1 This is doc/gcc.info, produced by makeinfo version 4.8 from
2 ../../../toolchain/android-toolchain/gcc-4.2.1/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 Software development
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 4.2.1. 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-4.2/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 GNU Fortran.
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, `-fmove-loop-invariants', `-Wformat' and so on. Most of these
311 have both positive and negative forms; the negative form of `-ffoo'
312 would be `-fno-foo'. This manual documents only one of these two
313 forms, 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 @FILE
364 *Note Options Controlling C Dialect: C Dialect Options.
365 -ansi -std=STANDARD -fgnu89-inline
367 -fno-asm -fno-builtin -fno-builtin-FUNCTION
368 -fhosted -ffreestanding -fopenmp -fms-extensions
369 -trigraphs -no-integrated-cpp -traditional -traditional-cpp
370 -fallow-single-precision -fcond-mismatch
371 -fsigned-bitfields -fsigned-char
372 -funsigned-bitfields -funsigned-char
374 _C++ Language Options_
375 *Note Options Controlling C++ Dialect: C++ Dialect Options.
376 -fabi-version=N -fno-access-control -fcheck-new
377 -fconserve-space -ffriend-injection
378 -fno-elide-constructors
379 -fno-enforce-eh-specs
380 -ffor-scope -fno-for-scope -fno-gnu-keywords
381 -fno-implicit-templates
382 -fno-implicit-inline-templates
383 -fno-implement-inlines -fms-extensions
384 -fno-nonansi-builtins -fno-operator-names
385 -fno-optional-diags -fpermissive
386 -frepo -fno-rtti -fstats -ftemplate-depth-N
387 -fno-threadsafe-statics -fuse-cxa-atexit -fno-weak -nostdinc++
388 -fno-default-inline -fvisibility-inlines-hidden
389 -Wabi -Wctor-dtor-privacy
390 -Wnon-virtual-dtor -Wreorder
391 -Weffc++ -Wno-deprecated -Wstrict-null-sentinel
392 -Wno-non-template-friend -Wold-style-cast
393 -Woverloaded-virtual -Wno-pmf-conversions
396 _Objective-C and Objective-C++ Language Options_
397 *Note Options Controlling Objective-C and Objective-C++ Dialects:
398 Objective-C and Objective-C++ Dialect Options.
399 -fconstant-string-class=CLASS-NAME
400 -fgnu-runtime -fnext-runtime
402 -fobjc-call-cxx-cdtors
403 -fobjc-direct-dispatch
406 -freplace-objc-classes
410 -Wno-protocol -Wselector
411 -Wstrict-selector-match
412 -Wundeclared-selector
414 _Language Independent Options_
415 *Note Options to Control Diagnostic Messages Formatting: Language
418 -fdiagnostics-show-location=[once|every-line]
419 -fdiagnostics-show-option
422 *Note Options to Request or Suppress Warnings: Warning Options.
423 -fsyntax-only -pedantic -pedantic-errors
424 -w -Wextra -Wall -Waddress -Waggregate-return -Wno-attributes
425 -Wc++-compat -Wcast-align -Wcast-qual -Wchar-subscripts -Wcomment
426 -Wconversion -Wno-deprecated-declarations
427 -Wdisabled-optimization -Wno-div-by-zero -Wno-endif-labels
428 -Werror -Werror=* -Werror-implicit-function-declaration
429 -Wfatal-errors -Wfloat-equal -Wformat -Wformat=2
430 -Wno-format-extra-args -Wformat-nonliteral
431 -Wformat-security -Wformat-y2k
432 -Wimplicit -Wimplicit-function-declaration -Wimplicit-int
433 -Wimport -Wno-import -Winit-self -Winline
434 -Wno-int-to-pointer-cast
435 -Wno-invalid-offsetof -Winvalid-pch
436 -Wlarger-than-LEN -Wframe-larger-than-LEN
437 -Wunsafe-loop-optimizations -Wlong-long
438 -Wmain -Wmissing-braces -Wmissing-field-initializers
439 -Wmissing-format-attribute -Wmissing-include-dirs
441 -Wno-multichar -Wnonnull -Wno-overflow
442 -Woverlength-strings -Wpacked -Wpadded
443 -Wparentheses -Wpointer-arith -Wno-pointer-to-int-cast
445 -Wreturn-type -Wsequence-point -Wshadow
446 -Wsign-compare -Wstack-protector
447 -Wstrict-aliasing -Wstrict-aliasing=n
448 -Wstrict-overflow -Wstrict-overflow=N
449 -Wswitch -Wswitch-default -Wswitch-enum
450 -Wsystem-headers -Wtrigraphs -Wundef -Wuninitialized
451 -Wunknown-pragmas -Wno-pragmas -Wunreachable-code
452 -Wunused -Wunused-function -Wunused-label -Wunused-parameter
453 -Wunused-value -Wunused-variable
454 -Wvariadic-macros -Wvla
455 -Wvolatile-register-var -Wwrite-strings
457 _C-only Warning Options_
458 -Wbad-function-cast -Wmissing-declarations
459 -Wmissing-prototypes -Wnested-externs -Wold-style-definition
460 -Wstrict-prototypes -Wtraditional
461 -Wdeclaration-after-statement -Wpointer-sign
464 *Note Options for Debugging Your Program or GCC: Debugging Options.
465 -dLETTERS -dumpspecs -dumpmachine -dumpversion
466 -fdump-noaddr -fdump-unnumbered -fdump-translation-unit[-N]
467 -fdump-class-hierarchy[-N]
468 -fdump-ipa-all -fdump-ipa-cgraph
470 -fdump-tree-original[-N]
471 -fdump-tree-optimized[-N]
472 -fdump-tree-inlined[-N]
473 -fdump-tree-cfg -fdump-tree-vcg -fdump-tree-alias
475 -fdump-tree-ssa[-N] -fdump-tree-pre[-N]
476 -fdump-tree-ccp[-N] -fdump-tree-dce[-N]
477 -fdump-tree-gimple[-raw] -fdump-tree-mudflap[-N]
480 -fdump-tree-phiopt[-N]
481 -fdump-tree-forwprop[-N]
482 -fdump-tree-copyrename[-N]
483 -fdump-tree-nrv -fdump-tree-vect
489 -ftree-vectorizer-verbose=N
490 -fdump-tree-storeccp[-N]
491 -feliminate-dwarf2-dups -feliminate-unused-debug-types
492 -feliminate-unused-debug-symbols -femit-class-debug-always
493 -fmem-report -fprofile-arcs
494 -frandom-seed=STRING -fsched-verbose=N
495 -ftest-coverage -ftime-report -fvar-tracking
496 -g -gLEVEL -gcoff -gdwarf-2
497 -ggdb -gstabs -gstabs+ -gvms -gxcoff -gxcoff+
498 -femit-struct-debug-baseonly -femit-struct-debug-reduced
499 -femit-struct-debug-detailed[=SPEC-LIST]
500 -p -pg -print-file-name=LIBRARY -print-libgcc-file-name
501 -print-multi-directory -print-multi-lib
502 -print-prog-name=PROGRAM -print-search-dirs -Q
505 _Optimization Options_
506 *Note Options that Control Optimization: Optimize Options.
507 -falign-functions=N -falign-jumps=N
508 -falign-labels=N -falign-loops=N
509 -fbounds-check -fmudflap -fmudflapth -fmudflapir
510 -fbranch-probabilities -fprofile-values -fvpt -fbranch-target-load-optimize
511 -fbranch-target-load-optimize2 -fbtr-bb-exclusive
512 -fcaller-saves -fcprop-registers -fcse-follow-jumps
513 -fcse-skip-blocks -fcx-limited-range -fdata-sections
514 -fdelayed-branch -fdelete-null-pointer-checks -fearly-inlining
515 -fexpensive-optimizations -ffast-math -ffloat-store
516 -fforce-addr -ffunction-sections
517 -fgcse -fgcse-lm -fgcse-sm -fgcse-las -fgcse-after-reload
518 -fcrossjumping -fif-conversion -fif-conversion2
519 -finline-functions -finline-functions-called-once
520 -finline-limit=N -fkeep-inline-functions
521 -fkeep-static-consts -fmerge-constants -fmerge-all-constants
522 -fmodulo-sched -fno-branch-count-reg
523 -fno-default-inline -fno-defer-pop -fmove-loop-invariants
524 -fno-function-cse -fno-guess-branch-probability
525 -fno-inline -fno-math-errno -fno-peephole -fno-peephole2
526 -funsafe-math-optimizations -funsafe-loop-optimizations -ffinite-math-only
527 -fno-toplevel-reorder -fno-trapping-math -fno-zero-initialized-in-bss
528 -fomit-frame-pointer -foptimize-register-move
529 -foptimize-sibling-calls -fprefetch-loop-arrays
530 -fprofile-generate -fprofile-use
531 -fregmove -frename-registers
532 -freorder-blocks -freorder-blocks-and-partition -freorder-functions
533 -frerun-cse-after-loop
534 -frounding-math -frtl-abstract-sequences
535 -fschedule-insns -fschedule-insns2
536 -fno-sched-interblock -fno-sched-spec -fsched-spec-load
537 -fsched-spec-load-dangerous
538 -fsched-stalled-insns=N -fsched-stalled-insns-dep=N
539 -fsched2-use-superblocks
540 -fsched2-use-traces -fsee -freschedule-modulo-scheduled-loops
541 -fsection-anchors -fsignaling-nans -fsingle-precision-constant
542 -fstack-protector -fstack-protector-all
543 -fstrict-aliasing -fstrict-overflow -ftracer -fthread-jumps
544 -funroll-all-loops -funroll-loops -fpeel-loops
545 -fsplit-ivs-in-unroller -funswitch-loops
546 -fvariable-expansion-in-unroller
547 -ftree-pre -ftree-ccp -ftree-dce -ftree-loop-optimize
548 -ftree-loop-linear -ftree-loop-im -ftree-loop-ivcanon -fivopts
549 -ftree-dominator-opts -ftree-dse -ftree-copyrename -ftree-sink
550 -ftree-ch -ftree-sra -ftree-ter -ftree-lrs -ftree-fre -ftree-vectorize
551 -ftree-vect-loop-version -ftree-salias -fipa-pta -fweb
552 -ftree-copy-prop -ftree-store-ccp -ftree-store-copy-prop -fwhole-program
554 -O -O0 -O1 -O2 -O3 -Os
556 _Preprocessor Options_
557 *Note Options Controlling the Preprocessor: Preprocessor Options.
563 -include FILE -imacros FILE
564 -iprefix FILE -iwithprefix DIR
565 -iwithprefixbefore DIR -isystem DIR
566 -imultilib DIR -isysroot DIR
567 -M -MM -MF -MG -MP -MQ -MT -nostdinc
568 -P -fworking-directory -remap
569 -trigraphs -undef -UMACRO -Wp,OPTION
570 -Xpreprocessor OPTION
573 *Note Passing Options to the Assembler: Assembler Options.
574 -Wa,OPTION -Xassembler OPTION
577 *Note Options for Linking: Link Options.
578 OBJECT-FILE-NAME -lLIBRARY
579 -nostartfiles -nodefaultlibs -nostdlib -pie -rdynamic
580 -s -static -static-libgcc -shared -shared-libgcc -symbolic
581 -Wl,OPTION -Xlinker OPTION
585 *Note Options for Directory Search: Directory Options.
586 -BPREFIX -IDIR -iquoteDIR -LDIR
587 -specs=FILE -I- --sysroot=DIR
590 *Note Target Options::.
591 -V VERSION -b MACHINE
593 _Machine Dependent Options_
594 *Note Hardware Models and Configurations: Submodel Options.
598 -mmangle-cpu -mcpu=CPU -mtext=TEXT-SECTION
599 -mdata=DATA-SECTION -mrodata=READONLY-DATA-SECTION
602 -mapcs-frame -mno-apcs-frame
604 -mapcs-stack-check -mno-apcs-stack-check
605 -mapcs-float -mno-apcs-float
606 -mapcs-reentrant -mno-apcs-reentrant
607 -msched-prolog -mno-sched-prolog
608 -mlittle-endian -mbig-endian -mwords-little-endian
609 -mfloat-abi=NAME -msoft-float -mhard-float -mfpe
610 -mthumb-interwork -mno-thumb-interwork
611 -mcpu=NAME -march=NAME -mfpu=NAME
612 -mstructure-size-boundary=N
614 -mlong-calls -mno-long-calls
615 -msingle-pic-base -mno-single-pic-base
618 -mcirrus-fix-invalid-insns -mno-cirrus-fix-invalid-insns
621 -mtpcs-frame -mtpcs-leaf-frame
622 -mcaller-super-interworking -mcallee-super-interworking
627 -mmcu=MCU -msize -minit-stack=N -mno-interrupts
628 -mcall-prologues -mno-tablejump -mtiny-stack -mint8
631 -momit-leaf-frame-pointer -mno-omit-leaf-frame-pointer
632 -mspecld-anomaly -mno-specld-anomaly -mcsync-anomaly -mno-csync-anomaly
633 -mlow-64k -mno-low64k -mid-shared-library
634 -mno-id-shared-library -mshared-library-id=N
635 -mlong-calls -mno-long-calls
638 -mcpu=CPU -march=CPU -mtune=CPU
639 -mmax-stack-frame=N -melinux-stacksize=N
640 -metrax4 -metrax100 -mpdebug -mcc-init -mno-side-effects
641 -mstack-align -mdata-align -mconst-align
642 -m32-bit -m16-bit -m8-bit -mno-prologue-epilogue -mno-gotplt
643 -melf -maout -melinux -mlinux -sim -sim2
644 -mmul-bug-workaround -mno-mul-bug-workaround
650 -all_load -allowable_client -arch -arch_errors_fatal
651 -arch_only -bind_at_load -bundle -bundle_loader
652 -client_name -compatibility_version -current_version
654 -dependency-file -dylib_file -dylinker_install_name
655 -dynamic -dynamiclib -exported_symbols_list
656 -filelist -flat_namespace -force_cpusubtype_ALL
657 -force_flat_namespace -headerpad_max_install_names
658 -image_base -init -install_name -keep_private_externs
659 -multi_module -multiply_defined -multiply_defined_unused
660 -noall_load -no_dead_strip_inits_and_terms
661 -nofixprebinding -nomultidefs -noprebind -noseglinkedit
662 -pagezero_size -prebind -prebind_all_twolevel_modules
663 -private_bundle -read_only_relocs -sectalign
664 -sectobjectsymbols -whyload -seg1addr
665 -sectcreate -sectobjectsymbols -sectorder
666 -segaddr -segs_read_only_addr -segs_read_write_addr
667 -seg_addr_table -seg_addr_table_filename -seglinkedit
668 -segprot -segs_read_only_addr -segs_read_write_addr
669 -single_module -static -sub_library -sub_umbrella
670 -twolevel_namespace -umbrella -undefined
671 -unexported_symbols_list -weak_reference_mismatches
672 -whatsloaded -F -gused -gfull -mmacosx-version-min=VERSION
673 -mkernel -mone-byte-bool
676 -mno-fp-regs -msoft-float -malpha-as -mgas
677 -mieee -mieee-with-inexact -mieee-conformant
678 -mfp-trap-mode=MODE -mfp-rounding-mode=MODE
679 -mtrap-precision=MODE -mbuild-constants
680 -mcpu=CPU-TYPE -mtune=CPU-TYPE
681 -mbwx -mmax -mfix -mcix
682 -mfloat-vax -mfloat-ieee
683 -mexplicit-relocs -msmall-data -mlarge-data
684 -msmall-text -mlarge-text
685 -mmemory-latency=TIME
687 _DEC Alpha/VMS Options_
691 -mgpr-32 -mgpr-64 -mfpr-32 -mfpr-64
692 -mhard-float -msoft-float
693 -malloc-cc -mfixed-cc -mdword -mno-dword
695 -mmedia -mno-media -mmuladd -mno-muladd
696 -mfdpic -minline-plt -mgprel-ro -multilib-library-pic
697 -mlinked-fp -mlong-calls -malign-labels
698 -mlibrary-pic -macc-4 -macc-8
699 -mpack -mno-pack -mno-eflags -mcond-move -mno-cond-move
700 -moptimize-membar -mno-optimize-membar
701 -mscc -mno-scc -mcond-exec -mno-cond-exec
702 -mvliw-branch -mno-vliw-branch
703 -mmulti-cond-exec -mno-multi-cond-exec -mnested-cond-exec
704 -mno-nested-cond-exec -mtomcat-stats
712 -mrelax -mh -ms -mn -mint32 -malign-300
715 -march=ARCHITECTURE-TYPE
716 -mbig-switch -mdisable-fpregs -mdisable-indexing
717 -mfast-indirect-calls -mgas -mgnu-ld -mhp-ld
718 -mfixed-range=REGISTER-RANGE
719 -mjump-in-delay -mlinker-opt -mlong-calls
720 -mlong-load-store -mno-big-switch -mno-disable-fpregs
721 -mno-disable-indexing -mno-fast-indirect-calls -mno-gas
722 -mno-jump-in-delay -mno-long-load-store
723 -mno-portable-runtime -mno-soft-float
724 -mno-space-regs -msoft-float -mpa-risc-1-0
725 -mpa-risc-1-1 -mpa-risc-2-0 -mportable-runtime
726 -mschedule=CPU-TYPE -mspace-regs -msio -mwsio
727 -munix=UNIX-STD -nolibdld -static -threads
729 _i386 and x86-64 Options_
730 -mtune=CPU-TYPE -march=CPU-TYPE
732 -masm=DIALECT -mno-fancy-math-387
733 -mno-fp-ret-in-387 -msoft-float -msvr3-shlib
734 -mno-wide-multiply -mrtd -malign-double
735 -mpreferred-stack-boundary=NUM
736 -mmmx -msse -msse2 -msse3 -m3dnow
737 -mthreads -mno-align-stringops -minline-all-stringops
738 -mpush-args -maccumulate-outgoing-args -m128bit-long-double
739 -m96bit-long-double -mregparm=NUM -msseregparm
741 -momit-leaf-frame-pointer -mno-red-zone -mno-tls-direct-seg-refs
743 -m32 -m64 -mlarge-data-threshold=NUM
746 -mbig-endian -mlittle-endian -mgnu-as -mgnu-ld -mno-pic
747 -mvolatile-asm-stop -mregister-names -mno-sdata
748 -mconstant-gp -mauto-pic -minline-float-divide-min-latency
749 -minline-float-divide-max-throughput
750 -minline-int-divide-min-latency
751 -minline-int-divide-max-throughput
752 -minline-sqrt-min-latency -minline-sqrt-max-throughput
753 -mno-dwarf2-asm -mearly-stop-bits
754 -mfixed-range=REGISTER-RANGE -mtls-size=TLS-SIZE
755 -mtune=CPU-TYPE -mt -pthread -milp32 -mlp64
756 -mno-sched-br-data-spec -msched-ar-data-spec -mno-sched-control-spec
757 -msched-br-in-data-spec -msched-ar-in-data-spec -msched-in-control-spec
758 -msched-ldc -mno-sched-control-ldc -mno-sched-spec-verbose
759 -mno-sched-prefer-non-data-spec-insns
760 -mno-sched-prefer-non-control-spec-insns
761 -mno-sched-count-spec-in-critical-path
766 -malign-loops -mno-align-loops
769 -mmodel=CODE-SIZE-MODEL-TYPE
771 -mno-flush-func -mflush-func=NAME
772 -mno-flush-trap -mflush-trap=NUMBER
776 -mcpu=CPU -msim -memregs=NUMBER
779 -m68000 -m68020 -m68020-40 -m68020-60 -m68030 -m68040
780 -m68060 -mcpu32 -m5200 -mcfv4e -m68881 -mbitfield
782 -mnobitfield -mrtd -mshort -msoft-float -mpcrel
783 -malign-int -mstrict-align -msep-data -mno-sep-data
784 -mshared-library-id=n -mid-shared-library -mno-id-shared-library
787 -m6811 -m6812 -m68hc11 -m68hc12 -m68hcs12
788 -mauto-incdec -minmax -mlong-calls -mshort
789 -msoft-reg-count=COUNT
792 -mhardlit -mno-hardlit -mdiv -mno-div -mrelax-immediates
793 -mno-relax-immediates -mwide-bitfields -mno-wide-bitfields
794 -m4byte-functions -mno-4byte-functions -mcallgraph-data
795 -mno-callgraph-data -mslow-bytes -mno-slow-bytes -mno-lsim
796 -mlittle-endian -mbig-endian -m210 -m340 -mstack-increment
799 -EL -EB -march=ARCH -mtune=ARCH
800 -mips1 -mips2 -mips3 -mips4 -mips32 -mips32r2 -mips64
801 -mips16 -mno-mips16 -mabi=ABI -mabicalls -mno-abicalls
802 -mshared -mno-shared -mxgot -mno-xgot -mgp32 -mgp64
803 -mfp32 -mfp64 -mhard-float -msoft-float
804 -msingle-float -mdouble-float -mdsp -mpaired-single -mips3d
805 -mlong64 -mlong32 -msym32 -mno-sym32
806 -GNUM -membedded-data -mno-embedded-data
807 -muninit-const-in-rodata -mno-uninit-const-in-rodata
808 -msplit-addresses -mno-split-addresses
809 -mexplicit-relocs -mno-explicit-relocs
810 -mcheck-zero-division -mno-check-zero-division
811 -mdivide-traps -mdivide-breaks
812 -mmemcpy -mno-memcpy -mlong-calls -mno-long-calls
813 -mmad -mno-mad -mfused-madd -mno-fused-madd -nocpp
814 -mfix-r4000 -mno-fix-r4000 -mfix-r4400 -mno-fix-r4400
815 -mfix-vr4120 -mno-fix-vr4120 -mfix-vr4130
816 -mfix-sb1 -mno-fix-sb1
817 -mflush-func=FUNC -mno-flush-func
818 -mbranch-likely -mno-branch-likely
819 -mfp-exceptions -mno-fp-exceptions
820 -mvr4130-align -mno-vr4130-align
823 -mlibfuncs -mno-libfuncs -mepsilon -mno-epsilon -mabi=gnu
824 -mabi=mmixware -mzero-extend -mknuthdiv -mtoplevel-symbols
825 -melf -mbranch-predict -mno-branch-predict -mbase-addresses
826 -mno-base-addresses -msingle-exit -mno-single-exit
829 -mmult-bug -mno-mult-bug
832 -mreturn-pointer-on-d0
836 -mno-crt0 -mbacc -msim
840 -mfpu -msoft-float -mac0 -mno-ac0 -m40 -m45 -m10
841 -mbcopy -mbcopy-builtin -mint32 -mno-int16
842 -mint16 -mno-int32 -mfloat32 -mno-float64
843 -mfloat64 -mno-float32 -mabshi -mno-abshi
844 -mbranch-expensive -mbranch-cheap
845 -msplit -mno-split -munix-asm -mdec-asm
847 _PowerPC Options_ See RS/6000 and PowerPC Options.
849 _RS/6000 and PowerPC Options_
852 -mpower -mno-power -mpower2 -mno-power2
853 -mpowerpc -mpowerpc64 -mno-powerpc
854 -maltivec -mno-altivec
855 -mpowerpc-gpopt -mno-powerpc-gpopt
856 -mpowerpc-gfxopt -mno-powerpc-gfxopt
857 -mmfcrf -mno-mfcrf -mpopcntb -mno-popcntb -mfprnd -mno-fprnd
858 -mnew-mnemonics -mold-mnemonics
859 -mfull-toc -mminimal-toc -mno-fp-in-toc -mno-sum-in-toc
860 -m64 -m32 -mxl-compat -mno-xl-compat -mpe
861 -malign-power -malign-natural
862 -msoft-float -mhard-float -mmultiple -mno-multiple
863 -mstring -mno-string -mupdate -mno-update
864 -mfused-madd -mno-fused-madd -mbit-align -mno-bit-align
865 -mstrict-align -mno-strict-align -mrelocatable
866 -mno-relocatable -mrelocatable-lib -mno-relocatable-lib
867 -mtoc -mno-toc -mlittle -mlittle-endian -mbig -mbig-endian
868 -mdynamic-no-pic -maltivec -mswdiv
869 -mprioritize-restricted-insns=PRIORITY
870 -msched-costly-dep=DEPENDENCE_TYPE
871 -minsert-sched-nops=SCHEME
872 -mcall-sysv -mcall-netbsd
873 -maix-struct-return -msvr4-struct-return
874 -mabi=ABI-TYPE -msecure-plt -mbss-plt
882 -mfloat-gprs=yes -mfloat-gprs=no -mfloat-gprs=single -mfloat-gprs=double
883 -mprototype -mno-prototype
884 -msim -mmvme -mads -myellowknife -memb -msdata
885 -msdata=OPT -mvxworks -mwindiss -G NUM -pthread
887 _S/390 and zSeries Options_
888 -mtune=CPU-TYPE -march=CPU-TYPE
889 -mhard-float -msoft-float -mlong-double-64 -mlong-double-128
890 -mbackchain -mno-backchain -mpacked-stack -mno-packed-stack
891 -msmall-exec -mno-small-exec -mmvcle -mno-mvcle
892 -m64 -m31 -mdebug -mno-debug -mesa -mzarch
893 -mtpf-trace -mno-tpf-trace -mfused-madd -mno-fused-madd
894 -mwarn-framesize -mwarn-dynamicstack -mstack-size -mstack-guard
901 -mscore5 -mscore5u -mscore7 -mscore7d
904 -m1 -m2 -m2e -m3 -m3e
905 -m4-nofpu -m4-single-only -m4-single -m4
906 -m4a-nofpu -m4a-single-only -m4a-single -m4a -m4al
907 -m5-64media -m5-64media-nofpu
908 -m5-32media -m5-32media-nofpu
909 -m5-compact -m5-compact-nofpu
910 -mb -ml -mdalign -mrelax
911 -mbigtable -mfmovd -mhitachi -mrenesas -mno-renesas -mnomacsave
912 -mieee -misize -mpadstruct -mspace
913 -mprefergot -musermode -multcost=NUMBER -mdiv=STRATEGY
914 -mdivsi3_libfunc=NAME
915 -madjust-unroll -mindexed-addressing -mgettrcost=NUMBER -mpt-fixed
922 -m32 -m64 -mapp-regs -mno-app-regs
923 -mfaster-structs -mno-faster-structs
924 -mfpu -mno-fpu -mhard-float -msoft-float
925 -mhard-quad-float -msoft-quad-float
926 -mimpure-text -mno-impure-text -mlittle-endian
927 -mstack-bias -mno-stack-bias
928 -munaligned-doubles -mno-unaligned-doubles
929 -mv8plus -mno-v8plus -mvis -mno-vis
930 -threads -pthreads -pthread
933 -Qy -Qn -YP,PATHS -Ym,DIR
935 _TMS320C3x/C4x Options_
936 -mcpu=CPU -mbig -msmall -mregparm -mmemparm
937 -mfast-fix -mmpyi -mbk -mti -mdp-isr-reload
938 -mrpts=COUNT -mrptb -mdb -mloop-unsigned
939 -mparallel-insns -mparallel-mpy -mpreserve-float
942 -mlong-calls -mno-long-calls -mep -mno-ep
943 -mprolog-function -mno-prolog-function -mspace
944 -mtda=N -msda=N -mzda=N
945 -mapp-regs -mno-app-regs
946 -mdisable-callt -mno-disable-callt
954 _x86-64 Options_ See i386 and x86-64 Options.
960 -mconst16 -mno-const16
961 -mfused-madd -mno-fused-madd
962 -mtext-section-literals -mno-text-section-literals
963 -mtarget-align -mno-target-align
964 -mlongcalls -mno-longcalls
966 _zSeries Options_ See S/390 and zSeries Options.
968 _Code Generation Options_
969 *Note Options for Code Generation Conventions: Code Gen Options.
970 -fcall-saved-REG -fcall-used-REG
971 -ffixed-REG -fexceptions
972 -fnon-call-exceptions -funwind-tables
973 -fasynchronous-unwind-tables
974 -finhibit-size-directive -finstrument-functions
975 -finstrument-functions-exclude-function-list=SYM,SYM,...
976 -finstrument-functions-exclude-file-list=FILE,FILE,...
977 -fno-common -fno-ident
978 -fpcc-struct-return -fpic -fPIC -fpie -fPIE
980 -freg-struct-return -fshort-enums
981 -fshort-double -fshort-wchar
982 -fverbose-asm -fpack-struct[=N] -fstack-check
983 -fstack-limit-register=REG -fstack-limit-symbol=SYM
984 -fargument-alias -fargument-noalias
985 -fargument-noalias-global -fargument-noalias-anything
986 -fleading-underscore -ftls-model=MODEL
987 -ftrapv -fwrapv -fbounds-check
993 * Overall Options:: Controlling the kind of output:
994 an executable, object files, assembler files,
995 or preprocessed source.
996 * C Dialect Options:: Controlling the variant of C language compiled.
997 * C++ Dialect Options:: Variations on C++.
998 * Objective-C and Objective-C++ Dialect Options:: Variations on Objective-C
1000 * Language Independent Options:: Controlling how diagnostics should be
1002 * Warning Options:: How picky should the compiler be?
1003 * Debugging Options:: Symbol tables, measurements, and debugging dumps.
1004 * Optimize Options:: How much optimization?
1005 * Preprocessor Options:: Controlling header files and macro definitions.
1006 Also, getting dependency information for Make.
1007 * Assembler Options:: Passing options to the assembler.
1008 * Link Options:: Specifying libraries and so on.
1009 * Directory Options:: Where to find header files and libraries.
1010 Where to find the compiler executable files.
1011 * Spec Files:: How to pass switches to sub-processes.
1012 * Target Options:: Running a cross-compiler, or an old version of GCC.
1015 File: gcc.info, Node: Overall Options, Next: Invoking G++, Prev: Option Summary, Up: Invoking GCC
1017 3.2 Options Controlling the Kind of Output
1018 ==========================================
1020 Compilation can involve up to four stages: preprocessing, compilation
1021 proper, assembly and linking, always in that order. GCC is capable of
1022 preprocessing and compiling several files either into several assembler
1023 input files, or into one assembler input file; then each assembler
1024 input file produces an object file, and linking combines all the object
1025 files (those newly compiled, and those specified as input) into an
1028 For any given input file, the file name suffix determines what kind of
1029 compilation is done:
1032 C source code which must be preprocessed.
1035 C source code which should not be preprocessed.
1038 C++ source code which should not be preprocessed.
1041 Objective-C source code. Note that you must link with the
1042 `libobjc' library to make an Objective-C program work.
1045 Objective-C source code which should not be preprocessed.
1049 Objective-C++ source code. Note that you must link with the
1050 `libobjc' library to make an Objective-C++ program work. Note
1051 that `.M' refers to a literal capital M.
1054 Objective-C++ source code which should not be preprocessed.
1057 C, C++, Objective-C or Objective-C++ header file to be turned into
1058 a precompiled header.
1067 C++ source code which must be preprocessed. Note that in `.cxx',
1068 the last two letters must both be literally `x'. Likewise, `.C'
1069 refers to a literal capital C.
1073 Objective-C++ source code which must be preprocessed.
1076 Objective-C++ source code which should not be preprocessed.
1080 C++ header file to be turned into a precompiled header.
1085 Fixed form Fortran source code which should not be preprocessed.
1090 Fixed form Fortran source code which must be preprocessed (with
1091 the traditional preprocessor).
1095 Free form Fortran source code which should not be preprocessed.
1099 Free form Fortran source code which must be preprocessed (with the
1100 traditional preprocessor).
1103 Ada source code file which contains a library unit declaration (a
1104 declaration of a package, subprogram, or generic, or a generic
1105 instantiation), or a library unit renaming declaration (a package,
1106 generic, or subprogram renaming declaration). Such files are also
1110 Ada source code file containing a library unit body (a subprogram
1111 or package body). Such files are also called "bodies".
1117 Assembler code which must be preprocessed.
1120 An object file to be fed straight into linking. Any file name
1121 with no recognized suffix is treated this way.
1123 You can specify the input language explicitly with the `-x' option:
1126 Specify explicitly the LANGUAGE for the following input files
1127 (rather than letting the compiler choose a default based on the
1128 file name suffix). This option applies to all following input
1129 files until the next `-x' option. Possible values for LANGUAGE
1131 c c-header c-cpp-output
1132 c++ c++-header c++-cpp-output
1133 objective-c objective-c-header objective-c-cpp-output
1134 objective-c++ objective-c++-header objective-c++-cpp-output
1135 assembler assembler-with-cpp
1142 Turn off any specification of a language, so that subsequent files
1143 are handled according to their file name suffixes (as they are if
1144 `-x' has not been used at all).
1147 Normally the `gcc' program will exit with the code of 1 if any
1148 phase of the compiler returns a non-success return code. If you
1149 specify `-pass-exit-codes', the `gcc' program will instead return
1150 with numerically highest error produced by any phase that returned
1151 an error indication. The C, C++, and Fortran frontends return 4,
1152 if an internal compiler error is encountered.
1154 If you only want some of the stages of compilation, you can use `-x'
1155 (or filename suffixes) to tell `gcc' where to start, and one of the
1156 options `-c', `-S', or `-E' to say where `gcc' is to stop. Note that
1157 some combinations (for example, `-x cpp-output -E') instruct `gcc' to
1161 Compile or assemble the source files, but do not link. The linking
1162 stage simply is not done. The ultimate output is in the form of an
1163 object file for each source file.
1165 By default, the object file name for a source file is made by
1166 replacing the suffix `.c', `.i', `.s', etc., with `.o'.
1168 Unrecognized input files, not requiring compilation or assembly,
1172 Stop after the stage of compilation proper; do not assemble. The
1173 output is in the form of an assembler code file for each
1174 non-assembler input file specified.
1176 By default, the assembler file name for a source file is made by
1177 replacing the suffix `.c', `.i', etc., with `.s'.
1179 Input files that don't require compilation are ignored.
1182 Stop after the preprocessing stage; do not run the compiler
1183 proper. The output is in the form of preprocessed source code,
1184 which is sent to the standard output.
1186 Input files which don't require preprocessing are ignored.
1189 Place output in file FILE. This applies regardless to whatever
1190 sort of output is being produced, whether it be an executable file,
1191 an object file, an assembler file or preprocessed C code.
1193 If `-o' is not specified, the default is to put an executable file
1194 in `a.out', the object file for `SOURCE.SUFFIX' in `SOURCE.o', its
1195 assembler file in `SOURCE.s', a precompiled header file in
1196 `SOURCE.SUFFIX.gch', and all preprocessed C source on standard
1200 Print (on standard error output) the commands executed to run the
1201 stages of compilation. Also print the version number of the
1202 compiler driver program and of the preprocessor and the compiler
1206 Like `-v' except the commands are not executed and all command
1207 arguments are quoted. This is useful for shell scripts to capture
1208 the driver-generated command lines.
1211 Use pipes rather than temporary files for communication between the
1212 various stages of compilation. This fails to work on some systems
1213 where the assembler is unable to read from a pipe; but the GNU
1214 assembler has no trouble.
1217 If you are compiling multiple source files, this option tells the
1218 driver to pass all the source files to the compiler at once (for
1219 those languages for which the compiler can handle this). This
1220 will allow intermodule analysis (IMA) to be performed by the
1221 compiler. Currently the only language for which this is supported
1222 is C. If you pass source files for multiple languages to the
1223 driver, using this option, the driver will invoke the compiler(s)
1224 that support IMA once each, passing each compiler all the source
1225 files appropriate for it. For those languages that do not support
1226 IMA this option will be ignored, and the compiler will be invoked
1227 once for each source file in that language. If you use this
1228 option in conjunction with `-save-temps', the compiler will
1229 generate multiple pre-processed files (one for each source file),
1230 but only one (combined) `.o' or `.s' file.
1233 Print (on the standard output) a description of the command line
1234 options understood by `gcc'. If the `-v' option is also specified
1235 then `--help' will also be passed on to the various processes
1236 invoked by `gcc', so that they can display the command line options
1237 they accept. If the `-Wextra' option is also specified then
1238 command line options which have no documentation associated with
1239 them will also be displayed.
1242 Print (on the standard output) a description of target specific
1243 command line options for each tool.
1246 Display the version number and copyrights of the invoked GCC.
1249 Read command-line options from FILE. The options read are
1250 inserted in place of the original @FILE option. If FILE does not
1251 exist, or cannot be read, then the option will be treated
1252 literally, and not removed.
1254 Options in FILE are separated by whitespace. A whitespace
1255 character may be included in an option by surrounding the entire
1256 option in either single or double quotes. Any character
1257 (including a backslash) may be included by prefixing the character
1258 to be included with a backslash. The FILE may itself contain
1259 additional @FILE options; any such options will be processed
1263 File: gcc.info, Node: Invoking G++, Next: C Dialect Options, Prev: Overall Options, Up: Invoking GCC
1265 3.3 Compiling C++ Programs
1266 ==========================
1268 C++ source files conventionally use one of the suffixes `.C', `.cc',
1269 `.cpp', `.CPP', `.c++', `.cp', or `.cxx'; C++ header files often use
1270 `.hh' or `.H'; and preprocessed C++ files use the suffix `.ii'. GCC
1271 recognizes files with these names and compiles them as C++ programs
1272 even if you call the compiler the same way as for compiling C programs
1273 (usually with the name `gcc').
1275 However, the use of `gcc' does not add the C++ library. `g++' is a
1276 program that calls GCC and treats `.c', `.h' and `.i' files as C++
1277 source files instead of C source files unless `-x' is used, and
1278 automatically specifies linking against the C++ library. This program
1279 is also useful when precompiling a C header file with a `.h' extension
1280 for use in C++ compilations. On many systems, `g++' is also installed
1281 with the name `c++'.
1283 When you compile C++ programs, you may specify many of the same
1284 command-line options that you use for compiling programs in any
1285 language; or command-line options meaningful for C and related
1286 languages; or options that are meaningful only for C++ programs. *Note
1287 Options Controlling C Dialect: C Dialect Options, for explanations of
1288 options for languages related to C. *Note Options Controlling C++
1289 Dialect: C++ Dialect Options, for explanations of options that are
1290 meaningful only for C++ programs.
1293 File: gcc.info, Node: C Dialect Options, Next: C++ Dialect Options, Prev: Invoking G++, Up: Invoking GCC
1295 3.4 Options Controlling C Dialect
1296 =================================
1298 The following options control the dialect of C (or languages derived
1299 from C, such as C++, Objective-C and Objective-C++) that the compiler
1303 In C mode, support all ISO C90 programs. In C++ mode, remove GNU
1304 extensions that conflict with ISO C++.
1306 This turns off certain features of GCC that are incompatible with
1307 ISO C90 (when compiling C code), or of standard C++ (when
1308 compiling C++ code), such as the `asm' and `typeof' keywords, and
1309 predefined macros such as `unix' and `vax' that identify the type
1310 of system you are using. It also enables the undesirable and
1311 rarely used ISO trigraph feature. For the C compiler, it disables
1312 recognition of C++ style `//' comments as well as the `inline'
1315 The alternate keywords `__asm__', `__extension__', `__inline__'
1316 and `__typeof__' continue to work despite `-ansi'. You would not
1317 want to use them in an ISO C program, of course, but it is useful
1318 to put them in header files that might be included in compilations
1319 done with `-ansi'. Alternate predefined macros such as `__unix__'
1320 and `__vax__' are also available, with or without `-ansi'.
1322 The `-ansi' option does not cause non-ISO programs to be rejected
1323 gratuitously. For that, `-pedantic' is required in addition to
1324 `-ansi'. *Note Warning Options::.
1326 The macro `__STRICT_ANSI__' is predefined when the `-ansi' option
1327 is used. Some header files may notice this macro and refrain from
1328 declaring certain functions or defining certain macros that the
1329 ISO standard doesn't call for; this is to avoid interfering with
1330 any programs that might use these names for other things.
1332 Functions which would normally be built in but do not have
1333 semantics defined by ISO C (such as `alloca' and `ffs') are not
1334 built-in functions with `-ansi' is used. *Note Other built-in
1335 functions provided by GCC: Other Builtins, for details of the
1339 Determine the language standard. This option is currently only
1340 supported when compiling C or C++. A value for this option must be
1341 provided; possible values are
1345 ISO C90 (same as `-ansi').
1348 ISO C90 as modified in amendment 1.
1354 ISO C99. Note that this standard is not yet fully supported;
1355 see `http://gcc.gnu.org/gcc-4.2/c99status.html' for more
1356 information. The names `c9x' and `iso9899:199x' are
1360 Default, ISO C90 plus GNU extensions (including some C99
1365 ISO C99 plus GNU extensions. When ISO C99 is fully
1366 implemented in GCC, this will become the default. The name
1367 `gnu9x' is deprecated.
1370 The 1998 ISO C++ standard plus amendments.
1373 The same as `-std=c++98' plus GNU extensions. This is the
1374 default for C++ code.
1376 Even when this option is not specified, you can still use some of
1377 the features of newer standards in so far as they do not conflict
1378 with previous C standards. For example, you may use
1379 `__restrict__' even when `-std=c99' is not specified.
1381 The `-std' options specifying some version of ISO C have the same
1382 effects as `-ansi', except that features that were not in ISO C90
1383 but are in the specified version (for example, `//' comments and
1384 the `inline' keyword in ISO C99) are not disabled.
1386 *Note Language Standards Supported by GCC: Standards, for details
1387 of these standard versions.
1390 The option `-fgnu89-inline' tells GCC to use the traditional GNU
1391 semantics for `inline' functions when in C99 mode. *Note An
1392 Inline Function is As Fast As a Macro: Inline. Using this option
1393 is roughly equivalent to adding the `gnu_inline' function
1394 attribute to all inline functions (*note Function Attributes::).
1396 This option is accepted by GCC versions 4.1.3 and up. In GCC
1397 versions prior to 4.3, C99 inline semantics are not supported, and
1398 thus this option is effectively assumed to be present regardless
1399 of whether or not it is specified; the only effect of specifying
1400 it explicitly is to disable warnings about using inline functions
1401 in C99 mode. Likewise, the option `-fno-gnu89-inline' is not
1402 supported in versions of GCC before 4.3. It will be supported
1403 only in C99 or gnu99 mode, not in C89 or gnu89 mode.
1405 The preprocesor macros `__GNUC_GNU_INLINE__' and
1406 `__GNUC_STDC_INLINE__' may be used to check which semantics are in
1407 effect for `inline' functions. *Note Common Predefined Macros:
1408 (cpp)Common Predefined Macros.
1410 `-aux-info FILENAME'
1411 Output to the given filename prototyped declarations for all
1412 functions declared and/or defined in a translation unit, including
1413 those in header files. This option is silently ignored in any
1414 language other than C.
1416 Besides declarations, the file indicates, in comments, the origin
1417 of each declaration (source file and line), whether the
1418 declaration was implicit, prototyped or unprototyped (`I', `N' for
1419 new or `O' for old, respectively, in the first character after the
1420 line number and the colon), and whether it came from a declaration
1421 or a definition (`C' or `F', respectively, in the following
1422 character). In the case of function definitions, a K&R-style list
1423 of arguments followed by their declarations is also provided,
1424 inside comments, after the declaration.
1427 Do not recognize `asm', `inline' or `typeof' as a keyword, so that
1428 code can use these words as identifiers. You can use the keywords
1429 `__asm__', `__inline__' and `__typeof__' instead. `-ansi' implies
1432 In C++, this switch only affects the `typeof' keyword, since `asm'
1433 and `inline' are standard keywords. You may want to use the
1434 `-fno-gnu-keywords' flag instead, which has the same effect. In
1435 C99 mode (`-std=c99' or `-std=gnu99'), this switch only affects
1436 the `asm' and `typeof' keywords, since `inline' is a standard
1440 `-fno-builtin-FUNCTION'
1441 Don't recognize built-in functions that do not begin with
1442 `__builtin_' as prefix. *Note Other built-in functions provided
1443 by GCC: Other Builtins, for details of the functions affected,
1444 including those which are not built-in functions when `-ansi' or
1445 `-std' options for strict ISO C conformance are used because they
1446 do not have an ISO standard meaning.
1448 GCC normally generates special code to handle certain built-in
1449 functions more efficiently; for instance, calls to `alloca' may
1450 become single instructions that adjust the stack directly, and
1451 calls to `memcpy' may become inline copy loops. The resulting
1452 code is often both smaller and faster, but since the function
1453 calls no longer appear as such, you cannot set a breakpoint on
1454 those calls, nor can you change the behavior of the functions by
1455 linking with a different library. In addition, when a function is
1456 recognized as a built-in function, GCC may use information about
1457 that function to warn about problems with calls to that function,
1458 or to generate more efficient code, even if the resulting code
1459 still contains calls to that function. For example, warnings are
1460 given with `-Wformat' for bad calls to `printf', when `printf' is
1461 built in, and `strlen' is known not to modify global memory.
1463 With the `-fno-builtin-FUNCTION' option only the built-in function
1464 FUNCTION is disabled. FUNCTION must not begin with `__builtin_'.
1465 If a function is named this is not built-in in this version of
1466 GCC, this option is ignored. There is no corresponding
1467 `-fbuiltin-FUNCTION' option; if you wish to enable built-in
1468 functions selectively when using `-fno-builtin' or
1469 `-ffreestanding', you may define macros such as:
1471 #define abs(n) __builtin_abs ((n))
1472 #define strcpy(d, s) __builtin_strcpy ((d), (s))
1475 Assert that compilation takes place in a hosted environment. This
1476 implies `-fbuiltin'. A hosted environment is one in which the
1477 entire standard library is available, and in which `main' has a
1478 return type of `int'. Examples are nearly everything except a
1479 kernel. This is equivalent to `-fno-freestanding'.
1482 Assert that compilation takes place in a freestanding environment.
1483 This implies `-fno-builtin'. A freestanding environment is one
1484 in which the standard library may not exist, and program startup
1485 may not necessarily be at `main'. The most obvious example is an
1486 OS kernel. This is equivalent to `-fno-hosted'.
1488 *Note Language Standards Supported by GCC: Standards, for details
1489 of freestanding and hosted environments.
1492 Enable handling of OpenMP directives `#pragma omp' in C/C++ and
1493 `!$omp' in Fortran. When `-fopenmp' is specified, the compiler
1494 generates parallel code according to the OpenMP Application
1495 Program Interface v2.5 `http://www.openmp.org/'.
1498 Accept some non-standard constructs used in Microsoft header files.
1500 Some cases of unnamed fields in structures and unions are only
1501 accepted with this option. *Note Unnamed struct/union fields
1502 within structs/unions: Unnamed Fields, for details.
1505 Support ISO C trigraphs. The `-ansi' option (and `-std' options
1506 for strict ISO C conformance) implies `-trigraphs'.
1508 `-no-integrated-cpp'
1509 Performs a compilation in two passes: preprocessing and compiling.
1510 This option allows a user supplied "cc1", "cc1plus", or "cc1obj"
1511 via the `-B' option. The user supplied compilation step can then
1512 add in an additional preprocessing step after normal preprocessing
1513 but before compiling. The default is to use the integrated cpp
1516 The semantics of this option will change if "cc1", "cc1plus", and
1517 "cc1obj" are merged.
1521 Formerly, these options caused GCC to attempt to emulate a
1522 pre-standard C compiler. They are now only supported with the
1523 `-E' switch. The preprocessor continues to support a pre-standard
1524 mode. See the GNU CPP manual for details.
1527 Allow conditional expressions with mismatched types in the second
1528 and third arguments. The value of such an expression is void.
1529 This option is not supported for C++.
1532 Let the type `char' be unsigned, like `unsigned char'.
1534 Each kind of machine has a default for what `char' should be. It
1535 is either like `unsigned char' by default or like `signed char' by
1538 Ideally, a portable program should always use `signed char' or
1539 `unsigned char' when it depends on the signedness of an object.
1540 But many programs have been written to use plain `char' and expect
1541 it to be signed, or expect it to be unsigned, depending on the
1542 machines they were written for. This option, and its inverse, let
1543 you make such a program work with the opposite default.
1545 The type `char' is always a distinct type from each of `signed
1546 char' or `unsigned char', even though its behavior is always just
1547 like one of those two.
1550 Let the type `char' be signed, like `signed char'.
1552 Note that this is equivalent to `-fno-unsigned-char', which is the
1553 negative form of `-funsigned-char'. Likewise, the option
1554 `-fno-signed-char' is equivalent to `-funsigned-char'.
1556 `-fsigned-bitfields'
1557 `-funsigned-bitfields'
1558 `-fno-signed-bitfields'
1559 `-fno-unsigned-bitfields'
1560 These options control whether a bit-field is signed or unsigned,
1561 when the declaration does not use either `signed' or `unsigned'.
1562 By default, such a bit-field is signed, because this is
1563 consistent: the basic integer types such as `int' are signed types.
1566 File: gcc.info, Node: C++ Dialect Options, Next: Objective-C and Objective-C++ Dialect Options, Prev: C Dialect Options, Up: Invoking GCC
1568 3.5 Options Controlling C++ Dialect
1569 ===================================
1571 This section describes the command-line options that are only meaningful
1572 for C++ programs; but you can also use most of the GNU compiler options
1573 regardless of what language your program is in. For example, you might
1574 compile a file `firstClass.C' like this:
1576 g++ -g -frepo -O -c firstClass.C
1578 In this example, only `-frepo' is an option meant only for C++
1579 programs; you can use the other options with any language supported by
1582 Here is a list of options that are _only_ for compiling C++ programs:
1585 Use version N of the C++ ABI. Version 2 is the version of the C++
1586 ABI that first appeared in G++ 3.4. Version 1 is the version of
1587 the C++ ABI that first appeared in G++ 3.2. Version 0 will always
1588 be the version that conforms most closely to the C++ ABI
1589 specification. Therefore, the ABI obtained using version 0 will
1590 change as ABI bugs are fixed.
1592 The default is version 2.
1594 `-fno-access-control'
1595 Turn off all access checking. This switch is mainly useful for
1596 working around bugs in the access control code.
1599 Check that the pointer returned by `operator new' is non-null
1600 before attempting to modify the storage allocated. This check is
1601 normally unnecessary because the C++ standard specifies that
1602 `operator new' will only return `0' if it is declared `throw()',
1603 in which case the compiler will always check the return value even
1604 without this option. In all other cases, when `operator new' has
1605 a non-empty exception specification, memory exhaustion is
1606 signalled by throwing `std::bad_alloc'. See also `new (nothrow)'.
1609 Put uninitialized or runtime-initialized global variables into the
1610 common segment, as C does. This saves space in the executable at
1611 the cost of not diagnosing duplicate definitions. If you compile
1612 with this flag and your program mysteriously crashes after
1613 `main()' has completed, you may have an object that is being
1614 destroyed twice because two definitions were merged.
1616 This option is no longer useful on most targets, now that support
1617 has been added for putting variables into BSS without making them
1620 `-ffriend-injection'
1621 Inject friend functions into the enclosing namespace, so that they
1622 are visible outside the scope of the class in which they are
1623 declared. Friend functions were documented to work this way in
1624 the old Annotated C++ Reference Manual, and versions of G++ before
1625 4.1 always worked that way. However, in ISO C++ a friend function
1626 which is not declared in an enclosing scope can only be found
1627 using argument dependent lookup. This option causes friends to be
1628 injected as they were in earlier releases.
1630 This option is for compatibility, and may be removed in a future
1633 `-fno-elide-constructors'
1634 The C++ standard allows an implementation to omit creating a
1635 temporary which is only used to initialize another object of the
1636 same type. Specifying this option disables that optimization, and
1637 forces G++ to call the copy constructor in all cases.
1639 `-fno-enforce-eh-specs'
1640 Don't generate code to check for violation of exception
1641 specifications at runtime. This option violates the C++ standard,
1642 but may be useful for reducing code size in production builds,
1643 much like defining `NDEBUG'. This does not give user code
1644 permission to throw exceptions in violation of the exception
1645 specifications; the compiler will still optimize based on the
1646 specifications, so throwing an unexpected exception will result in
1651 If `-ffor-scope' is specified, the scope of variables declared in
1652 a for-init-statement is limited to the `for' loop itself, as
1653 specified by the C++ standard. If `-fno-for-scope' is specified,
1654 the scope of variables declared in a for-init-statement extends to
1655 the end of the enclosing scope, as was the case in old versions of
1656 G++, and other (traditional) implementations of C++.
1658 The default if neither flag is given to follow the standard, but
1659 to allow and give a warning for old-style code that would
1660 otherwise be invalid, or have different behavior.
1663 Do not recognize `typeof' as a keyword, so that code can use this
1664 word as an identifier. You can use the keyword `__typeof__'
1665 instead. `-ansi' implies `-fno-gnu-keywords'.
1667 `-fno-implicit-templates'
1668 Never emit code for non-inline templates which are instantiated
1669 implicitly (i.e. by use); only emit code for explicit
1670 instantiations. *Note Template Instantiation::, for more
1673 `-fno-implicit-inline-templates'
1674 Don't emit code for implicit instantiations of inline templates,
1675 either. The default is to handle inlines differently so that
1676 compiles with and without optimization will need the same set of
1677 explicit instantiations.
1679 `-fno-implement-inlines'
1680 To save space, do not emit out-of-line copies of inline functions
1681 controlled by `#pragma implementation'. This will cause linker
1682 errors if these functions are not inlined everywhere they are
1686 Disable pedantic warnings about constructs used in MFC, such as
1687 implicit int and getting a pointer to member function via
1688 non-standard syntax.
1690 `-fno-nonansi-builtins'
1691 Disable built-in declarations of functions that are not mandated by
1692 ANSI/ISO C. These include `ffs', `alloca', `_exit', `index',
1693 `bzero', `conjf', and other related functions.
1695 `-fno-operator-names'
1696 Do not treat the operator name keywords `and', `bitand', `bitor',
1697 `compl', `not', `or' and `xor' as synonyms as keywords.
1699 `-fno-optional-diags'
1700 Disable diagnostics that the standard says a compiler does not
1701 need to issue. Currently, the only such diagnostic issued by G++
1702 is the one for a name having multiple meanings within a class.
1705 Downgrade some diagnostics about nonconformant code from errors to
1706 warnings. Thus, using `-fpermissive' will allow some
1707 nonconforming code to compile.
1710 Enable automatic template instantiation at link time. This option
1711 also implies `-fno-implicit-templates'. *Note Template
1712 Instantiation::, for more information.
1715 Disable generation of information about every class with virtual
1716 functions for use by the C++ runtime type identification features
1717 (`dynamic_cast' and `typeid'). If you don't use those parts of
1718 the language, you can save some space by using this flag. Note
1719 that exception handling uses the same information, but it will
1720 generate it as needed. The `dynamic_cast' operator can still be
1721 used for casts that do not require runtime type information, i.e.
1722 casts to `void *' or to unambiguous base classes.
1725 Emit statistics about front-end processing at the end of the
1726 compilation. This information is generally only useful to the G++
1729 `-ftemplate-depth-N'
1730 Set the maximum instantiation depth for template classes to N. A
1731 limit on the template instantiation depth is needed to detect
1732 endless recursions during template class instantiation. ANSI/ISO
1733 C++ conforming programs must not rely on a maximum depth greater
1736 `-fno-threadsafe-statics'
1737 Do not emit the extra code to use the routines specified in the C++
1738 ABI for thread-safe initialization of local statics. You can use
1739 this option to reduce code size slightly in code that doesn't need
1743 Register destructors for objects with static storage duration with
1744 the `__cxa_atexit' function rather than the `atexit' function.
1745 This option is required for fully standards-compliant handling of
1746 static destructors, but will only work if your C library supports
1749 `-fno-use-cxa-get-exception-ptr'
1750 Don't use the `__cxa_get_exception_ptr' runtime routine. This
1751 will cause `std::uncaught_exception' to be incorrect, but is
1752 necessary if the runtime routine is not available.
1754 `-fvisibility-inlines-hidden'
1755 This switch declares that the user does not attempt to compare
1756 pointers to inline methods where the addresses of the two functions
1757 were taken in different shared objects.
1759 The effect of this is that GCC may, effectively, mark inline
1760 methods with `__attribute__ ((visibility ("hidden")))' so that
1761 they do not appear in the export table of a DSO and do not require
1762 a PLT indirection when used within the DSO. Enabling this option
1763 can have a dramatic effect on load and link times of a DSO as it
1764 massively reduces the size of the dynamic export table when the
1765 library makes heavy use of templates.
1767 The behaviour of this switch is not quite the same as marking the
1768 methods as hidden directly, because it does not affect static
1769 variables local to the function or cause the compiler to deduce
1770 that the function is defined in only one shared object.
1772 You may mark a method as having a visibility explicitly to negate
1773 the effect of the switch for that method. For example, if you do
1774 want to compare pointers to a particular inline method, you might
1775 mark it as having default visibility. Marking the enclosing class
1776 with explicit visibility will have no effect.
1778 Explicitly instantiated inline methods are unaffected by this
1779 option as their linkage might otherwise cross a shared library
1780 boundary. *Note Template Instantiation::.
1783 Do not use weak symbol support, even if it is provided by the
1784 linker. By default, G++ will use weak symbols if they are
1785 available. This option exists only for testing, and should not be
1786 used by end-users; it will result in inferior code and has no
1787 benefits. This option may be removed in a future release of G++.
1790 Do not search for header files in the standard directories
1791 specific to C++, but do still search the other standard
1792 directories. (This option is used when building the C++ library.)
1794 In addition, these optimization, warning, and code generation options
1795 have meanings only for C++ programs:
1797 `-fno-default-inline'
1798 Do not assume `inline' for functions defined inside a class scope.
1799 *Note Options That Control Optimization: Optimize Options. Note
1800 that these functions will have linkage like inline functions; they
1801 just won't be inlined by default.
1804 Warn when G++ generates code that is probably not compatible with
1805 the vendor-neutral C++ ABI. Although an effort has been made to
1806 warn about all such cases, there are probably some cases that are
1807 not warned about, even though G++ is generating incompatible code.
1808 There may also be cases where warnings are emitted even though
1809 the code that is generated will be compatible.
1811 You should rewrite your code to avoid these warnings if you are
1812 concerned about the fact that code generated by G++ may not be
1813 binary compatible with code generated by other compilers.
1815 The known incompatibilities at this point include:
1817 * Incorrect handling of tail-padding for bit-fields. G++ may
1818 attempt to pack data into the same byte as a base class. For
1821 struct A { virtual void f(); int f1 : 1; };
1822 struct B : public A { int f2 : 1; };
1824 In this case, G++ will place `B::f2' into the same byte
1825 as`A::f1'; other compilers will not. You can avoid this
1826 problem by explicitly padding `A' so that its size is a
1827 multiple of the byte size on your platform; that will cause
1828 G++ and other compilers to layout `B' identically.
1830 * Incorrect handling of tail-padding for virtual bases. G++
1831 does not use tail padding when laying out virtual bases. For
1834 struct A { virtual void f(); char c1; };
1835 struct B { B(); char c2; };
1836 struct C : public A, public virtual B {};
1838 In this case, G++ will not place `B' into the tail-padding for
1839 `A'; other compilers will. You can avoid this problem by
1840 explicitly padding `A' so that its size is a multiple of its
1841 alignment (ignoring virtual base classes); that will cause
1842 G++ and other compilers to layout `C' identically.
1844 * Incorrect handling of bit-fields with declared widths greater
1845 than that of their underlying types, when the bit-fields
1846 appear in a union. For example:
1848 union U { int i : 4096; };
1850 Assuming that an `int' does not have 4096 bits, G++ will make
1851 the union too small by the number of bits in an `int'.
1853 * Empty classes can be placed at incorrect offsets. For
1863 struct C : public B, public A {};
1865 G++ will place the `A' base class of `C' at a nonzero offset;
1866 it should be placed at offset zero. G++ mistakenly believes
1867 that the `A' data member of `B' is already at offset zero.
1869 * Names of template functions whose types involve `typename' or
1870 template template parameters can be mangled incorrectly.
1872 template <typename Q>
1873 void f(typename Q::X) {}
1875 template <template <typename> class Q>
1876 void f(typename Q<int>::X) {}
1878 Instantiations of these templates may be mangled incorrectly.
1881 `-Wctor-dtor-privacy (C++ only)'
1882 Warn when a class seems unusable because all the constructors or
1883 destructors in that class are private, and it has neither friends
1884 nor public static member functions.
1886 `-Wnon-virtual-dtor (C++ only)'
1887 Warn when a class appears to be polymorphic, thereby requiring a
1888 virtual destructor, yet it declares a non-virtual one. This
1889 warning is also enabled if -Weffc++ is specified.
1891 `-Wreorder (C++ only)'
1892 Warn when the order of member initializers given in the code does
1893 not match the order in which they must be executed. For instance:
1898 A(): j (0), i (1) { }
1901 The compiler will rearrange the member initializers for `i' and
1902 `j' to match the declaration order of the members, emitting a
1903 warning to that effect. This warning is enabled by `-Wall'.
1905 The following `-W...' options are not affected by `-Wall'.
1907 `-Weffc++ (C++ only)'
1908 Warn about violations of the following style guidelines from Scott
1909 Meyers' `Effective C++' book:
1911 * Item 11: Define a copy constructor and an assignment
1912 operator for classes with dynamically allocated memory.
1914 * Item 12: Prefer initialization to assignment in constructors.
1916 * Item 14: Make destructors virtual in base classes.
1918 * Item 15: Have `operator=' return a reference to `*this'.
1920 * Item 23: Don't try to return a reference when you must
1924 Also warn about violations of the following style guidelines from
1925 Scott Meyers' `More Effective C++' book:
1927 * Item 6: Distinguish between prefix and postfix forms of
1928 increment and decrement operators.
1930 * Item 7: Never overload `&&', `||', or `,'.
1933 When selecting this option, be aware that the standard library
1934 headers do not obey all of these guidelines; use `grep -v' to
1935 filter out those warnings.
1937 `-Wno-deprecated (C++ only)'
1938 Do not warn about usage of deprecated features. *Note Deprecated
1941 `-Wstrict-null-sentinel (C++ only)'
1942 Warn also about the use of an uncasted `NULL' as sentinel. When
1943 compiling only with GCC this is a valid sentinel, as `NULL' is
1944 defined to `__null'. Although it is a null pointer constant not a
1945 null pointer, it is guaranteed to of the same size as a pointer.
1946 But this use is not portable across different compilers.
1948 `-Wno-non-template-friend (C++ only)'
1949 Disable warnings when non-templatized friend functions are declared
1950 within a template. Since the advent of explicit template
1951 specification support in G++, if the name of the friend is an
1952 unqualified-id (i.e., `friend foo(int)'), the C++ language
1953 specification demands that the friend declare or define an
1954 ordinary, nontemplate function. (Section 14.5.3). Before G++
1955 implemented explicit specification, unqualified-ids could be
1956 interpreted as a particular specialization of a templatized
1957 function. Because this non-conforming behavior is no longer the
1958 default behavior for G++, `-Wnon-template-friend' allows the
1959 compiler to check existing code for potential trouble spots and is
1960 on by default. This new compiler behavior can be turned off with
1961 `-Wno-non-template-friend' which keeps the conformant compiler code
1962 but disables the helpful warning.
1964 `-Wold-style-cast (C++ only)'
1965 Warn if an old-style (C-style) cast to a non-void type is used
1966 within a C++ program. The new-style casts (`dynamic_cast',
1967 `static_cast', `reinterpret_cast', and `const_cast') are less
1968 vulnerable to unintended effects and much easier to search for.
1970 `-Woverloaded-virtual (C++ only)'
1971 Warn when a function declaration hides virtual functions from a
1972 base class. For example, in:
1978 struct B: public A {
1982 the `A' class version of `f' is hidden in `B', and code like:
1987 will fail to compile.
1989 `-Wno-pmf-conversions (C++ only)'
1990 Disable the diagnostic for converting a bound pointer to member
1991 function to a plain pointer.
1993 `-Wsign-promo (C++ only)'
1994 Warn when overload resolution chooses a promotion from unsigned or
1995 enumerated type to a signed type, over a conversion to an unsigned
1996 type of the same size. Previous versions of G++ would try to
1997 preserve unsignedness, but the standard mandates the current
2002 A& operator = (int);
2011 In this example, G++ will synthesize a default `A& operator =
2012 (const A&);', while cfront will use the user-defined `operator ='.
2015 File: gcc.info, Node: Objective-C and Objective-C++ Dialect Options, Next: Language Independent Options, Prev: C++ Dialect Options, Up: Invoking GCC
2017 3.6 Options Controlling Objective-C and Objective-C++ Dialects
2018 ==============================================================
2020 (NOTE: This manual does not describe the Objective-C and Objective-C++
2021 languages themselves. See *Note Language Standards Supported by GCC:
2022 Standards, for references.)
2024 This section describes the command-line options that are only
2025 meaningful for Objective-C and Objective-C++ programs, but you can also
2026 use most of the language-independent GNU compiler options. For
2027 example, you might compile a file `some_class.m' like this:
2029 gcc -g -fgnu-runtime -O -c some_class.m
2031 In this example, `-fgnu-runtime' is an option meant only for
2032 Objective-C and Objective-C++ programs; you can use the other options
2033 with any language supported by GCC.
2035 Note that since Objective-C is an extension of the C language,
2036 Objective-C compilations may also use options specific to the C
2037 front-end (e.g., `-Wtraditional'). Similarly, Objective-C++
2038 compilations may use C++-specific options (e.g., `-Wabi').
2040 Here is a list of options that are _only_ for compiling Objective-C
2041 and Objective-C++ programs:
2043 `-fconstant-string-class=CLASS-NAME'
2044 Use CLASS-NAME as the name of the class to instantiate for each
2045 literal string specified with the syntax `@"..."'. The default
2046 class name is `NXConstantString' if the GNU runtime is being used,
2047 and `NSConstantString' if the NeXT runtime is being used (see
2048 below). The `-fconstant-cfstrings' option, if also present, will
2049 override the `-fconstant-string-class' setting and cause `@"..."'
2050 literals to be laid out as constant CoreFoundation strings.
2053 Generate object code compatible with the standard GNU Objective-C
2054 runtime. This is the default for most types of systems.
2057 Generate output compatible with the NeXT runtime. This is the
2058 default for NeXT-based systems, including Darwin and Mac OS X.
2059 The macro `__NEXT_RUNTIME__' is predefined if (and only if) this
2062 `-fno-nil-receivers'
2063 Assume that all Objective-C message dispatches (e.g., `[receiver
2064 message:arg]') in this translation unit ensure that the receiver
2065 is not `nil'. This allows for more efficient entry points in the
2066 runtime to be used. Currently, this option is only available in
2067 conjunction with the NeXT runtime on Mac OS X 10.3 and later.
2069 `-fobjc-call-cxx-cdtors'
2070 For each Objective-C class, check if any of its instance variables
2071 is a C++ object with a non-trivial default constructor. If so,
2072 synthesize a special `- (id) .cxx_construct' instance method that
2073 will run non-trivial default constructors on any such instance
2074 variables, in order, and then return `self'. Similarly, check if
2075 any instance variable is a C++ object with a non-trivial
2076 destructor, and if so, synthesize a special `- (void)
2077 .cxx_destruct' method that will run all such default destructors,
2080 The `- (id) .cxx_construct' and/or `- (void) .cxx_destruct' methods
2081 thusly generated will only operate on instance variables declared
2082 in the current Objective-C class, and not those inherited from
2083 superclasses. It is the responsibility of the Objective-C runtime
2084 to invoke all such methods in an object's inheritance hierarchy.
2085 The `- (id) .cxx_construct' methods will be invoked by the runtime
2086 immediately after a new object instance is allocated; the `-
2087 (void) .cxx_destruct' methods will be invoked immediately before
2088 the runtime deallocates an object instance.
2090 As of this writing, only the NeXT runtime on Mac OS X 10.4 and
2091 later has support for invoking the `- (id) .cxx_construct' and `-
2092 (void) .cxx_destruct' methods.
2094 `-fobjc-direct-dispatch'
2095 Allow fast jumps to the message dispatcher. On Darwin this is
2096 accomplished via the comm page.
2099 Enable syntactic support for structured exception handling in
2100 Objective-C, similar to what is offered by C++ and Java. This
2101 option is unavailable in conjunction with the NeXT runtime on Mac
2102 OS X 10.2 and earlier.
2109 @catch (AnObjCClass *exc) {
2116 @catch (AnotherClass *exc) {
2119 @catch (id allOthers) {
2128 The `@throw' statement may appear anywhere in an Objective-C or
2129 Objective-C++ program; when used inside of a `@catch' block, the
2130 `@throw' may appear without an argument (as shown above), in which
2131 case the object caught by the `@catch' will be rethrown.
2133 Note that only (pointers to) Objective-C objects may be thrown and
2134 caught using this scheme. When an object is thrown, it will be
2135 caught by the nearest `@catch' clause capable of handling objects
2136 of that type, analogously to how `catch' blocks work in C++ and
2137 Java. A `@catch(id ...)' clause (as shown above) may also be
2138 provided to catch any and all Objective-C exceptions not caught by
2139 previous `@catch' clauses (if any).
2141 The `@finally' clause, if present, will be executed upon exit from
2142 the immediately preceding `@try ... @catch' section. This will
2143 happen regardless of whether any exceptions are thrown, caught or
2144 rethrown inside the `@try ... @catch' section, analogously to the
2145 behavior of the `finally' clause in Java.
2147 There are several caveats to using the new exception mechanism:
2149 * Although currently designed to be binary compatible with
2150 `NS_HANDLER'-style idioms provided by the `NSException'
2151 class, the new exceptions can only be used on Mac OS X 10.3
2152 (Panther) and later systems, due to additional functionality
2153 needed in the (NeXT) Objective-C runtime.
2155 * As mentioned above, the new exceptions do not support handling
2156 types other than Objective-C objects. Furthermore, when
2157 used from Objective-C++, the Objective-C exception model does
2158 not interoperate with C++ exceptions at this time. This
2159 means you cannot `@throw' an exception from Objective-C and
2160 `catch' it in C++, or vice versa (i.e., `throw ... @catch').
2162 The `-fobjc-exceptions' switch also enables the use of
2163 synchronization blocks for thread-safe execution:
2165 @synchronized (ObjCClass *guard) {
2169 Upon entering the `@synchronized' block, a thread of execution
2170 shall first check whether a lock has been placed on the
2171 corresponding `guard' object by another thread. If it has, the
2172 current thread shall wait until the other thread relinquishes its
2173 lock. Once `guard' becomes available, the current thread will
2174 place its own lock on it, execute the code contained in the
2175 `@synchronized' block, and finally relinquish the lock (thereby
2176 making `guard' available to other threads).
2178 Unlike Java, Objective-C does not allow for entire methods to be
2179 marked `@synchronized'. Note that throwing exceptions out of
2180 `@synchronized' blocks is allowed, and will cause the guarding
2181 object to be unlocked properly.
2184 Enable garbage collection (GC) in Objective-C and Objective-C++
2187 `-freplace-objc-classes'
2188 Emit a special marker instructing `ld(1)' not to statically link in
2189 the resulting object file, and allow `dyld(1)' to load it in at
2190 run time instead. This is used in conjunction with the
2191 Fix-and-Continue debugging mode, where the object file in question
2192 may be recompiled and dynamically reloaded in the course of
2193 program execution, without the need to restart the program itself.
2194 Currently, Fix-and-Continue functionality is only available in
2195 conjunction with the NeXT runtime on Mac OS X 10.3 and later.
2198 When compiling for the NeXT runtime, the compiler ordinarily
2199 replaces calls to `objc_getClass("...")' (when the name of the
2200 class is known at compile time) with static class references that
2201 get initialized at load time, which improves run-time performance.
2202 Specifying the `-fzero-link' flag suppresses this behavior and
2203 causes calls to `objc_getClass("...")' to be retained. This is
2204 useful in Zero-Link debugging mode, since it allows for individual
2205 class implementations to be modified during program execution.
2208 Dump interface declarations for all classes seen in the source
2209 file to a file named `SOURCENAME.decl'.
2211 `-Wassign-intercept'
2212 Warn whenever an Objective-C assignment is being intercepted by the
2216 If a class is declared to implement a protocol, a warning is
2217 issued for every method in the protocol that is not implemented by
2218 the class. The default behavior is to issue a warning for every
2219 method not explicitly implemented in the class, even if a method
2220 implementation is inherited from the superclass. If you use the
2221 `-Wno-protocol' option, then methods inherited from the superclass
2222 are considered to be implemented, and no warning is issued for
2226 Warn if multiple methods of different types for the same selector
2227 are found during compilation. The check is performed on the list
2228 of methods in the final stage of compilation. Additionally, a
2229 check is performed for each selector appearing in a
2230 `@selector(...)' expression, and a corresponding method for that
2231 selector has been found during compilation. Because these checks
2232 scan the method table only at the end of compilation, these
2233 warnings are not produced if the final stage of compilation is not
2234 reached, for example because an error is found during compilation,
2235 or because the `-fsyntax-only' option is being used.
2237 `-Wstrict-selector-match'
2238 Warn if multiple methods with differing argument and/or return
2239 types are found for a given selector when attempting to send a
2240 message using this selector to a receiver of type `id' or `Class'.
2241 When this flag is off (which is the default behavior), the
2242 compiler will omit such warnings if any differences found are
2243 confined to types which share the same size and alignment.
2245 `-Wundeclared-selector'
2246 Warn if a `@selector(...)' expression referring to an undeclared
2247 selector is found. A selector is considered undeclared if no
2248 method with that name has been declared before the
2249 `@selector(...)' expression, either explicitly in an `@interface'
2250 or `@protocol' declaration, or implicitly in an `@implementation'
2251 section. This option always performs its checks as soon as a
2252 `@selector(...)' expression is found, while `-Wselector' only
2253 performs its checks in the final stage of compilation. This also
2254 enforces the coding style convention that methods and selectors
2255 must be declared before being used.
2257 `-print-objc-runtime-info'
2258 Generate C header describing the largest structure that is passed
2263 File: gcc.info, Node: Language Independent Options, Next: Warning Options, Prev: Objective-C and Objective-C++ Dialect Options, Up: Invoking GCC
2265 3.7 Options to Control Diagnostic Messages Formatting
2266 =====================================================
2268 Traditionally, diagnostic messages have been formatted irrespective of
2269 the output device's aspect (e.g. its width, ...). The options described
2270 below can be used to control the diagnostic messages formatting
2271 algorithm, e.g. how many characters per line, how often source location
2272 information should be reported. Right now, only the C++ front end can
2273 honor these options. However it is expected, in the near future, that
2274 the remaining front ends would be able to digest them correctly.
2276 `-fmessage-length=N'
2277 Try to format error messages so that they fit on lines of about N
2278 characters. The default is 72 characters for `g++' and 0 for the
2279 rest of the front ends supported by GCC. If N is zero, then no
2280 line-wrapping will be done; each error message will appear on a
2283 `-fdiagnostics-show-location=once'
2284 Only meaningful in line-wrapping mode. Instructs the diagnostic
2285 messages reporter to emit _once_ source location information; that
2286 is, in case the message is too long to fit on a single physical
2287 line and has to be wrapped, the source location won't be emitted
2288 (as prefix) again, over and over, in subsequent continuation
2289 lines. This is the default behavior.
2291 `-fdiagnostics-show-location=every-line'
2292 Only meaningful in line-wrapping mode. Instructs the diagnostic
2293 messages reporter to emit the same source location information (as
2294 prefix) for physical lines that result from the process of breaking
2295 a message which is too long to fit on a single line.
2297 `-fdiagnostics-show-option'
2298 This option instructs the diagnostic machinery to add text to each
2299 diagnostic emitted, which indicates which command line option
2300 directly controls that diagnostic, when such an option is known to
2301 the diagnostic machinery.
2305 File: gcc.info, Node: Warning Options, Next: Debugging Options, Prev: Language Independent Options, Up: Invoking GCC
2307 3.8 Options to Request or Suppress Warnings
2308 ===========================================
2310 Warnings are diagnostic messages that report constructions which are
2311 not inherently erroneous but which are risky or suggest there may have
2314 You can request many specific warnings with options beginning `-W',
2315 for example `-Wimplicit' to request warnings on implicit declarations.
2316 Each of these specific warning options also has a negative form
2317 beginning `-Wno-' to turn off warnings; for example, `-Wno-implicit'.
2318 This manual lists only one of the two forms, whichever is not the
2321 The following options control the amount and kinds of warnings produced
2322 by GCC; for further, language-specific options also refer to *Note C++
2323 Dialect Options:: and *Note Objective-C and Objective-C++ Dialect
2327 Check the code for syntax errors, but don't do anything beyond
2331 Issue all the warnings demanded by strict ISO C and ISO C++;
2332 reject all programs that use forbidden extensions, and some other
2333 programs that do not follow ISO C and ISO C++. For ISO C, follows
2334 the version of the ISO C standard specified by any `-std' option
2337 Valid ISO C and ISO C++ programs should compile properly with or
2338 without this option (though a rare few will require `-ansi' or a
2339 `-std' option specifying the required version of ISO C). However,
2340 without this option, certain GNU extensions and traditional C and
2341 C++ features are supported as well. With this option, they are
2344 `-pedantic' does not cause warning messages for use of the
2345 alternate keywords whose names begin and end with `__'. Pedantic
2346 warnings are also disabled in the expression that follows
2347 `__extension__'. However, only system header files should use
2348 these escape routes; application programs should avoid them.
2349 *Note Alternate Keywords::.
2351 Some users try to use `-pedantic' to check programs for strict ISO
2352 C conformance. They soon find that it does not do quite what they
2353 want: it finds some non-ISO practices, but not all--only those for
2354 which ISO C _requires_ a diagnostic, and some others for which
2355 diagnostics have been added.
2357 A feature to report any failure to conform to ISO C might be
2358 useful in some instances, but would require considerable
2359 additional work and would be quite different from `-pedantic'. We
2360 don't have plans to support such a feature in the near future.
2362 Where the standard specified with `-std' represents a GNU extended
2363 dialect of C, such as `gnu89' or `gnu99', there is a corresponding
2364 "base standard", the version of ISO C on which the GNU extended
2365 dialect is based. Warnings from `-pedantic' are given where they
2366 are required by the base standard. (It would not make sense for
2367 such warnings to be given only for features not in the specified
2368 GNU C dialect, since by definition the GNU dialects of C include
2369 all features the compiler supports with the given option, and
2370 there would be nothing to warn about.)
2373 Like `-pedantic', except that errors are produced rather than
2377 Inhibit all warning messages.
2380 Inhibit warning messages about the use of `#import'.
2383 Warn if an array subscript has type `char'. This is a common cause
2384 of error, as programmers often forget that this type is signed on
2385 some machines. This warning is enabled by `-Wall'.
2388 Warn whenever a comment-start sequence `/*' appears in a `/*'
2389 comment, or whenever a Backslash-Newline appears in a `//' comment.
2390 This warning is enabled by `-Wall'.
2393 This option causes the compiler to abort compilation on the first
2394 error occurred rather than trying to keep going and printing
2395 further error messages.
2398 Check calls to `printf' and `scanf', etc., to make sure that the
2399 arguments supplied have types appropriate to the format string
2400 specified, and that the conversions specified in the format string
2401 make sense. This includes standard functions, and others
2402 specified by format attributes (*note Function Attributes::), in
2403 the `printf', `scanf', `strftime' and `strfmon' (an X/Open
2404 extension, not in the C standard) families (or other
2405 target-specific families). Which functions are checked without
2406 format attributes having been specified depends on the standard
2407 version selected, and such checks of functions without the
2408 attribute specified are disabled by `-ffreestanding' or
2411 The formats are checked against the format features supported by
2412 GNU libc version 2.2. These include all ISO C90 and C99 features,
2413 as well as features from the Single Unix Specification and some
2414 BSD and GNU extensions. Other library implementations may not
2415 support all these features; GCC does not support warning about
2416 features that go beyond a particular library's limitations.
2417 However, if `-pedantic' is used with `-Wformat', warnings will be
2418 given about format features not in the selected standard version
2419 (but not for `strfmon' formats, since those are not in any version
2420 of the C standard). *Note Options Controlling C Dialect: C
2423 Since `-Wformat' also checks for null format arguments for several
2424 functions, `-Wformat' also implies `-Wnonnull'.
2426 `-Wformat' is included in `-Wall'. For more control over some
2427 aspects of format checking, the options `-Wformat-y2k',
2428 `-Wno-format-extra-args', `-Wno-format-zero-length',
2429 `-Wformat-nonliteral', `-Wformat-security', and `-Wformat=2' are
2430 available, but are not included in `-Wall'.
2433 If `-Wformat' is specified, also warn about `strftime' formats
2434 which may yield only a two-digit year.
2436 `-Wno-format-extra-args'
2437 If `-Wformat' is specified, do not warn about excess arguments to a
2438 `printf' or `scanf' format function. The C standard specifies
2439 that such arguments are ignored.
2441 Where the unused arguments lie between used arguments that are
2442 specified with `$' operand number specifications, normally
2443 warnings are still given, since the implementation could not know
2444 what type to pass to `va_arg' to skip the unused arguments.
2445 However, in the case of `scanf' formats, this option will suppress
2446 the warning if the unused arguments are all pointers, since the
2447 Single Unix Specification says that such unused arguments are
2450 `-Wno-format-zero-length'
2451 If `-Wformat' is specified, do not warn about zero-length formats.
2452 The C standard specifies that zero-length formats are allowed.
2454 `-Wformat-nonliteral'
2455 If `-Wformat' is specified, also warn if the format string is not a
2456 string literal and so cannot be checked, unless the format function
2457 takes its format arguments as a `va_list'.
2460 If `-Wformat' is specified, also warn about uses of format
2461 functions that represent possible security problems. At present,
2462 this warns about calls to `printf' and `scanf' functions where the
2463 format string is not a string literal and there are no format
2464 arguments, as in `printf (foo);'. This may be a security hole if
2465 the format string came from untrusted input and contains `%n'.
2466 (This is currently a subset of what `-Wformat-nonliteral' warns
2467 about, but in future warnings may be added to `-Wformat-security'
2468 that are not included in `-Wformat-nonliteral'.)
2471 Enable `-Wformat' plus format checks not included in `-Wformat'.
2472 Currently equivalent to `-Wformat -Wformat-nonliteral
2473 -Wformat-security -Wformat-y2k'.
2476 Warn about passing a null pointer for arguments marked as
2477 requiring a non-null value by the `nonnull' function attribute.
2479 `-Wnonnull' is included in `-Wall' and `-Wformat'. It can be
2480 disabled with the `-Wno-nonnull' option.
2482 `-Winit-self (C, C++, Objective-C and Objective-C++ only)'
2483 Warn about uninitialized variables which are initialized with
2484 themselves. Note this option can only be used with the
2485 `-Wuninitialized' option, which in turn only works with `-O1' and
2488 For example, GCC will warn about `i' being uninitialized in the
2489 following snippet only when `-Winit-self' has been specified:
2497 Warn when a declaration does not specify a type. This warning is
2500 `-Wimplicit-function-declaration'
2501 `-Werror-implicit-function-declaration'
2502 Give a warning (or error) whenever a function is used before being
2503 declared. The form `-Wno-error-implicit-function-declaration' is
2504 not supported. This warning is enabled by `-Wall' (as a warning,
2508 Same as `-Wimplicit-int' and `-Wimplicit-function-declaration'.
2509 This warning is enabled by `-Wall'.
2512 Warn if the type of `main' is suspicious. `main' should be a
2513 function with external linkage, returning int, taking either zero
2514 arguments, two, or three arguments of appropriate types. This
2515 warning is enabled by `-Wall'.
2518 Warn if an aggregate or union initializer is not fully bracketed.
2519 In the following example, the initializer for `a' is not fully
2520 bracketed, but that for `b' is fully bracketed.
2522 int a[2][2] = { 0, 1, 2, 3 };
2523 int b[2][2] = { { 0, 1 }, { 2, 3 } };
2525 This warning is enabled by `-Wall'.
2527 `-Wmissing-include-dirs (C, C++, Objective-C and Objective-C++ only)'
2528 Warn if a user-supplied include directory does not exist.
2531 Warn if parentheses are omitted in certain contexts, such as when
2532 there is an assignment in a context where a truth value is
2533 expected, or when operators are nested whose precedence people
2534 often get confused about.
2536 Also warn if a comparison like `x<=y<=z' appears; this is
2537 equivalent to `(x<=y ? 1 : 0) <= z', which is a different
2538 interpretation from that of ordinary mathematical notation.
2540 Also warn about constructions where there may be confusion to which
2541 `if' statement an `else' branch belongs. Here is an example of
2552 In C/C++, every `else' branch belongs to the innermost possible
2553 `if' statement, which in this example is `if (b)'. This is often
2554 not what the programmer expected, as illustrated in the above
2555 example by indentation the programmer chose. When there is the
2556 potential for this confusion, GCC will issue a warning when this
2557 flag is specified. To eliminate the warning, add explicit braces
2558 around the innermost `if' statement so there is no way the `else'
2559 could belong to the enclosing `if'. The resulting code would look
2572 This warning is enabled by `-Wall'.
2575 Warn about code that may have undefined semantics because of
2576 violations of sequence point rules in the C and C++ standards.
2578 The C and C++ standards defines the order in which expressions in
2579 a C/C++ program are evaluated in terms of "sequence points", which
2580 represent a partial ordering between the execution of parts of the
2581 program: those executed before the sequence point, and those
2582 executed after it. These occur after the evaluation of a full
2583 expression (one which is not part of a larger expression), after
2584 the evaluation of the first operand of a `&&', `||', `? :' or `,'
2585 (comma) operator, before a function is called (but after the
2586 evaluation of its arguments and the expression denoting the called
2587 function), and in certain other places. Other than as expressed
2588 by the sequence point rules, the order of evaluation of
2589 subexpressions of an expression is not specified. All these rules
2590 describe only a partial order rather than a total order, since,
2591 for example, if two functions are called within one expression
2592 with no sequence point between them, the order in which the
2593 functions are called is not specified. However, the standards
2594 committee have ruled that function calls do not overlap.
2596 It is not specified when between sequence points modifications to
2597 the values of objects take effect. Programs whose behavior
2598 depends on this have undefined behavior; the C and C++ standards
2599 specify that "Between the previous and next sequence point an
2600 object shall have its stored value modified at most once by the
2601 evaluation of an expression. Furthermore, the prior value shall
2602 be read only to determine the value to be stored.". If a program
2603 breaks these rules, the results on any particular implementation
2604 are entirely unpredictable.
2606 Examples of code with undefined behavior are `a = a++;', `a[n] =
2607 b[n++]' and `a[i++] = i;'. Some more complicated cases are not
2608 diagnosed by this option, and it may give an occasional false
2609 positive result, but in general it has been found fairly effective
2610 at detecting this sort of problem in programs.
2612 The standard is worded confusingly, therefore there is some debate
2613 over the precise meaning of the sequence point rules in subtle
2614 cases. Links to discussions of the problem, including proposed
2615 formal definitions, may be found on the GCC readings page, at
2616 `http://gcc.gnu.org/readings.html'.
2618 This warning is enabled by `-Wall' for C and C++.
2621 Warn whenever a function is defined with a return-type that
2622 defaults to `int'. Also warn about any `return' statement with no
2623 return-value in a function whose return-type is not `void'.
2625 For C, also warn if the return type of a function has a type
2626 qualifier such as `const'. Such a type qualifier has no effect,
2627 since the value returned by a function is not an lvalue. ISO C
2628 prohibits qualified `void' return types on function definitions,
2629 so such return types always receive a warning even without this
2632 For C++, a function without return type always produces a
2633 diagnostic message, even when `-Wno-return-type' is specified.
2634 The only exceptions are `main' and functions defined in system
2637 This warning is enabled by `-Wall'.
2640 Warn whenever a `switch' statement has an index of enumerated type
2641 and lacks a `case' for one or more of the named codes of that
2642 enumeration. (The presence of a `default' label prevents this
2643 warning.) `case' labels outside the enumeration range also
2644 provoke warnings when this option is used. This warning is
2648 Warn whenever a `switch' statement does not have a `default' case.
2651 Warn whenever a `switch' statement has an index of enumerated type
2652 and lacks a `case' for one or more of the named codes of that
2653 enumeration. `case' labels outside the enumeration range also
2654 provoke warnings when this option is used.
2657 Warn if any trigraphs are encountered that might change the
2658 meaning of the program (trigraphs within comments are not warned
2659 about). This warning is enabled by `-Wall'.
2662 Warn whenever a static function is declared but not defined or a
2663 non-inline static function is unused. This warning is enabled by
2667 Warn whenever a label is declared but not used. This warning is
2670 To suppress this warning use the `unused' attribute (*note
2671 Variable Attributes::).
2673 `-Wunused-parameter'
2674 Warn whenever a function parameter is unused aside from its
2677 To suppress this warning use the `unused' attribute (*note
2678 Variable Attributes::).
2681 Warn whenever a local variable or non-constant static variable is
2682 unused aside from its declaration. This warning is enabled by
2685 To suppress this warning use the `unused' attribute (*note
2686 Variable Attributes::).
2689 Warn whenever a statement computes a result that is explicitly not
2690 used. This warning is enabled by `-Wall'.
2692 To suppress this warning cast the expression to `void'.
2695 All the above `-Wunused' options combined.
2697 In order to get a warning about an unused function parameter, you
2698 must either specify `-Wextra -Wunused' (note that `-Wall' implies
2699 `-Wunused'), or separately specify `-Wunused-parameter'.
2702 Warn if an automatic variable is used without first being
2703 initialized or if a variable may be clobbered by a `setjmp' call.
2705 These warnings are possible only in optimizing compilation,
2706 because they require data flow information that is computed only
2707 when optimizing. If you do not specify `-O', you will not get
2708 these warnings. Instead, GCC will issue a warning about
2709 `-Wuninitialized' requiring `-O'.
2711 If you want to warn about code which uses the uninitialized value
2712 of the variable in its own initializer, use the `-Winit-self'
2715 These warnings occur for individual uninitialized or clobbered
2716 elements of structure, union or array variables as well as for
2717 variables which are uninitialized or clobbered as a whole. They do
2718 not occur for variables or elements declared `volatile'. Because
2719 these warnings depend on optimization, the exact variables or
2720 elements for which there are warnings will depend on the precise
2721 optimization options and version of GCC used.
2723 Note that there may be no warning about a variable that is used
2724 only to compute a value that itself is never used, because such
2725 computations may be deleted by data flow analysis before the
2726 warnings are printed.
2728 These warnings are made optional because GCC is not smart enough
2729 to see all the reasons why the code might be correct despite
2730 appearing to have an error. Here is one example of how this can
2746 If the value of `y' is always 1, 2 or 3, then `x' is always
2747 initialized, but GCC doesn't know this. Here is another common
2752 if (change_y) save_y = y, y = new_y;
2754 if (change_y) y = save_y;
2757 This has no bug because `save_y' is used only if it is set.
2759 This option also warns when a non-volatile automatic variable
2760 might be changed by a call to `longjmp'. These warnings as well
2761 are possible only in optimizing compilation.
2763 The compiler sees only the calls to `setjmp'. It cannot know
2764 where `longjmp' will be called; in fact, a signal handler could
2765 call it at any point in the code. As a result, you may get a
2766 warning even when there is in fact no problem because `longjmp'
2767 cannot in fact be called at the place which would cause a problem.
2769 Some spurious warnings can be avoided if you declare all the
2770 functions you use that never return as `noreturn'. *Note Function
2773 This warning is enabled by `-Wall'.
2776 Warn when a #pragma directive is encountered which is not
2777 understood by GCC. If this command line option is used, warnings
2778 will even be issued for unknown pragmas in system header files.
2779 This is not the case if the warnings were only enabled by the
2780 `-Wall' command line option.
2783 Do not warn about misuses of pragmas, such as incorrect parameters,
2784 invalid syntax, or conflicts between pragmas. See also
2785 `-Wunknown-pragmas'.
2788 This option is only active when `-fstrict-aliasing' is active. It
2789 warns about code which might break the strict aliasing rules that
2790 the compiler is using for optimization. The warning does not
2791 catch all cases, but does attempt to catch the more common
2792 pitfalls. It is included in `-Wall'. It is equivalent to
2795 `-Wstrict-aliasing=n'
2796 This option is only active when `-fstrict-aliasing' is active. It
2797 warns about code which might break the strict aliasing rules that
2798 the compiler is using for optimization. Higher levels correspond
2799 to higher accuracy (fewer false positives). Higher levels also
2800 correspond to more effort, similar to the way -O works.
2801 `-Wstrict-aliasing' is equivalent to `-Wstrict-aliasing=n', with
2804 Level 1: Most aggressive, quick, least accurate. Possibly useful
2805 when higher levels do not warn but -fstrict-aliasing still breaks
2806 the code, as it has very few false negatives. However, it has
2807 many false positives. Warns for all pointer conversions between
2808 possibly incompatible types, even if never dereferenced. Runs in
2811 Level 2: Aggressive, quick, not too precise. May still have many
2812 false positives (not as many as level 1 though), and few false
2813 negatives (but possibly more than level 1). Unlike level 1, it
2814 only warns when an address is taken. Warns about incomplete
2815 types. Runs in the frontend only.
2817 Level 3 (default for `-Wstrict-aliasing'): Should have very few
2818 false positives and few false negatives. Slightly slower than
2819 levels 1 or 2 when optimization is enabled. Takes care of the
2820 common punn+dereference pattern in the frontend:
2821 `*(int*)&some_float'. If optimization is enabled, it also runs in
2822 the backend, where it deals with multiple statement cases using
2823 flow-sensitive points-to information. Only warns when the
2824 converted pointer is dereferenced. Does not warn about incomplete
2829 `-Wstrict-overflow=N'
2830 This option is only active when `-fstrict-overflow' is active. It
2831 warns about cases where the compiler optimizes based on the
2832 assumption that signed overflow does not occur. Note that it does
2833 not warn about all cases where the code might overflow: it only
2834 warns about cases where the compiler implements some optimization.
2835 Thus this warning depends on the optimization level.
2837 An optimization which assumes that signed overflow does not occur
2838 is perfectly safe if the values of the variables involved are such
2839 that overflow never does, in fact, occur. Therefore this warning
2840 can easily give a false positive: a warning about code which is not
2841 actually a problem. To help focus on important issues, several
2842 warning levels are defined. No warnings are issued for the use of
2843 undefined signed overflow when estimating how many iterations a
2844 loop will require, in particular when determining whether a loop
2845 will be executed at all.
2847 `-Wstrict-overflow=1'
2848 Warn about cases which are both questionable and easy to
2849 avoid. For example: `x + 1 > x'; with `-fstrict-overflow',
2850 the compiler will simplify this to `1'. This level of
2851 `-Wstrict-overflow' is enabled by `-Wall'; higher levels are
2852 not, and must be explicitly requested.
2854 `-Wstrict-overflow=2'
2855 Also warn about other cases where a comparison is simplified
2856 to a constant. For example: `abs (x) >= 0'. This can only be
2857 simplified when `-fstrict-overflow' is in effect, because
2858 `abs (INT_MIN)' overflows to `INT_MIN', which is less than
2859 zero. `-Wstrict-overflow' (with no level) is the same as
2860 `-Wstrict-overflow=2'.
2862 `-Wstrict-overflow=3'
2863 Also warn about other cases where a comparison is simplified.
2864 For example: `x + 1 > 1' will be simplified to `x > 0'.
2866 `-Wstrict-overflow=4'
2867 Also warn about other simplifications not covered by the
2868 above cases. For example: `(x * 10) / 5' will be simplified
2871 `-Wstrict-overflow=5'
2872 Also warn about cases where the compiler reduces the
2873 magnitude of a constant involved in a comparison. For
2874 example: `x + 2 > y' will be simplified to `x + 1 >= y'.
2875 This is reported only at the highest warning level because
2876 this simplification applies to many comparisons, so this
2877 warning level will give a very large number of false
2881 All of the above `-W' options combined. This enables all the
2882 warnings about constructions that some users consider
2883 questionable, and that are easy to avoid (or modify to prevent the
2884 warning), even in conjunction with macros. This also enables some
2885 language-specific warnings described in *Note C++ Dialect
2886 Options:: and *Note Objective-C and Objective-C++ Dialect
2889 The following `-W...' options are not implied by `-Wall'. Some of
2890 them warn about constructions that users generally do not consider
2891 questionable, but which occasionally you might wish to check for;
2892 others warn about constructions that are necessary or hard to avoid in
2893 some cases, and there is no simple way to modify the code to suppress
2897 (This option used to be called `-W'. The older name is still
2898 supported, but the newer name is more descriptive.) Print extra
2899 warning messages for these events:
2901 * A function can return either with or without a value.
2902 (Falling off the end of the function body is considered
2903 returning without a value.) For example, this function would
2904 evoke such a warning:
2912 * An expression-statement or the left-hand side of a comma
2913 expression contains no side effects. To suppress the
2914 warning, cast the unused expression to void. For example, an
2915 expression such as `x[i,j]' will cause a warning, but
2916 `x[(void)i,j]' will not.
2918 * An unsigned value is compared against zero with `<' or `>='.
2920 * Storage-class specifiers like `static' are not the first
2921 things in a declaration. According to the C Standard, this
2922 usage is obsolescent.
2924 * If `-Wall' or `-Wunused' is also specified, warn about unused
2927 * A comparison between signed and unsigned values could produce
2928 an incorrect result when the signed value is converted to
2929 unsigned. (But don't warn if `-Wno-sign-compare' is also
2932 * An aggregate has an initializer which does not initialize all
2933 members. This warning can be independently controlled by
2934 `-Wmissing-field-initializers'.
2936 * An initialized field without side effects is overridden when
2937 using designated initializers (*note Designated Initializers:
2938 Designated Inits.). This warning can be independently
2939 controlled by `-Woverride-init'.
2941 * A function parameter is declared without a type specifier in
2942 K&R-style functions:
2946 * An empty body occurs in an `if' or `else' statement.
2948 * A pointer is compared against integer zero with `<', `<=',
2951 * A variable might be changed by `longjmp' or `vfork'.
2953 * (C++ only) An enumerator and a non-enumerator both appear in
2954 a conditional expression.
2956 * (C++ only) A non-static reference or non-static `const'
2957 member appears in a class without constructors.
2959 * (C++ only) Ambiguous virtual bases.
2961 * (C++ only) Subscripting an array which has been declared
2964 * (C++ only) Taking the address of a variable which has been
2965 declared `register'.
2967 * (C++ only) A base class is not initialized in a derived
2968 class' copy constructor.
2971 Do not warn about compile-time integer division by zero. Floating
2972 point division by zero is not warned about, as it can be a
2973 legitimate way of obtaining infinities and NaNs.
2976 Print warning messages for constructs found in system header files.
2977 Warnings from system headers are normally suppressed, on the
2978 assumption that they usually do not indicate real problems and
2979 would only make the compiler output harder to read. Using this
2980 command line option tells GCC to emit warnings from system headers
2981 as if they occurred in user code. However, note that using
2982 `-Wall' in conjunction with this option will _not_ warn about
2983 unknown pragmas in system headers--for that, `-Wunknown-pragmas'
2987 Warn if floating point values are used in equality comparisons.
2989 The idea behind this is that sometimes it is convenient (for the
2990 programmer) to consider floating-point values as approximations to
2991 infinitely precise real numbers. If you are doing this, then you
2992 need to compute (by analyzing the code, or in some other way) the
2993 maximum or likely maximum error that the computation introduces,
2994 and allow for it when performing comparisons (and when producing
2995 output, but that's a different problem). In particular, instead
2996 of testing for equality, you would check to see whether the two
2997 values have ranges that overlap; and this is done with the
2998 relational operators, so equality comparisons are probably
3001 `-Wtraditional (C only)'
3002 Warn about certain constructs that behave differently in
3003 traditional and ISO C. Also warn about ISO C constructs that have
3004 no traditional C equivalent, and/or problematic constructs which
3007 * Macro parameters that appear within string literals in the
3008 macro body. In traditional C macro replacement takes place
3009 within string literals, but does not in ISO C.
3011 * In traditional C, some preprocessor directives did not exist.
3012 Traditional preprocessors would only consider a line to be a
3013 directive if the `#' appeared in column 1 on the line.
3014 Therefore `-Wtraditional' warns about directives that
3015 traditional C understands but would ignore because the `#'
3016 does not appear as the first character on the line. It also
3017 suggests you hide directives like `#pragma' not understood by
3018 traditional C by indenting them. Some traditional
3019 implementations would not recognize `#elif', so it suggests
3020 avoiding it altogether.
3022 * A function-like macro that appears without arguments.
3024 * The unary plus operator.
3026 * The `U' integer constant suffix, or the `F' or `L' floating
3027 point constant suffixes. (Traditional C does support the `L'
3028 suffix on integer constants.) Note, these suffixes appear in
3029 macros defined in the system headers of most modern systems,
3030 e.g. the `_MIN'/`_MAX' macros in `<limits.h>'. Use of these
3031 macros in user code might normally lead to spurious warnings,
3032 however GCC's integrated preprocessor has enough context to
3033 avoid warning in these cases.
3035 * A function declared external in one block and then used after
3036 the end of the block.
3038 * A `switch' statement has an operand of type `long'.
3040 * A non-`static' function declaration follows a `static' one.
3041 This construct is not accepted by some traditional C
3044 * The ISO type of an integer constant has a different width or
3045 signedness from its traditional type. This warning is only
3046 issued if the base of the constant is ten. I.e. hexadecimal
3047 or octal values, which typically represent bit patterns, are
3050 * Usage of ISO string concatenation is detected.
3052 * Initialization of automatic aggregates.
3054 * Identifier conflicts with labels. Traditional C lacks a
3055 separate namespace for labels.
3057 * Initialization of unions. If the initializer is zero, the
3058 warning is omitted. This is done under the assumption that
3059 the zero initializer in user code appears conditioned on e.g.
3060 `__STDC__' to avoid missing initializer warnings and relies
3061 on default initialization to zero in the traditional C case.
3063 * Conversions by prototypes between fixed/floating point values
3064 and vice versa. The absence of these prototypes when
3065 compiling with traditional C would cause serious problems.
3066 This is a subset of the possible conversion warnings, for the
3067 full set use `-Wconversion'.
3069 * Use of ISO C style function definitions. This warning
3070 intentionally is _not_ issued for prototype declarations or
3071 variadic functions because these ISO C features will appear
3072 in your code when using libiberty's traditional C
3073 compatibility macros, `PARAMS' and `VPARAMS'. This warning
3074 is also bypassed for nested functions because that feature is
3075 already a GCC extension and thus not relevant to traditional
3078 `-Wdeclaration-after-statement (C only)'
3079 Warn when a declaration is found after a statement in a block.
3080 This construct, known from C++, was introduced with ISO C99 and is
3081 by default allowed in GCC. It is not supported by ISO C90 and was
3082 not supported by GCC versions before GCC 3.0. *Note Mixed
3086 Warn if an undefined identifier is evaluated in an `#if' directive.
3089 Do not warn whenever an `#else' or an `#endif' are followed by
3093 Warn whenever a local variable shadows another local variable,
3094 parameter or global variable or whenever a built-in function is
3098 Warn whenever an object of larger than LEN bytes is defined.
3100 `-Wframe-larger-than-LEN'
3101 Warn whenever the frame size of a function is larger than LEN
3104 `-Wunsafe-loop-optimizations'
3105 Warn if the loop cannot be optimized because the compiler could not
3106 assume anything on the bounds of the loop indices. With
3107 `-funsafe-loop-optimizations' warn if the compiler made such
3111 Warn about anything that depends on the "size of" a function type
3112 or of `void'. GNU C assigns these types a size of 1, for
3113 convenience in calculations with `void *' pointers and pointers to
3116 `-Wbad-function-cast (C only)'
3117 Warn whenever a function call is cast to a non-matching type. For
3118 example, warn if `int malloc()' is cast to `anything *'.
3121 Warn about ISO C constructs that are outside of the common subset
3122 of ISO C and ISO C++, e.g. request for implicit conversion from
3123 `void *' to a pointer to non-`void' type.
3126 Warn whenever a pointer is cast so as to remove a type qualifier
3127 from the target type. For example, warn if a `const char *' is
3128 cast to an ordinary `char *'.
3131 Warn whenever a pointer is cast such that the required alignment
3132 of the target is increased. For example, warn if a `char *' is
3133 cast to an `int *' on machines where integers can only be accessed
3134 at two- or four-byte boundaries.
3137 When compiling C, give string constants the type `const
3138 char[LENGTH]' so that copying the address of one into a
3139 non-`const' `char *' pointer will get a warning; when compiling
3140 C++, warn about the deprecated conversion from string literals to
3141 `char *'. This warning, by default, is enabled for C++ programs.
3142 These warnings will help you find at compile time code that can
3143 try to write into a string constant, but only if you have been
3144 very careful about using `const' in declarations and prototypes.
3145 Otherwise, it will just be a nuisance; this is why we did not make
3146 `-Wall' request these warnings.
3149 Warn if a prototype causes a type conversion that is different
3150 from what would happen to the same argument in the absence of a
3151 prototype. This includes conversions of fixed point to floating
3152 and vice versa, and conversions changing the width or signedness
3153 of a fixed point argument except when the same as the default
3156 Also, warn if a negative integer constant expression is implicitly
3157 converted to an unsigned type. For example, warn about the
3158 assignment `x = -1' if `x' is unsigned. But do not warn about
3159 explicit casts like `(unsigned) -1'.
3162 Warn when a comparison between signed and unsigned values could
3163 produce an incorrect result when the signed value is converted to
3164 unsigned. This warning is also enabled by `-Wextra'; to get the
3165 other warnings of `-Wextra' without this warning, use `-Wextra
3169 Warn about suspicious uses of memory addresses. These include using
3170 the address of a function in a conditional expression, such as
3171 `void func(void); if (func)', and comparisons against the memory
3172 address of a string literal, such as `if (x == "abc")'. Such uses
3173 typically indicate a programmer error: the address of a function
3174 always evaluates to true, so their use in a conditional usually
3175 indicate that the programmer forgot the parentheses in a function
3176 call; and comparisons against string literals result in unspecified
3177 behavior and are not portable in C, so they usually indicate that
3178 the programmer intended to use `strcmp'. This warning is enabled
3181 `-Waggregate-return'
3182 Warn if any functions that return structures or unions are defined
3183 or called. (In languages where you can return an array, this also
3187 Do not warn if an unexpected `__attribute__' is used, such as
3188 unrecognized attributes, function attributes applied to variables,
3189 etc. This will not stop errors for incorrect use of supported
3192 `-Wstrict-prototypes (C only)'
3193 Warn if a function is declared or defined without specifying the
3194 argument types. (An old-style function definition is permitted
3195 without a warning if preceded by a declaration which specifies the
3198 `-Wold-style-definition (C only)'
3199 Warn if an old-style function definition is used. A warning is
3200 given even if there is a previous prototype.
3202 `-Wmissing-prototypes (C only)'
3203 Warn if a global function is defined without a previous prototype
3204 declaration. This warning is issued even if the definition itself
3205 provides a prototype. The aim is to detect global functions that
3206 fail to be declared in header files.
3208 `-Wmissing-declarations (C only)'
3209 Warn if a global function is defined without a previous
3210 declaration. Do so even if the definition itself provides a
3211 prototype. Use this option to detect global functions that are
3212 not declared in header files.
3214 `-Wmissing-field-initializers'
3215 Warn if a structure's initializer has some fields missing. For
3216 example, the following code would cause such a warning, because
3217 `x.h' is implicitly zero:
3219 struct s { int f, g, h; };
3220 struct s x = { 3, 4 };
3222 This option does not warn about designated initializers, so the
3223 following modification would not trigger a warning:
3225 struct s { int f, g, h; };
3226 struct s x = { .f = 3, .g = 4 };
3228 This warning is included in `-Wextra'. To get other `-Wextra'
3229 warnings without this one, use `-Wextra
3230 -Wno-missing-field-initializers'.
3232 `-Wmissing-noreturn'
3233 Warn about functions which might be candidates for attribute
3234 `noreturn'. Note these are only possible candidates, not absolute
3235 ones. Care should be taken to manually verify functions actually
3236 do not ever return before adding the `noreturn' attribute,
3237 otherwise subtle code generation bugs could be introduced. You
3238 will not get a warning for `main' in hosted C environments.
3240 `-Wmissing-format-attribute'
3241 Warn about function pointers which might be candidates for `format'
3242 attributes. Note these are only possible candidates, not absolute
3243 ones. GCC will guess that function pointers with `format'
3244 attributes that are used in assignment, initialization, parameter
3245 passing or return statements should have a corresponding `format'
3246 attribute in the resulting type. I.e. the left-hand side of the
3247 assignment or initialization, the type of the parameter variable,
3248 or the return type of the containing function respectively should
3249 also have a `format' attribute to avoid the warning.
3251 GCC will also warn about function definitions which might be
3252 candidates for `format' attributes. Again, these are only
3253 possible candidates. GCC will guess that `format' attributes
3254 might be appropriate for any function that calls a function like
3255 `vprintf' or `vscanf', but this might not always be the case, and
3256 some functions for which `format' attributes are appropriate may
3260 Do not warn if a multicharacter constant (`'FOOF'') is used.
3261 Usually they indicate a typo in the user's code, as they have
3262 implementation-defined values, and should not be used in portable
3265 `-Wnormalized=<none|id|nfc|nfkc>'
3266 In ISO C and ISO C++, two identifiers are different if they are
3267 different sequences of characters. However, sometimes when
3268 characters outside the basic ASCII character set are used, you can
3269 have two different character sequences that look the same. To
3270 avoid confusion, the ISO 10646 standard sets out some
3271 "normalization rules" which when applied ensure that two sequences
3272 that look the same are turned into the same sequence. GCC can
3273 warn you if you are using identifiers which have not been
3274 normalized; this option controls that warning.
3276 There are four levels of warning that GCC supports. The default is
3277 `-Wnormalized=nfc', which warns about any identifier which is not
3278 in the ISO 10646 "C" normalized form, "NFC". NFC is the
3279 recommended form for most uses.
3281 Unfortunately, there are some characters which ISO C and ISO C++
3282 allow in identifiers that when turned into NFC aren't allowable as
3283 identifiers. That is, there's no way to use these symbols in
3284 portable ISO C or C++ and have all your identifiers in NFC.
3285 `-Wnormalized=id' suppresses the warning for these characters. It
3286 is hoped that future versions of the standards involved will
3287 correct this, which is why this option is not the default.
3289 You can switch the warning off for all characters by writing
3290 `-Wnormalized=none'. You would only want to do this if you were
3291 using some other normalization scheme (like "D"), because
3292 otherwise you can easily create bugs that are literally impossible
3295 Some characters in ISO 10646 have distinct meanings but look
3296 identical in some fonts or display methodologies, especially once
3297 formatting has been applied. For instance `\u207F', "SUPERSCRIPT
3298 LATIN SMALL LETTER N", will display just like a regular `n' which
3299 has been placed in a superscript. ISO 10646 defines the "NFKC"
3300 normalization scheme to convert all these into a standard form as
3301 well, and GCC will warn if your code is not in NFKC if you use
3302 `-Wnormalized=nfkc'. This warning is comparable to warning about
3303 every identifier that contains the letter O because it might be
3304 confused with the digit 0, and so is not the default, but may be
3305 useful as a local coding convention if the programming environment
3306 is unable to be fixed to display these characters distinctly.
3308 `-Wno-deprecated-declarations'
3309 Do not warn about uses of functions (*note Function Attributes::),
3310 variables (*note Variable Attributes::), and types (*note Type
3311 Attributes::) marked as deprecated by using the `deprecated'
3315 Do not warn about compile-time overflow in constant expressions.
3318 Warn if an initialized field without side effects is overridden
3319 when using designated initializers (*note Designated Initializers:
3322 This warning is included in `-Wextra'. To get other `-Wextra'
3323 warnings without this one, use `-Wextra -Wno-override-init'.
3326 Warn if a structure is given the packed attribute, but the packed
3327 attribute has no effect on the layout or size of the structure.
3328 Such structures may be mis-aligned for little benefit. For
3329 instance, in this code, the variable `f.x' in `struct bar' will be
3330 misaligned even though `struct bar' does not itself have the
3336 } __attribute__((packed));
3343 Warn if padding is included in a structure, either to align an
3344 element of the structure or to align the whole structure.
3345 Sometimes when this happens it is possible to rearrange the fields
3346 of the structure to reduce the padding and so make the structure
3350 Warn if anything is declared more than once in the same scope,
3351 even in cases where multiple declaration is valid and changes
3354 `-Wnested-externs (C only)'
3355 Warn if an `extern' declaration is encountered within a function.
3357 `-Wunreachable-code'
3358 Warn if the compiler detects that code will never be executed.
3360 This option is intended to warn when the compiler detects that at
3361 least a whole line of source code will never be executed, because
3362 some condition is never satisfied or because it is after a
3363 procedure that never returns.
3365 It is possible for this option to produce a warning even though
3366 there are circumstances under which part of the affected line can
3367 be executed, so care should be taken when removing
3368 apparently-unreachable code.
3370 For instance, when a function is inlined, a warning may mean that
3371 the line is unreachable in only one inlined copy of the function.
3373 This option is not made part of `-Wall' because in a debugging
3374 version of a program there is often substantial code which checks
3375 correct functioning of the program and is, hopefully, unreachable
3376 because the program does work. Another common use of unreachable
3377 code is to provide behavior which is selectable at compile-time.
3380 Warn if a function can not be inlined and it was declared as
3381 inline. Even with this option, the compiler will not warn about
3382 failures to inline functions declared in system headers.
3384 The compiler uses a variety of heuristics to determine whether or
3385 not to inline a function. For example, the compiler takes into
3386 account the size of the function being inlined and the amount of
3387 inlining that has already been done in the current function.
3388 Therefore, seemingly insignificant changes in the source program
3389 can cause the warnings produced by `-Winline' to appear or
3392 `-Wno-invalid-offsetof (C++ only)'
3393 Suppress warnings from applying the `offsetof' macro to a non-POD
3394 type. According to the 1998 ISO C++ standard, applying `offsetof'
3395 to a non-POD type is undefined. In existing C++ implementations,
3396 however, `offsetof' typically gives meaningful results even when
3397 applied to certain kinds of non-POD types. (Such as a simple
3398 `struct' that fails to be a POD type only by virtue of having a
3399 constructor.) This flag is for users who are aware that they are
3400 writing nonportable code and who have deliberately chosen to
3401 ignore the warning about it.
3403 The restrictions on `offsetof' may be relaxed in a future version
3404 of the C++ standard.
3406 `-Wno-int-to-pointer-cast (C only)'
3407 Suppress warnings from casts to pointer type of an integer of a
3410 `-Wno-pointer-to-int-cast (C only)'
3411 Suppress warnings from casts from a pointer to an integer type of a
3415 Warn if a precompiled header (*note Precompiled Headers::) is
3416 found in the search path but can't be used.
3419 Warn if `long long' type is used. This is default. To inhibit
3420 the warning messages, use `-Wno-long-long'. Flags `-Wlong-long'
3421 and `-Wno-long-long' are taken into account only when `-pedantic'
3425 Warn if variadic macros are used in pedantic ISO C90 mode, or the
3426 GNU alternate syntax when in pedantic ISO C99 mode. This is
3427 default. To inhibit the warning messages, use
3428 `-Wno-variadic-macros'.
3431 Warn if variable length array is used in the code. `-Wno-vla'
3432 will prevent the `-pedantic' warning of the variable length array.
3434 `-Wvolatile-register-var'
3435 Warn if a register variable is declared volatile. The volatile
3436 modifier does not inhibit all optimizations that may eliminate
3437 reads and/or writes to register variables.
3439 `-Wdisabled-optimization'
3440 Warn if a requested optimization pass is disabled. This warning
3441 does not generally indicate that there is anything wrong with your
3442 code; it merely indicates that GCC's optimizers were unable to
3443 handle the code effectively. Often, the problem is that your code
3444 is too big or too complex; GCC will refuse to optimize programs
3445 when the optimization itself is likely to take inordinate amounts
3449 Warn for pointer argument passing or assignment with different
3450 signedness. This option is only supported for C and Objective-C.
3451 It is implied by `-Wall' and by `-pedantic', which can be disabled
3452 with `-Wno-pointer-sign'.
3455 Make all warnings into errors.
3458 Make the specified warning into an errors. The specifier for a
3459 warning is appended, for example `-Werror=switch' turns the
3460 warnings controlled by `-Wswitch' into errors. This switch takes
3461 a negative form, to be used to negate `-Werror' for specific
3462 warnings, for example `-Wno-error=switch' makes `-Wswitch'
3463 warnings not be errors, even when `-Werror' is in effect. You can
3464 use the `-fdiagnostics-show-option' option to have each
3465 controllable warning amended with the option which controls it, to
3466 determine what to use with this option.
3468 Note that specifying `-Werror='FOO automatically implies `-W'FOO.
3469 However, `-Wno-error='FOO does not imply anything.
3472 This option is only active when `-fstack-protector' is active. It
3473 warns about functions that will not be protected against stack
3476 `-Woverlength-strings'
3477 Warn about string constants which are longer than the "minimum
3478 maximum" length specified in the C standard. Modern compilers
3479 generally allow string constants which are much longer than the
3480 standard's minimum limit, but very portable programs should avoid
3481 using longer strings.
3483 The limit applies _after_ string constant concatenation, and does
3484 not count the trailing NUL. In C89, the limit was 509 characters;
3485 in C99, it was raised to 4095. C++98 does not specify a normative
3486 minimum maximum, so we do not diagnose overlength strings in C++.
3488 This option is implied by `-pedantic', and can be disabled with
3489 `-Wno-overlength-strings'.
3492 File: gcc.info, Node: Debugging Options, Next: Optimize Options, Prev: Warning Options, Up: Invoking GCC
3494 3.9 Options for Debugging Your Program or GCC
3495 =============================================
3497 GCC has various special options that are used for debugging either your
3501 Produce debugging information in the operating system's native
3502 format (stabs, COFF, XCOFF, or DWARF 2). GDB can work with this
3503 debugging information.
3505 On most systems that use stabs format, `-g' enables use of extra
3506 debugging information that only GDB can use; this extra information
3507 makes debugging work better in GDB but will probably make other
3508 debuggers crash or refuse to read the program. If you want to
3509 control for certain whether to generate the extra information, use
3510 `-gstabs+', `-gstabs', `-gxcoff+', `-gxcoff', or `-gvms' (see
3513 GCC allows you to use `-g' with `-O'. The shortcuts taken by
3514 optimized code may occasionally produce surprising results: some
3515 variables you declared may not exist at all; flow of control may
3516 briefly move where you did not expect it; some statements may not
3517 be executed because they compute constant results or their values
3518 were already at hand; some statements may execute in different
3519 places because they were moved out of loops.
3521 Nevertheless it proves possible to debug optimized output. This
3522 makes it reasonable to use the optimizer for programs that might
3525 The following options are useful when GCC is generated with the
3526 capability for more than one debugging format.
3529 Produce debugging information for use by GDB. This means to use
3530 the most expressive format available (DWARF 2, stabs, or the
3531 native format if neither of those are supported), including GDB
3532 extensions if at all possible.
3535 Produce debugging information in stabs format (if that is
3536 supported), without GDB extensions. This is the format used by
3537 DBX on most BSD systems. On MIPS, Alpha and System V Release 4
3538 systems this option produces stabs debugging output which is not
3539 understood by DBX or SDB. On System V Release 4 systems this
3540 option requires the GNU assembler.
3542 `-feliminate-unused-debug-symbols'
3543 Produce debugging information in stabs format (if that is
3544 supported), for only symbols that are actually used.
3546 `-femit-class-debug-always'
3547 Instead of emitting debugging information for a C++ class in only
3548 one object file, emit it in all object files using the class.
3549 This option should be used only with debuggers that are unable to
3550 handle the way GCC normally emits debugging information for
3551 classes because using this option will increase the size of
3552 debugging information by as much as a factor of two.
3555 Produce debugging information in stabs format (if that is
3556 supported), using GNU extensions understood only by the GNU
3557 debugger (GDB). The use of these extensions is likely to make
3558 other debuggers crash or refuse to read the program.
3561 Produce debugging information in COFF format (if that is
3562 supported). This is the format used by SDB on most System V
3563 systems prior to System V Release 4.
3566 Produce debugging information in XCOFF format (if that is
3567 supported). This is the format used by the DBX debugger on IBM
3571 Produce debugging information in XCOFF format (if that is
3572 supported), using GNU extensions understood only by the GNU
3573 debugger (GDB). The use of these extensions is likely to make
3574 other debuggers crash or refuse to read the program, and may cause
3575 assemblers other than the GNU assembler (GAS) to fail with an
3579 Produce debugging information in DWARF version 2 format (if that is
3580 supported). This is the format used by DBX on IRIX 6. With this
3581 option, GCC uses features of DWARF version 3 when they are useful;
3582 version 3 is upward compatible with version 2, but may still cause
3583 problems for older debuggers.
3586 Produce debugging information in VMS debug format (if that is
3587 supported). This is the format used by DEBUG on VMS systems.
3595 Request debugging information and also use LEVEL to specify how
3596 much information. The default level is 2.
3598 Level 1 produces minimal information, enough for making backtraces
3599 in parts of the program that you don't plan to debug. This
3600 includes descriptions of functions and external variables, but no
3601 information about local variables and no line numbers.
3603 Level 3 includes extra information, such as all the macro
3604 definitions present in the program. Some debuggers support macro
3605 expansion when you use `-g3'.
3607 `-gdwarf-2' does not accept a concatenated debug level, because
3608 GCC used to support an option `-gdwarf' that meant to generate
3609 debug information in version 1 of the DWARF format (which is very
3610 different from version 2), and it would have been too confusing.
3611 That debug format is long obsolete, but the option cannot be
3612 changed now. Instead use an additional `-gLEVEL' option to change
3613 the debug level for DWARF2.
3615 `-feliminate-dwarf2-dups'
3616 Compress DWARF2 debugging information by eliminating duplicated
3617 information about each symbol. This option only makes sense when
3618 generating DWARF2 debugging information with `-gdwarf-2'.
3620 `-femit-struct-debug-baseonly'
3621 Emit debug information for struct-like types only when the base
3622 name of the compilation source file matches the base name of file
3623 in which the struct was defined.
3625 This option substantially reduces the size of debugging
3626 information, but at significant potential loss in type information
3627 to the debugger. See `-femit-struct-debug-reduced' for a less
3628 aggressive option. See `-femit-struct-debug-detailed' for more
3631 This option works only with DWARF 2.
3633 `-femit-struct-debug-reduced'
3634 Emit debug information for struct-like types only when the base
3635 name of the compilation source file matches the base name of file
3636 in which the type was defined, unless the struct is a template or
3637 defined in a system header.
3639 This option significantly reduces the size of debugging
3640 information, with some potential loss in type information to the
3641 debugger. See `-femit-struct-debug-baseonly' for a more
3642 aggressive option. See `-femit-struct-debug-detailed' for more
3645 This option works only with DWARF 2.
3647 `-femit-struct-debug-detailed[=SPEC-LIST]'
3648 Specify the struct-like types for which the compiler will generate
3649 debug information. The intent is to reduce duplicate struct debug
3650 information between different object files within the same program.
3652 This option is a detailed version of `-femit-struct-debug-reduced'
3653 and `-femit-struct-debug-baseonly', which will serve for most
3656 A specification has the syntax
3657 [`dir:'|`ind:'][`ord:'|`gen:'](`any'|`sys'|`base'|`none')
3659 The optional first word limits the specification to structs that
3660 are used directly (`dir:') or used indirectly (`ind:'). A struct
3661 type is used directly when it is the type of a variable, member.
3662 Indirect uses arise through pointers to structs. That is, when
3663 use of an incomplete struct would be legal, the use is indirect.
3664 An example is `struct one direct; struct two * indirect;'.
3666 The optional second word limits the specification to ordinary
3667 structs (`ord:') or generic structs (`gen:'). Generic structs are
3668 a bit complicated to explain. For C++, these are non-explicit
3669 specializations of template classes, or non-template classes
3670 within the above. Other programming languages have generics, but
3671 `-femit-struct-debug-detailed' does not yet implement them.
3673 The third word specifies the source files for those structs for
3674 which the compiler will emit debug information. The values `none'
3675 and `any' have the normal meaning. The value `base' means that
3676 the base of name of the file in which the type declaration appears
3677 must match the base of the name of the main compilation file. In
3678 practice, this means that types declared in `foo.c' and `foo.h'
3679 will have debug information, but types declared in other header
3680 will not. The value `sys' means those types satisfying `base' or
3681 declared in system or compiler headers.
3683 You may need to experiment to determine the best settings for your
3686 The default is `-femit-struct-debug-detailed=all'.
3688 This option works only with DWARF 2.
3691 Generate extra code to write profile information suitable for the
3692 analysis program `prof'. You must use this option when compiling
3693 the source files you want data about, and you must also use it when
3697 Generate extra code to write profile information suitable for the
3698 analysis program `gprof'. You must use this option when compiling
3699 the source files you want data about, and you must also use it when
3703 Makes the compiler print out each function name as it is compiled,
3704 and print some statistics about each pass when it finishes.
3707 Makes the compiler print some statistics about the time consumed
3708 by each pass when it finishes.
3711 Makes the compiler print some statistics about permanent memory
3712 allocation when it finishes.
3715 Add code so that program flow "arcs" are instrumented. During
3716 execution the program records how many times each branch and call
3717 is executed and how many times it is taken or returns. When the
3718 compiled program exits it saves this data to a file called
3719 `AUXNAME.gcda' for each source file. The data may be used for
3720 profile-directed optimizations (`-fbranch-probabilities'), or for
3721 test coverage analysis (`-ftest-coverage'). Each object file's
3722 AUXNAME is generated from the name of the output file, if
3723 explicitly specified and it is not the final executable, otherwise
3724 it is the basename of the source file. In both cases any suffix
3725 is removed (e.g. `foo.gcda' for input file `dir/foo.c', or
3726 `dir/foo.gcda' for output file specified as `-o dir/foo.o').
3727 *Note Cross-profiling::.
3730 This option is used to compile and link code instrumented for
3731 coverage analysis. The option is a synonym for `-fprofile-arcs'
3732 `-ftest-coverage' (when compiling) and `-lgcov' (when linking).
3733 See the documentation for those options for more details.
3735 * Compile the source files with `-fprofile-arcs' plus
3736 optimization and code generation options. For test coverage
3737 analysis, use the additional `-ftest-coverage' option. You
3738 do not need to profile every source file in a program.
3740 * Link your object files with `-lgcov' or `-fprofile-arcs' (the
3741 latter implies the former).
3743 * Run the program on a representative workload to generate the
3744 arc profile information. This may be repeated any number of
3745 times. You can run concurrent instances of your program, and
3746 provided that the file system supports locking, the data
3747 files will be correctly updated. Also `fork' calls are
3748 detected and correctly handled (double counting will not
3751 * For profile-directed optimizations, compile the source files
3752 again with the same optimization and code generation options
3753 plus `-fbranch-probabilities' (*note Options that Control
3754 Optimization: Optimize Options.).
3756 * For test coverage analysis, use `gcov' to produce human
3757 readable information from the `.gcno' and `.gcda' files.
3758 Refer to the `gcov' documentation for further information.
3761 With `-fprofile-arcs', for each function of your program GCC
3762 creates a program flow graph, then finds a spanning tree for the
3763 graph. Only arcs that are not on the spanning tree have to be
3764 instrumented: the compiler adds code to count the number of times
3765 that these arcs are executed. When an arc is the only exit or
3766 only entrance to a block, the instrumentation code can be added to
3767 the block; otherwise, a new basic block must be created to hold
3768 the instrumentation code.
3771 Produce a notes file that the `gcov' code-coverage utility (*note
3772 `gcov'--a Test Coverage Program: Gcov.) can use to show program
3773 coverage. Each source file's note file is called `AUXNAME.gcno'.
3774 Refer to the `-fprofile-arcs' option above for a description of
3775 AUXNAME and instructions on how to generate test coverage data.
3776 Coverage data will match the source files more closely, if you do
3782 Says to make debugging dumps during compilation at times specified
3783 by LETTERS. This is used for debugging the RTL-based passes of
3784 the compiler. The file names for most of the dumps are made by
3785 appending a pass number and a word to the DUMPNAME. DUMPNAME is
3786 generated from the name of the output file, if explicitly
3787 specified and it is not an executable, otherwise it is the
3788 basename of the source file.
3790 Most debug dumps can be enabled either passing a letter to the `-d'
3791 option, or with a long `-fdump-rtl' switch; here are the possible
3792 letters for use in LETTERS and PASS, and their meanings:
3795 Annotate the assembler output with miscellaneous debugging
3800 Dump after block reordering, to `FILE.148r.bbro'.
3803 `-fdump-rtl-combine'
3804 Dump after instruction combination, to the file
3805 `FILE.129r.combine'.
3810 `-dC' and `-fdump-rtl-ce1' enable dumping after the first if
3811 conversion, to the file `FILE.117r.ce1'. `-dC' and
3812 `-fdump-rtl-ce2' enable dumping after the second if
3813 conversion, to the file `FILE.130r.ce2'.
3818 `-dd' and `-fdump-rtl-btl' enable dumping after branch target
3819 load optimization, to `FILE.31.btl'. `-dd' and
3820 `-fdump-rtl-dbr' enable dumping after delayed branch
3821 scheduling, to `FILE.36.dbr'.
3824 Dump all macro definitions, at the end of preprocessing, in
3825 addition to normal output.
3829 Dump after the third if conversion, to `FILE.146r.ce3'.
3834 `-df' and `-fdump-rtl-cfg' enable dumping after control and
3835 data flow analysis, to `FILE.116r.cfg'. `-df' and
3836 `-fdump-rtl-cfg' enable dumping dump after life analysis, to
3837 `FILE.128r.life1' and `FILE.135r.life2'.
3841 Dump after global register allocation, to `FILE.139r.greg'.
3846 `-dG' and `-fdump-rtl-gcse' enable dumping after GCSE, to
3847 `FILE.114r.gcse'. `-dG' and `-fdump-rtl-bypass' enable
3848 dumping after jump bypassing and control flow optimizations,
3849 to `FILE.115r.bypass'.
3853 Dump after finalization of EH handling code, to `FILE.02.eh'.
3856 `-fdump-rtl-sibling'
3857 Dump after sibling call optimizations, to `FILE.106r.sibling'.
3861 Dump after the first jump optimization, to `FILE.112r.jump'.
3865 Dump after conversion from registers to stack, to
3870 Dump after local register allocation, to `FILE.138r.lreg'.
3874 `-dL' and `-fdump-rtl-loop2' enable dumping after the loop
3875 optimization pass, to `FILE.119r.loop2',
3876 `FILE.120r.loop2_init', `FILE.121r.loop2_invariant', and
3877 `FILE.125r.loop2_done'.
3881 Dump after modulo scheduling, to `FILE.136r.sms'.
3885 Dump after performing the machine dependent reorganization
3886 pass, to `FILE.155r.mach'.
3890 Dump after register renumbering, to `FILE.147r.rnreg'.
3893 `-fdump-rtl-regmove'
3894 Dump after the register move pass, to `FILE.132r.regmove'.
3897 `-fdump-rtl-postreload'
3898 Dump after post-reload optimizations, to `FILE.24.postreload'.
3902 Dump after RTL generation, to `FILE.104r.expand'.
3906 Dump after the second scheduling pass, to `FILE.150r.sched2'.
3910 Dump after CSE (including the jump optimization that
3911 sometimes follows CSE), to `FILE.113r.cse'.
3915 Dump after the first scheduling pass, to `FILE.21.sched'.
3919 Dump after the second CSE pass (including the jump
3920 optimization that sometimes follows CSE), to `FILE.127r.cse2'.
3924 Dump after running tracer, to `FILE.118r.tracer'.
3928 `-fdump-rtl-vartrack'
3929 `-dV' and `-fdump-rtl-vpt' enable dumping after the value
3930 profile transformations, to `FILE.10.vpt'. `-dV' and
3931 `-fdump-rtl-vartrack' enable dumping after variable tracking,
3932 to `FILE.154r.vartrack'.
3936 Dump after the second flow pass, to `FILE.142r.flow2'.
3939 `-fdump-rtl-peephole2'
3940 Dump after the peephole pass, to `FILE.145r.peephole2'.
3944 Dump after live range splitting, to `FILE.126r.web'.
3948 Produce all the dumps listed above.
3951 Produce a core dump whenever an error occurs.
3954 Print statistics on memory usage, at the end of the run, to
3958 Annotate the assembler output with a comment indicating which
3959 pattern and alternative was used. The length of each
3960 instruction is also printed.
3963 Dump the RTL in the assembler output as a comment before each
3964 instruction. Also turns on `-dp' annotation.
3967 For each of the other indicated dump files (either with `-d'
3968 or `-fdump-rtl-PASS'), dump a representation of the control
3969 flow graph suitable for viewing with VCG to `FILE.PASS.vcg'.
3972 Just generate RTL for a function instead of compiling it.
3973 Usually used with `r' (`-fdump-rtl-expand').
3976 Dump debugging information during parsing, to standard error.
3979 When doing debugging dumps (see `-d' option above), suppress
3980 address output. This makes it more feasible to use diff on
3981 debugging dumps for compiler invocations with different compiler
3982 binaries and/or different text / bss / data / heap / stack / dso
3986 When doing debugging dumps (see `-d' option above), suppress
3987 instruction numbers, line number note and address output. This
3988 makes it more feasible to use diff on debugging dumps for compiler
3989 invocations with different options, in particular with and without
3992 `-fdump-translation-unit (C++ only)'
3993 `-fdump-translation-unit-OPTIONS (C++ only)'
3994 Dump a representation of the tree structure for the entire
3995 translation unit to a file. The file name is made by appending
3996 `.tu' to the source file name. If the `-OPTIONS' form is used,
3997 OPTIONS controls the details of the dump as described for the
3998 `-fdump-tree' options.
4000 `-fdump-class-hierarchy (C++ only)'
4001 `-fdump-class-hierarchy-OPTIONS (C++ only)'
4002 Dump a representation of each class's hierarchy and virtual
4003 function table layout to a file. The file name is made by
4004 appending `.class' to the source file name. If the `-OPTIONS'
4005 form is used, OPTIONS controls the details of the dump as
4006 described for the `-fdump-tree' options.
4009 Control the dumping at various stages of inter-procedural analysis
4010 language tree to a file. The file name is generated by appending
4011 a switch specific suffix to the source file name. The following
4015 Enables all inter-procedural analysis dumps; currently the
4016 only produced dump is the `cgraph' dump.
4019 Dumps information about call-graph optimization, unused
4020 function removal, and inlining decisions.
4022 `-fdump-tree-SWITCH'
4023 `-fdump-tree-SWITCH-OPTIONS'
4024 Control the dumping at various stages of processing the
4025 intermediate language tree to a file. The file name is generated
4026 by appending a switch specific suffix to the source file name. If
4027 the `-OPTIONS' form is used, OPTIONS is a list of `-' separated
4028 options that control the details of the dump. Not all options are
4029 applicable to all dumps, those which are not meaningful will be
4030 ignored. The following options are available
4033 Print the address of each node. Usually this is not
4034 meaningful as it changes according to the environment and
4035 source file. Its primary use is for tying up a dump file
4036 with a debug environment.
4039 Inhibit dumping of members of a scope or body of a function
4040 merely because that scope has been reached. Only dump such
4041 items when they are directly reachable by some other path.
4042 When dumping pretty-printed trees, this option inhibits
4043 dumping the bodies of control structures.
4046 Print a raw representation of the tree. By default, trees are
4047 pretty-printed into a C-like representation.
4050 Enable more detailed dumps (not honored by every dump option).
4053 Enable dumping various statistics about the pass (not honored
4054 by every dump option).
4057 Enable showing basic block boundaries (disabled in raw dumps).
4060 Enable showing virtual operands for every statement.
4063 Enable showing line numbers for statements.
4066 Enable showing the unique ID (`DECL_UID') for each variable.
4069 Turn on all options, except `raw', `slim' and `lineno'.
4071 The following tree dumps are possible:
4073 Dump before any tree based optimization, to `FILE.original'.
4076 Dump after all tree based optimization, to `FILE.optimized'.
4079 Dump after function inlining, to `FILE.inlined'.
4082 Dump each function before and after the gimplification pass
4083 to a file. The file name is made by appending `.gimple' to
4084 the source file name.
4087 Dump the control flow graph of each function to a file. The
4088 file name is made by appending `.cfg' to the source file name.
4091 Dump the control flow graph of each function to a file in VCG
4092 format. The file name is made by appending `.vcg' to the
4093 source file name. Note that if the file contains more than
4094 one function, the generated file cannot be used directly by
4095 VCG. You will need to cut and paste each function's graph
4096 into its own separate file first.
4099 Dump each function after copying loop headers. The file name
4100 is made by appending `.ch' to the source file name.
4103 Dump SSA related information to a file. The file name is
4104 made by appending `.ssa' to the source file name.
4107 Dump structure aliasing variable information to a file. This
4108 file name is made by appending `.salias' to the source file
4112 Dump aliasing information for each function. The file name
4113 is made by appending `.alias' to the source file name.
4116 Dump each function after CCP. The file name is made by
4117 appending `.ccp' to the source file name.
4120 Dump each function after STORE-CCP. The file name is made by
4121 appending `.storeccp' to the source file name.
4124 Dump trees after partial redundancy elimination. The file
4125 name is made by appending `.pre' to the source file name.
4128 Dump trees after full redundancy elimination. The file name
4129 is made by appending `.fre' to the source file name.
4132 Dump trees after copy propagation. The file name is made by
4133 appending `.copyprop' to the source file name.
4136 Dump trees after store copy-propagation. The file name is
4137 made by appending `.store_copyprop' to the source file name.
4140 Dump each function after dead code elimination. The file
4141 name is made by appending `.dce' to the source file name.
4144 Dump each function after adding mudflap instrumentation. The
4145 file name is made by appending `.mudflap' to the source file
4149 Dump each function after performing scalar replacement of
4150 aggregates. The file name is made by appending `.sra' to the
4154 Dump each function after performing code sinking. The file
4155 name is made by appending `.sink' to the source file name.
4158 Dump each function after applying dominator tree
4159 optimizations. The file name is made by appending `.dom' to
4160 the source file name.
4163 Dump each function after applying dead store elimination.
4164 The file name is made by appending `.dse' to the source file
4168 Dump each function after optimizing PHI nodes into
4169 straightline code. The file name is made by appending
4170 `.phiopt' to the source file name.
4173 Dump each function after forward propagating single use
4174 variables. The file name is made by appending `.forwprop' to
4175 the source file name.
4178 Dump each function after applying the copy rename
4179 optimization. The file name is made by appending
4180 `.copyrename' to the source file name.
4183 Dump each function after applying the named return value
4184 optimization on generic trees. The file name is made by
4185 appending `.nrv' to the source file name.
4188 Dump each function after applying vectorization of loops.
4189 The file name is made by appending `.vect' to the source file
4193 Dump each function after Value Range Propagation (VRP). The
4194 file name is made by appending `.vrp' to the source file name.
4197 Enable all the available tree dumps with the flags provided
4200 `-ftree-vectorizer-verbose=N'
4201 This option controls the amount of debugging output the vectorizer
4202 prints. This information is written to standard error, unless
4203 `-fdump-tree-all' or `-fdump-tree-vect' is specified, in which
4204 case it is output to the usual dump listing file, `.vect'. For
4205 N=0 no diagnostic information is reported. If N=1 the vectorizer
4206 reports each loop that got vectorized, and the total number of
4207 loops that got vectorized. If N=2 the vectorizer also reports
4208 non-vectorized loops that passed the first analysis phase
4209 (vect_analyze_loop_form) - i.e. countable, inner-most, single-bb,
4210 single-entry/exit loops. This is the same verbosity level that
4211 `-fdump-tree-vect-stats' uses. Higher verbosity levels mean
4212 either more information dumped for each reported loop, or same
4213 amount of information reported for more loops: If N=3, alignment
4214 related information is added to the reports. If N=4,
4215 data-references related information (e.g. memory dependences,
4216 memory access-patterns) is added to the reports. If N=5, the
4217 vectorizer reports also non-vectorized inner-most loops that did
4218 not pass the first analysis phase (i.e. may not be countable, or
4219 may have complicated control-flow). If N=6, the vectorizer
4220 reports also non-vectorized nested loops. For N=7, all the
4221 information the vectorizer generates during its analysis and
4222 transformation is reported. This is the same verbosity level that
4223 `-fdump-tree-vect-details' uses.
4225 `-frandom-seed=STRING'
4226 This option provides a seed that GCC uses when it would otherwise
4227 use random numbers. It is used to generate certain symbol names
4228 that have to be different in every compiled file. It is also used
4229 to place unique stamps in coverage data files and the object files
4230 that produce them. You can use the `-frandom-seed' option to
4231 produce reproducibly identical object files.
4233 The STRING should be different for every file you compile.
4236 On targets that use instruction scheduling, this option controls
4237 the amount of debugging output the scheduler prints. This
4238 information is written to standard error, unless `-dS' or `-dR' is
4239 specified, in which case it is output to the usual dump listing
4240 file, `.sched' or `.sched2' respectively. However for N greater
4241 than nine, the output is always printed to standard error.
4243 For N greater than zero, `-fsched-verbose' outputs the same
4244 information as `-dRS'. For N greater than one, it also output
4245 basic block probabilities, detailed ready list information and
4246 unit/insn info. For N greater than two, it includes RTL at abort
4247 point, control-flow and regions info. And for N over four,
4248 `-fsched-verbose' also includes dependence info.
4251 Store the usual "temporary" intermediate files permanently; place
4252 them in the current directory and name them based on the source
4253 file. Thus, compiling `foo.c' with `-c -save-temps' would produce
4254 files `foo.i' and `foo.s', as well as `foo.o'. This creates a
4255 preprocessed `foo.i' output file even though the compiler now
4256 normally uses an integrated preprocessor.
4258 When used in combination with the `-x' command line option,
4259 `-save-temps' is sensible enough to avoid over writing an input
4260 source file with the same extension as an intermediate file. The
4261 corresponding intermediate file may be obtained by renaming the
4262 source file before using `-save-temps'.
4265 Report the CPU time taken by each subprocess in the compilation
4266 sequence. For C source files, this is the compiler proper and
4267 assembler (plus the linker if linking is done). The output looks
4273 The first number on each line is the "user time", that is time
4274 spent executing the program itself. The second number is "system
4275 time", time spent executing operating system routines on behalf of
4276 the program. Both numbers are in seconds.
4279 Run variable tracking pass. It computes where variables are
4280 stored at each position in code. Better debugging information is
4281 then generated (if the debugging information format supports this
4284 It is enabled by default when compiling with optimization (`-Os',
4285 `-O', `-O2', ...), debugging information (`-g') and the debug info
4288 `-print-file-name=LIBRARY'
4289 Print the full absolute name of the library file LIBRARY that
4290 would be used when linking--and don't do anything else. With this
4291 option, GCC does not compile or link anything; it just prints the
4294 `-print-multi-directory'
4295 Print the directory name corresponding to the multilib selected by
4296 any other switches present in the command line. This directory is
4297 supposed to exist in `GCC_EXEC_PREFIX'.
4300 Print the mapping from multilib directory names to compiler
4301 switches that enable them. The directory name is separated from
4302 the switches by `;', and each switch starts with an `@' instead of
4303 the `-', without spaces between multiple switches. This is
4304 supposed to ease shell-processing.
4306 `-print-prog-name=PROGRAM'
4307 Like `-print-file-name', but searches for a program such as `cpp'.
4309 `-print-libgcc-file-name'
4310 Same as `-print-file-name=libgcc.a'.
4312 This is useful when you use `-nostdlib' or `-nodefaultlibs' but
4313 you do want to link with `libgcc.a'. You can do
4315 gcc -nostdlib FILES... `gcc -print-libgcc-file-name`
4317 `-print-search-dirs'
4318 Print the name of the configured installation directory and a list
4319 of program and library directories `gcc' will search--and don't do
4322 This is useful when `gcc' prints the error message `installation
4323 problem, cannot exec cpp0: No such file or directory'. To resolve
4324 this you either need to put `cpp0' and the other compiler
4325 components where `gcc' expects to find them, or you can set the
4326 environment variable `GCC_EXEC_PREFIX' to the directory where you
4327 installed them. Don't forget the trailing `/'. *Note Environment
4331 Print the compiler's target machine (for example,
4332 `i686-pc-linux-gnu')--and don't do anything else.
4335 Print the compiler version (for example, `3.0')--and don't do
4339 Print the compiler's built-in specs--and don't do anything else.
4340 (This is used when GCC itself is being built.) *Note Spec Files::.
4342 `-feliminate-unused-debug-types'
4343 Normally, when producing DWARF2 output, GCC will emit debugging
4344 information for all types declared in a compilation unit,
4345 regardless of whether or not they are actually used in that
4346 compilation unit. Sometimes this is useful, such as if, in the
4347 debugger, you want to cast a value to a type that is not actually
4348 used in your program (but is declared). More often, however, this
4349 results in a significant amount of wasted space. With this
4350 option, GCC will avoid producing debug symbol output for types
4351 that are nowhere used in the source file being compiled.
4354 File: gcc.info, Node: Optimize Options, Next: Preprocessor Options, Prev: Debugging Options, Up: Invoking GCC
4356 3.10 Options That Control Optimization
4357 ======================================
4359 These options control various sorts of optimizations.
4361 Without any optimization option, the compiler's goal is to reduce the
4362 cost of compilation and to make debugging produce the expected results.
4363 Statements are independent: if you stop the program with a breakpoint
4364 between statements, you can then assign a new value to any variable or
4365 change the program counter to any other statement in the function and
4366 get exactly the results you would expect from the source code.
4368 Turning on optimization flags makes the compiler attempt to improve
4369 the performance and/or code size at the expense of compilation time and
4370 possibly the ability to debug the program.
4372 The compiler performs optimization based on the knowledge it has of
4373 the program. Optimization levels `-O' and above, in particular, enable
4374 _unit-at-a-time_ mode, which allows the compiler to consider
4375 information gained from later functions in the file when compiling a
4376 function. Compiling multiple files at once to a single output file in
4377 _unit-at-a-time_ mode allows the compiler to use information gained
4378 from all of the files when compiling each of them.
4380 Not all optimizations are controlled directly by a flag. Only
4381 optimizations that have a flag are listed.
4385 Optimize. Optimizing compilation takes somewhat more time, and a
4386 lot more memory for a large function.
4388 With `-O', the compiler tries to reduce code size and execution
4389 time, without performing any optimizations that take a great deal
4390 of compilation time.
4392 `-O' turns on the following optimization flags:
4395 -fguess-branch-probability
4401 -ftree-dominator-opts
4412 `-O' also turns on `-fomit-frame-pointer' on machines where doing
4413 so does not interfere with debugging.
4416 Optimize even more. GCC performs nearly all supported
4417 optimizations that do not involve a space-speed tradeoff. The
4418 compiler does not perform loop unrolling or function inlining when
4419 you specify `-O2'. As compared to `-O', this option increases
4420 both compilation time and the performance of the generated code.
4422 `-O2' turns on all optimization flags specified by `-O'. It also
4423 turns on the following optimization flags:
4426 -foptimize-sibling-calls
4427 -fcse-follow-jumps -fcse-skip-blocks
4429 -fexpensive-optimizations
4430 -frerun-cse-after-loop
4433 -fschedule-insns -fschedule-insns2
4434 -fsched-interblock -fsched-spec
4436 -fstrict-aliasing -fstrict-overflow
4437 -fdelete-null-pointer-checks
4438 -freorder-blocks -freorder-functions
4439 -falign-functions -falign-jumps
4440 -falign-loops -falign-labels
4444 Please note the warning under `-fgcse' about invoking `-O2' on
4445 programs that use computed gotos.
4447 `-O2' doesn't turn on `-ftree-vrp' for the Ada compiler. This
4448 option must be explicitly specified on the command line to be
4449 enabled for the Ada compiler.
4452 Optimize yet more. `-O3' turns on all optimizations specified by
4453 `-O2' and also turns on the `-finline-functions',
4454 `-funswitch-loops' and `-fgcse-after-reload' options.
4457 Do not optimize. This is the default.
4460 Optimize for size. `-Os' enables all `-O2' optimizations that do
4461 not typically increase code size. It also performs further
4462 optimizations designed to reduce code size.
4464 `-Os' disables the following optimization flags:
4465 -falign-functions -falign-jumps -falign-loops
4466 -falign-labels -freorder-blocks -freorder-blocks-and-partition
4467 -fprefetch-loop-arrays -ftree-vect-loop-version
4469 If you use multiple `-O' options, with or without level numbers,
4470 the last such option is the one that is effective.
4472 Options of the form `-fFLAG' specify machine-independent flags. Most
4473 flags have both positive and negative forms; the negative form of
4474 `-ffoo' would be `-fno-foo'. In the table below, only one of the forms
4475 is listed--the one you typically will use. You can figure out the
4476 other form by either removing `no-' or adding it.
4478 The following options control specific optimizations. They are either
4479 activated by `-O' options or are related to ones that are. You can use
4480 the following flags in the rare cases when "fine-tuning" of
4481 optimizations to be performed is desired.
4483 `-fno-default-inline'
4484 Do not make member functions inline by default merely because they
4485 are defined inside the class scope (C++ only). Otherwise, when
4486 you specify `-O', member functions defined inside class scope are
4487 compiled inline by default; i.e., you don't need to add `inline'
4488 in front of the member function name.
4491 Always pop the arguments to each function call as soon as that
4492 function returns. For machines which must pop arguments after a
4493 function call, the compiler normally lets arguments accumulate on
4494 the stack for several function calls and pops them all at once.
4496 Disabled at levels `-O', `-O2', `-O3', `-Os'.
4499 Force memory operands to be copied into registers before doing
4500 arithmetic on them. This produces better code by making all memory
4501 references potential common subexpressions. When they are not
4502 common subexpressions, instruction combination should eliminate
4503 the separate register-load. This option is now a nop and will be
4507 Force memory address constants to be copied into registers before
4508 doing arithmetic on them.
4510 `-fomit-frame-pointer'
4511 Don't keep the frame pointer in a register for functions that
4512 don't need one. This avoids the instructions to save, set up and
4513 restore frame pointers; it also makes an extra register available
4514 in many functions. *It also makes debugging impossible on some
4517 On some machines, such as the VAX, this flag has no effect, because
4518 the standard calling sequence automatically handles the frame
4519 pointer and nothing is saved by pretending it doesn't exist. The
4520 machine-description macro `FRAME_POINTER_REQUIRED' controls
4521 whether a target machine supports this flag. *Note Register
4522 Usage: (gccint)Registers.
4524 Enabled at levels `-O', `-O2', `-O3', `-Os'.
4526 `-foptimize-sibling-calls'
4527 Optimize sibling and tail recursive calls.
4529 Enabled at levels `-O2', `-O3', `-Os'.
4532 Don't pay attention to the `inline' keyword. Normally this option
4533 is used to keep the compiler from expanding any functions inline.
4534 Note that if you are not optimizing, no functions can be expanded
4537 `-finline-functions'
4538 Integrate all simple functions into their callers. The compiler
4539 heuristically decides which functions are simple enough to be worth
4540 integrating in this way.
4542 If all calls to a given function are integrated, and the function
4543 is declared `static', then the function is normally not output as
4544 assembler code in its own right.
4546 Enabled at level `-O3'.
4548 `-finline-functions-called-once'
4549 Consider all `static' functions called once for inlining into their
4550 caller even if they are not marked `inline'. If a call to a given
4551 function is integrated, then the function is not output as
4552 assembler code in its own right.
4554 Enabled if `-funit-at-a-time' is enabled.
4557 Inline functions marked by `always_inline' and functions whose
4558 body seems smaller than the function call overhead early before
4559 doing `-fprofile-generate' instrumentation and real inlining pass.
4560 Doing so makes profiling significantly cheaper and usually
4561 inlining faster on programs having large chains of nested wrapper
4567 By default, GCC limits the size of functions that can be inlined.
4568 This flag allows the control of this limit for functions that are
4569 explicitly marked as inline (i.e., marked with the inline keyword
4570 or defined within the class definition in c++). N is the size of
4571 functions that can be inlined in number of pseudo instructions
4572 (not counting parameter handling). The default value of N is 600.
4573 Increasing this value can result in more inlined code at the cost
4574 of compilation time and memory consumption. Decreasing usually
4575 makes the compilation faster and less code will be inlined (which
4576 presumably means slower programs). This option is particularly
4577 useful for programs that use inlining heavily such as those based
4578 on recursive templates with C++.
4580 Inlining is actually controlled by a number of parameters, which
4581 may be specified individually by using `--param NAME=VALUE'. The
4582 `-finline-limit=N' option sets some of these parameters as follows:
4584 `max-inline-insns-single'
4587 `max-inline-insns-auto'
4591 is set to 130 or N/4, whichever is smaller.
4593 `max-inline-insns-rtl'
4596 See below for a documentation of the individual parameters
4597 controlling inlining.
4599 _Note:_ pseudo instruction represents, in this particular context,
4600 an abstract measurement of function's size. In no way does it
4601 represent a count of assembly instructions and as such its exact
4602 meaning might change from one release to an another.
4604 `-fkeep-inline-functions'
4605 In C, emit `static' functions that are declared `inline' into the
4606 object file, even if the function has been inlined into all of its
4607 callers. This switch does not affect functions using the `extern
4608 inline' extension in GNU C. In C++, emit any and all inline
4609 functions into the object file.
4611 `-fkeep-static-consts'
4612 Emit variables declared `static const' when optimization isn't
4613 turned on, even if the variables aren't referenced.
4615 GCC enables this option by default. If you want to force the
4616 compiler to check if the variable was referenced, regardless of
4617 whether or not optimization is turned on, use the
4618 `-fno-keep-static-consts' option.
4621 Attempt to merge identical constants (string constants and
4622 floating point constants) across compilation units.
4624 This option is the default for optimized compilation if the
4625 assembler and linker support it. Use `-fno-merge-constants' to
4626 inhibit this behavior.
4628 Enabled at levels `-O', `-O2', `-O3', `-Os'.
4630 `-fmerge-all-constants'
4631 Attempt to merge identical constants and identical variables.
4633 This option implies `-fmerge-constants'. In addition to
4634 `-fmerge-constants' this considers e.g. even constant initialized
4635 arrays or initialized constant variables with integral or floating
4636 point types. Languages like C or C++ require each non-automatic
4637 variable to have distinct location, so using this option will
4638 result in non-conforming behavior.
4641 Perform swing modulo scheduling immediately before the first
4642 scheduling pass. This pass looks at innermost loops and reorders
4643 their instructions by overlapping different iterations.
4645 `-fno-branch-count-reg'
4646 Do not use "decrement and branch" instructions on a count register,
4647 but instead generate a sequence of instructions that decrement a
4648 register, compare it against zero, then branch based upon the
4649 result. This option is only meaningful on architectures that
4650 support such instructions, which include x86, PowerPC, IA-64 and
4653 The default is `-fbranch-count-reg'.
4656 Do not put function addresses in registers; make each instruction
4657 that calls a constant function contain the function's address
4660 This option results in less efficient code, but some strange hacks
4661 that alter the assembler output may be confused by the
4662 optimizations performed when this option is not used.
4664 The default is `-ffunction-cse'
4666 `-fno-zero-initialized-in-bss'
4667 If the target supports a BSS section, GCC by default puts
4668 variables that are initialized to zero into BSS. This can save
4669 space in the resulting code.
4671 This option turns off this behavior because some programs
4672 explicitly rely on variables going to the data section. E.g., so
4673 that the resulting executable can find the beginning of that
4674 section and/or make assumptions based on that.
4676 The default is `-fzero-initialized-in-bss'.
4679 For front-ends that support it, generate additional code to check
4680 that indices used to access arrays are within the declared range.
4681 This is currently only supported by the Java and Fortran
4682 front-ends, where this option defaults to true and false
4685 `-fmudflap -fmudflapth -fmudflapir'
4686 For front-ends that support it (C and C++), instrument all risky
4687 pointer/array dereferencing operations, some standard library
4688 string/heap functions, and some other associated constructs with
4689 range/validity tests. Modules so instrumented should be immune to
4690 buffer overflows, invalid heap use, and some other classes of C/C++
4691 programming errors. The instrumentation relies on a separate
4692 runtime library (`libmudflap'), which will be linked into a
4693 program if `-fmudflap' is given at link time. Run-time behavior
4694 of the instrumented program is controlled by the `MUDFLAP_OPTIONS'
4695 environment variable. See `env MUDFLAP_OPTIONS=-help a.out' for
4698 Use `-fmudflapth' instead of `-fmudflap' to compile and to link if
4699 your program is multi-threaded. Use `-fmudflapir', in addition to
4700 `-fmudflap' or `-fmudflapth', if instrumentation should ignore
4701 pointer reads. This produces less instrumentation (and therefore
4702 faster execution) and still provides some protection against
4703 outright memory corrupting writes, but allows erroneously read
4704 data to propagate within a program.
4707 Perform optimizations where we check to see if a jump branches to a
4708 location where another comparison subsumed by the first is found.
4709 If so, the first branch is redirected to either the destination of
4710 the second branch or a point immediately following it, depending
4711 on whether the condition is known to be true or false.
4713 Enabled at levels `-O2', `-O3', `-Os'.
4715 `-fcse-follow-jumps'
4716 In common subexpression elimination, scan through jump instructions
4717 when the target of the jump is not reached by any other path. For
4718 example, when CSE encounters an `if' statement with an `else'
4719 clause, CSE will follow the jump when the condition tested is
4722 Enabled at levels `-O2', `-O3', `-Os'.
4725 This is similar to `-fcse-follow-jumps', but causes CSE to follow
4726 jumps which conditionally skip over blocks. When CSE encounters a
4727 simple `if' statement with no else clause, `-fcse-skip-blocks'
4728 causes CSE to follow the jump around the body of the `if'.
4730 Enabled at levels `-O2', `-O3', `-Os'.
4732 `-frerun-cse-after-loop'
4733 Re-run common subexpression elimination after loop optimizations
4736 Enabled at levels `-O2', `-O3', `-Os'.
4739 Perform a global common subexpression elimination pass. This pass
4740 also performs global constant and copy propagation.
4742 _Note:_ When compiling a program using computed gotos, a GCC
4743 extension, you may get better runtime performance if you disable
4744 the global common subexpression elimination pass by adding
4745 `-fno-gcse' to the command line.
4747 Enabled at levels `-O2', `-O3', `-Os'.
4750 When `-fgcse-lm' is enabled, global common subexpression
4751 elimination will attempt to move loads which are only killed by
4752 stores into themselves. This allows a loop containing a
4753 load/store sequence to be changed to a load outside the loop, and
4754 a copy/store within the loop.
4756 Enabled by default when gcse is enabled.
4759 When `-fgcse-sm' is enabled, a store motion pass is run after
4760 global common subexpression elimination. This pass will attempt
4761 to move stores out of loops. When used in conjunction with
4762 `-fgcse-lm', loops containing a load/store sequence can be changed
4763 to a load before the loop and a store after the loop.
4765 Not enabled at any optimization level.
4768 When `-fgcse-las' is enabled, the global common subexpression
4769 elimination pass eliminates redundant loads that come after stores
4770 to the same memory location (both partial and full redundancies).
4772 Not enabled at any optimization level.
4774 `-fgcse-after-reload'
4775 When `-fgcse-after-reload' is enabled, a redundant load elimination
4776 pass is performed after reload. The purpose of this pass is to
4777 cleanup redundant spilling.
4779 `-funsafe-loop-optimizations'
4780 If given, the loop optimizer will assume that loop indices do not
4781 overflow, and that the loops with nontrivial exit condition are not
4782 infinite. This enables a wider range of loop optimizations even if
4783 the loop optimizer itself cannot prove that these assumptions are
4784 valid. Using `-Wunsafe-loop-optimizations', the compiler will
4785 warn you if it finds this kind of loop.
4788 Perform cross-jumping transformation. This transformation unifies
4789 equivalent code and save code size. The resulting code may or may
4790 not perform better than without cross-jumping.
4792 Enabled at levels `-O2', `-O3', `-Os'.
4795 Attempt to transform conditional jumps into branch-less
4796 equivalents. This include use of conditional moves, min, max, set
4797 flags and abs instructions, and some tricks doable by standard
4798 arithmetics. The use of conditional execution on chips where it
4799 is available is controlled by `if-conversion2'.
4801 Enabled at levels `-O', `-O2', `-O3', `-Os'.
4804 Use conditional execution (where available) to transform
4805 conditional jumps into branch-less equivalents.
4807 Enabled at levels `-O', `-O2', `-O3', `-Os'.
4809 `-fdelete-null-pointer-checks'
4810 Use global dataflow analysis to identify and eliminate useless
4811 checks for null pointers. The compiler assumes that dereferencing
4812 a null pointer would have halted the program. If a pointer is
4813 checked after it has already been dereferenced, it cannot be null.
4815 In some environments, this assumption is not true, and programs can
4816 safely dereference null pointers. Use
4817 `-fno-delete-null-pointer-checks' to disable this optimization for
4818 programs which depend on that behavior.
4820 Enabled at levels `-O2', `-O3', `-Os'.
4822 `-fexpensive-optimizations'
4823 Perform a number of minor optimizations that are relatively
4826 Enabled at levels `-O2', `-O3', `-Os'.
4828 `-foptimize-register-move'
4830 Attempt to reassign register numbers in move instructions and as
4831 operands of other simple instructions in order to maximize the
4832 amount of register tying. This is especially helpful on machines
4833 with two-operand instructions.
4835 Note `-fregmove' and `-foptimize-register-move' are the same
4838 Enabled at levels `-O2', `-O3', `-Os'.
4841 If supported for the target machine, attempt to reorder
4842 instructions to exploit instruction slots available after delayed
4843 branch instructions.
4845 Enabled at levels `-O', `-O2', `-O3', `-Os'.
4848 If supported for the target machine, attempt to reorder
4849 instructions to eliminate execution stalls due to required data
4850 being unavailable. This helps machines that have slow floating
4851 point or memory load instructions by allowing other instructions
4852 to be issued until the result of the load or floating point
4853 instruction is required.
4855 Enabled at levels `-O2', `-O3', `-Os'.
4858 Similar to `-fschedule-insns', but requests an additional pass of
4859 instruction scheduling after register allocation has been done.
4860 This is especially useful on machines with a relatively small
4861 number of registers and where memory load instructions take more
4864 Enabled at levels `-O2', `-O3', `-Os'.
4866 `-fno-sched-interblock'
4867 Don't schedule instructions across basic blocks. This is normally
4868 enabled by default when scheduling before register allocation, i.e.
4869 with `-fschedule-insns' or at `-O2' or higher.
4872 Don't allow speculative motion of non-load instructions. This is
4873 normally enabled by default when scheduling before register
4874 allocation, i.e. with `-fschedule-insns' or at `-O2' or higher.
4877 Allow speculative motion of some load instructions. This only
4878 makes sense when scheduling before register allocation, i.e. with
4879 `-fschedule-insns' or at `-O2' or higher.
4881 `-fsched-spec-load-dangerous'
4882 Allow speculative motion of more load instructions. This only
4883 makes sense when scheduling before register allocation, i.e. with
4884 `-fschedule-insns' or at `-O2' or higher.
4886 `-fsched-stalled-insns=N'
4887 Define how many insns (if any) can be moved prematurely from the
4888 queue of stalled insns into the ready list, during the second
4891 `-fsched-stalled-insns-dep=N'
4892 Define how many insn groups (cycles) will be examined for a
4893 dependency on a stalled insn that is candidate for premature
4894 removal from the queue of stalled insns. Has an effect only
4895 during the second scheduling pass, and only if
4896 `-fsched-stalled-insns' is used and its value is not zero.
4898 `-fsched2-use-superblocks'
4899 When scheduling after register allocation, do use superblock
4900 scheduling algorithm. Superblock scheduling allows motion across
4901 basic block boundaries resulting on faster schedules. This option
4902 is experimental, as not all machine descriptions used by GCC model
4903 the CPU closely enough to avoid unreliable results from the
4906 This only makes sense when scheduling after register allocation,
4907 i.e. with `-fschedule-insns2' or at `-O2' or higher.
4909 `-fsched2-use-traces'
4910 Use `-fsched2-use-superblocks' algorithm when scheduling after
4911 register allocation and additionally perform code duplication in
4912 order to increase the size of superblocks using tracer pass. See
4913 `-ftracer' for details on trace formation.
4915 This mode should produce faster but significantly longer programs.
4916 Also without `-fbranch-probabilities' the traces constructed may
4917 not match the reality and hurt the performance. This only makes
4918 sense when scheduling after register allocation, i.e. with
4919 `-fschedule-insns2' or at `-O2' or higher.
4922 Eliminates redundant extension instructions and move the non
4923 redundant ones to optimal placement using LCM.
4925 `-freschedule-modulo-scheduled-loops'
4926 The modulo scheduling comes before the traditional scheduling, if
4927 a loop was modulo scheduled we may want to prevent the later
4928 scheduling passes from changing its schedule, we use this option
4932 Enable values to be allocated in registers that will be clobbered
4933 by function calls, by emitting extra instructions to save and
4934 restore the registers around such calls. Such allocation is done
4935 only when it seems to result in better code than would otherwise
4938 This option is always enabled by default on certain machines,
4939 usually those which have no call-preserved registers to use
4942 Enabled at levels `-O2', `-O3', `-Os'.
4945 Perform Partial Redundancy Elimination (PRE) on trees. This flag
4946 is enabled by default at `-O2' and `-O3'.
4949 Perform Full Redundancy Elimination (FRE) on trees. The difference
4950 between FRE and PRE is that FRE only considers expressions that
4951 are computed on all paths leading to the redundant computation.
4952 This analysis faster than PRE, though it exposes fewer
4953 redundancies. This flag is enabled by default at `-O' and higher.
4956 Perform copy propagation on trees. This pass eliminates
4957 unnecessary copy operations. This flag is enabled by default at
4960 `-ftree-store-copy-prop'
4961 Perform copy propagation of memory loads and stores. This pass
4962 eliminates unnecessary copy operations in memory references
4963 (structures, global variables, arrays, etc). This flag is enabled
4964 by default at `-O2' and higher.
4967 Perform structural alias analysis on trees. This flag is enabled
4968 by default at `-O' and higher.
4971 Perform interprocedural pointer analysis.
4974 Perform forward store motion on trees. This flag is enabled by
4975 default at `-O' and higher.
4978 Perform sparse conditional constant propagation (CCP) on trees.
4979 This pass only operates on local scalar variables and is enabled
4980 by default at `-O' and higher.
4983 Perform sparse conditional constant propagation (CCP) on trees.
4984 This pass operates on both local scalar variables and memory
4985 stores and loads (global variables, structures, arrays, etc).
4986 This flag is enabled by default at `-O2' and higher.
4989 Perform dead code elimination (DCE) on trees. This flag is
4990 enabled by default at `-O' and higher.
4992 `-ftree-dominator-opts'
4993 Perform a variety of simple scalar cleanups (constant/copy
4994 propagation, redundancy elimination, range propagation and
4995 expression simplification) based on a dominator tree traversal.
4996 This also performs jump threading (to reduce jumps to jumps). This
4997 flag is enabled by default at `-O' and higher.
5000 Perform loop header copying on trees. This is beneficial since it
5001 increases effectiveness of code motion optimizations. It also
5002 saves one jump. This flag is enabled by default at `-O' and
5003 higher. It is not enabled for `-Os', since it usually increases
5006 `-ftree-loop-optimize'
5007 Perform loop optimizations on trees. This flag is enabled by
5008 default at `-O' and higher.
5010 `-ftree-loop-linear'
5011 Perform linear loop transformations on tree. This flag can
5012 improve cache performance and allow further loop optimizations to
5016 Perform loop invariant motion on trees. This pass moves only
5017 invariants that would be hard to handle at RTL level (function
5018 calls, operations that expand to nontrivial sequences of insns).
5019 With `-funswitch-loops' it also moves operands of conditions that
5020 are invariant out of the loop, so that we can use just trivial
5021 invariantness analysis in loop unswitching. The pass also includes
5024 `-ftree-loop-ivcanon'
5025 Create a canonical counter for number of iterations in the loop
5026 for that determining number of iterations requires complicated
5027 analysis. Later optimizations then may determine the number
5028 easily. Useful especially in connection with unrolling.
5031 Perform induction variable optimizations (strength reduction,
5032 induction variable merging and induction variable elimination) on
5036 Perform scalar replacement of aggregates. This pass replaces
5037 structure references with scalars to prevent committing structures
5038 to memory too early. This flag is enabled by default at `-O' and
5042 Perform copy renaming on trees. This pass attempts to rename
5043 compiler temporaries to other variables at copy locations, usually
5044 resulting in variable names which more closely resemble the
5045 original variables. This flag is enabled by default at `-O' and
5049 Perform temporary expression replacement during the SSA->normal
5050 phase. Single use/single def temporaries are replaced at their
5051 use location with their defining expression. This results in
5052 non-GIMPLE code, but gives the expanders much more complex trees
5053 to work on resulting in better RTL generation. This is enabled by
5054 default at `-O' and higher.
5057 Perform live range splitting during the SSA->normal phase.
5058 Distinct live ranges of a variable are split into unique
5059 variables, allowing for better optimization later. This is
5060 enabled by default at `-O' and higher.
5063 Perform loop vectorization on trees.
5065 `-ftree-vect-loop-version'
5066 Perform loop versioning when doing loop vectorization on trees.
5067 When a loop appears to be vectorizable except that data alignment
5068 or data dependence cannot be determined at compile time then
5069 vectorized and non-vectorized versions of the loop are generated
5070 along with runtime checks for alignment or dependence to control
5071 which version is executed. This option is enabled by default
5072 except at level `-Os' where it is disabled.
5075 Perform Value Range Propagation on trees. This is similar to the
5076 constant propagation pass, but instead of values, ranges of values
5077 are propagated. This allows the optimizers to remove unnecessary
5078 range checks like array bound checks and null pointer checks.
5079 This is enabled by default at `-O2' and higher. Null pointer check
5080 elimination is only done if `-fdelete-null-pointer-checks' is
5084 Perform tail duplication to enlarge superblock size. This
5085 transformation simplifies the control flow of the function
5086 allowing other optimizations to do better job.
5089 Unroll loops whose number of iterations can be determined at
5090 compile time or upon entry to the loop. `-funroll-loops' implies
5091 `-frerun-cse-after-loop'. This option makes code larger, and may
5092 or may not make it run faster.
5094 `-funroll-all-loops'
5095 Unroll all loops, even if their number of iterations is uncertain
5096 when the loop is entered. This usually makes programs run more
5097 slowly. `-funroll-all-loops' implies the same options as
5100 `-fsplit-ivs-in-unroller'
5101 Enables expressing of values of induction variables in later
5102 iterations of the unrolled loop using the value in the first
5103 iteration. This breaks long dependency chains, thus improving
5104 efficiency of the scheduling passes.
5106 Combination of `-fweb' and CSE is often sufficient to obtain the
5107 same effect. However in cases the loop body is more complicated
5108 than a single basic block, this is not reliable. It also does not
5109 work at all on some of the architectures due to restrictions in
5112 This optimization is enabled by default.
5114 `-fvariable-expansion-in-unroller'
5115 With this option, the compiler will create multiple copies of some
5116 local variables when unrolling a loop which can result in superior
5119 `-fprefetch-loop-arrays'
5120 If supported by the target machine, generate instructions to
5121 prefetch memory to improve the performance of loops that access
5124 This option may generate better or worse code; results are highly
5125 dependent on the structure of loops within the source code.
5127 Disabled at level `-Os'.
5131 Disable any machine-specific peephole optimizations. The
5132 difference between `-fno-peephole' and `-fno-peephole2' is in how
5133 they are implemented in the compiler; some targets use one, some
5134 use the other, a few use both.
5136 `-fpeephole' is enabled by default. `-fpeephole2' enabled at
5137 levels `-O2', `-O3', `-Os'.
5139 `-fno-guess-branch-probability'
5140 Do not guess branch probabilities using heuristics.
5142 GCC will use heuristics to guess branch probabilities if they are
5143 not provided by profiling feedback (`-fprofile-arcs'). These
5144 heuristics are based on the control flow graph. If some branch
5145 probabilities are specified by `__builtin_expect', then the
5146 heuristics will be used to guess branch probabilities for the rest
5147 of the control flow graph, taking the `__builtin_expect' info into
5148 account. The interactions between the heuristics and
5149 `__builtin_expect' can be complex, and in some cases, it may be
5150 useful to disable the heuristics so that the effects of
5151 `__builtin_expect' are easier to understand.
5153 The default is `-fguess-branch-probability' at levels `-O', `-O2',
5157 Reorder basic blocks in the compiled function in order to reduce
5158 number of taken branches and improve code locality.
5160 Enabled at levels `-O2', `-O3'.
5162 `-freorder-blocks-and-partition'
5163 In addition to reordering basic blocks in the compiled function,
5164 in order to reduce number of taken branches, partitions hot and
5165 cold basic blocks into separate sections of the assembly and .o
5166 files, to improve paging and cache locality performance.
5168 This optimization is automatically turned off in the presence of
5169 exception handling, for linkonce sections, for functions with a
5170 user-defined section attribute and on any architecture that does
5171 not support named sections.
5173 `-freorder-functions'
5174 Reorder functions in the object file in order to improve code
5175 locality. This is implemented by using special subsections
5176 `.text.hot' for most frequently executed functions and
5177 `.text.unlikely' for unlikely executed functions. Reordering is
5178 done by the linker so object file format must support named
5179 sections and linker must place them in a reasonable way.
5181 Also profile feedback must be available in to make this option
5182 effective. See `-fprofile-arcs' for details.
5184 Enabled at levels `-O2', `-O3', `-Os'.
5187 Allows the compiler to assume the strictest aliasing rules
5188 applicable to the language being compiled. For C (and C++), this
5189 activates optimizations based on the type of expressions. In
5190 particular, an object of one type is assumed never to reside at
5191 the same address as an object of a different type, unless the
5192 types are almost the same. For example, an `unsigned int' can
5193 alias an `int', but not a `void*' or a `double'. A character type
5194 may alias any other type.
5196 Pay special attention to code like this:
5207 The practice of reading from a different union member than the one
5208 most recently written to (called "type-punning") is common. Even
5209 with `-fstrict-aliasing', type-punning is allowed, provided the
5210 memory is accessed through the union type. So, the code above
5211 will work as expected. However, this code might not:
5220 Every language that wishes to perform language-specific alias
5221 analysis should define a function that computes, given an `tree'
5222 node, an alias set for the node. Nodes in different alias sets
5223 are not allowed to alias. For an example, see the C front-end
5224 function `c_get_alias_set'.
5226 Enabled at levels `-O2', `-O3', `-Os'.
5229 Allow the compiler to assume strict signed overflow rules,
5230 depending on the language being compiled. For C (and C++) this
5231 means that overflow when doing arithmetic with signed numbers is
5232 undefined, which means that the compiler may assume that it will
5233 not happen. This permits various optimizations. For example, the
5234 compiler will assume that an expression like `i + 10 > i' will
5235 always be true for signed `i'. This assumption is only valid if
5236 signed overflow is undefined, as the expression is false if `i +
5237 10' overflows when using twos complement arithmetic. When this
5238 option is in effect any attempt to determine whether an operation
5239 on signed numbers will overflow must be written carefully to not
5240 actually involve overflow.
5242 See also the `-fwrapv' option. Using `-fwrapv' means that signed
5243 overflow is fully defined: it wraps. When `-fwrapv' is used,
5244 there is no difference between `-fstrict-overflow' and
5245 `-fno-strict-overflow'. With `-fwrapv' certain types of overflow
5246 are permitted. For example, if the compiler gets an overflow when
5247 doing arithmetic on constants, the overflowed value can still be
5248 used with `-fwrapv', but not otherwise.
5250 The `-fstrict-overflow' option is enabled at levels `-O2', `-O3',
5254 `-falign-functions=N'
5255 Align the start of functions to the next power-of-two greater than
5256 N, skipping up to N bytes. For instance, `-falign-functions=32'
5257 aligns functions to the next 32-byte boundary, but
5258 `-falign-functions=24' would align to the next 32-byte boundary
5259 only if this can be done by skipping 23 bytes or less.
5261 `-fno-align-functions' and `-falign-functions=1' are equivalent
5262 and mean that functions will not be aligned.
5264 Some assemblers only support this flag when N is a power of two;
5265 in that case, it is rounded up.
5267 If N is not specified or is zero, use a machine-dependent default.
5269 Enabled at levels `-O2', `-O3'.
5273 Align all branch targets to a power-of-two boundary, skipping up to
5274 N bytes like `-falign-functions'. This option can easily make
5275 code slower, because it must insert dummy operations for when the
5276 branch target is reached in the usual flow of the code.
5278 `-fno-align-labels' and `-falign-labels=1' are equivalent and mean
5279 that labels will not be aligned.
5281 If `-falign-loops' or `-falign-jumps' are applicable and are
5282 greater than this value, then their values are used instead.
5284 If N is not specified or is zero, use a machine-dependent default
5285 which is very likely to be `1', meaning no alignment.
5287 Enabled at levels `-O2', `-O3'.
5291 Align loops to a power-of-two boundary, skipping up to N bytes
5292 like `-falign-functions'. The hope is that the loop will be
5293 executed many times, which will make up for any execution of the
5296 `-fno-align-loops' and `-falign-loops=1' are equivalent and mean
5297 that loops will not be aligned.
5299 If N is not specified or is zero, use a machine-dependent default.
5301 Enabled at levels `-O2', `-O3'.
5305 Align branch targets to a power-of-two boundary, for branch targets
5306 where the targets can only be reached by jumping, skipping up to N
5307 bytes like `-falign-functions'. In this case, no dummy operations
5310 `-fno-align-jumps' and `-falign-jumps=1' are equivalent and mean
5311 that loops will not be aligned.
5313 If N is not specified or is zero, use a machine-dependent default.
5315 Enabled at levels `-O2', `-O3'.
5318 Parse the whole compilation unit before starting to produce code.
5319 This allows some extra optimizations to take place but consumes
5320 more memory (in general). There are some compatibility issues
5321 with _unit-at-a-time_ mode:
5322 * enabling _unit-at-a-time_ mode may change the order in which
5323 functions, variables, and top-level `asm' statements are
5324 emitted, and will likely break code relying on some particular
5325 ordering. The majority of such top-level `asm' statements,
5326 though, can be replaced by `section' attributes. The
5327 `fno-toplevel-reorder' option may be used to keep the ordering
5328 used in the input file, at the cost of some optimizations.
5330 * _unit-at-a-time_ mode removes unreferenced static variables
5331 and functions. This may result in undefined references when
5332 an `asm' statement refers directly to variables or functions
5333 that are otherwise unused. In that case either the
5334 variable/function shall be listed as an operand of the `asm'
5335 statement operand or, in the case of top-level `asm'
5336 statements the attribute `used' shall be used on the
5339 * Static functions now can use non-standard passing conventions
5340 that may break `asm' statements calling functions directly.
5341 Again, attribute `used' will prevent this behavior.
5343 As a temporary workaround, `-fno-unit-at-a-time' can be used, but
5344 this scheme may not be supported by future releases of GCC.
5346 Enabled at levels `-O', `-O2', `-O3', `-Os'.
5348 `-fno-toplevel-reorder'
5349 Do not reorder top-level functions, variables, and `asm'
5350 statements. Output them in the same order that they appear in the
5351 input file. When this option is used, unreferenced static
5352 variables will not be removed. This option is intended to support
5353 existing code which relies on a particular ordering. For new
5354 code, it is better to use attributes.
5357 Constructs webs as commonly used for register allocation purposes
5358 and assign each web individual pseudo register. This allows the
5359 register allocation pass to operate on pseudos directly, but also
5360 strengthens several other optimization passes, such as CSE, loop
5361 optimizer and trivial dead code remover. It can, however, make
5362 debugging impossible, since variables will no longer stay in a
5365 Enabled by default with `-funroll-loops'.
5368 Assume that the current compilation unit represents whole program
5369 being compiled. All public functions and variables with the
5370 exception of `main' and those merged by attribute
5371 `externally_visible' become static functions and in a affect gets
5372 more aggressively optimized by interprocedural optimizers. While
5373 this option is equivalent to proper use of `static' keyword for
5374 programs consisting of single file, in combination with option
5375 `--combine' this flag can be used to compile most of smaller scale
5376 C programs since the functions and variables become local for the
5377 whole combined compilation unit, not for the single source file
5380 `-fno-cprop-registers'
5381 After register allocation and post-register allocation instruction
5382 splitting, we perform a copy-propagation pass to try to reduce
5383 scheduling dependencies and occasionally eliminate the copy.
5385 Disabled at levels `-O', `-O2', `-O3', `-Os'.
5387 `-fprofile-generate'
5388 Enable options usually used for instrumenting application to
5389 produce profile useful for later recompilation with profile
5390 feedback based optimization. You must use `-fprofile-generate'
5391 both when compiling and when linking your program.
5393 The following options are enabled: `-fprofile-arcs',
5394 `-fprofile-values', `-fvpt'.
5397 Enable profile feedback directed optimizations, and optimizations
5398 generally profitable only with profile feedback available.
5400 The following options are enabled: `-fbranch-probabilities',
5401 `-fvpt', `-funroll-loops', `-fpeel-loops', `-ftracer'
5404 The following options control compiler behavior regarding floating
5405 point arithmetic. These options trade off between speed and
5406 correctness. All must be specifically enabled.
5409 Do not store floating point variables in registers, and inhibit
5410 other options that might change whether a floating point value is
5411 taken from a register or memory.
5413 This option prevents undesirable excess precision on machines such
5414 as the 68000 where the floating registers (of the 68881) keep more
5415 precision than a `double' is supposed to have. Similarly for the
5416 x86 architecture. For most programs, the excess precision does
5417 only good, but a few programs rely on the precise definition of
5418 IEEE floating point. Use `-ffloat-store' for such programs, after
5419 modifying them to store all pertinent intermediate computations
5423 Sets `-fno-math-errno', `-funsafe-math-optimizations',
5424 `-fno-trapping-math', `-ffinite-math-only', `-fno-rounding-math',
5425 `-fno-signaling-nans' and `fcx-limited-range'.
5427 This option causes the preprocessor macro `__FAST_MATH__' to be
5430 This option should never be turned on by any `-O' option since it
5431 can result in incorrect output for programs which depend on an
5432 exact implementation of IEEE or ISO rules/specifications for math
5436 Do not set ERRNO after calling math functions that are executed
5437 with a single instruction, e.g., sqrt. A program that relies on
5438 IEEE exceptions for math error handling may want to use this flag
5439 for speed while maintaining IEEE arithmetic compatibility.
5441 This option should never be turned on by any `-O' option since it
5442 can result in incorrect output for programs which depend on an
5443 exact implementation of IEEE or ISO rules/specifications for math
5446 The default is `-fmath-errno'.
5448 On Darwin systems, the math library never sets `errno'. There is
5449 therefore no reason for the compiler to consider the possibility
5450 that it might, and `-fno-math-errno' is the default.
5452 `-funsafe-math-optimizations'
5453 Allow optimizations for floating-point arithmetic that (a) assume
5454 that arguments and results are valid and (b) may violate IEEE or
5455 ANSI standards. When used at link-time, it may include libraries
5456 or startup files that change the default FPU control word or other
5457 similar optimizations.
5459 This option should never be turned on by any `-O' option since it
5460 can result in incorrect output for programs which depend on an
5461 exact implementation of IEEE or ISO rules/specifications for math
5464 The default is `-fno-unsafe-math-optimizations'.
5466 `-ffinite-math-only'
5467 Allow optimizations for floating-point arithmetic that assume that
5468 arguments and results are not NaNs or +-Infs.
5470 This option should never be turned on by any `-O' option since it
5471 can result in incorrect output for programs which depend on an
5472 exact implementation of IEEE or ISO rules/specifications.
5474 The default is `-fno-finite-math-only'.
5476 `-fno-trapping-math'
5477 Compile code assuming that floating-point operations cannot
5478 generate user-visible traps. These traps include division by
5479 zero, overflow, underflow, inexact result and invalid operation.
5480 This option implies `-fno-signaling-nans'. Setting this option
5481 may allow faster code if one relies on "non-stop" IEEE arithmetic,
5484 This option should never be turned on by any `-O' option since it
5485 can result in incorrect output for programs which depend on an
5486 exact implementation of IEEE or ISO rules/specifications for math
5489 The default is `-ftrapping-math'.
5492 Disable transformations and optimizations that assume default
5493 floating point rounding behavior. This is round-to-zero for all
5494 floating point to integer conversions, and round-to-nearest for
5495 all other arithmetic truncations. This option should be specified
5496 for programs that change the FP rounding mode dynamically, or that
5497 may be executed with a non-default rounding mode. This option
5498 disables constant folding of floating point expressions at
5499 compile-time (which may be affected by rounding mode) and
5500 arithmetic transformations that are unsafe in the presence of
5501 sign-dependent rounding modes.
5503 The default is `-fno-rounding-math'.
5505 This option is experimental and does not currently guarantee to
5506 disable all GCC optimizations that are affected by rounding mode.
5507 Future versions of GCC may provide finer control of this setting
5508 using C99's `FENV_ACCESS' pragma. This command line option will
5509 be used to specify the default state for `FENV_ACCESS'.
5511 `-frtl-abstract-sequences'
5512 It is a size optimization method. This option is to find identical
5513 sequences of code, which can be turned into pseudo-procedures and
5514 then replace all occurrences with calls to the newly created
5515 subroutine. It is kind of an opposite of `-finline-functions'.
5516 This optimization runs at RTL level.
5519 Compile code assuming that IEEE signaling NaNs may generate
5520 user-visible traps during floating-point operations. Setting this
5521 option disables optimizations that may change the number of
5522 exceptions visible with signaling NaNs. This option implies
5525 This option causes the preprocessor macro `__SUPPORT_SNAN__' to be
5528 The default is `-fno-signaling-nans'.
5530 This option is experimental and does not currently guarantee to
5531 disable all GCC optimizations that affect signaling NaN behavior.
5533 `-fsingle-precision-constant'
5534 Treat floating point constant as single precision constant instead
5535 of implicitly converting it to double precision constant.
5537 `-fcx-limited-range'
5538 `-fno-cx-limited-range'
5539 When enabled, this option states that a range reduction step is not
5540 needed when performing complex division. The default is
5541 `-fno-cx-limited-range', but is enabled by `-ffast-math'.
5543 This option controls the default setting of the ISO C99
5544 `CX_LIMITED_RANGE' pragma. Nevertheless, the option applies to
5548 The following options control optimizations that may improve
5549 performance, but are not enabled by any `-O' options. This section
5550 includes experimental options that may produce broken code.
5552 `-fbranch-probabilities'
5553 After running a program compiled with `-fprofile-arcs' (*note
5554 Options for Debugging Your Program or `gcc': Debugging Options.),
5555 you can compile it a second time using `-fbranch-probabilities',
5556 to improve optimizations based on the number of times each branch
5557 was taken. When the program compiled with `-fprofile-arcs' exits
5558 it saves arc execution counts to a file called `SOURCENAME.gcda'
5559 for each source file The information in this data file is very
5560 dependent on the structure of the generated code, so you must use
5561 the same source code and the same optimization options for both
5564 With `-fbranch-probabilities', GCC puts a `REG_BR_PROB' note on
5565 each `JUMP_INSN' and `CALL_INSN'. These can be used to improve
5566 optimization. Currently, they are only used in one place: in
5567 `reorg.c', instead of guessing which path a branch is mostly to
5568 take, the `REG_BR_PROB' values are used to exactly determine which
5569 path is taken more often.
5572 If combined with `-fprofile-arcs', it adds code so that some data
5573 about values of expressions in the program is gathered.
5575 With `-fbranch-probabilities', it reads back the data gathered
5576 from profiling values of expressions and adds `REG_VALUE_PROFILE'
5577 notes to instructions for their later usage in optimizations.
5579 Enabled with `-fprofile-generate' and `-fprofile-use'.
5582 If combined with `-fprofile-arcs', it instructs the compiler to add
5583 a code to gather information about values of expressions.
5585 With `-fbranch-probabilities', it reads back the data gathered and
5586 actually performs the optimizations based on them. Currently the
5587 optimizations include specialization of division operation using
5588 the knowledge about the value of the denominator.
5590 `-frename-registers'
5591 Attempt to avoid false dependencies in scheduled code by making use
5592 of registers left over after register allocation. This
5593 optimization will most benefit processors with lots of registers.
5594 Depending on the debug information format adopted by the target,
5595 however, it can make debugging impossible, since variables will no
5596 longer stay in a "home register".
5598 Enabled by default with `-funroll-loops'.
5601 Perform tail duplication to enlarge superblock size. This
5602 transformation simplifies the control flow of the function
5603 allowing other optimizations to do better job.
5605 Enabled with `-fprofile-use'.
5608 Unroll loops whose number of iterations can be determined at
5609 compile time or upon entry to the loop. `-funroll-loops' implies
5610 `-frerun-cse-after-loop', `-fweb' and `-frename-registers'. It
5611 also turns on complete loop peeling (i.e. complete removal of
5612 loops with small constant number of iterations). This option
5613 makes code larger, and may or may not make it run faster.
5615 Enabled with `-fprofile-use'.
5617 `-funroll-all-loops'
5618 Unroll all loops, even if their number of iterations is uncertain
5619 when the loop is entered. This usually makes programs run more
5620 slowly. `-funroll-all-loops' implies the same options as
5624 Peels the loops for that there is enough information that they do
5625 not roll much (from profile feedback). It also turns on complete
5626 loop peeling (i.e. complete removal of loops with small constant
5627 number of iterations).
5629 Enabled with `-fprofile-use'.
5631 `-fmove-loop-invariants'
5632 Enables the loop invariant motion pass in the RTL loop optimizer.
5633 Enabled at level `-O1'
5636 Move branches with loop invariant conditions out of the loop, with
5637 duplicates of the loop on both branches (modified according to
5638 result of the condition).
5640 `-ffunction-sections'
5642 Place each function or data item into its own section in the output
5643 file if the target supports arbitrary sections. The name of the
5644 function or the name of the data item determines the section's name
5647 Use these options on systems where the linker can perform
5648 optimizations to improve locality of reference in the instruction
5649 space. Most systems using the ELF object format and SPARC
5650 processors running Solaris 2 have linkers with such optimizations.
5651 AIX may have these optimizations in the future.
5653 Only use these options when there are significant benefits from
5654 doing so. When you specify these options, the assembler and
5655 linker will create larger object and executable files and will
5656 also be slower. You will not be able to use `gprof' on all
5657 systems if you specify this option and you may have problems with
5658 debugging if you specify both this option and `-g'.
5660 `-fbranch-target-load-optimize'
5661 Perform branch target register load optimization before prologue /
5662 epilogue threading. The use of target registers can typically be
5663 exposed only during reload, thus hoisting loads out of loops and
5664 doing inter-block scheduling needs a separate optimization pass.
5666 `-fbranch-target-load-optimize2'
5667 Perform branch target register load optimization after prologue /
5670 `-fbtr-bb-exclusive'
5671 When performing branch target register load optimization, don't
5672 reuse branch target registers in within any basic block.
5675 Emit extra code to check for buffer overflows, such as stack
5676 smashing attacks. This is done by adding a guard variable to
5677 functions with vulnerable objects. This includes functions that
5678 call alloca, and functions with buffers larger than 8 bytes. The
5679 guards are initialized when a function is entered and then checked
5680 when the function exits. If a guard check fails, an error message
5681 is printed and the program exits.
5683 `-fstack-protector-all'
5684 Like `-fstack-protector' except that all functions are protected.
5687 Try to reduce the number of symbolic address calculations by using
5688 shared "anchor" symbols to address nearby objects. This
5689 transformation can help to reduce the number of GOT entries and
5690 GOT accesses on some targets.
5692 For example, the implementation of the following function `foo':
5695 int foo (void) { return a + b + c; }
5697 would usually calculate the addresses of all three variables, but
5698 if you compile it with `-fsection-anchors', it will access the
5699 variables from a common anchor point instead. The effect is
5700 similar to the following pseudocode (which isn't valid C):
5704 register int *xr = &x;
5705 return xr[&a - &x] + xr[&b - &x] + xr[&c - &x];
5708 Not all targets support this option.
5710 `--param NAME=VALUE'
5711 In some places, GCC uses various constants to control the amount of
5712 optimization that is done. For example, GCC will not inline
5713 functions that contain more that a certain number of instructions.
5714 You can control some of these constants on the command-line using
5715 the `--param' option.
5717 The names of specific parameters, and the meaning of the values,
5718 are tied to the internals of the compiler, and are subject to
5719 change without notice in future releases.
5721 In each case, the VALUE is an integer. The allowable choices for
5722 NAME are given in the following table:
5724 `salias-max-implicit-fields'
5725 The maximum number of fields in a variable without direct
5726 structure accesses for which structure aliasing will consider
5727 trying to track each field. The default is 5
5729 `salias-max-array-elements'
5730 The maximum number of elements an array can have and its
5731 elements still be tracked individually by structure aliasing.
5734 `sra-max-structure-size'
5735 The maximum structure size, in bytes, at which the scalar
5736 replacement of aggregates (SRA) optimization will perform
5737 block copies. The default value, 0, implies that GCC will
5738 select the most appropriate size itself.
5740 `sra-field-structure-ratio'
5741 The threshold ratio (as a percentage) between instantiated
5742 fields and the complete structure size. We say that if the
5743 ratio of the number of bytes in instantiated fields to the
5744 number of bytes in the complete structure exceeds this
5745 parameter, then block copies are not used. The default is 75.
5747 `max-crossjump-edges'
5748 The maximum number of incoming edges to consider for
5749 crossjumping. The algorithm used by `-fcrossjumping' is
5750 O(N^2) in the number of edges incoming to each block.
5751 Increasing values mean more aggressive optimization, making
5752 the compile time increase with probably small improvement in
5755 `min-crossjump-insns'
5756 The minimum number of instructions which must be matched at
5757 the end of two blocks before crossjumping will be performed
5758 on them. This value is ignored in the case where all
5759 instructions in the block being crossjumped from are matched.
5760 The default value is 5.
5762 `max-grow-copy-bb-insns'
5763 The maximum code size expansion factor when copying basic
5764 blocks instead of jumping. The expansion is relative to a
5765 jump instruction. The default value is 8.
5767 `max-goto-duplication-insns'
5768 The maximum number of instructions to duplicate to a block
5769 that jumps to a computed goto. To avoid O(N^2) behavior in a
5770 number of passes, GCC factors computed gotos early in the
5771 compilation process, and unfactors them as late as possible.
5772 Only computed jumps at the end of a basic blocks with no more
5773 than max-goto-duplication-insns are unfactored. The default
5776 `max-delay-slot-insn-search'
5777 The maximum number of instructions to consider when looking
5778 for an instruction to fill a delay slot. If more than this
5779 arbitrary number of instructions is searched, the time
5780 savings from filling the delay slot will be minimal so stop
5781 searching. Increasing values mean more aggressive
5782 optimization, making the compile time increase with probably
5783 small improvement in executable run time.
5785 `max-delay-slot-live-search'
5786 When trying to fill delay slots, the maximum number of
5787 instructions to consider when searching for a block with
5788 valid live register information. Increasing this arbitrarily
5789 chosen value means more aggressive optimization, increasing
5790 the compile time. This parameter should be removed when the
5791 delay slot code is rewritten to maintain the control-flow
5795 The approximate maximum amount of memory that will be
5796 allocated in order to perform the global common subexpression
5797 elimination optimization. If more memory than specified is
5798 required, the optimization will not be done.
5801 The maximum number of passes of GCSE to run. The default is
5804 `max-pending-list-length'
5805 The maximum number of pending dependencies scheduling will
5806 allow before flushing the current state and starting over.
5807 Large functions with few branches or calls can create
5808 excessively large lists which needlessly consume memory and
5811 `max-inline-insns-single'
5812 Several parameters control the tree inliner used in gcc.
5813 This number sets the maximum number of instructions (counted
5814 in GCC's internal representation) in a single function that
5815 the tree inliner will consider for inlining. This only
5816 affects functions declared inline and methods implemented in
5817 a class declaration (C++). The default value is 450.
5819 `max-inline-insns-auto'
5820 When you use `-finline-functions' (included in `-O3'), a lot
5821 of functions that would otherwise not be considered for
5822 inlining by the compiler will be investigated. To those
5823 functions, a different (more restrictive) limit compared to
5824 functions declared inline can be applied. The default value
5827 `large-function-insns'
5828 The limit specifying really large functions. For functions
5829 larger than this limit after inlining inlining is constrained
5830 by `--param large-function-growth'. This parameter is useful
5831 primarily to avoid extreme compilation time caused by
5832 non-linear algorithms used by the backend. This parameter is
5833 ignored when `-funit-at-a-time' is not used. The default
5836 `large-function-growth'
5837 Specifies maximal growth of large function caused by inlining
5838 in percents. This parameter is ignored when
5839 `-funit-at-a-time' is not used. The default value is 100
5840 which limits large function growth to 2.0 times the original
5844 The limit specifying large translation unit. Growth caused
5845 by inlining of units larger than this limit is limited by
5846 `--param inline-unit-growth'. For small units this might be
5847 too tight (consider unit consisting of function A that is
5848 inline and B that just calls A three time. If B is small
5849 relative to A, the growth of unit is 300\% and yet such
5850 inlining is very sane. For very large units consisting of
5851 small inlininable functions however the overall unit growth
5852 limit is needed to avoid exponential explosion of code size.
5853 Thus for smaller units, the size is increased to `--param
5854 large-unit-insns' before applying `--param
5855 inline-unit-growth'. The default is 10000
5857 `inline-unit-growth'
5858 Specifies maximal overall growth of the compilation unit
5859 caused by inlining. This parameter is ignored when
5860 `-funit-at-a-time' is not used. The default value is 50
5861 which limits unit growth to 1.5 times the original size.
5863 `max-inline-insns-recursive'
5864 `max-inline-insns-recursive-auto'
5865 Specifies maximum number of instructions out-of-line copy of
5866 self recursive inline function can grow into by performing
5869 For functions declared inline `--param
5870 max-inline-insns-recursive' is taken into account. For
5871 function not declared inline, recursive inlining happens only
5872 when `-finline-functions' (included in `-O3') is enabled and
5873 `--param max-inline-insns-recursive-auto' is used. The
5874 default value is 450.
5876 `max-inline-recursive-depth'
5877 `max-inline-recursive-depth-auto'
5878 Specifies maximum recursion depth used by the recursive
5881 For functions declared inline `--param
5882 max-inline-recursive-depth' is taken into account. For
5883 function not declared inline, recursive inlining happens only
5884 when `-finline-functions' (included in `-O3') is enabled and
5885 `--param max-inline-recursive-depth-auto' is used. The
5886 default value is 450.
5888 `min-inline-recursive-probability'
5889 Recursive inlining is profitable only for function having
5890 deep recursion in average and can hurt for function having
5891 little recursion depth by increasing the prologue size or
5892 complexity of function body to other optimizers.
5894 When profile feedback is available (see `-fprofile-generate')
5895 the actual recursion depth can be guessed from probability
5896 that function will recurse via given call expression. This
5897 parameter limits inlining only to call expression whose
5898 probability exceeds given threshold (in percents). The
5899 default value is 10.
5902 Specify cost of call instruction relative to simple
5903 arithmetics operations (having cost of 1). Increasing this
5904 cost disqualifies inlining of non-leaf functions and at the
5905 same time increases size of leaf function that is believed to
5906 reduce function size by being inlined. In effect it
5907 increases amount of inlining for code having large
5908 abstraction penalty (many functions that just pass the
5909 arguments to other functions) and decrease inlining for code
5910 with low abstraction penalty. The default value is 16.
5912 `max-unrolled-insns'
5913 The maximum number of instructions that a loop should have if
5914 that loop is unrolled, and if the loop is unrolled, it
5915 determines how many times the loop code is unrolled.
5917 `max-average-unrolled-insns'
5918 The maximum number of instructions biased by probabilities of
5919 their execution that a loop should have if that loop is
5920 unrolled, and if the loop is unrolled, it determines how many
5921 times the loop code is unrolled.
5924 The maximum number of unrollings of a single loop.
5927 The maximum number of instructions that a loop should have if
5928 that loop is peeled, and if the loop is peeled, it determines
5929 how many times the loop code is peeled.
5932 The maximum number of peelings of a single loop.
5934 `max-completely-peeled-insns'
5935 The maximum number of insns of a completely peeled loop.
5937 `max-completely-peel-times'
5938 The maximum number of iterations of a loop to be suitable for
5941 `max-unswitch-insns'
5942 The maximum number of insns of an unswitched loop.
5944 `max-unswitch-level'
5945 The maximum number of branches unswitched in a single loop.
5948 The minimum cost of an expensive expression in the loop
5951 `iv-consider-all-candidates-bound'
5952 Bound on number of candidates for induction variables below
5953 that all candidates are considered for each use in induction
5954 variable optimizations. Only the most relevant candidates
5955 are considered if there are more candidates, to avoid
5956 quadratic time complexity.
5958 `iv-max-considered-uses'
5959 The induction variable optimizations give up on loops that
5960 contain more induction variable uses.
5962 `iv-always-prune-cand-set-bound'
5963 If number of candidates in the set is smaller than this value,
5964 we always try to remove unnecessary ivs from the set during
5965 its optimization when a new iv is added to the set.
5967 `scev-max-expr-size'
5968 Bound on size of expressions used in the scalar evolutions
5969 analyzer. Large expressions slow the analyzer.
5971 `vect-max-version-checks'
5972 The maximum number of runtime checks that can be performed
5973 when doing loop versioning in the vectorizer. See option
5974 ftree-vect-loop-version for more information.
5976 `max-iterations-to-track'
5977 The maximum number of iterations of a loop the brute force
5978 algorithm for analysis of # of iterations of the loop tries
5981 `hot-bb-count-fraction'
5982 Select fraction of the maximal count of repetitions of basic
5983 block in program given basic block needs to have to be
5986 `hot-bb-frequency-fraction'
5987 Select fraction of the maximal frequency of executions of
5988 basic block in function given basic block needs to have to be
5991 `max-predicted-iterations'
5992 The maximum number of loop iterations we predict statically.
5993 This is useful in cases where function contain single loop
5994 with known bound and other loop with unknown. We predict the
5995 known number of iterations correctly, while the unknown
5996 number of iterations average to roughly 10. This means that
5997 the loop without bounds would appear artificially cold
5998 relative to the other one.
6000 `tracer-dynamic-coverage'
6001 `tracer-dynamic-coverage-feedback'
6002 This value is used to limit superblock formation once the
6003 given percentage of executed instructions is covered. This
6004 limits unnecessary code size expansion.
6006 The `tracer-dynamic-coverage-feedback' is used only when
6007 profile feedback is available. The real profiles (as opposed
6008 to statically estimated ones) are much less balanced allowing
6009 the threshold to be larger value.
6011 `tracer-max-code-growth'
6012 Stop tail duplication once code growth has reached given
6013 percentage. This is rather hokey argument, as most of the
6014 duplicates will be eliminated later in cross jumping, so it
6015 may be set to much higher values than is the desired code
6018 `tracer-min-branch-ratio'
6019 Stop reverse growth when the reverse probability of best edge
6020 is less than this threshold (in percent).
6022 `tracer-min-branch-ratio'
6023 `tracer-min-branch-ratio-feedback'
6024 Stop forward growth if the best edge do have probability
6025 lower than this threshold.
6027 Similarly to `tracer-dynamic-coverage' two values are
6028 present, one for compilation for profile feedback and one for
6029 compilation without. The value for compilation with profile
6030 feedback needs to be more conservative (higher) in order to
6031 make tracer effective.
6033 `max-cse-path-length'
6034 Maximum number of basic blocks on path that cse considers.
6038 The maximum instructions CSE process before flushing. The
6041 `global-var-threshold'
6042 Counts the number of function calls (N) and the number of
6043 call-clobbered variables (V). If NxV is larger than this
6044 limit, a single artificial variable will be created to
6045 represent all the call-clobbered variables at function call
6046 sites. This artificial variable will then be made to alias
6047 every call-clobbered variable. (done as `int * size_t' on
6048 the host machine; beware overflow).
6051 Maximum number of virtual operands allowed to represent
6052 aliases before triggering the alias grouping heuristic.
6053 Alias grouping reduces compile times and memory consumption
6054 needed for aliasing at the expense of precision loss in alias
6058 GCC uses a garbage collector to manage its own memory
6059 allocation. This parameter specifies the minimum percentage
6060 by which the garbage collector's heap should be allowed to
6061 expand between collections. Tuning this may improve
6062 compilation speed; it has no effect on code generation.
6064 The default is 30% + 70% * (RAM/1GB) with an upper bound of
6065 100% when RAM >= 1GB. If `getrlimit' is available, the
6066 notion of "RAM" is the smallest of actual RAM and
6067 `RLIMIT_DATA' or `RLIMIT_AS'. If GCC is not able to
6068 calculate RAM on a particular platform, the lower bound of
6069 30% is used. Setting this parameter and `ggc-min-heapsize'
6070 to zero causes a full collection to occur at every
6071 opportunity. This is extremely slow, but can be useful for
6075 Minimum size of the garbage collector's heap before it begins
6076 bothering to collect garbage. The first collection occurs
6077 after the heap expands by `ggc-min-expand'% beyond
6078 `ggc-min-heapsize'. Again, tuning this may improve
6079 compilation speed, and has no effect on code generation.
6081 The default is the smaller of RAM/8, RLIMIT_RSS, or a limit
6082 which tries to ensure that RLIMIT_DATA or RLIMIT_AS are not
6083 exceeded, but with a lower bound of 4096 (four megabytes) and
6084 an upper bound of 131072 (128 megabytes). If GCC is not able
6085 to calculate RAM on a particular platform, the lower bound is
6086 used. Setting this parameter very large effectively disables
6087 garbage collection. Setting this parameter and
6088 `ggc-min-expand' to zero causes a full collection to occur at
6091 `max-reload-search-insns'
6092 The maximum number of instruction reload should look backward
6093 for equivalent register. Increasing values mean more
6094 aggressive optimization, making the compile time increase
6095 with probably slightly better performance. The default value
6098 `max-cselib-memory-locations'
6099 The maximum number of memory locations cselib should take
6100 into account. Increasing values mean more aggressive
6101 optimization, making the compile time increase with probably
6102 slightly better performance. The default value is 500.
6104 `max-flow-memory-locations'
6105 Similar as `max-cselib-memory-locations' but for dataflow
6106 liveness. The default value is 100.
6108 `reorder-blocks-duplicate'
6109 `reorder-blocks-duplicate-feedback'
6110 Used by basic block reordering pass to decide whether to use
6111 unconditional branch or duplicate the code on its
6112 destination. Code is duplicated when its estimated size is
6113 smaller than this value multiplied by the estimated size of
6114 unconditional jump in the hot spots of the program.
6116 The `reorder-block-duplicate-feedback' is used only when
6117 profile feedback is available and may be set to higher values
6118 than `reorder-block-duplicate' since information about the
6119 hot spots is more accurate.
6121 `max-sched-ready-insns'
6122 The maximum number of instructions ready to be issued the
6123 scheduler should consider at any given time during the first
6124 scheduling pass. Increasing values mean more thorough
6125 searches, making the compilation time increase with probably
6126 little benefit. The default value is 100.
6128 `max-sched-region-blocks'
6129 The maximum number of blocks in a region to be considered for
6130 interblock scheduling. The default value is 10.
6132 `max-sched-region-insns'
6133 The maximum number of insns in a region to be considered for
6134 interblock scheduling. The default value is 100.
6137 The minimum probability (in percents) of reaching a source
6138 block for interblock speculative scheduling. The default
6141 `max-sched-extend-regions-iters'
6142 The maximum number of iterations through CFG to extend
6143 regions. 0 - disable region extension, N - do at most N
6144 iterations. The default value is 0.
6146 `max-sched-insn-conflict-delay'
6147 The maximum conflict delay for an insn to be considered for
6148 speculative motion. The default value is 3.
6150 `sched-spec-prob-cutoff'
6151 The minimal probability of speculation success (in percents),
6152 so that speculative insn will be scheduled. The default
6155 `max-last-value-rtl'
6156 The maximum size measured as number of RTLs that can be
6157 recorded in an expression in combiner for a pseudo register
6158 as last known value of that register. The default is 10000.
6160 `integer-share-limit'
6161 Small integer constants can use a shared data structure,
6162 reducing the compiler's memory usage and increasing its
6163 speed. This sets the maximum value of a shared integer
6164 constant's. The default value is 256.
6166 `min-virtual-mappings'
6167 Specifies the minimum number of virtual mappings in the
6168 incremental SSA updater that should be registered to trigger
6169 the virtual mappings heuristic defined by
6170 virtual-mappings-ratio. The default value is 100.
6172 `virtual-mappings-ratio'
6173 If the number of virtual mappings is virtual-mappings-ratio
6174 bigger than the number of virtual symbols to be updated, then
6175 the incremental SSA updater switches to a full update for
6176 those symbols. The default ratio is 3.
6179 The minimum size of buffers (i.e. arrays) that will receive
6180 stack smashing protection when `-fstack-protection' is used.
6182 `max-jump-thread-duplication-stmts'
6183 Maximum number of statements allowed in a block that needs to
6184 be duplicated when threading jumps.
6186 `max-fields-for-field-sensitive'
6187 Maximum number of fields in a structure we will treat in a
6188 field sensitive manner during pointer analysis.
6192 File: gcc.info, Node: Preprocessor Options, Next: Assembler Options, Prev: Optimize Options, Up: Invoking GCC
6194 3.11 Options Controlling the Preprocessor
6195 =========================================
6197 These options control the C preprocessor, which is run on each C source
6198 file before actual compilation.
6200 If you use the `-E' option, nothing is done except preprocessing.
6201 Some of these options make sense only together with `-E' because they
6202 cause the preprocessor output to be unsuitable for actual compilation.
6204 You can use `-Wp,OPTION' to bypass the compiler driver and pass
6205 OPTION directly through to the preprocessor. If OPTION contains
6206 commas, it is split into multiple options at the commas. However,
6207 many options are modified, translated or interpreted by the
6208 compiler driver before being passed to the preprocessor, and `-Wp'
6209 forcibly bypasses this phase. The preprocessor's direct interface
6210 is undocumented and subject to change, so whenever possible you
6211 should avoid using `-Wp' and let the driver handle the options
6214 `-Xpreprocessor OPTION'
6215 Pass OPTION as an option to the preprocessor. You can use this to
6216 supply system-specific preprocessor options which GCC does not
6217 know how to recognize.
6219 If you want to pass an option that takes an argument, you must use
6220 `-Xpreprocessor' twice, once for the option and once for the
6224 Predefine NAME as a macro, with definition `1'.
6226 `-D NAME=DEFINITION'
6227 The contents of DEFINITION are tokenized and processed as if they
6228 appeared during translation phase three in a `#define' directive.
6229 In particular, the definition will be truncated by embedded
6232 If you are invoking the preprocessor from a shell or shell-like
6233 program you may need to use the shell's quoting syntax to protect
6234 characters such as spaces that have a meaning in the shell syntax.
6236 If you wish to define a function-like macro on the command line,
6237 write its argument list with surrounding parentheses before the
6238 equals sign (if any). Parentheses are meaningful to most shells,
6239 so you will need to quote the option. With `sh' and `csh',
6240 `-D'NAME(ARGS...)=DEFINITION'' works.
6242 `-D' and `-U' options are processed in the order they are given on
6243 the command line. All `-imacros FILE' and `-include FILE' options
6244 are processed after all `-D' and `-U' options.
6247 Cancel any previous definition of NAME, either built in or
6248 provided with a `-D' option.
6251 Do not predefine any system-specific or GCC-specific macros. The
6252 standard predefined macros remain defined.
6255 Add the directory DIR to the list of directories to be searched
6256 for header files. Directories named by `-I' are searched before
6257 the standard system include directories. If the directory DIR is
6258 a standard system include directory, the option is ignored to
6259 ensure that the default search order for system directories and
6260 the special treatment of system headers are not defeated .
6263 Write output to FILE. This is the same as specifying FILE as the
6264 second non-option argument to `cpp'. `gcc' has a different
6265 interpretation of a second non-option argument, so you must use
6266 `-o' to specify the output file.
6269 Turns on all optional warnings which are desirable for normal code.
6270 At present this is `-Wcomment', `-Wtrigraphs', `-Wmultichar' and a
6271 warning about integer promotion causing a change of sign in `#if'
6272 expressions. Note that many of the preprocessor's warnings are on
6273 by default and have no options to control them.
6277 Warn whenever a comment-start sequence `/*' appears in a `/*'
6278 comment, or whenever a backslash-newline appears in a `//' comment.
6279 (Both forms have the same effect.)
6282 Most trigraphs in comments cannot affect the meaning of the
6283 program. However, a trigraph that would form an escaped newline
6284 (`??/' at the end of a line) can, by changing where the comment
6285 begins or ends. Therefore, only trigraphs that would form escaped
6286 newlines produce warnings inside a comment.
6288 This option is implied by `-Wall'. If `-Wall' is not given, this
6289 option is still enabled unless trigraphs are enabled. To get
6290 trigraph conversion without warnings, but get the other `-Wall'
6291 warnings, use `-trigraphs -Wall -Wno-trigraphs'.
6294 Warn about certain constructs that behave differently in
6295 traditional and ISO C. Also warn about ISO C constructs that have
6296 no traditional C equivalent, and problematic constructs which
6300 Warn the first time `#import' is used.
6303 Warn whenever an identifier which is not a macro is encountered in
6304 an `#if' directive, outside of `defined'. Such identifiers are
6308 Warn about macros defined in the main file that are unused. A
6309 macro is "used" if it is expanded or tested for existence at least
6310 once. The preprocessor will also warn if the macro has not been
6311 used at the time it is redefined or undefined.
6313 Built-in macros, macros defined on the command line, and macros
6314 defined in include files are not warned about.
6316 _Note:_ If a macro is actually used, but only used in skipped
6317 conditional blocks, then CPP will report it as unused. To avoid
6318 the warning in such a case, you might improve the scope of the
6319 macro's definition by, for example, moving it into the first
6320 skipped block. Alternatively, you could provide a dummy use with
6323 #if defined the_macro_causing_the_warning
6327 Warn whenever an `#else' or an `#endif' are followed by text.
6328 This usually happens in code of the form
6336 The second and third `FOO' should be in comments, but often are not
6337 in older programs. This warning is on by default.
6340 Make all warnings into hard errors. Source code which triggers
6341 warnings will be rejected.
6344 Issue warnings for code in system headers. These are normally
6345 unhelpful in finding bugs in your own code, therefore suppressed.
6346 If you are responsible for the system library, you may want to see
6350 Suppress all warnings, including those which GNU CPP issues by
6354 Issue all the mandatory diagnostics listed in the C standard.
6355 Some of them are left out by default, since they trigger
6356 frequently on harmless code.
6359 Issue all the mandatory diagnostics, and make all mandatory
6360 diagnostics into errors. This includes mandatory diagnostics that
6361 GCC issues without `-pedantic' but treats as warnings.
6364 Instead of outputting the result of preprocessing, output a rule
6365 suitable for `make' describing the dependencies of the main source
6366 file. The preprocessor outputs one `make' rule containing the
6367 object file name for that source file, a colon, and the names of
6368 all the included files, including those coming from `-include' or
6369 `-imacros' command line options.
6371 Unless specified explicitly (with `-MT' or `-MQ'), the object file
6372 name consists of the basename of the source file with any suffix
6373 replaced with object file suffix. If there are many included
6374 files then the rule is split into several lines using `\'-newline.
6375 The rule has no commands.
6377 This option does not suppress the preprocessor's debug output,
6378 such as `-dM'. To avoid mixing such debug output with the
6379 dependency rules you should explicitly specify the dependency
6380 output file with `-MF', or use an environment variable like
6381 `DEPENDENCIES_OUTPUT' (*note Environment Variables::). Debug
6382 output will still be sent to the regular output stream as normal.
6384 Passing `-M' to the driver implies `-E', and suppresses warnings
6385 with an implicit `-w'.
6388 Like `-M' but do not mention header files that are found in system
6389 header directories, nor header files that are included, directly
6390 or indirectly, from such a header.
6392 This implies that the choice of angle brackets or double quotes in
6393 an `#include' directive does not in itself determine whether that
6394 header will appear in `-MM' dependency output. This is a slight
6395 change in semantics from GCC versions 3.0 and earlier.
6398 When used with `-M' or `-MM', specifies a file to write the
6399 dependencies to. If no `-MF' switch is given the preprocessor
6400 sends the rules to the same place it would have sent preprocessed
6403 When used with the driver options `-MD' or `-MMD', `-MF' overrides
6404 the default dependency output file.
6407 In conjunction with an option such as `-M' requesting dependency
6408 generation, `-MG' assumes missing header files are generated files
6409 and adds them to the dependency list without raising an error.
6410 The dependency filename is taken directly from the `#include'
6411 directive without prepending any path. `-MG' also suppresses
6412 preprocessed output, as a missing header file renders this useless.
6414 This feature is used in automatic updating of makefiles.
6417 This option instructs CPP to add a phony target for each dependency
6418 other than the main file, causing each to depend on nothing. These
6419 dummy rules work around errors `make' gives if you remove header
6420 files without updating the `Makefile' to match.
6422 This is typical output:
6424 test.o: test.c test.h
6429 Change the target of the rule emitted by dependency generation. By
6430 default CPP takes the name of the main input file, including any
6431 path, deletes any file suffix such as `.c', and appends the
6432 platform's usual object suffix. The result is the target.
6434 An `-MT' option will set the target to be exactly the string you
6435 specify. If you want multiple targets, you can specify them as a
6436 single argument to `-MT', or use multiple `-MT' options.
6438 For example, `-MT '$(objpfx)foo.o'' might give
6440 $(objpfx)foo.o: foo.c
6443 Same as `-MT', but it quotes any characters which are special to
6444 Make. `-MQ '$(objpfx)foo.o'' gives
6446 $$(objpfx)foo.o: foo.c
6448 The default target is automatically quoted, as if it were given
6452 `-MD' is equivalent to `-M -MF FILE', except that `-E' is not
6453 implied. The driver determines FILE based on whether an `-o'
6454 option is given. If it is, the driver uses its argument but with
6455 a suffix of `.d', otherwise it take the basename of the input file
6456 and applies a `.d' suffix.
6458 If `-MD' is used in conjunction with `-E', any `-o' switch is
6459 understood to specify the dependency output file (*note -MF:
6460 dashMF.), but if used without `-E', each `-o' is understood to
6461 specify a target object file.
6463 Since `-E' is not implied, `-MD' can be used to generate a
6464 dependency output file as a side-effect of the compilation process.
6467 Like `-MD' except mention only user header files, not system
6471 When using precompiled headers (*note Precompiled Headers::), this
6472 flag will cause the dependency-output flags to also list the files
6473 from the precompiled header's dependencies. If not specified only
6474 the precompiled header would be listed and not the files that were
6475 used to create it because those files are not consulted when a
6476 precompiled header is used.
6479 This option allows use of a precompiled header (*note Precompiled
6480 Headers::) together with `-E'. It inserts a special `#pragma',
6481 `#pragma GCC pch_preprocess "<filename>"' in the output to mark
6482 the place where the precompiled header was found, and its
6483 filename. When `-fpreprocessed' is in use, GCC recognizes this
6484 `#pragma' and loads the PCH.
6486 This option is off by default, because the resulting preprocessed
6487 output is only really suitable as input to GCC. It is switched on
6490 You should not write this `#pragma' in your own code, but it is
6491 safe to edit the filename if the PCH file is available in a
6492 different location. The filename may be absolute or it may be
6493 relative to GCC's current directory.
6498 `-x assembler-with-cpp'
6499 Specify the source language: C, C++, Objective-C, or assembly.
6500 This has nothing to do with standards conformance or extensions;
6501 it merely selects which base syntax to expect. If you give none
6502 of these options, cpp will deduce the language from the extension
6503 of the source file: `.c', `.cc', `.m', or `.S'. Some other common
6504 extensions for C++ and assembly are also recognized. If cpp does
6505 not recognize the extension, it will treat the file as C; this is
6506 the most generic mode.
6508 _Note:_ Previous versions of cpp accepted a `-lang' option which
6509 selected both the language and the standards conformance level.
6510 This option has been removed, because it conflicts with the `-l'
6515 Specify the standard to which the code should conform. Currently
6516 CPP knows about C and C++ standards; others may be added in the
6519 STANDARD may be one of:
6522 The ISO C standard from 1990. `c89' is the customary
6523 shorthand for this version of the standard.
6525 The `-ansi' option is equivalent to `-std=c89'.
6528 The 1990 C standard, as amended in 1994.
6534 The revised ISO C standard, published in December 1999.
6535 Before publication, this was known as C9X.
6538 The 1990 C standard plus GNU extensions. This is the default.
6542 The 1999 C standard plus GNU extensions.
6545 The 1998 ISO C++ standard plus amendments.
6548 The same as `-std=c++98' plus GNU extensions. This is the
6549 default for C++ code.
6552 Split the include path. Any directories specified with `-I'
6553 options before `-I-' are searched only for headers requested with
6554 `#include "FILE"'; they are not searched for `#include <FILE>'.
6555 If additional directories are specified with `-I' options after
6556 the `-I-', those directories are searched for all `#include'
6559 In addition, `-I-' inhibits the use of the directory of the current
6560 file directory as the first search directory for `#include "FILE"'.
6561 This option has been deprecated.
6564 Do not search the standard system directories for header files.
6565 Only the directories you have specified with `-I' options (and the
6566 directory of the current file, if appropriate) are searched.
6569 Do not search for header files in the C++-specific standard
6570 directories, but do still search the other standard directories.
6571 (This option is used when building the C++ library.)
6574 Process FILE as if `#include "file"' appeared as the first line of
6575 the primary source file. However, the first directory searched
6576 for FILE is the preprocessor's working directory _instead of_ the
6577 directory containing the main source file. If not found there, it
6578 is searched for in the remainder of the `#include "..."' search
6581 If multiple `-include' options are given, the files are included
6582 in the order they appear on the command line.
6585 Exactly like `-include', except that any output produced by
6586 scanning FILE is thrown away. Macros it defines remain defined.
6587 This allows you to acquire all the macros from a header without
6588 also processing its declarations.
6590 All files specified by `-imacros' are processed before all files
6591 specified by `-include'.
6594 Search DIR for header files, but do it _after_ all directories
6595 specified with `-I' and the standard system directories have been
6596 exhausted. DIR is treated as a system include directory.
6599 Specify PREFIX as the prefix for subsequent `-iwithprefix'
6600 options. If the prefix represents a directory, you should include
6604 `-iwithprefixbefore DIR'
6605 Append DIR to the prefix specified previously with `-iprefix', and
6606 add the resulting directory to the include search path.
6607 `-iwithprefixbefore' puts it in the same place `-I' would;
6608 `-iwithprefix' puts it where `-idirafter' would.
6611 This option is like the `--sysroot' option, but applies only to
6612 header files. See the `--sysroot' option for more information.
6615 Use DIR as a subdirectory of the directory containing
6616 target-specific C++ headers.
6619 Search DIR for header files, after all directories specified by
6620 `-I' but before the standard system directories. Mark it as a
6621 system directory, so that it gets the same special treatment as is
6622 applied to the standard system directories.
6625 Search DIR only for header files requested with `#include "FILE"';
6626 they are not searched for `#include <FILE>', before all
6627 directories specified by `-I' and before the standard system
6631 This option provides a simplified preprocessor to improve the
6632 performance of distributed build systems such as distcc. It's
6633 behavior depends on a number of other flags.
6635 If the `-E' option is enabled, it suppresses things like macro
6636 expansion, trigraph conversion, and escaped newline splicing
6637 outside of directives. All directives are processed normally,
6638 except that macro definitions are output similar to the `-dD'
6641 If the `-fpreprocessed' option is enabled, it suppresses
6642 predefinition of most builtin and command line macros. This
6643 prevents duplicate definition of macros output with the `-E'
6646 `-fdollars-in-identifiers'
6647 Accept `$' in identifiers.
6649 `-fextended-identifiers'
6650 Accept universal character names in identifiers. This option is
6651 experimental; in a future version of GCC, it will be enabled by
6652 default for C99 and C++.
6655 Indicate to the preprocessor that the input file has already been
6656 preprocessed. This suppresses things like macro expansion,
6657 trigraph conversion, escaped newline splicing, and processing of
6658 most directives. The preprocessor still recognizes and removes
6659 comments, so that you can pass a file preprocessed with `-C' to
6660 the compiler without problems. In this mode the integrated
6661 preprocessor is little more than a tokenizer for the front ends.
6663 `-fpreprocessed' is implicit if the input file has one of the
6664 extensions `.i', `.ii' or `.mi'. These are the extensions that
6665 GCC uses for preprocessed files created by `-save-temps'.
6668 Set the distance between tab stops. This helps the preprocessor
6669 report correct column numbers in warnings or errors, even if tabs
6670 appear on the line. If the value is less than 1 or greater than
6671 100, the option is ignored. The default is 8.
6673 `-fexec-charset=CHARSET'
6674 Set the execution character set, used for string and character
6675 constants. The default is UTF-8. CHARSET can be any encoding
6676 supported by the system's `iconv' library routine.
6678 `-fwide-exec-charset=CHARSET'
6679 Set the wide execution character set, used for wide string and
6680 character constants. The default is UTF-32 or UTF-16, whichever
6681 corresponds to the width of `wchar_t'. As with `-fexec-charset',
6682 CHARSET can be any encoding supported by the system's `iconv'
6683 library routine; however, you will have problems with encodings
6684 that do not fit exactly in `wchar_t'.
6686 `-finput-charset=CHARSET'
6687 Set the input character set, used for translation from the
6688 character set of the input file to the source character set used
6689 by GCC. If the locale does not specify, or GCC cannot get this
6690 information from the locale, the default is UTF-8. This can be
6691 overridden by either the locale or this command line option.
6692 Currently the command line option takes precedence if there's a
6693 conflict. CHARSET can be any encoding supported by the system's
6694 `iconv' library routine.
6696 `-fworking-directory'
6697 Enable generation of linemarkers in the preprocessor output that
6698 will let the compiler know the current working directory at the
6699 time of preprocessing. When this option is enabled, the
6700 preprocessor will emit, after the initial linemarker, a second
6701 linemarker with the current working directory followed by two
6702 slashes. GCC will use this directory, when it's present in the
6703 preprocessed input, as the directory emitted as the current
6704 working directory in some debugging information formats. This
6705 option is implicitly enabled if debugging information is enabled,
6706 but this can be inhibited with the negated form
6707 `-fno-working-directory'. If the `-P' flag is present in the
6708 command line, this option has no effect, since no `#line'
6709 directives are emitted whatsoever.
6712 Do not print column numbers in diagnostics. This may be necessary
6713 if diagnostics are being scanned by a program that does not
6714 understand the column numbers, such as `dejagnu'.
6716 `-A PREDICATE=ANSWER'
6717 Make an assertion with the predicate PREDICATE and answer ANSWER.
6718 This form is preferred to the older form `-A PREDICATE(ANSWER)',
6719 which is still supported, because it does not use shell special
6722 `-A -PREDICATE=ANSWER'
6723 Cancel an assertion with the predicate PREDICATE and answer ANSWER.
6726 CHARS is a sequence of one or more of the following characters,
6727 and must not be preceded by a space. Other characters are
6728 interpreted by the compiler proper, or reserved for future
6729 versions of GCC, and so are silently ignored. If you specify
6730 characters whose behavior conflicts, the result is undefined.
6733 Instead of the normal output, generate a list of `#define'
6734 directives for all the macros defined during the execution of
6735 the preprocessor, including predefined macros. This gives
6736 you a way of finding out what is predefined in your version
6737 of the preprocessor. Assuming you have no file `foo.h', the
6740 touch foo.h; cpp -dM foo.h
6742 will show all the predefined macros.
6745 Like `M' except in two respects: it does _not_ include the
6746 predefined macros, and it outputs _both_ the `#define'
6747 directives and the result of preprocessing. Both kinds of
6748 output go to the standard output file.
6751 Like `D', but emit only the macro names, not their expansions.
6754 Output `#include' directives in addition to the result of
6758 Inhibit generation of linemarkers in the output from the
6759 preprocessor. This might be useful when running the preprocessor
6760 on something that is not C code, and will be sent to a program
6761 which might be confused by the linemarkers.
6764 Do not discard comments. All comments are passed through to the
6765 output file, except for comments in processed directives, which
6766 are deleted along with the directive.
6768 You should be prepared for side effects when using `-C'; it causes
6769 the preprocessor to treat comments as tokens in their own right.
6770 For example, comments appearing at the start of what would be a
6771 directive line have the effect of turning that line into an
6772 ordinary source line, since the first token on the line is no
6776 Do not discard comments, including during macro expansion. This is
6777 like `-C', except that comments contained within macros are also
6778 passed through to the output file where the macro is expanded.
6780 In addition to the side-effects of the `-C' option, the `-CC'
6781 option causes all C++-style comments inside a macro to be
6782 converted to C-style comments. This is to prevent later use of
6783 that macro from inadvertently commenting out the remainder of the
6786 The `-CC' option is generally used to support lint comments.
6789 Try to imitate the behavior of old-fashioned C preprocessors, as
6790 opposed to ISO C preprocessors.
6793 Process trigraph sequences. These are three-character sequences,
6794 all starting with `??', that are defined by ISO C to stand for
6795 single characters. For example, `??/' stands for `\', so `'??/n''
6796 is a character constant for a newline. By default, GCC ignores
6797 trigraphs, but in standard-conforming modes it converts them. See
6798 the `-std' and `-ansi' options.
6800 The nine trigraphs and their replacements are
6802 Trigraph: ??( ??) ??< ??> ??= ??/ ??' ??! ??-
6803 Replacement: [ ] { } # \ ^ | ~
6806 Enable special code to work around file systems which only permit
6807 very short file names, such as MS-DOS.
6811 Print text describing all the command line options instead of
6812 preprocessing anything.
6815 Verbose mode. Print out GNU CPP's version number at the beginning
6816 of execution, and report the final form of the include path.
6819 Print the name of each header file used, in addition to other
6820 normal activities. Each name is indented to show how deep in the
6821 `#include' stack it is. Precompiled header files are also
6822 printed, even if they are found to be invalid; an invalid
6823 precompiled header file is printed with `...x' and a valid one
6828 Print out GNU CPP's version number. With one dash, proceed to
6829 preprocess as normal. With two dashes, exit immediately.
6832 File: gcc.info, Node: Assembler Options, Next: Link Options, Prev: Preprocessor Options, Up: Invoking GCC
6834 3.12 Passing Options to the Assembler
6835 =====================================
6837 You can pass options to the assembler.
6840 Pass OPTION as an option to the assembler. If OPTION contains
6841 commas, it is split into multiple options at the commas.
6843 `-Xassembler OPTION'
6844 Pass OPTION as an option to the assembler. You can use this to
6845 supply system-specific assembler options which GCC does not know
6848 If you want to pass an option that takes an argument, you must use
6849 `-Xassembler' twice, once for the option and once for the argument.
6853 File: gcc.info, Node: Link Options, Next: Directory Options, Prev: Assembler Options, Up: Invoking GCC
6855 3.13 Options for Linking
6856 ========================
6858 These options come into play when the compiler links object files into
6859 an executable output file. They are meaningless if the compiler is not
6863 A file name that does not end in a special recognized suffix is
6864 considered to name an object file or library. (Object files are
6865 distinguished from libraries by the linker according to the file
6866 contents.) If linking is done, these object files are used as
6867 input to the linker.
6872 If any of these options is used, then the linker is not run, and
6873 object file names should not be used as arguments. *Note Overall
6878 Search the library named LIBRARY when linking. (The second
6879 alternative with the library as a separate argument is only for
6880 POSIX compliance and is not recommended.)
6882 It makes a difference where in the command you write this option;
6883 the linker searches and processes libraries and object files in
6884 the order they are specified. Thus, `foo.o -lz bar.o' searches
6885 library `z' after file `foo.o' but before `bar.o'. If `bar.o'
6886 refers to functions in `z', those functions may not be loaded.
6888 The linker searches a standard list of directories for the library,
6889 which is actually a file named `libLIBRARY.a'. The linker then
6890 uses this file as if it had been specified precisely by name.
6892 The directories searched include several standard system
6893 directories plus any that you specify with `-L'.
6895 Normally the files found this way are library files--archive files
6896 whose members are object files. The linker handles an archive
6897 file by scanning through it for members which define symbols that
6898 have so far been referenced but not defined. But if the file that
6899 is found is an ordinary object file, it is linked in the usual
6900 fashion. The only difference between using an `-l' option and
6901 specifying a file name is that `-l' surrounds LIBRARY with `lib'
6902 and `.a' and searches several directories.
6905 You need this special case of the `-l' option in order to link an
6906 Objective-C or Objective-C++ program.
6909 Do not use the standard system startup files when linking. The
6910 standard system libraries are used normally, unless `-nostdlib' or
6911 `-nodefaultlibs' is used.
6914 Do not use the standard system libraries when linking. Only the
6915 libraries you specify will be passed to the linker. The standard
6916 startup files are used normally, unless `-nostartfiles' is used.
6917 The compiler may generate calls to `memcmp', `memset', `memcpy'
6918 and `memmove'. These entries are usually resolved by entries in
6919 libc. These entry points should be supplied through some other
6920 mechanism when this option is specified.
6923 Do not use the standard system startup files or libraries when
6924 linking. No startup files and only the libraries you specify will
6925 be passed to the linker. The compiler may generate calls to
6926 `memcmp', `memset', `memcpy' and `memmove'. These entries are
6927 usually resolved by entries in libc. These entry points should be
6928 supplied through some other mechanism when this option is
6931 One of the standard libraries bypassed by `-nostdlib' and
6932 `-nodefaultlibs' is `libgcc.a', a library of internal subroutines
6933 that GCC uses to overcome shortcomings of particular machines, or
6934 special needs for some languages. (*Note Interfacing to GCC
6935 Output: (gccint)Interface, for more discussion of `libgcc.a'.) In
6936 most cases, you need `libgcc.a' even when you want to avoid other
6937 standard libraries. In other words, when you specify `-nostdlib'
6938 or `-nodefaultlibs' you should usually specify `-lgcc' as well.
6939 This ensures that you have no unresolved references to internal GCC
6940 library subroutines. (For example, `__main', used to ensure C++
6941 constructors will be called; *note `collect2': (gccint)Collect2.)
6944 Produce a position independent executable on targets which support
6945 it. For predictable results, you must also specify the same set
6946 of options that were used to generate code (`-fpie', `-fPIE', or
6947 model suboptions) when you specify this option.
6950 Pass the flag `-export-dynamic' to the ELF linker, on targets that
6951 support it. This instructs the linker to add all symbols, not only
6952 used ones, to the dynamic symbol table. This option is needed for
6953 some uses of `dlopen' or to allow obtaining backtraces from within
6957 Remove all symbol table and relocation information from the
6961 On systems that support dynamic linking, this prevents linking
6962 with the shared libraries. On other systems, this option has no
6966 Produce a shared object which can then be linked with other
6967 objects to form an executable. Not all systems support this
6968 option. For predictable results, you must also specify the same
6969 set of options that were used to generate code (`-fpic', `-fPIC',
6970 or model suboptions) when you specify this option.(1)
6974 On systems that provide `libgcc' as a shared library, these options
6975 force the use of either the shared or static version respectively.
6976 If no shared version of `libgcc' was built when the compiler was
6977 configured, these options have no effect.
6979 There are several situations in which an application should use the
6980 shared `libgcc' instead of the static version. The most common of
6981 these is when the application wishes to throw and catch exceptions
6982 across different shared libraries. In that case, each of the
6983 libraries as well as the application itself should use the shared
6986 Therefore, the G++ and GCJ drivers automatically add
6987 `-shared-libgcc' whenever you build a shared library or a main
6988 executable, because C++ and Java programs typically use
6989 exceptions, so this is the right thing to do.
6991 If, instead, you use the GCC driver to create shared libraries,
6992 you may find that they will not always be linked with the shared
6993 `libgcc'. If GCC finds, at its configuration time, that you have
6994 a non-GNU linker or a GNU linker that does not support option
6995 `--eh-frame-hdr', it will link the shared version of `libgcc' into
6996 shared libraries by default. Otherwise, it will take advantage of
6997 the linker and optimize away the linking with the shared version
6998 of `libgcc', linking with the static version of libgcc by default.
6999 This allows exceptions to propagate through such shared
7000 libraries, without incurring relocation costs at library load time.
7002 However, if a library or main executable is supposed to throw or
7003 catch exceptions, you must link it using the G++ or GCJ driver, as
7004 appropriate for the languages used in the program, or using the
7005 option `-shared-libgcc', such that it is linked with the shared
7009 Bind references to global symbols when building a shared object.
7010 Warn about any unresolved references (unless overridden by the
7011 link editor option `-Xlinker -z -Xlinker defs'). Only a few
7012 systems support this option.
7015 Pass OPTION as an option to the linker. You can use this to
7016 supply system-specific linker options which GCC does not know how
7019 If you want to pass an option that takes an argument, you must use
7020 `-Xlinker' twice, once for the option and once for the argument.
7021 For example, to pass `-assert definitions', you must write
7022 `-Xlinker -assert -Xlinker definitions'. It does not work to write
7023 `-Xlinker "-assert definitions"', because this passes the entire
7024 string as a single argument, which is not what the linker expects.
7027 Pass OPTION as an option to the linker. If OPTION contains
7028 commas, it is split into multiple options at the commas.
7031 Pretend the symbol SYMBOL is undefined, to force linking of
7032 library modules to define it. You can use `-u' multiple times with
7033 different symbols to force loading of additional library modules.
7035 ---------- Footnotes ----------
7037 (1) On some systems, `gcc -shared' needs to build supplementary stub
7038 code for constructors to work. On multi-libbed systems, `gcc -shared'
7039 must select the correct support libraries to link against. Failing to
7040 supply the correct flags may lead to subtle defects. Supplying them in
7041 cases where they are not necessary is innocuous.
7044 File: gcc.info, Node: Directory Options, Next: Spec Files, Prev: Link Options, Up: Invoking GCC
7046 3.14 Options for Directory Search
7047 =================================
7049 These options specify directories to search for header files, for
7050 libraries and for parts of the compiler:
7053 Add the directory DIR to the head of the list of directories to be
7054 searched for header files. This can be used to override a system
7055 header file, substituting your own version, since these
7056 directories are searched before the system header file
7057 directories. However, you should not use this option to add
7058 directories that contain vendor-supplied system header files (use
7059 `-isystem' for that). If you use more than one `-I' option, the
7060 directories are scanned in left-to-right order; the standard
7061 system directories come after.
7063 If a standard system include directory, or a directory specified
7064 with `-isystem', is also specified with `-I', the `-I' option will
7065 be ignored. The directory will still be searched but as a system
7066 directory at its normal position in the system include chain.
7067 This is to ensure that GCC's procedure to fix buggy system headers
7068 and the ordering for the include_next directive are not
7069 inadvertently changed. If you really need to change the search
7070 order for system directories, use the `-nostdinc' and/or
7074 Add the directory DIR to the head of the list of directories to be
7075 searched for header files only for the case of `#include "FILE"';
7076 they are not searched for `#include <FILE>', otherwise just like
7080 Add directory DIR to the list of directories to be searched for
7084 This option specifies where to find the executables, libraries,
7085 include files, and data files of the compiler itself.
7087 The compiler driver program runs one or more of the subprograms
7088 `cpp', `cc1', `as' and `ld'. It tries PREFIX as a prefix for each
7089 program it tries to run, both with and without `MACHINE/VERSION/'
7090 (*note Target Options::).
7092 For each subprogram to be run, the compiler driver first tries the
7093 `-B' prefix, if any. If that name is not found, or if `-B' was
7094 not specified, the driver tries two standard prefixes, which are
7095 `/usr/lib/gcc/' and `/usr/local/lib/gcc/'. If neither of those
7096 results in a file name that is found, the unmodified program name
7097 is searched for using the directories specified in your `PATH'
7098 environment variable.
7100 The compiler will check to see if the path provided by the `-B'
7101 refers to a directory, and if necessary it will add a directory
7102 separator character at the end of the path.
7104 `-B' prefixes that effectively specify directory names also apply
7105 to libraries in the linker, because the compiler translates these
7106 options into `-L' options for the linker. They also apply to
7107 includes files in the preprocessor, because the compiler
7108 translates these options into `-isystem' options for the
7109 preprocessor. In this case, the compiler appends `include' to the
7112 The run-time support file `libgcc.a' can also be searched for using
7113 the `-B' prefix, if needed. If it is not found there, the two
7114 standard prefixes above are tried, and that is all. The file is
7115 left out of the link if it is not found by those means.
7117 Another way to specify a prefix much like the `-B' prefix is to use
7118 the environment variable `GCC_EXEC_PREFIX'. *Note Environment
7121 As a special kludge, if the path provided by `-B' is
7122 `[dir/]stageN/', where N is a number in the range 0 to 9, then it
7123 will be replaced by `[dir/]include'. This is to help with
7124 boot-strapping the compiler.
7127 Process FILE after the compiler reads in the standard `specs'
7128 file, in order to override the defaults that the `gcc' driver
7129 program uses when determining what switches to pass to `cc1',
7130 `cc1plus', `as', `ld', etc. More than one `-specs=FILE' can be
7131 specified on the command line, and they are processed in order,
7135 Use DIR as the logical root directory for headers and libraries.
7136 For example, if the compiler would normally search for headers in
7137 `/usr/include' and libraries in `/usr/lib', it will instead search
7138 `DIR/usr/include' and `DIR/usr/lib'.
7140 If you use both this option and the `-isysroot' option, then the
7141 `--sysroot' option will apply to libraries, but the `-isysroot'
7142 option will apply to header files.
7144 The GNU linker (beginning with version 2.16) has the necessary
7145 support for this option. If your linker does not support this
7146 option, the header file aspect of `--sysroot' will still work, but
7147 the library aspect will not.
7150 This option has been deprecated. Please use `-iquote' instead for
7151 `-I' directories before the `-I-' and remove the `-I-'. Any
7152 directories you specify with `-I' options before the `-I-' option
7153 are searched only for the case of `#include "FILE"'; they are not
7154 searched for `#include <FILE>'.
7156 If additional directories are specified with `-I' options after
7157 the `-I-', these directories are searched for all `#include'
7158 directives. (Ordinarily _all_ `-I' directories are used this way.)
7160 In addition, the `-I-' option inhibits the use of the current
7161 directory (where the current input file came from) as the first
7162 search directory for `#include "FILE"'. There is no way to
7163 override this effect of `-I-'. With `-I.' you can specify
7164 searching the directory which was current when the compiler was
7165 invoked. That is not exactly the same as what the preprocessor
7166 does by default, but it is often satisfactory.
7168 `-I-' does not inhibit the use of the standard system directories
7169 for header files. Thus, `-I-' and `-nostdinc' are independent.
7172 File: gcc.info, Node: Spec Files, Next: Target Options, Prev: Directory Options, Up: Invoking GCC
7174 3.15 Specifying subprocesses and the switches to pass to them
7175 =============================================================
7177 `gcc' is a driver program. It performs its job by invoking a sequence
7178 of other programs to do the work of compiling, assembling and linking.
7179 GCC interprets its command-line parameters and uses these to deduce
7180 which programs it should invoke, and which command-line options it
7181 ought to place on their command lines. This behavior is controlled by
7182 "spec strings". In most cases there is one spec string for each
7183 program that GCC can invoke, but a few programs have multiple spec
7184 strings to control their behavior. The spec strings built into GCC can
7185 be overridden by using the `-specs=' command-line switch to specify a
7188 "Spec files" are plaintext files that are used to construct spec
7189 strings. They consist of a sequence of directives separated by blank
7190 lines. The type of directive is determined by the first non-whitespace
7191 character on the line and it can be one of the following:
7194 Issues a COMMAND to the spec file processor. The commands that can
7198 Search for FILE and insert its text at the current point in
7201 `%include_noerr <FILE>'
7202 Just like `%include', but do not generate an error message if
7203 the include file cannot be found.
7205 `%rename OLD_NAME NEW_NAME'
7206 Rename the spec string OLD_NAME to NEW_NAME.
7210 This tells the compiler to create, override or delete the named
7211 spec string. All lines after this directive up to the next
7212 directive or blank line are considered to be the text for the spec
7213 string. If this results in an empty string then the spec will be
7214 deleted. (Or, if the spec did not exist, then nothing will
7215 happened.) Otherwise, if the spec does not currently exist a new
7216 spec will be created. If the spec does exist then its contents
7217 will be overridden by the text of this directive, unless the first
7218 character of that text is the `+' character, in which case the
7219 text will be appended to the spec.
7222 Creates a new `[SUFFIX] spec' pair. All lines after this directive
7223 and up to the next directive or blank line are considered to make
7224 up the spec string for the indicated suffix. When the compiler
7225 encounters an input file with the named suffix, it will processes
7226 the spec string in order to work out how to compile that file.
7232 This says that any input file whose name ends in `.ZZ' should be
7233 passed to the program `z-compile', which should be invoked with the
7234 command-line switch `-input' and with the result of performing the
7235 `%i' substitution. (See below.)
7237 As an alternative to providing a spec string, the text that
7238 follows a suffix directive can be one of the following:
7241 This says that the suffix is an alias for a known LANGUAGE.
7242 This is similar to using the `-x' command-line switch to GCC
7243 to specify a language explicitly. For example:
7248 Says that .ZZ files are, in fact, C++ source files.
7251 This causes an error messages saying:
7253 NAME compiler not installed on this system.
7255 GCC already has an extensive list of suffixes built into it. This
7256 directive will add an entry to the end of the list of suffixes, but
7257 since the list is searched from the end backwards, it is
7258 effectively possible to override earlier entries using this
7262 GCC has the following spec strings built into it. Spec files can
7263 override these strings or create their own. Note that individual
7264 targets can also add their own spec strings to this list.
7266 asm Options to pass to the assembler
7267 asm_final Options to pass to the assembler post-processor
7268 cpp Options to pass to the C preprocessor
7269 cc1 Options to pass to the C compiler
7270 cc1plus Options to pass to the C++ compiler
7271 endfile Object files to include at the end of the link
7272 link Options to pass to the linker
7273 lib Libraries to include on the command line to the linker
7274 libgcc Decides which GCC support library to pass to the linker
7275 linker Sets the name of the linker
7276 predefines Defines to be passed to the C preprocessor
7277 signed_char Defines to pass to CPP to say whether `char' is signed
7279 startfile Object files to include at the start of the link
7281 Here is a small example of a spec file:
7286 --start-group -lgcc -lc -leval1 --end-group %(old_lib)
7288 This example renames the spec called `lib' to `old_lib' and then
7289 overrides the previous definition of `lib' with a new one. The new
7290 definition adds in some extra command-line options before including the
7291 text of the old definition.
7293 "Spec strings" are a list of command-line options to be passed to their
7294 corresponding program. In addition, the spec strings can contain
7295 `%'-prefixed sequences to substitute variable text or to conditionally
7296 insert text into the command line. Using these constructs it is
7297 possible to generate quite complex command lines.
7299 Here is a table of all defined `%'-sequences for spec strings. Note
7300 that spaces are not generated automatically around the results of
7301 expanding these sequences. Therefore you can concatenate them together
7302 or combine them with constant text in a single argument.
7305 Substitute one `%' into the program name or argument.
7308 Substitute the name of the input file being processed.
7311 Substitute the basename of the input file being processed. This
7312 is the substring up to (and not including) the last period and not
7313 including the directory.
7316 This is the same as `%b', but include the file suffix (text after
7320 Marks the argument containing or following the `%d' as a temporary
7321 file name, so that that file will be deleted if GCC exits
7322 successfully. Unlike `%g', this contributes no text to the
7326 Substitute a file name that has suffix SUFFIX and is chosen once
7327 per compilation, and mark the argument in the same way as `%d'.
7328 To reduce exposure to denial-of-service attacks, the file name is
7329 now chosen in a way that is hard to predict even when previously
7330 chosen file names are known. For example, `%g.s ... %g.o ... %g.s'
7331 might turn into `ccUVUUAU.s ccXYAXZ12.o ccUVUUAU.s'. SUFFIX
7332 matches the regexp `[.A-Za-z]*' or the special string `%O', which
7333 is treated exactly as if `%O' had been preprocessed. Previously,
7334 `%g' was simply substituted with a file name chosen once per
7335 compilation, without regard to any appended suffix (which was
7336 therefore treated just like ordinary text), making such attacks
7337 more likely to succeed.
7340 Like `%g', but generates a new temporary file name even if
7341 `%uSUFFIX' was already seen.
7344 Substitutes the last file name generated with `%uSUFFIX',
7345 generating a new one if there is no such last file name. In the
7346 absence of any `%uSUFFIX', this is just like `%gSUFFIX', except
7347 they don't share the same suffix _space_, so `%g.s ... %U.s ...
7348 %g.s ... %U.s' would involve the generation of two distinct file
7349 names, one for each `%g.s' and another for each `%U.s'.
7350 Previously, `%U' was simply substituted with a file name chosen
7351 for the previous `%u', without regard to any appended suffix.
7354 Substitutes the name of the `HOST_BIT_BUCKET', if any, and if it is
7355 writable, and if save-temps is off; otherwise, substitute the name
7356 of a temporary file, just like `%u'. This temporary file is not
7357 meant for communication between processes, but rather as a junk
7362 Like `%g', except if `-pipe' is in effect. In that case `%|'
7363 substitutes a single dash and `%m' substitutes nothing at all.
7364 These are the two most common ways to instruct a program that it
7365 should read from standard input or write to standard output. If
7366 you need something more elaborate you can use an `%{pipe:`X'}'
7367 construct: see for example `f/lang-specs.h'.
7370 Substitutes .SUFFIX for the suffixes of a matched switch's args
7371 when it is subsequently output with `%*'. SUFFIX is terminated by
7372 the next space or %.
7375 Marks the argument containing or following the `%w' as the
7376 designated output file of this compilation. This puts the argument
7377 into the sequence of arguments that `%o' will substitute later.
7380 Substitutes the names of all the output files, with spaces
7381 automatically placed around them. You should write spaces around
7382 the `%o' as well or the results are undefined. `%o' is for use in
7383 the specs for running the linker. Input files whose names have no
7384 recognized suffix are not compiled at all, but they are included
7385 among the output files, so they will be linked.
7388 Substitutes the suffix for object files. Note that this is
7389 handled specially when it immediately follows `%g, %u, or %U',
7390 because of the need for those to form complete file names. The
7391 handling is such that `%O' is treated exactly as if it had already
7392 been substituted, except that `%g, %u, and %U' do not currently
7393 support additional SUFFIX characters following `%O' as they would
7394 following, for example, `.o'.
7397 Substitutes the standard macro predefinitions for the current
7398 target machine. Use this when running `cpp'.
7401 Like `%p', but puts `__' before and after the name of each
7402 predefined macro, except for macros that start with `__' or with
7403 `_L', where L is an uppercase letter. This is for ISO C.
7406 Substitute any of `-iprefix' (made from `GCC_EXEC_PREFIX'),
7407 `-isysroot' (made from `TARGET_SYSTEM_ROOT'), `-isystem' (made
7408 from `COMPILER_PATH' and `-B' options) and `-imultilib' as
7412 Current argument is the name of a library or startup file of some
7413 sort. Search for that file in a standard list of directories and
7414 substitute the full name found.
7417 Print STR as an error message. STR is terminated by a newline.
7418 Use this when inconsistent options are detected.
7421 Substitute the contents of spec string NAME at this point.
7424 Like `%(...)' but put `__' around `-D' arguments.
7427 Accumulate an option for `%X'.
7430 Output the accumulated linker options specified by `-Wl' or a `%x'
7434 Output the accumulated assembler options specified by `-Wa'.
7437 Output the accumulated preprocessor options specified by `-Wp'.
7440 Process the `asm' spec. This is used to compute the switches to
7441 be passed to the assembler.
7444 Process the `asm_final' spec. This is a spec string for passing
7445 switches to an assembler post-processor, if such a program is
7449 Process the `link' spec. This is the spec for computing the
7450 command line passed to the linker. Typically it will make use of
7451 the `%L %G %S %D and %E' sequences.
7454 Dump out a `-L' option for each directory that GCC believes might
7455 contain startup files. If the target supports multilibs then the
7456 current multilib directory will be prepended to each of these
7460 Process the `lib' spec. This is a spec string for deciding which
7461 libraries should be included on the command line to the linker.
7464 Process the `libgcc' spec. This is a spec string for deciding
7465 which GCC support library should be included on the command line
7469 Process the `startfile' spec. This is a spec for deciding which
7470 object files should be the first ones passed to the linker.
7471 Typically this might be a file named `crt0.o'.
7474 Process the `endfile' spec. This is a spec string that specifies
7475 the last object files that will be passed to the linker.
7478 Process the `cpp' spec. This is used to construct the arguments
7479 to be passed to the C preprocessor.
7482 Process the `cc1' spec. This is used to construct the options to
7483 be passed to the actual C compiler (`cc1').
7486 Process the `cc1plus' spec. This is used to construct the options
7487 to be passed to the actual C++ compiler (`cc1plus').
7490 Substitute the variable part of a matched option. See below.
7491 Note that each comma in the substituted string is replaced by a
7495 Remove all occurrences of `-S' from the command line. Note--this
7496 command is position dependent. `%' commands in the spec string
7497 before this one will see `-S', `%' commands in the spec string
7498 after this one will not.
7501 Call the named function FUNCTION, passing it ARGS. ARGS is first
7502 processed as a nested spec string, then split into an argument
7503 vector in the usual fashion. The function returns a string which
7504 is processed as if it had appeared literally as part of the
7507 The following built-in spec functions are provided:
7510 The `if-exists' spec function takes one argument, an absolute
7511 pathname to a file. If the file exists, `if-exists' returns
7512 the pathname. Here is a small example of its usage:
7515 crt0%O%s %:if-exists(crti%O%s) crtbegin%O%s
7518 The `if-exists-else' spec function is similar to the
7519 `if-exists' spec function, except that it takes two
7520 arguments. The first argument is an absolute pathname to a
7521 file. If the file exists, `if-exists-else' returns the
7522 pathname. If it does not exist, it returns the second
7523 argument. This way, `if-exists-else' can be used to select
7524 one file or another, based on the existence of the first.
7525 Here is a small example of its usage:
7528 crt0%O%s %:if-exists(crti%O%s) \
7529 %:if-exists-else(crtbeginT%O%s crtbegin%O%s)
7532 The `replace-outfile' spec function takes two arguments. It
7533 looks for the first argument in the outfiles array and
7534 replaces it with the second argument. Here is a small
7535 example of its usage:
7537 %{fgnu-runtime:%:replace-outfile(-lobjc -lobjc-gnu)}
7541 Substitutes the `-S' switch, if that switch was given to GCC. If
7542 that switch was not specified, this substitutes nothing. Note that
7543 the leading dash is omitted when specifying this option, and it is
7544 automatically inserted if the substitution is performed. Thus the
7545 spec string `%{foo}' would match the command-line option `-foo'
7546 and would output the command line option `-foo'.
7549 Like %{`S'} but mark last argument supplied within as a file to be
7553 Substitutes all the switches specified to GCC whose names start
7554 with `-S', but which also take an argument. This is used for
7555 switches like `-o', `-D', `-I', etc. GCC considers `-o foo' as
7556 being one switch whose names starts with `o'. %{o*} would
7557 substitute this text, including the space. Thus two arguments
7561 Like %{`S'*}, but preserve order of `S' and `T' options (the order
7562 of `S' and `T' in the spec is not significant). There can be any
7563 number of ampersand-separated variables; for each the wild card is
7564 optional. Useful for CPP as `%{D*&U*&A*}'.
7567 Substitutes `X', if the `-S' switch was given to GCC.
7570 Substitutes `X', if the `-S' switch was _not_ given to GCC.
7573 Substitutes `X' if one or more switches whose names start with
7574 `-S' are specified to GCC. Normally `X' is substituted only once,
7575 no matter how many such switches appeared. However, if `%*'
7576 appears somewhere in `X', then `X' will be substituted once for
7577 each matching switch, with the `%*' replaced by the part of that
7578 switch that matched the `*'.
7581 Substitutes `X', if processing a file with suffix `S'.
7584 Substitutes `X', if _not_ processing a file with suffix `S'.
7587 Substitutes `X' if either `-S' or `-P' was given to GCC. This may
7588 be combined with `!', `.', and `*' sequences as well, although
7589 they have a stronger binding than the `|'. If `%*' appears in
7590 `X', all of the alternatives must be starred, and only the first
7591 matching alternative is substituted.
7593 For example, a spec string like this:
7595 %{.c:-foo} %{!.c:-bar} %{.c|d:-baz} %{!.c|d:-boggle}
7597 will output the following command-line options from the following
7598 input command-line options:
7602 -d fred.c -foo -baz -boggle
7603 -d jim.d -bar -baz -boggle
7606 If `S' was given to GCC, substitutes `X'; else if `T' was given to
7607 GCC, substitutes `Y'; else substitutes `D'. There can be as many
7608 clauses as you need. This may be combined with `.', `!', `|', and
7612 The conditional text `X' in a %{`S':`X'} or similar construct may
7613 contain other nested `%' constructs or spaces, or even newlines. They
7614 are processed as usual, as described above. Trailing white space in
7615 `X' is ignored. White space may also appear anywhere on the left side
7616 of the colon in these constructs, except between `.' or `*' and the
7619 The `-O', `-f', `-m', and `-W' switches are handled specifically in
7620 these constructs. If another value of `-O' or the negated form of a
7621 `-f', `-m', or `-W' switch is found later in the command line, the
7622 earlier switch value is ignored, except with {`S'*} where `S' is just
7623 one letter, which passes all matching options.
7625 The character `|' at the beginning of the predicate text is used to
7626 indicate that a command should be piped to the following command, but
7627 only if `-pipe' is specified.
7629 It is built into GCC which switches take arguments and which do not.
7630 (You might think it would be useful to generalize this to allow each
7631 compiler's spec to say which switches take arguments. But this cannot
7632 be done in a consistent fashion. GCC cannot even decide which input
7633 files have been specified without knowing which switches take arguments,
7634 and it must know which input files to compile in order to tell which
7637 GCC also knows implicitly that arguments starting in `-l' are to be
7638 treated as compiler output files, and passed to the linker in their
7639 proper position among the other output files.
7642 File: gcc.info, Node: Target Options, Next: Submodel Options, Prev: Spec Files, Up: Invoking GCC
7644 3.16 Specifying Target Machine and Compiler Version
7645 ===================================================
7647 The usual way to run GCC is to run the executable called `gcc', or
7648 `<machine>-gcc' when cross-compiling, or `<machine>-gcc-<version>' to
7649 run a version other than the one that was installed last. Sometimes
7650 this is inconvenient, so GCC provides options that will switch to
7651 another cross-compiler or version.
7654 The argument MACHINE specifies the target machine for compilation.
7656 The value to use for MACHINE is the same as was specified as the
7657 machine type when configuring GCC as a cross-compiler. For
7658 example, if a cross-compiler was configured with `configure
7659 arm-elf', meaning to compile for an arm processor with elf
7660 binaries, then you would specify `-b arm-elf' to run that cross
7661 compiler. Because there are other options beginning with `-b', the
7662 configuration must contain a hyphen.
7665 The argument VERSION specifies which version of GCC to run. This
7666 is useful when multiple versions are installed. For example,
7667 VERSION might be `4.0', meaning to run GCC version 4.0.
7669 The `-V' and `-b' options work by running the
7670 `<machine>-gcc-<version>' executable, so there's no real reason to use
7671 them if you can just run that directly.
7674 File: gcc.info, Node: Submodel Options, Next: Code Gen Options, Prev: Target Options, Up: Invoking GCC
7676 3.17 Hardware Models and Configurations
7677 =======================================
7679 Earlier we discussed the standard option `-b' which chooses among
7680 different installed compilers for completely different target machines,
7681 such as VAX vs. 68000 vs. 80386.
7683 In addition, each of these target machine types can have its own
7684 special options, starting with `-m', to choose among various hardware
7685 models or configurations--for example, 68010 vs 68020, floating
7686 coprocessor or none. A single installed version of the compiler can
7687 compile for any model or configuration, according to the options
7690 Some configurations of the compiler also support additional special
7691 options, usually for compatibility with other compilers on the same
7699 * Blackfin Options::
7703 * DEC Alpha Options::
7704 * DEC Alpha/VMS Options::
7706 * GNU/Linux Options::
7709 * i386 and x86-64 Options::
7722 * RS/6000 and PowerPC Options::
7723 * S/390 and zSeries Options::
7727 * System V Options::
7728 * TMS320C3x/C4x Options::
7732 * Xstormy16 Options::
7737 File: gcc.info, Node: ARC Options, Next: ARM Options, Up: Submodel Options
7742 These options are defined for ARC implementations:
7745 Compile code for little endian mode. This is the default.
7748 Compile code for big endian mode.
7751 Prepend the name of the cpu to all public symbol names. In
7752 multiple-processor systems, there are many ARC variants with
7753 different instruction and register set characteristics. This flag
7754 prevents code compiled for one cpu to be linked with code compiled
7755 for another. No facility exists for handling variants that are
7756 "almost identical". This is an all or nothing option.
7759 Compile code for ARC variant CPU. Which variants are supported
7760 depend on the configuration. All variants support `-mcpu=base',
7761 this is the default.
7763 `-mtext=TEXT-SECTION'
7764 `-mdata=DATA-SECTION'
7765 `-mrodata=READONLY-DATA-SECTION'
7766 Put functions, data, and readonly data in TEXT-SECTION,
7767 DATA-SECTION, and READONLY-DATA-SECTION respectively by default.
7768 This can be overridden with the `section' attribute. *Note
7769 Variable Attributes::.
7773 File: gcc.info, Node: ARM Options, Next: AVR Options, Prev: ARC Options, Up: Submodel Options
7778 These `-m' options are defined for Advanced RISC Machines (ARM)
7782 Generate code for the specified ABI. Permissible values are:
7783 `apcs-gnu', `atpcs', `aapcs', `aapcs-linux' and `iwmmxt'.
7786 Generate a stack frame that is compliant with the ARM Procedure
7787 Call Standard for all functions, even if this is not strictly
7788 necessary for correct execution of the code. Specifying
7789 `-fomit-frame-pointer' with this option will cause the stack
7790 frames not to be generated for leaf functions. The default is
7794 This is a synonym for `-mapcs-frame'.
7797 Generate code which supports calling between the ARM and Thumb
7798 instruction sets. Without this option the two instruction sets
7799 cannot be reliably used inside one program. The default is
7800 `-mno-thumb-interwork', since slightly larger code is generated
7801 when `-mthumb-interwork' is specified.
7804 Prevent the reordering of instructions in the function prolog, or
7805 the merging of those instruction with the instructions in the
7806 function's body. This means that all functions will start with a
7807 recognizable set of instructions (or in fact one of a choice from
7808 a small set of different function prologues), and this information
7809 can be used to locate the start if functions inside an executable
7810 piece of code. The default is `-msched-prolog'.
7813 Generate output containing floating point instructions. This is
7817 Generate output containing library calls for floating point.
7818 *Warning:* the requisite libraries are not available for all ARM
7819 targets. Normally the facilities of the machine's usual C
7820 compiler are used, but this cannot be done directly in
7821 cross-compilation. You must make your own arrangements to provide
7822 suitable library functions for cross-compilation.
7824 `-msoft-float' changes the calling convention in the output file;
7825 therefore, it is only useful if you compile _all_ of a program with
7826 this option. In particular, you need to compile `libgcc.a', the
7827 library that comes with GCC, with `-msoft-float' in order for this
7831 Specifies which ABI to use for floating point values. Permissible
7832 values are: `soft', `softfp' and `hard'.
7834 `soft' and `hard' are equivalent to `-msoft-float' and
7835 `-mhard-float' respectively. `softfp' allows the generation of
7836 floating point instructions, but still uses the soft-float calling
7840 Generate code for a processor running in little-endian mode. This
7841 is the default for all standard configurations.
7844 Generate code for a processor running in big-endian mode; the
7845 default is to compile code for a little-endian processor.
7847 `-mwords-little-endian'
7848 This option only applies when generating code for big-endian
7849 processors. Generate code for a little-endian word order but a
7850 big-endian byte order. That is, a byte order of the form
7851 `32107654'. Note: this option should only be used if you require
7852 compatibility with code for big-endian ARM processors generated by
7853 versions of the compiler prior to 2.8.
7856 This specifies the name of the target ARM processor. GCC uses
7857 this name to determine what kind of instructions it can emit when
7858 generating assembly code. Permissible names are: `arm2', `arm250',
7859 `arm3', `arm6', `arm60', `arm600', `arm610', `arm620', `arm7',
7860 `arm7m', `arm7d', `arm7dm', `arm7di', `arm7dmi', `arm70', `arm700',
7861 `arm700i', `arm710', `arm710c', `arm7100', `arm7500', `arm7500fe',
7862 `arm7tdmi', `arm7tdmi-s', `arm8', `strongarm', `strongarm110',
7863 `strongarm1100', `arm8', `arm810', `arm9', `arm9e', `arm920',
7864 `arm920t', `arm922t', `arm946e-s', `arm966e-s', `arm968e-s',
7865 `arm926ej-s', `arm940t', `arm9tdmi', `arm10tdmi', `arm1020t',
7866 `arm1026ej-s', `arm10e', `arm1020e', `arm1022e', `arm1136j-s',
7867 `arm1136jf-s', `mpcore', `mpcorenovfp', `arm1176jz-s',
7868 `arm1176jzf-s', `xscale', `iwmmxt', `ep9312'.
7871 This option is very similar to the `-mcpu=' option, except that
7872 instead of specifying the actual target processor type, and hence
7873 restricting which instructions can be used, it specifies that GCC
7874 should tune the performance of the code as if the target were of
7875 the type specified in this option, but still choosing the
7876 instructions that it will generate based on the cpu specified by a
7877 `-mcpu=' option. For some ARM implementations better performance
7878 can be obtained by using this option.
7881 This specifies the name of the target ARM architecture. GCC uses
7882 this name to determine what kind of instructions it can emit when
7883 generating assembly code. This option can be used in conjunction
7884 with or instead of the `-mcpu=' option. Permissible names are:
7885 `armv2', `armv2a', `armv3', `armv3m', `armv4', `armv4t', `armv5',
7886 `armv5t', `armv5te', `armv6', `armv6j', `iwmmxt', `ep9312'.
7891 This specifies what floating point hardware (or hardware
7892 emulation) is available on the target. Permissible names are:
7893 `fpa', `fpe2', `fpe3', `maverick', `vfp'. `-mfp' and `-mfpe' are
7894 synonyms for `-mfpu'=`fpe'NUMBER, for compatibility with older
7897 If `-msoft-float' is specified this specifies the format of
7898 floating point values.
7900 `-mstructure-size-boundary=N'
7901 The size of all structures and unions will be rounded up to a
7902 multiple of the number of bits set by this option. Permissible
7903 values are 8, 32 and 64. The default value varies for different
7904 toolchains. For the COFF targeted toolchain the default value is
7905 8. A value of 64 is only allowed if the underlying ABI supports
7908 Specifying the larger number can produce faster, more efficient
7909 code, but can also increase the size of the program. Different
7910 values are potentially incompatible. Code compiled with one value
7911 cannot necessarily expect to work with code or libraries compiled
7912 with another value, if they exchange information using structures
7915 `-mabort-on-noreturn'
7916 Generate a call to the function `abort' at the end of a `noreturn'
7917 function. It will be executed if the function tries to return.
7921 Tells the compiler to perform function calls by first loading the
7922 address of the function into a register and then performing a
7923 subroutine call on this register. This switch is needed if the
7924 target function will lie outside of the 64 megabyte addressing
7925 range of the offset based version of subroutine call instruction.
7927 Even if this switch is enabled, not all function calls will be
7928 turned into long calls. The heuristic is that static functions,
7929 functions which have the `short-call' attribute, functions that
7930 are inside the scope of a `#pragma no_long_calls' directive and
7931 functions whose definitions have already been compiled within the
7932 current compilation unit, will not be turned into long calls. The
7933 exception to this rule is that weak function definitions,
7934 functions with the `long-call' attribute or the `section'
7935 attribute, and functions that are within the scope of a `#pragma
7936 long_calls' directive, will always be turned into long calls.
7938 This feature is not enabled by default. Specifying
7939 `-mno-long-calls' will restore the default behavior, as will
7940 placing the function calls within the scope of a `#pragma
7941 long_calls_off' directive. Note these switches have no effect on
7942 how the compiler generates code to handle function calls via
7945 `-mnop-fun-dllimport'
7946 Disable support for the `dllimport' attribute.
7949 Treat the register used for PIC addressing as read-only, rather
7950 than loading it in the prologue for each function. The run-time
7951 system is responsible for initializing this register with an
7952 appropriate value before execution begins.
7954 `-mpic-register=REG'
7955 Specify the register to be used for PIC addressing. The default
7956 is R10 unless stack-checking is enabled, when R9 is used.
7958 `-mcirrus-fix-invalid-insns'
7959 Insert NOPs into the instruction stream to in order to work around
7960 problems with invalid Maverick instruction combinations. This
7961 option is only valid if the `-mcpu=ep9312' option has been used to
7962 enable generation of instructions for the Cirrus Maverick floating
7963 point co-processor. This option is not enabled by default, since
7964 the problem is only present in older Maverick implementations.
7965 The default can be re-enabled by use of the
7966 `-mno-cirrus-fix-invalid-insns' switch.
7968 `-mpoke-function-name'
7969 Write the name of each function into the text section, directly
7970 preceding the function prologue. The generated code is similar to
7974 .ascii "arm_poke_function_name", 0
7977 .word 0xff000000 + (t1 - t0)
7978 arm_poke_function_name
7980 stmfd sp!, {fp, ip, lr, pc}
7983 When performing a stack backtrace, code can inspect the value of
7984 `pc' stored at `fp + 0'. If the trace function then looks at
7985 location `pc - 12' and the top 8 bits are set, then we know that
7986 there is a function name embedded immediately preceding this
7987 location and has length `((pc[-3]) & 0xff000000)'.
7990 Generate code for the 16-bit Thumb instruction set. The default
7991 is to use the 32-bit ARM instruction set.
7994 Generate a stack frame that is compliant with the Thumb Procedure
7995 Call Standard for all non-leaf functions. (A leaf function is one
7996 that does not call any other functions.) The default is
8000 Generate a stack frame that is compliant with the Thumb Procedure
8001 Call Standard for all leaf functions. (A leaf function is one
8002 that does not call any other functions.) The default is
8003 `-mno-apcs-leaf-frame'.
8005 `-mcallee-super-interworking'
8006 Gives all externally visible functions in the file being compiled
8007 an ARM instruction set header which switches to Thumb mode before
8008 executing the rest of the function. This allows these functions
8009 to be called from non-interworking code.
8011 `-mcaller-super-interworking'
8012 Allows calls via function pointers (including virtual functions) to
8013 execute correctly regardless of whether the target code has been
8014 compiled for interworking or not. There is a small overhead in
8015 the cost of executing a function pointer if this option is enabled.
8018 Specify the access model for the thread local storage pointer.
8019 The valid models are `soft', which generates calls to
8020 `__aeabi_read_tp', `cp15', which fetches the thread pointer from
8021 `cp15' directly (supported in the arm6k architecture), and `auto',
8022 which uses the best available method for the selected processor.
8023 The default setting is `auto'.
8026 Enable Android specific compilier options.
8028 If this option is used, a preprocessor macro `__ANDROID__' is
8029 defined and has the value 1 during compilation. The option also
8030 implies C/C++ options `-fno-exceptions' `-fpic' `-mthumb-interwork'
8031 `-fno-short-enums' and C++ option `-fno-rtti'. These implied
8032 options can be overridden. For example RTTI in C++ code can still
8033 be enabled with -frtti even when -mandroid is also used.
8035 Linking options depend on whether a static executable, a dynamic
8036 executable or a shared library is built. When `-static' is given,
8037 `-mandroid' implies linking flag `-Bstatic', start file
8038 `crtbegin_static.o' and end file `crtend_android.o'.
8040 When `-shared' is given, `-mandroid' implies the linking flag
8041 `-Bsymbolic' and no start and end files.
8043 When none of `-static' and `-shared' is given, `-mandroid' implies
8044 linking flags `-Bdynamic -dynamic-linker /system/bin/linker',
8045 start file `crtbegin_dynamic.o' and end file `crtend_android.o'.
8046 The dynamic linker used can be overriden by another
8047 `-dynamic-linker' in command line.
8049 The linking option `-ldl' is also added if `-static' is not given.
8051 If more than one of `-dynamic', `-static' and `-shared' are given,
8052 behaviour of `-mandroid' is undefined.
8056 File: gcc.info, Node: AVR Options, Next: Blackfin Options, Prev: ARM Options, Up: Submodel Options
8061 These options are defined for AVR implementations:
8064 Specify ATMEL AVR instruction set or MCU type.
8066 Instruction set avr1 is for the minimal AVR core, not supported by
8067 the C compiler, only for assembler programs (MCU types: at90s1200,
8068 attiny10, attiny11, attiny12, attiny15, attiny28).
8070 Instruction set avr2 (default) is for the classic AVR core with up
8071 to 8K program memory space (MCU types: at90s2313, at90s2323,
8072 attiny22, at90s2333, at90s2343, at90s4414, at90s4433, at90s4434,
8073 at90s8515, at90c8534, at90s8535).
8075 Instruction set avr3 is for the classic AVR core with up to 128K
8076 program memory space (MCU types: atmega103, atmega603, at43usb320,
8079 Instruction set avr4 is for the enhanced AVR core with up to 8K
8080 program memory space (MCU types: atmega8, atmega83, atmega85).
8082 Instruction set avr5 is for the enhanced AVR core with up to 128K
8083 program memory space (MCU types: atmega16, atmega161, atmega163,
8084 atmega32, atmega323, atmega64, atmega128, at43usb355, at94k).
8087 Output instruction sizes to the asm file.
8090 Specify the initial stack address, which may be a symbol or
8091 numeric value, `__stack' is the default.
8094 Generated code is not compatible with hardware interrupts. Code
8095 size will be smaller.
8098 Functions prologues/epilogues expanded as call to appropriate
8099 subroutines. Code size will be smaller.
8102 Do not generate tablejump insns which sometimes increase code size.
8105 Change only the low 8 bits of the stack pointer.
8108 Assume int to be 8 bit integer. This affects the sizes of all
8109 types: A char will be 1 byte, an int will be 1 byte, an long will
8110 be 2 bytes and long long will be 4 bytes. Please note that this
8111 option does not comply to the C standards, but it will provide you
8112 with smaller code size.
8115 File: gcc.info, Node: Blackfin Options, Next: CRIS Options, Prev: AVR Options, Up: Submodel Options
8117 3.17.4 Blackfin Options
8118 -----------------------
8120 `-momit-leaf-frame-pointer'
8121 Don't keep the frame pointer in a register for leaf functions.
8122 This avoids the instructions to save, set up and restore frame
8123 pointers and makes an extra register available in leaf functions.
8124 The option `-fomit-frame-pointer' removes the frame pointer for
8125 all functions which might make debugging harder.
8128 When enabled, the compiler will ensure that the generated code
8129 does not contain speculative loads after jump instructions. This
8130 option is enabled by default.
8132 `-mno-specld-anomaly'
8133 Don't generate extra code to prevent speculative loads from
8137 When enabled, the compiler will ensure that the generated code
8138 does not contain CSYNC or SSYNC instructions too soon after
8139 conditional branches. This option is enabled by default.
8141 `-mno-csync-anomaly'
8142 Don't generate extra code to prevent CSYNC or SSYNC instructions
8143 from occurring too soon after a conditional branch.
8146 When enabled, the compiler is free to take advantage of the
8147 knowledge that the entire program fits into the low 64k of memory.
8150 Assume that the program is arbitrarily large. This is the default.
8152 `-mid-shared-library'
8153 Generate code that supports shared libraries via the library ID
8154 method. This allows for execute in place and shared libraries in
8155 an environment without virtual memory management. This option
8158 `-mno-id-shared-library'
8159 Generate code that doesn't assume ID based shared libraries are
8160 being used. This is the default.
8162 `-mshared-library-id=n'
8163 Specified the identification number of the ID based shared library
8164 being compiled. Specifying a value of 0 will generate more
8165 compact code, specifying other values will force the allocation of
8166 that number to the current library but is no more space or time
8167 efficient than omitting this option.
8171 Tells the compiler to perform function calls by first loading the
8172 address of the function into a register and then performing a
8173 subroutine call on this register. This switch is needed if the
8174 target function will lie outside of the 24 bit addressing range of
8175 the offset based version of subroutine call instruction.
8177 This feature is not enabled by default. Specifying
8178 `-mno-long-calls' will restore the default behavior. Note these
8179 switches have no effect on how the compiler generates code to
8180 handle function calls via function pointers.
8183 File: gcc.info, Node: CRIS Options, Next: CRX Options, Prev: Blackfin Options, Up: Submodel Options
8188 These options are defined specifically for the CRIS ports.
8190 `-march=ARCHITECTURE-TYPE'
8191 `-mcpu=ARCHITECTURE-TYPE'
8192 Generate code for the specified architecture. The choices for
8193 ARCHITECTURE-TYPE are `v3', `v8' and `v10' for respectively
8194 ETRAX 4, ETRAX 100, and ETRAX 100 LX. Default is `v0' except for
8195 cris-axis-linux-gnu, where the default is `v10'.
8197 `-mtune=ARCHITECTURE-TYPE'
8198 Tune to ARCHITECTURE-TYPE everything applicable about the generated
8199 code, except for the ABI and the set of available instructions.
8200 The choices for ARCHITECTURE-TYPE are the same as for
8201 `-march=ARCHITECTURE-TYPE'.
8203 `-mmax-stack-frame=N'
8204 Warn when the stack frame of a function exceeds N bytes.
8206 `-melinux-stacksize=N'
8207 Only available with the `cris-axis-aout' target. Arranges for
8208 indications in the program to the kernel loader that the stack of
8209 the program should be set to N bytes.
8213 The options `-metrax4' and `-metrax100' are synonyms for
8214 `-march=v3' and `-march=v8' respectively.
8216 `-mmul-bug-workaround'
8217 `-mno-mul-bug-workaround'
8218 Work around a bug in the `muls' and `mulu' instructions for CPU
8219 models where it applies. This option is active by default.
8222 Enable CRIS-specific verbose debug-related information in the
8223 assembly code. This option also has the effect to turn off the
8224 `#NO_APP' formatted-code indicator to the assembler at the
8225 beginning of the assembly file.
8228 Do not use condition-code results from previous instruction;
8229 always emit compare and test instructions before use of condition
8233 Do not emit instructions with side-effects in addressing modes
8234 other than post-increment.
8242 These options (no-options) arranges (eliminate arrangements) for
8243 the stack-frame, individual data and constants to be aligned for
8244 the maximum single data access size for the chosen CPU model. The
8245 default is to arrange for 32-bit alignment. ABI details such as
8246 structure layout are not affected by these options.
8251 Similar to the stack- data- and const-align options above, these
8252 options arrange for stack-frame, writable data and constants to
8253 all be 32-bit, 16-bit or 8-bit aligned. The default is 32-bit
8256 `-mno-prologue-epilogue'
8257 `-mprologue-epilogue'
8258 With `-mno-prologue-epilogue', the normal function prologue and
8259 epilogue that sets up the stack-frame are omitted and no return
8260 instructions or return sequences are generated in the code. Use
8261 this option only together with visual inspection of the compiled
8262 code: no warnings or errors are generated when call-saved
8263 registers must be saved, or storage for local variable needs to be
8268 With `-fpic' and `-fPIC', don't generate (do generate) instruction
8269 sequences that load addresses for functions from the PLT part of
8270 the GOT rather than (traditional on other architectures) calls to
8271 the PLT. The default is `-mgotplt'.
8274 Legacy no-op option only recognized with the cris-axis-aout target.
8277 Legacy no-op option only recognized with the cris-axis-elf and
8278 cris-axis-linux-gnu targets.
8281 Only recognized with the cris-axis-aout target, where it selects a
8282 GNU/linux-like multilib, include files and instruction set for
8286 Legacy no-op option only recognized with the cris-axis-linux-gnu
8290 This option, recognized for the cris-axis-aout and cris-axis-elf
8291 arranges to link with input-output functions from a simulator
8292 library. Code, initialized data and zero-initialized data are
8293 allocated consecutively.
8296 Like `-sim', but pass linker options to locate initialized data at
8297 0x40000000 and zero-initialized data at 0x80000000.
8300 File: gcc.info, Node: CRX Options, Next: Darwin Options, Prev: CRIS Options, Up: Submodel Options
8305 These options are defined specifically for the CRX ports.
8308 Enable the use of multiply-accumulate instructions. Disabled by
8312 Push instructions will be used to pass outgoing arguments when
8313 functions are called. Enabled by default.
8316 File: gcc.info, Node: Darwin Options, Next: DEC Alpha Options, Prev: CRX Options, Up: Submodel Options
8318 3.17.7 Darwin Options
8319 ---------------------
8321 These options are defined for all architectures running the Darwin
8324 FSF GCC on Darwin does not create "fat" object files; it will create
8325 an object file for the single architecture that it was built to target.
8326 Apple's GCC on Darwin does create "fat" files if multiple `-arch'
8327 options are used; it does so by running the compiler or linker multiple
8328 times and joining the results together with `lipo'.
8330 The subtype of the file created (like `ppc7400' or `ppc970' or `i686')
8331 is determined by the flags that specify the ISA that GCC is targetting,
8332 like `-mcpu' or `-march'. The `-force_cpusubtype_ALL' option can be
8333 used to override this.
8335 The Darwin tools vary in their behavior when presented with an ISA
8336 mismatch. The assembler, `as', will only permit instructions to be
8337 used that are valid for the subtype of the file it is generating, so
8338 you cannot put 64-bit instructions in an `ppc750' object file. The
8339 linker for shared libraries, `/usr/bin/libtool', will fail and print an
8340 error if asked to create a shared library with a less restrictive
8341 subtype than its input files (for instance, trying to put a `ppc970'
8342 object file in a `ppc7400' library). The linker for executables, `ld',
8343 will quietly give the executable the most restrictive subtype of any of
8347 Add the framework directory DIR to the head of the list of
8348 directories to be searched for header files. These directories are
8349 interleaved with those specified by `-I' options and are scanned
8350 in a left-to-right order.
8352 A framework directory is a directory with frameworks in it. A
8353 framework is a directory with a `"Headers"' and/or
8354 `"PrivateHeaders"' directory contained directly in it that ends in
8355 `".framework"'. The name of a framework is the name of this
8356 directory excluding the `".framework"'. Headers associated with
8357 the framework are found in one of those two directories, with
8358 `"Headers"' being searched first. A subframework is a framework
8359 directory that is in a framework's `"Frameworks"' directory.
8360 Includes of subframework headers can only appear in a header of a
8361 framework that contains the subframework, or in a sibling
8362 subframework header. Two subframeworks are siblings if they occur
8363 in the same framework. A subframework should not have the same
8364 name as a framework, a warning will be issued if this is violated.
8365 Currently a subframework cannot have subframeworks, in the
8366 future, the mechanism may be extended to support this. The
8367 standard frameworks can be found in `"/System/Library/Frameworks"'
8368 and `"/Library/Frameworks"'. An example include looks like
8369 `#include <Framework/header.h>', where `Framework' denotes the
8370 name of the framework and header.h is found in the
8371 `"PrivateHeaders"' or `"Headers"' directory.
8374 Emit debugging information for symbols that are used. For STABS
8375 debugging format, this enables `-feliminate-unused-debug-symbols'.
8376 This is by default ON.
8379 Emit debugging information for all symbols and types.
8381 `-mmacosx-version-min=VERSION'
8382 The earliest version of MacOS X that this executable will run on
8383 is VERSION. Typical values of VERSION include `10.1', `10.2', and
8386 The default for this option is to make choices that seem to be most
8390 Enable kernel development mode. The `-mkernel' option sets
8391 `-static', `-fno-common', `-fno-cxa-atexit', `-fno-exceptions',
8392 `-fno-non-call-exceptions', `-fapple-kext', `-fno-weak' and
8393 `-fno-rtti' where applicable. This mode also sets `-mno-altivec',
8394 `-msoft-float', `-fno-builtin' and `-mlong-branch' for PowerPC
8398 Override the defaults for `bool' so that `sizeof(bool)==1'. By
8399 default `sizeof(bool)' is `4' when compiling for Darwin/PowerPC
8400 and `1' when compiling for Darwin/x86, so this option has no
8403 *Warning:* The `-mone-byte-bool' switch causes GCC to generate
8404 code that is not binary compatible with code generated without
8405 that switch. Using this switch may require recompiling all other
8406 modules in a program, including system libraries. Use this switch
8407 to conform to a non-default data model.
8409 `-mfix-and-continue'
8410 `-ffix-and-continue'
8412 Generate code suitable for fast turn around development. Needed to
8413 enable gdb to dynamically load `.o' files into already running
8414 programs. `-findirect-data' and `-ffix-and-continue' are provided
8415 for backwards compatibility.
8418 Loads all members of static archive libraries. See man ld(1) for
8421 `-arch_errors_fatal'
8422 Cause the errors having to do with files that have the wrong
8423 architecture to be fatal.
8426 Causes the output file to be marked such that the dynamic linker
8427 will bind all undefined references when the file is loaded or
8431 Produce a Mach-o bundle format file. See man ld(1) for more
8434 `-bundle_loader EXECUTABLE'
8435 This option specifies the EXECUTABLE that will be loading the build
8436 output file being linked. See man ld(1) for more information.
8439 When passed this option, GCC will produce a dynamic library
8440 instead of an executable when linking, using the Darwin `libtool'
8443 `-force_cpusubtype_ALL'
8444 This causes GCC's output file to have the ALL subtype, instead of
8445 one controlled by the `-mcpu' or `-march' option.
8447 `-allowable_client CLIENT_NAME'
8449 `-compatibility_version'
8454 `-dylinker_install_name'
8456 `-exported_symbols_list'
8459 `-force_flat_namespace'
8460 `-headerpad_max_install_names'
8464 `-keep_private_externs'
8467 `-multiply_defined_unused'
8469 `-no_dead_strip_inits_and_terms'
8476 `-prebind_all_twolevel_modules'
8480 `-sectobjectsymbols'
8484 `-sectobjectsymbols'
8487 `-segs_read_only_addr'
8488 `-segs_read_write_addr'
8490 `-seg_addr_table_filename'
8493 `-segs_read_only_addr'
8494 `-segs_read_write_addr'
8499 `-twolevel_namespace'
8502 `-unexported_symbols_list'
8503 `-weak_reference_mismatches'
8505 These options are passed to the Darwin linker. The Darwin linker
8506 man page describes them in detail.
8509 File: gcc.info, Node: DEC Alpha Options, Next: DEC Alpha/VMS Options, Prev: Darwin Options, Up: Submodel Options
8511 3.17.8 DEC Alpha Options
8512 ------------------------
8514 These `-m' options are defined for the DEC Alpha implementations:
8518 Use (do not use) the hardware floating-point instructions for
8519 floating-point operations. When `-msoft-float' is specified,
8520 functions in `libgcc.a' will be used to perform floating-point
8521 operations. Unless they are replaced by routines that emulate the
8522 floating-point operations, or compiled in such a way as to call
8523 such emulations routines, these routines will issue floating-point
8524 operations. If you are compiling for an Alpha without
8525 floating-point operations, you must ensure that the library is
8526 built so as not to call them.
8528 Note that Alpha implementations without floating-point operations
8529 are required to have floating-point registers.
8533 Generate code that uses (does not use) the floating-point register
8534 set. `-mno-fp-regs' implies `-msoft-float'. If the floating-point
8535 register set is not used, floating point operands are passed in
8536 integer registers as if they were integers and floating-point
8537 results are passed in `$0' instead of `$f0'. This is a
8538 non-standard calling sequence, so any function with a
8539 floating-point argument or return value called by code compiled
8540 with `-mno-fp-regs' must also be compiled with that option.
8542 A typical use of this option is building a kernel that does not
8543 use, and hence need not save and restore, any floating-point
8547 The Alpha architecture implements floating-point hardware
8548 optimized for maximum performance. It is mostly compliant with
8549 the IEEE floating point standard. However, for full compliance,
8550 software assistance is required. This option generates code fully
8551 IEEE compliant code _except_ that the INEXACT-FLAG is not
8552 maintained (see below). If this option is turned on, the
8553 preprocessor macro `_IEEE_FP' is defined during compilation. The
8554 resulting code is less efficient but is able to correctly support
8555 denormalized numbers and exceptional IEEE values such as
8556 not-a-number and plus/minus infinity. Other Alpha compilers call
8557 this option `-ieee_with_no_inexact'.
8559 `-mieee-with-inexact'
8560 This is like `-mieee' except the generated code also maintains the
8561 IEEE INEXACT-FLAG. Turning on this option causes the generated
8562 code to implement fully-compliant IEEE math. In addition to
8563 `_IEEE_FP', `_IEEE_FP_EXACT' is defined as a preprocessor macro.
8564 On some Alpha implementations the resulting code may execute
8565 significantly slower than the code generated by default. Since
8566 there is very little code that depends on the INEXACT-FLAG, you
8567 should normally not specify this option. Other Alpha compilers
8568 call this option `-ieee_with_inexact'.
8570 `-mfp-trap-mode=TRAP-MODE'
8571 This option controls what floating-point related traps are enabled.
8572 Other Alpha compilers call this option `-fptm TRAP-MODE'. The
8573 trap mode can be set to one of four values:
8576 This is the default (normal) setting. The only traps that
8577 are enabled are the ones that cannot be disabled in software
8578 (e.g., division by zero trap).
8581 In addition to the traps enabled by `n', underflow traps are
8585 Like `u', but the instructions are marked to be safe for
8586 software completion (see Alpha architecture manual for
8590 Like `su', but inexact traps are enabled as well.
8592 `-mfp-rounding-mode=ROUNDING-MODE'
8593 Selects the IEEE rounding mode. Other Alpha compilers call this
8594 option `-fprm ROUNDING-MODE'. The ROUNDING-MODE can be one of:
8597 Normal IEEE rounding mode. Floating point numbers are
8598 rounded towards the nearest machine number or towards the
8599 even machine number in case of a tie.
8602 Round towards minus infinity.
8605 Chopped rounding mode. Floating point numbers are rounded
8609 Dynamic rounding mode. A field in the floating point control
8610 register (FPCR, see Alpha architecture reference manual)
8611 controls the rounding mode in effect. The C library
8612 initializes this register for rounding towards plus infinity.
8613 Thus, unless your program modifies the FPCR, `d' corresponds
8614 to round towards plus infinity.
8616 `-mtrap-precision=TRAP-PRECISION'
8617 In the Alpha architecture, floating point traps are imprecise.
8618 This means without software assistance it is impossible to recover
8619 from a floating trap and program execution normally needs to be
8620 terminated. GCC can generate code that can assist operating
8621 system trap handlers in determining the exact location that caused
8622 a floating point trap. Depending on the requirements of an
8623 application, different levels of precisions can be selected:
8626 Program precision. This option is the default and means a
8627 trap handler can only identify which program caused a
8628 floating point exception.
8631 Function precision. The trap handler can determine the
8632 function that caused a floating point exception.
8635 Instruction precision. The trap handler can determine the
8636 exact instruction that caused a floating point exception.
8638 Other Alpha compilers provide the equivalent options called
8639 `-scope_safe' and `-resumption_safe'.
8642 This option marks the generated code as IEEE conformant. You must
8643 not use this option unless you also specify `-mtrap-precision=i'
8644 and either `-mfp-trap-mode=su' or `-mfp-trap-mode=sui'. Its only
8645 effect is to emit the line `.eflag 48' in the function prologue of
8646 the generated assembly file. Under DEC Unix, this has the effect
8647 that IEEE-conformant math library routines will be linked in.
8650 Normally GCC examines a 32- or 64-bit integer constant to see if
8651 it can construct it from smaller constants in two or three
8652 instructions. If it cannot, it will output the constant as a
8653 literal and generate code to load it from the data segment at
8656 Use this option to require GCC to construct _all_ integer constants
8657 using code, even if it takes more instructions (the maximum is
8660 You would typically use this option to build a shared library
8661 dynamic loader. Itself a shared library, it must relocate itself
8662 in memory before it can find the variables and constants in its
8667 Select whether to generate code to be assembled by the
8668 vendor-supplied assembler (`-malpha-as') or by the GNU assembler
8679 Indicate whether GCC should generate code to use the optional BWX,
8680 CIX, FIX and MAX instruction sets. The default is to use the
8681 instruction sets supported by the CPU type specified via `-mcpu='
8682 option or that of the CPU on which GCC was built if none was
8687 Generate code that uses (does not use) VAX F and G floating point
8688 arithmetic instead of IEEE single and double precision.
8691 `-mno-explicit-relocs'
8692 Older Alpha assemblers provided no way to generate symbol
8693 relocations except via assembler macros. Use of these macros does
8694 not allow optimal instruction scheduling. GNU binutils as of
8695 version 2.12 supports a new syntax that allows the compiler to
8696 explicitly mark which relocations should apply to which
8697 instructions. This option is mostly useful for debugging, as GCC
8698 detects the capabilities of the assembler when it is built and
8699 sets the default accordingly.
8703 When `-mexplicit-relocs' is in effect, static data is accessed via
8704 "gp-relative" relocations. When `-msmall-data' is used, objects 8
8705 bytes long or smaller are placed in a "small data area" (the
8706 `.sdata' and `.sbss' sections) and are accessed via 16-bit
8707 relocations off of the `$gp' register. This limits the size of
8708 the small data area to 64KB, but allows the variables to be
8709 directly accessed via a single instruction.
8711 The default is `-mlarge-data'. With this option the data area is
8712 limited to just below 2GB. Programs that require more than 2GB of
8713 data must use `malloc' or `mmap' to allocate the data in the heap
8714 instead of in the program's data segment.
8716 When generating code for shared libraries, `-fpic' implies
8717 `-msmall-data' and `-fPIC' implies `-mlarge-data'.
8721 When `-msmall-text' is used, the compiler assumes that the code of
8722 the entire program (or shared library) fits in 4MB, and is thus
8723 reachable with a branch instruction. When `-msmall-data' is used,
8724 the compiler can assume that all local symbols share the same
8725 `$gp' value, and thus reduce the number of instructions required
8726 for a function call from 4 to 1.
8728 The default is `-mlarge-text'.
8731 Set the instruction set and instruction scheduling parameters for
8732 machine type CPU_TYPE. You can specify either the `EV' style name
8733 or the corresponding chip number. GCC supports scheduling
8734 parameters for the EV4, EV5 and EV6 family of processors and will
8735 choose the default values for the instruction set from the
8736 processor you specify. If you do not specify a processor type,
8737 GCC will default to the processor on which the compiler was built.
8739 Supported values for CPU_TYPE are
8744 Schedules as an EV4 and has no instruction set extensions.
8748 Schedules as an EV5 and has no instruction set extensions.
8752 Schedules as an EV5 and supports the BWX extension.
8757 Schedules as an EV5 and supports the BWX and MAX extensions.
8761 Schedules as an EV6 and supports the BWX, FIX, and MAX
8766 Schedules as an EV6 and supports the BWX, CIX, FIX, and MAX
8770 Set only the instruction scheduling parameters for machine type
8771 CPU_TYPE. The instruction set is not changed.
8773 `-mmemory-latency=TIME'
8774 Sets the latency the scheduler should assume for typical memory
8775 references as seen by the application. This number is highly
8776 dependent on the memory access patterns used by the application
8777 and the size of the external cache on the machine.
8779 Valid options for TIME are
8782 A decimal number representing clock cycles.
8788 The compiler contains estimates of the number of clock cycles
8789 for "typical" EV4 & EV5 hardware for the Level 1, 2 & 3 caches
8790 (also called Dcache, Scache, and Bcache), as well as to main
8791 memory. Note that L3 is only valid for EV5.
8795 File: gcc.info, Node: DEC Alpha/VMS Options, Next: FRV Options, Prev: DEC Alpha Options, Up: Submodel Options
8797 3.17.9 DEC Alpha/VMS Options
8798 ----------------------------
8800 These `-m' options are defined for the DEC Alpha/VMS implementations:
8802 `-mvms-return-codes'
8803 Return VMS condition codes from main. The default is to return
8804 POSIX style condition (e.g. error) codes.
8807 File: gcc.info, Node: FRV Options, Next: GNU/Linux Options, Prev: DEC Alpha/VMS Options, Up: Submodel Options
8813 Only use the first 32 general purpose registers.
8816 Use all 64 general purpose registers.
8819 Use only the first 32 floating point registers.
8822 Use all 64 floating point registers
8825 Use hardware instructions for floating point operations.
8828 Use library routines for floating point operations.
8831 Dynamically allocate condition code registers.
8834 Do not try to dynamically allocate condition code registers, only
8835 use `icc0' and `fcc0'.
8838 Change ABI to use double word insns.
8841 Do not use double word instructions.
8844 Use floating point double instructions.
8847 Do not use floating point double instructions.
8850 Use media instructions.
8853 Do not use media instructions.
8856 Use multiply and add/subtract instructions.
8859 Do not use multiply and add/subtract instructions.
8862 Select the FDPIC ABI, that uses function descriptors to represent
8863 pointers to functions. Without any PIC/PIE-related options, it
8864 implies `-fPIE'. With `-fpic' or `-fpie', it assumes GOT entries
8865 and small data are within a 12-bit range from the GOT base
8866 address; with `-fPIC' or `-fPIE', GOT offsets are computed with 32
8870 Enable inlining of PLT entries in function calls to functions that
8871 are not known to bind locally. It has no effect without `-mfdpic'.
8872 It's enabled by default if optimizing for speed and compiling for
8873 shared libraries (i.e., `-fPIC' or `-fpic'), or when an
8874 optimization option such as `-O3' or above is present in the
8878 Assume a large TLS segment when generating thread-local code.
8881 Do not assume a large TLS segment when generating thread-local
8885 Enable the use of `GPREL' relocations in the FDPIC ABI for data
8886 that is known to be in read-only sections. It's enabled by
8887 default, except for `-fpic' or `-fpie': even though it may help
8888 make the global offset table smaller, it trades 1 instruction for
8889 4. With `-fPIC' or `-fPIE', it trades 3 instructions for 4, one
8890 of which may be shared by multiple symbols, and it avoids the need
8891 for a GOT entry for the referenced symbol, so it's more likely to
8892 be a win. If it is not, `-mno-gprel-ro' can be used to disable it.
8894 `-multilib-library-pic'
8895 Link with the (library, not FD) pic libraries. It's implied by
8896 `-mlibrary-pic', as well as by `-fPIC' and `-fpic' without
8897 `-mfdpic'. You should never have to use it explicitly.
8900 Follow the EABI requirement of always creating a frame pointer
8901 whenever a stack frame is allocated. This option is enabled by
8902 default and can be disabled with `-mno-linked-fp'.
8905 Use indirect addressing to call functions outside the current
8906 compilation unit. This allows the functions to be placed anywhere
8907 within the 32-bit address space.
8910 Try to align labels to an 8-byte boundary by inserting nops into
8911 the previous packet. This option only has an effect when VLIW
8912 packing is enabled. It doesn't create new packets; it merely adds
8913 nops to existing ones.
8916 Generate position-independent EABI code.
8919 Use only the first four media accumulator registers.
8922 Use all eight media accumulator registers.
8925 Pack VLIW instructions.
8928 Do not pack VLIW instructions.
8931 Do not mark ABI switches in e_flags.
8934 Enable the use of conditional-move instructions (default).
8936 This switch is mainly for debugging the compiler and will likely
8937 be removed in a future version.
8940 Disable the use of conditional-move instructions.
8942 This switch is mainly for debugging the compiler and will likely
8943 be removed in a future version.
8946 Enable the use of conditional set instructions (default).
8948 This switch is mainly for debugging the compiler and will likely
8949 be removed in a future version.
8952 Disable the use of conditional set instructions.
8954 This switch is mainly for debugging the compiler and will likely
8955 be removed in a future version.
8958 Enable the use of conditional execution (default).
8960 This switch is mainly for debugging the compiler and will likely
8961 be removed in a future version.
8964 Disable the use of conditional execution.
8966 This switch is mainly for debugging the compiler and will likely
8967 be removed in a future version.
8970 Run a pass to pack branches into VLIW instructions (default).
8972 This switch is mainly for debugging the compiler and will likely
8973 be removed in a future version.
8976 Do not run a pass to pack branches into VLIW instructions.
8978 This switch is mainly for debugging the compiler and will likely
8979 be removed in a future version.
8982 Enable optimization of `&&' and `||' in conditional execution
8985 This switch is mainly for debugging the compiler and will likely
8986 be removed in a future version.
8988 `-mno-multi-cond-exec'
8989 Disable optimization of `&&' and `||' in conditional execution.
8991 This switch is mainly for debugging the compiler and will likely
8992 be removed in a future version.
8994 `-mnested-cond-exec'
8995 Enable nested conditional execution optimizations (default).
8997 This switch is mainly for debugging the compiler and will likely
8998 be removed in a future version.
9000 `-mno-nested-cond-exec'
9001 Disable nested conditional execution optimizations.
9003 This switch is mainly for debugging the compiler and will likely
9004 be removed in a future version.
9007 This switch removes redundant `membar' instructions from the
9008 compiler generated code. It is enabled by default.
9010 `-mno-optimize-membar'
9011 This switch disables the automatic removal of redundant `membar'
9012 instructions from the generated code.
9015 Cause gas to print out tomcat statistics.
9018 Select the processor type for which to generate code. Possible
9019 values are `frv', `fr550', `tomcat', `fr500', `fr450', `fr405',
9020 `fr400', `fr300' and `simple'.
9024 File: gcc.info, Node: GNU/Linux Options, Next: H8/300 Options, Prev: FRV Options, Up: Submodel Options
9026 3.17.11 GNU/Linux Options
9027 -------------------------
9029 These `-m' options are defined for GNU/Linux targets:
9032 Use the GNU C library instead of uClibc. This is the default
9033 except on `*-*-linux-*uclibc*' targets.
9036 Use uClibc instead of the GNU C library. This is the default on
9037 `*-*-linux-*uclibc*' targets.
9040 File: gcc.info, Node: H8/300 Options, Next: HPPA Options, Prev: GNU/Linux Options, Up: Submodel Options
9042 3.17.12 H8/300 Options
9043 ----------------------
9045 These `-m' options are defined for the H8/300 implementations:
9048 Shorten some address references at link time, when possible; uses
9049 the linker option `-relax'. *Note `ld' and the H8/300:
9050 (ld)H8/300, for a fuller description.
9053 Generate code for the H8/300H.
9056 Generate code for the H8S.
9059 Generate code for the H8S and H8/300H in the normal mode. This
9060 switch must be used either with `-mh' or `-ms'.
9063 Generate code for the H8S/2600. This switch must be used with
9067 Make `int' data 32 bits by default.
9070 On the H8/300H and H8S, use the same alignment rules as for the
9071 H8/300. The default for the H8/300H and H8S is to align longs and
9072 floats on 4 byte boundaries. `-malign-300' causes them to be
9073 aligned on 2 byte boundaries. This option has no effect on the
9077 File: gcc.info, Node: HPPA Options, Next: i386 and x86-64 Options, Prev: H8/300 Options, Up: Submodel Options
9079 3.17.13 HPPA Options
9080 --------------------
9082 These `-m' options are defined for the HPPA family of computers:
9084 `-march=ARCHITECTURE-TYPE'
9085 Generate code for the specified architecture. The choices for
9086 ARCHITECTURE-TYPE are `1.0' for PA 1.0, `1.1' for PA 1.1, and
9087 `2.0' for PA 2.0 processors. Refer to `/usr/lib/sched.models' on
9088 an HP-UX system to determine the proper architecture option for
9089 your machine. Code compiled for lower numbered architectures will
9090 run on higher numbered architectures, but not the other way around.
9095 Synonyms for `-march=1.0', `-march=1.1', and `-march=2.0'
9099 Generate code suitable for big switch tables. Use this option
9100 only if the assembler/linker complain about out of range branches
9101 within a switch table.
9104 Fill delay slots of function calls with unconditional jump
9105 instructions by modifying the return pointer for the function call
9106 to be the target of the conditional jump.
9109 Prevent floating point registers from being used in any manner.
9110 This is necessary for compiling kernels which perform lazy context
9111 switching of floating point registers. If you use this option and
9112 attempt to perform floating point operations, the compiler will
9115 `-mdisable-indexing'
9116 Prevent the compiler from using indexing address modes. This
9117 avoids some rather obscure problems when compiling MIG generated
9121 Generate code that assumes the target has no space registers.
9122 This allows GCC to generate faster indirect calls and use unscaled
9123 index address modes.
9125 Such code is suitable for level 0 PA systems and kernels.
9127 `-mfast-indirect-calls'
9128 Generate code that assumes calls never cross space boundaries.
9129 This allows GCC to emit code which performs faster indirect calls.
9131 This option will not work in the presence of shared libraries or
9134 `-mfixed-range=REGISTER-RANGE'
9135 Generate code treating the given register range as fixed registers.
9136 A fixed register is one that the register allocator can not use.
9137 This is useful when compiling kernel code. A register range is
9138 specified as two registers separated by a dash. Multiple register
9139 ranges can be specified separated by a comma.
9142 Generate 3-instruction load and store sequences as sometimes
9143 required by the HP-UX 10 linker. This is equivalent to the `+k'
9144 option to the HP compilers.
9146 `-mportable-runtime'
9147 Use the portable calling conventions proposed by HP for ELF
9151 Enable the use of assembler directives only GAS understands.
9153 `-mschedule=CPU-TYPE'
9154 Schedule code according to the constraints for the machine type
9155 CPU-TYPE. The choices for CPU-TYPE are `700' `7100', `7100LC',
9156 `7200', `7300' and `8000'. Refer to `/usr/lib/sched.models' on an
9157 HP-UX system to determine the proper scheduling option for your
9158 machine. The default scheduling is `8000'.
9161 Enable the optimization pass in the HP-UX linker. Note this makes
9162 symbolic debugging impossible. It also triggers a bug in the
9163 HP-UX 8 and HP-UX 9 linkers in which they give bogus error
9164 messages when linking some programs.
9167 Generate output containing library calls for floating point.
9168 *Warning:* the requisite libraries are not available for all HPPA
9169 targets. Normally the facilities of the machine's usual C
9170 compiler are used, but this cannot be done directly in
9171 cross-compilation. You must make your own arrangements to provide
9172 suitable library functions for cross-compilation. The embedded
9173 target `hppa1.1-*-pro' does provide software floating point
9176 `-msoft-float' changes the calling convention in the output file;
9177 therefore, it is only useful if you compile _all_ of a program with
9178 this option. In particular, you need to compile `libgcc.a', the
9179 library that comes with GCC, with `-msoft-float' in order for this
9183 Generate the predefine, `_SIO', for server IO. The default is
9184 `-mwsio'. This generates the predefines, `__hp9000s700',
9185 `__hp9000s700__' and `_WSIO', for workstation IO. These options
9186 are available under HP-UX and HI-UX.
9189 Use GNU ld specific options. This passes `-shared' to ld when
9190 building a shared library. It is the default when GCC is
9191 configured, explicitly or implicitly, with the GNU linker. This
9192 option does not have any affect on which ld is called, it only
9193 changes what parameters are passed to that ld. The ld that is
9194 called is determined by the `--with-ld' configure option, GCC's
9195 program search path, and finally by the user's `PATH'. The linker
9196 used by GCC can be printed using `which `gcc
9197 -print-prog-name=ld`'. This option is only available on the 64
9198 bit HP-UX GCC, i.e. configured with `hppa*64*-*-hpux*'.
9201 Use HP ld specific options. This passes `-b' to ld when building
9202 a shared library and passes `+Accept TypeMismatch' to ld on all
9203 links. It is the default when GCC is configured, explicitly or
9204 implicitly, with the HP linker. This option does not have any
9205 affect on which ld is called, it only changes what parameters are
9206 passed to that ld. The ld that is called is determined by the
9207 `--with-ld' configure option, GCC's program search path, and
9208 finally by the user's `PATH'. The linker used by GCC can be
9209 printed using `which `gcc -print-prog-name=ld`'. This option is
9210 only available on the 64 bit HP-UX GCC, i.e. configured with
9214 Generate code that uses long call sequences. This ensures that a
9215 call is always able to reach linker generated stubs. The default
9216 is to generate long calls only when the distance from the call
9217 site to the beginning of the function or translation unit, as the
9218 case may be, exceeds a predefined limit set by the branch type
9219 being used. The limits for normal calls are 7,600,000 and 240,000
9220 bytes, respectively for the PA 2.0 and PA 1.X architectures.
9221 Sibcalls are always limited at 240,000 bytes.
9223 Distances are measured from the beginning of functions when using
9224 the `-ffunction-sections' option, or when using the `-mgas' and
9225 `-mno-portable-runtime' options together under HP-UX with the SOM
9228 It is normally not desirable to use this option as it will degrade
9229 performance. However, it may be useful in large applications,
9230 particularly when partial linking is used to build the application.
9232 The types of long calls used depends on the capabilities of the
9233 assembler and linker, and the type of code being generated. The
9234 impact on systems that support long absolute calls, and long pic
9235 symbol-difference or pc-relative calls should be relatively small.
9236 However, an indirect call is used on 32-bit ELF systems in pic code
9237 and it is quite long.
9240 Generate compiler predefines and select a startfile for the
9241 specified UNIX standard. The choices for UNIX-STD are `93', `95'
9242 and `98'. `93' is supported on all HP-UX versions. `95' is
9243 available on HP-UX 10.10 and later. `98' is available on HP-UX
9244 11.11 and later. The default values are `93' for HP-UX 10.00,
9245 `95' for HP-UX 10.10 though to 11.00, and `98' for HP-UX 11.11 and
9248 `-munix=93' provides the same predefines as GCC 3.3 and 3.4.
9249 `-munix=95' provides additional predefines for `XOPEN_UNIX' and
9250 `_XOPEN_SOURCE_EXTENDED', and the startfile `unix95.o'.
9251 `-munix=98' provides additional predefines for `_XOPEN_UNIX',
9252 `_XOPEN_SOURCE_EXTENDED', `_INCLUDE__STDC_A1_SOURCE' and
9253 `_INCLUDE_XOPEN_SOURCE_500', and the startfile `unix98.o'.
9255 It is _important_ to note that this option changes the interfaces
9256 for various library routines. It also affects the operational
9257 behavior of the C library. Thus, _extreme_ care is needed in
9260 Library code that is intended to operate with more than one UNIX
9261 standard must test, set and restore the variable
9262 __XPG4_EXTENDED_MASK as appropriate. Most GNU software doesn't
9263 provide this capability.
9266 Suppress the generation of link options to search libdld.sl when
9267 the `-static' option is specified on HP-UX 10 and later.
9270 The HP-UX implementation of setlocale in libc has a dependency on
9271 libdld.sl. There isn't an archive version of libdld.sl. Thus,
9272 when the `-static' option is specified, special link options are
9273 needed to resolve this dependency.
9275 On HP-UX 10 and later, the GCC driver adds the necessary options to
9276 link with libdld.sl when the `-static' option is specified. This
9277 causes the resulting binary to be dynamic. On the 64-bit port,
9278 the linkers generate dynamic binaries by default in any case. The
9279 `-nolibdld' option can be used to prevent the GCC driver from
9280 adding these link options.
9283 Add support for multithreading with the "dce thread" library under
9284 HP-UX. This option sets flags for both the preprocessor and
9288 File: gcc.info, Node: i386 and x86-64 Options, Next: IA-64 Options, Prev: HPPA Options, Up: Submodel Options
9290 3.17.14 Intel 386 and AMD x86-64 Options
9291 ----------------------------------------
9293 These `-m' options are defined for the i386 and x86-64 family of
9297 Tune to CPU-TYPE everything applicable about the generated code,
9298 except for the ABI and the set of available instructions. The
9299 choices for CPU-TYPE are:
9301 Produce code optimized for the most common IA32/AMD64/EM64T
9302 processors. If you know the CPU on which your code will run,
9303 then you should use the corresponding `-mtune' option instead
9304 of `-mtune=generic'. But, if you do not know exactly what
9305 CPU users of your application will have, then you should use
9308 As new processors are deployed in the marketplace, the
9309 behavior of this option will change. Therefore, if you
9310 upgrade to a newer version of GCC, the code generated option
9311 will change to reflect the processors that were most common
9312 when that version of GCC was released.
9314 There is no `-march=generic' option because `-march'
9315 indicates the instruction set the compiler can use, and there
9316 is no generic instruction set applicable to all processors.
9317 In contrast, `-mtune' indicates the processor (or, in this
9318 case, collection of processors) for which the code is
9322 This selects the CPU to tune for at compilation time by
9323 determining the processor type of the compiling machine.
9324 Using `-mtune=native' will produce code optimized for the
9325 local machine under the constraints of the selected
9326 instruction set. Using `-march=native' will enable all
9327 instruction subsets supported by the local machine (hence the
9328 result might not run on different machines).
9331 Original Intel's i386 CPU.
9334 Intel's i486 CPU. (No scheduling is implemented for this
9338 Intel Pentium CPU with no MMX support.
9341 Intel PentiumMMX CPU based on Pentium core with MMX
9342 instruction set support.
9345 Intel PentiumPro CPU.
9348 Same as `generic', but when used as `march' option, PentiumPro
9349 instruction set will be used, so the code will run on all
9353 Intel Pentium2 CPU based on PentiumPro core with MMX
9354 instruction set support.
9356 _pentium3, pentium3m_
9357 Intel Pentium3 CPU based on PentiumPro core with MMX and SSE
9358 instruction set support.
9361 Low power version of Intel Pentium3 CPU with MMX, SSE and
9362 SSE2 instruction set support. Used by Centrino notebooks.
9364 _pentium4, pentium4m_
9365 Intel Pentium4 CPU with MMX, SSE and SSE2 instruction set
9369 Improved version of Intel Pentium4 CPU with MMX, SSE, SSE2
9370 and SSE3 instruction set support.
9373 Improved version of Intel Pentium4 CPU with 64-bit
9374 extensions, MMX, SSE, SSE2 and SSE3 instruction set support.
9377 AMD K6 CPU with MMX instruction set support.
9380 Improved versions of AMD K6 CPU with MMX and 3dNOW!
9381 instruction set support.
9383 _athlon, athlon-tbird_
9384 AMD Athlon CPU with MMX, 3dNOW!, enhanced 3dNOW! and SSE
9385 prefetch instructions support.
9387 _athlon-4, athlon-xp, athlon-mp_
9388 Improved AMD Athlon CPU with MMX, 3dNOW!, enhanced 3dNOW! and
9389 full SSE instruction set support.
9391 _k8, opteron, athlon64, athlon-fx_
9392 AMD K8 core based CPUs with x86-64 instruction set support.
9393 (This supersets MMX, SSE, SSE2, 3dNOW!, enhanced 3dNOW! and
9394 64-bit instruction set extensions.)
9397 IDT Winchip C6 CPU, dealt in same way as i486 with additional
9398 MMX instruction set support.
9401 IDT Winchip2 CPU, dealt in same way as i486 with additional
9402 MMX and 3dNOW! instruction set support.
9405 Via C3 CPU with MMX and 3dNOW! instruction set support. (No
9406 scheduling is implemented for this chip.)
9409 Via C3-2 CPU with MMX and SSE instruction set support. (No
9410 scheduling is implemented for this chip.)
9412 While picking a specific CPU-TYPE will schedule things
9413 appropriately for that particular chip, the compiler will not
9414 generate any code that does not run on the i386 without the
9415 `-march=CPU-TYPE' option being used.
9418 Generate instructions for the machine type CPU-TYPE. The choices
9419 for CPU-TYPE are the same as for `-mtune'. Moreover, specifying
9420 `-march=CPU-TYPE' implies `-mtune=CPU-TYPE'.
9423 A deprecated synonym for `-mtune'.
9429 These options are synonyms for `-mtune=i386', `-mtune=i486',
9430 `-mtune=pentium', and `-mtune=pentiumpro' respectively. These
9431 synonyms are deprecated.
9434 Generate floating point arithmetics for selected unit UNIT. The
9435 choices for UNIT are:
9438 Use the standard 387 floating point coprocessor present
9439 majority of chips and emulated otherwise. Code compiled with
9440 this option will run almost everywhere. The temporary
9441 results are computed in 80bit precision instead of precision
9442 specified by the type resulting in slightly different results
9443 compared to most of other chips. See `-ffloat-store' for
9444 more detailed description.
9446 This is the default choice for i386 compiler.
9449 Use scalar floating point instructions present in the SSE
9450 instruction set. This instruction set is supported by
9451 Pentium3 and newer chips, in the AMD line by Athlon-4,
9452 Athlon-xp and Athlon-mp chips. The earlier version of SSE
9453 instruction set supports only single precision arithmetics,
9454 thus the double and extended precision arithmetics is still
9455 done using 387. Later version, present only in Pentium4 and
9456 the future AMD x86-64 chips supports double precision
9459 For the i386 compiler, you need to use `-march=CPU-TYPE',
9460 `-msse' or `-msse2' switches to enable SSE extensions and
9461 make this option effective. For the x86-64 compiler, these
9462 extensions are enabled by default.
9464 The resulting code should be considerably faster in the
9465 majority of cases and avoid the numerical instability
9466 problems of 387 code, but may break some existing code that
9467 expects temporaries to be 80bit.
9469 This is the default choice for the x86-64 compiler.
9472 Attempt to utilize both instruction sets at once. This
9473 effectively double the amount of available registers and on
9474 chips with separate execution units for 387 and SSE the
9475 execution resources too. Use this option with care, as it is
9476 still experimental, because the GCC register allocator does
9477 not model separate functional units well resulting in
9478 instable performance.
9481 Output asm instructions using selected DIALECT. Supported choices
9482 are `intel' or `att' (the default one). Darwin does not support
9487 Control whether or not the compiler uses IEEE floating point
9488 comparisons. These handle correctly the case where the result of a
9489 comparison is unordered.
9492 Generate output containing library calls for floating point.
9493 *Warning:* the requisite libraries are not part of GCC. Normally
9494 the facilities of the machine's usual C compiler are used, but
9495 this can't be done directly in cross-compilation. You must make
9496 your own arrangements to provide suitable library functions for
9499 On machines where a function returns floating point results in the
9500 80387 register stack, some floating point opcodes may be emitted
9501 even if `-msoft-float' is used.
9503 `-mno-fp-ret-in-387'
9504 Do not use the FPU registers for return values of functions.
9506 The usual calling convention has functions return values of types
9507 `float' and `double' in an FPU register, even if there is no FPU.
9508 The idea is that the operating system should emulate an FPU.
9510 The option `-mno-fp-ret-in-387' causes such values to be returned
9511 in ordinary CPU registers instead.
9513 `-mno-fancy-math-387'
9514 Some 387 emulators do not support the `sin', `cos' and `sqrt'
9515 instructions for the 387. Specify this option to avoid generating
9516 those instructions. This option is the default on FreeBSD,
9517 OpenBSD and NetBSD. This option is overridden when `-march'
9518 indicates that the target cpu will always have an FPU and so the
9519 instruction will not need emulation. As of revision 2.6.1, these
9520 instructions are not generated unless you also use the
9521 `-funsafe-math-optimizations' switch.
9525 Control whether GCC aligns `double', `long double', and `long
9526 long' variables on a two word boundary or a one word boundary.
9527 Aligning `double' variables on a two word boundary will produce
9528 code that runs somewhat faster on a `Pentium' at the expense of
9531 On x86-64, `-malign-double' is enabled by default.
9533 *Warning:* if you use the `-malign-double' switch, structures
9534 containing the above types will be aligned differently than the
9535 published application binary interface specifications for the 386
9536 and will not be binary compatible with structures in code compiled
9537 without that switch.
9539 `-m96bit-long-double'
9540 `-m128bit-long-double'
9541 These switches control the size of `long double' type. The i386
9542 application binary interface specifies the size to be 96 bits, so
9543 `-m96bit-long-double' is the default in 32 bit mode.
9545 Modern architectures (Pentium and newer) would prefer `long double'
9546 to be aligned to an 8 or 16 byte boundary. In arrays or structures
9547 conforming to the ABI, this would not be possible. So specifying a
9548 `-m128bit-long-double' will align `long double' to a 16 byte
9549 boundary by padding the `long double' with an additional 32 bit
9552 In the x86-64 compiler, `-m128bit-long-double' is the default
9553 choice as its ABI specifies that `long double' is to be aligned on
9556 Notice that neither of these options enable any extra precision
9557 over the x87 standard of 80 bits for a `long double'.
9559 *Warning:* if you override the default value for your target ABI,
9560 the structures and arrays containing `long double' variables will
9561 change their size as well as function calling convention for
9562 function taking `long double' will be modified. Hence they will
9563 not be binary compatible with arrays or structures in code
9564 compiled without that switch.
9566 `-mmlarge-data-threshold=NUMBER'
9567 When `-mcmodel=medium' is specified, the data greater than
9568 THRESHOLD are placed in large data section. This value must be the
9569 same across all object linked into the binary and defaults to
9574 Control whether GCC places uninitialized local variables into the
9575 `bss' or `data' segments. `-msvr3-shlib' places them into `bss'.
9576 These options are meaningful only on System V Release 3.
9579 Use a different function-calling convention, in which functions
9580 that take a fixed number of arguments return with the `ret' NUM
9581 instruction, which pops their arguments while returning. This
9582 saves one instruction in the caller since there is no need to pop
9583 the arguments there.
9585 You can specify that an individual function is called with this
9586 calling sequence with the function attribute `stdcall'. You can
9587 also override the `-mrtd' option by using the function attribute
9588 `cdecl'. *Note Function Attributes::.
9590 *Warning:* this calling convention is incompatible with the one
9591 normally used on Unix, so you cannot use it if you need to call
9592 libraries compiled with the Unix compiler.
9594 Also, you must provide function prototypes for all functions that
9595 take variable numbers of arguments (including `printf'); otherwise
9596 incorrect code will be generated for calls to those functions.
9598 In addition, seriously incorrect code will result if you call a
9599 function with too many arguments. (Normally, extra arguments are
9600 harmlessly ignored.)
9603 Control how many registers are used to pass integer arguments. By
9604 default, no registers are used to pass arguments, and at most 3
9605 registers can be used. You can control this behavior for a
9606 specific function by using the function attribute `regparm'.
9607 *Note Function Attributes::.
9609 *Warning:* if you use this switch, and NUM is nonzero, then you
9610 must build all modules with the same value, including any
9611 libraries. This includes the system libraries and startup modules.
9614 Use SSE register passing conventions for float and double arguments
9615 and return values. You can control this behavior for a specific
9616 function by using the function attribute `sseregparm'. *Note
9617 Function Attributes::.
9619 *Warning:* if you use this switch then you must build all modules
9620 with the same value, including any libraries. This includes the
9621 system libraries and startup modules.
9624 Realign the stack at entry. On the Intel x86, the
9625 `-mstackrealign' option will generate an alternate prologue and
9626 epilogue that realigns the runtime stack. This supports mixing
9627 legacy codes that keep a 4-byte aligned stack with modern codes
9628 that keep a 16-byte stack for SSE compatibility. The alternate
9629 prologue and epilogue are slower and bigger than the regular ones,
9630 and the alternate prologue requires an extra scratch register;
9631 this lowers the number of registers available if used in
9632 conjunction with the `regparm' attribute. The `-mstackrealign'
9633 option is incompatible with the nested function prologue; this is
9634 considered a hard error. See also the attribute
9635 `force_align_arg_pointer', applicable to individual functions.
9637 `-mpreferred-stack-boundary=NUM'
9638 Attempt to keep the stack boundary aligned to a 2 raised to NUM
9639 byte boundary. If `-mpreferred-stack-boundary' is not specified,
9640 the default is 4 (16 bytes or 128 bits).
9642 On Pentium and PentiumPro, `double' and `long double' values
9643 should be aligned to an 8 byte boundary (see `-malign-double') or
9644 suffer significant run time performance penalties. On Pentium
9645 III, the Streaming SIMD Extension (SSE) data type `__m128' may not
9646 work properly if it is not 16 byte aligned.
9648 To ensure proper alignment of this values on the stack, the stack
9649 boundary must be as aligned as that required by any value stored
9650 on the stack. Further, every function must be generated such that
9651 it keeps the stack aligned. Thus calling a function compiled with
9652 a higher preferred stack boundary from a function compiled with a
9653 lower preferred stack boundary will most likely misalign the
9654 stack. It is recommended that libraries that use callbacks always
9655 use the default setting.
9657 This extra alignment does consume extra stack space, and generally
9658 increases code size. Code that is sensitive to stack space usage,
9659 such as embedded systems and operating system kernels, may want to
9660 reduce the preferred alignment to `-mpreferred-stack-boundary=2'.
9676 These switches enable or disable the use of instructions in the
9677 MMX, SSE, SSE2 or 3DNow! extended instruction sets. These
9678 extensions are also available as built-in functions: see *Note X86
9679 Built-in Functions::, for details of the functions enabled and
9680 disabled by these switches.
9682 To have SSE/SSE2 instructions generated automatically from
9683 floating-point code (as opposed to 387 instructions), see
9686 These options will enable GCC to use these extended instructions in
9687 generated code, even without `-mfpmath=sse'. Applications which
9688 perform runtime CPU detection must compile separate files for each
9689 supported architecture, using the appropriate flags. In
9690 particular, the file containing the CPU detection code should be
9691 compiled without these options.
9695 Use PUSH operations to store outgoing parameters. This method is
9696 shorter and usually equally fast as method using SUB/MOV
9697 operations and is enabled by default. In some cases disabling it
9698 may improve performance because of improved scheduling and reduced
9701 `-maccumulate-outgoing-args'
9702 If enabled, the maximum amount of space required for outgoing
9703 arguments will be computed in the function prologue. This is
9704 faster on most modern CPUs because of reduced dependencies,
9705 improved scheduling and reduced stack usage when preferred stack
9706 boundary is not equal to 2. The drawback is a notable increase in
9707 code size. This switch implies `-mno-push-args'.
9710 Support thread-safe exception handling on `Mingw32'. Code that
9711 relies on thread-safe exception handling must compile and link all
9712 code with the `-mthreads' option. When compiling, `-mthreads'
9713 defines `-D_MT'; when linking, it links in a special thread helper
9714 library `-lmingwthrd' which cleans up per thread exception
9717 `-mno-align-stringops'
9718 Do not align destination of inlined string operations. This
9719 switch reduces code size and improves performance in case the
9720 destination is already aligned, but GCC doesn't know about it.
9722 `-minline-all-stringops'
9723 By default GCC inlines string operations only when destination is
9724 known to be aligned at least to 4 byte boundary. This enables
9725 more inlining, increase code size, but may improve performance of
9726 code that depends on fast memcpy, strlen and memset for short
9729 `-momit-leaf-frame-pointer'
9730 Don't keep the frame pointer in a register for leaf functions.
9731 This avoids the instructions to save, set up and restore frame
9732 pointers and makes an extra register available in leaf functions.
9733 The option `-fomit-frame-pointer' removes the frame pointer for
9734 all functions which might make debugging harder.
9736 `-mtls-direct-seg-refs'
9737 `-mno-tls-direct-seg-refs'
9738 Controls whether TLS variables may be accessed with offsets from
9739 the TLS segment register (`%gs' for 32-bit, `%fs' for 64-bit), or
9740 whether the thread base pointer must be added. Whether or not this
9741 is legal depends on the operating system, and whether it maps the
9742 segment to cover the entire TLS area.
9744 For systems that use GNU libc, the default is on.
9746 These `-m' switches are supported in addition to the above on AMD
9747 x86-64 processors in 64-bit environments.
9751 Generate code for a 32-bit or 64-bit environment. The 32-bit
9752 environment sets int, long and pointer to 32 bits and generates
9753 code that runs on any i386 system. The 64-bit environment sets
9754 int to 32 bits and long and pointer to 64 bits and generates code
9755 for AMD's x86-64 architecture. For darwin only the -m64 option
9756 turns off the `-fno-pic' and `-mdynamic-no-pic' options.
9759 Do not use a so called red zone for x86-64 code. The red zone is
9760 mandated by the x86-64 ABI, it is a 128-byte area beyond the
9761 location of the stack pointer that will not be modified by signal
9762 or interrupt handlers and therefore can be used for temporary data
9763 without adjusting the stack pointer. The flag `-mno-red-zone'
9764 disables this red zone.
9767 Generate code for the small code model: the program and its
9768 symbols must be linked in the lower 2 GB of the address space.
9769 Pointers are 64 bits. Programs can be statically or dynamically
9770 linked. This is the default code model.
9773 Generate code for the kernel code model. The kernel runs in the
9774 negative 2 GB of the address space. This model has to be used for
9778 Generate code for the medium model: The program is linked in the
9779 lower 2 GB of the address space but symbols can be located
9780 anywhere in the address space. Programs can be statically or
9781 dynamically linked, but building of shared libraries are not
9782 supported with the medium model.
9785 Generate code for the large model: This model makes no assumptions
9786 about addresses and sizes of sections. Currently GCC does not
9787 implement this model.
9790 File: gcc.info, Node: IA-64 Options, Next: M32C Options, Prev: i386 and x86-64 Options, Up: Submodel Options
9792 3.17.15 IA-64 Options
9793 ---------------------
9795 These are the `-m' options defined for the Intel IA-64 architecture.
9798 Generate code for a big endian target. This is the default for
9802 Generate code for a little endian target. This is the default for
9807 Generate (or don't) code for the GNU assembler. This is the
9812 Generate (or don't) code for the GNU linker. This is the default.
9815 Generate code that does not use a global pointer register. The
9816 result is not position independent code, and violates the IA-64
9819 `-mvolatile-asm-stop'
9820 `-mno-volatile-asm-stop'
9821 Generate (or don't) a stop bit immediately before and after
9822 volatile asm statements.
9825 `-mno-register-names'
9826 Generate (or don't) `in', `loc', and `out' register names for the
9827 stacked registers. This may make assembler output more readable.
9831 Disable (or enable) optimizations that use the small data section.
9832 This may be useful for working around optimizer bugs.
9835 Generate code that uses a single constant global pointer value.
9836 This is useful when compiling kernel code.
9839 Generate code that is self-relocatable. This implies
9840 `-mconstant-gp'. This is useful when compiling firmware code.
9842 `-minline-float-divide-min-latency'
9843 Generate code for inline divides of floating point values using
9844 the minimum latency algorithm.
9846 `-minline-float-divide-max-throughput'
9847 Generate code for inline divides of floating point values using
9848 the maximum throughput algorithm.
9850 `-minline-int-divide-min-latency'
9851 Generate code for inline divides of integer values using the
9852 minimum latency algorithm.
9854 `-minline-int-divide-max-throughput'
9855 Generate code for inline divides of integer values using the
9856 maximum throughput algorithm.
9858 `-minline-sqrt-min-latency'
9859 Generate code for inline square roots using the minimum latency
9862 `-minline-sqrt-max-throughput'
9863 Generate code for inline square roots using the maximum throughput
9868 Don't (or do) generate assembler code for the DWARF2 line number
9869 debugging info. This may be useful when not using the GNU
9873 `-mno-early-stop-bits'
9874 Allow stop bits to be placed earlier than immediately preceding the
9875 instruction that triggered the stop bit. This can improve
9876 instruction scheduling, but does not always do so.
9878 `-mfixed-range=REGISTER-RANGE'
9879 Generate code treating the given register range as fixed registers.
9880 A fixed register is one that the register allocator can not use.
9881 This is useful when compiling kernel code. A register range is
9882 specified as two registers separated by a dash. Multiple register
9883 ranges can be specified separated by a comma.
9885 `-mtls-size=TLS-SIZE'
9886 Specify bit size of immediate TLS offsets. Valid values are 14,
9890 Tune the instruction scheduling for a particular CPU, Valid values
9891 are itanium, itanium1, merced, itanium2, and mckinley.
9895 Add support for multithreading using the POSIX threads library.
9896 This option sets flags for both the preprocessor and linker. It
9897 does not affect the thread safety of object code produced by the
9898 compiler or that of libraries supplied with it. These are HP-UX
9903 Generate code for a 32-bit or 64-bit environment. The 32-bit
9904 environment sets int, long and pointer to 32 bits. The 64-bit
9905 environment sets int to 32 bits and long and pointer to 64 bits.
9906 These are HP-UX specific flags.
9908 `-mno-sched-br-data-spec'
9909 `-msched-br-data-spec'
9910 (Dis/En)able data speculative scheduling before reload. This will
9911 result in generation of the ld.a instructions and the
9912 corresponding check instructions (ld.c / chk.a). The default is
9915 `-msched-ar-data-spec'
9916 `-mno-sched-ar-data-spec'
9917 (En/Dis)able data speculative scheduling after reload. This will
9918 result in generation of the ld.a instructions and the
9919 corresponding check instructions (ld.c / chk.a). The default is
9922 `-mno-sched-control-spec'
9923 `-msched-control-spec'
9924 (Dis/En)able control speculative scheduling. This feature is
9925 available only during region scheduling (i.e. before reload).
9926 This will result in generation of the ld.s instructions and the
9927 corresponding check instructions chk.s . The default is 'disable'.
9929 `-msched-br-in-data-spec'
9930 `-mno-sched-br-in-data-spec'
9931 (En/Dis)able speculative scheduling of the instructions that are
9932 dependent on the data speculative loads before reload. This is
9933 effective only with `-msched-br-data-spec' enabled. The default
9936 `-msched-ar-in-data-spec'
9937 `-mno-sched-ar-in-data-spec'
9938 (En/Dis)able speculative scheduling of the instructions that are
9939 dependent on the data speculative loads after reload. This is
9940 effective only with `-msched-ar-data-spec' enabled. The default
9943 `-msched-in-control-spec'
9944 `-mno-sched-in-control-spec'
9945 (En/Dis)able speculative scheduling of the instructions that are
9946 dependent on the control speculative loads. This is effective
9947 only with `-msched-control-spec' enabled. The default is 'enable'.
9951 (En/Dis)able use of simple data speculation checks ld.c . If
9952 disabled, only chk.a instructions will be emitted to check data
9953 speculative loads. The default is 'enable'.
9955 `-mno-sched-control-ldc'
9956 `-msched-control-ldc'
9957 (Dis/En)able use of ld.c instructions to check control speculative
9958 loads. If enabled, in case of control speculative load with no
9959 speculatively scheduled dependent instructions this load will be
9960 emitted as ld.sa and ld.c will be used to check it. The default
9963 `-mno-sched-spec-verbose'
9964 `-msched-spec-verbose'
9965 (Dis/En)able printing of the information about speculative motions.
9967 `-mno-sched-prefer-non-data-spec-insns'
9968 `-msched-prefer-non-data-spec-insns'
9969 If enabled, data speculative instructions will be chosen for
9970 schedule only if there are no other choices at the moment. This
9971 will make the use of the data speculation much more conservative.
9972 The default is 'disable'.
9974 `-mno-sched-prefer-non-control-spec-insns'
9975 `-msched-prefer-non-control-spec-insns'
9976 If enabled, control speculative instructions will be chosen for
9977 schedule only if there are no other choices at the moment. This
9978 will make the use of the control speculation much more
9979 conservative. The default is 'disable'.
9981 `-mno-sched-count-spec-in-critical-path'
9982 `-msched-count-spec-in-critical-path'
9983 If enabled, speculative dependencies will be considered during
9984 computation of the instructions priorities. This will make the
9985 use of the speculation a bit more conservative. The default is
9990 File: gcc.info, Node: M32C Options, Next: M32R/D Options, Prev: IA-64 Options, Up: Submodel Options
9992 3.17.16 M32C Options
9993 --------------------
9996 Select the CPU for which code is generated. NAME may be one of
9997 `r8c' for the R8C/Tiny series, `m16c' for the M16C (up to /60)
9998 series, `m32cm' for the M16C/80 series, or `m32c' for the M32C/80
10002 Specifies that the program will be run on the simulator. This
10003 causes an alternate runtime library to be linked in which
10004 supports, for example, file I/O. You must not use this option
10005 when generating programs that will run on real hardware; you must
10006 provide your own runtime library for whatever I/O functions are
10010 Specifies the number of memory-based pseudo-registers GCC will use
10011 during code generation. These pseudo-registers will be used like
10012 real registers, so there is a tradeoff between GCC's ability to
10013 fit the code into available registers, and the performance penalty
10014 of using memory instead of registers. Note that all modules in a
10015 program must be compiled with the same value for this option.
10016 Because of that, you must not use this option with the default
10017 runtime libraries gcc builds.
10021 File: gcc.info, Node: M32R/D Options, Next: M680x0 Options, Prev: M32C Options, Up: Submodel Options
10023 3.17.17 M32R/D Options
10024 ----------------------
10026 These `-m' options are defined for Renesas M32R/D architectures:
10029 Generate code for the M32R/2.
10032 Generate code for the M32R/X.
10035 Generate code for the M32R. This is the default.
10038 Assume all objects live in the lower 16MB of memory (so that their
10039 addresses can be loaded with the `ld24' instruction), and assume
10040 all subroutines are reachable with the `bl' instruction. This is
10043 The addressability of a particular object can be set with the
10047 Assume objects may be anywhere in the 32-bit address space (the
10048 compiler will generate `seth/add3' instructions to load their
10049 addresses), and assume all subroutines are reachable with the `bl'
10053 Assume objects may be anywhere in the 32-bit address space (the
10054 compiler will generate `seth/add3' instructions to load their
10055 addresses), and assume subroutines may not be reachable with the
10056 `bl' instruction (the compiler will generate the much slower
10057 `seth/add3/jl' instruction sequence).
10060 Disable use of the small data area. Variables will be put into
10061 one of `.data', `bss', or `.rodata' (unless the `section'
10062 attribute has been specified). This is the default.
10064 The small data area consists of sections `.sdata' and `.sbss'.
10065 Objects may be explicitly put in the small data area with the
10066 `section' attribute using one of these sections.
10069 Put small global and static data in the small data area, but do not
10070 generate special code to reference them.
10073 Put small global and static data in the small data area, and
10074 generate special instructions to reference them.
10077 Put global and static objects less than or equal to NUM bytes into
10078 the small data or bss sections instead of the normal data or bss
10079 sections. The default value of NUM is 8. The `-msdata' option
10080 must be set to one of `sdata' or `use' for this option to have any
10083 All modules should be compiled with the same `-G NUM' value.
10084 Compiling with different values of NUM may or may not work; if it
10085 doesn't the linker will give an error message--incorrect code will
10089 Makes the M32R specific code in the compiler display some
10090 statistics that might help in debugging programs.
10093 Align all loops to a 32-byte boundary.
10096 Do not enforce a 32-byte alignment for loops. This is the default.
10098 `-missue-rate=NUMBER'
10099 Issue NUMBER instructions per cycle. NUMBER can only be 1 or 2.
10101 `-mbranch-cost=NUMBER'
10102 NUMBER can only be 1 or 2. If it is 1 then branches will be
10103 preferred over conditional code, if it is 2, then the opposite will
10106 `-mflush-trap=NUMBER'
10107 Specifies the trap number to use to flush the cache. The default
10108 is 12. Valid numbers are between 0 and 15 inclusive.
10111 Specifies that the cache cannot be flushed by using a trap.
10113 `-mflush-func=NAME'
10114 Specifies the name of the operating system function to call to
10115 flush the cache. The default is __flush_cache_, but a function
10116 call will only be used if a trap is not available.
10119 Indicates that there is no OS function for flushing the cache.
10123 File: gcc.info, Node: M680x0 Options, Next: M68hc1x Options, Prev: M32R/D Options, Up: Submodel Options
10125 3.17.18 M680x0 Options
10126 ----------------------
10128 These are the `-m' options defined for the 68000 series. The default
10129 values for these options depends on which style of 68000 was selected
10130 when the compiler was configured; the defaults for the most common
10131 choices are given below.
10135 Generate output for a 68000. This is the default when the
10136 compiler is configured for 68000-based systems.
10138 Use this option for microcontrollers with a 68000 or EC000 core,
10139 including the 68008, 68302, 68306, 68307, 68322, 68328 and 68356.
10143 Generate output for a 68020. This is the default when the
10144 compiler is configured for 68020-based systems.
10147 Generate output containing 68881 instructions for floating point.
10148 This is the default for most 68020 systems unless `--nfp' was
10149 specified when the compiler was configured.
10152 Generate output for a 68030. This is the default when the
10153 compiler is configured for 68030-based systems.
10156 Generate output for a 68040. This is the default when the
10157 compiler is configured for 68040-based systems.
10159 This option inhibits the use of 68881/68882 instructions that have
10160 to be emulated by software on the 68040. Use this option if your
10161 68040 does not have code to emulate those instructions.
10164 Generate output for a 68060. This is the default when the
10165 compiler is configured for 68060-based systems.
10167 This option inhibits the use of 68020 and 68881/68882 instructions
10168 that have to be emulated by software on the 68060. Use this
10169 option if your 68060 does not have code to emulate those
10173 Generate output for a CPU32. This is the default when the
10174 compiler is configured for CPU32-based systems.
10176 Use this option for microcontrollers with a CPU32 or CPU32+ core,
10177 including the 68330, 68331, 68332, 68333, 68334, 68336, 68340,
10178 68341, 68349 and 68360.
10181 Generate output for a 520X "coldfire" family cpu. This is the
10182 default when the compiler is configured for 520X-based systems.
10184 Use this option for microcontroller with a 5200 core, including
10185 the MCF5202, MCF5203, MCF5204 and MCF5202.
10188 Generate output for a ColdFire V4e family cpu (e.g. 547x/548x).
10189 This includes use of hardware floating point instructions.
10192 Generate output for a 68040, without using any of the new
10193 instructions. This results in code which can run relatively
10194 efficiently on either a 68020/68881 or a 68030 or a 68040. The
10195 generated code does use the 68881 instructions that are emulated
10199 Generate output for a 68060, without using any of the new
10200 instructions. This results in code which can run relatively
10201 efficiently on either a 68020/68881 or a 68030 or a 68040. The
10202 generated code does use the 68881 instructions that are emulated
10206 Generate output containing library calls for floating point.
10207 *Warning:* the requisite libraries are not available for all m68k
10208 targets. Normally the facilities of the machine's usual C
10209 compiler are used, but this can't be done directly in
10210 cross-compilation. You must make your own arrangements to provide
10211 suitable library functions for cross-compilation. The embedded
10212 targets `m68k-*-aout' and `m68k-*-coff' do provide software
10213 floating point support.
10216 Consider type `int' to be 16 bits wide, like `short int'.
10217 Additionally, parameters passed on the stack are also aligned to a
10218 16-bit boundary even on targets whose API mandates promotion to
10222 Do not use the bit-field instructions. The `-m68000', `-mcpu32'
10223 and `-m5200' options imply `-mnobitfield'.
10226 Do use the bit-field instructions. The `-m68020' option implies
10227 `-mbitfield'. This is the default if you use a configuration
10228 designed for a 68020.
10231 Use a different function-calling convention, in which functions
10232 that take a fixed number of arguments return with the `rtd'
10233 instruction, which pops their arguments while returning. This
10234 saves one instruction in the caller since there is no need to pop
10235 the arguments there.
10237 This calling convention is incompatible with the one normally used
10238 on Unix, so you cannot use it if you need to call libraries
10239 compiled with the Unix compiler.
10241 Also, you must provide function prototypes for all functions that
10242 take variable numbers of arguments (including `printf'); otherwise
10243 incorrect code will be generated for calls to those functions.
10245 In addition, seriously incorrect code will result if you call a
10246 function with too many arguments. (Normally, extra arguments are
10247 harmlessly ignored.)
10249 The `rtd' instruction is supported by the 68010, 68020, 68030,
10250 68040, 68060 and CPU32 processors, but not by the 68000 or 5200.
10254 Control whether GCC aligns `int', `long', `long long', `float',
10255 `double', and `long double' variables on a 32-bit boundary
10256 (`-malign-int') or a 16-bit boundary (`-mno-align-int'). Aligning
10257 variables on 32-bit boundaries produces code that runs somewhat
10258 faster on processors with 32-bit busses at the expense of more
10261 *Warning:* if you use the `-malign-int' switch, GCC will align
10262 structures containing the above types differently than most
10263 published application binary interface specifications for the m68k.
10266 Use the pc-relative addressing mode of the 68000 directly, instead
10267 of using a global offset table. At present, this option implies
10268 `-fpic', allowing at most a 16-bit offset for pc-relative
10269 addressing. `-fPIC' is not presently supported with `-mpcrel',
10270 though this could be supported for 68020 and higher processors.
10272 `-mno-strict-align'
10274 Do not (do) assume that unaligned memory references will be
10275 handled by the system.
10278 Generate code that allows the data segment to be located in a
10279 different area of memory from the text segment. This allows for
10280 execute in place in an environment without virtual memory
10281 management. This option implies `-fPIC'.
10284 Generate code that assumes that the data segment follows the text
10285 segment. This is the default.
10287 `-mid-shared-library'
10288 Generate code that supports shared libraries via the library ID
10289 method. This allows for execute in place and shared libraries in
10290 an environment without virtual memory management. This option
10293 `-mno-id-shared-library'
10294 Generate code that doesn't assume ID based shared libraries are
10295 being used. This is the default.
10297 `-mshared-library-id=n'
10298 Specified the identification number of the ID based shared library
10299 being compiled. Specifying a value of 0 will generate more
10300 compact code, specifying other values will force the allocation of
10301 that number to the current library but is no more space or time
10302 efficient than omitting this option.
10306 File: gcc.info, Node: M68hc1x Options, Next: MCore Options, Prev: M680x0 Options, Up: Submodel Options
10308 3.17.19 M68hc1x Options
10309 -----------------------
10311 These are the `-m' options defined for the 68hc11 and 68hc12
10312 microcontrollers. The default values for these options depends on
10313 which style of microcontroller was selected when the compiler was
10314 configured; the defaults for the most common choices are given below.
10318 Generate output for a 68HC11. This is the default when the
10319 compiler is configured for 68HC11-based systems.
10323 Generate output for a 68HC12. This is the default when the
10324 compiler is configured for 68HC12-based systems.
10328 Generate output for a 68HCS12.
10331 Enable the use of 68HC12 pre and post auto-increment and
10332 auto-decrement addressing modes.
10336 Enable the use of 68HC12 min and max instructions.
10340 Treat all calls as being far away (near). If calls are assumed to
10341 be far away, the compiler will use the `call' instruction to call
10342 a function and the `rtc' instruction for returning.
10345 Consider type `int' to be 16 bits wide, like `short int'.
10347 `-msoft-reg-count=COUNT'
10348 Specify the number of pseudo-soft registers which are used for the
10349 code generation. The maximum number is 32. Using more pseudo-soft
10350 register may or may not result in better code depending on the
10351 program. The default is 4 for 68HC11 and 2 for 68HC12.
10355 File: gcc.info, Node: MCore Options, Next: MIPS Options, Prev: M68hc1x Options, Up: Submodel Options
10357 3.17.20 MCore Options
10358 ---------------------
10360 These are the `-m' options defined for the Motorola M*Core processors.
10364 Inline constants into the code stream if it can be done in two
10365 instructions or less.
10369 Use the divide instruction. (Enabled by default).
10371 `-mrelax-immediate'
10372 `-mno-relax-immediate'
10373 Allow arbitrary sized immediates in bit operations.
10376 `-mno-wide-bitfields'
10377 Always treat bit-fields as int-sized.
10379 `-m4byte-functions'
10380 `-mno-4byte-functions'
10381 Force all functions to be aligned to a four byte boundary.
10384 `-mno-callgraph-data'
10385 Emit callgraph information.
10389 Prefer word access when reading byte quantities.
10393 Generate code for a little endian target.
10397 Generate code for the 210 processor.
10400 File: gcc.info, Node: MIPS Options, Next: MMIX Options, Prev: MCore Options, Up: Submodel Options
10402 3.17.21 MIPS Options
10403 --------------------
10406 Generate big-endian code.
10409 Generate little-endian code. This is the default for `mips*el-*-*'
10413 Generate code that will run on ARCH, which can be the name of a
10414 generic MIPS ISA, or the name of a particular processor. The ISA
10415 names are: `mips1', `mips2', `mips3', `mips4', `mips32',
10416 `mips32r2', and `mips64'. The processor names are: `4kc', `4km',
10417 `4kp', `5kc', `5kf', `20kc', `24k', `24kc', `24kf', `24kx', `m4k',
10418 `orion', `r2000', `r3000', `r3900', `r4000', `r4400', `r4600',
10419 `r4650', `r6000', `r8000', `rm7000', `rm9000', `sb1', `sr71000',
10420 `vr4100', `vr4111', `vr4120', `vr4130', `vr4300', `vr5000',
10421 `vr5400' and `vr5500'. The special value `from-abi' selects the
10422 most compatible architecture for the selected ABI (that is,
10423 `mips1' for 32-bit ABIs and `mips3' for 64-bit ABIs).
10425 In processor names, a final `000' can be abbreviated as `k' (for
10426 example, `-march=r2k'). Prefixes are optional, and `vr' may be
10429 GCC defines two macros based on the value of this option. The
10430 first is `_MIPS_ARCH', which gives the name of target
10431 architecture, as a string. The second has the form
10432 `_MIPS_ARCH_FOO', where FOO is the capitalized value of
10433 `_MIPS_ARCH'. For example, `-march=r2000' will set `_MIPS_ARCH'
10434 to `"r2000"' and define the macro `_MIPS_ARCH_R2000'.
10436 Note that the `_MIPS_ARCH' macro uses the processor names given
10437 above. In other words, it will have the full prefix and will not
10438 abbreviate `000' as `k'. In the case of `from-abi', the macro
10439 names the resolved architecture (either `"mips1"' or `"mips3"').
10440 It names the default architecture when no `-march' option is given.
10443 Optimize for ARCH. Among other things, this option controls the
10444 way instructions are scheduled, and the perceived cost of
10445 arithmetic operations. The list of ARCH values is the same as for
10448 When this option is not used, GCC will optimize for the processor
10449 specified by `-march'. By using `-march' and `-mtune' together,
10450 it is possible to generate code that will run on a family of
10451 processors, but optimize the code for one particular member of
10454 `-mtune' defines the macros `_MIPS_TUNE' and `_MIPS_TUNE_FOO',
10455 which work in the same way as the `-march' ones described above.
10458 Equivalent to `-march=mips1'.
10461 Equivalent to `-march=mips2'.
10464 Equivalent to `-march=mips3'.
10467 Equivalent to `-march=mips4'.
10470 Equivalent to `-march=mips32'.
10473 Equivalent to `-march=mips32r2'.
10476 Equivalent to `-march=mips64'.
10480 Generate (do not generate) MIPS16 code. If GCC is targetting a
10481 MIPS32 or MIPS64 architecture, it will make use of the MIPS16e ASE.
10488 Generate code for the given ABI.
10490 Note that the EABI has a 32-bit and a 64-bit variant. GCC normally
10491 generates 64-bit code when you select a 64-bit architecture, but
10492 you can use `-mgp32' to get 32-bit code instead.
10494 For information about the O64 ABI, see
10495 `http://gcc.gnu.org/projects/mipso64-abi.html'.
10499 Generate (do not generate) code that is suitable for SVR4-style
10500 dynamic objects. `-mabicalls' is the default for SVR4-based
10505 Generate (do not generate) code that is fully position-independent,
10506 and that can therefore be linked into shared libraries. This
10507 option only affects `-mabicalls'.
10509 All `-mabicalls' code has traditionally been position-independent,
10510 regardless of options like `-fPIC' and `-fpic'. However, as an
10511 extension, the GNU toolchain allows executables to use absolute
10512 accesses for locally-binding symbols. It can also use shorter GP
10513 initialization sequences and generate direct calls to
10514 locally-defined functions. This mode is selected by `-mno-shared'.
10516 `-mno-shared' depends on binutils 2.16 or higher and generates
10517 objects that can only be linked by the GNU linker. However, the
10518 option does not affect the ABI of the final executable; it only
10519 affects the ABI of relocatable objects. Using `-mno-shared' will
10520 generally make executables both smaller and quicker.
10522 `-mshared' is the default.
10526 Lift (do not lift) the usual restrictions on the size of the global
10529 GCC normally uses a single instruction to load values from the GOT.
10530 While this is relatively efficient, it will only work if the GOT
10531 is smaller than about 64k. Anything larger will cause the linker
10532 to report an error such as:
10534 relocation truncated to fit: R_MIPS_GOT16 foobar
10536 If this happens, you should recompile your code with `-mxgot'. It
10537 should then work with very large GOTs, although it will also be
10538 less efficient, since it will take three instructions to fetch the
10539 value of a global symbol.
10541 Note that some linkers can create multiple GOTs. If you have such
10542 a linker, you should only need to use `-mxgot' when a single object
10543 file accesses more than 64k's worth of GOT entries. Very few do.
10545 These options have no effect unless GCC is generating position
10549 Assume that general-purpose registers are 32 bits wide.
10552 Assume that general-purpose registers are 64 bits wide.
10555 Assume that floating-point registers are 32 bits wide.
10558 Assume that floating-point registers are 64 bits wide.
10561 Use floating-point coprocessor instructions.
10564 Do not use floating-point coprocessor instructions. Implement
10565 floating-point calculations using library calls instead.
10568 Assume that the floating-point coprocessor only supports
10569 single-precision operations.
10572 Assume that the floating-point coprocessor supports
10573 double-precision operations. This is the default.
10577 Use (do not use) the MIPS DSP ASE. *Note MIPS DSP Built-in
10581 `-mno-paired-single'
10582 Use (do not use) paired-single floating-point instructions. *Note
10583 MIPS Paired-Single Support::. This option can only be used when
10584 generating 64-bit code and requires hardware floating-point
10585 support to be enabled.
10589 Use (do not use) the MIPS-3D ASE. *Note MIPS-3D Built-in
10590 Functions::. The option `-mips3d' implies `-mpaired-single'.
10593 Force `long' types to be 64 bits wide. See `-mlong32' for an
10594 explanation of the default and the way that the pointer size is
10598 Force `long', `int', and pointer types to be 32 bits wide.
10600 The default size of `int's, `long's and pointers depends on the
10601 ABI. All the supported ABIs use 32-bit `int's. The n64 ABI uses
10602 64-bit `long's, as does the 64-bit EABI; the others use 32-bit
10603 `long's. Pointers are the same size as `long's, or the same size
10604 as integer registers, whichever is smaller.
10608 Assume (do not assume) that all symbols have 32-bit values,
10609 regardless of the selected ABI. This option is useful in
10610 combination with `-mabi=64' and `-mno-abicalls' because it allows
10611 GCC to generate shorter and faster references to symbolic
10615 Put global and static items less than or equal to NUM bytes into
10616 the small data or bss section instead of the normal data or bss
10617 section. This allows the data to be accessed using a single
10620 All modules should be compiled with the same `-G NUM' value.
10623 `-mno-embedded-data'
10624 Allocate variables to the read-only data section first if
10625 possible, then next in the small data section if possible,
10626 otherwise in data. This gives slightly slower code than the
10627 default, but reduces the amount of RAM required when executing,
10628 and thus may be preferred for some embedded systems.
10630 `-muninit-const-in-rodata'
10631 `-mno-uninit-const-in-rodata'
10632 Put uninitialized `const' variables in the read-only data section.
10633 This option is only meaningful in conjunction with
10636 `-msplit-addresses'
10637 `-mno-split-addresses'
10638 Enable (disable) use of the `%hi()' and `%lo()' assembler
10639 relocation operators. This option has been superseded by
10640 `-mexplicit-relocs' but is retained for backwards compatibility.
10642 `-mexplicit-relocs'
10643 `-mno-explicit-relocs'
10644 Use (do not use) assembler relocation operators when dealing with
10645 symbolic addresses. The alternative, selected by
10646 `-mno-explicit-relocs', is to use assembler macros instead.
10648 `-mexplicit-relocs' is the default if GCC was configured to use an
10649 assembler that supports relocation operators.
10651 `-mcheck-zero-division'
10652 `-mno-check-zero-division'
10653 Trap (do not trap) on integer division by zero. The default is
10654 `-mcheck-zero-division'.
10658 MIPS systems check for division by zero by generating either a
10659 conditional trap or a break instruction. Using traps results in
10660 smaller code, but is only supported on MIPS II and later. Also,
10661 some versions of the Linux kernel have a bug that prevents trap
10662 from generating the proper signal (`SIGFPE'). Use
10663 `-mdivide-traps' to allow conditional traps on architectures that
10664 support them and `-mdivide-breaks' to force the use of breaks.
10666 The default is usually `-mdivide-traps', but this can be
10667 overridden at configure time using `--with-divide=breaks'.
10668 Divide-by-zero checks can be completely disabled using
10669 `-mno-check-zero-division'.
10673 Force (do not force) the use of `memcpy()' for non-trivial block
10674 moves. The default is `-mno-memcpy', which allows GCC to inline
10675 most constant-sized copies.
10679 Disable (do not disable) use of the `jal' instruction. Calling
10680 functions using `jal' is more efficient but requires the caller
10681 and callee to be in the same 256 megabyte segment.
10683 This option has no effect on abicalls code. The default is
10688 Enable (disable) use of the `mad', `madu' and `mul' instructions,
10689 as provided by the R4650 ISA.
10693 Enable (disable) use of the floating point multiply-accumulate
10694 instructions, when they are available. The default is
10697 When multiply-accumulate instructions are used, the intermediate
10698 product is calculated to infinite precision and is not subject to
10699 the FCSR Flush to Zero bit. This may be undesirable in some
10703 Tell the MIPS assembler to not run its preprocessor over user
10704 assembler files (with a `.s' suffix) when assembling them.
10708 Work around certain R4000 CPU errata:
10709 - A double-word or a variable shift may give an incorrect
10710 result if executed immediately after starting an integer
10713 - A double-word or a variable shift may give an incorrect
10714 result if executed while an integer multiplication is in
10717 - An integer division may give an incorrect result if started
10718 in a delay slot of a taken branch or a jump.
10722 Work around certain R4400 CPU errata:
10723 - A double-word or a variable shift may give an incorrect
10724 result if executed immediately after starting an integer
10729 Work around certain VR4120 errata:
10730 - `dmultu' does not always produce the correct result.
10732 - `div' and `ddiv' do not always produce the correct result if
10733 one of the operands is negative.
10734 The workarounds for the division errata rely on special functions
10735 in `libgcc.a'. At present, these functions are only provided by
10736 the `mips64vr*-elf' configurations.
10738 Other VR4120 errata require a nop to be inserted between certain
10739 pairs of instructions. These errata are handled by the assembler,
10743 Work around the VR4130 `mflo'/`mfhi' errata. The workarounds are
10744 implemented by the assembler rather than by GCC, although GCC will
10745 avoid using `mflo' and `mfhi' if the VR4130 `macc', `macchi',
10746 `dmacc' and `dmacchi' instructions are available instead.
10750 Work around certain SB-1 CPU core errata. (This flag currently
10751 works around the SB-1 revision 2 "F1" and "F2" floating point
10754 `-mflush-func=FUNC'
10756 Specifies the function to call to flush the I and D caches, or to
10757 not call any such function. If called, the function must take the
10758 same arguments as the common `_flush_func()', that is, the address
10759 of the memory range for which the cache is being flushed, the size
10760 of the memory range, and the number 3 (to flush both caches). The
10761 default depends on the target GCC was configured for, but commonly
10762 is either `_flush_func' or `__cpu_flush'.
10765 `-mno-branch-likely'
10766 Enable or disable use of Branch Likely instructions, regardless of
10767 the default for the selected architecture. By default, Branch
10768 Likely instructions may be generated if they are supported by the
10769 selected architecture. An exception is for the MIPS32 and MIPS64
10770 architectures and processors which implement those architectures;
10771 for those, Branch Likely instructions will not be generated by
10772 default because the MIPS32 and MIPS64 architectures specifically
10773 deprecate their use.
10776 `-mno-fp-exceptions'
10777 Specifies whether FP exceptions are enabled. This affects how we
10778 schedule FP instructions for some processors. The default is that
10779 FP exceptions are enabled.
10781 For instance, on the SB-1, if FP exceptions are disabled, and we
10782 are emitting 64-bit code, then we can use both FP pipes.
10783 Otherwise, we can only use one FP pipe.
10786 `-mno-vr4130-align'
10787 The VR4130 pipeline is two-way superscalar, but can only issue two
10788 instructions together if the first one is 8-byte aligned. When
10789 this option is enabled, GCC will align pairs of instructions that
10790 it thinks should execute in parallel.
10792 This option only has an effect when optimizing for the VR4130. It
10793 normally makes code faster, but at the expense of making it bigger.
10794 It is enabled by default at optimization level `-O3'.
10797 File: gcc.info, Node: MMIX Options, Next: MN10300 Options, Prev: MIPS Options, Up: Submodel Options
10799 3.17.22 MMIX Options
10800 --------------------
10802 These options are defined for the MMIX:
10806 Specify that intrinsic library functions are being compiled,
10807 passing all values in registers, no matter the size.
10811 Generate floating-point comparison instructions that compare with
10812 respect to the `rE' epsilon register.
10816 Generate code that passes function parameters and return values
10817 that (in the called function) are seen as registers `$0' and up,
10818 as opposed to the GNU ABI which uses global registers `$231' and
10823 When reading data from memory in sizes shorter than 64 bits, use
10824 (do not use) zero-extending load instructions by default, rather
10825 than sign-extending ones.
10829 Make the result of a division yielding a remainder have the same
10830 sign as the divisor. With the default, `-mno-knuthdiv', the sign
10831 of the remainder follows the sign of the dividend. Both methods
10832 are arithmetically valid, the latter being almost exclusively used.
10834 `-mtoplevel-symbols'
10835 `-mno-toplevel-symbols'
10836 Prepend (do not prepend) a `:' to all global symbols, so the
10837 assembly code can be used with the `PREFIX' assembly directive.
10840 Generate an executable in the ELF format, rather than the default
10841 `mmo' format used by the `mmix' simulator.
10844 `-mno-branch-predict'
10845 Use (do not use) the probable-branch instructions, when static
10846 branch prediction indicates a probable branch.
10849 `-mno-base-addresses'
10850 Generate (do not generate) code that uses _base addresses_. Using
10851 a base address automatically generates a request (handled by the
10852 assembler and the linker) for a constant to be set up in a global
10853 register. The register is used for one or more base address
10854 requests within the range 0 to 255 from the value held in the
10855 register. The generally leads to short and fast code, but the
10856 number of different data items that can be addressed is limited.
10857 This means that a program that uses lots of static data may
10858 require `-mno-base-addresses'.
10862 Force (do not force) generated code to have a single exit point in
10866 File: gcc.info, Node: MN10300 Options, Next: MT Options, Prev: MMIX Options, Up: Submodel Options
10868 3.17.23 MN10300 Options
10869 -----------------------
10871 These `-m' options are defined for Matsushita MN10300 architectures:
10874 Generate code to avoid bugs in the multiply instructions for the
10875 MN10300 processors. This is the default.
10878 Do not generate code to avoid bugs in the multiply instructions
10879 for the MN10300 processors.
10882 Generate code which uses features specific to the AM33 processor.
10885 Do not generate code which uses features specific to the AM33
10886 processor. This is the default.
10888 `-mreturn-pointer-on-d0'
10889 When generating a function which returns a pointer, return the
10890 pointer in both `a0' and `d0'. Otherwise, the pointer is returned
10891 only in a0, and attempts to call such functions without a prototype
10892 would result in errors. Note that this option is on by default;
10893 use `-mno-return-pointer-on-d0' to disable it.
10896 Do not link in the C run-time initialization object file.
10899 Indicate to the linker that it should perform a relaxation
10900 optimization pass to shorten branches, calls and absolute memory
10901 addresses. This option only has an effect when used on the
10902 command line for the final link step.
10904 This option makes symbolic debugging impossible.
10907 File: gcc.info, Node: MT Options, Next: PDP-11 Options, Prev: MN10300 Options, Up: Submodel Options
10912 These `-m' options are defined for Morpho MT architectures:
10915 Generate code that will run on CPU-TYPE, which is the name of a
10916 system representing a certain processor type. Possible values for
10917 CPU-TYPE are `ms1-64-001', `ms1-16-002', `ms1-16-003' and `ms2'.
10919 When this option is not used, the default is `-march=ms1-16-002'.
10922 Use byte loads and stores when generating code.
10925 Do not use byte loads and stores when generating code.
10928 Use simulator runtime
10931 Do not link in the C run-time initialization object file `crti.o'.
10932 Other run-time initialization and termination files such as
10933 `startup.o' and `exit.o' are still included on the linker command
10938 File: gcc.info, Node: PDP-11 Options, Next: PowerPC Options, Prev: MT Options, Up: Submodel Options
10940 3.17.25 PDP-11 Options
10941 ----------------------
10943 These options are defined for the PDP-11:
10946 Use hardware FPP floating point. This is the default. (FIS
10947 floating point on the PDP-11/40 is not supported.)
10950 Do not use hardware floating point.
10953 Return floating-point results in ac0 (fr0 in Unix assembler
10957 Return floating-point results in memory. This is the default.
10960 Generate code for a PDP-11/40.
10963 Generate code for a PDP-11/45. This is the default.
10966 Generate code for a PDP-11/10.
10969 Use inline `movmemhi' patterns for copying memory. This is the
10973 Do not use inline `movmemhi' patterns for copying memory.
10977 Use 16-bit `int'. This is the default.
10985 Use 64-bit `float'. This is the default.
10989 Use 32-bit `float'.
10992 Use `abshi2' pattern. This is the default.
10995 Do not use `abshi2' pattern.
10997 `-mbranch-expensive'
10998 Pretend that branches are expensive. This is for experimenting
10999 with code generation only.
11002 Do not pretend that branches are expensive. This is the default.
11005 Generate code for a system with split I&D.
11008 Generate code for a system without split I&D. This is the default.
11011 Use Unix assembler syntax. This is the default when configured for
11015 Use DEC assembler syntax. This is the default when configured for
11016 any PDP-11 target other than `pdp11-*-bsd'.
11019 File: gcc.info, Node: PowerPC Options, Next: RS/6000 and PowerPC Options, Prev: PDP-11 Options, Up: Submodel Options
11021 3.17.26 PowerPC Options
11022 -----------------------
11024 These are listed under *Note RS/6000 and PowerPC Options::.
11027 File: gcc.info, Node: RS/6000 and PowerPC Options, Next: S/390 and zSeries Options, Prev: PowerPC Options, Up: Submodel Options
11029 3.17.27 IBM RS/6000 and PowerPC Options
11030 ---------------------------------------
11032 These `-m' options are defined for the IBM RS/6000 and PowerPC:
11040 `-mno-powerpc-gpopt'
11042 `-mno-powerpc-gfxopt'
11051 GCC supports two related instruction set architectures for the
11052 RS/6000 and PowerPC. The "POWER" instruction set are those
11053 instructions supported by the `rios' chip set used in the original
11054 RS/6000 systems and the "PowerPC" instruction set is the
11055 architecture of the Freescale MPC5xx, MPC6xx, MPC8xx
11056 microprocessors, and the IBM 4xx, 6xx, and follow-on
11059 Neither architecture is a subset of the other. However there is a
11060 large common subset of instructions supported by both. An MQ
11061 register is included in processors supporting the POWER
11064 You use these options to specify which instructions are available
11065 on the processor you are using. The default value of these
11066 options is determined when configuring GCC. Specifying the
11067 `-mcpu=CPU_TYPE' overrides the specification of these options. We
11068 recommend you use the `-mcpu=CPU_TYPE' option rather than the
11069 options listed above.
11071 The `-mpower' option allows GCC to generate instructions that are
11072 found only in the POWER architecture and to use the MQ register.
11073 Specifying `-mpower2' implies `-power' and also allows GCC to
11074 generate instructions that are present in the POWER2 architecture
11075 but not the original POWER architecture.
11077 The `-mpowerpc' option allows GCC to generate instructions that
11078 are found only in the 32-bit subset of the PowerPC architecture.
11079 Specifying `-mpowerpc-gpopt' implies `-mpowerpc' and also allows
11080 GCC to use the optional PowerPC architecture instructions in the
11081 General Purpose group, including floating-point square root.
11082 Specifying `-mpowerpc-gfxopt' implies `-mpowerpc' and also allows
11083 GCC to use the optional PowerPC architecture instructions in the
11084 Graphics group, including floating-point select.
11086 The `-mmfcrf' option allows GCC to generate the move from
11087 condition register field instruction implemented on the POWER4
11088 processor and other processors that support the PowerPC V2.01
11089 architecture. The `-mpopcntb' option allows GCC to generate the
11090 popcount and double precision FP reciprocal estimate instruction
11091 implemented on the POWER5 processor and other processors that
11092 support the PowerPC V2.02 architecture. The `-mfprnd' option
11093 allows GCC to generate the FP round to integer instructions
11094 implemented on the POWER5+ processor and other processors that
11095 support the PowerPC V2.03 architecture.
11097 The `-mpowerpc64' option allows GCC to generate the additional
11098 64-bit instructions that are found in the full PowerPC64
11099 architecture and to treat GPRs as 64-bit, doubleword quantities.
11100 GCC defaults to `-mno-powerpc64'.
11102 If you specify both `-mno-power' and `-mno-powerpc', GCC will use
11103 only the instructions in the common subset of both architectures
11104 plus some special AIX common-mode calls, and will not use the MQ
11105 register. Specifying both `-mpower' and `-mpowerpc' permits GCC
11106 to use any instruction from either architecture and to allow use
11107 of the MQ register; specify this for the Motorola MPC601.
11111 Select which mnemonics to use in the generated assembler code.
11112 With `-mnew-mnemonics', GCC uses the assembler mnemonics defined
11113 for the PowerPC architecture. With `-mold-mnemonics' it uses the
11114 assembler mnemonics defined for the POWER architecture.
11115 Instructions defined in only one architecture have only one
11116 mnemonic; GCC uses that mnemonic irrespective of which of these
11117 options is specified.
11119 GCC defaults to the mnemonics appropriate for the architecture in
11120 use. Specifying `-mcpu=CPU_TYPE' sometimes overrides the value of
11121 these option. Unless you are building a cross-compiler, you
11122 should normally not specify either `-mnew-mnemonics' or
11123 `-mold-mnemonics', but should instead accept the default.
11126 Set architecture type, register usage, choice of mnemonics, and
11127 instruction scheduling parameters for machine type CPU_TYPE.
11128 Supported values for CPU_TYPE are `401', `403', `405', `405fp',
11129 `440', `440fp', `505', `601', `602', `603', `603e', `604', `604e',
11130 `620', `630', `740', `7400', `7450', `750', `801', `821', `823',
11131 `860', `970', `8540', `ec603e', `G3', `G4', `G5', `power',
11132 `power2', `power3', `power4', `power5', `power5+', `power6',
11133 `common', `powerpc', `powerpc64', `rios', `rios1', `rios2', `rsc',
11136 `-mcpu=common' selects a completely generic processor. Code
11137 generated under this option will run on any POWER or PowerPC
11138 processor. GCC will use only the instructions in the common
11139 subset of both architectures, and will not use the MQ register.
11140 GCC assumes a generic processor model for scheduling purposes.
11142 `-mcpu=power', `-mcpu=power2', `-mcpu=powerpc', and
11143 `-mcpu=powerpc64' specify generic POWER, POWER2, pure 32-bit
11144 PowerPC (i.e., not MPC601), and 64-bit PowerPC architecture machine
11145 types, with an appropriate, generic processor model assumed for
11146 scheduling purposes.
11148 The other options specify a specific processor. Code generated
11149 under those options will run best on that processor, and may not
11150 run at all on others.
11152 The `-mcpu' options automatically enable or disable the following
11153 options: `-maltivec', `-mfprnd', `-mhard-float', `-mmfcrf',
11154 `-mmultiple', `-mnew-mnemonics', `-mpopcntb', `-mpower',
11155 `-mpower2', `-mpowerpc64', `-mpowerpc-gpopt', `-mpowerpc-gfxopt',
11156 `-mstring', `-mmulhw', `-mdlmzb'. The particular options set for
11157 any particular CPU will vary between compiler versions, depending
11158 on what setting seems to produce optimal code for that CPU; it
11159 doesn't necessarily reflect the actual hardware's capabilities. If
11160 you wish to set an individual option to a particular value, you may
11161 specify it after the `-mcpu' option, like `-mcpu=970 -mno-altivec'.
11163 On AIX, the `-maltivec' and `-mpowerpc64' options are not enabled
11164 or disabled by the `-mcpu' option at present because AIX does not
11165 have full support for these options. You may still enable or
11166 disable them individually if you're sure it'll work in your
11170 Set the instruction scheduling parameters for machine type
11171 CPU_TYPE, but do not set the architecture type, register usage, or
11172 choice of mnemonics, as `-mcpu=CPU_TYPE' would. The same values
11173 for CPU_TYPE are used for `-mtune' as for `-mcpu'. If both are
11174 specified, the code generated will use the architecture,
11175 registers, and mnemonics set by `-mcpu', but the scheduling
11176 parameters set by `-mtune'.
11180 Generate code to compute division as reciprocal estimate and
11181 iterative refinement, creating opportunities for increased
11182 throughput. This feature requires: optional PowerPC Graphics
11183 instruction set for single precision and FRE instruction for
11184 double precision, assuming divides cannot generate user-visible
11185 traps, and the domain values not include Infinities, denormals or
11190 Generate code that uses (does not use) AltiVec instructions, and
11191 also enable the use of built-in functions that allow more direct
11192 access to the AltiVec instruction set. You may also need to set
11193 `-mabi=altivec' to adjust the current ABI with AltiVec ABI
11199 Generate VRSAVE instructions when generating AltiVec code.
11202 Generate code that allows ld and ld.so to build executables and
11203 shared libraries with non-exec .plt and .got sections. This is a
11204 PowerPC 32-bit SYSV ABI option.
11207 Generate code that uses a BSS .plt section that ld.so fills in, and
11208 requires .plt and .got sections that are both writable and
11209 executable. This is a PowerPC 32-bit SYSV ABI option.
11213 This switch enables or disables the generation of ISEL
11217 This switch has been deprecated. Use `-misel' and `-mno-isel'
11222 This switch enables or disables the generation of SPE simd
11226 This option has been deprecated. Use `-mspe' and `-mno-spe'
11229 `-mfloat-gprs=YES/SINGLE/DOUBLE/NO'
11231 This switch enables or disables the generation of floating point
11232 operations on the general purpose registers for architectures that
11235 The argument YES or SINGLE enables the use of single-precision
11236 floating point operations.
11238 The argument DOUBLE enables the use of single and double-precision
11239 floating point operations.
11241 The argument NO disables floating point operations on the general
11244 This option is currently only available on the MPC854x.
11248 Generate code for 32-bit or 64-bit environments of Darwin and SVR4
11249 targets (including GNU/Linux). The 32-bit environment sets int,
11250 long and pointer to 32 bits and generates code that runs on any
11251 PowerPC variant. The 64-bit environment sets int to 32 bits and
11252 long and pointer to 64 bits, and generates code for PowerPC64, as
11259 Modify generation of the TOC (Table Of Contents), which is created
11260 for every executable file. The `-mfull-toc' option is selected by
11261 default. In that case, GCC will allocate at least one TOC entry
11262 for each unique non-automatic variable reference in your program.
11263 GCC will also place floating-point constants in the TOC. However,
11264 only 16,384 entries are available in the TOC.
11266 If you receive a linker error message that saying you have
11267 overflowed the available TOC space, you can reduce the amount of
11268 TOC space used with the `-mno-fp-in-toc' and `-mno-sum-in-toc'
11269 options. `-mno-fp-in-toc' prevents GCC from putting floating-point
11270 constants in the TOC and `-mno-sum-in-toc' forces GCC to generate
11271 code to calculate the sum of an address and a constant at run-time
11272 instead of putting that sum into the TOC. You may specify one or
11273 both of these options. Each causes GCC to produce very slightly
11274 slower and larger code at the expense of conserving TOC space.
11276 If you still run out of space in the TOC even when you specify
11277 both of these options, specify `-mminimal-toc' instead. This
11278 option causes GCC to make only one TOC entry for every file. When
11279 you specify this option, GCC will produce code that is slower and
11280 larger but which uses extremely little TOC space. You may wish to
11281 use this option only on files that contain less frequently
11286 Enable 64-bit AIX ABI and calling convention: 64-bit pointers,
11287 64-bit `long' type, and the infrastructure needed to support them.
11288 Specifying `-maix64' implies `-mpowerpc64' and `-mpowerpc', while
11289 `-maix32' disables the 64-bit ABI and implies `-mno-powerpc64'.
11290 GCC defaults to `-maix32'.
11294 Produce code that conforms more closely to IBM XL compiler
11295 semantics when using AIX-compatible ABI. Pass floating-point
11296 arguments to prototyped functions beyond the register save area
11297 (RSA) on the stack in addition to argument FPRs. Do not assume
11298 that most significant double in 128-bit long double value is
11299 properly rounded when comparing values and converting to double.
11300 Use XL symbol names for long double support routines.
11302 The AIX calling convention was extended but not initially
11303 documented to handle an obscure K&R C case of calling a function
11304 that takes the address of its arguments with fewer arguments than
11305 declared. IBM XL compilers access floating point arguments which
11306 do not fit in the RSA from the stack when a subroutine is compiled
11307 without optimization. Because always storing floating-point
11308 arguments on the stack is inefficient and rarely needed, this
11309 option is not enabled by default and only is necessary when
11310 calling subroutines compiled by IBM XL compilers without
11314 Support "IBM RS/6000 SP" "Parallel Environment" (PE). Link an
11315 application written to use message passing with special startup
11316 code to enable the application to run. The system must have PE
11317 installed in the standard location (`/usr/lpp/ppe.poe/'), or the
11318 `specs' file must be overridden with the `-specs=' option to
11319 specify the appropriate directory location. The Parallel
11320 Environment does not support threads, so the `-mpe' option and the
11321 `-pthread' option are incompatible.
11325 On AIX, 32-bit Darwin, and 64-bit PowerPC GNU/Linux, the option
11326 `-malign-natural' overrides the ABI-defined alignment of larger
11327 types, such as floating-point doubles, on their natural size-based
11328 boundary. The option `-malign-power' instructs GCC to follow the
11329 ABI-specified alignment rules. GCC defaults to the standard
11330 alignment defined in the ABI.
11332 On 64-bit Darwin, natural alignment is the default, and
11333 `-malign-power' is not supported.
11337 Generate code that does not use (uses) the floating-point register
11338 set. Software floating point emulation is provided if you use the
11339 `-msoft-float' option, and pass the option to GCC when linking.
11343 Generate code that uses (does not use) the load multiple word
11344 instructions and the store multiple word instructions. These
11345 instructions are generated by default on POWER systems, and not
11346 generated on PowerPC systems. Do not use `-mmultiple' on little
11347 endian PowerPC systems, since those instructions do not work when
11348 the processor is in little endian mode. The exceptions are PPC740
11349 and PPC750 which permit the instructions usage in little endian
11354 Generate code that uses (does not use) the load string instructions
11355 and the store string word instructions to save multiple registers
11356 and do small block moves. These instructions are generated by
11357 default on POWER systems, and not generated on PowerPC systems.
11358 Do not use `-mstring' on little endian PowerPC systems, since those
11359 instructions do not work when the processor is in little endian
11360 mode. The exceptions are PPC740 and PPC750 which permit the
11361 instructions usage in little endian mode.
11365 Generate code that uses (does not use) the load or store
11366 instructions that update the base register to the address of the
11367 calculated memory location. These instructions are generated by
11368 default. If you use `-mno-update', there is a small window
11369 between the time that the stack pointer is updated and the address
11370 of the previous frame is stored, which means code that walks the
11371 stack frame across interrupts or signals may get corrupted data.
11375 Generate code that uses (does not use) the floating point multiply
11376 and accumulate instructions. These instructions are generated by
11377 default if hardware floating is used.
11381 Generate code that uses (does not use) the half-word multiply and
11382 multiply-accumulate instructions on the IBM 405 and 440 processors.
11383 These instructions are generated by default when targetting those
11388 Generate code that uses (does not use) the string-search `dlmzb'
11389 instruction on the IBM 405 and 440 processors. This instruction is
11390 generated by default when targetting those processors.
11394 On System V.4 and embedded PowerPC systems do not (do) force
11395 structures and unions that contain bit-fields to be aligned to the
11396 base type of the bit-field.
11398 For example, by default a structure containing nothing but 8
11399 `unsigned' bit-fields of length 1 would be aligned to a 4 byte
11400 boundary and have a size of 4 bytes. By using `-mno-bit-align',
11401 the structure would be aligned to a 1 byte boundary and be one
11404 `-mno-strict-align'
11406 On System V.4 and embedded PowerPC systems do not (do) assume that
11407 unaligned memory references will be handled by the system.
11411 On embedded PowerPC systems generate code that allows (does not
11412 allow) the program to be relocated to a different address at
11413 runtime. If you use `-mrelocatable' on any module, all objects
11414 linked together must be compiled with `-mrelocatable' or
11415 `-mrelocatable-lib'.
11417 `-mrelocatable-lib'
11418 `-mno-relocatable-lib'
11419 On embedded PowerPC systems generate code that allows (does not
11420 allow) the program to be relocated to a different address at
11421 runtime. Modules compiled with `-mrelocatable-lib' can be linked
11422 with either modules compiled without `-mrelocatable' and
11423 `-mrelocatable-lib' or with modules compiled with the
11424 `-mrelocatable' options.
11428 On System V.4 and embedded PowerPC systems do not (do) assume that
11429 register 2 contains a pointer to a global area pointing to the
11430 addresses used in the program.
11434 On System V.4 and embedded PowerPC systems compile code for the
11435 processor in little endian mode. The `-mlittle-endian' option is
11436 the same as `-mlittle'.
11440 On System V.4 and embedded PowerPC systems compile code for the
11441 processor in big endian mode. The `-mbig-endian' option is the
11445 On Darwin and Mac OS X systems, compile code so that it is not
11446 relocatable, but that its external references are relocatable. The
11447 resulting code is suitable for applications, but not shared
11450 `-mprioritize-restricted-insns=PRIORITY'
11451 This option controls the priority that is assigned to
11452 dispatch-slot restricted instructions during the second scheduling
11453 pass. The argument PRIORITY takes the value 0/1/2 to assign
11454 NO/HIGHEST/SECOND-HIGHEST priority to dispatch slot restricted
11457 `-msched-costly-dep=DEPENDENCE_TYPE'
11458 This option controls which dependences are considered costly by
11459 the target during instruction scheduling. The argument
11460 DEPENDENCE_TYPE takes one of the following values: NO: no
11461 dependence is costly, ALL: all dependences are costly,
11462 TRUE_STORE_TO_LOAD: a true dependence from store to load is costly,
11463 STORE_TO_LOAD: any dependence from store to load is costly,
11464 NUMBER: any dependence which latency >= NUMBER is costly.
11466 `-minsert-sched-nops=SCHEME'
11467 This option controls which nop insertion scheme will be used during
11468 the second scheduling pass. The argument SCHEME takes one of the
11469 following values: NO: Don't insert nops. PAD: Pad with nops any
11470 dispatch group which has vacant issue slots, according to the
11471 scheduler's grouping. REGROUP_EXACT: Insert nops to force costly
11472 dependent insns into separate groups. Insert exactly as many nops
11473 as needed to force an insn to a new group, according to the
11474 estimated processor grouping. NUMBER: Insert nops to force costly
11475 dependent insns into separate groups. Insert NUMBER nops to force
11476 an insn to a new group.
11479 On System V.4 and embedded PowerPC systems compile code using
11480 calling conventions that adheres to the March 1995 draft of the
11481 System V Application Binary Interface, PowerPC processor
11482 supplement. This is the default unless you configured GCC using
11483 `powerpc-*-eabiaix'.
11486 Specify both `-mcall-sysv' and `-meabi' options.
11488 `-mcall-sysv-noeabi'
11489 Specify both `-mcall-sysv' and `-mno-eabi' options.
11492 On System V.4 and embedded PowerPC systems compile code for the
11493 Solaris operating system.
11496 On System V.4 and embedded PowerPC systems compile code for the
11497 Linux-based GNU system.
11500 On System V.4 and embedded PowerPC systems compile code for the
11501 Hurd-based GNU system.
11504 On System V.4 and embedded PowerPC systems compile code for the
11505 NetBSD operating system.
11507 `-maix-struct-return'
11508 Return all structures in memory (as specified by the AIX ABI).
11510 `-msvr4-struct-return'
11511 Return structures smaller than 8 bytes in registers (as specified
11515 Extend the current ABI with a particular extension, or remove such
11516 extension. Valid values are ALTIVEC, NO-ALTIVEC, SPE, NO-SPE,
11517 IBMLONGDOUBLE, IEEELONGDOUBLE.
11520 Extend the current ABI with SPE ABI extensions. This does not
11521 change the default ABI, instead it adds the SPE ABI extensions to
11525 Disable Booke SPE ABI extensions for the current ABI.
11527 `-mabi=ibmlongdouble'
11528 Change the current ABI to use IBM extended precision long double.
11529 This is a PowerPC 32-bit SYSV ABI option.
11531 `-mabi=ieeelongdouble'
11532 Change the current ABI to use IEEE extended precision long double.
11533 This is a PowerPC 32-bit Linux ABI option.
11537 On System V.4 and embedded PowerPC systems assume that all calls to
11538 variable argument functions are properly prototyped. Otherwise,
11539 the compiler must insert an instruction before every non
11540 prototyped call to set or clear bit 6 of the condition code
11541 register (CR) to indicate whether floating point values were
11542 passed in the floating point registers in case the function takes
11543 a variable arguments. With `-mprototype', only calls to
11544 prototyped variable argument functions will set or clear the bit.
11547 On embedded PowerPC systems, assume that the startup module is
11548 called `sim-crt0.o' and that the standard C libraries are
11549 `libsim.a' and `libc.a'. This is the default for
11550 `powerpc-*-eabisim'. configurations.
11553 On embedded PowerPC systems, assume that the startup module is
11554 called `crt0.o' and the standard C libraries are `libmvme.a' and
11558 On embedded PowerPC systems, assume that the startup module is
11559 called `crt0.o' and the standard C libraries are `libads.a' and
11563 On embedded PowerPC systems, assume that the startup module is
11564 called `crt0.o' and the standard C libraries are `libyk.a' and
11568 On System V.4 and embedded PowerPC systems, specify that you are
11569 compiling for a VxWorks system.
11572 Specify that you are compiling for the WindISS simulation
11576 On embedded PowerPC systems, set the PPC_EMB bit in the ELF flags
11577 header to indicate that `eabi' extended relocations are used.
11581 On System V.4 and embedded PowerPC systems do (do not) adhere to
11582 the Embedded Applications Binary Interface (eabi) which is a set of
11583 modifications to the System V.4 specifications. Selecting `-meabi'
11584 means that the stack is aligned to an 8 byte boundary, a function
11585 `__eabi' is called to from `main' to set up the eabi environment,
11586 and the `-msdata' option can use both `r2' and `r13' to point to
11587 two separate small data areas. Selecting `-mno-eabi' means that
11588 the stack is aligned to a 16 byte boundary, do not call an
11589 initialization function from `main', and the `-msdata' option will
11590 only use `r13' to point to a single small data area. The `-meabi'
11591 option is on by default if you configured GCC using one of the
11592 `powerpc*-*-eabi*' options.
11595 On System V.4 and embedded PowerPC systems, put small initialized
11596 `const' global and static data in the `.sdata2' section, which is
11597 pointed to by register `r2'. Put small initialized non-`const'
11598 global and static data in the `.sdata' section, which is pointed
11599 to by register `r13'. Put small uninitialized global and static
11600 data in the `.sbss' section, which is adjacent to the `.sdata'
11601 section. The `-msdata=eabi' option is incompatible with the
11602 `-mrelocatable' option. The `-msdata=eabi' option also sets the
11606 On System V.4 and embedded PowerPC systems, put small global and
11607 static data in the `.sdata' section, which is pointed to by
11608 register `r13'. Put small uninitialized global and static data in
11609 the `.sbss' section, which is adjacent to the `.sdata' section.
11610 The `-msdata=sysv' option is incompatible with the `-mrelocatable'
11615 On System V.4 and embedded PowerPC systems, if `-meabi' is used,
11616 compile code the same as `-msdata=eabi', otherwise compile code the
11617 same as `-msdata=sysv'.
11620 On System V.4 and embedded PowerPC systems, put small global data
11621 in the `.sdata' section. Put small uninitialized global data in
11622 the `.sbss' section. Do not use register `r13' to address small
11623 data however. This is the default behavior unless other `-msdata'
11628 On embedded PowerPC systems, put all initialized global and static
11629 data in the `.data' section, and all uninitialized data in the
11633 On embedded PowerPC systems, put global and static items less than
11634 or equal to NUM bytes into the small data or bss sections instead
11635 of the normal data or bss section. By default, NUM is 8. The `-G
11636 NUM' switch is also passed to the linker. All modules should be
11637 compiled with the same `-G NUM' value.
11641 On System V.4 and embedded PowerPC systems do (do not) emit
11642 register names in the assembly language output using symbolic
11647 By default assume that all calls are far away so that a longer more
11648 expensive calling sequence is required. This is required for calls
11649 further than 32 megabytes (33,554,432 bytes) from the current
11650 location. A short call will be generated if the compiler knows
11651 the call cannot be that far away. This setting can be overridden
11652 by the `shortcall' function attribute, or by `#pragma longcall(0)'.
11654 Some linkers are capable of detecting out-of-range calls and
11655 generating glue code on the fly. On these systems, long calls are
11656 unnecessary and generate slower code. As of this writing, the AIX
11657 linker can do this, as can the GNU linker for PowerPC/64. It is
11658 planned to add this feature to the GNU linker for 32-bit PowerPC
11661 On Darwin/PPC systems, `#pragma longcall' will generate "jbsr
11662 callee, L42", plus a "branch island" (glue code). The two target
11663 addresses represent the callee and the "branch island". The
11664 Darwin/PPC linker will prefer the first address and generate a "bl
11665 callee" if the PPC "bl" instruction will reach the callee directly;
11666 otherwise, the linker will generate "bl L42" to call the "branch
11667 island". The "branch island" is appended to the body of the
11668 calling function; it computes the full 32-bit address of the callee
11671 On Mach-O (Darwin) systems, this option directs the compiler emit
11672 to the glue for every direct call, and the Darwin linker decides
11673 whether to use or discard it.
11675 In the future, we may cause GCC to ignore all longcall
11676 specifications when the linker is known to generate glue.
11679 Adds support for multithreading with the "pthreads" library. This
11680 option sets flags for both the preprocessor and linker.
11684 File: gcc.info, Node: S/390 and zSeries Options, Next: Score Options, Prev: RS/6000 and PowerPC Options, Up: Submodel Options
11686 3.17.28 S/390 and zSeries Options
11687 ---------------------------------
11689 These are the `-m' options defined for the S/390 and zSeries
11694 Use (do not use) the hardware floating-point instructions and
11695 registers for floating-point operations. When `-msoft-float' is
11696 specified, functions in `libgcc.a' will be used to perform
11697 floating-point operations. When `-mhard-float' is specified, the
11698 compiler generates IEEE floating-point instructions. This is the
11702 `-mlong-double-128'
11703 These switches control the size of `long double' type. A size of
11704 64bit makes the `long double' type equivalent to the `double'
11705 type. This is the default.
11709 Store (do not store) the address of the caller's frame as
11710 backchain pointer into the callee's stack frame. A backchain may
11711 be needed to allow debugging using tools that do not understand
11712 DWARF-2 call frame information. When `-mno-packed-stack' is in
11713 effect, the backchain pointer is stored at the bottom of the stack
11714 frame; when `-mpacked-stack' is in effect, the backchain is placed
11715 into the topmost word of the 96/160 byte register save area.
11717 In general, code compiled with `-mbackchain' is call-compatible
11718 with code compiled with `-mmo-backchain'; however, use of the
11719 backchain for debugging purposes usually requires that the whole
11720 binary is built with `-mbackchain'. Note that the combination of
11721 `-mbackchain', `-mpacked-stack' and `-mhard-float' is not
11722 supported. In order to build a linux kernel use `-msoft-float'.
11724 The default is to not maintain the backchain.
11728 `-mno-packed-stack'
11729 Use (do not use) the packed stack layout. When
11730 `-mno-packed-stack' is specified, the compiler uses the all fields
11731 of the 96/160 byte register save area only for their default
11732 purpose; unused fields still take up stack space. When
11733 `-mpacked-stack' is specified, register save slots are densely
11734 packed at the top of the register save area; unused space is
11735 reused for other purposes, allowing for more efficient use of the
11736 available stack space. However, when `-mbackchain' is also in
11737 effect, the topmost word of the save area is always used to store
11738 the backchain, and the return address register is always saved two
11739 words below the backchain.
11741 As long as the stack frame backchain is not used, code generated
11742 with `-mpacked-stack' is call-compatible with code generated with
11743 `-mno-packed-stack'. Note that some non-FSF releases of GCC 2.95
11744 for S/390 or zSeries generated code that uses the stack frame
11745 backchain at run time, not just for debugging purposes. Such code
11746 is not call-compatible with code compiled with `-mpacked-stack'.
11747 Also, note that the combination of `-mbackchain', `-mpacked-stack'
11748 and `-mhard-float' is not supported. In order to build a linux
11749 kernel use `-msoft-float'.
11751 The default is to not use the packed stack layout.
11755 Generate (or do not generate) code using the `bras' instruction to
11756 do subroutine calls. This only works reliably if the total
11757 executable size does not exceed 64k. The default is to use the
11758 `basr' instruction instead, which does not have this limitation.
11762 When `-m31' is specified, generate code compliant to the GNU/Linux
11763 for S/390 ABI. When `-m64' is specified, generate code compliant
11764 to the GNU/Linux for zSeries ABI. This allows GCC in particular
11765 to generate 64-bit instructions. For the `s390' targets, the
11766 default is `-m31', while the `s390x' targets default to `-m64'.
11770 When `-mzarch' is specified, generate code using the instructions
11771 available on z/Architecture. When `-mesa' is specified, generate
11772 code using the instructions available on ESA/390. Note that
11773 `-mesa' is not possible with `-m64'. When generating code
11774 compliant to the GNU/Linux for S/390 ABI, the default is `-mesa'.
11775 When generating code compliant to the GNU/Linux for zSeries ABI,
11776 the default is `-mzarch'.
11780 Generate (or do not generate) code using the `mvcle' instruction
11781 to perform block moves. When `-mno-mvcle' is specified, use a
11782 `mvc' loop instead. This is the default unless optimizing for
11787 Print (or do not print) additional debug information when
11788 compiling. The default is to not print debug information.
11791 Generate code that will run on CPU-TYPE, which is the name of a
11792 system representing a certain processor type. Possible values for
11793 CPU-TYPE are `g5', `g6', `z900', and `z990'. When generating code
11794 using the instructions available on z/Architecture, the default is
11795 `-march=z900'. Otherwise, the default is `-march=g5'.
11798 Tune to CPU-TYPE everything applicable about the generated code,
11799 except for the ABI and the set of available instructions. The
11800 list of CPU-TYPE values is the same as for `-march'. The default
11801 is the value used for `-march'.
11805 Generate code that adds (does not add) in TPF OS specific branches
11806 to trace routines in the operating system. This option is off by
11807 default, even when compiling for the TPF OS.
11811 Generate code that uses (does not use) the floating point multiply
11812 and accumulate instructions. These instructions are generated by
11813 default if hardware floating point is used.
11815 `-mwarn-framesize=FRAMESIZE'
11816 Emit a warning if the current function exceeds the given frame
11817 size. Because this is a compile time check it doesn't need to be
11818 a real problem when the program runs. It is intended to identify
11819 functions which most probably cause a stack overflow. It is
11820 useful to be used in an environment with limited stack size e.g.
11823 `-mwarn-dynamicstack'
11824 Emit a warning if the function calls alloca or uses dynamically
11825 sized arrays. This is generally a bad idea with a limited stack
11828 `-mstack-guard=STACK-GUARD'
11830 `-mstack-size=STACK-SIZE'
11831 These arguments always have to be used in conjunction. If they
11832 are present the s390 back end emits additional instructions in the
11833 function prologue which trigger a trap if the stack size is
11834 STACK-GUARD bytes above the STACK-SIZE (remember that the stack on
11835 s390 grows downward). These options are intended to be used to
11836 help debugging stack overflow problems. The additionally emitted
11837 code causes only little overhead and hence can also be used in
11838 production like systems without greater performance degradation.
11839 The given values have to be exact powers of 2 and STACK-SIZE has
11840 to be greater than STACK-GUARD without exceeding 64k. In order to
11841 be efficient the extra code makes the assumption that the stack
11842 starts at an address aligned to the value given by STACK-SIZE.
11845 File: gcc.info, Node: Score Options, Next: SH Options, Prev: S/390 and zSeries Options, Up: Submodel Options
11847 3.17.29 Score Options
11848 ---------------------
11850 These options are defined for Score implementations:
11853 Compile code for big endian mode. This is the default.
11856 Compile code for little endian mode.
11859 Disable generate bcnz instruction.
11862 Enable generate unaligned load and store instruction.
11865 Enable the use of multiply-accumulate instructions. Disabled by
11869 Specify the SCORE5 as the target architecture.
11872 Specify the SCORE5U of the target architecture.
11875 Specify the SCORE7 as the target architecture. This is the default.
11878 Specify the SCORE7D as the target architecture.
11881 File: gcc.info, Node: SH Options, Next: SPARC Options, Prev: Score Options, Up: Submodel Options
11886 These `-m' options are defined for the SH implementations:
11889 Generate code for the SH1.
11892 Generate code for the SH2.
11895 Generate code for the SH2e.
11898 Generate code for the SH3.
11901 Generate code for the SH3e.
11904 Generate code for the SH4 without a floating-point unit.
11907 Generate code for the SH4 with a floating-point unit that only
11908 supports single-precision arithmetic.
11911 Generate code for the SH4 assuming the floating-point unit is in
11912 single-precision mode by default.
11915 Generate code for the SH4.
11918 Generate code for the SH4al-dsp, or for a SH4a in such a way that
11919 the floating-point unit is not used.
11922 Generate code for the SH4a, in such a way that no double-precision
11923 floating point operations are used.
11926 Generate code for the SH4a assuming the floating-point unit is in
11927 single-precision mode by default.
11930 Generate code for the SH4a.
11933 Same as `-m4a-nofpu', except that it implicitly passes `-dsp' to
11934 the assembler. GCC doesn't generate any DSP instructions at the
11938 Compile code for the processor in big endian mode.
11941 Compile code for the processor in little endian mode.
11944 Align doubles at 64-bit boundaries. Note that this changes the
11945 calling conventions, and thus some functions from the standard C
11946 library will not work unless you recompile it first with
11950 Shorten some address references at link time, when possible; uses
11951 the linker option `-relax'.
11954 Use 32-bit offsets in `switch' tables. The default is to use
11958 Enable the use of the instruction `fmovd'.
11961 Comply with the calling conventions defined by Renesas.
11964 Comply with the calling conventions defined by Renesas.
11967 Comply with the calling conventions defined for GCC before the
11968 Renesas conventions were available. This option is the default
11969 for all targets of the SH toolchain except for `sh-symbianelf'.
11972 Mark the `MAC' register as call-clobbered, even if `-mhitachi' is
11976 Increase IEEE-compliance of floating-point code. At the moment,
11977 this is equivalent to `-fno-finite-math-only'. When generating 16
11978 bit SH opcodes, getting IEEE-conforming results for comparisons of
11979 NANs / infinities incurs extra overhead in every floating point
11980 comparison, therefore the default is set to `-ffinite-math-only'.
11983 Dump instruction size and location in the assembly code.
11986 This option is deprecated. It pads structures to multiple of 4
11987 bytes, which is incompatible with the SH ABI.
11990 Optimize for space instead of speed. Implied by `-Os'.
11993 When generating position-independent code, emit function calls
11994 using the Global Offset Table instead of the Procedure Linkage
11998 Generate a library function call to invalidate instruction cache
11999 entries, after fixing up a trampoline. This library function call
12000 doesn't assume it can write to the whole memory address space.
12001 This is the default when the target is `sh-*-linux*'.
12004 Set the cost to assume for a multiply insn.
12007 Set the division strategy to use for SHmedia code. STRATEGY must
12008 be one of: call, call2, fp, inv, inv:minlat, inv20u, inv20l,
12009 inv:call, inv:call2, inv:fp . "fp" performs the operation in
12010 floating point. This has a very high latency, but needs only a
12011 few instructions, so it might be a good choice if your code has
12012 enough easily exploitable ILP to allow the compiler to schedule
12013 the floating point instructions together with other instructions.
12014 Division by zero causes a floating point exception. "inv" uses
12015 integer operations to calculate the inverse of the divisor, and
12016 then multiplies the dividend with the inverse. This strategy
12017 allows cse and hoisting of the inverse calculation. Division by
12018 zero calculates an unspecified result, but does not trap.
12019 "inv:minlat" is a variant of "inv" where if no cse / hoisting
12020 opportunities have been found, or if the entire operation has been
12021 hoisted to the same place, the last stages of the inverse
12022 calculation are intertwined with the final multiply to reduce the
12023 overall latency, at the expense of using a few more instructions,
12024 and thus offering fewer scheduling opportunities with other code.
12025 "call" calls a library function that usually implements the
12026 inv:minlat strategy. This gives high code density for
12027 m5-*media-nofpu compilations. "call2" uses a different entry
12028 point of the same library function, where it assumes that a
12029 pointer to a lookup table has already been set up, which exposes
12030 the pointer load to cse / code hoisting optimizations.
12031 "inv:call", "inv:call2" and "inv:fp" all use the "inv" algorithm
12032 for initial code generation, but if the code stays unoptimized,
12033 revert to the "call", "call2", or "fp" strategies, respectively.
12034 Note that the potentially-trapping side effect of division by zero
12035 is carried by a separate instruction, so it is possible that all
12036 the integer instructions are hoisted out, but the marker for the
12037 side effect stays where it is. A recombination to fp operations
12038 or a call is not possible in that case. "inv20u" and "inv20l" are
12039 variants of the "inv:minlat" strategy. In the case that the
12040 inverse calculation was nor separated from the multiply, they speed
12041 up division where the dividend fits into 20 bits (plus sign where
12042 applicable), by inserting a test to skip a number of operations in
12043 this case; this test slows down the case of larger dividends.
12044 inv20u assumes the case of a such a small dividend to be unlikely,
12045 and inv20l assumes it to be likely.
12047 `-mdivsi3_libfunc=NAME'
12048 Set the name of the library function used for 32 bit signed
12049 division to NAME. This only affect the name used in the call and
12050 inv:call division strategies, and the compiler will still expect
12051 the same sets of input/output/clobbered registers as if this
12052 option was not present.
12055 Throttle unrolling to avoid thrashing target registers. This
12056 option only has an effect if the gcc code base supports the
12057 TARGET_ADJUST_UNROLL_MAX target hook.
12059 `-mindexed-addressing'
12060 Enable the use of the indexed addressing mode for
12061 SHmedia32/SHcompact. This is only safe if the hardware and/or OS
12062 implement 32 bit wrap-around semantics for the indexed addressing
12063 mode. The architecture allows the implementation of processors
12064 with 64 bit MMU, which the OS could use to get 32 bit addressing,
12065 but since no current hardware implementation supports this or any
12066 other way to make the indexed addressing mode safe to use in the
12067 32 bit ABI, the default is -mno-indexed-addressing.
12069 `-mgettrcost=NUMBER'
12070 Set the cost assumed for the gettr instruction to NUMBER. The
12071 default is 2 if `-mpt-fixed' is in effect, 100 otherwise.
12074 Assume pt* instructions won't trap. This will generally generate
12075 better scheduled code, but is unsafe on current hardware. The
12076 current architecture definition says that ptabs and ptrel trap
12077 when the target anded with 3 is 3. This has the unintentional
12078 effect of making it unsafe to schedule ptabs / ptrel before a
12079 branch, or hoist it out of a loop. For example,
12080 __do_global_ctors, a part of libgcc that runs constructors at
12081 program startup, calls functions in a list which is delimited by
12082 -1. With the -mpt-fixed option, the ptabs will be done before
12083 testing against -1. That means that all the constructors will be
12084 run a bit quicker, but when the loop comes to the end of the list,
12085 the program crashes because ptabs loads -1 into a target register.
12086 Since this option is unsafe for any hardware implementing the
12087 current architecture specification, the default is -mno-pt-fixed.
12088 Unless the user specifies a specific cost with `-mgettrcost',
12089 -mno-pt-fixed also implies `-mgettrcost=100'; this deters register
12090 allocation using target registers for storing ordinary integers.
12092 `-minvalid-symbols'
12093 Assume symbols might be invalid. Ordinary function symbols
12094 generated by the compiler will always be valid to load with
12095 movi/shori/ptabs or movi/shori/ptrel, but with assembler and/or
12096 linker tricks it is possible to generate symbols that will cause
12097 ptabs / ptrel to trap. This option is only meaningful when
12098 `-mno-pt-fixed' is in effect. It will then prevent
12099 cross-basic-block cse, hoisting and most scheduling of symbol
12100 loads. The default is `-mno-invalid-symbols'.
12103 File: gcc.info, Node: SPARC Options, Next: System V Options, Prev: SH Options, Up: Submodel Options
12105 3.17.31 SPARC Options
12106 ---------------------
12108 These `-m' options are supported on the SPARC:
12112 Specify `-mapp-regs' to generate output using the global registers
12113 2 through 4, which the SPARC SVR4 ABI reserves for applications.
12114 This is the default.
12116 To be fully SVR4 ABI compliant at the cost of some performance
12117 loss, specify `-mno-app-regs'. You should compile libraries and
12118 system software with this option.
12122 Generate output containing floating point instructions. This is
12127 Generate output containing library calls for floating point.
12128 *Warning:* the requisite libraries are not available for all SPARC
12129 targets. Normally the facilities of the machine's usual C
12130 compiler are used, but this cannot be done directly in
12131 cross-compilation. You must make your own arrangements to provide
12132 suitable library functions for cross-compilation. The embedded
12133 targets `sparc-*-aout' and `sparclite-*-*' do provide software
12134 floating point support.
12136 `-msoft-float' changes the calling convention in the output file;
12137 therefore, it is only useful if you compile _all_ of a program with
12138 this option. In particular, you need to compile `libgcc.a', the
12139 library that comes with GCC, with `-msoft-float' in order for this
12142 `-mhard-quad-float'
12143 Generate output containing quad-word (long double) floating point
12146 `-msoft-quad-float'
12147 Generate output containing library calls for quad-word (long
12148 double) floating point instructions. The functions called are
12149 those specified in the SPARC ABI. This is the default.
12151 As of this writing, there are no SPARC implementations that have
12152 hardware support for the quad-word floating point instructions.
12153 They all invoke a trap handler for one of these instructions, and
12154 then the trap handler emulates the effect of the instruction.
12155 Because of the trap handler overhead, this is much slower than
12156 calling the ABI library routines. Thus the `-msoft-quad-float'
12157 option is the default.
12159 `-mno-unaligned-doubles'
12160 `-munaligned-doubles'
12161 Assume that doubles have 8 byte alignment. This is the default.
12163 With `-munaligned-doubles', GCC assumes that doubles have 8 byte
12164 alignment only if they are contained in another type, or if they
12165 have an absolute address. Otherwise, it assumes they have 4 byte
12166 alignment. Specifying this option avoids some rare compatibility
12167 problems with code generated by other compilers. It is not the
12168 default because it results in a performance loss, especially for
12169 floating point code.
12171 `-mno-faster-structs'
12173 With `-mfaster-structs', the compiler assumes that structures
12174 should have 8 byte alignment. This enables the use of pairs of
12175 `ldd' and `std' instructions for copies in structure assignment,
12176 in place of twice as many `ld' and `st' pairs. However, the use
12177 of this changed alignment directly violates the SPARC ABI. Thus,
12178 it's intended only for use on targets where the developer
12179 acknowledges that their resulting code will not be directly in
12180 line with the rules of the ABI.
12183 `-mimpure-text', used in addition to `-shared', tells the compiler
12184 to not pass `-z text' to the linker when linking a shared object.
12185 Using this option, you can link position-dependent code into a
12188 `-mimpure-text' suppresses the "relocations remain against
12189 allocatable but non-writable sections" linker error message.
12190 However, the necessary relocations will trigger copy-on-write, and
12191 the shared object is not actually shared across processes.
12192 Instead of using `-mimpure-text', you should compile all source
12193 code with `-fpic' or `-fPIC'.
12195 This option is only available on SunOS and Solaris.
12198 Set the instruction set, register set, and instruction scheduling
12199 parameters for machine type CPU_TYPE. Supported values for
12200 CPU_TYPE are `v7', `cypress', `v8', `supersparc', `sparclite',
12201 `f930', `f934', `hypersparc', `sparclite86x', `sparclet',
12202 `tsc701', `v9', `ultrasparc', `ultrasparc3', and `niagara'.
12204 Default instruction scheduling parameters are used for values that
12205 select an architecture and not an implementation. These are `v7',
12206 `v8', `sparclite', `sparclet', `v9'.
12208 Here is a list of each supported architecture and their supported
12212 v8: supersparc, hypersparc
12213 sparclite: f930, f934, sparclite86x
12215 v9: ultrasparc, ultrasparc3, niagara
12217 By default (unless configured otherwise), GCC generates code for
12218 the V7 variant of the SPARC architecture. With `-mcpu=cypress',
12219 the compiler additionally optimizes it for the Cypress CY7C602
12220 chip, as used in the SPARCStation/SPARCServer 3xx series. This is
12221 also appropriate for the older SPARCStation 1, 2, IPX etc.
12223 With `-mcpu=v8', GCC generates code for the V8 variant of the SPARC
12224 architecture. The only difference from V7 code is that the
12225 compiler emits the integer multiply and integer divide
12226 instructions which exist in SPARC-V8 but not in SPARC-V7. With
12227 `-mcpu=supersparc', the compiler additionally optimizes it for the
12228 SuperSPARC chip, as used in the SPARCStation 10, 1000 and 2000
12231 With `-mcpu=sparclite', GCC generates code for the SPARClite
12232 variant of the SPARC architecture. This adds the integer
12233 multiply, integer divide step and scan (`ffs') instructions which
12234 exist in SPARClite but not in SPARC-V7. With `-mcpu=f930', the
12235 compiler additionally optimizes it for the Fujitsu MB86930 chip,
12236 which is the original SPARClite, with no FPU. With `-mcpu=f934',
12237 the compiler additionally optimizes it for the Fujitsu MB86934
12238 chip, which is the more recent SPARClite with FPU.
12240 With `-mcpu=sparclet', GCC generates code for the SPARClet variant
12241 of the SPARC architecture. This adds the integer multiply,
12242 multiply/accumulate, integer divide step and scan (`ffs')
12243 instructions which exist in SPARClet but not in SPARC-V7. With
12244 `-mcpu=tsc701', the compiler additionally optimizes it for the
12245 TEMIC SPARClet chip.
12247 With `-mcpu=v9', GCC generates code for the V9 variant of the SPARC
12248 architecture. This adds 64-bit integer and floating-point move
12249 instructions, 3 additional floating-point condition code registers
12250 and conditional move instructions. With `-mcpu=ultrasparc', the
12251 compiler additionally optimizes it for the Sun UltraSPARC I/II/IIi
12252 chips. With `-mcpu=ultrasparc3', the compiler additionally
12253 optimizes it for the Sun UltraSPARC III/III+/IIIi/IIIi+/IV/IV+
12254 chips. With `-mcpu=niagara', the compiler additionally optimizes
12255 it for Sun UltraSPARC T1 chips.
12258 Set the instruction scheduling parameters for machine type
12259 CPU_TYPE, but do not set the instruction set or register set that
12260 the option `-mcpu=CPU_TYPE' would.
12262 The same values for `-mcpu=CPU_TYPE' can be used for
12263 `-mtune=CPU_TYPE', but the only useful values are those that
12264 select a particular cpu implementation. Those are `cypress',
12265 `supersparc', `hypersparc', `f930', `f934', `sparclite86x',
12266 `tsc701', `ultrasparc', `ultrasparc3', and `niagara'.
12270 With `-mv8plus', GCC generates code for the SPARC-V8+ ABI. The
12271 difference from the V8 ABI is that the global and out registers are
12272 considered 64-bit wide. This is enabled by default on Solaris in
12273 32-bit mode for all SPARC-V9 processors.
12277 With `-mvis', GCC generates code that takes advantage of the
12278 UltraSPARC Visual Instruction Set extensions. The default is
12281 These `-m' options are supported in addition to the above on SPARC-V9
12282 processors in 64-bit environments:
12285 Generate code for a processor running in little-endian mode. It
12286 is only available for a few configurations and most notably not on
12291 Generate code for a 32-bit or 64-bit environment. The 32-bit
12292 environment sets int, long and pointer to 32 bits. The 64-bit
12293 environment sets int to 32 bits and long and pointer to 64 bits.
12296 Generate code for the Medium/Low code model: 64-bit addresses,
12297 programs must be linked in the low 32 bits of memory. Programs
12298 can be statically or dynamically linked.
12301 Generate code for the Medium/Middle code model: 64-bit addresses,
12302 programs must be linked in the low 44 bits of memory, the text and
12303 data segments must be less than 2GB in size and the data segment
12304 must be located within 2GB of the text segment.
12307 Generate code for the Medium/Anywhere code model: 64-bit
12308 addresses, programs may be linked anywhere in memory, the text and
12309 data segments must be less than 2GB in size and the data segment
12310 must be located within 2GB of the text segment.
12312 `-mcmodel=embmedany'
12313 Generate code for the Medium/Anywhere code model for embedded
12314 systems: 64-bit addresses, the text and data segments must be less
12315 than 2GB in size, both starting anywhere in memory (determined at
12316 link time). The global register %g4 points to the base of the
12317 data segment. Programs are statically linked and PIC is not
12322 With `-mstack-bias', GCC assumes that the stack pointer, and frame
12323 pointer if present, are offset by -2047 which must be added back
12324 when making stack frame references. This is the default in 64-bit
12325 mode. Otherwise, assume no such offset is present.
12327 These switches are supported in addition to the above on Solaris:
12330 Add support for multithreading using the Solaris threads library.
12331 This option sets flags for both the preprocessor and linker. This
12332 option does not affect the thread safety of object code produced
12333 by the compiler or that of libraries supplied with it.
12336 Add support for multithreading using the POSIX threads library.
12337 This option sets flags for both the preprocessor and linker. This
12338 option does not affect the thread safety of object code produced
12339 by the compiler or that of libraries supplied with it.
12342 This is a synonym for `-pthreads'.
12345 File: gcc.info, Node: System V Options, Next: TMS320C3x/C4x Options, Prev: SPARC Options, Up: Submodel Options
12347 3.17.32 Options for System V
12348 ----------------------------
12350 These additional options are available on System V Release 4 for
12351 compatibility with other compilers on those systems:
12354 Create a shared object. It is recommended that `-symbolic' or
12355 `-shared' be used instead.
12358 Identify the versions of each tool used by the compiler, in a
12359 `.ident' assembler directive in the output.
12362 Refrain from adding `.ident' directives to the output file (this is
12366 Search the directories DIRS, and no others, for libraries
12367 specified with `-l'.
12370 Look in the directory DIR to find the M4 preprocessor. The
12371 assembler uses this option.
12374 File: gcc.info, Node: TMS320C3x/C4x Options, Next: V850 Options, Prev: System V Options, Up: Submodel Options
12376 3.17.33 TMS320C3x/C4x Options
12377 -----------------------------
12379 These `-m' options are defined for TMS320C3x/C4x implementations:
12382 Set the instruction set, register set, and instruction scheduling
12383 parameters for machine type CPU_TYPE. Supported values for
12384 CPU_TYPE are `c30', `c31', `c32', `c40', and `c44'. The default
12385 is `c40' to generate code for the TMS320C40.
12391 Generates code for the big or small memory model. The small memory
12392 model assumed that all data fits into one 64K word page. At
12393 run-time the data page (DP) register must be set to point to the
12394 64K page containing the .bss and .data program sections. The big
12395 memory model is the default and requires reloading of the DP
12396 register for every direct memory access.
12400 Allow (disallow) allocation of general integer operands into the
12401 block count register BK.
12405 Enable (disable) generation of code using decrement and branch,
12406 DBcond(D), instructions. This is enabled by default for the C4x.
12407 To be on the safe side, this is disabled for the C3x, since the
12408 maximum iteration count on the C3x is 2^23 + 1 (but who iterates
12409 loops more than 2^23 times on the C3x?). Note that GCC will try
12410 to reverse a loop so that it can utilize the decrement and branch
12411 instruction, but will give up if there is more than one memory
12412 reference in the loop. Thus a loop where the loop counter is
12413 decremented can generate slightly more efficient code, in cases
12414 where the RPTB instruction cannot be utilized.
12418 Force the DP register to be saved on entry to an interrupt service
12419 routine (ISR), reloaded to point to the data section, and restored
12420 on exit from the ISR. This should not be required unless someone
12421 has violated the small memory model by modifying the DP register,
12422 say within an object library.
12426 For the C3x use the 24-bit MPYI instruction for integer multiplies
12427 instead of a library call to guarantee 32-bit results. Note that
12428 if one of the operands is a constant, then the multiplication will
12429 be performed using shifts and adds. If the `-mmpyi' option is not
12430 specified for the C3x, then squaring operations are performed
12431 inline instead of a library call.
12435 The C3x/C4x FIX instruction to convert a floating point value to an
12436 integer value chooses the nearest integer less than or equal to the
12437 floating point value rather than to the nearest integer. Thus if
12438 the floating point number is negative, the result will be
12439 incorrectly truncated an additional code is necessary to detect
12440 and correct this case. This option can be used to disable
12441 generation of the additional code required to correct the result.
12445 Enable (disable) generation of repeat block sequences using the
12446 RPTB instruction for zero overhead looping. The RPTB construct is
12447 only used for innermost loops that do not call functions or jump
12448 across the loop boundaries. There is no advantage having nested
12449 RPTB loops due to the overhead required to save and restore the
12450 RC, RS, and RE registers. This is enabled by default with `-O2'.
12454 Enable (disable) the use of the single instruction repeat
12455 instruction RPTS. If a repeat block contains a single
12456 instruction, and the loop count can be guaranteed to be less than
12457 the value COUNT, GCC will emit a RPTS instruction instead of a
12458 RPTB. If no value is specified, then a RPTS will be emitted even
12459 if the loop count cannot be determined at compile time. Note that
12460 the repeated instruction following RPTS does not have to be
12461 reloaded from memory each iteration, thus freeing up the CPU buses
12462 for operands. However, since interrupts are blocked by this
12463 instruction, it is disabled by default.
12466 `-mno-loop-unsigned'
12467 The maximum iteration count when using RPTS and RPTB (and DB on
12468 the C40) is 2^31 + 1 since these instructions test if the
12469 iteration count is negative to terminate the loop. If the
12470 iteration count is unsigned there is a possibility than the 2^31 +
12471 1 maximum iteration count may be exceeded. This switch allows an
12472 unsigned iteration count.
12475 Try to emit an assembler syntax that the TI assembler (asm30) is
12476 happy with. This also enforces compatibility with the API
12477 employed by the TI C3x C compiler. For example, long doubles are
12478 passed as structures rather than in floating point registers.
12482 Generate code that uses registers (stack) for passing arguments to
12483 functions. By default, arguments are passed in registers where
12484 possible rather than by pushing arguments on to the stack.
12487 `-mno-parallel-insns'
12488 Allow the generation of parallel instructions. This is enabled by
12489 default with `-O2'.
12492 `-mno-parallel-mpy'
12493 Allow the generation of MPY||ADD and MPY||SUB parallel
12494 instructions, provided `-mparallel-insns' is also specified.
12495 These instructions have tight register constraints which can
12496 pessimize the code generation of large functions.
12500 File: gcc.info, Node: V850 Options, Next: VAX Options, Prev: TMS320C3x/C4x Options, Up: Submodel Options
12502 3.17.34 V850 Options
12503 --------------------
12505 These `-m' options are defined for V850 implementations:
12509 Treat all calls as being far away (near). If calls are assumed to
12510 be far away, the compiler will always load the functions address
12511 up into a register, and call indirect through the pointer.
12515 Do not optimize (do optimize) basic blocks that use the same index
12516 pointer 4 or more times to copy pointer into the `ep' register, and
12517 use the shorter `sld' and `sst' instructions. The `-mep' option
12518 is on by default if you optimize.
12520 `-mno-prolog-function'
12521 `-mprolog-function'
12522 Do not use (do use) external functions to save and restore
12523 registers at the prologue and epilogue of a function. The
12524 external functions are slower, but use less code space if more
12525 than one function saves the same number of registers. The
12526 `-mprolog-function' option is on by default if you optimize.
12529 Try to make the code as small as possible. At present, this just
12530 turns on the `-mep' and `-mprolog-function' options.
12533 Put static or global variables whose size is N bytes or less into
12534 the tiny data area that register `ep' points to. The tiny data
12535 area can hold up to 256 bytes in total (128 bytes for byte
12539 Put static or global variables whose size is N bytes or less into
12540 the small data area that register `gp' points to. The small data
12541 area can hold up to 64 kilobytes.
12544 Put static or global variables whose size is N bytes or less into
12545 the first 32 kilobytes of memory.
12548 Specify that the target processor is the V850.
12551 Generate code suitable for big switch tables. Use this option
12552 only if the assembler/linker complain about out of range branches
12553 within a switch table.
12556 This option will cause r2 and r5 to be used in the code generated
12557 by the compiler. This setting is the default.
12560 This option will cause r2 and r5 to be treated as fixed registers.
12563 Specify that the target processor is the V850E1. The preprocessor
12564 constants `__v850e1__' and `__v850e__' will be defined if this
12568 Specify that the target processor is the V850E. The preprocessor
12569 constant `__v850e__' will be defined if this option is used.
12571 If neither `-mv850' nor `-mv850e' nor `-mv850e1' are defined then
12572 a default target processor will be chosen and the relevant
12573 `__v850*__' preprocessor constant will be defined.
12575 The preprocessor constants `__v850' and `__v851__' are always
12576 defined, regardless of which processor variant is the target.
12579 This option will suppress generation of the CALLT instruction for
12580 the v850e and v850e1 flavors of the v850 architecture. The
12581 default is `-mno-disable-callt' which allows the CALLT instruction
12586 File: gcc.info, Node: VAX Options, Next: x86-64 Options, Prev: V850 Options, Up: Submodel Options
12588 3.17.35 VAX Options
12589 -------------------
12591 These `-m' options are defined for the VAX:
12594 Do not output certain jump instructions (`aobleq' and so on) that
12595 the Unix assembler for the VAX cannot handle across long ranges.
12598 Do output those jump instructions, on the assumption that you will
12599 assemble with the GNU assembler.
12602 Output code for g-format floating point numbers instead of
12606 File: gcc.info, Node: x86-64 Options, Next: Xstormy16 Options, Prev: VAX Options, Up: Submodel Options
12608 3.17.36 x86-64 Options
12609 ----------------------
12611 These are listed under *Note i386 and x86-64 Options::.
12614 File: gcc.info, Node: Xstormy16 Options, Next: Xtensa Options, Prev: x86-64 Options, Up: Submodel Options
12616 3.17.37 Xstormy16 Options
12617 -------------------------
12619 These options are defined for Xstormy16:
12622 Choose startup files and linker script suitable for the simulator.
12625 File: gcc.info, Node: Xtensa Options, Next: zSeries Options, Prev: Xstormy16 Options, Up: Submodel Options
12627 3.17.38 Xtensa Options
12628 ----------------------
12630 These options are supported for Xtensa targets:
12634 Enable or disable use of `CONST16' instructions for loading
12635 constant values. The `CONST16' instruction is currently not a
12636 standard option from Tensilica. When enabled, `CONST16'
12637 instructions are always used in place of the standard `L32R'
12638 instructions. The use of `CONST16' is enabled by default only if
12639 the `L32R' instruction is not available.
12643 Enable or disable use of fused multiply/add and multiply/subtract
12644 instructions in the floating-point option. This has no effect if
12645 the floating-point option is not also enabled. Disabling fused
12646 multiply/add and multiply/subtract instructions forces the
12647 compiler to use separate instructions for the multiply and
12648 add/subtract operations. This may be desirable in some cases
12649 where strict IEEE 754-compliant results are required: the fused
12650 multiply add/subtract instructions do not round the intermediate
12651 result, thereby producing results with _more_ bits of precision
12652 than specified by the IEEE standard. Disabling fused multiply
12653 add/subtract instructions also ensures that the program output is
12654 not sensitive to the compiler's ability to combine multiply and
12655 add/subtract operations.
12657 `-mtext-section-literals'
12658 `-mno-text-section-literals'
12659 Control the treatment of literal pools. The default is
12660 `-mno-text-section-literals', which places literals in a separate
12661 section in the output file. This allows the literal pool to be
12662 placed in a data RAM/ROM, and it also allows the linker to combine
12663 literal pools from separate object files to remove redundant
12664 literals and improve code size. With `-mtext-section-literals',
12665 the literals are interspersed in the text section in order to keep
12666 them as close as possible to their references. This may be
12667 necessary for large assembly files.
12670 `-mno-target-align'
12671 When this option is enabled, GCC instructs the assembler to
12672 automatically align instructions to reduce branch penalties at the
12673 expense of some code density. The assembler attempts to widen
12674 density instructions to align branch targets and the instructions
12675 following call instructions. If there are not enough preceding
12676 safe density instructions to align a target, no widening will be
12677 performed. The default is `-mtarget-align'. These options do not
12678 affect the treatment of auto-aligned instructions like `LOOP',
12679 which the assembler will always align, either by widening density
12680 instructions or by inserting no-op instructions.
12684 When this option is enabled, GCC instructs the assembler to
12685 translate direct calls to indirect calls unless it can determine
12686 that the target of a direct call is in the range allowed by the
12687 call instruction. This translation typically occurs for calls to
12688 functions in other source files. Specifically, the assembler
12689 translates a direct `CALL' instruction into an `L32R' followed by
12690 a `CALLX' instruction. The default is `-mno-longcalls'. This
12691 option should be used in programs where the call target can
12692 potentially be out of range. This option is implemented in the
12693 assembler, not the compiler, so the assembly code generated by GCC
12694 will still show direct call instructions--look at the disassembled
12695 object code to see the actual instructions. Note that the
12696 assembler will use an indirect call for every cross-file call, not
12697 just those that really will be out of range.
12700 File: gcc.info, Node: zSeries Options, Prev: Xtensa Options, Up: Submodel Options
12702 3.17.39 zSeries Options
12703 -----------------------
12705 These are listed under *Note S/390 and zSeries Options::.
12708 File: gcc.info, Node: Code Gen Options, Next: Environment Variables, Prev: Submodel Options, Up: Invoking GCC
12710 3.18 Options for Code Generation Conventions
12711 ============================================
12713 These machine-independent options control the interface conventions
12714 used in code generation.
12716 Most of them have both positive and negative forms; the negative form
12717 of `-ffoo' would be `-fno-foo'. In the table below, only one of the
12718 forms is listed--the one which is not the default. You can figure out
12719 the other form by either removing `no-' or adding it.
12722 For front-ends that support it, generate additional code to check
12723 that indices used to access arrays are within the declared range.
12724 This is currently only supported by the Java and Fortran
12725 front-ends, where this option defaults to true and false
12729 This option generates traps for signed overflow on addition,
12730 subtraction, multiplication operations.
12733 This option instructs the compiler to assume that signed arithmetic
12734 overflow of addition, subtraction and multiplication wraps around
12735 using twos-complement representation. This flag enables some
12736 optimizations and disables others. This option is enabled by
12737 default for the Java front-end, as required by the Java language
12741 Enable exception handling. Generates extra code needed to
12742 propagate exceptions. For some targets, this implies GCC will
12743 generate frame unwind information for all functions, which can
12744 produce significant data size overhead, although it does not
12745 affect execution. If you do not specify this option, GCC will
12746 enable it by default for languages like C++ which normally require
12747 exception handling, and disable it for languages like C that do
12748 not normally require it. However, you may need to enable this
12749 option when compiling C code that needs to interoperate properly
12750 with exception handlers written in C++. You may also wish to
12751 disable this option if you are compiling older C++ programs that
12752 don't use exception handling.
12754 `-fnon-call-exceptions'
12755 Generate code that allows trapping instructions to throw
12756 exceptions. Note that this requires platform-specific runtime
12757 support that does not exist everywhere. Moreover, it only allows
12758 _trapping_ instructions to throw exceptions, i.e. memory
12759 references or floating point instructions. It does not allow
12760 exceptions to be thrown from arbitrary signal handlers such as
12764 Similar to `-fexceptions', except that it will just generate any
12765 needed static data, but will not affect the generated code in any
12766 other way. You will normally not enable this option; instead, a
12767 language processor that needs this handling would enable it on
12770 `-fasynchronous-unwind-tables'
12771 Generate unwind table in dwarf2 format, if supported by target
12772 machine. The table is exact at each instruction boundary, so it
12773 can be used for stack unwinding from asynchronous events (such as
12774 debugger or garbage collector).
12776 `-fpcc-struct-return'
12777 Return "short" `struct' and `union' values in memory like longer
12778 ones, rather than in registers. This convention is less
12779 efficient, but it has the advantage of allowing intercallability
12780 between GCC-compiled files and files compiled with other
12781 compilers, particularly the Portable C Compiler (pcc).
12783 The precise convention for returning structures in memory depends
12784 on the target configuration macros.
12786 Short structures and unions are those whose size and alignment
12787 match that of some integer type.
12789 *Warning:* code compiled with the `-fpcc-struct-return' switch is
12790 not binary compatible with code compiled with the
12791 `-freg-struct-return' switch. Use it to conform to a non-default
12792 application binary interface.
12794 `-freg-struct-return'
12795 Return `struct' and `union' values in registers when possible.
12796 This is more efficient for small structures than
12797 `-fpcc-struct-return'.
12799 If you specify neither `-fpcc-struct-return' nor
12800 `-freg-struct-return', GCC defaults to whichever convention is
12801 standard for the target. If there is no standard convention, GCC
12802 defaults to `-fpcc-struct-return', except on targets where GCC is
12803 the principal compiler. In those cases, we can choose the
12804 standard, and we chose the more efficient register return
12807 *Warning:* code compiled with the `-freg-struct-return' switch is
12808 not binary compatible with code compiled with the
12809 `-fpcc-struct-return' switch. Use it to conform to a non-default
12810 application binary interface.
12813 Allocate to an `enum' type only as many bytes as it needs for the
12814 declared range of possible values. Specifically, the `enum' type
12815 will be equivalent to the smallest integer type which has enough
12818 *Warning:* the `-fshort-enums' switch causes GCC to generate code
12819 that is not binary compatible with code generated without that
12820 switch. Use it to conform to a non-default application binary
12824 Use the same size for `double' as for `float'.
12826 *Warning:* the `-fshort-double' switch causes GCC to generate code
12827 that is not binary compatible with code generated without that
12828 switch. Use it to conform to a non-default application binary
12832 Override the underlying type for `wchar_t' to be `short unsigned
12833 int' instead of the default for the target. This option is useful
12834 for building programs to run under WINE.
12836 *Warning:* the `-fshort-wchar' switch causes GCC to generate code
12837 that is not binary compatible with code generated without that
12838 switch. Use it to conform to a non-default application binary
12842 In C, allocate even uninitialized global variables in the data
12843 section of the object file, rather than generating them as common
12844 blocks. This has the effect that if the same variable is declared
12845 (without `extern') in two different compilations, you will get an
12846 error when you link them. The only reason this might be useful is
12847 if you wish to verify that the program will work on other systems
12848 which always work this way.
12851 Ignore the `#ident' directive.
12853 `-finhibit-size-directive'
12854 Don't output a `.size' assembler directive, or anything else that
12855 would cause trouble if the function is split in the middle, and the
12856 two halves are placed at locations far apart in memory. This
12857 option is used when compiling `crtstuff.c'; you should not need to
12858 use it for anything else.
12861 Put extra commentary information in the generated assembly code to
12862 make it more readable. This option is generally only of use to
12863 those who actually need to read the generated assembly code
12864 (perhaps while debugging the compiler itself).
12866 `-fno-verbose-asm', the default, causes the extra information to
12867 be omitted and is useful when comparing two assembler files.
12870 Generate position-independent code (PIC) suitable for use in a
12871 shared library, if supported for the target machine. Such code
12872 accesses all constant addresses through a global offset table
12873 (GOT). The dynamic loader resolves the GOT entries when the
12874 program starts (the dynamic loader is not part of GCC; it is part
12875 of the operating system). If the GOT size for the linked
12876 executable exceeds a machine-specific maximum size, you get an
12877 error message from the linker indicating that `-fpic' does not
12878 work; in that case, recompile with `-fPIC' instead. (These
12879 maximums are 8k on the SPARC and 32k on the m68k and RS/6000. The
12880 386 has no such limit.)
12882 Position-independent code requires special support, and therefore
12883 works only on certain machines. For the 386, GCC supports PIC for
12884 System V but not for the Sun 386i. Code generated for the IBM
12885 RS/6000 is always position-independent.
12887 When this flag is set, the macros `__pic__' and `__PIC__' are
12891 If supported for the target machine, emit position-independent
12892 code, suitable for dynamic linking and avoiding any limit on the
12893 size of the global offset table. This option makes a difference
12894 on the m68k, PowerPC and SPARC.
12896 Position-independent code requires special support, and therefore
12897 works only on certain machines.
12899 When this flag is set, the macros `__pic__' and `__PIC__' are
12904 These options are similar to `-fpic' and `-fPIC', but generated
12905 position independent code can be only linked into executables.
12906 Usually these options are used when `-pie' GCC option will be used
12910 Do not use jump tables for switch statements even where it would be
12911 more efficient than other code generation strategies. This option
12912 is of use in conjunction with `-fpic' or `-fPIC' for building code
12913 which forms part of a dynamic linker and cannot reference the
12914 address of a jump table. On some targets, jump tables do not
12915 require a GOT and this option is not needed.
12918 Treat the register named REG as a fixed register; generated code
12919 should never refer to it (except perhaps as a stack pointer, frame
12920 pointer or in some other fixed role).
12922 REG must be the name of a register. The register names accepted
12923 are machine-specific and are defined in the `REGISTER_NAMES' macro
12924 in the machine description macro file.
12926 This flag does not have a negative form, because it specifies a
12930 Treat the register named REG as an allocable register that is
12931 clobbered by function calls. It may be allocated for temporaries
12932 or variables that do not live across a call. Functions compiled
12933 this way will not save and restore the register REG.
12935 It is an error to used this flag with the frame pointer or stack
12936 pointer. Use of this flag for other registers that have fixed
12937 pervasive roles in the machine's execution model will produce
12938 disastrous results.
12940 This flag does not have a negative form, because it specifies a
12944 Treat the register named REG as an allocable register saved by
12945 functions. It may be allocated even for temporaries or variables
12946 that live across a call. Functions compiled this way will save
12947 and restore the register REG if they use it.
12949 It is an error to used this flag with the frame pointer or stack
12950 pointer. Use of this flag for other registers that have fixed
12951 pervasive roles in the machine's execution model will produce
12952 disastrous results.
12954 A different sort of disaster will result from the use of this flag
12955 for a register in which function values may be returned.
12957 This flag does not have a negative form, because it specifies a
12960 `-fpack-struct[=N]'
12961 Without a value specified, pack all structure members together
12962 without holes. When a value is specified (which must be a small
12963 power of two), pack structure members according to this value,
12964 representing the maximum alignment (that is, objects with default
12965 alignment requirements larger than this will be output potentially
12966 unaligned at the next fitting location.
12968 *Warning:* the `-fpack-struct' switch causes GCC to generate code
12969 that is not binary compatible with code generated without that
12970 switch. Additionally, it makes the code suboptimal. Use it to
12971 conform to a non-default application binary interface.
12973 `-finstrument-functions'
12974 Generate instrumentation calls for entry and exit to functions.
12975 Just after function entry and just before function exit, the
12976 following profiling functions will be called with the address of
12977 the current function and its call site. (On some platforms,
12978 `__builtin_return_address' does not work beyond the current
12979 function, so the call site information may not be available to the
12980 profiling functions otherwise.)
12982 void __cyg_profile_func_enter (void *this_fn,
12984 void __cyg_profile_func_exit (void *this_fn,
12987 The first argument is the address of the start of the current
12988 function, which may be looked up exactly in the symbol table.
12990 This instrumentation is also done for functions expanded inline in
12991 other functions. The profiling calls will indicate where,
12992 conceptually, the inline function is entered and exited. This
12993 means that addressable versions of such functions must be
12994 available. If all your uses of a function are expanded inline,
12995 this may mean an additional expansion of code size. If you use
12996 `extern inline' in your C code, an addressable version of such
12997 functions must be provided. (This is normally the case anyways,
12998 but if you get lucky and the optimizer always expands the
12999 functions inline, you might have gotten away without providing
13002 A function may be given the attribute `no_instrument_function', in
13003 which case this instrumentation will not be done. This can be
13004 used, for example, for the profiling functions listed above,
13005 high-priority interrupt routines, and any functions from which the
13006 profiling functions cannot safely be called (perhaps signal
13007 handlers, if the profiling routines generate output or allocate
13010 `-finstrument-functions-exclude-file-list=FILE,FILE,...'
13011 Set the list of functions that are excluded from instrumentation
13012 (see the description of `-finstrument-functions'). If the file
13013 that contains a function definition matches with one of FILE, then
13014 that function is not instrumented. The match is done on
13015 substrings: if the FILE parameter is a substring of the file name,
13016 it is considered to be a match.
13019 `-finstrument-functions-exclude-file-list=/bits/stl,include/sys'
13020 will exclude any inline function defined in files whose pathnames
13021 contain `/bits/stl' or `include/sys'.
13023 If, for some reason, you want to include letter `','' in one of
13024 SYM, write `'\,''. For example,
13025 `-finstrument-functions-exclude-file-list='\,\,tmp'' (note the
13026 single quote surrounding the option).
13028 `-finstrument-functions-exclude-function-list=SYM,SYM,...'
13029 This is similar to `-finstrument-functions-exclude-file-list', but
13030 this option sets the list of function names to be excluded from
13031 instrumentation. The function name to be matched is its
13032 user-visible name, such as `vector<int> blah(const vector<int>
13033 &)', not the internal mangled name (e.g.,
13034 `_Z4blahRSt6vectorIiSaIiEE'). The match is done on substrings: if
13035 the SYM parameter is a substring of the function name, it is
13036 considered to be a match.
13039 Generate code to verify that you do not go beyond the boundary of
13040 the stack. You should specify this flag if you are running in an
13041 environment with multiple threads, but only rarely need to specify
13042 it in a single-threaded environment since stack overflow is
13043 automatically detected on nearly all systems if there is only one
13046 Note that this switch does not actually cause checking to be done;
13047 the operating system must do that. The switch causes generation
13048 of code to ensure that the operating system sees the stack being
13051 `-fstack-limit-register=REG'
13052 `-fstack-limit-symbol=SYM'
13054 Generate code to ensure that the stack does not grow beyond a
13055 certain value, either the value of a register or the address of a
13056 symbol. If the stack would grow beyond the value, a signal is
13057 raised. For most targets, the signal is raised before the stack
13058 overruns the boundary, so it is possible to catch the signal
13059 without taking special precautions.
13061 For instance, if the stack starts at absolute address `0x80000000'
13062 and grows downwards, you can use the flags
13063 `-fstack-limit-symbol=__stack_limit' and
13064 `-Wl,--defsym,__stack_limit=0x7ffe0000' to enforce a stack limit
13065 of 128KB. Note that this may only work with the GNU linker.
13068 `-fargument-noalias'
13069 `-fargument-noalias-global'
13070 `-fargument-noalias-anything'
13071 Specify the possible relationships among parameters and between
13072 parameters and global data.
13074 `-fargument-alias' specifies that arguments (parameters) may alias
13075 each other and may alias global storage.
13076 `-fargument-noalias' specifies that arguments do not alias each
13077 other, but may alias global storage.
13078 `-fargument-noalias-global' specifies that arguments do not alias
13079 each other and do not alias global storage.
13080 `-fargument-noalias-anything' specifies that arguments do not
13081 alias any other storage.
13083 Each language will automatically use whatever option is required by
13084 the language standard. You should not need to use these options
13087 `-fleading-underscore'
13088 This option and its counterpart, `-fno-leading-underscore',
13089 forcibly change the way C symbols are represented in the object
13090 file. One use is to help link with legacy assembly code.
13092 *Warning:* the `-fleading-underscore' switch causes GCC to
13093 generate code that is not binary compatible with code generated
13094 without that switch. Use it to conform to a non-default
13095 application binary interface. Not all targets provide complete
13096 support for this switch.
13098 `-ftls-model=MODEL'
13099 Alter the thread-local storage model to be used (*note
13100 Thread-Local::). The MODEL argument should be one of
13101 `global-dynamic', `local-dynamic', `initial-exec' or `local-exec'.
13103 The default without `-fpic' is `initial-exec'; with `-fpic' the
13104 default is `global-dynamic'.
13106 `-fvisibility=DEFAULT|INTERNAL|HIDDEN|PROTECTED'
13107 Set the default ELF image symbol visibility to the specified
13108 option--all symbols will be marked with this unless overridden
13109 within the code. Using this feature can very substantially
13110 improve linking and load times of shared object libraries, produce
13111 more optimized code, provide near-perfect API export and prevent
13112 symbol clashes. It is *strongly* recommended that you use this in
13113 any shared objects you distribute.
13115 Despite the nomenclature, `default' always means public ie;
13116 available to be linked against from outside the shared object.
13117 `protected' and `internal' are pretty useless in real-world usage
13118 so the only other commonly used option will be `hidden'. The
13119 default if `-fvisibility' isn't specified is `default', i.e., make
13120 every symbol public--this causes the same behavior as previous
13123 A good explanation of the benefits offered by ensuring ELF symbols
13124 have the correct visibility is given by "How To Write Shared
13125 Libraries" by Ulrich Drepper (which can be found at
13126 `http://people.redhat.com/~drepper/')--however a superior solution
13127 made possible by this option to marking things hidden when the
13128 default is public is to make the default hidden and mark things
13129 public. This is the norm with DLL's on Windows and with
13130 `-fvisibility=hidden' and `__attribute__
13131 ((visibility("default")))' instead of `__declspec(dllexport)' you
13132 get almost identical semantics with identical syntax. This is a
13133 great boon to those working with cross-platform projects.
13135 For those adding visibility support to existing code, you may find
13136 `#pragma GCC visibility' of use. This works by you enclosing the
13137 declarations you wish to set visibility for with (for example)
13138 `#pragma GCC visibility push(hidden)' and `#pragma GCC visibility
13139 pop'. Bear in mind that symbol visibility should be viewed *as
13140 part of the API interface contract* and thus all new code should
13141 always specify visibility when it is not the default ie;
13142 declarations only for use within the local DSO should *always* be
13143 marked explicitly as hidden as so to avoid PLT indirection
13144 overheads--making this abundantly clear also aids readability and
13145 self-documentation of the code. Note that due to ISO C++
13146 specification requirements, operator new and operator delete must
13147 always be of default visibility.
13149 Be aware that headers from outside your project, in particular
13150 system headers and headers from any other library you use, may not
13151 be expecting to be compiled with visibility other than the
13152 default. You may need to explicitly say `#pragma GCC visibility
13153 push(default)' before including any such headers.
13155 `extern' declarations are not affected by `-fvisibility', so a lot
13156 of code can be recompiled with `-fvisibility=hidden' with no
13157 modifications. However, this means that calls to `extern'
13158 functions with no explicit visibility will use the PLT, so it is
13159 more effective to use `__attribute ((visibility))' and/or `#pragma
13160 GCC visibility' to tell the compiler which `extern' declarations
13161 should be treated as hidden.
13163 Note that `-fvisibility' does affect C++ vague linkage entities.
13164 This means that, for instance, an exception class that will be
13165 thrown between DSOs must be explicitly marked with default
13166 visibility so that the `type_info' nodes will be unified between
13169 An overview of these techniques, their benefits and how to use them
13170 is at `http://gcc.gnu.org/wiki/Visibility'.
13174 File: gcc.info, Node: Environment Variables, Next: Precompiled Headers, Prev: Code Gen Options, Up: Invoking GCC
13176 3.19 Environment Variables Affecting GCC
13177 ========================================
13179 This section describes several environment variables that affect how GCC
13180 operates. Some of them work by specifying directories or prefixes to
13181 use when searching for various kinds of files. Some are used to
13182 specify other aspects of the compilation environment.
13184 Note that you can also specify places to search using options such as
13185 `-B', `-I' and `-L' (*note Directory Options::). These take precedence
13186 over places specified using environment variables, which in turn take
13187 precedence over those specified by the configuration of GCC. *Note
13188 Controlling the Compilation Driver `gcc': (gccint)Driver.
13194 These environment variables control the way that GCC uses
13195 localization information that allow GCC to work with different
13196 national conventions. GCC inspects the locale categories
13197 `LC_CTYPE' and `LC_MESSAGES' if it has been configured to do so.
13198 These locale categories can be set to any value supported by your
13199 installation. A typical value is `en_GB.UTF-8' for English in the
13200 United Kingdom encoded in UTF-8.
13202 The `LC_CTYPE' environment variable specifies character
13203 classification. GCC uses it to determine the character boundaries
13204 in a string; this is needed for some multibyte encodings that
13205 contain quote and escape characters that would otherwise be
13206 interpreted as a string end or escape.
13208 The `LC_MESSAGES' environment variable specifies the language to
13209 use in diagnostic messages.
13211 If the `LC_ALL' environment variable is set, it overrides the value
13212 of `LC_CTYPE' and `LC_MESSAGES'; otherwise, `LC_CTYPE' and
13213 `LC_MESSAGES' default to the value of the `LANG' environment
13214 variable. If none of these variables are set, GCC defaults to
13215 traditional C English behavior.
13218 If `TMPDIR' is set, it specifies the directory to use for temporary
13219 files. GCC uses temporary files to hold the output of one stage of
13220 compilation which is to be used as input to the next stage: for
13221 example, the output of the preprocessor, which is the input to the
13225 If `GCC_EXEC_PREFIX' is set, it specifies a prefix to use in the
13226 names of the subprograms executed by the compiler. No slash is
13227 added when this prefix is combined with the name of a subprogram,
13228 but you can specify a prefix that ends with a slash if you wish.
13230 If `GCC_EXEC_PREFIX' is not set, GCC will attempt to figure out an
13231 appropriate prefix to use based on the pathname it was invoked
13234 If GCC cannot find the subprogram using the specified prefix, it
13235 tries looking in the usual places for the subprogram.
13237 The default value of `GCC_EXEC_PREFIX' is `PREFIX/lib/gcc/' where
13238 PREFIX is the value of `prefix' when you ran the `configure'
13241 Other prefixes specified with `-B' take precedence over this
13244 This prefix is also used for finding files such as `crt0.o' that
13245 are used for linking.
13247 In addition, the prefix is used in an unusual way in finding the
13248 directories to search for header files. For each of the standard
13249 directories whose name normally begins with `/usr/local/lib/gcc'
13250 (more precisely, with the value of `GCC_INCLUDE_DIR'), GCC tries
13251 replacing that beginning with the specified prefix to produce an
13252 alternate directory name. Thus, with `-Bfoo/', GCC will search
13253 `foo/bar' where it would normally search `/usr/local/lib/bar'.
13254 These alternate directories are searched first; the standard
13255 directories come next.
13258 The value of `COMPILER_PATH' is a colon-separated list of
13259 directories, much like `PATH'. GCC tries the directories thus
13260 specified when searching for subprograms, if it can't find the
13261 subprograms using `GCC_EXEC_PREFIX'.
13264 The value of `LIBRARY_PATH' is a colon-separated list of
13265 directories, much like `PATH'. When configured as a native
13266 compiler, GCC tries the directories thus specified when searching
13267 for special linker files, if it can't find them using
13268 `GCC_EXEC_PREFIX'. Linking using GCC also uses these directories
13269 when searching for ordinary libraries for the `-l' option (but
13270 directories specified with `-L' come first).
13273 This variable is used to pass locale information to the compiler.
13274 One way in which this information is used is to determine the
13275 character set to be used when character literals, string literals
13276 and comments are parsed in C and C++. When the compiler is
13277 configured to allow multibyte characters, the following values for
13278 `LANG' are recognized:
13281 Recognize JIS characters.
13284 Recognize SJIS characters.
13287 Recognize EUCJP characters.
13289 If `LANG' is not defined, or if it has some other value, then the
13290 compiler will use mblen and mbtowc as defined by the default
13291 locale to recognize and translate multibyte characters.
13293 Some additional environments variables affect the behavior of the
13298 `CPLUS_INCLUDE_PATH'
13299 `OBJC_INCLUDE_PATH'
13300 Each variable's value is a list of directories separated by a
13301 special character, much like `PATH', in which to look for header
13302 files. The special character, `PATH_SEPARATOR', is
13303 target-dependent and determined at GCC build time. For Microsoft
13304 Windows-based targets it is a semicolon, and for almost all other
13305 targets it is a colon.
13307 `CPATH' specifies a list of directories to be searched as if
13308 specified with `-I', but after any paths given with `-I' options
13309 on the command line. This environment variable is used regardless
13310 of which language is being preprocessed.
13312 The remaining environment variables apply only when preprocessing
13313 the particular language indicated. Each specifies a list of
13314 directories to be searched as if specified with `-isystem', but
13315 after any paths given with `-isystem' options on the command line.
13317 In all these variables, an empty element instructs the compiler to
13318 search its current working directory. Empty elements can appear
13319 at the beginning or end of a path. For instance, if the value of
13320 `CPATH' is `:/special/include', that has the same effect as
13321 `-I. -I/special/include'.
13323 `DEPENDENCIES_OUTPUT'
13324 If this variable is set, its value specifies how to output
13325 dependencies for Make based on the non-system header files
13326 processed by the compiler. System header files are ignored in the
13329 The value of `DEPENDENCIES_OUTPUT' can be just a file name, in
13330 which case the Make rules are written to that file, guessing the
13331 target name from the source file name. Or the value can have the
13332 form `FILE TARGET', in which case the rules are written to file
13333 FILE using TARGET as the target name.
13335 In other words, this environment variable is equivalent to
13336 combining the options `-MM' and `-MF' (*note Preprocessor
13337 Options::), with an optional `-MT' switch too.
13339 `SUNPRO_DEPENDENCIES'
13340 This variable is the same as `DEPENDENCIES_OUTPUT' (see above),
13341 except that system header files are not ignored, so it implies
13342 `-M' rather than `-MM'. However, the dependence on the main input
13343 file is omitted. *Note Preprocessor Options::.
13346 File: gcc.info, Node: Precompiled Headers, Next: Running Protoize, Prev: Environment Variables, Up: Invoking GCC
13348 3.20 Using Precompiled Headers
13349 ==============================
13351 Often large projects have many header files that are included in every
13352 source file. The time the compiler takes to process these header files
13353 over and over again can account for nearly all of the time required to
13354 build the project. To make builds faster, GCC allows users to
13355 `precompile' a header file; then, if builds can use the precompiled
13356 header file they will be much faster.
13358 To create a precompiled header file, simply compile it as you would any
13359 other file, if necessary using the `-x' option to make the driver treat
13360 it as a C or C++ header file. You will probably want to use a tool
13361 like `make' to keep the precompiled header up-to-date when the headers
13362 it contains change.
13364 A precompiled header file will be searched for when `#include' is seen
13365 in the compilation. As it searches for the included file (*note Search
13366 Path: (cpp)Search Path.) the compiler looks for a precompiled header in
13367 each directory just before it looks for the include file in that
13368 directory. The name searched for is the name specified in the
13369 `#include' with `.gch' appended. If the precompiled header file can't
13370 be used, it is ignored.
13372 For instance, if you have `#include "all.h"', and you have `all.h.gch'
13373 in the same directory as `all.h', then the precompiled header file will
13374 be used if possible, and the original header will be used otherwise.
13376 Alternatively, you might decide to put the precompiled header file in a
13377 directory and use `-I' to ensure that directory is searched before (or
13378 instead of) the directory containing the original header. Then, if you
13379 want to check that the precompiled header file is always used, you can
13380 put a file of the same name as the original header in this directory
13381 containing an `#error' command.
13383 This also works with `-include'. So yet another way to use
13384 precompiled headers, good for projects not designed with precompiled
13385 header files in mind, is to simply take most of the header files used by
13386 a project, include them from another header file, precompile that header
13387 file, and `-include' the precompiled header. If the header files have
13388 guards against multiple inclusion, they will be skipped because they've
13389 already been included (in the precompiled header).
13391 If you need to precompile the same header file for different
13392 languages, targets, or compiler options, you can instead make a
13393 _directory_ named like `all.h.gch', and put each precompiled header in
13394 the directory, perhaps using `-o'. It doesn't matter what you call the
13395 files in the directory, every precompiled header in the directory will
13396 be considered. The first precompiled header encountered in the
13397 directory that is valid for this compilation will be used; they're
13398 searched in no particular order.
13400 There are many other possibilities, limited only by your imagination,
13401 good sense, and the constraints of your build system.
13403 A precompiled header file can be used only when these conditions apply:
13405 * Only one precompiled header can be used in a particular
13408 * A precompiled header can't be used once the first C token is seen.
13409 You can have preprocessor directives before a precompiled header;
13410 you can even include a precompiled header from inside another
13411 header, so long as there are no C tokens before the `#include'.
13413 * The precompiled header file must be produced for the same language
13414 as the current compilation. You can't use a C precompiled header
13415 for a C++ compilation.
13417 * The precompiled header file must have been produced by the same
13418 compiler binary as the current compilation is using.
13420 * Any macros defined before the precompiled header is included must
13421 either be defined in the same way as when the precompiled header
13422 was generated, or must not affect the precompiled header, which
13423 usually means that they don't appear in the precompiled header at
13426 The `-D' option is one way to define a macro before a precompiled
13427 header is included; using a `#define' can also do it. There are
13428 also some options that define macros implicitly, like `-O' and
13429 `-Wdeprecated'; the same rule applies to macros defined this way.
13431 * If debugging information is output when using the precompiled
13432 header, using `-g' or similar, the same kind of debugging
13433 information must have been output when building the precompiled
13434 header. However, a precompiled header built using `-g' can be
13435 used in a compilation when no debugging information is being
13438 * The same `-m' options must generally be used when building and
13439 using the precompiled header. *Note Submodel Options::, for any
13440 cases where this rule is relaxed.
13442 * Each of the following options must be the same when building and
13443 using the precompiled header:
13445 -fexceptions -funit-at-a-time
13447 * Some other command-line options starting with `-f', `-p', or `-O'
13448 must be defined in the same way as when the precompiled header was
13449 generated. At present, it's not clear which options are safe to
13450 change and which are not; the safest choice is to use exactly the
13451 same options when generating and using the precompiled header.
13452 The following are known to be safe:
13454 -fmessage-length= -fpreprocessed
13455 -fsched-interblock -fsched-spec -fsched-spec-load -fsched-spec-load-dangerous
13456 -fsched-verbose=<number> -fschedule-insns -fvisibility=
13460 For all of these except the last, the compiler will automatically
13461 ignore the precompiled header if the conditions aren't met. If you
13462 find an option combination that doesn't work and doesn't cause the
13463 precompiled header to be ignored, please consider filing a bug report,
13466 If you do use differing options when generating and using the
13467 precompiled header, the actual behavior will be a mixture of the
13468 behavior for the options. For instance, if you use `-g' to generate
13469 the precompiled header but not when using it, you may or may not get
13470 debugging information for routines in the precompiled header.
13473 File: gcc.info, Node: Running Protoize, Prev: Precompiled Headers, Up: Invoking GCC
13475 3.21 Running Protoize
13476 =====================
13478 The program `protoize' is an optional part of GCC. You can use it to
13479 add prototypes to a program, thus converting the program to ISO C in
13480 one respect. The companion program `unprotoize' does the reverse: it
13481 removes argument types from any prototypes that are found.
13483 When you run these programs, you must specify a set of source files as
13484 command line arguments. The conversion programs start out by compiling
13485 these files to see what functions they define. The information gathered
13486 about a file FOO is saved in a file named `FOO.X'.
13488 After scanning comes actual conversion. The specified files are all
13489 eligible to be converted; any files they include (whether sources or
13490 just headers) are eligible as well.
13492 But not all the eligible files are converted. By default, `protoize'
13493 and `unprotoize' convert only source and header files in the current
13494 directory. You can specify additional directories whose files should
13495 be converted with the `-d DIRECTORY' option. You can also specify
13496 particular files to exclude with the `-x FILE' option. A file is
13497 converted if it is eligible, its directory name matches one of the
13498 specified directory names, and its name within the directory has not
13501 Basic conversion with `protoize' consists of rewriting most function
13502 definitions and function declarations to specify the types of the
13503 arguments. The only ones not rewritten are those for varargs functions.
13505 `protoize' optionally inserts prototype declarations at the beginning
13506 of the source file, to make them available for any calls that precede
13507 the function's definition. Or it can insert prototype declarations
13508 with block scope in the blocks where undeclared functions are called.
13510 Basic conversion with `unprotoize' consists of rewriting most function
13511 declarations to remove any argument types, and rewriting function
13512 definitions to the old-style pre-ISO form.
13514 Both conversion programs print a warning for any function declaration
13515 or definition that they can't convert. You can suppress these warnings
13518 The output from `protoize' or `unprotoize' replaces the original
13519 source file. The original file is renamed to a name ending with
13520 `.save' (for DOS, the saved filename ends in `.sav' without the
13521 original `.c' suffix). If the `.save' (`.sav' for DOS) file already
13522 exists, then the source file is simply discarded.
13524 `protoize' and `unprotoize' both depend on GCC itself to scan the
13525 program and collect information about the functions it uses. So
13526 neither of these programs will work until GCC is installed.
13528 Here is a table of the options you can use with `protoize' and
13529 `unprotoize'. Each option works with both programs unless otherwise
13533 Look for the file `SYSCALLS.c.X' in DIRECTORY, instead of the
13534 usual directory (normally `/usr/local/lib'). This file contains
13535 prototype information about standard system functions. This option
13536 applies only to `protoize'.
13538 `-c COMPILATION-OPTIONS'
13539 Use COMPILATION-OPTIONS as the options when running `gcc' to
13540 produce the `.X' files. The special option `-aux-info' is always
13541 passed in addition, to tell `gcc' to write a `.X' file.
13543 Note that the compilation options must be given as a single
13544 argument to `protoize' or `unprotoize'. If you want to specify
13545 several `gcc' options, you must quote the entire set of
13546 compilation options to make them a single word in the shell.
13548 There are certain `gcc' arguments that you cannot use, because they
13549 would produce the wrong kind of output. These include `-g', `-O',
13550 `-c', `-S', and `-o' If you include these in the
13551 COMPILATION-OPTIONS, they are ignored.
13554 Rename files to end in `.C' (`.cc' for DOS-based file systems)
13555 instead of `.c'. This is convenient if you are converting a C
13556 program to C++. This option applies only to `protoize'.
13559 Add explicit global declarations. This means inserting explicit
13560 declarations at the beginning of each source file for each function
13561 that is called in the file and was not declared. These
13562 declarations precede the first function definition that contains a
13563 call to an undeclared function. This option applies only to
13567 Indent old-style parameter declarations with the string STRING.
13568 This option applies only to `protoize'.
13570 `unprotoize' converts prototyped function definitions to old-style
13571 function definitions, where the arguments are declared between the
13572 argument list and the initial `{'. By default, `unprotoize' uses
13573 five spaces as the indentation. If you want to indent with just
13574 one space instead, use `-i " "'.
13577 Keep the `.X' files. Normally, they are deleted after conversion
13581 Add explicit local declarations. `protoize' with `-l' inserts a
13582 prototype declaration for each function in each block which calls
13583 the function without any declaration. This option applies only to
13587 Make no real changes. This mode just prints information about the
13588 conversions that would have been done without `-n'.
13591 Make no `.save' files. The original files are simply deleted.
13592 Use this option with caution.
13595 Use the program PROGRAM as the compiler. Normally, the name `gcc'
13599 Work quietly. Most warnings are suppressed.
13602 Print the version number, just like `-v' for `gcc'.
13604 If you need special compiler options to compile one of your program's
13605 source files, then you should generate that file's `.X' file specially,
13606 by running `gcc' on that source file with the appropriate options and
13607 the option `-aux-info'. Then run `protoize' on the entire set of
13608 files. `protoize' will use the existing `.X' file because it is newer
13609 than the source file. For example:
13611 gcc -Dfoo=bar file1.c -aux-info file1.X
13614 You need to include the special files along with the rest in the
13615 `protoize' command, even though their `.X' files already exist, because
13616 otherwise they won't get converted.
13618 *Note Protoize Caveats::, for more information on how to use
13619 `protoize' successfully.
13622 File: gcc.info, Node: C Implementation, Next: C Extensions, Prev: Invoking GCC, Up: Top
13624 4 C Implementation-defined behavior
13625 ***********************************
13627 A conforming implementation of ISO C is required to document its choice
13628 of behavior in each of the areas that are designated "implementation
13629 defined". The following lists all such areas, along with the section
13630 numbers from the ISO/IEC 9899:1990 and ISO/IEC 9899:1999 standards.
13631 Some areas are only implementation-defined in one version of the
13634 Some choices depend on the externally determined ABI for the platform
13635 (including standard character encodings) which GCC follows; these are
13636 listed as "determined by ABI" below. *Note Binary Compatibility:
13637 Compatibility, and `http://gcc.gnu.org/readings.html'. Some choices
13638 are documented in the preprocessor manual. *Note
13639 Implementation-defined behavior: (cpp)Implementation-defined behavior.
13640 Some choices are made by the library and operating system (or other
13641 environment when compiling for a freestanding environment); refer to
13642 their documentation for details.
13646 * Translation implementation::
13647 * Environment implementation::
13648 * Identifiers implementation::
13649 * Characters implementation::
13650 * Integers implementation::
13651 * Floating point implementation::
13652 * Arrays and pointers implementation::
13653 * Hints implementation::
13654 * Structures unions enumerations and bit-fields implementation::
13655 * Qualifiers implementation::
13656 * Declarators implementation::
13657 * Statements implementation::
13658 * Preprocessing directives implementation::
13659 * Library functions implementation::
13660 * Architecture implementation::
13661 * Locale-specific behavior implementation::
13664 File: gcc.info, Node: Translation implementation, Next: Environment implementation, Up: C Implementation
13669 * `How a diagnostic is identified (C90 3.7, C99 3.10, C90 and C99
13672 Diagnostics consist of all the output sent to stderr by GCC.
13674 * `Whether each nonempty sequence of white-space characters other
13675 than new-line is retained or replaced by one space character in
13676 translation phase 3 (C90 and C99 5.1.1.2).'
13678 *Note Implementation-defined behavior: (cpp)Implementation-defined
13683 File: gcc.info, Node: Environment implementation, Next: Identifiers implementation, Prev: Translation implementation, Up: C Implementation
13688 The behavior of most of these points are dependent on the implementation
13689 of the C library, and are not defined by GCC itself.
13691 * `The mapping between physical source file multibyte characters and
13692 the source character set in translation phase 1 (C90 and C99
13695 *Note Implementation-defined behavior: (cpp)Implementation-defined
13700 File: gcc.info, Node: Identifiers implementation, Next: Characters implementation, Prev: Environment implementation, Up: C Implementation
13705 * `Which additional multibyte characters may appear in identifiers
13706 and their correspondence to universal character names (C99 6.4.2).'
13708 *Note Implementation-defined behavior: (cpp)Implementation-defined
13711 * `The number of significant initial characters in an identifier
13712 (C90 6.1.2, C90 and C99 5.2.4.1, C99 6.4.2).'
13714 For internal names, all characters are significant. For external
13715 names, the number of significant characters are defined by the
13716 linker; for almost all targets, all characters are significant.
13718 * `Whether case distinctions are significant in an identifier with
13719 external linkage (C90 6.1.2).'
13721 This is a property of the linker. C99 requires that case
13722 distinctions are always significant in identifiers with external
13723 linkage and systems without this property are not supported by GCC.
13727 File: gcc.info, Node: Characters implementation, Next: Integers implementation, Prev: Identifiers implementation, Up: C Implementation
13732 * `The number of bits in a byte (C90 3.4, C99 3.6).'
13736 * `The values of the members of the execution character set (C90 and
13741 * `The unique value of the member of the execution character set
13742 produced for each of the standard alphabetic escape sequences (C90
13747 * `The value of a `char' object into which has been stored any
13748 character other than a member of the basic execution character set
13749 (C90 6.1.2.5, C99 6.2.5).'
13753 * `Which of `signed char' or `unsigned char' has the same range,
13754 representation, and behavior as "plain" `char' (C90 6.1.2.5, C90
13755 6.2.1.1, C99 6.2.5, C99 6.3.1.1).'
13757 Determined by ABI. The options `-funsigned-char' and
13758 `-fsigned-char' change the default. *Note Options Controlling C
13759 Dialect: C Dialect Options.
13761 * `The mapping of members of the source character set (in character
13762 constants and string literals) to members of the execution
13763 character set (C90 6.1.3.4, C99 6.4.4.4, C90 and C99 5.1.1.2).'
13767 * `The value of an integer character constant containing more than
13768 one character or containing a character or escape sequence that
13769 does not map to a single-byte execution character (C90 6.1.3.4,
13772 *Note Implementation-defined behavior: (cpp)Implementation-defined
13775 * `The value of a wide character constant containing more than one
13776 multibyte character, or containing a multibyte character or escape
13777 sequence not represented in the extended execution character set
13778 (C90 6.1.3.4, C99 6.4.4.4).'
13780 *Note Implementation-defined behavior: (cpp)Implementation-defined
13783 * `The current locale used to convert a wide character constant
13784 consisting of a single multibyte character that maps to a member
13785 of the extended execution character set into a corresponding wide
13786 character code (C90 6.1.3.4, C99 6.4.4.4).'
13788 *Note Implementation-defined behavior: (cpp)Implementation-defined
13791 * `The current locale used to convert a wide string literal into
13792 corresponding wide character codes (C90 6.1.4, C99 6.4.5).'
13794 *Note Implementation-defined behavior: (cpp)Implementation-defined
13797 * `The value of a string literal containing a multibyte character or
13798 escape sequence not represented in the execution character set
13799 (C90 6.1.4, C99 6.4.5).'
13801 *Note Implementation-defined behavior: (cpp)Implementation-defined
13805 File: gcc.info, Node: Integers implementation, Next: Floating point implementation, Prev: Characters implementation, Up: C Implementation
13810 * `Any extended integer types that exist in the implementation (C99
13813 GCC does not support any extended integer types.
13815 * `Whether signed integer types are represented using sign and
13816 magnitude, two's complement, or one's complement, and whether the
13817 extraordinary value is a trap representation or an ordinary value
13820 GCC supports only two's complement integer types, and all bit
13821 patterns are ordinary values.
13823 * `The rank of any extended integer type relative to another extended
13824 integer type with the same precision (C99 6.3.1.1).'
13826 GCC does not support any extended integer types.
13828 * `The result of, or the signal raised by, converting an integer to a
13829 signed integer type when the value cannot be represented in an
13830 object of that type (C90 6.2.1.2, C99 6.3.1.3).'
13832 For conversion to a type of width N, the value is reduced modulo
13833 2^N to be within range of the type; no signal is raised.
13835 * `The results of some bitwise operations on signed integers (C90
13838 Bitwise operators act on the representation of the value including
13839 both the sign and value bits, where the sign bit is considered
13840 immediately above the highest-value value bit. Signed `>>' acts
13841 on negative numbers by sign extension.
13843 GCC does not use the latitude given in C99 only to treat certain
13844 aspects of signed `<<' as undefined, but this is subject to change.
13846 * `The sign of the remainder on integer division (C90 6.3.5).'
13848 GCC always follows the C99 requirement that the result of division
13849 is truncated towards zero.
13853 File: gcc.info, Node: Floating point implementation, Next: Arrays and pointers implementation, Prev: Integers implementation, Up: C Implementation
13858 * `The accuracy of the floating-point operations and of the library
13859 functions in `<math.h>' and `<complex.h>' that return
13860 floating-point results (C90 and C99 5.2.4.2.2).'
13862 The accuracy is unknown.
13864 * `The rounding behaviors characterized by non-standard values of
13865 `FLT_ROUNDS' (C90 and C99 5.2.4.2.2).'
13867 GCC does not use such values.
13869 * `The evaluation methods characterized by non-standard negative
13870 values of `FLT_EVAL_METHOD' (C99 5.2.4.2.2).'
13872 GCC does not use such values.
13874 * `The direction of rounding when an integer is converted to a
13875 floating-point number that cannot exactly represent the original
13876 value (C90 6.2.1.3, C99 6.3.1.4).'
13878 C99 Annex F is followed.
13880 * `The direction of rounding when a floating-point number is
13881 converted to a narrower floating-point number (C90 6.2.1.4, C99
13884 C99 Annex F is followed.
13886 * `How the nearest representable value or the larger or smaller
13887 representable value immediately adjacent to the nearest
13888 representable value is chosen for certain floating constants (C90
13889 6.1.3.1, C99 6.4.4.2).'
13891 C99 Annex F is followed.
13893 * `Whether and how floating expressions are contracted when not
13894 disallowed by the `FP_CONTRACT' pragma (C99 6.5).'
13896 Expressions are currently only contracted if
13897 `-funsafe-math-optimizations' or `-ffast-math' are used. This is
13900 * `The default state for the `FENV_ACCESS' pragma (C99 7.6.1).'
13902 This pragma is not implemented, but the default is to "off" unless
13903 `-frounding-math' is used in which case it is "on".
13905 * `Additional floating-point exceptions, rounding modes,
13906 environments, and classifications, and their macro names (C99 7.6,
13909 This is dependent on the implementation of the C library, and is
13910 not defined by GCC itself.
13912 * `The default state for the `FP_CONTRACT' pragma (C99 7.12.2).'
13914 This pragma is not implemented. Expressions are currently only
13915 contracted if `-funsafe-math-optimizations' or `-ffast-math' are
13916 used. This is subject to change.
13918 * `Whether the "inexact" floating-point exception can be raised when
13919 the rounded result actually does equal the mathematical result in
13920 an IEC 60559 conformant implementation (C99 F.9).'
13922 This is dependent on the implementation of the C library, and is
13923 not defined by GCC itself.
13925 * `Whether the "underflow" (and "inexact") floating-point exception
13926 can be raised when a result is tiny but not inexact in an IEC
13927 60559 conformant implementation (C99 F.9).'
13929 This is dependent on the implementation of the C library, and is
13930 not defined by GCC itself.
13934 File: gcc.info, Node: Arrays and pointers implementation, Next: Hints implementation, Prev: Floating point implementation, Up: C Implementation
13936 4.7 Arrays and pointers
13937 =======================
13939 * `The result of converting a pointer to an integer or vice versa
13940 (C90 6.3.4, C99 6.3.2.3).'
13942 A cast from pointer to integer discards most-significant bits if
13943 the pointer representation is larger than the integer type,
13944 sign-extends(1) if the pointer representation is smaller than the
13945 integer type, otherwise the bits are unchanged.
13947 A cast from integer to pointer discards most-significant bits if
13948 the pointer representation is smaller than the integer type,
13949 extends according to the signedness of the integer type if the
13950 pointer representation is larger than the integer type, otherwise
13951 the bits are unchanged.
13953 When casting from pointer to integer and back again, the resulting
13954 pointer must reference the same object as the original pointer,
13955 otherwise the behavior is undefined. That is, one may not use
13956 integer arithmetic to avoid the undefined behavior of pointer
13957 arithmetic as proscribed in C99 6.5.6/8.
13959 * `The size of the result of subtracting two pointers to elements of
13960 the same array (C90 6.3.6, C99 6.5.6).'
13962 The value is as specified in the standard and the type is
13963 determined by the ABI.
13966 ---------- Footnotes ----------
13968 (1) Future versions of GCC may zero-extend, or use a target-defined
13969 `ptr_extend' pattern. Do not rely on sign extension.
13972 File: gcc.info, Node: Hints implementation, Next: Structures unions enumerations and bit-fields implementation, Prev: Arrays and pointers implementation, Up: C Implementation
13977 * `The extent to which suggestions made by using the `register'
13978 storage-class specifier are effective (C90 6.5.1, C99 6.7.1).'
13980 The `register' specifier affects code generation only in these
13983 * When used as part of the register variable extension, see
13984 *Note Explicit Reg Vars::.
13986 * When `-O0' is in use, the compiler allocates distinct stack
13987 memory for all variables that do not have the `register'
13988 storage-class specifier; if `register' is specified, the
13989 variable may have a shorter lifespan than the code would
13990 indicate and may never be placed in memory.
13992 * On some rare x86 targets, `setjmp' doesn't save the registers
13993 in all circumstances. In those cases, GCC doesn't allocate
13994 any variables in registers unless they are marked `register'.
13997 * `The extent to which suggestions made by using the inline function
13998 specifier are effective (C99 6.7.4).'
14000 GCC will not inline any functions if the `-fno-inline' option is
14001 used or if `-O0' is used. Otherwise, GCC may still be unable to
14002 inline a function for many reasons; the `-Winline' option may be
14003 used to determine if a function has not been inlined and why not.
14007 File: gcc.info, Node: Structures unions enumerations and bit-fields implementation, Next: Qualifiers implementation, Prev: Hints implementation, Up: C Implementation
14009 4.9 Structures, unions, enumerations, and bit-fields
14010 ====================================================
14012 * `A member of a union object is accessed using a member of a
14013 different type (C90 6.3.2.3).'
14015 The relevant bytes of the representation of the object are treated
14016 as an object of the type used for the access. This may be a trap
14019 * `Whether a "plain" `int' bit-field is treated as a `signed int'
14020 bit-field or as an `unsigned int' bit-field (C90 6.5.2, C90
14021 6.5.2.1, C99 6.7.2, C99 6.7.2.1).'
14023 By default it is treated as `signed int' but this may be changed
14024 by the `-funsigned-bitfields' option.
14026 * `Allowable bit-field types other than `_Bool', `signed int', and
14027 `unsigned int' (C99 6.7.2.1).'
14029 No other types are permitted in strictly conforming mode.
14031 * `Whether a bit-field can straddle a storage-unit boundary (C90
14032 6.5.2.1, C99 6.7.2.1).'
14036 * `The order of allocation of bit-fields within a unit (C90 6.5.2.1,
14041 * `The alignment of non-bit-field members of structures (C90
14042 6.5.2.1, C99 6.7.2.1).'
14046 * `The integer type compatible with each enumerated type (C90
14047 6.5.2.2, C99 6.7.2.2).'
14049 Normally, the type is `unsigned int' if there are no negative
14050 values in the enumeration, otherwise `int'. If `-fshort-enums' is
14051 specified, then if there are negative values it is the first of
14052 `signed char', `short' and `int' that can represent all the
14053 values, otherwise it is the first of `unsigned char', `unsigned
14054 short' and `unsigned int' that can represent all the values.
14056 On some targets, `-fshort-enums' is the default; this is
14057 determined by the ABI.
14061 File: gcc.info, Node: Qualifiers implementation, Next: Declarators implementation, Prev: Structures unions enumerations and bit-fields implementation, Up: C Implementation
14066 * `What constitutes an access to an object that has
14067 volatile-qualified type (C90 6.5.3, C99 6.7.3).'
14069 Such an object is normally accessed by pointers and used for
14070 accessing hardware. In most expressions, it is intuitively
14071 obvious what is a read and what is a write. For example
14073 volatile int *dst = SOMEVALUE;
14074 volatile int *src = SOMEOTHERVALUE;
14077 will cause a read of the volatile object pointed to by SRC and
14078 store the value into the volatile object pointed to by DST. There
14079 is no guarantee that these reads and writes are atomic, especially
14080 for objects larger than `int'.
14082 However, if the volatile storage is not being modified, and the
14083 value of the volatile storage is not used, then the situation is
14084 less obvious. For example
14086 volatile int *src = SOMEVALUE;
14089 According to the C standard, such an expression is an rvalue whose
14090 type is the unqualified version of its original type, i.e. `int'.
14091 Whether GCC interprets this as a read of the volatile object being
14092 pointed to or only as a request to evaluate the expression for its
14093 side-effects depends on this type.
14095 If it is a scalar type, or on most targets an aggregate type whose
14096 only member object is of a scalar type, or a union type whose
14097 member objects are of scalar types, the expression is interpreted
14098 by GCC as a read of the volatile object; in the other cases, the
14099 expression is only evaluated for its side-effects.
14103 File: gcc.info, Node: Declarators implementation, Next: Statements implementation, Prev: Qualifiers implementation, Up: C Implementation
14108 * `The maximum number of declarators that may modify an arithmetic,
14109 structure or union type (C90 6.5.4).'
14111 GCC is only limited by available memory.
14115 File: gcc.info, Node: Statements implementation, Next: Preprocessing directives implementation, Prev: Declarators implementation, Up: C Implementation
14120 * `The maximum number of `case' values in a `switch' statement (C90
14123 GCC is only limited by available memory.
14127 File: gcc.info, Node: Preprocessing directives implementation, Next: Library functions implementation, Prev: Statements implementation, Up: C Implementation
14129 4.13 Preprocessing directives
14130 =============================
14132 *Note Implementation-defined behavior: (cpp)Implementation-defined
14133 behavior, for details of these aspects of implementation-defined
14136 * `How sequences in both forms of header names are mapped to headers
14137 or external source file names (C90 6.1.7, C99 6.4.7).'
14139 * `Whether the value of a character constant in a constant expression
14140 that controls conditional inclusion matches the value of the same
14141 character constant in the execution character set (C90 6.8.1, C99
14144 * `Whether the value of a single-character character constant in a
14145 constant expression that controls conditional inclusion may have a
14146 negative value (C90 6.8.1, C99 6.10.1).'
14148 * `The places that are searched for an included `<>' delimited
14149 header, and how the places are specified or the header is
14150 identified (C90 6.8.2, C99 6.10.2).'
14152 * `How the named source file is searched for in an included `""'
14153 delimited header (C90 6.8.2, C99 6.10.2).'
14155 * `The method by which preprocessing tokens (possibly resulting from
14156 macro expansion) in a `#include' directive are combined into a
14157 header name (C90 6.8.2, C99 6.10.2).'
14159 * `The nesting limit for `#include' processing (C90 6.8.2, C99
14162 * `Whether the `#' operator inserts a `\' character before the `\'
14163 character that begins a universal character name in a character
14164 constant or string literal (C99 6.10.3.2).'
14166 * `The behavior on each recognized non-`STDC #pragma' directive (C90
14167 6.8.6, C99 6.10.6).'
14169 *Note Pragmas: (cpp)Pragmas, for details of pragmas accepted by
14170 GCC on all targets. *Note Pragmas Accepted by GCC: Pragmas, for
14171 details of target-specific pragmas.
14173 * `The definitions for `__DATE__' and `__TIME__' when respectively,
14174 the date and time of translation are not available (C90 6.8.8, C99
14179 File: gcc.info, Node: Library functions implementation, Next: Architecture implementation, Prev: Preprocessing directives implementation, Up: C Implementation
14181 4.14 Library functions
14182 ======================
14184 The behavior of most of these points are dependent on the implementation
14185 of the C library, and are not defined by GCC itself.
14187 * `The null pointer constant to which the macro `NULL' expands (C90
14190 In `<stddef.h>', `NULL' expands to `((void *)0)'. GCC does not
14191 provide the other headers which define `NULL' and some library
14192 implementations may use other definitions in those headers.
14196 File: gcc.info, Node: Architecture implementation, Next: Locale-specific behavior implementation, Prev: Library functions implementation, Up: C Implementation
14201 * `The values or expressions assigned to the macros specified in the
14202 headers `<float.h>', `<limits.h>', and `<stdint.h>' (C90 and C99
14203 5.2.4.2, C99 7.18.2, C99 7.18.3).'
14207 * `The number, order, and encoding of bytes in any object (when not
14208 explicitly specified in this International Standard) (C99
14213 * `The value of the result of the `sizeof' operator (C90 6.3.3.4,
14220 File: gcc.info, Node: Locale-specific behavior implementation, Prev: Architecture implementation, Up: C Implementation
14222 4.16 Locale-specific behavior
14223 =============================
14225 The behavior of these points are dependent on the implementation of the
14226 C library, and are not defined by GCC itself.
14229 File: gcc.info, Node: C Extensions, Next: C++ Extensions, Prev: C Implementation, Up: Top
14231 5 Extensions to the C Language Family
14232 *************************************
14234 GNU C provides several language features not found in ISO standard C.
14235 (The `-pedantic' option directs GCC to print a warning message if any
14236 of these features is used.) To test for the availability of these
14237 features in conditional compilation, check for a predefined macro
14238 `__GNUC__', which is always defined under GCC.
14240 These extensions are available in C and Objective-C. Most of them are
14241 also available in C++. *Note Extensions to the C++ Language: C++
14242 Extensions, for extensions that apply _only_ to C++.
14244 Some features that are in ISO C99 but not C89 or C++ are also, as
14245 extensions, accepted by GCC in C89 mode and in C++.
14249 * Statement Exprs:: Putting statements and declarations inside expressions.
14250 * Local Labels:: Labels local to a block.
14251 * Labels as Values:: Getting pointers to labels, and computed gotos.
14252 * Nested Functions:: As in Algol and Pascal, lexical scoping of functions.
14253 * Constructing Calls:: Dispatching a call to another function.
14254 * Typeof:: `typeof': referring to the type of an expression.
14255 * Conditionals:: Omitting the middle operand of a `?:' expression.
14256 * Long Long:: Double-word integers---`long long int'.
14257 * Complex:: Data types for complex numbers.
14258 * Decimal Float:: Decimal Floating Types.
14259 * Hex Floats:: Hexadecimal floating-point constants.
14260 * Zero Length:: Zero-length arrays.
14261 * Variable Length:: Arrays whose length is computed at run time.
14262 * Empty Structures:: Structures with no members.
14263 * Variadic Macros:: Macros with a variable number of arguments.
14264 * Escaped Newlines:: Slightly looser rules for escaped newlines.
14265 * Subscripting:: Any array can be subscripted, even if not an lvalue.
14266 * Pointer Arith:: Arithmetic on `void'-pointers and function pointers.
14267 * Initializers:: Non-constant initializers.
14268 * Compound Literals:: Compound literals give structures, unions
14269 or arrays as values.
14270 * Designated Inits:: Labeling elements of initializers.
14271 * Cast to Union:: Casting to union type from any member of the union.
14272 * Case Ranges:: `case 1 ... 9' and such.
14273 * Mixed Declarations:: Mixing declarations and code.
14274 * Function Attributes:: Declaring that functions have no side effects,
14275 or that they can never return.
14276 * Attribute Syntax:: Formal syntax for attributes.
14277 * Function Prototypes:: Prototype declarations and old-style definitions.
14278 * C++ Comments:: C++ comments are recognized.
14279 * Dollar Signs:: Dollar sign is allowed in identifiers.
14280 * Character Escapes:: `\e' stands for the character <ESC>.
14281 * Variable Attributes:: Specifying attributes of variables.
14282 * Type Attributes:: Specifying attributes of types.
14283 * Alignment:: Inquiring about the alignment of a type or variable.
14284 * Inline:: Defining inline functions (as fast as macros).
14285 * Extended Asm:: Assembler instructions with C expressions as operands.
14286 (With them you can define ``built-in'' functions.)
14287 * Constraints:: Constraints for asm operands
14288 * Asm Labels:: Specifying the assembler name to use for a C symbol.
14289 * Explicit Reg Vars:: Defining variables residing in specified registers.
14290 * Alternate Keywords:: `__const__', `__asm__', etc., for header files.
14291 * Incomplete Enums:: `enum foo;', with details to follow.
14292 * Function Names:: Printable strings which are the name of the current
14294 * Return Address:: Getting the return or frame address of a function.
14295 * Vector Extensions:: Using vector instructions through built-in functions.
14296 * Offsetof:: Special syntax for implementing `offsetof'.
14297 * Atomic Builtins:: Built-in functions for atomic memory access.
14298 * Object Size Checking:: Built-in functions for limited buffer overflow
14300 * Other Builtins:: Other built-in functions.
14301 * Target Builtins:: Built-in functions specific to particular targets.
14302 * Target Format Checks:: Format checks specific to particular targets.
14303 * Pragmas:: Pragmas accepted by GCC.
14304 * Unnamed Fields:: Unnamed struct/union fields within structs/unions.
14305 * Thread-Local:: Per-thread variables.
14308 File: gcc.info, Node: Statement Exprs, Next: Local Labels, Up: C Extensions
14310 5.1 Statements and Declarations in Expressions
14311 ==============================================
14313 A compound statement enclosed in parentheses may appear as an expression
14314 in GNU C. This allows you to use loops, switches, and local variables
14315 within an expression.
14317 Recall that a compound statement is a sequence of statements surrounded
14318 by braces; in this construct, parentheses go around the braces. For
14321 ({ int y = foo (); int z;
14326 is a valid (though slightly more complex than necessary) expression for
14327 the absolute value of `foo ()'.
14329 The last thing in the compound statement should be an expression
14330 followed by a semicolon; the value of this subexpression serves as the
14331 value of the entire construct. (If you use some other kind of statement
14332 last within the braces, the construct has type `void', and thus
14333 effectively no value.)
14335 This feature is especially useful in making macro definitions "safe"
14336 (so that they evaluate each operand exactly once). For example, the
14337 "maximum" function is commonly defined as a macro in standard C as
14340 #define max(a,b) ((a) > (b) ? (a) : (b))
14342 But this definition computes either A or B twice, with bad results if
14343 the operand has side effects. In GNU C, if you know the type of the
14344 operands (here taken as `int'), you can define the macro safely as
14347 #define maxint(a,b) \
14348 ({int _a = (a), _b = (b); _a > _b ? _a : _b; })
14350 Embedded statements are not allowed in constant expressions, such as
14351 the value of an enumeration constant, the width of a bit-field, or the
14352 initial value of a static variable.
14354 If you don't know the type of the operand, you can still do this, but
14355 you must use `typeof' (*note Typeof::).
14357 In G++, the result value of a statement expression undergoes array and
14358 function pointer decay, and is returned by value to the enclosing
14359 expression. For instance, if `A' is a class, then
14365 will construct a temporary `A' object to hold the result of the
14366 statement expression, and that will be used to invoke `Foo'. Therefore
14367 the `this' pointer observed by `Foo' will not be the address of `a'.
14369 Any temporaries created within a statement within a statement
14370 expression will be destroyed at the statement's end. This makes
14371 statement expressions inside macros slightly different from function
14372 calls. In the latter case temporaries introduced during argument
14373 evaluation will be destroyed at the end of the statement that includes
14374 the function call. In the statement expression case they will be
14375 destroyed during the statement expression. For instance,
14377 #define macro(a) ({__typeof__(a) b = (a); b + 3; })
14378 template<typename T> T function(T a) { T b = a; return b + 3; }
14386 will have different places where temporaries are destroyed. For the
14387 `macro' case, the temporary `X' will be destroyed just after the
14388 initialization of `b'. In the `function' case that temporary will be
14389 destroyed when the function returns.
14391 These considerations mean that it is probably a bad idea to use
14392 statement-expressions of this form in header files that are designed to
14393 work with C++. (Note that some versions of the GNU C Library contained
14394 header files using statement-expression that lead to precisely this
14397 Jumping into a statement expression with `goto' or using a `switch'
14398 statement outside the statement expression with a `case' or `default'
14399 label inside the statement expression is not permitted. Jumping into a
14400 statement expression with a computed `goto' (*note Labels as Values::)
14401 yields undefined behavior. Jumping out of a statement expression is
14402 permitted, but if the statement expression is part of a larger
14403 expression then it is unspecified which other subexpressions of that
14404 expression have been evaluated except where the language definition
14405 requires certain subexpressions to be evaluated before or after the
14406 statement expression. In any case, as with a function call the
14407 evaluation of a statement expression is not interleaved with the
14408 evaluation of other parts of the containing expression. For example,
14410 foo (), (({ bar1 (); goto a; 0; }) + bar2 ()), baz();
14412 will call `foo' and `bar1' and will not call `baz' but may or may not
14413 call `bar2'. If `bar2' is called, it will be called after `foo' and
14417 File: gcc.info, Node: Local Labels, Next: Labels as Values, Prev: Statement Exprs, Up: C Extensions
14419 5.2 Locally Declared Labels
14420 ===========================
14422 GCC allows you to declare "local labels" in any nested block scope. A
14423 local label is just like an ordinary label, but you can only reference
14424 it (with a `goto' statement, or by taking its address) within the block
14425 in which it was declared.
14427 A local label declaration looks like this:
14433 __label__ LABEL1, LABEL2, /* ... */;
14435 Local label declarations must come at the beginning of the block,
14436 before any ordinary declarations or statements.
14438 The label declaration defines the label _name_, but does not define
14439 the label itself. You must do this in the usual way, with `LABEL:',
14440 within the statements of the statement expression.
14442 The local label feature is useful for complex macros. If a macro
14443 contains nested loops, a `goto' can be useful for breaking out of them.
14444 However, an ordinary label whose scope is the whole function cannot be
14445 used: if the macro can be expanded several times in one function, the
14446 label will be multiply defined in that function. A local label avoids
14447 this problem. For example:
14449 #define SEARCH(value, array, target) \
14452 typeof (target) _SEARCH_target = (target); \
14453 typeof (*(array)) *_SEARCH_array = (array); \
14456 for (i = 0; i < max; i++) \
14457 for (j = 0; j < max; j++) \
14458 if (_SEARCH_array[i][j] == _SEARCH_target) \
14459 { (value) = i; goto found; } \
14464 This could also be written using a statement-expression:
14466 #define SEARCH(array, target) \
14469 typeof (target) _SEARCH_target = (target); \
14470 typeof (*(array)) *_SEARCH_array = (array); \
14473 for (i = 0; i < max; i++) \
14474 for (j = 0; j < max; j++) \
14475 if (_SEARCH_array[i][j] == _SEARCH_target) \
14476 { value = i; goto found; } \
14482 Local label declarations also make the labels they declare visible to
14483 nested functions, if there are any. *Note Nested Functions::, for
14487 File: gcc.info, Node: Labels as Values, Next: Nested Functions, Prev: Local Labels, Up: C Extensions
14489 5.3 Labels as Values
14490 ====================
14492 You can get the address of a label defined in the current function (or
14493 a containing function) with the unary operator `&&'. The value has
14494 type `void *'. This value is a constant and can be used wherever a
14495 constant of that type is valid. For example:
14501 To use these values, you need to be able to jump to one. This is done
14502 with the computed goto statement(1), `goto *EXP;'. For example,
14506 Any expression of type `void *' is allowed.
14508 One way of using these constants is in initializing a static array that
14509 will serve as a jump table:
14511 static void *array[] = { &&foo, &&bar, &&hack };
14513 Then you can select a label with indexing, like this:
14517 Note that this does not check whether the subscript is in bounds--array
14518 indexing in C never does that.
14520 Such an array of label values serves a purpose much like that of the
14521 `switch' statement. The `switch' statement is cleaner, so use that
14522 rather than an array unless the problem does not fit a `switch'
14523 statement very well.
14525 Another use of label values is in an interpreter for threaded code.
14526 The labels within the interpreter function can be stored in the
14527 threaded code for super-fast dispatching.
14529 You may not use this mechanism to jump to code in a different function.
14530 If you do that, totally unpredictable things will happen. The best way
14531 to avoid this is to store the label address only in automatic variables
14532 and never pass it as an argument.
14534 An alternate way to write the above example is
14536 static const int array[] = { &&foo - &&foo, &&bar - &&foo,
14538 goto *(&&foo + array[i]);
14540 This is more friendly to code living in shared libraries, as it reduces
14541 the number of dynamic relocations that are needed, and by consequence,
14542 allows the data to be read-only.
14544 ---------- Footnotes ----------
14546 (1) The analogous feature in Fortran is called an assigned goto, but
14547 that name seems inappropriate in C, where one can do more than simply
14548 store label addresses in label variables.
14551 File: gcc.info, Node: Nested Functions, Next: Constructing Calls, Prev: Labels as Values, Up: C Extensions
14553 5.4 Nested Functions
14554 ====================
14556 A "nested function" is a function defined inside another function.
14557 (Nested functions are not supported for GNU C++.) The nested function's
14558 name is local to the block where it is defined. For example, here we
14559 define a nested function named `square', and call it twice:
14561 foo (double a, double b)
14563 double square (double z) { return z * z; }
14565 return square (a) + square (b);
14568 The nested function can access all the variables of the containing
14569 function that are visible at the point of its definition. This is
14570 called "lexical scoping". For example, here we show a nested function
14571 which uses an inherited variable named `offset':
14573 bar (int *array, int offset, int size)
14575 int access (int *array, int index)
14576 { return array[index + offset]; }
14579 for (i = 0; i < size; i++)
14580 /* ... */ access (array, i) /* ... */
14583 Nested function definitions are permitted within functions in the
14584 places where variable definitions are allowed; that is, in any block,
14585 mixed with the other declarations and statements in the block.
14587 It is possible to call the nested function from outside the scope of
14588 its name by storing its address or passing the address to another
14591 hack (int *array, int size)
14593 void store (int index, int value)
14594 { array[index] = value; }
14596 intermediate (store, size);
14599 Here, the function `intermediate' receives the address of `store' as
14600 an argument. If `intermediate' calls `store', the arguments given to
14601 `store' are used to store into `array'. But this technique works only
14602 so long as the containing function (`hack', in this example) does not
14605 If you try to call the nested function through its address after the
14606 containing function has exited, all hell will break loose. If you try
14607 to call it after a containing scope level has exited, and if it refers
14608 to some of the variables that are no longer in scope, you may be lucky,
14609 but it's not wise to take the risk. If, however, the nested function
14610 does not refer to anything that has gone out of scope, you should be
14613 GCC implements taking the address of a nested function using a
14614 technique called "trampolines". A paper describing them is available as
14616 `http://people.debian.org/~aaronl/Usenix88-lexic.pdf'.
14618 A nested function can jump to a label inherited from a containing
14619 function, provided the label was explicitly declared in the containing
14620 function (*note Local Labels::). Such a jump returns instantly to the
14621 containing function, exiting the nested function which did the `goto'
14622 and any intermediate functions as well. Here is an example:
14624 bar (int *array, int offset, int size)
14627 int access (int *array, int index)
14631 return array[index + offset];
14635 for (i = 0; i < size; i++)
14636 /* ... */ access (array, i) /* ... */
14640 /* Control comes here from `access'
14641 if it detects an error. */
14646 A nested function always has no linkage. Declaring one with `extern'
14647 or `static' is erroneous. If you need to declare the nested function
14648 before its definition, use `auto' (which is otherwise meaningless for
14649 function declarations).
14651 bar (int *array, int offset, int size)
14654 auto int access (int *, int);
14656 int access (int *array, int index)
14660 return array[index + offset];
14666 File: gcc.info, Node: Constructing Calls, Next: Typeof, Prev: Nested Functions, Up: C Extensions
14668 5.5 Constructing Function Calls
14669 ===============================
14671 Using the built-in functions described below, you can record the
14672 arguments a function received, and call another function with the same
14673 arguments, without knowing the number or types of the arguments.
14675 You can also record the return value of that function call, and later
14676 return that value, without knowing what data type the function tried to
14677 return (as long as your caller expects that data type).
14679 However, these built-in functions may interact badly with some
14680 sophisticated features or other extensions of the language. It is,
14681 therefore, not recommended to use them outside very simple functions
14682 acting as mere forwarders for their arguments.
14684 -- Built-in Function: void * __builtin_apply_args ()
14685 This built-in function returns a pointer to data describing how to
14686 perform a call with the same arguments as were passed to the
14689 The function saves the arg pointer register, structure value
14690 address, and all registers that might be used to pass arguments to
14691 a function into a block of memory allocated on the stack. Then it
14692 returns the address of that block.
14694 -- Built-in Function: void * __builtin_apply (void (*FUNCTION)(), void
14695 *ARGUMENTS, size_t SIZE)
14696 This built-in function invokes FUNCTION with a copy of the
14697 parameters described by ARGUMENTS and SIZE.
14699 The value of ARGUMENTS should be the value returned by
14700 `__builtin_apply_args'. The argument SIZE specifies the size of
14701 the stack argument data, in bytes.
14703 This function returns a pointer to data describing how to return
14704 whatever value was returned by FUNCTION. The data is saved in a
14705 block of memory allocated on the stack.
14707 It is not always simple to compute the proper value for SIZE. The
14708 value is used by `__builtin_apply' to compute the amount of data
14709 that should be pushed on the stack and copied from the incoming
14712 -- Built-in Function: void __builtin_return (void *RESULT)
14713 This built-in function returns the value described by RESULT from
14714 the containing function. You should specify, for RESULT, a value
14715 returned by `__builtin_apply'.
14718 File: gcc.info, Node: Typeof, Next: Conditionals, Prev: Constructing Calls, Up: C Extensions
14720 5.6 Referring to a Type with `typeof'
14721 =====================================
14723 Another way to refer to the type of an expression is with `typeof'.
14724 The syntax of using of this keyword looks like `sizeof', but the
14725 construct acts semantically like a type name defined with `typedef'.
14727 There are two ways of writing the argument to `typeof': with an
14728 expression or with a type. Here is an example with an expression:
14732 This assumes that `x' is an array of pointers to functions; the type
14733 described is that of the values of the functions.
14735 Here is an example with a typename as the argument:
14739 Here the type described is that of pointers to `int'.
14741 If you are writing a header file that must work when included in ISO C
14742 programs, write `__typeof__' instead of `typeof'. *Note Alternate
14745 A `typeof'-construct can be used anywhere a typedef name could be
14746 used. For example, you can use it in a declaration, in a cast, or
14747 inside of `sizeof' or `typeof'.
14749 `typeof' is often useful in conjunction with the
14750 statements-within-expressions feature. Here is how the two together can
14751 be used to define a safe "maximum" macro that operates on any
14752 arithmetic type and evaluates each of its arguments exactly once:
14755 ({ typeof (a) _a = (a); \
14756 typeof (b) _b = (b); \
14757 _a > _b ? _a : _b; })
14759 The reason for using names that start with underscores for the local
14760 variables is to avoid conflicts with variable names that occur within
14761 the expressions that are substituted for `a' and `b'. Eventually we
14762 hope to design a new form of declaration syntax that allows you to
14763 declare variables whose scopes start only after their initializers;
14764 this will be a more reliable way to prevent such conflicts.
14766 Some more examples of the use of `typeof':
14768 * This declares `y' with the type of what `x' points to.
14772 * This declares `y' as an array of such values.
14776 * This declares `y' as an array of pointers to characters:
14778 typeof (typeof (char *)[4]) y;
14780 It is equivalent to the following traditional C declaration:
14784 To see the meaning of the declaration using `typeof', and why it
14785 might be a useful way to write, rewrite it with these macros:
14787 #define pointer(T) typeof(T *)
14788 #define array(T, N) typeof(T [N])
14790 Now the declaration can be rewritten this way:
14792 array (pointer (char), 4) y;
14794 Thus, `array (pointer (char), 4)' is the type of arrays of 4
14795 pointers to `char'.
14797 _Compatibility Note:_ In addition to `typeof', GCC 2 supported a more
14798 limited extension which permitted one to write
14802 with the effect of declaring T to have the type of the expression EXPR.
14803 This extension does not work with GCC 3 (versions between 3.0 and 3.2
14804 will crash; 3.2.1 and later give an error). Code which relies on it
14805 should be rewritten to use `typeof':
14807 typedef typeof(EXPR) T;
14809 This will work with all versions of GCC.
14812 File: gcc.info, Node: Conditionals, Next: Long Long, Prev: Typeof, Up: C Extensions
14814 5.7 Conditionals with Omitted Operands
14815 ======================================
14817 The middle operand in a conditional expression may be omitted. Then if
14818 the first operand is nonzero, its value is the value of the conditional
14821 Therefore, the expression
14825 has the value of `x' if that is nonzero; otherwise, the value of `y'.
14827 This example is perfectly equivalent to
14831 In this simple case, the ability to omit the middle operand is not
14832 especially useful. When it becomes useful is when the first operand
14833 does, or may (if it is a macro argument), contain a side effect. Then
14834 repeating the operand in the middle would perform the side effect
14835 twice. Omitting the middle operand uses the value already computed
14836 without the undesirable effects of recomputing it.
14839 File: gcc.info, Node: Long Long, Next: Complex, Prev: Conditionals, Up: C Extensions
14841 5.8 Double-Word Integers
14842 ========================
14844 ISO C99 supports data types for integers that are at least 64 bits wide,
14845 and as an extension GCC supports them in C89 mode and in C++. Simply
14846 write `long long int' for a signed integer, or `unsigned long long int'
14847 for an unsigned integer. To make an integer constant of type `long
14848 long int', add the suffix `LL' to the integer. To make an integer
14849 constant of type `unsigned long long int', add the suffix `ULL' to the
14852 You can use these types in arithmetic like any other integer types.
14853 Addition, subtraction, and bitwise boolean operations on these types
14854 are open-coded on all types of machines. Multiplication is open-coded
14855 if the machine supports fullword-to-doubleword a widening multiply
14856 instruction. Division and shifts are open-coded only on machines that
14857 provide special support. The operations that are not open-coded use
14858 special library routines that come with GCC.
14860 There may be pitfalls when you use `long long' types for function
14861 arguments, unless you declare function prototypes. If a function
14862 expects type `int' for its argument, and you pass a value of type `long
14863 long int', confusion will result because the caller and the subroutine
14864 will disagree about the number of bytes for the argument. Likewise, if
14865 the function expects `long long int' and you pass `int'. The best way
14866 to avoid such problems is to use prototypes.
14869 File: gcc.info, Node: Complex, Next: Decimal Float, Prev: Long Long, Up: C Extensions
14871 5.9 Complex Numbers
14872 ===================
14874 ISO C99 supports complex floating data types, and as an extension GCC
14875 supports them in C89 mode and in C++, and supports complex integer data
14876 types which are not part of ISO C99. You can declare complex types
14877 using the keyword `_Complex'. As an extension, the older GNU keyword
14878 `__complex__' is also supported.
14880 For example, `_Complex double x;' declares `x' as a variable whose
14881 real part and imaginary part are both of type `double'. `_Complex
14882 short int y;' declares `y' to have real and imaginary parts of type
14883 `short int'; this is not likely to be useful, but it shows that the set
14884 of complex types is complete.
14886 To write a constant with a complex data type, use the suffix `i' or
14887 `j' (either one; they are equivalent). For example, `2.5fi' has type
14888 `_Complex float' and `3i' has type `_Complex int'. Such a constant
14889 always has a pure imaginary value, but you can form any complex value
14890 you like by adding one to a real constant. This is a GNU extension; if
14891 you have an ISO C99 conforming C library (such as GNU libc), and want
14892 to construct complex constants of floating type, you should include
14893 `<complex.h>' and use the macros `I' or `_Complex_I' instead.
14895 To extract the real part of a complex-valued expression EXP, write
14896 `__real__ EXP'. Likewise, use `__imag__' to extract the imaginary
14897 part. This is a GNU extension; for values of floating type, you should
14898 use the ISO C99 functions `crealf', `creal', `creall', `cimagf',
14899 `cimag' and `cimagl', declared in `<complex.h>' and also provided as
14900 built-in functions by GCC.
14902 The operator `~' performs complex conjugation when used on a value
14903 with a complex type. This is a GNU extension; for values of floating
14904 type, you should use the ISO C99 functions `conjf', `conj' and `conjl',
14905 declared in `<complex.h>' and also provided as built-in functions by
14908 GCC can allocate complex automatic variables in a noncontiguous
14909 fashion; it's even possible for the real part to be in a register while
14910 the imaginary part is on the stack (or vice-versa). Only the DWARF2
14911 debug info format can represent this, so use of DWARF2 is recommended.
14912 If you are using the stabs debug info format, GCC describes a
14913 noncontiguous complex variable as if it were two separate variables of
14914 noncomplex type. If the variable's actual name is `foo', the two
14915 fictitious variables are named `foo$real' and `foo$imag'. You can
14916 examine and set these two fictitious variables with your debugger.
14919 File: gcc.info, Node: Decimal Float, Next: Hex Floats, Prev: Complex, Up: C Extensions
14921 5.10 Decimal Floating Types
14922 ===========================
14924 As an extension, the GNU C compiler supports decimal floating types as
14925 defined in the N1176 draft of ISO/IEC WDTR24732. Support for decimal
14926 floating types in GCC will evolve as the draft technical report changes.
14927 Calling conventions for any target might also change. Not all targets
14928 support decimal floating types.
14930 The decimal floating types are `_Decimal32', `_Decimal64', and
14931 `_Decimal128'. They use a radix of ten, unlike the floating types
14932 `float', `double', and `long double' whose radix is not specified by
14933 the C standard but is usually two.
14935 Support for decimal floating types includes the arithmetic operators
14936 add, subtract, multiply, divide; unary arithmetic operators; relational
14937 operators; equality operators; and conversions to and from integer and
14938 other floating types. Use a suffix `df' or `DF' in a literal constant
14939 of type `_Decimal32', `dd' or `DD' for `_Decimal64', and `dl' or `DL'
14942 GCC support of decimal float as specified by the draft technical report
14945 * Translation time data type (TTDT) is not supported.
14947 * Characteristics of decimal floating types are defined in header
14948 file `decfloat.h' rather than `float.h'.
14950 * When the value of a decimal floating type cannot be represented in
14951 the integer type to which it is being converted, the result is
14952 undefined rather than the result value specified by the draft
14955 Types `_Decimal32', `_Decimal64', and `_Decimal128' are supported by
14956 the DWARF2 debug information format.
14959 File: gcc.info, Node: Hex Floats, Next: Zero Length, Prev: Decimal Float, Up: C Extensions
14964 ISO C99 supports floating-point numbers written not only in the usual
14965 decimal notation, such as `1.55e1', but also numbers such as `0x1.fp3'
14966 written in hexadecimal format. As a GNU extension, GCC supports this
14967 in C89 mode (except in some cases when strictly conforming) and in C++.
14968 In that format the `0x' hex introducer and the `p' or `P' exponent
14969 field are mandatory. The exponent is a decimal number that indicates
14970 the power of 2 by which the significant part will be multiplied. Thus
14971 `0x1.f' is 1 15/16, `p3' multiplies it by 8, and the value of `0x1.fp3'
14972 is the same as `1.55e1'.
14974 Unlike for floating-point numbers in the decimal notation the exponent
14975 is always required in the hexadecimal notation. Otherwise the compiler
14976 would not be able to resolve the ambiguity of, e.g., `0x1.f'. This
14977 could mean `1.0f' or `1.9375' since `f' is also the extension for
14978 floating-point constants of type `float'.
14981 File: gcc.info, Node: Zero Length, Next: Variable Length, Prev: Hex Floats, Up: C Extensions
14983 5.12 Arrays of Length Zero
14984 ==========================
14986 Zero-length arrays are allowed in GNU C. They are very useful as the
14987 last element of a structure which is really a header for a
14988 variable-length object:
14995 struct line *thisline = (struct line *)
14996 malloc (sizeof (struct line) + this_length);
14997 thisline->length = this_length;
14999 In ISO C90, you would have to give `contents' a length of 1, which
15000 means either you waste space or complicate the argument to `malloc'.
15002 In ISO C99, you would use a "flexible array member", which is slightly
15003 different in syntax and semantics:
15005 * Flexible array members are written as `contents[]' without the `0'.
15007 * Flexible array members have incomplete type, and so the `sizeof'
15008 operator may not be applied. As a quirk of the original
15009 implementation of zero-length arrays, `sizeof' evaluates to zero.
15011 * Flexible array members may only appear as the last member of a
15012 `struct' that is otherwise non-empty.
15014 * A structure containing a flexible array member, or a union
15015 containing such a structure (possibly recursively), may not be a
15016 member of a structure or an element of an array. (However, these
15017 uses are permitted by GCC as extensions.)
15019 GCC versions before 3.0 allowed zero-length arrays to be statically
15020 initialized, as if they were flexible arrays. In addition to those
15021 cases that were useful, it also allowed initializations in situations
15022 that would corrupt later data. Non-empty initialization of zero-length
15023 arrays is now treated like any case where there are more initializer
15024 elements than the array holds, in that a suitable warning about "excess
15025 elements in array" is given, and the excess elements (all of them, in
15026 this case) are ignored.
15028 Instead GCC allows static initialization of flexible array members.
15029 This is equivalent to defining a new structure containing the original
15030 structure followed by an array of sufficient size to contain the data.
15031 I.e. in the following, `f1' is constructed as if it were declared like
15036 } f1 = { 1, { 2, 3, 4 } };
15039 struct f1 f1; int data[3];
15040 } f2 = { { 1 }, { 2, 3, 4 } };
15042 The convenience of this extension is that `f1' has the desired type,
15043 eliminating the need to consistently refer to `f2.f1'.
15045 This has symmetry with normal static arrays, in that an array of
15046 unknown size is also written with `[]'.
15048 Of course, this extension only makes sense if the extra data comes at
15049 the end of a top-level object, as otherwise we would be overwriting
15050 data at subsequent offsets. To avoid undue complication and confusion
15051 with initialization of deeply nested arrays, we simply disallow any
15052 non-empty initialization except when the structure is the top-level
15053 object. For example:
15055 struct foo { int x; int y[]; };
15056 struct bar { struct foo z; };
15058 struct foo a = { 1, { 2, 3, 4 } }; // Valid.
15059 struct bar b = { { 1, { 2, 3, 4 } } }; // Invalid.
15060 struct bar c = { { 1, { } } }; // Valid.
15061 struct foo d[1] = { { 1 { 2, 3, 4 } } }; // Invalid.
15064 File: gcc.info, Node: Empty Structures, Next: Variadic Macros, Prev: Variable Length, Up: C Extensions
15066 5.13 Structures With No Members
15067 ===============================
15069 GCC permits a C structure to have no members:
15074 The structure will have size zero. In C++, empty structures are part
15075 of the language. G++ treats empty structures as if they had a single
15076 member of type `char'.
15079 File: gcc.info, Node: Variable Length, Next: Empty Structures, Prev: Zero Length, Up: C Extensions
15081 5.14 Arrays of Variable Length
15082 ==============================
15084 Variable-length automatic arrays are allowed in ISO C99, and as an
15085 extension GCC accepts them in C89 mode and in C++. (However, GCC's
15086 implementation of variable-length arrays does not yet conform in detail
15087 to the ISO C99 standard.) These arrays are declared like any other
15088 automatic arrays, but with a length that is not a constant expression.
15089 The storage is allocated at the point of declaration and deallocated
15090 when the brace-level is exited. For example:
15093 concat_fopen (char *s1, char *s2, char *mode)
15095 char str[strlen (s1) + strlen (s2) + 1];
15098 return fopen (str, mode);
15101 Jumping or breaking out of the scope of the array name deallocates the
15102 storage. Jumping into the scope is not allowed; you get an error
15105 You can use the function `alloca' to get an effect much like
15106 variable-length arrays. The function `alloca' is available in many
15107 other C implementations (but not in all). On the other hand,
15108 variable-length arrays are more elegant.
15110 There are other differences between these two methods. Space allocated
15111 with `alloca' exists until the containing _function_ returns. The
15112 space for a variable-length array is deallocated as soon as the array
15113 name's scope ends. (If you use both variable-length arrays and
15114 `alloca' in the same function, deallocation of a variable-length array
15115 will also deallocate anything more recently allocated with `alloca'.)
15117 You can also use variable-length arrays as arguments to functions:
15120 tester (int len, char data[len][len])
15125 The length of an array is computed once when the storage is allocated
15126 and is remembered for the scope of the array in case you access it with
15129 If you want to pass the array first and the length afterward, you can
15130 use a forward declaration in the parameter list--another GNU extension.
15133 tester (int len; char data[len][len], int len)
15138 The `int len' before the semicolon is a "parameter forward
15139 declaration", and it serves the purpose of making the name `len' known
15140 when the declaration of `data' is parsed.
15142 You can write any number of such parameter forward declarations in the
15143 parameter list. They can be separated by commas or semicolons, but the
15144 last one must end with a semicolon, which is followed by the "real"
15145 parameter declarations. Each forward declaration must match a "real"
15146 declaration in parameter name and data type. ISO C99 does not support
15147 parameter forward declarations.
15150 File: gcc.info, Node: Variadic Macros, Next: Escaped Newlines, Prev: Empty Structures, Up: C Extensions
15152 5.15 Macros with a Variable Number of Arguments.
15153 ================================================
15155 In the ISO C standard of 1999, a macro can be declared to accept a
15156 variable number of arguments much as a function can. The syntax for
15157 defining the macro is similar to that of a function. Here is an
15160 #define debug(format, ...) fprintf (stderr, format, __VA_ARGS__)
15162 Here `...' is a "variable argument". In the invocation of such a
15163 macro, it represents the zero or more tokens until the closing
15164 parenthesis that ends the invocation, including any commas. This set of
15165 tokens replaces the identifier `__VA_ARGS__' in the macro body wherever
15166 it appears. See the CPP manual for more information.
15168 GCC has long supported variadic macros, and used a different syntax
15169 that allowed you to give a name to the variable arguments just like any
15170 other argument. Here is an example:
15172 #define debug(format, args...) fprintf (stderr, format, args)
15174 This is in all ways equivalent to the ISO C example above, but arguably
15175 more readable and descriptive.
15177 GNU CPP has two further variadic macro extensions, and permits them to
15178 be used with either of the above forms of macro definition.
15180 In standard C, you are not allowed to leave the variable argument out
15181 entirely; but you are allowed to pass an empty argument. For example,
15182 this invocation is invalid in ISO C, because there is no comma after
15185 debug ("A message")
15187 GNU CPP permits you to completely omit the variable arguments in this
15188 way. In the above examples, the compiler would complain, though since
15189 the expansion of the macro still has the extra comma after the format
15192 To help solve this problem, CPP behaves specially for variable
15193 arguments used with the token paste operator, `##'. If instead you
15196 #define debug(format, ...) fprintf (stderr, format, ## __VA_ARGS__)
15198 and if the variable arguments are omitted or empty, the `##' operator
15199 causes the preprocessor to remove the comma before it. If you do
15200 provide some variable arguments in your macro invocation, GNU CPP does
15201 not complain about the paste operation and instead places the variable
15202 arguments after the comma. Just like any other pasted macro argument,
15203 these arguments are not macro expanded.
15206 File: gcc.info, Node: Escaped Newlines, Next: Subscripting, Prev: Variadic Macros, Up: C Extensions
15208 5.16 Slightly Looser Rules for Escaped Newlines
15209 ===============================================
15211 Recently, the preprocessor has relaxed its treatment of escaped
15212 newlines. Previously, the newline had to immediately follow a
15213 backslash. The current implementation allows whitespace in the form of
15214 spaces, horizontal and vertical tabs, and form feeds between the
15215 backslash and the subsequent newline. The preprocessor issues a
15216 warning, but treats it as a valid escaped newline and combines the two
15217 lines to form a single logical line. This works within comments and
15218 tokens, as well as between tokens. Comments are _not_ treated as
15219 whitespace for the purposes of this relaxation, since they have not yet
15220 been replaced with spaces.
15223 File: gcc.info, Node: Subscripting, Next: Pointer Arith, Prev: Escaped Newlines, Up: C Extensions
15225 5.17 Non-Lvalue Arrays May Have Subscripts
15226 ==========================================
15228 In ISO C99, arrays that are not lvalues still decay to pointers, and
15229 may be subscripted, although they may not be modified or used after the
15230 next sequence point and the unary `&' operator may not be applied to
15231 them. As an extension, GCC allows such arrays to be subscripted in C89
15232 mode, though otherwise they do not decay to pointers outside C99 mode.
15233 For example, this is valid in GNU C though not valid in C89:
15235 struct foo {int a[4];};
15241 return f().a[index];
15245 File: gcc.info, Node: Pointer Arith, Next: Initializers, Prev: Subscripting, Up: C Extensions
15247 5.18 Arithmetic on `void'- and Function-Pointers
15248 ================================================
15250 In GNU C, addition and subtraction operations are supported on pointers
15251 to `void' and on pointers to functions. This is done by treating the
15252 size of a `void' or of a function as 1.
15254 A consequence of this is that `sizeof' is also allowed on `void' and
15255 on function types, and returns 1.
15257 The option `-Wpointer-arith' requests a warning if these extensions
15261 File: gcc.info, Node: Initializers, Next: Compound Literals, Prev: Pointer Arith, Up: C Extensions
15263 5.19 Non-Constant Initializers
15264 ==============================
15266 As in standard C++ and ISO C99, the elements of an aggregate
15267 initializer for an automatic variable are not required to be constant
15268 expressions in GNU C. Here is an example of an initializer with
15269 run-time varying elements:
15271 foo (float f, float g)
15273 float beat_freqs[2] = { f-g, f+g };
15278 File: gcc.info, Node: Compound Literals, Next: Designated Inits, Prev: Initializers, Up: C Extensions
15280 5.20 Compound Literals
15281 ======================
15283 ISO C99 supports compound literals. A compound literal looks like a
15284 cast containing an initializer. Its value is an object of the type
15285 specified in the cast, containing the elements specified in the
15286 initializer; it is an lvalue. As an extension, GCC supports compound
15287 literals in C89 mode and in C++.
15289 Usually, the specified type is a structure. Assume that `struct foo'
15290 and `structure' are declared as shown:
15292 struct foo {int a; char b[2];} structure;
15294 Here is an example of constructing a `struct foo' with a compound
15297 structure = ((struct foo) {x + y, 'a', 0});
15299 This is equivalent to writing the following:
15302 struct foo temp = {x + y, 'a', 0};
15306 You can also construct an array. If all the elements of the compound
15307 literal are (made up of) simple constant expressions, suitable for use
15308 in initializers of objects of static storage duration, then the compound
15309 literal can be coerced to a pointer to its first element and used in
15310 such an initializer, as shown here:
15312 char **foo = (char *[]) { "x", "y", "z" };
15314 Compound literals for scalar types and union types are is also
15315 allowed, but then the compound literal is equivalent to a cast.
15317 As a GNU extension, GCC allows initialization of objects with static
15318 storage duration by compound literals (which is not possible in ISO
15319 C99, because the initializer is not a constant). It is handled as if
15320 the object was initialized only with the bracket enclosed list if the
15321 types of the compound literal and the object match. The initializer
15322 list of the compound literal must be constant. If the object being
15323 initialized has array type of unknown size, the size is determined by
15324 compound literal size.
15326 static struct foo x = (struct foo) {1, 'a', 'b'};
15327 static int y[] = (int []) {1, 2, 3};
15328 static int z[] = (int [3]) {1};
15330 The above lines are equivalent to the following:
15331 static struct foo x = {1, 'a', 'b'};
15332 static int y[] = {1, 2, 3};
15333 static int z[] = {1, 0, 0};
15336 File: gcc.info, Node: Designated Inits, Next: Cast to Union, Prev: Compound Literals, Up: C Extensions
15338 5.21 Designated Initializers
15339 ============================
15341 Standard C89 requires the elements of an initializer to appear in a
15342 fixed order, the same as the order of the elements in the array or
15343 structure being initialized.
15345 In ISO C99 you can give the elements in any order, specifying the array
15346 indices or structure field names they apply to, and GNU C allows this as
15347 an extension in C89 mode as well. This extension is not implemented in
15350 To specify an array index, write `[INDEX] =' before the element value.
15353 int a[6] = { [4] = 29, [2] = 15 };
15357 int a[6] = { 0, 0, 15, 0, 29, 0 };
15359 The index values must be constant expressions, even if the array being
15360 initialized is automatic.
15362 An alternative syntax for this which has been obsolete since GCC 2.5
15363 but GCC still accepts is to write `[INDEX]' before the element value,
15366 To initialize a range of elements to the same value, write `[FIRST ...
15367 LAST] = VALUE'. This is a GNU extension. For example,
15369 int widths[] = { [0 ... 9] = 1, [10 ... 99] = 2, [100] = 3 };
15371 If the value in it has side-effects, the side-effects will happen only
15372 once, not for each initialized field by the range initializer.
15374 Note that the length of the array is the highest value specified plus
15377 In a structure initializer, specify the name of a field to initialize
15378 with `.FIELDNAME =' before the element value. For example, given the
15379 following structure,
15381 struct point { int x, y; };
15383 the following initialization
15385 struct point p = { .y = yvalue, .x = xvalue };
15389 struct point p = { xvalue, yvalue };
15391 Another syntax which has the same meaning, obsolete since GCC 2.5, is
15392 `FIELDNAME:', as shown here:
15394 struct point p = { y: yvalue, x: xvalue };
15396 The `[INDEX]' or `.FIELDNAME' is known as a "designator". You can
15397 also use a designator (or the obsolete colon syntax) when initializing
15398 a union, to specify which element of the union should be used. For
15401 union foo { int i; double d; };
15403 union foo f = { .d = 4 };
15405 will convert 4 to a `double' to store it in the union using the second
15406 element. By contrast, casting 4 to type `union foo' would store it
15407 into the union as the integer `i', since it is an integer. (*Note Cast
15410 You can combine this technique of naming elements with ordinary C
15411 initialization of successive elements. Each initializer element that
15412 does not have a designator applies to the next consecutive element of
15413 the array or structure. For example,
15415 int a[6] = { [1] = v1, v2, [4] = v4 };
15419 int a[6] = { 0, v1, v2, 0, v4, 0 };
15421 Labeling the elements of an array initializer is especially useful
15422 when the indices are characters or belong to an `enum' type. For
15425 int whitespace[256]
15426 = { [' '] = 1, ['\t'] = 1, ['\h'] = 1,
15427 ['\f'] = 1, ['\n'] = 1, ['\r'] = 1 };
15429 You can also write a series of `.FIELDNAME' and `[INDEX]' designators
15430 before an `=' to specify a nested subobject to initialize; the list is
15431 taken relative to the subobject corresponding to the closest
15432 surrounding brace pair. For example, with the `struct point'
15435 struct point ptarray[10] = { [2].y = yv2, [2].x = xv2, [0].x = xv0 };
15437 If the same field is initialized multiple times, it will have value from
15438 the last initialization. If any such overridden initialization has
15439 side-effect, it is unspecified whether the side-effect happens or not.
15440 Currently, GCC will discard them and issue a warning.
15443 File: gcc.info, Node: Case Ranges, Next: Mixed Declarations, Prev: Cast to Union, Up: C Extensions
15448 You can specify a range of consecutive values in a single `case' label,
15453 This has the same effect as the proper number of individual `case'
15454 labels, one for each integer value from LOW to HIGH, inclusive.
15456 This feature is especially useful for ranges of ASCII character codes:
15460 *Be careful:* Write spaces around the `...', for otherwise it may be
15461 parsed wrong when you use it with integer values. For example, write
15471 File: gcc.info, Node: Cast to Union, Next: Case Ranges, Prev: Designated Inits, Up: C Extensions
15473 5.23 Cast to a Union Type
15474 =========================
15476 A cast to union type is similar to other casts, except that the type
15477 specified is a union type. You can specify the type either with `union
15478 TAG' or with a typedef name. A cast to union is actually a constructor
15479 though, not a cast, and hence does not yield an lvalue like normal
15480 casts. (*Note Compound Literals::.)
15482 The types that may be cast to the union type are those of the members
15483 of the union. Thus, given the following union and variables:
15485 union foo { int i; double d; };
15489 both `x' and `y' can be cast to type `union foo'.
15491 Using the cast as the right-hand side of an assignment to a variable of
15492 union type is equivalent to storing in a member of the union:
15496 u = (union foo) x == u.i = x
15497 u = (union foo) y == u.d = y
15499 You can also use the union cast as a function argument:
15501 void hack (union foo);
15503 hack ((union foo) x);
15506 File: gcc.info, Node: Mixed Declarations, Next: Function Attributes, Prev: Case Ranges, Up: C Extensions
15508 5.24 Mixed Declarations and Code
15509 ================================
15511 ISO C99 and ISO C++ allow declarations and code to be freely mixed
15512 within compound statements. As an extension, GCC also allows this in
15513 C89 mode. For example, you could do:
15520 Each identifier is visible from where it is declared until the end of
15521 the enclosing block.
15524 File: gcc.info, Node: Function Attributes, Next: Attribute Syntax, Prev: Mixed Declarations, Up: C Extensions
15526 5.25 Declaring Attributes of Functions
15527 ======================================
15529 In GNU C, you declare certain things about functions called in your
15530 program which help the compiler optimize function calls and check your
15531 code more carefully.
15533 The keyword `__attribute__' allows you to specify special attributes
15534 when making a declaration. This keyword is followed by an attribute
15535 specification inside double parentheses. The following attributes are
15536 currently defined for functions on all targets: `noreturn',
15537 `returns_twice', `noinline', `always_inline', `flatten', `pure',
15538 `const', `nothrow', `sentinel', `format', `format_arg',
15539 `no_instrument_function', `section', `constructor', `destructor',
15540 `used', `unused', `deprecated', `weak', `malloc', `alias',
15541 `warn_unused_result', `nonnull', `gnu_inline' and `externally_visible'.
15542 Several other attributes are defined for functions on particular
15543 target systems. Other attributes, including `section' are supported
15544 for variables declarations (*note Variable Attributes::) and for types
15545 (*note Type Attributes::).
15547 You may also specify attributes with `__' preceding and following each
15548 keyword. This allows you to use them in header files without being
15549 concerned about a possible macro of the same name. For example, you
15550 may use `__noreturn__' instead of `noreturn'.
15552 *Note Attribute Syntax::, for details of the exact syntax for using
15556 The `alias' attribute causes the declaration to be emitted as an
15557 alias for another symbol, which must be specified. For instance,
15559 void __f () { /* Do something. */; }
15560 void f () __attribute__ ((weak, alias ("__f")));
15562 defines `f' to be a weak alias for `__f'. In C++, the mangled
15563 name for the target must be used. It is an error if `__f' is not
15564 defined in the same translation unit.
15566 Not all target machines support this attribute.
15569 Generally, functions are not inlined unless optimization is
15570 specified. For functions declared inline, this attribute inlines
15571 the function even if no optimization level was specified.
15574 This attribute should be used with a function which is also
15575 declared with the `inline' keyword. It directs GCC to treat the
15576 function as if it were defined in gnu89 mode even when compiling
15577 in C99 or gnu99 mode.
15579 If the function is declared `extern', then this definition of the
15580 function is used only for inlining. In no case is the function
15581 compiled as a standalone function, not even if you take its address
15582 explicitly. Such an address becomes an external reference, as if
15583 you had only declared the function, and had not defined it. This
15584 has almost the effect of a macro. The way to use this is to put a
15585 function definition in a header file with this attribute, and put
15586 another copy of the function, without `extern', in a library file.
15587 The definition in the header file will cause most calls to the
15588 function to be inlined. If any uses of the function remain, they
15589 will refer to the single copy in the library. Note that the two
15590 definitions of the functions need not be precisely the same,
15591 although if they do not have the same effect your program may
15594 If the function is neither `extern' nor `static', then the
15595 function is compiled as a standalone function, as well as being
15596 inlined where possible.
15598 This is how GCC traditionally handled functions declared `inline'.
15599 Since ISO C99 specifies a different semantics for `inline', this
15600 function attribute is provided as a transition measure and as a
15601 useful feature in its own right. This attribute is available in
15602 GCC 4.1.3 and later. It is available if either of the
15603 preprocessor macros `__GNUC_GNU_INLINE__' or
15604 `__GNUC_STDC_INLINE__' are defined. *Note An Inline Function is
15605 As Fast As a Macro: Inline.
15607 Note that since the first version of GCC to support C99 inline
15608 semantics is 4.3, earlier versions of GCC which accept this
15609 attribute effectively assume that it is always present, whether or
15610 not it is given explicitly. In versions prior to 4.3, the only
15611 effect of explicitly including it is to disable warnings about
15612 using inline functions in C99 mode.
15615 Generally, inlining into a function is limited. For a function
15616 marked with this attribute, every call inside this function will
15617 be inlined, if possible. Whether the function itself is
15618 considered for inlining depends on its size and the current
15619 inlining parameters. The `flatten' attribute only works reliably
15620 in unit-at-a-time mode.
15623 On the Intel 386, the `cdecl' attribute causes the compiler to
15624 assume that the calling function will pop off the stack space used
15625 to pass arguments. This is useful to override the effects of the
15629 Many functions do not examine any values except their arguments,
15630 and have no effects except the return value. Basically this is
15631 just slightly more strict class than the `pure' attribute below,
15632 since function is not allowed to read global memory.
15634 Note that a function that has pointer arguments and examines the
15635 data pointed to must _not_ be declared `const'. Likewise, a
15636 function that calls a non-`const' function usually must not be
15637 `const'. It does not make sense for a `const' function to return
15640 The attribute `const' is not implemented in GCC versions earlier
15641 than 2.5. An alternative way to declare that a function has no
15642 side effects, which works in the current version and in some older
15643 versions, is as follows:
15645 typedef int intfn ();
15647 extern const intfn square;
15649 This approach does not work in GNU C++ from 2.6.0 on, since the
15650 language specifies that the `const' must be attached to the return
15655 The `constructor' attribute causes the function to be called
15656 automatically before execution enters `main ()'. Similarly, the
15657 `destructor' attribute causes the function to be called
15658 automatically after `main ()' has completed or `exit ()' has been
15659 called. Functions with these attributes are useful for
15660 initializing data that will be used implicitly during the
15661 execution of the program.
15663 These attributes are not currently implemented for Objective-C.
15666 The `deprecated' attribute results in a warning if the function is
15667 used anywhere in the source file. This is useful when identifying
15668 functions that are expected to be removed in a future version of a
15669 program. The warning also includes the location of the declaration
15670 of the deprecated function, to enable users to easily find further
15671 information about why the function is deprecated, or what they
15672 should do instead. Note that the warnings only occurs for uses:
15674 int old_fn () __attribute__ ((deprecated));
15676 int (*fn_ptr)() = old_fn;
15678 results in a warning on line 3 but not line 2.
15680 The `deprecated' attribute can also be used for variables and
15681 types (*note Variable Attributes::, *note Type Attributes::.)
15684 On Microsoft Windows targets and Symbian OS targets the
15685 `dllexport' attribute causes the compiler to provide a global
15686 pointer to a pointer in a DLL, so that it can be referenced with
15687 the `dllimport' attribute. On Microsoft Windows targets, the
15688 pointer name is formed by combining `_imp__' and the function or
15691 You can use `__declspec(dllexport)' as a synonym for
15692 `__attribute__ ((dllexport))' for compatibility with other
15695 On systems that support the `visibility' attribute, this attribute
15696 also implies "default" visibility, unless a `visibility' attribute
15697 is explicitly specified. You should avoid the use of `dllexport'
15698 with "hidden" or "internal" visibility; in the future GCC may
15699 issue an error for those cases.
15701 Currently, the `dllexport' attribute is ignored for inlined
15702 functions, unless the `-fkeep-inline-functions' flag has been
15703 used. The attribute is also ignored for undefined symbols.
15705 When applied to C++ classes, the attribute marks defined
15706 non-inlined member functions and static data members as exports.
15707 Static consts initialized in-class are not marked unless they are
15708 also defined out-of-class.
15710 For Microsoft Windows targets there are alternative methods for
15711 including the symbol in the DLL's export table such as using a
15712 `.def' file with an `EXPORTS' section or, with GNU ld, using the
15713 `--export-all' linker flag.
15716 On Microsoft Windows and Symbian OS targets, the `dllimport'
15717 attribute causes the compiler to reference a function or variable
15718 via a global pointer to a pointer that is set up by the DLL
15719 exporting the symbol. The attribute implies `extern' storage. On
15720 Microsoft Windows targets, the pointer name is formed by combining
15721 `_imp__' and the function or variable name.
15723 You can use `__declspec(dllimport)' as a synonym for
15724 `__attribute__ ((dllimport))' for compatibility with other
15727 Currently, the attribute is ignored for inlined functions. If the
15728 attribute is applied to a symbol _definition_, an error is
15729 reported. If a symbol previously declared `dllimport' is later
15730 defined, the attribute is ignored in subsequent references, and a
15731 warning is emitted. The attribute is also overridden by a
15732 subsequent declaration as `dllexport'.
15734 When applied to C++ classes, the attribute marks non-inlined
15735 member functions and static data members as imports. However, the
15736 attribute is ignored for virtual methods to allow creation of
15737 vtables using thunks.
15739 On the SH Symbian OS target the `dllimport' attribute also has
15740 another affect--it can cause the vtable and run-time type
15741 information for a class to be exported. This happens when the
15742 class has a dllimport'ed constructor or a non-inline, non-pure
15743 virtual function and, for either of those two conditions, the
15744 class also has a inline constructor or destructor and has a key
15745 function that is defined in the current translation unit.
15747 For Microsoft Windows based targets the use of the `dllimport'
15748 attribute on functions is not necessary, but provides a small
15749 performance benefit by eliminating a thunk in the DLL. The use of
15750 the `dllimport' attribute on imported variables was required on
15751 older versions of the GNU linker, but can now be avoided by
15752 passing the `--enable-auto-import' switch to the GNU linker. As
15753 with functions, using the attribute for a variable eliminates a
15756 One drawback to using this attribute is that a pointer to a
15757 function or variable marked as `dllimport' cannot be used as a
15758 constant address. On Microsoft Windows targets, the attribute can
15759 be disabled for functions by setting the `-mnop-fun-dllimport'
15763 Use this attribute on the H8/300, H8/300H, and H8S to indicate
15764 that the specified variable should be placed into the eight bit
15765 data section. The compiler will generate more efficient code for
15766 certain operations on data in the eight bit data area. Note the
15767 eight bit data area is limited to 256 bytes of data.
15769 You must use GAS and GLD from GNU binutils version 2.7 or later for
15770 this attribute to work correctly.
15772 `exception_handler'
15773 Use this attribute on the Blackfin to indicate that the specified
15774 function is an exception handler. The compiler will generate
15775 function entry and exit sequences suitable for use in an exception
15776 handler when this attribute is present.
15779 On 68HC11 and 68HC12 the `far' attribute causes the compiler to
15780 use a calling convention that takes care of switching memory banks
15781 when entering and leaving a function. This calling convention is
15782 also the default when using the `-mlong-calls' option.
15784 On 68HC12 the compiler will use the `call' and `rtc' instructions
15785 to call and return from a function.
15787 On 68HC11 the compiler will generate a sequence of instructions to
15788 invoke a board-specific routine to switch the memory bank and call
15789 the real function. The board-specific routine simulates a `call'.
15790 At the end of a function, it will jump to a board-specific routine
15791 instead of using `rts'. The board-specific return routine
15792 simulates the `rtc'.
15795 On the Intel 386, the `fastcall' attribute causes the compiler to
15796 pass the first argument (if of integral type) in the register ECX
15797 and the second argument (if of integral type) in the register EDX.
15798 Subsequent and other typed arguments are passed on the stack.
15799 The called function will pop the arguments off the stack. If the
15800 number of arguments is variable all arguments are pushed on the
15803 `format (ARCHETYPE, STRING-INDEX, FIRST-TO-CHECK)'
15804 The `format' attribute specifies that a function takes `printf',
15805 `scanf', `strftime' or `strfmon' style arguments which should be
15806 type-checked against a format string. For example, the
15810 my_printf (void *my_object, const char *my_format, ...)
15811 __attribute__ ((format (printf, 2, 3)));
15813 causes the compiler to check the arguments in calls to `my_printf'
15814 for consistency with the `printf' style format string argument
15817 The parameter ARCHETYPE determines how the format string is
15818 interpreted, and should be `printf', `scanf', `strftime' or
15819 `strfmon'. (You can also use `__printf__', `__scanf__',
15820 `__strftime__' or `__strfmon__'.) The parameter STRING-INDEX
15821 specifies which argument is the format string argument (starting
15822 from 1), while FIRST-TO-CHECK is the number of the first argument
15823 to check against the format string. For functions where the
15824 arguments are not available to be checked (such as `vprintf'),
15825 specify the third parameter as zero. In this case the compiler
15826 only checks the format string for consistency. For `strftime'
15827 formats, the third parameter is required to be zero. Since
15828 non-static C++ methods have an implicit `this' argument, the
15829 arguments of such methods should be counted from two, not one, when
15830 giving values for STRING-INDEX and FIRST-TO-CHECK.
15832 In the example above, the format string (`my_format') is the second
15833 argument of the function `my_print', and the arguments to check
15834 start with the third argument, so the correct parameters for the
15835 format attribute are 2 and 3.
15837 The `format' attribute allows you to identify your own functions
15838 which take format strings as arguments, so that GCC can check the
15839 calls to these functions for errors. The compiler always (unless
15840 `-ffreestanding' or `-fno-builtin' is used) checks formats for the
15841 standard library functions `printf', `fprintf', `sprintf',
15842 `scanf', `fscanf', `sscanf', `strftime', `vprintf', `vfprintf' and
15843 `vsprintf' whenever such warnings are requested (using
15844 `-Wformat'), so there is no need to modify the header file
15845 `stdio.h'. In C99 mode, the functions `snprintf', `vsnprintf',
15846 `vscanf', `vfscanf' and `vsscanf' are also checked. Except in
15847 strictly conforming C standard modes, the X/Open function
15848 `strfmon' is also checked as are `printf_unlocked' and
15849 `fprintf_unlocked'. *Note Options Controlling C Dialect: C
15852 The target may provide additional types of format checks. *Note
15853 Format Checks Specific to Particular Target Machines: Target
15856 `format_arg (STRING-INDEX)'
15857 The `format_arg' attribute specifies that a function takes a format
15858 string for a `printf', `scanf', `strftime' or `strfmon' style
15859 function and modifies it (for example, to translate it into
15860 another language), so the result can be passed to a `printf',
15861 `scanf', `strftime' or `strfmon' style function (with the
15862 remaining arguments to the format function the same as they would
15863 have been for the unmodified string). For example, the
15867 my_dgettext (char *my_domain, const char *my_format)
15868 __attribute__ ((format_arg (2)));
15870 causes the compiler to check the arguments in calls to a `printf',
15871 `scanf', `strftime' or `strfmon' type function, whose format
15872 string argument is a call to the `my_dgettext' function, for
15873 consistency with the format string argument `my_format'. If the
15874 `format_arg' attribute had not been specified, all the compiler
15875 could tell in such calls to format functions would be that the
15876 format string argument is not constant; this would generate a
15877 warning when `-Wformat-nonliteral' is used, but the calls could
15878 not be checked without the attribute.
15880 The parameter STRING-INDEX specifies which argument is the format
15881 string argument (starting from one). Since non-static C++ methods
15882 have an implicit `this' argument, the arguments of such methods
15883 should be counted from two.
15885 The `format-arg' attribute allows you to identify your own
15886 functions which modify format strings, so that GCC can check the
15887 calls to `printf', `scanf', `strftime' or `strfmon' type function
15888 whose operands are a call to one of your own function. The
15889 compiler always treats `gettext', `dgettext', and `dcgettext' in
15890 this manner except when strict ISO C support is requested by
15891 `-ansi' or an appropriate `-std' option, or `-ffreestanding' or
15892 `-fno-builtin' is used. *Note Options Controlling C Dialect: C
15896 Use this attribute on the H8/300, H8/300H, and H8S to indicate
15897 that the specified function should be called through the function
15898 vector. Calling a function through the function vector will
15899 reduce code size, however; the function vector has a limited size
15900 (maximum 128 entries on the H8/300 and 64 entries on the H8/300H
15901 and H8S) and shares space with the interrupt vector.
15903 You must use GAS and GLD from GNU binutils version 2.7 or later for
15904 this attribute to work correctly.
15907 Use this attribute on the ARM, AVR, C4x, CRX, M32C, M32R/D, MS1,
15908 and Xstormy16 ports to indicate that the specified function is an
15909 interrupt handler. The compiler will generate function entry and
15910 exit sequences suitable for use in an interrupt handler when this
15911 attribute is present.
15913 Note, interrupt handlers for the Blackfin, m68k, H8/300, H8/300H,
15914 H8S, and SH processors can be specified via the
15915 `interrupt_handler' attribute.
15917 Note, on the AVR, interrupts will be enabled inside the function.
15919 Note, for the ARM, you can specify the kind of interrupt to be
15920 handled by adding an optional parameter to the interrupt attribute
15923 void f () __attribute__ ((interrupt ("IRQ")));
15925 Permissible values for this parameter are: IRQ, FIQ, SWI, ABORT
15928 `interrupt_handler'
15929 Use this attribute on the Blackfin, m68k, H8/300, H8/300H, H8S,
15930 and SH to indicate that the specified function is an interrupt
15931 handler. The compiler will generate function entry and exit
15932 sequences suitable for use in an interrupt handler when this
15933 attribute is present.
15936 When used together with `interrupt_handler', `exception_handler'
15937 or `nmi_handler', code will be generated to load the stack pointer
15938 from the USP register in the function prologue.
15940 `long_call/short_call'
15941 This attribute specifies how a particular function is called on
15942 ARM. Both attributes override the `-mlong-calls' (*note ARM
15943 Options::) command line switch and `#pragma long_calls' settings.
15944 The `long_call' attribute indicates that the function might be far
15945 away from the call site and require a different (more expensive)
15946 calling sequence. The `short_call' attribute always places the
15947 offset to the function from the call site into the `BL'
15948 instruction directly.
15950 `longcall/shortcall'
15951 On the Blackfin, RS/6000 and PowerPC, the `longcall' attribute
15952 indicates that the function might be far away from the call site
15953 and require a different (more expensive) calling sequence. The
15954 `shortcall' attribute indicates that the function is always close
15955 enough for the shorter calling sequence to be used. These
15956 attributes override both the `-mlongcall' switch and, on the
15957 RS/6000 and PowerPC, the `#pragma longcall' setting.
15959 *Note RS/6000 and PowerPC Options::, for more information on
15960 whether long calls are necessary.
15963 This attribute specifies how a particular function is called on
15964 MIPS. The attribute overrides the `-mlong-calls' (*note MIPS
15965 Options::) command line switch. This attribute causes the
15966 compiler to always call the function by first loading its address
15967 into a register, and then using the contents of that register.
15970 The `malloc' attribute is used to tell the compiler that a function
15971 may be treated as if any non-`NULL' pointer it returns cannot
15972 alias any other pointer valid when the function returns. This
15973 will often improve optimization. Standard functions with this
15974 property include `malloc' and `calloc'. `realloc'-like functions
15975 have this property as long as the old pointer is never referred to
15976 (including comparing it to the new pointer) after the function
15977 returns a non-`NULL' value.
15979 `model (MODEL-NAME)'
15980 On the M32R/D, use this attribute to set the addressability of an
15981 object, and of the code generated for a function. The identifier
15982 MODEL-NAME is one of `small', `medium', or `large', representing
15983 each of the code models.
15985 Small model objects live in the lower 16MB of memory (so that their
15986 addresses can be loaded with the `ld24' instruction), and are
15987 callable with the `bl' instruction.
15989 Medium model objects may live anywhere in the 32-bit address space
15990 (the compiler will generate `seth/add3' instructions to load their
15991 addresses), and are callable with the `bl' instruction.
15993 Large model objects may live anywhere in the 32-bit address space
15994 (the compiler will generate `seth/add3' instructions to load their
15995 addresses), and may not be reachable with the `bl' instruction
15996 (the compiler will generate the much slower `seth/add3/jl'
15997 instruction sequence).
15999 On IA-64, use this attribute to set the addressability of an
16000 object. At present, the only supported identifier for MODEL-NAME
16001 is `small', indicating addressability via "small" (22-bit)
16002 addresses (so that their addresses can be loaded with the `addl'
16003 instruction). Caveat: such addressing is by definition not
16004 position independent and hence this attribute must not be used for
16005 objects defined by shared libraries.
16008 Use this attribute on the ARM, AVR, C4x and IP2K ports to indicate
16009 that the specified function does not need prologue/epilogue
16010 sequences generated by the compiler. It is up to the programmer
16011 to provide these sequences.
16014 On 68HC11 and 68HC12 the `near' attribute causes the compiler to
16015 use the normal calling convention based on `jsr' and `rts'. This
16016 attribute can be used to cancel the effect of the `-mlong-calls'
16020 Use this attribute together with `interrupt_handler',
16021 `exception_handler' or `nmi_handler' to indicate that the function
16022 entry code should enable nested interrupts or exceptions.
16025 Use this attribute on the Blackfin to indicate that the specified
16026 function is an NMI handler. The compiler will generate function
16027 entry and exit sequences suitable for use in an NMI handler when
16028 this attribute is present.
16030 `no_instrument_function'
16031 If `-finstrument-functions' is given, profiling function calls will
16032 be generated at entry and exit of most user-compiled functions.
16033 Functions with this attribute will not be so instrumented.
16036 This function attribute prevents a function from being considered
16039 `nonnull (ARG-INDEX, ...)'
16040 The `nonnull' attribute specifies that some function parameters
16041 should be non-null pointers. For instance, the declaration:
16044 my_memcpy (void *dest, const void *src, size_t len)
16045 __attribute__((nonnull (1, 2)));
16047 causes the compiler to check that, in calls to `my_memcpy',
16048 arguments DEST and SRC are non-null. If the compiler determines
16049 that a null pointer is passed in an argument slot marked as
16050 non-null, and the `-Wnonnull' option is enabled, a warning is
16051 issued. The compiler may also choose to make optimizations based
16052 on the knowledge that certain function arguments will not be null.
16054 If no argument index list is given to the `nonnull' attribute, all
16055 pointer arguments are marked as non-null. To illustrate, the
16056 following declaration is equivalent to the previous example:
16059 my_memcpy (void *dest, const void *src, size_t len)
16060 __attribute__((nonnull));
16063 A few standard library functions, such as `abort' and `exit',
16064 cannot return. GCC knows this automatically. Some programs define
16065 their own functions that never return. You can declare them
16066 `noreturn' to tell the compiler this fact. For example,
16068 void fatal () __attribute__ ((noreturn));
16073 /* ... */ /* Print error message. */ /* ... */
16077 The `noreturn' keyword tells the compiler to assume that `fatal'
16078 cannot return. It can then optimize without regard to what would
16079 happen if `fatal' ever did return. This makes slightly better
16080 code. More importantly, it helps avoid spurious warnings of
16081 uninitialized variables.
16083 The `noreturn' keyword does not affect the exceptional path when
16084 that applies: a `noreturn'-marked function may still return to the
16085 caller by throwing an exception or calling `longjmp'.
16087 Do not assume that registers saved by the calling function are
16088 restored before calling the `noreturn' function.
16090 It does not make sense for a `noreturn' function to have a return
16091 type other than `void'.
16093 The attribute `noreturn' is not implemented in GCC versions
16094 earlier than 2.5. An alternative way to declare that a function
16095 does not return, which works in the current version and in some
16096 older versions, is as follows:
16098 typedef void voidfn ();
16100 volatile voidfn fatal;
16102 This approach does not work in GNU C++.
16105 The `nothrow' attribute is used to inform the compiler that a
16106 function cannot throw an exception. For example, most functions in
16107 the standard C library can be guaranteed not to throw an exception
16108 with the notable exceptions of `qsort' and `bsearch' that take
16109 function pointer arguments. The `nothrow' attribute is not
16110 implemented in GCC versions earlier than 3.3.
16113 Many functions have no effects except the return value and their
16114 return value depends only on the parameters and/or global
16115 variables. Such a function can be subject to common subexpression
16116 elimination and loop optimization just as an arithmetic operator
16117 would be. These functions should be declared with the attribute
16118 `pure'. For example,
16120 int square (int) __attribute__ ((pure));
16122 says that the hypothetical function `square' is safe to call fewer
16123 times than the program says.
16125 Some of common examples of pure functions are `strlen' or `memcmp'.
16126 Interesting non-pure functions are functions with infinite loops
16127 or those depending on volatile memory or other system resource,
16128 that may change between two consecutive calls (such as `feof' in a
16129 multithreading environment).
16131 The attribute `pure' is not implemented in GCC versions earlier
16135 On the Intel 386, the `regparm' attribute causes the compiler to
16136 pass arguments number one to NUMBER if they are of integral type
16137 in registers EAX, EDX, and ECX instead of on the stack. Functions
16138 that take a variable number of arguments will continue to be
16139 passed all of their arguments on the stack.
16141 Beware that on some ELF systems this attribute is unsuitable for
16142 global functions in shared libraries with lazy binding (which is
16143 the default). Lazy binding will send the first call via resolving
16144 code in the loader, which might assume EAX, EDX and ECX can be
16145 clobbered, as per the standard calling conventions. Solaris 8 is
16146 affected by this. GNU systems with GLIBC 2.1 or higher, and
16147 FreeBSD, are believed to be safe since the loaders there save all
16148 registers. (Lazy binding can be disabled with the linker or the
16149 loader if desired, to avoid the problem.)
16152 On the Intel 386 with SSE support, the `sseregparm' attribute
16153 causes the compiler to pass up to 3 floating point arguments in
16154 SSE registers instead of on the stack. Functions that take a
16155 variable number of arguments will continue to pass all of their
16156 floating point arguments on the stack.
16158 `force_align_arg_pointer'
16159 On the Intel x86, the `force_align_arg_pointer' attribute may be
16160 applied to individual function definitions, generating an alternate
16161 prologue and epilogue that realigns the runtime stack. This
16162 supports mixing legacy codes that run with a 4-byte aligned stack
16163 with modern codes that keep a 16-byte stack for SSE compatibility.
16164 The alternate prologue and epilogue are slower and bigger than
16165 the regular ones, and the alternate prologue requires a scratch
16166 register; this lowers the number of registers available if used in
16167 conjunction with the `regparm' attribute. The
16168 `force_align_arg_pointer' attribute is incompatible with nested
16169 functions; this is considered a hard error.
16172 The `returns_twice' attribute tells the compiler that a function
16173 may return more than one time. The compiler will ensure that all
16174 registers are dead before calling such a function and will emit a
16175 warning about the variables that may be clobbered after the second
16176 return from the function. Examples of such functions are `setjmp'
16177 and `vfork'. The `longjmp'-like counterpart of such function, if
16178 any, might need to be marked with the `noreturn' attribute.
16181 Use this attribute on the Blackfin, H8/300, H8/300H, and H8S to
16182 indicate that all registers except the stack pointer should be
16183 saved in the prologue regardless of whether they are used or not.
16185 `section ("SECTION-NAME")'
16186 Normally, the compiler places the code it generates in the `text'
16187 section. Sometimes, however, you need additional sections, or you
16188 need certain particular functions to appear in special sections.
16189 The `section' attribute specifies that a function lives in a
16190 particular section. For example, the declaration:
16192 extern void foobar (void) __attribute__ ((section ("bar")));
16194 puts the function `foobar' in the `bar' section.
16196 Some file formats do not support arbitrary sections so the
16197 `section' attribute is not available on all platforms. If you
16198 need to map the entire contents of a module to a particular
16199 section, consider using the facilities of the linker instead.
16202 This function attribute ensures that a parameter in a function
16203 call is an explicit `NULL'. The attribute is only valid on
16204 variadic functions. By default, the sentinel is located at
16205 position zero, the last parameter of the function call. If an
16206 optional integer position argument P is supplied to the attribute,
16207 the sentinel must be located at position P counting backwards from
16208 the end of the argument list.
16210 __attribute__ ((sentinel))
16212 __attribute__ ((sentinel(0)))
16214 The attribute is automatically set with a position of 0 for the
16215 built-in functions `execl' and `execlp'. The built-in function
16216 `execle' has the attribute set with a position of 1.
16218 A valid `NULL' in this context is defined as zero with any pointer
16219 type. If your system defines the `NULL' macro with an integer type
16220 then you need to add an explicit cast. GCC replaces `stddef.h'
16221 with a copy that redefines NULL appropriately.
16223 The warnings for missing or incorrect sentinels are enabled with
16227 See long_call/short_call.
16230 See longcall/shortcall.
16233 Use this attribute on the AVR to indicate that the specified
16234 function is a signal handler. The compiler will generate function
16235 entry and exit sequences suitable for use in a signal handler when
16236 this attribute is present. Interrupts will be disabled inside the
16240 Use this attribute on the SH to indicate an `interrupt_handler'
16241 function should switch to an alternate stack. It expects a string
16242 argument that names a global variable holding the address of the
16246 void f () __attribute__ ((interrupt_handler,
16247 sp_switch ("alt_stack")));
16250 On the Intel 386, the `stdcall' attribute causes the compiler to
16251 assume that the called function will pop off the stack space used
16252 to pass arguments, unless it takes a variable number of arguments.
16255 Use this attribute on the H8/300H and H8S to indicate that the
16256 specified variable should be placed into the tiny data section.
16257 The compiler will generate more efficient code for loads and stores
16258 on data in the tiny data section. Note the tiny data area is
16259 limited to slightly under 32kbytes of data.
16262 Use this attribute on the SH for an `interrupt_handler' to return
16263 using `trapa' instead of `rte'. This attribute expects an integer
16264 argument specifying the trap number to be used.
16267 This attribute, attached to a function, means that the function is
16268 meant to be possibly unused. GCC will not produce a warning for
16272 This attribute, attached to a function, means that code must be
16273 emitted for the function even if it appears that the function is
16274 not referenced. This is useful, for example, when the function is
16275 referenced only in inline assembly.
16277 `visibility ("VISIBILITY_TYPE")'
16278 This attribute affects the linkage of the declaration to which it
16279 is attached. There are four supported VISIBILITY_TYPE values:
16280 default, hidden, protected or internal visibility.
16282 void __attribute__ ((visibility ("protected")))
16283 f () { /* Do something. */; }
16284 int i __attribute__ ((visibility ("hidden")));
16286 The possible values of VISIBILITY_TYPE correspond to the
16287 visibility settings in the ELF gABI.
16290 Default visibility is the normal case for the object file
16291 format. This value is available for the visibility attribute
16292 to override other options that may change the assumed
16293 visibility of entities.
16295 On ELF, default visibility means that the declaration is
16296 visible to other modules and, in shared libraries, means that
16297 the declared entity may be overridden.
16299 On Darwin, default visibility means that the declaration is
16300 visible to other modules.
16302 Default visibility corresponds to "external linkage" in the
16306 Hidden visibility indicates that the entity declared will
16307 have a new form of linkage, which we'll call "hidden
16308 linkage". Two declarations of an object with hidden linkage
16309 refer to the same object if they are in the same shared
16313 Internal visibility is like hidden visibility, but with
16314 additional processor specific semantics. Unless otherwise
16315 specified by the psABI, GCC defines internal visibility to
16316 mean that a function is _never_ called from another module.
16317 Compare this with hidden functions which, while they cannot
16318 be referenced directly by other modules, can be referenced
16319 indirectly via function pointers. By indicating that a
16320 function cannot be called from outside the module, GCC may
16321 for instance omit the load of a PIC register since it is known
16322 that the calling function loaded the correct value.
16325 Protected visibility is like default visibility except that it
16326 indicates that references within the defining module will
16327 bind to the definition in that module. That is, the declared
16328 entity cannot be overridden by another module.
16331 All visibilities are supported on many, but not all, ELF targets
16332 (supported when the assembler supports the `.visibility'
16333 pseudo-op). Default visibility is supported everywhere. Hidden
16334 visibility is supported on Darwin targets.
16336 The visibility attribute should be applied only to declarations
16337 which would otherwise have external linkage. The attribute should
16338 be applied consistently, so that the same entity should not be
16339 declared with different settings of the attribute.
16341 In C++, the visibility attribute applies to types as well as
16342 functions and objects, because in C++ types have linkage. A class
16343 must not have greater visibility than its non-static data member
16344 types and bases, and class members default to the visibility of
16345 their class. Also, a declaration without explicit visibility is
16346 limited to the visibility of its type.
16348 In C++, you can mark member functions and static member variables
16349 of a class with the visibility attribute. This is useful if if
16350 you know a particular method or static member variable should only
16351 be used from one shared object; then you can mark it hidden while
16352 the rest of the class has default visibility. Care must be taken
16353 to avoid breaking the One Definition Rule; for example, it is
16354 usually not useful to mark an inline method as hidden without
16355 marking the whole class as hidden.
16357 A C++ namespace declaration can also have the visibility attribute.
16358 This attribute applies only to the particular namespace body, not
16359 to other definitions of the same namespace; it is equivalent to
16360 using `#pragma GCC visibility' before and after the namespace
16361 definition (*note Visibility Pragmas::).
16363 In C++, if a template argument has limited visibility, this
16364 restriction is implicitly propagated to the template instantiation.
16365 Otherwise, template instantiations and specializations default to
16366 the visibility of their template.
16368 If both the template and enclosing class have explicit visibility,
16369 the visibility from the template is used.
16371 `warn_unused_result'
16372 The `warn_unused_result' attribute causes a warning to be emitted
16373 if a caller of the function with this attribute does not use its
16374 return value. This is useful for functions where not checking the
16375 result is either a security problem or always a bug, such as
16378 int fn () __attribute__ ((warn_unused_result));
16381 if (fn () < 0) return -1;
16386 results in warning on line 5.
16389 The `weak' attribute causes the declaration to be emitted as a weak
16390 symbol rather than a global. This is primarily useful in defining
16391 library functions which can be overridden in user code, though it
16392 can also be used with non-function declarations. Weak symbols are
16393 supported for ELF targets, and also for a.out targets when using
16394 the GNU assembler and linker.
16397 `weakref ("TARGET")'
16398 The `weakref' attribute marks a declaration as a weak reference.
16399 Without arguments, it should be accompanied by an `alias' attribute
16400 naming the target symbol. Optionally, the TARGET may be given as
16401 an argument to `weakref' itself. In either case, `weakref'
16402 implicitly marks the declaration as `weak'. Without a TARGET,
16403 given as an argument to `weakref' or to `alias', `weakref' is
16404 equivalent to `weak'.
16406 static int x() __attribute__ ((weakref ("y")));
16407 /* is equivalent to... */
16408 static int x() __attribute__ ((weak, weakref, alias ("y")));
16410 static int x() __attribute__ ((weakref));
16411 static int x() __attribute__ ((alias ("y")));
16413 A weak reference is an alias that does not by itself require a
16414 definition to be given for the target symbol. If the target
16415 symbol is only referenced through weak references, then the
16416 becomes a `weak' undefined symbol. If it is directly referenced,
16417 however, then such strong references prevail, and a definition
16418 will be required for the symbol, not necessarily in the same
16421 The effect is equivalent to moving all references to the alias to a
16422 separate translation unit, renaming the alias to the aliased
16423 symbol, declaring it as weak, compiling the two separate
16424 translation units and performing a reloadable link on them.
16426 At present, a declaration to which `weakref' is attached can only
16429 `externally_visible'
16430 This attribute, attached to a global variable or function nullify
16431 effect of `-fwhole-program' command line option, so the object
16432 remain visible outside the current compilation unit
16435 You can specify multiple attributes in a declaration by separating them
16436 by commas within the double parentheses or by immediately following an
16437 attribute declaration with another attribute declaration.
16439 Some people object to the `__attribute__' feature, suggesting that ISO
16440 C's `#pragma' should be used instead. At the time `__attribute__' was
16441 designed, there were two reasons for not doing this.
16443 1. It is impossible to generate `#pragma' commands from a macro.
16445 2. There is no telling what the same `#pragma' might mean in another
16448 These two reasons applied to almost any application that might have
16449 been proposed for `#pragma'. It was basically a mistake to use
16450 `#pragma' for _anything_.
16452 The ISO C99 standard includes `_Pragma', which now allows pragmas to
16453 be generated from macros. In addition, a `#pragma GCC' namespace is
16454 now in use for GCC-specific pragmas. However, it has been found
16455 convenient to use `__attribute__' to achieve a natural attachment of
16456 attributes to their corresponding declarations, whereas `#pragma GCC'
16457 is of use for constructs that do not naturally form part of the
16458 grammar. *Note Miscellaneous Preprocessing Directives: (cpp)Other
16462 File: gcc.info, Node: Attribute Syntax, Next: Function Prototypes, Prev: Function Attributes, Up: C Extensions
16464 5.26 Attribute Syntax
16465 =====================
16467 This section describes the syntax with which `__attribute__' may be
16468 used, and the constructs to which attribute specifiers bind, for the C
16469 language. Some details may vary for C++ and Objective-C. Because of
16470 infelicities in the grammar for attributes, some forms described here
16471 may not be successfully parsed in all cases.
16473 There are some problems with the semantics of attributes in C++. For
16474 example, there are no manglings for attributes, although they may affect
16475 code generation, so problems may arise when attributed types are used in
16476 conjunction with templates or overloading. Similarly, `typeid' does
16477 not distinguish between types with different attributes. Support for
16478 attributes in C++ may be restricted in future to attributes on
16479 declarations only, but not on nested declarators.
16481 *Note Function Attributes::, for details of the semantics of attributes
16482 applying to functions. *Note Variable Attributes::, for details of the
16483 semantics of attributes applying to variables. *Note Type Attributes::,
16484 for details of the semantics of attributes applying to structure, union
16485 and enumerated types.
16487 An "attribute specifier" is of the form `__attribute__
16488 ((ATTRIBUTE-LIST))'. An "attribute list" is a possibly empty
16489 comma-separated sequence of "attributes", where each attribute is one
16492 * Empty. Empty attributes are ignored.
16494 * A word (which may be an identifier such as `unused', or a reserved
16495 word such as `const').
16497 * A word, followed by, in parentheses, parameters for the attribute.
16498 These parameters take one of the following forms:
16500 * An identifier. For example, `mode' attributes use this form.
16502 * An identifier followed by a comma and a non-empty
16503 comma-separated list of expressions. For example, `format'
16504 attributes use this form.
16506 * A possibly empty comma-separated list of expressions. For
16507 example, `format_arg' attributes use this form with the list
16508 being a single integer constant expression, and `alias'
16509 attributes use this form with the list being a single string
16512 An "attribute specifier list" is a sequence of one or more attribute
16513 specifiers, not separated by any other tokens.
16515 In GNU C, an attribute specifier list may appear after the colon
16516 following a label, other than a `case' or `default' label. The only
16517 attribute it makes sense to use after a label is `unused'. This
16518 feature is intended for code generated by programs which contains labels
16519 that may be unused but which is compiled with `-Wall'. It would not
16520 normally be appropriate to use in it human-written code, though it
16521 could be useful in cases where the code that jumps to the label is
16522 contained within an `#ifdef' conditional. GNU C++ does not permit such
16523 placement of attribute lists, as it is permissible for a declaration,
16524 which could begin with an attribute list, to be labelled in C++.
16525 Declarations cannot be labelled in C90 or C99, so the ambiguity does
16528 An attribute specifier list may appear as part of a `struct', `union'
16529 or `enum' specifier. It may go either immediately after the `struct',
16530 `union' or `enum' keyword, or after the closing brace. The former
16531 syntax is preferred. Where attribute specifiers follow the closing
16532 brace, they are considered to relate to the structure, union or
16533 enumerated type defined, not to any enclosing declaration the type
16534 specifier appears in, and the type defined is not complete until after
16535 the attribute specifiers.
16537 Otherwise, an attribute specifier appears as part of a declaration,
16538 counting declarations of unnamed parameters and type names, and relates
16539 to that declaration (which may be nested in another declaration, for
16540 example in the case of a parameter declaration), or to a particular
16541 declarator within a declaration. Where an attribute specifier is
16542 applied to a parameter declared as a function or an array, it should
16543 apply to the function or array rather than the pointer to which the
16544 parameter is implicitly converted, but this is not yet correctly
16547 Any list of specifiers and qualifiers at the start of a declaration may
16548 contain attribute specifiers, whether or not such a list may in that
16549 context contain storage class specifiers. (Some attributes, however,
16550 are essentially in the nature of storage class specifiers, and only make
16551 sense where storage class specifiers may be used; for example,
16552 `section'.) There is one necessary limitation to this syntax: the
16553 first old-style parameter declaration in a function definition cannot
16554 begin with an attribute specifier, because such an attribute applies to
16555 the function instead by syntax described below (which, however, is not
16556 yet implemented in this case). In some other cases, attribute
16557 specifiers are permitted by this grammar but not yet supported by the
16558 compiler. All attribute specifiers in this place relate to the
16559 declaration as a whole. In the obsolescent usage where a type of `int'
16560 is implied by the absence of type specifiers, such a list of specifiers
16561 and qualifiers may be an attribute specifier list with no other
16562 specifiers or qualifiers.
16564 At present, the first parameter in a function prototype must have some
16565 type specifier which is not an attribute specifier; this resolves an
16566 ambiguity in the interpretation of `void f(int (__attribute__((foo))
16567 x))', but is subject to change. At present, if the parentheses of a
16568 function declarator contain only attributes then those attributes are
16569 ignored, rather than yielding an error or warning or implying a single
16570 parameter of type int, but this is subject to change.
16572 An attribute specifier list may appear immediately before a declarator
16573 (other than the first) in a comma-separated list of declarators in a
16574 declaration of more than one identifier using a single list of
16575 specifiers and qualifiers. Such attribute specifiers apply only to the
16576 identifier before whose declarator they appear. For example, in
16578 __attribute__((noreturn)) void d0 (void),
16579 __attribute__((format(printf, 1, 2))) d1 (const char *, ...),
16582 the `noreturn' attribute applies to all the functions declared; the
16583 `format' attribute only applies to `d1'.
16585 An attribute specifier list may appear immediately before the comma,
16586 `=' or semicolon terminating the declaration of an identifier other
16587 than a function definition. At present, such attribute specifiers apply
16588 to the declared object or function, but in future they may attach to the
16589 outermost adjacent declarator. In simple cases there is no difference,
16590 but, for example, in
16592 void (****f)(void) __attribute__((noreturn));
16594 at present the `noreturn' attribute applies to `f', which causes a
16595 warning since `f' is not a function, but in future it may apply to the
16596 function `****f'. The precise semantics of what attributes in such
16597 cases will apply to are not yet specified. Where an assembler name for
16598 an object or function is specified (*note Asm Labels::), at present the
16599 attribute must follow the `asm' specification; in future, attributes
16600 before the `asm' specification may apply to the adjacent declarator,
16601 and those after it to the declared object or function.
16603 An attribute specifier list may, in future, be permitted to appear
16604 after the declarator in a function definition (before any old-style
16605 parameter declarations or the function body).
16607 Attribute specifiers may be mixed with type qualifiers appearing inside
16608 the `[]' of a parameter array declarator, in the C99 construct by which
16609 such qualifiers are applied to the pointer to which the array is
16610 implicitly converted. Such attribute specifiers apply to the pointer,
16611 not to the array, but at present this is not implemented and they are
16614 An attribute specifier list may appear at the start of a nested
16615 declarator. At present, there are some limitations in this usage: the
16616 attributes correctly apply to the declarator, but for most individual
16617 attributes the semantics this implies are not implemented. When
16618 attribute specifiers follow the `*' of a pointer declarator, they may
16619 be mixed with any type qualifiers present. The following describes the
16620 formal semantics of this syntax. It will make the most sense if you
16621 are familiar with the formal specification of declarators in the ISO C
16624 Consider (as in C99 subclause 6.7.5 paragraph 4) a declaration `T D1',
16625 where `T' contains declaration specifiers that specify a type TYPE
16626 (such as `int') and `D1' is a declarator that contains an identifier
16627 IDENT. The type specified for IDENT for derived declarators whose type
16628 does not include an attribute specifier is as in the ISO C standard.
16630 If `D1' has the form `( ATTRIBUTE-SPECIFIER-LIST D )', and the
16631 declaration `T D' specifies the type "DERIVED-DECLARATOR-TYPE-LIST
16632 TYPE" for IDENT, then `T D1' specifies the type
16633 "DERIVED-DECLARATOR-TYPE-LIST ATTRIBUTE-SPECIFIER-LIST TYPE" for IDENT.
16635 If `D1' has the form `* TYPE-QUALIFIER-AND-ATTRIBUTE-SPECIFIER-LIST
16636 D', and the declaration `T D' specifies the type
16637 "DERIVED-DECLARATOR-TYPE-LIST TYPE" for IDENT, then `T D1' specifies
16638 the type "DERIVED-DECLARATOR-TYPE-LIST
16639 TYPE-QUALIFIER-AND-ATTRIBUTE-SPECIFIER-LIST TYPE" for IDENT.
16643 void (__attribute__((noreturn)) ****f) (void);
16645 specifies the type "pointer to pointer to pointer to pointer to
16646 non-returning function returning `void'". As another example,
16648 char *__attribute__((aligned(8))) *f;
16650 specifies the type "pointer to 8-byte-aligned pointer to `char'". Note
16651 again that this does not work with most attributes; for example, the
16652 usage of `aligned' and `noreturn' attributes given above is not yet
16655 For compatibility with existing code written for compiler versions that
16656 did not implement attributes on nested declarators, some laxity is
16657 allowed in the placing of attributes. If an attribute that only applies
16658 to types is applied to a declaration, it will be treated as applying to
16659 the type of that declaration. If an attribute that only applies to
16660 declarations is applied to the type of a declaration, it will be treated
16661 as applying to that declaration; and, for compatibility with code
16662 placing the attributes immediately before the identifier declared, such
16663 an attribute applied to a function return type will be treated as
16664 applying to the function type, and such an attribute applied to an array
16665 element type will be treated as applying to the array type. If an
16666 attribute that only applies to function types is applied to a
16667 pointer-to-function type, it will be treated as applying to the pointer
16668 target type; if such an attribute is applied to a function return type
16669 that is not a pointer-to-function type, it will be treated as applying
16670 to the function type.
16673 File: gcc.info, Node: Function Prototypes, Next: C++ Comments, Prev: Attribute Syntax, Up: C Extensions
16675 5.27 Prototypes and Old-Style Function Definitions
16676 ==================================================
16678 GNU C extends ISO C to allow a function prototype to override a later
16679 old-style non-prototype definition. Consider the following example:
16681 /* Use prototypes unless the compiler is old-fashioned. */
16688 /* Prototype function declaration. */
16689 int isroot P((uid_t));
16691 /* Old-style function definition. */
16693 isroot (x) /* ??? lossage here ??? */
16699 Suppose the type `uid_t' happens to be `short'. ISO C does not allow
16700 this example, because subword arguments in old-style non-prototype
16701 definitions are promoted. Therefore in this example the function
16702 definition's argument is really an `int', which does not match the
16703 prototype argument type of `short'.
16705 This restriction of ISO C makes it hard to write code that is portable
16706 to traditional C compilers, because the programmer does not know
16707 whether the `uid_t' type is `short', `int', or `long'. Therefore, in
16708 cases like these GNU C allows a prototype to override a later old-style
16709 definition. More precisely, in GNU C, a function prototype argument
16710 type overrides the argument type specified by a later old-style
16711 definition if the former type is the same as the latter type before
16712 promotion. Thus in GNU C the above example is equivalent to the
16715 int isroot (uid_t);
16723 GNU C++ does not support old-style function definitions, so this
16724 extension is irrelevant.
16727 File: gcc.info, Node: C++ Comments, Next: Dollar Signs, Prev: Function Prototypes, Up: C Extensions
16729 5.28 C++ Style Comments
16730 =======================
16732 In GNU C, you may use C++ style comments, which start with `//' and
16733 continue until the end of the line. Many other C implementations allow
16734 such comments, and they are included in the 1999 C standard. However,
16735 C++ style comments are not recognized if you specify an `-std' option
16736 specifying a version of ISO C before C99, or `-ansi' (equivalent to
16740 File: gcc.info, Node: Dollar Signs, Next: Character Escapes, Prev: C++ Comments, Up: C Extensions
16742 5.29 Dollar Signs in Identifier Names
16743 =====================================
16745 In GNU C, you may normally use dollar signs in identifier names. This
16746 is because many traditional C implementations allow such identifiers.
16747 However, dollar signs in identifiers are not supported on a few target
16748 machines, typically because the target assembler does not allow them.
16751 File: gcc.info, Node: Character Escapes, Next: Variable Attributes, Prev: Dollar Signs, Up: C Extensions
16753 5.30 The Character <ESC> in Constants
16754 =====================================
16756 You can use the sequence `\e' in a string or character constant to
16757 stand for the ASCII character <ESC>.
16760 File: gcc.info, Node: Alignment, Next: Inline, Prev: Type Attributes, Up: C Extensions
16762 5.31 Inquiring on Alignment of Types or Variables
16763 =================================================
16765 The keyword `__alignof__' allows you to inquire about how an object is
16766 aligned, or the minimum alignment usually required by a type. Its
16767 syntax is just like `sizeof'.
16769 For example, if the target machine requires a `double' value to be
16770 aligned on an 8-byte boundary, then `__alignof__ (double)' is 8. This
16771 is true on many RISC machines. On more traditional machine designs,
16772 `__alignof__ (double)' is 4 or even 2.
16774 Some machines never actually require alignment; they allow reference
16775 to any data type even at an odd address. For these machines,
16776 `__alignof__' reports the _recommended_ alignment of a type.
16778 If the operand of `__alignof__' is an lvalue rather than a type, its
16779 value is the required alignment for its type, taking into account any
16780 minimum alignment specified with GCC's `__attribute__' extension (*note
16781 Variable Attributes::). For example, after this declaration:
16783 struct foo { int x; char y; } foo1;
16785 the value of `__alignof__ (foo1.y)' is 1, even though its actual
16786 alignment is probably 2 or 4, the same as `__alignof__ (int)'.
16788 It is an error to ask for the alignment of an incomplete type.
16791 File: gcc.info, Node: Variable Attributes, Next: Type Attributes, Prev: Character Escapes, Up: C Extensions
16793 5.32 Specifying Attributes of Variables
16794 =======================================
16796 The keyword `__attribute__' allows you to specify special attributes of
16797 variables or structure fields. This keyword is followed by an
16798 attribute specification inside double parentheses. Some attributes are
16799 currently defined generically for variables. Other attributes are
16800 defined for variables on particular target systems. Other attributes
16801 are available for functions (*note Function Attributes::) and for types
16802 (*note Type Attributes::). Other front ends might define more
16803 attributes (*note Extensions to the C++ Language: C++ Extensions.).
16805 You may also specify attributes with `__' preceding and following each
16806 keyword. This allows you to use them in header files without being
16807 concerned about a possible macro of the same name. For example, you
16808 may use `__aligned__' instead of `aligned'.
16810 *Note Attribute Syntax::, for details of the exact syntax for using
16813 `aligned (ALIGNMENT)'
16814 This attribute specifies a minimum alignment for the variable or
16815 structure field, measured in bytes. For example, the declaration:
16817 int x __attribute__ ((aligned (16))) = 0;
16819 causes the compiler to allocate the global variable `x' on a
16820 16-byte boundary. On a 68040, this could be used in conjunction
16821 with an `asm' expression to access the `move16' instruction which
16822 requires 16-byte aligned operands.
16824 You can also specify the alignment of structure fields. For
16825 example, to create a double-word aligned `int' pair, you could
16828 struct foo { int x[2] __attribute__ ((aligned (8))); };
16830 This is an alternative to creating a union with a `double' member
16831 that forces the union to be double-word aligned.
16833 As in the preceding examples, you can explicitly specify the
16834 alignment (in bytes) that you wish the compiler to use for a given
16835 variable or structure field. Alternatively, you can leave out the
16836 alignment factor and just ask the compiler to align a variable or
16837 field to the maximum useful alignment for the target machine you
16838 are compiling for. For example, you could write:
16840 short array[3] __attribute__ ((aligned));
16842 Whenever you leave out the alignment factor in an `aligned'
16843 attribute specification, the compiler automatically sets the
16844 alignment for the declared variable or field to the largest
16845 alignment which is ever used for any data type on the target
16846 machine you are compiling for. Doing this can often make copy
16847 operations more efficient, because the compiler can use whatever
16848 instructions copy the biggest chunks of memory when performing
16849 copies to or from the variables or fields that you have aligned
16852 The `aligned' attribute can only increase the alignment; but you
16853 can decrease it by specifying `packed' as well. See below.
16855 Note that the effectiveness of `aligned' attributes may be limited
16856 by inherent limitations in your linker. On many systems, the
16857 linker is only able to arrange for variables to be aligned up to a
16858 certain maximum alignment. (For some linkers, the maximum
16859 supported alignment may be very very small.) If your linker is
16860 only able to align variables up to a maximum of 8 byte alignment,
16861 then specifying `aligned(16)' in an `__attribute__' will still
16862 only provide you with 8 byte alignment. See your linker
16863 documentation for further information.
16865 `cleanup (CLEANUP_FUNCTION)'
16866 The `cleanup' attribute runs a function when the variable goes out
16867 of scope. This attribute can only be applied to auto function
16868 scope variables; it may not be applied to parameters or variables
16869 with static storage duration. The function must take one
16870 parameter, a pointer to a type compatible with the variable. The
16871 return value of the function (if any) is ignored.
16873 If `-fexceptions' is enabled, then CLEANUP_FUNCTION will be run
16874 during the stack unwinding that happens during the processing of
16875 the exception. Note that the `cleanup' attribute does not allow
16876 the exception to be caught, only to perform an action. It is
16877 undefined what happens if CLEANUP_FUNCTION does not return
16882 The `common' attribute requests GCC to place a variable in
16883 "common" storage. The `nocommon' attribute requests the
16884 opposite--to allocate space for it directly.
16886 These attributes override the default chosen by the `-fno-common'
16887 and `-fcommon' flags respectively.
16890 The `deprecated' attribute results in a warning if the variable is
16891 used anywhere in the source file. This is useful when identifying
16892 variables that are expected to be removed in a future version of a
16893 program. The warning also includes the location of the declaration
16894 of the deprecated variable, to enable users to easily find further
16895 information about why the variable is deprecated, or what they
16896 should do instead. Note that the warning only occurs for uses:
16898 extern int old_var __attribute__ ((deprecated));
16899 extern int old_var;
16900 int new_fn () { return old_var; }
16902 results in a warning on line 3 but not line 2.
16904 The `deprecated' attribute can also be used for functions and
16905 types (*note Function Attributes::, *note Type Attributes::.)
16908 This attribute specifies the data type for the
16909 declaration--whichever type corresponds to the mode MODE. This in
16910 effect lets you request an integer or floating point type
16911 according to its width.
16913 You may also specify a mode of `byte' or `__byte__' to indicate
16914 the mode corresponding to a one-byte integer, `word' or `__word__'
16915 for the mode of a one-word integer, and `pointer' or `__pointer__'
16916 for the mode used to represent pointers.
16919 The `packed' attribute specifies that a variable or structure field
16920 should have the smallest possible alignment--one byte for a
16921 variable, and one bit for a field, unless you specify a larger
16922 value with the `aligned' attribute.
16924 Here is a structure in which the field `x' is packed, so that it
16925 immediately follows `a':
16930 int x[2] __attribute__ ((packed));
16933 `section ("SECTION-NAME")'
16934 Normally, the compiler places the objects it generates in sections
16935 like `data' and `bss'. Sometimes, however, you need additional
16936 sections, or you need certain particular variables to appear in
16937 special sections, for example to map to special hardware. The
16938 `section' attribute specifies that a variable (or function) lives
16939 in a particular section. For example, this small program uses
16940 several specific section names:
16942 struct duart a __attribute__ ((section ("DUART_A"))) = { 0 };
16943 struct duart b __attribute__ ((section ("DUART_B"))) = { 0 };
16944 char stack[10000] __attribute__ ((section ("STACK"))) = { 0 };
16945 int init_data __attribute__ ((section ("INITDATA"))) = 0;
16949 /* Initialize stack pointer */
16950 init_sp (stack + sizeof (stack));
16952 /* Initialize initialized data */
16953 memcpy (&init_data, &data, &edata - &data);
16955 /* Turn on the serial ports */
16960 Use the `section' attribute with an _initialized_ definition of a
16961 _global_ variable, as shown in the example. GCC issues a warning
16962 and otherwise ignores the `section' attribute in uninitialized
16963 variable declarations.
16965 You may only use the `section' attribute with a fully initialized
16966 global definition because of the way linkers work. The linker
16967 requires each object be defined once, with the exception that
16968 uninitialized variables tentatively go in the `common' (or `bss')
16969 section and can be multiply "defined". You can force a variable
16970 to be initialized with the `-fno-common' flag or the `nocommon'
16973 Some file formats do not support arbitrary sections so the
16974 `section' attribute is not available on all platforms. If you
16975 need to map the entire contents of a module to a particular
16976 section, consider using the facilities of the linker instead.
16979 On Microsoft Windows, in addition to putting variable definitions
16980 in a named section, the section can also be shared among all
16981 running copies of an executable or DLL. For example, this small
16982 program defines shared data by putting it in a named section
16983 `shared' and marking the section shareable:
16985 int foo __attribute__((section ("shared"), shared)) = 0;
16990 /* Read and write foo. All running
16991 copies see the same value. */
16995 You may only use the `shared' attribute along with `section'
16996 attribute with a fully initialized global definition because of
16997 the way linkers work. See `section' attribute for more
17000 The `shared' attribute is only available on Microsoft Windows.
17002 `tls_model ("TLS_MODEL")'
17003 The `tls_model' attribute sets thread-local storage model (*note
17004 Thread-Local::) of a particular `__thread' variable, overriding
17005 `-ftls-model=' command line switch on a per-variable basis. The
17006 TLS_MODEL argument should be one of `global-dynamic',
17007 `local-dynamic', `initial-exec' or `local-exec'.
17009 Not all targets support this attribute.
17012 This attribute, attached to a variable, means that the variable is
17013 meant to be possibly unused. GCC will not produce a warning for
17017 This attribute, attached to a variable, means that the variable
17018 must be emitted even if it appears that the variable is not
17021 `vector_size (BYTES)'
17022 This attribute specifies the vector size for the variable,
17023 measured in bytes. For example, the declaration:
17025 int foo __attribute__ ((vector_size (16)));
17027 causes the compiler to set the mode for `foo', to be 16 bytes,
17028 divided into `int' sized units. Assuming a 32-bit int (a vector of
17029 4 units of 4 bytes), the corresponding mode of `foo' will be V4SI.
17031 This attribute is only applicable to integral and float scalars,
17032 although arrays, pointers, and function return values are allowed
17033 in conjunction with this construct.
17035 Aggregates with this attribute are invalid, even if they are of
17036 the same size as a corresponding scalar. For example, the
17039 struct S { int a; };
17040 struct S __attribute__ ((vector_size (16))) foo;
17042 is invalid even if the size of the structure is the same as the
17046 The `selectany' attribute causes an initialized global variable to
17047 have link-once semantics. When multiple definitions of the
17048 variable are encountered by the linker, the first is selected and
17049 the remainder are discarded. Following usage by the Microsoft
17050 compiler, the linker is told _not_ to warn about size or content
17051 differences of the multiple definitions.
17053 Although the primary usage of this attribute is for POD types, the
17054 attribute can also be applied to global C++ objects that are
17055 initialized by a constructor. In this case, the static
17056 initialization and destruction code for the object is emitted in
17057 each translation defining the object, but the calls to the
17058 constructor and destructor are protected by a link-once guard
17061 The `selectany' attribute is only available on Microsoft Windows
17062 targets. You can use `__declspec (selectany)' as a synonym for
17063 `__attribute__ ((selectany))' for compatibility with other
17067 The `weak' attribute is described in *Note Function Attributes::.
17070 The `dllimport' attribute is described in *Note Function
17074 The `dllexport' attribute is described in *Note Function
17078 5.32.1 M32R/D Variable Attributes
17079 ---------------------------------
17081 One attribute is currently defined for the M32R/D.
17083 `model (MODEL-NAME)'
17084 Use this attribute on the M32R/D to set the addressability of an
17085 object. The identifier MODEL-NAME is one of `small', `medium', or
17086 `large', representing each of the code models.
17088 Small model objects live in the lower 16MB of memory (so that their
17089 addresses can be loaded with the `ld24' instruction).
17091 Medium and large model objects may live anywhere in the 32-bit
17092 address space (the compiler will generate `seth/add3' instructions
17093 to load their addresses).
17095 5.32.2 i386 Variable Attributes
17096 -------------------------------
17098 Two attributes are currently defined for i386 configurations:
17099 `ms_struct' and `gcc_struct'
17103 If `packed' is used on a structure, or if bit-fields are used it
17104 may be that the Microsoft ABI packs them differently than GCC
17105 would normally pack them. Particularly when moving packed data
17106 between functions compiled with GCC and the native Microsoft
17107 compiler (either via function call or as data in a file), it may
17108 be necessary to access either format.
17110 Currently `-m[no-]ms-bitfields' is provided for the Microsoft
17111 Windows X86 compilers to match the native Microsoft compiler.
17113 The Microsoft structure layout algorithm is fairly simple with the
17114 exception of the bitfield packing:
17116 The padding and alignment of members of structures and whether a
17117 bit field can straddle a storage-unit boundary
17119 1. Structure members are stored sequentially in the order in
17120 which they are declared: the first member has the lowest
17121 memory address and the last member the highest.
17123 2. Every data object has an alignment-requirement. The
17124 alignment-requirement for all data except structures, unions,
17125 and arrays is either the size of the object or the current
17126 packing size (specified with either the aligned attribute or
17127 the pack pragma), whichever is less. For structures, unions,
17128 and arrays, the alignment-requirement is the largest
17129 alignment-requirement of its members. Every object is
17130 allocated an offset so that:
17132 offset % alignment-requirement == 0
17134 3. Adjacent bit fields are packed into the same 1-, 2-, or
17135 4-byte allocation unit if the integral types are the same
17136 size and if the next bit field fits into the current
17137 allocation unit without crossing the boundary imposed by the
17138 common alignment requirements of the bit fields.
17140 Handling of zero-length bitfields:
17142 MSVC interprets zero-length bitfields in the following ways:
17144 1. If a zero-length bitfield is inserted between two bitfields
17145 that would normally be coalesced, the bitfields will not be
17152 unsigned long bf_1 : 12;
17154 unsigned long bf_2 : 12;
17157 The size of `t1' would be 8 bytes with the zero-length
17158 bitfield. If the zero-length bitfield were removed, `t1''s
17159 size would be 4 bytes.
17161 2. If a zero-length bitfield is inserted after a bitfield,
17162 `foo', and the alignment of the zero-length bitfield is
17163 greater than the member that follows it, `bar', `bar' will be
17164 aligned as the type of the zero-length bitfield.
17182 For `t2', `bar' will be placed at offset 2, rather than
17183 offset 1. Accordingly, the size of `t2' will be 4. For
17184 `t3', the zero-length bitfield will not affect the alignment
17185 of `bar' or, as a result, the size of the structure.
17187 Taking this into account, it is important to note the
17190 1. If a zero-length bitfield follows a normal bitfield, the
17191 type of the zero-length bitfield may affect the
17192 alignment of the structure as whole. For example, `t2'
17193 has a size of 4 bytes, since the zero-length bitfield
17194 follows a normal bitfield, and is of type short.
17196 2. Even if a zero-length bitfield is not followed by a
17197 normal bitfield, it may still affect the alignment of
17206 Here, `t4' will take up 4 bytes.
17208 3. Zero-length bitfields following non-bitfield members are
17218 Here, `t5' will take up 2 bytes.
17220 5.32.3 PowerPC Variable Attributes
17221 ----------------------------------
17223 Three attributes currently are defined for PowerPC configurations:
17224 `altivec', `ms_struct' and `gcc_struct'.
17226 For full documentation of the struct attributes please see the
17227 documentation in the *Note i386 Variable Attributes::, section.
17229 For documentation of `altivec' attribute please see the documentation
17230 in the *Note PowerPC Type Attributes::, section.
17232 5.32.4 Xstormy16 Variable Attributes
17233 ------------------------------------
17235 One attribute is currently defined for xstormy16 configurations:
17239 If a variable has the `below100' attribute (`BELOW100' is allowed
17240 also), GCC will place the variable in the first 0x100 bytes of
17241 memory and use special opcodes to access it. Such variables will
17242 be placed in either the `.bss_below100' section or the
17243 `.data_below100' section.
17247 File: gcc.info, Node: Type Attributes, Next: Alignment, Prev: Variable Attributes, Up: C Extensions
17249 5.33 Specifying Attributes of Types
17250 ===================================
17252 The keyword `__attribute__' allows you to specify special attributes of
17253 `struct' and `union' types when you define such types. This keyword is
17254 followed by an attribute specification inside double parentheses.
17255 Seven attributes are currently defined for types: `aligned', `packed',
17256 `transparent_union', `unused', `deprecated', `visibility', and
17257 `may_alias'. Other attributes are defined for functions (*note
17258 Function Attributes::) and for variables (*note Variable Attributes::).
17260 You may also specify any one of these attributes with `__' preceding
17261 and following its keyword. This allows you to use these attributes in
17262 header files without being concerned about a possible macro of the same
17263 name. For example, you may use `__aligned__' instead of `aligned'.
17265 You may specify type attributes either in a `typedef' declaration or
17266 in an enum, struct or union type declaration or definition.
17268 For an enum, struct or union type, you may specify attributes either
17269 between the enum, struct or union tag and the name of the type, or just
17270 past the closing curly brace of the _definition_. The former syntax is
17273 *Note Attribute Syntax::, for details of the exact syntax for using
17276 `aligned (ALIGNMENT)'
17277 This attribute specifies a minimum alignment (in bytes) for
17278 variables of the specified type. For example, the declarations:
17280 struct S { short f[3]; } __attribute__ ((aligned (8)));
17281 typedef int more_aligned_int __attribute__ ((aligned (8)));
17283 force the compiler to insure (as far as it can) that each variable
17284 whose type is `struct S' or `more_aligned_int' will be allocated
17285 and aligned _at least_ on a 8-byte boundary. On a SPARC, having
17286 all variables of type `struct S' aligned to 8-byte boundaries
17287 allows the compiler to use the `ldd' and `std' (doubleword load and
17288 store) instructions when copying one variable of type `struct S' to
17289 another, thus improving run-time efficiency.
17291 Note that the alignment of any given `struct' or `union' type is
17292 required by the ISO C standard to be at least a perfect multiple of
17293 the lowest common multiple of the alignments of all of the members
17294 of the `struct' or `union' in question. This means that you _can_
17295 effectively adjust the alignment of a `struct' or `union' type by
17296 attaching an `aligned' attribute to any one of the members of such
17297 a type, but the notation illustrated in the example above is a
17298 more obvious, intuitive, and readable way to request the compiler
17299 to adjust the alignment of an entire `struct' or `union' type.
17301 As in the preceding example, you can explicitly specify the
17302 alignment (in bytes) that you wish the compiler to use for a given
17303 `struct' or `union' type. Alternatively, you can leave out the
17304 alignment factor and just ask the compiler to align a type to the
17305 maximum useful alignment for the target machine you are compiling
17306 for. For example, you could write:
17308 struct S { short f[3]; } __attribute__ ((aligned));
17310 Whenever you leave out the alignment factor in an `aligned'
17311 attribute specification, the compiler automatically sets the
17312 alignment for the type to the largest alignment which is ever used
17313 for any data type on the target machine you are compiling for.
17314 Doing this can often make copy operations more efficient, because
17315 the compiler can use whatever instructions copy the biggest chunks
17316 of memory when performing copies to or from the variables which
17317 have types that you have aligned this way.
17319 In the example above, if the size of each `short' is 2 bytes, then
17320 the size of the entire `struct S' type is 6 bytes. The smallest
17321 power of two which is greater than or equal to that is 8, so the
17322 compiler sets the alignment for the entire `struct S' type to 8
17325 Note that although you can ask the compiler to select a
17326 time-efficient alignment for a given type and then declare only
17327 individual stand-alone objects of that type, the compiler's
17328 ability to select a time-efficient alignment is primarily useful
17329 only when you plan to create arrays of variables having the
17330 relevant (efficiently aligned) type. If you declare or use arrays
17331 of variables of an efficiently-aligned type, then it is likely
17332 that your program will also be doing pointer arithmetic (or
17333 subscripting, which amounts to the same thing) on pointers to the
17334 relevant type, and the code that the compiler generates for these
17335 pointer arithmetic operations will often be more efficient for
17336 efficiently-aligned types than for other types.
17338 The `aligned' attribute can only increase the alignment; but you
17339 can decrease it by specifying `packed' as well. See below.
17341 Note that the effectiveness of `aligned' attributes may be limited
17342 by inherent limitations in your linker. On many systems, the
17343 linker is only able to arrange for variables to be aligned up to a
17344 certain maximum alignment. (For some linkers, the maximum
17345 supported alignment may be very very small.) If your linker is
17346 only able to align variables up to a maximum of 8 byte alignment,
17347 then specifying `aligned(16)' in an `__attribute__' will still
17348 only provide you with 8 byte alignment. See your linker
17349 documentation for further information.
17352 This attribute, attached to `struct' or `union' type definition,
17353 specifies that each member (other than zero-width bitfields) of
17354 the structure or union is placed to minimize the memory required.
17355 When attached to an `enum' definition, it indicates that the
17356 smallest integral type should be used.
17358 Specifying this attribute for `struct' and `union' types is
17359 equivalent to specifying the `packed' attribute on each of the
17360 structure or union members. Specifying the `-fshort-enums' flag
17361 on the line is equivalent to specifying the `packed' attribute on
17362 all `enum' definitions.
17364 In the following example `struct my_packed_struct''s members are
17365 packed closely together, but the internal layout of its `s' member
17366 is not packed--to do that, `struct my_unpacked_struct' would need
17369 struct my_unpacked_struct
17375 struct __attribute__ ((__packed__)) my_packed_struct
17379 struct my_unpacked_struct s;
17382 You may only specify this attribute on the definition of a `enum',
17383 `struct' or `union', not on a `typedef' which does not also define
17384 the enumerated type, structure or union.
17386 `transparent_union'
17387 This attribute, attached to a `union' type definition, indicates
17388 that any function parameter having that union type causes calls to
17389 that function to be treated in a special way.
17391 First, the argument corresponding to a transparent union type can
17392 be of any type in the union; no cast is required. Also, if the
17393 union contains a pointer type, the corresponding argument can be a
17394 null pointer constant or a void pointer expression; and if the
17395 union contains a void pointer type, the corresponding argument can
17396 be any pointer expression. If the union member type is a pointer,
17397 qualifiers like `const' on the referenced type must be respected,
17398 just as with normal pointer conversions.
17400 Second, the argument is passed to the function using the calling
17401 conventions of the first member of the transparent union, not the
17402 calling conventions of the union itself. All members of the union
17403 must have the same machine representation; this is necessary for
17404 this argument passing to work properly.
17406 Transparent unions are designed for library functions that have
17407 multiple interfaces for compatibility reasons. For example,
17408 suppose the `wait' function must accept either a value of type
17409 `int *' to comply with Posix, or a value of type `union wait *' to
17410 comply with the 4.1BSD interface. If `wait''s parameter were
17411 `void *', `wait' would accept both kinds of arguments, but it
17412 would also accept any other pointer type and this would make
17413 argument type checking less useful. Instead, `<sys/wait.h>' might
17414 define the interface as follows:
17420 } wait_status_ptr_t __attribute__ ((__transparent_union__));
17422 pid_t wait (wait_status_ptr_t);
17424 This interface allows either `int *' or `union wait *' arguments
17425 to be passed, using the `int *' calling convention. The program
17426 can call `wait' with arguments of either type:
17428 int w1 () { int w; return wait (&w); }
17429 int w2 () { union wait w; return wait (&w); }
17431 With this interface, `wait''s implementation might look like this:
17433 pid_t wait (wait_status_ptr_t p)
17435 return waitpid (-1, p.__ip, 0);
17439 When attached to a type (including a `union' or a `struct'), this
17440 attribute means that variables of that type are meant to appear
17441 possibly unused. GCC will not produce a warning for any variables
17442 of that type, even if the variable appears to do nothing. This is
17443 often the case with lock or thread classes, which are usually
17444 defined and then not referenced, but contain constructors and
17445 destructors that have nontrivial bookkeeping functions.
17448 The `deprecated' attribute results in a warning if the type is
17449 used anywhere in the source file. This is useful when identifying
17450 types that are expected to be removed in a future version of a
17451 program. If possible, the warning also includes the location of
17452 the declaration of the deprecated type, to enable users to easily
17453 find further information about why the type is deprecated, or what
17454 they should do instead. Note that the warnings only occur for
17455 uses and then only if the type is being applied to an identifier
17456 that itself is not being declared as deprecated.
17458 typedef int T1 __attribute__ ((deprecated));
17462 typedef T1 T3 __attribute__ ((deprecated));
17463 T3 z __attribute__ ((deprecated));
17465 results in a warning on line 2 and 3 but not lines 4, 5, or 6. No
17466 warning is issued for line 4 because T2 is not explicitly
17467 deprecated. Line 5 has no warning because T3 is explicitly
17468 deprecated. Similarly for line 6.
17470 The `deprecated' attribute can also be used for functions and
17471 variables (*note Function Attributes::, *note Variable
17475 Accesses to objects with types with this attribute are not
17476 subjected to type-based alias analysis, but are instead assumed to
17477 be able to alias any other type of objects, just like the `char'
17478 type. See `-fstrict-aliasing' for more information on aliasing
17483 typedef short __attribute__((__may_alias__)) short_a;
17488 int a = 0x12345678;
17489 short_a *b = (short_a *) &a;
17493 if (a == 0x12345678)
17499 If you replaced `short_a' with `short' in the variable
17500 declaration, the above program would abort when compiled with
17501 `-fstrict-aliasing', which is on by default at `-O2' or above in
17502 recent GCC versions.
17505 In C++, attribute visibility (*note Function Attributes::) can
17506 also be applied to class, struct, union and enum types. Unlike
17507 other type attributes, the attribute must appear between the
17508 initial keyword and the name of the type; it cannot appear after
17509 the body of the type.
17511 Note that the type visibility is applied to vague linkage entities
17512 associated with the class (vtable, typeinfo node, etc.). In
17513 particular, if a class is thrown as an exception in one shared
17514 object and caught in another, the class must have default
17515 visibility. Otherwise the two shared objects will be unable to
17516 use the same typeinfo node and exception handling will break.
17518 5.33.1 ARM Type Attributes
17519 --------------------------
17521 On those ARM targets that support `dllimport' (such as Symbian
17522 OS), you can use the `notshared' attribute to indicate that the virtual
17523 table and other similar data for a class should not be exported from a
17526 class __declspec(notshared) C {
17528 __declspec(dllimport) C();
17532 __declspec(dllexport)
17535 In this code, `C::C' is exported from the current DLL, but the
17536 virtual table for `C' is not exported. (You can use `__attribute__'
17537 instead of `__declspec' if you prefer, but most Symbian OS code uses
17540 5.33.2 i386 Type Attributes
17541 ---------------------------
17543 Two attributes are currently defined for i386 configurations:
17544 `ms_struct' and `gcc_struct'
17548 If `packed' is used on a structure, or if bit-fields are used it
17549 may be that the Microsoft ABI packs them differently than GCC
17550 would normally pack them. Particularly when moving packed data
17551 between functions compiled with GCC and the native Microsoft
17552 compiler (either via function call or as data in a file), it may
17553 be necessary to access either format.
17555 Currently `-m[no-]ms-bitfields' is provided for the Microsoft
17556 Windows X86 compilers to match the native Microsoft compiler.
17558 To specify multiple attributes, separate them by commas within the
17559 double parentheses: for example, `__attribute__ ((aligned (16),
17562 5.33.3 PowerPC Type Attributes
17563 ------------------------------
17565 Three attributes currently are defined for PowerPC configurations:
17566 `altivec', `ms_struct' and `gcc_struct'.
17568 For full documentation of the struct attributes please see the
17569 documentation in the *Note i386 Type Attributes::, section.
17571 The `altivec' attribute allows one to declare AltiVec vector data
17572 types supported by the AltiVec Programming Interface Manual. The
17573 attribute requires an argument to specify one of three vector types:
17574 `vector__', `pixel__' (always followed by unsigned short), and `bool__'
17575 (always followed by unsigned).
17577 __attribute__((altivec(vector__)))
17578 __attribute__((altivec(pixel__))) unsigned short
17579 __attribute__((altivec(bool__))) unsigned
17581 These attributes mainly are intended to support the `__vector',
17582 `__pixel', and `__bool' AltiVec keywords.
17585 File: gcc.info, Node: Inline, Next: Extended Asm, Prev: Alignment, Up: C Extensions
17587 5.34 An Inline Function is As Fast As a Macro
17588 =============================================
17590 By declaring a function `inline', you can direct GCC to integrate that
17591 function's code into the code for its callers. This makes execution
17592 faster by eliminating the function-call overhead; in addition, if any
17593 of the actual argument values are constant, their known values may
17594 permit simplifications at compile time so that not all of the inline
17595 function's code needs to be included. The effect on code size is less
17596 predictable; object code may be larger or smaller with function
17597 inlining, depending on the particular case. Inlining of functions is an
17598 optimization and it really "works" only in optimizing compilation. If
17599 you don't use `-O', no function is really inline.
17601 Inline functions are included in the ISO C99 standard, but there are
17602 currently substantial differences between what GCC implements and what
17603 the ISO C99 standard requires. GCC will fully support C99 inline
17604 functions in version 4.3. The traditional GCC handling of inline
17605 functions will still be available with `-std=gnu89', `-fgnu89-inline'
17606 or when `gnu_inline' attribute is present on all inline declarations.
17607 The preprocessor macros `__GNUC_GNU_INLINE__' and
17608 `__GNUC_STDC_INLINE__' may be used to determine the handling of
17609 `inline' during a particular compilation (*note Common Predefined
17610 Macros: (cpp)Common Predefined Macros.).
17612 To declare a function inline, use the `inline' keyword in its
17613 declaration, like this:
17621 (If you are writing a header file to be included in ISO C programs,
17622 write `__inline__' instead of `inline'. *Note Alternate Keywords::.)
17623 You can also make all "simple enough" functions inline with the option
17624 `-finline-functions'.
17626 Note that certain usages in a function definition can make it
17627 unsuitable for inline substitution. Among these usages are: use of
17628 varargs, use of alloca, use of variable sized data types (*note
17629 Variable Length::), use of computed goto (*note Labels as Values::),
17630 use of nonlocal goto, and nested functions (*note Nested Functions::).
17631 Using `-Winline' will warn when a function marked `inline' could not be
17632 substituted, and will give the reason for the failure.
17634 Note that in C and Objective-C, unlike C++, the `inline' keyword does
17635 not affect the linkage of the function.
17637 GCC automatically inlines member functions defined within the class
17638 body of C++ programs even if they are not explicitly declared `inline'.
17639 (You can override this with `-fno-default-inline'; *note Options
17640 Controlling C++ Dialect: C++ Dialect Options.)
17642 When a function is both inline and `static', if all calls to the
17643 function are integrated into the caller, and the function's address is
17644 never used, then the function's own assembler code is never referenced.
17645 In this case, GCC does not actually output assembler code for the
17646 function, unless you specify the option `-fkeep-inline-functions'.
17647 Some calls cannot be integrated for various reasons (in particular,
17648 calls that precede the function's definition cannot be integrated, and
17649 neither can recursive calls within the definition). If there is a
17650 nonintegrated call, then the function is compiled to assembler code as
17651 usual. The function must also be compiled as usual if the program
17652 refers to its address, because that can't be inlined.
17654 When an inline function is not `static', then the compiler must assume
17655 that there may be calls from other source files; since a global symbol
17656 can be defined only once in any program, the function must not be
17657 defined in the other source files, so the calls therein cannot be
17658 integrated. Therefore, a non-`static' inline function is always
17659 compiled on its own in the usual fashion.
17661 If you specify both `inline' and `extern' in the function definition,
17662 then the definition is used only for inlining. In no case is the
17663 function compiled on its own, not even if you refer to its address
17664 explicitly. Such an address becomes an external reference, as if you
17665 had only declared the function, and had not defined it.
17667 This combination of `inline' and `extern' has almost the effect of a
17668 macro. The way to use it is to put a function definition in a header
17669 file with these keywords, and put another copy of the definition
17670 (lacking `inline' and `extern') in a library file. The definition in
17671 the header file will cause most calls to the function to be inlined.
17672 If any uses of the function remain, they will refer to the single copy
17675 Since GCC 4.3 will implement ISO C99 semantics for inline functions,
17676 it is simplest to use `static inline' only to guarantee compatibility.
17677 (The existing semantics will remain available when `-std=gnu89' is
17678 specified, but eventually the default will be `-std=gnu99'; that will
17679 implement the C99 semantics, though it does not do so in versions of
17680 GCC before 4.3. After the default changes, the existing semantics will
17681 still be available via the `-fgnu89-inline' option or the `gnu_inline'
17682 function attribute.)
17684 GCC does not inline any functions when not optimizing unless you
17685 specify the `always_inline' attribute for the function, like this:
17688 inline void foo (const char) __attribute__((always_inline));
17691 File: gcc.info, Node: Extended Asm, Next: Constraints, Prev: Inline, Up: C Extensions
17693 5.35 Assembler Instructions with C Expression Operands
17694 ======================================================
17696 In an assembler instruction using `asm', you can specify the operands
17697 of the instruction using C expressions. This means you need not guess
17698 which registers or memory locations will contain the data you want to
17701 You must specify an assembler instruction template much like what
17702 appears in a machine description, plus an operand constraint string for
17705 For example, here is how to use the 68881's `fsinx' instruction:
17707 asm ("fsinx %1,%0" : "=f" (result) : "f" (angle));
17709 Here `angle' is the C expression for the input operand while `result'
17710 is that of the output operand. Each has `"f"' as its operand
17711 constraint, saying that a floating point register is required. The `='
17712 in `=f' indicates that the operand is an output; all output operands'
17713 constraints must use `='. The constraints use the same language used
17714 in the machine description (*note Constraints::).
17716 Each operand is described by an operand-constraint string followed by
17717 the C expression in parentheses. A colon separates the assembler
17718 template from the first output operand and another separates the last
17719 output operand from the first input, if any. Commas separate the
17720 operands within each group. The total number of operands is currently
17721 limited to 30; this limitation may be lifted in some future version of
17724 If there are no output operands but there are input operands, you must
17725 place two consecutive colons surrounding the place where the output
17728 As of GCC version 3.1, it is also possible to specify input and output
17729 operands using symbolic names which can be referenced within the
17730 assembler code. These names are specified inside square brackets
17731 preceding the constraint string, and can be referenced inside the
17732 assembler code using `%[NAME]' instead of a percentage sign followed by
17733 the operand number. Using named operands the above example could look
17736 asm ("fsinx %[angle],%[output]"
17737 : [output] "=f" (result)
17738 : [angle] "f" (angle));
17740 Note that the symbolic operand names have no relation whatsoever to
17741 other C identifiers. You may use any name you like, even those of
17742 existing C symbols, but you must ensure that no two operands within the
17743 same assembler construct use the same symbolic name.
17745 Output operand expressions must be lvalues; the compiler can check
17746 this. The input operands need not be lvalues. The compiler cannot
17747 check whether the operands have data types that are reasonable for the
17748 instruction being executed. It does not parse the assembler instruction
17749 template and does not know what it means or even whether it is valid
17750 assembler input. The extended `asm' feature is most often used for
17751 machine instructions the compiler itself does not know exist. If the
17752 output expression cannot be directly addressed (for example, it is a
17753 bit-field), your constraint must allow a register. In that case, GCC
17754 will use the register as the output of the `asm', and then store that
17755 register into the output.
17757 The ordinary output operands must be write-only; GCC will assume that
17758 the values in these operands before the instruction are dead and need
17759 not be generated. Extended asm supports input-output or read-write
17760 operands. Use the constraint character `+' to indicate such an operand
17761 and list it with the output operands. You should only use read-write
17762 operands when the constraints for the operand (or the operand in which
17763 only some of the bits are to be changed) allow a register.
17765 You may, as an alternative, logically split its function into two
17766 separate operands, one input operand and one write-only output operand.
17767 The connection between them is expressed by constraints which say they
17768 need to be in the same location when the instruction executes. You can
17769 use the same C expression for both operands, or different expressions.
17770 For example, here we write the (fictitious) `combine' instruction with
17771 `bar' as its read-only source operand and `foo' as its read-write
17774 asm ("combine %2,%0" : "=r" (foo) : "0" (foo), "g" (bar));
17776 The constraint `"0"' for operand 1 says that it must occupy the same
17777 location as operand 0. A number in constraint is allowed only in an
17778 input operand and it must refer to an output operand.
17780 Only a number in the constraint can guarantee that one operand will be
17781 in the same place as another. The mere fact that `foo' is the value of
17782 both operands is not enough to guarantee that they will be in the same
17783 place in the generated assembler code. The following would not work
17786 asm ("combine %2,%0" : "=r" (foo) : "r" (foo), "g" (bar));
17788 Various optimizations or reloading could cause operands 0 and 1 to be
17789 in different registers; GCC knows no reason not to do so. For example,
17790 the compiler might find a copy of the value of `foo' in one register and
17791 use it for operand 1, but generate the output operand 0 in a different
17792 register (copying it afterward to `foo''s own address). Of course,
17793 since the register for operand 1 is not even mentioned in the assembler
17794 code, the result will not work, but GCC can't tell that.
17796 As of GCC version 3.1, one may write `[NAME]' instead of the operand
17797 number for a matching constraint. For example:
17799 asm ("cmoveq %1,%2,%[result]"
17800 : [result] "=r"(result)
17801 : "r" (test), "r"(new), "[result]"(old));
17803 Sometimes you need to make an `asm' operand be a specific register,
17804 but there's no matching constraint letter for that register _by
17805 itself_. To force the operand into that register, use a local variable
17806 for the operand and specify the register in the variable declaration.
17807 *Note Explicit Reg Vars::. Then for the `asm' operand, use any
17808 register constraint letter that matches the register:
17810 register int *p1 asm ("r0") = ...;
17811 register int *p2 asm ("r1") = ...;
17812 register int *result asm ("r0");
17813 asm ("sysint" : "=r" (result) : "0" (p1), "r" (p2));
17815 In the above example, beware that a register that is call-clobbered by
17816 the target ABI will be overwritten by any function call in the
17817 assignment, including library calls for arithmetic operators. Assuming
17818 it is a call-clobbered register, this may happen to `r0' above by the
17819 assignment to `p2'. If you have to use such a register, use temporary
17820 variables for expressions between the register assignment and use:
17823 register int *p1 asm ("r0") = ...;
17824 register int *p2 asm ("r1") = t1;
17825 register int *result asm ("r0");
17826 asm ("sysint" : "=r" (result) : "0" (p1), "r" (p2));
17828 Some instructions clobber specific hard registers. To describe this,
17829 write a third colon after the input operands, followed by the names of
17830 the clobbered hard registers (given as strings). Here is a realistic
17831 example for the VAX:
17833 asm volatile ("movc3 %0,%1,%2"
17835 : "g" (from), "g" (to), "g" (count)
17836 : "r0", "r1", "r2", "r3", "r4", "r5");
17838 You may not write a clobber description in a way that overlaps with an
17839 input or output operand. For example, you may not have an operand
17840 describing a register class with one member if you mention that register
17841 in the clobber list. Variables declared to live in specific registers
17842 (*note Explicit Reg Vars::), and used as asm input or output operands
17843 must have no part mentioned in the clobber description. There is no
17844 way for you to specify that an input operand is modified without also
17845 specifying it as an output operand. Note that if all the output
17846 operands you specify are for this purpose (and hence unused), you will
17847 then also need to specify `volatile' for the `asm' construct, as
17848 described below, to prevent GCC from deleting the `asm' statement as
17851 If you refer to a particular hardware register from the assembler code,
17852 you will probably have to list the register after the third colon to
17853 tell the compiler the register's value is modified. In some assemblers,
17854 the register names begin with `%'; to produce one `%' in the assembler
17855 code, you must write `%%' in the input.
17857 If your assembler instruction can alter the condition code register,
17858 add `cc' to the list of clobbered registers. GCC on some machines
17859 represents the condition codes as a specific hardware register; `cc'
17860 serves to name this register. On other machines, the condition code is
17861 handled differently, and specifying `cc' has no effect. But it is
17862 valid no matter what the machine.
17864 If your assembler instructions access memory in an unpredictable
17865 fashion, add `memory' to the list of clobbered registers. This will
17866 cause GCC to not keep memory values cached in registers across the
17867 assembler instruction and not optimize stores or loads to that memory.
17868 You will also want to add the `volatile' keyword if the memory affected
17869 is not listed in the inputs or outputs of the `asm', as the `memory'
17870 clobber does not count as a side-effect of the `asm'. If you know how
17871 large the accessed memory is, you can add it as input or output but if
17872 this is not known, you should add `memory'. As an example, if you
17873 access ten bytes of a string, you can use a memory input like:
17875 {"m"( ({ struct { char x[10]; } *p = (void *)ptr ; *p; }) )}.
17877 Note that in the following example the memory input is necessary,
17878 otherwise GCC might optimize the store to `x' away:
17884 asm ("magic stuff accessing an 'int' pointed to by '%1'"
17885 "=&d" (r) : "a" (y), "m" (*y));
17889 You can put multiple assembler instructions together in a single `asm'
17890 template, separated by the characters normally used in assembly code
17891 for the system. A combination that works in most places is a newline
17892 to break the line, plus a tab character to move to the instruction field
17893 (written as `\n\t'). Sometimes semicolons can be used, if the
17894 assembler allows semicolons as a line-breaking character. Note that
17895 some assembler dialects use semicolons to start a comment. The input
17896 operands are guaranteed not to use any of the clobbered registers, and
17897 neither will the output operands' addresses, so you can read and write
17898 the clobbered registers as many times as you like. Here is an example
17899 of multiple instructions in a template; it assumes the subroutine
17900 `_foo' accepts arguments in registers 9 and 10:
17902 asm ("movl %0,r9\n\tmovl %1,r10\n\tcall _foo"
17904 : "g" (from), "g" (to)
17907 Unless an output operand has the `&' constraint modifier, GCC may
17908 allocate it in the same register as an unrelated input operand, on the
17909 assumption the inputs are consumed before the outputs are produced.
17910 This assumption may be false if the assembler code actually consists of
17911 more than one instruction. In such a case, use `&' for each output
17912 operand that may not overlap an input. *Note Modifiers::.
17914 If you want to test the condition code produced by an assembler
17915 instruction, you must include a branch and a label in the `asm'
17916 construct, as follows:
17918 asm ("clr %0\n\tfrob %1\n\tbeq 0f\n\tmov #1,%0\n0:"
17922 This assumes your assembler supports local labels, as the GNU assembler
17923 and most Unix assemblers do.
17925 Speaking of labels, jumps from one `asm' to another are not supported.
17926 The compiler's optimizers do not know about these jumps, and therefore
17927 they cannot take account of them when deciding how to optimize.
17929 Usually the most convenient way to use these `asm' instructions is to
17930 encapsulate them in macros that look like functions. For example,
17933 ({ double __value, __arg = (x); \
17934 asm ("fsinx %1,%0": "=f" (__value): "f" (__arg)); \
17937 Here the variable `__arg' is used to make sure that the instruction
17938 operates on a proper `double' value, and to accept only those arguments
17939 `x' which can convert automatically to a `double'.
17941 Another way to make sure the instruction operates on the correct data
17942 type is to use a cast in the `asm'. This is different from using a
17943 variable `__arg' in that it converts more different types. For
17944 example, if the desired type were `int', casting the argument to `int'
17945 would accept a pointer with no complaint, while assigning the argument
17946 to an `int' variable named `__arg' would warn about using a pointer
17947 unless the caller explicitly casts it.
17949 If an `asm' has output operands, GCC assumes for optimization purposes
17950 the instruction has no side effects except to change the output
17951 operands. This does not mean instructions with a side effect cannot be
17952 used, but you must be careful, because the compiler may eliminate them
17953 if the output operands aren't used, or move them out of loops, or
17954 replace two with one if they constitute a common subexpression. Also,
17955 if your instruction does have a side effect on a variable that otherwise
17956 appears not to change, the old value of the variable may be reused later
17957 if it happens to be found in a register.
17959 You can prevent an `asm' instruction from being deleted by writing the
17960 keyword `volatile' after the `asm'. For example:
17962 #define get_and_set_priority(new) \
17964 asm volatile ("get_and_set_priority %0, %1" \
17965 : "=g" (__old) : "g" (new)); \
17968 The `volatile' keyword indicates that the instruction has important
17969 side-effects. GCC will not delete a volatile `asm' if it is reachable.
17970 (The instruction can still be deleted if GCC can prove that
17971 control-flow will never reach the location of the instruction.) Note
17972 that even a volatile `asm' instruction can be moved relative to other
17973 code, including across jump instructions. For example, on many targets
17974 there is a system register which can be set to control the rounding
17975 mode of floating point operations. You might try setting it with a
17976 volatile `asm', like this PowerPC example:
17978 asm volatile("mtfsf 255,%0" : : "f" (fpenv));
17981 This will not work reliably, as the compiler may move the addition back
17982 before the volatile `asm'. To make it work you need to add an
17983 artificial dependency to the `asm' referencing a variable in the code
17984 you don't want moved, for example:
17986 asm volatile ("mtfsf 255,%1" : "=X"(sum): "f"(fpenv));
17989 Similarly, you can't expect a sequence of volatile `asm' instructions
17990 to remain perfectly consecutive. If you want consecutive output, use a
17991 single `asm'. Also, GCC will perform some optimizations across a
17992 volatile `asm' instruction; GCC does not "forget everything" when it
17993 encounters a volatile `asm' instruction the way some other compilers do.
17995 An `asm' instruction without any output operands will be treated
17996 identically to a volatile `asm' instruction.
17998 It is a natural idea to look for a way to give access to the condition
17999 code left by the assembler instruction. However, when we attempted to
18000 implement this, we found no way to make it work reliably. The problem
18001 is that output operands might need reloading, which would result in
18002 additional following "store" instructions. On most machines, these
18003 instructions would alter the condition code before there was time to
18004 test it. This problem doesn't arise for ordinary "test" and "compare"
18005 instructions because they don't have any output operands.
18007 For reasons similar to those described above, it is not possible to
18008 give an assembler instruction access to the condition code left by
18009 previous instructions.
18011 If you are writing a header file that should be includable in ISO C
18012 programs, write `__asm__' instead of `asm'. *Note Alternate Keywords::.
18014 5.35.1 Size of an `asm'
18015 -----------------------
18017 Some targets require that GCC track the size of each instruction used in
18018 order to generate correct code. Because the final length of an `asm'
18019 is only known by the assembler, GCC must make an estimate as to how big
18020 it will be. The estimate is formed by counting the number of
18021 statements in the pattern of the `asm' and multiplying that by the
18022 length of the longest instruction on that processor. Statements in the
18023 `asm' are identified by newline characters and whatever statement
18024 separator characters are supported by the assembler; on most processors
18025 this is the ``;'' character.
18027 Normally, GCC's estimate is perfectly adequate to ensure that correct
18028 code is generated, but it is possible to confuse the compiler if you use
18029 pseudo instructions or assembler macros that expand into multiple real
18030 instructions or if you use assembler directives that expand to more
18031 space in the object file than would be needed for a single instruction.
18032 If this happens then the assembler will produce a diagnostic saying that
18033 a label is unreachable.
18035 5.35.2 i386 floating point asm operands
18036 ---------------------------------------
18038 There are several rules on the usage of stack-like regs in asm_operands
18039 insns. These rules apply only to the operands that are stack-like regs:
18041 1. Given a set of input regs that die in an asm_operands, it is
18042 necessary to know which are implicitly popped by the asm, and
18043 which must be explicitly popped by gcc.
18045 An input reg that is implicitly popped by the asm must be
18046 explicitly clobbered, unless it is constrained to match an output
18049 2. For any input reg that is implicitly popped by an asm, it is
18050 necessary to know how to adjust the stack to compensate for the
18051 pop. If any non-popped input is closer to the top of the
18052 reg-stack than the implicitly popped reg, it would not be possible
18053 to know what the stack looked like--it's not clear how the rest of
18054 the stack "slides up".
18056 All implicitly popped input regs must be closer to the top of the
18057 reg-stack than any input that is not implicitly popped.
18059 It is possible that if an input dies in an insn, reload might use
18060 the input reg for an output reload. Consider this example:
18062 asm ("foo" : "=t" (a) : "f" (b));
18064 This asm says that input B is not popped by the asm, and that the
18065 asm pushes a result onto the reg-stack, i.e., the stack is one
18066 deeper after the asm than it was before. But, it is possible that
18067 reload will think that it can use the same reg for both the input
18068 and the output, if input B dies in this insn.
18070 If any input operand uses the `f' constraint, all output reg
18071 constraints must use the `&' earlyclobber.
18073 The asm above would be written as
18075 asm ("foo" : "=&t" (a) : "f" (b));
18077 3. Some operands need to be in particular places on the stack. All
18078 output operands fall in this category--there is no other way to
18079 know which regs the outputs appear in unless the user indicates
18080 this in the constraints.
18082 Output operands must specifically indicate which reg an output
18083 appears in after an asm. `=f' is not allowed: the operand
18084 constraints must select a class with a single reg.
18086 4. Output operands may not be "inserted" between existing stack regs.
18087 Since no 387 opcode uses a read/write operand, all output operands
18088 are dead before the asm_operands, and are pushed by the
18089 asm_operands. It makes no sense to push anywhere but the top of
18092 Output operands must start at the top of the reg-stack: output
18093 operands may not "skip" a reg.
18095 5. Some asm statements may need extra stack space for internal
18096 calculations. This can be guaranteed by clobbering stack registers
18097 unrelated to the inputs and outputs.
18100 Here are a couple of reasonable asms to want to write. This asm takes
18101 one input, which is internally popped, and produces two outputs.
18103 asm ("fsincos" : "=t" (cos), "=u" (sin) : "0" (inp));
18105 This asm takes two inputs, which are popped by the `fyl2xp1' opcode,
18106 and replaces them with one output. The user must code the `st(1)'
18107 clobber for reg-stack.c to know that `fyl2xp1' pops both inputs.
18109 asm ("fyl2xp1" : "=t" (result) : "0" (x), "u" (y) : "st(1)");
18112 File: gcc.info, Node: Constraints, Next: Asm Labels, Prev: Extended Asm, Up: C Extensions
18114 5.36 Constraints for `asm' Operands
18115 ===================================
18117 Here are specific details on what constraint letters you can use with
18118 `asm' operands. Constraints can say whether an operand may be in a
18119 register, and which kinds of register; whether the operand can be a
18120 memory reference, and which kinds of address; whether the operand may
18121 be an immediate constant, and which possible values it may have.
18122 Constraints can also require two operands to match.
18126 * Simple Constraints:: Basic use of constraints.
18127 * Multi-Alternative:: When an insn has two alternative constraint-patterns.
18128 * Modifiers:: More precise control over effects of constraints.
18129 * Machine Constraints:: Special constraints for some particular machines.
18132 File: gcc.info, Node: Simple Constraints, Next: Multi-Alternative, Up: Constraints
18134 5.36.1 Simple Constraints
18135 -------------------------
18137 The simplest kind of constraint is a string full of letters, each of
18138 which describes one kind of operand that is permitted. Here are the
18139 letters that are allowed:
18142 Whitespace characters are ignored and can be inserted at any
18143 position except the first. This enables each alternative for
18144 different operands to be visually aligned in the machine
18145 description even if they have different number of constraints and
18149 A memory operand is allowed, with any kind of address that the
18150 machine supports in general.
18153 A memory operand is allowed, but only if the address is
18154 "offsettable". This means that adding a small integer (actually,
18155 the width in bytes of the operand, as determined by its machine
18156 mode) may be added to the address and the result is also a valid
18159 For example, an address which is constant is offsettable; so is an
18160 address that is the sum of a register and a constant (as long as a
18161 slightly larger constant is also within the range of
18162 address-offsets supported by the machine); but an autoincrement or
18163 autodecrement address is not offsettable. More complicated
18164 indirect/indexed addresses may or may not be offsettable depending
18165 on the other addressing modes that the machine supports.
18167 Note that in an output operand which can be matched by another
18168 operand, the constraint letter `o' is valid only when accompanied
18169 by both `<' (if the target machine has predecrement addressing)
18170 and `>' (if the target machine has preincrement addressing).
18173 A memory operand that is not offsettable. In other words,
18174 anything that would fit the `m' constraint but not the `o'
18178 A memory operand with autodecrement addressing (either
18179 predecrement or postdecrement) is allowed.
18182 A memory operand with autoincrement addressing (either
18183 preincrement or postincrement) is allowed.
18186 A register operand is allowed provided that it is in a general
18190 An immediate integer operand (one with constant value) is allowed.
18191 This includes symbolic constants whose values will be known only at
18192 assembly time or later.
18195 An immediate integer operand with a known numeric value is allowed.
18196 Many systems cannot support assembly-time constants for operands
18197 less than a word wide. Constraints for these operands should use
18198 `n' rather than `i'.
18200 `I', `J', `K', ... `P'
18201 Other letters in the range `I' through `P' may be defined in a
18202 machine-dependent fashion to permit immediate integer operands with
18203 explicit integer values in specified ranges. For example, on the
18204 68000, `I' is defined to stand for the range of values 1 to 8.
18205 This is the range permitted as a shift count in the shift
18209 An immediate floating operand (expression code `const_double') is
18210 allowed, but only if the target floating point format is the same
18211 as that of the host machine (on which the compiler is running).
18214 An immediate floating operand (expression code `const_double' or
18215 `const_vector') is allowed.
18218 `G' and `H' may be defined in a machine-dependent fashion to
18219 permit immediate floating operands in particular ranges of values.
18222 An immediate integer operand whose value is not an explicit
18223 integer is allowed.
18225 This might appear strange; if an insn allows a constant operand
18226 with a value not known at compile time, it certainly must allow
18227 any known value. So why use `s' instead of `i'? Sometimes it
18228 allows better code to be generated.
18230 For example, on the 68000 in a fullword instruction it is possible
18231 to use an immediate operand; but if the immediate value is between
18232 -128 and 127, better code results from loading the value into a
18233 register and using the register. This is because the load into
18234 the register can be done with a `moveq' instruction. We arrange
18235 for this to happen by defining the letter `K' to mean "any integer
18236 outside the range -128 to 127", and then specifying `Ks' in the
18237 operand constraints.
18240 Any register, memory or immediate integer operand is allowed,
18241 except for registers that are not general registers.
18244 Any operand whatsoever is allowed.
18246 `0', `1', `2', ... `9'
18247 An operand that matches the specified operand number is allowed.
18248 If a digit is used together with letters within the same
18249 alternative, the digit should come last.
18251 This number is allowed to be more than a single digit. If multiple
18252 digits are encountered consecutively, they are interpreted as a
18253 single decimal integer. There is scant chance for ambiguity,
18254 since to-date it has never been desirable that `10' be interpreted
18255 as matching either operand 1 _or_ operand 0. Should this be
18256 desired, one can use multiple alternatives instead.
18258 This is called a "matching constraint" and what it really means is
18259 that the assembler has only a single operand that fills two roles
18260 which `asm' distinguishes. For example, an add instruction uses
18261 two input operands and an output operand, but on most CISC
18262 machines an add instruction really has only two operands, one of
18263 them an input-output operand:
18267 Matching constraints are used in these circumstances. More
18268 precisely, the two operands that match must include one input-only
18269 operand and one output-only operand. Moreover, the digit must be a
18270 smaller number than the number of the operand that uses it in the
18274 An operand that is a valid memory address is allowed. This is for
18275 "load address" and "push address" instructions.
18277 `p' in the constraint must be accompanied by `address_operand' as
18278 the predicate in the `match_operand'. This predicate interprets
18279 the mode specified in the `match_operand' as the mode of the memory
18280 reference for which the address would be valid.
18283 Other letters can be defined in machine-dependent fashion to stand
18284 for particular classes of registers or other arbitrary operand
18285 types. `d', `a' and `f' are defined on the 68000/68020 to stand
18286 for data, address and floating point registers.
18289 File: gcc.info, Node: Multi-Alternative, Next: Modifiers, Prev: Simple Constraints, Up: Constraints
18291 5.36.2 Multiple Alternative Constraints
18292 ---------------------------------------
18294 Sometimes a single instruction has multiple alternative sets of possible
18295 operands. For example, on the 68000, a logical-or instruction can
18296 combine register or an immediate value into memory, or it can combine
18297 any kind of operand into a register; but it cannot combine one memory
18298 location into another.
18300 These constraints are represented as multiple alternatives. An
18301 alternative can be described by a series of letters for each operand.
18302 The overall constraint for an operand is made from the letters for this
18303 operand from the first alternative, a comma, the letters for this
18304 operand from the second alternative, a comma, and so on until the last
18307 If all the operands fit any one alternative, the instruction is valid.
18308 Otherwise, for each alternative, the compiler counts how many
18309 instructions must be added to copy the operands so that that
18310 alternative applies. The alternative requiring the least copying is
18311 chosen. If two alternatives need the same amount of copying, the one
18312 that comes first is chosen. These choices can be altered with the `?'
18313 and `!' characters:
18316 Disparage slightly the alternative that the `?' appears in, as a
18317 choice when no alternative applies exactly. The compiler regards
18318 this alternative as one unit more costly for each `?' that appears
18322 Disparage severely the alternative that the `!' appears in. This
18323 alternative can still be used if it fits without reloading, but if
18324 reloading is needed, some other alternative will be used.
18327 File: gcc.info, Node: Modifiers, Next: Machine Constraints, Prev: Multi-Alternative, Up: Constraints
18329 5.36.3 Constraint Modifier Characters
18330 -------------------------------------
18332 Here are constraint modifier characters.
18335 Means that this operand is write-only for this instruction: the
18336 previous value is discarded and replaced by output data.
18339 Means that this operand is both read and written by the
18342 When the compiler fixes up the operands to satisfy the constraints,
18343 it needs to know which operands are inputs to the instruction and
18344 which are outputs from it. `=' identifies an output; `+'
18345 identifies an operand that is both input and output; all other
18346 operands are assumed to be input only.
18348 If you specify `=' or `+' in a constraint, you put it in the first
18349 character of the constraint string.
18352 Means (in a particular alternative) that this operand is an
18353 "earlyclobber" operand, which is modified before the instruction is
18354 finished using the input operands. Therefore, this operand may
18355 not lie in a register that is used as an input operand or as part
18356 of any memory address.
18358 `&' applies only to the alternative in which it is written. In
18359 constraints with multiple alternatives, sometimes one alternative
18360 requires `&' while others do not. See, for example, the `movdf'
18363 An input operand can be tied to an earlyclobber operand if its only
18364 use as an input occurs before the early result is written. Adding
18365 alternatives of this form often allows GCC to produce better code
18366 when only some of the inputs can be affected by the earlyclobber.
18367 See, for example, the `mulsi3' insn of the ARM.
18369 `&' does not obviate the need to write `='.
18372 Declares the instruction to be commutative for this operand and the
18373 following operand. This means that the compiler may interchange
18374 the two operands if that is the cheapest way to make all operands
18375 fit the constraints. GCC can only handle one commutative pair in
18376 an asm; if you use more, the compiler may fail. Note that you
18377 need not use the modifier if the two alternatives are strictly
18378 identical; this would only waste time in the reload pass. The
18379 modifier is not operational after register allocation, so the
18380 result of `define_peephole2' and `define_split's performed after
18381 reload cannot rely on `%' to make the intended insn match.
18384 Says that all following characters, up to the next comma, are to be
18385 ignored as a constraint. They are significant only for choosing
18386 register preferences.
18389 Says that the following character should be ignored when choosing
18390 register preferences. `*' has no effect on the meaning of the
18391 constraint as a constraint, and no effect on reloading.
18395 File: gcc.info, Node: Machine Constraints, Prev: Modifiers, Up: Constraints
18397 5.36.4 Constraints for Particular Machines
18398 ------------------------------------------
18400 Whenever possible, you should use the general-purpose constraint letters
18401 in `asm' arguments, since they will convey meaning more readily to
18402 people reading your code. Failing that, use the constraint letters
18403 that usually have very similar meanings across architectures. The most
18404 commonly used constraints are `m' and `r' (for memory and
18405 general-purpose registers respectively; *note Simple Constraints::), and
18406 `I', usually the letter indicating the most common immediate-constant
18409 Each architecture defines additional constraints. These constraints
18410 are used by the compiler itself for instruction generation, as well as
18411 for `asm' statements; therefore, some of the constraints are not
18412 particularly useful for `asm'. Here is a summary of some of the
18413 machine-dependent constraints available on some particular machines; it
18414 includes both constraints that are useful for `asm' and constraints
18415 that aren't. The compiler source file mentioned in the table heading
18416 for each architecture is the definitive reference for the meanings of
18417 that architecture's constraints.
18419 _ARM family--`config/arm/arm.h'_
18422 Floating-point register
18425 VFP floating-point register
18428 One of the floating-point constants 0.0, 0.5, 1.0, 2.0, 3.0,
18432 Floating-point constant that would satisfy the constraint `F'
18436 Integer that is valid as an immediate operand in a data
18437 processing instruction. That is, an integer in the range 0
18438 to 255 rotated by a multiple of 2
18441 Integer in the range -4095 to 4095
18444 Integer that satisfies constraint `I' when inverted (ones
18448 Integer that satisfies constraint `I' when negated (twos
18452 Integer in the range 0 to 32
18455 A memory reference where the exact address is in a single
18456 register (``m'' is preferable for `asm' statements)
18459 An item in the constant pool
18462 A symbol in the text segment of the current file
18465 A memory reference suitable for VFP load/store insns
18466 (reg+constant offset)
18469 A memory reference suitable for iWMMXt load/store
18473 A memory reference suitable for the ARMv4 ldrsb instruction.
18475 _AVR family--`config/avr/constraints.md'_
18478 Registers from r0 to r15
18481 Registers from r16 to r23
18484 Registers from r16 to r31
18487 Registers from r24 to r31. These registers can be used in
18491 Pointer register (r26-r31)
18494 Base pointer register (r28-r31)
18497 Stack pointer register (SPH:SPL)
18500 Temporary register r0
18503 Register pair X (r27:r26)
18506 Register pair Y (r29:r28)
18509 Register pair Z (r31:r30)
18512 Constant greater than -1, less than 64
18515 Constant greater than -64, less than 1
18524 Constant that fits in 8 bits
18527 Constant integer -1
18530 Constant integer 8, 16, or 24
18536 A floating point constant 0.0
18538 _CRX Architecture--`config/crx/crx.h'_
18541 Registers from r0 to r14 (registers without stack pointer)
18544 Register r16 (64-bit accumulator lo register)
18547 Register r17 (64-bit accumulator hi register)
18550 Register pair r16-r17. (64-bit accumulator lo-hi pair)
18553 Constant that fits in 3 bits
18556 Constant that fits in 4 bits
18559 Constant that fits in 5 bits
18562 Constant that is one of -1, 4, -4, 7, 8, 12, 16, 20, 32, 48
18565 Floating point constant that is legal for store immediate
18567 _PowerPC and IBM RS6000--`config/rs6000/rs6000.h'_
18570 Address base register
18573 Floating point register
18579 `MQ', `CTR', or `LINK' register
18591 `CR' register (condition register) number 0
18594 `CR' register (condition register)
18597 `FPMEM' stack memory for FPR-GPR transfers
18600 Signed 16-bit constant
18603 Unsigned 16-bit constant shifted left 16 bits (use `L'
18604 instead for `SImode' constants)
18607 Unsigned 16-bit constant
18610 Signed 16-bit constant shifted left 16 bits
18613 Constant larger than 31
18622 Constant whose negation is a signed 16-bit constant
18625 Floating point constant that can be loaded into a register
18626 with one instruction per word
18629 Memory operand that is an offset from a register (`m' is
18630 preferable for `asm' statements)
18636 Constant suitable as a 64-bit mask operand
18639 Constant suitable as a 32-bit mask operand
18642 System V Release 4 small data area reference
18644 _MorphoTech family--`config/mt/mt.h'_
18647 Constant for an arithmetic insn (16-bit signed integer).
18653 Constant for a logical insn (16-bit zero-extended integer).
18656 A constant that can be loaded with `lui' (i.e. the bottom 16
18660 A constant that takes two words to load (i.e. not matched by
18664 Negative 16-bit constants other than -65536.
18667 A 15-bit signed integer constant.
18670 A positive 16-bit constant.
18672 _Intel 386--`config/i386/constraints.md'_
18675 Legacy register--the eight integer registers available on all
18676 i386 processors (`a', `b', `c', `d', `si', `di', `bp', `sp').
18679 Any register accessible as `Rl'. In 32-bit mode, `a', `b',
18680 `c', and `d'; in 64-bit mode, any integer register.
18683 Any register accessible as `Rh': `a', `b', `c', and `d'.
18704 The `a' and `d' registers, as a pair (for instructions that
18705 return half the result in one and half in the other).
18708 Any 80387 floating-point (stack) register.
18711 Top of 80387 floating-point stack (`%st(0)').
18714 Second from top of 80387 floating-point stack (`%st(1)').
18723 Integer constant in the range 0 ... 31, for 32-bit shifts.
18726 Integer constant in the range 0 ... 63, for 64-bit shifts.
18729 Signed 8-bit integer constant.
18732 `0xFF' or `0xFFFF', for andsi as a zero-extending move.
18735 0, 1, 2, or 3 (shifts for the `lea' instruction).
18738 Unsigned 8-bit integer constant (for `in' and `out'
18742 Standard 80387 floating point constant.
18745 Standard SSE floating point constant.
18748 32-bit signed integer constant, or a symbolic reference known
18749 to fit that range (for immediate operands in sign-extending
18750 x86-64 instructions).
18753 32-bit unsigned integer constant, or a symbolic reference
18754 known to fit that range (for immediate operands in
18755 zero-extending x86-64 instructions).
18758 _Intel IA-64--`config/ia64/ia64.h'_
18761 General register `r0' to `r3' for `addl' instruction
18767 Predicate register (`c' as in "conditional")
18770 Application register residing in M-unit
18773 Application register residing in I-unit
18776 Floating-point register
18779 Memory operand. Remember that `m' allows postincrement and
18780 postdecrement which require printing with `%Pn' on IA-64.
18781 Use `S' to disallow postincrement and postdecrement.
18784 Floating-point constant 0.0 or 1.0
18787 14-bit signed integer constant
18790 22-bit signed integer constant
18793 8-bit signed integer constant for logical instructions
18796 8-bit adjusted signed integer constant for compare pseudo-ops
18799 6-bit unsigned integer constant for shift counts
18802 9-bit signed integer constant for load and store
18809 0 or -1 for `dep' instruction
18812 Non-volatile memory for floating-point loads and stores
18815 Integer constant in the range 1 to 4 for `shladd' instruction
18818 Memory operand except postincrement and postdecrement
18820 _FRV--`config/frv/frv.h'_
18823 Register in the class `ACC_REGS' (`acc0' to `acc7').
18826 Register in the class `EVEN_ACC_REGS' (`acc0' to `acc7').
18829 Register in the class `CC_REGS' (`fcc0' to `fcc3' and `icc0'
18833 Register in the class `GPR_REGS' (`gr0' to `gr63').
18836 Register in the class `EVEN_REGS' (`gr0' to `gr63'). Odd
18837 registers are excluded not in the class but through the use
18838 of a machine mode larger than 4 bytes.
18841 Register in the class `FPR_REGS' (`fr0' to `fr63').
18844 Register in the class `FEVEN_REGS' (`fr0' to `fr63'). Odd
18845 registers are excluded not in the class but through the use
18846 of a machine mode larger than 4 bytes.
18849 Register in the class `LR_REG' (the `lr' register).
18852 Register in the class `QUAD_REGS' (`gr2' to `gr63').
18853 Register numbers not divisible by 4 are excluded not in the
18854 class but through the use of a machine mode larger than 8
18858 Register in the class `ICC_REGS' (`icc0' to `icc3').
18861 Register in the class `FCC_REGS' (`fcc0' to `fcc3').
18864 Register in the class `ICR_REGS' (`cc4' to `cc7').
18867 Register in the class `FCR_REGS' (`cc0' to `cc3').
18870 Register in the class `QUAD_FPR_REGS' (`fr0' to `fr63').
18871 Register numbers not divisible by 4 are excluded not in the
18872 class but through the use of a machine mode larger than 8
18876 Register in the class `SPR_REGS' (`lcr' and `lr').
18879 Register in the class `QUAD_ACC_REGS' (`acc0' to `acc7').
18882 Register in the class `ACCG_REGS' (`accg0' to `accg7').
18885 Register in the class `CR_REGS' (`cc0' to `cc7').
18888 Floating point constant zero
18891 6-bit signed integer constant
18894 10-bit signed integer constant
18897 16-bit signed integer constant
18900 16-bit unsigned integer constant
18903 12-bit signed integer constant that is negative--i.e. in the
18904 range of -2048 to -1
18910 12-bit signed integer constant that is greater than
18911 zero--i.e. in the range of 1 to 2047.
18914 _Blackfin family--`config/bfin/bfin.h'_
18923 A call clobbered P register.
18926 Even-numbered D register
18929 Odd-numbered D register
18932 Accumulator register.
18935 Even-numbered accumulator register.
18938 Odd-numbered accumulator register.
18950 Registers used for circular buffering, i.e. I, B, or L
18966 Any D, P, B, M, I or L register.
18969 Additional registers typically used only in prologues and
18970 epilogues: RETS, RETN, RETI, RETX, RETE, ASTAT, SEQSTAT and
18974 Any register except accumulators or CC.
18977 Signed 16 bit integer (in the range -32768 to 32767)
18980 Unsigned 16 bit integer (in the range 0 to 65535)
18983 Signed 7 bit integer (in the range -64 to 63)
18986 Unsigned 7 bit integer (in the range 0 to 127)
18989 Unsigned 5 bit integer (in the range 0 to 31)
18992 Signed 4 bit integer (in the range -8 to 7)
18995 Signed 3 bit integer (in the range -3 to 4)
18998 Unsigned 3 bit integer (in the range 0 to 7)
19001 Constant N, where N is a single-digit constant in the range 0
19011 An integer constant with exactly a single bit set.
19014 An integer constant with all bits set except exactly one.
19021 _M32C--`config/m32c/m32c.c'_
19026 `$sp', `$fb', `$sb'.
19029 Any control register, when they're 16 bits wide (nothing if
19030 control registers are 24 bits wide)
19033 Any control register, when they're 24 bits wide.
19039 $r0, $r1, $r2, $r3.
19042 $r0 or $r2, or $r2r0 for 32 bit values.
19045 $r1 or $r3, or $r3r1 for 32 bit values.
19048 A register that can hold a 64 bit value.
19051 $r0 or $r1 (registers with addressable high/low bytes)
19060 Address registers when they're 16 bits wide.
19063 Address registers when they're 24 bits wide.
19066 Registers that can hold QI values.
19069 Registers that can be used with displacements ($a0, $a1, $sb).
19072 Registers that can hold 32 bit values.
19075 Registers that can hold 16 bit values.
19078 Registers chat can hold 16 bit values, including all control
19082 $r0 through R1, plus $a0 and $a1.
19085 The flags register.
19088 The memory-based pseudo-registers $mem0 through $mem15.
19091 Registers that can hold pointers (16 bit registers for r8c,
19092 m16c; 24 bit registers for m32cm, m32c).
19095 Matches multiple registers in a PARALLEL to form a larger
19096 register. Used to match function return values.
19111 -8 ... -1 or 1 ... 8
19114 -16 ... -1 or 1 ... 16
19117 -32 ... -1 or 1 ... 32
19123 An 8 bit value with exactly one bit set.
19126 A 16 bit value with exactly one bit set.
19129 The common src/dest memory addressing modes.
19132 Memory addressed using $a0 or $a1.
19135 Memory addressed with immediate addresses.
19138 Memory addressed using the stack pointer ($sp).
19141 Memory addressed using the frame base register ($fb).
19144 Memory addressed using the small base register ($sb).
19149 _MIPS--`config/mips/constraints.md'_
19152 An address register. This is equivalent to `r' unless
19153 generating MIPS16 code.
19156 A floating-point register (if available).
19165 The `hi' and `lo' registers.
19168 A register suitable for use in an indirect jump. This will
19169 always be `$25' for `-mabicalls'.
19172 Equivalent to `r'; retained for backwards compatibility.
19175 A floating-point condition code register.
19178 A signed 16-bit constant (for arithmetic instructions).
19184 An unsigned 16-bit constant (for logic instructions).
19187 A signed 32-bit constant in which the lower 16 bits are zero.
19188 Such constants can be loaded using `lui'.
19191 A constant that cannot be loaded using `lui', `addiu' or
19195 A constant in the range -65535 to -1 (inclusive).
19198 A signed 15-bit constant.
19201 A constant in the range 1 to 65535 (inclusive).
19204 Floating-point zero.
19207 An address that can be used in a non-macro load or store.
19209 _Motorola 680x0--`config/m68k/m68k.h'_
19218 68881 floating-point register, if available
19221 Integer in the range 1 to 8
19224 16-bit signed number
19227 Signed number whose magnitude is greater than 0x80
19230 Integer in the range -8 to -1
19233 Signed number whose magnitude is greater than 0x100
19236 Floating point constant that is not a 68881 constant
19238 _Motorola 68HC11 & 68HC12 families--`config/m68hc11/m68hc11.h'_
19253 Temporary soft register _.tmp
19256 A soft register _.d1 to _.d31
19259 Stack pointer register
19268 Pseudo register `z' (replaced by `x' or `y' at the end)
19271 An address register: x, y or z
19274 An address register: x or y
19277 Register pair (x:d) to form a 32-bit value
19280 Constants in the range -65536 to 65535
19283 Constants whose 16-bit low part is zero
19286 Constant integer 1 or -1
19289 Constant integer 16
19292 Constants in the range -8 to 2
19295 _SPARC--`config/sparc/sparc.h'_
19298 Floating-point register on the SPARC-V8 architecture and
19299 lower floating-point register on the SPARC-V9 architecture.
19302 Floating-point register. It is equivalent to `f' on the
19303 SPARC-V8 architecture and contains both lower and upper
19304 floating-point registers on the SPARC-V9 architecture.
19307 Floating-point condition code register.
19310 Lower floating-point register. It is only valid on the
19311 SPARC-V9 architecture when the Visual Instruction Set is
19315 Floating-point register. It is only valid on the SPARC-V9
19316 architecture when the Visual Instruction Set is available.
19319 64-bit global or out register for the SPARC-V8+ architecture.
19322 Signed 13-bit constant
19328 32-bit constant with the low 12 bits clear (a constant that
19329 can be loaded with the `sethi' instruction)
19332 A constant in the range supported by `movcc' instructions
19335 A constant in the range supported by `movrcc' instructions
19338 Same as `K', except that it verifies that bits that are not
19339 in the lower 32-bit range are all zero. Must be used instead
19340 of `K' for modes wider than `SImode'
19346 Floating-point zero
19349 Signed 13-bit constant, sign-extended to 32 or 64 bits
19352 Floating-point constant whose integral representation can be
19353 moved into an integer register using a single sethi
19357 Floating-point constant whose integral representation can be
19358 moved into an integer register using a single mov instruction
19361 Floating-point constant whose integral representation can be
19362 moved into an integer register using a high/lo_sum
19363 instruction sequence
19366 Memory address aligned to an 8-byte boundary
19372 Memory address for `e' constraint registers
19378 _TMS320C3x/C4x--`config/c4x/c4x.h'_
19381 Auxiliary (address) register (ar0-ar7)
19384 Stack pointer register (sp)
19387 Standard (32-bit) precision integer register
19390 Extended (40-bit) precision register (r0-r11)
19393 Block count register (bk)
19396 Extended (40-bit) precision low register (r0-r7)
19399 Extended (40-bit) precision register (r0-r1)
19402 Extended (40-bit) precision register (r2-r3)
19405 Repeat count register (rc)
19408 Index register (ir0-ir1)
19411 Status (condition code) register (st)
19414 Data page register (dp)
19417 Floating-point zero
19420 Immediate 16-bit floating-point constant
19423 Signed 16-bit constant
19426 Signed 8-bit constant
19429 Signed 5-bit constant
19432 Unsigned 16-bit constant
19435 Unsigned 8-bit constant
19438 Ones complement of unsigned 16-bit constant
19441 High 16-bit constant (32-bit constant with 16 LSBs zero)
19444 Indirect memory reference with signed 8-bit or index register
19448 Indirect memory reference with unsigned 5-bit displacement
19451 Indirect memory reference with 1 bit or index register
19455 Direct memory reference
19461 _S/390 and zSeries--`config/s390/s390.h'_
19464 Address register (general purpose register except r0)
19467 Condition code register
19470 Data register (arbitrary general purpose register)
19473 Floating-point register
19476 Unsigned 8-bit constant (0-255)
19479 Unsigned 12-bit constant (0-4095)
19482 Signed 16-bit constant (-32768-32767)
19485 Value appropriate as displacement.
19487 for short displacement
19489 `(-524288..524287)'
19490 for long displacement
19493 Constant integer with a value of 0x7fffffff.
19496 Multiple letter constraint followed by 4 parameter letters.
19498 number of the part counting from most to least
19505 mode of the containing operand
19508 value of the other parts (F--all bits set)
19509 The constraint matches if the specified part of a constant
19510 has a value different from it's other parts.
19513 Memory reference without index register and with short
19517 Memory reference with index register and short displacement.
19520 Memory reference without index register but with long
19524 Memory reference with index register and long displacement.
19527 Pointer with short displacement.
19530 Pointer with long displacement.
19533 Shift count operand.
19536 _Score family--`config/score/score.h'_
19539 Registers from r0 to r32.
19542 Registers from r0 to r16.
19545 r8--r11 or r22--r27 registers.
19566 cnt + lcb + scb register.
19569 cr0--cr15 register.
19581 cp1 + cp2 + cp3 registers.
19584 High 16-bit constant (32-bit constant with 16 LSBs zero).
19587 Unsigned 5 bit integer (in the range 0 to 31).
19590 Unsigned 16 bit integer (in the range 0 to 65535).
19593 Signed 16 bit integer (in the range -32768 to 32767).
19596 Unsigned 14 bit integer (in the range 0 to 16383).
19599 Signed 14 bit integer (in the range -8192 to 8191).
19604 _Xstormy16--`config/stormy16/stormy16.h'_
19619 Registers r0 through r7.
19622 Registers r0 and r1.
19625 The carry register.
19628 Registers r8 and r9.
19631 A constant between 0 and 3 inclusive.
19634 A constant that has exactly one bit set.
19637 A constant that has exactly one bit clear.
19640 A constant between 0 and 255 inclusive.
19643 A constant between -255 and 0 inclusive.
19646 A constant between -3 and 0 inclusive.
19649 A constant between 1 and 4 inclusive.
19652 A constant between -4 and -1 inclusive.
19655 A memory reference that is a stack push.
19658 A memory reference that is a stack pop.
19661 A memory reference that refers to a constant address of known
19665 The register indicated by Rx (not implemented yet).
19668 A constant that is not between 2 and 15 inclusive.
19674 _Xtensa--`config/xtensa/xtensa.h'_
19677 General-purpose 32-bit register
19680 One-bit boolean register
19683 MAC16 40-bit accumulator register
19686 Signed 12-bit integer constant, for use in MOVI instructions
19689 Signed 8-bit integer constant, for use in ADDI instructions
19692 Integer constant valid for BccI instructions
19695 Unsigned constant valid for BccUI instructions
19700 File: gcc.info, Node: Asm Labels, Next: Explicit Reg Vars, Prev: Constraints, Up: C Extensions
19702 5.37 Controlling Names Used in Assembler Code
19703 =============================================
19705 You can specify the name to be used in the assembler code for a C
19706 function or variable by writing the `asm' (or `__asm__') keyword after
19707 the declarator as follows:
19709 int foo asm ("myfoo") = 2;
19711 This specifies that the name to be used for the variable `foo' in the
19712 assembler code should be `myfoo' rather than the usual `_foo'.
19714 On systems where an underscore is normally prepended to the name of a C
19715 function or variable, this feature allows you to define names for the
19716 linker that do not start with an underscore.
19718 It does not make sense to use this feature with a non-static local
19719 variable since such variables do not have assembler names. If you are
19720 trying to put the variable in a particular register, see *Note Explicit
19721 Reg Vars::. GCC presently accepts such code with a warning, but will
19722 probably be changed to issue an error, rather than a warning, in the
19725 You cannot use `asm' in this way in a function _definition_; but you
19726 can get the same effect by writing a declaration for the function
19727 before its definition and putting `asm' there, like this:
19729 extern func () asm ("FUNC");
19735 It is up to you to make sure that the assembler names you choose do not
19736 conflict with any other assembler symbols. Also, you must not use a
19737 register name; that would produce completely invalid assembler code.
19738 GCC does not as yet have the ability to store static variables in
19739 registers. Perhaps that will be added.
19742 File: gcc.info, Node: Explicit Reg Vars, Next: Alternate Keywords, Prev: Asm Labels, Up: C Extensions
19744 5.38 Variables in Specified Registers
19745 =====================================
19747 GNU C allows you to put a few global variables into specified hardware
19748 registers. You can also specify the register in which an ordinary
19749 register variable should be allocated.
19751 * Global register variables reserve registers throughout the program.
19752 This may be useful in programs such as programming language
19753 interpreters which have a couple of global variables that are
19754 accessed very often.
19756 * Local register variables in specific registers do not reserve the
19757 registers, except at the point where they are used as input or
19758 output operands in an `asm' statement and the `asm' statement
19759 itself is not deleted. The compiler's data flow analysis is
19760 capable of determining where the specified registers contain live
19761 values, and where they are available for other uses. Stores into
19762 local register variables may be deleted when they appear to be
19763 dead according to dataflow analysis. References to local register
19764 variables may be deleted or moved or simplified.
19766 These local variables are sometimes convenient for use with the
19767 extended `asm' feature (*note Extended Asm::), if you want to
19768 write one output of the assembler instruction directly into a
19769 particular register. (This will work provided the register you
19770 specify fits the constraints specified for that operand in the
19775 * Global Reg Vars::
19779 File: gcc.info, Node: Global Reg Vars, Next: Local Reg Vars, Up: Explicit Reg Vars
19781 5.38.1 Defining Global Register Variables
19782 -----------------------------------------
19784 You can define a global register variable in GNU C like this:
19786 register int *foo asm ("a5");
19788 Here `a5' is the name of the register which should be used. Choose a
19789 register which is normally saved and restored by function calls on your
19790 machine, so that library routines will not clobber it.
19792 Naturally the register name is cpu-dependent, so you would need to
19793 conditionalize your program according to cpu type. The register `a5'
19794 would be a good choice on a 68000 for a variable of pointer type. On
19795 machines with register windows, be sure to choose a "global" register
19796 that is not affected magically by the function call mechanism.
19798 In addition, operating systems on one type of cpu may differ in how
19799 they name the registers; then you would need additional conditionals.
19800 For example, some 68000 operating systems call this register `%a5'.
19802 Eventually there may be a way of asking the compiler to choose a
19803 register automatically, but first we need to figure out how it should
19804 choose and how to enable you to guide the choice. No solution is
19807 Defining a global register variable in a certain register reserves that
19808 register entirely for this use, at least within the current compilation.
19809 The register will not be allocated for any other purpose in the
19810 functions in the current compilation. The register will not be saved
19811 and restored by these functions. Stores into this register are never
19812 deleted even if they would appear to be dead, but references may be
19813 deleted or moved or simplified.
19815 It is not safe to access the global register variables from signal
19816 handlers, or from more than one thread of control, because the system
19817 library routines may temporarily use the register for other things
19818 (unless you recompile them specially for the task at hand).
19820 It is not safe for one function that uses a global register variable to
19821 call another such function `foo' by way of a third function `lose' that
19822 was compiled without knowledge of this variable (i.e. in a different
19823 source file in which the variable wasn't declared). This is because
19824 `lose' might save the register and put some other value there. For
19825 example, you can't expect a global register variable to be available in
19826 the comparison-function that you pass to `qsort', since `qsort' might
19827 have put something else in that register. (If you are prepared to
19828 recompile `qsort' with the same global register variable, you can solve
19831 If you want to recompile `qsort' or other source files which do not
19832 actually use your global register variable, so that they will not use
19833 that register for any other purpose, then it suffices to specify the
19834 compiler option `-ffixed-REG'. You need not actually add a global
19835 register declaration to their source code.
19837 A function which can alter the value of a global register variable
19838 cannot safely be called from a function compiled without this variable,
19839 because it could clobber the value the caller expects to find there on
19840 return. Therefore, the function which is the entry point into the part
19841 of the program that uses the global register variable must explicitly
19842 save and restore the value which belongs to its caller.
19844 On most machines, `longjmp' will restore to each global register
19845 variable the value it had at the time of the `setjmp'. On some
19846 machines, however, `longjmp' will not change the value of global
19847 register variables. To be portable, the function that called `setjmp'
19848 should make other arrangements to save the values of the global register
19849 variables, and to restore them in a `longjmp'. This way, the same
19850 thing will happen regardless of what `longjmp' does.
19852 All global register variable declarations must precede all function
19853 definitions. If such a declaration could appear after function
19854 definitions, the declaration would be too late to prevent the register
19855 from being used for other purposes in the preceding functions.
19857 Global register variables may not have initial values, because an
19858 executable file has no means to supply initial contents for a register.
19860 On the SPARC, there are reports that g3 ... g7 are suitable registers,
19861 but certain library functions, such as `getwd', as well as the
19862 subroutines for division and remainder, modify g3 and g4. g1 and g2
19863 are local temporaries.
19865 On the 68000, a2 ... a5 should be suitable, as should d2 ... d7. Of
19866 course, it will not do to use more than a few of those.
19869 File: gcc.info, Node: Local Reg Vars, Prev: Global Reg Vars, Up: Explicit Reg Vars
19871 5.38.2 Specifying Registers for Local Variables
19872 -----------------------------------------------
19874 You can define a local register variable with a specified register like
19877 register int *foo asm ("a5");
19879 Here `a5' is the name of the register which should be used. Note that
19880 this is the same syntax used for defining global register variables,
19881 but for a local variable it would appear within a function.
19883 Naturally the register name is cpu-dependent, but this is not a
19884 problem, since specific registers are most often useful with explicit
19885 assembler instructions (*note Extended Asm::). Both of these things
19886 generally require that you conditionalize your program according to cpu
19889 In addition, operating systems on one type of cpu may differ in how
19890 they name the registers; then you would need additional conditionals.
19891 For example, some 68000 operating systems call this register `%a5'.
19893 Defining such a register variable does not reserve the register; it
19894 remains available for other uses in places where flow control determines
19895 the variable's value is not live.
19897 This option does not guarantee that GCC will generate code that has
19898 this variable in the register you specify at all times. You may not
19899 code an explicit reference to this register in the _assembler
19900 instruction template_ part of an `asm' statement and assume it will
19901 always refer to this variable. However, using the variable as an `asm'
19902 _operand_ guarantees that the specified register is used for the
19905 Stores into local register variables may be deleted when they appear
19906 to be dead according to dataflow analysis. References to local
19907 register variables may be deleted or moved or simplified.
19909 As for global register variables, it's recommended that you choose a
19910 register which is normally saved and restored by function calls on your
19911 machine, so that library routines will not clobber it. A common
19912 pitfall is to initialize multiple call-clobbered registers with
19913 arbitrary expressions, where a function call or library call for an
19914 arithmetic operator will overwrite a register value from a previous
19915 assignment, for example `r0' below:
19916 register int *p1 asm ("r0") = ...;
19917 register int *p2 asm ("r1") = ...;
19918 In those cases, a solution is to use a temporary variable for each
19919 arbitrary expression. *Note Example of asm with clobbered asm reg::.
19922 File: gcc.info, Node: Alternate Keywords, Next: Incomplete Enums, Prev: Explicit Reg Vars, Up: C Extensions
19924 5.39 Alternate Keywords
19925 =======================
19927 `-ansi' and the various `-std' options disable certain keywords. This
19928 causes trouble when you want to use GNU C extensions, or a
19929 general-purpose header file that should be usable by all programs,
19930 including ISO C programs. The keywords `asm', `typeof' and `inline'
19931 are not available in programs compiled with `-ansi' or `-std' (although
19932 `inline' can be used in a program compiled with `-std=c99'). The ISO
19933 C99 keyword `restrict' is only available when `-std=gnu99' (which will
19934 eventually be the default) or `-std=c99' (or the equivalent
19935 `-std=iso9899:1999') is used.
19937 The way to solve these problems is to put `__' at the beginning and
19938 end of each problematical keyword. For example, use `__asm__' instead
19939 of `asm', and `__inline__' instead of `inline'.
19941 Other C compilers won't accept these alternative keywords; if you want
19942 to compile with another compiler, you can define the alternate keywords
19943 as macros to replace them with the customary keywords. It looks like
19947 #define __asm__ asm
19950 `-pedantic' and other options cause warnings for many GNU C extensions.
19951 You can prevent such warnings within one expression by writing
19952 `__extension__' before the expression. `__extension__' has no effect
19956 File: gcc.info, Node: Incomplete Enums, Next: Function Names, Prev: Alternate Keywords, Up: C Extensions
19958 5.40 Incomplete `enum' Types
19959 ============================
19961 You can define an `enum' tag without specifying its possible values.
19962 This results in an incomplete type, much like what you get if you write
19963 `struct foo' without describing the elements. A later declaration
19964 which does specify the possible values completes the type.
19966 You can't allocate variables or storage using the type while it is
19967 incomplete. However, you can work with pointers to that type.
19969 This extension may not be very useful, but it makes the handling of
19970 `enum' more consistent with the way `struct' and `union' are handled.
19972 This extension is not supported by GNU C++.
19975 File: gcc.info, Node: Function Names, Next: Return Address, Prev: Incomplete Enums, Up: C Extensions
19977 5.41 Function Names as Strings
19978 ==============================
19980 GCC provides three magic variables which hold the name of the current
19981 function, as a string. The first of these is `__func__', which is part
19982 of the C99 standard:
19984 The identifier `__func__' is implicitly declared by the translator
19985 as if, immediately following the opening brace of each function
19986 definition, the declaration
19987 static const char __func__[] = "function-name";
19989 appeared, where function-name is the name of the lexically-enclosing
19990 function. This name is the unadorned name of the function.
19992 `__FUNCTION__' is another name for `__func__'. Older versions of GCC
19993 recognize only this name. However, it is not standardized. For
19994 maximum portability, we recommend you use `__func__', but provide a
19995 fallback definition with the preprocessor:
19997 #if __STDC_VERSION__ < 199901L
19999 # define __func__ __FUNCTION__
20001 # define __func__ "<unknown>"
20005 In C, `__PRETTY_FUNCTION__' is yet another name for `__func__'.
20006 However, in C++, `__PRETTY_FUNCTION__' contains the type signature of
20007 the function as well as its bare name. For example, this program:
20010 extern int printf (char *, ...);
20017 printf ("__FUNCTION__ = %s\n", __FUNCTION__);
20018 printf ("__PRETTY_FUNCTION__ = %s\n", __PRETTY_FUNCTION__);
20033 __PRETTY_FUNCTION__ = void a::sub(int)
20035 These identifiers are not preprocessor macros. In GCC 3.3 and
20036 earlier, in C only, `__FUNCTION__' and `__PRETTY_FUNCTION__' were
20037 treated as string literals; they could be used to initialize `char'
20038 arrays, and they could be concatenated with other string literals. GCC
20039 3.4 and later treat them as variables, like `__func__'. In C++,
20040 `__FUNCTION__' and `__PRETTY_FUNCTION__' have always been variables.
20043 File: gcc.info, Node: Return Address, Next: Vector Extensions, Prev: Function Names, Up: C Extensions
20045 5.42 Getting the Return or Frame Address of a Function
20046 ======================================================
20048 These functions may be used to get information about the callers of a
20051 -- Built-in Function: void * __builtin_return_address (unsigned int
20053 This function returns the return address of the current function,
20054 or of one of its callers. The LEVEL argument is number of frames
20055 to scan up the call stack. A value of `0' yields the return
20056 address of the current function, a value of `1' yields the return
20057 address of the caller of the current function, and so forth. When
20058 inlining the expected behavior is that the function will return
20059 the address of the function that will be returned to. To work
20060 around this behavior use the `noinline' function attribute.
20062 The LEVEL argument must be a constant integer.
20064 On some machines it may be impossible to determine the return
20065 address of any function other than the current one; in such cases,
20066 or when the top of the stack has been reached, this function will
20067 return `0' or a random value. In addition,
20068 `__builtin_frame_address' may be used to determine if the top of
20069 the stack has been reached.
20071 This function should only be used with a nonzero argument for
20072 debugging purposes.
20074 -- Built-in Function: void * __builtin_frame_address (unsigned int
20076 This function is similar to `__builtin_return_address', but it
20077 returns the address of the function frame rather than the return
20078 address of the function. Calling `__builtin_frame_address' with a
20079 value of `0' yields the frame address of the current function, a
20080 value of `1' yields the frame address of the caller of the current
20081 function, and so forth.
20083 The frame is the area on the stack which holds local variables and
20084 saved registers. The frame address is normally the address of the
20085 first word pushed on to the stack by the function. However, the
20086 exact definition depends upon the processor and the calling
20087 convention. If the processor has a dedicated frame pointer
20088 register, and the function has a frame, then
20089 `__builtin_frame_address' will return the value of the frame
20092 On some machines it may be impossible to determine the frame
20093 address of any function other than the current one; in such cases,
20094 or when the top of the stack has been reached, this function will
20095 return `0' if the first frame pointer is properly initialized by
20098 This function should only be used with a nonzero argument for
20099 debugging purposes.
20102 File: gcc.info, Node: Vector Extensions, Next: Offsetof, Prev: Return Address, Up: C Extensions
20104 5.43 Using vector instructions through built-in functions
20105 =========================================================
20107 On some targets, the instruction set contains SIMD vector instructions
20108 that operate on multiple values contained in one large register at the
20109 same time. For example, on the i386 the MMX, 3Dnow! and SSE extensions
20110 can be used this way.
20112 The first step in using these extensions is to provide the necessary
20113 data types. This should be done using an appropriate `typedef':
20115 typedef int v4si __attribute__ ((vector_size (16)));
20117 The `int' type specifies the base type, while the attribute specifies
20118 the vector size for the variable, measured in bytes. For example, the
20119 declaration above causes the compiler to set the mode for the `v4si'
20120 type to be 16 bytes wide and divided into `int' sized units. For a
20121 32-bit `int' this means a vector of 4 units of 4 bytes, and the
20122 corresponding mode of `foo' will be V4SI.
20124 The `vector_size' attribute is only applicable to integral and float
20125 scalars, although arrays, pointers, and function return values are
20126 allowed in conjunction with this construct.
20128 All the basic integer types can be used as base types, both as signed
20129 and as unsigned: `char', `short', `int', `long', `long long'. In
20130 addition, `float' and `double' can be used to build floating-point
20133 Specifying a combination that is not valid for the current architecture
20134 will cause GCC to synthesize the instructions using a narrower mode.
20135 For example, if you specify a variable of type `V4SI' and your
20136 architecture does not allow for this specific SIMD type, GCC will
20137 produce code that uses 4 `SIs'.
20139 The types defined in this manner can be used with a subset of normal C
20140 operations. Currently, GCC will allow using the following operators on
20141 these types: `+, -, *, /, unary minus, ^, |, &, ~'.
20143 The operations behave like C++ `valarrays'. Addition is defined as
20144 the addition of the corresponding elements of the operands. For
20145 example, in the code below, each of the 4 elements in A will be added
20146 to the corresponding 4 elements in B and the resulting vector will be
20149 typedef int v4si __attribute__ ((vector_size (16)));
20155 Subtraction, multiplication, division, and the logical operations
20156 operate in a similar manner. Likewise, the result of using the unary
20157 minus or complement operators on a vector type is a vector whose
20158 elements are the negative or complemented values of the corresponding
20159 elements in the operand.
20161 You can declare variables and use them in function calls and returns,
20162 as well as in assignments and some casts. You can specify a vector
20163 type as a return type for a function. Vector types can also be used as
20164 function arguments. It is possible to cast from one vector type to
20165 another, provided they are of the same size (in fact, you can also cast
20166 vectors to and from other datatypes of the same size).
20168 You cannot operate between vectors of different lengths or different
20169 signedness without a cast.
20171 A port that supports hardware vector operations, usually provides a set
20172 of built-in functions that can be used to operate on vectors. For
20173 example, a function to add two vectors and multiply the result by a
20174 third could look like this:
20176 v4si f (v4si a, v4si b, v4si c)
20178 v4si tmp = __builtin_addv4si (a, b);
20179 return __builtin_mulv4si (tmp, c);
20183 File: gcc.info, Node: Offsetof, Next: Atomic Builtins, Prev: Vector Extensions, Up: C Extensions
20188 GCC implements for both C and C++ a syntactic extension to implement
20189 the `offsetof' macro.
20192 "__builtin_offsetof" "(" `typename' "," offsetof_member_designator ")"
20194 offsetof_member_designator:
20196 | offsetof_member_designator "." `identifier'
20197 | offsetof_member_designator "[" `expr' "]"
20199 This extension is sufficient such that
20201 #define offsetof(TYPE, MEMBER) __builtin_offsetof (TYPE, MEMBER)
20203 is a suitable definition of the `offsetof' macro. In C++, TYPE may be
20204 dependent. In either case, MEMBER may consist of a single identifier,
20205 or a sequence of member accesses and array references.
20208 File: gcc.info, Node: Atomic Builtins, Next: Object Size Checking, Prev: Offsetof, Up: C Extensions
20210 5.45 Built-in functions for atomic memory access
20211 ================================================
20213 The following builtins are intended to be compatible with those
20214 described in the `Intel Itanium Processor-specific Application Binary
20215 Interface', section 7.4. As such, they depart from the normal GCC
20216 practice of using the "__builtin_" prefix, and further that they are
20217 overloaded such that they work on multiple types.
20219 The definition given in the Intel documentation allows only for the
20220 use of the types `int', `long', `long long' as well as their unsigned
20221 counterparts. GCC will allow any integral scalar or pointer type that
20222 is 1, 2, 4 or 8 bytes in length.
20224 Not all operations are supported by all target processors. If a
20225 particular operation cannot be implemented on the target processor, a
20226 warning will be generated and a call an external function will be
20227 generated. The external function will carry the same name as the
20228 builtin, with an additional suffix `_N' where N is the size of the data
20231 In most cases, these builtins are considered a "full barrier". That
20232 is, no memory operand will be moved across the operation, either
20233 forward or backward. Further, instructions will be issued as necessary
20234 to prevent the processor from speculating loads across the operation
20235 and from queuing stores after the operation.
20237 All of the routines are are described in the Intel documentation to
20238 take "an optional list of variables protected by the memory barrier".
20239 It's not clear what is meant by that; it could mean that _only_ the
20240 following variables are protected, or it could mean that these variables
20241 should in addition be protected. At present GCC ignores this list and
20242 protects all variables which are globally accessible. If in the future
20243 we make some use of this list, an empty list will continue to mean all
20244 globally accessible variables.
20246 `TYPE __sync_fetch_and_add (TYPE *ptr, TYPE value, ...)'
20247 `TYPE __sync_fetch_and_sub (TYPE *ptr, TYPE value, ...)'
20248 `TYPE __sync_fetch_and_or (TYPE *ptr, TYPE value, ...)'
20249 `TYPE __sync_fetch_and_and (TYPE *ptr, TYPE value, ...)'
20250 `TYPE __sync_fetch_and_xor (TYPE *ptr, TYPE value, ...)'
20251 `TYPE __sync_fetch_and_nand (TYPE *ptr, TYPE value, ...)'
20252 These builtins perform the operation suggested by the name, and
20253 returns the value that had previously been in memory. That is,
20255 { tmp = *ptr; *ptr OP= value; return tmp; }
20256 { tmp = *ptr; *ptr = ~tmp & value; return tmp; } // nand
20258 `TYPE __sync_add_and_fetch (TYPE *ptr, TYPE value, ...)'
20259 `TYPE __sync_sub_and_fetch (TYPE *ptr, TYPE value, ...)'
20260 `TYPE __sync_or_and_fetch (TYPE *ptr, TYPE value, ...)'
20261 `TYPE __sync_and_and_fetch (TYPE *ptr, TYPE value, ...)'
20262 `TYPE __sync_xor_and_fetch (TYPE *ptr, TYPE value, ...)'
20263 `TYPE __sync_nand_and_fetch (TYPE *ptr, TYPE value, ...)'
20264 These builtins perform the operation suggested by the name, and
20265 return the new value. That is,
20267 { *ptr OP= value; return *ptr; }
20268 { *ptr = ~*ptr & value; return *ptr; } // nand
20270 `bool __sync_bool_compare_and_swap (TYPE *ptr, TYPE oldval TYPE newval, ...)'
20271 `TYPE __sync_val_compare_and_swap (TYPE *ptr, TYPE oldval TYPE newval, ...)'
20272 These builtins perform an atomic compare and swap. That is, if
20273 the current value of `*PTR' is OLDVAL, then write NEWVAL into
20276 The "bool" version returns true if the comparison is successful and
20277 NEWVAL was written. The "val" version returns the contents of
20278 `*PTR' before the operation.
20280 `__sync_synchronize (...)'
20281 This builtin issues a full memory barrier.
20283 `TYPE __sync_lock_test_and_set (TYPE *ptr, TYPE value, ...)'
20284 This builtin, as described by Intel, is not a traditional
20285 test-and-set operation, but rather an atomic exchange operation.
20286 It writes VALUE into `*PTR', and returns the previous contents of
20289 Many targets have only minimal support for such locks, and do not
20290 support a full exchange operation. In this case, a target may
20291 support reduced functionality here by which the _only_ valid value
20292 to store is the immediate constant 1. The exact value actually
20293 stored in `*PTR' is implementation defined.
20295 This builtin is not a full barrier, but rather an "acquire
20296 barrier". This means that references after the builtin cannot
20297 move to (or be speculated to) before the builtin, but previous
20298 memory stores may not be globally visible yet, and previous memory
20299 loads may not yet be satisfied.
20301 `void __sync_lock_release (TYPE *ptr, ...)'
20302 This builtin releases the lock acquired by
20303 `__sync_lock_test_and_set'. Normally this means writing the
20304 constant 0 to `*PTR'.
20306 This builtin is not a full barrier, but rather a "release barrier".
20307 This means that all previous memory stores are globally visible,
20308 and all previous memory loads have been satisfied, but following
20309 memory reads are not prevented from being speculated to before the
20313 File: gcc.info, Node: Object Size Checking, Next: Other Builtins, Prev: Atomic Builtins, Up: C Extensions
20315 5.46 Object Size Checking Builtins
20316 ==================================
20318 GCC implements a limited buffer overflow protection mechanism that can
20319 prevent some buffer overflow attacks.
20321 -- Built-in Function: size_t __builtin_object_size (void * PTR, int
20323 is a built-in construct that returns a constant number of bytes
20324 from PTR to the end of the object PTR pointer points to (if known
20325 at compile time). `__builtin_object_size' never evaluates its
20326 arguments for side-effects. If there are any side-effects in
20327 them, it returns `(size_t) -1' for TYPE 0 or 1 and `(size_t) 0'
20328 for TYPE 2 or 3. If there are multiple objects PTR can point to
20329 and all of them are known at compile time, the returned number is
20330 the maximum of remaining byte counts in those objects if TYPE & 2
20331 is 0 and minimum if nonzero. If it is not possible to determine
20332 which objects PTR points to at compile time,
20333 `__builtin_object_size' should return `(size_t) -1' for TYPE 0 or
20334 1 and `(size_t) 0' for TYPE 2 or 3.
20336 TYPE is an integer constant from 0 to 3. If the least significant
20337 bit is clear, objects are whole variables, if it is set, a closest
20338 surrounding subobject is considered the object a pointer points to.
20339 The second bit determines if maximum or minimum of remaining bytes
20342 struct V { char buf1[10]; int b; char buf2[10]; } var;
20343 char *p = &var.buf1[1], *q = &var.b;
20345 /* Here the object p points to is var. */
20346 assert (__builtin_object_size (p, 0) == sizeof (var) - 1);
20347 /* The subobject p points to is var.buf1. */
20348 assert (__builtin_object_size (p, 1) == sizeof (var.buf1) - 1);
20349 /* The object q points to is var. */
20350 assert (__builtin_object_size (q, 0)
20351 == (char *) (&var + 1) - (char *) &var.b);
20352 /* The subobject q points to is var.b. */
20353 assert (__builtin_object_size (q, 1) == sizeof (var.b));
20355 There are built-in functions added for many common string operation
20356 functions, e.g. for `memcpy' `__builtin___memcpy_chk' built-in is
20357 provided. This built-in has an additional last argument, which is the
20358 number of bytes remaining in object the DEST argument points to or
20359 `(size_t) -1' if the size is not known.
20361 The built-in functions are optimized into the normal string functions
20362 like `memcpy' if the last argument is `(size_t) -1' or if it is known
20363 at compile time that the destination object will not be overflown. If
20364 the compiler can determine at compile time the object will be always
20365 overflown, it issues a warning.
20367 The intended use can be e.g.
20370 #define bos0(dest) __builtin_object_size (dest, 0)
20371 #define memcpy(dest, src, n) \
20372 __builtin___memcpy_chk (dest, src, n, bos0 (dest))
20376 /* It is unknown what object p points to, so this is optimized
20377 into plain memcpy - no checking is possible. */
20378 memcpy (p, "abcde", n);
20379 /* Destination is known and length too. It is known at compile
20380 time there will be no overflow. */
20381 memcpy (&buf[5], "abcde", 5);
20382 /* Destination is known, but the length is not known at compile time.
20383 This will result in __memcpy_chk call that can check for overflow
20385 memcpy (&buf[5], "abcde", n);
20386 /* Destination is known and it is known at compile time there will
20387 be overflow. There will be a warning and __memcpy_chk call that
20388 will abort the program at runtime. */
20389 memcpy (&buf[6], "abcde", 5);
20391 Such built-in functions are provided for `memcpy', `mempcpy',
20392 `memmove', `memset', `strcpy', `stpcpy', `strncpy', `strcat' and
20395 There are also checking built-in functions for formatted output
20397 int __builtin___sprintf_chk (char *s, int flag, size_t os, const char *fmt, ...);
20398 int __builtin___snprintf_chk (char *s, size_t maxlen, int flag, size_t os,
20399 const char *fmt, ...);
20400 int __builtin___vsprintf_chk (char *s, int flag, size_t os, const char *fmt,
20402 int __builtin___vsnprintf_chk (char *s, size_t maxlen, int flag, size_t os,
20403 const char *fmt, va_list ap);
20405 The added FLAG argument is passed unchanged to `__sprintf_chk' etc.
20406 functions and can contain implementation specific flags on what
20407 additional security measures the checking function might take, such as
20408 handling `%n' differently.
20410 The OS argument is the object size S points to, like in the other
20411 built-in functions. There is a small difference in the behavior
20412 though, if OS is `(size_t) -1', the built-in functions are optimized
20413 into the non-checking functions only if FLAG is 0, otherwise the
20414 checking function is called with OS argument set to `(size_t) -1'.
20416 In addition to this, there are checking built-in functions
20417 `__builtin___printf_chk', `__builtin___vprintf_chk',
20418 `__builtin___fprintf_chk' and `__builtin___vfprintf_chk'. These have
20419 just one additional argument, FLAG, right before format string FMT. If
20420 the compiler is able to optimize them to `fputc' etc. functions, it
20421 will, otherwise the checking function should be called and the FLAG
20422 argument passed to it.
20425 File: gcc.info, Node: Other Builtins, Next: Target Builtins, Prev: Object Size Checking, Up: C Extensions
20427 5.47 Other built-in functions provided by GCC
20428 =============================================
20430 GCC provides a large number of built-in functions other than the ones
20431 mentioned above. Some of these are for internal use in the processing
20432 of exceptions or variable-length argument lists and will not be
20433 documented here because they may change from time to time; we do not
20434 recommend general use of these functions.
20436 The remaining functions are provided for optimization purposes.
20438 GCC includes built-in versions of many of the functions in the standard
20439 C library. The versions prefixed with `__builtin_' will always be
20440 treated as having the same meaning as the C library function even if you
20441 specify the `-fno-builtin' option. (*note C Dialect Options::) Many of
20442 these functions are only optimized in certain cases; if they are not
20443 optimized in a particular case, a call to the library function will be
20446 Outside strict ISO C mode (`-ansi', `-std=c89' or `-std=c99'), the
20447 functions `_exit', `alloca', `bcmp', `bzero', `dcgettext', `dgettext',
20448 `dremf', `dreml', `drem', `exp10f', `exp10l', `exp10', `ffsll', `ffsl',
20449 `ffs', `fprintf_unlocked', `fputs_unlocked', `gammaf', `gammal',
20450 `gamma', `gettext', `index', `isascii', `j0f', `j0l', `j0', `j1f',
20451 `j1l', `j1', `jnf', `jnl', `jn', `mempcpy', `pow10f', `pow10l', `pow10',
20452 `printf_unlocked', `rindex', `scalbf', `scalbl', `scalb', `signbit',
20453 `signbitf', `signbitl', `significandf', `significandl', `significand',
20454 `sincosf', `sincosl', `sincos', `stpcpy', `stpncpy', `strcasecmp',
20455 `strdup', `strfmon', `strncasecmp', `strndup', `toascii', `y0f', `y0l',
20456 `y0', `y1f', `y1l', `y1', `ynf', `ynl' and `yn' may be handled as
20457 built-in functions. All these functions have corresponding versions
20458 prefixed with `__builtin_', which may be used even in strict C89 mode.
20460 The ISO C99 functions `_Exit', `acoshf', `acoshl', `acosh', `asinhf',
20461 `asinhl', `asinh', `atanhf', `atanhl', `atanh', `cabsf', `cabsl',
20462 `cabs', `cacosf', `cacoshf', `cacoshl', `cacosh', `cacosl', `cacos',
20463 `cargf', `cargl', `carg', `casinf', `casinhf', `casinhl', `casinh',
20464 `casinl', `casin', `catanf', `catanhf', `catanhl', `catanh', `catanl',
20465 `catan', `cbrtf', `cbrtl', `cbrt', `ccosf', `ccoshf', `ccoshl',
20466 `ccosh', `ccosl', `ccos', `cexpf', `cexpl', `cexp', `cimagf', `cimagl',
20467 `cimag', `clogf', `clogl', `clog', `conjf', `conjl', `conj',
20468 `copysignf', `copysignl', `copysign', `cpowf', `cpowl', `cpow',
20469 `cprojf', `cprojl', `cproj', `crealf', `creall', `creal', `csinf',
20470 `csinhf', `csinhl', `csinh', `csinl', `csin', `csqrtf', `csqrtl',
20471 `csqrt', `ctanf', `ctanhf', `ctanhl', `ctanh', `ctanl', `ctan',
20472 `erfcf', `erfcl', `erfc', `erff', `erfl', `erf', `exp2f', `exp2l',
20473 `exp2', `expm1f', `expm1l', `expm1', `fdimf', `fdiml', `fdim', `fmaf',
20474 `fmal', `fmaxf', `fmaxl', `fmax', `fma', `fminf', `fminl', `fmin',
20475 `hypotf', `hypotl', `hypot', `ilogbf', `ilogbl', `ilogb', `imaxabs',
20476 `isblank', `iswblank', `lgammaf', `lgammal', `lgamma', `llabs',
20477 `llrintf', `llrintl', `llrint', `llroundf', `llroundl', `llround',
20478 `log1pf', `log1pl', `log1p', `log2f', `log2l', `log2', `logbf',
20479 `logbl', `logb', `lrintf', `lrintl', `lrint', `lroundf', `lroundl',
20480 `lround', `nearbyintf', `nearbyintl', `nearbyint', `nextafterf',
20481 `nextafterl', `nextafter', `nexttowardf', `nexttowardl', `nexttoward',
20482 `remainderf', `remainderl', `remainder', `remquof', `remquol',
20483 `remquo', `rintf', `rintl', `rint', `roundf', `roundl', `round',
20484 `scalblnf', `scalblnl', `scalbln', `scalbnf', `scalbnl', `scalbn',
20485 `snprintf', `tgammaf', `tgammal', `tgamma', `truncf', `truncl', `trunc',
20486 `vfscanf', `vscanf', `vsnprintf' and `vsscanf' are handled as built-in
20487 functions except in strict ISO C90 mode (`-ansi' or `-std=c89').
20489 There are also built-in versions of the ISO C99 functions `acosf',
20490 `acosl', `asinf', `asinl', `atan2f', `atan2l', `atanf', `atanl',
20491 `ceilf', `ceill', `cosf', `coshf', `coshl', `cosl', `expf', `expl',
20492 `fabsf', `fabsl', `floorf', `floorl', `fmodf', `fmodl', `frexpf',
20493 `frexpl', `ldexpf', `ldexpl', `log10f', `log10l', `logf', `logl',
20494 `modfl', `modf', `powf', `powl', `sinf', `sinhf', `sinhl', `sinl',
20495 `sqrtf', `sqrtl', `tanf', `tanhf', `tanhl' and `tanl' that are
20496 recognized in any mode since ISO C90 reserves these names for the
20497 purpose to which ISO C99 puts them. All these functions have
20498 corresponding versions prefixed with `__builtin_'.
20500 The ISO C94 functions `iswalnum', `iswalpha', `iswcntrl', `iswdigit',
20501 `iswgraph', `iswlower', `iswprint', `iswpunct', `iswspace', `iswupper',
20502 `iswxdigit', `towlower' and `towupper' are handled as built-in functions
20503 except in strict ISO C90 mode (`-ansi' or `-std=c89').
20505 The ISO C90 functions `abort', `abs', `acos', `asin', `atan2', `atan',
20506 `calloc', `ceil', `cosh', `cos', `exit', `exp', `fabs', `floor', `fmod',
20507 `fprintf', `fputs', `frexp', `fscanf', `isalnum', `isalpha', `iscntrl',
20508 `isdigit', `isgraph', `islower', `isprint', `ispunct', `isspace',
20509 `isupper', `isxdigit', `tolower', `toupper', `labs', `ldexp', `log10',
20510 `log', `malloc', `memcmp', `memcpy', `memset', `modf', `pow', `printf',
20511 `putchar', `puts', `scanf', `sinh', `sin', `snprintf', `sprintf',
20512 `sqrt', `sscanf', `strcat', `strchr', `strcmp', `strcpy', `strcspn',
20513 `strlen', `strncat', `strncmp', `strncpy', `strpbrk', `strrchr',
20514 `strspn', `strstr', `tanh', `tan', `vfprintf', `vprintf' and `vsprintf'
20515 are all recognized as built-in functions unless `-fno-builtin' is
20516 specified (or `-fno-builtin-FUNCTION' is specified for an individual
20517 function). All of these functions have corresponding versions prefixed
20520 GCC provides built-in versions of the ISO C99 floating point comparison
20521 macros that avoid raising exceptions for unordered operands. They have
20522 the same names as the standard macros ( `isgreater', `isgreaterequal',
20523 `isless', `islessequal', `islessgreater', and `isunordered') , with
20524 `__builtin_' prefixed. We intend for a library implementor to be able
20525 to simply `#define' each standard macro to its built-in equivalent.
20527 -- Built-in Function: int __builtin_types_compatible_p (TYPE1, TYPE2)
20528 You can use the built-in function `__builtin_types_compatible_p' to
20529 determine whether two types are the same.
20531 This built-in function returns 1 if the unqualified versions of the
20532 types TYPE1 and TYPE2 (which are types, not expressions) are
20533 compatible, 0 otherwise. The result of this built-in function can
20534 be used in integer constant expressions.
20536 This built-in function ignores top level qualifiers (e.g., `const',
20537 `volatile'). For example, `int' is equivalent to `const int'.
20539 The type `int[]' and `int[5]' are compatible. On the other hand,
20540 `int' and `char *' are not compatible, even if the size of their
20541 types, on the particular architecture are the same. Also, the
20542 amount of pointer indirection is taken into account when
20543 determining similarity. Consequently, `short *' is not similar to
20544 `short **'. Furthermore, two types that are typedefed are
20545 considered compatible if their underlying types are compatible.
20547 An `enum' type is not considered to be compatible with another
20548 `enum' type even if both are compatible with the same integer
20549 type; this is what the C standard specifies. For example, `enum
20550 {foo, bar}' is not similar to `enum {hot, dog}'.
20552 You would typically use this function in code whose execution
20553 varies depending on the arguments' types. For example:
20557 typeof (x) tmp = (x); \
20558 if (__builtin_types_compatible_p (typeof (x), long double)) \
20559 tmp = foo_long_double (tmp); \
20560 else if (__builtin_types_compatible_p (typeof (x), double)) \
20561 tmp = foo_double (tmp); \
20562 else if (__builtin_types_compatible_p (typeof (x), float)) \
20563 tmp = foo_float (tmp); \
20569 _Note:_ This construct is only available for C.
20572 -- Built-in Function: TYPE __builtin_choose_expr (CONST_EXP, EXP1,
20574 You can use the built-in function `__builtin_choose_expr' to
20575 evaluate code depending on the value of a constant expression.
20576 This built-in function returns EXP1 if CONST_EXP, which is a
20577 constant expression that must be able to be determined at compile
20578 time, is nonzero. Otherwise it returns 0.
20580 This built-in function is analogous to the `? :' operator in C,
20581 except that the expression returned has its type unaltered by
20582 promotion rules. Also, the built-in function does not evaluate
20583 the expression that was not chosen. For example, if CONST_EXP
20584 evaluates to true, EXP2 is not evaluated even if it has
20587 This built-in function can return an lvalue if the chosen argument
20590 If EXP1 is returned, the return type is the same as EXP1's type.
20591 Similarly, if EXP2 is returned, its return type is the same as
20597 __builtin_choose_expr ( \
20598 __builtin_types_compatible_p (typeof (x), double), \
20600 __builtin_choose_expr ( \
20601 __builtin_types_compatible_p (typeof (x), float), \
20603 /* The void expression results in a compile-time error \
20604 when assigning the result to something. */ \
20607 _Note:_ This construct is only available for C. Furthermore, the
20608 unused expression (EXP1 or EXP2 depending on the value of
20609 CONST_EXP) may still generate syntax errors. This may change in
20613 -- Built-in Function: int __builtin_constant_p (EXP)
20614 You can use the built-in function `__builtin_constant_p' to
20615 determine if a value is known to be constant at compile-time and
20616 hence that GCC can perform constant-folding on expressions
20617 involving that value. The argument of the function is the value
20618 to test. The function returns the integer 1 if the argument is
20619 known to be a compile-time constant and 0 if it is not known to be
20620 a compile-time constant. A return of 0 does not indicate that the
20621 value is _not_ a constant, but merely that GCC cannot prove it is
20622 a constant with the specified value of the `-O' option.
20624 You would typically use this function in an embedded application
20625 where memory was a critical resource. If you have some complex
20626 calculation, you may want it to be folded if it involves
20627 constants, but need to call a function if it does not. For
20630 #define Scale_Value(X) \
20631 (__builtin_constant_p (X) \
20632 ? ((X) * SCALE + OFFSET) : Scale (X))
20634 You may use this built-in function in either a macro or an inline
20635 function. However, if you use it in an inlined function and pass
20636 an argument of the function as the argument to the built-in, GCC
20637 will never return 1 when you call the inline function with a
20638 string constant or compound literal (*note Compound Literals::)
20639 and will not return 1 when you pass a constant numeric value to
20640 the inline function unless you specify the `-O' option.
20642 You may also use `__builtin_constant_p' in initializers for static
20643 data. For instance, you can write
20645 static const int table[] = {
20646 __builtin_constant_p (EXPRESSION) ? (EXPRESSION) : -1,
20650 This is an acceptable initializer even if EXPRESSION is not a
20651 constant expression. GCC must be more conservative about
20652 evaluating the built-in in this case, because it has no
20653 opportunity to perform optimization.
20655 Previous versions of GCC did not accept this built-in in data
20656 initializers. The earliest version where it is completely safe is
20659 -- Built-in Function: long __builtin_expect (long EXP, long C)
20660 You may use `__builtin_expect' to provide the compiler with branch
20661 prediction information. In general, you should prefer to use
20662 actual profile feedback for this (`-fprofile-arcs'), as
20663 programmers are notoriously bad at predicting how their programs
20664 actually perform. However, there are applications in which this
20665 data is hard to collect.
20667 The return value is the value of EXP, which should be an integral
20668 expression. The value of C must be a compile-time constant. The
20669 semantics of the built-in are that it is expected that EXP == C.
20672 if (__builtin_expect (x, 0))
20675 would indicate that we do not expect to call `foo', since we
20676 expect `x' to be zero. Since you are limited to integral
20677 expressions for EXP, you should use constructions such as
20679 if (__builtin_expect (ptr != NULL, 1))
20682 when testing pointer or floating-point values.
20684 -- Built-in Function: void __builtin_prefetch (const void *ADDR, ...)
20685 This function is used to minimize cache-miss latency by moving
20686 data into a cache before it is accessed. You can insert calls to
20687 `__builtin_prefetch' into code for which you know addresses of
20688 data in memory that is likely to be accessed soon. If the target
20689 supports them, data prefetch instructions will be generated. If
20690 the prefetch is done early enough before the access then the data
20691 will be in the cache by the time it is accessed.
20693 The value of ADDR is the address of the memory to prefetch. There
20694 are two optional arguments, RW and LOCALITY. The value of RW is a
20695 compile-time constant one or zero; one means that the prefetch is
20696 preparing for a write to the memory address and zero, the default,
20697 means that the prefetch is preparing for a read. The value
20698 LOCALITY must be a compile-time constant integer between zero and
20699 three. A value of zero means that the data has no temporal
20700 locality, so it need not be left in the cache after the access. A
20701 value of three means that the data has a high degree of temporal
20702 locality and should be left in all levels of cache possible.
20703 Values of one and two mean, respectively, a low or moderate degree
20704 of temporal locality. The default is three.
20706 for (i = 0; i < n; i++)
20708 a[i] = a[i] + b[i];
20709 __builtin_prefetch (&a[i+j], 1, 1);
20710 __builtin_prefetch (&b[i+j], 0, 1);
20714 Data prefetch does not generate faults if ADDR is invalid, but the
20715 address expression itself must be valid. For example, a prefetch
20716 of `p->next' will not fault if `p->next' is not a valid address,
20717 but evaluation will fault if `p' is not a valid address.
20719 If the target does not support data prefetch, the address
20720 expression is evaluated if it includes side effects but no other
20721 code is generated and GCC does not issue a warning.
20723 -- Built-in Function: double __builtin_huge_val (void)
20724 Returns a positive infinity, if supported by the floating-point
20725 format, else `DBL_MAX'. This function is suitable for
20726 implementing the ISO C macro `HUGE_VAL'.
20728 -- Built-in Function: float __builtin_huge_valf (void)
20729 Similar to `__builtin_huge_val', except the return type is `float'.
20731 -- Built-in Function: long double __builtin_huge_vall (void)
20732 Similar to `__builtin_huge_val', except the return type is `long
20735 -- Built-in Function: double __builtin_inf (void)
20736 Similar to `__builtin_huge_val', except a warning is generated if
20737 the target floating-point format does not support infinities.
20739 -- Built-in Function: _Decimal32 __builtin_infd32 (void)
20740 Similar to `__builtin_inf', except the return type is `_Decimal32'.
20742 -- Built-in Function: _Decimal64 __builtin_infd64 (void)
20743 Similar to `__builtin_inf', except the return type is `_Decimal64'.
20745 -- Built-in Function: _Decimal128 __builtin_infd128 (void)
20746 Similar to `__builtin_inf', except the return type is
20749 -- Built-in Function: float __builtin_inff (void)
20750 Similar to `__builtin_inf', except the return type is `float'.
20751 This function is suitable for implementing the ISO C99 macro
20754 -- Built-in Function: long double __builtin_infl (void)
20755 Similar to `__builtin_inf', except the return type is `long
20758 -- Built-in Function: double __builtin_nan (const char *str)
20759 This is an implementation of the ISO C99 function `nan'.
20761 Since ISO C99 defines this function in terms of `strtod', which we
20762 do not implement, a description of the parsing is in order. The
20763 string is parsed as by `strtol'; that is, the base is recognized by
20764 leading `0' or `0x' prefixes. The number parsed is placed in the
20765 significand such that the least significant bit of the number is
20766 at the least significant bit of the significand. The number is
20767 truncated to fit the significand field provided. The significand
20768 is forced to be a quiet NaN.
20770 This function, if given a string literal all of which would have
20771 been consumed by strtol, is evaluated early enough that it is
20772 considered a compile-time constant.
20774 -- Built-in Function: _Decimal32 __builtin_nand32 (const char *str)
20775 Similar to `__builtin_nan', except the return type is `_Decimal32'.
20777 -- Built-in Function: _Decimal64 __builtin_nand64 (const char *str)
20778 Similar to `__builtin_nan', except the return type is `_Decimal64'.
20780 -- Built-in Function: _Decimal128 __builtin_nand128 (const char *str)
20781 Similar to `__builtin_nan', except the return type is
20784 -- Built-in Function: float __builtin_nanf (const char *str)
20785 Similar to `__builtin_nan', except the return type is `float'.
20787 -- Built-in Function: long double __builtin_nanl (const char *str)
20788 Similar to `__builtin_nan', except the return type is `long
20791 -- Built-in Function: double __builtin_nans (const char *str)
20792 Similar to `__builtin_nan', except the significand is forced to be
20793 a signaling NaN. The `nans' function is proposed by WG14 N965.
20795 -- Built-in Function: float __builtin_nansf (const char *str)
20796 Similar to `__builtin_nans', except the return type is `float'.
20798 -- Built-in Function: long double __builtin_nansl (const char *str)
20799 Similar to `__builtin_nans', except the return type is `long
20802 -- Built-in Function: int __builtin_ffs (unsigned int x)
20803 Returns one plus the index of the least significant 1-bit of X, or
20804 if X is zero, returns zero.
20806 -- Built-in Function: int __builtin_clz (unsigned int x)
20807 Returns the number of leading 0-bits in X, starting at the most
20808 significant bit position. If X is 0, the result is undefined.
20810 -- Built-in Function: int __builtin_ctz (unsigned int x)
20811 Returns the number of trailing 0-bits in X, starting at the least
20812 significant bit position. If X is 0, the result is undefined.
20814 -- Built-in Function: int __builtin_popcount (unsigned int x)
20815 Returns the number of 1-bits in X.
20817 -- Built-in Function: int __builtin_parity (unsigned int x)
20818 Returns the parity of X, i.e. the number of 1-bits in X modulo 2.
20820 -- Built-in Function: int __builtin_ffsl (unsigned long)
20821 Similar to `__builtin_ffs', except the argument type is `unsigned
20824 -- Built-in Function: int __builtin_clzl (unsigned long)
20825 Similar to `__builtin_clz', except the argument type is `unsigned
20828 -- Built-in Function: int __builtin_ctzl (unsigned long)
20829 Similar to `__builtin_ctz', except the argument type is `unsigned
20832 -- Built-in Function: int __builtin_popcountl (unsigned long)
20833 Similar to `__builtin_popcount', except the argument type is
20836 -- Built-in Function: int __builtin_parityl (unsigned long)
20837 Similar to `__builtin_parity', except the argument type is
20840 -- Built-in Function: int __builtin_ffsll (unsigned long long)
20841 Similar to `__builtin_ffs', except the argument type is `unsigned
20844 -- Built-in Function: int __builtin_clzll (unsigned long long)
20845 Similar to `__builtin_clz', except the argument type is `unsigned
20848 -- Built-in Function: int __builtin_ctzll (unsigned long long)
20849 Similar to `__builtin_ctz', except the argument type is `unsigned
20852 -- Built-in Function: int __builtin_popcountll (unsigned long long)
20853 Similar to `__builtin_popcount', except the argument type is
20854 `unsigned long long'.
20856 -- Built-in Function: int __builtin_parityll (unsigned long long)
20857 Similar to `__builtin_parity', except the argument type is
20858 `unsigned long long'.
20860 -- Built-in Function: double __builtin_powi (double, int)
20861 Returns the first argument raised to the power of the second.
20862 Unlike the `pow' function no guarantees about precision and
20865 -- Built-in Function: float __builtin_powif (float, int)
20866 Similar to `__builtin_powi', except the argument and return types
20869 -- Built-in Function: long double __builtin_powil (long double, int)
20870 Similar to `__builtin_powi', except the argument and return types
20874 File: gcc.info, Node: Target Builtins, Next: Target Format Checks, Prev: Other Builtins, Up: C Extensions
20876 5.48 Built-in Functions Specific to Particular Target Machines
20877 ==============================================================
20879 On some target machines, GCC supports many built-in functions specific
20880 to those machines. Generally these generate calls to specific machine
20881 instructions, but allow the compiler to schedule those calls.
20885 * Alpha Built-in Functions::
20886 * ARM Built-in Functions::
20887 * Blackfin Built-in Functions::
20888 * FR-V Built-in Functions::
20889 * X86 Built-in Functions::
20890 * MIPS DSP Built-in Functions::
20891 * MIPS Paired-Single Support::
20892 * PowerPC AltiVec Built-in Functions::
20893 * SPARC VIS Built-in Functions::
20896 File: gcc.info, Node: Alpha Built-in Functions, Next: ARM Built-in Functions, Up: Target Builtins
20898 5.48.1 Alpha Built-in Functions
20899 -------------------------------
20901 These built-in functions are available for the Alpha family of
20902 processors, depending on the command-line switches used.
20904 The following built-in functions are always available. They all
20905 generate the machine instruction that is part of the name.
20907 long __builtin_alpha_implver (void)
20908 long __builtin_alpha_rpcc (void)
20909 long __builtin_alpha_amask (long)
20910 long __builtin_alpha_cmpbge (long, long)
20911 long __builtin_alpha_extbl (long, long)
20912 long __builtin_alpha_extwl (long, long)
20913 long __builtin_alpha_extll (long, long)
20914 long __builtin_alpha_extql (long, long)
20915 long __builtin_alpha_extwh (long, long)
20916 long __builtin_alpha_extlh (long, long)
20917 long __builtin_alpha_extqh (long, long)
20918 long __builtin_alpha_insbl (long, long)
20919 long __builtin_alpha_inswl (long, long)
20920 long __builtin_alpha_insll (long, long)
20921 long __builtin_alpha_insql (long, long)
20922 long __builtin_alpha_inswh (long, long)
20923 long __builtin_alpha_inslh (long, long)
20924 long __builtin_alpha_insqh (long, long)
20925 long __builtin_alpha_mskbl (long, long)
20926 long __builtin_alpha_mskwl (long, long)
20927 long __builtin_alpha_mskll (long, long)
20928 long __builtin_alpha_mskql (long, long)
20929 long __builtin_alpha_mskwh (long, long)
20930 long __builtin_alpha_msklh (long, long)
20931 long __builtin_alpha_mskqh (long, long)
20932 long __builtin_alpha_umulh (long, long)
20933 long __builtin_alpha_zap (long, long)
20934 long __builtin_alpha_zapnot (long, long)
20936 The following built-in functions are always with `-mmax' or
20937 `-mcpu=CPU' where CPU is `pca56' or later. They all generate the
20938 machine instruction that is part of the name.
20940 long __builtin_alpha_pklb (long)
20941 long __builtin_alpha_pkwb (long)
20942 long __builtin_alpha_unpkbl (long)
20943 long __builtin_alpha_unpkbw (long)
20944 long __builtin_alpha_minub8 (long, long)
20945 long __builtin_alpha_minsb8 (long, long)
20946 long __builtin_alpha_minuw4 (long, long)
20947 long __builtin_alpha_minsw4 (long, long)
20948 long __builtin_alpha_maxub8 (long, long)
20949 long __builtin_alpha_maxsb8 (long, long)
20950 long __builtin_alpha_maxuw4 (long, long)
20951 long __builtin_alpha_maxsw4 (long, long)
20952 long __builtin_alpha_perr (long, long)
20954 The following built-in functions are always with `-mcix' or
20955 `-mcpu=CPU' where CPU is `ev67' or later. They all generate the
20956 machine instruction that is part of the name.
20958 long __builtin_alpha_cttz (long)
20959 long __builtin_alpha_ctlz (long)
20960 long __builtin_alpha_ctpop (long)
20962 The following builtins are available on systems that use the OSF/1
20963 PALcode. Normally they invoke the `rduniq' and `wruniq' PAL calls, but
20964 when invoked with `-mtls-kernel', they invoke `rdval' and `wrval'.
20966 void *__builtin_thread_pointer (void)
20967 void __builtin_set_thread_pointer (void *)
20970 File: gcc.info, Node: ARM Built-in Functions, Next: Blackfin Built-in Functions, Prev: Alpha Built-in Functions, Up: Target Builtins
20972 5.48.2 ARM Built-in Functions
20973 -----------------------------
20975 These built-in functions are available for the ARM family of
20976 processors, when the `-mcpu=iwmmxt' switch is used:
20978 typedef int v2si __attribute__ ((vector_size (8)));
20979 typedef short v4hi __attribute__ ((vector_size (8)));
20980 typedef char v8qi __attribute__ ((vector_size (8)));
20982 int __builtin_arm_getwcx (int)
20983 void __builtin_arm_setwcx (int, int)
20984 int __builtin_arm_textrmsb (v8qi, int)
20985 int __builtin_arm_textrmsh (v4hi, int)
20986 int __builtin_arm_textrmsw (v2si, int)
20987 int __builtin_arm_textrmub (v8qi, int)
20988 int __builtin_arm_textrmuh (v4hi, int)
20989 int __builtin_arm_textrmuw (v2si, int)
20990 v8qi __builtin_arm_tinsrb (v8qi, int)
20991 v4hi __builtin_arm_tinsrh (v4hi, int)
20992 v2si __builtin_arm_tinsrw (v2si, int)
20993 long long __builtin_arm_tmia (long long, int, int)
20994 long long __builtin_arm_tmiabb (long long, int, int)
20995 long long __builtin_arm_tmiabt (long long, int, int)
20996 long long __builtin_arm_tmiaph (long long, int, int)
20997 long long __builtin_arm_tmiatb (long long, int, int)
20998 long long __builtin_arm_tmiatt (long long, int, int)
20999 int __builtin_arm_tmovmskb (v8qi)
21000 int __builtin_arm_tmovmskh (v4hi)
21001 int __builtin_arm_tmovmskw (v2si)
21002 long long __builtin_arm_waccb (v8qi)
21003 long long __builtin_arm_wacch (v4hi)
21004 long long __builtin_arm_waccw (v2si)
21005 v8qi __builtin_arm_waddb (v8qi, v8qi)
21006 v8qi __builtin_arm_waddbss (v8qi, v8qi)
21007 v8qi __builtin_arm_waddbus (v8qi, v8qi)
21008 v4hi __builtin_arm_waddh (v4hi, v4hi)
21009 v4hi __builtin_arm_waddhss (v4hi, v4hi)
21010 v4hi __builtin_arm_waddhus (v4hi, v4hi)
21011 v2si __builtin_arm_waddw (v2si, v2si)
21012 v2si __builtin_arm_waddwss (v2si, v2si)
21013 v2si __builtin_arm_waddwus (v2si, v2si)
21014 v8qi __builtin_arm_walign (v8qi, v8qi, int)
21015 long long __builtin_arm_wand(long long, long long)
21016 long long __builtin_arm_wandn (long long, long long)
21017 v8qi __builtin_arm_wavg2b (v8qi, v8qi)
21018 v8qi __builtin_arm_wavg2br (v8qi, v8qi)
21019 v4hi __builtin_arm_wavg2h (v4hi, v4hi)
21020 v4hi __builtin_arm_wavg2hr (v4hi, v4hi)
21021 v8qi __builtin_arm_wcmpeqb (v8qi, v8qi)
21022 v4hi __builtin_arm_wcmpeqh (v4hi, v4hi)
21023 v2si __builtin_arm_wcmpeqw (v2si, v2si)
21024 v8qi __builtin_arm_wcmpgtsb (v8qi, v8qi)
21025 v4hi __builtin_arm_wcmpgtsh (v4hi, v4hi)
21026 v2si __builtin_arm_wcmpgtsw (v2si, v2si)
21027 v8qi __builtin_arm_wcmpgtub (v8qi, v8qi)
21028 v4hi __builtin_arm_wcmpgtuh (v4hi, v4hi)
21029 v2si __builtin_arm_wcmpgtuw (v2si, v2si)
21030 long long __builtin_arm_wmacs (long long, v4hi, v4hi)
21031 long long __builtin_arm_wmacsz (v4hi, v4hi)
21032 long long __builtin_arm_wmacu (long long, v4hi, v4hi)
21033 long long __builtin_arm_wmacuz (v4hi, v4hi)
21034 v4hi __builtin_arm_wmadds (v4hi, v4hi)
21035 v4hi __builtin_arm_wmaddu (v4hi, v4hi)
21036 v8qi __builtin_arm_wmaxsb (v8qi, v8qi)
21037 v4hi __builtin_arm_wmaxsh (v4hi, v4hi)
21038 v2si __builtin_arm_wmaxsw (v2si, v2si)
21039 v8qi __builtin_arm_wmaxub (v8qi, v8qi)
21040 v4hi __builtin_arm_wmaxuh (v4hi, v4hi)
21041 v2si __builtin_arm_wmaxuw (v2si, v2si)
21042 v8qi __builtin_arm_wminsb (v8qi, v8qi)
21043 v4hi __builtin_arm_wminsh (v4hi, v4hi)
21044 v2si __builtin_arm_wminsw (v2si, v2si)
21045 v8qi __builtin_arm_wminub (v8qi, v8qi)
21046 v4hi __builtin_arm_wminuh (v4hi, v4hi)
21047 v2si __builtin_arm_wminuw (v2si, v2si)
21048 v4hi __builtin_arm_wmulsm (v4hi, v4hi)
21049 v4hi __builtin_arm_wmulul (v4hi, v4hi)
21050 v4hi __builtin_arm_wmulum (v4hi, v4hi)
21051 long long __builtin_arm_wor (long long, long long)
21052 v2si __builtin_arm_wpackdss (long long, long long)
21053 v2si __builtin_arm_wpackdus (long long, long long)
21054 v8qi __builtin_arm_wpackhss (v4hi, v4hi)
21055 v8qi __builtin_arm_wpackhus (v4hi, v4hi)
21056 v4hi __builtin_arm_wpackwss (v2si, v2si)
21057 v4hi __builtin_arm_wpackwus (v2si, v2si)
21058 long long __builtin_arm_wrord (long long, long long)
21059 long long __builtin_arm_wrordi (long long, int)
21060 v4hi __builtin_arm_wrorh (v4hi, long long)
21061 v4hi __builtin_arm_wrorhi (v4hi, int)
21062 v2si __builtin_arm_wrorw (v2si, long long)
21063 v2si __builtin_arm_wrorwi (v2si, int)
21064 v2si __builtin_arm_wsadb (v8qi, v8qi)
21065 v2si __builtin_arm_wsadbz (v8qi, v8qi)
21066 v2si __builtin_arm_wsadh (v4hi, v4hi)
21067 v2si __builtin_arm_wsadhz (v4hi, v4hi)
21068 v4hi __builtin_arm_wshufh (v4hi, int)
21069 long long __builtin_arm_wslld (long long, long long)
21070 long long __builtin_arm_wslldi (long long, int)
21071 v4hi __builtin_arm_wsllh (v4hi, long long)
21072 v4hi __builtin_arm_wsllhi (v4hi, int)
21073 v2si __builtin_arm_wsllw (v2si, long long)
21074 v2si __builtin_arm_wsllwi (v2si, int)
21075 long long __builtin_arm_wsrad (long long, long long)
21076 long long __builtin_arm_wsradi (long long, int)
21077 v4hi __builtin_arm_wsrah (v4hi, long long)
21078 v4hi __builtin_arm_wsrahi (v4hi, int)
21079 v2si __builtin_arm_wsraw (v2si, long long)
21080 v2si __builtin_arm_wsrawi (v2si, int)
21081 long long __builtin_arm_wsrld (long long, long long)
21082 long long __builtin_arm_wsrldi (long long, int)
21083 v4hi __builtin_arm_wsrlh (v4hi, long long)
21084 v4hi __builtin_arm_wsrlhi (v4hi, int)
21085 v2si __builtin_arm_wsrlw (v2si, long long)
21086 v2si __builtin_arm_wsrlwi (v2si, int)
21087 v8qi __builtin_arm_wsubb (v8qi, v8qi)
21088 v8qi __builtin_arm_wsubbss (v8qi, v8qi)
21089 v8qi __builtin_arm_wsubbus (v8qi, v8qi)
21090 v4hi __builtin_arm_wsubh (v4hi, v4hi)
21091 v4hi __builtin_arm_wsubhss (v4hi, v4hi)
21092 v4hi __builtin_arm_wsubhus (v4hi, v4hi)
21093 v2si __builtin_arm_wsubw (v2si, v2si)
21094 v2si __builtin_arm_wsubwss (v2si, v2si)
21095 v2si __builtin_arm_wsubwus (v2si, v2si)
21096 v4hi __builtin_arm_wunpckehsb (v8qi)
21097 v2si __builtin_arm_wunpckehsh (v4hi)
21098 long long __builtin_arm_wunpckehsw (v2si)
21099 v4hi __builtin_arm_wunpckehub (v8qi)
21100 v2si __builtin_arm_wunpckehuh (v4hi)
21101 long long __builtin_arm_wunpckehuw (v2si)
21102 v4hi __builtin_arm_wunpckelsb (v8qi)
21103 v2si __builtin_arm_wunpckelsh (v4hi)
21104 long long __builtin_arm_wunpckelsw (v2si)
21105 v4hi __builtin_arm_wunpckelub (v8qi)
21106 v2si __builtin_arm_wunpckeluh (v4hi)
21107 long long __builtin_arm_wunpckeluw (v2si)
21108 v8qi __builtin_arm_wunpckihb (v8qi, v8qi)
21109 v4hi __builtin_arm_wunpckihh (v4hi, v4hi)
21110 v2si __builtin_arm_wunpckihw (v2si, v2si)
21111 v8qi __builtin_arm_wunpckilb (v8qi, v8qi)
21112 v4hi __builtin_arm_wunpckilh (v4hi, v4hi)
21113 v2si __builtin_arm_wunpckilw (v2si, v2si)
21114 long long __builtin_arm_wxor (long long, long long)
21115 long long __builtin_arm_wzero ()
21118 File: gcc.info, Node: Blackfin Built-in Functions, Next: FR-V Built-in Functions, Prev: ARM Built-in Functions, Up: Target Builtins
21120 5.48.3 Blackfin Built-in Functions
21121 ----------------------------------
21123 Currently, there are two Blackfin-specific built-in functions. These
21124 are used for generating `CSYNC' and `SSYNC' machine insns without using
21125 inline assembly; by using these built-in functions the compiler can
21126 automatically add workarounds for hardware errata involving these
21127 instructions. These functions are named as follows:
21129 void __builtin_bfin_csync (void)
21130 void __builtin_bfin_ssync (void)
21133 File: gcc.info, Node: FR-V Built-in Functions, Next: X86 Built-in Functions, Prev: Blackfin Built-in Functions, Up: Target Builtins
21135 5.48.4 FR-V Built-in Functions
21136 ------------------------------
21138 GCC provides many FR-V-specific built-in functions. In general, these
21139 functions are intended to be compatible with those described by `FR-V
21140 Family, Softune C/C++ Compiler Manual (V6), Fujitsu Semiconductor'.
21141 The two exceptions are `__MDUNPACKH' and `__MBTOHE', the gcc forms of
21142 which pass 128-bit values by pointer rather than by value.
21144 Most of the functions are named after specific FR-V instructions.
21145 Such functions are said to be "directly mapped" and are summarized here
21151 * Directly-mapped Integer Functions::
21152 * Directly-mapped Media Functions::
21153 * Raw read/write Functions::
21154 * Other Built-in Functions::
21157 File: gcc.info, Node: Argument Types, Next: Directly-mapped Integer Functions, Up: FR-V Built-in Functions
21159 5.48.4.1 Argument Types
21160 .......................
21162 The arguments to the built-in functions can be divided into three
21163 groups: register numbers, compile-time constants and run-time values.
21164 In order to make this classification clear at a glance, the arguments
21165 and return values are given the following pseudo types:
21167 Pseudo type Real C type Constant? Description
21168 `uh' `unsigned short' No an unsigned halfword
21169 `uw1' `unsigned int' No an unsigned word
21170 `sw1' `int' No a signed word
21171 `uw2' `unsigned long long' No an unsigned doubleword
21172 `sw2' `long long' No a signed doubleword
21173 `const' `int' Yes an integer constant
21174 `acc' `int' Yes an ACC register number
21175 `iacc' `int' Yes an IACC register number
21177 These pseudo types are not defined by GCC, they are simply a notational
21178 convenience used in this manual.
21180 Arguments of type `uh', `uw1', `sw1', `uw2' and `sw2' are evaluated at
21181 run time. They correspond to register operands in the underlying FR-V
21184 `const' arguments represent immediate operands in the underlying FR-V
21185 instructions. They must be compile-time constants.
21187 `acc' arguments are evaluated at compile time and specify the number
21188 of an accumulator register. For example, an `acc' argument of 2 will
21189 select the ACC2 register.
21191 `iacc' arguments are similar to `acc' arguments but specify the number
21192 of an IACC register. See *note Other Built-in Functions:: for more
21196 File: gcc.info, Node: Directly-mapped Integer Functions, Next: Directly-mapped Media Functions, Prev: Argument Types, Up: FR-V Built-in Functions
21198 5.48.4.2 Directly-mapped Integer Functions
21199 ..........................................
21201 The functions listed below map directly to FR-V I-type instructions.
21203 Function prototype Example usage Assembly output
21204 `sw1 __ADDSS (sw1, sw1)' `C = __ADDSS (A, B)' `ADDSS A,B,C'
21205 `sw1 __SCAN (sw1, sw1)' `C = __SCAN (A, B)' `SCAN A,B,C'
21206 `sw1 __SCUTSS (sw1)' `B = __SCUTSS (A)' `SCUTSS A,B'
21207 `sw1 __SLASS (sw1, sw1)' `C = __SLASS (A, B)' `SLASS A,B,C'
21208 `void __SMASS (sw1, sw1)' `__SMASS (A, B)' `SMASS A,B'
21209 `void __SMSSS (sw1, sw1)' `__SMSSS (A, B)' `SMSSS A,B'
21210 `void __SMU (sw1, sw1)' `__SMU (A, B)' `SMU A,B'
21211 `sw2 __SMUL (sw1, sw1)' `C = __SMUL (A, B)' `SMUL A,B,C'
21212 `sw1 __SUBSS (sw1, sw1)' `C = __SUBSS (A, B)' `SUBSS A,B,C'
21213 `uw2 __UMUL (uw1, uw1)' `C = __UMUL (A, B)' `UMUL A,B,C'
21216 File: gcc.info, Node: Directly-mapped Media Functions, Next: Raw read/write Functions, Prev: Directly-mapped Integer Functions, Up: FR-V Built-in Functions
21218 5.48.4.3 Directly-mapped Media Functions
21219 ........................................
21221 The functions listed below map directly to FR-V M-type instructions.
21223 Function prototype Example usage Assembly output
21224 `uw1 __MABSHS (sw1)' `B = __MABSHS (A)' `MABSHS A,B'
21225 `void __MADDACCS (acc, acc)' `__MADDACCS (B, A)' `MADDACCS A,B'
21226 `sw1 __MADDHSS (sw1, sw1)' `C = __MADDHSS (A, B)' `MADDHSS A,B,C'
21227 `uw1 __MADDHUS (uw1, uw1)' `C = __MADDHUS (A, B)' `MADDHUS A,B,C'
21228 `uw1 __MAND (uw1, uw1)' `C = __MAND (A, B)' `MAND A,B,C'
21229 `void __MASACCS (acc, acc)' `__MASACCS (B, A)' `MASACCS A,B'
21230 `uw1 __MAVEH (uw1, uw1)' `C = __MAVEH (A, B)' `MAVEH A,B,C'
21231 `uw2 __MBTOH (uw1)' `B = __MBTOH (A)' `MBTOH A,B'
21232 `void __MBTOHE (uw1 *, uw1)' `__MBTOHE (&B, A)' `MBTOHE A,B'
21233 `void __MCLRACC (acc)' `__MCLRACC (A)' `MCLRACC A'
21234 `void __MCLRACCA (void)' `__MCLRACCA ()' `MCLRACCA'
21235 `uw1 __Mcop1 (uw1, uw1)' `C = __Mcop1 (A, B)' `Mcop1 A,B,C'
21236 `uw1 __Mcop2 (uw1, uw1)' `C = __Mcop2 (A, B)' `Mcop2 A,B,C'
21237 `uw1 __MCPLHI (uw2, const)' `C = __MCPLHI (A, B)' `MCPLHI A,#B,C'
21238 `uw1 __MCPLI (uw2, const)' `C = __MCPLI (A, B)' `MCPLI A,#B,C'
21239 `void __MCPXIS (acc, sw1, sw1)' `__MCPXIS (C, A, B)' `MCPXIS A,B,C'
21240 `void __MCPXIU (acc, uw1, uw1)' `__MCPXIU (C, A, B)' `MCPXIU A,B,C'
21241 `void __MCPXRS (acc, sw1, sw1)' `__MCPXRS (C, A, B)' `MCPXRS A,B,C'
21242 `void __MCPXRU (acc, uw1, uw1)' `__MCPXRU (C, A, B)' `MCPXRU A,B,C'
21243 `uw1 __MCUT (acc, uw1)' `C = __MCUT (A, B)' `MCUT A,B,C'
21244 `uw1 __MCUTSS (acc, sw1)' `C = __MCUTSS (A, B)' `MCUTSS A,B,C'
21245 `void __MDADDACCS (acc, acc)' `__MDADDACCS (B, A)' `MDADDACCS A,B'
21246 `void __MDASACCS (acc, acc)' `__MDASACCS (B, A)' `MDASACCS A,B'
21247 `uw2 __MDCUTSSI (acc, const)' `C = __MDCUTSSI (A, B)' `MDCUTSSI A,#B,C'
21248 `uw2 __MDPACKH (uw2, uw2)' `C = __MDPACKH (A, B)' `MDPACKH A,B,C'
21249 `uw2 __MDROTLI (uw2, const)' `C = __MDROTLI (A, B)' `MDROTLI A,#B,C'
21250 `void __MDSUBACCS (acc, acc)' `__MDSUBACCS (B, A)' `MDSUBACCS A,B'
21251 `void __MDUNPACKH (uw1 *, uw2)' `__MDUNPACKH (&B, A)' `MDUNPACKH A,B'
21252 `uw2 __MEXPDHD (uw1, const)' `C = __MEXPDHD (A, B)' `MEXPDHD A,#B,C'
21253 `uw1 __MEXPDHW (uw1, const)' `C = __MEXPDHW (A, B)' `MEXPDHW A,#B,C'
21254 `uw1 __MHDSETH (uw1, const)' `C = __MHDSETH (A, B)' `MHDSETH A,#B,C'
21255 `sw1 __MHDSETS (const)' `B = __MHDSETS (A)' `MHDSETS #A,B'
21256 `uw1 __MHSETHIH (uw1, const)' `B = __MHSETHIH (B, A)' `MHSETHIH #A,B'
21257 `sw1 __MHSETHIS (sw1, const)' `B = __MHSETHIS (B, A)' `MHSETHIS #A,B'
21258 `uw1 __MHSETLOH (uw1, const)' `B = __MHSETLOH (B, A)' `MHSETLOH #A,B'
21259 `sw1 __MHSETLOS (sw1, const)' `B = __MHSETLOS (B, A)' `MHSETLOS #A,B'
21260 `uw1 __MHTOB (uw2)' `B = __MHTOB (A)' `MHTOB A,B'
21261 `void __MMACHS (acc, sw1, sw1)' `__MMACHS (C, A, B)' `MMACHS A,B,C'
21262 `void __MMACHU (acc, uw1, uw1)' `__MMACHU (C, A, B)' `MMACHU A,B,C'
21263 `void __MMRDHS (acc, sw1, sw1)' `__MMRDHS (C, A, B)' `MMRDHS A,B,C'
21264 `void __MMRDHU (acc, uw1, uw1)' `__MMRDHU (C, A, B)' `MMRDHU A,B,C'
21265 `void __MMULHS (acc, sw1, sw1)' `__MMULHS (C, A, B)' `MMULHS A,B,C'
21266 `void __MMULHU (acc, uw1, uw1)' `__MMULHU (C, A, B)' `MMULHU A,B,C'
21267 `void __MMULXHS (acc, sw1, sw1)' `__MMULXHS (C, A, B)' `MMULXHS A,B,C'
21268 `void __MMULXHU (acc, uw1, uw1)' `__MMULXHU (C, A, B)' `MMULXHU A,B,C'
21269 `uw1 __MNOT (uw1)' `B = __MNOT (A)' `MNOT A,B'
21270 `uw1 __MOR (uw1, uw1)' `C = __MOR (A, B)' `MOR A,B,C'
21271 `uw1 __MPACKH (uh, uh)' `C = __MPACKH (A, B)' `MPACKH A,B,C'
21272 `sw2 __MQADDHSS (sw2, sw2)' `C = __MQADDHSS (A, B)' `MQADDHSS A,B,C'
21273 `uw2 __MQADDHUS (uw2, uw2)' `C = __MQADDHUS (A, B)' `MQADDHUS A,B,C'
21274 `void __MQCPXIS (acc, sw2, sw2)' `__MQCPXIS (C, A, B)' `MQCPXIS A,B,C'
21275 `void __MQCPXIU (acc, uw2, uw2)' `__MQCPXIU (C, A, B)' `MQCPXIU A,B,C'
21276 `void __MQCPXRS (acc, sw2, sw2)' `__MQCPXRS (C, A, B)' `MQCPXRS A,B,C'
21277 `void __MQCPXRU (acc, uw2, uw2)' `__MQCPXRU (C, A, B)' `MQCPXRU A,B,C'
21278 `sw2 __MQLCLRHS (sw2, sw2)' `C = __MQLCLRHS (A, B)' `MQLCLRHS A,B,C'
21279 `sw2 __MQLMTHS (sw2, sw2)' `C = __MQLMTHS (A, B)' `MQLMTHS A,B,C'
21280 `void __MQMACHS (acc, sw2, sw2)' `__MQMACHS (C, A, B)' `MQMACHS A,B,C'
21281 `void __MQMACHU (acc, uw2, uw2)' `__MQMACHU (C, A, B)' `MQMACHU A,B,C'
21282 `void __MQMACXHS (acc, sw2, `__MQMACXHS (C, A, B)' `MQMACXHS A,B,C'
21284 `void __MQMULHS (acc, sw2, sw2)' `__MQMULHS (C, A, B)' `MQMULHS A,B,C'
21285 `void __MQMULHU (acc, uw2, uw2)' `__MQMULHU (C, A, B)' `MQMULHU A,B,C'
21286 `void __MQMULXHS (acc, sw2, `__MQMULXHS (C, A, B)' `MQMULXHS A,B,C'
21288 `void __MQMULXHU (acc, uw2, `__MQMULXHU (C, A, B)' `MQMULXHU A,B,C'
21290 `sw2 __MQSATHS (sw2, sw2)' `C = __MQSATHS (A, B)' `MQSATHS A,B,C'
21291 `uw2 __MQSLLHI (uw2, int)' `C = __MQSLLHI (A, B)' `MQSLLHI A,B,C'
21292 `sw2 __MQSRAHI (sw2, int)' `C = __MQSRAHI (A, B)' `MQSRAHI A,B,C'
21293 `sw2 __MQSUBHSS (sw2, sw2)' `C = __MQSUBHSS (A, B)' `MQSUBHSS A,B,C'
21294 `uw2 __MQSUBHUS (uw2, uw2)' `C = __MQSUBHUS (A, B)' `MQSUBHUS A,B,C'
21295 `void __MQXMACHS (acc, sw2, `__MQXMACHS (C, A, B)' `MQXMACHS A,B,C'
21297 `void __MQXMACXHS (acc, sw2, `__MQXMACXHS (C, A, B)' `MQXMACXHS A,B,C'
21299 `uw1 __MRDACC (acc)' `B = __MRDACC (A)' `MRDACC A,B'
21300 `uw1 __MRDACCG (acc)' `B = __MRDACCG (A)' `MRDACCG A,B'
21301 `uw1 __MROTLI (uw1, const)' `C = __MROTLI (A, B)' `MROTLI A,#B,C'
21302 `uw1 __MROTRI (uw1, const)' `C = __MROTRI (A, B)' `MROTRI A,#B,C'
21303 `sw1 __MSATHS (sw1, sw1)' `C = __MSATHS (A, B)' `MSATHS A,B,C'
21304 `uw1 __MSATHU (uw1, uw1)' `C = __MSATHU (A, B)' `MSATHU A,B,C'
21305 `uw1 __MSLLHI (uw1, const)' `C = __MSLLHI (A, B)' `MSLLHI A,#B,C'
21306 `sw1 __MSRAHI (sw1, const)' `C = __MSRAHI (A, B)' `MSRAHI A,#B,C'
21307 `uw1 __MSRLHI (uw1, const)' `C = __MSRLHI (A, B)' `MSRLHI A,#B,C'
21308 `void __MSUBACCS (acc, acc)' `__MSUBACCS (B, A)' `MSUBACCS A,B'
21309 `sw1 __MSUBHSS (sw1, sw1)' `C = __MSUBHSS (A, B)' `MSUBHSS A,B,C'
21310 `uw1 __MSUBHUS (uw1, uw1)' `C = __MSUBHUS (A, B)' `MSUBHUS A,B,C'
21311 `void __MTRAP (void)' `__MTRAP ()' `MTRAP'
21312 `uw2 __MUNPACKH (uw1)' `B = __MUNPACKH (A)' `MUNPACKH A,B'
21313 `uw1 __MWCUT (uw2, uw1)' `C = __MWCUT (A, B)' `MWCUT A,B,C'
21314 `void __MWTACC (acc, uw1)' `__MWTACC (B, A)' `MWTACC A,B'
21315 `void __MWTACCG (acc, uw1)' `__MWTACCG (B, A)' `MWTACCG A,B'
21316 `uw1 __MXOR (uw1, uw1)' `C = __MXOR (A, B)' `MXOR A,B,C'
21319 File: gcc.info, Node: Raw read/write Functions, Next: Other Built-in Functions, Prev: Directly-mapped Media Functions, Up: FR-V Built-in Functions
21321 5.48.4.4 Raw read/write Functions
21322 .................................
21324 This sections describes built-in functions related to read and write
21325 instructions to access memory. These functions generate `membar'
21326 instructions to flush the I/O load and stores where appropriate, as
21327 described in Fujitsu's manual described above.
21329 `unsigned char __builtin_read8 (void *DATA)'
21331 `unsigned short __builtin_read16 (void *DATA)'
21333 `unsigned long __builtin_read32 (void *DATA)'
21335 `unsigned long long __builtin_read64 (void *DATA)'
21337 `void __builtin_write8 (void *DATA, unsigned char DATUM)'
21339 `void __builtin_write16 (void *DATA, unsigned short DATUM)'
21341 `void __builtin_write32 (void *DATA, unsigned long DATUM)'
21343 `void __builtin_write64 (void *DATA, unsigned long long DATUM)'
21346 File: gcc.info, Node: Other Built-in Functions, Prev: Raw read/write Functions, Up: FR-V Built-in Functions
21348 5.48.4.5 Other Built-in Functions
21349 .................................
21351 This section describes built-in functions that are not named after a
21352 specific FR-V instruction.
21354 `sw2 __IACCreadll (iacc REG)'
21355 Return the full 64-bit value of IACC0. The REG argument is
21356 reserved for future expansion and must be 0.
21358 `sw1 __IACCreadl (iacc REG)'
21359 Return the value of IACC0H if REG is 0 and IACC0L if REG is 1.
21360 Other values of REG are rejected as invalid.
21362 `void __IACCsetll (iacc REG, sw2 X)'
21363 Set the full 64-bit value of IACC0 to X. The REG argument is
21364 reserved for future expansion and must be 0.
21366 `void __IACCsetl (iacc REG, sw1 X)'
21367 Set IACC0H to X if REG is 0 and IACC0L to X if REG is 1. Other
21368 values of REG are rejected as invalid.
21370 `void __data_prefetch0 (const void *X)'
21371 Use the `dcpl' instruction to load the contents of address X into
21374 `void __data_prefetch (const void *X)'
21375 Use the `nldub' instruction to load the contents of address X into
21376 the data cache. The instruction will be issued in slot I1.
21379 File: gcc.info, Node: X86 Built-in Functions, Next: MIPS DSP Built-in Functions, Prev: FR-V Built-in Functions, Up: Target Builtins
21381 5.48.5 X86 Built-in Functions
21382 -----------------------------
21384 These built-in functions are available for the i386 and x86-64 family
21385 of computers, depending on the command-line switches used.
21387 Note that, if you specify command-line switches such as `-msse', the
21388 compiler could use the extended instruction sets even if the built-ins
21389 are not used explicitly in the program. For this reason, applications
21390 which perform runtime CPU detection must compile separate files for each
21391 supported architecture, using the appropriate flags. In particular,
21392 the file containing the CPU detection code should be compiled without
21395 The following machine modes are available for use with MMX built-in
21396 functions (*note Vector Extensions::): `V2SI' for a vector of two
21397 32-bit integers, `V4HI' for a vector of four 16-bit integers, and
21398 `V8QI' for a vector of eight 8-bit integers. Some of the built-in
21399 functions operate on MMX registers as a whole 64-bit entity, these use
21400 `DI' as their mode.
21402 If 3Dnow extensions are enabled, `V2SF' is used as a mode for a vector
21403 of two 32-bit floating point values.
21405 If SSE extensions are enabled, `V4SF' is used for a vector of four
21406 32-bit floating point values. Some instructions use a vector of four
21407 32-bit integers, these use `V4SI'. Finally, some instructions operate
21408 on an entire vector register, interpreting it as a 128-bit integer,
21409 these use mode `TI'.
21411 The following built-in functions are made available by `-mmmx'. All
21412 of them generate the machine instruction that is part of the name.
21414 v8qi __builtin_ia32_paddb (v8qi, v8qi)
21415 v4hi __builtin_ia32_paddw (v4hi, v4hi)
21416 v2si __builtin_ia32_paddd (v2si, v2si)
21417 v8qi __builtin_ia32_psubb (v8qi, v8qi)
21418 v4hi __builtin_ia32_psubw (v4hi, v4hi)
21419 v2si __builtin_ia32_psubd (v2si, v2si)
21420 v8qi __builtin_ia32_paddsb (v8qi, v8qi)
21421 v4hi __builtin_ia32_paddsw (v4hi, v4hi)
21422 v8qi __builtin_ia32_psubsb (v8qi, v8qi)
21423 v4hi __builtin_ia32_psubsw (v4hi, v4hi)
21424 v8qi __builtin_ia32_paddusb (v8qi, v8qi)
21425 v4hi __builtin_ia32_paddusw (v4hi, v4hi)
21426 v8qi __builtin_ia32_psubusb (v8qi, v8qi)
21427 v4hi __builtin_ia32_psubusw (v4hi, v4hi)
21428 v4hi __builtin_ia32_pmullw (v4hi, v4hi)
21429 v4hi __builtin_ia32_pmulhw (v4hi, v4hi)
21430 di __builtin_ia32_pand (di, di)
21431 di __builtin_ia32_pandn (di,di)
21432 di __builtin_ia32_por (di, di)
21433 di __builtin_ia32_pxor (di, di)
21434 v8qi __builtin_ia32_pcmpeqb (v8qi, v8qi)
21435 v4hi __builtin_ia32_pcmpeqw (v4hi, v4hi)
21436 v2si __builtin_ia32_pcmpeqd (v2si, v2si)
21437 v8qi __builtin_ia32_pcmpgtb (v8qi, v8qi)
21438 v4hi __builtin_ia32_pcmpgtw (v4hi, v4hi)
21439 v2si __builtin_ia32_pcmpgtd (v2si, v2si)
21440 v8qi __builtin_ia32_punpckhbw (v8qi, v8qi)
21441 v4hi __builtin_ia32_punpckhwd (v4hi, v4hi)
21442 v2si __builtin_ia32_punpckhdq (v2si, v2si)
21443 v8qi __builtin_ia32_punpcklbw (v8qi, v8qi)
21444 v4hi __builtin_ia32_punpcklwd (v4hi, v4hi)
21445 v2si __builtin_ia32_punpckldq (v2si, v2si)
21446 v8qi __builtin_ia32_packsswb (v4hi, v4hi)
21447 v4hi __builtin_ia32_packssdw (v2si, v2si)
21448 v8qi __builtin_ia32_packuswb (v4hi, v4hi)
21450 The following built-in functions are made available either with
21451 `-msse', or with a combination of `-m3dnow' and `-march=athlon'. All
21452 of them generate the machine instruction that is part of the name.
21454 v4hi __builtin_ia32_pmulhuw (v4hi, v4hi)
21455 v8qi __builtin_ia32_pavgb (v8qi, v8qi)
21456 v4hi __builtin_ia32_pavgw (v4hi, v4hi)
21457 v4hi __builtin_ia32_psadbw (v8qi, v8qi)
21458 v8qi __builtin_ia32_pmaxub (v8qi, v8qi)
21459 v4hi __builtin_ia32_pmaxsw (v4hi, v4hi)
21460 v8qi __builtin_ia32_pminub (v8qi, v8qi)
21461 v4hi __builtin_ia32_pminsw (v4hi, v4hi)
21462 int __builtin_ia32_pextrw (v4hi, int)
21463 v4hi __builtin_ia32_pinsrw (v4hi, int, int)
21464 int __builtin_ia32_pmovmskb (v8qi)
21465 void __builtin_ia32_maskmovq (v8qi, v8qi, char *)
21466 void __builtin_ia32_movntq (di *, di)
21467 void __builtin_ia32_sfence (void)
21469 The following built-in functions are available when `-msse' is used.
21470 All of them generate the machine instruction that is part of the name.
21472 int __builtin_ia32_comieq (v4sf, v4sf)
21473 int __builtin_ia32_comineq (v4sf, v4sf)
21474 int __builtin_ia32_comilt (v4sf, v4sf)
21475 int __builtin_ia32_comile (v4sf, v4sf)
21476 int __builtin_ia32_comigt (v4sf, v4sf)
21477 int __builtin_ia32_comige (v4sf, v4sf)
21478 int __builtin_ia32_ucomieq (v4sf, v4sf)
21479 int __builtin_ia32_ucomineq (v4sf, v4sf)
21480 int __builtin_ia32_ucomilt (v4sf, v4sf)
21481 int __builtin_ia32_ucomile (v4sf, v4sf)
21482 int __builtin_ia32_ucomigt (v4sf, v4sf)
21483 int __builtin_ia32_ucomige (v4sf, v4sf)
21484 v4sf __builtin_ia32_addps (v4sf, v4sf)
21485 v4sf __builtin_ia32_subps (v4sf, v4sf)
21486 v4sf __builtin_ia32_mulps (v4sf, v4sf)
21487 v4sf __builtin_ia32_divps (v4sf, v4sf)
21488 v4sf __builtin_ia32_addss (v4sf, v4sf)
21489 v4sf __builtin_ia32_subss (v4sf, v4sf)
21490 v4sf __builtin_ia32_mulss (v4sf, v4sf)
21491 v4sf __builtin_ia32_divss (v4sf, v4sf)
21492 v4si __builtin_ia32_cmpeqps (v4sf, v4sf)
21493 v4si __builtin_ia32_cmpltps (v4sf, v4sf)
21494 v4si __builtin_ia32_cmpleps (v4sf, v4sf)
21495 v4si __builtin_ia32_cmpgtps (v4sf, v4sf)
21496 v4si __builtin_ia32_cmpgeps (v4sf, v4sf)
21497 v4si __builtin_ia32_cmpunordps (v4sf, v4sf)
21498 v4si __builtin_ia32_cmpneqps (v4sf, v4sf)
21499 v4si __builtin_ia32_cmpnltps (v4sf, v4sf)
21500 v4si __builtin_ia32_cmpnleps (v4sf, v4sf)
21501 v4si __builtin_ia32_cmpngtps (v4sf, v4sf)
21502 v4si __builtin_ia32_cmpngeps (v4sf, v4sf)
21503 v4si __builtin_ia32_cmpordps (v4sf, v4sf)
21504 v4si __builtin_ia32_cmpeqss (v4sf, v4sf)
21505 v4si __builtin_ia32_cmpltss (v4sf, v4sf)
21506 v4si __builtin_ia32_cmpless (v4sf, v4sf)
21507 v4si __builtin_ia32_cmpunordss (v4sf, v4sf)
21508 v4si __builtin_ia32_cmpneqss (v4sf, v4sf)
21509 v4si __builtin_ia32_cmpnlts (v4sf, v4sf)
21510 v4si __builtin_ia32_cmpnless (v4sf, v4sf)
21511 v4si __builtin_ia32_cmpordss (v4sf, v4sf)
21512 v4sf __builtin_ia32_maxps (v4sf, v4sf)
21513 v4sf __builtin_ia32_maxss (v4sf, v4sf)
21514 v4sf __builtin_ia32_minps (v4sf, v4sf)
21515 v4sf __builtin_ia32_minss (v4sf, v4sf)
21516 v4sf __builtin_ia32_andps (v4sf, v4sf)
21517 v4sf __builtin_ia32_andnps (v4sf, v4sf)
21518 v4sf __builtin_ia32_orps (v4sf, v4sf)
21519 v4sf __builtin_ia32_xorps (v4sf, v4sf)
21520 v4sf __builtin_ia32_movss (v4sf, v4sf)
21521 v4sf __builtin_ia32_movhlps (v4sf, v4sf)
21522 v4sf __builtin_ia32_movlhps (v4sf, v4sf)
21523 v4sf __builtin_ia32_unpckhps (v4sf, v4sf)
21524 v4sf __builtin_ia32_unpcklps (v4sf, v4sf)
21525 v4sf __builtin_ia32_cvtpi2ps (v4sf, v2si)
21526 v4sf __builtin_ia32_cvtsi2ss (v4sf, int)
21527 v2si __builtin_ia32_cvtps2pi (v4sf)
21528 int __builtin_ia32_cvtss2si (v4sf)
21529 v2si __builtin_ia32_cvttps2pi (v4sf)
21530 int __builtin_ia32_cvttss2si (v4sf)
21531 v4sf __builtin_ia32_rcpps (v4sf)
21532 v4sf __builtin_ia32_rsqrtps (v4sf)
21533 v4sf __builtin_ia32_sqrtps (v4sf)
21534 v4sf __builtin_ia32_rcpss (v4sf)
21535 v4sf __builtin_ia32_rsqrtss (v4sf)
21536 v4sf __builtin_ia32_sqrtss (v4sf)
21537 v4sf __builtin_ia32_shufps (v4sf, v4sf, int)
21538 void __builtin_ia32_movntps (float *, v4sf)
21539 int __builtin_ia32_movmskps (v4sf)
21541 The following built-in functions are available when `-msse' is used.
21543 `v4sf __builtin_ia32_loadaps (float *)'
21544 Generates the `movaps' machine instruction as a load from memory.
21546 `void __builtin_ia32_storeaps (float *, v4sf)'
21547 Generates the `movaps' machine instruction as a store to memory.
21549 `v4sf __builtin_ia32_loadups (float *)'
21550 Generates the `movups' machine instruction as a load from memory.
21552 `void __builtin_ia32_storeups (float *, v4sf)'
21553 Generates the `movups' machine instruction as a store to memory.
21555 `v4sf __builtin_ia32_loadsss (float *)'
21556 Generates the `movss' machine instruction as a load from memory.
21558 `void __builtin_ia32_storess (float *, v4sf)'
21559 Generates the `movss' machine instruction as a store to memory.
21561 `v4sf __builtin_ia32_loadhps (v4sf, v2si *)'
21562 Generates the `movhps' machine instruction as a load from memory.
21564 `v4sf __builtin_ia32_loadlps (v4sf, v2si *)'
21565 Generates the `movlps' machine instruction as a load from memory
21567 `void __builtin_ia32_storehps (v4sf, v2si *)'
21568 Generates the `movhps' machine instruction as a store to memory.
21570 `void __builtin_ia32_storelps (v4sf, v2si *)'
21571 Generates the `movlps' machine instruction as a store to memory.
21573 The following built-in functions are available when `-msse2' is used.
21574 All of them generate the machine instruction that is part of the name.
21576 int __builtin_ia32_comisdeq (v2df, v2df)
21577 int __builtin_ia32_comisdlt (v2df, v2df)
21578 int __builtin_ia32_comisdle (v2df, v2df)
21579 int __builtin_ia32_comisdgt (v2df, v2df)
21580 int __builtin_ia32_comisdge (v2df, v2df)
21581 int __builtin_ia32_comisdneq (v2df, v2df)
21582 int __builtin_ia32_ucomisdeq (v2df, v2df)
21583 int __builtin_ia32_ucomisdlt (v2df, v2df)
21584 int __builtin_ia32_ucomisdle (v2df, v2df)
21585 int __builtin_ia32_ucomisdgt (v2df, v2df)
21586 int __builtin_ia32_ucomisdge (v2df, v2df)
21587 int __builtin_ia32_ucomisdneq (v2df, v2df)
21588 v2df __builtin_ia32_cmpeqpd (v2df, v2df)
21589 v2df __builtin_ia32_cmpltpd (v2df, v2df)
21590 v2df __builtin_ia32_cmplepd (v2df, v2df)
21591 v2df __builtin_ia32_cmpgtpd (v2df, v2df)
21592 v2df __builtin_ia32_cmpgepd (v2df, v2df)
21593 v2df __builtin_ia32_cmpunordpd (v2df, v2df)
21594 v2df __builtin_ia32_cmpneqpd (v2df, v2df)
21595 v2df __builtin_ia32_cmpnltpd (v2df, v2df)
21596 v2df __builtin_ia32_cmpnlepd (v2df, v2df)
21597 v2df __builtin_ia32_cmpngtpd (v2df, v2df)
21598 v2df __builtin_ia32_cmpngepd (v2df, v2df)
21599 v2df __builtin_ia32_cmpordpd (v2df, v2df)
21600 v2df __builtin_ia32_cmpeqsd (v2df, v2df)
21601 v2df __builtin_ia32_cmpltsd (v2df, v2df)
21602 v2df __builtin_ia32_cmplesd (v2df, v2df)
21603 v2df __builtin_ia32_cmpunordsd (v2df, v2df)
21604 v2df __builtin_ia32_cmpneqsd (v2df, v2df)
21605 v2df __builtin_ia32_cmpnltsd (v2df, v2df)
21606 v2df __builtin_ia32_cmpnlesd (v2df, v2df)
21607 v2df __builtin_ia32_cmpordsd (v2df, v2df)
21608 v2di __builtin_ia32_paddq (v2di, v2di)
21609 v2di __builtin_ia32_psubq (v2di, v2di)
21610 v2df __builtin_ia32_addpd (v2df, v2df)
21611 v2df __builtin_ia32_subpd (v2df, v2df)
21612 v2df __builtin_ia32_mulpd (v2df, v2df)
21613 v2df __builtin_ia32_divpd (v2df, v2df)
21614 v2df __builtin_ia32_addsd (v2df, v2df)
21615 v2df __builtin_ia32_subsd (v2df, v2df)
21616 v2df __builtin_ia32_mulsd (v2df, v2df)
21617 v2df __builtin_ia32_divsd (v2df, v2df)
21618 v2df __builtin_ia32_minpd (v2df, v2df)
21619 v2df __builtin_ia32_maxpd (v2df, v2df)
21620 v2df __builtin_ia32_minsd (v2df, v2df)
21621 v2df __builtin_ia32_maxsd (v2df, v2df)
21622 v2df __builtin_ia32_andpd (v2df, v2df)
21623 v2df __builtin_ia32_andnpd (v2df, v2df)
21624 v2df __builtin_ia32_orpd (v2df, v2df)
21625 v2df __builtin_ia32_xorpd (v2df, v2df)
21626 v2df __builtin_ia32_movsd (v2df, v2df)
21627 v2df __builtin_ia32_unpckhpd (v2df, v2df)
21628 v2df __builtin_ia32_unpcklpd (v2df, v2df)
21629 v16qi __builtin_ia32_paddb128 (v16qi, v16qi)
21630 v8hi __builtin_ia32_paddw128 (v8hi, v8hi)
21631 v4si __builtin_ia32_paddd128 (v4si, v4si)
21632 v2di __builtin_ia32_paddq128 (v2di, v2di)
21633 v16qi __builtin_ia32_psubb128 (v16qi, v16qi)
21634 v8hi __builtin_ia32_psubw128 (v8hi, v8hi)
21635 v4si __builtin_ia32_psubd128 (v4si, v4si)
21636 v2di __builtin_ia32_psubq128 (v2di, v2di)
21637 v8hi __builtin_ia32_pmullw128 (v8hi, v8hi)
21638 v8hi __builtin_ia32_pmulhw128 (v8hi, v8hi)
21639 v2di __builtin_ia32_pand128 (v2di, v2di)
21640 v2di __builtin_ia32_pandn128 (v2di, v2di)
21641 v2di __builtin_ia32_por128 (v2di, v2di)
21642 v2di __builtin_ia32_pxor128 (v2di, v2di)
21643 v16qi __builtin_ia32_pavgb128 (v16qi, v16qi)
21644 v8hi __builtin_ia32_pavgw128 (v8hi, v8hi)
21645 v16qi __builtin_ia32_pcmpeqb128 (v16qi, v16qi)
21646 v8hi __builtin_ia32_pcmpeqw128 (v8hi, v8hi)
21647 v4si __builtin_ia32_pcmpeqd128 (v4si, v4si)
21648 v16qi __builtin_ia32_pcmpgtb128 (v16qi, v16qi)
21649 v8hi __builtin_ia32_pcmpgtw128 (v8hi, v8hi)
21650 v4si __builtin_ia32_pcmpgtd128 (v4si, v4si)
21651 v16qi __builtin_ia32_pmaxub128 (v16qi, v16qi)
21652 v8hi __builtin_ia32_pmaxsw128 (v8hi, v8hi)
21653 v16qi __builtin_ia32_pminub128 (v16qi, v16qi)
21654 v8hi __builtin_ia32_pminsw128 (v8hi, v8hi)
21655 v16qi __builtin_ia32_punpckhbw128 (v16qi, v16qi)
21656 v8hi __builtin_ia32_punpckhwd128 (v8hi, v8hi)
21657 v4si __builtin_ia32_punpckhdq128 (v4si, v4si)
21658 v2di __builtin_ia32_punpckhqdq128 (v2di, v2di)
21659 v16qi __builtin_ia32_punpcklbw128 (v16qi, v16qi)
21660 v8hi __builtin_ia32_punpcklwd128 (v8hi, v8hi)
21661 v4si __builtin_ia32_punpckldq128 (v4si, v4si)
21662 v2di __builtin_ia32_punpcklqdq128 (v2di, v2di)
21663 v16qi __builtin_ia32_packsswb128 (v16qi, v16qi)
21664 v8hi __builtin_ia32_packssdw128 (v8hi, v8hi)
21665 v16qi __builtin_ia32_packuswb128 (v16qi, v16qi)
21666 v8hi __builtin_ia32_pmulhuw128 (v8hi, v8hi)
21667 void __builtin_ia32_maskmovdqu (v16qi, v16qi)
21668 v2df __builtin_ia32_loadupd (double *)
21669 void __builtin_ia32_storeupd (double *, v2df)
21670 v2df __builtin_ia32_loadhpd (v2df, double *)
21671 v2df __builtin_ia32_loadlpd (v2df, double *)
21672 int __builtin_ia32_movmskpd (v2df)
21673 int __builtin_ia32_pmovmskb128 (v16qi)
21674 void __builtin_ia32_movnti (int *, int)
21675 void __builtin_ia32_movntpd (double *, v2df)
21676 void __builtin_ia32_movntdq (v2df *, v2df)
21677 v4si __builtin_ia32_pshufd (v4si, int)
21678 v8hi __builtin_ia32_pshuflw (v8hi, int)
21679 v8hi __builtin_ia32_pshufhw (v8hi, int)
21680 v2di __builtin_ia32_psadbw128 (v16qi, v16qi)
21681 v2df __builtin_ia32_sqrtpd (v2df)
21682 v2df __builtin_ia32_sqrtsd (v2df)
21683 v2df __builtin_ia32_shufpd (v2df, v2df, int)
21684 v2df __builtin_ia32_cvtdq2pd (v4si)
21685 v4sf __builtin_ia32_cvtdq2ps (v4si)
21686 v4si __builtin_ia32_cvtpd2dq (v2df)
21687 v2si __builtin_ia32_cvtpd2pi (v2df)
21688 v4sf __builtin_ia32_cvtpd2ps (v2df)
21689 v4si __builtin_ia32_cvttpd2dq (v2df)
21690 v2si __builtin_ia32_cvttpd2pi (v2df)
21691 v2df __builtin_ia32_cvtpi2pd (v2si)
21692 int __builtin_ia32_cvtsd2si (v2df)
21693 int __builtin_ia32_cvttsd2si (v2df)
21694 long long __builtin_ia32_cvtsd2si64 (v2df)
21695 long long __builtin_ia32_cvttsd2si64 (v2df)
21696 v4si __builtin_ia32_cvtps2dq (v4sf)
21697 v2df __builtin_ia32_cvtps2pd (v4sf)
21698 v4si __builtin_ia32_cvttps2dq (v4sf)
21699 v2df __builtin_ia32_cvtsi2sd (v2df, int)
21700 v2df __builtin_ia32_cvtsi642sd (v2df, long long)
21701 v4sf __builtin_ia32_cvtsd2ss (v4sf, v2df)
21702 v2df __builtin_ia32_cvtss2sd (v2df, v4sf)
21703 void __builtin_ia32_clflush (const void *)
21704 void __builtin_ia32_lfence (void)
21705 void __builtin_ia32_mfence (void)
21706 v16qi __builtin_ia32_loaddqu (const char *)
21707 void __builtin_ia32_storedqu (char *, v16qi)
21708 unsigned long long __builtin_ia32_pmuludq (v2si, v2si)
21709 v2di __builtin_ia32_pmuludq128 (v4si, v4si)
21710 v8hi __builtin_ia32_psllw128 (v8hi, v2di)
21711 v4si __builtin_ia32_pslld128 (v4si, v2di)
21712 v2di __builtin_ia32_psllq128 (v4si, v2di)
21713 v8hi __builtin_ia32_psrlw128 (v8hi, v2di)
21714 v4si __builtin_ia32_psrld128 (v4si, v2di)
21715 v2di __builtin_ia32_psrlq128 (v2di, v2di)
21716 v8hi __builtin_ia32_psraw128 (v8hi, v2di)
21717 v4si __builtin_ia32_psrad128 (v4si, v2di)
21718 v2di __builtin_ia32_pslldqi128 (v2di, int)
21719 v8hi __builtin_ia32_psllwi128 (v8hi, int)
21720 v4si __builtin_ia32_pslldi128 (v4si, int)
21721 v2di __builtin_ia32_psllqi128 (v2di, int)
21722 v2di __builtin_ia32_psrldqi128 (v2di, int)
21723 v8hi __builtin_ia32_psrlwi128 (v8hi, int)
21724 v4si __builtin_ia32_psrldi128 (v4si, int)
21725 v2di __builtin_ia32_psrlqi128 (v2di, int)
21726 v8hi __builtin_ia32_psrawi128 (v8hi, int)
21727 v4si __builtin_ia32_psradi128 (v4si, int)
21728 v4si __builtin_ia32_pmaddwd128 (v8hi, v8hi)
21730 The following built-in functions are available when `-msse3' is used.
21731 All of them generate the machine instruction that is part of the name.
21733 v2df __builtin_ia32_addsubpd (v2df, v2df)
21734 v4sf __builtin_ia32_addsubps (v4sf, v4sf)
21735 v2df __builtin_ia32_haddpd (v2df, v2df)
21736 v4sf __builtin_ia32_haddps (v4sf, v4sf)
21737 v2df __builtin_ia32_hsubpd (v2df, v2df)
21738 v4sf __builtin_ia32_hsubps (v4sf, v4sf)
21739 v16qi __builtin_ia32_lddqu (char const *)
21740 void __builtin_ia32_monitor (void *, unsigned int, unsigned int)
21741 v2df __builtin_ia32_movddup (v2df)
21742 v4sf __builtin_ia32_movshdup (v4sf)
21743 v4sf __builtin_ia32_movsldup (v4sf)
21744 void __builtin_ia32_mwait (unsigned int, unsigned int)
21746 The following built-in functions are available when `-msse3' is used.
21748 `v2df __builtin_ia32_loadddup (double const *)'
21749 Generates the `movddup' machine instruction as a load from memory.
21751 The following built-in functions are available when `-m3dnow' is used.
21752 All of them generate the machine instruction that is part of the name.
21754 void __builtin_ia32_femms (void)
21755 v8qi __builtin_ia32_pavgusb (v8qi, v8qi)
21756 v2si __builtin_ia32_pf2id (v2sf)
21757 v2sf __builtin_ia32_pfacc (v2sf, v2sf)
21758 v2sf __builtin_ia32_pfadd (v2sf, v2sf)
21759 v2si __builtin_ia32_pfcmpeq (v2sf, v2sf)
21760 v2si __builtin_ia32_pfcmpge (v2sf, v2sf)
21761 v2si __builtin_ia32_pfcmpgt (v2sf, v2sf)
21762 v2sf __builtin_ia32_pfmax (v2sf, v2sf)
21763 v2sf __builtin_ia32_pfmin (v2sf, v2sf)
21764 v2sf __builtin_ia32_pfmul (v2sf, v2sf)
21765 v2sf __builtin_ia32_pfrcp (v2sf)
21766 v2sf __builtin_ia32_pfrcpit1 (v2sf, v2sf)
21767 v2sf __builtin_ia32_pfrcpit2 (v2sf, v2sf)
21768 v2sf __builtin_ia32_pfrsqrt (v2sf)
21769 v2sf __builtin_ia32_pfrsqrtit1 (v2sf, v2sf)
21770 v2sf __builtin_ia32_pfsub (v2sf, v2sf)
21771 v2sf __builtin_ia32_pfsubr (v2sf, v2sf)
21772 v2sf __builtin_ia32_pi2fd (v2si)
21773 v4hi __builtin_ia32_pmulhrw (v4hi, v4hi)
21775 The following built-in functions are available when both `-m3dnow' and
21776 `-march=athlon' are used. All of them generate the machine instruction
21777 that is part of the name.
21779 v2si __builtin_ia32_pf2iw (v2sf)
21780 v2sf __builtin_ia32_pfnacc (v2sf, v2sf)
21781 v2sf __builtin_ia32_pfpnacc (v2sf, v2sf)
21782 v2sf __builtin_ia32_pi2fw (v2si)
21783 v2sf __builtin_ia32_pswapdsf (v2sf)
21784 v2si __builtin_ia32_pswapdsi (v2si)
21787 File: gcc.info, Node: MIPS DSP Built-in Functions, Next: MIPS Paired-Single Support, Prev: X86 Built-in Functions, Up: Target Builtins
21789 5.48.6 MIPS DSP Built-in Functions
21790 ----------------------------------
21792 The MIPS DSP Application-Specific Extension (ASE) includes new
21793 instructions that are designed to improve the performance of DSP and
21794 media applications. It provides instructions that operate on packed
21795 8-bit integer data, Q15 fractional data and Q31 fractional data.
21797 GCC supports MIPS DSP operations using both the generic vector
21798 extensions (*note Vector Extensions::) and a collection of
21799 MIPS-specific built-in functions. Both kinds of support are enabled by
21800 the `-mdsp' command-line option.
21802 At present, GCC only provides support for operations on 32-bit
21803 vectors. The vector type associated with 8-bit integer data is usually
21804 called `v4i8' and the vector type associated with Q15 is usually called
21805 `v2q15'. They can be defined in C as follows:
21807 typedef char v4i8 __attribute__ ((vector_size(4)));
21808 typedef short v2q15 __attribute__ ((vector_size(4)));
21810 `v4i8' and `v2q15' values are initialized in the same way as
21811 aggregates. For example:
21813 v4i8 a = {1, 2, 3, 4};
21815 b = (v4i8) {5, 6, 7, 8};
21817 v2q15 c = {0x0fcb, 0x3a75};
21819 d = (v2q15) {0.1234 * 0x1.0p15, 0.4567 * 0x1.0p15};
21821 _Note:_ The CPU's endianness determines the order in which values are
21822 packed. On little-endian targets, the first value is the least
21823 significant and the last value is the most significant. The opposite
21824 order applies to big-endian targets. For example, the code above will
21825 set the lowest byte of `a' to `1' on little-endian targets and `4' on
21826 big-endian targets.
21828 _Note:_ Q15 and Q31 values must be initialized with their integer
21829 representation. As shown in this example, the integer representation
21830 of a Q15 value can be obtained by multiplying the fractional value by
21831 `0x1.0p15'. The equivalent for Q31 values is to multiply by `0x1.0p31'.
21833 The table below lists the `v4i8' and `v2q15' operations for which
21834 hardware support exists. `a' and `b' are `v4i8' values, and `c' and
21835 `d' are `v2q15' values.
21837 C code MIPS instruction
21843 It is easier to describe the DSP built-in functions if we first define
21844 the following types:
21848 typedef long long a64;
21850 `q31' and `i32' are actually the same as `int', but we use `q31' to
21851 indicate a Q31 fractional value and `i32' to indicate a 32-bit integer
21852 value. Similarly, `a64' is the same as `long long', but we use `a64'
21853 to indicate values that will be placed in one of the four DSP
21854 accumulators (`$ac0', `$ac1', `$ac2' or `$ac3').
21856 Also, some built-in functions prefer or require immediate numbers as
21857 parameters, because the corresponding DSP instructions accept both
21858 immediate numbers and register operands, or accept immediate numbers
21859 only. The immediate parameters are listed as follows.
21865 imm0_255: 0 to 255.
21866 imm_n32_31: -32 to 31.
21867 imm_n512_511: -512 to 511.
21869 The following built-in functions map directly to a particular MIPS DSP
21870 instruction. Please refer to the architecture specification for
21871 details on what each instruction does.
21873 v2q15 __builtin_mips_addq_ph (v2q15, v2q15)
21874 v2q15 __builtin_mips_addq_s_ph (v2q15, v2q15)
21875 q31 __builtin_mips_addq_s_w (q31, q31)
21876 v4i8 __builtin_mips_addu_qb (v4i8, v4i8)
21877 v4i8 __builtin_mips_addu_s_qb (v4i8, v4i8)
21878 v2q15 __builtin_mips_subq_ph (v2q15, v2q15)
21879 v2q15 __builtin_mips_subq_s_ph (v2q15, v2q15)
21880 q31 __builtin_mips_subq_s_w (q31, q31)
21881 v4i8 __builtin_mips_subu_qb (v4i8, v4i8)
21882 v4i8 __builtin_mips_subu_s_qb (v4i8, v4i8)
21883 i32 __builtin_mips_addsc (i32, i32)
21884 i32 __builtin_mips_addwc (i32, i32)
21885 i32 __builtin_mips_modsub (i32, i32)
21886 i32 __builtin_mips_raddu_w_qb (v4i8)
21887 v2q15 __builtin_mips_absq_s_ph (v2q15)
21888 q31 __builtin_mips_absq_s_w (q31)
21889 v4i8 __builtin_mips_precrq_qb_ph (v2q15, v2q15)
21890 v2q15 __builtin_mips_precrq_ph_w (q31, q31)
21891 v2q15 __builtin_mips_precrq_rs_ph_w (q31, q31)
21892 v4i8 __builtin_mips_precrqu_s_qb_ph (v2q15, v2q15)
21893 q31 __builtin_mips_preceq_w_phl (v2q15)
21894 q31 __builtin_mips_preceq_w_phr (v2q15)
21895 v2q15 __builtin_mips_precequ_ph_qbl (v4i8)
21896 v2q15 __builtin_mips_precequ_ph_qbr (v4i8)
21897 v2q15 __builtin_mips_precequ_ph_qbla (v4i8)
21898 v2q15 __builtin_mips_precequ_ph_qbra (v4i8)
21899 v2q15 __builtin_mips_preceu_ph_qbl (v4i8)
21900 v2q15 __builtin_mips_preceu_ph_qbr (v4i8)
21901 v2q15 __builtin_mips_preceu_ph_qbla (v4i8)
21902 v2q15 __builtin_mips_preceu_ph_qbra (v4i8)
21903 v4i8 __builtin_mips_shll_qb (v4i8, imm0_7)
21904 v4i8 __builtin_mips_shll_qb (v4i8, i32)
21905 v2q15 __builtin_mips_shll_ph (v2q15, imm0_15)
21906 v2q15 __builtin_mips_shll_ph (v2q15, i32)
21907 v2q15 __builtin_mips_shll_s_ph (v2q15, imm0_15)
21908 v2q15 __builtin_mips_shll_s_ph (v2q15, i32)
21909 q31 __builtin_mips_shll_s_w (q31, imm0_31)
21910 q31 __builtin_mips_shll_s_w (q31, i32)
21911 v4i8 __builtin_mips_shrl_qb (v4i8, imm0_7)
21912 v4i8 __builtin_mips_shrl_qb (v4i8, i32)
21913 v2q15 __builtin_mips_shra_ph (v2q15, imm0_15)
21914 v2q15 __builtin_mips_shra_ph (v2q15, i32)
21915 v2q15 __builtin_mips_shra_r_ph (v2q15, imm0_15)
21916 v2q15 __builtin_mips_shra_r_ph (v2q15, i32)
21917 q31 __builtin_mips_shra_r_w (q31, imm0_31)
21918 q31 __builtin_mips_shra_r_w (q31, i32)
21919 v2q15 __builtin_mips_muleu_s_ph_qbl (v4i8, v2q15)
21920 v2q15 __builtin_mips_muleu_s_ph_qbr (v4i8, v2q15)
21921 v2q15 __builtin_mips_mulq_rs_ph (v2q15, v2q15)
21922 q31 __builtin_mips_muleq_s_w_phl (v2q15, v2q15)
21923 q31 __builtin_mips_muleq_s_w_phr (v2q15, v2q15)
21924 a64 __builtin_mips_dpau_h_qbl (a64, v4i8, v4i8)
21925 a64 __builtin_mips_dpau_h_qbr (a64, v4i8, v4i8)
21926 a64 __builtin_mips_dpsu_h_qbl (a64, v4i8, v4i8)
21927 a64 __builtin_mips_dpsu_h_qbr (a64, v4i8, v4i8)
21928 a64 __builtin_mips_dpaq_s_w_ph (a64, v2q15, v2q15)
21929 a64 __builtin_mips_dpaq_sa_l_w (a64, q31, q31)
21930 a64 __builtin_mips_dpsq_s_w_ph (a64, v2q15, v2q15)
21931 a64 __builtin_mips_dpsq_sa_l_w (a64, q31, q31)
21932 a64 __builtin_mips_mulsaq_s_w_ph (a64, v2q15, v2q15)
21933 a64 __builtin_mips_maq_s_w_phl (a64, v2q15, v2q15)
21934 a64 __builtin_mips_maq_s_w_phr (a64, v2q15, v2q15)
21935 a64 __builtin_mips_maq_sa_w_phl (a64, v2q15, v2q15)
21936 a64 __builtin_mips_maq_sa_w_phr (a64, v2q15, v2q15)
21937 i32 __builtin_mips_bitrev (i32)
21938 i32 __builtin_mips_insv (i32, i32)
21939 v4i8 __builtin_mips_repl_qb (imm0_255)
21940 v4i8 __builtin_mips_repl_qb (i32)
21941 v2q15 __builtin_mips_repl_ph (imm_n512_511)
21942 v2q15 __builtin_mips_repl_ph (i32)
21943 void __builtin_mips_cmpu_eq_qb (v4i8, v4i8)
21944 void __builtin_mips_cmpu_lt_qb (v4i8, v4i8)
21945 void __builtin_mips_cmpu_le_qb (v4i8, v4i8)
21946 i32 __builtin_mips_cmpgu_eq_qb (v4i8, v4i8)
21947 i32 __builtin_mips_cmpgu_lt_qb (v4i8, v4i8)
21948 i32 __builtin_mips_cmpgu_le_qb (v4i8, v4i8)
21949 void __builtin_mips_cmp_eq_ph (v2q15, v2q15)
21950 void __builtin_mips_cmp_lt_ph (v2q15, v2q15)
21951 void __builtin_mips_cmp_le_ph (v2q15, v2q15)
21952 v4i8 __builtin_mips_pick_qb (v4i8, v4i8)
21953 v2q15 __builtin_mips_pick_ph (v2q15, v2q15)
21954 v2q15 __builtin_mips_packrl_ph (v2q15, v2q15)
21955 i32 __builtin_mips_extr_w (a64, imm0_31)
21956 i32 __builtin_mips_extr_w (a64, i32)
21957 i32 __builtin_mips_extr_r_w (a64, imm0_31)
21958 i32 __builtin_mips_extr_s_h (a64, i32)
21959 i32 __builtin_mips_extr_rs_w (a64, imm0_31)
21960 i32 __builtin_mips_extr_rs_w (a64, i32)
21961 i32 __builtin_mips_extr_s_h (a64, imm0_31)
21962 i32 __builtin_mips_extr_r_w (a64, i32)
21963 i32 __builtin_mips_extp (a64, imm0_31)
21964 i32 __builtin_mips_extp (a64, i32)
21965 i32 __builtin_mips_extpdp (a64, imm0_31)
21966 i32 __builtin_mips_extpdp (a64, i32)
21967 a64 __builtin_mips_shilo (a64, imm_n32_31)
21968 a64 __builtin_mips_shilo (a64, i32)
21969 a64 __builtin_mips_mthlip (a64, i32)
21970 void __builtin_mips_wrdsp (i32, imm0_63)
21971 i32 __builtin_mips_rddsp (imm0_63)
21972 i32 __builtin_mips_lbux (void *, i32)
21973 i32 __builtin_mips_lhx (void *, i32)
21974 i32 __builtin_mips_lwx (void *, i32)
21975 i32 __builtin_mips_bposge32 (void)
21978 File: gcc.info, Node: MIPS Paired-Single Support, Next: PowerPC AltiVec Built-in Functions, Prev: MIPS DSP Built-in Functions, Up: Target Builtins
21980 5.48.7 MIPS Paired-Single Support
21981 ---------------------------------
21983 The MIPS64 architecture includes a number of instructions that operate
21984 on pairs of single-precision floating-point values. Each pair is
21985 packed into a 64-bit floating-point register, with one element being
21986 designated the "upper half" and the other being designated the "lower
21989 GCC supports paired-single operations using both the generic vector
21990 extensions (*note Vector Extensions::) and a collection of
21991 MIPS-specific built-in functions. Both kinds of support are enabled by
21992 the `-mpaired-single' command-line option.
21994 The vector type associated with paired-single values is usually called
21995 `v2sf'. It can be defined in C as follows:
21997 typedef float v2sf __attribute__ ((vector_size (8)));
21999 `v2sf' values are initialized in the same way as aggregates. For
22002 v2sf a = {1.5, 9.1};
22007 _Note:_ The CPU's endianness determines which value is stored in the
22008 upper half of a register and which value is stored in the lower half.
22009 On little-endian targets, the first value is the lower one and the
22010 second value is the upper one. The opposite order applies to
22011 big-endian targets. For example, the code above will set the lower
22012 half of `a' to `1.5' on little-endian targets and `9.1' on big-endian
22017 * Paired-Single Arithmetic::
22018 * Paired-Single Built-in Functions::
22019 * MIPS-3D Built-in Functions::
22022 File: gcc.info, Node: Paired-Single Arithmetic, Next: Paired-Single Built-in Functions, Up: MIPS Paired-Single Support
22024 5.48.7.1 Paired-Single Arithmetic
22025 .................................
22027 The table below lists the `v2sf' operations for which hardware support
22028 exists. `a', `b' and `c' are `v2sf' values and `x' is an integral
22031 C code MIPS instruction
22036 `a * b + c' `madd.ps'
22037 `a * b - c' `msub.ps'
22038 `-(a * b + c)' `nmadd.ps'
22039 `-(a * b - c)' `nmsub.ps'
22040 `x ? a : b' `movn.ps'/`movz.ps'
22042 Note that the multiply-accumulate instructions can be disabled using
22043 the command-line option `-mno-fused-madd'.
22046 File: gcc.info, Node: Paired-Single Built-in Functions, Next: MIPS-3D Built-in Functions, Prev: Paired-Single Arithmetic, Up: MIPS Paired-Single Support
22048 5.48.7.2 Paired-Single Built-in Functions
22049 .........................................
22051 The following paired-single functions map directly to a particular MIPS
22052 instruction. Please refer to the architecture specification for
22053 details on what each instruction does.
22055 `v2sf __builtin_mips_pll_ps (v2sf, v2sf)'
22056 Pair lower lower (`pll.ps').
22058 `v2sf __builtin_mips_pul_ps (v2sf, v2sf)'
22059 Pair upper lower (`pul.ps').
22061 `v2sf __builtin_mips_plu_ps (v2sf, v2sf)'
22062 Pair lower upper (`plu.ps').
22064 `v2sf __builtin_mips_puu_ps (v2sf, v2sf)'
22065 Pair upper upper (`puu.ps').
22067 `v2sf __builtin_mips_cvt_ps_s (float, float)'
22068 Convert pair to paired single (`cvt.ps.s').
22070 `float __builtin_mips_cvt_s_pl (v2sf)'
22071 Convert pair lower to single (`cvt.s.pl').
22073 `float __builtin_mips_cvt_s_pu (v2sf)'
22074 Convert pair upper to single (`cvt.s.pu').
22076 `v2sf __builtin_mips_abs_ps (v2sf)'
22077 Absolute value (`abs.ps').
22079 `v2sf __builtin_mips_alnv_ps (v2sf, v2sf, int)'
22080 Align variable (`alnv.ps').
22082 _Note:_ The value of the third parameter must be 0 or 4 modulo 8,
22083 otherwise the result will be unpredictable. Please read the
22084 instruction description for details.
22086 The following multi-instruction functions are also available. In each
22087 case, COND can be any of the 16 floating-point conditions: `f', `un',
22088 `eq', `ueq', `olt', `ult', `ole', `ule', `sf', `ngle', `seq', `ngl',
22089 `lt', `nge', `le' or `ngt'.
22091 `v2sf __builtin_mips_movt_c_COND_ps (v2sf A, v2sf B, v2sf C, v2sf D)'
22092 `v2sf __builtin_mips_movf_c_COND_ps (v2sf A, v2sf B, v2sf C, v2sf D)'
22093 Conditional move based on floating point comparison (`c.COND.ps',
22094 `movt.ps'/`movf.ps').
22096 The `movt' functions return the value X computed by:
22102 The `movf' functions are similar but use `movf.ps' instead of
22105 `int __builtin_mips_upper_c_COND_ps (v2sf A, v2sf B)'
22106 `int __builtin_mips_lower_c_COND_ps (v2sf A, v2sf B)'
22107 Comparison of two paired-single values (`c.COND.ps',
22110 These functions compare A and B using `c.COND.ps' and return
22111 either the upper or lower half of the result. For example:
22114 if (__builtin_mips_upper_c_eq_ps (a, b))
22115 upper_halves_are_equal ();
22117 upper_halves_are_unequal ();
22119 if (__builtin_mips_lower_c_eq_ps (a, b))
22120 lower_halves_are_equal ();
22122 lower_halves_are_unequal ();
22125 File: gcc.info, Node: MIPS-3D Built-in Functions, Prev: Paired-Single Built-in Functions, Up: MIPS Paired-Single Support
22127 5.48.7.3 MIPS-3D Built-in Functions
22128 ...................................
22130 The MIPS-3D Application-Specific Extension (ASE) includes additional
22131 paired-single instructions that are designed to improve the performance
22132 of 3D graphics operations. Support for these instructions is controlled
22133 by the `-mips3d' command-line option.
22135 The functions listed below map directly to a particular MIPS-3D
22136 instruction. Please refer to the architecture specification for more
22137 details on what each instruction does.
22139 `v2sf __builtin_mips_addr_ps (v2sf, v2sf)'
22140 Reduction add (`addr.ps').
22142 `v2sf __builtin_mips_mulr_ps (v2sf, v2sf)'
22143 Reduction multiply (`mulr.ps').
22145 `v2sf __builtin_mips_cvt_pw_ps (v2sf)'
22146 Convert paired single to paired word (`cvt.pw.ps').
22148 `v2sf __builtin_mips_cvt_ps_pw (v2sf)'
22149 Convert paired word to paired single (`cvt.ps.pw').
22151 `float __builtin_mips_recip1_s (float)'
22152 `double __builtin_mips_recip1_d (double)'
22153 `v2sf __builtin_mips_recip1_ps (v2sf)'
22154 Reduced precision reciprocal (sequence step 1) (`recip1.FMT').
22156 `float __builtin_mips_recip2_s (float, float)'
22157 `double __builtin_mips_recip2_d (double, double)'
22158 `v2sf __builtin_mips_recip2_ps (v2sf, v2sf)'
22159 Reduced precision reciprocal (sequence step 2) (`recip2.FMT').
22161 `float __builtin_mips_rsqrt1_s (float)'
22162 `double __builtin_mips_rsqrt1_d (double)'
22163 `v2sf __builtin_mips_rsqrt1_ps (v2sf)'
22164 Reduced precision reciprocal square root (sequence step 1)
22167 `float __builtin_mips_rsqrt2_s (float, float)'
22168 `double __builtin_mips_rsqrt2_d (double, double)'
22169 `v2sf __builtin_mips_rsqrt2_ps (v2sf, v2sf)'
22170 Reduced precision reciprocal square root (sequence step 2)
22173 The following multi-instruction functions are also available. In each
22174 case, COND can be any of the 16 floating-point conditions: `f', `un',
22175 `eq', `ueq', `olt', `ult', `ole', `ule', `sf', `ngle', `seq', `ngl',
22176 `lt', `nge', `le' or `ngt'.
22178 `int __builtin_mips_cabs_COND_s (float A, float B)'
22179 `int __builtin_mips_cabs_COND_d (double A, double B)'
22180 Absolute comparison of two scalar values (`cabs.COND.FMT',
22183 These functions compare A and B using `cabs.COND.s' or
22184 `cabs.COND.d' and return the result as a boolean value. For
22188 if (__builtin_mips_cabs_eq_s (a, b))
22193 `int __builtin_mips_upper_cabs_COND_ps (v2sf A, v2sf B)'
22194 `int __builtin_mips_lower_cabs_COND_ps (v2sf A, v2sf B)'
22195 Absolute comparison of two paired-single values (`cabs.COND.ps',
22198 These functions compare A and B using `cabs.COND.ps' and return
22199 either the upper or lower half of the result. For example:
22202 if (__builtin_mips_upper_cabs_eq_ps (a, b))
22203 upper_halves_are_equal ();
22205 upper_halves_are_unequal ();
22207 if (__builtin_mips_lower_cabs_eq_ps (a, b))
22208 lower_halves_are_equal ();
22210 lower_halves_are_unequal ();
22212 `v2sf __builtin_mips_movt_cabs_COND_ps (v2sf A, v2sf B, v2sf C, v2sf D)'
22213 `v2sf __builtin_mips_movf_cabs_COND_ps (v2sf A, v2sf B, v2sf C, v2sf D)'
22214 Conditional move based on absolute comparison (`cabs.COND.ps',
22215 `movt.ps'/`movf.ps').
22217 The `movt' functions return the value X computed by:
22219 cabs.COND.ps CC,A,B
22223 The `movf' functions are similar but use `movf.ps' instead of
22226 `int __builtin_mips_any_c_COND_ps (v2sf A, v2sf B)'
22227 `int __builtin_mips_all_c_COND_ps (v2sf A, v2sf B)'
22228 `int __builtin_mips_any_cabs_COND_ps (v2sf A, v2sf B)'
22229 `int __builtin_mips_all_cabs_COND_ps (v2sf A, v2sf B)'
22230 Comparison of two paired-single values (`c.COND.ps'/`cabs.COND.ps',
22231 `bc1any2t'/`bc1any2f').
22233 These functions compare A and B using `c.COND.ps' or
22234 `cabs.COND.ps'. The `any' forms return true if either result is
22235 true and the `all' forms return true if both results are true.
22239 if (__builtin_mips_any_c_eq_ps (a, b))
22244 if (__builtin_mips_all_c_eq_ps (a, b))
22249 `int __builtin_mips_any_c_COND_4s (v2sf A, v2sf B, v2sf C, v2sf D)'
22250 `int __builtin_mips_all_c_COND_4s (v2sf A, v2sf B, v2sf C, v2sf D)'
22251 `int __builtin_mips_any_cabs_COND_4s (v2sf A, v2sf B, v2sf C, v2sf D)'
22252 `int __builtin_mips_all_cabs_COND_4s (v2sf A, v2sf B, v2sf C, v2sf D)'
22253 Comparison of four paired-single values
22254 (`c.COND.ps'/`cabs.COND.ps', `bc1any4t'/`bc1any4f').
22256 These functions use `c.COND.ps' or `cabs.COND.ps' to compare A
22257 with B and to compare C with D. The `any' forms return true if
22258 any of the four results are true and the `all' forms return true
22259 if all four results are true. For example:
22262 if (__builtin_mips_any_c_eq_4s (a, b, c, d))
22267 if (__builtin_mips_all_c_eq_4s (a, b, c, d))
22273 File: gcc.info, Node: PowerPC AltiVec Built-in Functions, Next: SPARC VIS Built-in Functions, Prev: MIPS Paired-Single Support, Up: Target Builtins
22275 5.48.8 PowerPC AltiVec Built-in Functions
22276 -----------------------------------------
22278 GCC provides an interface for the PowerPC family of processors to access
22279 the AltiVec operations described in Motorola's AltiVec Programming
22280 Interface Manual. The interface is made available by including
22281 `<altivec.h>' and using `-maltivec' and `-mabi=altivec'. The interface
22282 supports the following vector types.
22284 vector unsigned char
22288 vector unsigned short
22289 vector signed short
22293 vector unsigned int
22298 GCC's implementation of the high-level language interface available
22299 from C and C++ code differs from Motorola's documentation in several
22302 * A vector constant is a list of constant expressions within curly
22305 * A vector initializer requires no cast if the vector constant is of
22306 the same type as the variable it is initializing.
22308 * If `signed' or `unsigned' is omitted, the signedness of the vector
22309 type is the default signedness of the base type. The default
22310 varies depending on the operating system, so a portable program
22311 should always specify the signedness.
22313 * Compiling with `-maltivec' adds keywords `__vector', `__pixel',
22314 and `__bool'. Macros `vector', `pixel', and `bool' are defined in
22315 `<altivec.h>' and can be undefined.
22317 * GCC allows using a `typedef' name as the type specifier for a
22320 * For C, overloaded functions are implemented with macros so the
22321 following does not work:
22323 vec_add ((vector signed int){1, 2, 3, 4}, foo);
22325 Since `vec_add' is a macro, the vector constant in the example is
22326 treated as four separate arguments. Wrap the entire argument in
22327 parentheses for this to work.
22329 _Note:_ Only the `<altivec.h>' interface is supported. Internally,
22330 GCC uses built-in functions to achieve the functionality in the
22331 aforementioned header file, but they are not supported and are subject
22332 to change without notice.
22334 The following interfaces are supported for the generic and specific
22335 AltiVec operations and the AltiVec predicates. In cases where there is
22336 a direct mapping between generic and specific operations, only the
22337 generic names are shown here, although the specific operations can also
22340 Arguments that are documented as `const int' require literal integral
22341 values within the range required for that operation.
22343 vector signed char vec_abs (vector signed char);
22344 vector signed short vec_abs (vector signed short);
22345 vector signed int vec_abs (vector signed int);
22346 vector float vec_abs (vector float);
22348 vector signed char vec_abss (vector signed char);
22349 vector signed short vec_abss (vector signed short);
22350 vector signed int vec_abss (vector signed int);
22352 vector signed char vec_add (vector bool char, vector signed char);
22353 vector signed char vec_add (vector signed char, vector bool char);
22354 vector signed char vec_add (vector signed char, vector signed char);
22355 vector unsigned char vec_add (vector bool char, vector unsigned char);
22356 vector unsigned char vec_add (vector unsigned char, vector bool char);
22357 vector unsigned char vec_add (vector unsigned char,
22358 vector unsigned char);
22359 vector signed short vec_add (vector bool short, vector signed short);
22360 vector signed short vec_add (vector signed short, vector bool short);
22361 vector signed short vec_add (vector signed short, vector signed short);
22362 vector unsigned short vec_add (vector bool short,
22363 vector unsigned short);
22364 vector unsigned short vec_add (vector unsigned short,
22365 vector bool short);
22366 vector unsigned short vec_add (vector unsigned short,
22367 vector unsigned short);
22368 vector signed int vec_add (vector bool int, vector signed int);
22369 vector signed int vec_add (vector signed int, vector bool int);
22370 vector signed int vec_add (vector signed int, vector signed int);
22371 vector unsigned int vec_add (vector bool int, vector unsigned int);
22372 vector unsigned int vec_add (vector unsigned int, vector bool int);
22373 vector unsigned int vec_add (vector unsigned int, vector unsigned int);
22374 vector float vec_add (vector float, vector float);
22376 vector float vec_vaddfp (vector float, vector float);
22378 vector signed int vec_vadduwm (vector bool int, vector signed int);
22379 vector signed int vec_vadduwm (vector signed int, vector bool int);
22380 vector signed int vec_vadduwm (vector signed int, vector signed int);
22381 vector unsigned int vec_vadduwm (vector bool int, vector unsigned int);
22382 vector unsigned int vec_vadduwm (vector unsigned int, vector bool int);
22383 vector unsigned int vec_vadduwm (vector unsigned int,
22384 vector unsigned int);
22386 vector signed short vec_vadduhm (vector bool short,
22387 vector signed short);
22388 vector signed short vec_vadduhm (vector signed short,
22389 vector bool short);
22390 vector signed short vec_vadduhm (vector signed short,
22391 vector signed short);
22392 vector unsigned short vec_vadduhm (vector bool short,
22393 vector unsigned short);
22394 vector unsigned short vec_vadduhm (vector unsigned short,
22395 vector bool short);
22396 vector unsigned short vec_vadduhm (vector unsigned short,
22397 vector unsigned short);
22399 vector signed char vec_vaddubm (vector bool char, vector signed char);
22400 vector signed char vec_vaddubm (vector signed char, vector bool char);
22401 vector signed char vec_vaddubm (vector signed char, vector signed char);
22402 vector unsigned char vec_vaddubm (vector bool char,
22403 vector unsigned char);
22404 vector unsigned char vec_vaddubm (vector unsigned char,
22406 vector unsigned char vec_vaddubm (vector unsigned char,
22407 vector unsigned char);
22409 vector unsigned int vec_addc (vector unsigned int, vector unsigned int);
22411 vector unsigned char vec_adds (vector bool char, vector unsigned char);
22412 vector unsigned char vec_adds (vector unsigned char, vector bool char);
22413 vector unsigned char vec_adds (vector unsigned char,
22414 vector unsigned char);
22415 vector signed char vec_adds (vector bool char, vector signed char);
22416 vector signed char vec_adds (vector signed char, vector bool char);
22417 vector signed char vec_adds (vector signed char, vector signed char);
22418 vector unsigned short vec_adds (vector bool short,
22419 vector unsigned short);
22420 vector unsigned short vec_adds (vector unsigned short,
22421 vector bool short);
22422 vector unsigned short vec_adds (vector unsigned short,
22423 vector unsigned short);
22424 vector signed short vec_adds (vector bool short, vector signed short);
22425 vector signed short vec_adds (vector signed short, vector bool short);
22426 vector signed short vec_adds (vector signed short, vector signed short);
22427 vector unsigned int vec_adds (vector bool int, vector unsigned int);
22428 vector unsigned int vec_adds (vector unsigned int, vector bool int);
22429 vector unsigned int vec_adds (vector unsigned int, vector unsigned int);
22430 vector signed int vec_adds (vector bool int, vector signed int);
22431 vector signed int vec_adds (vector signed int, vector bool int);
22432 vector signed int vec_adds (vector signed int, vector signed int);
22434 vector signed int vec_vaddsws (vector bool int, vector signed int);
22435 vector signed int vec_vaddsws (vector signed int, vector bool int);
22436 vector signed int vec_vaddsws (vector signed int, vector signed int);
22438 vector unsigned int vec_vadduws (vector bool int, vector unsigned int);
22439 vector unsigned int vec_vadduws (vector unsigned int, vector bool int);
22440 vector unsigned int vec_vadduws (vector unsigned int,
22441 vector unsigned int);
22443 vector signed short vec_vaddshs (vector bool short,
22444 vector signed short);
22445 vector signed short vec_vaddshs (vector signed short,
22446 vector bool short);
22447 vector signed short vec_vaddshs (vector signed short,
22448 vector signed short);
22450 vector unsigned short vec_vadduhs (vector bool short,
22451 vector unsigned short);
22452 vector unsigned short vec_vadduhs (vector unsigned short,
22453 vector bool short);
22454 vector unsigned short vec_vadduhs (vector unsigned short,
22455 vector unsigned short);
22457 vector signed char vec_vaddsbs (vector bool char, vector signed char);
22458 vector signed char vec_vaddsbs (vector signed char, vector bool char);
22459 vector signed char vec_vaddsbs (vector signed char, vector signed char);
22461 vector unsigned char vec_vaddubs (vector bool char,
22462 vector unsigned char);
22463 vector unsigned char vec_vaddubs (vector unsigned char,
22465 vector unsigned char vec_vaddubs (vector unsigned char,
22466 vector unsigned char);
22468 vector float vec_and (vector float, vector float);
22469 vector float vec_and (vector float, vector bool int);
22470 vector float vec_and (vector bool int, vector float);
22471 vector bool int vec_and (vector bool int, vector bool int);
22472 vector signed int vec_and (vector bool int, vector signed int);
22473 vector signed int vec_and (vector signed int, vector bool int);
22474 vector signed int vec_and (vector signed int, vector signed int);
22475 vector unsigned int vec_and (vector bool int, vector unsigned int);
22476 vector unsigned int vec_and (vector unsigned int, vector bool int);
22477 vector unsigned int vec_and (vector unsigned int, vector unsigned int);
22478 vector bool short vec_and (vector bool short, vector bool short);
22479 vector signed short vec_and (vector bool short, vector signed short);
22480 vector signed short vec_and (vector signed short, vector bool short);
22481 vector signed short vec_and (vector signed short, vector signed short);
22482 vector unsigned short vec_and (vector bool short,
22483 vector unsigned short);
22484 vector unsigned short vec_and (vector unsigned short,
22485 vector bool short);
22486 vector unsigned short vec_and (vector unsigned short,
22487 vector unsigned short);
22488 vector signed char vec_and (vector bool char, vector signed char);
22489 vector bool char vec_and (vector bool char, vector bool char);
22490 vector signed char vec_and (vector signed char, vector bool char);
22491 vector signed char vec_and (vector signed char, vector signed char);
22492 vector unsigned char vec_and (vector bool char, vector unsigned char);
22493 vector unsigned char vec_and (vector unsigned char, vector bool char);
22494 vector unsigned char vec_and (vector unsigned char,
22495 vector unsigned char);
22497 vector float vec_andc (vector float, vector float);
22498 vector float vec_andc (vector float, vector bool int);
22499 vector float vec_andc (vector bool int, vector float);
22500 vector bool int vec_andc (vector bool int, vector bool int);
22501 vector signed int vec_andc (vector bool int, vector signed int);
22502 vector signed int vec_andc (vector signed int, vector bool int);
22503 vector signed int vec_andc (vector signed int, vector signed int);
22504 vector unsigned int vec_andc (vector bool int, vector unsigned int);
22505 vector unsigned int vec_andc (vector unsigned int, vector bool int);
22506 vector unsigned int vec_andc (vector unsigned int, vector unsigned int);
22507 vector bool short vec_andc (vector bool short, vector bool short);
22508 vector signed short vec_andc (vector bool short, vector signed short);
22509 vector signed short vec_andc (vector signed short, vector bool short);
22510 vector signed short vec_andc (vector signed short, vector signed short);
22511 vector unsigned short vec_andc (vector bool short,
22512 vector unsigned short);
22513 vector unsigned short vec_andc (vector unsigned short,
22514 vector bool short);
22515 vector unsigned short vec_andc (vector unsigned short,
22516 vector unsigned short);
22517 vector signed char vec_andc (vector bool char, vector signed char);
22518 vector bool char vec_andc (vector bool char, vector bool char);
22519 vector signed char vec_andc (vector signed char, vector bool char);
22520 vector signed char vec_andc (vector signed char, vector signed char);
22521 vector unsigned char vec_andc (vector bool char, vector unsigned char);
22522 vector unsigned char vec_andc (vector unsigned char, vector bool char);
22523 vector unsigned char vec_andc (vector unsigned char,
22524 vector unsigned char);
22526 vector unsigned char vec_avg (vector unsigned char,
22527 vector unsigned char);
22528 vector signed char vec_avg (vector signed char, vector signed char);
22529 vector unsigned short vec_avg (vector unsigned short,
22530 vector unsigned short);
22531 vector signed short vec_avg (vector signed short, vector signed short);
22532 vector unsigned int vec_avg (vector unsigned int, vector unsigned int);
22533 vector signed int vec_avg (vector signed int, vector signed int);
22535 vector signed int vec_vavgsw (vector signed int, vector signed int);
22537 vector unsigned int vec_vavguw (vector unsigned int,
22538 vector unsigned int);
22540 vector signed short vec_vavgsh (vector signed short,
22541 vector signed short);
22543 vector unsigned short vec_vavguh (vector unsigned short,
22544 vector unsigned short);
22546 vector signed char vec_vavgsb (vector signed char, vector signed char);
22548 vector unsigned char vec_vavgub (vector unsigned char,
22549 vector unsigned char);
22551 vector float vec_ceil (vector float);
22553 vector signed int vec_cmpb (vector float, vector float);
22555 vector bool char vec_cmpeq (vector signed char, vector signed char);
22556 vector bool char vec_cmpeq (vector unsigned char, vector unsigned char);
22557 vector bool short vec_cmpeq (vector signed short, vector signed short);
22558 vector bool short vec_cmpeq (vector unsigned short,
22559 vector unsigned short);
22560 vector bool int vec_cmpeq (vector signed int, vector signed int);
22561 vector bool int vec_cmpeq (vector unsigned int, vector unsigned int);
22562 vector bool int vec_cmpeq (vector float, vector float);
22564 vector bool int vec_vcmpeqfp (vector float, vector float);
22566 vector bool int vec_vcmpequw (vector signed int, vector signed int);
22567 vector bool int vec_vcmpequw (vector unsigned int, vector unsigned int);
22569 vector bool short vec_vcmpequh (vector signed short,
22570 vector signed short);
22571 vector bool short vec_vcmpequh (vector unsigned short,
22572 vector unsigned short);
22574 vector bool char vec_vcmpequb (vector signed char, vector signed char);
22575 vector bool char vec_vcmpequb (vector unsigned char,
22576 vector unsigned char);
22578 vector bool int vec_cmpge (vector float, vector float);
22580 vector bool char vec_cmpgt (vector unsigned char, vector unsigned char);
22581 vector bool char vec_cmpgt (vector signed char, vector signed char);
22582 vector bool short vec_cmpgt (vector unsigned short,
22583 vector unsigned short);
22584 vector bool short vec_cmpgt (vector signed short, vector signed short);
22585 vector bool int vec_cmpgt (vector unsigned int, vector unsigned int);
22586 vector bool int vec_cmpgt (vector signed int, vector signed int);
22587 vector bool int vec_cmpgt (vector float, vector float);
22589 vector bool int vec_vcmpgtfp (vector float, vector float);
22591 vector bool int vec_vcmpgtsw (vector signed int, vector signed int);
22593 vector bool int vec_vcmpgtuw (vector unsigned int, vector unsigned int);
22595 vector bool short vec_vcmpgtsh (vector signed short,
22596 vector signed short);
22598 vector bool short vec_vcmpgtuh (vector unsigned short,
22599 vector unsigned short);
22601 vector bool char vec_vcmpgtsb (vector signed char, vector signed char);
22603 vector bool char vec_vcmpgtub (vector unsigned char,
22604 vector unsigned char);
22606 vector bool int vec_cmple (vector float, vector float);
22608 vector bool char vec_cmplt (vector unsigned char, vector unsigned char);
22609 vector bool char vec_cmplt (vector signed char, vector signed char);
22610 vector bool short vec_cmplt (vector unsigned short,
22611 vector unsigned short);
22612 vector bool short vec_cmplt (vector signed short, vector signed short);
22613 vector bool int vec_cmplt (vector unsigned int, vector unsigned int);
22614 vector bool int vec_cmplt (vector signed int, vector signed int);
22615 vector bool int vec_cmplt (vector float, vector float);
22617 vector float vec_ctf (vector unsigned int, const int);
22618 vector float vec_ctf (vector signed int, const int);
22620 vector float vec_vcfsx (vector signed int, const int);
22622 vector float vec_vcfux (vector unsigned int, const int);
22624 vector signed int vec_cts (vector float, const int);
22626 vector unsigned int vec_ctu (vector float, const int);
22628 void vec_dss (const int);
22630 void vec_dssall (void);
22632 void vec_dst (const vector unsigned char *, int, const int);
22633 void vec_dst (const vector signed char *, int, const int);
22634 void vec_dst (const vector bool char *, int, const int);
22635 void vec_dst (const vector unsigned short *, int, const int);
22636 void vec_dst (const vector signed short *, int, const int);
22637 void vec_dst (const vector bool short *, int, const int);
22638 void vec_dst (const vector pixel *, int, const int);
22639 void vec_dst (const vector unsigned int *, int, const int);
22640 void vec_dst (const vector signed int *, int, const int);
22641 void vec_dst (const vector bool int *, int, const int);
22642 void vec_dst (const vector float *, int, const int);
22643 void vec_dst (const unsigned char *, int, const int);
22644 void vec_dst (const signed char *, int, const int);
22645 void vec_dst (const unsigned short *, int, const int);
22646 void vec_dst (const short *, int, const int);
22647 void vec_dst (const unsigned int *, int, const int);
22648 void vec_dst (const int *, int, const int);
22649 void vec_dst (const unsigned long *, int, const int);
22650 void vec_dst (const long *, int, const int);
22651 void vec_dst (const float *, int, const int);
22653 void vec_dstst (const vector unsigned char *, int, const int);
22654 void vec_dstst (const vector signed char *, int, const int);
22655 void vec_dstst (const vector bool char *, int, const int);
22656 void vec_dstst (const vector unsigned short *, int, const int);
22657 void vec_dstst (const vector signed short *, int, const int);
22658 void vec_dstst (const vector bool short *, int, const int);
22659 void vec_dstst (const vector pixel *, int, const int);
22660 void vec_dstst (const vector unsigned int *, int, const int);
22661 void vec_dstst (const vector signed int *, int, const int);
22662 void vec_dstst (const vector bool int *, int, const int);
22663 void vec_dstst (const vector float *, int, const int);
22664 void vec_dstst (const unsigned char *, int, const int);
22665 void vec_dstst (const signed char *, int, const int);
22666 void vec_dstst (const unsigned short *, int, const int);
22667 void vec_dstst (const short *, int, const int);
22668 void vec_dstst (const unsigned int *, int, const int);
22669 void vec_dstst (const int *, int, const int);
22670 void vec_dstst (const unsigned long *, int, const int);
22671 void vec_dstst (const long *, int, const int);
22672 void vec_dstst (const float *, int, const int);
22674 void vec_dststt (const vector unsigned char *, int, const int);
22675 void vec_dststt (const vector signed char *, int, const int);
22676 void vec_dststt (const vector bool char *, int, const int);
22677 void vec_dststt (const vector unsigned short *, int, const int);
22678 void vec_dststt (const vector signed short *, int, const int);
22679 void vec_dststt (const vector bool short *, int, const int);
22680 void vec_dststt (const vector pixel *, int, const int);
22681 void vec_dststt (const vector unsigned int *, int, const int);
22682 void vec_dststt (const vector signed int *, int, const int);
22683 void vec_dststt (const vector bool int *, int, const int);
22684 void vec_dststt (const vector float *, int, const int);
22685 void vec_dststt (const unsigned char *, int, const int);
22686 void vec_dststt (const signed char *, int, const int);
22687 void vec_dststt (const unsigned short *, int, const int);
22688 void vec_dststt (const short *, int, const int);
22689 void vec_dststt (const unsigned int *, int, const int);
22690 void vec_dststt (const int *, int, const int);
22691 void vec_dststt (const unsigned long *, int, const int);
22692 void vec_dststt (const long *, int, const int);
22693 void vec_dststt (const float *, int, const int);
22695 void vec_dstt (const vector unsigned char *, int, const int);
22696 void vec_dstt (const vector signed char *, int, const int);
22697 void vec_dstt (const vector bool char *, int, const int);
22698 void vec_dstt (const vector unsigned short *, int, const int);
22699 void vec_dstt (const vector signed short *, int, const int);
22700 void vec_dstt (const vector bool short *, int, const int);
22701 void vec_dstt (const vector pixel *, int, const int);
22702 void vec_dstt (const vector unsigned int *, int, const int);
22703 void vec_dstt (const vector signed int *, int, const int);
22704 void vec_dstt (const vector bool int *, int, const int);
22705 void vec_dstt (const vector float *, int, const int);
22706 void vec_dstt (const unsigned char *, int, const int);
22707 void vec_dstt (const signed char *, int, const int);
22708 void vec_dstt (const unsigned short *, int, const int);
22709 void vec_dstt (const short *, int, const int);
22710 void vec_dstt (const unsigned int *, int, const int);
22711 void vec_dstt (const int *, int, const int);
22712 void vec_dstt (const unsigned long *, int, const int);
22713 void vec_dstt (const long *, int, const int);
22714 void vec_dstt (const float *, int, const int);
22716 vector float vec_expte (vector float);
22718 vector float vec_floor (vector float);
22720 vector float vec_ld (int, const vector float *);
22721 vector float vec_ld (int, const float *);
22722 vector bool int vec_ld (int, const vector bool int *);
22723 vector signed int vec_ld (int, const vector signed int *);
22724 vector signed int vec_ld (int, const int *);
22725 vector signed int vec_ld (int, const long *);
22726 vector unsigned int vec_ld (int, const vector unsigned int *);
22727 vector unsigned int vec_ld (int, const unsigned int *);
22728 vector unsigned int vec_ld (int, const unsigned long *);
22729 vector bool short vec_ld (int, const vector bool short *);
22730 vector pixel vec_ld (int, const vector pixel *);
22731 vector signed short vec_ld (int, const vector signed short *);
22732 vector signed short vec_ld (int, const short *);
22733 vector unsigned short vec_ld (int, const vector unsigned short *);
22734 vector unsigned short vec_ld (int, const unsigned short *);
22735 vector bool char vec_ld (int, const vector bool char *);
22736 vector signed char vec_ld (int, const vector signed char *);
22737 vector signed char vec_ld (int, const signed char *);
22738 vector unsigned char vec_ld (int, const vector unsigned char *);
22739 vector unsigned char vec_ld (int, const unsigned char *);
22741 vector signed char vec_lde (int, const signed char *);
22742 vector unsigned char vec_lde (int, const unsigned char *);
22743 vector signed short vec_lde (int, const short *);
22744 vector unsigned short vec_lde (int, const unsigned short *);
22745 vector float vec_lde (int, const float *);
22746 vector signed int vec_lde (int, const int *);
22747 vector unsigned int vec_lde (int, const unsigned int *);
22748 vector signed int vec_lde (int, const long *);
22749 vector unsigned int vec_lde (int, const unsigned long *);
22751 vector float vec_lvewx (int, float *);
22752 vector signed int vec_lvewx (int, int *);
22753 vector unsigned int vec_lvewx (int, unsigned int *);
22754 vector signed int vec_lvewx (int, long *);
22755 vector unsigned int vec_lvewx (int, unsigned long *);
22757 vector signed short vec_lvehx (int, short *);
22758 vector unsigned short vec_lvehx (int, unsigned short *);
22760 vector signed char vec_lvebx (int, char *);
22761 vector unsigned char vec_lvebx (int, unsigned char *);
22763 vector float vec_ldl (int, const vector float *);
22764 vector float vec_ldl (int, const float *);
22765 vector bool int vec_ldl (int, const vector bool int *);
22766 vector signed int vec_ldl (int, const vector signed int *);
22767 vector signed int vec_ldl (int, const int *);
22768 vector signed int vec_ldl (int, const long *);
22769 vector unsigned int vec_ldl (int, const vector unsigned int *);
22770 vector unsigned int vec_ldl (int, const unsigned int *);
22771 vector unsigned int vec_ldl (int, const unsigned long *);
22772 vector bool short vec_ldl (int, const vector bool short *);
22773 vector pixel vec_ldl (int, const vector pixel *);
22774 vector signed short vec_ldl (int, const vector signed short *);
22775 vector signed short vec_ldl (int, const short *);
22776 vector unsigned short vec_ldl (int, const vector unsigned short *);
22777 vector unsigned short vec_ldl (int, const unsigned short *);
22778 vector bool char vec_ldl (int, const vector bool char *);
22779 vector signed char vec_ldl (int, const vector signed char *);
22780 vector signed char vec_ldl (int, const signed char *);
22781 vector unsigned char vec_ldl (int, const vector unsigned char *);
22782 vector unsigned char vec_ldl (int, const unsigned char *);
22784 vector float vec_loge (vector float);
22786 vector unsigned char vec_lvsl (int, const volatile unsigned char *);
22787 vector unsigned char vec_lvsl (int, const volatile signed char *);
22788 vector unsigned char vec_lvsl (int, const volatile unsigned short *);
22789 vector unsigned char vec_lvsl (int, const volatile short *);
22790 vector unsigned char vec_lvsl (int, const volatile unsigned int *);
22791 vector unsigned char vec_lvsl (int, const volatile int *);
22792 vector unsigned char vec_lvsl (int, const volatile unsigned long *);
22793 vector unsigned char vec_lvsl (int, const volatile long *);
22794 vector unsigned char vec_lvsl (int, const volatile float *);
22796 vector unsigned char vec_lvsr (int, const volatile unsigned char *);
22797 vector unsigned char vec_lvsr (int, const volatile signed char *);
22798 vector unsigned char vec_lvsr (int, const volatile unsigned short *);
22799 vector unsigned char vec_lvsr (int, const volatile short *);
22800 vector unsigned char vec_lvsr (int, const volatile unsigned int *);
22801 vector unsigned char vec_lvsr (int, const volatile int *);
22802 vector unsigned char vec_lvsr (int, const volatile unsigned long *);
22803 vector unsigned char vec_lvsr (int, const volatile long *);
22804 vector unsigned char vec_lvsr (int, const volatile float *);
22806 vector float vec_madd (vector float, vector float, vector float);
22808 vector signed short vec_madds (vector signed short,
22809 vector signed short,
22810 vector signed short);
22812 vector unsigned char vec_max (vector bool char, vector unsigned char);
22813 vector unsigned char vec_max (vector unsigned char, vector bool char);
22814 vector unsigned char vec_max (vector unsigned char,
22815 vector unsigned char);
22816 vector signed char vec_max (vector bool char, vector signed char);
22817 vector signed char vec_max (vector signed char, vector bool char);
22818 vector signed char vec_max (vector signed char, vector signed char);
22819 vector unsigned short vec_max (vector bool short,
22820 vector unsigned short);
22821 vector unsigned short vec_max (vector unsigned short,
22822 vector bool short);
22823 vector unsigned short vec_max (vector unsigned short,
22824 vector unsigned short);
22825 vector signed short vec_max (vector bool short, vector signed short);
22826 vector signed short vec_max (vector signed short, vector bool short);
22827 vector signed short vec_max (vector signed short, vector signed short);
22828 vector unsigned int vec_max (vector bool int, vector unsigned int);
22829 vector unsigned int vec_max (vector unsigned int, vector bool int);
22830 vector unsigned int vec_max (vector unsigned int, vector unsigned int);
22831 vector signed int vec_max (vector bool int, vector signed int);
22832 vector signed int vec_max (vector signed int, vector bool int);
22833 vector signed int vec_max (vector signed int, vector signed int);
22834 vector float vec_max (vector float, vector float);
22836 vector float vec_vmaxfp (vector float, vector float);
22838 vector signed int vec_vmaxsw (vector bool int, vector signed int);
22839 vector signed int vec_vmaxsw (vector signed int, vector bool int);
22840 vector signed int vec_vmaxsw (vector signed int, vector signed int);
22842 vector unsigned int vec_vmaxuw (vector bool int, vector unsigned int);
22843 vector unsigned int vec_vmaxuw (vector unsigned int, vector bool int);
22844 vector unsigned int vec_vmaxuw (vector unsigned int,
22845 vector unsigned int);
22847 vector signed short vec_vmaxsh (vector bool short, vector signed short);
22848 vector signed short vec_vmaxsh (vector signed short, vector bool short);
22849 vector signed short vec_vmaxsh (vector signed short,
22850 vector signed short);
22852 vector unsigned short vec_vmaxuh (vector bool short,
22853 vector unsigned short);
22854 vector unsigned short vec_vmaxuh (vector unsigned short,
22855 vector bool short);
22856 vector unsigned short vec_vmaxuh (vector unsigned short,
22857 vector unsigned short);
22859 vector signed char vec_vmaxsb (vector bool char, vector signed char);
22860 vector signed char vec_vmaxsb (vector signed char, vector bool char);
22861 vector signed char vec_vmaxsb (vector signed char, vector signed char);
22863 vector unsigned char vec_vmaxub (vector bool char,
22864 vector unsigned char);
22865 vector unsigned char vec_vmaxub (vector unsigned char,
22867 vector unsigned char vec_vmaxub (vector unsigned char,
22868 vector unsigned char);
22870 vector bool char vec_mergeh (vector bool char, vector bool char);
22871 vector signed char vec_mergeh (vector signed char, vector signed char);
22872 vector unsigned char vec_mergeh (vector unsigned char,
22873 vector unsigned char);
22874 vector bool short vec_mergeh (vector bool short, vector bool short);
22875 vector pixel vec_mergeh (vector pixel, vector pixel);
22876 vector signed short vec_mergeh (vector signed short,
22877 vector signed short);
22878 vector unsigned short vec_mergeh (vector unsigned short,
22879 vector unsigned short);
22880 vector float vec_mergeh (vector float, vector float);
22881 vector bool int vec_mergeh (vector bool int, vector bool int);
22882 vector signed int vec_mergeh (vector signed int, vector signed int);
22883 vector unsigned int vec_mergeh (vector unsigned int,
22884 vector unsigned int);
22886 vector float vec_vmrghw (vector float, vector float);
22887 vector bool int vec_vmrghw (vector bool int, vector bool int);
22888 vector signed int vec_vmrghw (vector signed int, vector signed int);
22889 vector unsigned int vec_vmrghw (vector unsigned int,
22890 vector unsigned int);
22892 vector bool short vec_vmrghh (vector bool short, vector bool short);
22893 vector signed short vec_vmrghh (vector signed short,
22894 vector signed short);
22895 vector unsigned short vec_vmrghh (vector unsigned short,
22896 vector unsigned short);
22897 vector pixel vec_vmrghh (vector pixel, vector pixel);
22899 vector bool char vec_vmrghb (vector bool char, vector bool char);
22900 vector signed char vec_vmrghb (vector signed char, vector signed char);
22901 vector unsigned char vec_vmrghb (vector unsigned char,
22902 vector unsigned char);
22904 vector bool char vec_mergel (vector bool char, vector bool char);
22905 vector signed char vec_mergel (vector signed char, vector signed char);
22906 vector unsigned char vec_mergel (vector unsigned char,
22907 vector unsigned char);
22908 vector bool short vec_mergel (vector bool short, vector bool short);
22909 vector pixel vec_mergel (vector pixel, vector pixel);
22910 vector signed short vec_mergel (vector signed short,
22911 vector signed short);
22912 vector unsigned short vec_mergel (vector unsigned short,
22913 vector unsigned short);
22914 vector float vec_mergel (vector float, vector float);
22915 vector bool int vec_mergel (vector bool int, vector bool int);
22916 vector signed int vec_mergel (vector signed int, vector signed int);
22917 vector unsigned int vec_mergel (vector unsigned int,
22918 vector unsigned int);
22920 vector float vec_vmrglw (vector float, vector float);
22921 vector signed int vec_vmrglw (vector signed int, vector signed int);
22922 vector unsigned int vec_vmrglw (vector unsigned int,
22923 vector unsigned int);
22924 vector bool int vec_vmrglw (vector bool int, vector bool int);
22926 vector bool short vec_vmrglh (vector bool short, vector bool short);
22927 vector signed short vec_vmrglh (vector signed short,
22928 vector signed short);
22929 vector unsigned short vec_vmrglh (vector unsigned short,
22930 vector unsigned short);
22931 vector pixel vec_vmrglh (vector pixel, vector pixel);
22933 vector bool char vec_vmrglb (vector bool char, vector bool char);
22934 vector signed char vec_vmrglb (vector signed char, vector signed char);
22935 vector unsigned char vec_vmrglb (vector unsigned char,
22936 vector unsigned char);
22938 vector unsigned short vec_mfvscr (void);
22940 vector unsigned char vec_min (vector bool char, vector unsigned char);
22941 vector unsigned char vec_min (vector unsigned char, vector bool char);
22942 vector unsigned char vec_min (vector unsigned char,
22943 vector unsigned char);
22944 vector signed char vec_min (vector bool char, vector signed char);
22945 vector signed char vec_min (vector signed char, vector bool char);
22946 vector signed char vec_min (vector signed char, vector signed char);
22947 vector unsigned short vec_min (vector bool short,
22948 vector unsigned short);
22949 vector unsigned short vec_min (vector unsigned short,
22950 vector bool short);
22951 vector unsigned short vec_min (vector unsigned short,
22952 vector unsigned short);
22953 vector signed short vec_min (vector bool short, vector signed short);
22954 vector signed short vec_min (vector signed short, vector bool short);
22955 vector signed short vec_min (vector signed short, vector signed short);
22956 vector unsigned int vec_min (vector bool int, vector unsigned int);
22957 vector unsigned int vec_min (vector unsigned int, vector bool int);
22958 vector unsigned int vec_min (vector unsigned int, vector unsigned int);
22959 vector signed int vec_min (vector bool int, vector signed int);
22960 vector signed int vec_min (vector signed int, vector bool int);
22961 vector signed int vec_min (vector signed int, vector signed int);
22962 vector float vec_min (vector float, vector float);
22964 vector float vec_vminfp (vector float, vector float);
22966 vector signed int vec_vminsw (vector bool int, vector signed int);
22967 vector signed int vec_vminsw (vector signed int, vector bool int);
22968 vector signed int vec_vminsw (vector signed int, vector signed int);
22970 vector unsigned int vec_vminuw (vector bool int, vector unsigned int);
22971 vector unsigned int vec_vminuw (vector unsigned int, vector bool int);
22972 vector unsigned int vec_vminuw (vector unsigned int,
22973 vector unsigned int);
22975 vector signed short vec_vminsh (vector bool short, vector signed short);
22976 vector signed short vec_vminsh (vector signed short, vector bool short);
22977 vector signed short vec_vminsh (vector signed short,
22978 vector signed short);
22980 vector unsigned short vec_vminuh (vector bool short,
22981 vector unsigned short);
22982 vector unsigned short vec_vminuh (vector unsigned short,
22983 vector bool short);
22984 vector unsigned short vec_vminuh (vector unsigned short,
22985 vector unsigned short);
22987 vector signed char vec_vminsb (vector bool char, vector signed char);
22988 vector signed char vec_vminsb (vector signed char, vector bool char);
22989 vector signed char vec_vminsb (vector signed char, vector signed char);
22991 vector unsigned char vec_vminub (vector bool char,
22992 vector unsigned char);
22993 vector unsigned char vec_vminub (vector unsigned char,
22995 vector unsigned char vec_vminub (vector unsigned char,
22996 vector unsigned char);
22998 vector signed short vec_mladd (vector signed short,
22999 vector signed short,
23000 vector signed short);
23001 vector signed short vec_mladd (vector signed short,
23002 vector unsigned short,
23003 vector unsigned short);
23004 vector signed short vec_mladd (vector unsigned short,
23005 vector signed short,
23006 vector signed short);
23007 vector unsigned short vec_mladd (vector unsigned short,
23008 vector unsigned short,
23009 vector unsigned short);
23011 vector signed short vec_mradds (vector signed short,
23012 vector signed short,
23013 vector signed short);
23015 vector unsigned int vec_msum (vector unsigned char,
23016 vector unsigned char,
23017 vector unsigned int);
23018 vector signed int vec_msum (vector signed char,
23019 vector unsigned char,
23020 vector signed int);
23021 vector unsigned int vec_msum (vector unsigned short,
23022 vector unsigned short,
23023 vector unsigned int);
23024 vector signed int vec_msum (vector signed short,
23025 vector signed short,
23026 vector signed int);
23028 vector signed int vec_vmsumshm (vector signed short,
23029 vector signed short,
23030 vector signed int);
23032 vector unsigned int vec_vmsumuhm (vector unsigned short,
23033 vector unsigned short,
23034 vector unsigned int);
23036 vector signed int vec_vmsummbm (vector signed char,
23037 vector unsigned char,
23038 vector signed int);
23040 vector unsigned int vec_vmsumubm (vector unsigned char,
23041 vector unsigned char,
23042 vector unsigned int);
23044 vector unsigned int vec_msums (vector unsigned short,
23045 vector unsigned short,
23046 vector unsigned int);
23047 vector signed int vec_msums (vector signed short,
23048 vector signed short,
23049 vector signed int);
23051 vector signed int vec_vmsumshs (vector signed short,
23052 vector signed short,
23053 vector signed int);
23055 vector unsigned int vec_vmsumuhs (vector unsigned short,
23056 vector unsigned short,
23057 vector unsigned int);
23059 void vec_mtvscr (vector signed int);
23060 void vec_mtvscr (vector unsigned int);
23061 void vec_mtvscr (vector bool int);
23062 void vec_mtvscr (vector signed short);
23063 void vec_mtvscr (vector unsigned short);
23064 void vec_mtvscr (vector bool short);
23065 void vec_mtvscr (vector pixel);
23066 void vec_mtvscr (vector signed char);
23067 void vec_mtvscr (vector unsigned char);
23068 void vec_mtvscr (vector bool char);
23070 vector unsigned short vec_mule (vector unsigned char,
23071 vector unsigned char);
23072 vector signed short vec_mule (vector signed char,
23073 vector signed char);
23074 vector unsigned int vec_mule (vector unsigned short,
23075 vector unsigned short);
23076 vector signed int vec_mule (vector signed short, vector signed short);
23078 vector signed int vec_vmulesh (vector signed short,
23079 vector signed short);
23081 vector unsigned int vec_vmuleuh (vector unsigned short,
23082 vector unsigned short);
23084 vector signed short vec_vmulesb (vector signed char,
23085 vector signed char);
23087 vector unsigned short vec_vmuleub (vector unsigned char,
23088 vector unsigned char);
23090 vector unsigned short vec_mulo (vector unsigned char,
23091 vector unsigned char);
23092 vector signed short vec_mulo (vector signed char, vector signed char);
23093 vector unsigned int vec_mulo (vector unsigned short,
23094 vector unsigned short);
23095 vector signed int vec_mulo (vector signed short, vector signed short);
23097 vector signed int vec_vmulosh (vector signed short,
23098 vector signed short);
23100 vector unsigned int vec_vmulouh (vector unsigned short,
23101 vector unsigned short);
23103 vector signed short vec_vmulosb (vector signed char,
23104 vector signed char);
23106 vector unsigned short vec_vmuloub (vector unsigned char,
23107 vector unsigned char);
23109 vector float vec_nmsub (vector float, vector float, vector float);
23111 vector float vec_nor (vector float, vector float);
23112 vector signed int vec_nor (vector signed int, vector signed int);
23113 vector unsigned int vec_nor (vector unsigned int, vector unsigned int);
23114 vector bool int vec_nor (vector bool int, vector bool int);
23115 vector signed short vec_nor (vector signed short, vector signed short);
23116 vector unsigned short vec_nor (vector unsigned short,
23117 vector unsigned short);
23118 vector bool short vec_nor (vector bool short, vector bool short);
23119 vector signed char vec_nor (vector signed char, vector signed char);
23120 vector unsigned char vec_nor (vector unsigned char,
23121 vector unsigned char);
23122 vector bool char vec_nor (vector bool char, vector bool char);
23124 vector float vec_or (vector float, vector float);
23125 vector float vec_or (vector float, vector bool int);
23126 vector float vec_or (vector bool int, vector float);
23127 vector bool int vec_or (vector bool int, vector bool int);
23128 vector signed int vec_or (vector bool int, vector signed int);
23129 vector signed int vec_or (vector signed int, vector bool int);
23130 vector signed int vec_or (vector signed int, vector signed int);
23131 vector unsigned int vec_or (vector bool int, vector unsigned int);
23132 vector unsigned int vec_or (vector unsigned int, vector bool int);
23133 vector unsigned int vec_or (vector unsigned int, vector unsigned int);
23134 vector bool short vec_or (vector bool short, vector bool short);
23135 vector signed short vec_or (vector bool short, vector signed short);
23136 vector signed short vec_or (vector signed short, vector bool short);
23137 vector signed short vec_or (vector signed short, vector signed short);
23138 vector unsigned short vec_or (vector bool short, vector unsigned short);
23139 vector unsigned short vec_or (vector unsigned short, vector bool short);
23140 vector unsigned short vec_or (vector unsigned short,
23141 vector unsigned short);
23142 vector signed char vec_or (vector bool char, vector signed char);
23143 vector bool char vec_or (vector bool char, vector bool char);
23144 vector signed char vec_or (vector signed char, vector bool char);
23145 vector signed char vec_or (vector signed char, vector signed char);
23146 vector unsigned char vec_or (vector bool char, vector unsigned char);
23147 vector unsigned char vec_or (vector unsigned char, vector bool char);
23148 vector unsigned char vec_or (vector unsigned char,
23149 vector unsigned char);
23151 vector signed char vec_pack (vector signed short, vector signed short);
23152 vector unsigned char vec_pack (vector unsigned short,
23153 vector unsigned short);
23154 vector bool char vec_pack (vector bool short, vector bool short);
23155 vector signed short vec_pack (vector signed int, vector signed int);
23156 vector unsigned short vec_pack (vector unsigned int,
23157 vector unsigned int);
23158 vector bool short vec_pack (vector bool int, vector bool int);
23160 vector bool short vec_vpkuwum (vector bool int, vector bool int);
23161 vector signed short vec_vpkuwum (vector signed int, vector signed int);
23162 vector unsigned short vec_vpkuwum (vector unsigned int,
23163 vector unsigned int);
23165 vector bool char vec_vpkuhum (vector bool short, vector bool short);
23166 vector signed char vec_vpkuhum (vector signed short,
23167 vector signed short);
23168 vector unsigned char vec_vpkuhum (vector unsigned short,
23169 vector unsigned short);
23171 vector pixel vec_packpx (vector unsigned int, vector unsigned int);
23173 vector unsigned char vec_packs (vector unsigned short,
23174 vector unsigned short);
23175 vector signed char vec_packs (vector signed short, vector signed short);
23176 vector unsigned short vec_packs (vector unsigned int,
23177 vector unsigned int);
23178 vector signed short vec_packs (vector signed int, vector signed int);
23180 vector signed short vec_vpkswss (vector signed int, vector signed int);
23182 vector unsigned short vec_vpkuwus (vector unsigned int,
23183 vector unsigned int);
23185 vector signed char vec_vpkshss (vector signed short,
23186 vector signed short);
23188 vector unsigned char vec_vpkuhus (vector unsigned short,
23189 vector unsigned short);
23191 vector unsigned char vec_packsu (vector unsigned short,
23192 vector unsigned short);
23193 vector unsigned char vec_packsu (vector signed short,
23194 vector signed short);
23195 vector unsigned short vec_packsu (vector unsigned int,
23196 vector unsigned int);
23197 vector unsigned short vec_packsu (vector signed int, vector signed int);
23199 vector unsigned short vec_vpkswus (vector signed int,
23200 vector signed int);
23202 vector unsigned char vec_vpkshus (vector signed short,
23203 vector signed short);
23205 vector float vec_perm (vector float,
23207 vector unsigned char);
23208 vector signed int vec_perm (vector signed int,
23210 vector unsigned char);
23211 vector unsigned int vec_perm (vector unsigned int,
23212 vector unsigned int,
23213 vector unsigned char);
23214 vector bool int vec_perm (vector bool int,
23216 vector unsigned char);
23217 vector signed short vec_perm (vector signed short,
23218 vector signed short,
23219 vector unsigned char);
23220 vector unsigned short vec_perm (vector unsigned short,
23221 vector unsigned short,
23222 vector unsigned char);
23223 vector bool short vec_perm (vector bool short,
23225 vector unsigned char);
23226 vector pixel vec_perm (vector pixel,
23228 vector unsigned char);
23229 vector signed char vec_perm (vector signed char,
23230 vector signed char,
23231 vector unsigned char);
23232 vector unsigned char vec_perm (vector unsigned char,
23233 vector unsigned char,
23234 vector unsigned char);
23235 vector bool char vec_perm (vector bool char,
23237 vector unsigned char);
23239 vector float vec_re (vector float);
23241 vector signed char vec_rl (vector signed char,
23242 vector unsigned char);
23243 vector unsigned char vec_rl (vector unsigned char,
23244 vector unsigned char);
23245 vector signed short vec_rl (vector signed short, vector unsigned short);
23246 vector unsigned short vec_rl (vector unsigned short,
23247 vector unsigned short);
23248 vector signed int vec_rl (vector signed int, vector unsigned int);
23249 vector unsigned int vec_rl (vector unsigned int, vector unsigned int);
23251 vector signed int vec_vrlw (vector signed int, vector unsigned int);
23252 vector unsigned int vec_vrlw (vector unsigned int, vector unsigned int);
23254 vector signed short vec_vrlh (vector signed short,
23255 vector unsigned short);
23256 vector unsigned short vec_vrlh (vector unsigned short,
23257 vector unsigned short);
23259 vector signed char vec_vrlb (vector signed char, vector unsigned char);
23260 vector unsigned char vec_vrlb (vector unsigned char,
23261 vector unsigned char);
23263 vector float vec_round (vector float);
23265 vector float vec_rsqrte (vector float);
23267 vector float vec_sel (vector float, vector float, vector bool int);
23268 vector float vec_sel (vector float, vector float, vector unsigned int);
23269 vector signed int vec_sel (vector signed int,
23272 vector signed int vec_sel (vector signed int,
23274 vector unsigned int);
23275 vector unsigned int vec_sel (vector unsigned int,
23276 vector unsigned int,
23278 vector unsigned int vec_sel (vector unsigned int,
23279 vector unsigned int,
23280 vector unsigned int);
23281 vector bool int vec_sel (vector bool int,
23284 vector bool int vec_sel (vector bool int,
23286 vector unsigned int);
23287 vector signed short vec_sel (vector signed short,
23288 vector signed short,
23289 vector bool short);
23290 vector signed short vec_sel (vector signed short,
23291 vector signed short,
23292 vector unsigned short);
23293 vector unsigned short vec_sel (vector unsigned short,
23294 vector unsigned short,
23295 vector bool short);
23296 vector unsigned short vec_sel (vector unsigned short,
23297 vector unsigned short,
23298 vector unsigned short);
23299 vector bool short vec_sel (vector bool short,
23301 vector bool short);
23302 vector bool short vec_sel (vector bool short,
23304 vector unsigned short);
23305 vector signed char vec_sel (vector signed char,
23306 vector signed char,
23308 vector signed char vec_sel (vector signed char,
23309 vector signed char,
23310 vector unsigned char);
23311 vector unsigned char vec_sel (vector unsigned char,
23312 vector unsigned char,
23314 vector unsigned char vec_sel (vector unsigned char,
23315 vector unsigned char,
23316 vector unsigned char);
23317 vector bool char vec_sel (vector bool char,
23320 vector bool char vec_sel (vector bool char,
23322 vector unsigned char);
23324 vector signed char vec_sl (vector signed char,
23325 vector unsigned char);
23326 vector unsigned char vec_sl (vector unsigned char,
23327 vector unsigned char);
23328 vector signed short vec_sl (vector signed short, vector unsigned short);
23329 vector unsigned short vec_sl (vector unsigned short,
23330 vector unsigned short);
23331 vector signed int vec_sl (vector signed int, vector unsigned int);
23332 vector unsigned int vec_sl (vector unsigned int, vector unsigned int);
23334 vector signed int vec_vslw (vector signed int, vector unsigned int);
23335 vector unsigned int vec_vslw (vector unsigned int, vector unsigned int);
23337 vector signed short vec_vslh (vector signed short,
23338 vector unsigned short);
23339 vector unsigned short vec_vslh (vector unsigned short,
23340 vector unsigned short);
23342 vector signed char vec_vslb (vector signed char, vector unsigned char);
23343 vector unsigned char vec_vslb (vector unsigned char,
23344 vector unsigned char);
23346 vector float vec_sld (vector float, vector float, const int);
23347 vector signed int vec_sld (vector signed int,
23350 vector unsigned int vec_sld (vector unsigned int,
23351 vector unsigned int,
23353 vector bool int vec_sld (vector bool int,
23356 vector signed short vec_sld (vector signed short,
23357 vector signed short,
23359 vector unsigned short vec_sld (vector unsigned short,
23360 vector unsigned short,
23362 vector bool short vec_sld (vector bool short,
23365 vector pixel vec_sld (vector pixel,
23368 vector signed char vec_sld (vector signed char,
23369 vector signed char,
23371 vector unsigned char vec_sld (vector unsigned char,
23372 vector unsigned char,
23374 vector bool char vec_sld (vector bool char,
23378 vector signed int vec_sll (vector signed int,
23379 vector unsigned int);
23380 vector signed int vec_sll (vector signed int,
23381 vector unsigned short);
23382 vector signed int vec_sll (vector signed int,
23383 vector unsigned char);
23384 vector unsigned int vec_sll (vector unsigned int,
23385 vector unsigned int);
23386 vector unsigned int vec_sll (vector unsigned int,
23387 vector unsigned short);
23388 vector unsigned int vec_sll (vector unsigned int,
23389 vector unsigned char);
23390 vector bool int vec_sll (vector bool int,
23391 vector unsigned int);
23392 vector bool int vec_sll (vector bool int,
23393 vector unsigned short);
23394 vector bool int vec_sll (vector bool int,
23395 vector unsigned char);
23396 vector signed short vec_sll (vector signed short,
23397 vector unsigned int);
23398 vector signed short vec_sll (vector signed short,
23399 vector unsigned short);
23400 vector signed short vec_sll (vector signed short,
23401 vector unsigned char);
23402 vector unsigned short vec_sll (vector unsigned short,
23403 vector unsigned int);
23404 vector unsigned short vec_sll (vector unsigned short,
23405 vector unsigned short);
23406 vector unsigned short vec_sll (vector unsigned short,
23407 vector unsigned char);
23408 vector bool short vec_sll (vector bool short, vector unsigned int);
23409 vector bool short vec_sll (vector bool short, vector unsigned short);
23410 vector bool short vec_sll (vector bool short, vector unsigned char);
23411 vector pixel vec_sll (vector pixel, vector unsigned int);
23412 vector pixel vec_sll (vector pixel, vector unsigned short);
23413 vector pixel vec_sll (vector pixel, vector unsigned char);
23414 vector signed char vec_sll (vector signed char, vector unsigned int);
23415 vector signed char vec_sll (vector signed char, vector unsigned short);
23416 vector signed char vec_sll (vector signed char, vector unsigned char);
23417 vector unsigned char vec_sll (vector unsigned char,
23418 vector unsigned int);
23419 vector unsigned char vec_sll (vector unsigned char,
23420 vector unsigned short);
23421 vector unsigned char vec_sll (vector unsigned char,
23422 vector unsigned char);
23423 vector bool char vec_sll (vector bool char, vector unsigned int);
23424 vector bool char vec_sll (vector bool char, vector unsigned short);
23425 vector bool char vec_sll (vector bool char, vector unsigned char);
23427 vector float vec_slo (vector float, vector signed char);
23428 vector float vec_slo (vector float, vector unsigned char);
23429 vector signed int vec_slo (vector signed int, vector signed char);
23430 vector signed int vec_slo (vector signed int, vector unsigned char);
23431 vector unsigned int vec_slo (vector unsigned int, vector signed char);
23432 vector unsigned int vec_slo (vector unsigned int, vector unsigned char);
23433 vector signed short vec_slo (vector signed short, vector signed char);
23434 vector signed short vec_slo (vector signed short, vector unsigned char);
23435 vector unsigned short vec_slo (vector unsigned short,
23436 vector signed char);
23437 vector unsigned short vec_slo (vector unsigned short,
23438 vector unsigned char);
23439 vector pixel vec_slo (vector pixel, vector signed char);
23440 vector pixel vec_slo (vector pixel, vector unsigned char);
23441 vector signed char vec_slo (vector signed char, vector signed char);
23442 vector signed char vec_slo (vector signed char, vector unsigned char);
23443 vector unsigned char vec_slo (vector unsigned char, vector signed char);
23444 vector unsigned char vec_slo (vector unsigned char,
23445 vector unsigned char);
23447 vector signed char vec_splat (vector signed char, const int);
23448 vector unsigned char vec_splat (vector unsigned char, const int);
23449 vector bool char vec_splat (vector bool char, const int);
23450 vector signed short vec_splat (vector signed short, const int);
23451 vector unsigned short vec_splat (vector unsigned short, const int);
23452 vector bool short vec_splat (vector bool short, const int);
23453 vector pixel vec_splat (vector pixel, const int);
23454 vector float vec_splat (vector float, const int);
23455 vector signed int vec_splat (vector signed int, const int);
23456 vector unsigned int vec_splat (vector unsigned int, const int);
23457 vector bool int vec_splat (vector bool int, const int);
23459 vector float vec_vspltw (vector float, const int);
23460 vector signed int vec_vspltw (vector signed int, const int);
23461 vector unsigned int vec_vspltw (vector unsigned int, const int);
23462 vector bool int vec_vspltw (vector bool int, const int);
23464 vector bool short vec_vsplth (vector bool short, const int);
23465 vector signed short vec_vsplth (vector signed short, const int);
23466 vector unsigned short vec_vsplth (vector unsigned short, const int);
23467 vector pixel vec_vsplth (vector pixel, const int);
23469 vector signed char vec_vspltb (vector signed char, const int);
23470 vector unsigned char vec_vspltb (vector unsigned char, const int);
23471 vector bool char vec_vspltb (vector bool char, const int);
23473 vector signed char vec_splat_s8 (const int);
23475 vector signed short vec_splat_s16 (const int);
23477 vector signed int vec_splat_s32 (const int);
23479 vector unsigned char vec_splat_u8 (const int);
23481 vector unsigned short vec_splat_u16 (const int);
23483 vector unsigned int vec_splat_u32 (const int);
23485 vector signed char vec_sr (vector signed char, vector unsigned char);
23486 vector unsigned char vec_sr (vector unsigned char,
23487 vector unsigned char);
23488 vector signed short vec_sr (vector signed short,
23489 vector unsigned short);
23490 vector unsigned short vec_sr (vector unsigned short,
23491 vector unsigned short);
23492 vector signed int vec_sr (vector signed int, vector unsigned int);
23493 vector unsigned int vec_sr (vector unsigned int, vector unsigned int);
23495 vector signed int vec_vsrw (vector signed int, vector unsigned int);
23496 vector unsigned int vec_vsrw (vector unsigned int, vector unsigned int);
23498 vector signed short vec_vsrh (vector signed short,
23499 vector unsigned short);
23500 vector unsigned short vec_vsrh (vector unsigned short,
23501 vector unsigned short);
23503 vector signed char vec_vsrb (vector signed char, vector unsigned char);
23504 vector unsigned char vec_vsrb (vector unsigned char,
23505 vector unsigned char);
23507 vector signed char vec_sra (vector signed char, vector unsigned char);
23508 vector unsigned char vec_sra (vector unsigned char,
23509 vector unsigned char);
23510 vector signed short vec_sra (vector signed short,
23511 vector unsigned short);
23512 vector unsigned short vec_sra (vector unsigned short,
23513 vector unsigned short);
23514 vector signed int vec_sra (vector signed int, vector unsigned int);
23515 vector unsigned int vec_sra (vector unsigned int, vector unsigned int);
23517 vector signed int vec_vsraw (vector signed int, vector unsigned int);
23518 vector unsigned int vec_vsraw (vector unsigned int,
23519 vector unsigned int);
23521 vector signed short vec_vsrah (vector signed short,
23522 vector unsigned short);
23523 vector unsigned short vec_vsrah (vector unsigned short,
23524 vector unsigned short);
23526 vector signed char vec_vsrab (vector signed char, vector unsigned char);
23527 vector unsigned char vec_vsrab (vector unsigned char,
23528 vector unsigned char);
23530 vector signed int vec_srl (vector signed int, vector unsigned int);
23531 vector signed int vec_srl (vector signed int, vector unsigned short);
23532 vector signed int vec_srl (vector signed int, vector unsigned char);
23533 vector unsigned int vec_srl (vector unsigned int, vector unsigned int);
23534 vector unsigned int vec_srl (vector unsigned int,
23535 vector unsigned short);
23536 vector unsigned int vec_srl (vector unsigned int, vector unsigned char);
23537 vector bool int vec_srl (vector bool int, vector unsigned int);
23538 vector bool int vec_srl (vector bool int, vector unsigned short);
23539 vector bool int vec_srl (vector bool int, vector unsigned char);
23540 vector signed short vec_srl (vector signed short, vector unsigned int);
23541 vector signed short vec_srl (vector signed short,
23542 vector unsigned short);
23543 vector signed short vec_srl (vector signed short, vector unsigned char);
23544 vector unsigned short vec_srl (vector unsigned short,
23545 vector unsigned int);
23546 vector unsigned short vec_srl (vector unsigned short,
23547 vector unsigned short);
23548 vector unsigned short vec_srl (vector unsigned short,
23549 vector unsigned char);
23550 vector bool short vec_srl (vector bool short, vector unsigned int);
23551 vector bool short vec_srl (vector bool short, vector unsigned short);
23552 vector bool short vec_srl (vector bool short, vector unsigned char);
23553 vector pixel vec_srl (vector pixel, vector unsigned int);
23554 vector pixel vec_srl (vector pixel, vector unsigned short);
23555 vector pixel vec_srl (vector pixel, vector unsigned char);
23556 vector signed char vec_srl (vector signed char, vector unsigned int);
23557 vector signed char vec_srl (vector signed char, vector unsigned short);
23558 vector signed char vec_srl (vector signed char, vector unsigned char);
23559 vector unsigned char vec_srl (vector unsigned char,
23560 vector unsigned int);
23561 vector unsigned char vec_srl (vector unsigned char,
23562 vector unsigned short);
23563 vector unsigned char vec_srl (vector unsigned char,
23564 vector unsigned char);
23565 vector bool char vec_srl (vector bool char, vector unsigned int);
23566 vector bool char vec_srl (vector bool char, vector unsigned short);
23567 vector bool char vec_srl (vector bool char, vector unsigned char);
23569 vector float vec_sro (vector float, vector signed char);
23570 vector float vec_sro (vector float, vector unsigned char);
23571 vector signed int vec_sro (vector signed int, vector signed char);
23572 vector signed int vec_sro (vector signed int, vector unsigned char);
23573 vector unsigned int vec_sro (vector unsigned int, vector signed char);
23574 vector unsigned int vec_sro (vector unsigned int, vector unsigned char);
23575 vector signed short vec_sro (vector signed short, vector signed char);
23576 vector signed short vec_sro (vector signed short, vector unsigned char);
23577 vector unsigned short vec_sro (vector unsigned short,
23578 vector signed char);
23579 vector unsigned short vec_sro (vector unsigned short,
23580 vector unsigned char);
23581 vector pixel vec_sro (vector pixel, vector signed char);
23582 vector pixel vec_sro (vector pixel, vector unsigned char);
23583 vector signed char vec_sro (vector signed char, vector signed char);
23584 vector signed char vec_sro (vector signed char, vector unsigned char);
23585 vector unsigned char vec_sro (vector unsigned char, vector signed char);
23586 vector unsigned char vec_sro (vector unsigned char,
23587 vector unsigned char);
23589 void vec_st (vector float, int, vector float *);
23590 void vec_st (vector float, int, float *);
23591 void vec_st (vector signed int, int, vector signed int *);
23592 void vec_st (vector signed int, int, int *);
23593 void vec_st (vector unsigned int, int, vector unsigned int *);
23594 void vec_st (vector unsigned int, int, unsigned int *);
23595 void vec_st (vector bool int, int, vector bool int *);
23596 void vec_st (vector bool int, int, unsigned int *);
23597 void vec_st (vector bool int, int, int *);
23598 void vec_st (vector signed short, int, vector signed short *);
23599 void vec_st (vector signed short, int, short *);
23600 void vec_st (vector unsigned short, int, vector unsigned short *);
23601 void vec_st (vector unsigned short, int, unsigned short *);
23602 void vec_st (vector bool short, int, vector bool short *);
23603 void vec_st (vector bool short, int, unsigned short *);
23604 void vec_st (vector pixel, int, vector pixel *);
23605 void vec_st (vector pixel, int, unsigned short *);
23606 void vec_st (vector pixel, int, short *);
23607 void vec_st (vector bool short, int, short *);
23608 void vec_st (vector signed char, int, vector signed char *);
23609 void vec_st (vector signed char, int, signed char *);
23610 void vec_st (vector unsigned char, int, vector unsigned char *);
23611 void vec_st (vector unsigned char, int, unsigned char *);
23612 void vec_st (vector bool char, int, vector bool char *);
23613 void vec_st (vector bool char, int, unsigned char *);
23614 void vec_st (vector bool char, int, signed char *);
23616 void vec_ste (vector signed char, int, signed char *);
23617 void vec_ste (vector unsigned char, int, unsigned char *);
23618 void vec_ste (vector bool char, int, signed char *);
23619 void vec_ste (vector bool char, int, unsigned char *);
23620 void vec_ste (vector signed short, int, short *);
23621 void vec_ste (vector unsigned short, int, unsigned short *);
23622 void vec_ste (vector bool short, int, short *);
23623 void vec_ste (vector bool short, int, unsigned short *);
23624 void vec_ste (vector pixel, int, short *);
23625 void vec_ste (vector pixel, int, unsigned short *);
23626 void vec_ste (vector float, int, float *);
23627 void vec_ste (vector signed int, int, int *);
23628 void vec_ste (vector unsigned int, int, unsigned int *);
23629 void vec_ste (vector bool int, int, int *);
23630 void vec_ste (vector bool int, int, unsigned int *);
23632 void vec_stvewx (vector float, int, float *);
23633 void vec_stvewx (vector signed int, int, int *);
23634 void vec_stvewx (vector unsigned int, int, unsigned int *);
23635 void vec_stvewx (vector bool int, int, int *);
23636 void vec_stvewx (vector bool int, int, unsigned int *);
23638 void vec_stvehx (vector signed short, int, short *);
23639 void vec_stvehx (vector unsigned short, int, unsigned short *);
23640 void vec_stvehx (vector bool short, int, short *);
23641 void vec_stvehx (vector bool short, int, unsigned short *);
23642 void vec_stvehx (vector pixel, int, short *);
23643 void vec_stvehx (vector pixel, int, unsigned short *);
23645 void vec_stvebx (vector signed char, int, signed char *);
23646 void vec_stvebx (vector unsigned char, int, unsigned char *);
23647 void vec_stvebx (vector bool char, int, signed char *);
23648 void vec_stvebx (vector bool char, int, unsigned char *);
23650 void vec_stl (vector float, int, vector float *);
23651 void vec_stl (vector float, int, float *);
23652 void vec_stl (vector signed int, int, vector signed int *);
23653 void vec_stl (vector signed int, int, int *);
23654 void vec_stl (vector unsigned int, int, vector unsigned int *);
23655 void vec_stl (vector unsigned int, int, unsigned int *);
23656 void vec_stl (vector bool int, int, vector bool int *);
23657 void vec_stl (vector bool int, int, unsigned int *);
23658 void vec_stl (vector bool int, int, int *);
23659 void vec_stl (vector signed short, int, vector signed short *);
23660 void vec_stl (vector signed short, int, short *);
23661 void vec_stl (vector unsigned short, int, vector unsigned short *);
23662 void vec_stl (vector unsigned short, int, unsigned short *);
23663 void vec_stl (vector bool short, int, vector bool short *);
23664 void vec_stl (vector bool short, int, unsigned short *);
23665 void vec_stl (vector bool short, int, short *);
23666 void vec_stl (vector pixel, int, vector pixel *);
23667 void vec_stl (vector pixel, int, unsigned short *);
23668 void vec_stl (vector pixel, int, short *);
23669 void vec_stl (vector signed char, int, vector signed char *);
23670 void vec_stl (vector signed char, int, signed char *);
23671 void vec_stl (vector unsigned char, int, vector unsigned char *);
23672 void vec_stl (vector unsigned char, int, unsigned char *);
23673 void vec_stl (vector bool char, int, vector bool char *);
23674 void vec_stl (vector bool char, int, unsigned char *);
23675 void vec_stl (vector bool char, int, signed char *);
23677 vector signed char vec_sub (vector bool char, vector signed char);
23678 vector signed char vec_sub (vector signed char, vector bool char);
23679 vector signed char vec_sub (vector signed char, vector signed char);
23680 vector unsigned char vec_sub (vector bool char, vector unsigned char);
23681 vector unsigned char vec_sub (vector unsigned char, vector bool char);
23682 vector unsigned char vec_sub (vector unsigned char,
23683 vector unsigned char);
23684 vector signed short vec_sub (vector bool short, vector signed short);
23685 vector signed short vec_sub (vector signed short, vector bool short);
23686 vector signed short vec_sub (vector signed short, vector signed short);
23687 vector unsigned short vec_sub (vector bool short,
23688 vector unsigned short);
23689 vector unsigned short vec_sub (vector unsigned short,
23690 vector bool short);
23691 vector unsigned short vec_sub (vector unsigned short,
23692 vector unsigned short);
23693 vector signed int vec_sub (vector bool int, vector signed int);
23694 vector signed int vec_sub (vector signed int, vector bool int);
23695 vector signed int vec_sub (vector signed int, vector signed int);
23696 vector unsigned int vec_sub (vector bool int, vector unsigned int);
23697 vector unsigned int vec_sub (vector unsigned int, vector bool int);
23698 vector unsigned int vec_sub (vector unsigned int, vector unsigned int);
23699 vector float vec_sub (vector float, vector float);
23701 vector float vec_vsubfp (vector float, vector float);
23703 vector signed int vec_vsubuwm (vector bool int, vector signed int);
23704 vector signed int vec_vsubuwm (vector signed int, vector bool int);
23705 vector signed int vec_vsubuwm (vector signed int, vector signed int);
23706 vector unsigned int vec_vsubuwm (vector bool int, vector unsigned int);
23707 vector unsigned int vec_vsubuwm (vector unsigned int, vector bool int);
23708 vector unsigned int vec_vsubuwm (vector unsigned int,
23709 vector unsigned int);
23711 vector signed short vec_vsubuhm (vector bool short,
23712 vector signed short);
23713 vector signed short vec_vsubuhm (vector signed short,
23714 vector bool short);
23715 vector signed short vec_vsubuhm (vector signed short,
23716 vector signed short);
23717 vector unsigned short vec_vsubuhm (vector bool short,
23718 vector unsigned short);
23719 vector unsigned short vec_vsubuhm (vector unsigned short,
23720 vector bool short);
23721 vector unsigned short vec_vsubuhm (vector unsigned short,
23722 vector unsigned short);
23724 vector signed char vec_vsububm (vector bool char, vector signed char);
23725 vector signed char vec_vsububm (vector signed char, vector bool char);
23726 vector signed char vec_vsububm (vector signed char, vector signed char);
23727 vector unsigned char vec_vsububm (vector bool char,
23728 vector unsigned char);
23729 vector unsigned char vec_vsububm (vector unsigned char,
23731 vector unsigned char vec_vsububm (vector unsigned char,
23732 vector unsigned char);
23734 vector unsigned int vec_subc (vector unsigned int, vector unsigned int);
23736 vector unsigned char vec_subs (vector bool char, vector unsigned char);
23737 vector unsigned char vec_subs (vector unsigned char, vector bool char);
23738 vector unsigned char vec_subs (vector unsigned char,
23739 vector unsigned char);
23740 vector signed char vec_subs (vector bool char, vector signed char);
23741 vector signed char vec_subs (vector signed char, vector bool char);
23742 vector signed char vec_subs (vector signed char, vector signed char);
23743 vector unsigned short vec_subs (vector bool short,
23744 vector unsigned short);
23745 vector unsigned short vec_subs (vector unsigned short,
23746 vector bool short);
23747 vector unsigned short vec_subs (vector unsigned short,
23748 vector unsigned short);
23749 vector signed short vec_subs (vector bool short, vector signed short);
23750 vector signed short vec_subs (vector signed short, vector bool short);
23751 vector signed short vec_subs (vector signed short, vector signed short);
23752 vector unsigned int vec_subs (vector bool int, vector unsigned int);
23753 vector unsigned int vec_subs (vector unsigned int, vector bool int);
23754 vector unsigned int vec_subs (vector unsigned int, vector unsigned int);
23755 vector signed int vec_subs (vector bool int, vector signed int);
23756 vector signed int vec_subs (vector signed int, vector bool int);
23757 vector signed int vec_subs (vector signed int, vector signed int);
23759 vector signed int vec_vsubsws (vector bool int, vector signed int);
23760 vector signed int vec_vsubsws (vector signed int, vector bool int);
23761 vector signed int vec_vsubsws (vector signed int, vector signed int);
23763 vector unsigned int vec_vsubuws (vector bool int, vector unsigned int);
23764 vector unsigned int vec_vsubuws (vector unsigned int, vector bool int);
23765 vector unsigned int vec_vsubuws (vector unsigned int,
23766 vector unsigned int);
23768 vector signed short vec_vsubshs (vector bool short,
23769 vector signed short);
23770 vector signed short vec_vsubshs (vector signed short,
23771 vector bool short);
23772 vector signed short vec_vsubshs (vector signed short,
23773 vector signed short);
23775 vector unsigned short vec_vsubuhs (vector bool short,
23776 vector unsigned short);
23777 vector unsigned short vec_vsubuhs (vector unsigned short,
23778 vector bool short);
23779 vector unsigned short vec_vsubuhs (vector unsigned short,
23780 vector unsigned short);
23782 vector signed char vec_vsubsbs (vector bool char, vector signed char);
23783 vector signed char vec_vsubsbs (vector signed char, vector bool char);
23784 vector signed char vec_vsubsbs (vector signed char, vector signed char);
23786 vector unsigned char vec_vsububs (vector bool char,
23787 vector unsigned char);
23788 vector unsigned char vec_vsububs (vector unsigned char,
23790 vector unsigned char vec_vsububs (vector unsigned char,
23791 vector unsigned char);
23793 vector unsigned int vec_sum4s (vector unsigned char,
23794 vector unsigned int);
23795 vector signed int vec_sum4s (vector signed char, vector signed int);
23796 vector signed int vec_sum4s (vector signed short, vector signed int);
23798 vector signed int vec_vsum4shs (vector signed short, vector signed int);
23800 vector signed int vec_vsum4sbs (vector signed char, vector signed int);
23802 vector unsigned int vec_vsum4ubs (vector unsigned char,
23803 vector unsigned int);
23805 vector signed int vec_sum2s (vector signed int, vector signed int);
23807 vector signed int vec_sums (vector signed int, vector signed int);
23809 vector float vec_trunc (vector float);
23811 vector signed short vec_unpackh (vector signed char);
23812 vector bool short vec_unpackh (vector bool char);
23813 vector signed int vec_unpackh (vector signed short);
23814 vector bool int vec_unpackh (vector bool short);
23815 vector unsigned int vec_unpackh (vector pixel);
23817 vector bool int vec_vupkhsh (vector bool short);
23818 vector signed int vec_vupkhsh (vector signed short);
23820 vector unsigned int vec_vupkhpx (vector pixel);
23822 vector bool short vec_vupkhsb (vector bool char);
23823 vector signed short vec_vupkhsb (vector signed char);
23825 vector signed short vec_unpackl (vector signed char);
23826 vector bool short vec_unpackl (vector bool char);
23827 vector unsigned int vec_unpackl (vector pixel);
23828 vector signed int vec_unpackl (vector signed short);
23829 vector bool int vec_unpackl (vector bool short);
23831 vector unsigned int vec_vupklpx (vector pixel);
23833 vector bool int vec_vupklsh (vector bool short);
23834 vector signed int vec_vupklsh (vector signed short);
23836 vector bool short vec_vupklsb (vector bool char);
23837 vector signed short vec_vupklsb (vector signed char);
23839 vector float vec_xor (vector float, vector float);
23840 vector float vec_xor (vector float, vector bool int);
23841 vector float vec_xor (vector bool int, vector float);
23842 vector bool int vec_xor (vector bool int, vector bool int);
23843 vector signed int vec_xor (vector bool int, vector signed int);
23844 vector signed int vec_xor (vector signed int, vector bool int);
23845 vector signed int vec_xor (vector signed int, vector signed int);
23846 vector unsigned int vec_xor (vector bool int, vector unsigned int);
23847 vector unsigned int vec_xor (vector unsigned int, vector bool int);
23848 vector unsigned int vec_xor (vector unsigned int, vector unsigned int);
23849 vector bool short vec_xor (vector bool short, vector bool short);
23850 vector signed short vec_xor (vector bool short, vector signed short);
23851 vector signed short vec_xor (vector signed short, vector bool short);
23852 vector signed short vec_xor (vector signed short, vector signed short);
23853 vector unsigned short vec_xor (vector bool short,
23854 vector unsigned short);
23855 vector unsigned short vec_xor (vector unsigned short,
23856 vector bool short);
23857 vector unsigned short vec_xor (vector unsigned short,
23858 vector unsigned short);
23859 vector signed char vec_xor (vector bool char, vector signed char);
23860 vector bool char vec_xor (vector bool char, vector bool char);
23861 vector signed char vec_xor (vector signed char, vector bool char);
23862 vector signed char vec_xor (vector signed char, vector signed char);
23863 vector unsigned char vec_xor (vector bool char, vector unsigned char);
23864 vector unsigned char vec_xor (vector unsigned char, vector bool char);
23865 vector unsigned char vec_xor (vector unsigned char,
23866 vector unsigned char);
23868 int vec_all_eq (vector signed char, vector bool char);
23869 int vec_all_eq (vector signed char, vector signed char);
23870 int vec_all_eq (vector unsigned char, vector bool char);
23871 int vec_all_eq (vector unsigned char, vector unsigned char);
23872 int vec_all_eq (vector bool char, vector bool char);
23873 int vec_all_eq (vector bool char, vector unsigned char);
23874 int vec_all_eq (vector bool char, vector signed char);
23875 int vec_all_eq (vector signed short, vector bool short);
23876 int vec_all_eq (vector signed short, vector signed short);
23877 int vec_all_eq (vector unsigned short, vector bool short);
23878 int vec_all_eq (vector unsigned short, vector unsigned short);
23879 int vec_all_eq (vector bool short, vector bool short);
23880 int vec_all_eq (vector bool short, vector unsigned short);
23881 int vec_all_eq (vector bool short, vector signed short);
23882 int vec_all_eq (vector pixel, vector pixel);
23883 int vec_all_eq (vector signed int, vector bool int);
23884 int vec_all_eq (vector signed int, vector signed int);
23885 int vec_all_eq (vector unsigned int, vector bool int);
23886 int vec_all_eq (vector unsigned int, vector unsigned int);
23887 int vec_all_eq (vector bool int, vector bool int);
23888 int vec_all_eq (vector bool int, vector unsigned int);
23889 int vec_all_eq (vector bool int, vector signed int);
23890 int vec_all_eq (vector float, vector float);
23892 int vec_all_ge (vector bool char, vector unsigned char);
23893 int vec_all_ge (vector unsigned char, vector bool char);
23894 int vec_all_ge (vector unsigned char, vector unsigned char);
23895 int vec_all_ge (vector bool char, vector signed char);
23896 int vec_all_ge (vector signed char, vector bool char);
23897 int vec_all_ge (vector signed char, vector signed char);
23898 int vec_all_ge (vector bool short, vector unsigned short);
23899 int vec_all_ge (vector unsigned short, vector bool short);
23900 int vec_all_ge (vector unsigned short, vector unsigned short);
23901 int vec_all_ge (vector signed short, vector signed short);
23902 int vec_all_ge (vector bool short, vector signed short);
23903 int vec_all_ge (vector signed short, vector bool short);
23904 int vec_all_ge (vector bool int, vector unsigned int);
23905 int vec_all_ge (vector unsigned int, vector bool int);
23906 int vec_all_ge (vector unsigned int, vector unsigned int);
23907 int vec_all_ge (vector bool int, vector signed int);
23908 int vec_all_ge (vector signed int, vector bool int);
23909 int vec_all_ge (vector signed int, vector signed int);
23910 int vec_all_ge (vector float, vector float);
23912 int vec_all_gt (vector bool char, vector unsigned char);
23913 int vec_all_gt (vector unsigned char, vector bool char);
23914 int vec_all_gt (vector unsigned char, vector unsigned char);
23915 int vec_all_gt (vector bool char, vector signed char);
23916 int vec_all_gt (vector signed char, vector bool char);
23917 int vec_all_gt (vector signed char, vector signed char);
23918 int vec_all_gt (vector bool short, vector unsigned short);
23919 int vec_all_gt (vector unsigned short, vector bool short);
23920 int vec_all_gt (vector unsigned short, vector unsigned short);
23921 int vec_all_gt (vector bool short, vector signed short);
23922 int vec_all_gt (vector signed short, vector bool short);
23923 int vec_all_gt (vector signed short, vector signed short);
23924 int vec_all_gt (vector bool int, vector unsigned int);
23925 int vec_all_gt (vector unsigned int, vector bool int);
23926 int vec_all_gt (vector unsigned int, vector unsigned int);
23927 int vec_all_gt (vector bool int, vector signed int);
23928 int vec_all_gt (vector signed int, vector bool int);
23929 int vec_all_gt (vector signed int, vector signed int);
23930 int vec_all_gt (vector float, vector float);
23932 int vec_all_in (vector float, vector float);
23934 int vec_all_le (vector bool char, vector unsigned char);
23935 int vec_all_le (vector unsigned char, vector bool char);
23936 int vec_all_le (vector unsigned char, vector unsigned char);
23937 int vec_all_le (vector bool char, vector signed char);
23938 int vec_all_le (vector signed char, vector bool char);
23939 int vec_all_le (vector signed char, vector signed char);
23940 int vec_all_le (vector bool short, vector unsigned short);
23941 int vec_all_le (vector unsigned short, vector bool short);
23942 int vec_all_le (vector unsigned short, vector unsigned short);
23943 int vec_all_le (vector bool short, vector signed short);
23944 int vec_all_le (vector signed short, vector bool short);
23945 int vec_all_le (vector signed short, vector signed short);
23946 int vec_all_le (vector bool int, vector unsigned int);
23947 int vec_all_le (vector unsigned int, vector bool int);
23948 int vec_all_le (vector unsigned int, vector unsigned int);
23949 int vec_all_le (vector bool int, vector signed int);
23950 int vec_all_le (vector signed int, vector bool int);
23951 int vec_all_le (vector signed int, vector signed int);
23952 int vec_all_le (vector float, vector float);
23954 int vec_all_lt (vector bool char, vector unsigned char);
23955 int vec_all_lt (vector unsigned char, vector bool char);
23956 int vec_all_lt (vector unsigned char, vector unsigned char);
23957 int vec_all_lt (vector bool char, vector signed char);
23958 int vec_all_lt (vector signed char, vector bool char);
23959 int vec_all_lt (vector signed char, vector signed char);
23960 int vec_all_lt (vector bool short, vector unsigned short);
23961 int vec_all_lt (vector unsigned short, vector bool short);
23962 int vec_all_lt (vector unsigned short, vector unsigned short);
23963 int vec_all_lt (vector bool short, vector signed short);
23964 int vec_all_lt (vector signed short, vector bool short);
23965 int vec_all_lt (vector signed short, vector signed short);
23966 int vec_all_lt (vector bool int, vector unsigned int);
23967 int vec_all_lt (vector unsigned int, vector bool int);
23968 int vec_all_lt (vector unsigned int, vector unsigned int);
23969 int vec_all_lt (vector bool int, vector signed int);
23970 int vec_all_lt (vector signed int, vector bool int);
23971 int vec_all_lt (vector signed int, vector signed int);
23972 int vec_all_lt (vector float, vector float);
23974 int vec_all_nan (vector float);
23976 int vec_all_ne (vector signed char, vector bool char);
23977 int vec_all_ne (vector signed char, vector signed char);
23978 int vec_all_ne (vector unsigned char, vector bool char);
23979 int vec_all_ne (vector unsigned char, vector unsigned char);
23980 int vec_all_ne (vector bool char, vector bool char);
23981 int vec_all_ne (vector bool char, vector unsigned char);
23982 int vec_all_ne (vector bool char, vector signed char);
23983 int vec_all_ne (vector signed short, vector bool short);
23984 int vec_all_ne (vector signed short, vector signed short);
23985 int vec_all_ne (vector unsigned short, vector bool short);
23986 int vec_all_ne (vector unsigned short, vector unsigned short);
23987 int vec_all_ne (vector bool short, vector bool short);
23988 int vec_all_ne (vector bool short, vector unsigned short);
23989 int vec_all_ne (vector bool short, vector signed short);
23990 int vec_all_ne (vector pixel, vector pixel);
23991 int vec_all_ne (vector signed int, vector bool int);
23992 int vec_all_ne (vector signed int, vector signed int);
23993 int vec_all_ne (vector unsigned int, vector bool int);
23994 int vec_all_ne (vector unsigned int, vector unsigned int);
23995 int vec_all_ne (vector bool int, vector bool int);
23996 int vec_all_ne (vector bool int, vector unsigned int);
23997 int vec_all_ne (vector bool int, vector signed int);
23998 int vec_all_ne (vector float, vector float);
24000 int vec_all_nge (vector float, vector float);
24002 int vec_all_ngt (vector float, vector float);
24004 int vec_all_nle (vector float, vector float);
24006 int vec_all_nlt (vector float, vector float);
24008 int vec_all_numeric (vector float);
24010 int vec_any_eq (vector signed char, vector bool char);
24011 int vec_any_eq (vector signed char, vector signed char);
24012 int vec_any_eq (vector unsigned char, vector bool char);
24013 int vec_any_eq (vector unsigned char, vector unsigned char);
24014 int vec_any_eq (vector bool char, vector bool char);
24015 int vec_any_eq (vector bool char, vector unsigned char);
24016 int vec_any_eq (vector bool char, vector signed char);
24017 int vec_any_eq (vector signed short, vector bool short);
24018 int vec_any_eq (vector signed short, vector signed short);
24019 int vec_any_eq (vector unsigned short, vector bool short);
24020 int vec_any_eq (vector unsigned short, vector unsigned short);
24021 int vec_any_eq (vector bool short, vector bool short);
24022 int vec_any_eq (vector bool short, vector unsigned short);
24023 int vec_any_eq (vector bool short, vector signed short);
24024 int vec_any_eq (vector pixel, vector pixel);
24025 int vec_any_eq (vector signed int, vector bool int);
24026 int vec_any_eq (vector signed int, vector signed int);
24027 int vec_any_eq (vector unsigned int, vector bool int);
24028 int vec_any_eq (vector unsigned int, vector unsigned int);
24029 int vec_any_eq (vector bool int, vector bool int);
24030 int vec_any_eq (vector bool int, vector unsigned int);
24031 int vec_any_eq (vector bool int, vector signed int);
24032 int vec_any_eq (vector float, vector float);
24034 int vec_any_ge (vector signed char, vector bool char);
24035 int vec_any_ge (vector unsigned char, vector bool char);
24036 int vec_any_ge (vector unsigned char, vector unsigned char);
24037 int vec_any_ge (vector signed char, vector signed char);
24038 int vec_any_ge (vector bool char, vector unsigned char);
24039 int vec_any_ge (vector bool char, vector signed char);
24040 int vec_any_ge (vector unsigned short, vector bool short);
24041 int vec_any_ge (vector unsigned short, vector unsigned short);
24042 int vec_any_ge (vector signed short, vector signed short);
24043 int vec_any_ge (vector signed short, vector bool short);
24044 int vec_any_ge (vector bool short, vector unsigned short);
24045 int vec_any_ge (vector bool short, vector signed short);
24046 int vec_any_ge (vector signed int, vector bool int);
24047 int vec_any_ge (vector unsigned int, vector bool int);
24048 int vec_any_ge (vector unsigned int, vector unsigned int);
24049 int vec_any_ge (vector signed int, vector signed int);
24050 int vec_any_ge (vector bool int, vector unsigned int);
24051 int vec_any_ge (vector bool int, vector signed int);
24052 int vec_any_ge (vector float, vector float);
24054 int vec_any_gt (vector bool char, vector unsigned char);
24055 int vec_any_gt (vector unsigned char, vector bool char);
24056 int vec_any_gt (vector unsigned char, vector unsigned char);
24057 int vec_any_gt (vector bool char, vector signed char);
24058 int vec_any_gt (vector signed char, vector bool char);
24059 int vec_any_gt (vector signed char, vector signed char);
24060 int vec_any_gt (vector bool short, vector unsigned short);
24061 int vec_any_gt (vector unsigned short, vector bool short);
24062 int vec_any_gt (vector unsigned short, vector unsigned short);
24063 int vec_any_gt (vector bool short, vector signed short);
24064 int vec_any_gt (vector signed short, vector bool short);
24065 int vec_any_gt (vector signed short, vector signed short);
24066 int vec_any_gt (vector bool int, vector unsigned int);
24067 int vec_any_gt (vector unsigned int, vector bool int);
24068 int vec_any_gt (vector unsigned int, vector unsigned int);
24069 int vec_any_gt (vector bool int, vector signed int);
24070 int vec_any_gt (vector signed int, vector bool int);
24071 int vec_any_gt (vector signed int, vector signed int);
24072 int vec_any_gt (vector float, vector float);
24074 int vec_any_le (vector bool char, vector unsigned char);
24075 int vec_any_le (vector unsigned char, vector bool char);
24076 int vec_any_le (vector unsigned char, vector unsigned char);
24077 int vec_any_le (vector bool char, vector signed char);
24078 int vec_any_le (vector signed char, vector bool char);
24079 int vec_any_le (vector signed char, vector signed char);
24080 int vec_any_le (vector bool short, vector unsigned short);
24081 int vec_any_le (vector unsigned short, vector bool short);
24082 int vec_any_le (vector unsigned short, vector unsigned short);
24083 int vec_any_le (vector bool short, vector signed short);
24084 int vec_any_le (vector signed short, vector bool short);
24085 int vec_any_le (vector signed short, vector signed short);
24086 int vec_any_le (vector bool int, vector unsigned int);
24087 int vec_any_le (vector unsigned int, vector bool int);
24088 int vec_any_le (vector unsigned int, vector unsigned int);
24089 int vec_any_le (vector bool int, vector signed int);
24090 int vec_any_le (vector signed int, vector bool int);
24091 int vec_any_le (vector signed int, vector signed int);
24092 int vec_any_le (vector float, vector float);
24094 int vec_any_lt (vector bool char, vector unsigned char);
24095 int vec_any_lt (vector unsigned char, vector bool char);
24096 int vec_any_lt (vector unsigned char, vector unsigned char);
24097 int vec_any_lt (vector bool char, vector signed char);
24098 int vec_any_lt (vector signed char, vector bool char);
24099 int vec_any_lt (vector signed char, vector signed char);
24100 int vec_any_lt (vector bool short, vector unsigned short);
24101 int vec_any_lt (vector unsigned short, vector bool short);
24102 int vec_any_lt (vector unsigned short, vector unsigned short);
24103 int vec_any_lt (vector bool short, vector signed short);
24104 int vec_any_lt (vector signed short, vector bool short);
24105 int vec_any_lt (vector signed short, vector signed short);
24106 int vec_any_lt (vector bool int, vector unsigned int);
24107 int vec_any_lt (vector unsigned int, vector bool int);
24108 int vec_any_lt (vector unsigned int, vector unsigned int);
24109 int vec_any_lt (vector bool int, vector signed int);
24110 int vec_any_lt (vector signed int, vector bool int);
24111 int vec_any_lt (vector signed int, vector signed int);
24112 int vec_any_lt (vector float, vector float);
24114 int vec_any_nan (vector float);
24116 int vec_any_ne (vector signed char, vector bool char);
24117 int vec_any_ne (vector signed char, vector signed char);
24118 int vec_any_ne (vector unsigned char, vector bool char);
24119 int vec_any_ne (vector unsigned char, vector unsigned char);
24120 int vec_any_ne (vector bool char, vector bool char);
24121 int vec_any_ne (vector bool char, vector unsigned char);
24122 int vec_any_ne (vector bool char, vector signed char);
24123 int vec_any_ne (vector signed short, vector bool short);
24124 int vec_any_ne (vector signed short, vector signed short);
24125 int vec_any_ne (vector unsigned short, vector bool short);
24126 int vec_any_ne (vector unsigned short, vector unsigned short);
24127 int vec_any_ne (vector bool short, vector bool short);
24128 int vec_any_ne (vector bool short, vector unsigned short);
24129 int vec_any_ne (vector bool short, vector signed short);
24130 int vec_any_ne (vector pixel, vector pixel);
24131 int vec_any_ne (vector signed int, vector bool int);
24132 int vec_any_ne (vector signed int, vector signed int);
24133 int vec_any_ne (vector unsigned int, vector bool int);
24134 int vec_any_ne (vector unsigned int, vector unsigned int);
24135 int vec_any_ne (vector bool int, vector bool int);
24136 int vec_any_ne (vector bool int, vector unsigned int);
24137 int vec_any_ne (vector bool int, vector signed int);
24138 int vec_any_ne (vector float, vector float);
24140 int vec_any_nge (vector float, vector float);
24142 int vec_any_ngt (vector float, vector float);
24144 int vec_any_nle (vector float, vector float);
24146 int vec_any_nlt (vector float, vector float);
24148 int vec_any_numeric (vector float);
24150 int vec_any_out (vector float, vector float);
24153 File: gcc.info, Node: SPARC VIS Built-in Functions, Prev: PowerPC AltiVec Built-in Functions, Up: Target Builtins
24155 5.48.9 SPARC VIS Built-in Functions
24156 -----------------------------------
24158 GCC supports SIMD operations on the SPARC using both the generic vector
24159 extensions (*note Vector Extensions::) as well as built-in functions for
24160 the SPARC Visual Instruction Set (VIS). When you use the `-mvis'
24161 switch, the VIS extension is exposed as the following built-in
24164 typedef int v2si __attribute__ ((vector_size (8)));
24165 typedef short v4hi __attribute__ ((vector_size (8)));
24166 typedef short v2hi __attribute__ ((vector_size (4)));
24167 typedef char v8qi __attribute__ ((vector_size (8)));
24168 typedef char v4qi __attribute__ ((vector_size (4)));
24170 void * __builtin_vis_alignaddr (void *, long);
24171 int64_t __builtin_vis_faligndatadi (int64_t, int64_t);
24172 v2si __builtin_vis_faligndatav2si (v2si, v2si);
24173 v4hi __builtin_vis_faligndatav4hi (v4si, v4si);
24174 v8qi __builtin_vis_faligndatav8qi (v8qi, v8qi);
24176 v4hi __builtin_vis_fexpand (v4qi);
24178 v4hi __builtin_vis_fmul8x16 (v4qi, v4hi);
24179 v4hi __builtin_vis_fmul8x16au (v4qi, v4hi);
24180 v4hi __builtin_vis_fmul8x16al (v4qi, v4hi);
24181 v4hi __builtin_vis_fmul8sux16 (v8qi, v4hi);
24182 v4hi __builtin_vis_fmul8ulx16 (v8qi, v4hi);
24183 v2si __builtin_vis_fmuld8sux16 (v4qi, v2hi);
24184 v2si __builtin_vis_fmuld8ulx16 (v4qi, v2hi);
24186 v4qi __builtin_vis_fpack16 (v4hi);
24187 v8qi __builtin_vis_fpack32 (v2si, v2si);
24188 v2hi __builtin_vis_fpackfix (v2si);
24189 v8qi __builtin_vis_fpmerge (v4qi, v4qi);
24191 int64_t __builtin_vis_pdist (v8qi, v8qi, int64_t);
24194 File: gcc.info, Node: Target Format Checks, Next: Pragmas, Prev: Target Builtins, Up: C Extensions
24196 5.49 Format Checks Specific to Particular Target Machines
24197 =========================================================
24199 For some target machines, GCC supports additional options to the format
24200 attribute (*note Declaring Attributes of Functions: Function
24205 * Solaris Format Checks::
24208 File: gcc.info, Node: Solaris Format Checks, Up: Target Format Checks
24210 5.49.1 Solaris Format Checks
24211 ----------------------------
24213 Solaris targets support the `cmn_err' (or `__cmn_err__') format check.
24214 `cmn_err' accepts a subset of the standard `printf' conversions, and
24215 the two-argument `%b' conversion for displaying bit-fields. See the
24216 Solaris man page for `cmn_err' for more information.
24219 File: gcc.info, Node: Pragmas, Next: Unnamed Fields, Prev: Target Format Checks, Up: C Extensions
24221 5.50 Pragmas Accepted by GCC
24222 ============================
24224 GCC supports several types of pragmas, primarily in order to compile
24225 code originally written for other compilers. Note that in general we
24226 do not recommend the use of pragmas; *Note Function Attributes::, for
24227 further explanation.
24233 * RS/6000 and PowerPC Pragmas::
24235 * Solaris Pragmas::
24236 * Symbol-Renaming Pragmas::
24237 * Structure-Packing Pragmas::
24239 * Diagnostic Pragmas::
24240 * Visibility Pragmas::
24243 File: gcc.info, Node: ARM Pragmas, Next: M32C Pragmas, Up: Pragmas
24248 The ARM target defines pragmas for controlling the default addition of
24249 `long_call' and `short_call' attributes to functions. *Note Function
24250 Attributes::, for information about the effects of these attributes.
24253 Set all subsequent functions to have the `long_call' attribute.
24256 Set all subsequent functions to have the `short_call' attribute.
24259 Do not affect the `long_call' or `short_call' attributes of
24260 subsequent functions.
24263 File: gcc.info, Node: M32C Pragmas, Next: RS/6000 and PowerPC Pragmas, Prev: ARM Pragmas, Up: Pragmas
24265 5.50.2 M32C Pragmas
24266 -------------------
24269 Overrides the command line option `-memregs=' for the current
24270 file. Use with care! This pragma must be before any function in
24271 the file, and mixing different memregs values in different objects
24272 may make them incompatible. This pragma is useful when a
24273 performance-critical function uses a memreg for temporary values,
24274 as it may allow you to reduce the number of memregs used.
24278 File: gcc.info, Node: RS/6000 and PowerPC Pragmas, Next: Darwin Pragmas, Prev: M32C Pragmas, Up: Pragmas
24280 5.50.3 RS/6000 and PowerPC Pragmas
24281 ----------------------------------
24283 The RS/6000 and PowerPC targets define one pragma for controlling
24284 whether or not the `longcall' attribute is added to function
24285 declarations by default. This pragma overrides the `-mlongcall'
24286 option, but not the `longcall' and `shortcall' attributes. *Note
24287 RS/6000 and PowerPC Options::, for more information about when long
24288 calls are and are not necessary.
24291 Apply the `longcall' attribute to all subsequent function
24295 Do not apply the `longcall' attribute to subsequent function
24299 File: gcc.info, Node: Darwin Pragmas, Next: Solaris Pragmas, Prev: RS/6000 and PowerPC Pragmas, Up: Pragmas
24301 5.50.4 Darwin Pragmas
24302 ---------------------
24304 The following pragmas are available for all architectures running the
24305 Darwin operating system. These are useful for compatibility with other
24309 This pragma is accepted, but has no effect.
24311 `options align=ALIGNMENT'
24312 This pragma sets the alignment of fields in structures. The
24313 values of ALIGNMENT may be `mac68k', to emulate m68k alignment, or
24314 `power', to emulate PowerPC alignment. Uses of this pragma nest
24315 properly; to restore the previous setting, use `reset' for the
24318 `segment TOKENS...'
24319 This pragma is accepted, but has no effect.
24321 `unused (VAR [, VAR]...)'
24322 This pragma declares variables to be possibly unused. GCC will not
24323 produce warnings for the listed variables. The effect is similar
24324 to that of the `unused' attribute, except that this pragma may
24325 appear anywhere within the variables' scopes.
24328 File: gcc.info, Node: Solaris Pragmas, Next: Symbol-Renaming Pragmas, Prev: Darwin Pragmas, Up: Pragmas
24330 5.50.5 Solaris Pragmas
24331 ----------------------
24333 The Solaris target supports `#pragma redefine_extname' (*note
24334 Symbol-Renaming Pragmas::). It also supports additional `#pragma'
24335 directives for compatibility with the system compiler.
24337 `align ALIGNMENT (VARIABLE [, VARIABLE]...)'
24338 Increase the minimum alignment of each VARIABLE to ALIGNMENT.
24339 This is the same as GCC's `aligned' attribute *note Variable
24340 Attributes::). Macro expansion occurs on the arguments to this
24341 pragma when compiling C and Objective-C. It does not currently
24342 occur when compiling C++, but this is a bug which may be fixed in
24345 `fini (FUNCTION [, FUNCTION]...)'
24346 This pragma causes each listed FUNCTION to be called after main,
24347 or during shared module unloading, by adding a call to the `.fini'
24350 `init (FUNCTION [, FUNCTION]...)'
24351 This pragma causes each listed FUNCTION to be called during
24352 initialization (before `main') or during shared module loading, by
24353 adding a call to the `.init' section.
24357 File: gcc.info, Node: Symbol-Renaming Pragmas, Next: Structure-Packing Pragmas, Prev: Solaris Pragmas, Up: Pragmas
24359 5.50.6 Symbol-Renaming Pragmas
24360 ------------------------------
24362 For compatibility with the Solaris and Tru64 UNIX system headers, GCC
24363 supports two `#pragma' directives which change the name used in
24364 assembly for a given declaration. These pragmas are only available on
24365 platforms whose system headers need them. To get this effect on all
24366 platforms supported by GCC, use the asm labels extension (*note Asm
24369 `redefine_extname OLDNAME NEWNAME'
24370 This pragma gives the C function OLDNAME the assembly symbol
24371 NEWNAME. The preprocessor macro `__PRAGMA_REDEFINE_EXTNAME' will
24372 be defined if this pragma is available (currently only on Solaris).
24374 `extern_prefix STRING'
24375 This pragma causes all subsequent external function and variable
24376 declarations to have STRING prepended to their assembly symbols.
24377 This effect may be terminated with another `extern_prefix' pragma
24378 whose argument is an empty string. The preprocessor macro
24379 `__PRAGMA_EXTERN_PREFIX' will be defined if this pragma is
24380 available (currently only on Tru64 UNIX).
24382 These pragmas and the asm labels extension interact in a complicated
24383 manner. Here are some corner cases you may want to be aware of.
24385 1. Both pragmas silently apply only to declarations with external
24386 linkage. Asm labels do not have this restriction.
24388 2. In C++, both pragmas silently apply only to declarations with "C"
24389 linkage. Again, asm labels do not have this restriction.
24391 3. If any of the three ways of changing the assembly name of a
24392 declaration is applied to a declaration whose assembly name has
24393 already been determined (either by a previous use of one of these
24394 features, or because the compiler needed the assembly name in
24395 order to generate code), and the new name is different, a warning
24396 issues and the name does not change.
24398 4. The OLDNAME used by `#pragma redefine_extname' is always the
24401 5. If `#pragma extern_prefix' is in effect, and a declaration occurs
24402 with an asm label attached, the prefix is silently ignored for
24405 6. If `#pragma extern_prefix' and `#pragma redefine_extname' apply to
24406 the same declaration, whichever triggered first wins, and a
24407 warning issues if they contradict each other. (We would like to
24408 have `#pragma redefine_extname' always win, for consistency with
24409 asm labels, but if `#pragma extern_prefix' triggers first we have
24410 no way of knowing that that happened.)
24413 File: gcc.info, Node: Structure-Packing Pragmas, Next: Weak Pragmas, Prev: Symbol-Renaming Pragmas, Up: Pragmas
24415 5.50.7 Structure-Packing Pragmas
24416 --------------------------------
24418 For compatibility with Win32, GCC supports a set of `#pragma'
24419 directives which change the maximum alignment of members of structures
24420 (other than zero-width bitfields), unions, and classes subsequently
24421 defined. The N value below always is required to be a small power of
24422 two and specifies the new alignment in bytes.
24424 1. `#pragma pack(N)' simply sets the new alignment.
24426 2. `#pragma pack()' sets the alignment to the one that was in effect
24427 when compilation started (see also command line option
24428 `-fpack-struct[=<n>]' *note Code Gen Options::).
24430 3. `#pragma pack(push[,N])' pushes the current alignment setting on
24431 an internal stack and then optionally sets the new alignment.
24433 4. `#pragma pack(pop)' restores the alignment setting to the one
24434 saved at the top of the internal stack (and removes that stack
24435 entry). Note that `#pragma pack([N])' does not influence this
24436 internal stack; thus it is possible to have `#pragma pack(push)'
24437 followed by multiple `#pragma pack(N)' instances and finalized by
24438 a single `#pragma pack(pop)'.
24440 Some targets, e.g. i386 and powerpc, support the `ms_struct' `#pragma'
24441 which lays out a structure as the documented `__attribute__
24443 1. `#pragma ms_struct on' turns on the layout for structures declared.
24445 2. `#pragma ms_struct off' turns off the layout for structures
24448 3. `#pragma ms_struct reset' goes back to the default layout.
24451 File: gcc.info, Node: Weak Pragmas, Next: Diagnostic Pragmas, Prev: Structure-Packing Pragmas, Up: Pragmas
24453 5.50.8 Weak Pragmas
24454 -------------------
24456 For compatibility with SVR4, GCC supports a set of `#pragma' directives
24457 for declaring symbols to be weak, and defining weak aliases.
24459 `#pragma weak SYMBOL'
24460 This pragma declares SYMBOL to be weak, as if the declaration had
24461 the attribute of the same name. The pragma may appear before or
24462 after the declaration of SYMBOL, but must appear before either its
24463 first use or its definition. It is not an error for SYMBOL to
24464 never be defined at all.
24466 `#pragma weak SYMBOL1 = SYMBOL2'
24467 This pragma declares SYMBOL1 to be a weak alias of SYMBOL2. It is
24468 an error if SYMBOL2 is not defined in the current translation unit.
24471 File: gcc.info, Node: Diagnostic Pragmas, Next: Visibility Pragmas, Prev: Weak Pragmas, Up: Pragmas
24473 5.50.9 Diagnostic Pragmas
24474 -------------------------
24476 GCC allows the user to selectively enable or disable certain types of
24477 diagnostics, and change the kind of the diagnostic. For example, a
24478 project's policy might require that all sources compile with `-Werror'
24479 but certain files might have exceptions allowing specific types of
24480 warnings. Or, a project might selectively enable diagnostics and treat
24481 them as errors depending on which preprocessor macros are defined.
24483 `#pragma GCC diagnostic KIND OPTION'
24484 Modifies the disposition of a diagnostic. Note that not all
24485 diagnostics are modifiable; at the moment only warnings (normally
24486 controlled by `-W...') can be controlled, and not all of them.
24487 Use `-fdiagnostics-show-option' to determine which diagnostics are
24488 controllable and which option controls them.
24490 KIND is `error' to treat this diagnostic as an error, `warning' to
24491 treat it like a warning (even if `-Werror' is in effect), or
24492 `ignored' if the diagnostic is to be ignored. OPTION is a double
24493 quoted string which matches the command line option.
24495 #pragma GCC diagnostic warning "-Wformat"
24496 #pragma GCC diagnostic error "-Wformat"
24497 #pragma GCC diagnostic ignored "-Wformat"
24499 Note that these pragmas override any command line options. Also,
24500 while it is syntactically valid to put these pragmas anywhere in
24501 your sources, the only supported location for them is before any
24502 data or functions are defined. Doing otherwise may result in
24503 unpredictable results depending on how the optimizer manages your
24504 sources. If the same option is listed multiple times, the last
24505 one specified is the one that is in effect. This pragma is not
24506 intended to be a general purpose replacement for command line
24507 options, but for implementing strict control over project policies.
24511 File: gcc.info, Node: Visibility Pragmas, Prev: Diagnostic Pragmas, Up: Pragmas
24513 5.50.10 Visibility Pragmas
24514 --------------------------
24516 `#pragma GCC visibility push(VISIBILITY)'
24517 `#pragma GCC visibility pop'
24518 This pragma allows the user to set the visibility for multiple
24519 declarations without having to give each a visibility attribute
24520 *Note Function Attributes::, for more information about visibility
24521 and the attribute syntax.
24523 In C++, `#pragma GCC visibility' affects only namespace-scope
24524 declarations. Class members and template specializations are not
24525 affected; if you want to override the visibility for a particular
24526 member or instantiation, you must use an attribute.
24530 File: gcc.info, Node: Unnamed Fields, Next: Thread-Local, Prev: Pragmas, Up: C Extensions
24532 5.51 Unnamed struct/union fields within structs/unions
24533 ======================================================
24535 For compatibility with other compilers, GCC allows you to define a
24536 structure or union that contains, as fields, structures and unions
24537 without names. For example:
24548 In this example, the user would be able to access members of the
24549 unnamed union with code like `foo.b'. Note that only unnamed structs
24550 and unions are allowed, you may not have, for example, an unnamed `int'.
24552 You must never create such structures that cause ambiguous field
24553 definitions. For example, this structure:
24562 It is ambiguous which `a' is being referred to with `foo.a'. Such
24563 constructs are not supported and must be avoided. In the future, such
24564 constructs may be detected and treated as compilation errors.
24566 Unless `-fms-extensions' is used, the unnamed field must be a
24567 structure or union definition without a tag (for example, `struct { int
24568 a; };'). If `-fms-extensions' is used, the field may also be a
24569 definition with a tag such as `struct foo { int a; };', a reference to
24570 a previously defined structure or union such as `struct foo;', or a
24571 reference to a `typedef' name for a previously defined structure or
24575 File: gcc.info, Node: Thread-Local, Prev: Unnamed Fields, Up: C Extensions
24577 5.52 Thread-Local Storage
24578 =========================
24580 Thread-local storage (TLS) is a mechanism by which variables are
24581 allocated such that there is one instance of the variable per extant
24582 thread. The run-time model GCC uses to implement this originates in
24583 the IA-64 processor-specific ABI, but has since been migrated to other
24584 processors as well. It requires significant support from the linker
24585 (`ld'), dynamic linker (`ld.so'), and system libraries (`libc.so' and
24586 `libpthread.so'), so it is not available everywhere.
24588 At the user level, the extension is visible with a new storage class
24589 keyword: `__thread'. For example:
24592 extern __thread struct state s;
24593 static __thread char *p;
24595 The `__thread' specifier may be used alone, with the `extern' or
24596 `static' specifiers, but with no other storage class specifier. When
24597 used with `extern' or `static', `__thread' must appear immediately
24598 after the other storage class specifier.
24600 The `__thread' specifier may be applied to any global, file-scoped
24601 static, function-scoped static, or static data member of a class. It
24602 may not be applied to block-scoped automatic or non-static data member.
24604 When the address-of operator is applied to a thread-local variable, it
24605 is evaluated at run-time and returns the address of the current thread's
24606 instance of that variable. An address so obtained may be used by any
24607 thread. When a thread terminates, any pointers to thread-local
24608 variables in that thread become invalid.
24610 No static initialization may refer to the address of a thread-local
24613 In C++, if an initializer is present for a thread-local variable, it
24614 must be a CONSTANT-EXPRESSION, as defined in 5.19.2 of the ANSI/ISO C++
24617 See ELF Handling For Thread-Local Storage
24618 (http://people.redhat.com/drepper/tls.pdf) for a detailed explanation of
24619 the four thread-local storage addressing models, and how the run-time
24620 is expected to function.
24624 * C99 Thread-Local Edits::
24625 * C++98 Thread-Local Edits::
24628 File: gcc.info, Node: C99 Thread-Local Edits, Next: C++98 Thread-Local Edits, Up: Thread-Local
24630 5.52.1 ISO/IEC 9899:1999 Edits for Thread-Local Storage
24631 -------------------------------------------------------
24633 The following are a set of changes to ISO/IEC 9899:1999 (aka C99) that
24634 document the exact semantics of the language extension.
24636 * `5.1.2 Execution environments'
24638 Add new text after paragraph 1
24640 Within either execution environment, a "thread" is a flow of
24641 control within a program. It is implementation defined
24642 whether or not there may be more than one thread associated
24643 with a program. It is implementation defined how threads
24644 beyond the first are created, the name and type of the
24645 function called at thread startup, and how threads may be
24646 terminated. However, objects with thread storage duration
24647 shall be initialized before thread startup.
24649 * `6.2.4 Storage durations of objects'
24651 Add new text before paragraph 3
24653 An object whose identifier is declared with the storage-class
24654 specifier `__thread' has "thread storage duration". Its
24655 lifetime is the entire execution of the thread, and its
24656 stored value is initialized only once, prior to thread
24663 * `6.7.1 Storage-class specifiers'
24665 Add `__thread' to the list of storage class specifiers in
24668 Change paragraph 2 to
24670 With the exception of `__thread', at most one storage-class
24671 specifier may be given [...]. The `__thread' specifier may
24672 be used alone, or immediately following `extern' or `static'.
24674 Add new text after paragraph 6
24676 The declaration of an identifier for a variable that has
24677 block scope that specifies `__thread' shall also specify
24678 either `extern' or `static'.
24680 The `__thread' specifier shall be used only with variables.
24683 File: gcc.info, Node: C++98 Thread-Local Edits, Prev: C99 Thread-Local Edits, Up: Thread-Local
24685 5.52.2 ISO/IEC 14882:1998 Edits for Thread-Local Storage
24686 --------------------------------------------------------
24688 The following are a set of changes to ISO/IEC 14882:1998 (aka C++98)
24689 that document the exact semantics of the language extension.
24691 * [intro.execution]
24693 New text after paragraph 4
24695 A "thread" is a flow of control within the abstract machine.
24696 It is implementation defined whether or not there may be more
24699 New text after paragraph 7
24701 It is unspecified whether additional action must be taken to
24702 ensure when and whether side effects are visible to other
24709 * [basic.start.main]
24711 Add after paragraph 5
24713 The thread that begins execution at the `main' function is
24714 called the "main thread". It is implementation defined how
24715 functions beginning threads other than the main thread are
24716 designated or typed. A function so designated, as well as
24717 the `main' function, is called a "thread startup function".
24718 It is implementation defined what happens if a thread startup
24719 function returns. It is implementation defined what happens
24720 to other threads when any thread calls `exit'.
24722 * [basic.start.init]
24724 Add after paragraph 4
24726 The storage for an object of thread storage duration shall be
24727 statically initialized before the first statement of the
24728 thread startup function. An object of thread storage
24729 duration shall not require dynamic initialization.
24731 * [basic.start.term]
24733 Add after paragraph 3
24735 The type of an object with thread storage duration shall not
24736 have a non-trivial destructor, nor shall it be an array type
24737 whose elements (directly or indirectly) have non-trivial
24742 Add "thread storage duration" to the list in paragraph 1.
24746 Thread, static, and automatic storage durations are
24747 associated with objects introduced by declarations [...].
24749 Add `__thread' to the list of specifiers in paragraph 3.
24751 * [basic.stc.thread]
24753 New section before [basic.stc.static]
24755 The keyword `__thread' applied to a non-local object gives the
24756 object thread storage duration.
24758 A local variable or class data member declared both `static'
24759 and `__thread' gives the variable or member thread storage
24762 * [basic.stc.static]
24766 All objects which have neither thread storage duration,
24767 dynamic storage duration nor are local [...].
24771 Add `__thread' to the list in paragraph 1.
24775 With the exception of `__thread', at most one
24776 STORAGE-CLASS-SPECIFIER shall appear in a given
24777 DECL-SPECIFIER-SEQ. The `__thread' specifier may be used
24778 alone, or immediately following the `extern' or `static'
24781 Add after paragraph 5
24783 The `__thread' specifier can be applied only to the names of
24784 objects and to anonymous unions.
24788 Add after paragraph 6
24790 Non-`static' members shall not be `__thread'.
24793 File: gcc.info, Node: C++ Extensions, Next: Objective-C, Prev: C Extensions, Up: Top
24795 6 Extensions to the C++ Language
24796 ********************************
24798 The GNU compiler provides these extensions to the C++ language (and you
24799 can also use most of the C language extensions in your C++ programs).
24800 If you want to write code that checks whether these features are
24801 available, you can test for the GNU compiler the same way as for C
24802 programs: check for a predefined macro `__GNUC__'. You can also use
24803 `__GNUG__' to test specifically for GNU C++ (*note Predefined Macros:
24804 (cpp)Common Predefined Macros.).
24808 * Volatiles:: What constitutes an access to a volatile object.
24809 * Restricted Pointers:: C99 restricted pointers and references.
24810 * Vague Linkage:: Where G++ puts inlines, vtables and such.
24811 * C++ Interface:: You can use a single C++ header file for both
24812 declarations and definitions.
24813 * Template Instantiation:: Methods for ensuring that exactly one copy of
24814 each needed template instantiation is emitted.
24815 * Bound member functions:: You can extract a function pointer to the
24816 method denoted by a `->*' or `.*' expression.
24817 * C++ Attributes:: Variable, function, and type attributes for C++ only.
24818 * Namespace Association:: Strong using-directives for namespace association.
24819 * Java Exceptions:: Tweaking exception handling to work with Java.
24820 * Deprecated Features:: Things will disappear from g++.
24821 * Backwards Compatibility:: Compatibilities with earlier definitions of C++.
24824 File: gcc.info, Node: Volatiles, Next: Restricted Pointers, Up: C++ Extensions
24826 6.1 When is a Volatile Object Accessed?
24827 =======================================
24829 Both the C and C++ standard have the concept of volatile objects. These
24830 are normally accessed by pointers and used for accessing hardware. The
24831 standards encourage compilers to refrain from optimizations concerning
24832 accesses to volatile objects. The C standard leaves it implementation
24833 defined as to what constitutes a volatile access. The C++ standard
24834 omits to specify this, except to say that C++ should behave in a
24835 similar manner to C with respect to volatiles, where possible. The
24836 minimum either standard specifies is that at a sequence point all
24837 previous accesses to volatile objects have stabilized and no subsequent
24838 accesses have occurred. Thus an implementation is free to reorder and
24839 combine volatile accesses which occur between sequence points, but
24840 cannot do so for accesses across a sequence point. The use of
24841 volatiles does not allow you to violate the restriction on updating
24842 objects multiple times within a sequence point.
24844 *Note Volatile qualifier and the C compiler: Qualifiers implementation.
24846 The behavior differs slightly between C and C++ in the non-obvious
24849 volatile int *src = SOMEVALUE;
24852 With C, such expressions are rvalues, and GCC interprets this either
24853 as a read of the volatile object being pointed to or only as request to
24854 evaluate the side-effects. The C++ standard specifies that such
24855 expressions do not undergo lvalue to rvalue conversion, and that the
24856 type of the dereferenced object may be incomplete. The C++ standard
24857 does not specify explicitly that it is this lvalue to rvalue conversion
24858 which may be responsible for causing an access. However, there is
24859 reason to believe that it is, because otherwise certain simple
24860 expressions become undefined. However, because it would surprise most
24861 programmers, G++ treats dereferencing a pointer to volatile object of
24862 complete type when the value is unused as GCC would do for an
24863 equivalent type in C. When the object has incomplete type, G++ issues
24864 a warning; if you wish to force an error, you must force a conversion
24865 to rvalue with, for instance, a static cast.
24867 When using a reference to volatile, G++ does not treat equivalent
24868 expressions as accesses to volatiles, but instead issues a warning that
24869 no volatile is accessed. The rationale for this is that otherwise it
24870 becomes difficult to determine where volatile access occur, and not
24871 possible to ignore the return value from functions returning volatile
24872 references. Again, if you wish to force a read, cast the reference to
24876 File: gcc.info, Node: Restricted Pointers, Next: Vague Linkage, Prev: Volatiles, Up: C++ Extensions
24878 6.2 Restricting Pointer Aliasing
24879 ================================
24881 As with the C front end, G++ understands the C99 feature of restricted
24882 pointers, specified with the `__restrict__', or `__restrict' type
24883 qualifier. Because you cannot compile C++ by specifying the `-std=c99'
24884 language flag, `restrict' is not a keyword in C++.
24886 In addition to allowing restricted pointers, you can specify restricted
24887 references, which indicate that the reference is not aliased in the
24890 void fn (int *__restrict__ rptr, int &__restrict__ rref)
24895 In the body of `fn', RPTR points to an unaliased integer and RREF
24896 refers to a (different) unaliased integer.
24898 You may also specify whether a member function's THIS pointer is
24899 unaliased by using `__restrict__' as a member function qualifier.
24901 void T::fn () __restrict__
24906 Within the body of `T::fn', THIS will have the effective definition `T
24907 *__restrict__ const this'. Notice that the interpretation of a
24908 `__restrict__' member function qualifier is different to that of
24909 `const' or `volatile' qualifier, in that it is applied to the pointer
24910 rather than the object. This is consistent with other compilers which
24911 implement restricted pointers.
24913 As with all outermost parameter qualifiers, `__restrict__' is ignored
24914 in function definition matching. This means you only need to specify
24915 `__restrict__' in a function definition, rather than in a function
24919 File: gcc.info, Node: Vague Linkage, Next: C++ Interface, Prev: Restricted Pointers, Up: C++ Extensions
24924 There are several constructs in C++ which require space in the object
24925 file but are not clearly tied to a single translation unit. We say that
24926 these constructs have "vague linkage". Typically such constructs are
24927 emitted wherever they are needed, though sometimes we can be more
24931 Inline functions are typically defined in a header file which can
24932 be included in many different compilations. Hopefully they can
24933 usually be inlined, but sometimes an out-of-line copy is
24934 necessary, if the address of the function is taken or if inlining
24935 fails. In general, we emit an out-of-line copy in all translation
24936 units where one is needed. As an exception, we only emit inline
24937 virtual functions with the vtable, since it will always require a
24940 Local static variables and string constants used in an inline
24941 function are also considered to have vague linkage, since they
24942 must be shared between all inlined and out-of-line instances of
24946 C++ virtual functions are implemented in most compilers using a
24947 lookup table, known as a vtable. The vtable contains pointers to
24948 the virtual functions provided by a class, and each object of the
24949 class contains a pointer to its vtable (or vtables, in some
24950 multiple-inheritance situations). If the class declares any
24951 non-inline, non-pure virtual functions, the first one is chosen as
24952 the "key method" for the class, and the vtable is only emitted in
24953 the translation unit where the key method is defined.
24955 _Note:_ If the chosen key method is later defined as inline, the
24956 vtable will still be emitted in every translation unit which
24957 defines it. Make sure that any inline virtuals are declared
24958 inline in the class body, even if they are not defined there.
24961 C++ requires information about types to be written out in order to
24962 implement `dynamic_cast', `typeid' and exception handling. For
24963 polymorphic classes (classes with virtual functions), the type_info
24964 object is written out along with the vtable so that `dynamic_cast'
24965 can determine the dynamic type of a class object at runtime. For
24966 all other types, we write out the type_info object when it is
24967 used: when applying `typeid' to an expression, throwing an object,
24968 or referring to a type in a catch clause or exception
24971 Template Instantiations
24972 Most everything in this section also applies to template
24973 instantiations, but there are other options as well. *Note
24974 Where's the Template?: Template Instantiation.
24977 When used with GNU ld version 2.8 or later on an ELF system such as
24978 GNU/Linux or Solaris 2, or on Microsoft Windows, duplicate copies of
24979 these constructs will be discarded at link time. This is known as
24982 On targets that don't support COMDAT, but do support weak symbols, GCC
24983 will use them. This way one copy will override all the others, but the
24984 unused copies will still take up space in the executable.
24986 For targets which do not support either COMDAT or weak symbols, most
24987 entities with vague linkage will be emitted as local symbols to avoid
24988 duplicate definition errors from the linker. This will not happen for
24989 local statics in inlines, however, as having multiple copies will
24990 almost certainly break things.
24992 *Note Declarations and Definitions in One Header: C++ Interface, for
24993 another way to control placement of these constructs.
24996 File: gcc.info, Node: C++ Interface, Next: Template Instantiation, Prev: Vague Linkage, Up: C++ Extensions
24998 6.4 #pragma interface and implementation
24999 ========================================
25001 `#pragma interface' and `#pragma implementation' provide the user with
25002 a way of explicitly directing the compiler to emit entities with vague
25003 linkage (and debugging information) in a particular translation unit.
25005 _Note:_ As of GCC 2.7.2, these `#pragma's are not useful in most
25006 cases, because of COMDAT support and the "key method" heuristic
25007 mentioned in *Note Vague Linkage::. Using them can actually cause your
25008 program to grow due to unnecessary out-of-line copies of inline
25009 functions. Currently (3.4) the only benefit of these `#pragma's is
25010 reduced duplication of debugging information, and that should be
25011 addressed soon on DWARF 2 targets with the use of COMDAT groups.
25013 `#pragma interface'
25014 `#pragma interface "SUBDIR/OBJECTS.h"'
25015 Use this directive in _header files_ that define object classes,
25016 to save space in most of the object files that use those classes.
25017 Normally, local copies of certain information (backup copies of
25018 inline member functions, debugging information, and the internal
25019 tables that implement virtual functions) must be kept in each
25020 object file that includes class definitions. You can use this
25021 pragma to avoid such duplication. When a header file containing
25022 `#pragma interface' is included in a compilation, this auxiliary
25023 information will not be generated (unless the main input source
25024 file itself uses `#pragma implementation'). Instead, the object
25025 files will contain references to be resolved at link time.
25027 The second form of this directive is useful for the case where you
25028 have multiple headers with the same name in different directories.
25029 If you use this form, you must specify the same string to `#pragma
25032 `#pragma implementation'
25033 `#pragma implementation "OBJECTS.h"'
25034 Use this pragma in a _main input file_, when you want full output
25035 from included header files to be generated (and made globally
25036 visible). The included header file, in turn, should use `#pragma
25037 interface'. Backup copies of inline member functions, debugging
25038 information, and the internal tables used to implement virtual
25039 functions are all generated in implementation files.
25041 If you use `#pragma implementation' with no argument, it applies to
25042 an include file with the same basename(1) as your source file.
25043 For example, in `allclass.cc', giving just `#pragma implementation'
25044 by itself is equivalent to `#pragma implementation "allclass.h"'.
25046 In versions of GNU C++ prior to 2.6.0 `allclass.h' was treated as
25047 an implementation file whenever you would include it from
25048 `allclass.cc' even if you never specified `#pragma
25049 implementation'. This was deemed to be more trouble than it was
25050 worth, however, and disabled.
25052 Use the string argument if you want a single implementation file to
25053 include code from multiple header files. (You must also use
25054 `#include' to include the header file; `#pragma implementation'
25055 only specifies how to use the file--it doesn't actually include
25058 There is no way to split up the contents of a single header file
25059 into multiple implementation files.
25061 `#pragma implementation' and `#pragma interface' also have an effect
25062 on function inlining.
25064 If you define a class in a header file marked with `#pragma
25065 interface', the effect on an inline function defined in that class is
25066 similar to an explicit `extern' declaration--the compiler emits no code
25067 at all to define an independent version of the function. Its
25068 definition is used only for inlining with its callers.
25070 Conversely, when you include the same header file in a main source file
25071 that declares it as `#pragma implementation', the compiler emits code
25072 for the function itself; this defines a version of the function that
25073 can be found via pointers (or by callers compiled without inlining).
25074 If all calls to the function can be inlined, you can avoid emitting the
25075 function by compiling with `-fno-implement-inlines'. If any calls were
25076 not inlined, you will get linker errors.
25078 ---------- Footnotes ----------
25080 (1) A file's "basename" was the name stripped of all leading path
25081 information and of trailing suffixes, such as `.h' or `.C' or `.cc'.
25084 File: gcc.info, Node: Template Instantiation, Next: Bound member functions, Prev: C++ Interface, Up: C++ Extensions
25086 6.5 Where's the Template?
25087 =========================
25089 C++ templates are the first language feature to require more
25090 intelligence from the environment than one usually finds on a UNIX
25091 system. Somehow the compiler and linker have to make sure that each
25092 template instance occurs exactly once in the executable if it is needed,
25093 and not at all otherwise. There are two basic approaches to this
25094 problem, which are referred to as the Borland model and the Cfront
25098 Borland C++ solved the template instantiation problem by adding
25099 the code equivalent of common blocks to their linker; the compiler
25100 emits template instances in each translation unit that uses them,
25101 and the linker collapses them together. The advantage of this
25102 model is that the linker only has to consider the object files
25103 themselves; there is no external complexity to worry about. This
25104 disadvantage is that compilation time is increased because the
25105 template code is being compiled repeatedly. Code written for this
25106 model tends to include definitions of all templates in the header
25107 file, since they must be seen to be instantiated.
25110 The AT&T C++ translator, Cfront, solved the template instantiation
25111 problem by creating the notion of a template repository, an
25112 automatically maintained place where template instances are
25113 stored. A more modern version of the repository works as follows:
25114 As individual object files are built, the compiler places any
25115 template definitions and instantiations encountered in the
25116 repository. At link time, the link wrapper adds in the objects in
25117 the repository and compiles any needed instances that were not
25118 previously emitted. The advantages of this model are more optimal
25119 compilation speed and the ability to use the system linker; to
25120 implement the Borland model a compiler vendor also needs to
25121 replace the linker. The disadvantages are vastly increased
25122 complexity, and thus potential for error; for some code this can be
25123 just as transparent, but in practice it can been very difficult to
25124 build multiple programs in one directory and one program in
25125 multiple directories. Code written for this model tends to
25126 separate definitions of non-inline member templates into a
25127 separate file, which should be compiled separately.
25129 When used with GNU ld version 2.8 or later on an ELF system such as
25130 GNU/Linux or Solaris 2, or on Microsoft Windows, G++ supports the
25131 Borland model. On other systems, G++ implements neither automatic
25134 A future version of G++ will support a hybrid model whereby the
25135 compiler will emit any instantiations for which the template definition
25136 is included in the compile, and store template definitions and
25137 instantiation context information into the object file for the rest.
25138 The link wrapper will extract that information as necessary and invoke
25139 the compiler to produce the remaining instantiations. The linker will
25140 then combine duplicate instantiations.
25142 In the mean time, you have the following options for dealing with
25143 template instantiations:
25145 1. Compile your template-using code with `-frepo'. The compiler will
25146 generate files with the extension `.rpo' listing all of the
25147 template instantiations used in the corresponding object files
25148 which could be instantiated there; the link wrapper, `collect2',
25149 will then update the `.rpo' files to tell the compiler where to
25150 place those instantiations and rebuild any affected object files.
25151 The link-time overhead is negligible after the first pass, as the
25152 compiler will continue to place the instantiations in the same
25155 This is your best option for application code written for the
25156 Borland model, as it will just work. Code written for the Cfront
25157 model will need to be modified so that the template definitions
25158 are available at one or more points of instantiation; usually this
25159 is as simple as adding `#include <tmethods.cc>' to the end of each
25162 For library code, if you want the library to provide all of the
25163 template instantiations it needs, just try to link all of its
25164 object files together; the link will fail, but cause the
25165 instantiations to be generated as a side effect. Be warned,
25166 however, that this may cause conflicts if multiple libraries try
25167 to provide the same instantiations. For greater control, use
25168 explicit instantiation as described in the next option.
25170 2. Compile your code with `-fno-implicit-templates' to disable the
25171 implicit generation of template instances, and explicitly
25172 instantiate all the ones you use. This approach requires more
25173 knowledge of exactly which instances you need than do the others,
25174 but it's less mysterious and allows greater control. You can
25175 scatter the explicit instantiations throughout your program,
25176 perhaps putting them in the translation units where the instances
25177 are used or the translation units that define the templates
25178 themselves; you can put all of the explicit instantiations you
25179 need into one big file; or you can create small files like
25184 template class Foo<int>;
25185 template ostream& operator <<
25186 (ostream&, const Foo<int>&);
25188 for each of the instances you need, and create a template
25189 instantiation library from those.
25191 If you are using Cfront-model code, you can probably get away with
25192 not using `-fno-implicit-templates' when compiling files that don't
25193 `#include' the member template definitions.
25195 If you use one big file to do the instantiations, you may want to
25196 compile it without `-fno-implicit-templates' so you get all of the
25197 instances required by your explicit instantiations (but not by any
25198 other files) without having to specify them as well.
25200 G++ has extended the template instantiation syntax given in the ISO
25201 standard to allow forward declaration of explicit instantiations
25202 (with `extern'), instantiation of the compiler support data for a
25203 template class (i.e. the vtable) without instantiating any of its
25204 members (with `inline'), and instantiation of only the static data
25205 members of a template class, without the support data or member
25206 functions (with (`static'):
25208 extern template int max (int, int);
25209 inline template class Foo<int>;
25210 static template class Foo<int>;
25212 3. Do nothing. Pretend G++ does implement automatic instantiation
25213 management. Code written for the Borland model will work fine, but
25214 each translation unit will contain instances of each of the
25215 templates it uses. In a large program, this can lead to an
25216 unacceptable amount of code duplication.
25219 File: gcc.info, Node: Bound member functions, Next: C++ Attributes, Prev: Template Instantiation, Up: C++ Extensions
25221 6.6 Extracting the function pointer from a bound pointer to member function
25222 ===========================================================================
25224 In C++, pointer to member functions (PMFs) are implemented using a wide
25225 pointer of sorts to handle all the possible call mechanisms; the PMF
25226 needs to store information about how to adjust the `this' pointer, and
25227 if the function pointed to is virtual, where to find the vtable, and
25228 where in the vtable to look for the member function. If you are using
25229 PMFs in an inner loop, you should really reconsider that decision. If
25230 that is not an option, you can extract the pointer to the function that
25231 would be called for a given object/PMF pair and call it directly inside
25232 the inner loop, to save a bit of time.
25234 Note that you will still be paying the penalty for the call through a
25235 function pointer; on most modern architectures, such a call defeats the
25236 branch prediction features of the CPU. This is also true of normal
25237 virtual function calls.
25239 The syntax for this extension is
25242 extern int (A::*fp)();
25243 typedef int (*fptr)(A *);
25245 fptr p = (fptr)(a.*fp);
25247 For PMF constants (i.e. expressions of the form `&Klasse::Member'), no
25248 object is needed to obtain the address of the function. They can be
25249 converted to function pointers directly:
25251 fptr p1 = (fptr)(&A::foo);
25253 You must specify `-Wno-pmf-conversions' to use this extension.
25256 File: gcc.info, Node: C++ Attributes, Next: Namespace Association, Prev: Bound member functions, Up: C++ Extensions
25258 6.7 C++-Specific Variable, Function, and Type Attributes
25259 ========================================================
25261 Some attributes only make sense for C++ programs.
25263 `init_priority (PRIORITY)'
25264 In Standard C++, objects defined at namespace scope are guaranteed
25265 to be initialized in an order in strict accordance with that of
25266 their definitions _in a given translation unit_. No guarantee is
25267 made for initializations across translation units. However, GNU
25268 C++ allows users to control the order of initialization of objects
25269 defined at namespace scope with the `init_priority' attribute by
25270 specifying a relative PRIORITY, a constant integral expression
25271 currently bounded between 101 and 65535 inclusive. Lower numbers
25272 indicate a higher priority.
25274 In the following example, `A' would normally be created before
25275 `B', but the `init_priority' attribute has reversed that order:
25277 Some_Class A __attribute__ ((init_priority (2000)));
25278 Some_Class B __attribute__ ((init_priority (543)));
25280 Note that the particular values of PRIORITY do not matter; only
25281 their relative ordering.
25284 This type attribute informs C++ that the class is a Java
25285 interface. It may only be applied to classes declared within an
25286 `extern "Java"' block. Calls to methods declared in this
25287 interface will be dispatched using GCJ's interface table
25288 mechanism, instead of regular virtual table dispatch.
25291 See also *Note Namespace Association::.
25294 File: gcc.info, Node: Namespace Association, Next: Java Exceptions, Prev: C++ Attributes, Up: C++ Extensions
25296 6.8 Namespace Association
25297 =========================
25299 *Caution:* The semantics of this extension are not fully defined.
25300 Users should refrain from using this extension as its semantics may
25301 change subtly over time. It is possible that this extension will be
25302 removed in future versions of G++.
25304 A using-directive with `__attribute ((strong))' is stronger than a
25305 normal using-directive in two ways:
25307 * Templates from the used namespace can be specialized and explicitly
25308 instantiated as though they were members of the using namespace.
25310 * The using namespace is considered an associated namespace of all
25311 templates in the used namespace for purposes of argument-dependent
25314 The used namespace must be nested within the using namespace so that
25315 normal unqualified lookup works properly.
25317 This is useful for composing a namespace transparently from
25318 implementation namespaces. For example:
25322 template <class T> struct A { };
25324 using namespace debug __attribute ((__strong__));
25325 template <> struct A<int> { }; // ok to specialize
25327 template <class T> void f (A<T>);
25332 f (std::A<float>()); // lookup finds std::f
25337 File: gcc.info, Node: Java Exceptions, Next: Deprecated Features, Prev: Namespace Association, Up: C++ Extensions
25339 6.9 Java Exceptions
25340 ===================
25342 The Java language uses a slightly different exception handling model
25343 from C++. Normally, GNU C++ will automatically detect when you are
25344 writing C++ code that uses Java exceptions, and handle them
25345 appropriately. However, if C++ code only needs to execute destructors
25346 when Java exceptions are thrown through it, GCC will guess incorrectly.
25347 Sample problematic code is:
25349 struct S { ~S(); };
25350 extern void bar(); // is written in Java, and may throw exceptions
25357 The usual effect of an incorrect guess is a link failure, complaining of
25358 a missing routine called `__gxx_personality_v0'.
25360 You can inform the compiler that Java exceptions are to be used in a
25361 translation unit, irrespective of what it might think, by writing
25362 `#pragma GCC java_exceptions' at the head of the file. This `#pragma'
25363 must appear before any functions that throw or catch exceptions, or run
25364 destructors when exceptions are thrown through them.
25366 You cannot mix Java and C++ exceptions in the same translation unit.
25367 It is believed to be safe to throw a C++ exception from one file through
25368 another file compiled for the Java exception model, or vice versa, but
25369 there may be bugs in this area.
25372 File: gcc.info, Node: Deprecated Features, Next: Backwards Compatibility, Prev: Java Exceptions, Up: C++ Extensions
25374 6.10 Deprecated Features
25375 ========================
25377 In the past, the GNU C++ compiler was extended to experiment with new
25378 features, at a time when the C++ language was still evolving. Now that
25379 the C++ standard is complete, some of those features are superseded by
25380 superior alternatives. Using the old features might cause a warning in
25381 some cases that the feature will be dropped in the future. In other
25382 cases, the feature might be gone already.
25384 While the list below is not exhaustive, it documents some of the
25385 options that are now deprecated:
25387 `-fexternal-templates'
25388 `-falt-external-templates'
25389 These are two of the many ways for G++ to implement template
25390 instantiation. *Note Template Instantiation::. The C++ standard
25391 clearly defines how template definitions have to be organized
25392 across implementation units. G++ has an implicit instantiation
25393 mechanism that should work just fine for standard-conforming code.
25395 `-fstrict-prototype'
25396 `-fno-strict-prototype'
25397 Previously it was possible to use an empty prototype parameter
25398 list to indicate an unspecified number of parameters (like C),
25399 rather than no parameters, as C++ demands. This feature has been
25400 removed, except where it is required for backwards compatibility
25401 *Note Backwards Compatibility::.
25403 G++ allows a virtual function returning `void *' to be overridden by
25404 one returning a different pointer type. This extension to the
25405 covariant return type rules is now deprecated and will be removed from a
25408 The G++ minimum and maximum operators (`<?' and `>?') and their
25409 compound forms (`<?=') and `>?=') have been deprecated and will be
25410 removed in a future version. Code using these operators should be
25411 modified to use `std::min' and `std::max' instead.
25413 The named return value extension has been deprecated, and is now
25416 The use of initializer lists with new expressions has been deprecated,
25417 and is now removed from G++.
25419 Floating and complex non-type template parameters have been deprecated,
25420 and are now removed from G++.
25422 The implicit typename extension has been deprecated and is now removed
25425 The use of default arguments in function pointers, function typedefs
25426 and other places where they are not permitted by the standard is
25427 deprecated and will be removed from a future version of G++.
25429 G++ allows floating-point literals to appear in integral constant
25430 expressions, e.g. ` enum E { e = int(2.2 * 3.7) } ' This extension is
25431 deprecated and will be removed from a future version.
25433 G++ allows static data members of const floating-point type to be
25434 declared with an initializer in a class definition. The standard only
25435 allows initializers for static members of const integral types and const
25436 enumeration types so this extension has been deprecated and will be
25437 removed from a future version.
25440 File: gcc.info, Node: Backwards Compatibility, Prev: Deprecated Features, Up: C++ Extensions
25442 6.11 Backwards Compatibility
25443 ============================
25445 Now that there is a definitive ISO standard C++, G++ has a specification
25446 to adhere to. The C++ language evolved over time, and features that
25447 used to be acceptable in previous drafts of the standard, such as the
25448 ARM [Annotated C++ Reference Manual], are no longer accepted. In order
25449 to allow compilation of C++ written to such drafts, G++ contains some
25450 backwards compatibilities. _All such backwards compatibility features
25451 are liable to disappear in future versions of G++._ They should be
25452 considered deprecated *Note Deprecated Features::.
25455 If a variable is declared at for scope, it used to remain in scope
25456 until the end of the scope which contained the for statement
25457 (rather than just within the for scope). G++ retains this, but
25458 issues a warning, if such a variable is accessed outside the for
25461 `Implicit C language'
25462 Old C system header files did not contain an `extern "C" {...}'
25463 scope to set the language. On such systems, all header files are
25464 implicitly scoped inside a C language scope. Also, an empty
25465 prototype `()' will be treated as an unspecified number of
25466 arguments, rather than no arguments, as C++ demands.
25469 File: gcc.info, Node: Objective-C, Next: Compatibility, Prev: C++ Extensions, Up: Top
25471 7 GNU Objective-C runtime features
25472 **********************************
25474 This document is meant to describe some of the GNU Objective-C runtime
25475 features. It is not intended to teach you Objective-C, there are
25476 several resources on the Internet that present the language. Questions
25477 and comments about this document to Ovidiu Predescu <ovidiu@cup.hp.com>.
25481 * Executing code before main::
25483 * Garbage Collection::
25484 * Constant string objects::
25485 * compatibility_alias::
25488 File: gcc.info, Node: Executing code before main, Next: Type encoding, Prev: Objective-C, Up: Objective-C
25490 7.1 `+load': Executing code before main
25491 =======================================
25493 The GNU Objective-C runtime provides a way that allows you to execute
25494 code before the execution of the program enters the `main' function.
25495 The code is executed on a per-class and a per-category basis, through a
25496 special class method `+load'.
25498 This facility is very useful if you want to initialize global variables
25499 which can be accessed by the program directly, without sending a message
25500 to the class first. The usual way to initialize global variables, in
25501 the `+initialize' method, might not be useful because `+initialize' is
25502 only called when the first message is sent to a class object, which in
25503 some cases could be too late.
25505 Suppose for example you have a `FileStream' class that declares
25506 `Stdin', `Stdout' and `Stderr' as global variables, like below:
25509 FileStream *Stdin = nil;
25510 FileStream *Stdout = nil;
25511 FileStream *Stderr = nil;
25513 @implementation FileStream
25517 Stdin = [[FileStream new] initWithFd:0];
25518 Stdout = [[FileStream new] initWithFd:1];
25519 Stderr = [[FileStream new] initWithFd:2];
25522 /* Other methods here */
25525 In this example, the initialization of `Stdin', `Stdout' and `Stderr'
25526 in `+initialize' occurs too late. The programmer can send a message to
25527 one of these objects before the variables are actually initialized,
25528 thus sending messages to the `nil' object. The `+initialize' method
25529 which actually initializes the global variables is not invoked until
25530 the first message is sent to the class object. The solution would
25531 require these variables to be initialized just before entering `main'.
25533 The correct solution of the above problem is to use the `+load' method
25534 instead of `+initialize':
25537 @implementation FileStream
25541 Stdin = [[FileStream new] initWithFd:0];
25542 Stdout = [[FileStream new] initWithFd:1];
25543 Stderr = [[FileStream new] initWithFd:2];
25546 /* Other methods here */
25549 The `+load' is a method that is not overridden by categories. If a
25550 class and a category of it both implement `+load', both methods are
25551 invoked. This allows some additional initializations to be performed in
25554 This mechanism is not intended to be a replacement for `+initialize'.
25555 You should be aware of its limitations when you decide to use it
25556 instead of `+initialize'.
25560 * What you can and what you cannot do in +load::
25563 File: gcc.info, Node: What you can and what you cannot do in +load, Prev: Executing code before main, Up: Executing code before main
25565 7.1.1 What you can and what you cannot do in `+load'
25566 ----------------------------------------------------
25568 The `+load' implementation in the GNU runtime guarantees you the
25571 * you can write whatever C code you like;
25573 * you can send messages to Objective-C constant strings (`@"this is a
25574 constant string"');
25576 * you can allocate and send messages to objects whose class is
25577 implemented in the same file;
25579 * the `+load' implementation of all super classes of a class are
25580 executed before the `+load' of that class is executed;
25582 * the `+load' implementation of a class is executed before the
25583 `+load' implementation of any category.
25586 In particular, the following things, even if they can work in a
25587 particular case, are not guaranteed:
25589 * allocation of or sending messages to arbitrary objects;
25591 * allocation of or sending messages to objects whose classes have a
25592 category implemented in the same file;
25595 You should make no assumptions about receiving `+load' in sibling
25596 classes when you write `+load' of a class. The order in which sibling
25597 classes receive `+load' is not guaranteed.
25599 The order in which `+load' and `+initialize' are called could be
25600 problematic if this matters. If you don't allocate objects inside
25601 `+load', it is guaranteed that `+load' is called before `+initialize'.
25602 If you create an object inside `+load' the `+initialize' method of
25603 object's class is invoked even if `+load' was not invoked. Note if you
25604 explicitly call `+load' on a class, `+initialize' will be called first.
25605 To avoid possible problems try to implement only one of these methods.
25607 The `+load' method is also invoked when a bundle is dynamically loaded
25608 into your running program. This happens automatically without any
25609 intervening operation from you. When you write bundles and you need to
25610 write `+load' you can safely create and send messages to objects whose
25611 classes already exist in the running program. The same restrictions as
25612 above apply to classes defined in bundle.
25615 File: gcc.info, Node: Type encoding, Next: Garbage Collection, Prev: Executing code before main, Up: Objective-C
25620 The Objective-C compiler generates type encodings for all the types.
25621 These type encodings are used at runtime to find out information about
25622 selectors and methods and about objects and classes.
25624 The types are encoded in the following way:
25628 `unsigned char' `C'
25630 `unsigned short' `S'
25634 `unsigned long' `L'
25646 Complex types `j' followed by the inner type. For example
25647 `_Complex double' is encoded as "jd".
25648 bit-fields `b' followed by the starting position of the
25649 bit-field, the type of the bit-field and the size of
25650 the bit-field (the bit-fields encoding was changed
25651 from the NeXT's compiler encoding, see below)
25653 The encoding of bit-fields has changed to allow bit-fields to be
25654 properly handled by the runtime functions that compute sizes and
25655 alignments of types that contain bit-fields. The previous encoding
25656 contained only the size of the bit-field. Using only this information
25657 it is not possible to reliably compute the size occupied by the
25658 bit-field. This is very important in the presence of the Boehm's
25659 garbage collector because the objects are allocated using the typed
25660 memory facility available in this collector. The typed memory
25661 allocation requires information about where the pointers are located
25664 The position in the bit-field is the position, counting in bits, of the
25665 bit closest to the beginning of the structure.
25667 The non-atomic types are encoded as follows:
25669 pointers `^' followed by the pointed type.
25670 arrays `[' followed by the number of elements in the array
25671 followed by the type of the elements followed by `]'
25672 structures `{' followed by the name of the structure (or `?' if the
25673 structure is unnamed), the `=' sign, the type of the
25675 unions `(' followed by the name of the structure (or `?' if the
25676 union is unnamed), the `=' sign, the type of the members
25679 Here are some types and their encodings, as they are generated by the
25680 compiler on an i386 machine:
25683 Objective-C type Compiler encoding
25685 struct { `{?=i[3f]b128i3b131i2c}'
25694 In addition to the types the compiler also encodes the type
25695 specifiers. The table below describes the encoding of the current
25696 Objective-C type specifiers:
25708 The type specifiers are encoded just before the type. Unlike types
25709 however, the type specifiers are only encoded when they appear in method
25713 File: gcc.info, Node: Garbage Collection, Next: Constant string objects, Prev: Type encoding, Up: Objective-C
25715 7.3 Garbage Collection
25716 ======================
25718 Support for a new memory management policy has been added by using a
25719 powerful conservative garbage collector, known as the
25720 Boehm-Demers-Weiser conservative garbage collector. It is available
25721 from `http://www.hpl.hp.com/personal/Hans_Boehm/gc/'.
25723 To enable the support for it you have to configure the compiler using
25724 an additional argument, `--enable-objc-gc'. You need to have garbage
25725 collector installed before building the compiler. This will build an
25726 additional runtime library which has several enhancements to support
25727 the garbage collector. The new library has a new name, `libobjc_gc.a'
25728 to not conflict with the non-garbage-collected library.
25730 When the garbage collector is used, the objects are allocated using the
25731 so-called typed memory allocation mechanism available in the
25732 Boehm-Demers-Weiser collector. This mode requires precise information
25733 on where pointers are located inside objects. This information is
25734 computed once per class, immediately after the class has been
25737 There is a new runtime function `class_ivar_set_gcinvisible()' which
25738 can be used to declare a so-called "weak pointer" reference. Such a
25739 pointer is basically hidden for the garbage collector; this can be
25740 useful in certain situations, especially when you want to keep track of
25741 the allocated objects, yet allow them to be collected. This kind of
25742 pointers can only be members of objects, you cannot declare a global
25743 pointer as a weak reference. Every type which is a pointer type can be
25744 declared a weak pointer, including `id', `Class' and `SEL'.
25746 Here is an example of how to use this feature. Suppose you want to
25747 implement a class whose instances hold a weak pointer reference; the
25748 following class does this:
25751 @interface WeakPointer : Object
25753 const void* weakPointer;
25756 - initWithPointer:(const void*)p;
25757 - (const void*)weakPointer;
25761 @implementation WeakPointer
25765 class_ivar_set_gcinvisible (self, "weakPointer", YES);
25768 - initWithPointer:(const void*)p
25774 - (const void*)weakPointer
25776 return weakPointer;
25781 Weak pointers are supported through a new type character specifier
25782 represented by the `!' character. The `class_ivar_set_gcinvisible()'
25783 function adds or removes this specifier to the string type description
25784 of the instance variable named as argument.
25787 File: gcc.info, Node: Constant string objects, Next: compatibility_alias, Prev: Garbage Collection, Up: Objective-C
25789 7.4 Constant string objects
25790 ===========================
25792 GNU Objective-C provides constant string objects that are generated
25793 directly by the compiler. You declare a constant string object by
25794 prefixing a C constant string with the character `@':
25796 id myString = @"this is a constant string object";
25798 The constant string objects are by default instances of the
25799 `NXConstantString' class which is provided by the GNU Objective-C
25800 runtime. To get the definition of this class you must include the
25801 `objc/NXConstStr.h' header file.
25803 User defined libraries may want to implement their own constant string
25804 class. To be able to support them, the GNU Objective-C compiler
25805 provides a new command line options
25806 `-fconstant-string-class=CLASS-NAME'. The provided class should adhere
25807 to a strict structure, the same as `NXConstantString''s structure:
25810 @interface MyConstantStringClass
25818 `NXConstantString' inherits from `Object'; user class libraries may
25819 choose to inherit the customized constant string class from a different
25820 class than `Object'. There is no requirement in the methods the
25821 constant string class has to implement, but the final ivar layout of
25822 the class must be the compatible with the given structure.
25824 When the compiler creates the statically allocated constant string
25825 object, the `c_string' field will be filled by the compiler with the
25826 string; the `length' field will be filled by the compiler with the
25827 string length; the `isa' pointer will be filled with `NULL' by the
25828 compiler, and it will later be fixed up automatically at runtime by the
25829 GNU Objective-C runtime library to point to the class which was set by
25830 the `-fconstant-string-class' option when the object file is loaded (if
25831 you wonder how it works behind the scenes, the name of the class to
25832 use, and the list of static objects to fixup, are stored by the
25833 compiler in the object file in a place where the GNU runtime library
25834 will find them at runtime).
25836 As a result, when a file is compiled with the
25837 `-fconstant-string-class' option, all the constant string objects will
25838 be instances of the class specified as argument to this option. It is
25839 possible to have multiple compilation units referring to different
25840 constant string classes, neither the compiler nor the linker impose any
25841 restrictions in doing this.
25844 File: gcc.info, Node: compatibility_alias, Prev: Constant string objects, Up: Objective-C
25846 7.5 compatibility_alias
25847 =======================
25849 This is a feature of the Objective-C compiler rather than of the
25850 runtime, anyway since it is documented nowhere and its existence was
25851 forgotten, we are documenting it here.
25853 The keyword `@compatibility_alias' allows you to define a class name
25854 as equivalent to another class name. For example:
25856 @compatibility_alias WOApplication GSWApplication;
25858 tells the compiler that each time it encounters `WOApplication' as a
25859 class name, it should replace it with `GSWApplication' (that is,
25860 `WOApplication' is just an alias for `GSWApplication').
25862 There are some constraints on how this can be used--
25864 * `WOApplication' (the alias) must not be an existing class;
25866 * `GSWApplication' (the real class) must be an existing class.
25870 File: gcc.info, Node: Compatibility, Next: Gcov, Prev: Objective-C, Up: Top
25872 8 Binary Compatibility
25873 **********************
25875 Binary compatibility encompasses several related concepts:
25877 "application binary interface (ABI)"
25878 The set of runtime conventions followed by all of the tools that
25879 deal with binary representations of a program, including
25880 compilers, assemblers, linkers, and language runtime support.
25881 Some ABIs are formal with a written specification, possibly
25882 designed by multiple interested parties. Others are simply the
25883 way things are actually done by a particular set of tools.
25886 A compiler conforms to an ABI if it generates code that follows
25887 all of the specifications enumerated by that ABI. A library
25888 conforms to an ABI if it is implemented according to that ABI. An
25889 application conforms to an ABI if it is built using tools that
25890 conform to that ABI and does not contain source code that
25891 specifically changes behavior specified by the ABI.
25893 "calling conventions"
25894 Calling conventions are a subset of an ABI that specify of how
25895 arguments are passed and function results are returned.
25898 Different sets of tools are interoperable if they generate files
25899 that can be used in the same program. The set of tools includes
25900 compilers, assemblers, linkers, libraries, header files, startup
25901 files, and debuggers. Binaries produced by different sets of
25902 tools are not interoperable unless they implement the same ABI.
25903 This applies to different versions of the same tools as well as
25904 tools from different vendors.
25907 Whether a function in a binary built by one set of tools can call a
25908 function in a binary built by a different set of tools is a subset
25909 of interoperability.
25911 "implementation-defined features"
25912 Language standards include lists of implementation-defined
25913 features whose behavior can vary from one implementation to
25914 another. Some of these features are normally covered by a
25915 platform's ABI and others are not. The features that are not
25916 covered by an ABI generally affect how a program behaves, but not
25920 Conformance to the same ABI and the same behavior of
25921 implementation-defined features are both relevant for
25924 The application binary interface implemented by a C or C++ compiler
25925 affects code generation and runtime support for:
25927 * size and alignment of data types
25929 * layout of structured types
25931 * calling conventions
25933 * register usage conventions
25935 * interfaces for runtime arithmetic support
25937 * object file formats
25939 In addition, the application binary interface implemented by a C++
25940 compiler affects code generation and runtime support for:
25943 * exception handling
25945 * invoking constructors and destructors
25947 * layout, alignment, and padding of classes
25949 * layout and alignment of virtual tables
25951 Some GCC compilation options cause the compiler to generate code that
25952 does not conform to the platform's default ABI. Other options cause
25953 different program behavior for implementation-defined features that are
25954 not covered by an ABI. These options are provided for consistency with
25955 other compilers that do not follow the platform's default ABI or the
25956 usual behavior of implementation-defined features for the platform. Be
25957 very careful about using such options.
25959 Most platforms have a well-defined ABI that covers C code, but ABIs
25960 that cover C++ functionality are not yet common.
25962 Starting with GCC 3.2, GCC binary conventions for C++ are based on a
25963 written, vendor-neutral C++ ABI that was designed to be specific to
25964 64-bit Itanium but also includes generic specifications that apply to
25965 any platform. This C++ ABI is also implemented by other compiler
25966 vendors on some platforms, notably GNU/Linux and BSD systems. We have
25967 tried hard to provide a stable ABI that will be compatible with future
25968 GCC releases, but it is possible that we will encounter problems that
25969 make this difficult. Such problems could include different
25970 interpretations of the C++ ABI by different vendors, bugs in the ABI, or
25971 bugs in the implementation of the ABI in different compilers. GCC's
25972 `-Wabi' switch warns when G++ generates code that is probably not
25973 compatible with the C++ ABI.
25975 The C++ library used with a C++ compiler includes the Standard C++
25976 Library, with functionality defined in the C++ Standard, plus language
25977 runtime support. The runtime support is included in a C++ ABI, but
25978 there is no formal ABI for the Standard C++ Library. Two
25979 implementations of that library are interoperable if one follows the
25980 de-facto ABI of the other and if they are both built with the same
25981 compiler, or with compilers that conform to the same ABI for C++
25982 compiler and runtime support.
25984 When G++ and another C++ compiler conform to the same C++ ABI, but the
25985 implementations of the Standard C++ Library that they normally use do
25986 not follow the same ABI for the Standard C++ Library, object files
25987 built with those compilers can be used in the same program only if they
25988 use the same C++ library. This requires specifying the location of the
25989 C++ library header files when invoking the compiler whose usual library
25990 is not being used. The location of GCC's C++ header files depends on
25991 how the GCC build was configured, but can be seen by using the G++ `-v'
25992 option. With default configuration options for G++ 3.3 the compile
25993 line for a different C++ compiler needs to include
25995 -IGCC_INSTALL_DIRECTORY/include/c++/3.3
25997 Similarly, compiling code with G++ that must use a C++ library other
25998 than the GNU C++ library requires specifying the location of the header
25999 files for that other library.
26001 The most straightforward way to link a program to use a particular C++
26002 library is to use a C++ driver that specifies that C++ library by
26003 default. The `g++' driver, for example, tells the linker where to find
26004 GCC's C++ library (`libstdc++') plus the other libraries and startup
26005 files it needs, in the proper order.
26007 If a program must use a different C++ library and it's not possible to
26008 do the final link using a C++ driver that uses that library by default,
26009 it is necessary to tell `g++' the location and name of that library.
26010 It might also be necessary to specify different startup files and other
26011 runtime support libraries, and to suppress the use of GCC's support
26012 libraries with one or more of the options `-nostdlib', `-nostartfiles',
26013 and `-nodefaultlibs'.
26016 File: gcc.info, Node: Gcov, Next: Trouble, Prev: Compatibility, Up: Top
26018 9 `gcov'--a Test Coverage Program
26019 *********************************
26021 `gcov' is a tool you can use in conjunction with GCC to test code
26022 coverage in your programs.
26026 * Gcov Intro:: Introduction to gcov.
26027 * Invoking Gcov:: How to use gcov.
26028 * Gcov and Optimization:: Using gcov with GCC optimization.
26029 * Gcov Data Files:: The files used by gcov.
26030 * Cross-profiling:: Data file relocation.
26033 File: gcc.info, Node: Gcov Intro, Next: Invoking Gcov, Up: Gcov
26035 9.1 Introduction to `gcov'
26036 ==========================
26038 `gcov' is a test coverage program. Use it in concert with GCC to
26039 analyze your programs to help create more efficient, faster running
26040 code and to discover untested parts of your program. You can use
26041 `gcov' as a profiling tool to help discover where your optimization
26042 efforts will best affect your code. You can also use `gcov' along with
26043 the other profiling tool, `gprof', to assess which parts of your code
26044 use the greatest amount of computing time.
26046 Profiling tools help you analyze your code's performance. Using a
26047 profiler such as `gcov' or `gprof', you can find out some basic
26048 performance statistics, such as:
26050 * how often each line of code executes
26052 * what lines of code are actually executed
26054 * how much computing time each section of code uses
26056 Once you know these things about how your code works when compiled, you
26057 can look at each module to see which modules should be optimized.
26058 `gcov' helps you determine where to work on optimization.
26060 Software developers also use coverage testing in concert with
26061 testsuites, to make sure software is actually good enough for a release.
26062 Testsuites can verify that a program works as expected; a coverage
26063 program tests to see how much of the program is exercised by the
26064 testsuite. Developers can then determine what kinds of test cases need
26065 to be added to the testsuites to create both better testing and a better
26068 You should compile your code without optimization if you plan to use
26069 `gcov' because the optimization, by combining some lines of code into
26070 one function, may not give you as much information as you need to look
26071 for `hot spots' where the code is using a great deal of computer time.
26072 Likewise, because `gcov' accumulates statistics by line (at the lowest
26073 resolution), it works best with a programming style that places only
26074 one statement on each line. If you use complicated macros that expand
26075 to loops or to other control structures, the statistics are less
26076 helpful--they only report on the line where the macro call appears. If
26077 your complex macros behave like functions, you can replace them with
26078 inline functions to solve this problem.
26080 `gcov' creates a logfile called `SOURCEFILE.gcov' which indicates how
26081 many times each line of a source file `SOURCEFILE.c' has executed. You
26082 can use these logfiles along with `gprof' to aid in fine-tuning the
26083 performance of your programs. `gprof' gives timing information you can
26084 use along with the information you get from `gcov'.
26086 `gcov' works only on code compiled with GCC. It is not compatible
26087 with any other profiling or test coverage mechanism.
26090 File: gcc.info, Node: Invoking Gcov, Next: Gcov and Optimization, Prev: Gcov Intro, Up: Gcov
26092 9.2 Invoking `gcov'
26093 ===================
26095 gcov [OPTIONS] SOURCEFILE
26097 `gcov' accepts the following options:
26101 Display help about using `gcov' (on the standard output), and exit
26102 without doing any further processing.
26106 Display the `gcov' version number (on the standard output), and
26107 exit without doing any further processing.
26111 Write individual execution counts for every basic block. Normally
26112 gcov outputs execution counts only for the main blocks of a line.
26113 With this option you can determine if blocks within a single line
26114 are not being executed.
26117 `--branch-probabilities'
26118 Write branch frequencies to the output file, and write branch
26119 summary info to the standard output. This option allows you to
26120 see how often each branch in your program was taken.
26121 Unconditional branches will not be shown, unless the `-u' option
26126 Write branch frequencies as the number of branches taken, rather
26127 than the percentage of branches taken.
26131 Do not create the `gcov' output file.
26134 `--long-file-names'
26135 Create long file names for included source files. For example, if
26136 the header file `x.h' contains code, and was included in the file
26137 `a.c', then running `gcov' on the file `a.c' will produce an
26138 output file called `a.c##x.h.gcov' instead of `x.h.gcov'. This
26139 can be useful if `x.h' is included in multiple source files. If
26140 you use the `-p' option, both the including and included file
26141 names will be complete path names.
26145 Preserve complete path information in the names of generated
26146 `.gcov' files. Without this option, just the filename component is
26147 used. With this option, all directories are used, with `/'
26148 characters translated to `#' characters, `.' directory components
26149 removed and `..' components renamed to `^'. This is useful if
26150 sourcefiles are in several different directories. It also affects
26154 `--function-summaries'
26155 Output summaries for each function in addition to the file level
26158 `-o DIRECTORY|FILE'
26159 `--object-directory DIRECTORY'
26160 `--object-file FILE'
26161 Specify either the directory containing the gcov data files, or the
26162 object path name. The `.gcno', and `.gcda' data files are
26163 searched for using this option. If a directory is specified, the
26164 data files are in that directory and named after the source file
26165 name, without its extension. If a file is specified here, the
26166 data files are named after that file, without its extension. If
26167 this option is not supplied, it defaults to the current directory.
26170 `--unconditional-branches'
26171 When branch probabilities are given, include those of
26172 unconditional branches. Unconditional branches are normally not
26176 `gcov' should be run with the current directory the same as that when
26177 you invoked the compiler. Otherwise it will not be able to locate the
26178 source files. `gcov' produces files called `MANGLEDNAME.gcov' in the
26179 current directory. These contain the coverage information of the
26180 source file they correspond to. One `.gcov' file is produced for each
26181 source file containing code, which was compiled to produce the data
26182 files. The MANGLEDNAME part of the output file name is usually simply
26183 the source file name, but can be something more complicated if the `-l'
26184 or `-p' options are given. Refer to those options for details.
26186 The `.gcov' files contain the `:' separated fields along with program
26187 source code. The format is
26189 EXECUTION_COUNT:LINE_NUMBER:SOURCE LINE TEXT
26191 Additional block information may succeed each line, when requested by
26192 command line option. The EXECUTION_COUNT is `-' for lines containing
26193 no code and `#####' for lines which were never executed. Some lines of
26194 information at the start have LINE_NUMBER of zero.
26196 The preamble lines are of the form
26200 The ordering and number of these preamble lines will be augmented as
26201 `gcov' development progresses -- do not rely on them remaining
26202 unchanged. Use TAG to locate a particular preamble line.
26204 The additional block information is of the form
26208 The INFORMATION is human readable, but designed to be simple enough
26209 for machine parsing too.
26211 When printing percentages, 0% and 100% are only printed when the values
26212 are _exactly_ 0% and 100% respectively. Other values which would
26213 conventionally be rounded to 0% or 100% are instead printed as the
26214 nearest non-boundary value.
26216 When using `gcov', you must first compile your program with two
26217 special GCC options: `-fprofile-arcs -ftest-coverage'. This tells the
26218 compiler to generate additional information needed by gcov (basically a
26219 flow graph of the program) and also includes additional code in the
26220 object files for generating the extra profiling information needed by
26221 gcov. These additional files are placed in the directory where the
26222 object file is located.
26224 Running the program will cause profile output to be generated. For
26225 each source file compiled with `-fprofile-arcs', an accompanying
26226 `.gcda' file will be placed in the object file directory.
26228 Running `gcov' with your program's source file names as arguments will
26229 now produce a listing of the code along with frequency of execution for
26230 each line. For example, if your program is called `tmp.c', this is
26231 what you see when you use the basic `gcov' facility:
26233 $ gcc -fprofile-arcs -ftest-coverage tmp.c
26236 90.00% of 10 source lines executed in file tmp.c
26237 Creating tmp.c.gcov.
26239 The file `tmp.c.gcov' contains output from `gcov'. Here is a sample:
26242 -: 0:Graph:tmp.gcno
26246 -: 1:#include <stdio.h>
26248 -: 3:int main (void)
26250 1: 5: int i, total;
26254 11: 9: for (i = 0; i < 10; i++)
26255 10: 10: total += i;
26257 1: 12: if (total != 45)
26258 #####: 13: printf ("Failure\n");
26260 1: 15: printf ("Success\n");
26264 When you use the `-a' option, you will get individual block counts,
26265 and the output looks like this:
26268 -: 0:Graph:tmp.gcno
26272 -: 1:#include <stdio.h>
26274 -: 3:int main (void)
26277 1: 5: int i, total;
26281 11: 9: for (i = 0; i < 10; i++)
26283 10: 10: total += i;
26286 1: 12: if (total != 45)
26288 #####: 13: printf ("Failure\n");
26291 1: 15: printf ("Success\n");
26297 In this mode, each basic block is only shown on one line - the last
26298 line of the block. A multi-line block will only contribute to the
26299 execution count of that last line, and other lines will not be shown to
26300 contain code, unless previous blocks end on those lines. The total
26301 execution count of a line is shown and subsequent lines show the
26302 execution counts for individual blocks that end on that line. After
26303 each block, the branch and call counts of the block will be shown, if
26304 the `-b' option is given.
26306 Because of the way GCC instruments calls, a call count can be shown
26307 after a line with no individual blocks. As you can see, line 13
26308 contains a basic block that was not executed.
26310 When you use the `-b' option, your output looks like this:
26313 90.00% of 10 source lines executed in file tmp.c
26314 80.00% of 5 branches executed in file tmp.c
26315 80.00% of 5 branches taken at least once in file tmp.c
26316 50.00% of 2 calls executed in file tmp.c
26317 Creating tmp.c.gcov.
26319 Here is a sample of a resulting `tmp.c.gcov' file:
26322 -: 0:Graph:tmp.gcno
26326 -: 1:#include <stdio.h>
26328 -: 3:int main (void)
26329 function main called 1 returned 1 blocks executed 75%
26331 1: 5: int i, total;
26335 11: 9: for (i = 0; i < 10; i++)
26336 branch 0 taken 91% (fallthrough)
26338 10: 10: total += i;
26340 1: 12: if (total != 45)
26341 branch 0 taken 0% (fallthrough)
26342 branch 1 taken 100%
26343 #####: 13: printf ("Failure\n");
26344 call 0 never executed
26346 1: 15: printf ("Success\n");
26347 call 0 called 1 returned 100%
26351 For each function, a line is printed showing how many times the
26352 function is called, how many times it returns and what percentage of the
26353 function's blocks were executed.
26355 For each basic block, a line is printed after the last line of the
26356 basic block describing the branch or call that ends the basic block.
26357 There can be multiple branches and calls listed for a single source
26358 line if there are multiple basic blocks that end on that line. In this
26359 case, the branches and calls are each given a number. There is no
26360 simple way to map these branches and calls back to source constructs.
26361 In general, though, the lowest numbered branch or call will correspond
26362 to the leftmost construct on the source line.
26364 For a branch, if it was executed at least once, then a percentage
26365 indicating the number of times the branch was taken divided by the
26366 number of times the branch was executed will be printed. Otherwise, the
26367 message "never executed" is printed.
26369 For a call, if it was executed at least once, then a percentage
26370 indicating the number of times the call returned divided by the number
26371 of times the call was executed will be printed. This will usually be
26372 100%, but may be less for functions that call `exit' or `longjmp', and
26373 thus may not return every time they are called.
26375 The execution counts are cumulative. If the example program were
26376 executed again without removing the `.gcda' file, the count for the
26377 number of times each line in the source was executed would be added to
26378 the results of the previous run(s). This is potentially useful in
26379 several ways. For example, it could be used to accumulate data over a
26380 number of program runs as part of a test verification suite, or to
26381 provide more accurate long-term information over a large number of
26384 The data in the `.gcda' files is saved immediately before the program
26385 exits. For each source file compiled with `-fprofile-arcs', the
26386 profiling code first attempts to read in an existing `.gcda' file; if
26387 the file doesn't match the executable (differing number of basic block
26388 counts) it will ignore the contents of the file. It then adds in the
26389 new execution counts and finally writes the data to the file.
26392 File: gcc.info, Node: Gcov and Optimization, Next: Gcov Data Files, Prev: Invoking Gcov, Up: Gcov
26394 9.3 Using `gcov' with GCC Optimization
26395 ======================================
26397 If you plan to use `gcov' to help optimize your code, you must first
26398 compile your program with two special GCC options: `-fprofile-arcs
26399 -ftest-coverage'. Aside from that, you can use any other GCC options;
26400 but if you want to prove that every single line in your program was
26401 executed, you should not compile with optimization at the same time.
26402 On some machines the optimizer can eliminate some simple code lines by
26403 combining them with other lines. For example, code like this:
26410 can be compiled into one instruction on some machines. In this case,
26411 there is no way for `gcov' to calculate separate execution counts for
26412 each line because there isn't separate code for each line. Hence the
26413 `gcov' output looks like this if you compiled the program with
26416 100: 12:if (a != b)
26421 The output shows that this block of code, combined by optimization,
26422 executed 100 times. In one sense this result is correct, because there
26423 was only one instruction representing all four of these lines. However,
26424 the output does not indicate how many times the result was 0 and how
26425 many times the result was 1.
26427 Inlineable functions can create unexpected line counts. Line counts
26428 are shown for the source code of the inlineable function, but what is
26429 shown depends on where the function is inlined, or if it is not inlined
26432 If the function is not inlined, the compiler must emit an out of line
26433 copy of the function, in any object file that needs it. If `fileA.o'
26434 and `fileB.o' both contain out of line bodies of a particular
26435 inlineable function, they will also both contain coverage counts for
26436 that function. When `fileA.o' and `fileB.o' are linked together, the
26437 linker will, on many systems, select one of those out of line bodies
26438 for all calls to that function, and remove or ignore the other.
26439 Unfortunately, it will not remove the coverage counters for the unused
26440 function body. Hence when instrumented, all but one use of that
26441 function will show zero counts.
26443 If the function is inlined in several places, the block structure in
26444 each location might not be the same. For instance, a condition might
26445 now be calculable at compile time in some instances. Because the
26446 coverage of all the uses of the inline function will be shown for the
26447 same source lines, the line counts themselves might seem inconsistent.
26450 File: gcc.info, Node: Gcov Data Files, Next: Cross-profiling, Prev: Gcov and Optimization, Up: Gcov
26452 9.4 Brief description of `gcov' data files
26453 ==========================================
26455 `gcov' uses two files for profiling. The names of these files are
26456 derived from the original _object_ file by substituting the file suffix
26457 with either `.gcno', or `.gcda'. All of these files are placed in the
26458 same directory as the object file, and contain data stored in a
26459 platform-independent format.
26461 The `.gcno' file is generated when the source file is compiled with
26462 the GCC `-ftest-coverage' option. It contains information to
26463 reconstruct the basic block graphs and assign source line numbers to
26466 The `.gcda' file is generated when a program containing object files
26467 built with the GCC `-fprofile-arcs' option is executed. A separate
26468 `.gcda' file is created for each object file compiled with this option.
26469 It contains arc transition counts, and some summary information.
26471 The full details of the file format is specified in `gcov-io.h', and
26472 functions provided in that header file should be used to access the
26476 File: gcc.info, Node: Cross-profiling, Prev: Gcov Data Files, Up: Gcov
26478 9.5 Data file relocation to support cross-profiling
26479 ===================================================
26481 Running the program will cause profile output to be generated. For each
26482 source file compiled with `-fprofile-arcs', an accompanying `.gcda'
26483 file will be placed in the object file directory. That implicitly
26484 requires running the program on the same system as it was built or
26485 having the same absolute directory structure on the target system. The
26486 program will try to create the needed directory structure, if it is not
26489 To support cross-profiling, a program compiled with `-fprofile-arcs'
26490 can relocate the data files based on two environment variables:
26492 * GCOV_PREFIX contains the prefix to add to the absolute paths in
26493 the object file. Prefix must be absolute as well, otherwise its
26494 value is ignored. The default is no prefix.
26496 * GCOV_PREFIX_STRIP indicates the how many initial directory names
26497 to strip off the hardwired absolute paths. Default value is 0.
26499 _Note:_ GCOV_PREFIX_STRIP has no effect if GCOV_PREFIX is
26500 undefined, empty or non-absolute.
26502 For example, if the object file `/user/build/foo.o' was built with
26503 `-fprofile-arcs', the final executable will try to create the data file
26504 `/user/build/foo.gcda' when running on the target system. This will
26505 fail if the corresponding directory does not exist and it is unable to
26506 create it. This can be overcome by, for example, setting the
26507 environment as `GCOV_PREFIX=/target/run' and `GCOV_PREFIX_STRIP=1'.
26508 Such a setting will name the data file `/target/run/build/foo.gcda'.
26510 You must move the data files to the expected directory tree in order to
26511 use them for profile directed optimizations (`--use-profile'), or to
26512 use the `gcov' tool.
26515 File: gcc.info, Node: Trouble, Next: Bugs, Prev: Gcov, Up: Top
26517 10 Known Causes of Trouble with GCC
26518 ***********************************
26520 This section describes known problems that affect users of GCC. Most
26521 of these are not GCC bugs per se--if they were, we would fix them. But
26522 the result for a user may be like the result of a bug.
26524 Some of these problems are due to bugs in other software, some are
26525 missing features that are too much work to add, and some are places
26526 where people's opinions differ as to what is best.
26530 * Actual Bugs:: Bugs we will fix later.
26531 * Cross-Compiler Problems:: Common problems of cross compiling with GCC.
26532 * Interoperation:: Problems using GCC with other compilers,
26533 and with certain linkers, assemblers and debuggers.
26534 * Incompatibilities:: GCC is incompatible with traditional C.
26535 * Fixed Headers:: GCC uses corrected versions of system header files.
26536 This is necessary, but doesn't always work smoothly.
26537 * Standard Libraries:: GCC uses the system C library, which might not be
26538 compliant with the ISO C standard.
26539 * Disappointments:: Regrettable things we can't change, but not quite bugs.
26540 * C++ Misunderstandings:: Common misunderstandings with GNU C++.
26541 * Protoize Caveats:: Things to watch out for when using `protoize'.
26542 * Non-bugs:: Things we think are right, but some others disagree.
26543 * Warnings and Errors:: Which problems in your code get warnings,
26544 and which get errors.
26547 File: gcc.info, Node: Actual Bugs, Next: Cross-Compiler Problems, Up: Trouble
26549 10.1 Actual Bugs We Haven't Fixed Yet
26550 =====================================
26552 * The `fixincludes' script interacts badly with automounters; if the
26553 directory of system header files is automounted, it tends to be
26554 unmounted while `fixincludes' is running. This would seem to be a
26555 bug in the automounter. We don't know any good way to work around
26558 * The `fixproto' script will sometimes add prototypes for the
26559 `sigsetjmp' and `siglongjmp' functions that reference the
26560 `jmp_buf' type before that type is defined. To work around this,
26561 edit the offending file and place the typedef in front of the
26565 File: gcc.info, Node: Cross-Compiler Problems, Next: Interoperation, Prev: Actual Bugs, Up: Trouble
26567 10.2 Cross-Compiler Problems
26568 ============================
26570 You may run into problems with cross compilation on certain machines,
26571 for several reasons.
26573 * At present, the program `mips-tfile' which adds debug support to
26574 object files on MIPS systems does not work in a cross compile
26578 File: gcc.info, Node: Interoperation, Next: Incompatibilities, Prev: Cross-Compiler Problems, Up: Trouble
26580 10.3 Interoperation
26581 ===================
26583 This section lists various difficulties encountered in using GCC
26584 together with other compilers or with the assemblers, linkers,
26585 libraries and debuggers on certain systems.
26587 * On many platforms, GCC supports a different ABI for C++ than do
26588 other compilers, so the object files compiled by GCC cannot be
26589 used with object files generated by another C++ compiler.
26591 An area where the difference is most apparent is name mangling.
26592 The use of different name mangling is intentional, to protect you
26593 from more subtle problems. Compilers differ as to many internal
26594 details of C++ implementation, including: how class instances are
26595 laid out, how multiple inheritance is implemented, and how virtual
26596 function calls are handled. If the name encoding were made the
26597 same, your programs would link against libraries provided from
26598 other compilers--but the programs would then crash when run.
26599 Incompatible libraries are then detected at link time, rather than
26602 * On some BSD systems, including some versions of Ultrix, use of
26603 profiling causes static variable destructors (currently used only
26604 in C++) not to be run.
26606 * On some SGI systems, when you use `-lgl_s' as an option, it gets
26607 translated magically to `-lgl_s -lX11_s -lc_s'. Naturally, this
26608 does not happen when you use GCC. You must specify all three
26609 options explicitly.
26611 * On a SPARC, GCC aligns all values of type `double' on an 8-byte
26612 boundary, and it expects every `double' to be so aligned. The Sun
26613 compiler usually gives `double' values 8-byte alignment, with one
26614 exception: function arguments of type `double' may not be aligned.
26616 As a result, if a function compiled with Sun CC takes the address
26617 of an argument of type `double' and passes this pointer of type
26618 `double *' to a function compiled with GCC, dereferencing the
26619 pointer may cause a fatal signal.
26621 One way to solve this problem is to compile your entire program
26622 with GCC. Another solution is to modify the function that is
26623 compiled with Sun CC to copy the argument into a local variable;
26624 local variables are always properly aligned. A third solution is
26625 to modify the function that uses the pointer to dereference it via
26626 the following function `access_double' instead of directly with
26630 access_double (double *unaligned_ptr)
26632 union d2i { double d; int i[2]; };
26634 union d2i *p = (union d2i *) unaligned_ptr;
26643 Storing into the pointer can be done likewise with the same union.
26645 * On Solaris, the `malloc' function in the `libmalloc.a' library may
26646 allocate memory that is only 4 byte aligned. Since GCC on the
26647 SPARC assumes that doubles are 8 byte aligned, this may result in a
26648 fatal signal if doubles are stored in memory allocated by the
26649 `libmalloc.a' library.
26651 The solution is to not use the `libmalloc.a' library. Use instead
26652 `malloc' and related functions from `libc.a'; they do not have
26655 * On the HP PA machine, ADB sometimes fails to work on functions
26656 compiled with GCC. Specifically, it fails to work on functions
26657 that use `alloca' or variable-size arrays. This is because GCC
26658 doesn't generate HP-UX unwind descriptors for such functions. It
26659 may even be impossible to generate them.
26661 * Debugging (`-g') is not supported on the HP PA machine, unless you
26662 use the preliminary GNU tools.
26664 * Taking the address of a label may generate errors from the HP-UX
26665 PA assembler. GAS for the PA does not have this problem.
26667 * Using floating point parameters for indirect calls to static
26668 functions will not work when using the HP assembler. There simply
26669 is no way for GCC to specify what registers hold arguments for
26670 static functions when using the HP assembler. GAS for the PA does
26671 not have this problem.
26673 * In extremely rare cases involving some very large functions you may
26674 receive errors from the HP linker complaining about an out of
26675 bounds unconditional branch offset. This used to occur more often
26676 in previous versions of GCC, but is now exceptionally rare. If
26677 you should run into it, you can work around by making your
26680 * GCC compiled code sometimes emits warnings from the HP-UX
26681 assembler of the form:
26683 (warning) Use of GR3 when
26684 frame >= 8192 may cause conflict.
26686 These warnings are harmless and can be safely ignored.
26688 * In extremely rare cases involving some very large functions you may
26689 receive errors from the AIX Assembler complaining about a
26690 displacement that is too large. If you should run into it, you
26691 can work around by making your function smaller.
26693 * The `libstdc++.a' library in GCC relies on the SVR4 dynamic linker
26694 semantics which merges global symbols between libraries and
26695 applications, especially necessary for C++ streams functionality.
26696 This is not the default behavior of AIX shared libraries and
26697 dynamic linking. `libstdc++.a' is built on AIX with
26698 "runtime-linking" enabled so that symbol merging can occur. To
26699 utilize this feature, the application linked with `libstdc++.a'
26700 must include the `-Wl,-brtl' flag on the link line. G++ cannot
26701 impose this because this option may interfere with the semantics
26702 of the user program and users may not always use `g++' to link his
26703 or her application. Applications are not required to use the
26704 `-Wl,-brtl' flag on the link line--the rest of the `libstdc++.a'
26705 library which is not dependent on the symbol merging semantics
26706 will continue to function correctly.
26708 * An application can interpose its own definition of functions for
26709 functions invoked by `libstdc++.a' with "runtime-linking" enabled
26710 on AIX. To accomplish this the application must be linked with
26711 "runtime-linking" option and the functions explicitly must be
26712 exported by the application (`-Wl,-brtl,-bE:exportfile').
26714 * AIX on the RS/6000 provides support (NLS) for environments outside
26715 of the United States. Compilers and assemblers use NLS to support
26716 locale-specific representations of various objects including
26717 floating-point numbers (`.' vs `,' for separating decimal
26718 fractions). There have been problems reported where the library
26719 linked with GCC does not produce the same floating-point formats
26720 that the assembler accepts. If you have this problem, set the
26721 `LANG' environment variable to `C' or `En_US'.
26723 * Even if you specify `-fdollars-in-identifiers', you cannot
26724 successfully use `$' in identifiers on the RS/6000 due to a
26725 restriction in the IBM assembler. GAS supports these identifiers.
26727 * On Ultrix, the Fortran compiler expects registers 2 through 5 to
26728 be saved by function calls. However, the C compiler uses
26729 conventions compatible with BSD Unix: registers 2 through 5 may be
26730 clobbered by function calls.
26732 GCC uses the same convention as the Ultrix C compiler. You can use
26733 these options to produce code compatible with the Fortran compiler:
26735 -fcall-saved-r2 -fcall-saved-r3 -fcall-saved-r4 -fcall-saved-r5
26738 File: gcc.info, Node: Incompatibilities, Next: Fixed Headers, Prev: Interoperation, Up: Trouble
26740 10.4 Incompatibilities of GCC
26741 =============================
26743 There are several noteworthy incompatibilities between GNU C and K&R
26744 (non-ISO) versions of C.
26746 * GCC normally makes string constants read-only. If several
26747 identical-looking string constants are used, GCC stores only one
26748 copy of the string.
26750 One consequence is that you cannot call `mktemp' with a string
26751 constant argument. The function `mktemp' always alters the string
26752 its argument points to.
26754 Another consequence is that `sscanf' does not work on some very
26755 old systems when passed a string constant as its format control
26756 string or input. This is because `sscanf' incorrectly tries to
26757 write into the string constant. Likewise `fscanf' and `scanf'.
26759 The solution to these problems is to change the program to use
26760 `char'-array variables with initialization strings for these
26761 purposes instead of string constants.
26763 * `-2147483648' is positive.
26765 This is because 2147483648 cannot fit in the type `int', so
26766 (following the ISO C rules) its data type is `unsigned long int'.
26767 Negating this value yields 2147483648 again.
26769 * GCC does not substitute macro arguments when they appear inside of
26770 string constants. For example, the following macro in GCC
26774 will produce output `"a"' regardless of what the argument A is.
26776 * When you use `setjmp' and `longjmp', the only automatic variables
26777 guaranteed to remain valid are those declared `volatile'. This is
26778 a consequence of automatic register allocation. Consider this
26792 /* `longjmp (j)' may occur in `fun3'. */
26793 return a + fun3 ();
26796 Here `a' may or may not be restored to its first value when the
26797 `longjmp' occurs. If `a' is allocated in a register, then its
26798 first value is restored; otherwise, it keeps the last value stored
26801 If you use the `-W' option with the `-O' option, you will get a
26802 warning when GCC thinks such a problem might be possible.
26804 * Programs that use preprocessing directives in the middle of macro
26805 arguments do not work with GCC. For example, a program like this
26812 ISO C does not permit such a construct.
26814 * K&R compilers allow comments to cross over an inclusion boundary
26815 (i.e. started in an include file and ended in the including file).
26817 * Declarations of external variables and functions within a block
26818 apply only to the block containing the declaration. In other
26819 words, they have the same scope as any other declaration in the
26822 In some other C compilers, a `extern' declaration affects all the
26823 rest of the file even if it happens within a block.
26825 * In traditional C, you can combine `long', etc., with a typedef
26826 name, as shown here:
26829 typedef long foo bar;
26831 In ISO C, this is not allowed: `long' and other type modifiers
26832 require an explicit `int'.
26834 * PCC allows typedef names to be used as function parameters.
26836 * Traditional C allows the following erroneous pair of declarations
26837 to appear together in a given scope:
26842 * GCC treats all characters of identifiers as significant.
26843 According to K&R-1 (2.2), "No more than the first eight characters
26844 are significant, although more may be used.". Also according to
26845 K&R-1 (2.2), "An identifier is a sequence of letters and digits;
26846 the first character must be a letter. The underscore _ counts as
26847 a letter.", but GCC also allows dollar signs in identifiers.
26849 * PCC allows whitespace in the middle of compound assignment
26850 operators such as `+='. GCC, following the ISO standard, does not
26853 * GCC complains about unterminated character constants inside of
26854 preprocessing conditionals that fail. Some programs have English
26855 comments enclosed in conditionals that are guaranteed to fail; if
26856 these comments contain apostrophes, GCC will probably report an
26857 error. For example, this code would produce an error:
26860 You can't expect this to work.
26863 The best solution to such a problem is to put the text into an
26864 actual C comment delimited by `/*...*/'.
26866 * Many user programs contain the declaration `long time ();'. In the
26867 past, the system header files on many systems did not actually
26868 declare `time', so it did not matter what type your program
26869 declared it to return. But in systems with ISO C headers, `time'
26870 is declared to return `time_t', and if that is not the same as
26871 `long', then `long time ();' is erroneous.
26873 The solution is to change your program to use appropriate system
26874 headers (`<time.h>' on systems with ISO C headers) and not to
26875 declare `time' if the system header files declare it, or failing
26876 that to use `time_t' as the return type of `time'.
26878 * When compiling functions that return `float', PCC converts it to a
26879 double. GCC actually returns a `float'. If you are concerned
26880 with PCC compatibility, you should declare your functions to return
26881 `double'; you might as well say what you mean.
26883 * When compiling functions that return structures or unions, GCC
26884 output code normally uses a method different from that used on most
26885 versions of Unix. As a result, code compiled with GCC cannot call
26886 a structure-returning function compiled with PCC, and vice versa.
26888 The method used by GCC is as follows: a structure or union which is
26889 1, 2, 4 or 8 bytes long is returned like a scalar. A structure or
26890 union with any other size is stored into an address supplied by
26891 the caller (usually in a special, fixed register, but on some
26892 machines it is passed on the stack). The target hook
26893 `TARGET_STRUCT_VALUE_RTX' tells GCC where to pass this address.
26895 By contrast, PCC on most target machines returns structures and
26896 unions of any size by copying the data into an area of static
26897 storage, and then returning the address of that storage as if it
26898 were a pointer value. The caller must copy the data from that
26899 memory area to the place where the value is wanted. GCC does not
26900 use this method because it is slower and nonreentrant.
26902 On some newer machines, PCC uses a reentrant convention for all
26903 structure and union returning. GCC on most of these machines uses
26904 a compatible convention when returning structures and unions in
26905 memory, but still returns small structures and unions in registers.
26907 You can tell GCC to use a compatible convention for all structure
26908 and union returning with the option `-fpcc-struct-return'.
26910 * GCC complains about program fragments such as `0x74ae-0x4000'
26911 which appear to be two hexadecimal constants separated by the minus
26912 operator. Actually, this string is a single "preprocessing token".
26913 Each such token must correspond to one token in C. Since this
26914 does not, GCC prints an error message. Although it may appear
26915 obvious that what is meant is an operator and two values, the ISO
26916 C standard specifically requires that this be treated as erroneous.
26918 A "preprocessing token" is a "preprocessing number" if it begins
26919 with a digit and is followed by letters, underscores, digits,
26920 periods and `e+', `e-', `E+', `E-', `p+', `p-', `P+', or `P-'
26921 character sequences. (In strict C89 mode, the sequences `p+',
26922 `p-', `P+' and `P-' cannot appear in preprocessing numbers.)
26924 To make the above program fragment valid, place whitespace in
26925 front of the minus sign. This whitespace will end the
26926 preprocessing number.
26929 File: gcc.info, Node: Fixed Headers, Next: Standard Libraries, Prev: Incompatibilities, Up: Trouble
26931 10.5 Fixed Header Files
26932 =======================
26934 GCC needs to install corrected versions of some system header files.
26935 This is because most target systems have some header files that won't
26936 work with GCC unless they are changed. Some have bugs, some are
26937 incompatible with ISO C, and some depend on special features of other
26940 Installing GCC automatically creates and installs the fixed header
26941 files, by running a program called `fixincludes'. Normally, you don't
26942 need to pay attention to this. But there are cases where it doesn't do
26943 the right thing automatically.
26945 * If you update the system's header files, such as by installing a
26946 new system version, the fixed header files of GCC are not
26947 automatically updated. They can be updated using the `mkheaders'
26948 script installed in `LIBEXECDIR/gcc/TARGET/VERSION/install-tools/'.
26950 * On some systems, header file directories contain machine-specific
26951 symbolic links in certain places. This makes it possible to share
26952 most of the header files among hosts running the same version of
26953 the system on different machine models.
26955 The programs that fix the header files do not understand this
26956 special way of using symbolic links; therefore, the directory of
26957 fixed header files is good only for the machine model used to
26960 It is possible to make separate sets of fixed header files for the
26961 different machine models, and arrange a structure of symbolic
26962 links so as to use the proper set, but you'll have to do this by
26966 File: gcc.info, Node: Standard Libraries, Next: Disappointments, Prev: Fixed Headers, Up: Trouble
26968 10.6 Standard Libraries
26969 =======================
26971 GCC by itself attempts to be a conforming freestanding implementation.
26972 *Note Language Standards Supported by GCC: Standards, for details of
26973 what this means. Beyond the library facilities required of such an
26974 implementation, the rest of the C library is supplied by the vendor of
26975 the operating system. If that C library doesn't conform to the C
26976 standards, then your programs might get warnings (especially when using
26977 `-Wall') that you don't expect.
26979 For example, the `sprintf' function on SunOS 4.1.3 returns `char *'
26980 while the C standard says that `sprintf' returns an `int'. The
26981 `fixincludes' program could make the prototype for this function match
26982 the Standard, but that would be wrong, since the function will still
26985 If you need a Standard compliant library, then you need to find one, as
26986 GCC does not provide one. The GNU C library (called `glibc') provides
26987 ISO C, POSIX, BSD, SystemV and X/Open compatibility for GNU/Linux and
26988 HURD-based GNU systems; no recent version of it supports other systems,
26989 though some very old versions did. Version 2.2 of the GNU C library
26990 includes nearly complete C99 support. You could also ask your
26991 operating system vendor if newer libraries are available.
26994 File: gcc.info, Node: Disappointments, Next: C++ Misunderstandings, Prev: Standard Libraries, Up: Trouble
26996 10.7 Disappointments and Misunderstandings
26997 ==========================================
26999 These problems are perhaps regrettable, but we don't know any practical
27002 * Certain local variables aren't recognized by debuggers when you
27003 compile with optimization.
27005 This occurs because sometimes GCC optimizes the variable out of
27006 existence. There is no way to tell the debugger how to compute the
27007 value such a variable "would have had", and it is not clear that
27008 would be desirable anyway. So GCC simply does not mention the
27009 eliminated variable when it writes debugging information.
27011 You have to expect a certain amount of disagreement between the
27012 executable and your source code, when you use optimization.
27014 * Users often think it is a bug when GCC reports an error for code
27017 int foo (struct mumble *);
27019 struct mumble { ... };
27021 int foo (struct mumble *x)
27024 This code really is erroneous, because the scope of `struct
27025 mumble' in the prototype is limited to the argument list
27026 containing it. It does not refer to the `struct mumble' defined
27027 with file scope immediately below--they are two unrelated types
27028 with similar names in different scopes.
27030 But in the definition of `foo', the file-scope type is used
27031 because that is available to be inherited. Thus, the definition
27032 and the prototype do not match, and you get an error.
27034 This behavior may seem silly, but it's what the ISO standard
27035 specifies. It is easy enough for you to make your code work by
27036 moving the definition of `struct mumble' above the prototype.
27037 It's not worth being incompatible with ISO C just to avoid an
27038 error for the example shown above.
27040 * Accesses to bit-fields even in volatile objects works by accessing
27041 larger objects, such as a byte or a word. You cannot rely on what
27042 size of object is accessed in order to read or write the
27043 bit-field; it may even vary for a given bit-field according to the
27046 If you care about controlling the amount of memory that is
27047 accessed, use volatile but do not use bit-fields.
27049 * GCC comes with shell scripts to fix certain known problems in
27050 system header files. They install corrected copies of various
27051 header files in a special directory where only GCC will normally
27052 look for them. The scripts adapt to various systems by searching
27053 all the system header files for the problem cases that we know
27056 If new system header files are installed, nothing automatically
27057 arranges to update the corrected header files. They can be
27058 updated using the `mkheaders' script installed in
27059 `LIBEXECDIR/gcc/TARGET/VERSION/install-tools/'.
27061 * On 68000 and x86 systems, for instance, you can get paradoxical
27062 results if you test the precise values of floating point numbers.
27063 For example, you can find that a floating point value which is not
27064 a NaN is not equal to itself. This results from the fact that the
27065 floating point registers hold a few more bits of precision than
27066 fit in a `double' in memory. Compiled code moves values between
27067 memory and floating point registers at its convenience, and moving
27068 them into memory truncates them.
27070 You can partially avoid this problem by using the `-ffloat-store'
27071 option (*note Optimize Options::).
27073 * On AIX and other platforms without weak symbol support, templates
27074 need to be instantiated explicitly and symbols for static members
27075 of templates will not be generated.
27077 * On AIX, GCC scans object files and library archives for static
27078 constructors and destructors when linking an application before the
27079 linker prunes unreferenced symbols. This is necessary to prevent
27080 the AIX linker from mistakenly assuming that static constructor or
27081 destructor are unused and removing them before the scanning can
27082 occur. All static constructors and destructors found will be
27083 referenced even though the modules in which they occur may not be
27084 used by the program. This may lead to both increased executable
27085 size and unexpected symbol references.
27088 File: gcc.info, Node: C++ Misunderstandings, Next: Protoize Caveats, Prev: Disappointments, Up: Trouble
27090 10.8 Common Misunderstandings with GNU C++
27091 ==========================================
27093 C++ is a complex language and an evolving one, and its standard
27094 definition (the ISO C++ standard) was only recently completed. As a
27095 result, your C++ compiler may occasionally surprise you, even when its
27096 behavior is correct. This section discusses some areas that frequently
27097 give rise to questions of this sort.
27101 * Static Definitions:: Static member declarations are not definitions
27102 * Name lookup:: Name lookup, templates, and accessing members of base classes
27103 * Temporaries:: Temporaries may vanish before you expect
27104 * Copy Assignment:: Copy Assignment operators copy virtual bases twice
27107 File: gcc.info, Node: Static Definitions, Next: Name lookup, Up: C++ Misunderstandings
27109 10.8.1 Declare _and_ Define Static Members
27110 ------------------------------------------
27112 When a class has static data members, it is not enough to _declare_ the
27113 static member; you must also _define_ it. For example:
27122 This declaration only establishes that the class `Foo' has an `int'
27123 named `Foo::bar', and a member function named `Foo::method'. But you
27124 still need to define _both_ `method' and `bar' elsewhere. According to
27125 the ISO standard, you must supply an initializer in one (and only one)
27126 source file, such as:
27130 Other C++ compilers may not correctly implement the standard behavior.
27131 As a result, when you switch to `g++' from one of these compilers, you
27132 may discover that a program that appeared to work correctly in fact
27133 does not conform to the standard: `g++' reports as undefined symbols
27134 any static data members that lack definitions.
27137 File: gcc.info, Node: Name lookup, Next: Temporaries, Prev: Static Definitions, Up: C++ Misunderstandings
27139 10.8.2 Name lookup, templates, and accessing members of base classes
27140 --------------------------------------------------------------------
27142 The C++ standard prescribes that all names that are not dependent on
27143 template parameters are bound to their present definitions when parsing
27144 a template function or class.(1) Only names that are dependent are
27145 looked up at the point of instantiation. For example, consider
27150 template <typename T>
27159 static const int N;
27162 Here, the names `foo' and `N' appear in a context that does not depend
27163 on the type of `T'. The compiler will thus require that they are
27164 defined in the context of use in the template, not only before the
27165 point of instantiation, and will here use `::foo(double)' and `A::N',
27166 respectively. In particular, it will convert the integer value to a
27167 `double' when passing it to `::foo(double)'.
27169 Conversely, `bar' and the call to `foo' in the fourth marked line are
27170 used in contexts that do depend on the type of `T', so they are only
27171 looked up at the point of instantiation, and you can provide
27172 declarations for them after declaring the template, but before
27173 instantiating it. In particular, if you instantiate `A::f<int>', the
27174 last line will call an overloaded `::foo(int)' if one was provided,
27175 even if after the declaration of `struct A'.
27177 This distinction between lookup of dependent and non-dependent names is
27178 called two-stage (or dependent) name lookup. G++ implements it since
27181 Two-stage name lookup sometimes leads to situations with behavior
27182 different from non-template codes. The most common is probably this:
27184 template <typename T> struct Base {
27188 template <typename T> struct Derived : public Base<T> {
27189 int get_i() { return i; }
27192 In `get_i()', `i' is not used in a dependent context, so the compiler
27193 will look for a name declared at the enclosing namespace scope (which
27194 is the global scope here). It will not look into the base class, since
27195 that is dependent and you may declare specializations of `Base' even
27196 after declaring `Derived', so the compiler can't really know what `i'
27197 would refer to. If there is no global variable `i', then you will get
27200 In order to make it clear that you want the member of the base class,
27201 you need to defer lookup until instantiation time, at which the base
27202 class is known. For this, you need to access `i' in a dependent
27203 context, by either using `this->i' (remember that `this' is of type
27204 `Derived<T>*', so is obviously dependent), or using `Base<T>::i'.
27205 Alternatively, `Base<T>::i' might be brought into scope by a
27206 `using'-declaration.
27208 Another, similar example involves calling member functions of a base
27211 template <typename T> struct Base {
27215 template <typename T> struct Derived : Base<T> {
27216 int g() { return f(); };
27219 Again, the call to `f()' is not dependent on template arguments (there
27220 are no arguments that depend on the type `T', and it is also not
27221 otherwise specified that the call should be in a dependent context).
27222 Thus a global declaration of such a function must be available, since
27223 the one in the base class is not visible until instantiation time. The
27224 compiler will consequently produce the following error message:
27226 x.cc: In member function `int Derived<T>::g()':
27227 x.cc:6: error: there are no arguments to `f' that depend on a template
27228 parameter, so a declaration of `f' must be available
27229 x.cc:6: error: (if you use `-fpermissive', G++ will accept your code, but
27230 allowing the use of an undeclared name is deprecated)
27232 To make the code valid either use `this->f()', or `Base<T>::f()'.
27233 Using the `-fpermissive' flag will also let the compiler accept the
27234 code, by marking all function calls for which no declaration is visible
27235 at the time of definition of the template for later lookup at
27236 instantiation time, as if it were a dependent call. We do not
27237 recommend using `-fpermissive' to work around invalid code, and it will
27238 also only catch cases where functions in base classes are called, not
27239 where variables in base classes are used (as in the example above).
27241 Note that some compilers (including G++ versions prior to 3.4) get
27242 these examples wrong and accept above code without an error. Those
27243 compilers do not implement two-stage name lookup correctly.
27245 ---------- Footnotes ----------
27247 (1) The C++ standard just uses the term "dependent" for names that
27248 depend on the type or value of template parameters. This shorter term
27249 will also be used in the rest of this section.
27252 File: gcc.info, Node: Temporaries, Next: Copy Assignment, Prev: Name lookup, Up: C++ Misunderstandings
27254 10.8.3 Temporaries May Vanish Before You Expect
27255 -----------------------------------------------
27257 It is dangerous to use pointers or references to _portions_ of a
27258 temporary object. The compiler may very well delete the object before
27259 you expect it to, leaving a pointer to garbage. The most common place
27260 where this problem crops up is in classes like string classes,
27261 especially ones that define a conversion function to type `char *' or
27262 `const char *'--which is one reason why the standard `string' class
27263 requires you to call the `c_str' member function. However, any class
27264 that returns a pointer to some internal structure is potentially
27265 subject to this problem.
27267 For example, a program may use a function `strfunc' that returns
27268 `string' objects, and another function `charfunc' that operates on
27269 pointers to `char':
27272 void charfunc (const char *);
27277 const char *p = strfunc().c_str();
27284 In this situation, it may seem reasonable to save a pointer to the C
27285 string returned by the `c_str' member function and use that rather than
27286 call `c_str' repeatedly. However, the temporary string created by the
27287 call to `strfunc' is destroyed after `p' is initialized, at which point
27288 `p' is left pointing to freed memory.
27290 Code like this may run successfully under some other compilers,
27291 particularly obsolete cfront-based compilers that delete temporaries
27292 along with normal local variables. However, the GNU C++ behavior is
27293 standard-conforming, so if your program depends on late destruction of
27294 temporaries it is not portable.
27296 The safe way to write such code is to give the temporary a name, which
27297 forces it to remain until the end of the scope of the name. For
27300 const string& tmp = strfunc ();
27301 charfunc (tmp.c_str ());
27304 File: gcc.info, Node: Copy Assignment, Prev: Temporaries, Up: C++ Misunderstandings
27306 10.8.4 Implicit Copy-Assignment for Virtual Bases
27307 -------------------------------------------------
27309 When a base class is virtual, only one subobject of the base class
27310 belongs to each full object. Also, the constructors and destructors are
27311 invoked only once, and called from the most-derived class. However,
27312 such objects behave unspecified when being assigned. For example:
27316 Base(char *n) : name(strdup(n)){}
27317 Base& operator= (const Base& other){
27319 name = strdup (other.name);
27323 struct A:virtual Base{
27328 struct B:virtual Base{
27333 struct Derived:public A, public B{
27334 Derived():Base("Derived"){}
27337 void func(Derived &d1, Derived &d2)
27342 The C++ standard specifies that `Base::Base' is only called once when
27343 constructing or copy-constructing a Derived object. It is unspecified
27344 whether `Base::operator=' is called more than once when the implicit
27345 copy-assignment for Derived objects is invoked (as it is inside `func'
27348 G++ implements the "intuitive" algorithm for copy-assignment: assign
27349 all direct bases, then assign all members. In that algorithm, the
27350 virtual base subobject can be encountered more than once. In the
27351 example, copying proceeds in the following order: `val', `name' (via
27352 `strdup'), `bval', and `name' again.
27354 If application code relies on copy-assignment, a user-defined
27355 copy-assignment operator removes any uncertainties. With such an
27356 operator, the application can define whether and how the virtual base
27357 subobject is assigned.
27360 File: gcc.info, Node: Protoize Caveats, Next: Non-bugs, Prev: C++ Misunderstandings, Up: Trouble
27362 10.9 Caveats of using `protoize'
27363 ================================
27365 The conversion programs `protoize' and `unprotoize' can sometimes
27366 change a source file in a way that won't work unless you rearrange it.
27368 * `protoize' can insert references to a type name or type tag before
27369 the definition, or in a file where they are not defined.
27371 If this happens, compiler error messages should show you where the
27372 new references are, so fixing the file by hand is straightforward.
27374 * There are some C constructs which `protoize' cannot figure out.
27375 For example, it can't determine argument types for declaring a
27376 pointer-to-function variable; this you must do by hand. `protoize'
27377 inserts a comment containing `???' each time it finds such a
27378 variable; so you can find all such variables by searching for this
27379 string. ISO C does not require declaring the argument types of
27380 pointer-to-function types.
27382 * Using `unprotoize' can easily introduce bugs. If the program
27383 relied on prototypes to bring about conversion of arguments, these
27384 conversions will not take place in the program without prototypes.
27385 One case in which you can be sure `unprotoize' is safe is when you
27386 are removing prototypes that were made with `protoize'; if the
27387 program worked before without any prototypes, it will work again
27390 You can find all the places where this problem might occur by
27391 compiling the program with the `-Wconversion' option. It prints a
27392 warning whenever an argument is converted.
27394 * Both conversion programs can be confused if there are macro calls
27395 in and around the text to be converted. In other words, the
27396 standard syntax for a declaration or definition must not result
27397 from expanding a macro. This problem is inherent in the design of
27398 C and cannot be fixed. If only a few functions have confusing
27399 macro calls, you can easily convert them manually.
27401 * `protoize' cannot get the argument types for a function whose
27402 definition was not actually compiled due to preprocessing
27403 conditionals. When this happens, `protoize' changes nothing in
27404 regard to such a function. `protoize' tries to detect such
27405 instances and warn about them.
27407 You can generally work around this problem by using `protoize' step
27408 by step, each time specifying a different set of `-D' options for
27409 compilation, until all of the functions have been converted.
27410 There is no automatic way to verify that you have got them all,
27413 * Confusion may result if there is an occasion to convert a function
27414 declaration or definition in a region of source code where there
27415 is more than one formal parameter list present. Thus, attempts to
27416 convert code containing multiple (conditionally compiled) versions
27417 of a single function header (in the same vicinity) may not produce
27418 the desired (or expected) results.
27420 If you plan on converting source files which contain such code, it
27421 is recommended that you first make sure that each conditionally
27422 compiled region of source code which contains an alternative
27423 function header also contains at least one additional follower
27424 token (past the final right parenthesis of the function header).
27425 This should circumvent the problem.
27427 * `unprotoize' can become confused when trying to convert a function
27428 definition or declaration which contains a declaration for a
27429 pointer-to-function formal argument which has the same name as the
27430 function being defined or declared. We recommend you avoid such
27431 choices of formal parameter names.
27433 * You might also want to correct some of the indentation by hand and
27434 break long lines. (The conversion programs don't write lines
27435 longer than eighty characters in any case.)
27438 File: gcc.info, Node: Non-bugs, Next: Warnings and Errors, Prev: Protoize Caveats, Up: Trouble
27440 10.10 Certain Changes We Don't Want to Make
27441 ===========================================
27443 This section lists changes that people frequently request, but which we
27444 do not make because we think GCC is better without them.
27446 * Checking the number and type of arguments to a function which has
27447 an old-fashioned definition and no prototype.
27449 Such a feature would work only occasionally--only for calls that
27450 appear in the same file as the called function, following the
27451 definition. The only way to check all calls reliably is to add a
27452 prototype for the function. But adding a prototype eliminates the
27453 motivation for this feature. So the feature is not worthwhile.
27455 * Warning about using an expression whose type is signed as a shift
27458 Shift count operands are probably signed more often than unsigned.
27459 Warning about this would cause far more annoyance than good.
27461 * Warning about assigning a signed value to an unsigned variable.
27463 Such assignments must be very common; warning about them would
27464 cause more annoyance than good.
27466 * Warning when a non-void function value is ignored.
27468 C contains many standard functions that return a value that most
27469 programs choose to ignore. One obvious example is `printf'.
27470 Warning about this practice only leads the defensive programmer to
27471 clutter programs with dozens of casts to `void'. Such casts are
27472 required so frequently that they become visual noise. Writing
27473 those casts becomes so automatic that they no longer convey useful
27474 information about the intentions of the programmer. For functions
27475 where the return value should never be ignored, use the
27476 `warn_unused_result' function attribute (*note Function
27479 * Making `-fshort-enums' the default.
27481 This would cause storage layout to be incompatible with most other
27482 C compilers. And it doesn't seem very important, given that you
27483 can get the same result in other ways. The case where it matters
27484 most is when the enumeration-valued object is inside a structure,
27485 and in that case you can specify a field width explicitly.
27487 * Making bit-fields unsigned by default on particular machines where
27488 "the ABI standard" says to do so.
27490 The ISO C standard leaves it up to the implementation whether a
27491 bit-field declared plain `int' is signed or not. This in effect
27492 creates two alternative dialects of C.
27494 The GNU C compiler supports both dialects; you can specify the
27495 signed dialect with `-fsigned-bitfields' and the unsigned dialect
27496 with `-funsigned-bitfields'. However, this leaves open the
27497 question of which dialect to use by default.
27499 Currently, the preferred dialect makes plain bit-fields signed,
27500 because this is simplest. Since `int' is the same as `signed int'
27501 in every other context, it is cleanest for them to be the same in
27502 bit-fields as well.
27504 Some computer manufacturers have published Application Binary
27505 Interface standards which specify that plain bit-fields should be
27506 unsigned. It is a mistake, however, to say anything about this
27507 issue in an ABI. This is because the handling of plain bit-fields
27508 distinguishes two dialects of C. Both dialects are meaningful on
27509 every type of machine. Whether a particular object file was
27510 compiled using signed bit-fields or unsigned is of no concern to
27511 other object files, even if they access the same bit-fields in the
27512 same data structures.
27514 A given program is written in one or the other of these two
27515 dialects. The program stands a chance to work on most any machine
27516 if it is compiled with the proper dialect. It is unlikely to work
27517 at all if compiled with the wrong dialect.
27519 Many users appreciate the GNU C compiler because it provides an
27520 environment that is uniform across machines. These users would be
27521 inconvenienced if the compiler treated plain bit-fields
27522 differently on certain machines.
27524 Occasionally users write programs intended only for a particular
27525 machine type. On these occasions, the users would benefit if the
27526 GNU C compiler were to support by default the same dialect as the
27527 other compilers on that machine. But such applications are rare.
27528 And users writing a program to run on more than one type of
27529 machine cannot possibly benefit from this kind of compatibility.
27531 This is why GCC does and will treat plain bit-fields in the same
27532 fashion on all types of machines (by default).
27534 There are some arguments for making bit-fields unsigned by default
27535 on all machines. If, for example, this becomes a universal de
27536 facto standard, it would make sense for GCC to go along with it.
27537 This is something to be considered in the future.
27539 (Of course, users strongly concerned about portability should
27540 indicate explicitly in each bit-field whether it is signed or not.
27541 In this way, they write programs which have the same meaning in
27544 * Undefining `__STDC__' when `-ansi' is not used.
27546 Currently, GCC defines `__STDC__' unconditionally. This provides
27547 good results in practice.
27549 Programmers normally use conditionals on `__STDC__' to ask whether
27550 it is safe to use certain features of ISO C, such as function
27551 prototypes or ISO token concatenation. Since plain `gcc' supports
27552 all the features of ISO C, the correct answer to these questions is
27555 Some users try to use `__STDC__' to check for the availability of
27556 certain library facilities. This is actually incorrect usage in
27557 an ISO C program, because the ISO C standard says that a conforming
27558 freestanding implementation should define `__STDC__' even though it
27559 does not have the library facilities. `gcc -ansi -pedantic' is a
27560 conforming freestanding implementation, and it is therefore
27561 required to define `__STDC__', even though it does not come with
27564 Sometimes people say that defining `__STDC__' in a compiler that
27565 does not completely conform to the ISO C standard somehow violates
27566 the standard. This is illogical. The standard is a standard for
27567 compilers that claim to support ISO C, such as `gcc -ansi'--not
27568 for other compilers such as plain `gcc'. Whatever the ISO C
27569 standard says is relevant to the design of plain `gcc' without
27570 `-ansi' only for pragmatic reasons, not as a requirement.
27572 GCC normally defines `__STDC__' to be 1, and in addition defines
27573 `__STRICT_ANSI__' if you specify the `-ansi' option, or a `-std'
27574 option for strict conformance to some version of ISO C. On some
27575 hosts, system include files use a different convention, where
27576 `__STDC__' is normally 0, but is 1 if the user specifies strict
27577 conformance to the C Standard. GCC follows the host convention
27578 when processing system include files, but when processing user
27579 files it follows the usual GNU C convention.
27581 * Undefining `__STDC__' in C++.
27583 Programs written to compile with C++-to-C translators get the
27584 value of `__STDC__' that goes with the C compiler that is
27585 subsequently used. These programs must test `__STDC__' to
27586 determine what kind of C preprocessor that compiler uses: whether
27587 they should concatenate tokens in the ISO C fashion or in the
27588 traditional fashion.
27590 These programs work properly with GNU C++ if `__STDC__' is defined.
27591 They would not work otherwise.
27593 In addition, many header files are written to provide prototypes
27594 in ISO C but not in traditional C. Many of these header files can
27595 work without change in C++ provided `__STDC__' is defined. If
27596 `__STDC__' is not defined, they will all fail, and will all need
27597 to be changed to test explicitly for C++ as well.
27599 * Deleting "empty" loops.
27601 Historically, GCC has not deleted "empty" loops under the
27602 assumption that the most likely reason you would put one in a
27603 program is to have a delay, so deleting them will not make real
27604 programs run any faster.
27606 However, the rationale here is that optimization of a nonempty loop
27607 cannot produce an empty one. This held for carefully written C
27608 compiled with less powerful optimizers but is not always the case
27609 for carefully written C++ or with more powerful optimizers. Thus
27610 GCC will remove operations from loops whenever it can determine
27611 those operations are not externally visible (apart from the time
27612 taken to execute them, of course). In case the loop can be proved
27613 to be finite, GCC will also remove the loop itself.
27615 Be aware of this when performing timing tests, for instance the
27616 following loop can be completely removed, provided
27617 `some_expression' can provably not change any global state.
27623 for (ix = 0; ix != 10000; ix++)
27624 sum += some_expression;
27627 Even though `sum' is accumulated in the loop, no use is made of
27628 that summation, so the accumulation can be removed.
27630 * Making side effects happen in the same order as in some other
27633 It is never safe to depend on the order of evaluation of side
27634 effects. For example, a function call like this may very well
27635 behave differently from one compiler to another:
27637 void func (int, int);
27642 There is no guarantee (in either the C or the C++ standard language
27643 definitions) that the increments will be evaluated in any
27644 particular order. Either increment might happen first. `func'
27645 might get the arguments `2, 3', or it might get `3, 2', or even
27648 * Making certain warnings into errors by default.
27650 Some ISO C testsuites report failure when the compiler does not
27651 produce an error message for a certain program.
27653 ISO C requires a "diagnostic" message for certain kinds of invalid
27654 programs, but a warning is defined by GCC to count as a
27655 diagnostic. If GCC produces a warning but not an error, that is
27656 correct ISO C support. If testsuites call this "failure", they
27657 should be run with the GCC option `-pedantic-errors', which will
27658 turn these warnings into errors.
27662 File: gcc.info, Node: Warnings and Errors, Prev: Non-bugs, Up: Trouble
27664 10.11 Warning Messages and Error Messages
27665 =========================================
27667 The GNU compiler can produce two kinds of diagnostics: errors and
27668 warnings. Each kind has a different purpose:
27670 "Errors" report problems that make it impossible to compile your
27671 program. GCC reports errors with the source file name and line
27672 number where the problem is apparent.
27674 "Warnings" report other unusual conditions in your code that _may_
27675 indicate a problem, although compilation can (and does) proceed.
27676 Warning messages also report the source file name and line number,
27677 but include the text `warning:' to distinguish them from error
27680 Warnings may indicate danger points where you should check to make sure
27681 that your program really does what you intend; or the use of obsolete
27682 features; or the use of nonstandard features of GNU C or C++. Many
27683 warnings are issued only if you ask for them, with one of the `-W'
27684 options (for instance, `-Wall' requests a variety of useful warnings).
27686 GCC always tries to compile your program if possible; it never
27687 gratuitously rejects a program whose meaning is clear merely because
27688 (for instance) it fails to conform to a standard. In some cases,
27689 however, the C and C++ standards specify that certain extensions are
27690 forbidden, and a diagnostic _must_ be issued by a conforming compiler.
27691 The `-pedantic' option tells GCC to issue warnings in such cases;
27692 `-pedantic-errors' says to make them errors instead. This does not
27693 mean that _all_ non-ISO constructs get warnings or errors.
27695 *Note Options to Request or Suppress Warnings: Warning Options, for
27696 more detail on these and related command-line options.
27699 File: gcc.info, Node: Bugs, Next: Service, Prev: Trouble, Up: Top
27704 Your bug reports play an essential role in making GCC reliable.
27706 When you encounter a problem, the first thing to do is to see if it is
27707 already known. *Note Trouble::. If it isn't known, then you should
27708 report the problem.
27712 * Criteria: Bug Criteria. Have you really found a bug?
27713 * Reporting: Bug Reporting. How to report a bug effectively.
27714 * Known: Trouble. Known problems.
27715 * Help: Service. Where to ask for help.
27718 File: gcc.info, Node: Bug Criteria, Next: Bug Reporting, Up: Bugs
27720 11.1 Have You Found a Bug?
27721 ==========================
27723 If you are not sure whether you have found a bug, here are some
27726 * If the compiler gets a fatal signal, for any input whatever, that
27727 is a compiler bug. Reliable compilers never crash.
27729 * If the compiler produces invalid assembly code, for any input
27730 whatever (except an `asm' statement), that is a compiler bug,
27731 unless the compiler reports errors (not just warnings) which would
27732 ordinarily prevent the assembler from being run.
27734 * If the compiler produces valid assembly code that does not
27735 correctly execute the input source code, that is a compiler bug.
27737 However, you must double-check to make sure, because you may have a
27738 program whose behavior is undefined, which happened by chance to
27739 give the desired results with another C or C++ compiler.
27741 For example, in many nonoptimizing compilers, you can write `x;'
27742 at the end of a function instead of `return x;', with the same
27743 results. But the value of the function is undefined if `return'
27744 is omitted; it is not a bug when GCC produces different results.
27746 Problems often result from expressions with two increment
27747 operators, as in `f (*p++, *p++)'. Your previous compiler might
27748 have interpreted that expression the way you intended; GCC might
27749 interpret it another way. Neither compiler is wrong. The bug is
27752 After you have localized the error to a single source line, it
27753 should be easy to check for these things. If your program is
27754 correct and well defined, you have found a compiler bug.
27756 * If the compiler produces an error message for valid input, that is
27759 * If the compiler does not produce an error message for invalid
27760 input, that is a compiler bug. However, you should note that your
27761 idea of "invalid input" might be someone else's idea of "an
27762 extension" or "support for traditional practice".
27764 * If you are an experienced user of one of the languages GCC
27765 supports, your suggestions for improvement of GCC are welcome in
27769 File: gcc.info, Node: Bug Reporting, Prev: Bug Criteria, Up: Bugs
27771 11.2 How and where to Report Bugs
27772 =================================
27774 Bugs should be reported to the GCC bug database. Please refer to
27775 `http://gcc.gnu.org/bugs.html' for up-to-date instructions how to
27776 submit bug reports. Copies of this file in HTML (`bugs.html') and
27777 plain text (`BUGS') are also part of GCC releases.
27780 File: gcc.info, Node: Service, Next: Contributing, Prev: Bugs, Up: Top
27782 12 How To Get Help with GCC
27783 ***************************
27785 If you need help installing, using or changing GCC, there are two ways
27788 * Send a message to a suitable network mailing list. First try
27789 <gcc-help@gcc.gnu.org> (for help installing or using GCC), and if
27790 that brings no response, try <gcc@gcc.gnu.org>. For help changing
27791 GCC, ask <gcc@gcc.gnu.org>. If you think you have found a bug in
27792 GCC, please report it following the instructions at *note Bug
27795 * Look in the service directory for someone who might help you for a
27796 fee. The service directory is found at
27797 `http://www.gnu.org/prep/service.html'.
27799 For further information, see `http://gcc.gnu.org/faq.html#support'.
27802 File: gcc.info, Node: Contributing, Next: Funding, Prev: Service, Up: Top
27804 13 Contributing to GCC Development
27805 **********************************
27807 If you would like to help pretest GCC releases to assure they work well,
27808 current development sources are available by SVN (see
27809 `http://gcc.gnu.org/svn.html'). Source and binary snapshots are also
27810 available for FTP; see `http://gcc.gnu.org/snapshots.html'.
27812 If you would like to work on improvements to GCC, please read the
27813 advice at these URLs:
27815 `http://gcc.gnu.org/contribute.html'
27816 `http://gcc.gnu.org/contributewhy.html'
27818 for information on how to make useful contributions and avoid
27819 duplication of effort. Suggested projects are listed at
27820 `http://gcc.gnu.org/projects/'.
27823 File: gcc.info, Node: Funding, Next: GNU Project, Prev: Contributing, Up: Top
27825 Funding Free Software
27826 *********************
27828 If you want to have more free software a few years from now, it makes
27829 sense for you to help encourage people to contribute funds for its
27830 development. The most effective approach known is to encourage
27831 commercial redistributors to donate.
27833 Users of free software systems can boost the pace of development by
27834 encouraging for-a-fee distributors to donate part of their selling price
27835 to free software developers--the Free Software Foundation, and others.
27837 The way to convince distributors to do this is to demand it and expect
27838 it from them. So when you compare distributors, judge them partly by
27839 how much they give to free software development. Show distributors
27840 they must compete to be the one who gives the most.
27842 To make this approach work, you must insist on numbers that you can
27843 compare, such as, "We will donate ten dollars to the Frobnitz project
27844 for each disk sold." Don't be satisfied with a vague promise, such as
27845 "A portion of the profits are donated," since it doesn't give a basis
27848 Even a precise fraction "of the profits from this disk" is not very
27849 meaningful, since creative accounting and unrelated business decisions
27850 can greatly alter what fraction of the sales price counts as profit.
27851 If the price you pay is $50, ten percent of the profit is probably less
27852 than a dollar; it might be a few cents, or nothing at all.
27854 Some redistributors do development work themselves. This is useful
27855 too; but to keep everyone honest, you need to inquire how much they do,
27856 and what kind. Some kinds of development make much more long-term
27857 difference than others. For example, maintaining a separate version of
27858 a program contributes very little; maintaining the standard version of a
27859 program for the whole community contributes much. Easy new ports
27860 contribute little, since someone else would surely do them; difficult
27861 ports such as adding a new CPU to the GNU Compiler Collection
27862 contribute more; major new features or packages contribute the most.
27864 By establishing the idea that supporting further development is "the
27865 proper thing to do" when distributing free software for a fee, we can
27866 assure a steady flow of resources into making more free software.
27868 Copyright (C) 1994 Free Software Foundation, Inc.
27869 Verbatim copying and redistribution of this section is permitted
27870 without royalty; alteration is not permitted.
27873 File: gcc.info, Node: GNU Project, Next: Copying, Prev: Funding, Up: Top
27875 The GNU Project and GNU/Linux
27876 *****************************
27878 The GNU Project was launched in 1984 to develop a complete Unix-like
27879 operating system which is free software: the GNU system. (GNU is a
27880 recursive acronym for "GNU's Not Unix"; it is pronounced "guh-NEW".)
27881 Variants of the GNU operating system, which use the kernel Linux, are
27882 now widely used; though these systems are often referred to as "Linux",
27883 they are more accurately called GNU/Linux systems.
27885 For more information, see:
27886 `http://www.gnu.org/'
27887 `http://www.gnu.org/gnu/linux-and-gnu.html'
27890 File: gcc.info, Node: Copying, Next: GNU Free Documentation License, Prev: GNU Project, Up: Top
27892 GNU GENERAL PUBLIC LICENSE
27893 **************************
27895 Version 2, June 1991
27897 Copyright (C) 1989, 1991 Free Software Foundation, Inc.
27898 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA
27900 Everyone is permitted to copy and distribute verbatim copies
27901 of this license document, but changing it is not allowed.
27906 The licenses for most software are designed to take away your freedom
27907 to share and change it. By contrast, the GNU General Public License is
27908 intended to guarantee your freedom to share and change free
27909 software--to make sure the software is free for all its users. This
27910 General Public License applies to most of the Free Software
27911 Foundation's software and to any other program whose authors commit to
27912 using it. (Some other Free Software Foundation software is covered by
27913 the GNU Library General Public License instead.) You can apply it to
27914 your programs, too.
27916 When we speak of free software, we are referring to freedom, not
27917 price. Our General Public Licenses are designed to make sure that you
27918 have the freedom to distribute copies of free software (and charge for
27919 this service if you wish), that you receive source code or can get it
27920 if you want it, that you can change the software or use pieces of it in
27921 new free programs; and that you know you can do these things.
27923 To protect your rights, we need to make restrictions that forbid
27924 anyone to deny you these rights or to ask you to surrender the rights.
27925 These restrictions translate to certain responsibilities for you if you
27926 distribute copies of the software, or if you modify it.
27928 For example, if you distribute copies of such a program, whether
27929 gratis or for a fee, you must give the recipients all the rights that
27930 you have. You must make sure that they, too, receive or can get the
27931 source code. And you must show them these terms so they know their
27934 We protect your rights with two steps: (1) copyright the software, and
27935 (2) offer you this license which gives you legal permission to copy,
27936 distribute and/or modify the software.
27938 Also, for each author's protection and ours, we want to make certain
27939 that everyone understands that there is no warranty for this free
27940 software. If the software is modified by someone else and passed on, we
27941 want its recipients to know that what they have is not the original, so
27942 that any problems introduced by others will not reflect on the original
27943 authors' reputations.
27945 Finally, any free program is threatened constantly by software
27946 patents. We wish to avoid the danger that redistributors of a free
27947 program will individually obtain patent licenses, in effect making the
27948 program proprietary. To prevent this, we have made it clear that any
27949 patent must be licensed for everyone's free use or not licensed at all.
27951 The precise terms and conditions for copying, distribution and
27952 modification follow.
27954 TERMS AND CONDITIONS FOR COPYING, DISTRIBUTION AND MODIFICATION
27955 0. This License applies to any program or other work which contains a
27956 notice placed by the copyright holder saying it may be distributed
27957 under the terms of this General Public License. The "Program",
27958 below, refers to any such program or work, and a "work based on
27959 the Program" means either the Program or any derivative work under
27960 copyright law: that is to say, a work containing the Program or a
27961 portion of it, either verbatim or with modifications and/or
27962 translated into another language. (Hereinafter, translation is
27963 included without limitation in the term "modification".) Each
27964 licensee is addressed as "you".
27966 Activities other than copying, distribution and modification are
27967 not covered by this License; they are outside its scope. The act
27968 of running the Program is not restricted, and the output from the
27969 Program is covered only if its contents constitute a work based on
27970 the Program (independent of having been made by running the
27971 Program). Whether that is true depends on what the Program does.
27973 1. You may copy and distribute verbatim copies of the Program's
27974 source code as you receive it, in any medium, provided that you
27975 conspicuously and appropriately publish on each copy an appropriate
27976 copyright notice and disclaimer of warranty; keep intact all the
27977 notices that refer to this License and to the absence of any
27978 warranty; and give any other recipients of the Program a copy of
27979 this License along with the Program.
27981 You may charge a fee for the physical act of transferring a copy,
27982 and you may at your option offer warranty protection in exchange
27985 2. You may modify your copy or copies of the Program or any portion
27986 of it, thus forming a work based on the Program, and copy and
27987 distribute such modifications or work under the terms of Section 1
27988 above, provided that you also meet all of these conditions:
27990 a. You must cause the modified files to carry prominent notices
27991 stating that you changed the files and the date of any change.
27993 b. You must cause any work that you distribute or publish, that
27994 in whole or in part contains or is derived from the Program
27995 or any part thereof, to be licensed as a whole at no charge
27996 to all third parties under the terms of this License.
27998 c. If the modified program normally reads commands interactively
27999 when run, you must cause it, when started running for such
28000 interactive use in the most ordinary way, to print or display
28001 an announcement including an appropriate copyright notice and
28002 a notice that there is no warranty (or else, saying that you
28003 provide a warranty) and that users may redistribute the
28004 program under these conditions, and telling the user how to
28005 view a copy of this License. (Exception: if the Program
28006 itself is interactive but does not normally print such an
28007 announcement, your work based on the Program is not required
28008 to print an announcement.)
28010 These requirements apply to the modified work as a whole. If
28011 identifiable sections of that work are not derived from the
28012 Program, and can be reasonably considered independent and separate
28013 works in themselves, then this License, and its terms, do not
28014 apply to those sections when you distribute them as separate
28015 works. But when you distribute the same sections as part of a
28016 whole which is a work based on the Program, the distribution of
28017 the whole must be on the terms of this License, whose permissions
28018 for other licensees extend to the entire whole, and thus to each
28019 and every part regardless of who wrote it.
28021 Thus, it is not the intent of this section to claim rights or
28022 contest your rights to work written entirely by you; rather, the
28023 intent is to exercise the right to control the distribution of
28024 derivative or collective works based on the Program.
28026 In addition, mere aggregation of another work not based on the
28027 Program with the Program (or with a work based on the Program) on
28028 a volume of a storage or distribution medium does not bring the
28029 other work under the scope of this License.
28031 3. You may copy and distribute the Program (or a work based on it,
28032 under Section 2) in object code or executable form under the terms
28033 of Sections 1 and 2 above provided that you also do one of the
28036 a. Accompany it with the complete corresponding machine-readable
28037 source code, which must be distributed under the terms of
28038 Sections 1 and 2 above on a medium customarily used for
28039 software interchange; or,
28041 b. Accompany it with a written offer, valid for at least three
28042 years, to give any third party, for a charge no more than your
28043 cost of physically performing source distribution, a complete
28044 machine-readable copy of the corresponding source code, to be
28045 distributed under the terms of Sections 1 and 2 above on a
28046 medium customarily used for software interchange; or,
28048 c. Accompany it with the information you received as to the offer
28049 to distribute corresponding source code. (This alternative is
28050 allowed only for noncommercial distribution and only if you
28051 received the program in object code or executable form with
28052 such an offer, in accord with Subsection b above.)
28054 The source code for a work means the preferred form of the work for
28055 making modifications to it. For an executable work, complete
28056 source code means all the source code for all modules it contains,
28057 plus any associated interface definition files, plus the scripts
28058 used to control compilation and installation of the executable.
28059 However, as a special exception, the source code distributed need
28060 not include anything that is normally distributed (in either
28061 source or binary form) with the major components (compiler,
28062 kernel, and so on) of the operating system on which the executable
28063 runs, unless that component itself accompanies the executable.
28065 If distribution of executable or object code is made by offering
28066 access to copy from a designated place, then offering equivalent
28067 access to copy the source code from the same place counts as
28068 distribution of the source code, even though third parties are not
28069 compelled to copy the source along with the object code.
28071 4. You may not copy, modify, sublicense, or distribute the Program
28072 except as expressly provided under this License. Any attempt
28073 otherwise to copy, modify, sublicense or distribute the Program is
28074 void, and will automatically terminate your rights under this
28075 License. However, parties who have received copies, or rights,
28076 from you under this License will not have their licenses
28077 terminated so long as such parties remain in full compliance.
28079 5. You are not required to accept this License, since you have not
28080 signed it. However, nothing else grants you permission to modify
28081 or distribute the Program or its derivative works. These actions
28082 are prohibited by law if you do not accept this License.
28083 Therefore, by modifying or distributing the Program (or any work
28084 based on the Program), you indicate your acceptance of this
28085 License to do so, and all its terms and conditions for copying,
28086 distributing or modifying the Program or works based on it.
28088 6. Each time you redistribute the Program (or any work based on the
28089 Program), the recipient automatically receives a license from the
28090 original licensor to copy, distribute or modify the Program
28091 subject to these terms and conditions. You may not impose any
28092 further restrictions on the recipients' exercise of the rights
28093 granted herein. You are not responsible for enforcing compliance
28094 by third parties to this License.
28096 7. If, as a consequence of a court judgment or allegation of patent
28097 infringement or for any other reason (not limited to patent
28098 issues), conditions are imposed on you (whether by court order,
28099 agreement or otherwise) that contradict the conditions of this
28100 License, they do not excuse you from the conditions of this
28101 License. If you cannot distribute so as to satisfy simultaneously
28102 your obligations under this License and any other pertinent
28103 obligations, then as a consequence you may not distribute the
28104 Program at all. For example, if a patent license would not permit
28105 royalty-free redistribution of the Program by all those who
28106 receive copies directly or indirectly through you, then the only
28107 way you could satisfy both it and this License would be to refrain
28108 entirely from distribution of the Program.
28110 If any portion of this section is held invalid or unenforceable
28111 under any particular circumstance, the balance of the section is
28112 intended to apply and the section as a whole is intended to apply
28113 in other circumstances.
28115 It is not the purpose of this section to induce you to infringe any
28116 patents or other property right claims or to contest validity of
28117 any such claims; this section has the sole purpose of protecting
28118 the integrity of the free software distribution system, which is
28119 implemented by public license practices. Many people have made
28120 generous contributions to the wide range of software distributed
28121 through that system in reliance on consistent application of that
28122 system; it is up to the author/donor to decide if he or she is
28123 willing to distribute software through any other system and a
28124 licensee cannot impose that choice.
28126 This section is intended to make thoroughly clear what is believed
28127 to be a consequence of the rest of this License.
28129 8. If the distribution and/or use of the Program is restricted in
28130 certain countries either by patents or by copyrighted interfaces,
28131 the original copyright holder who places the Program under this
28132 License may add an explicit geographical distribution limitation
28133 excluding those countries, so that distribution is permitted only
28134 in or among countries not thus excluded. In such case, this
28135 License incorporates the limitation as if written in the body of
28138 9. The Free Software Foundation may publish revised and/or new
28139 versions of the General Public License from time to time. Such
28140 new versions will be similar in spirit to the present version, but
28141 may differ in detail to address new problems or concerns.
28143 Each version is given a distinguishing version number. If the
28144 Program specifies a version number of this License which applies
28145 to it and "any later version", you have the option of following
28146 the terms and conditions either of that version or of any later
28147 version published by the Free Software Foundation. If the Program
28148 does not specify a version number of this License, you may choose
28149 any version ever published by the Free Software Foundation.
28151 10. If you wish to incorporate parts of the Program into other free
28152 programs whose distribution conditions are different, write to the
28153 author to ask for permission. For software which is copyrighted
28154 by the Free Software Foundation, write to the Free Software
28155 Foundation; we sometimes make exceptions for this. Our decision
28156 will be guided by the two goals of preserving the free status of
28157 all derivatives of our free software and of promoting the sharing
28158 and reuse of software generally.
28161 11. BECAUSE THE PROGRAM IS LICENSED FREE OF CHARGE, THERE IS NO
28162 WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY APPLICABLE
28163 LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT
28164 HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT
28165 WARRANTY OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT
28166 NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
28167 FITNESS FOR A PARTICULAR PURPOSE. THE ENTIRE RISK AS TO THE
28168 QUALITY AND PERFORMANCE OF THE PROGRAM IS WITH YOU. SHOULD THE
28169 PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF ALL NECESSARY
28170 SERVICING, REPAIR OR CORRECTION.
28172 12. IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN
28173 WRITING WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MAY
28174 MODIFY AND/OR REDISTRIBUTE THE PROGRAM AS PERMITTED ABOVE, BE
28175 LIABLE TO YOU FOR DAMAGES, INCLUDING ANY GENERAL, SPECIAL,
28176 INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE USE OR
28177 INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF
28178 DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU
28179 OR THIRD PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY
28180 OTHER PROGRAMS), EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN
28181 ADVISED OF THE POSSIBILITY OF SUCH DAMAGES.
28183 END OF TERMS AND CONDITIONS
28184 Appendix: How to Apply These Terms to Your New Programs
28185 =======================================================
28187 If you develop a new program, and you want it to be of the greatest
28188 possible use to the public, the best way to achieve this is to make it
28189 free software which everyone can redistribute and change under these
28192 To do so, attach the following notices to the program. It is safest
28193 to attach them to the start of each source file to most effectively
28194 convey the exclusion of warranty; and each file should have at least
28195 the "copyright" line and a pointer to where the full notice is found.
28197 ONE LINE TO GIVE THE PROGRAM'S NAME AND A BRIEF IDEA OF WHAT IT DOES.
28198 Copyright (C) YEAR NAME OF AUTHOR
28200 This program is free software; you can redistribute it and/or modify
28201 it under the terms of the GNU General Public License as published by
28202 the Free Software Foundation; either version 2 of the License, or
28203 (at your option) any later version.
28205 This program is distributed in the hope that it will be useful,
28206 but WITHOUT ANY WARRANTY; without even the implied warranty of
28207 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
28208 GNU General Public License for more details.
28210 You should have received a copy of the GNU General Public License
28211 along with this program; if not, write to the Free Software
28212 Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA
28214 Also add information on how to contact you by electronic and paper
28217 If the program is interactive, make it output a short notice like this
28218 when it starts in an interactive mode:
28220 Gnomovision version 69, Copyright (C) YEAR NAME OF AUTHOR
28221 Gnomovision comes with ABSOLUTELY NO WARRANTY; for details
28223 This is free software, and you are welcome to redistribute it
28224 under certain conditions; type `show c' for details.
28226 The hypothetical commands `show w' and `show c' should show the
28227 appropriate parts of the General Public License. Of course, the
28228 commands you use may be called something other than `show w' and `show
28229 c'; they could even be mouse-clicks or menu items--whatever suits your
28232 You should also get your employer (if you work as a programmer) or your
28233 school, if any, to sign a "copyright disclaimer" for the program, if
28234 necessary. Here is a sample; alter the names:
28236 Yoyodyne, Inc., hereby disclaims all copyright interest in the program
28237 `Gnomovision' (which makes passes at compilers) written by James Hacker.
28239 SIGNATURE OF TY COON, 1 April 1989
28240 Ty Coon, President of Vice
28242 This General Public License does not permit incorporating your program
28243 into proprietary programs. If your program is a subroutine library,
28244 you may consider it more useful to permit linking proprietary
28245 applications with the library. If this is what you want to do, use the
28246 GNU Library General Public License instead of this License.
28249 File: gcc.info, Node: GNU Free Documentation License, Next: Contributors, Prev: Copying, Up: Top
28251 GNU Free Documentation License
28252 ******************************
28254 Version 1.2, November 2002
28256 Copyright (C) 2000,2001,2002 Free Software Foundation, Inc.
28257 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA
28259 Everyone is permitted to copy and distribute verbatim copies
28260 of this license document, but changing it is not allowed.
28264 The purpose of this License is to make a manual, textbook, or other
28265 functional and useful document "free" in the sense of freedom: to
28266 assure everyone the effective freedom to copy and redistribute it,
28267 with or without modifying it, either commercially or
28268 noncommercially. Secondarily, this License preserves for the
28269 author and publisher a way to get credit for their work, while not
28270 being considered responsible for modifications made by others.
28272 This License is a kind of "copyleft", which means that derivative
28273 works of the document must themselves be free in the same sense.
28274 It complements the GNU General Public License, which is a copyleft
28275 license designed for free software.
28277 We have designed this License in order to use it for manuals for
28278 free software, because free software needs free documentation: a
28279 free program should come with manuals providing the same freedoms
28280 that the software does. But this License is not limited to
28281 software manuals; it can be used for any textual work, regardless
28282 of subject matter or whether it is published as a printed book.
28283 We recommend this License principally for works whose purpose is
28284 instruction or reference.
28286 1. APPLICABILITY AND DEFINITIONS
28288 This License applies to any manual or other work, in any medium,
28289 that contains a notice placed by the copyright holder saying it
28290 can be distributed under the terms of this License. Such a notice
28291 grants a world-wide, royalty-free license, unlimited in duration,
28292 to use that work under the conditions stated herein. The
28293 "Document", below, refers to any such manual or work. Any member
28294 of the public is a licensee, and is addressed as "you". You
28295 accept the license if you copy, modify or distribute the work in a
28296 way requiring permission under copyright law.
28298 A "Modified Version" of the Document means any work containing the
28299 Document or a portion of it, either copied verbatim, or with
28300 modifications and/or translated into another language.
28302 A "Secondary Section" is a named appendix or a front-matter section
28303 of the Document that deals exclusively with the relationship of the
28304 publishers or authors of the Document to the Document's overall
28305 subject (or to related matters) and contains nothing that could
28306 fall directly within that overall subject. (Thus, if the Document
28307 is in part a textbook of mathematics, a Secondary Section may not
28308 explain any mathematics.) The relationship could be a matter of
28309 historical connection with the subject or with related matters, or
28310 of legal, commercial, philosophical, ethical or political position
28313 The "Invariant Sections" are certain Secondary Sections whose
28314 titles are designated, as being those of Invariant Sections, in
28315 the notice that says that the Document is released under this
28316 License. If a section does not fit the above definition of
28317 Secondary then it is not allowed to be designated as Invariant.
28318 The Document may contain zero Invariant Sections. If the Document
28319 does not identify any Invariant Sections then there are none.
28321 The "Cover Texts" are certain short passages of text that are
28322 listed, as Front-Cover Texts or Back-Cover Texts, in the notice
28323 that says that the Document is released under this License. A
28324 Front-Cover Text may be at most 5 words, and a Back-Cover Text may
28325 be at most 25 words.
28327 A "Transparent" copy of the Document means a machine-readable copy,
28328 represented in a format whose specification is available to the
28329 general public, that is suitable for revising the document
28330 straightforwardly with generic text editors or (for images
28331 composed of pixels) generic paint programs or (for drawings) some
28332 widely available drawing editor, and that is suitable for input to
28333 text formatters or for automatic translation to a variety of
28334 formats suitable for input to text formatters. A copy made in an
28335 otherwise Transparent file format whose markup, or absence of
28336 markup, has been arranged to thwart or discourage subsequent
28337 modification by readers is not Transparent. An image format is
28338 not Transparent if used for any substantial amount of text. A
28339 copy that is not "Transparent" is called "Opaque".
28341 Examples of suitable formats for Transparent copies include plain
28342 ASCII without markup, Texinfo input format, LaTeX input format,
28343 SGML or XML using a publicly available DTD, and
28344 standard-conforming simple HTML, PostScript or PDF designed for
28345 human modification. Examples of transparent image formats include
28346 PNG, XCF and JPG. Opaque formats include proprietary formats that
28347 can be read and edited only by proprietary word processors, SGML or
28348 XML for which the DTD and/or processing tools are not generally
28349 available, and the machine-generated HTML, PostScript or PDF
28350 produced by some word processors for output purposes only.
28352 The "Title Page" means, for a printed book, the title page itself,
28353 plus such following pages as are needed to hold, legibly, the
28354 material this License requires to appear in the title page. For
28355 works in formats which do not have any title page as such, "Title
28356 Page" means the text near the most prominent appearance of the
28357 work's title, preceding the beginning of the body of the text.
28359 A section "Entitled XYZ" means a named subunit of the Document
28360 whose title either is precisely XYZ or contains XYZ in parentheses
28361 following text that translates XYZ in another language. (Here XYZ
28362 stands for a specific section name mentioned below, such as
28363 "Acknowledgements", "Dedications", "Endorsements", or "History".)
28364 To "Preserve the Title" of such a section when you modify the
28365 Document means that it remains a section "Entitled XYZ" according
28366 to this definition.
28368 The Document may include Warranty Disclaimers next to the notice
28369 which states that this License applies to the Document. These
28370 Warranty Disclaimers are considered to be included by reference in
28371 this License, but only as regards disclaiming warranties: any other
28372 implication that these Warranty Disclaimers may have is void and
28373 has no effect on the meaning of this License.
28375 2. VERBATIM COPYING
28377 You may copy and distribute the Document in any medium, either
28378 commercially or noncommercially, provided that this License, the
28379 copyright notices, and the license notice saying this License
28380 applies to the Document are reproduced in all copies, and that you
28381 add no other conditions whatsoever to those of this License. You
28382 may not use technical measures to obstruct or control the reading
28383 or further copying of the copies you make or distribute. However,
28384 you may accept compensation in exchange for copies. If you
28385 distribute a large enough number of copies you must also follow
28386 the conditions in section 3.
28388 You may also lend copies, under the same conditions stated above,
28389 and you may publicly display copies.
28391 3. COPYING IN QUANTITY
28393 If you publish printed copies (or copies in media that commonly
28394 have printed covers) of the Document, numbering more than 100, and
28395 the Document's license notice requires Cover Texts, you must
28396 enclose the copies in covers that carry, clearly and legibly, all
28397 these Cover Texts: Front-Cover Texts on the front cover, and
28398 Back-Cover Texts on the back cover. Both covers must also clearly
28399 and legibly identify you as the publisher of these copies. The
28400 front cover must present the full title with all words of the
28401 title equally prominent and visible. You may add other material
28402 on the covers in addition. Copying with changes limited to the
28403 covers, as long as they preserve the title of the Document and
28404 satisfy these conditions, can be treated as verbatim copying in
28407 If the required texts for either cover are too voluminous to fit
28408 legibly, you should put the first ones listed (as many as fit
28409 reasonably) on the actual cover, and continue the rest onto
28412 If you publish or distribute Opaque copies of the Document
28413 numbering more than 100, you must either include a
28414 machine-readable Transparent copy along with each Opaque copy, or
28415 state in or with each Opaque copy a computer-network location from
28416 which the general network-using public has access to download
28417 using public-standard network protocols a complete Transparent
28418 copy of the Document, free of added material. If you use the
28419 latter option, you must take reasonably prudent steps, when you
28420 begin distribution of Opaque copies in quantity, to ensure that
28421 this Transparent copy will remain thus accessible at the stated
28422 location until at least one year after the last time you
28423 distribute an Opaque copy (directly or through your agents or
28424 retailers) of that edition to the public.
28426 It is requested, but not required, that you contact the authors of
28427 the Document well before redistributing any large number of
28428 copies, to give them a chance to provide you with an updated
28429 version of the Document.
28433 You may copy and distribute a Modified Version of the Document
28434 under the conditions of sections 2 and 3 above, provided that you
28435 release the Modified Version under precisely this License, with
28436 the Modified Version filling the role of the Document, thus
28437 licensing distribution and modification of the Modified Version to
28438 whoever possesses a copy of it. In addition, you must do these
28439 things in the Modified Version:
28441 A. Use in the Title Page (and on the covers, if any) a title
28442 distinct from that of the Document, and from those of
28443 previous versions (which should, if there were any, be listed
28444 in the History section of the Document). You may use the
28445 same title as a previous version if the original publisher of
28446 that version gives permission.
28448 B. List on the Title Page, as authors, one or more persons or
28449 entities responsible for authorship of the modifications in
28450 the Modified Version, together with at least five of the
28451 principal authors of the Document (all of its principal
28452 authors, if it has fewer than five), unless they release you
28453 from this requirement.
28455 C. State on the Title page the name of the publisher of the
28456 Modified Version, as the publisher.
28458 D. Preserve all the copyright notices of the Document.
28460 E. Add an appropriate copyright notice for your modifications
28461 adjacent to the other copyright notices.
28463 F. Include, immediately after the copyright notices, a license
28464 notice giving the public permission to use the Modified
28465 Version under the terms of this License, in the form shown in
28466 the Addendum below.
28468 G. Preserve in that license notice the full lists of Invariant
28469 Sections and required Cover Texts given in the Document's
28472 H. Include an unaltered copy of this License.
28474 I. Preserve the section Entitled "History", Preserve its Title,
28475 and add to it an item stating at least the title, year, new
28476 authors, and publisher of the Modified Version as given on
28477 the Title Page. If there is no section Entitled "History" in
28478 the Document, create one stating the title, year, authors,
28479 and publisher of the Document as given on its Title Page,
28480 then add an item describing the Modified Version as stated in
28481 the previous sentence.
28483 J. Preserve the network location, if any, given in the Document
28484 for public access to a Transparent copy of the Document, and
28485 likewise the network locations given in the Document for
28486 previous versions it was based on. These may be placed in
28487 the "History" section. You may omit a network location for a
28488 work that was published at least four years before the
28489 Document itself, or if the original publisher of the version
28490 it refers to gives permission.
28492 K. For any section Entitled "Acknowledgements" or "Dedications",
28493 Preserve the Title of the section, and preserve in the
28494 section all the substance and tone of each of the contributor
28495 acknowledgements and/or dedications given therein.
28497 L. Preserve all the Invariant Sections of the Document,
28498 unaltered in their text and in their titles. Section numbers
28499 or the equivalent are not considered part of the section
28502 M. Delete any section Entitled "Endorsements". Such a section
28503 may not be included in the Modified Version.
28505 N. Do not retitle any existing section to be Entitled
28506 "Endorsements" or to conflict in title with any Invariant
28509 O. Preserve any Warranty Disclaimers.
28511 If the Modified Version includes new front-matter sections or
28512 appendices that qualify as Secondary Sections and contain no
28513 material copied from the Document, you may at your option
28514 designate some or all of these sections as invariant. To do this,
28515 add their titles to the list of Invariant Sections in the Modified
28516 Version's license notice. These titles must be distinct from any
28517 other section titles.
28519 You may add a section Entitled "Endorsements", provided it contains
28520 nothing but endorsements of your Modified Version by various
28521 parties--for example, statements of peer review or that the text
28522 has been approved by an organization as the authoritative
28523 definition of a standard.
28525 You may add a passage of up to five words as a Front-Cover Text,
28526 and a passage of up to 25 words as a Back-Cover Text, to the end
28527 of the list of Cover Texts in the Modified Version. Only one
28528 passage of Front-Cover Text and one of Back-Cover Text may be
28529 added by (or through arrangements made by) any one entity. If the
28530 Document already includes a cover text for the same cover,
28531 previously added by you or by arrangement made by the same entity
28532 you are acting on behalf of, you may not add another; but you may
28533 replace the old one, on explicit permission from the previous
28534 publisher that added the old one.
28536 The author(s) and publisher(s) of the Document do not by this
28537 License give permission to use their names for publicity for or to
28538 assert or imply endorsement of any Modified Version.
28540 5. COMBINING DOCUMENTS
28542 You may combine the Document with other documents released under
28543 this License, under the terms defined in section 4 above for
28544 modified versions, provided that you include in the combination
28545 all of the Invariant Sections of all of the original documents,
28546 unmodified, and list them all as Invariant Sections of your
28547 combined work in its license notice, and that you preserve all
28548 their Warranty Disclaimers.
28550 The combined work need only contain one copy of this License, and
28551 multiple identical Invariant Sections may be replaced with a single
28552 copy. If there are multiple Invariant Sections with the same name
28553 but different contents, make the title of each such section unique
28554 by adding at the end of it, in parentheses, the name of the
28555 original author or publisher of that section if known, or else a
28556 unique number. Make the same adjustment to the section titles in
28557 the list of Invariant Sections in the license notice of the
28560 In the combination, you must combine any sections Entitled
28561 "History" in the various original documents, forming one section
28562 Entitled "History"; likewise combine any sections Entitled
28563 "Acknowledgements", and any sections Entitled "Dedications". You
28564 must delete all sections Entitled "Endorsements."
28566 6. COLLECTIONS OF DOCUMENTS
28568 You may make a collection consisting of the Document and other
28569 documents released under this License, and replace the individual
28570 copies of this License in the various documents with a single copy
28571 that is included in the collection, provided that you follow the
28572 rules of this License for verbatim copying of each of the
28573 documents in all other respects.
28575 You may extract a single document from such a collection, and
28576 distribute it individually under this License, provided you insert
28577 a copy of this License into the extracted document, and follow
28578 this License in all other respects regarding verbatim copying of
28581 7. AGGREGATION WITH INDEPENDENT WORKS
28583 A compilation of the Document or its derivatives with other
28584 separate and independent documents or works, in or on a volume of
28585 a storage or distribution medium, is called an "aggregate" if the
28586 copyright resulting from the compilation is not used to limit the
28587 legal rights of the compilation's users beyond what the individual
28588 works permit. When the Document is included in an aggregate, this
28589 License does not apply to the other works in the aggregate which
28590 are not themselves derivative works of the Document.
28592 If the Cover Text requirement of section 3 is applicable to these
28593 copies of the Document, then if the Document is less than one half
28594 of the entire aggregate, the Document's Cover Texts may be placed
28595 on covers that bracket the Document within the aggregate, or the
28596 electronic equivalent of covers if the Document is in electronic
28597 form. Otherwise they must appear on printed covers that bracket
28598 the whole aggregate.
28602 Translation is considered a kind of modification, so you may
28603 distribute translations of the Document under the terms of section
28604 4. Replacing Invariant Sections with translations requires special
28605 permission from their copyright holders, but you may include
28606 translations of some or all Invariant Sections in addition to the
28607 original versions of these Invariant Sections. You may include a
28608 translation of this License, and all the license notices in the
28609 Document, and any Warranty Disclaimers, provided that you also
28610 include the original English version of this License and the
28611 original versions of those notices and disclaimers. In case of a
28612 disagreement between the translation and the original version of
28613 this License or a notice or disclaimer, the original version will
28616 If a section in the Document is Entitled "Acknowledgements",
28617 "Dedications", or "History", the requirement (section 4) to
28618 Preserve its Title (section 1) will typically require changing the
28623 You may not copy, modify, sublicense, or distribute the Document
28624 except as expressly provided for under this License. Any other
28625 attempt to copy, modify, sublicense or distribute the Document is
28626 void, and will automatically terminate your rights under this
28627 License. However, parties who have received copies, or rights,
28628 from you under this License will not have their licenses
28629 terminated so long as such parties remain in full compliance.
28631 10. FUTURE REVISIONS OF THIS LICENSE
28633 The Free Software Foundation may publish new, revised versions of
28634 the GNU Free Documentation License from time to time. Such new
28635 versions will be similar in spirit to the present version, but may
28636 differ in detail to address new problems or concerns. See
28637 `http://www.gnu.org/copyleft/'.
28639 Each version of the License is given a distinguishing version
28640 number. If the Document specifies that a particular numbered
28641 version of this License "or any later version" applies to it, you
28642 have the option of following the terms and conditions either of
28643 that specified version or of any later version that has been
28644 published (not as a draft) by the Free Software Foundation. If
28645 the Document does not specify a version number of this License,
28646 you may choose any version ever published (not as a draft) by the
28647 Free Software Foundation.
28649 ADDENDUM: How to use this License for your documents
28650 ====================================================
28652 To use this License in a document you have written, include a copy of
28653 the License in the document and put the following copyright and license
28654 notices just after the title page:
28656 Copyright (C) YEAR YOUR NAME.
28657 Permission is granted to copy, distribute and/or modify this document
28658 under the terms of the GNU Free Documentation License, Version 1.2
28659 or any later version published by the Free Software Foundation;
28660 with no Invariant Sections, no Front-Cover Texts, and no Back-Cover
28661 Texts. A copy of the license is included in the section entitled ``GNU
28662 Free Documentation License''.
28664 If you have Invariant Sections, Front-Cover Texts and Back-Cover Texts,
28665 replace the "with...Texts." line with this:
28667 with the Invariant Sections being LIST THEIR TITLES, with
28668 the Front-Cover Texts being LIST, and with the Back-Cover Texts
28671 If you have Invariant Sections without Cover Texts, or some other
28672 combination of the three, merge those two alternatives to suit the
28675 If your document contains nontrivial examples of program code, we
28676 recommend releasing these examples in parallel under your choice of
28677 free software license, such as the GNU General Public License, to
28678 permit their use in free software.
28681 File: gcc.info, Node: Contributors, Next: Option Index, Prev: GNU Free Documentation License, Up: Top
28683 Contributors to GCC
28684 *******************
28686 The GCC project would like to thank its many contributors. Without
28687 them the project would not have been nearly as successful as it has
28688 been. Any omissions in this list are accidental. Feel free to contact
28689 <law@redhat.com> or <gerald@pfeifer.com> if you have been left out or
28690 some of your contributions are not listed. Please keep this list in
28691 alphabetical order.
28693 * Analog Devices helped implement the support for complex data types
28696 * John David Anglin for threading-related fixes and improvements to
28697 libstdc++-v3, and the HP-UX port.
28699 * James van Artsdalen wrote the code that makes efficient use of the
28700 Intel 80387 register stack.
28702 * Abramo and Roberto Bagnara for the SysV68 Motorola 3300 Delta
28705 * Alasdair Baird for various bug fixes.
28707 * Giovanni Bajo for analyzing lots of complicated C++ problem
28710 * Peter Barada for his work to improve code generation for new
28713 * Gerald Baumgartner added the signature extension to the C++ front
28716 * Godmar Back for his Java improvements and encouragement.
28718 * Scott Bambrough for help porting the Java compiler.
28720 * Wolfgang Bangerth for processing tons of bug reports.
28722 * Jon Beniston for his Microsoft Windows port of Java.
28724 * Daniel Berlin for better DWARF2 support, faster/better
28725 optimizations, improved alias analysis, plus migrating GCC to
28728 * Geoff Berry for his Java object serialization work and various
28731 * Uros Bizjak for the implementation of x87 math built-in functions
28732 and for various middle end and i386 back end improvements and
28735 * Eric Blake for helping to make GCJ and libgcj conform to the
28738 * Janne Blomqvist for contributions to GNU Fortran.
28740 * Segher Boessenkool for various fixes.
28742 * Hans-J. Boehm for his garbage collector, IA-64 libffi port, and
28745 * Neil Booth for work on cpplib, lang hooks, debug hooks and other
28746 miscellaneous clean-ups.
28748 * Steven Bosscher for integrating the GNU Fortran front end into GCC
28749 and for contributing to the tree-ssa branch.
28751 * Eric Botcazou for fixing middle- and backend bugs left and right.
28753 * Per Bothner for his direction via the steering committee and
28754 various improvements to the infrastructure for supporting new
28755 languages. Chill front end implementation. Initial
28756 implementations of cpplib, fix-header, config.guess, libio, and
28757 past C++ library (libg++) maintainer. Dreaming up, designing and
28758 implementing much of GCJ.
28760 * Devon Bowen helped port GCC to the Tahoe.
28762 * Don Bowman for mips-vxworks contributions.
28764 * Dave Brolley for work on cpplib and Chill.
28766 * Paul Brook for work on the ARM architecture and maintaining GNU
28769 * Robert Brown implemented the support for Encore 32000 systems.
28771 * Christian Bruel for improvements to local store elimination.
28773 * Herman A.J. ten Brugge for various fixes.
28775 * Joerg Brunsmann for Java compiler hacking and help with the GCJ
28778 * Joe Buck for his direction via the steering committee.
28780 * Craig Burley for leadership of the G77 Fortran effort.
28782 * Stephan Buys for contributing Doxygen notes for libstdc++.
28784 * Paolo Carlini for libstdc++ work: lots of efficiency improvements
28785 to the C++ strings, streambufs and formatted I/O, hard detective
28786 work on the frustrating localization issues, and keeping up with
28787 the problem reports.
28789 * John Carr for his alias work, SPARC hacking, infrastructure
28790 improvements, previous contributions to the steering committee,
28791 loop optimizations, etc.
28793 * Stephane Carrez for 68HC11 and 68HC12 ports.
28795 * Steve Chamberlain for support for the Renesas SH and H8 processors
28796 and the PicoJava processor, and for GCJ config fixes.
28798 * Glenn Chambers for help with the GCJ FAQ.
28800 * John-Marc Chandonia for various libgcj patches.
28802 * Scott Christley for his Objective-C contributions.
28804 * Eric Christopher for his Java porting help and clean-ups.
28806 * Branko Cibej for more warning contributions.
28808 * The GNU Classpath project for all of their merged runtime code.
28810 * Nick Clifton for arm, mcore, fr30, v850, m32r work, `--help', and
28811 other random hacking.
28813 * Michael Cook for libstdc++ cleanup patches to reduce warnings.
28815 * R. Kelley Cook for making GCC buildable from a read-only directory
28816 as well as other miscellaneous build process and documentation
28819 * Ralf Corsepius for SH testing and minor bugfixing.
28821 * Stan Cox for care and feeding of the x86 port and lots of behind
28822 the scenes hacking.
28824 * Alex Crain provided changes for the 3b1.
28826 * Ian Dall for major improvements to the NS32k port.
28828 * Paul Dale for his work to add uClinux platform support to the m68k
28831 * Dario Dariol contributed the four varieties of sample programs
28832 that print a copy of their source.
28834 * Russell Davidson for fstream and stringstream fixes in libstdc++.
28836 * Bud Davis for work on the G77 and GNU Fortran compilers.
28838 * Mo DeJong for GCJ and libgcj bug fixes.
28840 * DJ Delorie for the DJGPP port, build and libiberty maintenance,
28841 various bug fixes, and the M32C port.
28843 * Arnaud Desitter for helping to debug GNU Fortran.
28845 * Gabriel Dos Reis for contributions to G++, contributions and
28846 maintenance of GCC diagnostics infrastructure, libstdc++-v3,
28847 including `valarray<>', `complex<>', maintaining the numerics
28848 library (including that pesky `<limits>' :-) and keeping
28849 up-to-date anything to do with numbers.
28851 * Ulrich Drepper for his work on glibc, testing of GCC using glibc,
28852 ISO C99 support, CFG dumping support, etc., plus support of the
28853 C++ runtime libraries including for all kinds of C interface
28854 issues, contributing and maintaining `complex<>', sanity checking
28855 and disbursement, configuration architecture, libio maintenance,
28856 and early math work.
28858 * Zdenek Dvorak for a new loop unroller and various fixes.
28860 * Richard Earnshaw for his ongoing work with the ARM.
28862 * David Edelsohn for his direction via the steering committee,
28863 ongoing work with the RS6000/PowerPC port, help cleaning up Haifa
28864 loop changes, doing the entire AIX port of libstdc++ with his bare
28865 hands, and for ensuring GCC properly keeps working on AIX.
28867 * Kevin Ediger for the floating point formatting of num_put::do_put
28870 * Phil Edwards for libstdc++ work including configuration hackery,
28871 documentation maintainer, chief breaker of the web pages, the
28872 occasional iostream bug fix, and work on shared library symbol
28875 * Paul Eggert for random hacking all over GCC.
28877 * Mark Elbrecht for various DJGPP improvements, and for libstdc++
28878 configuration support for locales and fstream-related fixes.
28880 * Vadim Egorov for libstdc++ fixes in strings, streambufs, and
28883 * Christian Ehrhardt for dealing with bug reports.
28885 * Ben Elliston for his work to move the Objective-C runtime into its
28886 own subdirectory and for his work on autoconf.
28888 * Marc Espie for OpenBSD support.
28890 * Doug Evans for much of the global optimization framework, arc,
28891 m32r, and SPARC work.
28893 * Christopher Faylor for his work on the Cygwin port and for caring
28894 and feeding the gcc.gnu.org box and saving its users tons of spam.
28896 * Fred Fish for BeOS support and Ada fixes.
28898 * Ivan Fontes Garcia for the Portuguese translation of the GCJ FAQ.
28900 * Peter Gerwinski for various bug fixes and the Pascal front end.
28902 * Kaveh R. Ghazi for his direction via the steering committee,
28903 amazing work to make `-W -Wall -W* -Werror' useful, and
28904 continuously testing GCC on a plethora of platforms. Kaveh
28905 extends his gratitude to the CAIP Center at Rutgers University for
28906 providing him with computing resources to work on Free Software
28907 since the late 1980s.
28909 * John Gilmore for a donation to the FSF earmarked improving GNU
28912 * Judy Goldberg for c++ contributions.
28914 * Torbjorn Granlund for various fixes and the c-torture testsuite,
28915 multiply- and divide-by-constant optimization, improved long long
28916 support, improved leaf function register allocation, and his
28917 direction via the steering committee.
28919 * Anthony Green for his `-Os' contributions and Java front end work.
28921 * Stu Grossman for gdb hacking, allowing GCJ developers to debug
28924 * Michael K. Gschwind contributed the port to the PDP-11.
28926 * Ron Guilmette implemented the `protoize' and `unprotoize' tools,
28927 the support for Dwarf symbolic debugging information, and much of
28928 the support for System V Release 4. He has also worked heavily on
28929 the Intel 386 and 860 support.
28931 * Mostafa Hagog for Swing Modulo Scheduling (SMS) and post reload
28934 * Bruno Haible for improvements in the runtime overhead for EH, new
28935 warnings and assorted bug fixes.
28937 * Andrew Haley for his amazing Java compiler and library efforts.
28939 * Chris Hanson assisted in making GCC work on HP-UX for the 9000
28942 * Michael Hayes for various thankless work he's done trying to get
28943 the c30/c40 ports functional. Lots of loop and unroll
28944 improvements and fixes.
28946 * Dara Hazeghi for wading through myriads of target-specific bug
28949 * Kate Hedstrom for staking the G77 folks with an initial testsuite.
28951 * Richard Henderson for his ongoing SPARC, alpha, ia32, and ia64
28952 work, loop opts, and generally fixing lots of old problems we've
28953 ignored for years, flow rewrite and lots of further stuff,
28954 including reviewing tons of patches.
28956 * Aldy Hernandez for working on the PowerPC port, SIMD support, and
28959 * Nobuyuki Hikichi of Software Research Associates, Tokyo,
28960 contributed the support for the Sony NEWS machine.
28962 * Kazu Hirata for caring and feeding the Renesas H8/300 port and
28965 * Katherine Holcomb for work on GNU Fortran.
28967 * Manfred Hollstein for his ongoing work to keep the m88k alive, lots
28968 of testing and bug fixing, particularly of GCC configury code.
28970 * Steve Holmgren for MachTen patches.
28972 * Jan Hubicka for his x86 port improvements.
28974 * Falk Hueffner for working on C and optimization bug reports.
28976 * Bernardo Innocenti for his m68k work, including merging of
28977 ColdFire improvements and uClinux support.
28979 * Christian Iseli for various bug fixes.
28981 * Kamil Iskra for general m68k hacking.
28983 * Lee Iverson for random fixes and MIPS testing.
28985 * Andreas Jaeger for testing and benchmarking of GCC and various bug
28988 * Jakub Jelinek for his SPARC work and sibling call optimizations as
28989 well as lots of bug fixes and test cases, and for improving the
28992 * Janis Johnson for ia64 testing and fixes, her quality improvement
28993 sidetracks, and web page maintenance.
28995 * Kean Johnston for SCO OpenServer support and various fixes.
28997 * Tim Josling for the sample language treelang based originally on
28998 Richard Kenner's "toy" language.
29000 * Nicolai Josuttis for additional libstdc++ documentation.
29002 * Klaus Kaempf for his ongoing work to make alpha-vms a viable
29005 * Steven G. Kargl for work on GNU Fortran.
29007 * David Kashtan of SRI adapted GCC to VMS.
29009 * Ryszard Kabatek for many, many libstdc++ bug fixes and
29010 optimizations of strings, especially member functions, and for
29013 * Geoffrey Keating for his ongoing work to make the PPC work for
29014 GNU/Linux and his automatic regression tester.
29016 * Brendan Kehoe for his ongoing work with G++ and for a lot of early
29017 work in just about every part of libstdc++.
29019 * Oliver M. Kellogg of Deutsche Aerospace contributed the port to the
29022 * Richard Kenner of the New York University Ultracomputer Research
29023 Laboratory wrote the machine descriptions for the AMD 29000, the
29024 DEC Alpha, the IBM RT PC, and the IBM RS/6000 as well as the
29025 support for instruction attributes. He also made changes to
29026 better support RISC processors including changes to common
29027 subexpression elimination, strength reduction, function calling
29028 sequence handling, and condition code support, in addition to
29029 generalizing the code for frame pointer elimination and delay slot
29030 scheduling. Richard Kenner was also the head maintainer of GCC
29033 * Mumit Khan for various contributions to the Cygwin and Mingw32
29034 ports and maintaining binary releases for Microsoft Windows hosts,
29035 and for massive libstdc++ porting work to Cygwin/Mingw32.
29037 * Robin Kirkham for cpu32 support.
29039 * Mark Klein for PA improvements.
29041 * Thomas Koenig for various bug fixes.
29043 * Bruce Korb for the new and improved fixincludes code.
29045 * Benjamin Kosnik for his G++ work and for leading the libstdc++-v3
29048 * Charles LaBrec contributed the support for the Integrated Solutions
29051 * Asher Langton and Mike Kumbera for contributing Cray pointer
29052 support to GNU Fortran, and for other GNU Fortran improvements.
29054 * Jeff Law for his direction via the steering committee,
29055 coordinating the entire egcs project and GCC 2.95, rolling out
29056 snapshots and releases, handling merges from GCC2, reviewing tons
29057 of patches that might have fallen through the cracks else, and
29058 random but extensive hacking.
29060 * Marc Lehmann for his direction via the steering committee and
29061 helping with analysis and improvements of x86 performance.
29063 * Victor Leikehman for work on GNU Fortran.
29065 * Ted Lemon wrote parts of the RTL reader and printer.
29067 * Kriang Lerdsuwanakij for C++ improvements including template as
29068 template parameter support, and many C++ fixes.
29070 * Warren Levy for tremendous work on libgcj (Java Runtime Library)
29071 and random work on the Java front end.
29073 * Alain Lichnewsky ported GCC to the MIPS CPU.
29075 * Oskar Liljeblad for hacking on AWT and his many Java bug reports
29078 * Robert Lipe for OpenServer support, new testsuites, testing, etc.
29080 * Weiwen Liu for testing and various bug fixes.
29082 * Dave Love for his ongoing work with the Fortran front end and
29085 * Martin von Lo"wis for internal consistency checking infrastructure,
29086 various C++ improvements including namespace support, and tons of
29087 assistance with libstdc++/compiler merges.
29089 * H.J. Lu for his previous contributions to the steering committee,
29090 many x86 bug reports, prototype patches, and keeping the GNU/Linux
29093 * Greg McGary for random fixes and (someday) bounded pointers.
29095 * Andrew MacLeod for his ongoing work in building a real EH system,
29096 various code generation improvements, work on the global
29099 * Vladimir Makarov for hacking some ugly i960 problems, PowerPC
29100 hacking improvements to compile-time performance, overall
29101 knowledge and direction in the area of instruction scheduling, and
29102 design and implementation of the automaton based instruction
29105 * Bob Manson for his behind the scenes work on dejagnu.
29107 * Philip Martin for lots of libstdc++ string and vector iterator
29108 fixes and improvements, and string clean up and testsuites.
29110 * All of the Mauve project contributors, for Java test code.
29112 * Bryce McKinlay for numerous GCJ and libgcj fixes and improvements.
29114 * Adam Megacz for his work on the Microsoft Windows port of GCJ.
29116 * Michael Meissner for LRS framework, ia32, m32r, v850, m88k, MIPS,
29117 powerpc, haifa, ECOFF debug support, and other assorted hacking.
29119 * Jason Merrill for his direction via the steering committee and
29120 leading the G++ effort.
29122 * Martin Michlmayr for testing GCC on several architectures using the
29123 entire Debian archive.
29125 * David Miller for his direction via the steering committee, lots of
29126 SPARC work, improvements in jump.c and interfacing with the Linux
29129 * Gary Miller ported GCC to Charles River Data Systems machines.
29131 * Alfred Minarik for libstdc++ string and ios bug fixes, and turning
29132 the entire libstdc++ testsuite namespace-compatible.
29134 * Mark Mitchell for his direction via the steering committee,
29135 mountains of C++ work, load/store hoisting out of loops, alias
29136 analysis improvements, ISO C `restrict' support, and serving as
29137 release manager for GCC 3.x.
29139 * Alan Modra for various GNU/Linux bits and testing.
29141 * Toon Moene for his direction via the steering committee, Fortran
29142 maintenance, and his ongoing work to make us make Fortran run fast.
29144 * Jason Molenda for major help in the care and feeding of all the
29145 services on the gcc.gnu.org (formerly egcs.cygnus.com)
29146 machine--mail, web services, ftp services, etc etc. Doing all
29147 this work on scrap paper and the backs of envelopes would have
29150 * Catherine Moore for fixing various ugly problems we have sent her
29151 way, including the haifa bug which was killing the Alpha & PowerPC
29154 * Mike Moreton for his various Java patches.
29156 * David Mosberger-Tang for various Alpha improvements, and for the
29157 initial IA-64 port.
29159 * Stephen Moshier contributed the floating point emulator that
29160 assists in cross-compilation and permits support for floating
29161 point numbers wider than 64 bits and for ISO C99 support.
29163 * Bill Moyer for his behind the scenes work on various issues.
29165 * Philippe De Muyter for his work on the m68k port.
29167 * Joseph S. Myers for his work on the PDP-11 port, format checking
29168 and ISO C99 support, and continuous emphasis on (and contributions
29171 * Nathan Myers for his work on libstdc++-v3: architecture and
29172 authorship through the first three snapshots, including
29173 implementation of locale infrastructure, string, shadow C headers,
29174 and the initial project documentation (DESIGN, CHECKLIST, and so
29175 forth). Later, more work on MT-safe string and shadow headers.
29177 * Felix Natter for documentation on porting libstdc++.
29179 * Nathanael Nerode for cleaning up the configuration/build process.
29181 * NeXT, Inc. donated the front end that supports the Objective-C
29184 * Hans-Peter Nilsson for the CRIS and MMIX ports, improvements to
29185 the search engine setup, various documentation fixes and other
29188 * Geoff Noer for his work on getting cygwin native builds working.
29190 * Diego Novillo for his work on Tree SSA, OpenMP, SPEC performance
29191 tracking web pages and assorted fixes.
29193 * David O'Brien for the FreeBSD/alpha, FreeBSD/AMD x86-64,
29194 FreeBSD/ARM, FreeBSD/PowerPC, and FreeBSD/SPARC64 ports and
29195 related infrastructure improvements.
29197 * Alexandre Oliva for various build infrastructure improvements,
29198 scripts and amazing testing work, including keeping libtool issues
29201 * Stefan Olsson for work on mt_alloc.
29203 * Melissa O'Neill for various NeXT fixes.
29205 * Rainer Orth for random MIPS work, including improvements to GCC's
29206 o32 ABI support, improvements to dejagnu's MIPS support, Java
29207 configuration clean-ups and porting work, etc.
29209 * Hartmut Penner for work on the s390 port.
29211 * Paul Petersen wrote the machine description for the Alliant FX/8.
29213 * Alexandre Petit-Bianco for implementing much of the Java compiler
29214 and continued Java maintainership.
29216 * Matthias Pfaller for major improvements to the NS32k port.
29218 * Gerald Pfeifer for his direction via the steering committee,
29219 pointing out lots of problems we need to solve, maintenance of the
29220 web pages, and taking care of documentation maintenance in general.
29222 * Andrew Pinski for processing bug reports by the dozen.
29224 * Ovidiu Predescu for his work on the Objective-C front end and
29227 * Jerry Quinn for major performance improvements in C++ formatted
29230 * Ken Raeburn for various improvements to checker, MIPS ports and
29231 various cleanups in the compiler.
29233 * Rolf W. Rasmussen for hacking on AWT.
29235 * David Reese of Sun Microsystems contributed to the Solaris on
29238 * Volker Reichelt for keeping up with the problem reports.
29240 * Joern Rennecke for maintaining the sh port, loop, regmove & reload
29243 * Loren J. Rittle for improvements to libstdc++-v3 including the
29244 FreeBSD port, threading fixes, thread-related configury changes,
29245 critical threading documentation, and solutions to really tricky
29246 I/O problems, as well as keeping GCC properly working on FreeBSD
29247 and continuous testing.
29249 * Craig Rodrigues for processing tons of bug reports.
29251 * Ola Ro"nnerup for work on mt_alloc.
29253 * Gavin Romig-Koch for lots of behind the scenes MIPS work.
29255 * David Ronis inspired and encouraged Craig to rewrite the G77
29256 documentation in texinfo format by contributing a first pass at a
29257 translation of the old `g77-0.5.16/f/DOC' file.
29259 * Ken Rose for fixes to GCC's delay slot filling code.
29261 * Paul Rubin wrote most of the preprocessor.
29263 * Pe'tur Runo'lfsson for major performance improvements in C++
29264 formatted I/O and large file support in C++ filebuf.
29266 * Chip Salzenberg for libstdc++ patches and improvements to locales,
29267 traits, Makefiles, libio, libtool hackery, and "long long" support.
29269 * Juha Sarlin for improvements to the H8 code generator.
29271 * Greg Satz assisted in making GCC work on HP-UX for the 9000 series
29274 * Roger Sayle for improvements to constant folding and GCC's RTL
29275 optimizers as well as for fixing numerous bugs.
29277 * Bradley Schatz for his work on the GCJ FAQ.
29279 * Peter Schauer wrote the code to allow debugging to work on the
29282 * William Schelter did most of the work on the Intel 80386 support.
29284 * Tobias Schlu"ter for work on GNU Fortran.
29286 * Bernd Schmidt for various code generation improvements and major
29287 work in the reload pass as well a serving as release manager for
29290 * Peter Schmid for constant testing of libstdc++--especially
29291 application testing, going above and beyond what was requested for
29292 the release criteria--and libstdc++ header file tweaks.
29294 * Jason Schroeder for jcf-dump patches.
29296 * Andreas Schwab for his work on the m68k port.
29298 * Lars Segerlund for work on GNU Fortran.
29300 * Joel Sherrill for his direction via the steering committee, RTEMS
29301 contributions and RTEMS testing.
29303 * Nathan Sidwell for many C++ fixes/improvements.
29305 * Jeffrey Siegal for helping RMS with the original design of GCC,
29306 some code which handles the parse tree and RTL data structures,
29307 constant folding and help with the original VAX & m68k ports.
29309 * Kenny Simpson for prompting libstdc++ fixes due to defect reports
29310 from the LWG (thereby keeping GCC in line with updates from the
29313 * Franz Sirl for his ongoing work with making the PPC port stable
29316 * Andrey Slepuhin for assorted AIX hacking.
29318 * Christopher Smith did the port for Convex machines.
29320 * Danny Smith for his major efforts on the Mingw (and Cygwin) ports.
29322 * Randy Smith finished the Sun FPA support.
29324 * Scott Snyder for queue, iterator, istream, and string fixes and
29325 libstdc++ testsuite entries. Also for providing the patch to G77
29326 to add rudimentary support for `INTEGER*1', `INTEGER*2', and
29329 * Brad Spencer for contributions to the GLIBCPP_FORCE_NEW technique.
29331 * Richard Stallman, for writing the original GCC and launching the
29334 * Jan Stein of the Chalmers Computer Society provided support for
29335 Genix, as well as part of the 32000 machine description.
29337 * Nigel Stephens for various mips16 related fixes/improvements.
29339 * Jonathan Stone wrote the machine description for the Pyramid
29342 * Graham Stott for various infrastructure improvements.
29344 * John Stracke for his Java HTTP protocol fixes.
29346 * Mike Stump for his Elxsi port, G++ contributions over the years
29347 and more recently his vxworks contributions
29349 * Jeff Sturm for Java porting help, bug fixes, and encouragement.
29351 * Shigeya Suzuki for this fixes for the bsdi platforms.
29353 * Ian Lance Taylor for his mips16 work, general configury hacking,
29356 * Holger Teutsch provided the support for the Clipper CPU.
29358 * Gary Thomas for his ongoing work to make the PPC work for
29361 * Philipp Thomas for random bug fixes throughout the compiler
29363 * Jason Thorpe for thread support in libstdc++ on NetBSD.
29365 * Kresten Krab Thorup wrote the run time support for the Objective-C
29366 language and the fantastic Java bytecode interpreter.
29368 * Michael Tiemann for random bug fixes, the first instruction
29369 scheduler, initial C++ support, function integration, NS32k, SPARC
29370 and M88k machine description work, delay slot scheduling.
29372 * Andreas Tobler for his work porting libgcj to Darwin.
29374 * Teemu Torma for thread safe exception handling support.
29376 * Leonard Tower wrote parts of the parser, RTL generator, and RTL
29377 definitions, and of the VAX machine description.
29379 * Tom Tromey for internationalization support and for his many Java
29380 contributions and libgcj maintainership.
29382 * Lassi Tuura for improvements to config.guess to determine HP
29385 * Petter Urkedal for libstdc++ CXXFLAGS, math, and algorithms fixes.
29387 * Andy Vaught for the design and initial implementation of the GNU
29390 * Brent Verner for work with the libstdc++ cshadow files and their
29391 associated configure steps.
29393 * Todd Vierling for contributions for NetBSD ports.
29395 * Jonathan Wakely for contributing libstdc++ Doxygen notes and XHTML
29398 * Dean Wakerley for converting the install documentation from HTML
29399 to texinfo in time for GCC 3.0.
29401 * Krister Walfridsson for random bug fixes.
29403 * Feng Wang for contributions to GNU Fortran.
29405 * Stephen M. Webb for time and effort on making libstdc++ shadow
29406 files work with the tricky Solaris 8+ headers, and for pushing the
29407 build-time header tree.
29409 * John Wehle for various improvements for the x86 code generator,
29410 related infrastructure improvements to help x86 code generation,
29411 value range propagation and other work, WE32k port.
29413 * Ulrich Weigand for work on the s390 port.
29415 * Zack Weinberg for major work on cpplib and various other bug fixes.
29417 * Matt Welsh for help with Linux Threads support in GCJ.
29419 * Urban Widmark for help fixing java.io.
29421 * Mark Wielaard for new Java library code and his work integrating
29424 * Dale Wiles helped port GCC to the Tahoe.
29426 * Bob Wilson from Tensilica, Inc. for the Xtensa port.
29428 * Jim Wilson for his direction via the steering committee, tackling
29429 hard problems in various places that nobody else wanted to work
29430 on, strength reduction and other loop optimizations.
29432 * Paul Woegerer and Tal Agmon for the CRX port.
29434 * Carlo Wood for various fixes.
29436 * Tom Wood for work on the m88k port.
29438 * Canqun Yang for work on GNU Fortran.
29440 * Masanobu Yuhara of Fujitsu Laboratories implemented the machine
29441 description for the Tron architecture (specifically, the Gmicro).
29443 * Kevin Zachmann helped port GCC to the Tahoe.
29445 * Ayal Zaks for Swing Modulo Scheduling (SMS).
29447 * Xiaoqiang Zhang for work on GNU Fortran.
29449 * Gilles Zunino for help porting Java to Irix.
29452 The following people are recognized for their contributions to GNAT,
29453 the Ada front end of GCC:
29456 * Romain Berrendonner
29506 * Hristian Kirtchev
29549 The following people are recognized for their contributions of new
29550 features, bug reports, testing and integration of classpath/libgcj for
29552 * Lillian Angel for `JTree' implementation and lots Free Swing
29553 additions and bugfixes.
29555 * Wolfgang Baer for `GapContent' bugfixes.
29557 * Anthony Balkissoon for `JList', Free Swing 1.5 updates and mouse
29558 event fixes, lots of Free Swing work including `JTable' editing.
29560 * Stuart Ballard for RMI constant fixes.
29562 * Goffredo Baroncelli for `HTTPURLConnection' fixes.
29564 * Gary Benson for `MessageFormat' fixes.
29566 * Daniel Bonniot for `Serialization' fixes.
29568 * Chris Burdess for lots of gnu.xml and http protocol fixes, `StAX'
29569 and `DOM xml:id' support.
29571 * Ka-Hing Cheung for `TreePath' and `TreeSelection' fixes.
29573 * Archie Cobbs for build fixes, VM interface updates,
29574 `URLClassLoader' updates.
29576 * Kelley Cook for build fixes.
29578 * Martin Cordova for Suggestions for better `SocketTimeoutException'.
29580 * David Daney for `BitSet' bugfixes, `HttpURLConnection' rewrite and
29583 * Thomas Fitzsimmons for lots of upgrades to the gtk+ AWT and Cairo
29584 2D support. Lots of imageio framework additions, lots of AWT and
29585 Free Swing bugfixes.
29587 * Jeroen Frijters for `ClassLoader' and nio cleanups, serialization
29588 fixes, better `Proxy' support, bugfixes and IKVM integration.
29590 * Santiago Gala for `AccessControlContext' fixes.
29592 * Nicolas Geoffray for `VMClassLoader' and `AccessController'
29595 * David Gilbert for `basic' and `metal' icon and plaf support and
29596 lots of documenting, Lots of Free Swing and metal theme additions.
29597 `MetalIconFactory' implementation.
29599 * Anthony Green for `MIDI' framework, `ALSA' and `DSSI' providers.
29601 * Andrew Haley for `Serialization' and `URLClassLoader' fixes, gcj
29604 * Kim Ho for `JFileChooser' implementation.
29606 * Andrew John Hughes for `Locale' and net fixes, URI RFC2986
29607 updates, `Serialization' fixes, `Properties' XML support and
29608 generic branch work, VMIntegration guide update.
29610 * Bastiaan Huisman for `TimeZone' bugfixing.
29612 * Andreas Jaeger for mprec updates.
29614 * Paul Jenner for better `-Werror' support.
29616 * Ito Kazumitsu for `NetworkInterface' implementation and updates.
29618 * Roman Kennke for `BoxLayout', `GrayFilter' and `SplitPane', plus
29619 bugfixes all over. Lots of Free Swing work including styled text.
29621 * Simon Kitching for `String' cleanups and optimization suggestions.
29623 * Michael Koch for configuration fixes, `Locale' updates, bug and
29626 * Guilhem Lavaux for configuration, thread and channel fixes and
29627 Kaffe integration. JCL native `Pointer' updates. Logger bugfixes.
29629 * David Lichteblau for JCL support library global/local reference
29632 * Aaron Luchko for JDWP updates and documentation fixes.
29634 * Ziga Mahkovec for `Graphics2D' upgraded to Cairo 0.5 and new regex
29637 * Sven de Marothy for BMP imageio support, CSS and `TextLayout'
29638 fixes. `GtkImage' rewrite, 2D, awt, free swing and date/time fixes
29639 and implementing the Qt4 peers.
29641 * Casey Marshall for crypto algorithm fixes, `FileChannel' lock,
29642 `SystemLogger' and `FileHandler' rotate implementations, NIO
29643 `FileChannel.map' support, security and policy updates.
29645 * Bryce McKinlay for RMI work.
29647 * Audrius Meskauskas for lots of Free Corba, RMI and HTML work plus
29648 testing and documenting.
29650 * Kalle Olavi Niemitalo for build fixes.
29652 * Rainer Orth for build fixes.
29654 * Andrew Overholt for `File' locking fixes.
29656 * Ingo Proetel for `Image', `Logger' and `URLClassLoader' updates.
29658 * Olga Rodimina for `MenuSelectionManager' implementation.
29660 * Jan Roehrich for `BasicTreeUI' and `JTree' fixes.
29662 * Julian Scheid for documentation updates and gjdoc support.
29664 * Christian Schlichtherle for zip fixes and cleanups.
29666 * Robert Schuster for documentation updates and beans fixes,
29667 `TreeNode' enumerations and `ActionCommand' and various fixes, XML
29668 and URL, AWT and Free Swing bugfixes.
29670 * Keith Seitz for lots of JDWP work.
29672 * Christian Thalinger for 64-bit cleanups, Configuration and VM
29673 interface fixes and `CACAO' integration, `fdlibm' updates.
29675 * Gael Thomas for `VMClassLoader' boot packages support suggestions.
29677 * Andreas Tobler for Darwin and Solaris testing and fixing, `Qt4'
29678 support for Darwin/OS X, `Graphics2D' support, `gtk+' updates.
29680 * Dalibor Topic for better `DEBUG' support, build cleanups and Kaffe
29681 integration. `Qt4' build infrastructure, `SHA1PRNG' and
29682 `GdkPixbugDecoder' updates.
29684 * Tom Tromey for Eclipse integration, generics work, lots of bugfixes
29685 and gcj integration including coordinating The Big Merge.
29687 * Mark Wielaard for bugfixes, packaging and release management,
29688 `Clipboard' implementation, system call interrupts and network
29689 timeouts and `GdkPixpufDecoder' fixes.
29692 In addition to the above, all of which also contributed time and
29693 energy in testing GCC, we would like to thank the following for their
29694 contributions to testing:
29696 * Michael Abd-El-Malek
29706 * David Billinghurst
29710 * Stephane Bortzmeyer
29720 * Bradford Castalia
29740 * Charles-Antoine Gauthier
29762 * Kevin B. Hendricks
29766 * Christian Joensson
29774 * Anand Krishnaswamy
29776 * A. O. V. Le Blanc
29840 * Pedro A. M. Vazquez
29850 And finally we'd like to thank everyone who uses the compiler, submits
29851 bug reports and generally reminds us why we're doing this work in the
29855 File: gcc.info, Node: Option Index, Next: Keyword Index, Prev: Contributors, Up: Top
29860 GCC's command line options are indexed here without any initial `-' or
29861 `--'. Where an option has both positive and negative forms (such as
29862 `-fOPTION' and `-fno-OPTION'), relevant entries in the manual are
29863 indexed under the most appropriate form; it may sometimes be useful to
29864 look up both forms.
29869 * ###: Overall Options. (line 192)
29870 * A: Preprocessor Options.
29872 * all_load: Darwin Options. (line 103)
29873 * allowable_client: Darwin Options. (line 190)
29874 * ansi <1>: Non-bugs. (line 107)
29875 * ansi <2>: Other Builtins. (line 22)
29876 * ansi <3>: Preprocessor Options.
29878 * ansi <4>: C Dialect Options. (line 11)
29879 * ansi: Standards. (line 13)
29880 * arch_errors_fatal: Darwin Options. (line 107)
29881 * aux-info: C Dialect Options. (line 119)
29882 * b: Target Options. (line 13)
29883 * B: Directory Options. (line 41)
29884 * bcopy-builtin: PDP-11 Options. (line 32)
29885 * bind_at_load: Darwin Options. (line 111)
29886 * bundle: Darwin Options. (line 116)
29887 * bundle_loader: Darwin Options. (line 120)
29888 * c: Link Options. (line 20)
29889 * C: Preprocessor Options.
29891 * c: Overall Options. (line 147)
29892 * client_name: Darwin Options. (line 190)
29893 * combine: Overall Options. (line 203)
29894 * compatibility_version: Darwin Options. (line 190)
29895 * coverage: Debugging Options. (line 239)
29896 * crossjumping: Optimize Options. (line 435)
29897 * current_version: Darwin Options. (line 190)
29898 * D: Preprocessor Options.
29900 * d: Debugging Options. (line 291)
29901 * da: Debugging Options. (line 457)
29902 * dA: Debugging Options. (line 304)
29903 * dB: Debugging Options. (line 309)
29904 * dC: Debugging Options. (line 319)
29905 * dc: Debugging Options. (line 313)
29906 * dD <1>: Preprocessor Options.
29908 * dD: Debugging Options. (line 333)
29909 * dd: Debugging Options. (line 327)
29910 * dE: Debugging Options. (line 338)
29911 * dead_strip: Darwin Options. (line 190)
29912 * dependency-file: Darwin Options. (line 190)
29913 * df: Debugging Options. (line 343)
29914 * dG: Debugging Options. (line 355)
29915 * dg: Debugging Options. (line 350)
29916 * dH: Debugging Options. (line 460)
29917 * dh: Debugging Options. (line 362)
29918 * dI: Preprocessor Options.
29920 * di: Debugging Options. (line 366)
29921 * dj: Debugging Options. (line 370)
29922 * dk: Debugging Options. (line 374)
29923 * dL: Debugging Options. (line 383)
29924 * dl: Debugging Options. (line 379)
29925 * dM: Preprocessor Options.
29927 * dm: Debugging Options. (line 463)
29928 * dM: Debugging Options. (line 394)
29929 * dm: Debugging Options. (line 390)
29930 * dN <1>: Preprocessor Options.
29932 * dN: Debugging Options. (line 403)
29933 * dn: Debugging Options. (line 399)
29934 * do: Debugging Options. (line 407)
29935 * dP: Debugging Options. (line 472)
29936 * dp: Debugging Options. (line 467)
29937 * dR: Debugging Options. (line 415)
29938 * dr: Debugging Options. (line 411)
29939 * dS: Debugging Options. (line 424)
29940 * ds: Debugging Options. (line 419)
29941 * dT: Debugging Options. (line 433)
29942 * dt: Debugging Options. (line 428)
29943 * dumpmachine: Debugging Options. (line 840)
29944 * dumpspecs: Debugging Options. (line 848)
29945 * dumpversion: Debugging Options. (line 844)
29946 * dv: Debugging Options. (line 476)
29947 * dV: Debugging Options. (line 438)
29948 * dw: Debugging Options. (line 445)
29949 * dx: Debugging Options. (line 481)
29950 * dy: Debugging Options. (line 485)
29951 * dylib_file: Darwin Options. (line 190)
29952 * dylinker_install_name: Darwin Options. (line 190)
29953 * dynamic: Darwin Options. (line 190)
29954 * dynamiclib: Darwin Options. (line 124)
29955 * dZ: Debugging Options. (line 453)
29956 * dz: Debugging Options. (line 449)
29957 * E <1>: Link Options. (line 20)
29958 * E: Overall Options. (line 168)
29959 * EB <1>: MIPS Options. (line 7)
29960 * EB: ARC Options. (line 12)
29961 * EL <1>: MIPS Options. (line 10)
29962 * EL: ARC Options. (line 9)
29963 * exported_symbols_list: Darwin Options. (line 190)
29964 * F: Darwin Options. (line 32)
29965 * fabi-version: C++ Dialect Options.
29967 * falign-functions: Optimize Options. (line 902)
29968 * falign-jumps: Optimize Options. (line 952)
29969 * falign-labels: Optimize Options. (line 920)
29970 * falign-loops: Optimize Options. (line 938)
29971 * fargument-alias: Code Gen Options. (line 364)
29972 * fargument-noalias: Code Gen Options. (line 364)
29973 * fargument-noalias-anything: Code Gen Options. (line 364)
29974 * fargument-noalias-global: Code Gen Options. (line 364)
29975 * fasynchronous-unwind-tables: Code Gen Options. (line 64)
29976 * fbounds-check <1>: Code Gen Options. (line 15)
29977 * fbounds-check: Optimize Options. (line 326)
29978 * fbranch-probabilities: Optimize Options. (line 1200)
29979 * fbranch-target-load-optimize: Optimize Options. (line 1308)
29980 * fbranch-target-load-optimize2: Optimize Options. (line 1314)
29981 * fbtr-bb-exclusive: Optimize Options. (line 1318)
29982 * fcall-saved <1>: Interoperation. (line 150)
29983 * fcall-saved: Code Gen Options. (line 237)
29984 * fcall-used: Code Gen Options. (line 223)
29985 * fcaller-saves: Optimize Options. (line 579)
29986 * fcheck-new: C++ Dialect Options.
29988 * fcommon: Variable Attributes.
29990 * fcond-mismatch: C Dialect Options. (line 235)
29991 * fconserve-space: C++ Dialect Options.
29993 * fconstant-string-class: Objective-C and Objective-C++ Dialect Options.
29995 * fcse-follow-jumps: Optimize Options. (line 363)
29996 * fcse-skip-blocks: Optimize Options. (line 372)
29997 * fcx-limited-range: Optimize Options. (line 1186)
29998 * fdata-sections: Optimize Options. (line 1289)
29999 * fdelayed-branch: Optimize Options. (line 488)
30000 * fdelete-null-pointer-checks: Optimize Options. (line 457)
30001 * fdiagnostics-show-location: Language Independent Options.
30003 * fdiagnostics-show-option: Language Independent Options.
30005 * fdirectives-only: Preprocessor Options.
30007 * fdollars-in-identifiers <1>: Interoperation. (line 146)
30008 * fdollars-in-identifiers: Preprocessor Options.
30010 * fdump-class-hierarchy: Debugging Options. (line 511)
30011 * fdump-ipa: Debugging Options. (line 518)
30012 * fdump-noaddr: Debugging Options. (line 488)
30013 * fdump-rtl-all: Debugging Options. (line 457)
30014 * fdump-rtl-bbro: Debugging Options. (line 309)
30015 * fdump-rtl-btl: Debugging Options. (line 327)
30016 * fdump-rtl-bypass: Debugging Options. (line 355)
30017 * fdump-rtl-ce1: Debugging Options. (line 319)
30018 * fdump-rtl-ce2: Debugging Options. (line 319)
30019 * fdump-rtl-ce3: Debugging Options. (line 338)
30020 * fdump-rtl-cfg: Debugging Options. (line 343)
30021 * fdump-rtl-combine: Debugging Options. (line 313)
30022 * fdump-rtl-cse: Debugging Options. (line 419)
30023 * fdump-rtl-cse2: Debugging Options. (line 428)
30024 * fdump-rtl-dbr: Debugging Options. (line 327)
30025 * fdump-rtl-eh: Debugging Options. (line 362)
30026 * fdump-rtl-expand: Debugging Options. (line 411)
30027 * fdump-rtl-flow2: Debugging Options. (line 445)
30028 * fdump-rtl-gcse: Debugging Options. (line 355)
30029 * fdump-rtl-greg: Debugging Options. (line 350)
30030 * fdump-rtl-jump: Debugging Options. (line 370)
30031 * fdump-rtl-life: Debugging Options. (line 343)
30032 * fdump-rtl-loop2: Debugging Options. (line 383)
30033 * fdump-rtl-lreg: Debugging Options. (line 379)
30034 * fdump-rtl-mach: Debugging Options. (line 394)
30035 * fdump-rtl-peephole2: Debugging Options. (line 449)
30036 * fdump-rtl-postreload: Debugging Options. (line 407)
30037 * fdump-rtl-regmove: Debugging Options. (line 403)
30038 * fdump-rtl-rnreg: Debugging Options. (line 399)
30039 * fdump-rtl-sched: Debugging Options. (line 424)
30040 * fdump-rtl-sched2: Debugging Options. (line 415)
30041 * fdump-rtl-sibling: Debugging Options. (line 366)
30042 * fdump-rtl-sms: Debugging Options. (line 390)
30043 * fdump-rtl-stack: Debugging Options. (line 374)
30044 * fdump-rtl-tracer: Debugging Options. (line 433)
30045 * fdump-rtl-vartrack: Debugging Options. (line 438)
30046 * fdump-rtl-vpt: Debugging Options. (line 438)
30047 * fdump-rtl-web: Debugging Options. (line 453)
30048 * fdump-translation-unit: Debugging Options. (line 503)
30049 * fdump-tree: Debugging Options. (line 533)
30050 * fdump-tree-alias: Debugging Options. (line 621)
30051 * fdump-tree-all: Debugging Options. (line 706)
30052 * fdump-tree-ccp: Debugging Options. (line 625)
30053 * fdump-tree-cfg: Debugging Options. (line 596)
30054 * fdump-tree-ch: Debugging Options. (line 608)
30055 * fdump-tree-copyprop: Debugging Options. (line 641)
30056 * fdump-tree-copyrename: Debugging Options. (line 687)
30057 * fdump-tree-dce: Debugging Options. (line 649)
30058 * fdump-tree-dom: Debugging Options. (line 667)
30059 * fdump-tree-dse: Debugging Options. (line 672)
30060 * fdump-tree-forwprop: Debugging Options. (line 682)
30061 * fdump-tree-fre: Debugging Options. (line 637)
30062 * fdump-tree-gimple: Debugging Options. (line 591)
30063 * fdump-tree-mudflap: Debugging Options. (line 653)
30064 * fdump-tree-nrv: Debugging Options. (line 692)
30065 * fdump-tree-phiopt: Debugging Options. (line 677)
30066 * fdump-tree-pre: Debugging Options. (line 633)
30067 * fdump-tree-salias: Debugging Options. (line 616)
30068 * fdump-tree-sink: Debugging Options. (line 663)
30069 * fdump-tree-sra: Debugging Options. (line 658)
30070 * fdump-tree-ssa: Debugging Options. (line 612)
30071 * fdump-tree-store_copyprop: Debugging Options. (line 645)
30072 * fdump-tree-storeccp: Debugging Options. (line 629)
30073 * fdump-tree-vcg: Debugging Options. (line 600)
30074 * fdump-tree-vect: Debugging Options. (line 697)
30075 * fdump-tree-vrp: Debugging Options. (line 702)
30076 * fdump-unnumbered: Debugging Options. (line 495)
30077 * fearly-inlining: Optimize Options. (line 204)
30078 * feliminate-dwarf2-dups: Debugging Options. (line 125)
30079 * feliminate-unused-debug-symbols: Debugging Options. (line 52)
30080 * feliminate-unused-debug-types: Debugging Options. (line 852)
30081 * fexceptions: Code Gen Options. (line 34)
30082 * fexec-charset: Preprocessor Options.
30084 * fexpensive-optimizations: Optimize Options. (line 470)
30085 * fextended-identifiers: Preprocessor Options.
30087 * ffast-math: Optimize Options. (line 1070)
30088 * ffinite-math-only: Optimize Options. (line 1114)
30089 * ffix-and-continue: Darwin Options. (line 97)
30090 * ffixed: Code Gen Options. (line 211)
30091 * ffloat-store <1>: Disappointments. (line 77)
30092 * ffloat-store: Optimize Options. (line 1056)
30093 * ffor-scope: C++ Dialect Options.
30095 * fforce-addr: Optimize Options. (line 154)
30096 * fforce-mem: Optimize Options. (line 146)
30097 * ffreestanding <1>: Function Attributes.
30099 * ffreestanding <2>: Warning Options. (line 94)
30100 * ffreestanding <3>: C Dialect Options. (line 190)
30101 * ffreestanding: Standards. (line 81)
30102 * ffriend-injection: C++ Dialect Options.
30104 * ffunction-sections: Optimize Options. (line 1289)
30105 * fgcse: Optimize Options. (line 386)
30106 * fgcse-after-reload: Optimize Options. (line 422)
30107 * fgcse-las: Optimize Options. (line 415)
30108 * fgcse-lm: Optimize Options. (line 397)
30109 * fgcse-sm: Optimize Options. (line 406)
30110 * fgnu-runtime: Objective-C and Objective-C++ Dialect Options.
30112 * fgnu89-inline: C Dialect Options. (line 98)
30113 * fhosted: C Dialect Options. (line 183)
30114 * filelist: Darwin Options. (line 190)
30115 * findirect-data: Darwin Options. (line 97)
30116 * finhibit-size-directive: Code Gen Options. (line 147)
30117 * finline-functions: Optimize Options. (line 185)
30118 * finline-functions-called-once: Optimize Options. (line 196)
30119 * finline-limit: Optimize Options. (line 214)
30120 * finput-charset: Preprocessor Options.
30122 * finstrument-functions <1>: Function Attributes.
30124 * finstrument-functions: Code Gen Options. (line 267)
30125 * finstrument-functions-exclude-file-list: Code Gen Options. (line 304)
30126 * finstrument-functions-exclude-function-list: Code Gen Options.
30128 * fkeep-inline-functions <1>: Inline. (line 58)
30129 * fkeep-inline-functions: Optimize Options. (line 252)
30130 * fkeep-static-consts: Optimize Options. (line 259)
30131 * flat_namespace: Darwin Options. (line 190)
30132 * fleading-underscore: Code Gen Options. (line 381)
30133 * fmem-report: Debugging Options. (line 220)
30134 * fmessage-length: Language Independent Options.
30136 * fmodulo-sched: Optimize Options. (line 288)
30137 * fmove-loop-invariants: Optimize Options. (line 1279)
30138 * fms-extensions <1>: Unnamed Fields. (line 37)
30139 * fms-extensions <2>: C++ Dialect Options.
30141 * fms-extensions: C Dialect Options. (line 206)
30142 * fmudflap: Optimize Options. (line 333)
30143 * fmudflapir: Optimize Options. (line 333)
30144 * fmudflapth: Optimize Options. (line 333)
30145 * fnext-runtime: Objective-C and Objective-C++ Dialect Options.
30147 * fno-access-control: C++ Dialect Options.
30149 * fno-asm: C Dialect Options. (line 135)
30150 * fno-branch-count-reg: Optimize Options. (line 293)
30151 * fno-builtin <1>: Other Builtins. (line 14)
30152 * fno-builtin <2>: Function Attributes.
30154 * fno-builtin <3>: Warning Options. (line 94)
30155 * fno-builtin: C Dialect Options. (line 149)
30156 * fno-common <1>: Variable Attributes.
30158 * fno-common: Code Gen Options. (line 135)
30159 * fno-cprop-registers: Optimize Options. (line 1028)
30160 * fno-cx-limited-range: Optimize Options. (line 1186)
30161 * fno-default-inline <1>: Inline. (line 53)
30162 * fno-default-inline <2>: Optimize Options. (line 131)
30163 * fno-default-inline: C++ Dialect Options.
30165 * fno-defer-pop: Optimize Options. (line 138)
30166 * fno-elide-constructors: C++ Dialect Options.
30168 * fno-enforce-eh-specs: C++ Dialect Options.
30170 * fno-for-scope: C++ Dialect Options.
30172 * fno-function-cse: Optimize Options. (line 303)
30173 * fno-gnu-keywords: C++ Dialect Options.
30175 * fno-guess-branch-probability: Optimize Options. (line 787)
30176 * fno-ident: Code Gen Options. (line 144)
30177 * fno-implement-inlines <1>: C++ Interface. (line 75)
30178 * fno-implement-inlines: C++ Dialect Options.
30180 * fno-implicit-inline-templates: C++ Dialect Options.
30182 * fno-implicit-templates <1>: Template Instantiation.
30184 * fno-implicit-templates: C++ Dialect Options.
30186 * fno-inline: Optimize Options. (line 179)
30187 * fno-jump-tables: Code Gen Options. (line 203)
30188 * fno-math-errno: Optimize Options. (line 1083)
30189 * fno-nil-receivers: Objective-C and Objective-C++ Dialect Options.
30191 * fno-nonansi-builtins: C++ Dialect Options.
30193 * fno-operator-names: C++ Dialect Options.
30195 * fno-optional-diags: C++ Dialect Options.
30197 * fno-peephole: Optimize Options. (line 778)
30198 * fno-peephole2: Optimize Options. (line 778)
30199 * fno-rtti: C++ Dialect Options.
30201 * fno-sched-interblock: Optimize Options. (line 514)
30202 * fno-sched-spec: Optimize Options. (line 519)
30203 * fno-show-column: Preprocessor Options.
30205 * fno-signed-bitfields: C Dialect Options. (line 268)
30206 * fno-stack-limit: Code Gen Options. (line 347)
30207 * fno-threadsafe-statics: C++ Dialect Options.
30209 * fno-trapping-math: Optimize Options. (line 1124)
30210 * fno-unsigned-bitfields: C Dialect Options. (line 268)
30211 * fno-use-cxa-get-exception-ptr: C++ Dialect Options.
30213 * fno-weak: C++ Dialect Options.
30215 * fno-working-directory: Preprocessor Options.
30217 * fno-zero-initialized-in-bss: Optimize Options. (line 314)
30218 * fnon-call-exceptions: Code Gen Options. (line 48)
30219 * fobjc-call-cxx-cdtors: Objective-C and Objective-C++ Dialect Options.
30221 * fobjc-direct-dispatch: Objective-C and Objective-C++ Dialect Options.
30223 * fobjc-exceptions: Objective-C and Objective-C++ Dialect Options.
30225 * fobjc-gc: Objective-C and Objective-C++ Dialect Options.
30227 * fomit-frame-pointer: Optimize Options. (line 158)
30228 * fopenmp: C Dialect Options. (line 200)
30229 * foptimize-register-move: Optimize Options. (line 477)
30230 * foptimize-sibling-calls: Optimize Options. (line 174)
30231 * force_cpusubtype_ALL: Darwin Options. (line 129)
30232 * force_flat_namespace: Darwin Options. (line 190)
30233 * fpack-struct: Code Gen Options. (line 254)
30234 * fpcc-struct-return <1>: Incompatibilities. (line 170)
30235 * fpcc-struct-return: Code Gen Options. (line 70)
30236 * fpch-deps: Preprocessor Options.
30238 * fpch-preprocess: Preprocessor Options.
30240 * fpeel-loops: Optimize Options. (line 1271)
30241 * fpermissive: C++ Dialect Options.
30243 * fPIC: Code Gen Options. (line 184)
30244 * fpic: Code Gen Options. (line 163)
30245 * fPIE: Code Gen Options. (line 197)
30246 * fpie: Code Gen Options. (line 197)
30247 * fprefetch-loop-arrays: Optimize Options. (line 767)
30248 * fpreprocessed: Preprocessor Options.
30250 * fprofile-arcs <1>: Other Builtins. (line 236)
30251 * fprofile-arcs: Debugging Options. (line 224)
30252 * fprofile-generate: Optimize Options. (line 1035)
30253 * fprofile-use: Optimize Options. (line 1044)
30254 * fprofile-values: Optimize Options. (line 1219)
30255 * frandom-string: Debugging Options. (line 735)
30256 * freg-struct-return: Code Gen Options. (line 88)
30257 * fregmove: Optimize Options. (line 477)
30258 * frename-registers: Optimize Options. (line 1238)
30259 * freorder-blocks: Optimize Options. (line 804)
30260 * freorder-blocks-and-partition: Optimize Options. (line 810)
30261 * freorder-functions: Optimize Options. (line 821)
30262 * freplace-objc-classes: Objective-C and Objective-C++ Dialect Options.
30264 * frepo <1>: Template Instantiation.
30266 * frepo: C++ Dialect Options.
30268 * frerun-cse-after-loop: Optimize Options. (line 380)
30269 * frounding-math: Optimize Options. (line 1139)
30270 * frtl-abstract-sequences: Optimize Options. (line 1159)
30271 * fsched-spec-load: Optimize Options. (line 524)
30272 * fsched-spec-load-dangerous: Optimize Options. (line 529)
30273 * fsched-stalled-insns: Optimize Options. (line 534)
30274 * fsched-stalled-insns-dep: Optimize Options. (line 539)
30275 * fsched-verbose: Debugging Options. (line 745)
30276 * fsched2-use-superblocks: Optimize Options. (line 546)
30277 * fsched2-use-traces: Optimize Options. (line 557)
30278 * fschedule-insns: Optimize Options. (line 495)
30279 * fschedule-insns2: Optimize Options. (line 505)
30280 * fscheduling-in-modulo-scheduled-loops: Optimize Options. (line 573)
30281 * fsection-anchors: Optimize Options. (line 1334)
30282 * fsee: Optimize Options. (line 569)
30283 * fshort-double: Code Gen Options. (line 117)
30284 * fshort-enums <1>: Non-bugs. (line 42)
30285 * fshort-enums <2>: Type Attributes. (line 112)
30286 * fshort-enums <3>: Structures unions enumerations and bit-fields implementation.
30288 * fshort-enums: Code Gen Options. (line 106)
30289 * fshort-wchar: Code Gen Options. (line 125)
30290 * fsignaling-nans: Optimize Options. (line 1166)
30291 * fsigned-bitfields <1>: Non-bugs. (line 57)
30292 * fsigned-bitfields: C Dialect Options. (line 268)
30293 * fsigned-char <1>: Characters implementation.
30295 * fsigned-char: C Dialect Options. (line 258)
30296 * fsingle-precision-constant: Optimize Options. (line 1181)
30297 * fsplit-ivs-in-unroller: Optimize Options. (line 748)
30298 * fstack-check: Code Gen Options. (line 332)
30299 * fstack-limit-register: Code Gen Options. (line 347)
30300 * fstack-limit-symbol: Code Gen Options. (line 347)
30301 * fstats: C++ Dialect Options.
30303 * fstrict-aliasing: Optimize Options. (line 834)
30304 * fstrict-overflow: Optimize Options. (line 876)
30305 * fsyntax-only: Warning Options. (line 23)
30306 * ftabstop: Preprocessor Options.
30308 * ftemplate-depth: C++ Dialect Options.
30310 * ftest-coverage: Debugging Options. (line 280)
30311 * fthread-jumps: Optimize Options. (line 354)
30312 * ftime-report: Debugging Options. (line 216)
30313 * ftracer: Optimize Options. (line 731)
30314 * ftrapv: Code Gen Options. (line 22)
30315 * ftree-vect-loop-version: Optimize Options. (line 713)
30316 * ftree-vectorizer-verbose: Debugging Options. (line 710)
30317 * funit-at-a-time: Optimize Options. (line 965)
30318 * funroll-all-loops: Optimize Options. (line 742)
30319 * funroll-loops: Optimize Options. (line 736)
30320 * funsafe-loop-optimizations: Optimize Options. (line 427)
30321 * funsafe-math-optimizations: Optimize Options. (line 1100)
30322 * funsigned-bitfields <1>: Non-bugs. (line 57)
30323 * funsigned-bitfields <2>: Structures unions enumerations and bit-fields implementation.
30325 * funsigned-bitfields: C Dialect Options. (line 268)
30326 * funsigned-char <1>: Characters implementation.
30328 * funsigned-char: C Dialect Options. (line 240)
30329 * funswitch-loops: Optimize Options. (line 1283)
30330 * funwind-tables: Code Gen Options. (line 57)
30331 * fuse-cxa-atexit: C++ Dialect Options.
30333 * fvar-tracking: Debugging Options. (line 788)
30334 * fvariable-expansion-in-unroller: Optimize Options. (line 762)
30335 * fverbose-asm: Code Gen Options. (line 154)
30336 * fvisibility: Code Gen Options. (line 400)
30337 * fvisibility-inlines-hidden: C++ Dialect Options.
30339 * fvpt: Optimize Options. (line 1229)
30340 * fweb: Optimize Options. (line 1004)
30341 * fwhole-program: Optimize Options. (line 1015)
30342 * fwide-exec-charset: Preprocessor Options.
30344 * fworking-directory: Preprocessor Options.
30346 * fwrapv: Code Gen Options. (line 26)
30347 * fzero-link: Objective-C and Objective-C++ Dialect Options.
30349 * G <1>: System V Options. (line 10)
30350 * G <2>: RS/6000 and PowerPC Options.
30352 * G <3>: MIPS Options. (line 216)
30353 * G: M32R/D Options. (line 57)
30354 * g: Debugging Options. (line 10)
30355 * gcoff: Debugging Options. (line 70)
30356 * gdwarf-2: Debugging Options. (line 88)
30357 * gen-decls: Objective-C and Objective-C++ Dialect Options.
30359 * gfull: Darwin Options. (line 64)
30360 * ggdb: Debugging Options. (line 38)
30361 * gnu-ld: HPPA Options. (line 113)
30362 * gstabs: Debugging Options. (line 44)
30363 * gstabs+: Debugging Options. (line 64)
30364 * gused: Darwin Options. (line 59)
30365 * gvms: Debugging Options. (line 95)
30366 * gxcoff: Debugging Options. (line 75)
30367 * gxcoff+: Debugging Options. (line 80)
30368 * H: Preprocessor Options.
30370 * headerpad_max_install_names: Darwin Options. (line 190)
30371 * help <1>: Preprocessor Options.
30373 * help: Overall Options. (line 219)
30374 * hp-ld: HPPA Options. (line 125)
30375 * I <1>: Directory Options. (line 10)
30376 * I: Preprocessor Options.
30378 * I- <1>: Directory Options. (line 107)
30379 * I-: Preprocessor Options.
30381 * idirafter: Preprocessor Options.
30383 * if-conversion: Optimize Options. (line 442)
30384 * if-conversion2: Optimize Options. (line 451)
30385 * imacros: Preprocessor Options.
30387 * image_base: Darwin Options. (line 190)
30388 * imultilib: Preprocessor Options.
30390 * include: Preprocessor Options.
30392 * init: Darwin Options. (line 190)
30393 * install_name: Darwin Options. (line 190)
30394 * iprefix: Preprocessor Options.
30396 * iquote <1>: Directory Options. (line 31)
30397 * iquote: Preprocessor Options.
30399 * isysroot: Preprocessor Options.
30401 * isystem: Preprocessor Options.
30403 * iwithprefix: Preprocessor Options.
30405 * iwithprefixbefore: Preprocessor Options.
30407 * keep_private_externs: Darwin Options. (line 190)
30408 * L: Directory Options. (line 37)
30409 * l: Link Options. (line 26)
30410 * lobjc: Link Options. (line 53)
30411 * M: Preprocessor Options.
30413 * m1: SH Options. (line 9)
30414 * m10: PDP-11 Options. (line 29)
30415 * m128bit-long-double: i386 and x86-64 Options.
30417 * m16-bit: CRIS Options. (line 69)
30418 * m2: SH Options. (line 12)
30419 * m210: MCore Options. (line 43)
30420 * m3: SH Options. (line 18)
30421 * m31: S/390 and zSeries Options.
30423 * m32 <1>: SPARC Options. (line 189)
30424 * m32 <2>: RS/6000 and PowerPC Options.
30426 * m32: i386 and x86-64 Options.
30428 * m32-bit: CRIS Options. (line 69)
30429 * m32r: M32R/D Options. (line 15)
30430 * m32r2: M32R/D Options. (line 9)
30431 * m32rx: M32R/D Options. (line 12)
30432 * m340: MCore Options. (line 43)
30433 * m386: i386 and x86-64 Options.
30435 * m3dnow: i386 and x86-64 Options.
30437 * m3e: SH Options. (line 21)
30438 * m4: SH Options. (line 35)
30439 * m4-nofpu: SH Options. (line 24)
30440 * m4-single: SH Options. (line 31)
30441 * m4-single-only: SH Options. (line 27)
30442 * m40: PDP-11 Options. (line 23)
30443 * m45: PDP-11 Options. (line 26)
30444 * m486: i386 and x86-64 Options.
30446 * m4a: SH Options. (line 50)
30447 * m4a-nofpu: SH Options. (line 38)
30448 * m4a-single: SH Options. (line 46)
30449 * m4a-single-only: SH Options. (line 42)
30450 * m4al: SH Options. (line 53)
30451 * m4byte-functions: MCore Options. (line 27)
30452 * m5200: M680x0 Options. (line 59)
30453 * m64 <1>: SPARC Options. (line 189)
30454 * m64 <2>: S/390 and zSeries Options.
30456 * m64 <3>: RS/6000 and PowerPC Options.
30458 * m64: i386 and x86-64 Options.
30460 * m68000: M680x0 Options. (line 13)
30461 * m68020: M680x0 Options. (line 21)
30462 * m68020-40: M680x0 Options. (line 70)
30463 * m68020-60: M680x0 Options. (line 77)
30464 * m68030: M680x0 Options. (line 30)
30465 * m68040: M680x0 Options. (line 34)
30466 * m68060: M680x0 Options. (line 42)
30467 * m6811: M68hc1x Options. (line 13)
30468 * m6812: M68hc1x Options. (line 18)
30469 * m68881: M680x0 Options. (line 25)
30470 * m68hc11: M68hc1x Options. (line 13)
30471 * m68hc12: M68hc1x Options. (line 18)
30472 * m68hcs12: M68hc1x Options. (line 23)
30473 * m68S12: M68hc1x Options. (line 23)
30474 * m8-bit: CRIS Options. (line 69)
30475 * m96bit-long-double: i386 and x86-64 Options.
30477 * mabi <1>: RS/6000 and PowerPC Options.
30479 * mabi: ARM Options. (line 10)
30480 * mabi-mmixware: MMIX Options. (line 20)
30481 * mabi=32: MIPS Options. (line 89)
30482 * mabi=64: MIPS Options. (line 89)
30483 * mabi=eabi: MIPS Options. (line 89)
30484 * mabi=gnu: MMIX Options. (line 20)
30485 * mabi=ibmlongdouble: RS/6000 and PowerPC Options.
30487 * mabi=ieeelongdouble: RS/6000 and PowerPC Options.
30489 * mabi=n32: MIPS Options. (line 89)
30490 * mabi=no-spe: RS/6000 and PowerPC Options.
30492 * mabi=o64: MIPS Options. (line 89)
30493 * mabi=spe: RS/6000 and PowerPC Options.
30495 * mabicalls: MIPS Options. (line 100)
30496 * mabort-on-noreturn: ARM Options. (line 144)
30497 * mabshi: PDP-11 Options. (line 55)
30498 * mac0: PDP-11 Options. (line 16)
30499 * macc-4: FRV Options. (line 113)
30500 * macc-8: FRV Options. (line 116)
30501 * maccumulate-outgoing-args: i386 and x86-64 Options.
30503 * madjust-unroll: SH Options. (line 175)
30504 * mads: RS/6000 and PowerPC Options.
30506 * maix-struct-return: RS/6000 and PowerPC Options.
30508 * maix32: RS/6000 and PowerPC Options.
30510 * maix64: RS/6000 and PowerPC Options.
30512 * malign-300: H8/300 Options. (line 31)
30513 * malign-double: i386 and x86-64 Options.
30515 * malign-int: M680x0 Options. (line 132)
30516 * malign-labels: FRV Options. (line 104)
30517 * malign-loops: M32R/D Options. (line 73)
30518 * malign-natural: RS/6000 and PowerPC Options.
30520 * malign-power: RS/6000 and PowerPC Options.
30522 * malloc-cc: FRV Options. (line 25)
30523 * malpha-as: DEC Alpha Options. (line 159)
30524 * maltivec: RS/6000 and PowerPC Options.
30526 * mam33: MN10300 Options. (line 17)
30527 * mandroid: ARM Options. (line 254)
30528 * maout: CRIS Options. (line 92)
30529 * mapcs: ARM Options. (line 22)
30530 * mapcs-frame: ARM Options. (line 14)
30531 * mapp-regs <1>: V850 Options. (line 57)
30532 * mapp-regs: SPARC Options. (line 10)
30533 * march <1>: S/390 and zSeries Options.
30535 * march <2>: MT Options. (line 9)
30536 * march <3>: MIPS Options. (line 14)
30537 * march <4>: i386 and x86-64 Options.
30539 * march <5>: HPPA Options. (line 9)
30540 * march <6>: CRIS Options. (line 10)
30541 * march: ARM Options. (line 109)
30542 * masm=DIALECT: i386 and x86-64 Options.
30544 * mauto-incdec: M68hc1x Options. (line 26)
30545 * mauto-pic: IA-64 Options. (line 50)
30546 * mb: SH Options. (line 58)
30547 * mbacc: MT Options. (line 16)
30548 * mbackchain: S/390 and zSeries Options.
30550 * mbase-addresses: MMIX Options. (line 54)
30551 * mbcopy: PDP-11 Options. (line 36)
30552 * mbig <1>: TMS320C3x/C4x Options.
30554 * mbig: RS/6000 and PowerPC Options.
30556 * mbig-endian <1>: RS/6000 and PowerPC Options.
30558 * mbig-endian <2>: MCore Options. (line 39)
30559 * mbig-endian <3>: IA-64 Options. (line 9)
30560 * mbig-endian: ARM Options. (line 72)
30561 * mbig-memory: TMS320C3x/C4x Options.
30563 * mbig-switch <1>: V850 Options. (line 52)
30564 * mbig-switch: HPPA Options. (line 23)
30565 * mbigtable: SH Options. (line 74)
30566 * mbit-align: RS/6000 and PowerPC Options.
30568 * mbitfield: M680x0 Options. (line 104)
30569 * mbk: TMS320C3x/C4x Options.
30571 * mbranch-cheap: PDP-11 Options. (line 65)
30572 * mbranch-cost=NUMBER: M32R/D Options. (line 82)
30573 * mbranch-expensive: PDP-11 Options. (line 61)
30574 * mbranch-likely: MIPS Options. (line 367)
30575 * mbranch-predict: MMIX Options. (line 49)
30576 * mbss-plt: RS/6000 and PowerPC Options.
30578 * mbuild-constants: DEC Alpha Options. (line 142)
30579 * mbwx: DEC Alpha Options. (line 171)
30580 * mc68000: M680x0 Options. (line 13)
30581 * mc68020: M680x0 Options. (line 21)
30582 * mcall-gnu: RS/6000 and PowerPC Options.
30584 * mcall-linux: RS/6000 and PowerPC Options.
30586 * mcall-netbsd: RS/6000 and PowerPC Options.
30588 * mcall-prologues: AVR Options. (line 43)
30589 * mcall-solaris: RS/6000 and PowerPC Options.
30591 * mcall-sysv: RS/6000 and PowerPC Options.
30593 * mcall-sysv-eabi: RS/6000 and PowerPC Options.
30595 * mcall-sysv-noeabi: RS/6000 and PowerPC Options.
30597 * mcallee-super-interworking: ARM Options. (line 234)
30598 * mcaller-super-interworking: ARM Options. (line 240)
30599 * mcallgraph-data: MCore Options. (line 31)
30600 * mcc-init: CRIS Options. (line 46)
30601 * mcfv4e: M680x0 Options. (line 66)
30602 * mcheck-zero-division: MIPS Options. (line 254)
30603 * mcirrus-fix-invalid-insns: ARM Options. (line 187)
30604 * mcix: DEC Alpha Options. (line 171)
30605 * mcmodel=embmedany: SPARC Options. (line 211)
30606 * mcmodel=kernel: i386 and x86-64 Options.
30608 * mcmodel=large: i386 and x86-64 Options.
30610 * mcmodel=medany: SPARC Options. (line 205)
30611 * mcmodel=medium: i386 and x86-64 Options.
30613 * mcmodel=medlow: SPARC Options. (line 194)
30614 * mcmodel=medmid: SPARC Options. (line 199)
30615 * mcmodel=small: i386 and x86-64 Options.
30617 * mcond-exec: FRV Options. (line 152)
30618 * mcond-move: FRV Options. (line 128)
30619 * mconst-align: CRIS Options. (line 60)
30620 * mconst16: Xtensa Options. (line 10)
30621 * mconstant-gp: IA-64 Options. (line 46)
30622 * mcpu <1>: TMS320C3x/C4x Options.
30624 * mcpu <2>: SPARC Options. (line 96)
30625 * mcpu <3>: RS/6000 and PowerPC Options.
30627 * mcpu <4>: i386 and x86-64 Options.
30629 * mcpu <5>: FRV Options. (line 212)
30630 * mcpu <6>: DEC Alpha Options. (line 223)
30631 * mcpu <7>: CRIS Options. (line 10)
30632 * mcpu <8>: ARM Options. (line 84)
30633 * mcpu: ARC Options. (line 23)
30634 * mcpu32: M680x0 Options. (line 51)
30635 * mcpu=: M32C Options. (line 7)
30636 * mcsync-anomaly: Blackfin Options. (line 23)
30637 * MD: Preprocessor Options.
30639 * mdalign: SH Options. (line 64)
30640 * mdata: ARC Options. (line 30)
30641 * mdata-align: CRIS Options. (line 60)
30642 * mdb: TMS320C3x/C4x Options.
30644 * mdebug <1>: S/390 and zSeries Options.
30646 * mdebug: M32R/D Options. (line 69)
30647 * mdec-asm: PDP-11 Options. (line 78)
30648 * mdisable-callt: V850 Options. (line 80)
30649 * mdisable-fpregs: HPPA Options. (line 33)
30650 * mdisable-indexing: HPPA Options. (line 40)
30651 * mdiv: MCore Options. (line 15)
30652 * mdiv=STRATEGY: SH Options. (line 127)
30653 * mdivide-breaks: MIPS Options. (line 259)
30654 * mdivide-traps: MIPS Options. (line 259)
30655 * mdivsi3_libfunc=NAME: SH Options. (line 168)
30656 * mdlmzb: RS/6000 and PowerPC Options.
30658 * mdouble: FRV Options. (line 38)
30659 * mdouble-float: MIPS Options. (line 173)
30660 * mdp-isr-reload: TMS320C3x/C4x Options.
30662 * mdsp: MIPS Options. (line 178)
30663 * mdwarf2-asm: IA-64 Options. (line 79)
30664 * mdword: FRV Options. (line 32)
30665 * mdynamic-no-pic: RS/6000 and PowerPC Options.
30667 * meabi: RS/6000 and PowerPC Options.
30669 * mearly-stop-bits: IA-64 Options. (line 85)
30670 * meb: Score Options. (line 9)
30671 * mel: Score Options. (line 12)
30672 * melf <1>: MMIX Options. (line 44)
30673 * melf: CRIS Options. (line 95)
30674 * melinux: CRIS Options. (line 99)
30675 * melinux-stacksize: CRIS Options. (line 25)
30676 * memb: RS/6000 and PowerPC Options.
30678 * membedded-data: MIPS Options. (line 225)
30679 * memregs=: M32C Options. (line 21)
30680 * mep: V850 Options. (line 16)
30681 * mepsilon: MMIX Options. (line 15)
30682 * mesa: S/390 and zSeries Options.
30684 * metrax100: CRIS Options. (line 31)
30685 * metrax4: CRIS Options. (line 31)
30686 * mexplicit-relocs <1>: MIPS Options. (line 245)
30687 * mexplicit-relocs: DEC Alpha Options. (line 184)
30688 * MF: Preprocessor Options.
30690 * mfast-fix: TMS320C3x/C4x Options.
30692 * mfast-indirect-calls: HPPA Options. (line 52)
30693 * mfaster-structs: SPARC Options. (line 71)
30694 * mfdpic: FRV Options. (line 56)
30695 * mfix: DEC Alpha Options. (line 171)
30696 * mfix-and-continue: Darwin Options. (line 97)
30697 * mfix-r4000: MIPS Options. (line 309)
30698 * mfix-r4400: MIPS Options. (line 323)
30699 * mfix-sb1: MIPS Options. (line 351)
30700 * mfix-vr4120: MIPS Options. (line 330)
30701 * mfix-vr4130: MIPS Options. (line 344)
30702 * mfixed-cc: FRV Options. (line 28)
30703 * mfixed-range <1>: IA-64 Options. (line 90)
30704 * mfixed-range: HPPA Options. (line 59)
30705 * mfloat-abi: ARM Options. (line 59)
30706 * mfloat-gprs: RS/6000 and PowerPC Options.
30708 * mfloat-ieee: DEC Alpha Options. (line 179)
30709 * mfloat-vax: DEC Alpha Options. (line 179)
30710 * mfloat32: PDP-11 Options. (line 52)
30711 * mfloat64: PDP-11 Options. (line 48)
30712 * mflush-func: MIPS Options. (line 357)
30713 * mflush-func=NAME: M32R/D Options. (line 94)
30714 * mflush-trap=NUMBER: M32R/D Options. (line 87)
30715 * mfmovd: SH Options. (line 78)
30716 * mfp: ARM Options. (line 119)
30717 * mfp-exceptions: MIPS Options. (line 378)
30718 * mfp-reg: DEC Alpha Options. (line 25)
30719 * mfp-rounding-mode: DEC Alpha Options. (line 85)
30720 * mfp-trap-mode: DEC Alpha Options. (line 63)
30721 * mfp32: MIPS Options. (line 156)
30722 * mfp64: MIPS Options. (line 159)
30723 * mfpe: ARM Options. (line 119)
30724 * mfpr-32: FRV Options. (line 13)
30725 * mfpr-64: FRV Options. (line 16)
30726 * mfprnd: RS/6000 and PowerPC Options.
30728 * mfpu <1>: SPARC Options. (line 20)
30729 * mfpu <2>: PDP-11 Options. (line 9)
30730 * mfpu: ARM Options. (line 119)
30731 * mfull-toc: RS/6000 and PowerPC Options.
30733 * mfused-madd <1>: Xtensa Options. (line 19)
30734 * mfused-madd <2>: S/390 and zSeries Options.
30736 * mfused-madd <3>: RS/6000 and PowerPC Options.
30738 * mfused-madd: MIPS Options. (line 294)
30739 * mg: VAX Options. (line 17)
30740 * MG: Preprocessor Options.
30742 * mgas <1>: HPPA Options. (line 75)
30743 * mgas: DEC Alpha Options. (line 159)
30744 * mgettrcost=NUMBER: SH Options. (line 190)
30745 * mglibc: GNU/Linux Options. (line 9)
30746 * mgnu: VAX Options. (line 13)
30747 * mgnu-as: IA-64 Options. (line 18)
30748 * mgnu-ld: IA-64 Options. (line 23)
30749 * mgotplt: CRIS Options. (line 86)
30750 * mgp32: MIPS Options. (line 150)
30751 * mgp64: MIPS Options. (line 153)
30752 * mgpr-32: FRV Options. (line 7)
30753 * mgpr-64: FRV Options. (line 10)
30754 * mgprel-ro: FRV Options. (line 79)
30755 * mh: H8/300 Options. (line 14)
30756 * mhard-float <1>: SPARC Options. (line 20)
30757 * mhard-float <2>: S/390 and zSeries Options.
30759 * mhard-float <3>: RS/6000 and PowerPC Options.
30761 * mhard-float <4>: MIPS Options. (line 162)
30762 * mhard-float <5>: FRV Options. (line 19)
30763 * mhard-float: ARM Options. (line 41)
30764 * mhard-quad-float: SPARC Options. (line 41)
30765 * mhardlit: MCore Options. (line 10)
30766 * mhitachi: SH Options. (line 81)
30767 * mid-shared-library: Blackfin Options. (line 39)
30768 * mieee <1>: SH Options. (line 96)
30769 * mieee: DEC Alpha Options. (line 39)
30770 * mieee-conformant: DEC Alpha Options. (line 134)
30771 * mieee-fp: i386 and x86-64 Options.
30773 * mieee-with-inexact: DEC Alpha Options. (line 52)
30774 * milp32: IA-64 Options. (line 114)
30775 * mimpure-text: SPARC Options. (line 81)
30776 * mindexed-addressing: SH Options. (line 180)
30777 * minit-stack: AVR Options. (line 35)
30778 * minline-all-stringops: i386 and x86-64 Options.
30780 * minline-float-divide-max-throughput: IA-64 Options. (line 58)
30781 * minline-float-divide-min-latency: IA-64 Options. (line 54)
30782 * minline-int-divide-max-throughput: IA-64 Options. (line 66)
30783 * minline-int-divide-min-latency: IA-64 Options. (line 62)
30784 * minline-plt: FRV Options. (line 64)
30785 * minline-sqrt-max-throughput: IA-64 Options. (line 74)
30786 * minline-sqrt-min-latency: IA-64 Options. (line 70)
30787 * minmax: M68hc1x Options. (line 31)
30788 * minsert-sched-nops: RS/6000 and PowerPC Options.
30790 * mint16: PDP-11 Options. (line 40)
30791 * mint32 <1>: PDP-11 Options. (line 44)
30792 * mint32: H8/300 Options. (line 28)
30793 * mint8: AVR Options. (line 53)
30794 * minvalid-symbols: SH Options. (line 213)
30795 * mips1: MIPS Options. (line 59)
30796 * mips16: MIPS Options. (line 81)
30797 * mips2: MIPS Options. (line 62)
30798 * mips3: MIPS Options. (line 65)
30799 * mips32: MIPS Options. (line 71)
30800 * mips32r2: MIPS Options. (line 74)
30801 * mips3d: MIPS Options. (line 190)
30802 * mips4: MIPS Options. (line 68)
30803 * mips64: MIPS Options. (line 77)
30804 * misel: RS/6000 and PowerPC Options.
30806 * misize: SH Options. (line 103)
30807 * missue-rate=NUMBER: M32R/D Options. (line 79)
30808 * mjump-in-delay: HPPA Options. (line 28)
30809 * mkernel: Darwin Options. (line 75)
30810 * mknuthdiv: MMIX Options. (line 33)
30811 * ml: SH Options. (line 61)
30812 * mlarge-data: DEC Alpha Options. (line 195)
30813 * mlarge-data-threshold=NUMBER: i386 and x86-64 Options.
30815 * mlarge-text: DEC Alpha Options. (line 213)
30816 * mlibfuncs: MMIX Options. (line 10)
30817 * mlibrary-pic: FRV Options. (line 110)
30818 * mlinked-fp: FRV Options. (line 94)
30819 * mlinker-opt: HPPA Options. (line 85)
30820 * mlinux: CRIS Options. (line 104)
30821 * mlittle: RS/6000 and PowerPC Options.
30823 * mlittle-endian <1>: SPARC Options. (line 183)
30824 * mlittle-endian <2>: RS/6000 and PowerPC Options.
30826 * mlittle-endian <3>: MCore Options. (line 39)
30827 * mlittle-endian <4>: IA-64 Options. (line 13)
30828 * mlittle-endian: ARM Options. (line 68)
30829 * mlong-calls <1>: V850 Options. (line 10)
30830 * mlong-calls <2>: MIPS Options. (line 280)
30831 * mlong-calls <3>: M68hc1x Options. (line 35)
30832 * mlong-calls <4>: FRV Options. (line 99)
30833 * mlong-calls <5>: Blackfin Options. (line 57)
30834 * mlong-calls: ARM Options. (line 149)
30835 * mlong-double-128: S/390 and zSeries Options.
30837 * mlong-double-64: S/390 and zSeries Options.
30839 * mlong-load-store: HPPA Options. (line 66)
30840 * mlong32: MIPS Options. (line 199)
30841 * mlong64: MIPS Options. (line 194)
30842 * mlongcall: RS/6000 and PowerPC Options.
30844 * mlongcalls: Xtensa Options. (line 60)
30845 * mloop-unsigned: TMS320C3x/C4x Options.
30847 * mlow-64k: Blackfin Options. (line 32)
30848 * mlp64: IA-64 Options. (line 114)
30849 * MM: Preprocessor Options.
30851 * mmac <1>: Score Options. (line 21)
30852 * mmac: CRX Options. (line 9)
30853 * mmad: MIPS Options. (line 289)
30854 * mmangle-cpu: ARC Options. (line 15)
30855 * mmax: DEC Alpha Options. (line 171)
30856 * mmax-stack-frame: CRIS Options. (line 22)
30857 * mmcu: AVR Options. (line 9)
30858 * MMD: Preprocessor Options.
30860 * mmedia: FRV Options. (line 44)
30861 * mmemcpy: MIPS Options. (line 274)
30862 * mmemory-latency: DEC Alpha Options. (line 266)
30863 * mmemparm: TMS320C3x/C4x Options.
30865 * mmfcrf: RS/6000 and PowerPC Options.
30867 * mminimal-toc: RS/6000 and PowerPC Options.
30869 * mmmx: i386 and x86-64 Options.
30871 * mmodel=large: M32R/D Options. (line 33)
30872 * mmodel=medium: M32R/D Options. (line 27)
30873 * mmodel=small: M32R/D Options. (line 18)
30874 * mmpyi: TMS320C3x/C4x Options.
30876 * mmul-bug-workaround: CRIS Options. (line 36)
30877 * mmuladd: FRV Options. (line 50)
30878 * mmulhw: RS/6000 and PowerPC Options.
30880 * mmult-bug: MN10300 Options. (line 9)
30881 * mmulti-cond-exec: FRV Options. (line 176)
30882 * mmultiple: RS/6000 and PowerPC Options.
30884 * mmvcle: S/390 and zSeries Options.
30886 * mmvme: RS/6000 and PowerPC Options.
30888 * mn: H8/300 Options. (line 20)
30889 * mnested-cond-exec: FRV Options. (line 189)
30890 * mnew-mnemonics: RS/6000 and PowerPC Options.
30892 * mnhwloop: Score Options. (line 15)
30893 * mno-3dnow: i386 and x86-64 Options.
30895 * mno-4byte-functions: MCore Options. (line 27)
30896 * mno-abicalls: MIPS Options. (line 100)
30897 * mno-abshi: PDP-11 Options. (line 58)
30898 * mno-ac0: PDP-11 Options. (line 20)
30899 * mno-align-double: i386 and x86-64 Options.
30901 * mno-align-int: M680x0 Options. (line 132)
30902 * mno-align-loops: M32R/D Options. (line 76)
30903 * mno-align-stringops: i386 and x86-64 Options.
30905 * mno-altivec: RS/6000 and PowerPC Options.
30907 * mno-am33: MN10300 Options. (line 20)
30908 * mno-app-regs <1>: V850 Options. (line 61)
30909 * mno-app-regs: SPARC Options. (line 10)
30910 * mno-bacc: MT Options. (line 19)
30911 * mno-backchain: S/390 and zSeries Options.
30913 * mno-base-addresses: MMIX Options. (line 54)
30914 * mno-bit-align: RS/6000 and PowerPC Options.
30916 * mno-bk: TMS320C3x/C4x Options.
30918 * mno-branch-likely: MIPS Options. (line 367)
30919 * mno-branch-predict: MMIX Options. (line 49)
30920 * mno-bwx: DEC Alpha Options. (line 171)
30921 * mno-callgraph-data: MCore Options. (line 31)
30922 * mno-check-zero-division: MIPS Options. (line 254)
30923 * mno-cirrus-fix-invalid-insns: ARM Options. (line 187)
30924 * mno-cix: DEC Alpha Options. (line 171)
30925 * mno-cond-exec: FRV Options. (line 158)
30926 * mno-cond-move: FRV Options. (line 134)
30927 * mno-const-align: CRIS Options. (line 60)
30928 * mno-const16: Xtensa Options. (line 10)
30929 * mno-crt0 <1>: MT Options. (line 25)
30930 * mno-crt0: MN10300 Options. (line 31)
30931 * mno-csync-anomaly: Blackfin Options. (line 28)
30932 * mno-data-align: CRIS Options. (line 60)
30933 * mno-db: TMS320C3x/C4x Options.
30935 * mno-debug: S/390 and zSeries Options.
30937 * mno-div: MCore Options. (line 15)
30938 * mno-dlmzb: RS/6000 and PowerPC Options.
30940 * mno-double: FRV Options. (line 41)
30941 * mno-dsp: MIPS Options. (line 178)
30942 * mno-dwarf2-asm: IA-64 Options. (line 79)
30943 * mno-dword: FRV Options. (line 35)
30944 * mno-eabi: RS/6000 and PowerPC Options.
30946 * mno-early-stop-bits: IA-64 Options. (line 85)
30947 * mno-eflags: FRV Options. (line 125)
30948 * mno-embedded-data: MIPS Options. (line 225)
30949 * mno-ep: V850 Options. (line 16)
30950 * mno-epsilon: MMIX Options. (line 15)
30951 * mno-explicit-relocs <1>: MIPS Options. (line 245)
30952 * mno-explicit-relocs: DEC Alpha Options. (line 184)
30953 * mno-fancy-math-387: i386 and x86-64 Options.
30955 * mno-fast-fix: TMS320C3x/C4x Options.
30957 * mno-faster-structs: SPARC Options. (line 71)
30958 * mno-fix: DEC Alpha Options. (line 171)
30959 * mno-fix-r4000: MIPS Options. (line 309)
30960 * mno-fix-r4400: MIPS Options. (line 323)
30961 * mno-float32: PDP-11 Options. (line 48)
30962 * mno-float64: PDP-11 Options. (line 52)
30963 * mno-flush-func: M32R/D Options. (line 99)
30964 * mno-flush-trap: M32R/D Options. (line 91)
30965 * mno-fp-in-toc: RS/6000 and PowerPC Options.
30967 * mno-fp-regs: DEC Alpha Options. (line 25)
30968 * mno-fp-ret-in-387: i386 and x86-64 Options.
30970 * mno-fprnd: RS/6000 and PowerPC Options.
30972 * mno-fpu: SPARC Options. (line 25)
30973 * mno-fused-madd <1>: Xtensa Options. (line 19)
30974 * mno-fused-madd <2>: S/390 and zSeries Options.
30976 * mno-fused-madd <3>: RS/6000 and PowerPC Options.
30978 * mno-fused-madd: MIPS Options. (line 294)
30979 * mno-gnu-as: IA-64 Options. (line 18)
30980 * mno-gnu-ld: IA-64 Options. (line 23)
30981 * mno-gotplt: CRIS Options. (line 86)
30982 * mno-hardlit: MCore Options. (line 10)
30983 * mno-id-shared-library: Blackfin Options. (line 45)
30984 * mno-ieee-fp: i386 and x86-64 Options.
30986 * mno-int16: PDP-11 Options. (line 44)
30987 * mno-int32: PDP-11 Options. (line 40)
30988 * mno-interrupts: AVR Options. (line 39)
30989 * mno-isel: RS/6000 and PowerPC Options.
30991 * mno-knuthdiv: MMIX Options. (line 33)
30992 * mno-libfuncs: MMIX Options. (line 10)
30993 * mno-long-calls <1>: V850 Options. (line 10)
30994 * mno-long-calls <2>: MIPS Options. (line 280)
30995 * mno-long-calls <3>: M68hc1x Options. (line 35)
30996 * mno-long-calls <4>: HPPA Options. (line 138)
30997 * mno-long-calls <5>: Blackfin Options. (line 57)
30998 * mno-long-calls: ARM Options. (line 149)
30999 * mno-longcall: RS/6000 and PowerPC Options.
31001 * mno-longcalls: Xtensa Options. (line 60)
31002 * mno-loop-unsigned: TMS320C3x/C4x Options.
31004 * mno-low-64k: Blackfin Options. (line 36)
31005 * mno-mad: MIPS Options. (line 289)
31006 * mno-max: DEC Alpha Options. (line 171)
31007 * mno-media: FRV Options. (line 47)
31008 * mno-memcpy: MIPS Options. (line 274)
31009 * mno-mfcrf: RS/6000 and PowerPC Options.
31011 * mno-mips16: MIPS Options. (line 81)
31012 * mno-mips3d: MIPS Options. (line 190)
31013 * mno-mmx: i386 and x86-64 Options.
31015 * mno-mpyi: TMS320C3x/C4x Options.
31017 * mno-mul-bug-workaround: CRIS Options. (line 36)
31018 * mno-muladd: FRV Options. (line 53)
31019 * mno-mulhw: RS/6000 and PowerPC Options.
31021 * mno-mult-bug: MN10300 Options. (line 13)
31022 * mno-multi-cond-exec: FRV Options. (line 183)
31023 * mno-multiple: RS/6000 and PowerPC Options.
31025 * mno-mvcle: S/390 and zSeries Options.
31027 * mno-nested-cond-exec: FRV Options. (line 195)
31028 * mno-optimize-membar: FRV Options. (line 205)
31029 * mno-pack: FRV Options. (line 122)
31030 * mno-packed-stack: S/390 and zSeries Options.
31032 * mno-paired-single: MIPS Options. (line 183)
31033 * mno-parallel-insns: TMS320C3x/C4x Options.
31035 * mno-parallel-mpy: TMS320C3x/C4x Options.
31037 * mno-pic: IA-64 Options. (line 26)
31038 * mno-popcntb: RS/6000 and PowerPC Options.
31040 * mno-power: RS/6000 and PowerPC Options.
31042 * mno-power2: RS/6000 and PowerPC Options.
31044 * mno-powerpc: RS/6000 and PowerPC Options.
31046 * mno-powerpc-gfxopt: RS/6000 and PowerPC Options.
31048 * mno-powerpc-gpopt: RS/6000 and PowerPC Options.
31050 * mno-powerpc64: RS/6000 and PowerPC Options.
31052 * mno-prolog-function: V850 Options. (line 23)
31053 * mno-prologue-epilogue: CRIS Options. (line 76)
31054 * mno-prototype: RS/6000 and PowerPC Options.
31056 * mno-push-args: i386 and x86-64 Options.
31058 * mno-register-names: IA-64 Options. (line 37)
31059 * mno-regnames: RS/6000 and PowerPC Options.
31061 * mno-relax-immediate: MCore Options. (line 19)
31062 * mno-relocatable: RS/6000 and PowerPC Options.
31064 * mno-relocatable-lib: RS/6000 and PowerPC Options.
31066 * mno-rptb: TMS320C3x/C4x Options.
31068 * mno-rpts: TMS320C3x/C4x Options.
31070 * mno-scc: FRV Options. (line 146)
31071 * mno-sched-ar-data-spec: IA-64 Options. (line 128)
31072 * mno-sched-ar-in-data-spec: IA-64 Options. (line 149)
31073 * mno-sched-br-data-spec: IA-64 Options. (line 121)
31074 * mno-sched-br-in-data-spec: IA-64 Options. (line 142)
31075 * mno-sched-control-ldc: IA-64 Options. (line 168)
31076 * mno-sched-control-spec: IA-64 Options. (line 135)
31077 * mno-sched-count-spec-in-critical-path: IA-64 Options. (line 194)
31078 * mno-sched-in-control-spec: IA-64 Options. (line 156)
31079 * mno-sched-ldc: IA-64 Options. (line 162)
31080 * mno-sched-prefer-non-control-spec-insns: IA-64 Options. (line 187)
31081 * mno-sched-prefer-non-data-spec-insns: IA-64 Options. (line 180)
31082 * mno-sched-prolog: ARM Options. (line 32)
31083 * mno-sched-spec-verbose: IA-64 Options. (line 176)
31084 * mno-sdata <1>: RS/6000 and PowerPC Options.
31086 * mno-sdata: IA-64 Options. (line 42)
31087 * mno-side-effects: CRIS Options. (line 51)
31088 * mno-single-exit: MMIX Options. (line 66)
31089 * mno-slow-bytes: MCore Options. (line 35)
31090 * mno-small-exec: S/390 and zSeries Options.
31092 * mno-soft-float: DEC Alpha Options. (line 10)
31093 * mno-space-regs: HPPA Options. (line 45)
31094 * mno-spe: RS/6000 and PowerPC Options.
31096 * mno-specld-anomaly: Blackfin Options. (line 19)
31097 * mno-split: PDP-11 Options. (line 71)
31098 * mno-split-addresses: MIPS Options. (line 239)
31099 * mno-sse: i386 and x86-64 Options.
31101 * mno-stack-align: CRIS Options. (line 60)
31102 * mno-stack-bias: SPARC Options. (line 220)
31103 * mno-strict-align <1>: RS/6000 and PowerPC Options.
31105 * mno-strict-align: M680x0 Options. (line 152)
31106 * mno-string: RS/6000 and PowerPC Options.
31108 * mno-sum-in-toc: RS/6000 and PowerPC Options.
31110 * mno-svr3-shlib: i386 and x86-64 Options.
31112 * mno-swdiv: RS/6000 and PowerPC Options.
31114 * mno-sym32: MIPS Options. (line 209)
31115 * mno-tablejump: AVR Options. (line 47)
31116 * mno-target-align: Xtensa Options. (line 47)
31117 * mno-text-section-literals: Xtensa Options. (line 35)
31118 * mno-toc: RS/6000 and PowerPC Options.
31120 * mno-toplevel-symbols: MMIX Options. (line 40)
31121 * mno-tpf-trace: S/390 and zSeries Options.
31123 * mno-unaligned-doubles: SPARC Options. (line 59)
31124 * mno-uninit-const-in-rodata: MIPS Options. (line 233)
31125 * mno-update: RS/6000 and PowerPC Options.
31127 * mno-v8plus: SPARC Options. (line 168)
31128 * mno-vis: SPARC Options. (line 175)
31129 * mno-vliw-branch: FRV Options. (line 170)
31130 * mno-volatile-asm-stop: IA-64 Options. (line 32)
31131 * mno-vrsave: RS/6000 and PowerPC Options.
31133 * mno-wide-bitfields: MCore Options. (line 23)
31134 * mno-xgot: MIPS Options. (line 127)
31135 * mno-xl-compat: RS/6000 and PowerPC Options.
31137 * mno-zero-extend: MMIX Options. (line 27)
31138 * mnobitfield: M680x0 Options. (line 100)
31139 * mnomacsave: SH Options. (line 92)
31140 * mnominmax: M68hc1x Options. (line 31)
31141 * mnop-fun-dllimport: ARM Options. (line 174)
31142 * mold-mnemonics: RS/6000 and PowerPC Options.
31144 * momit-leaf-frame-pointer <1>: i386 and x86-64 Options.
31146 * momit-leaf-frame-pointer: Blackfin Options. (line 7)
31147 * mone-byte-bool: Darwin Options. (line 83)
31148 * moptimize-membar: FRV Options. (line 201)
31149 * MP: Preprocessor Options.
31151 * mpa-risc-1-0: HPPA Options. (line 19)
31152 * mpa-risc-1-1: HPPA Options. (line 19)
31153 * mpa-risc-2-0: HPPA Options. (line 19)
31154 * mpack: FRV Options. (line 119)
31155 * mpacked-stack: S/390 and zSeries Options.
31157 * mpadstruct: SH Options. (line 106)
31158 * mpaired-single: MIPS Options. (line 183)
31159 * mparallel-insns: TMS320C3x/C4x Options.
31161 * mparallel-mpy: TMS320C3x/C4x Options.
31163 * mparanoid: TMS320C3x/C4x Options.
31165 * mpcrel: M680x0 Options. (line 144)
31166 * mpdebug: CRIS Options. (line 40)
31167 * mpe: RS/6000 and PowerPC Options.
31169 * mpentium: i386 and x86-64 Options.
31171 * mpentiumpro: i386 and x86-64 Options.
31173 * mpic-register: ARM Options. (line 183)
31174 * mpoke-function-name: ARM Options. (line 197)
31175 * mpopcntb: RS/6000 and PowerPC Options.
31177 * mportable-runtime: HPPA Options. (line 71)
31178 * mpower: RS/6000 and PowerPC Options.
31180 * mpower2: RS/6000 and PowerPC Options.
31182 * mpowerpc: RS/6000 and PowerPC Options.
31184 * mpowerpc-gfxopt: RS/6000 and PowerPC Options.
31186 * mpowerpc-gpopt: RS/6000 and PowerPC Options.
31188 * mpowerpc64: RS/6000 and PowerPC Options.
31190 * mprefergot: SH Options. (line 113)
31191 * mpreferred-stack-boundary: i386 and x86-64 Options.
31193 * mprioritize-restricted-insns: RS/6000 and PowerPC Options.
31195 * mprolog-function: V850 Options. (line 23)
31196 * mprologue-epilogue: CRIS Options. (line 76)
31197 * mprototype: RS/6000 and PowerPC Options.
31199 * mpt-fixed: SH Options. (line 194)
31200 * mpush-args <1>: i386 and x86-64 Options.
31202 * mpush-args: CRX Options. (line 13)
31203 * MQ: Preprocessor Options.
31205 * mregister-names: IA-64 Options. (line 37)
31206 * mregnames: RS/6000 and PowerPC Options.
31208 * mregparm <1>: TMS320C3x/C4x Options.
31210 * mregparm: i386 and x86-64 Options.
31212 * mrelax <1>: SH Options. (line 70)
31213 * mrelax <2>: MN10300 Options. (line 34)
31214 * mrelax: H8/300 Options. (line 9)
31215 * mrelax-immediate: MCore Options. (line 19)
31216 * mrelocatable: RS/6000 and PowerPC Options.
31218 * mrelocatable-lib: RS/6000 and PowerPC Options.
31220 * mreturn-pointer-on-d0: MN10300 Options. (line 24)
31221 * mrodata: ARC Options. (line 30)
31222 * mrptb: TMS320C3x/C4x Options.
31224 * mrpts: TMS320C3x/C4x Options.
31226 * mrtd <1>: Function Attributes.
31228 * mrtd <2>: M680x0 Options. (line 109)
31229 * mrtd: i386 and x86-64 Options.
31231 * ms: H8/300 Options. (line 17)
31232 * ms2600: H8/300 Options. (line 24)
31233 * mscc: FRV Options. (line 140)
31234 * msched-ar-data-spec: IA-64 Options. (line 128)
31235 * msched-ar-in-data-spec: IA-64 Options. (line 149)
31236 * msched-br-data-spec: IA-64 Options. (line 121)
31237 * msched-br-in-data-spec: IA-64 Options. (line 142)
31238 * msched-control-ldc: IA-64 Options. (line 168)
31239 * msched-control-spec: IA-64 Options. (line 135)
31240 * msched-costly-dep: RS/6000 and PowerPC Options.
31242 * msched-count-spec-in-critical-path: IA-64 Options. (line 194)
31243 * msched-in-control-spec: IA-64 Options. (line 156)
31244 * msched-ldc: IA-64 Options. (line 162)
31245 * msched-prefer-non-control-spec-insns: IA-64 Options. (line 187)
31246 * msched-prefer-non-data-spec-insns: IA-64 Options. (line 180)
31247 * msched-spec-verbose: IA-64 Options. (line 176)
31248 * mschedule: HPPA Options. (line 78)
31249 * mscore5: Score Options. (line 25)
31250 * mscore5u: Score Options. (line 28)
31251 * mscore7: Score Options. (line 31)
31252 * mscore7d: Score Options. (line 34)
31253 * msda: V850 Options. (line 40)
31254 * msdata <1>: RS/6000 and PowerPC Options.
31256 * msdata: IA-64 Options. (line 42)
31257 * msdata-data: RS/6000 and PowerPC Options.
31259 * msdata=default: RS/6000 and PowerPC Options.
31261 * msdata=eabi: RS/6000 and PowerPC Options.
31263 * msdata=none <1>: RS/6000 and PowerPC Options.
31265 * msdata=none: M32R/D Options. (line 40)
31266 * msdata=sdata: M32R/D Options. (line 49)
31267 * msdata=sysv: RS/6000 and PowerPC Options.
31269 * msdata=use: M32R/D Options. (line 53)
31270 * msecure-plt: RS/6000 and PowerPC Options.
31272 * mshared-library-id: Blackfin Options. (line 49)
31273 * mshort <1>: M68hc1x Options. (line 40)
31274 * mshort: M680x0 Options. (line 94)
31275 * msim <1>: Xstormy16 Options. (line 9)
31276 * msim <2>: RS/6000 and PowerPC Options.
31278 * msim <3>: MT Options. (line 22)
31279 * msim: M32C Options. (line 13)
31280 * msingle-exit: MMIX Options. (line 66)
31281 * msingle-float: MIPS Options. (line 169)
31282 * msingle-pic-base: ARM Options. (line 177)
31283 * msio: HPPA Options. (line 107)
31284 * msize: AVR Options. (line 32)
31285 * mslow-bytes: MCore Options. (line 35)
31286 * msmall: TMS320C3x/C4x Options.
31288 * msmall-data: DEC Alpha Options. (line 195)
31289 * msmall-exec: S/390 and zSeries Options.
31291 * msmall-memory: TMS320C3x/C4x Options.
31293 * msmall-text: DEC Alpha Options. (line 213)
31294 * msoft-float <1>: SPARC Options. (line 25)
31295 * msoft-float <2>: S/390 and zSeries Options.
31297 * msoft-float <3>: RS/6000 and PowerPC Options.
31299 * msoft-float <4>: PDP-11 Options. (line 13)
31300 * msoft-float <5>: MIPS Options. (line 165)
31301 * msoft-float <6>: M680x0 Options. (line 84)
31302 * msoft-float <7>: i386 and x86-64 Options.
31304 * msoft-float <8>: HPPA Options. (line 91)
31305 * msoft-float <9>: FRV Options. (line 22)
31306 * msoft-float <10>: DEC Alpha Options. (line 10)
31307 * msoft-float: ARM Options. (line 45)
31308 * msoft-quad-float: SPARC Options. (line 45)
31309 * msoft-reg-count: M68hc1x Options. (line 43)
31310 * mspace <1>: V850 Options. (line 30)
31311 * mspace: SH Options. (line 110)
31312 * mspe: RS/6000 and PowerPC Options.
31314 * mspecld-anomaly: Blackfin Options. (line 14)
31315 * msplit: PDP-11 Options. (line 68)
31316 * msplit-addresses: MIPS Options. (line 239)
31317 * msse: i386 and x86-64 Options.
31319 * msseregparm: i386 and x86-64 Options.
31321 * mstack-align: CRIS Options. (line 60)
31322 * mstack-bias: SPARC Options. (line 220)
31323 * mstack-guard: S/390 and zSeries Options.
31325 * mstack-size: S/390 and zSeries Options.
31327 * mstackrealign: i386 and x86-64 Options.
31329 * mstrict-align <1>: RS/6000 and PowerPC Options.
31331 * mstrict-align: M680x0 Options. (line 152)
31332 * mstring: RS/6000 and PowerPC Options.
31334 * mstructure-size-boundary: ARM Options. (line 129)
31335 * msvr3-shlib: i386 and x86-64 Options.
31337 * msvr4-struct-return: RS/6000 and PowerPC Options.
31339 * mswdiv: RS/6000 and PowerPC Options.
31341 * msym32: MIPS Options. (line 209)
31342 * mt: IA-64 Options. (line 106)
31343 * MT: Preprocessor Options.
31345 * mtarget-align: Xtensa Options. (line 47)
31346 * mtda: V850 Options. (line 34)
31347 * mtext: ARC Options. (line 30)
31348 * mtext-section-literals: Xtensa Options. (line 35)
31349 * mthreads: i386 and x86-64 Options.
31351 * mthumb: ARM Options. (line 218)
31352 * mthumb-interwork: ARM Options. (line 25)
31353 * mti: TMS320C3x/C4x Options.
31355 * mtiny-stack: AVR Options. (line 50)
31356 * mtls-direct-seg-refs: i386 and x86-64 Options.
31358 * mtls-size: IA-64 Options. (line 97)
31359 * mtoc: RS/6000 and PowerPC Options.
31361 * mtomcat-stats: FRV Options. (line 209)
31362 * mtoplevel-symbols: MMIX Options. (line 40)
31363 * mtp: ARM Options. (line 246)
31364 * mtpcs-frame: ARM Options. (line 222)
31365 * mtpcs-leaf-frame: ARM Options. (line 228)
31366 * mtpf-trace: S/390 and zSeries Options.
31368 * mtrap-precision: DEC Alpha Options. (line 109)
31369 * mtune <1>: SPARC Options. (line 156)
31370 * mtune <2>: S/390 and zSeries Options.
31372 * mtune <3>: RS/6000 and PowerPC Options.
31374 * mtune <4>: MIPS Options. (line 44)
31375 * mtune <5>: IA-64 Options. (line 101)
31376 * mtune <6>: i386 and x86-64 Options.
31378 * mtune <7>: DEC Alpha Options. (line 262)
31379 * mtune <8>: CRIS Options. (line 16)
31380 * mtune: ARM Options. (line 99)
31381 * muclibc: GNU/Linux Options. (line 13)
31382 * muls: Score Options. (line 18)
31383 * multcost=NUMBER: SH Options. (line 124)
31384 * multi_module: Darwin Options. (line 190)
31385 * multilib-library-pic: FRV Options. (line 89)
31386 * multiply_defined: Darwin Options. (line 190)
31387 * multiply_defined_unused: Darwin Options. (line 190)
31388 * munaligned-doubles: SPARC Options. (line 59)
31389 * muninit-const-in-rodata: MIPS Options. (line 233)
31390 * munix: VAX Options. (line 9)
31391 * munix-asm: PDP-11 Options. (line 74)
31392 * mupdate: RS/6000 and PowerPC Options.
31394 * musermode: SH Options. (line 118)
31395 * mv850: V850 Options. (line 49)
31396 * mv850e: V850 Options. (line 69)
31397 * mv850e1: V850 Options. (line 64)
31398 * mv8plus: SPARC Options. (line 168)
31399 * mvis: SPARC Options. (line 175)
31400 * mvliw-branch: FRV Options. (line 164)
31401 * mvms-return-codes: DEC Alpha/VMS Options.
31403 * mvolatile-asm-stop: IA-64 Options. (line 32)
31404 * mvr4130-align: MIPS Options. (line 388)
31405 * mvrsave: RS/6000 and PowerPC Options.
31407 * mvxworks: RS/6000 and PowerPC Options.
31409 * mwarn-dynamicstack: S/390 and zSeries Options.
31411 * mwarn-framesize: S/390 and zSeries Options.
31413 * mwide-bitfields: MCore Options. (line 23)
31414 * mwindiss: RS/6000 and PowerPC Options.
31416 * mwords-little-endian: ARM Options. (line 76)
31417 * mxgot: MIPS Options. (line 127)
31418 * mxl-compat: RS/6000 and PowerPC Options.
31420 * myellowknife: RS/6000 and PowerPC Options.
31422 * mzarch: S/390 and zSeries Options.
31424 * mzda: V850 Options. (line 45)
31425 * mzero-extend: MMIX Options. (line 27)
31426 * no-integrated-cpp: C Dialect Options. (line 217)
31427 * no-red-zone: i386 and x86-64 Options.
31429 * no_dead_strip_inits_and_terms: Darwin Options. (line 190)
31430 * noall_load: Darwin Options. (line 190)
31431 * nocpp: MIPS Options. (line 304)
31432 * nodefaultlibs: Link Options. (line 62)
31433 * nofixprebinding: Darwin Options. (line 190)
31434 * nolibdld: HPPA Options. (line 190)
31435 * nomultidefs: Darwin Options. (line 190)
31436 * noprebind: Darwin Options. (line 190)
31437 * noseglinkedit: Darwin Options. (line 190)
31438 * nostartfiles: Link Options. (line 57)
31439 * nostdinc: Preprocessor Options.
31441 * nostdinc++ <1>: Preprocessor Options.
31443 * nostdinc++: C++ Dialect Options.
31445 * nostdlib: Link Options. (line 71)
31446 * o: Preprocessor Options.
31448 * O: Optimize Options. (line 32)
31449 * o: Overall Options. (line 175)
31450 * O0: Optimize Options. (line 104)
31451 * O1: Optimize Options. (line 32)
31452 * O2: Optimize Options. (line 63)
31453 * O3: Optimize Options. (line 99)
31454 * Os: Optimize Options. (line 107)
31455 * P: Preprocessor Options.
31457 * p: Debugging Options. (line 200)
31458 * pagezero_size: Darwin Options. (line 190)
31459 * param: Optimize Options. (line 1358)
31460 * pass-exit-codes: Overall Options. (line 133)
31461 * pedantic <1>: Warnings and Errors.
31463 * pedantic <2>: Alternate Keywords. (line 29)
31464 * pedantic <3>: C Extensions. (line 6)
31465 * pedantic <4>: Preprocessor Options.
31467 * pedantic <5>: Warning Options. (line 27)
31468 * pedantic: Standards. (line 13)
31469 * pedantic-errors <1>: Warnings and Errors.
31471 * pedantic-errors <2>: Non-bugs. (line 216)
31472 * pedantic-errors <3>: Preprocessor Options.
31474 * pedantic-errors <4>: Warning Options. (line 69)
31475 * pedantic-errors: Standards. (line 13)
31476 * pg: Debugging Options. (line 206)
31477 * pie: Link Options. (line 92)
31478 * pipe: Overall Options. (line 197)
31479 * prebind: Darwin Options. (line 190)
31480 * prebind_all_twolevel_modules: Darwin Options. (line 190)
31481 * preprocessor: Preprocessor Options.
31483 * print-file-name: Debugging Options. (line 798)
31484 * print-libgcc-file-name: Debugging Options. (line 819)
31485 * print-multi-directory: Debugging Options. (line 804)
31486 * print-multi-lib: Debugging Options. (line 809)
31487 * print-objc-runtime-info: Objective-C and Objective-C++ Dialect Options.
31489 * print-prog-name: Debugging Options. (line 816)
31490 * print-search-dirs: Debugging Options. (line 827)
31491 * private_bundle: Darwin Options. (line 190)
31492 * pthread <1>: SPARC Options. (line 240)
31493 * pthread <2>: RS/6000 and PowerPC Options.
31495 * pthread: IA-64 Options. (line 106)
31496 * pthreads: SPARC Options. (line 234)
31497 * Q: Debugging Options. (line 212)
31498 * Qn: System V Options. (line 18)
31499 * Qy: System V Options. (line 14)
31500 * rdynamic: Link Options. (line 98)
31501 * read_only_relocs: Darwin Options. (line 190)
31502 * remap: Preprocessor Options.
31504 * s: Link Options. (line 105)
31505 * S <1>: Link Options. (line 20)
31506 * S: Overall Options. (line 158)
31507 * save-temps: Debugging Options. (line 760)
31508 * sectalign: Darwin Options. (line 190)
31509 * sectcreate: Darwin Options. (line 190)
31510 * sectobjectsymbols: Darwin Options. (line 190)
31511 * sectorder: Darwin Options. (line 190)
31512 * seg1addr: Darwin Options. (line 190)
31513 * seg_addr_table: Darwin Options. (line 190)
31514 * seg_addr_table_filename: Darwin Options. (line 190)
31515 * segaddr: Darwin Options. (line 190)
31516 * seglinkedit: Darwin Options. (line 190)
31517 * segprot: Darwin Options. (line 190)
31518 * segs_read_only_addr: Darwin Options. (line 190)
31519 * segs_read_write_addr: Darwin Options. (line 190)
31520 * shared: Link Options. (line 114)
31521 * shared-libgcc: Link Options. (line 122)
31522 * sim: CRIS Options. (line 108)
31523 * sim2: CRIS Options. (line 114)
31524 * single_module: Darwin Options. (line 190)
31525 * specs: Directory Options. (line 84)
31526 * static <1>: HPPA Options. (line 194)
31527 * static <2>: Darwin Options. (line 190)
31528 * static: Link Options. (line 109)
31529 * static-libgcc: Link Options. (line 122)
31530 * std <1>: Non-bugs. (line 107)
31531 * std <2>: Other Builtins. (line 22)
31532 * std <3>: C Dialect Options. (line 47)
31533 * std: Standards. (line 13)
31534 * std=: Preprocessor Options.
31536 * sub_library: Darwin Options. (line 190)
31537 * sub_umbrella: Darwin Options. (line 190)
31538 * symbolic: Link Options. (line 157)
31539 * sysroot: Directory Options. (line 92)
31540 * target-help <1>: Preprocessor Options.
31542 * target-help: Overall Options. (line 228)
31543 * threads <1>: SPARC Options. (line 228)
31544 * threads: HPPA Options. (line 207)
31545 * time: Debugging Options. (line 774)
31546 * tls: FRV Options. (line 75)
31547 * TLS: FRV Options. (line 72)
31548 * traditional <1>: Incompatibilities. (line 6)
31549 * traditional: C Dialect Options. (line 229)
31550 * traditional-cpp <1>: Preprocessor Options.
31552 * traditional-cpp: C Dialect Options. (line 229)
31553 * trigraphs <1>: Preprocessor Options.
31555 * trigraphs: C Dialect Options. (line 213)
31556 * twolevel_namespace: Darwin Options. (line 190)
31557 * u: Link Options. (line 179)
31558 * U: Preprocessor Options.
31560 * umbrella: Darwin Options. (line 190)
31561 * undef: Preprocessor Options.
31563 * undefined: Darwin Options. (line 190)
31564 * unexported_symbols_list: Darwin Options. (line 190)
31565 * V: Target Options. (line 24)
31566 * v <1>: Preprocessor Options.
31568 * v: Overall Options. (line 186)
31569 * version <1>: Preprocessor Options.
31571 * version: Overall Options. (line 232)
31572 * W: Incompatibilities. (line 64)
31573 * w: Preprocessor Options.
31575 * W: Warning Options. (line 593)
31576 * w: Warning Options. (line 73)
31577 * Wa: Assembler Options. (line 9)
31578 * Wabi: C++ Dialect Options.
31580 * Waddress: Warning Options. (line 865)
31581 * Waggregate-return: Warning Options. (line 878)
31582 * Wall <1>: Standard Libraries. (line 6)
31583 * Wall <2>: Preprocessor Options.
31585 * Wall: Warning Options. (line 577)
31586 * Wassign-intercept: Objective-C and Objective-C++ Dialect Options.
31588 * Wattributes: Warning Options. (line 883)
31589 * Wbad-function-cast: Warning Options. (line 813)
31590 * Wcast-align: Warning Options. (line 827)
31591 * Wcast-qual: Warning Options. (line 822)
31592 * Wchar-subscripts: Warning Options. (line 79)
31593 * Wcomment <1>: Preprocessor Options.
31595 * Wcomment: Warning Options. (line 84)
31596 * Wcomments: Preprocessor Options.
31598 * Wconversion <1>: Protoize Caveats. (line 31)
31599 * Wconversion: Warning Options. (line 845)
31600 * Wctor-dtor-privacy: C++ Dialect Options.
31602 * Wdeclaration-after-statement: Warning Options. (line 775)
31603 * Wdisabled-optimization: Warning Options. (line 1136)
31604 * Wdiv-by-zero: Warning Options. (line 667)
31605 * weak_reference_mismatches: Darwin Options. (line 190)
31606 * Weffc++: C++ Dialect Options.
31608 * Wendif-labels <1>: Preprocessor Options.
31610 * Wendif-labels: Warning Options. (line 785)
31611 * Werror <1>: Preprocessor Options.
31613 * Werror: Warning Options. (line 1151)
31614 * Werror-implicit-function-declaration: Warning Options. (line 198)
31615 * Werror=: Warning Options. (line 1154)
31616 * Wextra: Warning Options. (line 593)
31617 * Wfatal-errors: Warning Options. (line 89)
31618 * Wfloat-equal: Warning Options. (line 683)
31619 * Wformat <1>: Function Attributes.
31621 * Wformat: Warning Options. (line 94)
31622 * Wformat-nonliteral <1>: Function Attributes.
31624 * Wformat-nonliteral: Warning Options. (line 151)
31625 * Wformat-security: Warning Options. (line 156)
31626 * Wformat-y2k: Warning Options. (line 129)
31627 * Wformat=2: Warning Options. (line 167)
31628 * Wframe-larger-than: Warning Options. (line 797)
31629 * whatsloaded: Darwin Options. (line 190)
31630 * whyload: Darwin Options. (line 190)
31631 * Wimplicit: Warning Options. (line 204)
31632 * Wimplicit-function-declaration: Warning Options. (line 198)
31633 * Wimplicit-int: Warning Options. (line 193)
31634 * Wimport: Preprocessor Options.
31636 * Winit-self: Warning Options. (line 179)
31637 * Winline <1>: Inline. (line 42)
31638 * Winline: Warning Options. (line 1076)
31639 * Winvalid-pch: Warning Options. (line 1111)
31640 * Wl: Link Options. (line 175)
31641 * Wlarger-than: Warning Options. (line 794)
31642 * Wlong-long: Warning Options. (line 1115)
31643 * Wmain: Warning Options. (line 208)
31644 * Wmissing-braces: Warning Options. (line 214)
31645 * Wmissing-declarations: Warning Options. (line 905)
31646 * Wmissing-field-initializers: Warning Options. (line 911)
31647 * Wmissing-format-attribute: Warning Options. (line 937)
31648 * Wmissing-include-dirs: Warning Options. (line 224)
31649 * Wmissing-noreturn: Warning Options. (line 929)
31650 * Wmissing-prototypes: Warning Options. (line 899)
31651 * Wmultichar: Warning Options. (line 956)
31652 * Wnested-externs: Warning Options. (line 1051)
31653 * Wno-address: Warning Options. (line 865)
31654 * Wno-attributes: Warning Options. (line 883)
31655 * Wno-deprecated: C++ Dialect Options.
31657 * Wno-deprecated-declarations: Warning Options. (line 1005)
31658 * Wno-div-by-zero: Warning Options. (line 667)
31659 * Wno-endif-labels: Warning Options. (line 785)
31660 * Wno-format-extra-args: Warning Options. (line 133)
31661 * Wno-format-zero-length: Warning Options. (line 147)
31662 * Wno-import: Warning Options. (line 76)
31663 * Wno-int-to-pointer-cast: Warning Options. (line 1103)
31664 * Wno-invalid-offsetof: Warning Options. (line 1089)
31665 * Wno-long-long: Warning Options. (line 1115)
31666 * Wno-multichar: Warning Options. (line 956)
31667 * Wno-non-template-friend: C++ Dialect Options.
31669 * Wno-overflow: Warning Options. (line 1011)
31670 * Wno-pmf-conversions <1>: Bound member functions.
31672 * Wno-pmf-conversions: C++ Dialect Options.
31674 * Wno-pointer-sign: Warning Options. (line 1145)
31675 * Wno-pointer-to-int-cast: Warning Options. (line 1107)
31676 * Wno-pragmas: Warning Options. (line 479)
31677 * Wno-protocol: Objective-C and Objective-C++ Dialect Options.
31679 * Wno-variadic-macros: Warning Options. (line 1121)
31680 * Wno-vla: Warning Options. (line 1127)
31681 * Wno-volatile-register-var: Warning Options. (line 1131)
31682 * Wnon-virtual-dtor: C++ Dialect Options.
31684 * Wnonnull: Warning Options. (line 172)
31685 * Wnormalized: Warning Options. (line 962)
31686 * Wold-style-cast: C++ Dialect Options.
31688 * Wold-style-definition: Warning Options. (line 895)
31689 * Woverlength-strings: Warning Options. (line 1173)
31690 * Woverloaded-virtual: C++ Dialect Options.
31692 * Woverride-init: Warning Options. (line 1014)
31693 * Wp: Preprocessor Options.
31695 * Wpacked: Warning Options. (line 1022)
31696 * Wpadded: Warning Options. (line 1039)
31697 * Wparentheses: Warning Options. (line 227)
31698 * Wpointer-arith <1>: Pointer Arith. (line 13)
31699 * Wpointer-arith: Warning Options. (line 807)
31700 * Wpointer-sign: Warning Options. (line 1145)
31701 * Wpragmas: Warning Options. (line 479)
31702 * Wredundant-decls: Warning Options. (line 1046)
31703 * Wreorder: C++ Dialect Options.
31705 * Wreturn-type: Warning Options. (line 317)
31706 * Wselector: Objective-C and Objective-C++ Dialect Options.
31708 * Wsequence-point: Warning Options. (line 271)
31709 * Wshadow: Warning Options. (line 789)
31710 * Wsign-compare: Warning Options. (line 858)
31711 * Wsign-promo: C++ Dialect Options.
31713 * Wstack-protector: Warning Options. (line 1168)
31714 * Wstrict-aliasing: Warning Options. (line 484)
31715 * Wstrict-aliasing=n: Warning Options. (line 492)
31716 * Wstrict-null-sentinel: C++ Dialect Options.
31718 * Wstrict-overflow: Warning Options. (line 526)
31719 * Wstrict-prototypes: Warning Options. (line 889)
31720 * Wstrict-selector-match: Objective-C and Objective-C++ Dialect Options.
31722 * Wswitch: Warning Options. (line 336)
31723 * Wswitch-enum: Warning Options. (line 347)
31724 * Wswitch-switch: Warning Options. (line 344)
31725 * Wsystem-headers <1>: Preprocessor Options.
31727 * Wsystem-headers: Warning Options. (line 672)
31728 * Wtraditional <1>: Preprocessor Options.
31730 * Wtraditional: Warning Options. (line 698)
31731 * Wtrigraphs <1>: Preprocessor Options.
31733 * Wtrigraphs: Warning Options. (line 353)
31734 * Wundeclared-selector: Objective-C and Objective-C++ Dialect Options.
31736 * Wundef <1>: Preprocessor Options.
31738 * Wundef: Warning Options. (line 782)
31739 * Wuninitialized: Warning Options. (line 398)
31740 * Wunknown-pragmas: Warning Options. (line 472)
31741 * Wunreachable-code: Warning Options. (line 1054)
31742 * Wunsafe-loop-optimizations: Warning Options. (line 801)
31743 * Wunused: Warning Options. (line 391)
31744 * Wunused-function: Warning Options. (line 358)
31745 * Wunused-label: Warning Options. (line 363)
31746 * Wunused-macros: Preprocessor Options.
31748 * Wunused-parameter: Warning Options. (line 370)
31749 * Wunused-value: Warning Options. (line 385)
31750 * Wunused-variable: Warning Options. (line 377)
31751 * Wvariadic-macros: Warning Options. (line 1121)
31752 * Wvla: Warning Options. (line 1127)
31753 * Wvolatile-register-var: Warning Options. (line 1131)
31754 * Wwrite-strings: Warning Options. (line 833)
31755 * x <1>: Preprocessor Options.
31757 * x: Overall Options. (line 109)
31758 * Xassembler: Assembler Options. (line 13)
31759 * Xlinker: Link Options. (line 163)
31760 * Ym: System V Options. (line 26)
31761 * YP: System V Options. (line 22)
31764 File: gcc.info, Node: Keyword Index, Prev: Option Index, Up: Top
31772 * ! in constraint: Multi-Alternative. (line 33)
31773 * # in constraint: Modifiers. (line 57)
31774 * #pragma: Pragmas. (line 6)
31775 * #pragma implementation: C++ Interface. (line 39)
31776 * #pragma implementation, implied: C++ Interface. (line 46)
31777 * #pragma interface: C++ Interface. (line 20)
31778 * #pragma, reason for not using: Function Attributes.
31780 * $: Dollar Signs. (line 6)
31781 * % in constraint: Modifiers. (line 45)
31782 * %include: Spec Files. (line 27)
31783 * %include_noerr: Spec Files. (line 31)
31784 * %rename: Spec Files. (line 35)
31785 * & in constraint: Modifiers. (line 25)
31786 * ': Incompatibilities. (line 116)
31787 * * in constraint: Modifiers. (line 62)
31788 * + in constraint: Modifiers. (line 12)
31789 * -lgcc, use with -nodefaultlibs: Link Options. (line 79)
31790 * -lgcc, use with -nostdlib: Link Options. (line 79)
31791 * -nodefaultlibs and unresolved references: Link Options. (line 79)
31792 * -nostdlib and unresolved references: Link Options. (line 79)
31793 * .sdata/.sdata2 references (PowerPC): RS/6000 and PowerPC Options.
31795 * //: C++ Comments. (line 6)
31796 * 0 in constraint: Simple Constraints. (line 115)
31797 * < in constraint: Simple Constraints. (line 46)
31798 * = in constraint: Modifiers. (line 8)
31799 * > in constraint: Simple Constraints. (line 50)
31800 * ? in constraint: Multi-Alternative. (line 27)
31801 * ?: extensions: Conditionals. (line 6)
31802 * ?: side effect: Conditionals. (line 20)
31803 * _ in variables in macros: Typeof. (line 42)
31804 * __builtin___fprintf_chk: Object Size Checking.
31806 * __builtin___memcpy_chk: Object Size Checking.
31808 * __builtin___memmove_chk: Object Size Checking.
31810 * __builtin___mempcpy_chk: Object Size Checking.
31812 * __builtin___memset_chk: Object Size Checking.
31814 * __builtin___printf_chk: Object Size Checking.
31816 * __builtin___snprintf_chk: Object Size Checking.
31818 * __builtin___sprintf_chk: Object Size Checking.
31820 * __builtin___stpcpy_chk: Object Size Checking.
31822 * __builtin___strcat_chk: Object Size Checking.
31824 * __builtin___strcpy_chk: Object Size Checking.
31826 * __builtin___strncat_chk: Object Size Checking.
31828 * __builtin___strncpy_chk: Object Size Checking.
31830 * __builtin___vfprintf_chk: Object Size Checking.
31832 * __builtin___vprintf_chk: Object Size Checking.
31834 * __builtin___vsnprintf_chk: Object Size Checking.
31836 * __builtin___vsprintf_chk: Object Size Checking.
31838 * __builtin_apply: Constructing Calls. (line 31)
31839 * __builtin_apply_args: Constructing Calls. (line 20)
31840 * __builtin_choose_expr: Other Builtins. (line 150)
31841 * __builtin_clz: Other Builtins. (line 383)
31842 * __builtin_clzl: Other Builtins. (line 401)
31843 * __builtin_clzll: Other Builtins. (line 421)
31844 * __builtin_constant_p: Other Builtins. (line 190)
31845 * __builtin_ctz: Other Builtins. (line 387)
31846 * __builtin_ctzl: Other Builtins. (line 405)
31847 * __builtin_ctzll: Other Builtins. (line 425)
31848 * __builtin_expect: Other Builtins. (line 236)
31849 * __builtin_ffs: Other Builtins. (line 379)
31850 * __builtin_ffsl: Other Builtins. (line 397)
31851 * __builtin_ffsll: Other Builtins. (line 417)
31852 * __builtin_frame_address: Return Address. (line 34)
31853 * __builtin_huge_val: Other Builtins. (line 300)
31854 * __builtin_huge_valf: Other Builtins. (line 305)
31855 * __builtin_huge_vall: Other Builtins. (line 308)
31856 * __builtin_inf: Other Builtins. (line 312)
31857 * __builtin_infd128: Other Builtins. (line 322)
31858 * __builtin_infd32: Other Builtins. (line 316)
31859 * __builtin_infd64: Other Builtins. (line 319)
31860 * __builtin_inff: Other Builtins. (line 326)
31861 * __builtin_infl: Other Builtins. (line 331)
31862 * __builtin_isgreater: Other Builtins. (line 6)
31863 * __builtin_isgreaterequal: Other Builtins. (line 6)
31864 * __builtin_isless: Other Builtins. (line 6)
31865 * __builtin_islessequal: Other Builtins. (line 6)
31866 * __builtin_islessgreater: Other Builtins. (line 6)
31867 * __builtin_isunordered: Other Builtins. (line 6)
31868 * __builtin_nan: Other Builtins. (line 335)
31869 * __builtin_nand128: Other Builtins. (line 357)
31870 * __builtin_nand32: Other Builtins. (line 351)
31871 * __builtin_nand64: Other Builtins. (line 354)
31872 * __builtin_nanf: Other Builtins. (line 361)
31873 * __builtin_nanl: Other Builtins. (line 364)
31874 * __builtin_nans: Other Builtins. (line 368)
31875 * __builtin_nansf: Other Builtins. (line 372)
31876 * __builtin_nansl: Other Builtins. (line 375)
31877 * __builtin_object_size: Object Size Checking.
31879 * __builtin_offsetof: Offsetof. (line 6)
31880 * __builtin_parity: Other Builtins. (line 394)
31881 * __builtin_parityl: Other Builtins. (line 413)
31882 * __builtin_parityll: Other Builtins. (line 433)
31883 * __builtin_popcount: Other Builtins. (line 391)
31884 * __builtin_popcountl: Other Builtins. (line 409)
31885 * __builtin_popcountll: Other Builtins. (line 429)
31886 * __builtin_powi: Other Builtins. (line 6)
31887 * __builtin_powif: Other Builtins. (line 6)
31888 * __builtin_powil: Other Builtins. (line 6)
31889 * __builtin_prefetch: Other Builtins. (line 261)
31890 * __builtin_return: Constructing Calls. (line 48)
31891 * __builtin_return_address: Return Address. (line 11)
31892 * __builtin_types_compatible_p: Other Builtins. (line 104)
31893 * __complex__ keyword: Complex. (line 6)
31894 * __declspec(dllexport): Function Attributes.
31896 * __declspec(dllimport): Function Attributes.
31898 * __extension__: Alternate Keywords. (line 29)
31899 * __func__ identifier: Function Names. (line 6)
31900 * __FUNCTION__ identifier: Function Names. (line 6)
31901 * __imag__ keyword: Complex. (line 27)
31902 * __PRETTY_FUNCTION__ identifier: Function Names. (line 6)
31903 * __real__ keyword: Complex. (line 27)
31904 * __STDC_HOSTED__: Standards. (line 6)
31905 * __sync_add_and_fetch: Atomic Builtins. (line 57)
31906 * __sync_and_and_fetch: Atomic Builtins. (line 57)
31907 * __sync_bool_compare_and_swap: Atomic Builtins. (line 65)
31908 * __sync_fetch_and_add: Atomic Builtins. (line 45)
31909 * __sync_fetch_and_and: Atomic Builtins. (line 45)
31910 * __sync_fetch_and_nand: Atomic Builtins. (line 45)
31911 * __sync_fetch_and_or: Atomic Builtins. (line 45)
31912 * __sync_fetch_and_sub: Atomic Builtins. (line 45)
31913 * __sync_fetch_and_xor: Atomic Builtins. (line 45)
31914 * __sync_lock_release: Atomic Builtins. (line 95)
31915 * __sync_lock_test_and_set: Atomic Builtins. (line 77)
31916 * __sync_nand_and_fetch: Atomic Builtins. (line 57)
31917 * __sync_or_and_fetch: Atomic Builtins. (line 57)
31918 * __sync_sub_and_fetch: Atomic Builtins. (line 57)
31919 * __sync_synchronize: Atomic Builtins. (line 74)
31920 * __sync_val_compare_and_swap: Atomic Builtins. (line 65)
31921 * __sync_xor_and_fetch: Atomic Builtins. (line 57)
31922 * __thread: Thread-Local. (line 6)
31923 * _Complex keyword: Complex. (line 6)
31924 * _Decimal128 data type: Decimal Float. (line 6)
31925 * _Decimal32 data type: Decimal Float. (line 6)
31926 * _Decimal64 data type: Decimal Float. (line 6)
31927 * _exit: Other Builtins. (line 6)
31928 * _Exit: Other Builtins. (line 6)
31929 * ABI: Compatibility. (line 6)
31930 * abort: Other Builtins. (line 6)
31931 * abs: Other Builtins. (line 6)
31932 * accessing volatiles: Volatiles. (line 6)
31933 * acos: Other Builtins. (line 6)
31934 * acosf: Other Builtins. (line 6)
31935 * acosh: Other Builtins. (line 6)
31936 * acoshf: Other Builtins. (line 6)
31937 * acoshl: Other Builtins. (line 6)
31938 * acosl: Other Builtins. (line 6)
31939 * Ada: G++ and GCC. (line 6)
31940 * address constraints: Simple Constraints. (line 142)
31941 * address of a label: Labels as Values. (line 6)
31942 * address_operand: Simple Constraints. (line 146)
31943 * alias attribute: Function Attributes.
31945 * aliasing of parameters: Code Gen Options. (line 360)
31946 * aligned attribute <1>: Type Attributes. (line 30)
31947 * aligned attribute: Variable Attributes.
31949 * alignment: Alignment. (line 6)
31950 * alloca: Other Builtins. (line 6)
31951 * alloca vs variable-length arrays: Variable Length. (line 27)
31952 * Allow nesting in an interrupt handler on the Blackfin processor.: Function Attributes.
31954 * alternate keywords: Alternate Keywords. (line 6)
31955 * always_inline function attribute: Function Attributes.
31957 * AMD x86-64 Options: i386 and x86-64 Options.
31959 * AMD1: Standards. (line 6)
31960 * ANSI C: Standards. (line 6)
31961 * ANSI C standard: Standards. (line 6)
31962 * ANSI C89: Standards. (line 6)
31963 * ANSI support: C Dialect Options. (line 10)
31964 * ANSI X3.159-1989: Standards. (line 6)
31965 * apostrophes: Incompatibilities. (line 116)
31966 * application binary interface: Compatibility. (line 6)
31967 * ARC Options: ARC Options. (line 6)
31968 * ARM [Annotated C++ Reference Manual]: Backwards Compatibility.
31970 * ARM options: ARM Options. (line 6)
31971 * arrays of length zero: Zero Length. (line 6)
31972 * arrays of variable length: Variable Length. (line 6)
31973 * arrays, non-lvalue: Subscripting. (line 6)
31974 * asin: Other Builtins. (line 6)
31975 * asinf: Other Builtins. (line 6)
31976 * asinh: Other Builtins. (line 6)
31977 * asinhf: Other Builtins. (line 6)
31978 * asinhl: Other Builtins. (line 6)
31979 * asinl: Other Builtins. (line 6)
31980 * asm constraints: Constraints. (line 6)
31981 * asm expressions: Extended Asm. (line 6)
31982 * assembler instructions: Extended Asm. (line 6)
31983 * assembler names for identifiers: Asm Labels. (line 6)
31984 * assembly code, invalid: Bug Criteria. (line 12)
31985 * atan: Other Builtins. (line 6)
31986 * atan2: Other Builtins. (line 6)
31987 * atan2f: Other Builtins. (line 6)
31988 * atan2l: Other Builtins. (line 6)
31989 * atanf: Other Builtins. (line 6)
31990 * atanh: Other Builtins. (line 6)
31991 * atanhf: Other Builtins. (line 6)
31992 * atanhl: Other Builtins. (line 6)
31993 * atanl: Other Builtins. (line 6)
31994 * attribute of types: Type Attributes. (line 6)
31995 * attribute of variables: Variable Attributes.
31997 * attribute syntax: Attribute Syntax. (line 6)
31998 * autoincrement/decrement addressing: Simple Constraints. (line 28)
31999 * automatic inline for C++ member fns: Inline. (line 53)
32000 * AVR Options: AVR Options. (line 6)
32001 * Backwards Compatibility: Backwards Compatibility.
32003 * base class members: Name lookup. (line 6)
32004 * bcmp: Other Builtins. (line 6)
32005 * below100 attribute: Variable Attributes.
32007 * binary compatibility: Compatibility. (line 6)
32008 * Blackfin Options: Blackfin Options. (line 6)
32009 * bound pointer to member function: Bound member functions.
32011 * bounds checking: Optimize Options. (line 333)
32012 * bug criteria: Bug Criteria. (line 6)
32013 * bugs: Bugs. (line 6)
32014 * bugs, known: Trouble. (line 6)
32015 * built-in functions <1>: Other Builtins. (line 6)
32016 * built-in functions: C Dialect Options. (line 149)
32017 * bzero: Other Builtins. (line 6)
32018 * C compilation options: Invoking GCC. (line 17)
32019 * C intermediate output, nonexistent: G++ and GCC. (line 35)
32020 * C language extensions: C Extensions. (line 6)
32021 * C language, traditional: C Dialect Options. (line 227)
32022 * C standard: Standards. (line 6)
32023 * C standards: Standards. (line 6)
32024 * c++: Invoking G++. (line 13)
32025 * C++: G++ and GCC. (line 30)
32026 * C++ comments: C++ Comments. (line 6)
32027 * C++ compilation options: Invoking GCC. (line 23)
32028 * C++ interface and implementation headers: C++ Interface. (line 6)
32029 * C++ language extensions: C++ Extensions. (line 6)
32030 * C++ member fns, automatically inline: Inline. (line 53)
32031 * C++ misunderstandings: C++ Misunderstandings.
32033 * C++ options, command line: C++ Dialect Options.
32035 * C++ pragmas, effect on inlining: C++ Interface. (line 66)
32036 * C++ source file suffixes: Invoking G++. (line 6)
32037 * C++ static data, declaring and defining: Static Definitions.
32039 * C89: Standards. (line 6)
32040 * C90: Standards. (line 6)
32041 * C94: Standards. (line 6)
32042 * C95: Standards. (line 6)
32043 * C99: Standards. (line 6)
32044 * C9X: Standards. (line 6)
32045 * C_INCLUDE_PATH: Environment Variables.
32047 * cabs: Other Builtins. (line 6)
32048 * cabsf: Other Builtins. (line 6)
32049 * cabsl: Other Builtins. (line 6)
32050 * cacos: Other Builtins. (line 6)
32051 * cacosf: Other Builtins. (line 6)
32052 * cacosh: Other Builtins. (line 6)
32053 * cacoshf: Other Builtins. (line 6)
32054 * cacoshl: Other Builtins. (line 6)
32055 * cacosl: Other Builtins. (line 6)
32056 * calling functions through the function vector on the H8/300 processors: Function Attributes.
32058 * calloc: Other Builtins. (line 6)
32059 * carg: Other Builtins. (line 6)
32060 * cargf: Other Builtins. (line 6)
32061 * cargl: Other Builtins. (line 6)
32062 * case labels in initializers: Designated Inits. (line 6)
32063 * case ranges: Case Ranges. (line 6)
32064 * casin: Other Builtins. (line 6)
32065 * casinf: Other Builtins. (line 6)
32066 * casinh: Other Builtins. (line 6)
32067 * casinhf: Other Builtins. (line 6)
32068 * casinhl: Other Builtins. (line 6)
32069 * casinl: Other Builtins. (line 6)
32070 * cast to a union: Cast to Union. (line 6)
32071 * catan: Other Builtins. (line 6)
32072 * catanf: Other Builtins. (line 6)
32073 * catanh: Other Builtins. (line 6)
32074 * catanhf: Other Builtins. (line 6)
32075 * catanhl: Other Builtins. (line 6)
32076 * catanl: Other Builtins. (line 6)
32077 * cbrt: Other Builtins. (line 6)
32078 * cbrtf: Other Builtins. (line 6)
32079 * cbrtl: Other Builtins. (line 6)
32080 * ccos: Other Builtins. (line 6)
32081 * ccosf: Other Builtins. (line 6)
32082 * ccosh: Other Builtins. (line 6)
32083 * ccoshf: Other Builtins. (line 6)
32084 * ccoshl: Other Builtins. (line 6)
32085 * ccosl: Other Builtins. (line 6)
32086 * ceil: Other Builtins. (line 6)
32087 * ceilf: Other Builtins. (line 6)
32088 * ceill: Other Builtins. (line 6)
32089 * cexp: Other Builtins. (line 6)
32090 * cexpf: Other Builtins. (line 6)
32091 * cexpl: Other Builtins. (line 6)
32092 * character set, execution: Preprocessor Options.
32094 * character set, input: Preprocessor Options.
32096 * character set, input normalization: Warning Options. (line 962)
32097 * character set, wide execution: Preprocessor Options.
32099 * cimag: Other Builtins. (line 6)
32100 * cimagf: Other Builtins. (line 6)
32101 * cimagl: Other Builtins. (line 6)
32102 * cleanup attribute: Variable Attributes.
32104 * clog: Other Builtins. (line 6)
32105 * clogf: Other Builtins. (line 6)
32106 * clogl: Other Builtins. (line 6)
32107 * COBOL: G++ and GCC. (line 23)
32108 * code generation conventions: Code Gen Options. (line 6)
32109 * code, mixed with declarations: Mixed Declarations. (line 6)
32110 * command options: Invoking GCC. (line 6)
32111 * comments, C++ style: C++ Comments. (line 6)
32112 * common attribute: Variable Attributes.
32114 * comparison of signed and unsigned values, warning: Warning Options.
32116 * compiler bugs, reporting: Bug Reporting. (line 6)
32117 * compiler compared to C++ preprocessor: G++ and GCC. (line 35)
32118 * compiler options, C++: C++ Dialect Options.
32120 * compiler options, Objective-C and Objective-C++: Objective-C and Objective-C++ Dialect Options.
32122 * compiler version, specifying: Target Options. (line 6)
32123 * COMPILER_PATH: Environment Variables.
32125 * complex conjugation: Complex. (line 34)
32126 * complex numbers: Complex. (line 6)
32127 * compound literals: Compound Literals. (line 6)
32128 * computed gotos: Labels as Values. (line 6)
32129 * conditional expressions, extensions: Conditionals. (line 6)
32130 * conflicting types: Disappointments. (line 21)
32131 * conj: Other Builtins. (line 6)
32132 * conjf: Other Builtins. (line 6)
32133 * conjl: Other Builtins. (line 6)
32134 * const applied to function: Function Attributes.
32136 * const function attribute: Function Attributes.
32138 * constants in constraints: Simple Constraints. (line 58)
32139 * constraint modifier characters: Modifiers. (line 6)
32140 * constraint, matching: Simple Constraints. (line 127)
32141 * constraints, asm: Constraints. (line 6)
32142 * constraints, machine specific: Machine Constraints.
32144 * constructing calls: Constructing Calls. (line 6)
32145 * constructor expressions: Compound Literals. (line 6)
32146 * constructor function attribute: Function Attributes.
32148 * contributors: Contributors. (line 6)
32149 * copysign: Other Builtins. (line 6)
32150 * copysignf: Other Builtins. (line 6)
32151 * copysignl: Other Builtins. (line 6)
32152 * core dump: Bug Criteria. (line 9)
32153 * cos: Other Builtins. (line 6)
32154 * cosf: Other Builtins. (line 6)
32155 * cosh: Other Builtins. (line 6)
32156 * coshf: Other Builtins. (line 6)
32157 * coshl: Other Builtins. (line 6)
32158 * cosl: Other Builtins. (line 6)
32159 * CPATH: Environment Variables.
32161 * CPLUS_INCLUDE_PATH: Environment Variables.
32163 * cpow: Other Builtins. (line 6)
32164 * cpowf: Other Builtins. (line 6)
32165 * cpowl: Other Builtins. (line 6)
32166 * cproj: Other Builtins. (line 6)
32167 * cprojf: Other Builtins. (line 6)
32168 * cprojl: Other Builtins. (line 6)
32169 * creal: Other Builtins. (line 6)
32170 * crealf: Other Builtins. (line 6)
32171 * creall: Other Builtins. (line 6)
32172 * CRIS Options: CRIS Options. (line 6)
32173 * cross compiling: Target Options. (line 6)
32174 * CRX Options: CRX Options. (line 6)
32175 * csin: Other Builtins. (line 6)
32176 * csinf: Other Builtins. (line 6)
32177 * csinh: Other Builtins. (line 6)
32178 * csinhf: Other Builtins. (line 6)
32179 * csinhl: Other Builtins. (line 6)
32180 * csinl: Other Builtins. (line 6)
32181 * csqrt: Other Builtins. (line 6)
32182 * csqrtf: Other Builtins. (line 6)
32183 * csqrtl: Other Builtins. (line 6)
32184 * ctan: Other Builtins. (line 6)
32185 * ctanf: Other Builtins. (line 6)
32186 * ctanh: Other Builtins. (line 6)
32187 * ctanhf: Other Builtins. (line 6)
32188 * ctanhl: Other Builtins. (line 6)
32189 * ctanl: Other Builtins. (line 6)
32190 * Darwin options: Darwin Options. (line 6)
32191 * dcgettext: Other Builtins. (line 6)
32192 * DD integer suffix: Decimal Float. (line 6)
32193 * dd integer suffix: Decimal Float. (line 6)
32194 * deallocating variable length arrays: Variable Length. (line 23)
32195 * debugging information options: Debugging Options. (line 6)
32196 * decimal floating types: Decimal Float. (line 6)
32197 * declaration scope: Incompatibilities. (line 80)
32198 * declarations inside expressions: Statement Exprs. (line 6)
32199 * declarations, mixed with code: Mixed Declarations. (line 6)
32200 * declaring attributes of functions: Function Attributes.
32202 * declaring static data in C++: Static Definitions. (line 6)
32203 * defining static data in C++: Static Definitions. (line 6)
32204 * dependencies for make as output: Environment Variables.
32206 * dependencies, make: Preprocessor Options.
32208 * DEPENDENCIES_OUTPUT: Environment Variables.
32210 * dependent name lookup: Name lookup. (line 6)
32211 * deprecated attribute: Variable Attributes.
32213 * deprecated attribute.: Function Attributes.
32215 * designated initializers: Designated Inits. (line 6)
32216 * designator lists: Designated Inits. (line 94)
32217 * designators: Designated Inits. (line 61)
32218 * destructor function attribute: Function Attributes.
32220 * DF integer suffix: Decimal Float. (line 6)
32221 * df integer suffix: Decimal Float. (line 6)
32222 * dgettext: Other Builtins. (line 6)
32223 * diagnostic messages: Language Independent Options.
32225 * dialect options: C Dialect Options. (line 6)
32226 * digits in constraint: Simple Constraints. (line 115)
32227 * directory options: Directory Options. (line 6)
32228 * DL integer suffix: Decimal Float. (line 6)
32229 * dl integer suffix: Decimal Float. (line 6)
32230 * dollar signs in identifier names: Dollar Signs. (line 6)
32231 * double-word arithmetic: Long Long. (line 6)
32232 * downward funargs: Nested Functions. (line 6)
32233 * drem: Other Builtins. (line 6)
32234 * dremf: Other Builtins. (line 6)
32235 * dreml: Other Builtins. (line 6)
32236 * E in constraint: Simple Constraints. (line 77)
32237 * earlyclobber operand: Modifiers. (line 25)
32238 * eight bit data on the H8/300, H8/300H, and H8S: Function Attributes.
32240 * empty structures: Empty Structures. (line 6)
32241 * environment variables: Environment Variables.
32243 * erf: Other Builtins. (line 6)
32244 * erfc: Other Builtins. (line 6)
32245 * erfcf: Other Builtins. (line 6)
32246 * erfcl: Other Builtins. (line 6)
32247 * erff: Other Builtins. (line 6)
32248 * erfl: Other Builtins. (line 6)
32249 * error messages: Warnings and Errors.
32251 * escaped newlines: Escaped Newlines. (line 6)
32252 * exception handler functions on the Blackfin processor: Function Attributes.
32254 * exclamation point: Multi-Alternative. (line 33)
32255 * exit: Other Builtins. (line 6)
32256 * exp: Other Builtins. (line 6)
32257 * exp10: Other Builtins. (line 6)
32258 * exp10f: Other Builtins. (line 6)
32259 * exp10l: Other Builtins. (line 6)
32260 * exp2: Other Builtins. (line 6)
32261 * exp2f: Other Builtins. (line 6)
32262 * exp2l: Other Builtins. (line 6)
32263 * expf: Other Builtins. (line 6)
32264 * expl: Other Builtins. (line 6)
32265 * explicit register variables: Explicit Reg Vars. (line 6)
32266 * expm1: Other Builtins. (line 6)
32267 * expm1f: Other Builtins. (line 6)
32268 * expm1l: Other Builtins. (line 6)
32269 * expressions containing statements: Statement Exprs. (line 6)
32270 * expressions, constructor: Compound Literals. (line 6)
32271 * extended asm: Extended Asm. (line 6)
32272 * extensible constraints: Simple Constraints. (line 151)
32273 * extensions, ?:: Conditionals. (line 6)
32274 * extensions, C language: C Extensions. (line 6)
32275 * extensions, C++ language: C++ Extensions. (line 6)
32276 * external declaration scope: Incompatibilities. (line 80)
32277 * externally_visible attribute.: Function Attributes.
32279 * F in constraint: Simple Constraints. (line 82)
32280 * fabs: Other Builtins. (line 6)
32281 * fabsf: Other Builtins. (line 6)
32282 * fabsl: Other Builtins. (line 6)
32283 * fatal signal: Bug Criteria. (line 9)
32284 * fdim: Other Builtins. (line 6)
32285 * fdimf: Other Builtins. (line 6)
32286 * fdiml: Other Builtins. (line 6)
32287 * FDL, GNU Free Documentation License: GNU Free Documentation License.
32289 * ffs: Other Builtins. (line 6)
32290 * file name suffix: Overall Options. (line 14)
32291 * file names: Link Options. (line 10)
32292 * flatten function attribute: Function Attributes.
32294 * flexible array members: Zero Length. (line 6)
32295 * float as function value type: Incompatibilities. (line 141)
32296 * floating point precision <1>: Disappointments. (line 68)
32297 * floating point precision: Optimize Options. (line 1060)
32298 * floor: Other Builtins. (line 6)
32299 * floorf: Other Builtins. (line 6)
32300 * floorl: Other Builtins. (line 6)
32301 * fma: Other Builtins. (line 6)
32302 * fmaf: Other Builtins. (line 6)
32303 * fmal: Other Builtins. (line 6)
32304 * fmax: Other Builtins. (line 6)
32305 * fmaxf: Other Builtins. (line 6)
32306 * fmaxl: Other Builtins. (line 6)
32307 * fmin: Other Builtins. (line 6)
32308 * fminf: Other Builtins. (line 6)
32309 * fminl: Other Builtins. (line 6)
32310 * fmod: Other Builtins. (line 6)
32311 * fmodf: Other Builtins. (line 6)
32312 * fmodl: Other Builtins. (line 6)
32313 * force_align_arg_pointer attribute: Function Attributes.
32315 * format function attribute: Function Attributes.
32317 * format_arg function attribute: Function Attributes.
32319 * Fortran: G++ and GCC. (line 6)
32320 * forwarding calls: Constructing Calls. (line 6)
32321 * fprintf: Other Builtins. (line 6)
32322 * fprintf_unlocked: Other Builtins. (line 6)
32323 * fputs: Other Builtins. (line 6)
32324 * fputs_unlocked: Other Builtins. (line 6)
32325 * freestanding environment: Standards. (line 6)
32326 * freestanding implementation: Standards. (line 6)
32327 * frexp: Other Builtins. (line 6)
32328 * frexpf: Other Builtins. (line 6)
32329 * frexpl: Other Builtins. (line 6)
32330 * FRV Options: FRV Options. (line 6)
32331 * fscanf: Other Builtins. (line 6)
32332 * fscanf, and constant strings: Incompatibilities. (line 17)
32333 * function addressability on the M32R/D: Function Attributes.
32335 * function attributes: Function Attributes.
32337 * function pointers, arithmetic: Pointer Arith. (line 6)
32338 * function prototype declarations: Function Prototypes.
32340 * function without a prologue/epilogue code: Function Attributes.
32342 * function, size of pointer to: Pointer Arith. (line 6)
32343 * functions called via pointer on the RS/6000 and PowerPC: Function Attributes.
32345 * functions in arbitrary sections: Function Attributes.
32347 * functions that are passed arguments in registers on the 386: Function Attributes.
32349 * functions that behave like malloc: Function Attributes.
32351 * functions that do not pop the argument stack on the 386: Function Attributes.
32353 * functions that do pop the argument stack on the 386: Function Attributes.
32355 * functions that have no side effects: Function Attributes.
32357 * functions that never return: Function Attributes.
32359 * functions that pop the argument stack on the 386: Function Attributes.
32361 * functions that return more than once: Function Attributes.
32363 * functions which do not handle memory bank switching on 68HC11/68HC12: Function Attributes.
32365 * functions which handle memory bank switching: Function Attributes.
32367 * functions with non-null pointer arguments: Function Attributes.
32369 * functions with printf, scanf, strftime or strfmon style arguments: Function Attributes.
32371 * g in constraint: Simple Constraints. (line 108)
32372 * G in constraint: Simple Constraints. (line 86)
32373 * g++: Invoking G++. (line 13)
32374 * G++: G++ and GCC. (line 30)
32375 * gamma: Other Builtins. (line 6)
32376 * gammaf: Other Builtins. (line 6)
32377 * gammal: Other Builtins. (line 6)
32378 * GCC: G++ and GCC. (line 6)
32379 * GCC command options: Invoking GCC. (line 6)
32380 * GCC_EXEC_PREFIX: Environment Variables.
32382 * gcc_struct: Type Attributes. (line 302)
32383 * gcc_struct attribute: Variable Attributes.
32385 * gcov: Debugging Options. (line 238)
32386 * gettext: Other Builtins. (line 6)
32387 * global offset table: Code Gen Options. (line 163)
32388 * global register after longjmp: Global Reg Vars. (line 66)
32389 * global register variables: Global Reg Vars. (line 6)
32390 * GNAT: G++ and GCC. (line 30)
32391 * GNU C Compiler: G++ and GCC. (line 6)
32392 * GNU Compiler Collection: G++ and GCC. (line 6)
32393 * gnu_inline function attribute: Function Attributes.
32395 * goto with computed label: Labels as Values. (line 6)
32396 * gp-relative references (MIPS): MIPS Options. (line 216)
32397 * gprof: Debugging Options. (line 205)
32398 * grouping options: Invoking GCC. (line 26)
32399 * H in constraint: Simple Constraints. (line 86)
32400 * hardware models and configurations, specifying: Submodel Options.
32402 * hex floats: Hex Floats. (line 6)
32403 * hosted environment <1>: C Dialect Options. (line 183)
32404 * hosted environment: Standards. (line 6)
32405 * hosted implementation: Standards. (line 6)
32406 * HPPA Options: HPPA Options. (line 6)
32407 * hypot: Other Builtins. (line 6)
32408 * hypotf: Other Builtins. (line 6)
32409 * hypotl: Other Builtins. (line 6)
32410 * I in constraint: Simple Constraints. (line 69)
32411 * i in constraint: Simple Constraints. (line 58)
32412 * i386 Options: i386 and x86-64 Options.
32414 * IA-64 Options: IA-64 Options. (line 6)
32415 * IBM RS/6000 and PowerPC Options: RS/6000 and PowerPC Options.
32417 * identifier names, dollar signs in: Dollar Signs. (line 6)
32418 * identifiers, names in assembler code: Asm Labels. (line 6)
32419 * ilogb: Other Builtins. (line 6)
32420 * ilogbf: Other Builtins. (line 6)
32421 * ilogbl: Other Builtins. (line 6)
32422 * imaxabs: Other Builtins. (line 6)
32423 * implementation-defined behavior, C language: C Implementation.
32425 * implied #pragma implementation: C++ Interface. (line 46)
32426 * incompatibilities of GCC: Incompatibilities. (line 6)
32427 * increment operators: Bug Criteria. (line 17)
32428 * index: Other Builtins. (line 6)
32429 * indirect calls on ARM: Function Attributes.
32431 * indirect calls on MIPS: Function Attributes.
32433 * init_priority attribute: C++ Attributes. (line 9)
32434 * initializations in expressions: Compound Literals. (line 6)
32435 * initializers with labeled elements: Designated Inits. (line 6)
32436 * initializers, non-constant: Initializers. (line 6)
32437 * inline automatic for C++ member fns: Inline. (line 53)
32438 * inline functions: Inline. (line 6)
32439 * inline functions, omission of: Inline. (line 58)
32440 * inlining and C++ pragmas: C++ Interface. (line 66)
32441 * installation trouble: Trouble. (line 6)
32442 * integrating function code: Inline. (line 6)
32443 * Intel 386 Options: i386 and x86-64 Options.
32445 * interface and implementation headers, C++: C++ Interface. (line 6)
32446 * intermediate C version, nonexistent: G++ and GCC. (line 35)
32447 * interrupt handler functions: Function Attributes.
32449 * interrupt handler functions on the Blackfin, m68k, H8/300 and SH processors: Function Attributes.
32451 * introduction: Top. (line 6)
32452 * invalid assembly code: Bug Criteria. (line 12)
32453 * invalid input: Bug Criteria. (line 42)
32454 * invoking g++: Invoking G++. (line 21)
32455 * isalnum: Other Builtins. (line 6)
32456 * isalpha: Other Builtins. (line 6)
32457 * isascii: Other Builtins. (line 6)
32458 * isblank: Other Builtins. (line 6)
32459 * iscntrl: Other Builtins. (line 6)
32460 * isdigit: Other Builtins. (line 6)
32461 * isgraph: Other Builtins. (line 6)
32462 * islower: Other Builtins. (line 6)
32463 * ISO 9899: Standards. (line 6)
32464 * ISO C: Standards. (line 6)
32465 * ISO C standard: Standards. (line 6)
32466 * ISO C90: Standards. (line 6)
32467 * ISO C94: Standards. (line 6)
32468 * ISO C95: Standards. (line 6)
32469 * ISO C99: Standards. (line 6)
32470 * ISO C9X: Standards. (line 6)
32471 * ISO support: C Dialect Options. (line 10)
32472 * ISO/IEC 9899: Standards. (line 6)
32473 * isprint: Other Builtins. (line 6)
32474 * ispunct: Other Builtins. (line 6)
32475 * isspace: Other Builtins. (line 6)
32476 * isupper: Other Builtins. (line 6)
32477 * iswalnum: Other Builtins. (line 6)
32478 * iswalpha: Other Builtins. (line 6)
32479 * iswblank: Other Builtins. (line 6)
32480 * iswcntrl: Other Builtins. (line 6)
32481 * iswdigit: Other Builtins. (line 6)
32482 * iswgraph: Other Builtins. (line 6)
32483 * iswlower: Other Builtins. (line 6)
32484 * iswprint: Other Builtins. (line 6)
32485 * iswpunct: Other Builtins. (line 6)
32486 * iswspace: Other Builtins. (line 6)
32487 * iswupper: Other Builtins. (line 6)
32488 * iswxdigit: Other Builtins. (line 6)
32489 * isxdigit: Other Builtins. (line 6)
32490 * j0: Other Builtins. (line 6)
32491 * j0f: Other Builtins. (line 6)
32492 * j0l: Other Builtins. (line 6)
32493 * j1: Other Builtins. (line 6)
32494 * j1f: Other Builtins. (line 6)
32495 * j1l: Other Builtins. (line 6)
32496 * Java: G++ and GCC. (line 6)
32497 * java_interface attribute: C++ Attributes. (line 29)
32498 * jn: Other Builtins. (line 6)
32499 * jnf: Other Builtins. (line 6)
32500 * jnl: Other Builtins. (line 6)
32501 * keywords, alternate: Alternate Keywords. (line 6)
32502 * known causes of trouble: Trouble. (line 6)
32503 * labeled elements in initializers: Designated Inits. (line 6)
32504 * labels as values: Labels as Values. (line 6)
32505 * labs: Other Builtins. (line 6)
32506 * LANG: Environment Variables.
32508 * language dialect options: C Dialect Options. (line 6)
32509 * LC_ALL: Environment Variables.
32511 * LC_CTYPE: Environment Variables.
32513 * LC_MESSAGES: Environment Variables.
32515 * ldexp: Other Builtins. (line 6)
32516 * ldexpf: Other Builtins. (line 6)
32517 * ldexpl: Other Builtins. (line 6)
32518 * length-zero arrays: Zero Length. (line 6)
32519 * lgamma: Other Builtins. (line 6)
32520 * lgammaf: Other Builtins. (line 6)
32521 * lgammal: Other Builtins. (line 6)
32522 * Libraries: Link Options. (line 24)
32523 * LIBRARY_PATH: Environment Variables.
32525 * link options: Link Options. (line 6)
32526 * LL integer suffix: Long Long. (line 6)
32527 * llabs: Other Builtins. (line 6)
32528 * llrint: Other Builtins. (line 6)
32529 * llrintf: Other Builtins. (line 6)
32530 * llrintl: Other Builtins. (line 6)
32531 * llround: Other Builtins. (line 6)
32532 * llroundf: Other Builtins. (line 6)
32533 * llroundl: Other Builtins. (line 6)
32534 * load address instruction: Simple Constraints. (line 142)
32535 * local labels: Local Labels. (line 6)
32536 * local variables in macros: Typeof. (line 42)
32537 * local variables, specifying registers: Local Reg Vars. (line 6)
32538 * locale: Environment Variables.
32540 * locale definition: Environment Variables.
32542 * log: Other Builtins. (line 6)
32543 * log10: Other Builtins. (line 6)
32544 * log10f: Other Builtins. (line 6)
32545 * log10l: Other Builtins. (line 6)
32546 * log1p: Other Builtins. (line 6)
32547 * log1pf: Other Builtins. (line 6)
32548 * log1pl: Other Builtins. (line 6)
32549 * log2: Other Builtins. (line 6)
32550 * log2f: Other Builtins. (line 6)
32551 * log2l: Other Builtins. (line 6)
32552 * logb: Other Builtins. (line 6)
32553 * logbf: Other Builtins. (line 6)
32554 * logbl: Other Builtins. (line 6)
32555 * logf: Other Builtins. (line 6)
32556 * logl: Other Builtins. (line 6)
32557 * long long data types: Long Long. (line 6)
32558 * longjmp: Global Reg Vars. (line 66)
32559 * longjmp incompatibilities: Incompatibilities. (line 39)
32560 * longjmp warnings: Warning Options. (line 455)
32561 * lrint: Other Builtins. (line 6)
32562 * lrintf: Other Builtins. (line 6)
32563 * lrintl: Other Builtins. (line 6)
32564 * lround: Other Builtins. (line 6)
32565 * lroundf: Other Builtins. (line 6)
32566 * lroundl: Other Builtins. (line 6)
32567 * m in constraint: Simple Constraints. (line 17)
32568 * M32C options: M32C Options. (line 6)
32569 * M32R/D options: M32R/D Options. (line 6)
32570 * M680x0 options: M680x0 Options. (line 6)
32571 * M68hc1x options: M68hc1x Options. (line 6)
32572 * machine dependent options: Submodel Options. (line 6)
32573 * machine specific constraints: Machine Constraints.
32575 * macro with variable arguments: Variadic Macros. (line 6)
32576 * macros containing asm: Extended Asm. (line 239)
32577 * macros, inline alternative: Inline. (line 6)
32578 * macros, local labels: Local Labels. (line 6)
32579 * macros, local variables in: Typeof. (line 42)
32580 * macros, statements in expressions: Statement Exprs. (line 6)
32581 * macros, types of arguments: Typeof. (line 6)
32582 * make: Preprocessor Options.
32584 * malloc: Other Builtins. (line 6)
32585 * malloc attribute: Function Attributes.
32587 * matching constraint: Simple Constraints. (line 127)
32588 * MCore options: MCore Options. (line 6)
32589 * member fns, automatically inline: Inline. (line 53)
32590 * memcmp: Other Builtins. (line 6)
32591 * memcpy: Other Builtins. (line 6)
32592 * memory references in constraints: Simple Constraints. (line 17)
32593 * mempcpy: Other Builtins. (line 6)
32594 * memset: Other Builtins. (line 6)
32595 * Mercury: G++ and GCC. (line 23)
32596 * message formatting: Language Independent Options.
32598 * messages, warning: Warning Options. (line 6)
32599 * messages, warning and error: Warnings and Errors.
32601 * middle-operands, omitted: Conditionals. (line 6)
32602 * MIPS options: MIPS Options. (line 6)
32603 * misunderstandings in C++: C++ Misunderstandings.
32605 * mixed declarations and code: Mixed Declarations. (line 6)
32606 * mktemp, and constant strings: Incompatibilities. (line 13)
32607 * MMIX Options: MMIX Options. (line 6)
32608 * MN10300 options: MN10300 Options. (line 6)
32609 * mode attribute: Variable Attributes.
32611 * modf: Other Builtins. (line 6)
32612 * modff: Other Builtins. (line 6)
32613 * modfl: Other Builtins. (line 6)
32614 * modifiers in constraints: Modifiers. (line 6)
32615 * ms_struct: Type Attributes. (line 302)
32616 * ms_struct attribute: Variable Attributes.
32618 * MT options: MT Options. (line 6)
32619 * mudflap: Optimize Options. (line 333)
32620 * multiple alternative constraints: Multi-Alternative. (line 6)
32621 * multiprecision arithmetic: Long Long. (line 6)
32622 * n in constraint: Simple Constraints. (line 63)
32623 * names used in assembler code: Asm Labels. (line 6)
32624 * naming convention, implementation headers: C++ Interface. (line 46)
32625 * nearbyint: Other Builtins. (line 6)
32626 * nearbyintf: Other Builtins. (line 6)
32627 * nearbyintl: Other Builtins. (line 6)
32628 * nested functions: Nested Functions. (line 6)
32629 * newlines (escaped): Escaped Newlines. (line 6)
32630 * nextafter: Other Builtins. (line 6)
32631 * nextafterf: Other Builtins. (line 6)
32632 * nextafterl: Other Builtins. (line 6)
32633 * nexttoward: Other Builtins. (line 6)
32634 * nexttowardf: Other Builtins. (line 6)
32635 * nexttowardl: Other Builtins. (line 6)
32636 * NFC: Warning Options. (line 962)
32637 * NFKC: Warning Options. (line 962)
32638 * NMI handler functions on the Blackfin processor: Function Attributes.
32640 * no_instrument_function function attribute: Function Attributes.
32642 * nocommon attribute: Variable Attributes.
32644 * noinline function attribute: Function Attributes.
32646 * non-constant initializers: Initializers. (line 6)
32647 * non-static inline function: Inline. (line 70)
32648 * nonnull function attribute: Function Attributes.
32650 * noreturn function attribute: Function Attributes.
32652 * nothrow function attribute: Function Attributes.
32654 * o in constraint: Simple Constraints. (line 21)
32655 * OBJC_INCLUDE_PATH: Environment Variables.
32657 * Objective-C <1>: Standards. (line 110)
32658 * Objective-C: G++ and GCC. (line 6)
32659 * Objective-C and Objective-C++ options, command line: Objective-C and Objective-C++ Dialect Options.
32661 * Objective-C++ <1>: Standards. (line 110)
32662 * Objective-C++: G++ and GCC. (line 6)
32663 * offsettable address: Simple Constraints. (line 21)
32664 * old-style function definitions: Function Prototypes.
32666 * omitted middle-operands: Conditionals. (line 6)
32667 * open coding: Inline. (line 6)
32668 * openmp parallel: C Dialect Options. (line 200)
32669 * operand constraints, asm: Constraints. (line 6)
32670 * optimize options: Optimize Options. (line 6)
32671 * options to control diagnostics formatting: Language Independent Options.
32673 * options to control warnings: Warning Options. (line 6)
32674 * options, C++: C++ Dialect Options.
32676 * options, code generation: Code Gen Options. (line 6)
32677 * options, debugging: Debugging Options. (line 6)
32678 * options, dialect: C Dialect Options. (line 6)
32679 * options, directory search: Directory Options. (line 6)
32680 * options, GCC command: Invoking GCC. (line 6)
32681 * options, grouping: Invoking GCC. (line 26)
32682 * options, linking: Link Options. (line 6)
32683 * options, Objective-C and Objective-C++: Objective-C and Objective-C++ Dialect Options.
32685 * options, optimization: Optimize Options. (line 6)
32686 * options, order: Invoking GCC. (line 30)
32687 * options, preprocessor: Preprocessor Options.
32689 * order of evaluation, side effects: Non-bugs. (line 196)
32690 * order of options: Invoking GCC. (line 30)
32691 * other register constraints: Simple Constraints. (line 151)
32692 * output file option: Overall Options. (line 174)
32693 * overloaded virtual fn, warning: C++ Dialect Options.
32695 * p in constraint: Simple Constraints. (line 142)
32696 * packed attribute: Variable Attributes.
32698 * parameter forward declaration: Variable Length. (line 60)
32699 * parameters, aliased: Code Gen Options. (line 360)
32700 * Pascal: G++ and GCC. (line 23)
32701 * PDP-11 Options: PDP-11 Options. (line 6)
32702 * PIC: Code Gen Options. (line 163)
32703 * pmf: Bound member functions.
32705 * pointer arguments: Function Attributes.
32707 * pointer to member function: Bound member functions.
32709 * portions of temporary objects, pointers to: Temporaries. (line 6)
32710 * pow: Other Builtins. (line 6)
32711 * pow10: Other Builtins. (line 6)
32712 * pow10f: Other Builtins. (line 6)
32713 * pow10l: Other Builtins. (line 6)
32714 * PowerPC options: PowerPC Options. (line 6)
32715 * powf: Other Builtins. (line 6)
32716 * powl: Other Builtins. (line 6)
32717 * pragma, align: Solaris Pragmas. (line 11)
32718 * pragma, diagnostic: Diagnostic Pragmas. (line 14)
32719 * pragma, extern_prefix: Symbol-Renaming Pragmas.
32721 * pragma, fini: Solaris Pragmas. (line 19)
32722 * pragma, init: Solaris Pragmas. (line 24)
32723 * pragma, long_calls: ARM Pragmas. (line 11)
32724 * pragma, long_calls_off: ARM Pragmas. (line 17)
32725 * pragma, longcall: RS/6000 and PowerPC Pragmas.
32727 * pragma, mark: Darwin Pragmas. (line 11)
32728 * pragma, memregs: M32C Pragmas. (line 7)
32729 * pragma, no_long_calls: ARM Pragmas. (line 14)
32730 * pragma, options align: Darwin Pragmas. (line 14)
32731 * pragma, reason for not using: Function Attributes.
32733 * pragma, redefine_extname: Symbol-Renaming Pragmas.
32735 * pragma, segment: Darwin Pragmas. (line 21)
32736 * pragma, unused: Darwin Pragmas. (line 24)
32737 * pragma, visibility: Visibility Pragmas. (line 8)
32738 * pragma, weak: Weak Pragmas. (line 10)
32739 * pragmas: Pragmas. (line 6)
32740 * pragmas in C++, effect on inlining: C++ Interface. (line 66)
32741 * pragmas, interface and implementation: C++ Interface. (line 6)
32742 * pragmas, warning of unknown: Warning Options. (line 472)
32743 * precompiled headers: Precompiled Headers.
32745 * preprocessing numbers: Incompatibilities. (line 173)
32746 * preprocessing tokens: Incompatibilities. (line 173)
32747 * preprocessor options: Preprocessor Options.
32749 * printf: Other Builtins. (line 6)
32750 * printf_unlocked: Other Builtins. (line 6)
32751 * prof: Debugging Options. (line 199)
32752 * promotion of formal parameters: Function Prototypes.
32754 * pure function attribute: Function Attributes.
32756 * push address instruction: Simple Constraints. (line 142)
32757 * putchar: Other Builtins. (line 6)
32758 * puts: Other Builtins. (line 6)
32759 * qsort, and global register variables: Global Reg Vars. (line 42)
32760 * question mark: Multi-Alternative. (line 27)
32761 * r in constraint: Simple Constraints. (line 54)
32762 * ranges in case statements: Case Ranges. (line 6)
32763 * read-only strings: Incompatibilities. (line 9)
32764 * register variable after longjmp: Global Reg Vars. (line 66)
32765 * registers: Extended Asm. (line 6)
32766 * registers for local variables: Local Reg Vars. (line 6)
32767 * registers in constraints: Simple Constraints. (line 54)
32768 * registers, global allocation: Explicit Reg Vars. (line 6)
32769 * registers, global variables in: Global Reg Vars. (line 6)
32770 * regparm attribute: Function Attributes.
32772 * relocation truncated to fit (MIPS): MIPS Options. (line 135)
32773 * remainder: Other Builtins. (line 6)
32774 * remainderf: Other Builtins. (line 6)
32775 * remainderl: Other Builtins. (line 6)
32776 * remquo: Other Builtins. (line 6)
32777 * remquof: Other Builtins. (line 6)
32778 * remquol: Other Builtins. (line 6)
32779 * reordering, warning: C++ Dialect Options.
32781 * reporting bugs: Bugs. (line 6)
32782 * rest argument (in macro): Variadic Macros. (line 6)
32783 * restricted pointers: Restricted Pointers.
32785 * restricted references: Restricted Pointers.
32787 * restricted this pointer: Restricted Pointers.
32789 * returns_twice attribute: Function Attributes.
32791 * rindex: Other Builtins. (line 6)
32792 * rint: Other Builtins. (line 6)
32793 * rintf: Other Builtins. (line 6)
32794 * rintl: Other Builtins. (line 6)
32795 * round: Other Builtins. (line 6)
32796 * roundf: Other Builtins. (line 6)
32797 * roundl: Other Builtins. (line 6)
32798 * RS/6000 and PowerPC Options: RS/6000 and PowerPC Options.
32800 * RTTI: Vague Linkage. (line 43)
32801 * run-time options: Code Gen Options. (line 6)
32802 * s in constraint: Simple Constraints. (line 90)
32803 * S/390 and zSeries Options: S/390 and zSeries Options.
32805 * save all registers on the Blackfin, H8/300, H8/300H, and H8S: Function Attributes.
32807 * scalb: Other Builtins. (line 6)
32808 * scalbf: Other Builtins. (line 6)
32809 * scalbl: Other Builtins. (line 6)
32810 * scalbln: Other Builtins. (line 6)
32811 * scalblnf: Other Builtins. (line 6)
32812 * scalbn: Other Builtins. (line 6)
32813 * scalbnf: Other Builtins. (line 6)
32814 * scanf, and constant strings: Incompatibilities. (line 17)
32815 * scanfnl: Other Builtins. (line 6)
32816 * scope of a variable length array: Variable Length. (line 23)
32817 * scope of declaration: Disappointments. (line 21)
32818 * scope of external declarations: Incompatibilities. (line 80)
32819 * Score Options: Score Options. (line 6)
32820 * search path: Directory Options. (line 6)
32821 * section function attribute: Function Attributes.
32823 * section variable attribute: Variable Attributes.
32825 * sentinel function attribute: Function Attributes.
32827 * setjmp: Global Reg Vars. (line 66)
32828 * setjmp incompatibilities: Incompatibilities. (line 39)
32829 * shared strings: Incompatibilities. (line 9)
32830 * shared variable attribute: Variable Attributes.
32832 * side effect in ?:: Conditionals. (line 20)
32833 * side effects, macro argument: Statement Exprs. (line 35)
32834 * side effects, order of evaluation: Non-bugs. (line 196)
32835 * signal handler functions on the AVR processors: Function Attributes.
32837 * signbit: Other Builtins. (line 6)
32838 * signbitf: Other Builtins. (line 6)
32839 * signbitl: Other Builtins. (line 6)
32840 * signed and unsigned values, comparison warning: Warning Options.
32842 * significand: Other Builtins. (line 6)
32843 * significandf: Other Builtins. (line 6)
32844 * significandl: Other Builtins. (line 6)
32845 * simple constraints: Simple Constraints. (line 6)
32846 * sin: Other Builtins. (line 6)
32847 * sincos: Other Builtins. (line 6)
32848 * sincosf: Other Builtins. (line 6)
32849 * sincosl: Other Builtins. (line 6)
32850 * sinf: Other Builtins. (line 6)
32851 * sinh: Other Builtins. (line 6)
32852 * sinhf: Other Builtins. (line 6)
32853 * sinhl: Other Builtins. (line 6)
32854 * sinl: Other Builtins. (line 6)
32855 * sizeof: Typeof. (line 6)
32856 * smaller data references: M32R/D Options. (line 57)
32857 * smaller data references (MIPS): MIPS Options. (line 216)
32858 * smaller data references (PowerPC): RS/6000 and PowerPC Options.
32860 * snprintf: Other Builtins. (line 6)
32861 * SPARC options: SPARC Options. (line 6)
32862 * Spec Files: Spec Files. (line 6)
32863 * specified registers: Explicit Reg Vars. (line 6)
32864 * specifying compiler version and target machine: Target Options.
32866 * specifying hardware config: Submodel Options. (line 6)
32867 * specifying machine version: Target Options. (line 6)
32868 * specifying registers for local variables: Local Reg Vars. (line 6)
32869 * speed of compilation: Precompiled Headers.
32871 * sprintf: Other Builtins. (line 6)
32872 * sqrt: Other Builtins. (line 6)
32873 * sqrtf: Other Builtins. (line 6)
32874 * sqrtl: Other Builtins. (line 6)
32875 * sscanf: Other Builtins. (line 6)
32876 * sscanf, and constant strings: Incompatibilities. (line 17)
32877 * sseregparm attribute: Function Attributes.
32879 * statements inside expressions: Statement Exprs. (line 6)
32880 * static data in C++, declaring and defining: Static Definitions.
32882 * stpcpy: Other Builtins. (line 6)
32883 * stpncpy: Other Builtins. (line 6)
32884 * strcasecmp: Other Builtins. (line 6)
32885 * strcat: Other Builtins. (line 6)
32886 * strchr: Other Builtins. (line 6)
32887 * strcmp: Other Builtins. (line 6)
32888 * strcpy: Other Builtins. (line 6)
32889 * strcspn: Other Builtins. (line 6)
32890 * strdup: Other Builtins. (line 6)
32891 * strfmon: Other Builtins. (line 6)
32892 * strftime: Other Builtins. (line 6)
32893 * string constants: Incompatibilities. (line 9)
32894 * strlen: Other Builtins. (line 6)
32895 * strncasecmp: Other Builtins. (line 6)
32896 * strncat: Other Builtins. (line 6)
32897 * strncmp: Other Builtins. (line 6)
32898 * strncpy: Other Builtins. (line 6)
32899 * strndup: Other Builtins. (line 6)
32900 * strpbrk: Other Builtins. (line 6)
32901 * strrchr: Other Builtins. (line 6)
32902 * strspn: Other Builtins. (line 6)
32903 * strstr: Other Builtins. (line 6)
32904 * struct: Unnamed Fields. (line 6)
32905 * structures: Incompatibilities. (line 146)
32906 * structures, constructor expression: Compound Literals. (line 6)
32907 * submodel options: Submodel Options. (line 6)
32908 * subscripting: Subscripting. (line 6)
32909 * subscripting and function values: Subscripting. (line 6)
32910 * suffixes for C++ source: Invoking G++. (line 6)
32911 * SUNPRO_DEPENDENCIES: Environment Variables.
32913 * suppressing warnings: Warning Options. (line 6)
32914 * surprises in C++: C++ Misunderstandings.
32916 * syntax checking: Warning Options. (line 22)
32917 * system headers, warnings from: Warning Options. (line 672)
32918 * tan: Other Builtins. (line 6)
32919 * tanf: Other Builtins. (line 6)
32920 * tanh: Other Builtins. (line 6)
32921 * tanhf: Other Builtins. (line 6)
32922 * tanhl: Other Builtins. (line 6)
32923 * tanl: Other Builtins. (line 6)
32924 * target machine, specifying: Target Options. (line 6)
32925 * target options: Target Options. (line 6)
32926 * TC1: Standards. (line 6)
32927 * TC2: Standards. (line 6)
32928 * Technical Corrigenda: Standards. (line 6)
32929 * Technical Corrigendum 1: Standards. (line 6)
32930 * Technical Corrigendum 2: Standards. (line 6)
32931 * template instantiation: Template Instantiation.
32933 * temporaries, lifetime of: Temporaries. (line 6)
32934 * tgamma: Other Builtins. (line 6)
32935 * tgammaf: Other Builtins. (line 6)
32936 * tgammal: Other Builtins. (line 6)
32937 * Thread-Local Storage: Thread-Local. (line 6)
32938 * thunks: Nested Functions. (line 6)
32939 * tiny data section on the H8/300H and H8S: Function Attributes.
32941 * TLS: Thread-Local. (line 6)
32942 * tls_model attribute: Variable Attributes.
32944 * TMPDIR: Environment Variables.
32946 * TMS320C3x/C4x Options: TMS320C3x/C4x Options.
32948 * toascii: Other Builtins. (line 6)
32949 * tolower: Other Builtins. (line 6)
32950 * toupper: Other Builtins. (line 6)
32951 * towlower: Other Builtins. (line 6)
32952 * towupper: Other Builtins. (line 6)
32953 * traditional C language: C Dialect Options. (line 227)
32954 * treelang <1>: Standards. (line 123)
32955 * treelang: G++ and GCC. (line 6)
32956 * trunc: Other Builtins. (line 6)
32957 * truncf: Other Builtins. (line 6)
32958 * truncl: Other Builtins. (line 6)
32959 * two-stage name lookup: Name lookup. (line 6)
32960 * type alignment: Alignment. (line 6)
32961 * type attributes: Type Attributes. (line 6)
32962 * type_info: Vague Linkage. (line 43)
32963 * typedef names as function parameters: Incompatibilities. (line 97)
32964 * typeof: Typeof. (line 6)
32965 * ULL integer suffix: Long Long. (line 6)
32966 * Ultrix calling convention: Interoperation. (line 150)
32967 * undefined behavior: Bug Criteria. (line 17)
32968 * undefined function value: Bug Criteria. (line 17)
32969 * underscores in variables in macros: Typeof. (line 42)
32970 * union: Unnamed Fields. (line 6)
32971 * union, casting to a: Cast to Union. (line 6)
32972 * unions: Incompatibilities. (line 146)
32973 * unknown pragmas, warning: Warning Options. (line 472)
32974 * unresolved references and -nodefaultlibs: Link Options. (line 79)
32975 * unresolved references and -nostdlib: Link Options. (line 79)
32976 * unused attribute.: Function Attributes.
32978 * used attribute.: Function Attributes.
32980 * User stack pointer in interrupts on the Blackfin: Function Attributes.
32982 * V in constraint: Simple Constraints. (line 41)
32983 * V850 Options: V850 Options. (line 6)
32984 * vague linkage: Vague Linkage. (line 6)
32985 * value after longjmp: Global Reg Vars. (line 66)
32986 * variable addressability on the IA-64: Function Attributes.
32988 * variable addressability on the M32R/D: Variable Attributes.
32990 * variable alignment: Alignment. (line 6)
32991 * variable attributes: Variable Attributes.
32993 * variable number of arguments: Variadic Macros. (line 6)
32994 * variable-length array scope: Variable Length. (line 23)
32995 * variable-length arrays: Variable Length. (line 6)
32996 * variables in specified registers: Explicit Reg Vars. (line 6)
32997 * variables, local, in macros: Typeof. (line 42)
32998 * variadic macros: Variadic Macros. (line 6)
32999 * VAX calling convention: Interoperation. (line 150)
33000 * VAX options: VAX Options. (line 6)
33001 * vfprintf: Other Builtins. (line 6)
33002 * vfscanf: Other Builtins. (line 6)
33003 * visibility attribute: Function Attributes.
33005 * VLAs: Variable Length. (line 6)
33006 * void pointers, arithmetic: Pointer Arith. (line 6)
33007 * void, size of pointer to: Pointer Arith. (line 6)
33008 * volatile access: Volatiles. (line 6)
33009 * volatile applied to function: Function Attributes.
33011 * volatile read: Volatiles. (line 6)
33012 * volatile write: Volatiles. (line 6)
33013 * vprintf: Other Builtins. (line 6)
33014 * vscanf: Other Builtins. (line 6)
33015 * vsnprintf: Other Builtins. (line 6)
33016 * vsprintf: Other Builtins. (line 6)
33017 * vsscanf: Other Builtins. (line 6)
33018 * vtable: Vague Linkage. (line 28)
33019 * warn_unused_result attribute: Function Attributes.
33021 * warning for comparison of signed and unsigned values: Warning Options.
33023 * warning for overloaded virtual fn: C++ Dialect Options.
33025 * warning for reordering of member initializers: C++ Dialect Options.
33027 * warning for unknown pragmas: Warning Options. (line 472)
33028 * warning messages: Warning Options. (line 6)
33029 * warnings from system headers: Warning Options. (line 672)
33030 * warnings vs errors: Warnings and Errors.
33032 * weak attribute: Function Attributes.
33034 * weakref attribute: Function Attributes.
33036 * whitespace: Incompatibilities. (line 112)
33037 * X in constraint: Simple Constraints. (line 112)
33038 * X3.159-1989: Standards. (line 6)
33039 * x86-64 options: x86-64 Options. (line 6)
33040 * x86-64 Options: i386 and x86-64 Options.
33042 * Xstormy16 Options: Xstormy16 Options. (line 6)
33043 * Xtensa Options: Xtensa Options. (line 6)
33044 * y0: Other Builtins. (line 6)
33045 * y0f: Other Builtins. (line 6)
33046 * y0l: Other Builtins. (line 6)
33047 * y1: Other Builtins. (line 6)
33048 * y1f: Other Builtins. (line 6)
33049 * y1l: Other Builtins. (line 6)
33050 * yn: Other Builtins. (line 6)
33051 * ynf: Other Builtins. (line 6)
33052 * ynl: Other Builtins. (line 6)
33053 * zero-length arrays: Zero Length. (line 6)
33054 * zero-size structures: Empty Structures. (line 6)
33055 * zSeries options: zSeries Options. (line 6)
33061 Node: G++ and GCC
\7f3745
33062 Node: Standards
\7f5810
33063 Node: Invoking GCC
\7f12938
33064 Node: Option Summary
\7f16699
33065 Node: Overall Options
\7f45578
33066 Node: Invoking G++
\7f54804
33067 Node: C Dialect Options
\7f56283
33068 Node: C++ Dialect Options
\7f68696
33069 Node: Objective-C and Objective-C++ Dialect Options
\7f87746
33070 Node: Language Independent Options
\7f99342
33071 Node: Warning Options
\7f101424
33072 Node: Debugging Options
\7f154290
33073 Node: Optimize Options
\7f189446
33074 Node: Preprocessor Options
\7f269720
33075 Ref: Wtrigraphs
\7f273684
33076 Ref: dashMF
\7f278441
33077 Ref: fdollars-in-identifiers
\7f288296
33078 Node: Assembler Options
\7f296352
33079 Node: Link Options
\7f297057
33080 Ref: Link Options-Footnote-1
\7f305625
33081 Node: Directory Options
\7f305959
33082 Node: Spec Files
\7f312021
33083 Node: Target Options
\7f331327
33084 Node: Submodel Options
\7f332751
33085 Node: ARC Options
\7f334381
33086 Node: ARM Options
\7f335571
33087 Node: AVR Options
\7f348527
33088 Node: Blackfin Options
\7f350660
33089 Node: CRIS Options
\7f353428
33090 Node: CRX Options
\7f357647
33091 Node: Darwin Options
\7f358072
33092 Node: DEC Alpha Options
\7f365025
33093 Node: DEC Alpha/VMS Options
\7f376502
33094 Node: FRV Options
\7f376887
33095 Node: GNU/Linux Options
\7f383557
33096 Node: H8/300 Options
\7f384015
33097 Node: HPPA Options
\7f385082
33098 Node: i386 and x86-64 Options
\7f394675
33099 Node: IA-64 Options
\7f416120
33100 Node: M32C Options
\7f423437
33101 Node: M32R/D Options
\7f424728
33102 Node: M680x0 Options
\7f428315
33103 Node: M68hc1x Options
\7f435692
33104 Node: MCore Options
\7f437260
33105 Node: MIPS Options
\7f438281
33106 Node: MMIX Options
\7f453364
33107 Node: MN10300 Options
\7f455846
33108 Node: MT Options
\7f457264
33109 Node: PDP-11 Options
\7f458178
33110 Node: PowerPC Options
\7f460012
33111 Node: RS/6000 and PowerPC Options
\7f460246
33112 Node: S/390 and zSeries Options
\7f488915
33113 Node: Score Options
\7f496230
33114 Node: SH Options
\7f497058
33115 Node: SPARC Options
\7f506282
33116 Node: System V Options
\7f517125
33117 Node: TMS320C3x/C4x Options
\7f517959
33118 Node: V850 Options
\7f523484
33119 Node: VAX Options
\7f526629
33120 Node: x86-64 Options
\7f527176
33121 Node: Xstormy16 Options
\7f527390
33122 Node: Xtensa Options
\7f527679
33123 Node: zSeries Options
\7f531519
33124 Node: Code Gen Options
\7f531715
33125 Node: Environment Variables
\7f553772
33126 Node: Precompiled Headers
\7f561444
33127 Node: Running Protoize
\7f567681
33128 Node: C Implementation
\7f574018
33129 Node: Translation implementation
\7f575681
33130 Node: Environment implementation
\7f576255
33131 Node: Identifiers implementation
\7f576805
33132 Node: Characters implementation
\7f577859
33133 Node: Integers implementation
\7f580665
33134 Node: Floating point implementation
\7f582490
33135 Node: Arrays and pointers implementation
\7f585419
33136 Ref: Arrays and pointers implementation-Footnote-1
\7f586854
33137 Node: Hints implementation
\7f586978
33138 Node: Structures unions enumerations and bit-fields implementation
\7f588444
33139 Node: Qualifiers implementation
\7f590407
33140 Node: Declarators implementation
\7f592179
33141 Node: Statements implementation
\7f592521
33142 Node: Preprocessing directives implementation
\7f592848
33143 Node: Library functions implementation
\7f594953
33144 Node: Architecture implementation
\7f595593
33145 Node: Locale-specific behavior implementation
\7f596296
33146 Node: C Extensions
\7f596601
33147 Node: Statement Exprs
\7f600999
33148 Node: Local Labels
\7f605512
33149 Node: Labels as Values
\7f608491
33150 Ref: Labels as Values-Footnote-1
\7f610545
33151 Node: Nested Functions
\7f610728
33152 Node: Constructing Calls
\7f614622
33153 Node: Typeof
\7f616958
33154 Node: Conditionals
\7f620124
33155 Node: Long Long
\7f621015
33156 Node: Complex
\7f622516
33157 Node: Decimal Float
\7f625085
33158 Node: Hex Floats
\7f626766
33159 Node: Zero Length
\7f627807
33160 Node: Empty Structures
\7f631084
33161 Node: Variable Length
\7f631500
33162 Node: Variadic Macros
\7f634267
33163 Node: Escaped Newlines
\7f636649
33164 Node: Subscripting
\7f637488
33165 Node: Pointer Arith
\7f638211
33166 Node: Initializers
\7f638779
33167 Node: Compound Literals
\7f639275
33168 Node: Designated Inits
\7f641450
33169 Node: Case Ranges
\7f645105
33170 Node: Cast to Union
\7f645788
33171 Node: Mixed Declarations
\7f646884
33172 Node: Function Attributes
\7f647390
33173 Node: Attribute Syntax
\7f691235
33174 Node: Function Prototypes
\7f702106
33175 Node: C++ Comments
\7f703887
33176 Node: Dollar Signs
\7f704406
33177 Node: Character Escapes
\7f704871
33178 Node: Alignment
\7f705165
33179 Node: Variable Attributes
\7f706482
33180 Ref: i386 Variable Attributes
\7f719505
33181 Node: Type Attributes
\7f725002
33182 Ref: i386 Type Attributes
\7f738304
33183 Ref: PowerPC Type Attributes
\7f739148
33184 Node: Inline
\7f740001
33185 Node: Extended Asm
\7f745333
33186 Ref: Example of asm with clobbered asm reg
\7f751419
33187 Node: Constraints
\7f765515
33188 Node: Simple Constraints
\7f766365
33189 Node: Multi-Alternative
\7f772892
33190 Node: Modifiers
\7f774609
33191 Node: Machine Constraints
\7f777503
33192 Node: Asm Labels
\7f804750
33193 Node: Explicit Reg Vars
\7f806426
33194 Node: Global Reg Vars
\7f808034
33195 Node: Local Reg Vars
\7f812584
33196 Node: Alternate Keywords
\7f815025
33197 Node: Incomplete Enums
\7f816453
33198 Node: Function Names
\7f817210
33199 Node: Return Address
\7f819400
33200 Node: Vector Extensions
\7f822197
33201 Node: Offsetof
\7f825699
33202 Node: Atomic Builtins
\7f826485
33203 Node: Object Size Checking
\7f831570
33204 Node: Other Builtins
\7f836927
33205 Node: Target Builtins
\7f859015
33206 Node: Alpha Built-in Functions
\7f859748
33207 Node: ARM Built-in Functions
\7f862740
33208 Node: Blackfin Built-in Functions
\7f869447
33209 Node: FR-V Built-in Functions
\7f870064
33210 Node: Argument Types
\7f870923
33211 Node: Directly-mapped Integer Functions
\7f872679
33212 Node: Directly-mapped Media Functions
\7f873761
33213 Node: Raw read/write Functions
\7f880793
33214 Node: Other Built-in Functions
\7f881705
33215 Node: X86 Built-in Functions
\7f882894
33216 Node: MIPS DSP Built-in Functions
\7f901017
33217 Node: MIPS Paired-Single Support
\7f909442
33218 Node: Paired-Single Arithmetic
\7f911052
33219 Node: Paired-Single Built-in Functions
\7f911992
33220 Node: MIPS-3D Built-in Functions
\7f914656
33221 Node: PowerPC AltiVec Built-in Functions
\7f920025
33222 Node: SPARC VIS Built-in Functions
\7f1021329
33223 Node: Target Format Checks
\7f1022988
33224 Node: Solaris Format Checks
\7f1023395
33225 Node: Pragmas
\7f1023792
33226 Node: ARM Pragmas
\7f1024422
33227 Node: M32C Pragmas
\7f1025025
33228 Node: RS/6000 and PowerPC Pragmas
\7f1025601
33229 Node: Darwin Pragmas
\7f1026343
33230 Node: Solaris Pragmas
\7f1027410
33231 Node: Symbol-Renaming Pragmas
\7f1028571
33232 Node: Structure-Packing Pragmas
\7f1031193
33233 Node: Weak Pragmas
\7f1032824
33234 Node: Diagnostic Pragmas
\7f1033626
33235 Node: Visibility Pragmas
\7f1035619
33236 Node: Unnamed Fields
\7f1036340
33237 Node: Thread-Local
\7f1037850
33238 Node: C99 Thread-Local Edits
\7f1039934
33239 Node: C++98 Thread-Local Edits
\7f1041946
33240 Node: C++ Extensions
\7f1045391
33241 Node: Volatiles
\7f1046967
33242 Node: Restricted Pointers
\7f1049643
33243 Node: Vague Linkage
\7f1051237
33244 Node: C++ Interface
\7f1054893
33245 Ref: C++ Interface-Footnote-1
\7f1059190
33246 Node: Template Instantiation
\7f1059327
33247 Node: Bound member functions
\7f1066339
33248 Node: C++ Attributes
\7f1067882
33249 Node: Namespace Association
\7f1069540
33250 Node: Java Exceptions
\7f1070958
33251 Node: Deprecated Features
\7f1072363
33252 Node: Backwards Compatibility
\7f1075338
33253 Node: Objective-C
\7f1076693
33254 Node: Executing code before main
\7f1077274
33255 Node: What you can and what you cannot do in +load
\7f1079880
33256 Node: Type encoding
\7f1082047
33257 Node: Garbage Collection
\7f1085434
33258 Node: Constant string objects
\7f1088058
33259 Node: compatibility_alias
\7f1090566
33260 Node: Compatibility
\7f1091444
33261 Node: Gcov
\7f1098011
33262 Node: Gcov Intro
\7f1098535
33263 Node: Invoking Gcov
\7f1101251
33264 Node: Gcov and Optimization
\7f1113111
33265 Node: Gcov Data Files
\7f1115764
33266 Node: Cross-profiling
\7f1116902
33267 Node: Trouble
\7f1118728
33268 Node: Actual Bugs
\7f1120268
33269 Node: Cross-Compiler Problems
\7f1121008
33270 Node: Interoperation
\7f1121422
33271 Node: Incompatibilities
\7f1129020
33272 Node: Fixed Headers
\7f1137170
33273 Node: Standard Libraries
\7f1138833
33274 Node: Disappointments
\7f1140205
33275 Node: C++ Misunderstandings
\7f1144563
33276 Node: Static Definitions
\7f1145382
33277 Node: Name lookup
\7f1146435
33278 Ref: Name lookup-Footnote-1
\7f1151213
33279 Node: Temporaries
\7f1151400
33280 Node: Copy Assignment
\7f1153376
33281 Node: Protoize Caveats
\7f1155183
33282 Node: Non-bugs
\7f1159145
33283 Node: Warnings and Errors
\7f1169649
33284 Node: Bugs
\7f1171413
33285 Node: Bug Criteria
\7f1171977
33286 Node: Bug Reporting
\7f1174187
33287 Node: Service
\7f1174579
33288 Node: Contributing
\7f1175398
33289 Node: Funding
\7f1176138
33290 Node: GNU Project
\7f1178627
33291 Node: Copying
\7f1179273
33292 Node: GNU Free Documentation License
\7f1198450
33293 Node: Contributors
\7f1220856
33294 Node: Option Index
\7f1256712
33295 Node: Keyword Index
\7f1393229