprocessed: Preprocessor Options.
29057 * fprofile-arcs <1>: Other Builtins. (line 236)
29058 * fprofile-arcs: Debugging Options. (line 146)
29059 * fprofile-generate: Optimize Options. (line 1017)
29060 * fprofile-use: Optimize Options. (line 1026)
29061 * fprofile-values: Optimize Options. (line 1195)
29062 * frandom-string: Debugging Options. (line 636)
29063 * freg-struct-return: Code Gen Options. (line 88)
29064 * fregmove: Optimize Options. (line 503)
29065 * frename-registers: Optimize Options. (line 1214)
29066 * freorder-blocks: Optimize Options. (line 821)
29067 * freorder-blocks-and-partition: Optimize Options. (line 827)
29068 * freorder-functions: Optimize Options. (line 838)
29069 * freplace-objc-classes: Objective-C and Objective-C++ Dialect Options.
29071 * frepo <1>: Template Instantiation.
29073 * frepo: C++ Dialect Options.
29075 * frerun-cse-after-loop: Optimize Options. (line 389)
29076 * frerun-loop-opt: Optimize Options. (line 395)
29077 * frounding-math: Optimize Options. (line 1122)
29078 * fsched-spec-load: Optimize Options. (line 550)
29079 * fsched-spec-load-dangerous: Optimize Options. (line 555)
29080 * fsched-stalled-insns: Optimize Options. (line 560)
29081 * fsched-stalled-insns-dep: Optimize Options. (line 565)
29082 * fsched-verbose: Debugging Options. (line 646)
29083 * fsched2-use-superblocks: Optimize Options. (line 572)
29084 * fsched2-use-traces: Optimize Options. (line 583)
29085 * fschedule-insns: Optimize Options. (line 521)
29086 * fschedule-insns2: Optimize Options. (line 531)
29087 * fscheduling-in-modulo-scheduled-loops: Optimize Options. (line 595)
29088 * fshared-data: Code Gen Options. (line 135)
29089 * fshort-double: Code Gen Options. (line 117)
29090 * fshort-enums <1>: Non-bugs. (line 42)
29091 * fshort-enums <2>: Type Attributes. (line 112)
29092 * fshort-enums <3>: Structures unions enumerations and bit-fields implementation.
29094 * fshort-enums: Code Gen Options. (line 106)
29095 * fshort-wchar: Code Gen Options. (line 125)
29096 * fsignaling-nans: Optimize Options. (line 1142)
29097 * fsigned-bitfields <1>: Non-bugs. (line 57)
29098 * fsigned-bitfields: C Dialect Options. (line 241)
29099 * fsigned-char <1>: Characters implementation.
29101 * fsigned-char: C Dialect Options. (line 231)
29102 * fsingle-precision-constant: Optimize Options. (line 1157)
29103 * fstack-check: Code Gen Options. (line 305)
29104 * fstack-limit-register: Code Gen Options. (line 320)
29105 * fstack-limit-symbol: Code Gen Options. (line 320)
29106 * fstats: C++ Dialect Options.
29108 * fstrength-reduce: Optimize Options. (line 357)
29109 * fstrict-aliasing: Optimize Options. (line 851)
29110 * fsyntax-only: Warning Options. (line 23)
29111 * ftabstop: Preprocessor Options.
29113 * ftemplate-depth: C++ Dialect Options.
29115 * ftest-coverage: Debugging Options. (line 202)
29116 * fthread-jumps: Optimize Options. (line 363)
29117 * ftime-report: Debugging Options. (line 138)
29118 * ftracer: Optimize Options. (line 750)
29119 * ftrapv: Code Gen Options. (line 22)
29120 * ftree-vect-loop-version: Optimize Options. (line 732)
29121 * ftree-vectorizer-verbose: Debugging Options. (line 630)
29122 * funit-at-a-time: Optimize Options. (line 957)
29123 * funroll-all-loops: Optimize Options. (line 761)
29124 * funroll-loops: Optimize Options. (line 755)
29125 * funsafe-loop-optimizations: Optimize Options. (line 453)
29126 * funsafe-math-optimizations: Optimize Options. (line 1083)
29127 * funsigned-bitfields <1>: Non-bugs. (line 57)
29128 * funsigned-bitfields <2>: Structures unions enumerations and bit-fields implementation.
29130 * funsigned-bitfields: C Dialect Options. (line 241)
29131 * funsigned-char <1>: Characters implementation.
29133 * funsigned-char: C Dialect Options. (line 213)
29134 * funswitch-loops: Optimize Options. (line 1259)
29135 * funwind-tables: Code Gen Options. (line 57)
29136 * fuse-cxa-atexit: C++ Dialect Options.
29138 * fvar-tracking: Debugging Options. (line 689)
29139 * fverbose-asm: Code Gen Options. (line 161)
29140 * fvisibility: Code Gen Options. (line 370)
29141 * fvisibility-inlines-hidden: C++ Dialect Options.
29143 * fvpt: Optimize Options. (line 1205)
29144 * fweb: Optimize Options. (line 986)
29145 * fwhole-program: Optimize Options. (line 997)
29146 * fwide-exec-charset: Preprocessor Options.
29148 * fworking-directory: Preprocessor Options.
29150 * fwrapv: Code Gen Options. (line 26)
29151 * fzero-link: Objective-C and Objective-C++ Dialect Options.
29153 * G <1>: System V Options. (line 10)
29154 * G <2>: RS/6000 and PowerPC Options.
29156 * G <3>: MIPS Options. (line 194)
29157 * G: M32R/D Options. (line 57)
29158 * g: Debugging Options. (line 10)
29159 * gcoff: Debugging Options. (line 62)
29160 * gdwarf-2: Debugging Options. (line 80)
29161 * gen-decls: Objective-C and Objective-C++ Dialect Options.
29163 * ggdb: Debugging Options. (line 38)
29164 * gnu-ld: HPPA Options. (line 113)
29165 * gstabs: Debugging Options. (line 44)
29166 * gstabs+: Debugging Options. (line 56)
29167 * gvms: Debugging Options. (line 87)
29168 * gxcoff: Debugging Options. (line 67)
29169 * gxcoff+: Debugging Options. (line 72)
29170 * H: Preprocessor Options.
29172 * headerpad_max_install_names: Darwin Options. (line 182)
29173 * help <1>: Preprocessor Options.
29175 * help: Overall Options. (line 218)
29176 * hp-ld: HPPA Options. (line 125)
29177 * I <1>: Directory Options. (line 10)
29178 * I: Preprocessor Options.
29180 * I- <1>: Directory Options. (line 107)
29181 * I-: Preprocessor Options.
29183 * idirafter: Preprocessor Options.
29185 * if-conversion: Optimize Options. (line 468)
29186 * if-conversion2: Optimize Options. (line 477)
29187 * imacros: Preprocessor Options.
29189 * image_base: Darwin Options. (line 182)
29190 * include: Preprocessor Options.
29192 * init: Darwin Options. (line 182)
29193 * install_name: Darwin Options. (line 182)
29194 * iprefix: Preprocessor Options.
29196 * iquote <1>: Directory Options. (line 31)
29197 * iquote: Preprocessor Options.
29199 * isysroot: Preprocessor Options.
29201 * isystem: Preprocessor Options.
29203 * iwithprefix: Preprocessor Options.
29205 * iwithprefixbefore: Preprocessor Options.
29207 * keep_private_externs: Darwin Options. (line 182)
29208 * L: Directory Options. (line 37)
29209 * l: Link Options. (line 26)
29210 * lobjc: Link Options. (line 53)
29211 * M: Preprocessor Options.
29213 * m1: SH Options. (line 9)
29214 * m10: PDP-11 Options. (line 29)
29215 * m128bit-long-double: i386 and x86-64 Options.
29217 * m16-bit: CRIS Options. (line 69)
29218 * m2: SH Options. (line 12)
29219 * m210: MCore Options. (line 43)
29220 * m3: SH Options. (line 18)
29221 * m31: S/390 and zSeries Options.
29223 * m32 <1>: SPARC Options. (line 187)
29224 * m32 <2>: RS/6000 and PowerPC Options.
29226 * m32: i386 and x86-64 Options.
29228 * m32-bit: CRIS Options. (line 69)
29229 * m32r: M32R/D Options. (line 15)
29230 * m32r2: M32R/D Options. (line 9)
29231 * m32rx: M32R/D Options. (line 12)
29232 * m340: MCore Options. (line 43)
29233 * m386: i386 and x86-64 Options.
29235 * m3dnow: i386 and x86-64 Options.
29237 * m3e: SH Options. (line 21)
29238 * m4: SH Options. (line 35)
29239 * m4-nofpu: SH Options. (line 24)
29240 * m4-single: SH Options. (line 31)
29241 * m4-single-only: SH Options. (line 27)
29242 * m40: PDP-11 Options. (line 23)
29243 * m45: PDP-11 Options. (line 26)
29244 * m486: i386 and x86-64 Options.
29246 * m4a: SH Options. (line 50)
29247 * m4a-nofpu: SH Options. (line 38)
29248 * m4a-single: SH Options. (line 46)
29249 * m4a-single-only: SH Options. (line 42)
29250 * m4al: SH Options. (line 53)
29251 * m4byte-functions: MCore Options. (line 27)
29252 * m5200: M680x0 Options. (line 59)
29253 * m64 <1>: SPARC Options. (line 187)
29254 * m64 <2>: S/390 and zSeries Options.
29256 * m64 <3>: RS/6000 and PowerPC Options.
29258 * m64: i386 and x86-64 Options.
29260 * m68000: M680x0 Options. (line 13)
29261 * m68020: M680x0 Options. (line 21)
29262 * m68020-40: M680x0 Options. (line 66)
29263 * m68020-60: M680x0 Options. (line 73)
29264 * m68030: M680x0 Options. (line 30)
29265 * m68040: M680x0 Options. (line 34)
29266 * m68060: M680x0 Options. (line 42)
29267 * m6811: M68hc1x Options. (line 13)
29268 * m6812: M68hc1x Options. (line 18)
29269 * m68881: M680x0 Options. (line 25)
29270 * m68hc11: M68hc1x Options. (line 13)
29271 * m68hc12: M68hc1x Options. (line 18)
29272 * m68hcs12: M68hc1x Options. (line 23)
29273 * m68S12: M68hc1x Options. (line 23)
29274 * m8-bit: CRIS Options. (line 69)
29275 * m96bit-long-double: i386 and x86-64 Options.
29277 * mabi <1>: RS/6000 and PowerPC Options.
29279 * mabi: ARM Options. (line 10)
29280 * mabi-mmixware: MMIX Options. (line 20)
29281 * mabi=32: MIPS Options. (line 89)
29282 * mabi=64: MIPS Options. (line 89)
29283 * mabi=eabi: MIPS Options. (line 89)
29284 * mabi=gnu: MMIX Options. (line 20)
29285 * mabi=ibmlongdouble: RS/6000 and PowerPC Options.
29287 * mabi=ieeelongdouble: RS/6000 and PowerPC Options.
29289 * mabi=n32: MIPS Options. (line 89)
29290 * mabi=no-spe: RS/6000 and PowerPC Options.
29292 * mabi=o64: MIPS Options. (line 89)
29293 * mabi=spe: RS/6000 and PowerPC Options.
29295 * mabicalls: MIPS Options. (line 100)
29296 * mabort-on-noreturn: ARM Options. (line 144)
29297 * mabshi: PDP-11 Options. (line 55)
29298 * mac0: PDP-11 Options. (line 16)
29299 * macc-4: FRV Options. (line 113)
29300 * macc-8: FRV Options. (line 116)
29301 * maccumulate-outgoing-args: i386 and x86-64 Options.
29303 * madjust-unroll: SH Options. (line 175)
29304 * mads: RS/6000 and PowerPC Options.
29306 * maix-struct-return: RS/6000 and PowerPC Options.
29308 * maix32: RS/6000 and PowerPC Options.
29310 * maix64: RS/6000 and PowerPC Options.
29312 * malign-300: H8/300 Options. (line 31)
29313 * malign-double: i386 and x86-64 Options.
29315 * malign-int: M680x0 Options. (line 128)
29316 * malign-labels: FRV Options. (line 104)
29317 * malign-loops: M32R/D Options. (line 73)
29318 * malign-natural: RS/6000 and PowerPC Options.
29320 * malign-power: RS/6000 and PowerPC Options.
29322 * malloc-cc: FRV Options. (line 25)
29323 * malpha-as: DEC Alpha Options. (line 159)
29324 * maltivec: RS/6000 and PowerPC Options.
29326 * mam33: MN10300 Options. (line 17)
29327 * maout: CRIS Options. (line 92)
29328 * mapcs: ARM Options. (line 22)
29329 * mapcs-frame: ARM Options. (line 14)
29330 * mapp-regs <1>: V850 Options. (line 57)
29331 * mapp-regs: SPARC Options. (line 10)
29332 * march <1>: S/390 and zSeries Options.
29334 * march <2>: MT Options. (line 9)
29335 * march <3>: MIPS Options. (line 14)
29336 * march <4>: i386 and x86-64 Options.
29338 * march <5>: HPPA Options. (line 9)
29339 * march <6>: CRIS Options. (line 10)
29340 * march: ARM Options. (line 109)
29341 * masm=DIALECT: i386 and x86-64 Options.
29343 * mauto-incdec: M68hc1x Options. (line 26)
29344 * mauto-pic: IA-64 Options. (line 50)
29345 * mb: SH Options. (line 58)
29346 * mbacc: MT Options. (line 16)
29347 * mbackchain: S/390 and zSeries Options.
29349 * mbase-addresses: MMIX Options. (line 54)
29350 * mbcopy: PDP-11 Options. (line 36)
29351 * mbig <1>: TMS320C3x/C4x Options.
29353 * mbig: RS/6000 and PowerPC Options.
29355 * mbig-endian <1>: RS/6000 and PowerPC Options.
29357 * mbig-endian <2>: MCore Options. (line 39)
29358 * mbig-endian <3>: IA-64 Options. (line 9)
29359 * mbig-endian: ARM Options. (line 72)
29360 * mbig-memory: TMS320C3x/C4x Options.
29362 * mbig-switch <1>: V850 Options. (line 52)
29363 * mbig-switch: HPPA Options. (line 23)
29364 * mbigtable: SH Options. (line 74)
29365 * mbit-align: RS/6000 and PowerPC Options.
29367 * mbitfield: M680x0 Options. (line 100)
29368 * mbk: TMS320C3x/C4x Options.
29370 * mbranch-cheap: PDP-11 Options. (line 65)
29371 * mbranch-cost=NUMBER: M32R/D Options. (line 82)
29372 * mbranch-expensive: PDP-11 Options. (line 61)
29373 * mbranch-likely: MIPS Options. (line 345)
29374 * mbranch-predict: MMIX Options. (line 49)
29375 * mbss-plt: RS/6000 and PowerPC Options.
29377 * mbuild-constants: DEC Alpha Options. (line 142)
29378 * mbwx: DEC Alpha Options. (line 171)
29379 * mc68000: M680x0 Options. (line 13)
29380 * mc68020: M680x0 Options. (line 21)
29381 * mcall-gnu: RS/6000 and PowerPC Options.
29383 * mcall-linux: RS/6000 and PowerPC Options.
29385 * mcall-netbsd: RS/6000 and PowerPC Options.
29387 * mcall-prologues: AVR Options. (line 43)
29388 * mcall-solaris: RS/6000 and PowerPC Options.
29390 * mcall-sysv: RS/6000 and PowerPC Options.
29392 * mcall-sysv-eabi: RS/6000 and PowerPC Options.
29394 * mcall-sysv-noeabi: RS/6000 and PowerPC Options.
29396 * mcallee-super-interworking: ARM Options. (line 234)
29397 * mcaller-super-interworking: ARM Options. (line 240)
29398 * mcallgraph-data: MCore Options. (line 31)
29399 * mcc-init: CRIS Options. (line 46)
29400 * mcheck-zero-division: MIPS Options. (line 232)
29401 * mcirrus-fix-invalid-insns: ARM Options. (line 187)
29402 * mcix: DEC Alpha Options. (line 171)
29403 * mcmodel=embmedany: SPARC Options. (line 209)
29404 * mcmodel=kernel: i386 and x86-64 Options.
29406 * mcmodel=large: i386 and x86-64 Options.
29408 * mcmodel=medany: SPARC Options. (line 203)
29409 * mcmodel=medium: i386 and x86-64 Options.
29411 * mcmodel=medlow: SPARC Options. (line 192)
29412 * mcmodel=medmid: SPARC Options. (line 197)
29413 * mcmodel=small: i386 and x86-64 Options.
29415 * mcond-exec: FRV Options. (line 152)
29416 * mcond-move: FRV Options. (line 128)
29417 * mconst-align: CRIS Options. (line 60)
29418 * mconst16: Xtensa Options. (line 10)
29419 * mconstant-gp: IA-64 Options. (line 46)
29420 * mcpu <1>: TMS320C3x/C4x Options.
29422 * mcpu <2>: SPARC Options. (line 96)
29423 * mcpu <3>: RS/6000 and PowerPC Options.
29425 * mcpu <4>: i386 and x86-64 Options.
29427 * mcpu <5>: FRV Options. (line 212)
29428 * mcpu <6>: DEC Alpha Options. (line 223)
29429 * mcpu <7>: CRIS Options. (line 10)
29430 * mcpu <8>: ARM Options. (line 84)
29431 * mcpu: ARC Options. (line 23)
29432 * mcpu32: M680x0 Options. (line 51)
29433 * mcpu=: M32C Options. (line 7)
29434 * mcsync-anomaly: Blackfin Options. (line 23)
29435 * MD: Preprocessor Options.
29437 * mdalign: SH Options. (line 64)
29438 * mdata: ARC Options. (line 30)
29439 * mdata-align: CRIS Options. (line 60)
29440 * mdb: TMS320C3x/C4x Options.
29442 * mdebug <1>: S/390 and zSeries Options.
29444 * mdebug: M32R/D Options. (line 69)
29445 * mdec-asm: PDP-11 Options. (line 78)
29446 * mdisable-callt: V850 Options. (line 80)
29447 * mdisable-fpregs: HPPA Options. (line 33)
29448 * mdisable-indexing: HPPA Options. (line 40)
29449 * mdiv: MCore Options. (line 15)
29450 * mdiv=STRATEGY: SH Options. (line 127)
29451 * mdivide-breaks: MIPS Options. (line 237)
29452 * mdivide-traps: MIPS Options. (line 237)
29453 * mdivsi3_libfunc=NAME: SH Options. (line 168)
29454 * mdouble: FRV Options. (line 38)
29455 * mdouble-float: MIPS Options. (line 151)
29456 * mdp-isr-reload: TMS320C3x/C4x Options.
29458 * mdsp: MIPS Options. (line 156)
29459 * mdwarf2-asm: IA-64 Options. (line 79)
29460 * mdword: FRV Options. (line 32)
29461 * mdynamic-no-pic: RS/6000 and PowerPC Options.
29463 * meabi: RS/6000 and PowerPC Options.
29465 * mearly-stop-bits: IA-64 Options. (line 85)
29466 * melf <1>: MMIX Options. (line 44)
29467 * melf: CRIS Options. (line 95)
29468 * melinux: CRIS Options. (line 99)
29469 * melinux-stacksize: CRIS Options. (line 25)
29470 * memb: RS/6000 and PowerPC Options.
29472 * membedded-data: MIPS Options. (line 203)
29473 * memregs=: M32C Options. (line 21)
29474 * mep: V850 Options. (line 16)
29475 * mepsilon: MMIX Options. (line 15)
29476 * mesa: S/390 and zSeries Options.
29478 * metrax100: CRIS Options. (line 31)
29479 * metrax4: CRIS Options. (line 31)
29480 * mexplicit-relocs <1>: MIPS Options. (line 223)
29481 * mexplicit-relocs: DEC Alpha Options. (line 184)
29482 * MF: Preprocessor Options.
29484 * mfast-fix: TMS320C3x/C4x Options.
29486 * mfast-indirect-calls: HPPA Options. (line 52)
29487 * mfaster-structs: SPARC Options. (line 71)
29488 * mfdpic: FRV Options. (line 56)
29489 * mfix: DEC Alpha Options. (line 171)
29490 * mfix-and-continue: Darwin Options. (line 89)
29491 * mfix-r4000: MIPS Options. (line 287)
29492 * mfix-r4400: MIPS Options. (line 301)
29493 * mfix-sb1: MIPS Options. (line 329)
29494 * mfix-vr4120: MIPS Options. (line 308)
29495 * mfix-vr4130: MIPS Options. (line 322)
29496 * mfixed-cc: FRV Options. (line 28)
29497 * mfixed-range <1>: IA-64 Options. (line 90)
29498 * mfixed-range: HPPA Options. (line 59)
29499 * mfloat-abi: ARM Options. (line 59)
29500 * mfloat-gprs: RS/6000 and PowerPC Options.
29502 * mfloat-ieee: DEC Alpha Options. (line 179)
29503 * mfloat-vax: DEC Alpha Options. (line 179)
29504 * mfloat32: PDP-11 Options. (line 52)
29505 * mfloat64: PDP-11 Options. (line 48)
29506 * mflush-func: MIPS Options. (line 335)
29507 * mflush-func=NAME: M32R/D Options. (line 94)
29508 * mflush-trap=NUMBER: M32R/D Options. (line 87)
29509 * mfmovd: SH Options. (line 78)
29510 * mfp: ARM Options. (line 119)
29511 * mfp-exceptions: MIPS Options. (line 356)
29512 * mfp-reg: DEC Alpha Options. (line 25)
29513 * mfp-rounding-mode: DEC Alpha Options. (line 85)
29514 * mfp-trap-mode: DEC Alpha Options. (line 63)
29515 * mfp32: MIPS Options. (line 134)
29516 * mfp64: MIPS Options. (line 137)
29517 * mfpe: ARM Options. (line 119)
29518 * mfpr-32: FRV Options. (line 13)
29519 * mfpr-64: FRV Options. (line 16)
29520 * mfprnd: RS/6000 and PowerPC Options.
29522 * mfpu <1>: SPARC Options. (line 20)
29523 * mfpu <2>: PDP-11 Options. (line 9)
29524 * mfpu: ARM Options. (line 119)
29525 * mfull-toc: RS/6000 and PowerPC Options.
29527 * mfused-madd <1>: Xtensa Options. (line 19)
29528 * mfused-madd <2>: S/390 and zSeries Options.
29530 * mfused-madd <3>: RS/6000 and PowerPC Options.
29532 * mfused-madd: MIPS Options. (line 272)
29533 * mg: VAX Options. (line 17)
29534 * MG: Preprocessor Options.
29536 * mgas <1>: HPPA Options. (line 75)
29537 * mgas: DEC Alpha Options. (line 159)
29538 * mgettrcost=NUMBER: SH Options. (line 190)
29539 * mgnu: VAX Options. (line 13)
29540 * mgnu-as: IA-64 Options. (line 18)
29541 * mgnu-ld: IA-64 Options. (line 23)
29542 * mgotplt: CRIS Options. (line 86)
29543 * mgp32: MIPS Options. (line 128)
29544 * mgp64: MIPS Options. (line 131)
29545 * mgpr-32: FRV Options. (line 7)
29546 * mgpr-64: FRV Options. (line 10)
29547 * mgprel-ro: FRV Options. (line 79)
29548 * mh: H8/300 Options. (line 14)
29549 * mhard-float <1>: SPARC Options. (line 20)
29550 * mhard-float <2>: S/390 and zSeries Options.
29552 * mhard-float <3>: RS/6000 and PowerPC Options.
29554 * mhard-float <4>: MIPS Options. (line 140)
29555 * mhard-float <5>: FRV Options. (line 19)
29556 * mhard-float: ARM Options. (line 41)
29557 * mhard-quad-float: SPARC Options. (line 41)
29558 * mhardlit: MCore Options. (line 10)
29559 * mhitachi: SH Options. (line 81)
29560 * mid-shared-library: Blackfin Options. (line 39)
29561 * mieee <1>: SH Options. (line 96)
29562 * mieee: DEC Alpha Options. (line 39)
29563 * mieee-conformant: DEC Alpha Options. (line 134)
29564 * mieee-fp: i386 and x86-64 Options.
29566 * mieee-with-inexact: DEC Alpha Options. (line 52)
29567 * milp32: IA-64 Options. (line 114)
29568 * mimpure-text: SPARC Options. (line 81)
29569 * mindexed-addressing: SH Options. (line 180)
29570 * minit-stack: AVR Options. (line 35)
29571 * minline-all-stringops: i386 and x86-64 Options.
29573 * minline-float-divide-max-throughput: IA-64 Options. (line 58)
29574 * minline-float-divide-min-latency: IA-64 Options. (line 54)
29575 * minline-int-divide-max-throughput: IA-64 Options. (line 66)
29576 * minline-int-divide-min-latency: IA-64 Options. (line 62)
29577 * minline-plt: FRV Options. (line 64)
29578 * minline-sqrt-max-throughput: IA-64 Options. (line 74)
29579 * minline-sqrt-min-latency: IA-64 Options. (line 70)
29580 * minmax: M68hc1x Options. (line 31)
29581 * minsert-sched-nops: RS/6000 and PowerPC Options.
29583 * mint16: PDP-11 Options. (line 40)
29584 * mint32 <1>: PDP-11 Options. (line 44)
29585 * mint32: H8/300 Options. (line 28)
29586 * mint8: AVR Options. (line 53)
29587 * minvalid-symbols: SH Options. (line 213)
29588 * mips1: MIPS Options. (line 59)
29589 * mips16: MIPS Options. (line 81)
29590 * mips2: MIPS Options. (line 62)
29591 * mips3: MIPS Options. (line 65)
29592 * mips32: MIPS Options. (line 71)
29593 * mips32r2: MIPS Options. (line 74)
29594 * mips3d: MIPS Options. (line 168)
29595 * mips4: MIPS Options. (line 68)
29596 * mips64: MIPS Options. (line 77)
29597 * misel: RS/6000 and PowerPC Options.
29599 * misize: SH Options. (line 103)
29600 * missue-rate=NUMBER: M32R/D Options. (line 79)
29601 * mjump-in-delay: HPPA Options. (line 28)
29602 * mknuthdiv: MMIX Options. (line 33)
29603 * ml: SH Options. (line 61)
29604 * mlarge-data: DEC Alpha Options. (line 195)
29605 * mlarge-data-threshold=NUMBER: i386 and x86-64 Options.
29607 * mlarge-text: DEC Alpha Options. (line 213)
29608 * mlibfuncs: MMIX Options. (line 10)
29609 * mlibrary-pic: FRV Options. (line 110)
29610 * mlinked-fp: FRV Options. (line 94)
29611 * mlinker-opt: HPPA Options. (line 85)
29612 * mlinux: CRIS Options. (line 104)
29613 * mlittle: RS/6000 and PowerPC Options.
29615 * mlittle-endian <1>: SPARC Options. (line 181)
29616 * mlittle-endian <2>: RS/6000 and PowerPC Options.
29618 * mlittle-endian <3>: MCore Options. (line 39)
29619 * mlittle-endian <4>: IA-64 Options. (line 13)
29620 * mlittle-endian: ARM Options. (line 68)
29621 * mlong-calls <1>: V850 Options. (line 10)
29622 * mlong-calls <2>: MIPS Options. (line 258)
29623 * mlong-calls <3>: M68hc1x Options. (line 35)
29624 * mlong-calls <4>: FRV Options. (line 99)
29625 * mlong-calls <5>: Blackfin Options. (line 57)
29626 * mlong-calls: ARM Options. (line 149)
29627 * mlong-double-128: S/390 and zSeries Options.
29629 * mlong-double-64: S/390 and zSeries Options.
29631 * mlong-load-store: HPPA Options. (line 66)
29632 * mlong32: MIPS Options. (line 177)
29633 * mlong64: MIPS Options. (line 172)
29634 * mlongcall: RS/6000 and PowerPC Options.
29636 * mlongcalls: Xtensa Options. (line 60)
29637 * mloop-unsigned: TMS320C3x/C4x Options.
29639 * mlow-64k: Blackfin Options. (line 32)
29640 * mlp64: IA-64 Options. (line 114)
29641 * MM: Preprocessor Options.
29643 * mmac: CRX Options. (line 9)
29644 * mmad: MIPS Options. (line 267)
29645 * mmangle-cpu: ARC Options. (line 15)
29646 * mmax: DEC Alpha Options. (line 171)
29647 * mmax-stack-frame: CRIS Options. (line 22)
29648 * mmcu: AVR Options. (line 9)
29649 * MMD: Preprocessor Options.
29651 * mmedia: FRV Options. (line 44)
29652 * mmemcpy: MIPS Options. (line 252)
29653 * mmemory-latency: DEC Alpha Options. (line 266)
29654 * mmemparm: TMS320C3x/C4x Options.
29656 * mmfcrf: RS/6000 and PowerPC Options.
29658 * mminimal-toc: RS/6000 and PowerPC Options.
29660 * mmmx: i386 and x86-64 Options.
29662 * mmodel=large: M32R/D Options. (line 33)
29663 * mmodel=medium: M32R/D Options. (line 27)
29664 * mmodel=small: M32R/D Options. (line 18)
29665 * mmpyi: TMS320C3x/C4x Options.
29667 * mmul-bug-workaround: CRIS Options. (line 36)
29668 * mmuladd: FRV Options. (line 50)
29669 * mmult-bug: MN10300 Options. (line 9)
29670 * mmulti-cond-exec: FRV Options. (line 176)
29671 * mmultiple: RS/6000 and PowerPC Options.
29673 * mmvcle: S/390 and zSeries Options.
29675 * mmvme: RS/6000 and PowerPC Options.
29677 * mn: H8/300 Options. (line 20)
29678 * mnested-cond-exec: FRV Options. (line 189)
29679 * mnew-mnemonics: RS/6000 and PowerPC Options.
29681 * mno-3dnow: i386 and x86-64 Options.
29683 * mno-4byte-functions: MCore Options. (line 27)
29684 * mno-abicalls: MIPS Options. (line 100)
29685 * mno-abshi: PDP-11 Options. (line 58)
29686 * mno-ac0: PDP-11 Options. (line 20)
29687 * mno-align-double: i386 and x86-64 Options.
29689 * mno-align-int: M680x0 Options. (line 128)
29690 * mno-align-loops: M32R/D Options. (line 76)
29691 * mno-align-stringops: i386 and x86-64 Options.
29693 * mno-altivec: RS/6000 and PowerPC Options.
29695 * mno-am33: MN10300 Options. (line 20)
29696 * mno-app-regs <1>: V850 Options. (line 61)
29697 * mno-app-regs: SPARC Options. (line 10)
29698 * mno-bacc: MT Options. (line 19)
29699 * mno-backchain: S/390 and zSeries Options.
29701 * mno-base-addresses: MMIX Options. (line 54)
29702 * mno-bit-align: RS/6000 and PowerPC Options.
29704 * mno-bk: TMS320C3x/C4x Options.
29706 * mno-branch-likely: MIPS Options. (line 345)
29707 * mno-branch-predict: MMIX Options. (line 49)
29708 * mno-bwx: DEC Alpha Options. (line 171)
29709 * mno-callgraph-data: MCore Options. (line 31)
29710 * mno-check-zero-division: MIPS Options. (line 232)
29711 * mno-cirrus-fix-invalid-insns: ARM Options. (line 187)
29712 * mno-cix: DEC Alpha Options. (line 171)
29713 * mno-cond-exec: FRV Options. (line 158)
29714 * mno-cond-move: FRV Options. (line 134)
29715 * mno-const-align: CRIS Options. (line 60)
29716 * mno-const16: Xtensa Options. (line 10)
29717 * mno-crt0 <1>: MT Options. (line 25)
29718 * mno-crt0: MN10300 Options. (line 31)
29719 * mno-csync-anomaly: Blackfin Options. (line 28)
29720 * mno-data-align: CRIS Options. (line 60)
29721 * mno-db: TMS320C3x/C4x Options.
29723 * mno-debug: S/390 and zSeries Options.
29725 * mno-div: MCore Options. (line 15)
29726 * mno-double: FRV Options. (line 41)
29727 * mno-dsp: MIPS Options. (line 156)
29728 * mno-dwarf2-asm: IA-64 Options. (line 79)
29729 * mno-dword: FRV Options. (line 35)
29730 * mno-eabi: RS/6000 and PowerPC Options.
29732 * mno-early-stop-bits: IA-64 Options. (line 85)
29733 * mno-eflags: FRV Options. (line 125)
29734 * mno-embedded-data: MIPS Options. (line 203)
29735 * mno-ep: V850 Options. (line 16)
29736 * mno-epsilon: MMIX Options. (line 15)
29737 * mno-explicit-relocs <1>: MIPS Options. (line 223)
29738 * mno-explicit-relocs: DEC Alpha Options. (line 184)
29739 * mno-fancy-math-387: i386 and x86-64 Options.
29741 * mno-fast-fix: TMS320C3x/C4x Options.
29743 * mno-faster-structs: SPARC Options. (line 71)
29744 * mno-fix: DEC Alpha Options. (line 171)
29745 * mno-fix-r4000: MIPS Options. (line 287)
29746 * mno-fix-r4400: MIPS Options. (line 301)
29747 * mno-float32: PDP-11 Options. (line 48)
29748 * mno-float64: PDP-11 Options. (line 52)
29749 * mno-flush-func: M32R/D Options. (line 99)
29750 * mno-flush-trap: M32R/D Options. (line 91)
29751 * mno-fp-in-toc: RS/6000 and PowerPC Options.
29753 * mno-fp-regs: DEC Alpha Options. (line 25)
29754 * mno-fp-ret-in-387: i386 and x86-64 Options.
29756 * mno-fprnd: RS/6000 and PowerPC Options.
29758 * mno-fpu: SPARC Options. (line 25)
29759 * mno-fused-madd <1>: Xtensa Options. (line 19)
29760 * mno-fused-madd <2>: S/390 and zSeries Options.
29762 * mno-fused-madd <3>: RS/6000 and PowerPC Options.
29764 * mno-fused-madd: MIPS Options. (line 272)
29765 * mno-gnu-as: IA-64 Options. (line 18)
29766 * mno-gnu-ld: IA-64 Options. (line 23)
29767 * mno-gotplt: CRIS Options. (line 86)
29768 * mno-hardlit: MCore Options. (line 10)
29769 * mno-id-shared-library: Blackfin Options. (line 45)
29770 * mno-ieee-fp: i386 and x86-64 Options.
29772 * mno-int16: PDP-11 Options. (line 44)
29773 * mno-int32: PDP-11 Options. (line 40)
29774 * mno-interrupts: AVR Options. (line 39)
29775 * mno-isel: RS/6000 and PowerPC Options.
29777 * mno-knuthdiv: MMIX Options. (line 33)
29778 * mno-libfuncs: MMIX Options. (line 10)
29779 * mno-long-calls <1>: V850 Options. (line 10)
29780 * mno-long-calls <2>: MIPS Options. (line 258)
29781 * mno-long-calls <3>: M68hc1x Options. (line 35)
29782 * mno-long-calls <4>: HPPA Options. (line 138)
29783 * mno-long-calls <5>: Blackfin Options. (line 57)
29784 * mno-long-calls: ARM Options. (line 149)
29785 * mno-longcall: RS/6000 and PowerPC Options.
29787 * mno-longcalls: Xtensa Options. (line 60)
29788 * mno-loop-unsigned: TMS320C3x/C4x Options.
29790 * mno-low-64k: Blackfin Options. (line 36)
29791 * mno-mad: MIPS Options. (line 267)
29792 * mno-max: DEC Alpha Options. (line 171)
29793 * mno-media: FRV Options. (line 47)
29794 * mno-memcpy: MIPS Options. (line 252)
29795 * mno-mfcrf: RS/6000 and PowerPC Options.
29797 * mno-mips16: MIPS Options. (line 81)
29798 * mno-mips3d: MIPS Options. (line 168)
29799 * mno-mmx: i386 and x86-64 Options.
29801 * mno-mpyi: TMS320C3x/C4x Options.
29803 * mno-mul-bug-workaround: CRIS Options. (line 36)
29804 * mno-muladd: FRV Options. (line 53)
29805 * mno-mult-bug: MN10300 Options. (line 13)
29806 * mno-multi-cond-exec: FRV Options. (line 183)
29807 * mno-multiple: RS/6000 and PowerPC Options.
29809 * mno-mvcle: S/390 and zSeries Options.
29811 * mno-nested-cond-exec: FRV Options. (line 195)
29812 * mno-optimize-membar: FRV Options. (line 205)
29813 * mno-pack: FRV Options. (line 122)
29814 * mno-packed-stack: S/390 and zSeries Options.
29816 * mno-paired-single: MIPS Options. (line 161)
29817 * mno-parallel-insns: TMS320C3x/C4x Options.
29819 * mno-parallel-mpy: TMS320C3x/C4x Options.
29821 * mno-pic: IA-64 Options. (line 26)
29822 * mno-popcntb: RS/6000 and PowerPC Options.
29824 * mno-power: RS/6000 and PowerPC Options.
29826 * mno-power2: RS/6000 and PowerPC Options.
29828 * mno-powerpc: RS/6000 and PowerPC Options.
29830 * mno-powerpc-gfxopt: RS/6000 and PowerPC Options.
29832 * mno-powerpc-gpopt: RS/6000 and PowerPC Options.
29834 * mno-powerpc64: RS/6000 and PowerPC Options.
29836 * mno-prolog-function: V850 Options. (line 23)
29837 * mno-prologue-epilogue: CRIS Options. (line 76)
29838 * mno-prototype: RS/6000 and PowerPC Options.
29840 * mno-push-args: i386 and x86-64 Options.
29842 * mno-register-names: IA-64 Options. (line 37)
29843 * mno-regnames: RS/6000 and PowerPC Options.
29845 * mno-relax-immediate: MCore Options. (line 19)
29846 * mno-relocatable: RS/6000 and PowerPC Options.
29848 * mno-relocatable-lib: RS/6000 and PowerPC Options.
29850 * mno-rptb: TMS320C3x/C4x Options.
29852 * mno-rpts: TMS320C3x/C4x Options.
29854 * mno-scc: FRV Options. (line 146)
29855 * mno-sched-prolog: ARM Options. (line 32)
29856 * mno-sdata <1>: RS/6000 and PowerPC Options.
29858 * mno-sdata: IA-64 Options. (line 42)
29859 * mno-side-effects: CRIS Options. (line 51)
29860 * mno-single-exit: MMIX Options. (line 66)
29861 * mno-slow-bytes: MCore Options. (line 35)
29862 * mno-small-exec: S/390 and zSeries Options.
29864 * mno-soft-float: DEC Alpha Options. (line 10)
29865 * mno-space-regs: HPPA Options. (line 45)
29866 * mno-spe: RS/6000 and PowerPC Options.
29868 * mno-specld-anomaly: Blackfin Options. (line 19)
29869 * mno-split: PDP-11 Options. (line 71)
29870 * mno-split-addresses: MIPS Options. (line 217)
29871 * mno-sse: i386 and x86-64 Options.
29873 * mno-stack-align: CRIS Options. (line 60)
29874 * mno-stack-bias: SPARC Options. (line 218)
29875 * mno-strict-align <1>: RS/6000 and PowerPC Options.
29877 * mno-strict-align: M680x0 Options. (line 148)
29878 * mno-string: RS/6000 and PowerPC Options.
29880 * mno-sum-in-toc: RS/6000 and PowerPC Options.
29882 * mno-svr3-shlib: i386 and x86-64 Options.
29884 * mno-swdiv: RS/6000 and PowerPC Options.
29886 * mno-sym32: MIPS Options. (line 187)
29887 * mno-tablejump: AVR Options. (line 47)
29888 * mno-target-align: Xtensa Options. (line 47)
29889 * mno-text-section-literals: Xtensa Options. (line 35)
29890 * mno-toc: RS/6000 and PowerPC Options.
29892 * mno-toplevel-symbols: MMIX Options. (line 40)
29893 * mno-tpf-trace: S/390 and zSeries Options.
29895 * mno-unaligned-doubles: SPARC Options. (line 59)
29896 * mno-uninit-const-in-rodata: MIPS Options. (line 211)
29897 * mno-update: RS/6000 and PowerPC Options.
29899 * mno-v8plus: SPARC Options. (line 166)
29900 * mno-vis: SPARC Options. (line 173)
29901 * mno-vliw-branch: FRV Options. (line 170)
29902 * mno-volatile-asm-stop: IA-64 Options. (line 32)
29903 * mno-vrsave: RS/6000 and PowerPC Options.
29905 * mno-wide-bitfields: MCore Options. (line 23)
29906 * mno-xgot: MIPS Options. (line 105)
29907 * mno-xl-compat: RS/6000 and PowerPC Options.
29909 * mno-zero-extend: MMIX Options. (line 27)
29910 * mnobitfield: M680x0 Options. (line 96)
29911 * mnomacsave: SH Options. (line 92)
29912 * mnominmax: M68hc1x Options. (line 31)
29913 * mnop-fun-dllimport: ARM Options. (line 174)
29914 * mold-mnemonics: RS/6000 and PowerPC Options.
29916 * momit-leaf-frame-pointer <1>: i386 and x86-64 Options.
29918 * momit-leaf-frame-pointer: Blackfin Options. (line 7)
29919 * moptimize-membar: FRV Options. (line 201)
29920 * MP: Preprocessor Options.
29922 * mpa-risc-1-0: HPPA Options. (line 19)
29923 * mpa-risc-1-1: HPPA Options. (line 19)
29924 * mpa-risc-2-0: HPPA Options. (line 19)
29925 * mpack: FRV Options. (line 119)
29926 * mpacked-stack: S/390 and zSeries Options.
29928 * mpadstruct: SH Options. (line 106)
29929 * mpaired-single: MIPS Options. (line 161)
29930 * mparallel-insns: TMS320C3x/C4x Options.
29932 * mparallel-mpy: TMS320C3x/C4x Options.
29934 * mparanoid: TMS320C3x/C4x Options.
29936 * mpcrel: M680x0 Options. (line 140)
29937 * mpdebug: CRIS Options. (line 40)
29938 * mpe: RS/6000 and PowerPC Options.
29940 * mpentium: i386 and x86-64 Options.
29942 * mpentiumpro: i386 and x86-64 Options.
29944 * mpic-register: ARM Options. (line 183)
29945 * mpoke-function-name: ARM Options. (line 197)
29946 * mpopcntb: RS/6000 and PowerPC Options.
29948 * mportable-runtime: HPPA Options. (line 71)
29949 * mpower: RS/6000 and PowerPC Options.
29951 * mpower2: RS/6000 and PowerPC Options.
29953 * mpowerpc: RS/6000 and PowerPC Options.
29955 * mpowerpc-gfxopt: RS/6000 and PowerPC Options.
29957 * mpowerpc-gpopt: RS/6000 and PowerPC Options.
29959 * mpowerpc64: RS/6000 and PowerPC Options.
29961 * mprefergot: SH Options. (line 113)
29962 * mpreferred-stack-boundary: i386 and x86-64 Options.
29964 * mprioritize-restricted-insns: RS/6000 and PowerPC Options.
29966 * mprolog-function: V850 Options. (line 23)
29967 * mprologue-epilogue: CRIS Options. (line 76)
29968 * mprototype: RS/6000 and PowerPC Options.
29970 * mpt-fixed: SH Options. (line 194)
29971 * mpush-args <1>: i386 and x86-64 Options.
29973 * mpush-args: CRX Options. (line 13)
29974 * MQ: Preprocessor Options.
29976 * mregister-names: IA-64 Options. (line 37)
29977 * mregnames: RS/6000 and PowerPC Options.
29979 * mregparm <1>: TMS320C3x/C4x Options.
29981 * mregparm: i386 and x86-64 Options.
29983 * mrelax <1>: SH Options. (line 70)
29984 * mrelax <2>: MN10300 Options. (line 34)
29985 * mrelax: H8/300 Options. (line 9)
29986 * mrelax-immediate: MCore Options. (line 19)
29987 * mrelocatable: RS/6000 and PowerPC Options.
29989 * mrelocatable-lib: RS/6000 and PowerPC Options.
29991 * mreturn-pointer-on-d0: MN10300 Options. (line 24)
29992 * mrodata: ARC Options. (line 30)
29993 * mrptb: TMS320C3x/C4x Options.
29995 * mrpts: TMS320C3x/C4x Options.
29997 * mrtd <1>: Function Attributes.
29999 * mrtd <2>: M680x0 Options. (line 105)
30000 * mrtd: i386 and x86-64 Options.
30002 * ms: H8/300 Options. (line 17)
30003 * ms2600: H8/300 Options. (line 24)
30004 * mscc: FRV Options. (line 140)
30005 * msched-costly-dep: RS/6000 and PowerPC Options.
30007 * mschedule: HPPA Options. (line 78)
30008 * msda: V850 Options. (line 40)
30009 * msdata <1>: RS/6000 and PowerPC Options.
30011 * msdata: IA-64 Options. (line 42)
30012 * msdata-data: RS/6000 and PowerPC Options.
30014 * msdata=default: RS/6000 and PowerPC Options.
30016 * msdata=eabi: RS/6000 and PowerPC Options.
30018 * msdata=none <1>: RS/6000 and PowerPC Options.
30020 * msdata=none: M32R/D Options. (line 40)
30021 * msdata=sdata: M32R/D Options. (line 49)
30022 * msdata=sysv: RS/6000 and PowerPC Options.
30024 * msdata=use: M32R/D Options. (line 53)
30025 * msecure-plt: RS/6000 and PowerPC Options.
30027 * mshared-library-id: Blackfin Options. (line 49)
30028 * mshort <1>: M68hc1x Options. (line 40)
30029 * mshort: M680x0 Options. (line 90)
30030 * msim <1>: Xstormy16 Options. (line 9)
30031 * msim <2>: RS/6000 and PowerPC Options.
30033 * msim <3>: MT Options. (line 22)
30034 * msim: M32C Options. (line 13)
30035 * msingle-exit: MMIX Options. (line 66)
30036 * msingle-float: MIPS Options. (line 147)
30037 * msingle-pic-base: ARM Options. (line 177)
30038 * msio: HPPA Options. (line 107)
30039 * msize: AVR Options. (line 32)
30040 * mslow-bytes: MCore Options. (line 35)
30041 * msmall: TMS320C3x/C4x Options.
30043 * msmall-data: DEC Alpha Options. (line 195)
30044 * msmall-exec: S/390 and zSeries Options.
30046 * msmall-memory: TMS320C3x/C4x Options.
30048 * msmall-text: DEC Alpha Options. (line 213)
30049 * msoft-float <1>: SPARC Options. (line 25)
30050 * msoft-float <2>: S/390 and zSeries Options.
30052 * msoft-float <3>: RS/6000 and PowerPC Options.
30054 * msoft-float <4>: PDP-11 Options. (line 13)
30055 * msoft-float <5>: MIPS Options. (line 143)
30056 * msoft-float <6>: M680x0 Options. (line 80)
30057 * msoft-float <7>: i386 and x86-64 Options.
30059 * msoft-float <8>: HPPA Options. (line 91)
30060 * msoft-float <9>: FRV Options. (line 22)
30061 * msoft-float <10>: DEC Alpha Options. (line 10)
30062 * msoft-float: ARM Options. (line 45)
30063 * msoft-quad-float: SPARC Options. (line 45)
30064 * msoft-reg-count: M68hc1x Options. (line 43)
30065 * mspace <1>: V850 Options. (line 30)
30066 * mspace: SH Options. (line 110)
30067 * mspe: RS/6000 and PowerPC Options.
30069 * mspecld-anomaly: Blackfin Options. (line 14)
30070 * msplit: PDP-11 Options. (line 68)
30071 * msplit-addresses: MIPS Options. (line 217)
30072 * msse: i386 and x86-64 Options.
30074 * msseregparm: i386 and x86-64 Options.
30076 * mstack-align: CRIS Options. (line 60)
30077 * mstack-bias: SPARC Options. (line 218)
30078 * mstack-guard: S/390 and zSeries Options.
30080 * mstack-size: S/390 and zSeries Options.
30082 * mstrict-align <1>: RS/6000 and PowerPC Options.
30084 * mstrict-align: M680x0 Options. (line 148)
30085 * mstring: RS/6000 and PowerPC Options.
30087 * mstructure-size-boundary: ARM Options. (line 129)
30088 * msvr3-shlib: i386 and x86-64 Options.
30090 * msvr4-struct-return: RS/6000 and PowerPC Options.
30092 * mswdiv: RS/6000 and PowerPC Options.
30094 * msym32: MIPS Options. (line 187)
30095 * mt: IA-64 Options. (line 106)
30096 * MT: Preprocessor Options.
30098 * mtarget-align: Xtensa Options. (line 47)
30099 * mtda: V850 Options. (line 34)
30100 * mtext: ARC Options. (line 30)
30101 * mtext-section-literals: Xtensa Options. (line 35)
30102 * mthreads: i386 and x86-64 Options.
30104 * mthumb: ARM Options. (line 218)
30105 * mthumb-interwork: ARM Options. (line 25)
30106 * mti: TMS320C3x/C4x Options.
30108 * mtiny-stack: AVR Options. (line 50)
30109 * mtls-direct-seg-refs: i386 and x86-64 Options.
30111 * mtls-size: IA-64 Options. (line 97)
30112 * mtoc: RS/6000 and PowerPC Options.
30114 * mtomcat-stats: FRV Options. (line 209)
30115 * mtoplevel-symbols: MMIX Options. (line 40)
30116 * mtp: ARM Options. (line 246)
30117 * mtpcs-frame: ARM Options. (line 222)
30118 * mtpcs-leaf-frame: ARM Options. (line 228)
30119 * mtpf-trace: S/390 and zSeries Options.
30121 * mtrap-precision: DEC Alpha Options. (line 109)
30122 * mtune <1>: SPARC Options. (line 154)
30123 * mtune <2>: S/390 and zSeries Options.
30125 * mtune <3>: RS/6000 and PowerPC Options.
30127 * mtune <4>: MIPS Options. (line 44)
30128 * mtune <5>: IA-64 Options. (line 101)
30129 * mtune <6>: i386 and x86-64 Options.
30131 * mtune <7>: DEC Alpha Options. (line 262)
30132 * mtune <8>: CRIS Options. (line 16)
30133 * mtune: ARM Options. (line 99)
30134 * multcost=NUMBER: SH Options. (line 124)
30135 * multi_module: Darwin Options. (line 182)
30136 * multilib-library-pic: FRV Options. (line 89)
30137 * multiply_defined: Darwin Options. (line 182)
30138 * multiply_defined_unused: Darwin Options. (line 182)
30139 * munaligned-doubles: SPARC Options. (line 59)
30140 * muninit-const-in-rodata: MIPS Options. (line 211)
30141 * munix: VAX Options. (line 9)
30142 * munix-asm: PDP-11 Options. (line 74)
30143 * mupdate: RS/6000 and PowerPC Options.
30145 * musermode: SH Options. (line 118)
30146 * mv850: V850 Options. (line 49)
30147 * mv850e: V850 Options. (line 69)
30148 * mv850e1: V850 Options. (line 64)
30149 * mv8plus: SPARC Options. (line 166)
30150 * mvis: SPARC Options. (line 173)
30151 * mvliw-branch: FRV Options. (line 164)
30152 * mvms-return-codes: DEC Alpha/VMS Options.
30154 * mvolatile-asm-stop: IA-64 Options. (line 32)
30155 * mvr4130-align: MIPS Options. (line 366)
30156 * mvrsave: RS/6000 and PowerPC Options.
30158 * mvxworks: RS/6000 and PowerPC Options.
30160 * mwarn-dynamicstack: S/390 and zSeries Options.
30162 * mwarn-framesize: S/390 and zSeries Options.
30164 * mwide-bitfields: MCore Options. (line 23)
30165 * mwindiss: RS/6000 and PowerPC Options.
30167 * mwords-little-endian: ARM Options. (line 76)
30168 * mxgot: MIPS Options. (line 105)
30169 * mxl-compat: RS/6000 and PowerPC Options.
30171 * myellowknife: RS/6000 and PowerPC Options.
30173 * mzarch: S/390 and zSeries Options.
30175 * mzda: V850 Options. (line 45)
30176 * mzero-extend: MMIX Options. (line 27)
30177 * no-integrated-cpp: C Dialect Options. (line 190)
30178 * no-red-zone: i386 and x86-64 Options.
30180 * no_dead_strip_inits_and_terms: Darwin Options. (line 182)
30181 * noall_load: Darwin Options. (line 182)
30182 * nocpp: MIPS Options. (line 282)
30183 * nodefaultlibs: Link Options. (line 62)
30184 * nofixprebinding: Darwin Options. (line 182)
30185 * nolibdld: HPPA Options. (line 190)
30186 * nomultidefs: Darwin Options. (line 182)
30187 * noprebind: Darwin Options. (line 182)
30188 * noseglinkedit: Darwin Options. (line 182)
30189 * nostartfiles: Link Options. (line 57)
30190 * nostdinc: Preprocessor Options.
30192 * nostdinc++ <1>: Preprocessor Options.
30194 * nostdinc++: C++ Dialect Options.
30196 * nostdlib: Link Options. (line 71)
30197 * o: Preprocessor Options.
30199 * O: Optimize Options. (line 32)
30200 * o: Overall Options. (line 174)
30201 * O0: Optimize Options. (line 106)
30202 * O1: Optimize Options. (line 32)
30203 * O2: Optimize Options. (line 68)
30204 * O3: Optimize Options. (line 101)
30205 * Os: Optimize Options. (line 109)
30206 * P: Preprocessor Options.
30208 * p: Debugging Options. (line 122)
30209 * pagezero_size: Darwin Options. (line 182)
30210 * param: Optimize Options. (line 1317)
30211 * pass-exit-codes: Overall Options. (line 133)
30212 * pedantic <1>: Warnings and Errors.
30214 * pedantic <2>: Alternate Keywords. (line 29)
30215 * pedantic <3>: C Extensions. (line 6)
30216 * pedantic <4>: Preprocessor Options.
30218 * pedantic <5>: Warning Options. (line 27)
30219 * pedantic: Standards. (line 13)
30220 * pedantic-errors <1>: Warnings and Errors.
30222 * pedantic-errors <2>: Non-bugs. (line 216)
30223 * pedantic-errors <3>: Preprocessor Options.
30225 * pedantic-errors <4>: Warning Options. (line 69)
30226 * pedantic-errors: Standards. (line 13)
30227 * pg: Debugging Options. (line 128)
30228 * pie: Link Options. (line 92)
30229 * pipe: Overall Options. (line 196)
30230 * prebind: Darwin Options. (line 182)
30231 * prebind_all_twolevel_modules: Darwin Options. (line 182)
30232 * preprocessor: Preprocessor Options.
30234 * print-file-name: Debugging Options. (line 699)
30235 * print-libgcc-file-name: Debugging Options. (line 720)
30236 * print-multi-directory: Debugging Options. (line 705)
30237 * print-multi-lib: Debugging Options. (line 710)
30238 * print-objc-runtime-info: Objective-C and Objective-C++ Dialect Options.
30240 * print-prog-name: Debugging Options. (line 717)
30241 * print-search-dirs: Debugging Options. (line 728)
30242 * private_bundle: Darwin Options. (line 182)
30243 * pthread <1>: SPARC Options. (line 238)
30244 * pthread <2>: RS/6000 and PowerPC Options.
30246 * pthread: IA-64 Options. (line 106)
30247 * pthreads: SPARC Options. (line 232)
30248 * Q: Debugging Options. (line 134)
30249 * Qn: System V Options. (line 18)
30250 * Qy: System V Options. (line 14)
30251 * rdynamic: Link Options. (line 98)
30252 * read_only_relocs: Darwin Options. (line 182)
30253 * remap: Preprocessor Options.
30255 * s: Link Options. (line 105)
30256 * S <1>: Link Options. (line 20)
30257 * S: Overall Options. (line 157)
30258 * save-temps: Debugging Options. (line 661)
30259 * sectalign: Darwin Options. (line 182)
30260 * sectcreate: Darwin Options. (line 182)
30261 * sectobjectsymbols: Darwin Options. (line 182)
30262 * sectorder: Darwin Options. (line 182)
30263 * seg1addr: Darwin Options. (line 182)
30264 * seg_addr_table: Darwin Options. (line 182)
30265 * seg_addr_table_filename: Darwin Options. (line 182)
30266 * segaddr: Darwin Options. (line 182)
30267 * seglinkedit: Darwin Options. (line 182)
30268 * segprot: Darwin Options. (line 182)
30269 * segs_read_only_addr: Darwin Options. (line 182)
30270 * segs_read_write_addr: Darwin Options. (line 182)
30271 * shared: Link Options. (line 114)
30272 * shared-libgcc: Link Options. (line 122)
30273 * sim: CRIS Options. (line 108)
30274 * sim2: CRIS Options. (line 114)
30275 * single_module: Darwin Options. (line 182)
30276 * specs: Directory Options. (line 84)
30277 * static <1>: HPPA Options. (line 194)
30278 * static <2>: Darwin Options. (line 182)
30279 * static: Link Options. (line 109)
30280 * static-libgcc: Link Options. (line 122)
30281 * std <1>: Non-bugs. (line 107)
30282 * std <2>: Other Builtins. (line 22)
30283 * std <3>: C Dialect Options. (line 47)
30284 * std: Standards. (line 13)
30285 * std=: Preprocessor Options.
30287 * sub_library: Darwin Options. (line 182)
30288 * sub_umbrella: Darwin Options. (line 182)
30289 * symbolic: Link Options. (line 157)
30290 * sysroot: Directory Options. (line 92)
30291 * target-help <1>: Preprocessor Options.
30293 * target-help: Overall Options. (line 227)
30294 * threads <1>: SPARC Options. (line 226)
30295 * threads: HPPA Options. (line 207)
30296 * time: Debugging Options. (line 675)
30297 * tls: FRV Options. (line 75)
30298 * TLS: FRV Options. (line 72)
30299 * traditional <1>: Incompatibilities. (line 6)
30300 * traditional: C Dialect Options. (line 202)
30301 * traditional-cpp <1>: Preprocessor Options.
30303 * traditional-cpp: C Dialect Options. (line 202)
30304 * trigraphs <1>: Preprocessor Options.
30306 * trigraphs: C Dialect Options. (line 186)
30307 * twolevel_namespace: Darwin Options. (line 182)
30308 * u: Link Options. (line 179)
30309 * U: Preprocessor Options.
30311 * umbrella: Darwin Options. (line 182)
30312 * undef: Preprocessor Options.
30314 * undefined: Darwin Options. (line 182)
30315 * unexported_symbols_list: Darwin Options. (line 182)
30316 * V: Target Options. (line 24)
30317 * v <1>: Preprocessor Options.
30319 * v: Overall Options. (line 185)
30320 * version <1>: Preprocessor Options.
30322 * version: Overall Options. (line 231)
30323 * W: Incompatibilities. (line 64)
30324 * w: Preprocessor Options.
30326 * W: Warning Options. (line 518)
30327 * w: Warning Options. (line 73)
30328 * Wa: Assembler Options. (line 9)
30329 * Wabi: C++ Dialect Options.
30331 * Waggregate-return: Warning Options. (line 783)
30332 * Wall <1>: Standard Libraries. (line 6)
30333 * Wall <2>: Preprocessor Options.
30335 * Wall: Warning Options. (line 502)
30336 * Wassign-intercept: Objective-C and Objective-C++ Dialect Options.
30338 * Wattributes: Warning Options. (line 788)
30339 * Wbad-function-cast: Warning Options. (line 732)
30340 * Wcast-align: Warning Options. (line 746)
30341 * Wcast-qual: Warning Options. (line 741)
30342 * Wchar-subscripts: Warning Options. (line 79)
30343 * Wcomment <1>: Preprocessor Options.
30345 * Wcomment: Warning Options. (line 84)
30346 * Wcomments: Preprocessor Options.
30348 * Wconversion <1>: Protoize Caveats. (line 31)
30349 * Wconversion: Warning Options. (line 763)
30350 * Wctor-dtor-privacy: C++ Dialect Options.
30352 * Wdeclaration-after-statement: Warning Options. (line 698)
30353 * Wdisabled-optimization: Warning Options. (line 1026)
30354 * Wdiv-by-zero: Warning Options. (line 590)
30355 * weak_reference_mismatches: Darwin Options. (line 182)
30356 * Weffc++: C++ Dialect Options.
30358 * Wendif-labels <1>: Preprocessor Options.
30360 * Wendif-labels: Warning Options. (line 708)
30361 * Werror <1>: Preprocessor Options.
30363 * Werror: Warning Options. (line 1041)
30364 * Werror-implicit-function-declaration: Warning Options. (line 198)
30365 * Wextra: Warning Options. (line 518)
30366 * Wfatal-errors: Warning Options. (line 89)
30367 * Wfloat-equal: Warning Options. (line 606)
30368 * Wformat <1>: Function Attributes.
30370 * Wformat: Warning Options. (line 94)
30371 * Wformat-nonliteral <1>: Function Attributes.
30373 * Wformat-nonliteral: Warning Options. (line 151)
30374 * Wformat-security: Warning Options. (line 156)
30375 * Wformat-y2k: Warning Options. (line 129)
30376 * Wformat=2: Warning Options. (line 167)
30377 * whatsloaded: Darwin Options. (line 182)
30378 * whyload: Darwin Options. (line 182)
30379 * Wimplicit: Warning Options. (line 204)
30380 * Wimplicit-function-declaration: Warning Options. (line 198)
30381 * Wimplicit-int: Warning Options. (line 193)
30382 * Wimport: Preprocessor Options.
30384 * Winit-self: Warning Options. (line 179)
30385 * Winline <1>: Inline. (line 35)
30386 * Winline: Warning Options. (line 970)
30387 * Winvalid-pch: Warning Options. (line 1005)
30388 * Wl: Link Options. (line 175)
30389 * Wlarger-than: Warning Options. (line 717)
30390 * Wlong-long: Warning Options. (line 1009)
30391 * Wmain: Warning Options. (line 208)
30392 * Wmissing-braces: Warning Options. (line 214)
30393 * Wmissing-declarations: Warning Options. (line 810)
30394 * Wmissing-field-initializers: Warning Options. (line 816)
30395 * Wmissing-format-attribute: Warning Options. (line 842)
30396 * Wmissing-include-dirs: Warning Options. (line 224)
30397 * Wmissing-noreturn: Warning Options. (line 834)
30398 * Wmissing-prototypes: Warning Options. (line 804)
30399 * Wmultichar: Warning Options. (line 861)
30400 * Wnested-externs: Warning Options. (line 945)
30401 * Wno-attributes: Warning Options. (line 788)
30402 * Wno-deprecated: C++ Dialect Options.
30404 * Wno-deprecated-declarations: Warning Options. (line 910)
30405 * Wno-div-by-zero: Warning Options. (line 590)
30406 * Wno-endif-labels: Warning Options. (line 708)
30407 * Wno-format-extra-args: Warning Options. (line 133)
30408 * Wno-format-zero-length: Warning Options. (line 147)
30409 * Wno-import: Warning Options. (line 76)
30410 * Wno-int-to-pointer-cast: Warning Options. (line 997)
30411 * Wno-invalid-offsetof: Warning Options. (line 983)
30412 * Wno-long-long: Warning Options. (line 1009)
30413 * Wno-multichar: Warning Options. (line 861)
30414 * Wno-non-template-friend: C++ Dialect Options.
30416 * Wno-pmf-conversions <1>: Bound member functions.
30418 * Wno-pmf-conversions: C++ Dialect Options.
30420 * Wno-pointer-sign: Warning Options. (line 1035)
30421 * Wno-pointer-to-int-cast: Warning Options. (line 1001)
30422 * Wno-pragmas: Warning Options. (line 483)
30423 * Wno-protocol: Objective-C and Objective-C++ Dialect Options.
30425 * Wno-variadic-macros: Warning Options. (line 1015)
30426 * Wno-volatile-register-var: Warning Options. (line 1021)
30427 * Wnon-virtual-dtor: C++ Dialect Options.
30429 * Wnonnull: Warning Options. (line 172)
30430 * Wnormalized: Warning Options. (line 867)
30431 * Wold-style-cast: C++ Dialect Options.
30433 * Wold-style-definition: Warning Options. (line 800)
30434 * Woverloaded-virtual: C++ Dialect Options.
30436 * Wp: Preprocessor Options.
30438 * Wpacked: Warning Options. (line 916)
30439 * Wpadded: Warning Options. (line 933)
30440 * Wparentheses: Warning Options. (line 227)
30441 * Wpointer-arith <1>: Pointer Arith. (line 13)
30442 * Wpointer-arith: Warning Options. (line 726)
30443 * Wpointer-sign: Warning Options. (line 1035)
30444 * Wpragmas: Warning Options. (line 483)
30445 * Wredundant-decls: Warning Options. (line 940)
30446 * Wreorder: C++ Dialect Options.
30448 * Wreturn-type: Warning Options. (line 322)
30449 * Wselector: Objective-C and Objective-C++ Dialect Options.
30451 * Wsequence-point: Warning Options. (line 273)
30452 * Wshadow: Warning Options. (line 712)
30453 * Wsign-compare: Warning Options. (line 776)
30454 * Wsign-promo: C++ Dialect Options.
30456 * Wstrict-aliasing: Warning Options. (line 488)
30457 * Wstrict-aliasing=2: Warning Options. (line 495)
30458 * Wstrict-null-sentinel: C++ Dialect Options.
30460 * Wstrict-prototypes: Warning Options. (line 794)
30461 * Wstrict-selector-match: Objective-C and Objective-C++ Dialect Options.
30463 * Wswitch: Warning Options. (line 341)
30464 * Wswitch-enum: Warning Options. (line 352)
30465 * Wswitch-switch: Warning Options. (line 349)
30466 * Wsystem-headers <1>: Preprocessor Options.
30468 * Wsystem-headers: Warning Options. (line 595)
30469 * Wtraditional <1>: Preprocessor Options.
30471 * Wtraditional: Warning Options. (line 621)
30472 * Wtrigraphs <1>: Preprocessor Options.
30474 * Wtrigraphs: Warning Options. (line 358)
30475 * Wundeclared-selector: Objective-C and Objective-C++ Dialect Options.
30477 * Wundef <1>: Preprocessor Options.
30479 * Wundef: Warning Options. (line 705)
30480 * Wuninitialized: Warning Options. (line 403)
30481 * Wunknown-pragmas: Warning Options. (line 476)
30482 * Wunreachable-code: Warning Options. (line 948)
30483 * Wunsafe-loop-optimizations: Warning Options. (line 720)
30484 * Wunused: Warning Options. (line 396)
30485 * Wunused-function: Warning Options. (line 363)
30486 * Wunused-label: Warning Options. (line 368)
30487 * Wunused-macros: Preprocessor Options.
30489 * Wunused-parameter: Warning Options. (line 375)
30490 * Wunused-value: Warning Options. (line 390)
30491 * Wunused-variable: Warning Options. (line 382)
30492 * Wvariadic-macros: Warning Options. (line 1015)
30493 * Wvolatile-register-var: Warning Options. (line 1021)
30494 * Wwrite-strings: Warning Options. (line 752)
30495 * x <1>: Preprocessor Options.
30497 * x: Overall Options. (line 109)
30498 * Xassembler: Assembler Options. (line 13)
30499 * Xlinker: Link Options. (line 163)
30500 * Ym: System V Options. (line 26)
30501 * YP: System V Options. (line 22)
30504 File: gcc.info, Node: Keyword Index, Prev: Option Index, Up: Top
30512 * ! in constraint: Multi-Alternative. (line 33)
30513 * # in constraint: Modifiers. (line 57)
30514 * #pragma: Pragmas. (line 6)
30515 * #pragma implementation: C++ Interface. (line 39)
30516 * #pragma implementation, implied: C++ Interface. (line 46)
30517 * #pragma interface: C++ Interface. (line 20)
30518 * #pragma, reason for not using: Function Attributes.
30520 * $: Dollar Signs. (line 6)
30521 * % in constraint: Modifiers. (line 45)
30522 * %include: Spec Files. (line 27)
30523 * %include_noerr: Spec Files. (line 31)
30524 * %rename: Spec Files. (line 35)
30525 * & in constraint: Modifiers. (line 25)
30526 * ': Incompatibilities. (line 116)
30527 * * in constraint: Modifiers. (line 62)
30528 * + in constraint: Modifiers. (line 12)
30529 * -lgcc, use with -nodefaultlibs: Link Options. (line 79)
30530 * -lgcc, use with -nostdlib: Link Options. (line 79)
30531 * -nodefaultlibs and unresolved references: Link Options. (line 79)
30532 * -nostdlib and unresolved references: Link Options. (line 79)
30533 * .sdata/.sdata2 references (PowerPC): RS/6000 and PowerPC Options.
30535 * //: C++ Comments. (line 6)
30536 * 0 in constraint: Simple Constraints. (line 115)
30537 * < in constraint: Simple Constraints. (line 46)
30538 * = in constraint: Modifiers. (line 8)
30539 * > in constraint: Simple Constraints. (line 50)
30540 * ? in constraint: Multi-Alternative. (line 27)
30541 * ?: extensions: Conditionals. (line 6)
30542 * ?: side effect: Conditionals. (line 20)
30543 * _ in variables in macros: Typeof. (line 42)
30544 * __builtin___fprintf_chk: Object Size Checking.
30546 * __builtin___memcpy_chk: Object Size Checking.
30548 * __builtin___memmove_chk: Object Size Checking.
30550 * __builtin___mempcpy_chk: Object Size Checking.
30552 * __builtin___memset_chk: Object Size Checking.
30554 * __builtin___printf_chk: Object Size Checking.
30556 * __builtin___snprintf_chk: Object Size Checking.
30558 * __builtin___sprintf_chk: Object Size Checking.
30560 * __builtin___stpcpy_chk: Object Size Checking.
30562 * __builtin___strcat_chk: Object Size Checking.
30564 * __builtin___strcpy_chk: Object Size Checking.
30566 * __builtin___strncat_chk: Object Size Checking.
30568 * __builtin___strncpy_chk: Object Size Checking.
30570 * __builtin___vfprintf_chk: Object Size Checking.
30572 * __builtin___vprintf_chk: Object Size Checking.
30574 * __builtin___vsnprintf_chk: Object Size Checking.
30576 * __builtin___vsprintf_chk: Object Size Checking.
30578 * __builtin_apply: Constructing Calls. (line 31)
30579 * __builtin_apply_args: Constructing Calls. (line 20)
30580 * __builtin_choose_expr: Other Builtins. (line 150)
30581 * __builtin_clz: Other Builtins. (line 362)
30582 * __builtin_clzl: Other Builtins. (line 380)
30583 * __builtin_clzll: Other Builtins. (line 400)
30584 * __builtin_constant_p: Other Builtins. (line 190)
30585 * __builtin_ctz: Other Builtins. (line 366)
30586 * __builtin_ctzl: Other Builtins. (line 384)
30587 * __builtin_ctzll: Other Builtins. (line 404)
30588 * __builtin_expect: Other Builtins. (line 236)
30589 * __builtin_ffs: Other Builtins. (line 358)
30590 * __builtin_ffsl: Other Builtins. (line 376)
30591 * __builtin_ffsll: Other Builtins. (line 396)
30592 * __builtin_frame_address: Return Address. (line 34)
30593 * __builtin_huge_val: Other Builtins. (line 300)
30594 * __builtin_huge_valf: Other Builtins. (line 305)
30595 * __builtin_huge_vall: Other Builtins. (line 308)
30596 * __builtin_inf: Other Builtins. (line 312)
30597 * __builtin_inff: Other Builtins. (line 316)
30598 * __builtin_infl: Other Builtins. (line 321)
30599 * __builtin_isgreater: Other Builtins. (line 6)
30600 * __builtin_isgreaterequal: Other Builtins. (line 6)
30601 * __builtin_isless: Other Builtins. (line 6)
30602 * __builtin_islessequal: Other Builtins. (line 6)
30603 * __builtin_islessgreater: Other Builtins. (line 6)
30604 * __builtin_isunordered: Other Builtins. (line 6)
30605 * __builtin_nan: Other Builtins. (line 325)
30606 * __builtin_nanf: Other Builtins. (line 340)
30607 * __builtin_nanl: Other Builtins. (line 343)
30608 * __builtin_nans: Other Builtins. (line 347)
30609 * __builtin_nansf: Other Builtins. (line 351)
30610 * __builtin_nansl: Other Builtins. (line 354)
30611 * __builtin_object_size: Object Size Checking.
30613 * __builtin_offsetof: Offsetof. (line 6)
30614 * __builtin_parity: Other Builtins. (line 373)
30615 * __builtin_parityl: Other Builtins. (line 392)
30616 * __builtin_parityll: Other Builtins. (line 412)
30617 * __builtin_popcount: Other Builtins. (line 370)
30618 * __builtin_popcountl: Other Builtins. (line 388)
30619 * __builtin_popcountll: Other Builtins. (line 408)
30620 * __builtin_powi: Other Builtins. (line 6)
30621 * __builtin_powif: Other Builtins. (line 6)
30622 * __builtin_powil: Other Builtins. (line 6)
30623 * __builtin_prefetch: Other Builtins. (line 261)
30624 * __builtin_return: Constructing Calls. (line 48)
30625 * __builtin_return_address: Return Address. (line 11)
30626 * __builtin_types_compatible_p: Other Builtins. (line 104)
30627 * __complex__ keyword: Complex. (line 6)
30628 * __declspec(dllexport): Function Attributes.
30630 * __declspec(dllimport): Function Attributes.
30632 * __extension__: Alternate Keywords. (line 29)
30633 * __func__ identifier: Function Names. (line 6)
30634 * __FUNCTION__ identifier: Function Names. (line 6)
30635 * __imag__ keyword: Complex. (line 27)
30636 * __PRETTY_FUNCTION__ identifier: Function Names. (line 6)
30637 * __real__ keyword: Complex. (line 27)
30638 * __STDC_HOSTED__: Standards. (line 6)
30639 * __sync_add_and_fetch: Atomic Builtins. (line 57)
30640 * __sync_and_and_fetch: Atomic Builtins. (line 57)
30641 * __sync_bool_compare_and_swap: Atomic Builtins. (line 65)
30642 * __sync_fetch_and_add: Atomic Builtins. (line 45)
30643 * __sync_fetch_and_and: Atomic Builtins. (line 45)
30644 * __sync_fetch_and_nand: Atomic Builtins. (line 45)
30645 * __sync_fetch_and_or: Atomic Builtins. (line 45)
30646 * __sync_fetch_and_sub: Atomic Builtins. (line 45)
30647 * __sync_fetch_and_xor: Atomic Builtins. (line 45)
30648 * __sync_lock_release: Atomic Builtins. (line 95)
30649 * __sync_lock_test_and_set: Atomic Builtins. (line 77)
30650 * __sync_nand_and_fetch: Atomic Builtins. (line 57)
30651 * __sync_or_and_fetch: Atomic Builtins. (line 57)
30652 * __sync_sub_and_fetch: Atomic Builtins. (line 57)
30653 * __sync_synchronize: Atomic Builtins. (line 74)
30654 * __sync_val_compare_and_swap: Atomic Builtins. (line 65)
30655 * __sync_xor_and_fetch: Atomic Builtins. (line 57)
30656 * __thread: Thread-Local. (line 6)
30657 * _Complex keyword: Complex. (line 6)
30658 * _exit: Other Builtins. (line 6)
30659 * _Exit: Other Builtins. (line 6)
30660 * ABI: Compatibility. (line 6)
30661 * abort: Other Builtins. (line 6)
30662 * abs: Other Builtins. (line 6)
30663 * accessing volatiles: Volatiles. (line 6)
30664 * acos: Other Builtins. (line 6)
30665 * acosf: Other Builtins. (line 6)
30666 * acosh: Other Builtins. (line 6)
30667 * acoshf: Other Builtins. (line 6)
30668 * acoshl: Other Builtins. (line 6)
30669 * acosl: Other Builtins. (line 6)
30670 * Ada: G++ and GCC. (line 6)
30671 * address constraints: Simple Constraints. (line 142)
30672 * address of a label: Labels as Values. (line 6)
30673 * address_operand: Simple Constraints. (line 146)
30674 * alias attribute: Function Attributes.
30676 * aliasing of parameters: Code Gen Options. (line 333)
30677 * aligned attribute <1>: Type Attributes. (line 30)
30678 * aligned attribute: Variable Attributes.
30680 * alignment: Alignment. (line 6)
30681 * alloca: Other Builtins. (line 6)
30682 * alloca vs variable-length arrays: Variable Length. (line 27)
30683 * Allow nesting in an interrupt handler on the Blackfin processor.: Function Attributes.
30685 * alternate keywords: Alternate Keywords. (line 6)
30686 * always_inline function attribute: Function Attributes.
30688 * AMD x86-64 Options: i386 and x86-64 Options.
30690 * AMD1: Standards. (line 6)
30691 * ANSI C: Standards. (line 6)
30692 * ANSI C standard: Standards. (line 6)
30693 * ANSI C89: Standards. (line 6)
30694 * ANSI support: C Dialect Options. (line 10)
30695 * ANSI X3.159-1989: Standards. (line 6)
30696 * apostrophes: Incompatibilities. (line 116)
30697 * application binary interface: Compatibility. (line 6)
30698 * ARC Options: ARC Options. (line 6)
30699 * ARM [Annotated C++ Reference Manual]: Backwards Compatibility.
30701 * ARM options: ARM Options. (line 6)
30702 * arrays of length zero: Zero Length. (line 6)
30703 * arrays of variable length: Variable Length. (line 6)
30704 * arrays, non-lvalue: Subscripting. (line 6)
30705 * asin: Other Builtins. (line 6)
30706 * asinf: Other Builtins. (line 6)
30707 * asinh: Other Builtins. (line 6)
30708 * asinhf: Other Builtins. (line 6)
30709 * asinhl: Other Builtins. (line 6)
30710 * asinl: Other Builtins. (line 6)
30711 * asm constraints: Constraints. (line 6)
30712 * asm expressions: Extended Asm. (line 6)
30713 * assembler instructions: Extended Asm. (line 6)
30714 * assembler names for identifiers: Asm Labels. (line 6)
30715 * assembly code, invalid: Bug Criteria. (line 12)
30716 * atan: Other Builtins. (line 6)
30717 * atan2: Other Builtins. (line 6)
30718 * atan2f: Other Builtins. (line 6)
30719 * atan2l: Other Builtins. (line 6)
30720 * atanf: Other Builtins. (line 6)
30721 * atanh: Other Builtins. (line 6)
30722 * atanhf: Other Builtins. (line 6)
30723 * atanhl: Other Builtins. (line 6)
30724 * atanl: Other Builtins. (line 6)
30725 * attribute of types: Type Attributes. (line 6)
30726 * attribute of variables: Variable Attributes.
30728 * attribute syntax: Attribute Syntax. (line 6)
30729 * autoincrement/decrement addressing: Simple Constraints. (line 28)
30730 * automatic inline for C++ member fns: Inline. (line 46)
30731 * AVR Options: AVR Options. (line 6)
30732 * Backwards Compatibility: Backwards Compatibility.
30734 * base class members: Name lookup. (line 6)
30735 * bcmp: Other Builtins. (line 6)
30736 * below100 attribute: Variable Attributes.
30738 * binary compatibility: Compatibility. (line 6)
30739 * Blackfin Options: Blackfin Options. (line 6)
30740 * bound pointer to member function: Bound member functions.
30742 * bounds checking: Optimize Options. (line 336)
30743 * bug criteria: Bug Criteria. (line 6)
30744 * bugs: Bugs. (line 6)
30745 * bugs, known: Trouble. (line 6)
30746 * built-in functions <1>: Other Builtins. (line 6)
30747 * built-in functions: C Dialect Options. (line 128)
30748 * bzero: Other Builtins. (line 6)
30749 * C compilation options: Invoking GCC. (line 17)
30750 * C intermediate output, nonexistent: G++ and GCC. (line 35)
30751 * C language extensions: C Extensions. (line 6)
30752 * C language, traditional: C Dialect Options. (line 200)
30753 * C standard: Standards. (line 6)
30754 * C standards: Standards. (line 6)
30755 * c++: Invoking G++. (line 13)
30756 * C++: G++ and GCC. (line 30)
30757 * C++ comments: C++ Comments. (line 6)
30758 * C++ compilation options: Invoking GCC. (line 23)
30759 * C++ interface and implementation headers: C++ Interface. (line 6)
30760 * C++ language extensions: C++ Extensions. (line 6)
30761 * C++ member fns, automatically inline: Inline. (line 46)
30762 * C++ misunderstandings: C++ Misunderstandings.
30764 * C++ options, command line: C++ Dialect Options.
30766 * C++ pragmas, effect on inlining: C++ Interface. (line 66)
30767 * C++ source file suffixes: Invoking G++. (line 6)
30768 * C++ static data, declaring and defining: Static Definitions.
30770 * C89: Standards. (line 6)
30771 * C90: Standards. (line 6)
30772 * C94: Standards. (line 6)
30773 * C95: Standards. (line 6)
30774 * C99: Standards. (line 6)
30775 * C9X: Standards. (line 6)
30776 * C_INCLUDE_PATH: Environment Variables.
30778 * cabs: Other Builtins. (line 6)
30779 * cabsf: Other Builtins. (line 6)
30780 * cabsl: Other Builtins. (line 6)
30781 * cacos: Other Builtins. (line 6)
30782 * cacosf: Other Builtins. (line 6)
30783 * cacosh: Other Builtins. (line 6)
30784 * cacoshf: Other Builtins. (line 6)
30785 * cacoshl: Other Builtins. (line 6)
30786 * cacosl: Other Builtins. (line 6)
30787 * calling functions through the function vector on the H8/300 processors: Function Attributes.
30789 * calloc: Other Builtins. (line 6)
30790 * carg: Other Builtins. (line 6)
30791 * cargf: Other Builtins. (line 6)
30792 * cargl: Other Builtins. (line 6)
30793 * case labels in initializers: Designated Inits. (line 6)
30794 * case ranges: Case Ranges. (line 6)
30795 * casin: Other Builtins. (line 6)
30796 * casinf: Other Builtins. (line 6)
30797 * casinh: Other Builtins. (line 6)
30798 * casinhf: Other Builtins. (line 6)
30799 * casinhl: Other Builtins. (line 6)
30800 * casinl: Other Builtins. (line 6)
30801 * cast to a union: Cast to Union. (line 6)
30802 * catan: Other Builtins. (line 6)
30803 * catanf: Other Builtins. (line 6)
30804 * catanh: Other Builtins. (line 6)
30805 * catanhf: Other Builtins. (line 6)
30806 * catanhl: Other Builtins. (line 6)
30807 * catanl: Other Builtins. (line 6)
30808 * cbrt: Other Builtins. (line 6)
30809 * cbrtf: Other Builtins. (line 6)
30810 * cbrtl: Other Builtins. (line 6)
30811 * ccos: Other Builtins. (line 6)
30812 * ccosf: Other Builtins. (line 6)
30813 * ccosh: Other Builtins. (line 6)
30814 * ccoshf: Other Builtins. (line 6)
30815 * ccoshl: Other Builtins. (line 6)
30816 * ccosl: Other Builtins. (line 6)
30817 * ceil: Other Builtins. (line 6)
30818 * ceilf: Other Builtins. (line 6)
30819 * ceill: Other Builtins. (line 6)
30820 * cexp: Other Builtins. (line 6)
30821 * cexpf: Other Builtins. (line 6)
30822 * cexpl: Other Builtins. (line 6)
30823 * character set, execution: Preprocessor Options.
30825 * character set, input: Preprocessor Options.
30827 * character set, input normalization: Warning Options. (line 867)
30828 * character set, wide execution: Preprocessor Options.
30830 * cimag: Other Builtins. (line 6)
30831 * cimagf: Other Builtins. (line 6)
30832 * cimagl: Other Builtins. (line 6)
30833 * cleanup attribute: Variable Attributes.
30835 * clog: Other Builtins. (line 6)
30836 * clogf: Other Builtins. (line 6)
30837 * clogl: Other Builtins. (line 6)
30838 * COBOL: G++ and GCC. (line 23)
30839 * code generation conventions: Code Gen Options. (line 6)
30840 * code, mixed with declarations: Mixed Declarations. (line 6)
30841 * command options: Invoking GCC. (line 6)
30842 * comments, C++ style: C++ Comments. (line 6)
30843 * common attribute: Variable Attributes.
30845 * comparison of signed and unsigned values, warning: Warning Options.
30847 * compiler bugs, reporting: Bug Reporting. (line 6)
30848 * compiler compared to C++ preprocessor: G++ and GCC. (line 35)
30849 * compiler options, C++: C++ Dialect Options.
30851 * compiler options, Objective-C and Objective-C++: Objective-C and Objective-C++ Dialect Options.
30853 * compiler version, specifying: Target Options. (line 6)
30854 * COMPILER_PATH: Environment Variables.
30856 * complex conjugation: Complex. (line 34)
30857 * complex numbers: Complex. (line 6)
30858 * compound literals: Compound Literals. (line 6)
30859 * computed gotos: Labels as Values. (line 6)
30860 * conditional expressions, extensions: Conditionals. (line 6)
30861 * conflicting types: Disappointments. (line 21)
30862 * conj: Other Builtins. (line 6)
30863 * conjf: Other Builtins. (line 6)
30864 * conjl: Other Builtins. (line 6)
30865 * const applied to function: Function Attributes.
30867 * const function attribute: Function Attributes.
30869 * constants in constraints: Simple Constraints. (line 58)
30870 * constraint modifier characters: Modifiers. (line 6)
30871 * constraint, matching: Simple Constraints. (line 127)
30872 * constraints, asm: Constraints. (line 6)
30873 * constraints, machine specific: Machine Constraints.
30875 * constructing calls: Constructing Calls. (line 6)
30876 * constructor expressions: Compound Literals. (line 6)
30877 * constructor function attribute: Function Attributes.
30879 * contributors: Contributors. (line 6)
30880 * copysign: Other Builtins. (line 6)
30881 * copysignf: Other Builtins. (line 6)
30882 * copysignl: Other Builtins. (line 6)
30883 * core dump: Bug Criteria. (line 9)
30884 * cos: Other Builtins. (line 6)
30885 * cosf: Other Builtins. (line 6)
30886 * cosh: Other Builtins. (line 6)
30887 * coshf: Other Builtins. (line 6)
30888 * coshl: Other Builtins. (line 6)
30889 * cosl: Other Builtins. (line 6)
30890 * CPATH: Environment Variables.
30892 * CPLUS_INCLUDE_PATH: Environment Variables.
30894 * cpow: Other Builtins. (line 6)
30895 * cpowf: Other Builtins. (line 6)
30896 * cpowl: Other Builtins. (line 6)
30897 * cproj: Other Builtins. (line 6)
30898 * cprojf: Other Builtins. (line 6)
30899 * cprojl: Other Builtins. (line 6)
30900 * creal: Other Builtins. (line 6)
30901 * crealf: Other Builtins. (line 6)
30902 * creall: Other Builtins. (line 6)
30903 * CRIS Options: CRIS Options. (line 6)
30904 * cross compiling: Target Options. (line 6)
30905 * CRX Options: CRX Options. (line 6)
30906 * csin: Other Builtins. (line 6)
30907 * csinf: Other Builtins. (line 6)
30908 * csinh: Other Builtins. (line 6)
30909 * csinhf: Other Builtins. (line 6)
30910 * csinhl: Other Builtins. (line 6)
30911 * csinl: Other Builtins. (line 6)
30912 * csqrt: Other Builtins. (line 6)
30913 * csqrtf: Other Builtins. (line 6)
30914 * csqrtl: Other Builtins. (line 6)
30915 * ctan: Other Builtins. (line 6)
30916 * ctanf: Other Builtins. (line 6)
30917 * ctanh: Other Builtins. (line 6)
30918 * ctanhf: Other Builtins. (line 6)
30919 * ctanhl: Other Builtins. (line 6)
30920 * ctanl: Other Builtins. (line 6)
30921 * Darwin options: Darwin Options. (line 6)
30922 * dcgettext: Other Builtins. (line 6)
30923 * deallocating variable length arrays: Variable Length. (line 23)
30924 * debugging information options: Debugging Options. (line 6)
30925 * declaration scope: Incompatibilities. (line 80)
30926 * declarations inside expressions: Statement Exprs. (line 6)
30927 * declarations, mixed with code: Mixed Declarations. (line 6)
30928 * declaring attributes of functions: Function Attributes.
30930 * declaring static data in C++: Static Definitions. (line 6)
30931 * defining static data in C++: Static Definitions. (line 6)
30932 * dependencies for make as output: Environment Variables.
30934 * dependencies, make: Preprocessor Options.
30936 * DEPENDENCIES_OUTPUT: Environment Variables.
30938 * dependent name lookup: Name lookup. (line 6)
30939 * deprecated attribute: Variable Attributes.
30941 * deprecated attribute.: Function Attributes.
30943 * designated initializers: Designated Inits. (line 6)
30944 * designator lists: Designated Inits. (line 94)
30945 * designators: Designated Inits. (line 61)
30946 * destructor function attribute: Function Attributes.
30948 * dgettext: Other Builtins. (line 6)
30949 * diagnostic messages: Language Independent Options.
30951 * dialect options: C Dialect Options. (line 6)
30952 * digits in constraint: Simple Constraints. (line 115)
30953 * directory options: Directory Options. (line 6)
30954 * dollar signs in identifier names: Dollar Signs. (line 6)
30955 * double-word arithmetic: Long Long. (line 6)
30956 * downward funargs: Nested Functions. (line 6)
30957 * drem: Other Builtins. (line 6)
30958 * dremf: Other Builtins. (line 6)
30959 * dreml: Other Builtins. (line 6)
30960 * E in constraint: Simple Constraints. (line 77)
30961 * earlyclobber operand: Modifiers. (line 25)
30962 * eight bit data on the H8/300, H8/300H, and H8S: Function Attributes.
30964 * empty structures: Empty Structures. (line 6)
30965 * environment variables: Environment Variables.
30967 * erf: Other Builtins. (line 6)
30968 * erfc: Other Builtins. (line 6)
30969 * erfcf: Other Builtins. (line 6)
30970 * erfcl: Other Builtins. (line 6)
30971 * erff: Other Builtins. (line 6)
30972 * erfl: Other Builtins. (line 6)
30973 * error messages: Warnings and Errors.
30975 * escaped newlines: Escaped Newlines. (line 6)
30976 * exception handler functions on the Blackfin processor: Function Attributes.
30978 * exclamation point: Multi-Alternative. (line 33)
30979 * exit: Other Builtins. (line 6)
30980 * exp: Other Builtins. (line 6)
30981 * exp10: Other Builtins. (line 6)
30982 * exp10f: Other Builtins. (line 6)
30983 * exp10l: Other Builtins. (line 6)
30984 * exp2: Other Builtins. (line 6)
30985 * exp2f: Other Builtins. (line 6)
30986 * exp2l: Other Builtins. (line 6)
30987 * expf: Other Builtins. (line 6)
30988 * expl: Other Builtins. (line 6)
30989 * explicit register variables: Explicit Reg Vars. (line 6)
30990 * expm1: Other Builtins. (line 6)
30991 * expm1f: Other Builtins. (line 6)
30992 * expm1l: Other Builtins. (line 6)
30993 * expressions containing statements: Statement Exprs. (line 6)
30994 * expressions, constructor: Compound Literals. (line 6)
30995 * extended asm: Extended Asm. (line 6)
30996 * extensible constraints: Simple Constraints. (line 151)
30997 * extensions, ?:: Conditionals. (line 6)
30998 * extensions, C language: C Extensions. (line 6)
30999 * extensions, C++ language: C++ Extensions. (line 6)
31000 * external declaration scope: Incompatibilities. (line 80)
31001 * externally_visible attribute.: Function Attributes.
31003 * F in constraint: Simple Constraints. (line 82)
31004 * fabs: Other Builtins. (line 6)
31005 * fabsf: Other Builtins. (line 6)
31006 * fabsl: Other Builtins. (line 6)
31007 * fatal signal: Bug Criteria. (line 9)
31008 * fdim: Other Builtins. (line 6)
31009 * fdimf: Other Builtins. (line 6)
31010 * fdiml: Other Builtins. (line 6)
31011 * FDL, GNU Free Documentation License: GNU Free Documentation License.
31013 * ffs: Other Builtins. (line 6)
31014 * file name suffix: Overall Options. (line 14)
31015 * file names: Link Options. (line 10)
31016 * flatten function attribute: Function Attributes.
31018 * flexible array members: Zero Length. (line 6)
31019 * float as function value type: Incompatibilities. (line 141)
31020 * floating point precision <1>: Disappointments. (line 68)
31021 * floating point precision: Optimize Options. (line 1043)
31022 * floor: Other Builtins. (line 6)
31023 * floorf: Other Builtins. (line 6)
31024 * floorl: Other Builtins. (line 6)
31025 * fma: Other Builtins. (line 6)
31026 * fmaf: Other Builtins. (line 6)
31027 * fmal: Other Builtins. (line 6)
31028 * fmax: Other Builtins. (line 6)
31029 * fmaxf: Other Builtins. (line 6)
31030 * fmaxl: Other Builtins. (line 6)
31031 * fmin: Other Builtins. (line 6)
31032 * fminf: Other Builtins. (line 6)
31033 * fminl: Other Builtins. (line 6)
31034 * fmod: Other Builtins. (line 6)
31035 * fmodf: Other Builtins. (line 6)
31036 * fmodl: Other Builtins. (line 6)
31037 * format function attribute: Function Attributes.
31039 * format_arg function attribute: Function Attributes.
31041 * Fortran: G++ and GCC. (line 6)
31042 * forwarding calls: Constructing Calls. (line 6)
31043 * fprintf: Other Builtins. (line 6)
31044 * fprintf_unlocked: Other Builtins. (line 6)
31045 * fputs: Other Builtins. (line 6)
31046 * fputs_unlocked: Other Builtins. (line 6)
31047 * freestanding environment: Standards. (line 6)
31048 * freestanding implementation: Standards. (line 6)
31049 * frexp: Other Builtins. (line 6)
31050 * frexpf: Other Builtins. (line 6)
31051 * frexpl: Other Builtins. (line 6)
31052 * FRV Options: FRV Options. (line 6)
31053 * fscanf: Other Builtins. (line 6)
31054 * fscanf, and constant strings: Incompatibilities. (line 17)
31055 * function addressability on the M32R/D: Function Attributes.
31057 * function attributes: Function Attributes.
31059 * function pointers, arithmetic: Pointer Arith. (line 6)
31060 * function prototype declarations: Function Prototypes.
31062 * function without a prologue/epilogue code: Function Attributes.
31064 * function, size of pointer to: Pointer Arith. (line 6)
31065 * functions called via pointer on the RS/6000 and PowerPC: Function Attributes.
31067 * functions in arbitrary sections: Function Attributes.
31069 * functions that are passed arguments in registers on the 386: Function Attributes.
31071 * functions that behave like malloc: Function Attributes.
31073 * functions that do not pop the argument stack on the 386: Function Attributes.
31075 * functions that do pop the argument stack on the 386: Function Attributes.
31077 * functions that have no side effects: Function Attributes.
31079 * functions that never return: Function Attributes.
31081 * functions that pop the argument stack on the 386: Function Attributes.
31083 * functions that return more than once: Function Attributes.
31085 * functions which do not handle memory bank switching on 68HC11/68HC12: Function Attributes.
31087 * functions which handle memory bank switching: Function Attributes.
31089 * functions with non-null pointer arguments: Function Attributes.
31091 * functions with printf, scanf, strftime or strfmon style arguments: Function Attributes.
31093 * g in constraint: Simple Constraints. (line 108)
31094 * G in constraint: Simple Constraints. (line 86)
31095 * g++: Invoking G++. (line 13)
31096 * G++: G++ and GCC. (line 30)
31097 * gamma: Other Builtins. (line 6)
31098 * gammaf: Other Builtins. (line 6)
31099 * gammal: Other Builtins. (line 6)
31100 * GCC: G++ and GCC. (line 6)
31101 * GCC command options: Invoking GCC. (line 6)
31102 * GCC_EXEC_PREFIX: Environment Variables.
31104 * gcc_struct: Type Attributes. (line 288)
31105 * gcc_struct attribute: Variable Attributes.
31107 * gcov: Debugging Options. (line 160)
31108 * gettext: Other Builtins. (line 6)
31109 * global offset table: Code Gen Options. (line 170)
31110 * global register after longjmp: Global Reg Vars. (line 66)
31111 * global register variables: Global Reg Vars. (line 6)
31112 * GNAT: G++ and GCC. (line 30)
31113 * GNU C Compiler: G++ and GCC. (line 6)
31114 * GNU Compiler Collection: G++ and GCC. (line 6)
31115 * goto with computed label: Labels as Values. (line 6)
31116 * gp-relative references (MIPS): MIPS Options. (line 194)
31117 * gprof: Debugging Options. (line 127)
31118 * grouping options: Invoking GCC. (line 26)
31119 * H in constraint: Simple Constraints. (line 86)
31120 * hardware models and configurations, specifying: Submodel Options.
31122 * hex floats: Hex Floats. (line 6)
31123 * hosted environment <1>: C Dialect Options. (line 162)
31124 * hosted environment: Standards. (line 6)
31125 * hosted implementation: Standards. (line 6)
31126 * HPPA Options: HPPA Options. (line 6)
31127 * hypot: Other Builtins. (line 6)
31128 * hypotf: Other Builtins. (line 6)
31129 * hypotl: Other Builtins. (line 6)
31130 * I in constraint: Simple Constraints. (line 69)
31131 * i in constraint: Simple Constraints. (line 58)
31132 * i386 Options: i386 and x86-64 Options.
31134 * IA-64 Options: IA-64 Options. (line 6)
31135 * IBM RS/6000 and PowerPC Options: RS/6000 and PowerPC Options.
31137 * identifier names, dollar signs in: Dollar Signs. (line 6)
31138 * identifiers, names in assembler code: Asm Labels. (line 6)
31139 * ilogb: Other Builtins. (line 6)
31140 * ilogbf: Other Builtins. (line 6)
31141 * ilogbl: Other Builtins. (line 6)
31142 * imaxabs: Other Builtins. (line 6)
31143 * implementation-defined behavior, C language: C Implementation.
31145 * implied #pragma implementation: C++ Interface. (line 46)
31146 * incompatibilities of GCC: Incompatibilities. (line 6)
31147 * increment operators: Bug Criteria. (line 17)
31148 * index: Other Builtins. (line 6)
31149 * indirect calls on ARM: Function Attributes.
31151 * indirect calls on MIPS: Function Attributes.
31153 * init_priority attribute: C++ Attributes. (line 9)
31154 * initializations in expressions: Compound Literals. (line 6)
31155 * initializers with labeled elements: Designated Inits. (line 6)
31156 * initializers, non-constant: Initializers. (line 6)
31157 * inline automatic for C++ member fns: Inline. (line 46)
31158 * inline functions: Inline. (line 6)
31159 * inline functions, omission of: Inline. (line 51)
31160 * inlining and C++ pragmas: C++ Interface. (line 66)
31161 * installation trouble: Trouble. (line 6)
31162 * integrating function code: Inline. (line 6)
31163 * Intel 386 Options: i386 and x86-64 Options.
31165 * interface and implementation headers, C++: C++ Interface. (line 6)
31166 * intermediate C version, nonexistent: G++ and GCC. (line 35)
31167 * interrupt handler functions: Function Attributes.
31169 * interrupt handler functions on the Blackfin, m68k, H8/300 and SH processors: Function Attributes.
31171 * introduction: Top. (line 6)
31172 * invalid assembly code: Bug Criteria. (line 12)
31173 * invalid input: Bug Criteria. (line 42)
31174 * invoking g++: Invoking G++. (line 23)
31175 * isalnum: Other Builtins. (line 6)
31176 * isalpha: Other Builtins. (line 6)
31177 * isascii: Other Builtins. (line 6)
31178 * isblank: Other Builtins. (line 6)
31179 * iscntrl: Other Builtins. (line 6)
31180 * isdigit: Other Builtins. (line 6)
31181 * isgraph: Other Builtins. (line 6)
31182 * islower: Other Builtins. (line 6)
31183 * ISO 9899: Standards. (line 6)
31184 * ISO C: Standards. (line 6)
31185 * ISO C standard: Standards. (line 6)
31186 * ISO C90: Standards. (line 6)
31187 * ISO C94: Standards. (line 6)
31188 * ISO C95: Standards. (line 6)
31189 * ISO C99: Standards. (line 6)
31190 * ISO C9X: Standards. (line 6)
31191 * ISO support: C Dialect Options. (line 10)
31192 * ISO/IEC 9899: Standards. (line 6)
31193 * isprint: Other Builtins. (line 6)
31194 * ispunct: Other Builtins. (line 6)
31195 * isspace: Other Builtins. (line 6)
31196 * isupper: Other Builtins. (line 6)
31197 * iswalnum: Other Builtins. (line 6)
31198 * iswalpha: Other Builtins. (line 6)
31199 * iswblank: Other Builtins. (line 6)
31200 * iswcntrl: Other Builtins. (line 6)
31201 * iswdigit: Other Builtins. (line 6)
31202 * iswgraph: Other Builtins. (line 6)
31203 * iswlower: Other Builtins. (line 6)
31204 * iswprint: Other Builtins. (line 6)
31205 * iswpunct: Other Builtins. (line 6)
31206 * iswspace: Other Builtins. (line 6)
31207 * iswupper: Other Builtins. (line 6)
31208 * iswxdigit: Other Builtins. (line 6)
31209 * isxdigit: Other Builtins. (line 6)
31210 * j0: Other Builtins. (line 6)
31211 * j0f: Other Builtins. (line 6)
31212 * j0l: Other Builtins. (line 6)
31213 * j1: Other Builtins. (line 6)
31214 * j1f: Other Builtins. (line 6)
31215 * j1l: Other Builtins. (line 6)
31216 * Java: G++ and GCC. (line 6)
31217 * java_interface attribute: C++ Attributes. (line 29)
31218 * jn: Other Builtins. (line 6)
31219 * jnf: Other Builtins. (line 6)
31220 * jnl: Other Builtins. (line 6)
31221 * keywords, alternate: Alternate Keywords. (line 6)
31222 * known causes of trouble: Trouble. (line 6)
31223 * labeled elements in initializers: Designated Inits. (line 6)
31224 * labels as values: Labels as Values. (line 6)
31225 * labs: Other Builtins. (line 6)
31226 * LANG: Environment Variables.
31228 * language dialect options: C Dialect Options. (line 6)
31229 * LC_ALL: Environment Variables.
31231 * LC_CTYPE: Environment Variables.
31233 * LC_MESSAGES: Environment Variables.
31235 * ldexp: Other Builtins. (line 6)
31236 * ldexpf: Other Builtins. (line 6)
31237 * ldexpl: Other Builtins. (line 6)
31238 * length-zero arrays: Zero Length. (line 6)
31239 * lgamma: Other Builtins. (line 6)
31240 * lgammaf: Other Builtins. (line 6)
31241 * lgammal: Other Builtins. (line 6)
31242 * Libraries: Link Options. (line 24)
31243 * LIBRARY_PATH: Environment Variables.
31245 * link options: Link Options. (line 6)
31246 * LL integer suffix: Long Long. (line 6)
31247 * llabs: Other Builtins. (line 6)
31248 * llrint: Other Builtins. (line 6)
31249 * llrintf: Other Builtins. (line 6)
31250 * llrintl: Other Builtins. (line 6)
31251 * llround: Other Builtins. (line 6)
31252 * llroundf: Other Builtins. (line 6)
31253 * llroundl: Other Builtins. (line 6)
31254 * load address instruction: Simple Constraints. (line 142)
31255 * local labels: Local Labels. (line 6)
31256 * local variables in macros: Typeof. (line 42)
31257 * local variables, specifying registers: Local Reg Vars. (line 6)
31258 * locale: Environment Variables.
31260 * locale definition: Environment Variables.
31262 * log: Other Builtins. (line 6)
31263 * log10: Other Builtins. (line 6)
31264 * log10f: Other Builtins. (line 6)
31265 * log10l: Other Builtins. (line 6)
31266 * log1p: Other Builtins. (line 6)
31267 * log1pf: Other Builtins. (line 6)
31268 * log1pl: Other Builtins. (line 6)
31269 * log2: Other Builtins. (line 6)
31270 * log2f: Other Builtins. (line 6)
31271 * log2l: Other Builtins. (line 6)
31272 * logb: Other Builtins. (line 6)
31273 * logbf: Other Builtins. (line 6)
31274 * logbl: Other Builtins. (line 6)
31275 * logf: Other Builtins. (line 6)
31276 * logl: Other Builtins. (line 6)
31277 * long long data types: Long Long. (line 6)
31278 * longjmp: Global Reg Vars. (line 66)
31279 * longjmp incompatibilities: Incompatibilities. (line 39)
31280 * longjmp warnings: Warning Options. (line 459)
31281 * lrint: Other Builtins. (line 6)
31282 * lrintf: Other Builtins. (line 6)
31283 * lrintl: Other Builtins. (line 6)
31284 * lround: Other Builtins. (line 6)
31285 * lroundf: Other Builtins. (line 6)
31286 * lroundl: Other Builtins. (line 6)
31287 * m in constraint: Simple Constraints. (line 17)
31288 * M32C options: M32C Options. (line 6)
31289 * M32R/D options: M32R/D Options. (line 6)
31290 * M680x0 options: M680x0 Options. (line 6)
31291 * M68hc1x options: M68hc1x Options. (line 6)
31292 * machine dependent options: Submodel Options. (line 6)
31293 * machine specific constraints: Machine Constraints.
31295 * macro with variable arguments: Variadic Macros. (line 6)
31296 * macros containing asm: Extended Asm. (line 239)
31297 * macros, inline alternative: Inline. (line 6)
31298 * macros, local labels: Local Labels. (line 6)
31299 * macros, local variables in: Typeof. (line 42)
31300 * macros, statements in expressions: Statement Exprs. (line 6)
31301 * macros, types of arguments: Typeof. (line 6)
31302 * make: Preprocessor Options.
31304 * malloc: Other Builtins. (line 6)
31305 * malloc attribute: Function Attributes.
31307 * matching constraint: Simple Constraints. (line 127)
31308 * MCore options: MCore Options. (line 6)
31309 * member fns, automatically inline: Inline. (line 46)
31310 * memcmp: Other Builtins. (line 6)
31311 * memcpy: Other Builtins. (line 6)
31312 * memory references in constraints: Simple Constraints. (line 17)
31313 * mempcpy: Other Builtins. (line 6)
31314 * memset: Other Builtins. (line 6)
31315 * Mercury: G++ and GCC. (line 23)
31316 * message formatting: Language Independent Options.
31318 * messages, warning: Warning Options. (line 6)
31319 * messages, warning and error: Warnings and Errors.
31321 * middle-operands, omitted: Conditionals. (line 6)
31322 * MIPS options: MIPS Options. (line 6)
31323 * misunderstandings in C++: C++ Misunderstandings.
31325 * mixed declarations and code: Mixed Declarations. (line 6)
31326 * mktemp, and constant strings: Incompatibilities. (line 13)
31327 * MMIX Options: MMIX Options. (line 6)
31328 * MN10300 options: MN10300 Options. (line 6)
31329 * mode attribute: Variable Attributes.
31331 * modf: Other Builtins. (line 6)
31332 * modff: Other Builtins. (line 6)
31333 * modfl: Other Builtins. (line 6)
31334 * modifiers in constraints: Modifiers. (line 6)
31335 * ms_struct: Type Attributes. (line 288)
31336 * ms_struct attribute: Variable Attributes.
31338 * MT options: MT Options. (line 6)
31339 * mudflap: Optimize Options. (line 336)
31340 * multiple alternative constraints: Multi-Alternative. (line 6)
31341 * multiprecision arithmetic: Long Long. (line 6)
31342 * n in constraint: Simple Constraints. (line 63)
31343 * names used in assembler code: Asm Labels. (line 6)
31344 * naming convention, implementation headers: C++ Interface. (line 46)
31345 * nearbyint: Other Builtins. (line 6)
31346 * nearbyintf: Other Builtins. (line 6)
31347 * nearbyintl: Other Builtins. (line 6)
31348 * nested functions: Nested Functions. (line 6)
31349 * newlines (escaped): Escaped Newlines. (line 6)
31350 * nextafter: Other Builtins. (line 6)
31351 * nextafterf: Other Builtins. (line 6)
31352 * nextafterl: Other Builtins. (line 6)
31353 * nexttoward: Other Builtins. (line 6)
31354 * nexttowardf: Other Builtins. (line 6)
31355 * nexttowardl: Other Builtins. (line 6)
31356 * NFC: Warning Options. (line 867)
31357 * NFKC: Warning Options. (line 867)
31358 * NMI handler functions on the Blackfin processor: Function Attributes.
31360 * no_instrument_function function attribute: Function Attributes.
31362 * nocommon attribute: Variable Attributes.
31364 * noinline function attribute: Function Attributes.
31366 * non-constant initializers: Initializers. (line 6)
31367 * non-static inline function: Inline. (line 63)
31368 * nonnull function attribute: Function Attributes.
31370 * noreturn function attribute: Function Attributes.
31372 * nothrow function attribute: Function Attributes.
31374 * o in constraint: Simple Constraints. (line 21)
31375 * OBJC_INCLUDE_PATH: Environment Variables.
31377 * Objective-C <1>: Standards. (line 110)
31378 * Objective-C: G++ and GCC. (line 6)
31379 * Objective-C and Objective-C++ options, command line: Objective-C and Objective-C++ Dialect Options.
31381 * Objective-C++ <1>: Standards. (line 110)
31382 * Objective-C++: G++ and GCC. (line 6)
31383 * offsettable address: Simple Constraints. (line 21)
31384 * old-style function definitions: Function Prototypes.
31386 * omitted middle-operands: Conditionals. (line 6)
31387 * open coding: Inline. (line 6)
31388 * operand constraints, asm: Constraints. (line 6)
31389 * optimize options: Optimize Options. (line 6)
31390 * options to control diagnostics formatting: Language Independent Options.
31392 * options to control warnings: Warning Options. (line 6)
31393 * options, C++: C++ Dialect Options.
31395 * options, code generation: Code Gen Options. (line 6)
31396 * options, debugging: Debugging Options. (line 6)
31397 * options, dialect: C Dialect Options. (line 6)
31398 * options, directory search: Directory Options. (line 6)
31399 * options, GCC command: Invoking GCC. (line 6)
31400 * options, grouping: Invoking GCC. (line 26)
31401 * options, linking: Link Options. (line 6)
31402 * options, Objective-C and Objective-C++: Objective-C and Objective-C++ Dialect Options.
31404 * options, optimization: Optimize Options. (line 6)
31405 * options, order: Invoking GCC. (line 30)
31406 * options, preprocessor: Preprocessor Options.
31408 * order of evaluation, side effects: Non-bugs. (line 196)
31409 * order of options: Invoking GCC. (line 30)
31410 * other register constraints: Simple Constraints. (line 151)
31411 * output file option: Overall Options. (line 173)
31412 * overloaded virtual fn, warning: C++ Dialect Options.
31414 * p in constraint: Simple Constraints. (line 142)
31415 * packed attribute: Variable Attributes.
31417 * parameter forward declaration: Variable Length. (line 60)
31418 * parameters, aliased: Code Gen Options. (line 333)
31419 * Pascal: G++ and GCC. (line 23)
31420 * PDP-11 Options: PDP-11 Options. (line 6)
31421 * PIC: Code Gen Options. (line 170)
31422 * pmf: Bound member functions.
31424 * pointer arguments: Function Attributes.
31426 * pointer to member function: Bound member functions.
31428 * portions of temporary objects, pointers to: Temporaries. (line 6)
31429 * pow: Other Builtins. (line 6)
31430 * pow10: Other Builtins. (line 6)
31431 * pow10f: Other Builtins. (line 6)
31432 * pow10l: Other Builtins. (line 6)
31433 * PowerPC options: PowerPC Options. (line 6)
31434 * powf: Other Builtins. (line 6)
31435 * powl: Other Builtins. (line 6)
31436 * pragma, align: Solaris Pragmas. (line 11)
31437 * pragma, extern_prefix: Symbol-Renaming Pragmas.
31439 * pragma, fini: Solaris Pragmas. (line 19)
31440 * pragma, init: Solaris Pragmas. (line 24)
31441 * pragma, long_calls: ARM Pragmas. (line 11)
31442 * pragma, long_calls_off: ARM Pragmas. (line 17)
31443 * pragma, longcall: RS/6000 and PowerPC Pragmas.
31445 * pragma, mark: Darwin Pragmas. (line 11)
31446 * pragma, memregs: M32C Pragmas. (line 7)
31447 * pragma, no_long_calls: ARM Pragmas. (line 14)
31448 * pragma, options align: Darwin Pragmas. (line 14)
31449 * pragma, reason for not using: Function Attributes.
31451 * pragma, redefine_extname: Symbol-Renaming Pragmas.
31453 * pragma, segment: Darwin Pragmas. (line 21)
31454 * pragma, unused: Darwin Pragmas. (line 24)
31455 * pragma, weak: Weak Pragmas. (line 10)
31456 * pragmas: Pragmas. (line 6)
31457 * pragmas in C++, effect on inlining: C++ Interface. (line 66)
31458 * pragmas, interface and implementation: C++ Interface. (line 6)
31459 * pragmas, warning of unknown: Warning Options. (line 476)
31460 * precompiled headers: Precompiled Headers.
31462 * preprocessing numbers: Incompatibilities. (line 173)
31463 * preprocessing tokens: Incompatibilities. (line 173)
31464 * preprocessor options: Preprocessor Options.
31466 * printf: Other Builtins. (line 6)
31467 * printf_unlocked: Other Builtins. (line 6)
31468 * prof: Debugging Options. (line 121)
31469 * promotion of formal parameters: Function Prototypes.
31471 * pure function attribute: Function Attributes.
31473 * push address instruction: Simple Constraints. (line 142)
31474 * putchar: Other Builtins. (line 6)
31475 * puts: Other Builtins. (line 6)
31476 * qsort, and global register variables: Global Reg Vars. (line 42)
31477 * question mark: Multi-Alternative. (line 27)
31478 * r in constraint: Simple Constraints. (line 54)
31479 * ranges in case statements: Case Ranges. (line 6)
31480 * read-only strings: Incompatibilities. (line 9)
31481 * register variable after longjmp: Global Reg Vars. (line 66)
31482 * registers: Extended Asm. (line 6)
31483 * registers for local variables: Local Reg Vars. (line 6)
31484 * registers in constraints: Simple Constraints. (line 54)
31485 * registers, global allocation: Explicit Reg Vars. (line 6)
31486 * registers, global variables in: Global Reg Vars. (line 6)
31487 * regparm attribute: Function Attributes.
31489 * relocation truncated to fit (MIPS): MIPS Options. (line 113)
31490 * remainder: Other Builtins. (line 6)
31491 * remainderf: Other Builtins. (line 6)
31492 * remainderl: Other Builtins. (line 6)
31493 * remquo: Other Builtins. (line 6)
31494 * remquof: Other Builtins. (line 6)
31495 * remquol: Other Builtins. (line 6)
31496 * reordering, warning: C++ Dialect Options.
31498 * reporting bugs: Bugs. (line 6)
31499 * rest argument (in macro): Variadic Macros. (line 6)
31500 * restricted pointers: Restricted Pointers.
31502 * restricted references: Restricted Pointers.
31504 * restricted this pointer: Restricted Pointers.
31506 * returns_twice attribute: Function Attributes.
31508 * rindex: Other Builtins. (line 6)
31509 * rint: Other Builtins. (line 6)
31510 * rintf: Other Builtins. (line 6)
31511 * rintl: Other Builtins. (line 6)
31512 * round: Other Builtins. (line 6)
31513 * roundf: Other Builtins. (line 6)
31514 * roundl: Other Builtins. (line 6)
31515 * RS/6000 and PowerPC Options: RS/6000 and PowerPC Options.
31517 * RTTI: Vague Linkage. (line 43)
31518 * run-time options: Code Gen Options. (line 6)
31519 * s in constraint: Simple Constraints. (line 90)
31520 * S/390 and zSeries Options: S/390 and zSeries Options.
31522 * save all registers on the Blackfin, H8/300, H8/300H, and H8S: Function Attributes.
31524 * scalb: Other Builtins. (line 6)
31525 * scalbf: Other Builtins. (line 6)
31526 * scalbl: Other Builtins. (line 6)
31527 * scalbln: Other Builtins. (line 6)
31528 * scalblnf: Other Builtins. (line 6)
31529 * scalbn: Other Builtins. (line 6)
31530 * scalbnf: Other Builtins. (line 6)
31531 * scanf, and constant strings: Incompatibilities. (line 17)
31532 * scanfnl: Other Builtins. (line 6)
31533 * scope of a variable length array: Variable Length. (line 23)
31534 * scope of declaration: Disappointments. (line 21)
31535 * scope of external declarations: Incompatibilities. (line 80)
31536 * search path: Directory Options. (line 6)
31537 * section function attribute: Function Attributes.
31539 * section variable attribute: Variable Attributes.
31541 * sentinel function attribute: Function Attributes.
31543 * setjmp: Global Reg Vars. (line 66)
31544 * setjmp incompatibilities: Incompatibilities. (line 39)
31545 * shared strings: Incompatibilities. (line 9)
31546 * shared variable attribute: Variable Attributes.
31548 * side effect in ?:: Conditionals. (line 20)
31549 * side effects, macro argument: Statement Exprs. (line 35)
31550 * side effects, order of evaluation: Non-bugs. (line 196)
31551 * signal handler functions on the AVR processors: Function Attributes.
31553 * signbit: Other Builtins. (line 6)
31554 * signbitf: Other Builtins. (line 6)
31555 * signbitl: Other Builtins. (line 6)
31556 * signed and unsigned values, comparison warning: Warning Options.
31558 * significand: Other Builtins. (line 6)
31559 * significandf: Other Builtins. (line 6)
31560 * significandl: Other Builtins. (line 6)
31561 * simple constraints: Simple Constraints. (line 6)
31562 * sin: Other Builtins. (line 6)
31563 * sincos: Other Builtins. (line 6)
31564 * sincosf: Other Builtins. (line 6)
31565 * sincosl: Other Builtins. (line 6)
31566 * sinf: Other Builtins. (line 6)
31567 * sinh: Other Builtins. (line 6)
31568 * sinhf: Other Builtins. (line 6)
31569 * sinhl: Other Builtins. (line 6)
31570 * sinl: Other Builtins. (line 6)
31571 * sizeof: Typeof. (line 6)
31572 * smaller data references: M32R/D Options. (line 57)
31573 * smaller data references (MIPS): MIPS Options. (line 194)
31574 * smaller data references (PowerPC): RS/6000 and PowerPC Options.
31576 * snprintf: Other Builtins. (line 6)
31577 * SPARC options: SPARC Options. (line 6)
31578 * Spec Files: Spec Files. (line 6)
31579 * specified registers: Explicit Reg Vars. (line 6)
31580 * specifying compiler version and target machine: Target Options.
31582 * specifying hardware config: Submodel Options. (line 6)
31583 * specifying machine version: Target Options. (line 6)
31584 * specifying registers for local variables: Local Reg Vars. (line 6)
31585 * speed of compilation: Precompiled Headers.
31587 * sprintf: Other Builtins. (line 6)
31588 * sqrt: Other Builtins. (line 6)
31589 * sqrtf: Other Builtins. (line 6)
31590 * sqrtl: Other Builtins. (line 6)
31591 * sscanf: Other Builtins. (line 6)
31592 * sscanf, and constant strings: Incompatibilities. (line 17)
31593 * sseregparm attribute: Function Attributes.
31595 * statements inside expressions: Statement Exprs. (line 6)
31596 * static data in C++, declaring and defining: Static Definitions.
31598 * stpcpy: Other Builtins. (line 6)
31599 * stpncpy: Other Builtins. (line 6)
31600 * strcasecmp: Other Builtins. (line 6)
31601 * strcat: Other Builtins. (line 6)
31602 * strchr: Other Builtins. (line 6)
31603 * strcmp: Other Builtins. (line 6)
31604 * strcpy: Other Builtins. (line 6)
31605 * strcspn: Other Builtins. (line 6)
31606 * strdup: Other Builtins. (line 6)
31607 * strfmon: Other Builtins. (line 6)
31608 * strftime: Other Builtins. (line 6)
31609 * string constants: Incompatibilities. (line 9)
31610 * strlen: Other Builtins. (line 6)
31611 * strncasecmp: Other Builtins. (line 6)
31612 * strncat: Other Builtins. (line 6)
31613 * strncmp: Other Builtins. (line 6)
31614 * strncpy: Other Builtins. (line 6)
31615 * strndup: Other Builtins. (line 6)
31616 * strpbrk: Other Builtins. (line 6)
31617 * strrchr: Other Builtins. (line 6)
31618 * strspn: Other Builtins. (line 6)
31619 * strstr: Other Builtins. (line 6)
31620 * struct: Unnamed Fields. (line 6)
31621 * structures: Incompatibilities. (line 146)
31622 * structures, constructor expression: Compound Literals. (line 6)
31623 * submodel options: Submodel Options. (line 6)
31624 * subscripting: Subscripting. (line 6)
31625 * subscripting and function values: Subscripting. (line 6)
31626 * suffixes for C++ source: Invoking G++. (line 6)
31627 * SUNPRO_DEPENDENCIES: Environment Variables.
31629 * suppressing warnings: Warning Options. (line 6)
31630 * surprises in C++: C++ Misunderstandings.
31632 * syntax checking: Warning Options. (line 22)
31633 * system headers, warnings from: Warning Options. (line 595)
31634 * tan: Other Builtins. (line 6)
31635 * tanf: Other Builtins. (line 6)
31636 * tanh: Other Builtins. (line 6)
31637 * tanhf: Other Builtins. (line 6)
31638 * tanhl: Other Builtins. (line 6)
31639 * tanl: Other Builtins. (line 6)
31640 * target machine, specifying: Target Options. (line 6)
31641 * target options: Target Options. (line 6)
31642 * TC1: Standards. (line 6)
31643 * TC2: Standards. (line 6)
31644 * Technical Corrigenda: Standards. (line 6)
31645 * Technical Corrigendum 1: Standards. (line 6)
31646 * Technical Corrigendum 2: Standards. (line 6)
31647 * template instantiation: Template Instantiation.
31649 * temporaries, lifetime of: Temporaries. (line 6)
31650 * tgamma: Other Builtins. (line 6)
31651 * tgammaf: Other Builtins. (line 6)
31652 * tgammal: Other Builtins. (line 6)
31653 * Thread-Local Storage: Thread-Local. (line 6)
31654 * thunks: Nested Functions. (line 6)
31655 * tiny data section on the H8/300H and H8S: Function Attributes.
31657 * TLS: Thread-Local. (line 6)
31658 * tls_model attribute: Variable Attributes.
31660 * TMPDIR: Environment Variables.
31662 * TMS320C3x/C4x Options: TMS320C3x/C4x Options.
31664 * toascii: Other Builtins. (line 6)
31665 * tolower: Other Builtins. (line 6)
31666 * toupper: Other Builtins. (line 6)
31667 * towlower: Other Builtins. (line 6)
31668 * towupper: Other Builtins. (line 6)
31669 * traditional C language: C Dialect Options. (line 200)
31670 * treelang <1>: Standards. (line 123)
31671 * treelang: G++ and GCC. (line 6)
31672 * trunc: Other Builtins. (line 6)
31673 * truncf: Other Builtins. (line 6)
31674 * truncl: Other Builtins. (line 6)
31675 * two-stage name lookup: Name lookup. (line 6)
31676 * type alignment: Alignment. (line 6)
31677 * type attributes: Type Attributes. (line 6)
31678 * type_info: Vague Linkage. (line 43)
31679 * typedef names as function parameters: Incompatibilities. (line 97)
31680 * typeof: Typeof. (line 6)
31681 * ULL integer suffix: Long Long. (line 6)
31682 * Ultrix calling convention: Interoperation. (line 150)
31683 * undefined behavior: Bug Criteria. (line 17)
31684 * undefined function value: Bug Criteria. (line 17)
31685 * underscores in variables in macros: Typeof. (line 42)
31686 * union: Unnamed Fields. (line 6)
31687 * union, casting to a: Cast to Union. (line 6)
31688 * unions: Incompatibilities. (line 146)
31689 * unknown pragmas, warning: Warning Options. (line 476)
31690 * unresolved references and -nodefaultlibs: Link Options. (line 79)
31691 * unresolved references and -nostdlib: Link Options. (line 79)
31692 * unused attribute.: Function Attributes.
31694 * used attribute.: Function Attributes.
31696 * User stack pointer in interrupts on the Blackfin: Function Attributes.
31698 * V in constraint: Simple Constraints. (line 41)
31699 * V850 Options: V850 Options. (line 6)
31700 * vague linkage: Vague Linkage. (line 6)
31701 * value after longjmp: Global Reg Vars. (line 66)
31702 * variable addressability on the IA-64: Function Attributes.
31704 * variable addressability on the M32R/D: Variable Attributes.
31706 * variable alignment: Alignment. (line 6)
31707 * variable attributes: Variable Attributes.
31709 * variable number of arguments: Variadic Macros. (line 6)
31710 * variable-length array scope: Variable Length. (line 23)
31711 * variable-length arrays: Variable Length. (line 6)
31712 * variables in specified registers: Explicit Reg Vars. (line 6)
31713 * variables, local, in macros: Typeof. (line 42)
31714 * variadic macros: Variadic Macros. (line 6)
31715 * VAX calling convention: Interoperation. (line 150)
31716 * VAX options: VAX Options. (line 6)
31717 * vfprintf: Other Builtins. (line 6)
31718 * vfscanf: Other Builtins. (line 6)
31719 * visibility attribute: Function Attributes.
31721 * VLAs: Variable Length. (line 6)
31722 * void pointers, arithmetic: Pointer Arith. (line 6)
31723 * void, size of pointer to: Pointer Arith. (line 6)
31724 * volatile access: Volatiles. (line 6)
31725 * volatile applied to function: Function Attributes.
31727 * volatile read: Volatiles. (line 6)
31728 * volatile write: Volatiles. (line 6)
31729 * vprintf: Other Builtins. (line 6)
31730 * vscanf: Other Builtins. (line 6)
31731 * vsnprintf: Other Builtins. (line 6)
31732 * vsprintf: Other Builtins. (line 6)
31733 * vsscanf: Other Builtins. (line 6)
31734 * vtable: Vague Linkage. (line 28)
31735 * warn_unused_result attribute: Function Attributes.
31737 * warning for comparison of signed and unsigned values: Warning Options.
31739 * warning for overloaded virtual fn: C++ Dialect Options.
31741 * warning for reordering of member initializers: C++ Dialect Options.
31743 * warning for unknown pragmas: Warning Options. (line 476)
31744 * warning messages: Warning Options. (line 6)
31745 * warnings from system headers: Warning Options. (line 595)
31746 * warnings vs errors: Warnings and Errors.
31748 * weak attribute: Function Attributes.
31750 * weakref attribute: Function Attributes.
31752 * whitespace: Incompatibilities. (line 112)
31753 * X in constraint: Simple Constraints. (line 112)
31754 * X3.159-1989: Standards. (line 6)
31755 * x86-64 options: x86-64 Options. (line 6)
31756 * x86-64 Options: i386 and x86-64 Options.
31758 * Xstormy16 Options: Xstormy16 Options. (line 6)
31759 * Xtensa Options: Xtensa Options. (line 6)
31760 * y0: Other Builtins. (line 6)
31761 * y0f: Other Builtins. (line 6)
31762 * y0l: Other Builtins. (line 6)
31763 * y1: Other Builtins. (line 6)
31764 * y1f: Other Builtins. (line 6)
31765 * y1l: Other Builtins. (line 6)
31766 * yn: Other Builtins. (line 6)
31767 * ynf: Other Builtins. (line 6)
31768 * ynl: Other Builtins. (line 6)
31769 * zero-length arrays: Zero Length. (line 6)
31770 * zero-size structures: Empty Structures. (line 6)
31771 * zSeries options: zSeries Options. (line 6)
31777 Node: G++ and GCC
\7f3740
31778 Node: Standards
\7f5805
31779 Node: Invoking GCC
\7f12932
31780 Node: Option Summary
\7f16688
31781 Node: Overall Options
\7f44294
31782 Node: Invoking G++
\7f52764
31783 Node: C Dialect Options
\7f54386
31784 Node: C++ Dialect Options
\7f65396
31785 Node: Objective-C and Objective-C++ Dialect Options
\7f83854
31786 Node: Language Independent Options
\7f95461
31787 Node: Warning Options
\7f97544
31788 Node: Debugging Options
\7f143589
31789 Node: Optimize Options
\7f173509
31790 Node: Preprocessor Options
\7f250361
31791 Ref: Wtrigraphs
\7f254325
31792 Ref: dashMF
\7f259082
31793 Ref: fdollars-in-identifiers
\7f268133
31794 Node: Assembler Options
\7f276189
31795 Node: Link Options
\7f276894
31796 Ref: Link Options-Footnote-1
\7f285462
31797 Node: Directory Options
\7f285796
31798 Node: Spec Files
\7f291858
31799 Node: Target Options
\7f311146
31800 Node: Submodel Options
\7f312570
31801 Node: ARC Options
\7f314160
31802 Node: ARM Options
\7f315350
31803 Node: AVR Options
\7f326961
31804 Node: Blackfin Options
\7f329094
31805 Node: CRIS Options
\7f331862
31806 Node: CRX Options
\7f336081
31807 Node: Darwin Options
\7f336506
31808 Node: DEC Alpha Options
\7f343094
31809 Node: DEC Alpha/VMS Options
\7f354572
31810 Node: FRV Options
\7f354957
31811 Node: H8/300 Options
\7f361624
31812 Node: HPPA Options
\7f362685
31813 Node: i386 and x86-64 Options
\7f372278
31814 Node: IA-64 Options
\7f391095
31815 Node: M32C Options
\7f395086
31816 Node: M32R/D Options
\7f396377
31817 Node: M680x0 Options
\7f399964
31818 Node: M68hc1x Options
\7f407197
31819 Node: MCore Options
\7f408765
31820 Node: MIPS Options
\7f409786
31821 Node: MMIX Options
\7f423858
31822 Node: MN10300 Options
\7f426340
31823 Node: MT Options
\7f427758
31824 Node: PDP-11 Options
\7f428672
31825 Node: PowerPC Options
\7f430506
31826 Node: RS/6000 and PowerPC Options
\7f430740
31827 Node: S/390 and zSeries Options
\7f458810
31828 Node: SH Options
\7f466122
31829 Node: SPARC Options
\7f475358
31830 Node: System V Options
\7f486041
31831 Node: TMS320C3x/C4x Options
\7f486875
31832 Node: V850 Options
\7f492400
31833 Node: VAX Options
\7f495545
31834 Node: x86-64 Options
\7f496092
31835 Node: Xstormy16 Options
\7f496306
31836 Node: Xtensa Options
\7f496595
31837 Node: zSeries Options
\7f500435
31838 Node: Code Gen Options
\7f500631
31839 Node: Environment Variables
\7f520251
31840 Node: Precompiled Headers
\7f527923
31841 Node: Running Protoize
\7f534160
31842 Node: C Implementation
\7f540497
31843 Node: Translation implementation
\7f542160
31844 Node: Environment implementation
\7f542734
31845 Node: Identifiers implementation
\7f543284
31846 Node: Characters implementation
\7f544338
31847 Node: Integers implementation
\7f547144
31848 Node: Floating point implementation
\7f548969
31849 Node: Arrays and pointers implementation
\7f551898
31850 Ref: Arrays and pointers implementation-Footnote-1
\7f553333
31851 Node: Hints implementation
\7f553457
31852 Node: Structures unions enumerations and bit-fields implementation
\7f554923
31853 Node: Qualifiers implementation
\7f556886
31854 Node: Declarators implementation
\7f558658
31855 Node: Statements implementation
\7f559000
31856 Node: Preprocessing directives implementation
\7f559327
31857 Node: Library functions implementation
\7f561432
31858 Node: Architecture implementation
\7f562072
31859 Node: Locale-specific behavior implementation
\7f562775
31860 Node: C Extensions
\7f563080
31861 Node: Statement Exprs
\7f567430
31862 Node: Local Labels
\7f571943
31863 Node: Labels as Values
\7f574922
31864 Ref: Labels as Values-Footnote-1
\7f576976
31865 Node: Nested Functions
\7f577159
31866 Node: Constructing Calls
\7f581053
31867 Node: Typeof
\7f583389
31868 Node: Conditionals
\7f586555
31869 Node: Long Long
\7f587446
31870 Node: Complex
\7f588947
31871 Node: Hex Floats
\7f591513
31872 Node: Zero Length
\7f592548
31873 Node: Empty Structures
\7f595825
31874 Node: Variable Length
\7f596241
31875 Node: Variadic Macros
\7f599008
31876 Node: Escaped Newlines
\7f601390
31877 Node: Subscripting
\7f602229
31878 Node: Pointer Arith
\7f602952
31879 Node: Initializers
\7f603520
31880 Node: Compound Literals
\7f604016
31881 Node: Designated Inits
\7f606191
31882 Node: Case Ranges
\7f609846
31883 Node: Cast to Union
\7f610529
31884 Node: Mixed Declarations
\7f611625
31885 Node: Function Attributes
\7f612131
31886 Node: Attribute Syntax
\7f650166
31887 Node: Function Prototypes
\7f661250
31888 Node: C++ Comments
\7f663031
31889 Node: Dollar Signs
\7f663550
31890 Node: Character Escapes
\7f664015
31891 Node: Alignment
\7f664309
31892 Node: Variable Attributes
\7f665626
31893 Node: Type Attributes
\7f679658
31894 Node: Inline
\7f693109
31895 Node: Extended Asm
\7f697813
31896 Ref: Example of asm with clobbered asm reg
\7f703899
31897 Node: Constraints
\7f717995
31898 Node: Simple Constraints
\7f718845
31899 Node: Multi-Alternative
\7f725373
31900 Node: Modifiers
\7f727090
31901 Node: Machine Constraints
\7f729984
31902 Node: Asm Labels
\7f756119
31903 Node: Explicit Reg Vars
\7f757795
31904 Node: Global Reg Vars
\7f759403
31905 Node: Local Reg Vars
\7f763953
31906 Node: Alternate Keywords
\7f766394
31907 Node: Incomplete Enums
\7f767822
31908 Node: Function Names
\7f768579
31909 Node: Return Address
\7f770769
31910 Node: Vector Extensions
\7f773566
31911 Node: Offsetof
\7f777068
31912 Node: Atomic Builtins
\7f777854
31913 Node: Object Size Checking
\7f782939
31914 Node: Other Builtins
\7f788296
31915 Node: Target Builtins
\7f809490
31916 Node: Alpha Built-in Functions
\7f810223
31917 Node: ARM Built-in Functions
\7f813215
31918 Node: Blackfin Built-in Functions
\7f819922
31919 Node: FR-V Built-in Functions
\7f820539
31920 Node: Argument Types
\7f821398
31921 Node: Directly-mapped Integer Functions
\7f823154
31922 Node: Directly-mapped Media Functions
\7f824236
31923 Node: Raw read/write Functions
\7f831268
31924 Node: Other Built-in Functions
\7f832180
31925 Node: X86 Built-in Functions
\7f833369
31926 Node: MIPS DSP Built-in Functions
\7f851492
31927 Node: MIPS Paired-Single Support
\7f859917
31928 Node: Paired-Single Arithmetic
\7f861527
31929 Node: Paired-Single Built-in Functions
\7f862467
31930 Node: MIPS-3D Built-in Functions
\7f865131
31931 Node: PowerPC AltiVec Built-in Functions
\7f870500
31932 Node: SPARC VIS Built-in Functions
\7f971804
31933 Node: Target Format Checks
\7f973463
31934 Node: Solaris Format Checks
\7f973870
31935 Node: Pragmas
\7f974267
31936 Node: ARM Pragmas
\7f974851
31937 Node: M32C Pragmas
\7f975454
31938 Node: RS/6000 and PowerPC Pragmas
\7f976030
31939 Node: Darwin Pragmas
\7f976772
31940 Node: Solaris Pragmas
\7f977839
31941 Node: Symbol-Renaming Pragmas
\7f979000
31942 Node: Structure-Packing Pragmas
\7f981622
31943 Node: Weak Pragmas
\7f982885
31944 Node: Unnamed Fields
\7f983660
31945 Node: Thread-Local
\7f985170
31946 Node: C99 Thread-Local Edits
\7f987254
31947 Node: C++98 Thread-Local Edits
\7f989266
31948 Node: C++ Extensions
\7f992711
31949 Node: Volatiles
\7f994283
31950 Node: Restricted Pointers
\7f996959
31951 Node: Vague Linkage
\7f998553
31952 Node: C++ Interface
\7f1002209
31953 Ref: C++ Interface-Footnote-1
\7f1006506
31954 Node: Template Instantiation
\7f1006643
31955 Node: Bound member functions
\7f1013655
31956 Node: C++ Attributes
\7f1015198
31957 Node: Strong Using
\7f1016838
31958 Node: Java Exceptions
\7f1018087
31959 Node: Deprecated Features
\7f1019483
31960 Node: Backwards Compatibility
\7f1022458
31961 Node: Objective-C
\7f1023813
31962 Node: Executing code before main
\7f1024394
31963 Node: What you can and what you cannot do in +load
\7f1027000
31964 Node: Type encoding
\7f1029167
31965 Node: Garbage Collection
\7f1032410
31966 Node: Constant string objects
\7f1035034
31967 Node: compatibility_alias
\7f1037542
31968 Node: Compatibility
\7f1038420
31969 Node: Gcov
\7f1044987
31970 Node: Gcov Intro
\7f1045511
31971 Node: Invoking Gcov
\7f1048227
31972 Node: Gcov and Optimization
\7f1060083
31973 Node: Gcov Data Files
\7f1062736
31974 Node: Cross-profiling
\7f1063874
31975 Node: Trouble
\7f1065700
31976 Node: Actual Bugs
\7f1067240
31977 Node: Cross-Compiler Problems
\7f1067980
31978 Node: Interoperation
\7f1068394
31979 Node: Incompatibilities
\7f1075992
31980 Node: Fixed Headers
\7f1084142
31981 Node: Standard Libraries
\7f1085805
31982 Node: Disappointments
\7f1087177
31983 Node: C++ Misunderstandings
\7f1091535
31984 Node: Static Definitions
\7f1092354
31985 Node: Name lookup
\7f1093407
31986 Ref: Name lookup-Footnote-1
\7f1098185
31987 Node: Temporaries
\7f1098372
31988 Node: Copy Assignment
\7f1100348
31989 Node: Protoize Caveats
\7f1102155
31990 Node: Non-bugs
\7f1106117
31991 Node: Warnings and Errors
\7f1116621
31992 Node: Bugs
\7f1118385
31993 Node: Bug Criteria
\7f1118949
31994 Node: Bug Reporting
\7f1121159
31995 Node: Service
\7f1121551
31996 Node: Contributing
\7f1122370
31997 Node: Funding
\7f1123110
31998 Node: GNU Project
\7f1125599
31999 Node: Copying
\7f1126245
32000 Node: GNU Free Documentation License
\7f1145422
32001 Node: Contributors
\7f1167828
32002 Node: Option Index
\7f1203392
32003 Node: Keyword Index
\7f1335138