1 @c Copyright (C) 1988,89,92,93,94,95,96,97,1998 Free Software Foundation, Inc.
2 @c This is part of the GCC manual.
3 @c For copying conditions, see the file gcc.texi.
6 @chapter GNU CC Command Options
7 @cindex GNU CC command options
8 @cindex command options
9 @cindex options, GNU CC command
11 When you invoke GNU CC, it normally does preprocessing, compilation,
12 assembly and linking. The ``overall options'' allow you to stop this
13 process at an intermediate stage. For example, the @samp{-c} option
14 says not to run the linker. Then the output consists of object files
15 output by the assembler.
17 Other options are passed on to one stage of processing. Some options
18 control the preprocessor and others the compiler itself. Yet other
19 options control the assembler and linker; most of these are not
20 documented here, since you rarely need to use any of them.
22 @cindex C compilation options
23 Most of the command line options that you can use with GNU CC are useful
24 for C programs; when an option is only useful with another language
25 (usually C++), the explanation says so explicitly. If the description
26 for a particular option does not mention a source language, you can use
27 that option with all supported languages.
29 @cindex C++ compilation options
30 @xref{Invoking G++,,Compiling C++ Programs}, for a summary of special
31 options for compiling C++ programs.
33 @cindex grouping options
34 @cindex options, grouping
35 The @code{gcc} program accepts options and file names as operands. Many
36 options have multiletter names; therefore multiple single-letter options
37 may @emph{not} be grouped: @samp{-dr} is very different from @w{@samp{-d
40 @cindex order of options
41 @cindex options, order
42 You can mix options and other arguments. For the most part, the order
43 you use doesn't matter. Order does matter when you use several options
44 of the same kind; for example, if you specify @samp{-L} more than once,
45 the directories are searched in the order specified.
47 Many options have long names starting with @samp{-f} or with
48 @samp{-W}---for example, @samp{-fforce-mem},
49 @samp{-fstrength-reduce}, @samp{-Wformat} and so on. Most of
50 these have both positive and negative forms; the negative form of
51 @samp{-ffoo} would be @samp{-fno-foo}. This manual documents
52 only one of these two forms, whichever one is not the default.
55 * Option Summary:: Brief list of all options, without explanations.
56 * Overall Options:: Controlling the kind of output:
57 an executable, object files, assembler files,
58 or preprocessed source.
59 * Invoking G++:: Compiling C++ programs.
60 * C Dialect Options:: Controlling the variant of C language compiled.
61 * C++ Dialect Options:: Variations on C++.
62 * Warning Options:: How picky should the compiler be?
63 * Debugging Options:: Symbol tables, measurements, and debugging dumps.
64 * Optimize Options:: How much optimization?
65 * Preprocessor Options:: Controlling header files and macro definitions.
66 Also, getting dependency information for Make.
67 * Assembler Options:: Passing options to the assembler.
68 * Link Options:: Specifying libraries and so on.
69 * Directory Options:: Where to find header files and libraries.
70 Where to find the compiler executable files.
71 * Target Options:: Running a cross-compiler, or an old version of GNU CC.
72 * Submodel Options:: Specifying minor hardware or convention variations,
73 such as 68010 vs 68020.
74 * Code Gen Options:: Specifying conventions for function calls, data layout
76 * Environment Variables:: Env vars that affect GNU CC.
77 * Running Protoize:: Automatically adding or removing function prototypes.
81 @section Option Summary
83 Here is a summary of all the options, grouped by type. Explanations are
84 in the following sections.
88 @xref{Overall Options,,Options Controlling the Kind of Output}.
90 -c -S -E -o @var{file} -pipe -v -x @var{language}
93 @item C Language Options
94 @xref{C Dialect Options,,Options Controlling C Dialect}.
96 -ansi -fallow-single-precision -fcond-mismatch -fno-asm
97 -fno-builtin -ffreestanding -fhosted -fsigned-bitfields -fsigned-char
98 -funsigned-bitfields -funsigned-char -fwritable-strings
99 -traditional -traditional-cpp -trigraphs
102 @item C++ Language Options
103 @xref{C++ Dialect Options,,Options Controlling C++ Dialect}.
105 -fall-virtual -fdollars-in-identifiers -felide-constructors
106 -fenum-int-equiv -fexternal-templates -ffor-scope
107 -fno-for-scope -fhandle-signatures -fmemoize-lookups
108 -fname-mangling-version-@var{n} -fno-default-inline
109 -fno-gnu-keywords -fnonnull-objects -fguiding-decls
110 -foperator-names -fstrict-prototype -fthis-is-variable
111 -ftemplate-depth-@var{n} -nostdinc++ -traditional +e@var{n}
114 @item Warning Options
115 @xref{Warning Options,,Options to Request or Suppress Warnings}.
117 -fsyntax-only -pedantic -pedantic-errors
118 -w -W -Wall -Waggregate-return -Wbad-function-cast
119 -Wcast-align -Wcast-qual -Wchar-subscript -Wcomment
120 -Wconversion -Werror -Wformat
121 -Wid-clash-@var{len} -Wimplicit -Wimplicit-int
122 -Wimplicit-function-declarations -Wimport -Winline
123 -Wlarger-than-@var{len} -Wmain -Wmissing-declarations
124 -Wmissing-prototypes -Wnested-externs -Wno-import
125 -Wold-style-cast -Woverloaded-virtual -Wparentheses
126 -Wpointer-arith -Wredundant-decls -Wreorder -Wreturn-type
127 -Wshadow -Wsign-compare -Wstrict-prototypes -Wswitch
128 -Wsynth -Wtemplate-debugging -Wtraditional -Wtrigraphs
129 -Wundef -Wuninitialized -Wunused -Wwrite-strings
133 @item Debugging Options
134 @xref{Debugging Options,,Options for Debugging Your Program or GCC}.
136 -a -ax -d@var{letters} -fpretend-float
137 -fprofile-arcs -ftest-coverage
138 -g -g@var{level} -gcoff -gdwarf -gdwarf-1 -gdwarf-1+ -gdwarf-2
139 -ggdb -gstabs -gstabs+ -gxcoff -gxcoff+
140 -p -pg -print-file-name=@var{library} -print-libgcc-file-name
141 -print-prog-name=@var{program} -print-search-dirs -save-temps
144 @item Optimization Options
145 @xref{Optimize Options,,Options that Control Optimization}.
147 -fbranch-probabilities
148 -fcaller-saves -fcse-follow-jumps -fcse-skip-blocks
149 -fdelayed-branch -fexpensive-optimizations
150 -ffast-math -ffloat-store -fforce-addr -fforce-mem
151 -ffunction-sections -finline-functions
152 -fkeep-inline-functions -fno-default-inline
153 -fno-defer-pop -fno-function-cse
154 -fno-inline -fno-peephole -fomit-frame-pointer -fregmove
155 -frerun-cse-after-loop -frerun-loop-opt -fschedule-insns
156 -fschedule-insns2 -fstrength-reduce -fthread-jumps
157 -funroll-all-loops -funroll-loops
158 -fmove-all-movables -freduce-all-givs
159 -O -O0 -O1 -O2 -O3 -Os
162 @item Preprocessor Options
163 @xref{Preprocessor Options,,Options Controlling the Preprocessor}.
165 -A@var{question}(@var{answer}) -C -dD -dM -dN
166 -D@var{macro}@r{[}=@var{defn}@r{]} -E -H
168 -include @var{file} -imacros @var{file}
169 -iprefix @var{file} -iwithprefix @var{dir}
170 -iwithprefixbefore @var{dir} -isystem @var{dir}
171 -M -MD -MM -MMD -MG -nostdinc -P -trigraphs
172 -undef -U@var{macro} -Wp,@var{option}
175 @item Assembler Option
176 @xref{Assembler Options,,Passing Options to the Assembler}.
182 @xref{Link Options,,Options for Linking}.
184 @var{object-file-name} -l@var{library}
185 -nostartfiles -nodefaultlibs -nostdlib
186 -s -static -shared -symbolic
187 -Wl,@var{option} -Xlinker @var{option}
191 @item Directory Options
192 @xref{Directory Options,,Options for Directory Search}.
194 -B@var{prefix} -I@var{dir} -I- -L@var{dir} -specs=@var{file}
198 @c I wrote this xref this way to avoid overfull hbox. -- rms
199 @xref{Target Options}.
201 -b @var{machine} -V @var{version}
204 @item Machine Dependent Options
205 @xref{Submodel Options,,Hardware Models and Configurations}.
207 @emph{M680x0 Options}
208 -m68000 -m68020 -m68020-40 -m68020-60 -m68030 -m68040
209 -m68060 -mcpu32 -m5200 -m68881 -mbitfield -mc68000 -mc68020
210 -mfpa -mnobitfield -mrtd -mshort -msoft-float
218 -mtune=@var{cpu type}
219 -mcmodel=@var{code model}
220 -malign-jumps=@var{num} -malign-loops=@var{num}
221 -malign-functions=@var{num}
223 -mapp-regs -mbroken-saverestore -mcypress -mepilogue
224 -mflat -mfpu -mhard-float -mhard-quad-float
225 -mimpure-text -mlive-g0 -mno-app-regs -mno-epilogue
226 -mno-flat -mno-fpu -mno-impure-text
227 -mno-stack-bias -mno-unaligned-doubles
228 -msoft-float -msoft-quad-float -msparclite -mstack-bias
229 -msupersparc -munaligned-doubles -mv8
231 @emph{Convex Options}
232 -mc1 -mc2 -mc32 -mc34 -mc38
233 -margcount -mnoargcount
235 -mvolatile-cache -mvolatile-nocache
237 @emph{AMD29K Options}
238 -m29000 -m29050 -mbw -mnbw -mdw -mndw
239 -mlarge -mnormal -msmall
240 -mkernel-registers -mno-reuse-arg-regs
241 -mno-stack-check -mno-storem-bug
242 -mreuse-arg-regs -msoft-float -mstack-check
243 -mstorem-bug -muser-registers
246 -mapcs-frame -mno-apcs-frame
248 -mapcs-stack-check -mno-apcs-stack-check
249 -mapcs-float -mno-apcs-float
250 -mapcs-reentrant -mno-apcs-reentrant
251 -msched-prolog -mno-sched-prolog
252 -mlittle-endian -mbig-endian -mwords-little-endian
253 -mshort-load-bytes -mno-short-load-bytes -mshort-load-words -mno-short-load-words
254 -msoft-float -mhard-float -mfpe
255 -mthumb-interwork -mno-thumb-interwork
256 -mcpu= -march= -mfpe=
257 -mstructure-size-boundary=
258 -mbsd -mxopen -mno-symrename
261 -mtpcs-frame -mno-tpcs-frame
262 -mtpcs-leaf-frame -mno-tpcs-leaf-frame
263 -mlittle-endian -mbig-endian
264 -mthumb-interwork -mno-thumb-interwork
265 -mstructure-size-boundary=
267 @emph{MN10300 Options}
271 @emph{M32R/D Options}
272 -mcode-model=@var{model type} -msdata=@var{sdata type}
276 -m88000 -m88100 -m88110 -mbig-pic
277 -mcheck-zero-division -mhandle-large-shift
278 -midentify-revision -mno-check-zero-division
279 -mno-ocs-debug-info -mno-ocs-frame-position
280 -mno-optimize-arg-area -mno-serialize-volatile
281 -mno-underscores -mocs-debug-info
282 -mocs-frame-position -moptimize-arg-area
283 -mserialize-volatile -mshort-data-@var{num} -msvr3
284 -msvr4 -mtrap-large-shift -muse-div-instruction
285 -mversion-03.00 -mwarn-passed-structs
287 @emph{RS/6000 and PowerPC Options}
289 -mtune=@var{cpu type}
290 -mpower -mno-power -mpower2 -mno-power2
291 -mpowerpc -mno-powerpc
292 -mpowerpc-gpopt -mno-powerpc-gpopt
293 -mpowerpc-gfxopt -mno-powerpc-gfxopt
294 -mnew-mnemonics -mno-new-mnemonics
295 -mfull-toc -mminimal-toc -mno-fop-in-toc -mno-sum-in-toc
296 -mxl-call -mno-xl-call -mthreads -mpe
297 -msoft-float -mhard-float -mmultiple -mno-multiple
298 -mstring -mno-string -mupdate -mno-update
299 -mfused-madd -mno-fused-madd -mbit-align -mno-bit-align
300 -mstrict-align -mno-strict-align -mrelocatable
301 -mno-relocatable -mrelocatable-lib -mno-relocatable-lib
302 -mtoc -mno-toc -mtraceback -mno-traceback
303 -mlittle -mlittle-endian -mbig -mbig-endian
304 -mcall-aix -mcall-sysv -mprototype -mno-prototype
305 -msim -mmvme -mads -myellowknife -memb
306 -msdata -msdata=@var{opt} -G @var{num}
309 -mcall-lib-mul -mfp-arg-in-fpregs -mfp-arg-in-gregs
310 -mfull-fp-blocks -mhc-struct-return -min-line-mul
311 -mminimum-fp-blocks -mnohc-struct-return
314 -mabicalls -mcpu=@var{cpu type} -membedded-data
315 -membedded-pic -mfp32 -mfp64 -mgas -mgp32 -mgp64
316 -mgpopt -mhalf-pic -mhard-float -mint64 -mips1
317 -mips2 -mips3 -mlong64 -mlong-calls -mmemcpy
318 -mmips-as -mmips-tfile -mno-abicalls
319 -mno-embedded-data -mno-embedded-pic
320 -mno-gpopt -mno-long-calls
321 -mno-memcpy -mno-mips-tfile -mno-rnames -mno-stats
322 -mrnames -msoft-float
323 -m4650 -msingle-float -mmad
324 -mstats -EL -EB -G @var{num} -nocpp
328 -march=@var{cpu type}
329 -mieee-fp -mno-fancy-math-387
330 -mno-fp-ret-in-387 -msoft-float -msvr3-shlib
331 -mno-wide-multiply -mrtd -malign-double
332 -mreg-alloc=@var{list} -mregparm=@var{num}
333 -malign-jumps=@var{num} -malign-loops=@var{num}
334 -malign-functions=@var{num}
337 -mbig-switch -mdisable-fpregs -mdisable-indexing
338 -mfast-indirect-calls -mgas -mjump-in-delay
339 -mlong-load-store -mno-big-switch -mno-disable-fpregs
340 -mno-disable-indexing -mno-fast-indirect-calls -mno-gas
341 -mno-jump-in-delay -mno-long-load-store
342 -mno-portable-runtime -mno-soft-float -mno-space
343 -mno-space-regs -msoft-float -mpa-risc-1-0
344 -mpa-risc-1-1 -mportable-runtime
345 -mschedule=@var{list} -mspace -mspace-regs
347 @emph{Intel 960 Options}
348 -m@var{cpu type} -masm-compat -mclean-linkage
349 -mcode-align -mcomplex-addr -mleaf-procedures
350 -mic-compat -mic2.0-compat -mic3.0-compat
351 -mintel-asm -mno-clean-linkage -mno-code-align
352 -mno-complex-addr -mno-leaf-procedures
353 -mno-old-align -mno-strict-align -mno-tail-call
354 -mnumerics -mold-align -msoft-float -mstrict-align
357 @emph{DEC Alpha Options}
358 -mfp-regs -mno-fp-regs -mno-soft-float -msoft-float
360 -mieee -mieee-with-inexact -mieee-conformant
361 -mfp-trap-mode=@var{mode} -mfp-rounding-mode=@var{mode}
362 -mtrap-precision=@var{mode} -mbuild-constants
364 -mbwx -mno-bwx -mcix -mno-cix -mmax -mno-max
365 -mmemory-latency=@var{time}
367 @emph{Clipper Options}
370 @emph{H8/300 Options}
371 -mrelax -mh -ms -mint32 -malign-300
374 -m1 -m2 -m3 -m3e -mb -ml -mrelax
376 @emph{System V Options}
377 -Qy -Qn -YP,@var{paths} -Ym,@var{dir}
381 -mmangle-cpu -mcpu=@var{cpu} -mtext=@var{text section}
382 -mdata=@var{data section} -mrodata=@var{readonly data section}
385 -mlong-calls -mno-long-calls -mep -mno-ep
386 -mprolog-function -mno-prolog-function -mspace
387 -mtda=@var{n} -msda=@var{n} -mzda=@var{n}
391 @item Code Generation Options
392 @xref{Code Gen Options,,Options for Code Generation Conventions}.
394 -fcall-saved-@var{reg} -fcall-used-@var{reg}
395 -fexceptions -ffixed-@var{reg} -finhibit-size-directive
396 -fcheck-memory-usage -fprefix-function-name
397 -fno-common -fno-ident -fno-gnu-linker
398 -fpcc-struct-return -fpic -fPIC
399 -freg-struct-return -fshared-data -fshort-enums
400 -fshort-double -fvolatile -fvolatile-global
401 -fverbose-asm -fpack-struct -fstack-check +e0 +e1
402 -fargument-alias -fargument-noalias
403 -fargument-noalias-global
408 * Overall Options:: Controlling the kind of output:
409 an executable, object files, assembler files,
410 or preprocessed source.
411 * C Dialect Options:: Controlling the variant of C language compiled.
412 * C++ Dialect Options:: Variations on C++.
413 * Warning Options:: How picky should the compiler be?
414 * Debugging Options:: Symbol tables, measurements, and debugging dumps.
415 * Optimize Options:: How much optimization?
416 * Preprocessor Options:: Controlling header files and macro definitions.
417 Also, getting dependency information for Make.
418 * Assembler Options:: Passing options to the assembler.
419 * Link Options:: Specifying libraries and so on.
420 * Directory Options:: Where to find header files and libraries.
421 Where to find the compiler executable files.
422 * Target Options:: Running a cross-compiler, or an old version of GNU CC.
425 @node Overall Options
426 @section Options Controlling the Kind of Output
428 Compilation can involve up to four stages: preprocessing, compilation
429 proper, assembly and linking, always in that order. The first three
430 stages apply to an individual source file, and end by producing an
431 object file; linking combines all the object files (those newly
432 compiled, and those specified as input) into an executable file.
434 @cindex file name suffix
435 For any given input file, the file name suffix determines what kind of
440 C source code which must be preprocessed.
443 C source code which should not be preprocessed.
446 C++ source code which should not be preprocessed.
449 Objective-C source code. Note that you must link with the library
450 @file{libobjc.a} to make an Objective-C program work.
453 C header file (not to be compiled or linked).
456 @itemx @var{file}.cxx
457 @itemx @var{file}.cpp
459 C++ source code which must be preprocessed. Note that in @samp{.cxx},
460 the last two letters must both be literally @samp{x}. Likewise,
461 @samp{.C} refers to a literal capital C.
467 Assembler code which must be preprocessed.
470 An object file to be fed straight into linking.
471 Any file name with no recognized suffix is treated this way.
474 You can specify the input language explicitly with the @samp{-x} option:
477 @item -x @var{language}
478 Specify explicitly the @var{language} for the following input files
479 (rather than letting the compiler choose a default based on the file
480 name suffix). This option applies to all following input files until
481 the next @samp{-x} option. Possible values for @var{language} are:
484 c-header cpp-output c++-cpp-output
485 assembler assembler-with-cpp
489 Turn off any specification of a language, so that subsequent files are
490 handled according to their file name suffixes (as they are if @samp{-x}
491 has not been used at all).
494 If you only want some of the stages of compilation, you can use
495 @samp{-x} (or filename suffixes) to tell @code{gcc} where to start, and
496 one of the options @samp{-c}, @samp{-S}, or @samp{-E} to say where
497 @code{gcc} is to stop. Note that some combinations (for example,
498 @samp{-x cpp-output -E} instruct @code{gcc} to do nothing at all.
502 Compile or assemble the source files, but do not link. The linking
503 stage simply is not done. The ultimate output is in the form of an
504 object file for each source file.
506 By default, the object file name for a source file is made by replacing
507 the suffix @samp{.c}, @samp{.i}, @samp{.s}, etc., with @samp{.o}.
509 Unrecognized input files, not requiring compilation or assembly, are
513 Stop after the stage of compilation proper; do not assemble. The output
514 is in the form of an assembler code file for each non-assembler input
517 By default, the assembler file name for a source file is made by
518 replacing the suffix @samp{.c}, @samp{.i}, etc., with @samp{.s}.
520 Input files that don't require compilation are ignored.
523 Stop after the preprocessing stage; do not run the compiler proper. The
524 output is in the form of preprocessed source code, which is sent to the
527 Input files which don't require preprocessing are ignored.
529 @cindex output file option
531 Place output in file @var{file}. This applies regardless to whatever
532 sort of output is being produced, whether it be an executable file,
533 an object file, an assembler file or preprocessed C code.
535 Since only one output file can be specified, it does not make sense to
536 use @samp{-o} when compiling more than one input file, unless you are
537 producing an executable file as output.
539 If @samp{-o} is not specified, the default is to put an executable file
540 in @file{a.out}, the object file for @file{@var{source}.@var{suffix}} in
541 @file{@var{source}.o}, its assembler file in @file{@var{source}.s}, and
542 all preprocessed C source on standard output.@refill
545 Print (on standard error output) the commands executed to run the stages
546 of compilation. Also print the version number of the compiler driver
547 program and of the preprocessor and the compiler proper.
550 Use pipes rather than temporary files for communication between the
551 various stages of compilation. This fails to work on some systems where
552 the assembler is unable to read from a pipe; but the GNU assembler has
557 @section Compiling C++ Programs
559 @cindex suffixes for C++ source
560 @cindex C++ source file suffixes
561 C++ source files conventionally use one of the suffixes @samp{.C},
562 @samp{.cc}, @samp{cpp}, or @samp{.cxx}; preprocessed C++ files use the
563 suffix @samp{.ii}. GNU CC recognizes files with these names and
564 compiles them as C++ programs even if you call the compiler the same way
565 as for compiling C programs (usually with the name @code{gcc}).
569 However, C++ programs often require class libraries as well as a
570 compiler that understands the C++ language---and under some
571 circumstances, you might want to compile programs from standard input,
572 or otherwise without a suffix that flags them as C++ programs.
573 @code{g++} is a program that calls GNU CC with the default language
574 set to C++, and automatically specifies linking against the C++
576 @cindex @code{g++ 1.@var{xx}}
577 @cindex @code{g++}, separate compiler
578 @cindex @code{g++} older version
579 @footnote{Prior to release 2 of the compiler,
580 there was a separate @code{g++} compiler. That version was based on GNU
581 CC, but not integrated with it. Versions of @code{g++} with a
582 @samp{1.@var{xx}} version number---for example, @code{g++} version 1.37
583 or 1.42---are much less reliable than the versions integrated with GCC
584 2. Moreover, combining G++ @samp{1.@var{xx}} with a version 2 GCC will
585 simply not work.} On many systems, the script @code{g++} is also
586 installed with the name @code{c++}.
588 @cindex invoking @code{g++}
589 When you compile C++ programs, you may specify many of the same
590 command-line options that you use for compiling programs in any
591 language; or command-line options meaningful for C and related
592 languages; or options that are meaningful only for C++ programs.
593 @xref{C Dialect Options,,Options Controlling C Dialect}, for
594 explanations of options for languages related to C.
595 @xref{C++ Dialect Options,,Options Controlling C++ Dialect}, for
596 explanations of options that are meaningful only for C++ programs.
598 @node C Dialect Options
599 @section Options Controlling C Dialect
600 @cindex dialect options
601 @cindex language dialect options
602 @cindex options, dialect
604 The following options control the dialect of C (or languages derived
605 from C, such as C++ and Objective C) that the compiler accepts:
610 Support all ANSI standard C programs.
612 This turns off certain features of GNU C that are incompatible with ANSI
613 C, such as the @code{asm}, @code{inline} and @code{typeof} keywords, and
614 predefined macros such as @code{unix} and @code{vax} that identify the
615 type of system you are using. It also enables the undesirable and
616 rarely used ANSI trigraph feature, and it disables recognition of C++
617 style @samp{//} comments.
619 The alternate keywords @code{__asm__}, @code{__extension__},
620 @code{__inline__} and @code{__typeof__} continue to work despite
621 @samp{-ansi}. You would not want to use them in an ANSI C program, of
622 course, but it is useful to put them in header files that might be included
623 in compilations done with @samp{-ansi}. Alternate predefined macros
624 such as @code{__unix__} and @code{__vax__} are also available, with or
625 without @samp{-ansi}.
627 The @samp{-ansi} option does not cause non-ANSI programs to be
628 rejected gratuitously. For that, @samp{-pedantic} is required in
629 addition to @samp{-ansi}. @xref{Warning Options}.
631 The macro @code{__STRICT_ANSI__} is predefined when the @samp{-ansi}
632 option is used. Some header files may notice this macro and refrain
633 from declaring certain functions or defining certain macros that the
634 ANSI standard doesn't call for; this is to avoid interfering with any
635 programs that might use these names for other things.
637 The functions @code{alloca}, @code{abort}, @code{exit}, and
638 @code{_exit} are not builtin functions when @samp{-ansi} is used.
641 Do not recognize @code{asm}, @code{inline} or @code{typeof} as a
642 keyword, so that code can use these words as identifiers. You can use
643 the keywords @code{__asm__}, @code{__inline__} and @code{__typeof__}
644 instead. @samp{-ansi} implies @samp{-fno-asm}.
646 In C++, this switch only affects the @code{typeof} keyword, since
647 @code{asm} and @code{inline} are standard keywords. You may want to
648 use the @samp{-fno-gnu-keywords} flag instead, as it also disables the
649 other, C++-specific, extension keywords such as @code{headof}.
652 @cindex builtin functions
668 Don't recognize builtin functions that do not begin with two leading
669 underscores. Currently, the functions affected include @code{abort},
670 @code{abs}, @code{alloca}, @code{cos}, @code{exit}, @code{fabs},
671 @code{ffs}, @code{labs}, @code{memcmp}, @code{memcpy}, @code{sin},
672 @code{sqrt}, @code{strcmp}, @code{strcpy}, and @code{strlen}.
674 GCC normally generates special code to handle certain builtin functions
675 more efficiently; for instance, calls to @code{alloca} may become single
676 instructions that adjust the stack directly, and calls to @code{memcpy}
677 may become inline copy loops. The resulting code is often both smaller
678 and faster, but since the function calls no longer appear as such, you
679 cannot set a breakpoint on those calls, nor can you change the behavior
680 of the functions by linking with a different library.
682 The @samp{-ansi} option prevents @code{alloca} and @code{ffs} from being
683 builtin functions, since these functions do not have an ANSI standard
687 @cindex hosted environment
689 Assert that compilation takes place in a hosted environment. This implies
690 @samp{-fbuiltin}. A hosted environment is one in which the
691 entire standard library is available, and in which @code{main} has a return
692 type of @code{int}. Examples are nearly everything except a kernel.
693 This is equivalent to @samp{-fno-freestanding}.
696 @cindex hosted environment
698 Assert that compilation takes place in a freestanding environment. This
699 implies @samp{-fno-builtin}. A freestanding environment
700 is one in which the standard library may not exist, and program startup may
701 not necessarily be at @code{main}. The most obvious example is an OS kernel.
702 This is equivalent to @samp{-fno-hosted}.
705 Support ANSI C trigraphs. You don't want to know about this
706 brain-damage. The @samp{-ansi} option implies @samp{-trigraphs}.
708 @cindex traditional C language
709 @cindex C language, traditional
711 Attempt to support some aspects of traditional C compilers.
716 All @code{extern} declarations take effect globally even if they
717 are written inside of a function definition. This includes implicit
718 declarations of functions.
721 The newer keywords @code{typeof}, @code{inline}, @code{signed}, @code{const}
722 and @code{volatile} are not recognized. (You can still use the
723 alternative keywords such as @code{__typeof__}, @code{__inline__}, and
727 Comparisons between pointers and integers are always allowed.
730 Integer types @code{unsigned short} and @code{unsigned char} promote
731 to @code{unsigned int}.
734 Out-of-range floating point literals are not an error.
737 Certain constructs which ANSI regards as a single invalid preprocessing
738 number, such as @samp{0xe-0xd}, are treated as expressions instead.
741 String ``constants'' are not necessarily constant; they are stored in
742 writable space, and identical looking constants are allocated
743 separately. (This is the same as the effect of
744 @samp{-fwritable-strings}.)
746 @cindex @code{longjmp} and automatic variables
748 All automatic variables not declared @code{register} are preserved by
749 @code{longjmp}. Ordinarily, GNU C follows ANSI C: automatic variables
750 not declared @code{volatile} may be clobbered.
755 @cindex escape sequences, traditional
756 The character escape sequences @samp{\x} and @samp{\a} evaluate as the
757 literal characters @samp{x} and @samp{a} respectively. Without
758 @w{@samp{-traditional}}, @samp{\x} is a prefix for the hexadecimal
759 representation of a character, and @samp{\a} produces a bell.
762 In C++ programs, assignment to @code{this} is permitted with
763 @samp{-traditional}. (The option @samp{-fthis-is-variable} also has
767 You may wish to use @samp{-fno-builtin} as well as @samp{-traditional}
768 if your program uses names that are normally GNU C builtin functions for
769 other purposes of its own.
771 You cannot use @samp{-traditional} if you include any header files that
772 rely on ANSI C features. Some vendors are starting to ship systems with
773 ANSI C header files and you cannot use @samp{-traditional} on such
774 systems to compile files that include any system headers.
776 The @samp{-traditional} option also enables @samp{-traditional-cpp},
777 which is described next.
779 @item -traditional-cpp
780 Attempt to support some aspects of traditional C preprocessors.
785 Comments convert to nothing at all, rather than to a space. This allows
786 traditional token concatenation.
789 In a preprocessing directive, the @samp{#} symbol must appear as the first
793 Macro arguments are recognized within string constants in a macro
794 definition (and their values are stringified, though without additional
795 quote marks, when they appear in such a context). The preprocessor
796 always considers a string constant to end at a newline.
799 @cindex detecting @w{@samp{-traditional}}
800 The predefined macro @code{__STDC__} is not defined when you use
801 @samp{-traditional}, but @code{__GNUC__} is (since the GNU extensions
802 which @code{__GNUC__} indicates are not affected by
803 @samp{-traditional}). If you need to write header files that work
804 differently depending on whether @samp{-traditional} is in use, by
805 testing both of these predefined macros you can distinguish four
806 situations: GNU C, traditional GNU C, other ANSI C compilers, and other
807 old C compilers. The predefined macro @code{__STDC_VERSION__} is also
808 not defined when you use @samp{-traditional}. @xref{Standard
809 Predefined,,Standard Predefined Macros,cpp.info,The C Preprocessor},
810 for more discussion of these and other predefined macros.
813 @cindex string constants vs newline
814 @cindex newline vs string constants
815 The preprocessor considers a string constant to end at a newline (unless
816 the newline is escaped with @samp{\}). (Without @w{@samp{-traditional}},
817 string constants can contain the newline character as typed.)
820 @item -fcond-mismatch
821 Allow conditional expressions with mismatched types in the second and
822 third arguments. The value of such an expression is void.
824 @item -funsigned-char
825 Let the type @code{char} be unsigned, like @code{unsigned char}.
827 Each kind of machine has a default for what @code{char} should
828 be. It is either like @code{unsigned char} by default or like
829 @code{signed char} by default.
831 Ideally, a portable program should always use @code{signed char} or
832 @code{unsigned char} when it depends on the signedness of an object.
833 But many programs have been written to use plain @code{char} and
834 expect it to be signed, or expect it to be unsigned, depending on the
835 machines they were written for. This option, and its inverse, let you
836 make such a program work with the opposite default.
838 The type @code{char} is always a distinct type from each of
839 @code{signed char} or @code{unsigned char}, even though its behavior
840 is always just like one of those two.
843 Let the type @code{char} be signed, like @code{signed char}.
845 Note that this is equivalent to @samp{-fno-unsigned-char}, which is
846 the negative form of @samp{-funsigned-char}. Likewise, the option
847 @samp{-fno-signed-char} is equivalent to @samp{-funsigned-char}.
849 You may wish to use @samp{-fno-builtin} as well as @samp{-traditional}
850 if your program uses names that are normally GNU C builtin functions for
851 other purposes of its own.
853 You cannot use @samp{-traditional} if you include any header files that
854 rely on ANSI C features. Some vendors are starting to ship systems with
855 ANSI C header files and you cannot use @samp{-traditional} on such
856 systems to compile files that include any system headers.
858 @item -fsigned-bitfields
859 @itemx -funsigned-bitfields
860 @itemx -fno-signed-bitfields
861 @itemx -fno-unsigned-bitfields
862 These options control whether a bitfield is signed or unsigned, when the
863 declaration does not use either @code{signed} or @code{unsigned}. By
864 default, such a bitfield is signed, because this is consistent: the
865 basic integer types such as @code{int} are signed types.
867 However, when @samp{-traditional} is used, bitfields are all unsigned
870 @item -fwritable-strings
871 Store string constants in the writable data segment and don't uniquize
872 them. This is for compatibility with old programs which assume they can
873 write into string constants. The option @samp{-traditional} also has
876 Writing into string constants is a very bad idea; ``constants'' should
879 @item -fallow-single-precision
880 Do not promote single precision math operations to double precision,
881 even when compiling with @samp{-traditional}.
883 Traditional K&R C promotes all floating point operations to double
884 precision, regardless of the sizes of the operands. On the
885 architecture for which you are compiling, single precision may be faster
886 than double precision. If you must use @samp{-traditional}, but want
887 to use single precision operations when the operands are single
888 precision, use this option. This option has no effect when compiling
889 with ANSI or GNU C conventions (the default).
893 @node C++ Dialect Options
894 @section Options Controlling C++ Dialect
896 @cindex compiler options, C++
897 @cindex C++ options, command line
899 This section describes the command-line options that are only meaningful
900 for C++ programs; but you can also use most of the GNU compiler options
901 regardless of what language your program is in. For example, you
902 might compile a file @code{firstClass.C} like this:
905 g++ -g -felide-constructors -O -c firstClass.C
909 In this example, only @samp{-felide-constructors} is an option meant
910 only for C++ programs; you can use the other options with any
911 language supported by GNU CC.
913 Here is a list of options that are @emph{only} for compiling C++ programs:
916 @item -fno-access-control
917 Turn off all access checking. This switch is mainly useful for working
918 around bugs in the access control code.
921 Treat all possible member functions as virtual, implicitly.
922 All member functions (except for constructor functions and @code{new} or
923 @code{delete} member operators) are treated as virtual functions of the
924 class where they appear.
926 This does not mean that all calls to these member functions will be made
927 through the internal table of virtual functions. Under some
928 circumstances, the compiler can determine that a call to a given virtual
929 function can be made directly; in these cases the calls are direct in
933 Check that the pointer returned by @code{operator new} is non-null
934 before attempting to modify the storage allocated. The current Working
935 Paper requires that @code{operator new} never return a null pointer, so
936 this check is normally unnecessary.
938 @item -fconserve-space
939 Put uninitialized or runtime-initialized global variables into the
940 common segment, as C does. This saves space in the executable at the
941 cost of not diagnosing duplicate definitions. If you compile with this
942 flag and your program mysteriously crashes after @code{main()} has
943 completed, you may have an object that is being destroyed twice because
944 two definitions were merged.
946 @item -fdollars-in-identifiers
947 Accept @samp{$} in identifiers. You can also explicitly prohibit use of
948 @samp{$} with the option @samp{-fno-dollars-in-identifiers}. (GNU C allows
949 @samp{$} by default on most target systems, but there are a few exceptions.)
950 Traditional C allowed the character @samp{$} to form part of
951 identifiers. However, ANSI C and C++ forbid @samp{$} in identifiers.
953 @item -fenum-int-equiv
954 Anachronistically permit implicit conversion of @code{int} to
955 enumeration types. Current C++ allows conversion of @code{enum} to
956 @code{int}, but not the other way around.
958 @item -fexternal-templates
959 Cause template instantiations to obey @samp{#pragma interface} and
960 @samp{implementation}; template instances are emitted or not according
961 to the location of the template definition. @xref{Template
962 Instantiation}, for more information.
964 This option is deprecated.
966 @item -falt-external-templates
967 Similar to -fexternal-templates, but template instances are emitted or
968 not according to the place where they are first instantiated.
969 @xref{Template Instantiation}, for more information.
971 This option is deprecated.
974 @itemx -fno-for-scope
975 If -ffor-scope is specified, the scope of variables declared in
976 a @i{for-init-statement} is limited to the @samp{for} loop itself,
977 as specified by the draft C++ standard.
978 If -fno-for-scope is specified, the scope of variables declared in
979 a @i{for-init-statement} extends to the end of the enclosing scope,
980 as was the case in old versions of gcc, and other (traditional)
981 implementations of C++.
983 The default if neither flag is given to follow the standard,
984 but to allow and give a warning for old-style code that would
985 otherwise be invalid, or have different behavior.
987 @item -fno-gnu-keywords
988 Do not recognize @code{classof}, @code{headof}, @code{signature},
989 @code{sigof} or @code{typeof} as a keyword, so that code can use these
990 words as identifiers. You can use the keywords @code{__classof__},
991 @code{__headof__}, @code{__signature__}, @code{__sigof__}, and
992 @code{__typeof__} instead. @samp{-ansi} implies
993 @samp{-fno-gnu-keywords}.
995 @item -fguiding-decls
996 Treat a function declaration with the same type as a potential function
997 template instantiation as though it declares that instantiation, not a
998 normal function. If a definition is given for the function later in the
999 translation unit (or another translation unit if the target supports
1000 weak symbols), that definition will be used; otherwise the template will
1001 be instantiated. This behavior reflects the C++ language prior to
1002 September 1996, when guiding declarations were removed.
1004 This option implies @samp{-fname-mangling-version-0}, and will not work
1005 with other name mangling versions.
1007 @item -fno-implicit-templates
1008 Never emit code for templates which are instantiated implicitly (i.e. by
1009 use); only emit code for explicit instantiations. @xref{Template
1010 Instantiation}, for more information.
1012 @item -fhandle-signatures
1013 Recognize the @code{signature} and @code{sigof} keywords for specifying
1014 abstract types. The default (@samp{-fno-handle-signatures}) is not to
1015 recognize them. @xref{C++ Signatures, Type Abstraction using
1018 @item -fhuge-objects
1019 Support virtual function calls for objects that exceed the size
1020 representable by a @samp{short int}. Users should not use this flag by
1021 default; if you need to use it, the compiler will tell you so. If you
1022 compile any of your code with this flag, you must compile @emph{all} of
1023 your code with this flag (including the C++ library, if you use it).
1025 This flag is not useful when compiling with -fvtable-thunks.
1027 @item -fno-implement-inlines
1028 To save space, do not emit out-of-line copies of inline functions
1029 controlled by @samp{#pragma implementation}. This will cause linker
1030 errors if these functions are not inlined everywhere they are called.
1032 @item -fmemoize-lookups
1033 @itemx -fsave-memoized
1034 Use heuristics to compile faster. These heuristics are not enabled by
1035 default, since they are only effective for certain input files. Other
1036 input files compile more slowly.
1038 The first time the compiler must build a call to a member function (or
1039 reference to a data member), it must (1) determine whether the class
1040 implements member functions of that name; (2) resolve which member
1041 function to call (which involves figuring out what sorts of type
1042 conversions need to be made); and (3) check the visibility of the member
1043 function to the caller. All of this adds up to slower compilation.
1044 Normally, the second time a call is made to that member function (or
1045 reference to that data member), it must go through the same lengthy
1046 process again. This means that code like this:
1049 cout << "This " << p << " has " << n << " legs.\n";
1053 makes six passes through all three steps. By using a software cache, a
1054 ``hit'' significantly reduces this cost. Unfortunately, using the cache
1055 introduces another layer of mechanisms which must be implemented, and so
1056 incurs its own overhead. @samp{-fmemoize-lookups} enables the software
1059 Because access privileges (visibility) to members and member functions
1060 may differ from one function context to the next, G++ may need to flush
1061 the cache. With the @samp{-fmemoize-lookups} flag, the cache is flushed
1062 after every function that is compiled. The @samp{-fsave-memoized} flag
1063 enables the same software cache, but when the compiler determines that
1064 the context of the last function compiled would yield the same access
1065 privileges of the next function to compile, it preserves the cache.
1066 This is most helpful when defining many member functions for the same
1067 class: with the exception of member functions which are friends of other
1068 classes, each member function has exactly the same access privileges as
1069 every other, and the cache need not be flushed.
1071 The code that implements these flags has rotted; you should probably
1074 @item -fstrict-prototype
1075 Within an @samp{extern "C"} linkage specification, treat a function
1076 declaration with no arguments, such as @samp{int foo ();}, as declaring
1077 the function to take no arguments. Normally, such a declaration means
1078 that the function @code{foo} can take any combination of arguments, as
1079 in C. @samp{-pedantic} implies @samp{-fstrict-prototype} unless
1080 overridden with @samp{-fno-strict-prototype}.
1082 This flag no longer affects declarations with C++ linkage.
1084 @item -fname-mangling-version-@var{n}
1085 Control the way in which names are mangled. Version 0 is compatible
1086 with versions of g++ before 2.8. Version 1 is the default. Version 1
1087 will allow correct mangling of function templates. For example,
1088 version 0 mangling does not mangle foo<int, double> and foo<int, char>
1089 given this declaration:
1092 template <class T, class U> void foo(T t);
1095 @item -fno-nonnull-objects
1096 Don't assume that a reference is initialized to refer to a valid object.
1097 Although the current C++ Working Paper prohibits null references, some
1098 old code may rely on them, and you can use @samp{-fno-nonnull-objects}
1099 to turn on checking.
1101 At the moment, the compiler only does this checking for conversions to
1102 virtual base classes.
1104 @item -foperator-names
1105 Recognize the operator name keywords @code{and}, @code{bitand},
1106 @code{bitor}, @code{compl}, @code{not}, @code{or} and @code{xor} as
1107 synonyms for the symbols they refer to. @samp{-ansi} implies
1108 @samp{-foperator-names}.
1111 Enable automatic template instantiation. This option also implies
1112 @samp{-fno-implicit-templates}. @xref{Template Instantiation}, for more
1116 @itemx -fno-squangle
1117 @samp{-fsquangle} will enable a compressed form of name mangling for
1118 identifiers. In particular, it helps to shorten very long names by recognizing
1119 types and class names which occur more than once, replacing them with special
1120 short ID codes. This option also requires any C++ libraries being used to
1121 be compiled with this option as well. The compiler has this disabled (the
1122 equivalent of @samp{-fno-squangle}) by default.
1124 @item -fthis-is-variable
1125 Permit assignment to @code{this}. The incorporation of user-defined
1126 free store management into C++ has made assignment to @samp{this} an
1127 anachronism. Therefore, by default it is invalid to assign to
1128 @code{this} within a class member function; that is, GNU C++ treats
1129 @samp{this} in a member function of class @code{X} as a non-lvalue of
1130 type @samp{X *}. However, for backwards compatibility, you can make it
1131 valid with @samp{-fthis-is-variable}.
1133 @item -fvtable-thunks
1134 Use @samp{thunks} to implement the virtual function dispatch table
1135 (@samp{vtable}). The traditional (cfront-style) approach to
1136 implementing vtables was to store a pointer to the function and two
1137 offsets for adjusting the @samp{this} pointer at the call site. Newer
1138 implementations store a single pointer to a @samp{thunk} function which
1139 does any necessary adjustment and then calls the target function.
1141 This option also enables a heuristic for controlling emission of
1142 vtables; if a class has any non-inline virtual functions, the vtable
1143 will be emitted in the translation unit containing the first one of
1146 @item -ftemplate-depth-@var{n}
1147 Set the maximum instantiation depth for template classes to @var{n}.
1148 A limit on the template instantiation depth is needed to detect
1149 endless recursions during template class instantiation. ANSI/ISO C++
1150 conforming programs must not rely on a maximum depth greater than 17.
1153 Do not search for header files in the standard directories specific to
1154 C++, but do still search the other standard directories. (This option
1155 is used when building the C++ library.)
1158 For C++ programs (in addition to the effects that apply to both C and
1159 C++), this has the same effect as @samp{-fthis-is-variable}.
1160 @xref{C Dialect Options,, Options Controlling C Dialect}.
1163 In addition, these optimization, warning, and code generation options
1164 have meanings only for C++ programs:
1167 @item -fno-default-inline
1168 Do not assume @samp{inline} for functions defined inside a class scope.
1169 @xref{Optimize Options,,Options That Control Optimization}.
1171 @item -Wold-style-cast
1172 @itemx -Woverloaded-virtual
1173 @itemx -Wtemplate-debugging
1174 Warnings that apply only to C++ programs. @xref{Warning
1175 Options,,Options to Request or Suppress Warnings}.
1178 Warn about violation of some style rules from Effective C++ by Scott Myers.
1181 Control how virtual function definitions are used, in a fashion
1182 compatible with @code{cfront} 1.x. @xref{Code Gen Options,,Options for
1183 Code Generation Conventions}.
1186 @node Warning Options
1187 @section Options to Request or Suppress Warnings
1188 @cindex options to control warnings
1189 @cindex warning messages
1190 @cindex messages, warning
1191 @cindex suppressing warnings
1193 Warnings are diagnostic messages that report constructions which
1194 are not inherently erroneous but which are risky or suggest there
1195 may have been an error.
1197 You can request many specific warnings with options beginning @samp{-W},
1198 for example @samp{-Wimplicit} to request warnings on implicit
1199 declarations. Each of these specific warning options also has a
1200 negative form beginning @samp{-Wno-} to turn off warnings;
1201 for example, @samp{-Wno-implicit}. This manual lists only one of the
1202 two forms, whichever is not the default.
1204 These options control the amount and kinds of warnings produced by GNU
1208 @cindex syntax checking
1210 Check the code for syntax errors, but don't do anything beyond that.
1213 Issue all the warnings demanded by strict ANSI C and ISO C++;
1214 reject all programs that use forbidden extensions.
1216 Valid ANSI C and ISO C++ programs should compile properly with or without
1217 this option (though a rare few will require @samp{-ansi}). However,
1218 without this option, certain GNU extensions and traditional C and C++
1219 features are supported as well. With this option, they are rejected.
1221 @samp{-pedantic} does not cause warning messages for use of the
1222 alternate keywords whose names begin and end with @samp{__}. Pedantic
1223 warnings are also disabled in the expression that follows
1224 @code{__extension__}. However, only system header files should use
1225 these escape routes; application programs should avoid them.
1226 @xref{Alternate Keywords}.
1228 This option is not intended to be @i{useful}; it exists only to satisfy
1229 pedants who would otherwise claim that GNU CC fails to support the ANSI
1232 Some users try to use @samp{-pedantic} to check programs for strict ANSI
1233 C conformance. They soon find that it does not do quite what they want:
1234 it finds some non-ANSI practices, but not all---only those for which
1235 ANSI C @emph{requires} a diagnostic.
1237 A feature to report any failure to conform to ANSI C might be useful in
1238 some instances, but would require considerable additional work and would
1239 be quite different from @samp{-pedantic}. We recommend, rather, that
1240 users take advantage of the extensions of GNU C and disregard the
1241 limitations of other compilers. Aside from certain supercomputers and
1242 obsolete small machines, there is less and less reason ever to use any
1243 other C compiler other than for bootstrapping GNU CC.
1245 @item -pedantic-errors
1246 Like @samp{-pedantic}, except that errors are produced rather than
1250 Inhibit all warning messages.
1253 Inhibit warning messages about the use of @samp{#import}.
1255 @item -Wchar-subscripts
1256 Warn if an array subscript has type @code{char}. This is a common cause
1257 of error, as programmers often forget that this type is signed on some
1261 Warn whenever a comment-start sequence @samp{/*} appears in a @samp{/*}
1262 comment, or whenever a Backslash-Newline appears in a @samp{//} comment.
1265 Check calls to @code{printf} and @code{scanf}, etc., to make sure that
1266 the arguments supplied have types appropriate to the format string
1269 @item -Wimplicit-int
1270 Warn when a declaration does not specify a type.
1272 @item -Wimplicit-function-declarations
1273 Warn whenever a function is used before being declared.
1276 Same as @samp{-Wimplicit-int} and @samp{-Wimplicit-function-}@*
1280 Warn if the type of @samp{main} is suspicious. @samp{main} should be a
1281 function with external linkage, returning int, taking either zero
1282 arguments, two, or three arguments of appropriate types.
1285 Warn if parentheses are omitted in certain contexts, such
1286 as when there is an assignment in a context where a truth value
1287 is expected, or when operators are nested whose precedence people
1288 often get confused about.
1290 Also warn about constructions where there may be confusion to which
1291 @code{if} statement an @code{else} branch belongs. Here is an example of
1304 In C, every @code{else} branch belongs to the innermost possible @code{if}
1305 statement, which in this example is @code{if (b)}. This is often not
1306 what the programmer expected, as illustrated in the above example by
1307 indentation the programmer chose. When there is the potential for this
1308 confusion, GNU C will issue a warning when this flag is specified.
1309 To eliminate the warning, add explicit braces around the innermost
1310 @code{if} statement so there is no way the @code{else} could belong to
1311 the enclosing @code{if}. The resulting code would look like this:
1326 Warn whenever a function is defined with a return-type that defaults
1327 to @code{int}. Also warn about any @code{return} statement with no
1328 return-value in a function whose return-type is not @code{void}.
1331 Warn whenever a @code{switch} statement has an index of enumeral type
1332 and lacks a @code{case} for one or more of the named codes of that
1333 enumeration. (The presence of a @code{default} label prevents this
1334 warning.) @code{case} labels outside the enumeration range also
1335 provoke warnings when this option is used.
1338 Warn if any trigraphs are encountered (assuming they are enabled).
1341 Warn whenever a variable is unused aside from its declaration,
1342 whenever a function is declared static but never defined, whenever a
1343 label is declared but not used, and whenever a statement computes a
1344 result that is explicitly not used.
1346 In order to get a warning about an unused function parameter, you must
1347 specify both @samp{-W} and @samp{-Wunused}.
1349 To suppress this warning for an expression, simply cast it to void. For
1350 unused variables and parameters, use the @samp{unused} attribute
1351 (@pxref{Variable Attributes}).
1353 @item -Wuninitialized
1354 An automatic variable is used without first being initialized.
1356 These warnings are possible only in optimizing compilation,
1357 because they require data flow information that is computed only
1358 when optimizing. If you don't specify @samp{-O}, you simply won't
1361 These warnings occur only for variables that are candidates for
1362 register allocation. Therefore, they do not occur for a variable that
1363 is declared @code{volatile}, or whose address is taken, or whose size
1364 is other than 1, 2, 4 or 8 bytes. Also, they do not occur for
1365 structures, unions or arrays, even when they are in registers.
1367 Note that there may be no warning about a variable that is used only
1368 to compute a value that itself is never used, because such
1369 computations may be deleted by data flow analysis before the warnings
1372 These warnings are made optional because GNU CC is not smart
1373 enough to see all the reasons why the code might be correct
1374 despite appearing to have an error. Here is one example of how
1393 If the value of @code{y} is always 1, 2 or 3, then @code{x} is
1394 always initialized, but GNU CC doesn't know this. Here is
1395 another common case:
1400 if (change_y) save_y = y, y = new_y;
1402 if (change_y) y = save_y;
1407 This has no bug because @code{save_y} is used only if it is set.
1409 Some spurious warnings can be avoided if you declare all the functions
1410 you use that never return as @code{noreturn}. @xref{Function
1413 @item -Wreorder (C++ only)
1414 @cindex reordering, warning
1415 @cindex warning for reordering of member initializers
1416 Warn when the order of member initializers given in the code does not
1417 match the order in which they must be executed. For instance:
1423 A(): j (0), i (1) @{ @}
1427 Here the compiler will warn that the member initializers for @samp{i}
1428 and @samp{j} will be rearranged to match the declaration order of the
1431 @item -Wtemplate-debugging
1432 @cindex template debugging
1433 When using templates in a C++ program, warn if debugging is not yet
1434 fully available (C++ only).
1436 @item -Wunknown-pragmas
1437 @cindex warning for unknown pragmas
1438 @cindex unknown pragmas, warning
1439 @cindex pragmas, warning of unknown
1440 Warn when a #pragma directive is encountered which is not understood by
1441 GCC. If this command line option is used, warnings will even be issued
1442 for unknown pragmas in system header files. This is not the case if
1443 the warnings were only enabled by the @samp{-Wall} command line option.
1446 All of the above @samp{-W} options combined. This enables all the
1447 warnings about constructions that some users consider questionable, and
1448 that are easy to avoid (or modify to prevent the warning), even in
1449 conjunction with macros.
1452 The following @samp{-W@dots{}} options are not implied by @samp{-Wall}.
1453 Some of them warn about constructions that users generally do not
1454 consider questionable, but which occasionally you might wish to check
1455 for; others warn about constructions that are necessary or hard to avoid
1456 in some cases, and there is no simple way to modify the code to suppress
1461 Print extra warning messages for these events:
1464 @cindex @code{longjmp} warnings
1466 A nonvolatile automatic variable might be changed by a call to
1467 @code{longjmp}. These warnings as well are possible only in
1468 optimizing compilation.
1470 The compiler sees only the calls to @code{setjmp}. It cannot know
1471 where @code{longjmp} will be called; in fact, a signal handler could
1472 call it at any point in the code. As a result, you may get a warning
1473 even when there is in fact no problem because @code{longjmp} cannot
1474 in fact be called at the place which would cause a problem.
1477 A function can return either with or without a value. (Falling
1478 off the end of the function body is considered returning without
1479 a value.) For example, this function would evoke such a
1493 An expression-statement or the left-hand side of a comma expression
1494 contains no side effects.
1495 To suppress the warning, cast the unused expression to void.
1496 For example, an expression such as @samp{x[i,j]} will cause a warning,
1497 but @samp{x[(void)i,j]} will not.
1500 An unsigned value is compared against zero with @samp{<} or @samp{<=}.
1503 A comparison like @samp{x<=y<=z} appears; this is equivalent to
1504 @samp{(x<=y ? 1 : 0) <= z}, which is a different interpretation from
1505 that of ordinary mathematical notation.
1508 Storage-class specifiers like @code{static} are not the first things in
1509 a declaration. According to the C Standard, this usage is obsolescent.
1512 If @samp{-Wall} or @samp{-Wunused} is also specified, warn about unused
1516 A comparison between signed and unsigned values could produce an
1517 incorrect result when the signed value is converted to unsigned.
1518 (But don't warn if @samp{-Wno-sign-compare} is also specified.)
1521 An aggregate has a partly bracketed initializer.
1522 For example, the following code would evoke such a warning,
1523 because braces are missing around the initializer for @code{x.h}:
1526 struct s @{ int f, g; @};
1527 struct t @{ struct s h; int i; @};
1528 struct t x = @{ 1, 2, 3 @};
1533 Warn about certain constructs that behave differently in traditional and
1538 Macro arguments occurring within string constants in the macro body.
1539 These would substitute the argument in traditional C, but are part of
1540 the constant in ANSI C.
1543 A function declared external in one block and then used after the end of
1547 A @code{switch} statement has an operand of type @code{long}.
1551 Warn if an undefined identifier is evaluated in an @samp{#if} directive.
1554 Warn whenever a local variable shadows another local variable.
1556 @item -Wid-clash-@var{len}
1557 Warn whenever two distinct identifiers match in the first @var{len}
1558 characters. This may help you prepare a program that will compile
1559 with certain obsolete, brain-damaged compilers.
1561 @item -Wlarger-than-@var{len}
1562 Warn whenever an object of larger than @var{len} bytes is defined.
1564 @item -Wpointer-arith
1565 Warn about anything that depends on the ``size of'' a function type or
1566 of @code{void}. GNU C assigns these types a size of 1, for
1567 convenience in calculations with @code{void *} pointers and pointers
1570 @item -Wbad-function-cast
1571 Warn whenever a function call is cast to a non-matching type.
1572 For example, warn if @code{int malloc()} is cast to @code{anything *}.
1575 Warn whenever a pointer is cast so as to remove a type qualifier from
1576 the target type. For example, warn if a @code{const char *} is cast
1577 to an ordinary @code{char *}.
1580 Warn whenever a pointer is cast such that the required alignment of the
1581 target is increased. For example, warn if a @code{char *} is cast to
1582 an @code{int *} on machines where integers can only be accessed at
1583 two- or four-byte boundaries.
1585 @item -Wwrite-strings
1586 Give string constants the type @code{const char[@var{length}]} so that
1587 copying the address of one into a non-@code{const} @code{char *}
1588 pointer will get a warning. These warnings will help you find at
1589 compile time code that can try to write into a string constant, but
1590 only if you have been very careful about using @code{const} in
1591 declarations and prototypes. Otherwise, it will just be a nuisance;
1592 this is why we did not make @samp{-Wall} request these warnings.
1595 Warn if a prototype causes a type conversion that is different from what
1596 would happen to the same argument in the absence of a prototype. This
1597 includes conversions of fixed point to floating and vice versa, and
1598 conversions changing the width or signedness of a fixed point argument
1599 except when the same as the default promotion.
1601 Also, warn if a negative integer constant expression is implicitly
1602 converted to an unsigned type. For example, warn about the assignment
1603 @code{x = -1} if @code{x} is unsigned. But do not warn about explicit
1604 casts like @code{(unsigned) -1}.
1606 @item -Wsign-compare
1607 @cindex warning for comparison of signed and unsigned values
1608 @cindex comparison of signed and unsigned values, warning
1609 @cindex signed and unsigned values, comparison warning
1610 Warn when a comparison between signed and unsigned values could produce
1611 an incorrect result when the signed value is converted to unsigned.
1612 This warning is also enabled by @samp{-W}; to get the other warnings
1613 of @samp{-W} without this warning, use @samp{-W -Wno-sign-compare}.
1615 @item -Waggregate-return
1616 Warn if any functions that return structures or unions are defined or
1617 called. (In languages where you can return an array, this also elicits
1620 @item -Wstrict-prototypes
1621 Warn if a function is declared or defined without specifying the
1622 argument types. (An old-style function definition is permitted without
1623 a warning if preceded by a declaration which specifies the argument
1626 @item -Wmissing-prototypes
1627 Warn if a global function is defined without a previous prototype
1628 declaration. This warning is issued even if the definition itself
1629 provides a prototype. The aim is to detect global functions that fail
1630 to be declared in header files.
1632 @item -Wmissing-declarations
1633 Warn if a global function is defined without a previous declaration.
1634 Do so even if the definition itself provides a prototype.
1635 Use this option to detect global functions that are not declared in
1638 @item -Wredundant-decls
1639 Warn if anything is declared more than once in the same scope, even in
1640 cases where multiple declaration is valid and changes nothing.
1642 @item -Wnested-externs
1643 Warn if an @code{extern} declaration is encountered within an function.
1646 Warn if a function can not be inlined, and either it was declared as inline,
1647 or else the @samp{-finline-functions} option was given.
1649 @item -Wold-style-cast
1650 Warn if an old-style (C-style) cast is used within a program.
1652 @item -Woverloaded-virtual
1653 @cindex overloaded virtual fn, warning
1654 @cindex warning for overloaded virtual fn
1655 Warn when a derived class function declaration may be an error in
1656 defining a virtual function (C++ only). In a derived class, the
1657 definitions of virtual functions must match the type signature of a
1658 virtual function declared in the base class. With this option, the
1659 compiler warns when you define a function with the same name as a
1660 virtual function, but with a type signature that does not match any
1661 declarations from the base class.
1663 @item -Wsynth (C++ only)
1664 @cindex warning for synthesized methods
1665 @cindex synthesized methods, warning
1666 Warn when g++'s synthesis behavior does not match that of cfront. For
1672 A& operator = (int);
1682 In this example, g++ will synthesize a default @samp{A& operator =
1683 (const A&);}, while cfront will use the user-defined @samp{operator =}.
1686 Make all warnings into errors.
1689 @node Debugging Options
1690 @section Options for Debugging Your Program or GNU CC
1691 @cindex options, debugging
1692 @cindex debugging information options
1694 GNU CC has various special options that are used for debugging
1695 either your program or GCC:
1699 Produce debugging information in the operating system's native format
1700 (stabs, COFF, XCOFF, or DWARF). GDB can work with this debugging
1703 On most systems that use stabs format, @samp{-g} enables use of extra
1704 debugging information that only GDB can use; this extra information
1705 makes debugging work better in GDB but will probably make other debuggers
1707 refuse to read the program. If you want to control for certain whether
1708 to generate the extra information, use @samp{-gstabs+}, @samp{-gstabs},
1709 @samp{-gxcoff+}, @samp{-gxcoff}, @samp{-gdwarf-1+}, or @samp{-gdwarf-1}
1712 Unlike most other C compilers, GNU CC allows you to use @samp{-g} with
1713 @samp{-O}. The shortcuts taken by optimized code may occasionally
1714 produce surprising results: some variables you declared may not exist
1715 at all; flow of control may briefly move where you did not expect it;
1716 some statements may not be executed because they compute constant
1717 results or their values were already at hand; some statements may
1718 execute in different places because they were moved out of loops.
1720 Nevertheless it proves possible to debug optimized output. This makes
1721 it reasonable to use the optimizer for programs that might have bugs.
1723 The following options are useful when GNU CC is generated with the
1724 capability for more than one debugging format.
1727 Produce debugging information for use by GDB. This means to use the
1728 most expressive format available (DWARF 2, stabs, or the native format
1729 if neither of those are supported), including GDB extensions if at all
1733 Produce debugging information in stabs format (if that is supported),
1734 without GDB extensions. This is the format used by DBX on most BSD
1735 systems. On MIPS, Alpha and System V Release 4 systems this option
1736 produces stabs debugging output which is not understood by DBX or SDB.
1737 On System V Release 4 systems this option requires the GNU assembler.
1740 Produce debugging information in stabs format (if that is supported),
1741 using GNU extensions understood only by the GNU debugger (GDB). The
1742 use of these extensions is likely to make other debuggers crash or
1743 refuse to read the program.
1746 Produce debugging information in COFF format (if that is supported).
1747 This is the format used by SDB on most System V systems prior to
1751 Produce debugging information in XCOFF format (if that is supported).
1752 This is the format used by the DBX debugger on IBM RS/6000 systems.
1755 Produce debugging information in XCOFF format (if that is supported),
1756 using GNU extensions understood only by the GNU debugger (GDB). The
1757 use of these extensions is likely to make other debuggers crash or
1758 refuse to read the program, and may cause assemblers other than the GNU
1759 assembler (GAS) to fail with an error.
1762 Produce debugging information in DWARF version 1 format (if that is
1763 supported). This is the format used by SDB on most System V Release 4
1767 Produce debugging information in DWARF version 1 format (if that is
1768 supported), using GNU extensions understood only by the GNU debugger
1769 (GDB). The use of these extensions is likely to make other debuggers
1770 crash or refuse to read the program.
1773 Produce debugging information in DWARF version 2 format (if that is
1774 supported). This is the format used by DBX on IRIX 6.
1777 @itemx -ggdb@var{level}
1778 @itemx -gstabs@var{level}
1779 @itemx -gcoff@var{level}
1780 @itemx -gxcoff@var{level}
1781 @itemx -gdwarf@var{level}
1782 @itemx -gdwarf-2@var{level}
1783 Request debugging information and also use @var{level} to specify how
1784 much information. The default level is 2.
1786 Level 1 produces minimal information, enough for making backtraces in
1787 parts of the program that you don't plan to debug. This includes
1788 descriptions of functions and external variables, but no information
1789 about local variables and no line numbers.
1791 Level 3 includes extra information, such as all the macro definitions
1792 present in the program. Some debuggers support macro expansion when
1797 Generate extra code to write profile information suitable for the
1798 analysis program @code{prof}. You must use this option when compiling
1799 the source files you want data about, and you must also use it when
1802 @cindex @code{gprof}
1804 Generate extra code to write profile information suitable for the
1805 analysis program @code{gprof}. You must use this option when compiling
1806 the source files you want data about, and you must also use it when
1811 Generate extra code to write profile information for basic blocks, which will
1812 record the number of times each basic block is executed, the basic block start
1813 address, and the function name containing the basic block. If @samp{-g} is
1814 used, the line number and filename of the start of the basic block will also be
1815 recorded. If not overridden by the machine description, the default action is
1816 to append to the text file @file{bb.out}.
1818 This data could be analyzed by a program like @code{tcov}. Note,
1819 however, that the format of the data is not what @code{tcov} expects.
1820 Eventually GNU @code{gprof} should be extended to process this data.
1823 Makes the compiler print out each function name as it is compiled, and
1824 print some statistics about each pass when it finishes.
1827 Generate extra code to profile basic blocks. Your executable will
1828 produce output that is a superset of that produced when @samp{-a} is
1829 used. Additional output is the source and target address of the basic
1830 blocks where a jump takes place, the number of times a jump is executed,
1831 and (optionally) the complete sequence of basic blocks being executed.
1832 The output is appended to file @file{bb.out}.
1834 You can examine different profiling aspects without recompilation. Your
1835 executable will read a list of function names from file @file{bb.in}.
1836 Profiling starts when a function on the list is entered and stops when
1837 that invocation is exited. To exclude a function from profiling, prefix
1838 its name with `-'. If a function name is not unique, you can
1839 disambiguate it by writing it in the form
1840 @samp{/path/filename.d:functionname}. Your executable will write the
1841 available paths and filenames in file @file{bb.out}.
1843 Several function names have a special meaning:
1846 Write source, target and frequency of jumps to file @file{bb.out}.
1847 @item __bb_hidecall__
1848 Exclude function calls from frequency count.
1849 @item __bb_showret__
1850 Include function returns in frequency count.
1852 Write the sequence of basic blocks executed to file @file{bbtrace.gz}.
1853 The file will be compressed using the program @samp{gzip}, which must
1854 exist in your @code{PATH}. On systems without the @samp{popen}
1855 function, the file will be named @file{bbtrace} and will not be
1856 compressed. @strong{Profiling for even a few seconds on these systems
1857 will produce a very large file.} Note: @code{__bb_hidecall__} and
1858 @code{__bb_showret__} will not affect the sequence written to
1862 Here's a short example using different profiling parameters
1863 in file @file{bb.in}. Assume function @code{foo} consists of basic blocks
1864 1 and 2 and is called twice from block 3 of function @code{main}. After
1865 the calls, block 3 transfers control to block 4 of @code{main}.
1867 With @code{__bb_trace__} and @code{main} contained in file @file{bb.in},
1868 the following sequence of blocks is written to file @file{bbtrace.gz}:
1869 0 3 1 2 1 2 4. The return from block 2 to block 3 is not shown, because
1870 the return is to a point inside the block and not to the top. The
1871 block address 0 always indicates, that control is transferred
1872 to the trace from somewhere outside the observed functions. With
1873 @samp{-foo} added to @file{bb.in}, the blocks of function
1874 @code{foo} are removed from the trace, so only 0 3 4 remains.
1876 With @code{__bb_jumps__} and @code{main} contained in file @file{bb.in},
1877 jump frequencies will be written to file @file{bb.out}. The
1878 frequencies are obtained by constructing a trace of blocks
1879 and incrementing a counter for every neighbouring pair of blocks
1880 in the trace. The trace 0 3 1 2 1 2 4 displays the following
1884 Jump from block 0x0 to block 0x3 executed 1 time(s)
1885 Jump from block 0x3 to block 0x1 executed 1 time(s)
1886 Jump from block 0x1 to block 0x2 executed 2 time(s)
1887 Jump from block 0x2 to block 0x1 executed 1 time(s)
1888 Jump from block 0x2 to block 0x4 executed 1 time(s)
1891 With @code{__bb_hidecall__}, control transfer due to call instructions
1892 is removed from the trace, that is the trace is cut into three parts: 0
1893 3 4, 0 1 2 and 0 1 2. With @code{__bb_showret__}, control transfer due
1894 to return instructions is added to the trace. The trace becomes: 0 3 1
1895 2 3 1 2 3 4. Note, that this trace is not the same, as the sequence
1896 written to @file{bbtrace.gz}. It is solely used for counting jump
1899 @item -fprofile-arcs
1900 Instrument @dfn{arcs} during compilation. For each function of your
1901 program, GNU CC creates a program flow graph, then finds a spanning tree
1902 for the graph. Only arcs that are not on the spanning tree have to be
1903 instrumented: the compiler adds code to count the number of times that these
1904 arcs are executed. When an arc is the only exit or only entrance to a
1905 block, the instrumentation code can be added to the block; otherwise, a
1906 new basic block must be created to hold the instrumentation code.
1908 Since not every arc in the program must be instrumented, programs
1909 compiled with this option run faster than programs compiled with
1910 @samp{-a}, which adds instrumentation code to every basic block in the
1911 program. The tradeoff: since @code{gcov} does not have
1912 execution counts for all branches, it must start with the execution
1913 counts for the instrumented branches, and then iterate over the program
1914 flow graph until the entire graph has been solved. Hence, @code{gcov}
1915 runs a little more slowly than a program which uses information from
1918 @samp{-fprofile-arcs} also makes it possible to estimate branch
1919 probabilities, and to calculate basic block execution counts. In
1920 general, basic block execution counts do not give enough information to
1921 estimate all branch probabilities. When the compiled program exits, it
1922 saves the arc execution counts to a file called
1923 @file{@var{sourcename}.da}. Use the compiler option
1924 @samp{-fbranch-probabilities} (@pxref{Optimize Options,,Options that
1925 Control Optimization}) when recompiling, to optimize using estimated
1926 branch probabilities.
1929 @item -ftest-coverage
1930 Create data files for the @code{gcov} code-coverage utility
1931 (@pxref{Gcov,, @code{gcov}: a GNU CC Test Coverage Program}).
1932 The data file names begin with the name of your source file:
1935 @item @var{sourcename}.bb
1936 A mapping from basic blocks to line numbers, which @code{gcov} uses to
1937 associate basic block execution counts with line numbers.
1939 @item @var{sourcename}.bbg
1940 A list of all arcs in the program flow graph. This allows @code{gcov}
1941 to reconstruct the program flow graph, so that it can compute all basic
1942 block and arc execution counts from the information in the
1943 @code{@var{sourcename}.da} file (this last file is the output from
1944 @samp{-fprofile-arcs}).
1948 Makes the compiler print out each function name as it is compiled, and
1949 print some statistics about each pass when it finishes.
1951 @item -d@var{letters}
1952 Says to make debugging dumps during compilation at times specified by
1953 @var{letters}. This is used for debugging the compiler. The file names
1954 for most of the dumps are made by appending a word to the source file
1955 name (e.g. @file{foo.c.rtl} or @file{foo.c.jump}). Here are the
1956 possible letters for use in @var{letters}, and their meanings:
1960 Dump after computing branch probabilities, to @file{@var{file}.bp}.
1962 Dump after instruction combination, to the file @file{@var{file}.combine}.
1964 Dump after delayed branch scheduling, to @file{@var{file}.dbr}.
1966 Dump after purging ADDRESSOF, to @file{@var{file}.addressof}.
1968 Dump after flow analysis, to @file{@var{file}.flow}.
1970 Dump after global register allocation, to @file{@var{file}.greg}.
1972 Dump after first jump optimization, to @file{@var{file}.jump}.
1974 Dump after last jump optimization, to @file{@var{file}.jump2}.
1976 Dump after conversion from registers to stack, to @file{@var{file}.stack}.
1978 Dump after local register allocation, to @file{@var{file}.lreg}.
1980 Dump after loop optimization, to @file{@var{file}.loop}.
1982 Dump after performing the machine dependent reorganisation pass, to
1983 @file{@var{file}.mach}.
1985 Dump after the register move pass, to @file{@var{file}.regmove}.
1987 Dump after RTL generation, to @file{@var{file}.rtl}.
1989 Dump after the second instruction scheduling pass, to @file{@var{file}.sched2}.
1991 Dump after CSE (including the jump optimization that sometimes follows
1992 CSE), to @file{@var{file}.cse}.
1994 Dump after the first instruction scheduling pass, to @file{@var{file}.sched}.
1996 Dump after the second CSE pass (including the jump optimization that
1997 sometimes follows CSE), to @file{@var{file}.cse2}.
1999 Just generate RTL for a function instead of compiling it. Usually used
2002 Produce all the dumps listed above.
2004 Print statistics on memory usage, at the end of the run, to
2007 Annotate the assembler output with a comment indicating which
2008 pattern and alternative was used.
2010 Dump debugging information during parsing, to standard error.
2012 Annotate the assembler output with miscellaneous debugging information.
2015 @item -fpretend-float
2016 When running a cross-compiler, pretend that the target machine uses the
2017 same floating point format as the host machine. This causes incorrect
2018 output of the actual floating constants, but the actual instruction
2019 sequence will probably be the same as GNU CC would make when running on
2023 Store the usual ``temporary'' intermediate files permanently; place them
2024 in the current directory and name them based on the source file. Thus,
2025 compiling @file{foo.c} with @samp{-c -save-temps} would produce files
2026 @file{foo.i} and @file{foo.s}, as well as @file{foo.o}.
2028 @item -print-file-name=@var{library}
2029 Print the full absolute name of the library file @var{library} that
2030 would be used when linking---and don't do anything else. With this
2031 option, GNU CC does not compile or link anything; it just prints the
2034 @item -print-prog-name=@var{program}
2035 Like @samp{-print-file-name}, but searches for a program such as @samp{cpp}.
2037 @item -print-libgcc-file-name
2038 Same as @samp{-print-file-name=libgcc.a}.
2040 This is useful when you use @samp{-nostdlib} or @samp{-nodefaultlibs}
2041 but you do want to link with @file{libgcc.a}. You can do
2044 gcc -nostdlib @var{files}@dots{} `gcc -print-libgcc-file-name`
2047 @item -print-search-dirs
2048 Print the name of the configured installation directory and a list of
2049 program and library directories gcc will search---and don't do anything else.
2051 This is useful when gcc prints the error message
2052 @samp{installation problem, cannot exec cpp: No such file or directory}.
2053 To resolve this you either need to put @file{cpp} and the other compiler
2054 components where gcc expects to find them, or you can set the environment
2055 variable @code{GCC_EXEC_PREFIX} to the directory where you installed them.
2056 Don't forget the trailing '/'.
2057 @xref{Environment Variables}.
2060 @node Optimize Options
2061 @section Options That Control Optimization
2062 @cindex optimize options
2063 @cindex options, optimization
2065 These options control various sorts of optimizations:
2070 Optimize. Optimizing compilation takes somewhat more time, and a lot
2071 more memory for a large function.
2073 Without @samp{-O}, the compiler's goal is to reduce the cost of
2074 compilation and to make debugging produce the expected results.
2075 Statements are independent: if you stop the program with a breakpoint
2076 between statements, you can then assign a new value to any variable or
2077 change the program counter to any other statement in the function and
2078 get exactly the results you would expect from the source code.
2080 Without @samp{-O}, the compiler only allocates variables declared
2081 @code{register} in registers. The resulting compiled code is a little
2082 worse than produced by PCC without @samp{-O}.
2084 With @samp{-O}, the compiler tries to reduce code size and execution
2087 When you specify @samp{-O}, the compiler turns on @samp{-fthread-jumps}
2088 and @samp{-fdefer-pop} on all machines. The compiler turns on
2089 @samp{-fdelayed-branch} on machines that have delay slots, and
2090 @samp{-fomit-frame-pointer} on machines that can support debugging even
2091 without a frame pointer. On some machines the compiler also turns
2092 on other flags.@refill
2095 Optimize even more. GNU CC performs nearly all supported optimizations
2096 that do not involve a space-speed tradeoff. The compiler does not
2097 perform loop unrolling or function inlining when you specify @samp{-O2}.
2098 As compared to @samp{-O}, this option increases both compilation time
2099 and the performance of the generated code.
2101 @samp{-O2} turns on all optional optimizations except for loop unrolling
2102 and function inlining. It also turns on the @samp{-fforce-mem} option
2103 on all machines and frame pointer elimination on machines where doing so
2104 does not interfere with debugging.
2107 Optimize yet more. @samp{-O3} turns on all optimizations specified by
2108 @samp{-O2} and also turns on the @samp{inline-functions} option.
2114 Optimize for size. @samp{-Os} enables all @samp{-O2} optimizations that
2115 do not typically increase code size. It also performs further
2116 optimizations designed to reduce code size.
2118 If you use multiple @samp{-O} options, with or without level numbers,
2119 the last such option is the one that is effective.
2122 Options of the form @samp{-f@var{flag}} specify machine-independent
2123 flags. Most flags have both positive and negative forms; the negative
2124 form of @samp{-ffoo} would be @samp{-fno-foo}. In the table below,
2125 only one of the forms is listed---the one which is not the default.
2126 You can figure out the other form by either removing @samp{no-} or
2131 Do not store floating point variables in registers, and inhibit other
2132 options that might change whether a floating point value is taken from a
2135 @cindex floating point precision
2136 This option prevents undesirable excess precision on machines such as
2137 the 68000 where the floating registers (of the 68881) keep more
2138 precision than a @code{double} is supposed to have. Similarly for the
2139 x86 architecture. For most programs, the excess precision does only
2140 good, but a few programs rely on the precise definition of IEEE floating
2141 point. Use @samp{-ffloat-store} for such programs.
2143 @item -fno-default-inline
2144 Do not make member functions inline by default merely because they are
2145 defined inside the class scope (C++ only). Otherwise, when you specify
2146 @w{@samp{-O}}, member functions defined inside class scope are compiled
2147 inline by default; i.e., you don't need to add @samp{inline} in front of
2148 the member function name.
2150 @item -fno-defer-pop
2151 Always pop the arguments to each function call as soon as that function
2152 returns. For machines which must pop arguments after a function call,
2153 the compiler normally lets arguments accumulate on the stack for several
2154 function calls and pops them all at once.
2157 Force memory operands to be copied into registers before doing
2158 arithmetic on them. This produces better code by making all memory
2159 references potential common subexpressions. When they are not common
2160 subexpressions, instruction combination should eliminate the separate
2161 register-load. The @samp{-O2} option turns on this option.
2164 Force memory address constants to be copied into registers before
2165 doing arithmetic on them. This may produce better code just as
2166 @samp{-fforce-mem} may.
2168 @item -fomit-frame-pointer
2169 Don't keep the frame pointer in a register for functions that
2170 don't need one. This avoids the instructions to save, set up and
2171 restore frame pointers; it also makes an extra register available
2172 in many functions. @strong{It also makes debugging impossible on
2176 On some machines, such as the Vax, this flag has no effect, because
2177 the standard calling sequence automatically handles the frame pointer
2178 and nothing is saved by pretending it doesn't exist. The
2179 machine-description macro @code{FRAME_POINTER_REQUIRED} controls
2180 whether a target machine supports this flag. @xref{Registers}.@refill
2183 On some machines, such as the Vax, this flag has no effect, because
2184 the standard calling sequence automatically handles the frame pointer
2185 and nothing is saved by pretending it doesn't exist. The
2186 machine-description macro @code{FRAME_POINTER_REQUIRED} controls
2187 whether a target machine supports this flag. @xref{Registers,,Register
2188 Usage, gcc.info, Using and Porting GCC}.@refill
2192 Don't pay attention to the @code{inline} keyword. Normally this option
2193 is used to keep the compiler from expanding any functions inline.
2194 Note that if you are not optimizing, no functions can be expanded inline.
2196 @item -finline-functions
2197 Integrate all simple functions into their callers. The compiler
2198 heuristically decides which functions are simple enough to be worth
2199 integrating in this way.
2201 If all calls to a given function are integrated, and the function is
2202 declared @code{static}, then the function is normally not output as
2203 assembler code in its own right.
2205 @item -fkeep-inline-functions
2206 Even if all calls to a given function are integrated, and the function
2207 is declared @code{static}, nevertheless output a separate run-time
2208 callable version of the function. This switch does not affect
2209 @code{extern inline} functions.
2211 @item -fkeep-static-consts
2212 Emit variables declared @code{static const} when optimization isn't turned
2213 on, even if the variables aren't referenced.
2215 GNU CC enables this option by default. If you want to force the compiler to
2216 check if the variable was referenced, regardless of whether or not
2217 optimization is turned on, use the @samp{-fno-keep-static-consts} option.
2219 @item -fno-function-cse
2220 Do not put function addresses in registers; make each instruction that
2221 calls a constant function contain the function's address explicitly.
2223 This option results in less efficient code, but some strange hacks
2224 that alter the assembler output may be confused by the optimizations
2225 performed when this option is not used.
2228 This option allows GCC to violate some ANSI or IEEE rules and/or
2229 specifications in the interest of optimizing code for speed. For
2230 example, it allows the compiler to assume arguments to the @code{sqrt}
2231 function are non-negative numbers and that no floating-point values
2234 This option should never be turned on by any @samp{-O} option since
2235 it can result in incorrect output for programs which depend on
2236 an exact implementation of IEEE or ANSI rules/specifications for
2240 @c following causes underfulls.. they don't look great, but we deal.
2242 The following options control specific optimizations. The @samp{-O2}
2243 option turns on all of these optimizations except @samp{-funroll-loops}
2244 and @samp{-funroll-all-loops}. On most machines, the @samp{-O} option
2245 turns on the @samp{-fthread-jumps} and @samp{-fdelayed-branch} options,
2246 but specific machines may handle it differently.
2248 You can use the following flags in the rare cases when ``fine-tuning''
2249 of optimizations to be performed is desired.
2252 @item -fstrength-reduce
2253 Perform the optimizations of loop strength reduction and
2254 elimination of iteration variables.
2256 @item -fthread-jumps
2257 Perform optimizations where we check to see if a jump branches to a
2258 location where another comparison subsumed by the first is found. If
2259 so, the first branch is redirected to either the destination of the
2260 second branch or a point immediately following it, depending on whether
2261 the condition is known to be true or false.
2263 @item -fcse-follow-jumps
2264 In common subexpression elimination, scan through jump instructions
2265 when the target of the jump is not reached by any other path. For
2266 example, when CSE encounters an @code{if} statement with an
2267 @code{else} clause, CSE will follow the jump when the condition
2270 @item -fcse-skip-blocks
2271 This is similar to @samp{-fcse-follow-jumps}, but causes CSE to
2272 follow jumps which conditionally skip over blocks. When CSE
2273 encounters a simple @code{if} statement with no else clause,
2274 @samp{-fcse-skip-blocks} causes CSE to follow the jump around the
2275 body of the @code{if}.
2277 @item -frerun-cse-after-loop
2278 Re-run common subexpression elimination after loop optimizations has been
2281 @item -frerun-loop-opt
2282 Run the loop optimizer twice.
2284 @item -fexpensive-optimizations
2285 Perform a number of minor optimizations that are relatively expensive.
2287 @item -fdelayed-branch
2288 If supported for the target machine, attempt to reorder instructions
2289 to exploit instruction slots available after delayed branch
2292 @item -fschedule-insns
2293 If supported for the target machine, attempt to reorder instructions to
2294 eliminate execution stalls due to required data being unavailable. This
2295 helps machines that have slow floating point or memory load instructions
2296 by allowing other instructions to be issued until the result of the load
2297 or floating point instruction is required.
2299 @item -fschedule-insns2
2300 Similar to @samp{-fschedule-insns}, but requests an additional pass of
2301 instruction scheduling after register allocation has been done. This is
2302 especially useful on machines with a relatively small number of
2303 registers and where memory load instructions take more than one cycle.
2305 @item -ffunction-sections
2306 Place each function into its own section in the output file if the
2307 target supports arbitrary sections. The function's name determines
2308 the section's name in the output file.
2310 Use this option on systems where the linker can perform optimizations
2311 to improve locality of reference in the instruction space. HPPA
2312 processors running HP-UX and Sparc processors running Solaris 2 have
2313 linkers with such optimizations. Other systems using the ELF object format
2314 as well as AIX may have these optimizations in the future.
2316 Only use this option when there are significant benefits from doing
2317 so. When you specify this option, the assembler and linker will
2318 create larger object and executable files and will also be slower.
2319 You will not be able to use @code{gprof} on all systems if you
2320 specify this option and you may have problems with debugging if
2321 you specify both this option and @samp{-g}.
2323 @item -fcaller-saves
2324 Enable values to be allocated in registers that will be clobbered by
2325 function calls, by emitting extra instructions to save and restore the
2326 registers around such calls. Such allocation is done only when it
2327 seems to result in better code than would otherwise be produced.
2329 This option is enabled by default on certain machines, usually those
2330 which have no call-preserved registers to use instead.
2332 @item -funroll-loops
2333 Perform the optimization of loop unrolling. This is only done for loops
2334 whose number of iterations can be determined at compile time or run time.
2335 @samp{-funroll-loop} implies both @samp{-fstrength-reduce} and
2336 @samp{-frerun-cse-after-loop}.
2338 @item -funroll-all-loops
2339 Perform the optimization of loop unrolling. This is done for all loops
2340 and usually makes programs run more slowly. @samp{-funroll-all-loops}
2341 implies @samp{-fstrength-reduce} as well as @samp{-frerun-cse-after-loop}.
2343 @item -fmove-all-movables
2344 Forces all invariant computations in loops to be moved
2347 @item -freduce-all-givs
2348 Forces all general-induction variables in loops to be
2351 @emph{Note:} When compiling programs written in Fortran,
2352 @samp{-fmove-all-moveables} and @samp{-freduce-all-givs} are enabled
2353 by default when you use the optimizer.
2355 These options may generate better or worse code; results are highly
2356 dependent on the structure of loops within the source code.
2358 These two options are intended to be removed someday, once
2359 they have helped determine the efficacy of various
2360 approaches to improving loop optimizations.
2362 Please let us (@code{egcs@@cygnus.com} and @code{fortran@@gnu.org})
2363 know how use of these options affects
2364 the performance of your production code.
2365 We're very interested in code that runs @emph{slower}
2366 when these options are @emph{enabled}.
2369 Disable any machine-specific peephole optimizations.
2371 @item -fbranch-probabilities
2372 After running a program compiled with @samp{-fprofile-arcs}
2373 (@pxref{Debugging Options,, Options for Debugging Your Program or
2374 @code{gcc}}), you can compile it a second time using
2375 @samp{-fbranch-probabilities}, to improve optimizations based on
2376 guessing the path a branch might take.
2379 With @samp{-fbranch-probabilities}, GCC puts a @samp{REG_EXEC_COUNT}
2380 note on the first instruction of each basic block, and a
2381 @samp{REG_BR_PROB} note on each @samp{JUMP_INSN} and @samp{CALL_INSN}.
2382 These can be used to improve optimization. Currently, they are only
2383 used in one place: in @file{reorg.c}, instead of guessing which path a
2384 branch is mostly to take, the @samp{REG_BR_PROB} values are used to
2385 exactly determine which path is taken more often.
2389 Some machines only support 2 operands per instruction. On such
2390 machines, GNU CC might have to do extra copies. The @samp{-fregmove}
2391 option overrides the default for the machine to do the copy before
2392 register allocation.
2395 @node Preprocessor Options
2396 @section Options Controlling the Preprocessor
2397 @cindex preprocessor options
2398 @cindex options, preprocessor
2400 These options control the C preprocessor, which is run on each C source
2401 file before actual compilation.
2403 If you use the @samp{-E} option, nothing is done except preprocessing.
2404 Some of these options make sense only together with @samp{-E} because
2405 they cause the preprocessor output to be unsuitable for actual
2409 @item -include @var{file}
2410 Process @var{file} as input before processing the regular input file.
2411 In effect, the contents of @var{file} are compiled first. Any @samp{-D}
2412 and @samp{-U} options on the command line are always processed before
2413 @samp{-include @var{file}}, regardless of the order in which they are
2414 written. All the @samp{-include} and @samp{-imacros} options are
2415 processed in the order in which they are written.
2417 @item -imacros @var{file}
2418 Process @var{file} as input, discarding the resulting output, before
2419 processing the regular input file. Because the output generated from
2420 @var{file} is discarded, the only effect of @samp{-imacros @var{file}}
2421 is to make the macros defined in @var{file} available for use in the
2424 Any @samp{-D} and @samp{-U} options on the command line are always
2425 processed before @samp{-imacros @var{file}}, regardless of the order in
2426 which they are written. All the @samp{-include} and @samp{-imacros}
2427 options are processed in the order in which they are written.
2429 @item -idirafter @var{dir}
2430 @cindex second include path
2431 Add the directory @var{dir} to the second include path. The directories
2432 on the second include path are searched when a header file is not found
2433 in any of the directories in the main include path (the one that
2436 @item -iprefix @var{prefix}
2437 Specify @var{prefix} as the prefix for subsequent @samp{-iwithprefix}
2440 @item -iwithprefix @var{dir}
2441 Add a directory to the second include path. The directory's name is
2442 made by concatenating @var{prefix} and @var{dir}, where @var{prefix} was
2443 specified previously with @samp{-iprefix}. If you have not specified a
2444 prefix yet, the directory containing the installed passes of the
2445 compiler is used as the default.
2447 @item -iwithprefixbefore @var{dir}
2448 Add a directory to the main include path. The directory's name is made
2449 by concatenating @var{prefix} and @var{dir}, as in the case of
2450 @samp{-iwithprefix}.
2452 @item -isystem @var{dir}
2453 Add a directory to the beginning of the second include path, marking it
2454 as a system directory, so that it gets the same special treatment as
2455 is applied to the standard system directories.
2458 Do not search the standard system directories for header files. Only
2459 the directories you have specified with @samp{-I} options (and the
2460 current directory, if appropriate) are searched. @xref{Directory
2461 Options}, for information on @samp{-I}.
2463 By using both @samp{-nostdinc} and @samp{-I-}, you can limit the include-file
2464 search path to only those directories you specify explicitly.
2467 Do not predefine any nonstandard macros. (Including architecture flags).
2470 Run only the C preprocessor. Preprocess all the C source files
2471 specified and output the results to standard output or to the
2472 specified output file.
2475 Tell the preprocessor not to discard comments. Used with the
2479 Tell the preprocessor not to generate @samp{#line} directives.
2480 Used with the @samp{-E} option.
2483 @cindex dependencies, make
2485 Tell the preprocessor to output a rule suitable for @code{make}
2486 describing the dependencies of each object file. For each source file,
2487 the preprocessor outputs one @code{make}-rule whose target is the object
2488 file name for that source file and whose dependencies are all the
2489 @code{#include} header files it uses. This rule may be a single line or
2490 may be continued with @samp{\}-newline if it is long. The list of rules
2491 is printed on standard output instead of the preprocessed C program.
2493 @samp{-M} implies @samp{-E}.
2495 Another way to specify output of a @code{make} rule is by setting
2496 the environment variable @code{DEPENDENCIES_OUTPUT} (@pxref{Environment
2500 Like @samp{-M} but the output mentions only the user header files
2501 included with @samp{#include "@var{file}"}. System header files
2502 included with @samp{#include <@var{file}>} are omitted.
2505 Like @samp{-M} but the dependency information is written to a file made by
2506 replacing ".c" with ".d" at the end of the input file names.
2507 This is in addition to compiling the file as specified---@samp{-MD} does
2508 not inhibit ordinary compilation the way @samp{-M} does.
2510 In Mach, you can use the utility @code{md} to merge multiple dependency
2511 files into a single dependency file suitable for using with the @samp{make}
2515 Like @samp{-MD} except mention only user header files, not system
2519 Treat missing header files as generated files and assume they live in the
2520 same directory as the source file. If you specify @samp{-MG}, you
2521 must also specify either @samp{-M} or @samp{-MM}. @samp{-MG} is not
2522 supported with @samp{-MD} or @samp{-MMD}.
2525 Print the name of each header file used, in addition to other normal
2528 @item -A@var{question}(@var{answer})
2529 Assert the answer @var{answer} for @var{question}, in case it is tested
2530 with a preprocessing conditional such as @samp{#if
2531 #@var{question}(@var{answer})}. @samp{-A-} disables the standard
2532 assertions that normally describe the target machine.
2535 Define macro @var{macro} with the string @samp{1} as its definition.
2537 @item -D@var{macro}=@var{defn}
2538 Define macro @var{macro} as @var{defn}. All instances of @samp{-D} on
2539 the command line are processed before any @samp{-U} options.
2542 Undefine macro @var{macro}. @samp{-U} options are evaluated after all
2543 @samp{-D} options, but before any @samp{-include} and @samp{-imacros}
2547 Tell the preprocessor to output only a list of the macro definitions
2548 that are in effect at the end of preprocessing. Used with the @samp{-E}
2552 Tell the preprocessing to pass all macro definitions into the output, in
2553 their proper sequence in the rest of the output.
2556 Like @samp{-dD} except that the macro arguments and contents are omitted.
2557 Only @samp{#define @var{name}} is included in the output.
2560 Support ANSI C trigraphs. The @samp{-ansi} option also has this effect.
2562 @item -Wp,@var{option}
2563 Pass @var{option} as an option to the preprocessor. If @var{option}
2564 contains commas, it is split into multiple options at the commas.
2567 @node Assembler Options
2568 @section Passing Options to the Assembler
2570 @c prevent bad page break with this line
2571 You can pass options to the assembler.
2574 @item -Wa,@var{option}
2575 Pass @var{option} as an option to the assembler. If @var{option}
2576 contains commas, it is split into multiple options at the commas.
2580 @section Options for Linking
2581 @cindex link options
2582 @cindex options, linking
2584 These options come into play when the compiler links object files into
2585 an executable output file. They are meaningless if the compiler is
2586 not doing a link step.
2590 @item @var{object-file-name}
2591 A file name that does not end in a special recognized suffix is
2592 considered to name an object file or library. (Object files are
2593 distinguished from libraries by the linker according to the file
2594 contents.) If linking is done, these object files are used as input
2600 If any of these options is used, then the linker is not run, and
2601 object file names should not be used as arguments. @xref{Overall
2605 @item -l@var{library}
2606 Search the library named @var{library} when linking.
2608 It makes a difference where in the command you write this option; the
2609 linker searches processes libraries and object files in the order they
2610 are specified. Thus, @samp{foo.o -lz bar.o} searches library @samp{z}
2611 after file @file{foo.o} but before @file{bar.o}. If @file{bar.o} refers
2612 to functions in @samp{z}, those functions may not be loaded.
2614 The linker searches a standard list of directories for the library,
2615 which is actually a file named @file{lib@var{library}.a}. The linker
2616 then uses this file as if it had been specified precisely by name.
2618 The directories searched include several standard system directories
2619 plus any that you specify with @samp{-L}.
2621 Normally the files found this way are library files---archive files
2622 whose members are object files. The linker handles an archive file by
2623 scanning through it for members which define symbols that have so far
2624 been referenced but not defined. But if the file that is found is an
2625 ordinary object file, it is linked in the usual fashion. The only
2626 difference between using an @samp{-l} option and specifying a file name
2627 is that @samp{-l} surrounds @var{library} with @samp{lib} and @samp{.a}
2628 and searches several directories.
2631 You need this special case of the @samp{-l} option in order to
2632 link an Objective C program.
2635 Do not use the standard system startup files when linking.
2636 The standard system libraries are used normally, unless @code{-nostdlib}
2637 or @code{-nodefaultlibs} is used.
2639 @item -nodefaultlibs
2640 Do not use the standard system libraries when linking.
2641 Only the libraries you specify will be passed to the linker.
2642 The standard startup files are used normally, unless @code{-nostartfiles}
2646 Do not use the standard system startup files or libraries when linking.
2647 No startup files and only the libraries you specify will be passed to
2650 @cindex @code{-lgcc}, use with @code{-nostdlib}
2651 @cindex @code{-nostdlib} and unresolved references
2652 @cindex unresolved references and @code{-nostdlib}
2653 @cindex @code{-lgcc}, use with @code{-nodefaultlibs}
2654 @cindex @code{-nodefaultlibs} and unresolved references
2655 @cindex unresolved references and @code{-nodefaultlibs}
2656 One of the standard libraries bypassed by @samp{-nostdlib} and
2657 @samp{-nodefaultlibs} is @file{libgcc.a}, a library of internal subroutines
2658 that GNU CC uses to overcome shortcomings of particular machines, or special
2659 needs for some languages.
2661 (@xref{Interface,,Interfacing to GNU CC Output}, for more discussion of
2665 (@xref{Interface,,Interfacing to GNU CC Output,gcc.info,Porting GNU CC},
2666 for more discussion of @file{libgcc.a}.)
2668 In most cases, you need @file{libgcc.a} even when you want to avoid
2669 other standard libraries. In other words, when you specify @samp{-nostdlib}
2670 or @samp{-nodefaultlibs} you should usually specify @samp{-lgcc} as well.
2671 This ensures that you have no unresolved references to internal GNU CC
2672 library subroutines. (For example, @samp{__main}, used to ensure C++
2673 constructors will be called; @pxref{Collect2,,@code{collect2}}.)
2676 Remove all symbol table and relocation information from the executable.
2679 On systems that support dynamic linking, this prevents linking with the shared
2680 libraries. On other systems, this option has no effect.
2683 Produce a shared object which can then be linked with other objects to
2684 form an executable. Not all systems support this option. You must
2685 also specify @samp{-fpic} or @samp{-fPIC} on some systems when
2686 you specify this option.
2689 Bind references to global symbols when building a shared object. Warn
2690 about any unresolved references (unless overridden by the link editor
2691 option @samp{-Xlinker -z -Xlinker defs}). Only a few systems support
2694 @item -Xlinker @var{option}
2695 Pass @var{option} as an option to the linker. You can use this to
2696 supply system-specific linker options which GNU CC does not know how to
2699 If you want to pass an option that takes an argument, you must use
2700 @samp{-Xlinker} twice, once for the option and once for the argument.
2701 For example, to pass @samp{-assert definitions}, you must write
2702 @samp{-Xlinker -assert -Xlinker definitions}. It does not work to write
2703 @samp{-Xlinker "-assert definitions"}, because this passes the entire
2704 string as a single argument, which is not what the linker expects.
2706 @item -Wl,@var{option}
2707 Pass @var{option} as an option to the linker. If @var{option} contains
2708 commas, it is split into multiple options at the commas.
2710 @item -u @var{symbol}
2711 Pretend the symbol @var{symbol} is undefined, to force linking of
2712 library modules to define it. You can use @samp{-u} multiple times with
2713 different symbols to force loading of additional library modules.
2716 @node Directory Options
2717 @section Options for Directory Search
2718 @cindex directory options
2719 @cindex options, directory search
2722 These options specify directories to search for header files, for
2723 libraries and for parts of the compiler:
2727 Add the directory @var{dir} to the head of the list of directories to be
2728 searched for header files. This can be used to override a system header
2729 file, substituting your own version, since these directories are
2730 searched before the system header file directories. If you use more
2731 than one @samp{-I} option, the directories are scanned in left-to-right
2732 order; the standard system directories come after.
2735 Any directories you specify with @samp{-I} options before the @samp{-I-}
2736 option are searched only for the case of @samp{#include "@var{file}"};
2737 they are not searched for @samp{#include <@var{file}>}.
2739 If additional directories are specified with @samp{-I} options after
2740 the @samp{-I-}, these directories are searched for all @samp{#include}
2741 directives. (Ordinarily @emph{all} @samp{-I} directories are used
2744 In addition, the @samp{-I-} option inhibits the use of the current
2745 directory (where the current input file came from) as the first search
2746 directory for @samp{#include "@var{file}"}. There is no way to
2747 override this effect of @samp{-I-}. With @samp{-I.} you can specify
2748 searching the directory which was current when the compiler was
2749 invoked. That is not exactly the same as what the preprocessor does
2750 by default, but it is often satisfactory.
2752 @samp{-I-} does not inhibit the use of the standard system directories
2753 for header files. Thus, @samp{-I-} and @samp{-nostdinc} are
2757 Add directory @var{dir} to the list of directories to be searched
2760 @item -B@var{prefix}
2761 This option specifies where to find the executables, libraries,
2762 include files, and data files of the compiler itself.
2764 The compiler driver program runs one or more of the subprograms
2765 @file{cpp}, @file{cc1}, @file{as} and @file{ld}. It tries
2766 @var{prefix} as a prefix for each program it tries to run, both with and
2767 without @samp{@var{machine}/@var{version}/} (@pxref{Target Options}).
2769 For each subprogram to be run, the compiler driver first tries the
2770 @samp{-B} prefix, if any. If that name is not found, or if @samp{-B}
2771 was not specified, the driver tries two standard prefixes, which are
2772 @file{/usr/lib/gcc/} and @file{/usr/local/lib/gcc-lib/}. If neither of
2773 those results in a file name that is found, the unmodified program
2774 name is searched for using the directories specified in your
2775 @samp{PATH} environment variable.
2777 @samp{-B} prefixes that effectively specify directory names also apply
2778 to libraries in the linker, because the compiler translates these
2779 options into @samp{-L} options for the linker. They also apply to
2780 includes files in the preprocessor, because the compiler translates these
2781 options into @samp{-isystem} options for the preprocessor. In this case,
2782 the compiler appends @samp{include} to the prefix.
2784 The run-time support file @file{libgcc.a} can also be searched for using
2785 the @samp{-B} prefix, if needed. If it is not found there, the two
2786 standard prefixes above are tried, and that is all. The file is left
2787 out of the link if it is not found by those means.
2789 Another way to specify a prefix much like the @samp{-B} prefix is to use
2790 the environment variable @code{GCC_EXEC_PREFIX}. @xref{Environment
2793 @item -specs=@var{file}
2794 Process @var{file} after the compiler reads in the standard @file{specs}
2795 file, in order to override the defaults that the @file{gcc} driver
2796 program uses when determining what switches to pass to @file{cc1},
2797 @file{cc1plus}, @file{as}, @file{ld}, etc. More than one
2798 @samp{-specs=}@var{file} can be specified on the command line, and they
2799 are processed in order, from left to right.
2802 @node Target Options
2803 @section Specifying Target Machine and Compiler Version
2804 @cindex target options
2805 @cindex cross compiling
2806 @cindex specifying machine version
2807 @cindex specifying compiler version and target machine
2808 @cindex compiler version, specifying
2809 @cindex target machine, specifying
2811 By default, GNU CC compiles code for the same type of machine that you
2812 are using. However, it can also be installed as a cross-compiler, to
2813 compile for some other type of machine. In fact, several different
2814 configurations of GNU CC, for different target machines, can be
2815 installed side by side. Then you specify which one to use with the
2818 In addition, older and newer versions of GNU CC can be installed side
2819 by side. One of them (probably the newest) will be the default, but
2820 you may sometimes wish to use another.
2823 @item -b @var{machine}
2824 The argument @var{machine} specifies the target machine for compilation.
2825 This is useful when you have installed GNU CC as a cross-compiler.
2827 The value to use for @var{machine} is the same as was specified as the
2828 machine type when configuring GNU CC as a cross-compiler. For
2829 example, if a cross-compiler was configured with @samp{configure
2830 i386v}, meaning to compile for an 80386 running System V, then you
2831 would specify @samp{-b i386v} to run that cross compiler.
2833 When you do not specify @samp{-b}, it normally means to compile for
2834 the same type of machine that you are using.
2836 @item -V @var{version}
2837 The argument @var{version} specifies which version of GNU CC to run.
2838 This is useful when multiple versions are installed. For example,
2839 @var{version} might be @samp{2.0}, meaning to run GNU CC version 2.0.
2841 The default version, when you do not specify @samp{-V}, is the last
2842 version of GNU CC that you installed.
2845 The @samp{-b} and @samp{-V} options actually work by controlling part of
2846 the file name used for the executable files and libraries used for
2847 compilation. A given version of GNU CC, for a given target machine, is
2848 normally kept in the directory @file{/usr/local/lib/gcc-lib/@var{machine}/@var{version}}.@refill
2850 Thus, sites can customize the effect of @samp{-b} or @samp{-V} either by
2851 changing the names of these directories or adding alternate names (or
2852 symbolic links). If in directory @file{/usr/local/lib/gcc-lib/} the
2853 file @file{80386} is a link to the file @file{i386v}, then @samp{-b
2854 80386} becomes an alias for @samp{-b i386v}.
2856 In one respect, the @samp{-b} or @samp{-V} do not completely change
2857 to a different compiler: the top-level driver program @code{gcc}
2858 that you originally invoked continues to run and invoke the other
2859 executables (preprocessor, compiler per se, assembler and linker)
2860 that do the real work. However, since no real work is done in the
2861 driver program, it usually does not matter that the driver program
2862 in use is not the one for the specified target and version.
2864 The only way that the driver program depends on the target machine is
2865 in the parsing and handling of special machine-specific options.
2866 However, this is controlled by a file which is found, along with the
2867 other executables, in the directory for the specified version and
2868 target machine. As a result, a single installed driver program adapts
2869 to any specified target machine and compiler version.
2871 The driver program executable does control one significant thing,
2872 however: the default version and target machine. Therefore, you can
2873 install different instances of the driver program, compiled for
2874 different targets or versions, under different names.
2876 For example, if the driver for version 2.0 is installed as @code{ogcc}
2877 and that for version 2.1 is installed as @code{gcc}, then the command
2878 @code{gcc} will use version 2.1 by default, while @code{ogcc} will use
2879 2.0 by default. However, you can choose either version with either
2880 command with the @samp{-V} option.
2882 @node Submodel Options
2883 @section Hardware Models and Configurations
2884 @cindex submodel options
2885 @cindex specifying hardware config
2886 @cindex hardware models and configurations, specifying
2887 @cindex machine dependent options
2889 Earlier we discussed the standard option @samp{-b} which chooses among
2890 different installed compilers for completely different target
2891 machines, such as Vax vs. 68000 vs. 80386.
2893 In addition, each of these target machine types can have its own
2894 special options, starting with @samp{-m}, to choose among various
2895 hardware models or configurations---for example, 68010 vs 68020,
2896 floating coprocessor or none. A single installed version of the
2897 compiler can compile for any model or configuration, according to the
2900 Some configurations of the compiler also support additional special
2901 options, usually for compatibility with other compilers on the same
2905 These options are defined by the macro @code{TARGET_SWITCHES} in the
2906 machine description. The default for the options is also defined by
2907 that macro, which enables you to change the defaults.
2921 * RS/6000 and PowerPC Options::
2926 * Intel 960 Options::
2927 * DEC Alpha Options::
2931 * System V Options::
2936 @node M680x0 Options
2937 @subsection M680x0 Options
2938 @cindex M680x0 options
2940 These are the @samp{-m} options defined for the 68000 series. The default
2941 values for these options depends on which style of 68000 was selected when
2942 the compiler was configured; the defaults for the most common choices are
2948 Generate output for a 68000. This is the default
2949 when the compiler is configured for 68000-based systems.
2951 Use this option for microcontrollers with a 68000 or EC000 core,
2952 including the 68008, 68302, 68306, 68307, 68322, 68328 and 68356.
2956 Generate output for a 68020. This is the default
2957 when the compiler is configured for 68020-based systems.
2960 Generate output containing 68881 instructions for floating point.
2961 This is the default for most 68020 systems unless @samp{-nfp} was
2962 specified when the compiler was configured.
2965 Generate output for a 68030. This is the default when the compiler is
2966 configured for 68030-based systems.
2969 Generate output for a 68040. This is the default when the compiler is
2970 configured for 68040-based systems.
2972 This option inhibits the use of 68881/68882 instructions that have to be
2973 emulated by software on the 68040. Use this option if your 68040 does not
2974 have code to emulate those instructions.
2977 Generate output for a 68060. This is the default when the compiler is
2978 configured for 68060-based systems.
2980 This option inhibits the use of 68020 and 68881/68882 instructions that
2981 have to be emulated by software on the 68060. Use this option if your 68060
2982 does not have code to emulate those instructions.
2985 Generate output for a CPU32. This is the default
2986 when the compiler is configured for CPU32-based systems.
2988 Use this option for microcontrollers with a
2989 CPU32 or CPU32+ core, including the 68330, 68331, 68332, 68333, 68334,
2990 68336, 68340, 68341, 68349 and 68360.
2993 Generate output for a 520X "coldfire" family cpu. This is the default
2994 when the compiler is configured for 520X-based systems.
2996 Use this option for microcontroller with a 5200 core, including
2997 the MCF5202, MCF5203, MCF5204 and MCF5202.
3001 Generate output for a 68040, without using any of the new instructions.
3002 This results in code which can run relatively efficiently on either a
3003 68020/68881 or a 68030 or a 68040. The generated code does use the
3004 68881 instructions that are emulated on the 68040.
3007 Generate output for a 68060, without using any of the new instructions.
3008 This results in code which can run relatively efficiently on either a
3009 68020/68881 or a 68030 or a 68040. The generated code does use the
3010 68881 instructions that are emulated on the 68060.
3013 Generate output containing Sun FPA instructions for floating point.
3016 Generate output containing library calls for floating point.
3017 @strong{Warning:} the requisite libraries are not available for all m68k
3018 targets. Normally the facilities of the machine's usual C compiler are
3019 used, but this can't be done directly in cross-compilation. You must
3020 make your own arrangements to provide suitable library functions for
3021 cross-compilation. The embedded targets @samp{m68k-*-aout} and
3022 @samp{m68k-*-coff} do provide software floating point support.
3025 Consider type @code{int} to be 16 bits wide, like @code{short int}.
3028 Do not use the bit-field instructions. The @samp{-m68000}, @samp{-mcpu32}
3029 and @samp{-m5200} options imply @w{@samp{-mnobitfield}}.
3032 Do use the bit-field instructions. The @samp{-m68020} option implies
3033 @samp{-mbitfield}. This is the default if you use a configuration
3034 designed for a 68020.
3037 Use a different function-calling convention, in which functions
3038 that take a fixed number of arguments return with the @code{rtd}
3039 instruction, which pops their arguments while returning. This
3040 saves one instruction in the caller since there is no need to pop
3041 the arguments there.
3043 This calling convention is incompatible with the one normally
3044 used on Unix, so you cannot use it if you need to call libraries
3045 compiled with the Unix compiler.
3047 Also, you must provide function prototypes for all functions that
3048 take variable numbers of arguments (including @code{printf});
3049 otherwise incorrect code will be generated for calls to those
3052 In addition, seriously incorrect code will result if you call a
3053 function with too many arguments. (Normally, extra arguments are
3054 harmlessly ignored.)
3056 The @code{rtd} instruction is supported by the 68010, 68020, 68030,
3057 68040, 68060 and CPU32 processors, but not by the 68000 or 5200.
3060 @itemx -mno-align-int
3061 Control whether GNU CC aligns @code{int}, @code{long}, @code{long long},
3062 @code{float}, @code{double}, and @code{long double} variables on a 32-bit
3063 boundary (@samp{-malign-int}) or a 16-bit boundary (@samp{-mno-align-int}).
3064 Aligning variables on 32-bit boundaries produces code that runs somewhat
3065 faster on processors with 32-bit busses at the expense of more memory.
3067 @strong{Warning:} if you use the @samp{-malign-int} switch, GNU CC will
3068 align structures containing the above types differently than
3069 most published application binary interface specifications for the m68k.
3074 @subsection VAX Options
3077 These @samp{-m} options are defined for the Vax:
3081 Do not output certain jump instructions (@code{aobleq} and so on)
3082 that the Unix assembler for the Vax cannot handle across long
3086 Do output those jump instructions, on the assumption that you
3087 will assemble with the GNU assembler.
3090 Output code for g-format floating point numbers instead of d-format.
3094 @subsection SPARC Options
3095 @cindex SPARC options
3097 These @samp{-m} switches are supported on the SPARC:
3102 Specify @samp{-mapp-regs} to generate output using the global registers
3103 2 through 4, which the SPARC SVR4 ABI reserves for applications. This
3106 To be fully SVR4 ABI compliant at the cost of some performance loss,
3107 specify @samp{-mno-app-regs}. You should compile libraries and system
3108 software with this option.
3112 Generate output containing floating point instructions. This is the
3117 Generate output containing library calls for floating point.
3118 @strong{Warning:} the requisite libraries are not available for all SPARC
3119 targets. Normally the facilities of the machine's usual C compiler are
3120 used, but this cannot be done directly in cross-compilation. You must make
3121 your own arrangements to provide suitable library functions for
3122 cross-compilation. The embedded targets @samp{sparc-*-aout} and
3123 @samp{sparclite-*-*} do provide software floating point support.
3125 @samp{-msoft-float} changes the calling convention in the output file;
3126 therefore, it is only useful if you compile @emph{all} of a program with
3127 this option. In particular, you need to compile @file{libgcc.a}, the
3128 library that comes with GNU CC, with @samp{-msoft-float} in order for
3131 @item -mhard-quad-float
3132 Generate output containing quad-word (long double) floating point
3135 @item -msoft-quad-float
3136 Generate output containing library calls for quad-word (long double)
3137 floating point instructions. The functions called are those specified
3138 in the SPARC ABI. This is the default.
3140 As of this writing, there are no sparc implementations that have hardware
3141 support for the quad-word floating point instructions. They all invoke
3142 a trap handler for one of these instructions, and then the trap handler
3143 emulates the effect of the instruction. Because of the trap handler overhead,
3144 this is much slower than calling the ABI library routines. Thus the
3145 @samp{-msoft-quad-float} option is the default.
3149 With @samp{-mepilogue} (the default), the compiler always emits code for
3150 function exit at the end of each function. Any function exit in
3151 the middle of the function (such as a return statement in C) will
3152 generate a jump to the exit code at the end of the function.
3154 With @samp{-mno-epilogue}, the compiler tries to emit exit code inline
3155 at every function exit.
3159 With @samp{-mflat}, the compiler does not generate save/restore instructions
3160 and will use a "flat" or single register window calling convention.
3161 This model uses %i7 as the frame pointer and is compatible with the normal
3162 register window model. Code from either may be intermixed.
3163 The local registers and the input registers (0-5) are still treated as
3164 "call saved" registers and will be saved on the stack as necessary.
3166 With @samp{-mno-flat} (the default), the compiler emits save/restore
3167 instructions (except for leaf functions) and is the normal mode of operation.
3169 @item -mno-unaligned-doubles
3170 @itemx -munaligned-doubles
3171 Assume that doubles have 8 byte alignment. This is the default.
3173 With @samp{-munaligned-doubles}, GNU CC assumes that doubles have 8 byte
3174 alignment only if they are contained in another type, or if they have an
3175 absolute address. Otherwise, it assumes they have 4 byte alignment.
3176 Specifying this option avoids some rare compatibility problems with code
3177 generated by other compilers. It is not the default because it results
3178 in a performance loss, especially for floating point code.
3182 These two options select variations on the SPARC architecture.
3184 By default (unless specifically configured for the Fujitsu SPARClite),
3185 GCC generates code for the v7 variant of the SPARC architecture.
3187 @samp{-mv8} will give you SPARC v8 code. The only difference from v7
3188 code is that the compiler emits the integer multiply and integer
3189 divide instructions which exist in SPARC v8 but not in SPARC v7.
3191 @samp{-msparclite} will give you SPARClite code. This adds the integer
3192 multiply, integer divide step and scan (@code{ffs}) instructions which
3193 exist in SPARClite but not in SPARC v7.
3195 These options are deprecated and will be deleted in GNU CC 2.9.
3196 They have been replaced with @samp{-mcpu=xxx}.
3200 These two options select the processor for which the code is optimised.
3202 With @samp{-mcypress} (the default), the compiler optimizes code for the
3203 Cypress CY7C602 chip, as used in the SparcStation/SparcServer 3xx series.
3204 This is also appropriate for the older SparcStation 1, 2, IPX etc.
3206 With @samp{-msupersparc} the compiler optimizes code for the SuperSparc cpu, as
3207 used in the SparcStation 10, 1000 and 2000 series. This flag also enables use
3208 of the full SPARC v8 instruction set.
3210 These options are deprecated and will be deleted in GNU CC 2.9.
3211 They have been replaced with @samp{-mcpu=xxx}.
3213 @item -mcpu=@var{cpu_type}
3214 Set the instruction set, register set, and instruction scheduling parameters
3215 for machine type @var{cpu_type}. Supported values for @var{cpu_type} are
3216 @samp{v7}, @samp{cypress}, @samp{v8}, @samp{supersparc}, @samp{sparclite},
3217 @samp{f930}, @samp{f934}, @samp{sparclet}, @samp{tsc701}, @samp{v9}, and
3220 Default instruction scheduling parameters are used for values that select
3221 an architecture and not an implementation. These are @samp{v7}, @samp{v8},
3222 @samp{sparclite}, @samp{sparclet}, @samp{v9}.
3224 Here is a list of each supported architecture and their supported
3230 sparclite: f930, f934
3235 @item -mtune=@var{cpu_type}
3236 Set the instruction scheduling parameters for machine type
3237 @var{cpu_type}, but do not set the instruction set or register set that the
3238 option @samp{-mcpu=}@var{cpu_type} would.
3240 The same values for @samp{-mcpu=}@var{cpu_type} are used for
3241 @samp{-mtune=}@*@var{cpu_type}, though the only useful values are those that
3242 select a particular cpu implementation: @samp{cypress}, @samp{supersparc},
3243 @samp{f930}, @samp{f934}, @samp{tsc701}, @samp{ultrasparc}.
3245 @item -malign-loops=@var{num}
3246 Align loops to a 2 raised to a @var{num} byte boundary. If
3247 @samp{-malign-loops} is not specified, the default is 2.
3249 @item -malign-jumps=@var{num}
3250 Align instructions that are only jumped to to a 2 raised to a @var{num}
3251 byte boundary. If @samp{-malign-jumps} is not specified, the default is 2.
3253 @item -malign-functions=@var{num}
3254 Align the start of functions to a 2 raised to @var{num} byte boundary.
3255 If @samp{-malign-functions} is not specified, the default is 2 if compiling
3256 for 32 bit sparc, and 5 if compiling for 64 bit sparc.
3260 These @samp{-m} switches are supported in addition to the above
3261 on the SPARCLET processor.
3264 @item -mlittle-endian
3265 Generate code for a processor running in little-endian mode.
3268 Treat register @code{%g0} as a normal register.
3269 GCC will continue to clobber it as necessary but will not assume
3270 it always reads as 0.
3272 @item -mbroken-saverestore
3273 Generate code that does not use non-trivial forms of the @code{save} and
3274 @code{restore} instructions. Early versions of the SPARCLET processor do
3275 not correctly handle @code{save} and @code{restore} instructions used with
3276 arguments. They correctly handle them used without arguments. A @code{save}
3277 instruction used without arguments increments the current window pointer
3278 but does not allocate a new stack frame. It is assumed that the window
3279 overflow trap handler will properly handle this case as will interrupt
3283 These @samp{-m} switches are supported in addition to the above
3284 on SPARC V9 processors in 64 bit environments.
3287 @item -mlittle-endian
3288 Generate code for a processor running in little-endian mode.
3292 Generate code for a 32 bit or 64 bit environment.
3293 The 32 bit environment sets int, long and pointer to 32 bits.
3294 The 64 bit environment sets int to 32 bits and long and pointer
3297 @item -mcmodel=medlow
3298 Generate code for the Medium/Low code model: the program must be linked
3299 in the low 32 bits of the address space. Pointers are 64 bits.
3300 Programs can be statically or dynamically linked.
3302 @item -mcmodel=medmid
3303 Generate code for the Medium/Middle code model: the program must be linked
3304 in the low 44 bits of the address space, the text segment must be less than
3305 2G bytes, and data segment must be within 2G of the text segment.
3306 Pointers are 64 bits.
3308 @item -mcmodel=medany
3309 Generate code for the Medium/Anywhere code model: the program may be linked
3310 anywhere in the address space, the text segment must be less than
3311 2G bytes, and data segment must be within 2G of the text segment.
3312 Pointers are 64 bits.
3314 @item -mcmodel=embmedany
3315 Generate code for the Medium/Anywhere code model for embedded systems:
3316 assume a 32 bit text and a 32 bit data segment, both starting anywhere
3317 (determined at link time). Register %g4 points to the base of the
3318 data segment. Pointers still 64 bits.
3319 Programs are statically linked, PIC is not supported.
3322 @itemx -mno-stack-bias
3323 With @samp{-mstack-bias}, GNU CC assumes that the stack pointer, and
3324 frame pointer if present, are offset by -2047 which must be added back
3325 when making stack frame references.
3326 Otherwise, assume no such offset is present.
3329 @node Convex Options
3330 @subsection Convex Options
3331 @cindex Convex options
3333 These @samp{-m} options are defined for Convex:
3337 Generate output for C1. The code will run on any Convex machine.
3338 The preprocessor symbol @code{__convex__c1__} is defined.
3341 Generate output for C2. Uses instructions not available on C1.
3342 Scheduling and other optimizations are chosen for max performance on C2.
3343 The preprocessor symbol @code{__convex_c2__} is defined.
3346 Generate output for C32xx. Uses instructions not available on C1.
3347 Scheduling and other optimizations are chosen for max performance on C32.
3348 The preprocessor symbol @code{__convex_c32__} is defined.
3351 Generate output for C34xx. Uses instructions not available on C1.
3352 Scheduling and other optimizations are chosen for max performance on C34.
3353 The preprocessor symbol @code{__convex_c34__} is defined.
3356 Generate output for C38xx. Uses instructions not available on C1.
3357 Scheduling and other optimizations are chosen for max performance on C38.
3358 The preprocessor symbol @code{__convex_c38__} is defined.
3361 Generate code which puts an argument count in the word preceding each
3362 argument list. This is compatible with regular CC, and a few programs
3363 may need the argument count word. GDB and other source-level debuggers
3364 do not need it; this info is in the symbol table.
3367 Omit the argument count word. This is the default.
3369 @item -mvolatile-cache
3370 Allow volatile references to be cached. This is the default.
3372 @item -mvolatile-nocache
3373 Volatile references bypass the data cache, going all the way to memory.
3374 This is only needed for multi-processor code that does not use standard
3375 synchronization instructions. Making non-volatile references to volatile
3376 locations will not necessarily work.
3379 Type long is 32 bits, the same as type int. This is the default.
3382 Type long is 64 bits, the same as type long long. This option is useless,
3383 because no library support exists for it.
3386 @node AMD29K Options
3387 @subsection AMD29K Options
3388 @cindex AMD29K options
3390 These @samp{-m} options are defined for the AMD Am29000:
3395 @cindex DW bit (29k)
3396 Generate code that assumes the @code{DW} bit is set, i.e., that byte and
3397 halfword operations are directly supported by the hardware. This is the
3402 Generate code that assumes the @code{DW} bit is not set.
3406 @cindex byte writes (29k)
3407 Generate code that assumes the system supports byte and halfword write
3408 operations. This is the default.
3412 Generate code that assumes the systems does not support byte and
3413 halfword write operations. @samp{-mnbw} implies @samp{-mndw}.
3417 @cindex memory model (29k)
3418 Use a small memory model that assumes that all function addresses are
3419 either within a single 256 KB segment or at an absolute address of less
3420 than 256k. This allows the @code{call} instruction to be used instead
3421 of a @code{const}, @code{consth}, @code{calli} sequence.
3425 Use the normal memory model: Generate @code{call} instructions only when
3426 calling functions in the same file and @code{calli} instructions
3427 otherwise. This works if each file occupies less than 256 KB but allows
3428 the entire executable to be larger than 256 KB. This is the default.
3431 Always use @code{calli} instructions. Specify this option if you expect
3432 a single file to compile into more than 256 KB of code.
3436 @cindex processor selection (29k)
3437 Generate code for the Am29050.
3441 Generate code for the Am29000. This is the default.
3443 @item -mkernel-registers
3444 @kindex -mkernel-registers
3445 @cindex kernel and user registers (29k)
3446 Generate references to registers @code{gr64-gr95} instead of to
3447 registers @code{gr96-gr127}. This option can be used when compiling
3448 kernel code that wants a set of global registers disjoint from that used
3451 Note that when this option is used, register names in @samp{-f} flags
3452 must use the normal, user-mode, names.
3454 @item -muser-registers
3455 @kindex -muser-registers
3456 Use the normal set of global registers, @code{gr96-gr127}. This is the
3460 @itemx -mno-stack-check
3461 @kindex -mstack-check
3462 @cindex stack checks (29k)
3463 Insert (or do not insert) a call to @code{__msp_check} after each stack
3464 adjustment. This is often used for kernel code.
3467 @itemx -mno-storem-bug
3468 @kindex -mstorem-bug
3469 @cindex storem bug (29k)
3470 @samp{-mstorem-bug} handles 29k processors which cannot handle the
3471 separation of a mtsrim insn and a storem instruction (most 29000 chips
3472 to date, but not the 29050).
3474 @item -mno-reuse-arg-regs
3475 @itemx -mreuse-arg-regs
3476 @kindex -mreuse-arg-regs
3477 @samp{-mno-reuse-arg-regs} tells the compiler to only use incoming argument
3478 registers for copying out arguments. This helps detect calling a function
3479 with fewer arguments than it was declared with.
3481 @item -mno-impure-text
3482 @itemx -mimpure-text
3483 @kindex -mimpure-text
3484 @samp{-mimpure-text}, used in addition to @samp{-shared}, tells the compiler to
3485 not pass @samp{-assert pure-text} to the linker when linking a shared object.
3488 @kindex -msoft-float
3489 Generate output containing library calls for floating point.
3490 @strong{Warning:} the requisite libraries are not part of GNU CC.
3491 Normally the facilities of the machine's usual C compiler are used, but
3492 this can't be done directly in cross-compilation. You must make your
3493 own arrangements to provide suitable library functions for
3498 @subsection ARM Options
3501 These @samp{-m} options are defined for Advanced RISC Machines (ARM)
3506 @kindex -mapcs-frame
3507 Generate a stack frame that is compliant with the ARM Procedure Call
3508 Standard for all functions, even if this is not strictly necessary for
3509 correct execution of the code. Specifying @samp{-fomit-frame-pointer}
3510 with this option will cause the stack frames not to be generated for
3511 leaf functions. The default is @samp{-mno-apcs-frame}.
3515 This is a synonym for @samp{-mapcs-frame}.
3519 Generate code for a processor running with a 26-bit program counter,
3520 and conforming to the function calling standards for the APCS 26-bit
3521 option. This option replaces the @samp{-m2} and @samp{-m3} options
3522 of previous releases of the compiler.
3526 Generate code for a processor running with a 32-bit program counter,
3527 and conforming to the function calling standards for the APCS 32-bit
3528 option. This option replaces the @samp{-m6} option of previous releases
3531 @item -mapcs-stack-check
3532 @kindex -mapcs-stack-check
3533 @kindex -mno-apcs-stack-check
3534 Generate code to check the amount of stack space available upon entry to
3535 every function (that actually uses some stack space). If there is
3536 insufficient space available then either the function
3537 @samp{__rt_stkovf_split_small} or @samp{__rt_stkovf_split_big} will be
3538 called, depending upon the amount of stack space required. The run time
3539 system is required to provide these functions. The default is
3540 @samp{-mno-apcs-stack-check}, since this produces smaller code.
3543 @kindex -mapcs-float
3544 @kindex -mno-apcs-float
3545 Pass floating point arguments using the float point registers. This is
3546 one of the variants of the APCS. This option is reccommended if the
3547 target hardware has a floating point unit or if a lot of floating point
3548 arithmetic is going to be performed by the code. The default is
3549 @samp{-mno-apcs-float}, since integer only code is slightly increased in
3550 size if @samp{-mapcs-float} is used.
3552 @item -mapcs-reentrant
3553 @kindex -mapcs-reentrant
3554 @kindex -mno-apcs-reentrant
3555 Generate reentrant, position independent code. This is the equivalent
3556 to specifying the @samp{-fpic} option. The default is
3557 @samp{-mno-apcs-reentrant}.
3559 @item -mthumb-interwork
3560 @kindex -mthumb-interwork
3561 @kindex -mno-thumb-interwork
3562 Generate code which supports calling between the ARM and THUMB
3563 instruction sets. Without this option the two instruction sets cannot
3564 be reliably used inside one program. The default is
3565 @samp{-mno-thumb-interwork}, since slightly larger code is generated
3566 when @samp{-mthumb-interwork} is specified.
3568 @item -mno-sched-prolog
3569 @kindex -mno-sched-prolog
3570 @kindex -msched-prolog
3571 Prevent the reordering of instructions in the function prolog, or the
3572 merging of those instruction with the instructions in the function's
3573 body. This means that all functions will start with a recognisable set
3574 of instructions (or in fact one of a chioce from a small set of
3575 different function prologues), and this information can be used to
3576 locate the start if functions inside an executable piece of code. The
3577 default is @samp{-msched-prolog}.
3580 Generate output containing floating point instructions. This is the
3584 Generate output containing library calls for floating point.
3585 @strong{Warning:} the requisite libraries are not available for all ARM
3586 targets. Normally the facilities of the machine's usual C compiler are
3587 used, but this cannot be done directly in cross-compilation. You must make
3588 your own arrangements to provide suitable library functions for
3591 @samp{-msoft-float} changes the calling convention in the output file;
3592 therefore, it is only useful if you compile @emph{all} of a program with
3593 this option. In particular, you need to compile @file{libgcc.a}, the
3594 library that comes with GNU CC, with @samp{-msoft-float} in order for
3597 @item -mlittle-endian
3598 Generate code for a processor running in little-endian mode. This is
3599 the default for all standard configurations.
3602 Generate code for a processor running in big-endian mode; the default is
3603 to compile code for a little-endian processor.
3605 @item -mwords-little-endian
3606 This option only applies when generating code for big-endian processors.
3607 Generate code for a little-endian word order but a big-endian byte
3608 order. That is, a byte order of the form @samp{32107654}. Note: this
3609 option should only be used if you require compatibility with code for
3610 big-endian ARM processors generated by versions of the compiler prior to
3613 @item -mshort-load-bytes
3614 @kindex -mshort-load-bytes
3615 Do not try to load half-words (eg @samp{short}s) by loading a word from
3616 an unaligned address. For some targets the MMU is configured to trap
3617 unaligned loads; use this option to generate code that is safe in these
3620 @item -mno-short-load-bytes
3621 @kindex -mno-short-load-bytes
3622 Use unaligned word loads to load half-words (eg @samp{short}s). This
3623 option produces more efficient code, but the MMU is sometimes configured
3624 to trap these instructions.
3626 @item -mshort-load-words
3627 @kindex -mshort-load-words
3628 This is a synonym for the @samp{-mno-short-load-bytes}.
3630 @item -mno-short-load-words
3631 @kindex -mno-short-load-words
3632 This is a synonym for the @samp{-mshort-load-bytes}.
3636 This option only applies to RISC iX. Emulate the native BSD-mode
3637 compiler. This is the default if @samp{-ansi} is not specified.
3641 This option only applies to RISC iX. Emulate the native X/Open-mode
3644 @item -mno-symrename
3645 @kindex -mno-symrename
3646 This option only applies to RISC iX. Do not run the assembler
3647 post-processor, @samp{symrename}, after code has been assembled.
3648 Normally it is necessary to modify some of the standard symbols in
3649 preparation for linking with the RISC iX C library; this option
3650 suppresses this pass. The post-processor is never run when the
3651 compiler is built for cross-compilation.
3655 This specifies the name of the target ARM processor. GCC uses this name
3656 to determine what kind of instructions it can use when generating
3657 assembly code. Permissable names are: arm2, arm250, arm3, arm6, arm60,
3658 arm600, arm610, arm620, arm7, arm7m, arm7d, arm7dm, arm7di, arm7dmi,
3659 arm70, arm700, arm700i, arm710, arm710c, arm7100, arm7500, arm7500fe,
3660 arm7tdmi, arm8, strongarm, strongarm110
3664 This specifies the name of the target ARM architecture. GCC uses this
3665 name to determine what kind of instructions it can use when generating
3666 assembly code. This option can be used in conjunction with or instead
3667 of the @samp{-mcpu=} option. Permissable names are: armv2, armv2a,
3668 armv3, armv3m, armv4, armv4t
3670 @item -mfpe=<number>
3672 This specifes the version of the floating point emulation available on
3673 the target. Permissable values are 2 and 3.
3675 @item -mstructure-size-boundary=<n>
3676 @kindex -mstructure-size-boundary
3677 The size of all structures and unions will be rounded up to a multiple
3678 of the number of bits set by this option. Permissable values are 8 and
3679 32. The default value varies for different toolchains. For the COFF
3680 targeted toolchain the default value is 8. Specifying the larger number
3681 can produced faster, more efficient code, but can also increase the size
3682 of the program. The two values are potentially incompatible. Code
3683 compiled with one value cannot necessarily expect to work with code or
3684 libraries compiled with the other value, if they exchange information
3685 using structures or unions. Programmers are encouraged to use the 32
3686 value as future versions of the toolchain may default to this value.
3691 @subsection Thumb Options
3692 @cindex Thumb Options
3696 @item -mthumb-interwork
3697 @kindex -mthumb-interwork
3698 @kindex -mno-thumb-interwork
3699 Generate code which supports calling between the THUMB and ARM
3700 instruction sets. Without this option the two instruction sets cannot
3701 be reliably used inside one program. The default is
3702 @samp{-mno-thumb-interwork}, since slightly smaller code is generated
3706 @kindex -mtpcs-frame
3707 @kindex -mno-tpcs-frame
3708 Generate a stack frame that is compliant with the Thumb Procedure Call
3709 Standard for all non-leaf functions. (A leaf function is one that does
3710 not call any other functions). The default is @samp{-mno-apcs-frame}.
3712 @item -mtpcs-leaf-frame
3713 @kindex -mtpcs-leaf-frame
3714 @kindex -mno-tpcs-leaf-frame
3715 Generate a stack frame that is compliant with the Thumb Procedure Call
3716 Standard for all leaf functions. (A leaf function is one that does
3717 not call any other functions). The default is @samp{-mno-apcs-leaf-frame}.
3719 @item -mlittle-endian
3720 @kindex -mlittle-endian
3721 Generate code for a processor running in little-endian mode. This is
3722 the default for all standard configurations.
3725 @kindex -mbig-endian
3726 Generate code for a processor running in big-endian mode.
3728 @item -mstructure-size-boundary=<n>
3729 @kindex -mstructure-size-boundary
3730 The size of all structures and unions will be rounded up to a multiple
3731 of the number of bits set by this option. Permissable values are 8 and
3732 32. The default value varies for different toolchains. For the COFF
3733 targeted toolchain the default value is 8. Specifying the larger number
3734 can produced faster, more efficient code, but can also increase the size
3735 of the program. The two values are potentially incompatible. Code
3736 compiled with one value cannot necessarily expect to work with code or
3737 libraries compiled with the other value, if they exchange information
3738 using structures or unions. Programmers are encouraged to use the 32
3739 value as future versions of the toolchain may default to this value.
3744 @node MN10300 Options
3745 @subsection MN10300 Options
3746 @cindex MN10300 options
3747 These @samp{-m} options are defined for Matsushita MN10300 architectures:
3751 Generate code to avoid bugs in the multiply instructions for the MN10300
3752 processors. This is the default.
3755 Do not generate code to avoid bugs in the multiply instructions for the
3759 @node M32R/D Options
3760 @subsection M32R/D Options
3761 @cindex M32R/D options
3763 These @samp{-m} options are defined for Mitsubishi M32R/D architectures:
3766 @item -mcode-model=small
3767 Assume all objects live in the lower 16MB of memory (so that their addresses
3768 can be loaded with the @code{ld24} instruction), and assume all subroutines
3769 are reachable with the @code{bl} instruction.
3770 This is the default.
3772 The addressability of a particular object can be set with the
3773 @code{model} attribute.
3775 @item -mcode-model=medium
3776 Assume objects may be anywhere in the 32 bit address space (the compiler
3777 will generate @code{seth/add3} instructions to load their addresses), and
3778 assume all subroutines are reachable with the @code{bl} instruction.
3780 @item -mcode-model=large
3781 Assume objects may be anywhere in the 32 bit address space (the compiler
3782 will generate @code{seth/add3} instructions to load their addresses), and
3783 assume subroutines may not be reachable with the @code{bl} instruction
3784 (the compiler will generate the much slower @code{seth/add3/jl}
3785 instruction sequence).
3788 Disable use of the small data area. Variables will be put into
3789 one of @samp{.data}, @samp{bss}, or @samp{.rodata} (unless the
3790 @code{section} attribute has been specified).
3791 This is the default.
3793 The small data area consists of sections @samp{.sdata} and @samp{.sbss}.
3794 Objects may be explicitly put in the small data area with the
3795 @code{section} attribute using one of these sections.
3798 Put small global and static data in the small data area, but do not
3799 generate special code to reference them.
3802 Put small global and static data in the small data area, and generate
3803 special instructions to reference them.
3806 @cindex smaller data references
3807 Put global and static objects less than or equal to @var{num} bytes
3808 into the small data or bss sections instead of the normal data or bss
3809 sections. The default value of @var{num} is 8.
3810 The @samp{-msdata} option must be set to one of @samp{sdata} or @samp{use}
3811 for this option to have any effect.
3813 All modules should be compiled with the same @samp{-G @var{num}} value.
3814 Compiling with different values of @var{num} may or may not work; if it
3815 doesn't the linker will give an error message - incorrect code will not be
3821 @subsection M88K Options
3822 @cindex M88k options
3824 These @samp{-m} options are defined for Motorola 88k architectures:
3829 Generate code that works well on both the m88100 and the
3834 Generate code that works best for the m88100, but that also
3839 Generate code that works best for the m88110, and may not run
3844 Obsolete option to be removed from the next revision.
3847 @item -midentify-revision
3848 @kindex -midentify-revision
3850 @cindex identifying source, compiler (88k)
3851 Include an @code{ident} directive in the assembler output recording the
3852 source file name, compiler name and version, timestamp, and compilation
3855 @item -mno-underscores
3856 @kindex -mno-underscores
3857 @cindex underscores, avoiding (88k)
3858 In assembler output, emit symbol names without adding an underscore
3859 character at the beginning of each name. The default is to use an
3860 underscore as prefix on each name.
3862 @item -mocs-debug-info
3863 @itemx -mno-ocs-debug-info
3864 @kindex -mocs-debug-info
3865 @kindex -mno-ocs-debug-info
3867 @cindex debugging, 88k OCS
3868 Include (or omit) additional debugging information (about registers used
3869 in each stack frame) as specified in the 88open Object Compatibility
3870 Standard, ``OCS''. This extra information allows debugging of code that
3871 has had the frame pointer eliminated. The default for DG/UX, SVr4, and
3872 Delta 88 SVr3.2 is to include this information; other 88k configurations
3873 omit this information by default.
3875 @item -mocs-frame-position
3876 @kindex -mocs-frame-position
3877 @cindex register positions in frame (88k)
3878 When emitting COFF debugging information for automatic variables and
3879 parameters stored on the stack, use the offset from the canonical frame
3880 address, which is the stack pointer (register 31) on entry to the
3881 function. The DG/UX, SVr4, Delta88 SVr3.2, and BCS configurations use
3882 @samp{-mocs-frame-position}; other 88k configurations have the default
3883 @samp{-mno-ocs-frame-position}.
3885 @item -mno-ocs-frame-position
3886 @kindex -mno-ocs-frame-position
3887 @cindex register positions in frame (88k)
3888 When emitting COFF debugging information for automatic variables and
3889 parameters stored on the stack, use the offset from the frame pointer
3890 register (register 30). When this option is in effect, the frame
3891 pointer is not eliminated when debugging information is selected by the
3894 @item -moptimize-arg-area
3895 @itemx -mno-optimize-arg-area
3896 @kindex -moptimize-arg-area
3897 @kindex -mno-optimize-arg-area
3898 @cindex arguments in frame (88k)
3899 Control how function arguments are stored in stack frames.
3900 @samp{-moptimize-arg-area} saves space by optimizing them, but this
3901 conflicts with the 88open specifications. The opposite alternative,
3902 @samp{-mno-optimize-arg-area}, agrees with 88open standards. By default
3903 GNU CC does not optimize the argument area.
3905 @item -mshort-data-@var{num}
3906 @kindex -mshort-data-@var{num}
3907 @cindex smaller data references (88k)
3908 @cindex r0-relative references (88k)
3909 Generate smaller data references by making them relative to @code{r0},
3910 which allows loading a value using a single instruction (rather than the
3911 usual two). You control which data references are affected by
3912 specifying @var{num} with this option. For example, if you specify
3913 @samp{-mshort-data-512}, then the data references affected are those
3914 involving displacements of less than 512 bytes.
3915 @samp{-mshort-data-@var{num}} is not effective for @var{num} greater
3918 @item -mserialize-volatile
3919 @kindex -mserialize-volatile
3920 @itemx -mno-serialize-volatile
3921 @kindex -mno-serialize-volatile
3922 @cindex sequential consistency on 88k
3923 Do, or don't, generate code to guarantee sequential consistency
3924 of volatile memory references. By default, consistency is
3927 The order of memory references made by the MC88110 processor does
3928 not always match the order of the instructions requesting those
3929 references. In particular, a load instruction may execute before
3930 a preceding store instruction. Such reordering violates
3931 sequential consistency of volatile memory references, when there
3932 are multiple processors. When consistency must be guaranteed,
3933 GNU C generates special instructions, as needed, to force
3934 execution in the proper order.
3936 The MC88100 processor does not reorder memory references and so
3937 always provides sequential consistency. However, by default, GNU
3938 C generates the special instructions to guarantee consistency
3939 even when you use @samp{-m88100}, so that the code may be run on an
3940 MC88110 processor. If you intend to run your code only on the
3941 MC88100 processor, you may use @samp{-mno-serialize-volatile}.
3943 The extra code generated to guarantee consistency may affect the
3944 performance of your application. If you know that you can safely
3945 forgo this guarantee, you may use @samp{-mno-serialize-volatile}.
3951 @cindex assembler syntax, 88k
3953 Turn on (@samp{-msvr4}) or off (@samp{-msvr3}) compiler extensions
3954 related to System V release 4 (SVr4). This controls the following:
3958 Which variant of the assembler syntax to emit.
3960 @samp{-msvr4} makes the C preprocessor recognize @samp{#pragma weak}
3961 that is used on System V release 4.
3963 @samp{-msvr4} makes GNU CC issue additional declaration directives used in
3967 @samp{-msvr4} is the default for the m88k-motorola-sysv4 and
3968 m88k-dg-dgux m88k configurations. @samp{-msvr3} is the default for all
3969 other m88k configurations.
3971 @item -mversion-03.00
3972 @kindex -mversion-03.00
3973 This option is obsolete, and is ignored.
3974 @c ??? which asm syntax better for GAS? option there too?
3976 @item -mno-check-zero-division
3977 @itemx -mcheck-zero-division
3978 @kindex -mno-check-zero-division
3979 @kindex -mcheck-zero-division
3980 @cindex zero division on 88k
3981 Do, or don't, generate code to guarantee that integer division by
3982 zero will be detected. By default, detection is guaranteed.
3984 Some models of the MC88100 processor fail to trap upon integer
3985 division by zero under certain conditions. By default, when
3986 compiling code that might be run on such a processor, GNU C
3987 generates code that explicitly checks for zero-valued divisors
3988 and traps with exception number 503 when one is detected. Use of
3989 mno-check-zero-division suppresses such checking for code
3990 generated to run on an MC88100 processor.
3992 GNU C assumes that the MC88110 processor correctly detects all
3993 instances of integer division by zero. When @samp{-m88110} is
3994 specified, both @samp{-mcheck-zero-division} and
3995 @samp{-mno-check-zero-division} are ignored, and no explicit checks for
3996 zero-valued divisors are generated.
3998 @item -muse-div-instruction
3999 @kindex -muse-div-instruction
4000 @cindex divide instruction, 88k
4001 Use the div instruction for signed integer division on the
4002 MC88100 processor. By default, the div instruction is not used.
4004 On the MC88100 processor the signed integer division instruction
4005 div) traps to the operating system on a negative operand. The
4006 operating system transparently completes the operation, but at a
4007 large cost in execution time. By default, when compiling code
4008 that might be run on an MC88100 processor, GNU C emulates signed
4009 integer division using the unsigned integer division instruction
4010 divu), thereby avoiding the large penalty of a trap to the
4011 operating system. Such emulation has its own, smaller, execution
4012 cost in both time and space. To the extent that your code's
4013 important signed integer division operations are performed on two
4014 nonnegative operands, it may be desirable to use the div
4015 instruction directly.
4017 On the MC88110 processor the div instruction (also known as the
4018 divs instruction) processes negative operands without trapping to
4019 the operating system. When @samp{-m88110} is specified,
4020 @samp{-muse-div-instruction} is ignored, and the div instruction is used
4021 for signed integer division.
4023 Note that the result of dividing INT_MIN by -1 is undefined. In
4024 particular, the behavior of such a division with and without
4025 @samp{-muse-div-instruction} may differ.
4027 @item -mtrap-large-shift
4028 @itemx -mhandle-large-shift
4029 @kindex -mtrap-large-shift
4030 @kindex -mhandle-large-shift
4031 @cindex bit shift overflow (88k)
4032 @cindex large bit shifts (88k)
4033 Include code to detect bit-shifts of more than 31 bits; respectively,
4034 trap such shifts or emit code to handle them properly. By default GNU CC
4035 makes no special provision for large bit shifts.
4037 @item -mwarn-passed-structs
4038 @kindex -mwarn-passed-structs
4039 @cindex structure passing (88k)
4040 Warn when a function passes a struct as an argument or result.
4041 Structure-passing conventions have changed during the evolution of the C
4042 language, and are often the source of portability problems. By default,
4043 GNU CC issues no such warning.
4046 @node RS/6000 and PowerPC Options
4047 @subsection IBM RS/6000 and PowerPC Options
4048 @cindex RS/6000 and PowerPC Options
4049 @cindex IBM RS/6000 and PowerPC Options
4051 These @samp{-m} options are defined for the IBM RS/6000 and PowerPC:
4059 @itemx -mpowerpc-gpopt
4060 @itemx -mno-powerpc-gpopt
4061 @itemx -mpowerpc-gfxopt
4062 @itemx -mno-powerpc-gfxopt
4066 @kindex -mpowerpc-gpopt
4067 @kindex -mpowerpc-gfxopt
4068 GNU CC supports two related instruction set architectures for the
4069 RS/6000 and PowerPC. The @dfn{POWER} instruction set are those
4070 instructions supported by the @samp{rios} chip set used in the original
4071 RS/6000 systems and the @dfn{PowerPC} instruction set is the
4072 architecture of the Motorola MPC5xx, MPC6xx, MPC8xx microprocessors, and
4073 the IBM 4xx microprocessors.
4075 Neither architecture is a subset of the other. However there is a
4076 large common subset of instructions supported by both. An MQ
4077 register is included in processors supporting the POWER architecture.
4079 You use these options to specify which instructions are available on the
4080 processor you are using. The default value of these options is
4081 determined when configuring GNU CC. Specifying the
4082 @samp{-mcpu=@var{cpu_type}} overrides the specification of these
4083 options. We recommend you use the @samp{-mcpu=@var{cpu_type}} option
4084 rather than the options listed above.
4086 The @samp{-mpower} option allows GNU CC to generate instructions that
4087 are found only in the POWER architecture and to use the MQ register.
4088 Specifying @samp{-mpower2} implies @samp{-power} and also allows GNU CC
4089 to generate instructions that are present in the POWER2 architecture but
4090 not the original POWER architecture.
4092 The @samp{-mpowerpc} option allows GNU CC to generate instructions that
4093 are found only in the 32-bit subset of the PowerPC architecture.
4094 Specifying @samp{-mpowerpc-gpopt} implies @samp{-mpowerpc} and also allows
4095 GNU CC to use the optional PowerPC architecture instructions in the
4096 General Purpose group, including floating-point square root. Specifying
4097 @samp{-mpowerpc-gfxopt} implies @samp{-mpowerpc} and also allows GNU CC to
4098 use the optional PowerPC architecture instructions in the Graphics
4099 group, including floating-point select.
4101 If you specify both @samp{-mno-power} and @samp{-mno-powerpc}, GNU CC
4102 will use only the instructions in the common subset of both
4103 architectures plus some special AIX common-mode calls, and will not use
4104 the MQ register. Specifying both @samp{-mpower} and @samp{-mpowerpc}
4105 permits GNU CC to use any instruction from either architecture and to
4106 allow use of the MQ register; specify this for the Motorola MPC601.
4108 @item -mnew-mnemonics
4109 @itemx -mold-mnemonics
4110 @kindex -mnew-mnemonics
4111 @kindex -mold-mnemonics
4112 Select which mnemonics to use in the generated assembler code.
4113 @samp{-mnew-mnemonics} requests output that uses the assembler mnemonics
4114 defined for the PowerPC architecture, while @samp{-mold-mnemonics}
4115 requests the assembler mnemonics defined for the POWER architecture.
4116 Instructions defined in only one architecture have only one mnemonic;
4117 GNU CC uses that mnemonic irrespective of which of these options is
4120 PowerPC assemblers support both the old and new mnemonics, as will later
4121 POWER assemblers. Current POWER assemblers only support the old
4122 mnemonics. Specify @samp{-mnew-mnemonics} if you have an assembler that
4123 supports them, otherwise specify @samp{-mold-mnemonics}.
4125 The default value of these options depends on how GNU CC was configured.
4126 Specifying @samp{-mcpu=@var{cpu_type}} sometimes overrides the value of
4127 these option. Unless you are building a cross-compiler, you should
4128 normally not specify either @samp{-mnew-mnemonics} or
4129 @samp{-mold-mnemonics}, but should instead accept the default.
4131 @item -mcpu=@var{cpu_type}
4132 Set architecture type, register usage, choice of mnemonics, and
4133 instruction scheduling parameters for machine type @var{cpu_type}.
4134 Supported values for @var{cpu_type} are @samp{rs6000}, @samp{rios1},
4135 @samp{rios2}, @samp{rsc}, @samp{601}, @samp{602}, @samp{603},
4136 @samp{603e}, @samp{604}, @samp{604e}, @samp{620}, @samp{power},
4137 @samp{power2}, @samp{powerpc}, @samp{403}, @samp{505}, @samp{801},
4138 @samp{821}, @samp{823}, and @samp{860} and @samp{common}.
4139 @samp{-mcpu=power}, @samp{-mcpu=power2}, and @samp{-mcpu=powerpc}
4140 specify generic POWER, POWER2 and pure PowerPC (i.e., not MPC601)
4141 architecture machine types, with an appropriate, generic processor model
4142 assumed for scheduling purposes.@refill
4144 @c overfull hbox here --bob 22 jul96
4145 @c original text between ignore ... end ignore
4147 Specifying any of the @samp{-mcpu=rios1}, @samp{-mcpu=rios2},
4148 @samp{-mcpu=rsc}, @samp{-mcpu=power}, or @samp{-mcpu=power2} options
4149 enables the @samp{-mpower} option and disables the @samp{-mpowerpc}
4150 option; @samp{-mcpu=601} enables both the @samp{-mpower} and
4151 @samp{-mpowerpc} options; all of @samp{-mcpu=602}, @samp{-mcpu=603},
4152 @samp{-mcpu=603e}, @samp{-mcpu=604}, @samp{-mcpu=604e},
4153 @samp{-mcpu=620}, @samp{-mcpu=403}, @samp{-mcpu=505}, @samp{-mcpu=801},
4154 @samp{-mcpu=821}, @samp{-mcpu=823}, @samp{-mcpu=860} and
4155 @samp{-mcpu=powerpc} enable the @samp{-mpowerpc} option and disable the
4156 @samp{-mpower} option; @samp{-mcpu=common} disables both the
4157 @samp{-mpower} and @samp{-mpowerpc} options.@refill
4159 @c changed paragraph
4160 Specifying any of the following options:
4161 @samp{-mcpu=rios1}, @samp{-mcpu=rios2}, @samp{-mcpu=rsc},
4162 @samp{-mcpu=power}, or @samp{-mcpu=power2}
4163 enables the @samp{-mpower} option and disables the @samp{-mpowerpc} option;
4164 @samp{-mcpu=601} enables both the @samp{-mpower} and @samp{-mpowerpc} options.
4165 All of @samp{-mcpu=602}, @samp{-mcpu=603}, @samp{-mcpu=603e},
4166 @samp{-mcpu=604}, @samp{-mcpu=620},
4167 enable the @samp{-mpowerpc} option and disable the @samp{-mpower} option.
4168 Exactly similarly, all of @samp{-mcpu=403},
4169 @samp{-mcpu=505}, @samp{-mcpu=821}, @samp{-mcpu=860} and @samp{-mcpu=powerpc}
4170 enable the @samp{-mpowerpc} option and disable the @samp{-mpower} option.
4171 @samp{-mcpu=common} disables both the
4172 @samp{-mpower} and @samp{-mpowerpc} options.@refill
4173 @c end changes to prevent overfull hboxes
4175 AIX versions 4 or greater selects @samp{-mcpu=common} by default, so
4176 that code will operate on all members of the RS/6000 and PowerPC
4177 families. In that case, GNU CC will use only the instructions in the
4178 common subset of both architectures plus some special AIX common-mode
4179 calls, and will not use the MQ register. GNU CC assumes a generic
4180 processor model for scheduling purposes.
4182 Specifying any of the options @samp{-mcpu=rios1}, @samp{-mcpu=rios2},
4183 @samp{-mcpu=rsc}, @samp{-mcpu=power}, or @samp{-mcpu=power2} also
4184 disables the @samp{new-mnemonics} option. Specifying @samp{-mcpu=601},
4185 @samp{-mcpu=602}, @samp{-mcpu=603}, @samp{-mcpu=603e}, @samp{-mcpu=604},
4186 @samp{620}, @samp{403}, or @samp{-mcpu=powerpc} also enables the
4187 @samp{new-mnemonics} option.@refill
4189 Specifying @samp{-mcpu=403}, @samp{-mcpu=821}, or @samp{-mcpu=860} also
4190 enables the @samp{-msoft-float} option.
4192 @item -mtune=@var{cpu_type}
4193 Set the instruction scheduling parameters for machine type
4194 @var{cpu_type}, but do not set the architecture type, register usage,
4195 choice of mnemonics like @samp{-mcpu=}@var{cpu_type} would. The same
4196 values for @var{cpu_type} are used for @samp{-mtune=}@var{cpu_type} as
4197 for @samp{-mcpu=}@var{cpu_type}. The @samp{-mtune=}@var{cpu_type}
4198 option overrides the @samp{-mcpu=}@var{cpu_type} option in terms of
4199 instruction scheduling parameters.
4202 @itemx -mno-fp-in-toc
4203 @itemx -mno-sum-in-toc
4204 @itemx -mminimal-toc
4205 Modify generation of the TOC (Table Of Contents), which is created for
4206 every executable file. The @samp{-mfull-toc} option is selected by
4207 default. In that case, GNU CC will allocate at least one TOC entry for
4208 each unique non-automatic variable reference in your program. GNU CC
4209 will also place floating-point constants in the TOC. However, only
4210 16,384 entries are available in the TOC.
4212 If you receive a linker error message that saying you have overflowed
4213 the available TOC space, you can reduce the amount of TOC space used
4214 with the @samp{-mno-fp-in-toc} and @samp{-mno-sum-in-toc} options.
4215 @samp{-mno-fp-in-toc} prevents GNU CC from putting floating-point
4216 constants in the TOC and @samp{-mno-sum-in-toc} forces GNU CC to
4217 generate code to calculate the sum of an address and a constant at
4218 run-time instead of putting that sum into the TOC. You may specify one
4219 or both of these options. Each causes GNU CC to produce very slightly
4220 slower and larger code at the expense of conserving TOC space.
4222 If you still run out of space in the TOC even when you specify both of
4223 these options, specify @samp{-mminimal-toc} instead. This option causes
4224 GNU CC to make only one TOC entry for every file. When you specify this
4225 option, GNU CC will produce code that is slower and larger but which
4226 uses extremely little TOC space. You may wish to use this option
4227 only on files that contain less frequently executed code. @refill
4231 On AIX, pass floating-point arguments to prototyped functions beyond the
4232 register save area (RSA) on the stack in addition to argument FPRs. The
4233 AIX calling convention was extended but not initially documented to
4234 handle an obscure K&R C case of calling a function that takes the
4235 address of its arguments with fewer arguments than declared. AIX XL
4236 compilers assume that floating point arguments which do not fit in the
4237 RSA are on the stack when they compile a subroutine without
4238 optimization. Because always storing floating-point arguments on the
4239 stack is inefficient and rarely needed, this option is not enabled by
4240 default and only is necessary when calling subroutines compiled by AIX
4241 XL compilers without optimization.
4244 Support @dfn{AIX Threads}. Link an application written to use
4245 @dfn{pthreads} with special libraries and startup code to enable the
4249 Support @dfn{IBM RS/6000 SP} @dfn{Parallel Environment} (PE). Link an
4250 application written to use message passing with special startup code to
4251 enable the application to run. The system must have PE installed in the
4252 standard location (@file{/usr/lpp/ppe.poe/}), or the @file{specs} file
4253 must be overridden with the @samp{-specs=} option to specify the
4254 appropriate directory location. The Parallel Environment does not
4255 support threads, so the @samp{-mpe} option and the @samp{-mthreads}
4256 option are incompatible.
4260 Generate code that does not use (uses) the floating-point register set.
4261 Software floating point emulation is provided if you use the
4262 @samp{-msoft-float} option, and pass the option to GNU CC when linking.
4265 @itemx -mno-multiple
4266 Generate code that uses (does not use) the load multiple word
4267 instructions and the store multiple word instructions. These
4268 instructions are generated by default on POWER systems, and not
4269 generated on PowerPC systems. Do not use @samp{-mmultiple} on little
4270 endian PowerPC systems, since those instructions do not work when the
4271 processor is in little endian mode.
4275 Generate code that uses (does not use) the load string instructions and the
4276 store string word instructions to save multiple registers and do small block
4277 moves. These instructions are generated by default on POWER systems, and not
4278 generated on PowerPC systems. Do not use @samp{-mstring} on little endian
4279 PowerPC systems, since those instructions do not work when the processor is in
4284 Generate code that uses (does not use) the load or store instructions
4285 that update the base register to the address of the calculated memory
4286 location. These instructions are generated by default. If you use
4287 @samp{-mno-update}, there is a small window between the time that the
4288 stack pointer is updated and the address of the previous frame is
4289 stored, which means code that walks the stack frame across interrupts or
4290 signals may get corrupted data.
4293 @itemx -mno-fused-madd
4294 Generate code that uses (does not use) the floating point multiply and
4295 accumulate instructions. These instructions are generated by default if
4296 hardware floating is used.
4298 @item -mno-bit-align
4300 On System V.4 and embedded PowerPC systems do not (do) force structures
4301 and unions that contain bit fields to be aligned to the base type of the
4304 For example, by default a structure containing nothing but 8
4305 @code{unsigned} bitfields of length 1 would be aligned to a 4 byte
4306 boundary and have a size of 4 bytes. By using @samp{-mno-bit-align},
4307 the structure would be aligned to a 1 byte boundary and be one byte in
4310 @item -mno-strict-align
4311 @itemx -mstrict-align
4312 On System V.4 and embedded PowerPC systems do not (do) assume that
4313 unaligned memory references will be handled by the system.
4316 @itemx -mno-relocatable
4317 On embedded PowerPC systems generate code that allows (does not allow)
4318 the program to be relocated to a different address at runtime. If you
4319 use @samp{-mrelocatable} on any module, all objects linked together must
4320 be compiled with @samp{-mrelocatable} or @samp{-mrelocatable-lib}.
4322 @item -mrelocatable-lib
4323 @itemx -mno-relocatable-lib
4324 On embedded PowerPC systems generate code that allows (does not allow)
4325 the program to be relocated to a different address at runtime. Modules
4326 compiled with @samp{-mrelocatable-lib} can be linked with either modules
4327 compiled without @samp{-mrelocatable} and @samp{-mrelocatable-lib} or
4328 with modules compiled with the @samp{-mrelocatable} options.
4332 On System V.4 and embedded PowerPC systems do not (do) assume that
4333 register 2 contains a pointer to a global area pointing to the addresses
4334 used in the program.
4336 @item -mno-traceback
4338 On embedded PowerPC systems do not (do) generate a traceback tag before
4339 the start of the function. This tag can be used by the debugger to
4340 identify where the start of a function is.
4343 @itemx -mlittle-endian
4344 On System V.4 and embedded PowerPC systems compile code for the
4345 processor in little endian mode. The @samp{-mlittle-endian} option is
4346 the same as @samp{-mlittle}.
4350 On System V.4 and embedded PowerPC systems compile code for the
4351 processor in big endian mode. The @samp{-mbig-endian} option is
4352 the same as @samp{-mbig}.
4355 On System V.4 and embedded PowerPC systems compile code using calling
4356 conventions that adheres to the March 1995 draft of the System V
4357 Application Binary Interface, PowerPC processor supplement. This is the
4358 default unless you configured GCC using @samp{powerpc-*-eabiaix}.
4360 @item -mcall-sysv-eabi
4361 Specify both @samp{-mcall-sysv} and @samp{-meabi} options.
4363 @item -mcall-sysv-noeabi
4364 Specify both @samp{-mcall-sysv} and @samp{-mno-eabi} options.
4367 On System V.4 and embedded PowerPC systems compile code using calling
4368 conventions that are similar to those used on AIX. This is the
4369 default if you configured GCC using @samp{powerpc-*-eabiaix}.
4371 @item -mcall-solaris
4372 On System V.4 and embedded PowerPC systems compile code for the Solaris
4376 On System V.4 and embedded PowerPC systems compile code for the
4377 Linux-based GNU system.
4380 @itemx -mno-prototype
4381 On System V.4 and embedded PowerPC systems assume that all calls to
4382 variable argument functions are properly prototyped. Otherwise, the
4383 compiler must insert an instruction before every non prototyped call to
4384 set or clear bit 6 of the condition code register (@var{CR}) to
4385 indicate whether floating point values were passed in the floating point
4386 registers in case the function takes a variable arguments. With
4387 @samp{-mprototype}, only calls to prototyped variable argument functions
4388 will set or clear the bit.
4391 On embedded PowerPC systems, assume that the startup module is called
4392 @file{sim-crt0.o} and that the standard C libraries are @file{libsim.a} and
4393 @file{libc.a}. This is the default for @samp{powerpc-*-eabisim}.
4397 On embedded PowerPC systems, assume that the startup module is called
4398 @file{crt0.o} and the standard C libraries are @file{libmvme.a} and
4402 On embedded PowerPC systems, assume that the startup module is called
4403 @file{crt0.o} and the standard C libraries are @file{libads.a} and
4407 On embedded PowerPC systems, assume that the startup module is called
4408 @file{crt0.o} and the standard C libraries are @file{libyk.a} and
4412 On embedded PowerPC systems, set the @var{PPC_EMB} bit in the ELF flags
4413 header to indicate that @samp{eabi} extended relocations are used.
4417 On System V.4 and embedded PowerPC systems do (do not) adhere to the
4418 Embedded Applications Binary Interface (eabi) which is a set of
4419 modifications to the System V.4 specifications. Selecting @code{-meabi}
4420 means that the stack is aligned to an 8 byte boundary, a function
4421 @code{__eabi} is called to from @code{main} to set up the eabi
4422 environment, and the @samp{-msdata} option can use both @code{r2} and
4423 @code{r13} to point to two separate small data areas. Selecting
4424 @code{-mno-eabi} means that the stack is aligned to a 16 byte boundary,
4425 do not call an initialization function from @code{main}, and the
4426 @samp{-msdata} option will only use @code{r13} to point to a single
4427 small data area. The @samp{-meabi} option is on by default if you
4428 configured GCC using one of the @samp{powerpc*-*-eabi*} options.
4431 On System V.4 and embedded PowerPC systems, put small initialized
4432 @code{const} global and static data in the @samp{.sdata2} section, which
4433 is pointed to by register @code{r2}. Put small initialized
4434 non-@code{const} global and static data in the @samp{.sdata} section,
4435 which is pointed to by register @code{r13}. Put small uninitialized
4436 global and static data in the @samp{.sbss} section, which is adjacent to
4437 the @samp{.sdata} section. The @samp{-msdata=eabi} option is
4438 incompatible with the @samp{-mrelocatable} option. The
4439 @samp{-msdata=eabi} option also sets the @samp{-memb} option.
4442 On System V.4 and embedded PowerPC systems, put small global and static
4443 data in the @samp{.sdata} section, which is pointed to by register
4444 @code{r13}. Put small uninitialized global and static data in the
4445 @samp{.sbss} section, which is adjacent to the @samp{.sdata} section.
4446 The @samp{-msdata=sysv} option is incompatible with the
4447 @samp{-mrelocatable} option.
4449 @item -msdata=default
4451 On System V.4 and embedded PowerPC systems, if @samp{-meabi} is used,
4452 compile code the same as @samp{-msdata=eabi}, otherwise compile code the
4453 same as @samp{-msdata=sysv}.
4456 On System V.4 and embedded PowerPC systems, put small global and static
4457 data in the @samp{.sdata} section. Put small uninitialized global and
4458 static data in the @samp{.sbss} section. Do not use register @code{r13}
4459 to address small data however. This is the default behavior unless
4460 other @samp{-msdata} options are used.
4464 On embedded PowerPC systems, put all initialized global and static data
4465 in the @samp{.data} section, and all uninitialized data in the
4466 @samp{.bss} section.
4469 @cindex smaller data references (PowerPC)
4470 @cindex .sdata/.sdata2 references (PowerPC)
4471 On embedded PowerPC systems, put global and static items less than or
4472 equal to @var{num} bytes into the small data or bss sections instead of
4473 the normal data or bss section. By default, @var{num} is 8. The
4474 @samp{-G @var{num}} switch is also passed to the linker.
4475 All modules should be compiled with the same @samp{-G @var{num}} value.
4478 @itemx -mno-regnames
4479 On System V.4 and embedded PowerPC systems do (do not) emit register
4480 names in the assembly language output using symbolic forms.
4483 @subsection IBM RT Options
4485 @cindex IBM RT options
4487 These @samp{-m} options are defined for the IBM RT PC:
4491 Use an in-line code sequence for integer multiplies. This is the
4494 @item -mcall-lib-mul
4495 Call @code{lmul$$} for integer multiples.
4497 @item -mfull-fp-blocks
4498 Generate full-size floating point data blocks, including the minimum
4499 amount of scratch space recommended by IBM. This is the default.
4501 @item -mminimum-fp-blocks
4502 Do not include extra scratch space in floating point data blocks. This
4503 results in smaller code, but slower execution, since scratch space must
4504 be allocated dynamically.
4506 @cindex @file{varargs.h} and RT PC
4507 @cindex @file{stdarg.h} and RT PC
4508 @item -mfp-arg-in-fpregs
4509 Use a calling sequence incompatible with the IBM calling convention in
4510 which floating point arguments are passed in floating point registers.
4511 Note that @code{varargs.h} and @code{stdargs.h} will not work with
4512 floating point operands if this option is specified.
4514 @item -mfp-arg-in-gregs
4515 Use the normal calling convention for floating point arguments. This is
4518 @item -mhc-struct-return
4519 Return structures of more than one word in memory, rather than in a
4520 register. This provides compatibility with the MetaWare HighC (hc)
4521 compiler. Use the option @samp{-fpcc-struct-return} for compatibility
4522 with the Portable C Compiler (pcc).
4524 @item -mnohc-struct-return
4525 Return some structures of more than one word in registers, when
4526 convenient. This is the default. For compatibility with the
4527 IBM-supplied compilers, use the option @samp{-fpcc-struct-return} or the
4528 option @samp{-mhc-struct-return}.
4532 @subsection MIPS Options
4533 @cindex MIPS options
4535 These @samp{-m} options are defined for the MIPS family of computers:
4538 @item -mcpu=@var{cpu type}
4539 Assume the defaults for the machine type @var{cpu type} when scheduling
4540 instructions. The choices for @var{cpu type} are @samp{r2000}, @samp{r3000},
4541 @samp{r4000}, @samp{r4400}, @samp{r4600}, and @samp{r6000}. While picking a
4542 specific @var{cpu type} will schedule things appropriately for that
4543 particular chip, the compiler will not generate any code that does not
4544 meet level 1 of the MIPS ISA (instruction set architecture) without
4545 the @samp{-mips2} or @samp{-mips3} switches being used.
4548 Issue instructions from level 1 of the MIPS ISA. This is the default.
4549 @samp{r3000} is the default @var{cpu type} at this ISA level.
4552 Issue instructions from level 2 of the MIPS ISA (branch likely, square
4553 root instructions). @samp{r6000} is the default @var{cpu type} at this
4557 Issue instructions from level 3 of the MIPS ISA (64 bit instructions).
4558 @samp{r4000} is the default @var{cpu type} at this ISA level.
4559 This option does not change the sizes of any of the C data types.
4562 Assume that 32 32-bit floating point registers are available. This is
4566 Assume that 32 64-bit floating point registers are available. This is
4567 the default when the @samp{-mips3} option is used.
4570 Assume that 32 32-bit general purpose registers are available. This is
4574 Assume that 32 64-bit general purpose registers are available. This is
4575 the default when the @samp{-mips3} option is used.
4578 Types long, int, and pointer are 64 bits. This works only if @samp{-mips3}
4582 Types long and pointer are 64 bits, and type int is 32 bits.
4583 This works only if @samp{-mips3} is also specified.
4586 Generate code for the MIPS assembler, and invoke @file{mips-tfile} to
4587 add normal debug information. This is the default for all
4588 platforms except for the OSF/1 reference platform, using the OSF/rose
4589 object format. If the either of the @samp{-gstabs} or @samp{-gstabs+}
4590 switches are used, the @file{mips-tfile} program will encapsulate the
4591 stabs within MIPS ECOFF.
4594 Generate code for the GNU assembler. This is the default on the OSF/1
4595 reference platform, using the OSF/rose object format. Also, this is
4596 the default if the configure option @samp{--with-gnu-as} is used.
4598 @item -msplit-addresses
4599 @itemx -mno-split-addresses
4600 Generate code to load the high and low parts of address constants separately.
4601 This allows @code{gcc} to optimize away redundant loads of the high order
4602 bits of addresses. This optimization requires GNU as and GNU ld.
4603 This optimization is enabled by default for some embedded targets where
4604 GNU as and GNU ld are standard.
4608 The @samp{-mrnames} switch says to output code using the MIPS software
4609 names for the registers, instead of the hardware names (ie, @var{a0}
4610 instead of @var{$4}). The only known assembler that supports this option
4611 is the Algorithmics assembler.
4615 The @samp{-mgpopt} switch says to write all of the data declarations
4616 before the instructions in the text section, this allows the MIPS
4617 assembler to generate one word memory references instead of using two
4618 words for short global or static data items. This is on by default if
4619 optimization is selected.
4623 For each non-inline function processed, the @samp{-mstats} switch
4624 causes the compiler to emit one line to the standard error file to
4625 print statistics about the program (number of registers saved, stack
4630 The @samp{-mmemcpy} switch makes all block moves call the appropriate
4631 string function (@samp{memcpy} or @samp{bcopy}) instead of possibly
4632 generating inline code.
4635 @itemx -mno-mips-tfile
4636 The @samp{-mno-mips-tfile} switch causes the compiler not
4637 postprocess the object file with the @file{mips-tfile} program,
4638 after the MIPS assembler has generated it to add debug support. If
4639 @file{mips-tfile} is not run, then no local variables will be
4640 available to the debugger. In addition, @file{stage2} and
4641 @file{stage3} objects will have the temporary file names passed to the
4642 assembler embedded in the object file, which means the objects will
4643 not compare the same. The @samp{-mno-mips-tfile} switch should only
4644 be used when there are bugs in the @file{mips-tfile} program that
4645 prevents compilation.
4648 Generate output containing library calls for floating point.
4649 @strong{Warning:} the requisite libraries are not part of GNU CC.
4650 Normally the facilities of the machine's usual C compiler are used, but
4651 this can't be done directly in cross-compilation. You must make your
4652 own arrangements to provide suitable library functions for
4656 Generate output containing floating point instructions. This is the
4657 default if you use the unmodified sources.
4660 @itemx -mno-abicalls
4661 Emit (or do not emit) the pseudo operations @samp{.abicalls},
4662 @samp{.cpload}, and @samp{.cprestore} that some System V.4 ports use for
4663 position independent code.
4666 @itemx -mno-long-calls
4667 Do all calls with the @samp{JALR} instruction, which requires
4668 loading up a function's address into a register before the call.
4669 You need to use this switch, if you call outside of the current
4670 512 megabyte segment to functions that are not through pointers.
4673 @itemx -mno-half-pic
4674 Put pointers to extern references into the data section and load them
4675 up, rather than put the references in the text section.
4677 @item -membedded-pic
4678 @itemx -mno-embedded-pic
4679 Generate PIC code suitable for some embedded systems. All calls are made
4680 using PC relative address, and all data is addressed using the $gp register.
4681 This requires GNU as and GNU ld which do most of the work.
4683 @item -membedded-data
4684 @itemx -mno-embedded-data
4685 Allocate variables to the read-only data section first if possible, then
4686 next in the small data section if possible, otherwise in data. This gives
4687 slightly slower code than the default, but reduces the amount of RAM required
4688 when executing, and thus may be preferred for some embedded systems.
4690 @item -msingle-float
4691 @itemx -mdouble-float
4692 The @samp{-msingle-float} switch tells gcc to assume that the floating
4693 point coprocessor only supports single precision operations, as on the
4694 @samp{r4650} chip. The @samp{-mdouble-float} switch permits gcc to use
4695 double precision operations. This is the default.
4699 Permit use of the @samp{mad}, @samp{madu} and @samp{mul} instructions,
4700 as on the @samp{r4650} chip.
4703 Turns on @samp{-msingle-float}, @samp{-mmad}, and, at least for now,
4707 Compile code for the processor in little endian mode.
4708 The requisite libraries are assumed to exist.
4711 Compile code for the processor in big endian mode.
4712 The requisite libraries are assumed to exist.
4715 @cindex smaller data references (MIPS)
4716 @cindex gp-relative references (MIPS)
4717 Put global and static items less than or equal to @var{num} bytes into
4718 the small data or bss sections instead of the normal data or bss
4719 section. This allows the assembler to emit one word memory reference
4720 instructions based on the global pointer (@var{gp} or @var{$28}),
4721 instead of the normal two words used. By default, @var{num} is 8 when
4722 the MIPS assembler is used, and 0 when the GNU assembler is used. The
4723 @samp{-G @var{num}} switch is also passed to the assembler and linker.
4724 All modules should be compiled with the same @samp{-G @var{num}}
4728 Tell the MIPS assembler to not run it's preprocessor over user
4729 assembler files (with a @samp{.s} suffix) when assembling them.
4733 These options are defined by the macro
4734 @code{TARGET_SWITCHES} in the machine description. The default for the
4735 options is also defined by that macro, which enables you to change the
4740 @subsection Intel 386 Options
4741 @cindex i386 Options
4742 @cindex Intel 386 Options
4744 These @samp{-m} options are defined for the i386 family of computers:
4747 @item -mcpu=@var{cpu type}
4748 Assume the defaults for the machine type @var{cpu type} when scheduling
4749 instructions. The choices for @var{cpu type} are: @samp{i386},
4750 @samp{i486}, @samp{i586} (@samp{pentium}), @samp{pentium}, @samp{i686}
4751 (@samp{pentiumpro}) and @samp{pentiumpro}. While picking a specific
4752 @var{cpu type} will schedule things appropriately for that particular
4753 chip, the compiler will not generate any code that does not run on the
4754 i386 without the @samp{-march=@var{cpu type}} option being used.
4756 @item -march=@var{cpu type}
4757 Generate instructions for the machine type @var{cpu type}. The choices
4758 for @var{cpu type} are: @samp{i386}, @samp{i486}, @samp{pentium}, and
4759 @samp{pentiumpro}. Specifying @samp{-march=@var{cpu type}} implies
4760 @samp{-mcpu=@var{cpu type}}.
4766 Synonyms for -mcpu=i386, -mcpu=i486, -mcpu=pentium, and -mcpu=pentiumpro
4771 Control whether or not the compiler uses IEEE floating point
4772 comparisons. These handle correctly the case where the result of a
4773 comparison is unordered.
4776 Generate output containing library calls for floating point.
4777 @strong{Warning:} the requisite libraries are not part of GNU CC.
4778 Normally the facilities of the machine's usual C compiler are used, but
4779 this can't be done directly in cross-compilation. You must make your
4780 own arrangements to provide suitable library functions for
4783 On machines where a function returns floating point results in the 80387
4784 register stack, some floating point opcodes may be emitted even if
4785 @samp{-msoft-float} is used.
4787 @item -mno-fp-ret-in-387
4788 Do not use the FPU registers for return values of functions.
4790 The usual calling convention has functions return values of types
4791 @code{float} and @code{double} in an FPU register, even if there
4792 is no FPU. The idea is that the operating system should emulate
4795 The option @samp{-mno-fp-ret-in-387} causes such values to be returned
4796 in ordinary CPU registers instead.
4798 @item -mno-fancy-math-387
4799 Some 387 emulators do not support the @code{sin}, @code{cos} and
4800 @code{sqrt} instructions for the 387. Specify this option to avoid
4801 generating those instructions. This option is the default on FreeBSD.
4802 As of revision 2.6.1, these instructions are not generated unless you
4803 also use the @samp{-ffast-math} switch.
4805 @item -malign-double
4806 @itemx -mno-align-double
4807 Control whether GNU CC aligns @code{double}, @code{long double}, and
4808 @code{long long} variables on a two word boundary or a one word
4809 boundary. Aligning @code{double} variables on a two word boundary will
4810 produce code that runs somewhat faster on a @samp{Pentium} at the
4811 expense of more memory.
4813 @strong{Warning:} if you use the @samp{-malign-double} switch,
4814 structures containing the above types will be aligned differently than
4815 the published application binary interface specifications for the 386.
4818 @itemx -mno-svr3-shlib
4819 Control whether GNU CC places uninitialized locals into @code{bss} or
4820 @code{data}. @samp{-msvr3-shlib} places these locals into @code{bss}.
4821 These options are meaningful only on System V Release 3.
4823 @item -mno-wide-multiply
4824 @itemx -mwide-multiply
4825 Control whether GNU CC uses the @code{mul} and @code{imul} that produce
4826 64 bit results in @code{eax:edx} from 32 bit operands to do @code{long
4827 long} multiplies and 32-bit division by constants.
4830 Use a different function-calling convention, in which functions that
4831 take a fixed number of arguments return with the @code{ret} @var{num}
4832 instruction, which pops their arguments while returning. This saves one
4833 instruction in the caller since there is no need to pop the arguments
4836 You can specify that an individual function is called with this calling
4837 sequence with the function attribute @samp{stdcall}. You can also
4838 override the @samp{-mrtd} option by using the function attribute
4839 @samp{cdecl}. @xref{Function Attributes}
4841 @strong{Warning:} this calling convention is incompatible with the one
4842 normally used on Unix, so you cannot use it if you need to call
4843 libraries compiled with the Unix compiler.
4845 Also, you must provide function prototypes for all functions that
4846 take variable numbers of arguments (including @code{printf});
4847 otherwise incorrect code will be generated for calls to those
4850 In addition, seriously incorrect code will result if you call a
4851 function with too many arguments. (Normally, extra arguments are
4852 harmlessly ignored.)
4854 @item -mreg-alloc=@var{regs}
4855 Control the default allocation order of integer registers. The
4856 string @var{regs} is a series of letters specifying a register. The
4857 supported letters are: @code{a} allocate EAX; @code{b} allocate EBX;
4858 @code{c} allocate ECX; @code{d} allocate EDX; @code{S} allocate ESI;
4859 @code{D} allocate EDI; @code{B} allocate EBP.
4861 @item -mregparm=@var{num}
4862 Control how many registers are used to pass integer arguments. By
4863 default, no registers are used to pass arguments, and at most 3
4864 registers can be used. You can control this behavior for a specific
4865 function by using the function attribute @samp{regparm}. @xref{Function Attributes}
4867 @strong{Warning:} if you use this switch, and
4868 @var{num} is nonzero, then you must build all modules with the same
4869 value, including any libraries. This includes the system libraries and
4872 @item -malign-loops=@var{num}
4873 Align loops to a 2 raised to a @var{num} byte boundary. If
4874 @samp{-malign-loops} is not specified, the default is 2.
4876 @item -malign-jumps=@var{num}
4877 Align instructions that are only jumped to to a 2 raised to a @var{num}
4878 byte boundary. If @samp{-malign-jumps} is not specified, the default is
4879 2 if optimizing for a 386, and 4 if optimizing for a 486.
4881 @item -malign-functions=@var{num}
4882 Align the start of functions to a 2 raised to @var{num} byte boundary.
4883 If @samp{-malign-functions} is not specified, the default is 2 if optimizing
4884 for a 386, and 4 if optimizing for a 486.
4888 @subsection HPPA Options
4889 @cindex HPPA Options
4891 These @samp{-m} options are defined for the HPPA family of computers:
4895 Generate code for a PA 1.0 processor.
4898 Generate code for a PA 1.1 processor.
4901 Generate code suitable for big switch tables. Use this option only if
4902 the assembler/linker complain about out of range branches within a switch
4905 @item -mjump-in-delay
4906 Fill delay slots of function calls with unconditional jump instructions
4907 by modifying the return pointer for the function call to be the target
4908 of the conditional jump.
4910 @item -mdisable-fpregs
4911 Prevent floating point registers from being used in any manner. This is
4912 necessary for compiling kernels which perform lazy context switching of
4913 floating point registers. If you use this option and attempt to perform
4914 floating point operations, the compiler will abort.
4916 @item -mdisable-indexing
4917 Prevent the compiler from using indexing address modes. This avoids some
4918 rather obscure problems when compiling MIG generated code under MACH.
4920 @item -mno-space-regs
4921 Generate code that assumes the target has no space registers. This allows
4922 GCC to generate faster indirect calls and use unscaled index address modes.
4924 Such code is suitable for level 0 PA systems and kernels.
4926 @item -mfast-indirect-calls
4927 Generate code that assumes calls never cross space boundaries. This
4928 allows GCC to emit code which performs faster indirect calls.
4930 This option will not work in the presense of shared libraries or nested
4934 Optimize for space rather than execution time. Currently this only
4935 enables out of line function prologues and epilogues. This option is
4936 incompatible with PIC code generation and profiling.
4938 @item -mlong-load-store
4939 Generate 3-instruction load and store sequences as sometimes required by
4940 the HP-UX 10 linker. This is equivalent to the @samp{+k} option to
4943 @item -mportable-runtime
4944 Use the portable calling conventions proposed by HP for ELF systems.
4947 Enable the use of assembler directives only GAS understands.
4949 @item -mschedule=@var{cpu type}
4950 Schedule code according to the constraints for the machine type
4951 @var{cpu type}. The choices for @var{cpu type} are @samp{700} for
4952 7@var{n}0 machines, @samp{7100} for 7@var{n}5 machines, and @samp{7100LC}
4953 for 7@var{n}2 machines. @samp{7100} is the default for @var{cpu type}.
4955 Note the @samp{7100LC} scheduling information is incomplete and using
4956 @samp{7100LC} often leads to bad schedules. For now it's probably best
4957 to use @samp{7100} instead of @samp{7100LC} for the 7@var{n}2 machines.
4960 Enable the optimization pass in the HPUX linker. Note this makes symbolic
4961 debugging impossible. It also triggers a bug in the HPUX 8 and HPUX 9 linkers
4962 in which they give bogus error messages when linking some programs.
4965 Generate output containing library calls for floating point.
4966 @strong{Warning:} the requisite libraries are not available for all HPPA
4967 targets. Normally the facilities of the machine's usual C compiler are
4968 used, but this cannot be done directly in cross-compilation. You must make
4969 your own arrangements to provide suitable library functions for
4970 cross-compilation. The embedded target @samp{hppa1.1-*-pro}
4971 does provide software floating point support.
4973 @samp{-msoft-float} changes the calling convention in the output file;
4974 therefore, it is only useful if you compile @emph{all} of a program with
4975 this option. In particular, you need to compile @file{libgcc.a}, the
4976 library that comes with GNU CC, with @samp{-msoft-float} in order for
4980 @node Intel 960 Options
4981 @subsection Intel 960 Options
4983 These @samp{-m} options are defined for the Intel 960 implementations:
4986 @item -m@var{cpu type}
4987 Assume the defaults for the machine type @var{cpu type} for some of
4988 the other options, including instruction scheduling, floating point
4989 support, and addressing modes. The choices for @var{cpu type} are
4990 @samp{ka}, @samp{kb}, @samp{mc}, @samp{ca}, @samp{cf},
4991 @samp{sa}, and @samp{sb}.
4997 The @samp{-mnumerics} option indicates that the processor does support
4998 floating-point instructions. The @samp{-msoft-float} option indicates
4999 that floating-point support should not be assumed.
5001 @item -mleaf-procedures
5002 @itemx -mno-leaf-procedures
5003 Do (or do not) attempt to alter leaf procedures to be callable with the
5004 @code{bal} instruction as well as @code{call}. This will result in more
5005 efficient code for explicit calls when the @code{bal} instruction can be
5006 substituted by the assembler or linker, but less efficient code in other
5007 cases, such as calls via function pointers, or using a linker that doesn't
5008 support this optimization.
5011 @itemx -mno-tail-call
5012 Do (or do not) make additional attempts (beyond those of the
5013 machine-independent portions of the compiler) to optimize tail-recursive
5014 calls into branches. You may not want to do this because the detection of
5015 cases where this is not valid is not totally complete. The default is
5016 @samp{-mno-tail-call}.
5018 @item -mcomplex-addr
5019 @itemx -mno-complex-addr
5020 Assume (or do not assume) that the use of a complex addressing mode is a
5021 win on this implementation of the i960. Complex addressing modes may not
5022 be worthwhile on the K-series, but they definitely are on the C-series.
5023 The default is currently @samp{-mcomplex-addr} for all processors except
5027 @itemx -mno-code-align
5028 Align code to 8-byte boundaries for faster fetching (or don't bother).
5029 Currently turned on by default for C-series implementations only.
5032 @item -mclean-linkage
5033 @itemx -mno-clean-linkage
5034 These options are not fully implemented.
5038 @itemx -mic2.0-compat
5039 @itemx -mic3.0-compat
5040 Enable compatibility with iC960 v2.0 or v3.0.
5044 Enable compatibility with the iC960 assembler.
5046 @item -mstrict-align
5047 @itemx -mno-strict-align
5048 Do not permit (do permit) unaligned accesses.
5051 Enable structure-alignment compatibility with Intel's gcc release version
5052 1.3 (based on gcc 1.37). This option implies @samp{-mstrict-align}.
5055 @node DEC Alpha Options
5056 @subsection DEC Alpha Options
5058 These @samp{-m} options are defined for the DEC Alpha implementations:
5061 @item -mno-soft-float
5063 Use (do not use) the hardware floating-point instructions for
5064 floating-point operations. When @code{-msoft-float} is specified,
5065 functions in @file{libgcc1.c} will be used to perform floating-point
5066 operations. Unless they are replaced by routines that emulate the
5067 floating-point operations, or compiled in such a way as to call such
5068 emulations routines, these routines will issue floating-point
5069 operations. If you are compiling for an Alpha without floating-point
5070 operations, you must ensure that the library is built so as not to call
5073 Note that Alpha implementations without floating-point operations are
5074 required to have floating-point registers.
5078 Generate code that uses (does not use) the floating-point register set.
5079 @code{-mno-fp-regs} implies @code{-msoft-float}. If the floating-point
5080 register set is not used, floating point operands are passed in integer
5081 registers as if they were integers and floating-point results are passed
5082 in $0 instead of $f0. This is a non-standard calling sequence, so any
5083 function with a floating-point argument or return value called by code
5084 compiled with @code{-mno-fp-regs} must also be compiled with that
5087 A typical use of this option is building a kernel that does not use,
5088 and hence need not save and restore, any floating-point registers.
5091 The Alpha architecture implements floating-point hardware optimized for
5092 maximum performance. It is mostly compliant with the IEEE floating
5093 point standard. However, for full compliance, software assistance is
5094 required. This option generates code fully IEEE compliant code
5095 @emph{except} that the @var{inexact flag} is not maintained (see below).
5096 If this option is turned on, the CPP macro @code{_IEEE_FP} is defined
5097 during compilation. The option is a shorthand for: @samp{-D_IEEE_FP
5098 -mfp-trap-mode=su -mtrap-precision=i -mieee-conformant}. The resulting
5099 code is less efficient but is able to correctly support denormalized
5100 numbers and exceptional IEEE values such as not-a-number and plus/minus
5101 infinity. Other Alpha compilers call this option
5102 @code{-ieee_with_no_inexact}.
5104 @item -mieee-with-inexact
5105 @c overfull hbox here --bob 22 jul96
5106 @c original text between ignore ... end ignore
5108 This is like @samp{-mieee} except the generated code also maintains the
5109 IEEE @var{inexact flag}. Turning on this option causes the generated
5110 code to implement fully-compliant IEEE math. The option is a shorthand
5111 for @samp{-D_IEEE_FP -D_IEEE_FP_INEXACT} plus @samp{-mieee-conformant},
5112 @samp{-mfp-trap-mode=sui}, and @samp{-mtrap-precision=i}. On some Alpha
5113 implementations the resulting code may execute significantly slower than
5114 the code generated by default. Since there is very little code that
5115 depends on the @var{inexact flag}, you should normally not specify this
5116 option. Other Alpha compilers call this option
5117 @samp{-ieee_with_inexact}.
5119 @c changed paragraph
5120 This is like @samp{-mieee} except the generated code also maintains the
5121 IEEE @var{inexact flag}. Turning on this option causes the generated
5122 code to implement fully-compliant IEEE math. The option is a shorthand
5123 for @samp{-D_IEEE_FP -D_IEEE_FP_INEXACT} plus the three following:
5124 @samp{-mieee-conformant},
5125 @samp{-mfp-trap-mode=sui},
5126 and @samp{-mtrap-precision=i}.
5127 On some Alpha implementations the resulting code may execute
5128 significantly slower than the code generated by default. Since there
5129 is very little code that depends on the @var{inexact flag}, you should
5130 normally not specify this option. Other Alpha compilers call this
5131 option @samp{-ieee_with_inexact}.
5132 @c end changes to prevent overfull hboxes
5134 @item -mfp-trap-mode=@var{trap mode}
5135 This option controls what floating-point related traps are enabled.
5136 Other Alpha compilers call this option @samp{-fptm }@var{trap mode}.
5137 The trap mode can be set to one of four values:
5141 This is the default (normal) setting. The only traps that are enabled
5142 are the ones that cannot be disabled in software (e.g., division by zero
5146 In addition to the traps enabled by @samp{n}, underflow traps are enabled
5150 Like @samp{su}, but the instructions are marked to be safe for software
5151 completion (see Alpha architecture manual for details).
5154 Like @samp{su}, but inexact traps are enabled as well.
5157 @item -mfp-rounding-mode=@var{rounding mode}
5158 Selects the IEEE rounding mode. Other Alpha compilers call this option
5159 @samp{-fprm }@var{rounding mode}. The @var{rounding mode} can be one
5164 Normal IEEE rounding mode. Floating point numbers are rounded towards
5165 the nearest machine number or towards the even machine number in case
5169 Round towards minus infinity.
5172 Chopped rounding mode. Floating point numbers are rounded towards zero.
5175 Dynamic rounding mode. A field in the floating point control register
5176 (@var{fpcr}, see Alpha architecture reference manual) controls the
5177 rounding mode in effect. The C library initializes this register for
5178 rounding towards plus infinity. Thus, unless your program modifies the
5179 @var{fpcr}, @samp{d} corresponds to round towards plus infinity.@end table
5181 @item -mtrap-precision=@var{trap precision}
5182 In the Alpha architecture, floating point traps are imprecise. This
5183 means without software assistance it is impossible to recover from a
5184 floating trap and program execution normally needs to be terminated.
5185 GNU CC can generate code that can assist operating system trap handlers
5186 in determining the exact location that caused a floating point trap.
5187 Depending on the requirements of an application, different levels of
5188 precisions can be selected:
5192 Program precision. This option is the default and means a trap handler
5193 can only identify which program caused a floating point exception.
5196 Function precision. The trap handler can determine the function that
5197 caused a floating point exception.
5200 Instruction precision. The trap handler can determine the exact
5201 instruction that caused a floating point exception.
5204 Other Alpha compilers provide the equivalent options called
5205 @samp{-scope_safe} and @samp{-resumption_safe}.
5207 @item -mieee-conformant
5208 This option marks the generated code as IEEE conformant. You must not
5209 use this option unless you also specify @samp{-mtrap-precision=i} and either
5210 @samp{-mfp-trap-mode=su} or @samp{-mfp-trap-mode=sui}. Its only effect
5211 is to emit the line @samp{.eflag 48} in the function prologue of the
5212 generated assembly file. Under DEC Unix, this has the effect that
5213 IEEE-conformant math library routines will be linked in.
5215 @item -mbuild-constants
5216 Normally GNU CC examines a 32- or 64-bit integer constant to
5217 see if it can construct it from smaller constants in two or three
5218 instructions. If it cannot, it will output the constant as a literal and
5219 generate code to load it from the data segment at runtime.
5221 Use this option to require GNU CC to construct @emph{all} integer constants
5222 using code, even if it takes more instructions (the maximum is six).
5224 You would typically use this option to build a shared library dynamic
5225 loader. Itself a shared library, it must relocate itself in memory
5226 before it can find the variables and constants in its own data segment.
5230 Select whether to generate code to be assembled by the vendor-supplied
5231 assembler (@samp{-malpha-as}) or by the GNU assembler @samp{-mgas}.
5239 Indicate whether GNU CC should generate code to use the optional BWX,
5240 CIX, and MAX instruction sets. The default is to use the instruction sets
5241 supported by the CPU type specified via @samp{-mcpu=} option or that
5242 of the CPU on which GNU CC was built if none was specified.
5244 @item -mcpu=@var{cpu_type}
5245 Set the instruction set, register set, and instruction scheduling
5246 parameters for machine type @var{cpu_type}. You can specify either the
5247 @samp{EV} style name or the corresponding chip number. GNU CC
5248 supports scheduling parameters for the EV4 and EV5 family of processors
5249 and will choose the default values for the instruction set from
5250 the processor you specify. If you do not specify a processor type,
5251 GNU CC will default to the processor on which the compiler was built.
5253 Supported values for @var{cpu_type} are
5258 Schedules as an EV4 and has no instruction set extensions.
5262 Schedules as an EV5 and has no instruction set extensions.
5266 Schedules as an EV5 and supports the BWX extension.
5271 Schedules as an EV5 and supports the BWX and MAX extensions.
5275 Schedules as an EV5 (until Digital releases the scheduling parameters
5276 for the EV6) and supports the BWX, CIX, and MAX extensions.
5279 @item -mmemory-latency=@var{time}
5280 Sets the latency the scheduler should assume for typical memory
5281 references as seen by the application. This number is highly
5282 dependant on the memory access patterns used by the application
5283 and the size of the external cache on the machine.
5285 Valid options for @var{time} are
5289 A decimal number representing clock cycles.
5295 The compiler contains estimates of the number of clock cycles for
5296 ``typical'' EV4 & EV5 hardware for the Level 1, 2 & 3 caches
5297 (also called Dcache, Scache, and Bcache), as well as to main memory.
5298 Note that L3 is only valid for EV5.
5303 @node Clipper Options
5304 @subsection Clipper Options
5306 These @samp{-m} options are defined for the Clipper implementations:
5310 Produce code for a C300 Clipper processor. This is the default.
5313 Produce code for a C400 Clipper processor i.e. use floating point
5317 @node H8/300 Options
5318 @subsection H8/300 Options
5320 These @samp{-m} options are defined for the H8/300 implementations:
5324 Shorten some address references at link time, when possible; uses the
5325 linker option @samp{-relax}. @xref{H8/300,, @code{ld} and the H8/300,
5326 ld.info, Using ld}, for a fuller description.
5329 Generate code for the H8/300H.
5332 Generate code for the H8/S.
5335 Make @code{int} data 32 bits by default.
5338 On the h8/300h, use the same alignment rules as for the h8/300.
5339 The default for the h8/300h is to align longs and floats on 4 byte boundaries.
5340 @samp{-malign-300} causes them to be aligned on 2 byte boundaries.
5341 This option has no effect on the h8/300.
5345 @subsection SH Options
5347 These @samp{-m} options are defined for the SH implementations:
5351 Generate code for the SH1.
5354 Generate code for the SH2.
5357 Generate code for the SH3.
5360 Generate code for the SH3e.
5363 Compile code for the processor in big endian mode.
5366 Compile code for the processor in little endian mode.
5369 Shorten some address references at link time, when possible; uses the
5370 linker option @samp{-relax}.
5373 @node System V Options
5374 @subsection Options for System V
5376 These additional options are available on System V Release 4 for
5377 compatibility with other compilers on those systems:
5381 Create a shared object.
5382 It is recommended that @samp{-symbolic} or @samp{-shared} be used instead.
5385 Identify the versions of each tool used by the compiler, in a
5386 @code{.ident} assembler directive in the output.
5389 Refrain from adding @code{.ident} directives to the output file (this is
5392 @item -YP,@var{dirs}
5393 Search the directories @var{dirs}, and no others, for libraries
5394 specified with @samp{-l}.
5397 Look in the directory @var{dir} to find the M4 preprocessor.
5398 The assembler uses this option.
5399 @c This is supposed to go with a -Yd for predefined M4 macro files, but
5400 @c the generic assembler that comes with Solaris takes just -Ym.
5404 @subsection V850 Options
5405 @cindex V850 Options
5407 These @samp{-m} options are defined for V850 implementations:
5411 @itemx -mno-long-calls
5412 Treat all calls as being far away (near). If calls are assumed to be
5413 far away, the compiler will always load the functions address up into a
5414 register, and call indirect through the pointer.
5418 Do not optimize (do optimize) basic blocks that use the same index
5419 pointer 4 or more times to copy pointer into the @code{ep} register, and
5420 use the shorter @code{sld} and @code{sst} instructions. The @samp{-mep}
5421 option is on by default if you optimize.
5423 @item -mno-prolog-function
5424 @itemx -mprolog-function
5425 Do not use (do use) external functions to save and restore registers at
5426 the prolog and epilog of a function. The external functions are slower,
5427 but use less code space if more than one function saves the same number
5428 of registers. The @samp{-mprolog-function} option is on by default if
5432 Try to make the code as small as possible. At present, this just turns
5433 on the @samp{-mep} and @samp{-mprolog-function} options.
5436 Put static or global variables whose size is @var{n} bytes or less into
5437 the tiny data area that register @code{ep} points to. The tiny data
5438 area can hold up to 256 bytes in total (128 bytes for byte references).
5441 Put static or global variables whose size is @var{n} bytes or less into
5442 the small data area that register @code{gp} points to. The small data
5443 area can hold up to 64 kilobytes.
5446 Put static or global variables whose size is @var{n} bytes or less into
5447 the first 32 kilobytes of memory.
5450 Specify that the target processor is the V850.
5453 Generate code suitable for big switch tables. Use this option only if
5454 the assembler/linker complain about out of range branches within a switch
5459 @subsection ARC Options
5462 These options are defined for ARC implementations:
5466 Compile code for little endian mode. This is the default.
5469 Compile code for big endian mode.
5472 Prepend the name of the cpu to all public symbol names.
5473 In multiple-processor systems, there are many ARC variants with different
5474 instruction and register set characteristics. This flag prevents code
5475 compiled for one cpu to be linked with code compiled for another.
5476 No facility exists for handling variants that are "almost identical".
5477 This is an all or nothing option.
5479 @item -mcpu=@var{cpu}
5480 Compile code for ARC variant @var{cpu}.
5481 Which variants are supported depend on the configuration.
5482 All variants support @samp{-mcpu=base}, this is the default.
5484 @item -mtext=@var{text section}
5485 @item -mdata=@var{data section}
5486 @item -mrodata=@var{readonly data section}
5487 Put functions, data, and readonly data in @var{text section},
5488 @var{data section}, and @var{readonly data section} respectively
5489 by default. This can be overridden with the @code{section} attribute.
5490 @xref{Variable Attributes}
5495 @node Code Gen Options
5496 @section Options for Code Generation Conventions
5497 @cindex code generation conventions
5498 @cindex options, code generation
5499 @cindex run-time options
5501 These machine-independent options control the interface conventions
5502 used in code generation.
5504 Most of them have both positive and negative forms; the negative form
5505 of @samp{-ffoo} would be @samp{-fno-foo}. In the table below, only
5506 one of the forms is listed---the one which is not the default. You
5507 can figure out the other form by either removing @samp{no-} or adding
5512 Enable exception handling, and generate extra code needed to propagate
5513 exceptions. If you do not specify this option, GNU CC enables it by
5514 default for languages like C++ that normally require exception handling,
5515 and disabled for languages like C that do not normally require it.
5516 However, when compiling C code that needs to interoperate properly with
5517 exception handlers written in C++, you may need to enable this option.
5518 You may also wish to disable this option is you are compiling older C++
5519 programs that don't use exception handling.
5521 @item -fpcc-struct-return
5522 Return ``short'' @code{struct} and @code{union} values in memory like
5523 longer ones, rather than in registers. This convention is less
5524 efficient, but it has the advantage of allowing intercallability between
5525 GNU CC-compiled files and files compiled with other compilers.
5527 The precise convention for returning structures in memory depends
5528 on the target configuration macros.
5530 Short structures and unions are those whose size and alignment match
5531 that of some integer type.
5533 @item -freg-struct-return
5534 Use the convention that @code{struct} and @code{union} values are
5535 returned in registers when possible. This is more efficient for small
5536 structures than @samp{-fpcc-struct-return}.
5538 If you specify neither @samp{-fpcc-struct-return} nor its contrary
5539 @samp{-freg-struct-return}, GNU CC defaults to whichever convention is
5540 standard for the target. If there is no standard convention, GNU CC
5541 defaults to @samp{-fpcc-struct-return}, except on targets where GNU CC
5542 is the principal compiler. In those cases, we can choose the standard,
5543 and we chose the more efficient register return alternative.
5546 Allocate to an @code{enum} type only as many bytes as it needs for the
5547 declared range of possible values. Specifically, the @code{enum} type
5548 will be equivalent to the smallest integer type which has enough room.
5550 @item -fshort-double
5551 Use the same size for @code{double} as for @code{float}.
5554 Requests that the data and non-@code{const} variables of this
5555 compilation be shared data rather than private data. The distinction
5556 makes sense only on certain operating systems, where shared data is
5557 shared between processes running the same program, while private data
5558 exists in one copy per process.
5561 Allocate even uninitialized global variables in the bss section of the
5562 object file, rather than generating them as common blocks. This has the
5563 effect that if the same variable is declared (without @code{extern}) in
5564 two different compilations, you will get an error when you link them.
5565 The only reason this might be useful is if you wish to verify that the
5566 program will work on other systems which always work this way.
5569 Ignore the @samp{#ident} directive.
5571 @item -fno-gnu-linker
5572 Do not output global initializations (such as C++ constructors and
5573 destructors) in the form used by the GNU linker (on systems where the GNU
5574 linker is the standard method of handling them). Use this option when
5575 you want to use a non-GNU linker, which also requires using the
5576 @code{collect2} program to make sure the system linker includes
5577 constructors and destructors. (@code{collect2} is included in the GNU CC
5578 distribution.) For systems which @emph{must} use @code{collect2}, the
5579 compiler driver @code{gcc} is configured to do this automatically.
5581 @item -finhibit-size-directive
5582 Don't output a @code{.size} assembler directive, or anything else that
5583 would cause trouble if the function is split in the middle, and the
5584 two halves are placed at locations far apart in memory. This option is
5585 used when compiling @file{crtstuff.c}; you should not need to use it
5589 Put extra commentary information in the generated assembly code to
5590 make it more readable. This option is generally only of use to those
5591 who actually need to read the generated assembly code (perhaps while
5592 debugging the compiler itself).
5594 @samp{-fno-verbose-asm}, the default, causes the
5595 extra information to be omitted and is useful when comparing two assembler
5599 Consider all memory references through pointers to be volatile.
5601 @item -fvolatile-global
5602 Consider all memory references to extern and global data items to
5606 @cindex global offset table
5608 Generate position-independent code (PIC) suitable for use in a shared
5609 library, if supported for the target machine. Such code accesses all
5610 constant addresses through a global offset table (GOT). The dynamic
5611 loader resolves the GOT entries when the program starts (the dynamic
5612 loader is not part of GNU CC; it is part of the operating system). If
5613 the GOT size for the linked executable exceeds a machine-specific
5614 maximum size, you get an error message from the linker indicating that
5615 @samp{-fpic} does not work; in that case, recompile with @samp{-fPIC}
5616 instead. (These maximums are 16k on the m88k, 8k on the Sparc, and 32k
5617 on the m68k and RS/6000. The 386 has no such limit.)
5619 Position-independent code requires special support, and therefore works
5620 only on certain machines. For the 386, GNU CC supports PIC for System V
5621 but not for the Sun 386i. Code generated for the IBM RS/6000 is always
5622 position-independent.
5625 If supported for the target machine, emit position-independent code,
5626 suitable for dynamic linking and avoiding any limit on the size of the
5627 global offset table. This option makes a difference on the m68k, m88k,
5630 Position-independent code requires special support, and therefore works
5631 only on certain machines.
5633 @item -ffixed-@var{reg}
5634 Treat the register named @var{reg} as a fixed register; generated code
5635 should never refer to it (except perhaps as a stack pointer, frame
5636 pointer or in some other fixed role).
5638 @var{reg} must be the name of a register. The register names accepted
5639 are machine-specific and are defined in the @code{REGISTER_NAMES}
5640 macro in the machine description macro file.
5642 This flag does not have a negative form, because it specifies a
5645 @item -fcall-used-@var{reg}
5646 Treat the register named @var{reg} as an allocable register that is
5647 clobbered by function calls. It may be allocated for temporaries or
5648 variables that do not live across a call. Functions compiled this way
5649 will not save and restore the register @var{reg}.
5651 Use of this flag for a register that has a fixed pervasive role in the
5652 machine's execution model, such as the stack pointer or frame pointer,
5653 will produce disastrous results.
5655 This flag does not have a negative form, because it specifies a
5658 @item -fcall-saved-@var{reg}
5659 Treat the register named @var{reg} as an allocable register saved by
5660 functions. It may be allocated even for temporaries or variables that
5661 live across a call. Functions compiled this way will save and restore
5662 the register @var{reg} if they use it.
5664 Use of this flag for a register that has a fixed pervasive role in the
5665 machine's execution model, such as the stack pointer or frame pointer,
5666 will produce disastrous results.
5668 A different sort of disaster will result from the use of this flag for
5669 a register in which function values may be returned.
5671 This flag does not have a negative form, because it specifies a
5675 Pack all structure members together without holes. Usually you would
5676 not want to use this option, since it makes the code suboptimal, and
5677 the offsets of structure members won't agree with system libraries.
5679 @item -fcheck-memory-usage
5680 Generate extra code to check each memory access. GNU CC will generate
5681 code that is suitable for a detector of bad memory accesses such as
5682 @file{Checker}. If you specify this option, you can not use the
5683 @code{asm} or @code{__asm__} keywords.
5685 You must also specify this option when you compile functions you call that
5686 have side effects. If you do not, you may get erroneous messages from
5687 the detector. Normally, you should compile all your code with this option.
5688 If you use functions from a library that have side-effects (such as
5689 @code{read}), you may not be able to recompile the library and
5690 specify this option. In that case, you can enable the
5691 @samp{-fprefix-function-name} option, which requests GNU CC to encapsulate
5692 your code and make other functions look as if they were compiled with
5693 @samp{-fcheck-memory-usage}. This is done by calling ``stubs'',
5694 which are provided by the detector. If you cannot find or build
5695 stubs for every function you call, you may have to specify
5696 @samp{-fcheck-memory-usage} without @samp{-fprefix-function-name}.
5698 @item -fprefix-function-name
5699 Request GNU CC to add a prefix to the symbols generated for function names.
5700 GNU CC adds a prefix to the names of functions defined as well as
5701 functions called. Code compiled with this option and code compiled
5702 without the option can't be linked together, unless or stubs are used.
5704 If you compile the following code with @samp{-fprefix-function-name}
5706 extern void bar (int);
5716 GNU CC will compile the code as if it was written:
5718 extern void prefix_bar (int);
5722 return prefix_bar (a + 5);
5725 This option is designed to be used with @samp{-fcheck-memory-usage}.
5728 Generate code to verify that you do not go beyond the boundary of the
5729 stack. You should specify this flag if you are running in an
5730 environment with multiple threads, but only rarely need to specify it in
5731 a single-threaded environment since stack overflow is automatically
5732 detected on nearly all systems if there is only one stack.
5735 Enable exception handling. For some targets, this implies
5736 generation of frame unwind information for all functions, which can produce
5737 significant data size overhead, though it does not affect execution.
5739 This option is on by default for languages that support exception
5740 handling (such as C++), and off for those that don't (such as C).
5744 Control whether virtual function definitions in classes are used to
5745 generate code, or only to define interfaces for their callers. (C++
5748 These options are provided for compatibility with @code{cfront} 1.x
5749 usage; the recommended alternative GNU C++ usage is in flux. @xref{C++
5750 Interface,,Declarations and Definitions in One Header}.
5752 With @samp{+e0}, virtual function definitions in classes are declared
5753 @code{extern}; the declaration is used only as an interface
5754 specification, not to generate code for the virtual functions (in this
5757 With @samp{+e1}, G++ actually generates the code implementing virtual
5758 functions defined in the code, and makes them publicly visible.
5760 @cindex aliasing of parameters
5761 @cindex parameters, aliased
5762 @item -fargument-alias
5763 @itemx -fargument-noalias
5764 @itemx -fargument-noalias-global
5765 Specify the possible relationships among parameters and between
5766 parameters and global data.
5768 @samp{-fargument-alias} specifies that arguments (parameters) may
5769 alias each other and may alias global storage.
5770 @samp{-fargument-noalias} specifies that arguments do not alias
5771 each other, but may alias global storage.
5772 @samp{-fargument-noalias-global} specifies that arguments do not
5773 alias each other and do not alias global storage.
5775 Each language will automatically use whatever option is required by
5776 the language standard. You should not need to use these options yourself.
5779 @node Environment Variables
5780 @section Environment Variables Affecting GNU CC
5781 @cindex environment variables
5783 This section describes several environment variables that affect how GNU
5784 CC operates. They work by specifying directories or prefixes to use
5785 when searching for various kinds of files.
5788 Note that you can also specify places to search using options such as
5789 @samp{-B}, @samp{-I} and @samp{-L} (@pxref{Directory Options}). These
5790 take precedence over places specified using environment variables, which
5791 in turn take precedence over those specified by the configuration of GNU
5795 Note that you can also specify places to search using options such as
5796 @samp{-B}, @samp{-I} and @samp{-L} (@pxref{Directory Options}). These
5797 take precedence over places specified using environment variables, which
5798 in turn take precedence over those specified by the configuration of GNU
5805 If @code{TMPDIR} is set, it specifies the directory to use for temporary
5806 files. GNU CC uses temporary files to hold the output of one stage of
5807 compilation which is to be used as input to the next stage: for example,
5808 the output of the preprocessor, which is the input to the compiler
5811 @item GCC_EXEC_PREFIX
5812 @findex GCC_EXEC_PREFIX
5813 If @code{GCC_EXEC_PREFIX} is set, it specifies a prefix to use in the
5814 names of the subprograms executed by the compiler. No slash is added
5815 when this prefix is combined with the name of a subprogram, but you can
5816 specify a prefix that ends with a slash if you wish.
5818 If GNU CC cannot find the subprogram using the specified prefix, it
5819 tries looking in the usual places for the subprogram.
5821 The default value of @code{GCC_EXEC_PREFIX} is
5822 @file{@var{prefix}/lib/gcc-lib/} where @var{prefix} is the value
5823 of @code{prefix} when you ran the @file{configure} script.
5825 Other prefixes specified with @samp{-B} take precedence over this prefix.
5827 This prefix is also used for finding files such as @file{crt0.o} that are
5830 In addition, the prefix is used in an unusual way in finding the
5831 directories to search for header files. For each of the standard
5832 directories whose name normally begins with @samp{/usr/local/lib/gcc-lib}
5833 (more precisely, with the value of @code{GCC_INCLUDE_DIR}), GNU CC tries
5834 replacing that beginning with the specified prefix to produce an
5835 alternate directory name. Thus, with @samp{-Bfoo/}, GNU CC will search
5836 @file{foo/bar} where it would normally search @file{/usr/local/lib/bar}.
5837 These alternate directories are searched first; the standard directories
5841 @findex COMPILER_PATH
5842 The value of @code{COMPILER_PATH} is a colon-separated list of
5843 directories, much like @code{PATH}. GNU CC tries the directories thus
5844 specified when searching for subprograms, if it can't find the
5845 subprograms using @code{GCC_EXEC_PREFIX}.
5848 @findex LIBRARY_PATH
5849 The value of @code{LIBRARY_PATH} is a colon-separated list of
5850 directories, much like @code{PATH}. When configured as a native compiler,
5851 GNU CC tries the directories thus specified when searching for special
5852 linker files, if it can't find them using @code{GCC_EXEC_PREFIX}. Linking
5853 using GNU CC also uses these directories when searching for ordinary
5854 libraries for the @samp{-l} option (but directories specified with
5855 @samp{-L} come first).
5857 @item C_INCLUDE_PATH
5858 @itemx CPLUS_INCLUDE_PATH
5859 @itemx OBJC_INCLUDE_PATH
5860 @findex C_INCLUDE_PATH
5861 @findex CPLUS_INCLUDE_PATH
5862 @findex OBJC_INCLUDE_PATH
5863 @c @itemx OBJCPLUS_INCLUDE_PATH
5864 These environment variables pertain to particular languages. Each
5865 variable's value is a colon-separated list of directories, much like
5866 @code{PATH}. When GNU CC searches for header files, it tries the
5867 directories listed in the variable for the language you are using, after
5868 the directories specified with @samp{-I} but before the standard header
5871 @item DEPENDENCIES_OUTPUT
5872 @findex DEPENDENCIES_OUTPUT
5873 @cindex dependencies for make as output
5874 If this variable is set, its value specifies how to output dependencies
5875 for Make based on the header files processed by the compiler. This
5876 output looks much like the output from the @samp{-M} option
5877 (@pxref{Preprocessor Options}), but it goes to a separate file, and is
5878 in addition to the usual results of compilation.
5880 The value of @code{DEPENDENCIES_OUTPUT} can be just a file name, in
5881 which case the Make rules are written to that file, guessing the target
5882 name from the source file name. Or the value can have the form
5883 @samp{@var{file} @var{target}}, in which case the rules are written to
5884 file @var{file} using @var{target} as the target name.
5887 @node Running Protoize
5888 @section Running Protoize
5890 The program @code{protoize} is an optional part of GNU C. You can use
5891 it to add prototypes to a program, thus converting the program to ANSI
5892 C in one respect. The companion program @code{unprotoize} does the
5893 reverse: it removes argument types from any prototypes that are found.
5895 When you run these programs, you must specify a set of source files as
5896 command line arguments. The conversion programs start out by compiling
5897 these files to see what functions they define. The information gathered
5898 about a file @var{foo} is saved in a file named @file{@var{foo}.X}.
5900 After scanning comes actual conversion. The specified files are all
5901 eligible to be converted; any files they include (whether sources or
5902 just headers) are eligible as well.
5904 But not all the eligible files are converted. By default,
5905 @code{protoize} and @code{unprotoize} convert only source and header
5906 files in the current directory. You can specify additional directories
5907 whose files should be converted with the @samp{-d @var{directory}}
5908 option. You can also specify particular files to exclude with the
5909 @samp{-x @var{file}} option. A file is converted if it is eligible, its
5910 directory name matches one of the specified directory names, and its
5911 name within the directory has not been excluded.
5913 Basic conversion with @code{protoize} consists of rewriting most
5914 function definitions and function declarations to specify the types of
5915 the arguments. The only ones not rewritten are those for varargs
5918 @code{protoize} optionally inserts prototype declarations at the
5919 beginning of the source file, to make them available for any calls that
5920 precede the function's definition. Or it can insert prototype
5921 declarations with block scope in the blocks where undeclared functions
5924 Basic conversion with @code{unprotoize} consists of rewriting most
5925 function declarations to remove any argument types, and rewriting
5926 function definitions to the old-style pre-ANSI form.
5928 Both conversion programs print a warning for any function declaration or
5929 definition that they can't convert. You can suppress these warnings
5932 The output from @code{protoize} or @code{unprotoize} replaces the
5933 original source file. The original file is renamed to a name ending
5934 with @samp{.save}. If the @samp{.save} file already exists, then
5935 the source file is simply discarded.
5937 @code{protoize} and @code{unprotoize} both depend on GNU CC itself to
5938 scan the program and collect information about the functions it uses.
5939 So neither of these programs will work until GNU CC is installed.
5941 Here is a table of the options you can use with @code{protoize} and
5942 @code{unprotoize}. Each option works with both programs unless
5946 @item -B @var{directory}
5947 Look for the file @file{SYSCALLS.c.X} in @var{directory}, instead of the
5948 usual directory (normally @file{/usr/local/lib}). This file contains
5949 prototype information about standard system functions. This option
5950 applies only to @code{protoize}.
5952 @item -c @var{compilation-options}
5953 Use @var{compilation-options} as the options when running @code{gcc} to
5954 produce the @samp{.X} files. The special option @samp{-aux-info} is
5955 always passed in addition, to tell @code{gcc} to write a @samp{.X} file.
5957 Note that the compilation options must be given as a single argument to
5958 @code{protoize} or @code{unprotoize}. If you want to specify several
5959 @code{gcc} options, you must quote the entire set of compilation options
5960 to make them a single word in the shell.
5962 There are certain @code{gcc} arguments that you cannot use, because they
5963 would produce the wrong kind of output. These include @samp{-g},
5964 @samp{-O}, @samp{-c}, @samp{-S}, and @samp{-o} If you include these in
5965 the @var{compilation-options}, they are ignored.
5968 Rename files to end in @samp{.C} instead of @samp{.c}.
5969 This is convenient if you are converting a C program to C++.
5970 This option applies only to @code{protoize}.
5973 Add explicit global declarations. This means inserting explicit
5974 declarations at the beginning of each source file for each function
5975 that is called in the file and was not declared. These declarations
5976 precede the first function definition that contains a call to an
5977 undeclared function. This option applies only to @code{protoize}.
5979 @item -i @var{string}
5980 Indent old-style parameter declarations with the string @var{string}.
5981 This option applies only to @code{protoize}.
5983 @code{unprotoize} converts prototyped function definitions to old-style
5984 function definitions, where the arguments are declared between the
5985 argument list and the initial @samp{@{}. By default, @code{unprotoize}
5986 uses five spaces as the indentation. If you want to indent with just
5987 one space instead, use @samp{-i " "}.
5990 Keep the @samp{.X} files. Normally, they are deleted after conversion
5994 Add explicit local declarations. @code{protoize} with @samp{-l} inserts
5995 a prototype declaration for each function in each block which calls the
5996 function without any declaration. This option applies only to
6000 Make no real changes. This mode just prints information about the conversions
6001 that would have been done without @samp{-n}.
6004 Make no @samp{.save} files. The original files are simply deleted.
6005 Use this option with caution.
6007 @item -p @var{program}
6008 Use the program @var{program} as the compiler. Normally, the name
6012 Work quietly. Most warnings are suppressed.
6015 Print the version number, just like @samp{-v} for @code{gcc}.
6018 If you need special compiler options to compile one of your program's
6019 source files, then you should generate that file's @samp{.X} file
6020 specially, by running @code{gcc} on that source file with the
6021 appropriate options and the option @samp{-aux-info}. Then run
6022 @code{protoize} on the entire set of files. @code{protoize} will use
6023 the existing @samp{.X} file because it is newer than the source file.
6027 gcc -Dfoo=bar file1.c -aux-info
6032 You need to include the special files along with the rest in the
6033 @code{protoize} command, even though their @samp{.X} files already
6034 exist, because otherwise they won't get converted.
6036 @xref{Protoize Caveats}, for more information on how to use
6037 @code{protoize} successfully.